778 files changed, 425928 insertions, 172747 deletions

diff --git a/xlators/Makefile.am b/xlators/Makefile.am
index 4c94f5e44c1..ef20cbb64fa 100644
--- a/xlators/Makefile.am
+++ b/xlators/Makefile.am
@@ -1,3 +1,13 @@
-SUBDIRS = cluster storage protocol performance debug features encryption mount nfs mgmt
+if BUILD_GNFS
+  GNFS_DIR = nfs
+endif
+
+DIST_SUBDIRS = cluster storage protocol performance debug features \
+          mount nfs mgmt system playground meta
+
+SUBDIRS = cluster storage protocol performance debug features \
+          mount ${GNFS_DIR} mgmt system playground meta
+
+EXTRA_DIST = xlator.sym
 
 CLEANFILES =
diff --git a/xlators/bindings/Makefile.am b/xlators/bindings/Makefile.am
deleted file mode 100644
index f7766580257..00000000000
--- a/xlators/bindings/Makefile.am
+++ /dev/null
@@ -1 +0,0 @@
-SUBDIRS = $(BINDINGS_SUBDIRS)
diff --git a/xlators/bindings/python/src/Makefile.am b/xlators/bindings/python/src/Makefile.am
deleted file mode 100644
index c0b9141c667..00000000000
--- a/xlators/bindings/python/src/Makefile.am
+++ /dev/null
@@ -1,19 +0,0 @@
-
-xlator_PROGRAMS = python.so
-
-xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/bindings
-
-python_PYTHON = gluster.py glustertypes.py glusterstack.py
-
-pythondir = $(xlatordir)/python
-
-python_so_SOURCES = python.c
-
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall \
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles \
-	$(PYTHON_CPPLAGS) -DGLUSTER_PYTHON_PATH=\"$(pythondir)\"
-
-AM_LDFLAGS = $(PYTHON_LDFLAGS)
-
-CLEANFILES = 
-
diff --git a/xlators/bindings/python/src/gluster.py b/xlators/bindings/python/src/gluster.py
deleted file mode 100644
index ee0eb131011..00000000000
--- a/xlators/bindings/python/src/gluster.py
+++ /dev/null
@@ -1,47 +0,0 @@
-#   Copyright (c) 2007 Chris AtLee <chris@atlee.ca>
-#   This file is part of GlusterFS.
-#
-#   GlusterFS is free software; you can redistribute it and/or modify
-#   it under the terms of the GNU General Public License as published
-#   by the Free Software Foundation; either version 3 of the License,
-#   or (at your option) any later version.
-#
-#   GlusterFS is distributed in the hope that it will be useful, but
-#   WITHOUT ANY WARRANTY; without even the implied warranty of
-#   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-#   General Public License for more details.
-#
-#   You should have received a copy of the GNU General Public License
-#   along with this program.  If not, see
-#   <http://www.gnu.org/licenses/>.
-from ctypes import *
-from glustertypes import *
-from glusterstack import *
-import sys
-import inspect
-
-libglusterfs = CDLL("libglusterfs.so")
-_gf_log = libglusterfs._gf_log
-_gf_log.restype = c_int32
-_gf_log.argtypes = [c_char_p, c_char_p, c_char_p, c_int32, c_int, c_char_p]
-
-gf_log_loglevel = c_int.in_dll(libglusterfs, "gf_log_loglevel")
-
-GF_LOG_NONE = 0
-GF_LOG_CRITICAL = 1
-GF_LOG_ERROR = 2
-GF_LOG_WARNING = 3
-GF_LOG_DEBUG = 4
-
-def gf_log(module, level, fmt, *params):
-    if level <= gf_log_loglevel:
-        frame = sys._getframe(1)
-        _gf_log(module, frame.f_code.co_filename, frame.f_code.co_name,
-                frame.f_lineno, level, fmt, *params)
-
-class ComplexTranslator(object):
-    def __init__(self, xlator):
-        self.xlator = xlator_t.from_address(xlator)
-
-    def __getattr__(self, item):
-        return getattr(self.xlator, item)
diff --git a/xlators/bindings/python/src/glusterstack.py b/xlators/bindings/python/src/glusterstack.py
deleted file mode 100644
index ba24c81652e..00000000000
--- a/xlators/bindings/python/src/glusterstack.py
+++ /dev/null
@@ -1,55 +0,0 @@
-#   Copyright (c) 2007 Chris AtLee <chris@atlee.ca>
-#   This file is part of GlusterFS.
-#
-#   GlusterFS is free software; you can redistribute it and/or modify
-#   it under the terms of the GNU General Public License as published
-#   by the Free Software Foundation; either version 3 of the License,
-#   or (at your option) any later version.
-#
-#   GlusterFS is distributed in the hope that it will be useful, but
-#   WITHOUT ANY WARRANTY; without even the implied warranty of
-#   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-#   General Public License for more details.
-#
-#   You should have received a copy of the GNU General Public License
-#   along with this program.  If not, see
-#   <http://www.gnu.org/licenses/>.
-from ctypes import *
-from glustertypes import *
-
-libc = CDLL("libc.so.6")
-calloc = libc.calloc
-calloc.argtypes = [c_int, c_int]
-calloc.restype = c_void_p
-
-# TODO: Can these be done in C somehow?
-def stack_wind(frame, rfn, obj, fn, *params):
-    """Frame is a frame object"""
-    _new = cast(calloc(1, sizeof(call_frame_t)), POINTER(call_frame_t))
-    _new[0].root = frame.root
-    _new[0].next = frame.root[0].frames.next
-    _new[0].prev = pointer(frame.root[0].frames)
-    if frame.root[0].frames.next:
-        frame.root[0].frames.next[0].prev = _new
-    frame.root[0].frames.next = _new
-    _new[0].this = obj
-    # TODO: Type checking like tmp_cbk?
-    _new[0].ret = rfn
-    _new[0].parent = pointer(frame)
-    _new[0].cookie = cast(_new, c_void_p)
-    # TODO: Initialize lock
-    #_new.lock.init()
-    frame.ref_count += 1
-    fn(_new, obj, *params)
-
-def stack_unwind(frame, *params):
-    """Frame is a frame object"""
-    fn = frame[0].ret
-    parent = frame[0].parent[0]
-    parent.ref_count -= 1
-
-    op_ret = params[0]
-    op_err = params[1]
-    params = params[2:]
-    fn(parent, call_frame_t.from_address(frame[0].cookie), parent.this,
-            op_ret, op_err, *params)
diff --git a/xlators/bindings/python/src/glustertypes.py b/xlators/bindings/python/src/glustertypes.py
deleted file mode 100644
index e9069d07c72..00000000000
--- a/xlators/bindings/python/src/glustertypes.py
+++ /dev/null
@@ -1,167 +0,0 @@
-#   Copyright (c) 2007 Chris AtLee <chris@atlee.ca>
-#   This file is part of GlusterFS.
-#
-#   GlusterFS is free software; you can redistribute it and/or modify
-#   it under the terms of the GNU General Public License as published
-#   by the Free Software Foundation; either version 3 of the License,
-#   or (at your option) any later version.
-#
-#   GlusterFS is distributed in the hope that it will be useful, but
-#   WITHOUT ANY WARRANTY; without even the implied warranty of
-#   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-#   General Public License for more details.
-#
-#   You should have received a copy of the GNU General Public License
-#   along with this program.  If not, see
-#   <http://www.gnu.org/licenses/>.
-from ctypes import *
-import collections
-
-#
-# Forward declaration of some gluster types
-#
-class call_frame_t(Structure):
-    pass
-
-class call_ctx_t(Structure):
-    pass
-
-class call_pool_t(Structure):
-    pass
-
-class xlator_t(Structure):
-    def _getFirstChild(self):
-        return self.children[0].xlator
-    firstChild = property(_getFirstChild)
-
-class xlator_list_t(Structure):
-    pass
-
-class xlator_fops(Structure):
-    pass
-
-class xlator_mops(Structure):
-    pass
-
-class glusterfs_ctx_t(Structure):
-    pass
-
-class list_head(Structure):
-    pass
-
-class dict_t(Structure):
-    pass
-
-class inode_table_t(Structure):
-    pass
-
-class fd_t(Structure):
-    pass
-
-class iovec(Structure):
-    _fields_ = [
-            ("iov_base", c_void_p),
-            ("iov_len", c_size_t),
-            ]
-
-    def __init__(self, s):
-        self.iov_base = cast(c_char_p(s), c_void_p)
-        self.iov_len = len(s)
-
-    def getBytes(self):
-        return string_at(self.iov_base, self.iov_len)
-
-# This is a pthread_spinlock_t
-# TODO: what happens to volatile-ness?
-gf_lock_t = c_int
-
-uid_t = c_uint32
-gid_t = c_uint32
-pid_t = c_int32
-
-off_t = c_int64
-
-#
-# Function pointer types
-#
-ret_fn_t = CFUNCTYPE(c_int32, POINTER(call_frame_t), POINTER(call_frame_t),
-                              POINTER(xlator_t), c_int32, c_int32)
-
-fini_fn_t = CFUNCTYPE(None, POINTER(xlator_t))
-init_fn_t = CFUNCTYPE(c_int32, POINTER(xlator_t))
-event_notify_fn_t = CFUNCTYPE(c_int32, POINTER(xlator_t), c_int32, c_void_p)
-
-list_head._fields_ = [
-        ("next", POINTER(list_head)),
-        ("prev", POINTER(list_head)),
-        ]
-
-call_frame_t._fields_ = [
-        ("root", POINTER(call_ctx_t)),
-        ("parent", POINTER(call_frame_t)),
-        ("next", POINTER(call_frame_t)),
-        ("prev", POINTER(call_frame_t)),
-        ("local", c_void_p),
-        ("this", POINTER(xlator_t)),
-        ("ret", ret_fn_t),
-        ("ref_count", c_int32),
-        ("lock", gf_lock_t),
-        ("cookie", c_void_p),
-        ("op", c_int32),
-        ("type", c_int8),
-        ]
-
-call_ctx_t._fields_ = [
-        ("all_frames", list_head),
-        ("trans", c_void_p),
-        ("pool", call_pool_t),
-        ("unique", c_uint64),
-        ("state", c_void_p),
-        ("uid", uid_t),
-        ("gid", gid_t),
-        ("pid", pid_t),
-        ("frames", call_frame_t),
-        ("req_refs", POINTER(dict_t)),
-        ("rsp_refs", POINTER(dict_t)),
-        ]
-
-xlator_t._fields_ = [
-        ("name", c_char_p),
-        ("type", c_char_p),
-        ("next", POINTER(xlator_t)),
-        ("prev", POINTER(xlator_t)),
-        ("parent", POINTER(xlator_t)),
-        ("children", POINTER(xlator_list_t)),
-        ("fops", POINTER(xlator_fops)),
-        ("mops", POINTER(xlator_mops)),
-        ("fini", fini_fn_t),
-        ("init", init_fn_t),
-        ("notify", event_notify_fn_t),
-        ("options", POINTER(dict_t)),
-        ("ctx", POINTER(glusterfs_ctx_t)),
-        ("itable", POINTER(inode_table_t)),
-        ("ready", c_char),
-        ("private", c_void_p),
-        ]
-
-xlator_list_t._fields_ = [
-        ("xlator", POINTER(xlator_t)),
-        ("next", POINTER(xlator_list_t)),
-        ]
-
-fop_functions = collections.defaultdict(lambda: c_void_p)
-fop_function_names = ['lookup', 'forget', 'stat', 'fstat', 'chmod', 'fchmod',
-        'chown', 'fchown', 'truncate', 'ftruncate', 'utimens', 'access',
-        'readlink', 'mknod', 'mkdir', 'unlink', 'rmdir', 'symlink',
-        'rename', 'link', 'create', 'open', 'readv', 'writev', 'flush',
-        'close', 'fsync', 'opendir', 'readdir', 'closedir', 'fsyncdir',
-        'statfs', 'setxattr', 'getxattr', 'removexattr', 'lk', 'writedir',
-        # TODO: Call backs?
-        ]
-
-fop_writev_t = CFUNCTYPE(c_int32, POINTER(call_frame_t), POINTER(xlator_t),
-                                  POINTER(fd_t), POINTER(iovec), c_int32,
-                                  off_t)
-
-fop_functions['writev'] = fop_writev_t
-xlator_fops._fields_ = [(f, fop_functions[f]) for f in fop_function_names]
diff --git a/xlators/bindings/python/src/python.c b/xlators/bindings/python/src/python.c
deleted file mode 100644
index c11323cda4f..00000000000
--- a/xlators/bindings/python/src/python.c
+++ /dev/null
@@ -1,232 +0,0 @@
-/*
-   Copyright (c) 2007-2009 Chris AtLee <chris@atlee.ca>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   General Public License for more details.
-
-   You should have received a copy of the GNU General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-#include <Python.h>
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "xlator.h"
-#include "logging.h"
-#include "defaults.h"
-
-typedef struct
-{
-    char        *scriptname;
-    PyObject    *pXlator;
-    PyObject    *pScriptModule;
-    PyObject    *pGlusterModule;
-    PyThreadState *pInterp;
-
-    PyObject    *pFrameType, *pVectorType, *pFdType;
-} python_private_t;
-
-int32_t
-python_writev (call_frame_t *frame,
-              xlator_t *this,
-              fd_t *fd,
-              struct iovec *vector,
-              int32_t count, 
-              off_t offset)
-{
-  python_private_t *priv = (python_private_t *)this->private;
-  gf_log("python", GF_LOG_DEBUG, "In writev");
-  if (PyObject_HasAttrString(priv->pXlator, "writev"))
-  {
-
-      PyObject *retval = PyObject_CallMethod(priv->pXlator, "writev",
-              "O O O i l",
-              PyObject_CallMethod(priv->pFrameType, "from_address", "O&", PyLong_FromVoidPtr, frame),
-              PyObject_CallMethod(priv->pFdType, "from_address", "O&", PyLong_FromVoidPtr, fd),
-              PyObject_CallMethod(priv->pVectorType, "from_address", "O&", PyLong_FromVoidPtr, vector),
-              count,
-              offset);
-      if (PyErr_Occurred())
-      {
-          PyErr_Print();
-      }
-      Py_XDECREF(retval);
-  }
-  else
-  {
-      return default_writev(frame, this, fd, vector, count, offset);
-  }
-  return 0;
-}
-
-struct xlator_fops fops = {
-    .writev       = python_writev
-};
-
-static PyObject *
-AnonModule_FromFile (const char* fname)
-{
-    // Get the builtins
-    PyThreadState* pThread = PyThreadState_Get();
-    PyObject *pBuiltins = pThread->interp->builtins;
-
-    if (PyErr_Occurred())
-    {
-        PyErr_Print();
-        return NULL;
-    }
-
-    // Create a new dictionary for running code in
-    PyObject *pModuleDict = PyDict_New();
-    PyDict_SetItemString(pModuleDict, "__builtins__", pBuiltins);
-    Py_INCREF(pBuiltins);
-
-    // Run the file in the new context
-    FILE* fp = fopen(fname, "r");
-    PyRun_File(fp, fname, Py_file_input, pModuleDict, pModuleDict);
-    fclose(fp);
-    if (PyErr_Occurred())
-    {
-        PyErr_Print();
-        Py_DECREF(pModuleDict);
-        Py_DECREF(pBuiltins);
-        return NULL;
-    }
-
-    // Create an object to hold the new context
-    PyRun_String("class ModuleWrapper(object):\n\tpass\n", Py_single_input, pModuleDict, pModuleDict);
-    if (PyErr_Occurred())
-    {
-        PyErr_Print();
-        Py_DECREF(pModuleDict);
-        Py_DECREF(pBuiltins);
-        return NULL;
-    }
-    PyObject *pModule = PyRun_String("ModuleWrapper()", Py_eval_input, pModuleDict, pModuleDict);
-    if (PyErr_Occurred())
-    {
-        PyErr_Print();
-        Py_DECREF(pModuleDict);
-        Py_DECREF(pBuiltins);
-        Py_XDECREF(pModule);
-        return NULL;
-    }
-
-    // Set the new context's dictionary to the one we used to run the code
-    // inside
-    PyObject_SetAttrString(pModule, "__dict__", pModuleDict);
-    if (PyErr_Occurred())
-    {
-        PyErr_Print();
-        Py_DECREF(pModuleDict);
-        Py_DECREF(pBuiltins);
-        Py_DECREF(pModule);
-        return NULL;
-    }
-
-    return pModule;
-}
-
-int32_t
-init (xlator_t *this)
-{
-  // This is ok to call more than once per process
-  Py_InitializeEx(0);
-
-  if (!this->children) {
-    gf_log ("python", GF_LOG_ERROR, 
-            "FATAL: python should have exactly one child");
-    return -1;
-  }
-
-  python_private_t *priv = CALLOC (sizeof (python_private_t), 1);
-  ERR_ABORT (priv);
-
-  data_t *scriptname = dict_get (this->options, "scriptname");
-  if (scriptname) {
-      priv->scriptname = data_to_str(scriptname);
-  } else {
-      gf_log("python", GF_LOG_ERROR,
-              "FATAL: python requires the scriptname parameter");
-      return -1;
-  }
-
-  priv->pInterp = Py_NewInterpreter();
-    
-  // Adjust python's path
-  PyObject *syspath = PySys_GetObject("path");
-  PyObject *path = PyString_FromString(GLUSTER_PYTHON_PATH);
-  PyList_Append(syspath, path);
-  Py_DECREF(path);
-
-  gf_log("python", GF_LOG_DEBUG,
-          "Loading gluster module");
-
-  priv->pGlusterModule = PyImport_ImportModule("gluster");
-  if (PyErr_Occurred())
-  {
-      PyErr_Print();
-      return -1;
-  }
-
-  priv->pFrameType = PyObject_GetAttrString(priv->pGlusterModule, "call_frame_t");
-  priv->pFdType = PyObject_GetAttrString(priv->pGlusterModule, "fd_t");
-  priv->pVectorType = PyObject_GetAttrString(priv->pGlusterModule, "iovec");
-
-  gf_log("python", GF_LOG_DEBUG, "Loading script...%s", priv->scriptname);
-  
-  priv->pScriptModule = AnonModule_FromFile(priv->scriptname);
-  if (!priv->pScriptModule || PyErr_Occurred())
-  {
-      gf_log("python", GF_LOG_ERROR, "Error loading %s", priv->scriptname);
-      PyErr_Print();
-      return -1;
-  }
-
-  if (!PyObject_HasAttrString(priv->pScriptModule, "xlator"))
-  {
-      gf_log("python", GF_LOG_ERROR, "%s does not have a xlator attribute", priv->scriptname);
-      return -1;
-  }
-  gf_log("python", GF_LOG_DEBUG, "Instantiating translator");
-  priv->pXlator = PyObject_CallMethod(priv->pScriptModule, "xlator", "O&",
-          PyLong_FromVoidPtr, this);
-  if (PyErr_Occurred() || !priv->pXlator)
-  {
-      PyErr_Print();
-      return -1;
-  }
-
-  this->private = priv;
-
-  gf_log ("python", GF_LOG_DEBUG, "python xlator loaded");
-  return 0;
-}
-
-void 
-fini (xlator_t *this)
-{
-  python_private_t *priv = (python_private_t*)(this->private);
-  Py_DECREF(priv->pXlator);
-  Py_DECREF(priv->pScriptModule);
-  Py_DECREF(priv->pGlusterModule);
-  Py_DECREF(priv->pFrameType);
-  Py_DECREF(priv->pFdType);
-  Py_DECREF(priv->pVectorType);
-  Py_EndInterpreter(priv->pInterp);
-  return;
-}
diff --git a/xlators/bindings/python/src/testxlator.py b/xlators/bindings/python/src/testxlator.py
deleted file mode 100644
index 507455c856a..00000000000
--- a/xlators/bindings/python/src/testxlator.py
+++ /dev/null
@@ -1,56 +0,0 @@
-#   Copyright (c) 2007 Chris AtLee <chris@atlee.ca>
-#   This file is part of GlusterFS.
-#
-#   GlusterFS is free software; you can redistribute it and/or modify
-#   it under the terms of the GNU General Public License as published
-#   by the Free Software Foundation; either version 3 of the License,
-#   or (at your option) any later version.
-#
-#   GlusterFS is distributed in the hope that it will be useful, but
-#   WITHOUT ANY WARRANTY; without even the implied warranty of
-#   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-#   General Public License for more details.
-#
-#   You should have received a copy of the GNU General Public License
-#   along with this program.  If not, see
-#   <http://www.gnu.org/licenses/>.
-
-"""
-This is a test translator written in python.
-
-Important things to note:
-    This file must be import-able from glusterfsd.  This probably means
-    setting PYTHONPATH to where this file is located.
-
-    This file must have a top-level xlator class object that will be
-    used to instantiate individual translators.
-"""
-from gluster import *
-
-class MyXlator(ComplexTranslator):
-    name = "MyXlator"
-    def writev_cbk(self, frame, cookie, op_ret, op_errno, buf):
-        stack_unwind(frame, op_ret, op_errno, buf)
-        return 0
-
-    def writev(self, frame, fd, vector, count, offset):
-        gf_log(self.name, GF_LOG_WARNING, "writev %i bytes", vector.iov_len)
-        # TODO: Use cookie to pass this to writev_cbk
-        old_count = vector.iov_len
-
-        data = vector.getBytes().encode("zlib")
-
-        vector = iovec(data)
-        gf_log(self.name, GF_LOG_WARNING, "writev %i bytes", vector.iov_len)
-
-        @ret_fn_t
-        def rfn(frame, prev, this, op_ret, op_errno, *params):
-            if len(params) == 0:
-                params = [0]
-            return self.writev_cbk(frame, prev, old_count, op_errno, *params)
-
-        stack_wind(frame, rfn, self.firstChild,
-                self.firstChild[0].fops[0].writev, fd, vector, count, offset)
-        return 0
-
-xlator = MyXlator
diff --git a/xlators/cluster/Makefile.am b/xlators/cluster/Makefile.am
index 0990822a7d3..8e067d5ab58 100644
--- a/xlators/cluster/Makefile.am
+++ b/xlators/cluster/Makefile.am
@@ -1,3 +1,3 @@
-SUBDIRS = stripe afr dht
+SUBDIRS = afr dht ec
 
 CLEANFILES = 
diff --git a/xlators/cluster/afr/src/Makefile.am b/xlators/cluster/afr/src/Makefile.am
index 4e4b4c75260..610819b28fc 100644
--- a/xlators/cluster/afr/src/Makefile.am
+++ b/xlators/cluster/afr/src/Makefile.am
@@ -1,26 +1,35 @@
-xlator_LTLIBRARIES = afr.la pump.la
+xlator_LTLIBRARIES = afr.la
 xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/cluster
 
-afr_common_source = afr-dir-read.c afr-dir-write.c afr-inode-read.c afr-inode-write.c afr-open.c afr-transaction.c afr-self-heal-data.c afr-self-heal-common.c afr-self-heal-metadata.c afr-self-heal-entry.c afr-self-heal-algorithm.c afr-lk-common.c
+afr_common_source = afr-dir-read.c afr-dir-write.c afr-inode-read.c \
+	afr-inode-write.c afr-open.c afr-transaction.c afr-lk-common.c \
+	afr-read-txn.c \
+	$(top_builddir)/xlators/lib/src/libxlator.c
 
-afr_la_LDFLAGS = -module -avoidversion
-afr_la_SOURCES = $(afr_common_source) afr.c
+AFR_SELFHEAL_SOURCES = afr-self-heal-common.c afr-self-heal-data.c \
+	afr-self-heal-entry.c afr-self-heal-metadata.c afr-self-heald.c \
+	afr-self-heal-name.c
+
+afr_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+afr_la_SOURCES = $(afr_common_source) $(AFR_SELFHEAL_SOURCES) afr.c
 afr_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
-pump_la_LDFLAGS = -module -avoidversion
-pump_la_SOURCES =  $(afr_common_source) pump.c
-pump_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+noinst_HEADERS = afr.h afr-transaction.h afr-inode-write.h afr-inode-read.h \
+	afr-dir-read.h afr-dir-write.h afr-self-heal.h afr-mem-types.h \
+	afr-common.c afr-self-heald.h \
+	$(top_builddir)/xlators/lib/src/libxlator.h afr-messages.h
 
-noinst_HEADERS = afr.h afr-transaction.h afr-inode-write.h afr-inode-read.h afr-dir-read.h afr-dir-write.h afr-self-heal.h afr-self-heal-common.h afr-self-heal-algorithm.h pump.h afr-mem-types.h afr-common.c
+AM_CPPFLAGS = $(GF_CPPFLAGS) \
+	-I$(top_srcdir)/libglusterfs/src -I$(top_srcdir)/xlators/lib/src \
+	-I$(top_srcdir)/rpc/rpc-lib/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS) \
-	    -I$(top_srcdir)/libglusterfs/src -I$(top_srcdir)/contrib/md5 -shared -nostartfiles $(GF_CFLAGS)
+AM_CFLAGS = -Wall $(GF_CFLAGS)
 
 CLEANFILES =
 
 uninstall-local:
 	rm -f $(DESTDIR)$(xlatordir)/replicate.so
-	rm -f $(DESTDIR)$(xlatordir)/pump.so
 
 install-data-hook:
 	ln -sf afr.so $(DESTDIR)$(xlatordir)/replicate.so
diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c
index bd4e9d66d35..032ab5c8001 100644
--- a/xlators/cluster/afr/src/afr-common.c
+++ b/xlators/cluster/afr/src/afr-common.c
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2007-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #include <libgen.h>
@@ -24,28 +15,20 @@
 #include <stdlib.h>
 #include <signal.h>
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
+#include <glusterfs/glusterfs.h>
 #include "afr.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-#include "byte-order.h"
-#include "statedump.h"
-
-#include "fd.h"
+#include <glusterfs/dict.h>
+#include <glusterfs/hashfn.h>
+#include <glusterfs/list.h>
+#include <glusterfs/call-stub.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/byte-order.h>
+#include <glusterfs/statedump.h>
+#include <glusterfs/events.h>
+#include <glusterfs/upcall-utils.h>
 
 #include "afr-inode-read.h"
 #include "afr-inode-write.h"
@@ -53,2558 +36,7843 @@
 #include "afr-dir-write.h"
 #include "afr-transaction.h"
 #include "afr-self-heal.h"
-#include "afr-self-heal-common.h"
-#include "pump.h"
-
-#define AFR_ICTX_OPENDIR_DONE_MASK     0x0000000200000000ULL
-#define AFR_ICTX_SPLIT_BRAIN_MASK      0x0000000100000000ULL
-#define AFR_ICTX_READ_CHILD_MASK       0x00000000FFFFFFFFULL
+#include "afr-self-heald.h"
+#include "afr-messages.h"
 
 int32_t
-afr_set_dict_gfid (dict_t *dict, uuid_t gfid)
+afr_quorum_errno(afr_private_t *priv)
 {
-        int ret = 0;
-
-        GF_ASSERT (gfid);
+    return ENOTCONN;
+}
 
-        ret = dict_set_static_bin (dict, "gfid-req", gfid, 16);
-        if (ret)
-                gf_log (THIS->name, GF_LOG_DEBUG, "gfid set failed");
+gf_boolean_t
+afr_is_private_directory(afr_private_t *priv, uuid_t pargfid, const char *name,
+                         pid_t pid)
+{
+    if (!__is_root_gfid(pargfid)) {
+        return _gf_false;
+    }
+
+    if (strcmp(name, GF_REPLICATE_TRASH_DIR) == 0) {
+        /*For backward compatibility /.landfill is private*/
+        return _gf_true;
+    }
+
+    if (pid == GF_CLIENT_PID_GSYNCD) {
+        /*geo-rep needs to create/sync private directory on slave because
+         * it appears in changelog*/
+        return _gf_false;
+    }
+
+    if (pid == GF_CLIENT_PID_GLFS_HEAL || pid == GF_CLIENT_PID_SELF_HEALD) {
+        if (strcmp(name, priv->anon_inode_name) == 0) {
+            /* anonymous-inode dir is private*/
+            return _gf_true;
+        }
+    } else {
+        if (strncmp(name, AFR_ANON_DIR_PREFIX, strlen(AFR_ANON_DIR_PREFIX)) ==
+            0) {
+            /* anonymous-inode dir prefix is private for geo-rep to work*/
+            return _gf_true;
+        }
+    }
 
-        return ret;
+    return _gf_false;
 }
 
-uint64_t
-afr_is_split_brain (xlator_t *this, inode_t *inode)
+void
+afr_fill_success_replies(afr_local_t *local, afr_private_t *priv,
+                         unsigned char *replies)
 {
-        int ret = 0;
+    int i = 0;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->replies[i].valid && local->replies[i].op_ret == 0) {
+            replies[i] = 1;
+        } else {
+            replies[i] = 0;
+        }
+    }
+}
 
-        uint64_t ctx         = 0;
-        uint64_t split_brain = 0;
+int
+afr_fav_child_reset_sink_xattrs(void *opaque);
 
-        VALIDATE_OR_GOTO (inode, out);
+int
+afr_fav_child_reset_sink_xattrs_cbk(int ret, call_frame_t *frame, void *opaque);
 
-        LOCK (&inode->lock);
-        {
-                ret = __inode_ctx_get (inode, this, &ctx);
+static void
+afr_discover_done(call_frame_t *frame, xlator_t *this);
 
-                if (ret < 0)
-                        goto unlock;
+int
+afr_dom_lock_acquire_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int op_ret, int op_errno, dict_t *xdata)
+{
+    afr_local_t *local = frame->local;
+    afr_private_t *priv = this->private;
+    int i = (long)cookie;
+
+    local->cont.lk.dom_lock_op_ret[i] = op_ret;
+    local->cont.lk.dom_lock_op_errno[i] = op_errno;
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, AFR_MSG_LK_HEAL_DOM,
+               "%s: Failed to acquire %s on %s",
+               uuid_utoa(local->fd->inode->gfid), AFR_LK_HEAL_DOM,
+               priv->children[i]->name);
+    } else {
+        local->cont.lk.dom_locked_nodes[i] = 1;
+    }
+
+    syncbarrier_wake(&local->barrier);
+
+    return 0;
+}
 
-                split_brain = ctx & AFR_ICTX_SPLIT_BRAIN_MASK;
+int
+afr_dom_lock_acquire(call_frame_t *frame)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    struct gf_flock flock = {
+        0,
+    };
+    int i = 0;
+
+    priv = frame->this->private;
+    local = frame->local;
+    local->cont.lk.dom_locked_nodes = GF_CALLOC(
+        priv->child_count, sizeof(*local->cont.lk.locked_nodes),
+        gf_afr_mt_char);
+    if (!local->cont.lk.dom_locked_nodes) {
+        return -ENOMEM;
+    }
+    local->cont.lk.dom_lock_op_ret = GF_CALLOC(
+        priv->child_count, sizeof(*local->cont.lk.dom_lock_op_ret),
+        gf_afr_mt_int32_t);
+    if (!local->cont.lk.dom_lock_op_ret) {
+        return -ENOMEM; /* CALLOC'd members are freed in afr_local_cleanup. */
+    }
+    local->cont.lk.dom_lock_op_errno = GF_CALLOC(
+        priv->child_count, sizeof(*local->cont.lk.dom_lock_op_errno),
+        gf_afr_mt_int32_t);
+    if (!local->cont.lk.dom_lock_op_errno) {
+        return -ENOMEM; /* CALLOC'd members are freed in afr_local_cleanup. */
+    }
+    flock.l_type = F_WRLCK;
+
+    AFR_ONALL(frame, afr_dom_lock_acquire_cbk, finodelk, AFR_LK_HEAL_DOM,
+              local->fd, F_SETLK, &flock, NULL);
+
+    if (!afr_has_quorum(local->cont.lk.dom_locked_nodes, frame->this, NULL))
+        goto blocking_lock;
+
+    /*If any of the bricks returned EAGAIN, we still need blocking locks.*/
+    if (AFR_COUNT(local->cont.lk.dom_locked_nodes, priv->child_count) !=
+        priv->child_count) {
+        for (i = 0; i < priv->child_count; i++) {
+            if (local->cont.lk.dom_lock_op_ret[i] == -1 &&
+                local->cont.lk.dom_lock_op_errno[i] == EAGAIN)
+                goto blocking_lock;
         }
-unlock:
-        UNLOCK (&inode->lock);
+    }
 
-out:
-        return split_brain;
+    return 0;
+
+blocking_lock:
+    afr_dom_lock_release(frame);
+    AFR_ONALL(frame, afr_dom_lock_acquire_cbk, finodelk, AFR_LK_HEAL_DOM,
+              local->fd, F_SETLKW, &flock, NULL);
+    if (!afr_has_quorum(local->cont.lk.dom_locked_nodes, frame->this, NULL)) {
+        afr_dom_lock_release(frame);
+        return -afr_quorum_errno(priv);
+    }
+
+    return 0;
 }
 
+int
+afr_dom_lock_release_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int op_ret, int op_errno, dict_t *xdata)
+{
+    afr_local_t *local = frame->local;
+    afr_private_t *priv = this->private;
+    int i = (long)cookie;
+
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, AFR_MSG_LK_HEAL_DOM,
+               "%s: Failed to release %s on %s", local->loc.path,
+               AFR_LK_HEAL_DOM, priv->children[i]->name);
+    }
+    local->cont.lk.dom_locked_nodes[i] = 0;
+
+    syncbarrier_wake(&local->barrier);
+
+    return 0;
+}
 
 void
-afr_set_split_brain (xlator_t *this, inode_t *inode, gf_boolean_t set)
+afr_dom_lock_release(call_frame_t *frame)
 {
-        uint64_t ctx = 0;
-        int      ret = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    unsigned char *locked_on = NULL;
+    struct gf_flock flock = {
+        0,
+    };
+
+    local = frame->local;
+    priv = frame->this->private;
+    locked_on = local->cont.lk.dom_locked_nodes;
+    if (AFR_COUNT(locked_on, priv->child_count) == 0)
+        return;
+    flock.l_type = F_UNLCK;
 
-        VALIDATE_OR_GOTO (inode, out);
+    AFR_ONLIST(locked_on, frame, afr_dom_lock_release_cbk, finodelk,
+               AFR_LK_HEAL_DOM, local->fd, F_SETLK, &flock, NULL);
 
-        LOCK (&inode->lock);
-        {
-                ret = __inode_ctx_get (inode, this, &ctx);
+    return;
+}
 
-                if (ret < 0) {
-                        ctx = 0;
-                }
+static void
+afr_lk_heal_info_cleanup(afr_lk_heal_info_t *info)
+{
+    if (!info)
+        return;
+    if (info->xdata_req)
+        dict_unref(info->xdata_req);
+    if (info->fd)
+        fd_unref(info->fd);
+    GF_FREE(info->locked_nodes);
+    GF_FREE(info->child_up_event_gen);
+    GF_FREE(info->child_down_event_gen);
+    GF_FREE(info);
+}
 
-                if (set) {
-                        ctx = (~AFR_ICTX_SPLIT_BRAIN_MASK & ctx)
-                                | (0xFFFFFFFFFFFFFFFFULL & AFR_ICTX_SPLIT_BRAIN_MASK);
-                } else {
-                        ctx = (~AFR_ICTX_SPLIT_BRAIN_MASK & ctx);
-                }
-                __inode_ctx_put (inode, this, ctx);
+static int
+afr_add_lock_to_saved_locks(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = this->private;
+    afr_local_t *local = frame->local;
+    afr_lk_heal_info_t *info = NULL;
+    afr_fd_ctx_t *fd_ctx = NULL;
+    int ret = -ENOMEM;
+
+    info = GF_CALLOC(sizeof(*info), 1, gf_afr_mt_lk_heal_info_t);
+    if (!info) {
+        goto cleanup;
+    }
+    INIT_LIST_HEAD(&info->pos);
+    info->fd = fd_ref(local->fd);
+    info->cmd = local->cont.lk.cmd;
+    info->pid = frame->root->pid;
+    info->flock = local->cont.lk.user_flock;
+    info->xdata_req = dict_copy_with_ref(local->xdata_req, NULL);
+    if (!info->xdata_req) {
+        goto cleanup;
+    }
+    info->lk_owner = frame->root->lk_owner;
+    info->locked_nodes = GF_MALLOC(
+        sizeof(*info->locked_nodes) * priv->child_count, gf_afr_mt_char);
+    if (!info->locked_nodes) {
+        goto cleanup;
+    }
+    memcpy(info->locked_nodes, local->cont.lk.locked_nodes,
+           sizeof(*info->locked_nodes) * priv->child_count);
+    info->child_up_event_gen = GF_CALLOC(sizeof(*info->child_up_event_gen),
+                                         priv->child_count, gf_afr_mt_int32_t);
+    if (!info->child_up_event_gen) {
+        goto cleanup;
+    }
+    info->child_down_event_gen = GF_CALLOC(sizeof(*info->child_down_event_gen),
+                                           priv->child_count,
+                                           gf_afr_mt_int32_t);
+    if (!info->child_down_event_gen) {
+        goto cleanup;
+    }
+
+    LOCK(&local->fd->lock);
+    {
+        fd_ctx = __afr_fd_ctx_get(local->fd, this);
+        if (fd_ctx)
+            fd_ctx->lk_heal_info = info;
+    }
+    UNLOCK(&local->fd->lock);
+    if (!fd_ctx) {
+        goto cleanup;
+    }
+
+    LOCK(&priv->lock);
+    {
+        list_add_tail(&info->pos, &priv->saved_locks);
+    }
+    UNLOCK(&priv->lock);
+
+    return 0;
+cleanup:
+    gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_LK_HEAL_DOM,
+           "%s: Failed to add lock to healq",
+           uuid_utoa(local->fd->inode->gfid));
+    if (info) {
+        afr_lk_heal_info_cleanup(info);
+        if (fd_ctx) {
+            LOCK(&local->fd->lock);
+            {
+                fd_ctx->lk_heal_info = NULL;
+            }
+            UNLOCK(&local->fd->lock);
         }
-        UNLOCK (&inode->lock);
+    }
+    return ret;
+}
+
+static int
+afr_remove_lock_from_saved_locks(afr_local_t *local, xlator_t *this)
+{
+    afr_private_t *priv = this->private;
+    struct gf_flock flock = local->cont.lk.user_flock;
+    afr_lk_heal_info_t *info = NULL;
+    afr_fd_ctx_t *fd_ctx = NULL;
+    int ret = -EINVAL;
+
+    fd_ctx = afr_fd_ctx_get(local->fd, this);
+    if (!fd_ctx || !fd_ctx->lk_heal_info) {
+        goto out;
+    }
+
+    info = fd_ctx->lk_heal_info;
+    if ((info->flock.l_start != flock.l_start) ||
+        (info->flock.l_whence != flock.l_whence) ||
+        (info->flock.l_len != flock.l_len)) {
+        /*TODO: Compare lkowners too.*/
+        goto out;
+    }
+
+    LOCK(&priv->lock);
+    {
+        list_del(&fd_ctx->lk_heal_info->pos);
+    }
+    UNLOCK(&priv->lock);
+
+    afr_lk_heal_info_cleanup(info);
+    fd_ctx->lk_heal_info = NULL;
+    ret = 0;
 out:
-        return;
+    if (ret)
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_LK_HEAL_DOM,
+               "%s: Failed to remove lock from healq",
+               uuid_utoa(local->fd->inode->gfid));
+    return ret;
 }
 
+int
+afr_lock_heal_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct gf_flock *lock,
+                  dict_t *xdata)
+{
+    afr_local_t *local = frame->local;
+    int i = (long)cookie;
+
+    local->replies[i].valid = 1;
+    local->replies[i].op_ret = op_ret;
+    local->replies[i].op_errno = op_errno;
+    if (op_ret != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, AFR_MSG_LK_HEAL_DOM,
+               "Failed to heal lock on child %d for %s", i,
+               uuid_utoa(local->fd->inode->gfid));
+    }
+    syncbarrier_wake(&local->barrier);
+    return 0;
+}
 
-uint64_t
-afr_is_opendir_done (xlator_t *this, inode_t *inode)
+int
+afr_getlk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct gf_flock *lock, dict_t *xdata)
 {
-        int ret = 0;
+    afr_local_t *local = frame->local;
+    int i = (long)cookie;
+
+    local->replies[i].valid = 1;
+    local->replies[i].op_ret = op_ret;
+    local->replies[i].op_errno = op_errno;
+    if (op_ret != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, AFR_MSG_LK_HEAL_DOM,
+               "Failed getlk for %s", uuid_utoa(local->fd->inode->gfid));
+    } else {
+        local->cont.lk.getlk_rsp[i] = *lock;
+    }
+
+    syncbarrier_wake(&local->barrier);
+    return 0;
+}
 
-        uint64_t ctx          = 0;
-        uint64_t opendir_done = 0;
+static gf_boolean_t
+afr_does_lk_owner_match(call_frame_t *frame, afr_private_t *priv,
+                        afr_lk_heal_info_t *info)
+{
+    int i = 0;
+    afr_local_t *local = frame->local;
+    struct gf_flock flock = {
+        0,
+    };
+    gf_boolean_t ret = _gf_true;
+    char *wind_on = alloca0(priv->child_count);
+    unsigned char *success_replies = alloca0(priv->child_count);
+    local->cont.lk.getlk_rsp = GF_CALLOC(sizeof(*local->cont.lk.getlk_rsp),
+                                         priv->child_count, gf_afr_mt_gf_lock);
+
+    flock = info->flock;
+    for (i = 0; i < priv->child_count; i++) {
+        if (info->locked_nodes[i])
+            wind_on[i] = 1;
+    }
+
+    AFR_ONLIST(wind_on, frame, afr_getlk_cbk, lk, info->fd, F_GETLK, &flock,
+               info->xdata_req);
+
+    afr_fill_success_replies(local, priv, success_replies);
+    if (AFR_COUNT(success_replies, priv->child_count) == 0) {
+        ret = _gf_false;
+        goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid || local->replies[i].op_ret != 0)
+            continue;
+        if (local->cont.lk.getlk_rsp[i].l_type == F_UNLCK)
+            continue;
+        /*TODO: Do we really need to compare lkowner if F_UNLCK is true?*/
+        if (!is_same_lkowner(&local->cont.lk.getlk_rsp[i].l_owner,
+                             &info->lk_owner)) {
+            ret = _gf_false;
+            break;
+        }
+    }
+out:
+    afr_local_replies_wipe(local, priv);
+    GF_FREE(local->cont.lk.getlk_rsp);
+    local->cont.lk.getlk_rsp = NULL;
+    return ret;
+}
 
-        VALIDATE_OR_GOTO (inode, out);
+static void
+afr_mark_fd_bad(fd_t *fd, xlator_t *this)
+{
+    afr_fd_ctx_t *fd_ctx = NULL;
 
-        LOCK (&inode->lock);
-        {
-                ret = __inode_ctx_get (inode, this, &ctx);
+    if (!fd)
+        return;
+    LOCK(&fd->lock);
+    {
+        fd_ctx = __afr_fd_ctx_get(fd, this);
+        if (fd_ctx) {
+            fd_ctx->is_fd_bad = _gf_true;
+            fd_ctx->lk_heal_info = NULL;
+        }
+    }
+    UNLOCK(&fd->lock);
+}
 
-                if (ret < 0)
-                        goto unlock;
+static void
+afr_add_lock_to_lkhealq(afr_private_t *priv, afr_lk_heal_info_t *info)
+{
+    LOCK(&priv->lock);
+    {
+        list_del(&info->pos);
+        list_add_tail(&info->pos, &priv->lk_healq);
+    }
+    UNLOCK(&priv->lock);
+}
 
-                opendir_done = ctx & AFR_ICTX_OPENDIR_DONE_MASK;
+static void
+afr_lock_heal_do(call_frame_t *frame, afr_private_t *priv,
+                 afr_lk_heal_info_t *info)
+{
+    int i = 0;
+    int op_errno = 0;
+    int32_t *current_event_gen = NULL;
+    afr_local_t *local = frame->local;
+    xlator_t *this = frame->this;
+    char *wind_on = alloca0(priv->child_count);
+    gf_boolean_t retry = _gf_true;
+
+    frame->root->pid = info->pid;
+    lk_owner_copy(&frame->root->lk_owner, &info->lk_owner);
+
+    op_errno = -afr_dom_lock_acquire(frame);
+    if ((op_errno != 0)) {
+        goto release;
+    }
+
+    if (!afr_does_lk_owner_match(frame, priv, info)) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_LK_HEAL_DOM,
+               "Ignoring lock heal for %s since lk-onwers mismatch. "
+               "Lock possibly pre-empted by another client.",
+               uuid_utoa(info->fd->inode->gfid));
+        goto release;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (info->locked_nodes[i])
+            continue;
+        wind_on[i] = 1;
+    }
+
+    current_event_gen = alloca(priv->child_count);
+    memcpy(current_event_gen, info->child_up_event_gen,
+           priv->child_count * sizeof *current_event_gen);
+    AFR_ONLIST(wind_on, frame, afr_lock_heal_cbk, lk, info->fd, info->cmd,
+               &info->flock, info->xdata_req);
+
+    LOCK(&priv->lock);
+    {
+        for (i = 0; i < priv->child_count; i++) {
+            if (!wind_on[i])
+                continue;
+            if ((!local->replies[i].valid) || (local->replies[i].op_ret != 0)) {
+                continue;
+            }
+
+            if ((current_event_gen[i] == info->child_up_event_gen[i]) &&
+                (current_event_gen[i] > info->child_down_event_gen[i])) {
+                info->locked_nodes[i] = 1;
+                retry = _gf_false;
+                list_del_init(&info->pos);
+                list_add_tail(&info->pos, &priv->saved_locks);
+            } else {
+                /*We received subsequent child up/down events while heal was in
+                 * progress; don't mark child as healed. Attempt again on the
+                 * new child up*/
+                gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_LK_HEAL_DOM,
+                       "Event gen mismatch: skipped healing lock on child %d "
+                       "for %s.",
+                       i, uuid_utoa(info->fd->inode->gfid));
+            }
         }
-unlock:
-        UNLOCK (&inode->lock);
+    }
+    UNLOCK(&priv->lock);
+
+release:
+    afr_dom_lock_release(frame);
+    if (retry)
+        afr_add_lock_to_lkhealq(priv, info);
+    return;
+}
 
-out:
-        return opendir_done;
+static int
+afr_lock_heal_done(int ret, call_frame_t *frame, void *opaque)
+{
+    STACK_DESTROY(frame->root);
+    return 0;
 }
 
+static int
+afr_lock_heal(void *opaque)
+{
+    call_frame_t *frame = (call_frame_t *)opaque;
+    call_frame_t *iter_frame = NULL;
+    xlator_t *this = frame->this;
+    afr_private_t *priv = this->private;
+    afr_lk_heal_info_t *info = NULL;
+    afr_lk_heal_info_t *tmp = NULL;
+    struct list_head healq = {
+        0,
+    };
+    int ret = 0;
+
+    iter_frame = afr_copy_frame(frame);
+    if (!iter_frame) {
+        return ENOMEM;
+    }
+
+    INIT_LIST_HEAD(&healq);
+    LOCK(&priv->lock);
+    {
+        list_splice_init(&priv->lk_healq, &healq);
+    }
+    UNLOCK(&priv->lock);
+
+    list_for_each_entry_safe(info, tmp, &healq, pos)
+    {
+        GF_ASSERT((AFR_COUNT(info->locked_nodes, priv->child_count) <
+                   priv->child_count));
+        ((afr_local_t *)(iter_frame->local))->fd = fd_ref(info->fd);
+        afr_lock_heal_do(iter_frame, priv, info);
+        AFR_STACK_RESET(iter_frame);
+        if (iter_frame->local == NULL) {
+            ret = ENOTCONN;
+            gf_msg(frame->this->name, GF_LOG_ERROR, ENOTCONN,
+                   AFR_MSG_LK_HEAL_DOM,
+                   "Aborting processing of lk_healq."
+                   "Healing will be reattempted on next child up for locks "
+                   "that are still in quorum.");
+            LOCK(&priv->lock);
+            {
+                list_add_tail(&healq, &priv->lk_healq);
+            }
+            UNLOCK(&priv->lock);
+            break;
+        }
+    }
+
+    AFR_STACK_DESTROY(iter_frame);
+    return ret;
+}
 
-void
-afr_set_opendir_done (xlator_t *this, inode_t *inode)
+static int
+__afr_lock_heal_synctask(xlator_t *this, afr_private_t *priv, int child)
 {
-        uint64_t ctx = 0;
-        int      ret = 0;
+    int ret = 0;
+    call_frame_t *frame = NULL;
+    afr_lk_heal_info_t *info = NULL;
+    afr_lk_heal_info_t *tmp = NULL;
 
-        VALIDATE_OR_GOTO (inode, out);
+    if (priv->shd.iamshd)
+        return 0;
 
-        LOCK (&inode->lock);
-        {
-                ret = __inode_ctx_get (inode, this, &ctx);
+    list_for_each_entry_safe(info, tmp, &priv->saved_locks, pos)
+    {
+        info->child_up_event_gen[child] = priv->event_generation;
+        list_del_init(&info->pos);
+        list_add_tail(&info->pos, &priv->lk_healq);
+    }
 
-                if (ret < 0) {
-                        ctx = 0;
-                }
+    frame = create_frame(this, this->ctx->pool);
+    if (!frame)
+        return -1;
 
-                ctx = (~AFR_ICTX_OPENDIR_DONE_MASK & ctx)
-                        | (0xFFFFFFFFFFFFFFFFULL & AFR_ICTX_OPENDIR_DONE_MASK);
+    ret = synctask_new(this->ctx->env, afr_lock_heal, afr_lock_heal_done, frame,
+                       frame);
+    if (ret)
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_LK_HEAL_DOM,
+               "Failed to launch lock heal synctask");
 
-                __inode_ctx_put (inode, this, ctx);
-        }
-        UNLOCK (&inode->lock);
-out:
-        return;
+    return ret;
 }
 
-
-uint64_t
-afr_read_child (xlator_t *this, inode_t *inode)
+static int
+__afr_mark_pending_lk_heal(xlator_t *this, afr_private_t *priv, int child)
 {
-        int ret = 0;
-
-        uint64_t ctx         = 0;
-        uint64_t read_child  = 0;
+    afr_lk_heal_info_t *info = NULL;
+    afr_lk_heal_info_t *tmp = NULL;
 
-        VALIDATE_OR_GOTO (inode, out);
+    if (priv->shd.iamshd)
+        return 0;
+    list_for_each_entry_safe(info, tmp, &priv->saved_locks, pos)
+    {
+        info->child_down_event_gen[child] = priv->event_generation;
+        if (info->locked_nodes[child] == 1)
+            info->locked_nodes[child] = 0;
+        if (!afr_has_quorum(info->locked_nodes, this, NULL)) {
+            /* Since the lock was lost on quorum no. of nodes, we should
+             * not attempt to heal it anymore. Some other client could have
+             * acquired the lock, modified data and released it and this
+             * client wouldn't know about it if we heal it.*/
+            afr_mark_fd_bad(info->fd, this);
+            list_del(&info->pos);
+            afr_lk_heal_info_cleanup(info);
+            /* We're not winding an unlock on the node where the lock is still
+             * present because when fencing logic switches over to the new
+             * client (since we marked the fd bad), it should preempt any
+             * existing lock. */
+        }
+    }
+    return 0;
+}
 
-        LOCK (&inode->lock);
-        {
-                ret = __inode_ctx_get (inode, this, &ctx);
+gf_boolean_t
+afr_is_consistent_io_possible(afr_local_t *local, afr_private_t *priv,
+                              int32_t *op_errno)
+{
+    if (priv->consistent_io && local->call_count != priv->child_count) {
+        gf_msg(THIS->name, GF_LOG_INFO, 0, AFR_MSG_SUBVOLS_DOWN,
+               "All subvolumes are not up");
+        if (op_errno)
+            *op_errno = ENOTCONN;
+        return _gf_false;
+    }
+    return _gf_true;
+}
 
-                if (ret < 0)
-                        goto unlock;
+gf_boolean_t
+afr_is_lock_mode_mandatory(dict_t *xdata)
+{
+    int ret = 0;
+    uint32_t lk_mode = GF_LK_ADVISORY;
 
-                read_child = ctx & AFR_ICTX_READ_CHILD_MASK;
-        }
-unlock:
-        UNLOCK (&inode->lock);
+    ret = dict_get_uint32(xdata, GF_LOCK_MODE, &lk_mode);
+    if (!ret && lk_mode == GF_LK_MANDATORY)
+        return _gf_true;
 
-out:
-        return read_child;
+    return _gf_false;
 }
 
+call_frame_t *
+afr_copy_frame(call_frame_t *base)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *frame = NULL;
+    int op_errno = 0;
+
+    frame = copy_frame(base);
+    if (!frame)
+        return NULL;
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local) {
+        AFR_STACK_DESTROY(frame);
+        return NULL;
+    }
+
+    return frame;
+}
 
-void
-afr_set_read_child (xlator_t *this, inode_t *inode, int32_t read_child)
+/* Check if an entry or inode could be undergoing a transaction. */
+gf_boolean_t
+afr_is_possibly_under_txn(afr_transaction_type type, afr_local_t *local,
+                          xlator_t *this)
 {
-        uint64_t ctx = 0;
-        int      ret = 0;
+    int i = 0;
+    int tmp = 0;
+    afr_private_t *priv = NULL;
+    GF_UNUSED char *key = NULL;
+    int keylen = 0;
+
+    priv = this->private;
+
+    if (type == AFR_ENTRY_TRANSACTION) {
+        key = GLUSTERFS_PARENT_ENTRYLK;
+        keylen = SLEN(GLUSTERFS_PARENT_ENTRYLK);
+    } else if (type == AFR_DATA_TRANSACTION) {
+        /*FIXME: Use GLUSTERFS_INODELK_DOM_COUNT etc. once
+         * pl_inodelk_xattr_fill supports separate keys for different
+         * domains.*/
+        key = GLUSTERFS_INODELK_COUNT;
+        keylen = SLEN(GLUSTERFS_INODELK_COUNT);
+    }
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].xdata)
+            continue;
+        if (dict_get_int32n(local->replies[i].xdata, key, keylen, &tmp) == 0)
+            if (tmp)
+                return _gf_true;
+    }
+
+    return _gf_false;
+}
 
-        VALIDATE_OR_GOTO (inode, out);
+static void
+afr_inode_ctx_destroy(afr_inode_ctx_t *ctx)
+{
+    int i = 0;
 
-        LOCK (&inode->lock);
-        {
-                ret = __inode_ctx_get (inode, this, &ctx);
+    if (!ctx)
+        return;
 
-                if (ret < 0) {
-                        ctx = 0;
-                }
+    for (i = 0; i < AFR_NUM_CHANGE_LOGS; i++) {
+        GF_FREE(ctx->pre_op_done[i]);
+    }
 
-                ctx = (~AFR_ICTX_READ_CHILD_MASK & ctx)
-                        | (AFR_ICTX_READ_CHILD_MASK & read_child);
+    GF_FREE(ctx);
+}
 
-                __inode_ctx_put (inode, this, ctx);
+int
+__afr_inode_ctx_get(xlator_t *this, inode_t *inode, afr_inode_ctx_t **ctx)
+{
+    uint64_t ctx_int = 0;
+    int ret = -1;
+    int i = -1;
+    int num_locks = -1;
+    afr_inode_ctx_t *ictx = NULL;
+    afr_lock_t *lock = NULL;
+    afr_private_t *priv = this->private;
+
+    ret = __inode_ctx_get(inode, this, &ctx_int);
+    if (ret == 0) {
+        *ctx = (afr_inode_ctx_t *)(uintptr_t)ctx_int;
+        return 0;
+    }
+
+    ictx = GF_CALLOC(1, sizeof(afr_inode_ctx_t), gf_afr_mt_inode_ctx_t);
+    if (!ictx)
+        goto out;
+
+    for (i = 0; i < AFR_NUM_CHANGE_LOGS; i++) {
+        ictx->pre_op_done[i] = GF_CALLOC(sizeof *ictx->pre_op_done[i],
+                                         priv->child_count, gf_afr_mt_int32_t);
+        if (!ictx->pre_op_done[i]) {
+            ret = -ENOMEM;
+            goto out;
         }
-        UNLOCK (&inode->lock);
-
+    }
+
+    num_locks = sizeof(ictx->lock) / sizeof(afr_lock_t);
+    for (i = 0; i < num_locks; i++) {
+        lock = &ictx->lock[i];
+        INIT_LIST_HEAD(&lock->post_op);
+        INIT_LIST_HEAD(&lock->frozen);
+        INIT_LIST_HEAD(&lock->waiting);
+        INIT_LIST_HEAD(&lock->owners);
+    }
+
+    ctx_int = (uint64_t)(uintptr_t)ictx;
+    ret = __inode_ctx_set(inode, this, &ctx_int);
+    if (ret) {
+        goto out;
+    }
+
+    ictx->spb_choice = -1;
+    ictx->read_subvol = 0;
+    ictx->write_subvol = 0;
+    ictx->lock_count = 0;
+    ret = 0;
+    *ctx = ictx;
 out:
-        return;
+    if (ret) {
+        afr_inode_ctx_destroy(ictx);
+    }
+    return ret;
 }
 
-
-/**
- * afr_local_cleanup - cleanup everything in frame->local
+/*
+ * INODE CTX 64-bit VALUE FORMAT FOR SMALL (<= 16) SUBVOL COUNTS:
+ *
+ * |<----------   64bit   ------------>|
+ *  63           32 31    16 15       0
+ * |   EVENT_GEN   |  DATA  | METADATA |
+ *
+ *
+ *  METADATA (bit-0 .. bit-15): bitmap representing subvolumes from which
+ *                              metadata can be attempted to be read.
+ *
+ *                              bit-0 => priv->subvolumes[0]
+ *                              bit-1 => priv->subvolumes[1]
+ *                              ... etc. till bit-15
+ *
+ *  DATA (bit-16 .. bit-31): bitmap representing subvolumes from which data
+ *                           can be attempted to be read.
+ *
+ *                           bit-16 => priv->subvolumes[0]
+ *                           bit-17 => priv->subvolumes[1]
+ *                           ... etc. till bit-31
+ *
+ *  EVENT_GEN (bit-32 .. bit-63): event generation (i.e priv->event_generation)
+ *                                when DATA and METADATA was last updated.
+ *
+ *                                If EVENT_GEN is < priv->event_generation,
+ *                                or is 0, it means afr_inode_refresh() needs
+ *                                to be called to recalculate the bitmaps.
  */
 
-void
-afr_local_sh_cleanup (afr_local_t *local, xlator_t *this)
+int
+__afr_set_in_flight_sb_status(xlator_t *this, afr_local_t *local,
+                              inode_t *inode)
 {
-        afr_self_heal_t *sh = NULL;
-        afr_private_t   *priv = NULL;
-        int              i = 0;
+    int i = 0;
+    int txn_type = 0;
+    int count = 0;
+    int index = -1;
+    uint16_t datamap_old = 0;
+    uint16_t metadatamap_old = 0;
+    uint16_t datamap = 0;
+    uint16_t metadatamap = 0;
+    uint16_t tmp_map = 0;
+    uint16_t mask = 0;
+    uint32_t event = 0;
+    uint64_t val = 0;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    txn_type = local->transaction.type;
+
+    if (txn_type == AFR_DATA_TRANSACTION)
+        val = local->inode_ctx->write_subvol;
+    else
+        val = local->inode_ctx->read_subvol;
+
+    metadatamap_old = metadatamap = (val & 0x000000000000ffff);
+    datamap_old = datamap = (val & 0x00000000ffff0000) >> 16;
+    event = (val & 0xffffffff00000000) >> 32;
+
+    if (txn_type == AFR_DATA_TRANSACTION)
+        tmp_map = datamap;
+    else if (txn_type == AFR_METADATA_TRANSACTION)
+        tmp_map = metadatamap;
+
+    count = gf_bits_count(tmp_map);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->transaction.failed_subvols[i])
+            continue;
+
+        mask = 1 << i;
+        if (txn_type == AFR_METADATA_TRANSACTION)
+            metadatamap &= ~mask;
+        else if (txn_type == AFR_DATA_TRANSACTION)
+            datamap &= ~mask;
+    }
+
+    switch (txn_type) {
+        case AFR_METADATA_TRANSACTION:
+            if ((metadatamap_old != 0) && (metadatamap == 0) && (count == 1)) {
+                index = gf_bits_index(tmp_map);
+                local->transaction.in_flight_sb_errno = local->replies[index]
+                                                            .op_errno;
+                local->transaction.in_flight_sb = _gf_true;
+                metadatamap |= (1 << index);
+            }
+            if (metadatamap_old != metadatamap) {
+                __afr_inode_need_refresh_set(inode, this);
+            }
+            break;
+
+        case AFR_DATA_TRANSACTION:
+            if ((datamap_old != 0) && (datamap == 0) && (count == 1)) {
+                index = gf_bits_index(tmp_map);
+                local->transaction.in_flight_sb_errno = local->replies[index]
+                                                            .op_errno;
+                local->transaction.in_flight_sb = _gf_true;
+                datamap |= (1 << index);
+            }
+            if (datamap_old != datamap)
+                __afr_inode_need_refresh_set(inode, this);
+            break;
 
+        default:
+            break;
+    }
 
-        sh = &local->self_heal;
-        priv = this->private;
+    val = ((uint64_t)metadatamap) | (((uint64_t)datamap) << 16) |
+          (((uint64_t)event) << 32);
 
-        if (sh->buf)
-                GF_FREE (sh->buf);
+    if (txn_type == AFR_DATA_TRANSACTION)
+        local->inode_ctx->write_subvol = val;
+    local->inode_ctx->read_subvol = val;
 
-        if (sh->xattr) {
-                for (i = 0; i < priv->child_count; i++) {
-                        if (sh->xattr[i]) {
-                                dict_unref (sh->xattr[i]);
-                                sh->xattr[i] = NULL;
-                        }
-                }
-                GF_FREE (sh->xattr);
-        }
-
-        if (sh->child_errno)
-                GF_FREE (sh->child_errno);
+    return 0;
+}
 
-        if (sh->pending_matrix) {
-                for (i = 0; i < priv->child_count; i++) {
-                        GF_FREE (sh->pending_matrix[i]);
-                }
-                GF_FREE (sh->pending_matrix);
+gf_boolean_t
+afr_is_symmetric_error(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int op_errno = 0;
+    int i_errno = 0;
+    gf_boolean_t matching_errors = _gf_true;
+    int i = 0;
+
+    priv = this->private;
+    local = frame->local;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
+        if (local->replies[i].op_ret != -1) {
+            /* Operation succeeded on at least one subvol,
+               so it is not a failed-everywhere situation.
+            */
+            matching_errors = _gf_false;
+            break;
+        }
+        i_errno = local->replies[i].op_errno;
+
+        if (i_errno == ENOTCONN) {
+            /* ENOTCONN is not a symmetric error. We do not
+               know if the operation was performed on the
+               backend or not.
+            */
+            matching_errors = _gf_false;
+            break;
         }
 
-        if (sh->delta_matrix) {
-                for (i = 0; i < priv->child_count; i++) {
-                        GF_FREE (sh->delta_matrix[i]);
-                }
-                GF_FREE (sh->delta_matrix);
+        if (!op_errno) {
+            op_errno = i_errno;
+        } else if (op_errno != i_errno) {
+            /* Mismatching op_errno's */
+            matching_errors = _gf_false;
+            break;
         }
+    }
 
-        if (sh->sources)
-                GF_FREE (sh->sources);
+    return matching_errors;
+}
 
-        if (sh->success)
-                GF_FREE (sh->success);
+int
+afr_set_in_flight_sb_status(xlator_t *this, call_frame_t *frame, inode_t *inode)
+{
+    int ret = -1;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
 
-        if (sh->locked_nodes)
-                GF_FREE (sh->locked_nodes);
+    priv = this->private;
+    local = frame->local;
 
-        if (sh->healing_fd && !sh->healing_fd_opened) {
-                fd_unref (sh->healing_fd);
-                sh->healing_fd = NULL;
-        }
+    /* If this transaction saw no failures, then exit. */
+    if (AFR_COUNT(local->transaction.failed_subvols, priv->child_count) == 0)
+        return 0;
 
-        if (sh->linkname)
-                GF_FREE ((char *)sh->linkname);
+    if (afr_is_symmetric_error(frame, this))
+        return 0;
 
-        loc_wipe (&sh->parent_loc);
-}
+    LOCK(&inode->lock);
+    {
+        ret = __afr_set_in_flight_sb_status(this, local, inode);
+    }
+    UNLOCK(&inode->lock);
 
+    return ret;
+}
 
-void
-afr_local_transaction_cleanup (afr_local_t *local, xlator_t *this)
+int
+__afr_inode_read_subvol_get_small(inode_t *inode, xlator_t *this,
+                                  unsigned char *data, unsigned char *metadata,
+                                  int *event_p)
 {
-        int             i = 0;
-        afr_private_t * priv = NULL;
+    afr_private_t *priv = NULL;
+    int ret = -1;
+    uint16_t datamap = 0;
+    uint16_t metadatamap = 0;
+    uint32_t event = 0;
+    uint64_t val = 0;
+    int i = 0;
+    afr_inode_ctx_t *ctx = NULL;
+
+    priv = this->private;
+
+    ret = __afr_inode_ctx_get(this, inode, &ctx);
+    if (ret < 0)
+        return ret;
 
-        priv = this->private;
+    val = ctx->read_subvol;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->pending && local->pending[i])
-                        GF_FREE (local->pending[i]);
-        }
+    metadatamap = (val & 0x000000000000ffff);
+    datamap = (val & 0x00000000ffff0000) >> 16;
+    event = (val & 0xffffffff00000000) >> 32;
 
-        GF_FREE (local->pending);
+    for (i = 0; i < priv->child_count; i++) {
+        if (metadata)
+            metadata[i] = (metadatamap >> i) & 1;
+        if (data)
+            data[i] = (datamap >> i) & 1;
+    }
 
-        if (local->internal_lock.locked_nodes)
-                GF_FREE (local->internal_lock.locked_nodes);
+    if (event_p)
+        *event_p = event;
+    return ret;
+}
 
-        if (local->internal_lock.inode_locked_nodes)
-                GF_FREE (local->internal_lock.inode_locked_nodes);
+int
+__afr_inode_read_subvol_set_small(inode_t *inode, xlator_t *this,
+                                  unsigned char *data, unsigned char *metadata,
+                                  int event)
+{
+    afr_private_t *priv = NULL;
+    uint16_t datamap = 0;
+    uint16_t metadatamap = 0;
+    uint64_t val = 0;
+    int i = 0;
+    int ret = -1;
+    afr_inode_ctx_t *ctx = NULL;
 
-        if (local->internal_lock.entry_locked_nodes)
-                GF_FREE (local->internal_lock.entry_locked_nodes);
+    priv = this->private;
 
-        if (local->internal_lock.lower_locked_nodes)
-                GF_FREE (local->internal_lock.lower_locked_nodes);
+    ret = __afr_inode_ctx_get(this, inode, &ctx);
+    if (ret)
+        goto out;
 
+    for (i = 0; i < priv->child_count; i++) {
+        if (data[i])
+            datamap |= (1 << i);
+        if (metadata[i])
+            metadatamap |= (1 << i);
+    }
 
-        GF_FREE (local->transaction.child_errno);
-        GF_FREE (local->child_errno);
+    val = ((uint64_t)metadatamap) | (((uint64_t)datamap) << 16) |
+          (((uint64_t)event) << 32);
 
-        GF_FREE (local->transaction.basename);
-        GF_FREE (local->transaction.new_basename);
+    ctx->read_subvol = val;
 
-        loc_wipe (&local->transaction.parent_loc);
-        loc_wipe (&local->transaction.new_parent_loc);
+    ret = 0;
+out:
+    return ret;
 }
 
-
-void
-afr_local_cleanup (afr_local_t *local, xlator_t *this)
+int
+__afr_inode_read_subvol_get(inode_t *inode, xlator_t *this, unsigned char *data,
+                            unsigned char *metadata, int *event_p)
 {
-        int i;
-        afr_private_t * priv = NULL;
+    afr_private_t *priv = NULL;
+    int ret = -1;
 
-        if (!local)
-                return;
+    priv = this->private;
 
-        afr_local_sh_cleanup (local, this);
+    if (priv->child_count <= 16)
+        ret = __afr_inode_read_subvol_get_small(inode, this, data, metadata,
+                                                event_p);
+    else
+        /* TBD: allocate structure with array and read from it */
+        ret = -1;
 
-        afr_local_transaction_cleanup (local, this);
+    return ret;
+}
 
-        priv = this->private;
+int
+__afr_inode_split_brain_choice_get(inode_t *inode, xlator_t *this,
+                                   int *spb_choice)
+{
+    afr_inode_ctx_t *ctx = NULL;
+    int ret = -1;
 
-        loc_wipe (&local->loc);
-        loc_wipe (&local->newloc);
+    ret = __afr_inode_ctx_get(this, inode, &ctx);
+    if (ret < 0)
+        return ret;
 
-        if (local->fd)
-                fd_unref (local->fd);
-
-        if (local->xattr_req)
-                dict_unref (local->xattr_req);
-
-        GF_FREE (local->child_up);
-
-        { /* lookup */
-                if (local->cont.lookup.xattrs) {
-                        for (i = 0; i < priv->child_count; i++) {
-                                if (local->cont.lookup.xattrs[i]) {
-                                        dict_unref (local->cont.lookup.xattrs[i]);
-                                        local->cont.lookup.xattrs[i] = NULL;
-                                }
-                        }
-                        GF_FREE (local->cont.lookup.xattrs);
-                        local->cont.lookup.xattrs = NULL;
-                }
+    *spb_choice = ctx->spb_choice;
+    return 0;
+}
 
-                if (local->cont.lookup.xattr) {
-                        dict_unref (local->cont.lookup.xattr);
-                }
+int
+__afr_inode_read_subvol_set(inode_t *inode, xlator_t *this, unsigned char *data,
+                            unsigned char *metadata, int event)
+{
+    afr_private_t *priv = NULL;
+    int ret = -1;
 
-                if (local->cont.lookup.inode) {
-                        inode_unref (local->cont.lookup.inode);
-                }
-        }
+    priv = this->private;
 
-        { /* getxattr */
-                if (local->cont.getxattr.name)
-                        GF_FREE (local->cont.getxattr.name);
-        }
+    if (priv->child_count <= 16)
+        ret = __afr_inode_read_subvol_set_small(inode, this, data, metadata,
+                                                event);
+    else
+        ret = -1;
 
-        { /* lk */
-                if (local->cont.lk.locked_nodes)
-                        GF_FREE (local->cont.lk.locked_nodes);
-        }
+    return ret;
+}
 
-        { /* create */
-                if (local->cont.create.fd)
-                        fd_unref (local->cont.create.fd);
-                if (local->cont.create.params)
-                        dict_unref (local->cont.create.params);
-        }
+int
+__afr_inode_split_brain_choice_set(inode_t *inode, xlator_t *this,
+                                   int spb_choice)
+{
+    afr_inode_ctx_t *ctx = NULL;
+    int ret = -1;
 
-        { /* mknod */
-                if (local->cont.mknod.params)
-                        dict_unref (local->cont.mknod.params);
-        }
+    ret = __afr_inode_ctx_get(this, inode, &ctx);
+    if (ret)
+        goto out;
 
-        { /* mkdir */
-                if (local->cont.mkdir.params)
-                        dict_unref (local->cont.mkdir.params);
-        }
+    ctx->spb_choice = spb_choice;
 
-        { /* symlink */
-                if (local->cont.symlink.params)
-                        dict_unref (local->cont.symlink.params);
-        }
+    ret = 0;
+out:
+    return ret;
+}
 
-        { /* writev */
-                GF_FREE (local->cont.writev.vector);
-        }
+int
+afr_inode_read_subvol_get(inode_t *inode, xlator_t *this, unsigned char *data,
+                          unsigned char *metadata, int *event_p)
+{
+    int ret = -1;
 
-        { /* setxattr */
-                if (local->cont.setxattr.dict)
-                        dict_unref (local->cont.setxattr.dict);
-        }
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
 
-        { /* removexattr */
-                GF_FREE (local->cont.removexattr.name);
-        }
+    LOCK(&inode->lock);
+    {
+        ret = __afr_inode_read_subvol_get(inode, this, data, metadata, event_p);
+    }
+    UNLOCK(&inode->lock);
+out:
+    return ret;
+}
 
-        { /* symlink */
-                GF_FREE (local->cont.symlink.linkpath);
+int
+afr_inode_get_readable(call_frame_t *frame, inode_t *inode, xlator_t *this,
+                       unsigned char *readable, int *event_p, int type)
+{
+    afr_private_t *priv = this->private;
+    afr_local_t *local = frame->local;
+    unsigned char *data = alloca0(priv->child_count);
+    unsigned char *metadata = alloca0(priv->child_count);
+    int data_count = 0;
+    int metadata_count = 0;
+    int event_generation = 0;
+    int ret = 0;
+
+    ret = afr_inode_read_subvol_get(inode, this, data, metadata,
+                                    &event_generation);
+    if (ret == -1)
+        return -EIO;
+
+    data_count = AFR_COUNT(data, priv->child_count);
+    metadata_count = AFR_COUNT(metadata, priv->child_count);
+
+    if (inode->ia_type == IA_IFDIR) {
+        /* For directories, allow even if it is in data split-brain. */
+        if (type == AFR_METADATA_TRANSACTION || local->op == GF_FOP_STAT ||
+            local->op == GF_FOP_FSTAT) {
+            if (!metadata_count)
+                return -EIO;
         }
-
-        { /* opendir */
-                if (local->cont.opendir.checksum)
-                        GF_FREE (local->cont.opendir.checksum);
+    } else {
+        /* For files, abort in case of data/metadata split-brain. */
+        if (!data_count || !metadata_count) {
+            return -EIO;
         }
+    }
+
+    if (type == AFR_METADATA_TRANSACTION && readable)
+        memcpy(readable, metadata, priv->child_count * sizeof *metadata);
+    if (type == AFR_DATA_TRANSACTION && readable) {
+        if (!data_count)
+            memcpy(readable, local->child_up,
+                   priv->child_count * sizeof *readable);
+        else
+            memcpy(readable, data, priv->child_count * sizeof *data);
+    }
+    if (event_p)
+        *event_p = event_generation;
+    return 0;
 }
 
+static int
+afr_inode_split_brain_choice_get(inode_t *inode, xlator_t *this,
+                                 int *spb_choice)
+{
+    int ret = -1;
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+
+    LOCK(&inode->lock);
+    {
+        ret = __afr_inode_split_brain_choice_get(inode, this, spb_choice);
+    }
+    UNLOCK(&inode->lock);
+out:
+    return ret;
+}
 
+/*
+ * frame is used to get the favourite policy. Since
+ * afr_inode_split_brain_choice_get was called with afr_open, it is possible to
+ * have a frame with out local->replies. So in that case, frame is passed as
+ * null, hence this function will handle the frame NULL case.
+ */
 int
-afr_frame_return (call_frame_t *frame)
+afr_split_brain_read_subvol_get(inode_t *inode, xlator_t *this,
+                                call_frame_t *frame, int *spb_subvol)
 {
-        afr_local_t *local = NULL;
-        int          call_count = 0;
+    int ret = -1;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        local = frame->local;
+    GF_VALIDATE_OR_GOTO("afr", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+    GF_VALIDATE_OR_GOTO(this->name, spb_subvol, out);
 
-        LOCK (&frame->lock);
-        {
-                call_count = --local->call_count;
+    priv = this->private;
+
+    ret = afr_inode_split_brain_choice_get(inode, this, spb_subvol);
+    if (*spb_subvol < 0 && priv->fav_child_policy && frame && frame->local) {
+        local = frame->local;
+        *spb_subvol = afr_sh_get_fav_by_policy(this, local->replies, inode,
+                                               NULL);
+        if (*spb_subvol >= 0) {
+            ret = 0;
         }
-        UNLOCK (&frame->lock);
+    }
 
-        return call_count;
+out:
+    return ret;
 }
+int
+afr_inode_read_subvol_set(inode_t *inode, xlator_t *this, unsigned char *data,
+                          unsigned char *metadata, int event)
+{
+    int ret = -1;
 
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
 
-/**
- * up_children_count - return the number of children that are up
- */
+    LOCK(&inode->lock);
+    {
+        ret = __afr_inode_read_subvol_set(inode, this, data, metadata, event);
+    }
+    UNLOCK(&inode->lock);
+out:
+    return ret;
+}
 
 int
-afr_up_children_count (int child_count, unsigned char *child_up)
+afr_inode_split_brain_choice_set(inode_t *inode, xlator_t *this, int spb_choice)
 {
-        int i   = 0;
-        int ret = 0;
+    int ret = -1;
 
-        for (i = 0; i < child_count; i++)
-                if (child_up[i])
-                        ret++;
-        return ret;
-}
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
 
+    LOCK(&inode->lock);
+    {
+        ret = __afr_inode_split_brain_choice_set(inode, this, spb_choice);
+    }
+    UNLOCK(&inode->lock);
+out:
+    return ret;
+}
 
-ino64_t
-afr_itransform (ino64_t ino, int child_count, int child_index)
+/* The caller of this should perform afr_inode_refresh, if this function
+ * returns _gf_true
+ */
+gf_boolean_t
+afr_is_inode_refresh_reqd(inode_t *inode, xlator_t *this, int event_gen1,
+                          int event_gen2)
 {
-        ino64_t scaled_ino = -1;
+    gf_boolean_t need_refresh = _gf_false;
+    afr_inode_ctx_t *ctx = NULL;
+    int ret = -1;
 
-        if (ino == ((uint64_t) -1)) {
-                scaled_ino = ((uint64_t) -1);
-                goto out;
-        }
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
 
-        scaled_ino = (ino * child_count) + child_index;
+    LOCK(&inode->lock);
+    {
+        ret = __afr_inode_ctx_get(this, inode, &ctx);
+        if (ret)
+            goto unlock;
+
+        need_refresh = ctx->need_refresh;
+        /* Hoping that the caller will do inode_refresh followed by
+         * this, hence setting the need_refresh to false */
+        ctx->need_refresh = _gf_false;
+    }
+unlock:
+    UNLOCK(&inode->lock);
 
+    if (event_gen1 != event_gen2)
+        need_refresh = _gf_true;
 out:
-        return scaled_ino;
+    return need_refresh;
 }
 
-
 int
-afr_deitransform_orig (ino64_t ino, int child_count)
+__afr_inode_need_refresh_set(inode_t *inode, xlator_t *this)
 {
-        int index = -1;
+    int ret = -1;
+    afr_inode_ctx_t *ctx = NULL;
 
-        index = ino % child_count;
+    ret = __afr_inode_ctx_get(this, inode, &ctx);
+    if (ret == 0) {
+        ctx->need_refresh = _gf_true;
+    }
 
-        return index;
+    return ret;
 }
 
-
 int
-afr_deitransform (ino64_t ino, int child_count)
+afr_inode_need_refresh_set(inode_t *inode, xlator_t *this)
 {
-        return 0;
-}
+    int ret = -1;
 
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+
+    LOCK(&inode->lock);
+    {
+        ret = __afr_inode_need_refresh_set(inode, this);
+    }
+    UNLOCK(&inode->lock);
+out:
+    return ret;
+}
 
 int
-afr_self_heal_lookup_unwind (call_frame_t *frame, xlator_t *this)
+afr_spb_choice_timeout_cancel(xlator_t *this, inode_t *inode)
 {
-        afr_local_t *local = NULL;
+    afr_inode_ctx_t *ctx = NULL;
+    int ret = -1;
 
-        local = frame->local;
+    if (!inode)
+        return ret;
 
-        if (local->govinda_gOvinda) {
-                afr_set_split_brain (this, local->cont.lookup.inode, _gf_true);
+    LOCK(&inode->lock);
+    {
+        ret = __afr_inode_ctx_get(this, inode, &ctx);
+        if (ret < 0 || !ctx) {
+            UNLOCK(&inode->lock);
+            gf_msg(this->name, GF_LOG_WARNING, 0,
+                   AFR_MSG_SPLIT_BRAIN_CHOICE_ERROR,
+                   "Failed to cancel split-brain choice timer.");
+            goto out;
         }
-
-        AFR_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno,
-                          local->cont.lookup.inode,
-                          &local->cont.lookup.buf,
-                          local->cont.lookup.xattr,
-                          &local->cont.lookup.postparent);
-
-        return 0;
+        ctx->spb_choice = -1;
+        if (ctx->timer) {
+            gf_timer_call_cancel(this->ctx, ctx->timer);
+            ctx->timer = NULL;
+        }
+        ret = 0;
+    }
+    UNLOCK(&inode->lock);
+out:
+    return ret;
 }
 
-
-static void
-afr_lookup_collect_xattr (afr_local_t *local, xlator_t *this,
-                          int child_index, dict_t *xattr)
+void
+afr_set_split_brain_choice_cbk(void *data)
 {
-        uint32_t        inodelk_count = 0;
-        uint32_t        entrylk_count = 0;
+    inode_t *inode = data;
+    xlator_t *this = THIS;
 
-        int ret = 0;
+    afr_spb_choice_timeout_cancel(this, inode);
+    inode_invalidate(inode);
+    inode_unref(inode);
+    return;
+}
 
-        if (afr_sh_has_metadata_pending (xattr, child_index, this)) {
-                local->self_heal.need_metadata_self_heal = _gf_true;
-                gf_log(this->name, GF_LOG_DEBUG,
-                       "metadata self-heal is pending for %s.",
-                       local->loc.path);
+int
+afr_set_split_brain_choice(int ret, call_frame_t *frame, void *opaque)
+{
+    int op_errno = ENOMEM;
+    afr_private_t *priv = NULL;
+    afr_inode_ctx_t *ctx = NULL;
+    inode_t *inode = NULL;
+    loc_t *loc = NULL;
+    xlator_t *this = NULL;
+    afr_spbc_timeout_t *data = opaque;
+    struct timespec delta = {
+        0,
+    };
+    gf_boolean_t timer_set = _gf_false;
+    gf_boolean_t timer_cancelled = _gf_false;
+    gf_boolean_t timer_reset = _gf_false;
+    int old_spb_choice = -1;
+
+    frame = data->frame;
+    loc = data->loc;
+    this = frame->this;
+    priv = this->private;
+
+    if (ret) {
+        op_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    delta.tv_sec = priv->spb_choice_timeout;
+    delta.tv_nsec = 0;
+
+    if (!loc->inode) {
+        ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    if (!(data->d_spb || data->m_spb)) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_SPLIT_BRAIN_CHOICE_ERROR,
+               "Cannot set "
+               "replica.split-brain-choice on %s. File is"
+               " not in data/metadata split-brain.",
+               uuid_utoa(loc->gfid));
+        ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    /*
+     * we're ref'ing the inode before LOCK like it is done elsewhere in the
+     * code. If we ref after LOCK, coverity complains of possible deadlocks.
+     */
+    inode = inode_ref(loc->inode);
+
+    LOCK(&inode->lock);
+    {
+        ret = __afr_inode_ctx_get(this, inode, &ctx);
+        if (ret) {
+            UNLOCK(&inode->lock);
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   AFR_MSG_SPLIT_BRAIN_CHOICE_ERROR,
+                   "Failed to get inode_ctx for %s", loc->name);
+            goto post_unlock;
         }
 
-        if (afr_sh_has_entry_pending (xattr, child_index, this)) {
-                local->self_heal.need_entry_self_heal = _gf_true;
-                gf_log(this->name, GF_LOG_DEBUG,
-                       "entry self-heal is pending for %s.", local->loc.path);
+        old_spb_choice = ctx->spb_choice;
+        ctx->spb_choice = data->spb_child_index;
+
+        /* Possible changes in spb-choice :
+         *         valid to -1    : cancel timer and unref
+         *         valid to valid : cancel timer and inject new one
+         *         -1    to -1    : unref and do not do anything
+         *         -1 to valid    : inject timer
+         */
+
+        /* ctx->timer is NULL iff previous value of
+         * ctx->spb_choice is -1
+         */
+        if (ctx->timer) {
+            if (ctx->spb_choice == -1) {
+                if (!gf_timer_call_cancel(this->ctx, ctx->timer)) {
+                    ctx->timer = NULL;
+                    timer_cancelled = _gf_true;
+                }
+                /* If timer cancel failed here it means that the
+                 *  previous cbk will be executed which will set
+                 *  spb_choice to -1. So we can consider the
+                 *  'valid to -1' case to be a success
+                 *  (i.e. ret = 0) and goto unlock.
+                 */
+                goto unlock;
+            }
+            goto reset_timer;
+        } else {
+            if (ctx->spb_choice == -1)
+                goto unlock;
+            goto set_timer;
         }
 
-        if (afr_sh_has_data_pending (xattr, child_index, this)) {
-                local->self_heal.need_data_self_heal = _gf_true;
-                gf_log(this->name, GF_LOG_DEBUG,
-                       "data self-heal is pending for %s.", local->loc.path);
+    reset_timer:
+        ret = gf_timer_call_cancel(this->ctx, ctx->timer);
+        if (ret != 0) {
+            /* We need to bail out now instead of launching a new
+             * timer. Otherwise the cbk of the previous timer event
+             * will cancel the new ctx->timer.
+             */
+            ctx->spb_choice = old_spb_choice;
+            ret = -1;
+            op_errno = EAGAIN;
+            goto unlock;
         }
-
-        ret = dict_get_uint32 (xattr, GLUSTERFS_INODELK_COUNT,
-                               &inodelk_count);
-        if (ret == 0)
-                local->inodelk_count += inodelk_count;
-
-        ret = dict_get_uint32 (xattr, GLUSTERFS_ENTRYLK_COUNT,
-                               &entrylk_count);
-        if (ret == 0)
-                local->entrylk_count += entrylk_count;
+        ctx->timer = NULL;
+        timer_reset = _gf_true;
+
+    set_timer:
+        ctx->timer = gf_timer_call_after(this->ctx, delta,
+                                         afr_set_split_brain_choice_cbk, inode);
+        if (!ctx->timer) {
+            ctx->spb_choice = old_spb_choice;
+            ret = -1;
+            op_errno = ENOMEM;
+        }
+        if (!timer_reset && ctx->timer)
+            timer_set = _gf_true;
+        if (timer_reset && !ctx->timer)
+            timer_cancelled = _gf_true;
+    }
+unlock:
+    UNLOCK(&inode->lock);
+post_unlock:
+    if (!timer_set)
+        inode_unref(inode);
+    if (timer_cancelled)
+        inode_unref(inode);
+    /*
+     * We need to invalidate the inode to prevent the kernel from serving
+     * reads from an older cached value despite a change in spb_choice to
+     * a new value.
+     */
+    inode_invalidate(inode);
+out:
+    GF_FREE(data);
+    AFR_STACK_UNWIND(setxattr, frame, ret, op_errno, NULL);
+    return 0;
 }
 
-
-static void
-afr_lookup_self_heal_check (xlator_t *this, afr_local_t *local,
-                            struct iatt *buf, struct iatt *lookup_buf)
+int
+afr_accused_fill(xlator_t *this, dict_t *xdata, unsigned char *accused,
+                 afr_transaction_type type)
 {
-        if (FILETYPE_DIFFERS (buf, lookup_buf)) {
-                /* mismatching filetypes with same name
-                */
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int idx = afr_index_for_transaction_type(type);
+    void *pending_raw = NULL;
+    int pending[3];
+    int ret = 0;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        ret = dict_get_ptr(xdata, priv->pending_key[i], &pending_raw);
+        if (ret) /* no pending flags */
+            continue;
+        memcpy(pending, pending_raw, sizeof(pending));
+
+        if (ntoh32(pending[idx]))
+            accused[i] = 1;
+    }
+
+    return 0;
+}
 
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "filetype differs for %s ", local->loc.path);
+int
+afr_accuse_smallfiles(xlator_t *this, struct afr_reply *replies,
+                      unsigned char *data_accused)
+{
+    int i = 0;
+    afr_private_t *priv = NULL;
+    uint64_t maxsize = 0;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (replies[i].valid && replies[i].xdata &&
+            dict_get_sizen(replies[i].xdata, GLUSTERFS_BAD_INODE))
+            continue;
+        if (data_accused[i])
+            continue;
+        if (replies[i].poststat.ia_size > maxsize)
+            maxsize = replies[i].poststat.ia_size;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (data_accused[i])
+            continue;
+        if (AFR_IS_ARBITER_BRICK(priv, i))
+            continue;
+        if (replies[i].poststat.ia_size < maxsize)
+            data_accused[i] = 1;
+    }
+
+    return 0;
+}
 
-                local->govinda_gOvinda = 1;
+int
+afr_readables_fill(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                   unsigned char *data_accused, unsigned char *metadata_accused,
+                   unsigned char *data_readable,
+                   unsigned char *metadata_readable, struct afr_reply *replies)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    dict_t *xdata = NULL;
+    int i = 0;
+    int ret = 0;
+    ia_type_t ia_type = IA_INVAL;
+
+    local = frame->local;
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        data_readable[i] = 1;
+        metadata_readable[i] = 1;
+    }
+    if (AFR_IS_ARBITER_BRICK(priv, ARBITER_BRICK_INDEX)) {
+        data_readable[ARBITER_BRICK_INDEX] = 0;
+        metadata_readable[ARBITER_BRICK_INDEX] = 0;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (replies) { /* Lookup */
+            if (!replies[i].valid || replies[i].op_ret == -1 ||
+                (replies[i].xdata &&
+                 dict_get_sizen(replies[i].xdata, GLUSTERFS_BAD_INODE))) {
+                data_readable[i] = 0;
+                metadata_readable[i] = 0;
+                continue;
+            }
+
+            xdata = replies[i].xdata;
+            ia_type = replies[i].poststat.ia_type;
+        } else { /* pre-op xattrop */
+            xdata = local->transaction.changelog_xdata[i];
+            ia_type = inode->ia_type;
         }
 
-        if (PERMISSION_DIFFERS (buf, lookup_buf)) {
-                /* mismatching permissions */
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "permissions differ for %s ", local->loc.path);
-                local->self_heal.need_metadata_self_heal = _gf_true;
+        if (!xdata)
+            continue; /* mkdir_cbk sends NULL xdata_rsp. */
+        afr_accused_fill(this, xdata, data_accused,
+                         (ia_type == IA_IFDIR) ? AFR_ENTRY_TRANSACTION
+                                               : AFR_DATA_TRANSACTION);
+
+        afr_accused_fill(this, xdata, metadata_accused,
+                         AFR_METADATA_TRANSACTION);
+    }
+
+    if (replies && ia_type != IA_INVAL && ia_type != IA_IFDIR &&
+        /* We want to accuse small files only when we know for
+         * sure that there is no IO happening. Otherwise, the
+         * ia_sizes obtained in post-refresh replies may
+         * mismatch due to a race between inode-refresh and
+         * ongoing writes, causing spurious heal launches*/
+        !afr_is_possibly_under_txn(AFR_DATA_TRANSACTION, local, this)) {
+        afr_accuse_smallfiles(this, replies, data_accused);
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (data_accused[i]) {
+            data_readable[i] = 0;
+            ret = 1;
         }
-
-        if (OWNERSHIP_DIFFERS (buf, lookup_buf)) {
-                /* mismatching permissions */
-                local->self_heal.need_metadata_self_heal = _gf_true;
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "ownership differs for %s ", local->loc.path);
+        if (metadata_accused[i]) {
+            metadata_readable[i] = 0;
+            ret = 1;
         }
+    }
+    return ret;
+}
 
-        if (SIZE_DIFFERS (buf, lookup_buf)
-            && IA_ISREG (buf->ia_type)) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "size differs for %s ", local->loc.path);
-                local->self_heal.need_data_self_heal = _gf_true;
+int
+afr_replies_interpret(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                      gf_boolean_t *start_heal)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    struct afr_reply *replies = NULL;
+    int event_generation = 0;
+    int i = 0;
+    unsigned char *data_accused = NULL;
+    unsigned char *metadata_accused = NULL;
+    unsigned char *data_readable = NULL;
+    unsigned char *metadata_readable = NULL;
+    int ret = 0;
+
+    local = frame->local;
+    priv = this->private;
+    replies = local->replies;
+    event_generation = local->event_generation;
+
+    data_accused = alloca0(priv->child_count);
+    data_readable = alloca0(priv->child_count);
+    metadata_accused = alloca0(priv->child_count);
+    metadata_readable = alloca0(priv->child_count);
+
+    ret = afr_readables_fill(frame, this, inode, data_accused, metadata_accused,
+                             data_readable, metadata_readable, replies);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (start_heal && priv->child_up[i] &&
+            (data_accused[i] || metadata_accused[i])) {
+            *start_heal = _gf_true;
+            break;
         }
-
+    }
+    afr_inode_read_subvol_set(inode, this, data_readable, metadata_readable,
+                              event_generation);
+    return ret;
 }
 
+int
+afr_refresh_selfheal_done(int ret, call_frame_t *heal, void *opaque)
+{
+    if (heal)
+        AFR_STACK_DESTROY(heal);
+    return 0;
+}
 
-static void
-afr_lookup_done (call_frame_t *frame, xlator_t *this, struct iatt *lookup_buf)
+int
+afr_inode_refresh_err(call_frame_t *frame, xlator_t *this)
 {
-        int  unwind = 1;
-        int  source = -1;
-        int  up_count = 0;
-        char sh_type_str[256] = {0,};
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int err = 0;
+
+    local = frame->local;
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->replies[i].valid && !local->replies[i].op_ret) {
+            err = 0;
+            goto ret;
+        }
+    }
 
-        afr_private_t *priv  = NULL;
-        afr_local_t   *local = NULL;
+    err = afr_final_errno(local, priv);
+ret:
+    return err;
+}
 
-        priv  = this->private;
-        local = frame->local;
+gf_boolean_t
+afr_selfheal_enabled(const xlator_t *this)
+{
+    const afr_private_t *priv = this->private;
 
-        local->cont.lookup.postparent.ia_ino  = local->cont.lookup.parent_ino;
+    return priv->data_self_heal || priv->metadata_self_heal ||
+           priv->entry_self_heal;
+}
 
-        if (local->cont.lookup.ino) {
-                local->cont.lookup.buf.ia_ino = local->cont.lookup.ino;
+int
+afr_txn_refresh_done(call_frame_t *frame, xlator_t *this, int err)
+{
+    call_frame_t *heal_frame = NULL;
+    afr_local_t *heal_local = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    inode_t *inode = NULL;
+    int event_generation = 0;
+    int read_subvol = -1;
+    int ret = 0;
+
+    local = frame->local;
+    inode = local->inode;
+    priv = this->private;
+
+    if (err)
+        goto refresh_done;
+
+    if (local->op == GF_FOP_LOOKUP)
+        goto refresh_done;
+
+    ret = afr_inode_get_readable(frame, inode, this, local->readable,
+                                 &event_generation, local->transaction.type);
+
+    if (ret == -EIO) {
+        /* No readable subvolume even after refresh ==> splitbrain.*/
+        if (!priv->fav_child_policy) {
+            err = EIO;
+            goto refresh_done;
         }
-
-        up_count = afr_up_children_count (priv->child_count, priv->child_up);
-        if (up_count == 1) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Only 1 child up - do not attempt to detect self heal");
-                goto unwind;
+        read_subvol = afr_sh_get_fav_by_policy(this, local->replies, inode,
+                                               NULL);
+        if (read_subvol == -1) {
+            err = EIO;
+            goto refresh_done;
         }
 
-        if (local->op_ret == 0) {
-                /* KLUDGE: assuming DHT will not itransform in
-                   revalidate */
-                if (local->cont.lookup.inode->ino) {
-                        local->cont.lookup.buf.ia_ino =
-                                local->cont.lookup.inode->ino;
-                }
+        heal_frame = afr_frame_create(this, NULL);
+        if (!heal_frame) {
+            err = EIO;
+            goto refresh_done;
         }
-
-        if (local->success_count && local->enoent_count) {
-                local->self_heal.need_metadata_self_heal = _gf_true;
-                local->self_heal.need_data_self_heal     = _gf_true;
-                local->self_heal.need_entry_self_heal    = _gf_true;
-                gf_log(this->name, GF_LOG_NORMAL,
-                       "entries are missing in lookup of %s.",
-                       local->loc.path);
+        heal_local = heal_frame->local;
+        heal_local->xdata_req = dict_new();
+        if (!heal_local->xdata_req) {
+            err = EIO;
+            AFR_STACK_DESTROY(heal_frame);
+            goto refresh_done;
         }
+        heal_local->heal_frame = frame;
+        ret = synctask_new(this->ctx->env, afr_fav_child_reset_sink_xattrs,
+                           afr_fav_child_reset_sink_xattrs_cbk, heal_frame,
+                           heal_frame);
+        return 0;
+    }
 
-        if (local->success_count) {
-                /* check for split-brain case in previous lookup */
-                if (afr_is_split_brain (this,
-                                        local->cont.lookup.inode)) {
-                        local->self_heal.need_data_self_heal = _gf_true;
-                        gf_log(this->name, GF_LOG_NORMAL,
-                               "split brain detected during lookup of "
-                               "%s.", local->loc.path);
-                }
+refresh_done:
+    afr_local_replies_wipe(local, this->private);
+    local->refreshfn(frame, this, err);
+
+    return 0;
+}
+
+int
+afr_inode_refresh_done(call_frame_t *frame, xlator_t *this, int error)
+{
+    call_frame_t *heal_frame = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    gf_boolean_t start_heal = _gf_false;
+    afr_local_t *heal_local = NULL;
+    unsigned char *success_replies = NULL;
+    int ret = 0;
+
+    if (error != 0) {
+        goto refresh_done;
+    }
+
+    local = frame->local;
+    priv = this->private;
+    success_replies = alloca0(priv->child_count);
+    afr_fill_success_replies(local, priv, success_replies);
+
+    if (priv->thin_arbiter_count && local->is_read_txn &&
+        AFR_COUNT(success_replies, priv->child_count) != priv->child_count) {
+        /* We need to query the good bricks and/or thin-arbiter.*/
+        if (success_replies[0]) {
+            local->read_txn_query_child = AFR_CHILD_ZERO;
+        } else if (success_replies[1]) {
+            local->read_txn_query_child = AFR_CHILD_ONE;
         }
+        error = EINVAL;
+        goto refresh_done;
+    }
+
+    if (!afr_has_quorum(success_replies, this, frame)) {
+        error = afr_final_errno(frame->local, this->private);
+        if (!error)
+            error = afr_quorum_errno(priv);
+        goto refresh_done;
+    }
+
+    ret = afr_replies_interpret(frame, this, local->refreshinode, &start_heal);
+
+    if (ret && afr_selfheal_enabled(this) && start_heal) {
+        heal_frame = afr_frame_create(this, NULL);
+        if (!heal_frame)
+            goto refresh_done;
+        heal_local = heal_frame->local;
+        heal_local->refreshinode = inode_ref(local->refreshinode);
+        heal_local->heal_frame = heal_frame;
+        if (!afr_throttled_selfheal(heal_frame, this)) {
+            AFR_STACK_DESTROY(heal_frame);
+            goto refresh_done;
+        }
+    }
 
-        if ((local->self_heal.need_metadata_self_heal
-             || local->self_heal.need_data_self_heal
-             || local->self_heal.need_entry_self_heal)
-            && ((!local->cont.lookup.is_revalidate)
-                || (local->op_ret != -1))) {
+refresh_done:
+    afr_txn_refresh_done(frame, this, error);
 
-                if (local->inodelk_count || local->entrylk_count) {
+    return 0;
+}
 
-                        /* Someone else is doing self-heal on this file.
-                           So just make a best effort to set the read-subvolume
-                           and return */
+void
+afr_inode_refresh_subvol_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                             int op_ret, int op_errno, struct iatt *buf,
+                             dict_t *xdata, struct iatt *par)
+{
+    afr_local_t *local = NULL;
+    int call_child = (long)cookie;
+    int8_t need_heal = 1;
+    int call_count = 0;
+    int ret = 0;
+
+    local = frame->local;
+    local->replies[call_child].valid = 1;
+    local->replies[call_child].op_ret = op_ret;
+    local->replies[call_child].op_errno = op_errno;
+    if (op_ret != -1) {
+        local->replies[call_child].poststat = *buf;
+        if (par)
+            local->replies[call_child].postparent = *par;
+        if (xdata)
+            local->replies[call_child].xdata = dict_ref(xdata);
+    }
+
+    if (xdata) {
+        ret = dict_get_int8(xdata, "link-count", &need_heal);
+        if (ret) {
+            gf_msg_debug(this->name, -ret, "Unable to get link count");
+        }
+    }
+
+    local->replies[call_child].need_heal = need_heal;
+    call_count = afr_frame_return(frame);
+    if (call_count == 0) {
+        afr_set_need_heal(this, local);
+        ret = afr_inode_refresh_err(frame, this);
+        if (ret) {
+            gf_msg_debug(this->name, ret, "afr_inode_refresh_err failed");
+        }
+        afr_inode_refresh_done(frame, this, ret);
+    }
+}
 
-                        if (IA_ISREG (local->cont.lookup.inode->ia_type)) {
-                                source = afr_self_heal_get_source (this, local, local->cont.lookup.xattrs);
+int
+afr_inode_refresh_subvol_with_lookup_cbk(call_frame_t *frame, void *cookie,
+                                         xlator_t *this, int op_ret,
+                                         int op_errno, inode_t *inode,
+                                         struct iatt *buf, dict_t *xdata,
+                                         struct iatt *par)
+{
+    afr_inode_refresh_subvol_cbk(frame, cookie, this, op_ret, op_errno, buf,
+                                 xdata, par);
+    return 0;
+}
 
-                                if (source >= 0) {
-                                        afr_set_read_child (this,
-                                                            local->cont.lookup.inode,
-                                                            source);
-                                }
-                        }
-                } else {
-                        if (!local->cont.lookup.inode->ia_type) {
-                                /* fix for RT #602 */
-                                local->cont.lookup.inode->ia_type =
-                                        lookup_buf->ia_type;
-                        }
+int
+afr_inode_refresh_subvol_with_lookup(call_frame_t *frame, xlator_t *this, int i,
+                                     inode_t *inode, uuid_t gfid, dict_t *xdata)
+{
+    loc_t loc = {
+        0,
+    };
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+
+    loc.inode = inode;
+    if (gf_uuid_is_null(inode->gfid) && gfid) {
+        /* To handle setattr/setxattr on yet to be linked inode from
+         * dht */
+        gf_uuid_copy(loc.gfid, gfid);
+    } else {
+        gf_uuid_copy(loc.gfid, inode->gfid);
+    }
+
+    STACK_WIND_COOKIE(frame, afr_inode_refresh_subvol_with_lookup_cbk,
+                      (void *)(long)i, priv->children[i],
+                      priv->children[i]->fops->lookup, &loc, xdata);
+    return 0;
+}
 
-                        local->self_heal.background = _gf_true;
-                        local->self_heal.type       = local->cont.lookup.buf.ia_type;
-                        local->self_heal.unwind     = afr_self_heal_lookup_unwind;
+int
+afr_inode_refresh_subvol_with_fstat_cbk(call_frame_t *frame, void *cookie,
+                                        xlator_t *this, int32_t op_ret,
+                                        int32_t op_errno, struct iatt *buf,
+                                        dict_t *xdata)
+{
+    afr_inode_refresh_subvol_cbk(frame, cookie, this, op_ret, op_errno, buf,
+                                 xdata, NULL);
+    return 0;
+}
 
-                        unwind = 0;
+int
+afr_inode_refresh_subvol_with_fstat(call_frame_t *frame, xlator_t *this, int i,
+                                    dict_t *xdata)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
 
-                        afr_self_heal_type_str_get(&local->self_heal,
-                                                   sh_type_str,
-                                                   sizeof(sh_type_str));
+    priv = this->private;
+    local = frame->local;
 
-                        gf_log (this->name, GF_LOG_NORMAL, "background %s "
-                                "self-heal triggered. path: %s",
-                                sh_type_str, local->loc.path);
+    STACK_WIND_COOKIE(frame, afr_inode_refresh_subvol_with_fstat_cbk,
+                      (void *)(long)i, priv->children[i],
+                      priv->children[i]->fops->fstat, local->fd, xdata);
+    return 0;
+}
 
-                        afr_self_heal (frame, this);
-                }
+int
+afr_inode_refresh_do(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int call_count = 0;
+    int i = 0;
+    int ret = 0;
+    dict_t *xdata = NULL;
+    afr_fd_ctx_t *fd_ctx = NULL;
+    unsigned char *wind_subvols = NULL;
+
+    priv = this->private;
+    local = frame->local;
+    wind_subvols = alloca0(priv->child_count);
+
+    afr_local_replies_wipe(local, priv);
+
+    if (local->fd) {
+        fd_ctx = afr_fd_ctx_get(local->fd, this);
+        if (!fd_ctx) {
+            afr_inode_refresh_done(frame, this, EINVAL);
+            return 0;
         }
+    }
 
-unwind:
-        if (unwind) {
-                AFR_STACK_UNWIND (lookup, frame, local->op_ret,
-                                  local->op_errno,
-                                  local->cont.lookup.inode,
-                                  &local->cont.lookup.buf,
-                                  local->cont.lookup.xattr,
-                                  &local->cont.lookup.postparent);
-        }
-}
+    xdata = dict_new();
+    if (!xdata) {
+        afr_inode_refresh_done(frame, this, ENOMEM);
+        return 0;
+    }
 
+    ret = afr_xattr_req_prepare(this, xdata);
+    if (ret != 0) {
+        dict_unref(xdata);
+        afr_inode_refresh_done(frame, this, -ret);
+        return 0;
+    }
 
-/*
- * During a lookup, some errors are more "important" than
- * others in that they must be given higher priority while
- * returning to the user.
- *
- * The hierarchy is ESTALE > ENOENT > others
- *
- */
+    ret = dict_set_sizen_str_sizen(xdata, "link-count", GF_XATTROP_INDEX_COUNT);
+    if (ret) {
+        gf_msg_debug(this->name, -ret, "Unable to set link-count in dict ");
+    }
 
-static gf_boolean_t
-__error_more_important (int32_t old_errno, int32_t new_errno)
-{
-        gf_boolean_t ret = _gf_true;
+    ret = dict_set_str_sizen(xdata, GLUSTERFS_INODELK_DOM_COUNT, this->name);
+    if (ret) {
+        gf_msg_debug(this->name, -ret,
+                     "Unable to set inodelk-dom-count in dict ");
+    }
 
-        /* Nothing should ever overwrite ESTALE */
-        if (old_errno == ESTALE)
-                ret = _gf_false;
+    if (local->fd) {
+        for (i = 0; i < priv->child_count; i++) {
+            if (local->child_up[i] && fd_ctx->opened_on[i] == AFR_FD_OPENED)
+                wind_subvols[i] = 1;
+        }
+    } else {
+        memcpy(wind_subvols, local->child_up,
+               sizeof(*local->child_up) * priv->child_count);
+    }
+
+    local->call_count = AFR_COUNT(wind_subvols, priv->child_count);
+
+    call_count = local->call_count;
+    if (!call_count) {
+        dict_unref(xdata);
+        if (local->fd && AFR_COUNT(local->child_up, priv->child_count))
+            afr_inode_refresh_done(frame, this, EBADFD);
+        else
+            afr_inode_refresh_done(frame, this, ENOTCONN);
+        return 0;
+    }
+    for (i = 0; i < priv->child_count; i++) {
+        if (!wind_subvols[i])
+            continue;
 
-        /* Nothing should overwrite ENOENT, except ESTALE */
-        else if ((old_errno == ENOENT) && (new_errno != ESTALE))
-                ret = _gf_false;
+        if (local->fd)
+            afr_inode_refresh_subvol_with_fstat(frame, this, i, xdata);
+        else
+            afr_inode_refresh_subvol_with_lookup(
+                frame, this, i, local->refreshinode, local->refreshgfid, xdata);
 
-        return ret;
-}
+        if (!--call_count)
+            break;
+    }
 
+    dict_unref(xdata);
+
+    return 0;
+}
 
 int
-afr_fresh_lookup_cbk (call_frame_t *frame, void *cookie,
-                      xlator_t *this,  int32_t op_ret,  int32_t op_errno,
-                      inode_t *inode,   struct iatt *buf, dict_t *xattr,
-                      struct iatt *postparent)
+afr_inode_refresh(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                  uuid_t gfid, afr_inode_refresh_cbk_t refreshfn)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-        struct iatt *   lookup_buf = NULL;
+    afr_local_t *local = NULL;
 
-        int             call_count      = -1;
-        int             child_index     = -1;
-        int             first_up_child  = -1;
+    local = frame->local;
 
-        child_index = (long) cookie;
-        priv = this->private;
+    local->refreshfn = refreshfn;
 
-        LOCK (&frame->lock);
-        {
-                local = frame->local;
+    if (local->refreshinode) {
+        inode_unref(local->refreshinode);
+        local->refreshinode = NULL;
+    }
 
-                lookup_buf = &local->cont.lookup.buf;
+    local->refreshinode = inode_ref(inode);
 
-                if (op_ret == -1) {
-                        if (op_errno == ENOENT)
-                                local->enoent_count++;
+    if (gfid)
+        gf_uuid_copy(local->refreshgfid, gfid);
+    else
+        gf_uuid_clear(local->refreshgfid);
 
-                        if (__error_more_important (local->op_errno, op_errno))
-                                local->op_errno = op_errno;
+    afr_inode_refresh_do(frame, this);
 
-                        if (local->op_errno == ESTALE) {
-                                local->op_ret = -1;
-                        }
+    return 0;
+}
 
-                        goto unlock;
-                }
+int
+afr_xattr_req_prepare(xlator_t *this, dict_t *xattr_req)
+{
+    int i = 0;
+    afr_private_t *priv = NULL;
+    int ret = 0;
 
-                afr_lookup_collect_xattr (local, this, child_index, xattr);
+    priv = this->private;
 
-                first_up_child = afr_first_up_child (priv);
+    for (i = 0; i < priv->child_count; i++) {
+        ret = dict_set_uint64(xattr_req, priv->pending_key[i],
+                              AFR_NUM_CHANGE_LOGS * sizeof(int));
+        if (ret < 0)
+            gf_msg(this->name, GF_LOG_WARNING, -ret, AFR_MSG_DICT_SET_FAILED,
+                   "Unable to set dict value for %s", priv->pending_key[i]);
+        /* 3 = data+metadata+entry */
+    }
+    ret = dict_set_uint64(xattr_req, AFR_DIRTY,
+                          AFR_NUM_CHANGE_LOGS * sizeof(int));
+    if (ret) {
+        gf_msg_debug(this->name, -ret,
+                     "failed to set dirty "
+                     "query flag");
+    }
+
+    ret = dict_set_int32_sizen(xattr_req, "list-xattr", 1);
+    if (ret) {
+        gf_msg_debug(this->name, -ret, "Unable to set list-xattr in dict ");
+    }
+
+    return ret;
+}
 
-                if (child_index == first_up_child) {
-                        local->cont.lookup.ino =
-                                afr_itransform (buf->ia_ino,
-                                                priv->child_count,
-                                                first_up_child);
-                }
+int
+afr_lookup_xattr_req_prepare(afr_local_t *local, xlator_t *this,
+                             dict_t *xattr_req, loc_t *loc)
+{
+    int ret = -ENOMEM;
+
+    if (!local->xattr_req)
+        local->xattr_req = dict_new();
+
+    if (!local->xattr_req)
+        goto out;
+
+    if (xattr_req && (xattr_req != local->xattr_req))
+        dict_copy(xattr_req, local->xattr_req);
+
+    ret = afr_xattr_req_prepare(this, local->xattr_req);
+
+    ret = dict_set_uint64(local->xattr_req, GLUSTERFS_INODELK_COUNT, 0);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, AFR_MSG_DICT_SET_FAILED,
+               "%s: Unable to set dict value for %s", loc->path,
+               GLUSTERFS_INODELK_COUNT);
+    }
+    ret = dict_set_uint64(local->xattr_req, GLUSTERFS_ENTRYLK_COUNT, 0);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, AFR_MSG_DICT_SET_FAILED,
+               "%s: Unable to set dict value for %s", loc->path,
+               GLUSTERFS_ENTRYLK_COUNT);
+    }
+
+    ret = dict_set_uint32(local->xattr_req, GLUSTERFS_PARENT_ENTRYLK, 0);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, AFR_MSG_DICT_SET_FAILED,
+               "%s: Unable to set dict value for %s", loc->path,
+               GLUSTERFS_PARENT_ENTRYLK);
+    }
+
+    ret = dict_set_sizen_str_sizen(local->xattr_req, "link-count",
+                                   GF_XATTROP_INDEX_COUNT);
+    if (ret) {
+        gf_msg_debug(this->name, -ret, "Unable to set link-count in dict ");
+    }
+
+    ret = 0;
+out:
+    return ret;
+}
 
-                if (local->success_count == 0) {
-                        if (local->op_errno != ESTALE)
-                                local->op_ret = op_ret;
-
-                        local->cont.lookup.inode               = inode_ref (inode);
-                        local->cont.lookup.xattr               = dict_ref (xattr);
-                        local->cont.lookup.xattrs[child_index] = dict_ref (xattr);
-                        local->cont.lookup.postparent          = *postparent;
-
-                        if (priv->first_lookup && inode->ino == 1) {
-                                gf_log (this->name, GF_LOG_NORMAL,
-                                        "added root inode");
-                                priv->root_inode = inode_ref (inode);
-                                priv->first_lookup = 0;
-                        }
-
-                        *lookup_buf = *buf;
-
-                        lookup_buf->ia_ino = afr_itransform (buf->ia_ino,
-                                                             priv->child_count,
-                                                             child_index);
-                        if (priv->read_child >= 0) {
-                                afr_set_read_child (this,
-                                                    local->cont.lookup.inode,
-                                                    priv->read_child);
-                        } else {
-                                afr_set_read_child (this,
-                                                    local->cont.lookup.inode,
-                                                    child_index);
-                        }
-
-                } else {
-                        afr_lookup_self_heal_check (this, local, buf, lookup_buf);
-
-                        if (child_index == local->read_child_index) {
-                                /*
-                                   lookup has succeeded on the read child.
-                                   So use its inode number
-                                */
-                                if (local->cont.lookup.xattr)
-                                        dict_unref (local->cont.lookup.xattr);
-
-                                local->cont.lookup.xattr = dict_ref (xattr);
-                                local->cont.lookup.xattrs[child_index] = dict_ref (xattr);
-                                local->cont.lookup.postparent          = *postparent;
-
-                                *lookup_buf = *buf;
-
-                                if (priv->read_child >= 0) {
-                                        afr_set_read_child (this,
-                                                            local->cont.lookup.inode,
-                                                            priv->read_child);
-                                } else {
-                                        afr_set_read_child (this,
-                                                            local->cont.lookup.inode,
-                                                            local->read_child_index);
-                                }
-                        }
+int
+afr_least_pending_reads_child(afr_private_t *priv, unsigned char *readable)
+{
+    int i = 0;
+    int child = -1;
+    int64_t read_iter = -1;
+    int64_t pending_read = -1;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (AFR_IS_ARBITER_BRICK(priv, i) || !readable[i])
+            continue;
+        read_iter = GF_ATOMIC_GET(priv->pending_reads[i]);
+        if (child == -1 || read_iter < pending_read) {
+            pending_read = read_iter;
+            child = i;
+        }
+    }
 
-                }
+    return child;
+}
 
-                local->success_count++;
-        }
-unlock:
-        UNLOCK (&frame->lock);
+static int32_t
+afr_least_latency_child(afr_private_t *priv, unsigned char *readable)
+{
+    int32_t i = 0;
+    int child = -1;
 
-        call_count = afr_frame_return (frame);
+    for (i = 0; i < priv->child_count; i++) {
+        if (AFR_IS_ARBITER_BRICK(priv, i) || !readable[i] ||
+            priv->child_latency[i] < 0)
+            continue;
 
-        if (call_count == 0) {
-                afr_lookup_done (frame, this, lookup_buf);
+        if (child == -1 ||
+            priv->child_latency[i] < priv->child_latency[child]) {
+            child = i;
         }
-
-        return 0;
+    }
+    return child;
 }
 
+static int32_t
+afr_least_latency_times_pending_reads_child(afr_private_t *priv,
+                                            unsigned char *readable)
+{
+    int32_t i = 0;
+    int child = -1;
+    int64_t pending_read = 0;
+    int64_t latency = -1;
+    int64_t least_latency = -1;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (AFR_IS_ARBITER_BRICK(priv, i) || !readable[i] ||
+            priv->child_latency[i] < 0)
+            continue;
+
+        pending_read = GF_ATOMIC_GET(priv->pending_reads[i]);
+        latency = (pending_read + 1) * priv->child_latency[i];
+
+        if (child == -1 || latency < least_latency) {
+            least_latency = latency;
+            child = i;
+        }
+    }
+    return child;
+}
 
 int
-afr_revalidate_lookup_cbk (call_frame_t *frame, void *cookie,
-                           xlator_t *this, int32_t op_ret, int32_t op_errno,
-                           inode_t *inode, struct iatt *buf, dict_t *xattr,
-                           struct iatt *postparent)
+afr_hash_child(afr_read_subvol_args_t *args, afr_private_t *priv,
+               unsigned char *readable)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-        struct iatt *   lookup_buf = NULL;
-
-        int             call_count      = -1;
-        int             child_index     = -1;
-        int             first_up_child  = -1;
-
-        child_index = (long) cookie;
-        priv = this->private;
+    uuid_t gfid_copy = {
+        0,
+    };
+    pid_t pid;
+    int child = -1;
+
+    switch (priv->hash_mode) {
+        case AFR_READ_POLICY_FIRST_UP:
+            break;
+        case AFR_READ_POLICY_GFID_HASH:
+            gf_uuid_copy(gfid_copy, args->gfid);
+            child = SuperFastHash((char *)gfid_copy, sizeof(gfid_copy)) %
+                    priv->child_count;
+            break;
+        case AFR_READ_POLICY_GFID_PID_HASH:
+            if (args->ia_type != IA_IFDIR) {
+                /*
+                 * Why getpid?  Because it's one of the cheapest calls
+                 * available - faster than gethostname etc. - and
+                 * returns a constant-length value that's sure to be
+                 * shorter than a UUID. It's still very unlikely to be
+                 * the same across clients, so it still provides good
+                 * mixing.  We're not trying for perfection here. All we
+                 * need is a low probability that multiple clients
+                 * won't converge on the same subvolume.
+                 */
+                gf_uuid_copy(gfid_copy, args->gfid);
+                pid = getpid();
+                *(pid_t *)gfid_copy ^= pid;
+            }
+            child = SuperFastHash((char *)gfid_copy, sizeof(gfid_copy)) %
+                    priv->child_count;
+            break;
+        case AFR_READ_POLICY_LESS_LOAD:
+            child = afr_least_pending_reads_child(priv, readable);
+            break;
+        case AFR_READ_POLICY_LEAST_LATENCY:
+            child = afr_least_latency_child(priv, readable);
+            break;
+        case AFR_READ_POLICY_LOAD_LATENCY_HYBRID:
+            child = afr_least_latency_times_pending_reads_child(priv, readable);
+            break;
+    }
+
+    return child;
+}
 
-        LOCK (&frame->lock);
-        {
-                local = frame->local;
+int
+afr_read_subvol_select_by_policy(inode_t *inode, xlator_t *this,
+                                 unsigned char *readable,
+                                 afr_read_subvol_args_t *args)
+{
+    int i = 0;
+    int read_subvol = -1;
+    afr_private_t *priv = NULL;
+    afr_read_subvol_args_t local_args = {
+        0,
+    };
+
+    priv = this->private;
+
+    /* first preference - explicitly specified or local subvolume */
+    if (priv->read_child >= 0 && readable[priv->read_child])
+        return priv->read_child;
+
+    if (inode_is_linked(inode)) {
+        gf_uuid_copy(local_args.gfid, inode->gfid);
+        local_args.ia_type = inode->ia_type;
+    } else if (args) {
+        local_args = *args;
+    }
+
+    /* second preference - use hashed mode */
+    read_subvol = afr_hash_child(&local_args, priv, readable);
+    if (read_subvol >= 0 && readable[read_subvol])
+        return read_subvol;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (readable[i])
+            return i;
+    }
+
+    /* no readable subvolumes, either split brain or all subvols down */
+
+    return -1;
+}
 
-                lookup_buf = &local->cont.lookup.buf;
+int
+afr_inode_read_subvol_type_get(inode_t *inode, xlator_t *this,
+                               unsigned char *readable, int *event_p, int type)
+{
+    int ret = -1;
 
-                if (op_ret == -1) {
-                        if (op_errno == ENOENT)
-                                local->enoent_count++;
+    if (type == AFR_METADATA_TRANSACTION)
+        ret = afr_inode_read_subvol_get(inode, this, 0, readable, event_p);
+    else
+        ret = afr_inode_read_subvol_get(inode, this, readable, 0, event_p);
+    return ret;
+}
 
-                        if (__error_more_important (local->op_errno, op_errno))
-                            local->op_errno = op_errno;
+void
+afr_readables_intersect_get(inode_t *inode, xlator_t *this, int *event,
+                            unsigned char *intersection)
+{
+    afr_private_t *priv = NULL;
+    unsigned char *data_readable = NULL;
+    unsigned char *metadata_readable = NULL;
+    unsigned char *intersect = NULL;
+
+    priv = this->private;
+    data_readable = alloca0(priv->child_count);
+    metadata_readable = alloca0(priv->child_count);
+    intersect = alloca0(priv->child_count);
+
+    afr_inode_read_subvol_get(inode, this, data_readable, metadata_readable,
+                              event);
+
+    AFR_INTERSECT(intersect, data_readable, metadata_readable,
+                  priv->child_count);
+    if (intersection)
+        memcpy(intersection, intersect,
+               sizeof(*intersection) * priv->child_count);
+}
 
-                            if (local->op_errno == ESTALE) {
-                                    local->op_ret = -1;
-                            }
+int
+afr_read_subvol_get(inode_t *inode, xlator_t *this, int *subvol_p,
+                    unsigned char *readables, int *event_p,
+                    afr_transaction_type type, afr_read_subvol_args_t *args)
+{
+    afr_private_t *priv = NULL;
+    unsigned char *readable = NULL;
+    unsigned char *intersection = NULL;
+    int subvol = -1;
+    int event = 0;
+
+    priv = this->private;
+
+    readable = alloca0(priv->child_count);
+    intersection = alloca0(priv->child_count);
+
+    afr_inode_read_subvol_type_get(inode, this, readable, &event, type);
+
+    afr_readables_intersect_get(inode, this, &event, intersection);
+
+    if (AFR_COUNT(intersection, priv->child_count) > 0)
+        subvol = afr_read_subvol_select_by_policy(inode, this, intersection,
+                                                  args);
+    else
+        subvol = afr_read_subvol_select_by_policy(inode, this, readable, args);
+    if (subvol_p)
+        *subvol_p = subvol;
+    if (event_p)
+        *event_p = event;
+    if (readables)
+        memcpy(readables, readable, sizeof(*readables) * priv->child_count);
+    return subvol;
+}
 
-                            goto unlock;
-                }
+void
+afr_local_transaction_cleanup(afr_local_t *local, xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    int i = 0;
 
-                afr_lookup_collect_xattr (local, this, child_index, xattr);
+    priv = this->private;
 
-                first_up_child = afr_first_up_child (priv);
+    afr_matrix_cleanup(local->pending, priv->child_count);
 
-                if (child_index == first_up_child) {
-                        local->cont.lookup.ino =
-                                afr_itransform (buf->ia_ino,
-                                                priv->child_count,
-                                                first_up_child);
-                }
+    GF_FREE(local->internal_lock.lower_locked_nodes);
 
-                /* in case of revalidate, we need to send stat of the
-                 * child whose stat was sent during the first lookup.
-                 * (so that time stamp does not vary with revalidate.
-                 * in case it is down, stat of the fist success will
-                 * be replied */
-
-                /* inode number should be preserved across revalidates */
-
-                if (local->success_count == 0) {
-                        if (local->op_errno != ESTALE)
-                                local->op_ret = op_ret;
-
-                        local->cont.lookup.inode               = inode_ref (inode);
-                        local->cont.lookup.xattr               = dict_ref (xattr);
-                        local->cont.lookup.xattrs[child_index] = dict_ref (xattr);
-                        local->cont.lookup.postparent          = *postparent;
-
-                        *lookup_buf = *buf;
-
-                        lookup_buf->ia_ino = afr_itransform (buf->ia_ino,
-                                                             priv->child_count,
-                                                             child_index);
-
-                        if (priv->read_child >= 0) {
-                                afr_set_read_child (this,
-                                                    local->cont.lookup.inode,
-                                                    priv->read_child);
-                        } else {
-                                afr_set_read_child (this,
-                                                    local->cont.lookup.inode,
-                                                    child_index);
-                        }
-
-                } else {
-                        afr_lookup_self_heal_check (this, local, buf, lookup_buf);
-
-                        if (child_index == local->read_child_index) {
-
-                                /*
-                                   lookup has succeeded on the read child.
-                                   So use its inode number
-                                */
-
-                                if (local->cont.lookup.xattr)
-                                        dict_unref (local->cont.lookup.xattr);
-
-                                local->cont.lookup.xattr               = dict_ref (xattr);
-                                local->cont.lookup.xattrs[child_index] = dict_ref (xattr);
-                                local->cont.lookup.postparent          = *postparent;
-
-                                *lookup_buf = *buf;
-
-                                if (priv->read_child >= 0) {
-                                        afr_set_read_child (this,
-                                                            local->cont.lookup.inode,
-                                                            priv->read_child);
-                                } else {
-                                        afr_set_read_child (this,
-                                                            local->cont.lookup.inode,
-                                                            local->read_child_index);
-                                }
-                        }
+    afr_lockees_cleanup(&local->internal_lock);
 
-                }
+    GF_FREE(local->transaction.pre_op);
 
-                local->success_count++;
+    GF_FREE(local->transaction.pre_op_sources);
+    if (local->transaction.changelog_xdata) {
+        for (i = 0; i < priv->child_count; i++) {
+            if (!local->transaction.changelog_xdata[i])
+                continue;
+            dict_unref(local->transaction.changelog_xdata[i]);
         }
-unlock:
-        UNLOCK (&frame->lock);
+        GF_FREE(local->transaction.changelog_xdata);
+    }
 
-        call_count = afr_frame_return (frame);
+    GF_FREE(local->transaction.failed_subvols);
 
-        if (call_count == 0) {
-                afr_lookup_done (frame, this, lookup_buf);
-        }
+    GF_FREE(local->transaction.basename);
+    GF_FREE(local->transaction.new_basename);
 
-        return 0;
+    loc_wipe(&local->transaction.parent_loc);
+    loc_wipe(&local->transaction.new_parent_loc);
 }
 
+void
+afr_reply_wipe(struct afr_reply *reply)
+{
+    if (reply->xdata) {
+        dict_unref(reply->xdata);
+        reply->xdata = NULL;
+    }
+
+    if (reply->xattr) {
+        dict_unref(reply->xattr);
+        reply->xattr = NULL;
+    }
+}
 
-int
-afr_lookup (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, dict_t *xattr_req)
+void
+afr_replies_wipe(struct afr_reply *replies, int count)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t   *local = NULL;
-        int            ret = -1;
-        int            i = 0;
+    int i = 0;
 
-        fop_lookup_cbk_t callback;
+    for (i = 0; i < count; i++) {
+        afr_reply_wipe(&replies[i]);
+    }
+}
 
-        int call_count = 0;
+void
+afr_local_replies_wipe(afr_local_t *local, afr_private_t *priv)
+{
+    if (!local->replies)
+        return;
 
-        uint64_t       ctx;
+    afr_replies_wipe(local->replies, priv->child_count);
 
-        int32_t        op_errno = 0;
+    memset(local->replies, 0, sizeof(*local->replies) * priv->child_count);
+}
 
-        priv = this->private;
+static gf_boolean_t
+afr_fop_lock_is_unlock(call_frame_t *frame)
+{
+    afr_local_t *local = frame->local;
+    switch (local->op) {
+        case GF_FOP_INODELK:
+        case GF_FOP_FINODELK:
+            if ((F_UNLCK == local->cont.inodelk.in_flock.l_type) &&
+                (local->cont.inodelk.in_cmd == F_SETLKW ||
+                 local->cont.inodelk.in_cmd == F_SETLK))
+                return _gf_true;
+            break;
+        case GF_FOP_ENTRYLK:
+        case GF_FOP_FENTRYLK:
+            if (ENTRYLK_UNLOCK == local->cont.entrylk.in_cmd)
+                return _gf_true;
+            break;
+        default:
+            return _gf_false;
+    }
+    return _gf_false;
+}
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+static gf_boolean_t
+afr_lk_is_unlock(int32_t cmd, struct gf_flock *flock)
+{
+    switch (cmd) {
+        case F_RESLK_UNLCK:
+            return _gf_true;
+            break;
 
-        local->op_ret = -1;
+#if F_SETLKW != F_SETLKW64
+        case F_SETLKW64:
+#endif
+        case F_SETLKW:
 
-        frame->local = local;
+#if F_SETLK != F_SETLK64
+        case F_SETLK64:
+#endif
+        case F_SETLK:
+            if (F_UNLCK == flock->l_type)
+                return _gf_true;
+            break;
+        default:
+            return _gf_false;
+    }
+    return _gf_false;
+}
 
-        if (!strcmp (loc->path, "/" GF_REPLICATE_TRASH_DIR)) {
-                op_errno = ENOENT;
-                goto out;
-        }
+void
+afr_handle_inconsistent_fop(call_frame_t *frame, int32_t *op_ret,
+                            int32_t *op_errno)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
 
-        loc_copy (&local->loc, loc);
+    if (!frame || !frame->this || !frame->local || !frame->this->private)
+        return;
 
-        ret = inode_ctx_get (loc->inode, this, &ctx);
-        if (ret == 0) {
-                /* lookup is a revalidate */
+    if (*op_ret < 0)
+        return;
 
-                callback = afr_revalidate_lookup_cbk;
+    /* Failing inodelk/entrylk/lk here is not a good idea because we
+     * need to cleanup the locks on the other bricks if we choose to fail
+     * the fop here. The brick may go down just after unwind happens as well
+     * so anyways the fop will fail when the next fop is sent so leaving
+     * it like this for now.*/
+    local = frame->local;
+    switch (local->op) {
+        case GF_FOP_LOOKUP:
+        case GF_FOP_INODELK:
+        case GF_FOP_FINODELK:
+        case GF_FOP_ENTRYLK:
+        case GF_FOP_FENTRYLK:
+        case GF_FOP_LK:
+            return;
+        default:
+            break;
+    }
 
-                local->cont.lookup.is_revalidate = _gf_true;
-                local->read_child_index          = afr_read_child (this,
-                                                                   loc->inode);
-        } else {
-                callback = afr_fresh_lookup_cbk;
+    priv = frame->this->private;
+    if (!priv->consistent_io)
+        return;
 
-                LOCK (&priv->read_child_lock);
-                {
-                        local->read_child_index = (++priv->read_child_rr)
-                                % (priv->child_count);
-                }
-                UNLOCK (&priv->read_child_lock);
-        }
+    if (local->event_generation &&
+        (local->event_generation != priv->event_generation))
+        goto inconsistent;
 
-        if (loc->parent)
-                local->cont.lookup.parent_ino = loc->parent->ino;
+    return;
+inconsistent:
+    *op_ret = -1;
+    *op_errno = ENOTCONN;
+}
 
-        local->child_up = memdup (priv->child_up, priv->child_count);
+void
+afr_local_cleanup(afr_local_t *local, xlator_t *this)
+{
+    afr_private_t *priv = NULL;
 
-        local->cont.lookup.xattrs = GF_CALLOC (priv->child_count,
-                                    sizeof (*local->cont.lookup.xattr),
-                                    gf_afr_mt_dict_t);
+    if (!local)
+        return;
 
-        local->call_count = afr_up_children_count (priv->child_count,
-                                                   local->child_up);
-        call_count = local->call_count;
+    syncbarrier_destroy(&local->barrier);
 
-        if (local->call_count == 0) {
-                ret      = -1;
-                op_errno = ENOTCONN;
-                goto out;
-        }
+    afr_local_transaction_cleanup(local, this);
 
-        /* By default assume ENOTCONN. On success it will be set to 0. */
-        local->op_errno = ENOTCONN;
+    priv = this->private;
 
-        if (xattr_req == NULL)
-                local->xattr_req = dict_new ();
-        else
-                local->xattr_req = dict_ref (xattr_req);
+    loc_wipe(&local->loc);
+    loc_wipe(&local->newloc);
 
-        for (i = 0; i < priv->child_count; i++) {
-                ret = dict_set_uint64 (local->xattr_req, priv->pending_key[i],
-                                       3 * sizeof(int32_t));
-                if (ret < 0)
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Unable to set dict value.");
-                /* 3 = data+metadata+entry */
-        }
+    if (local->fd)
+        fd_unref(local->fd);
 
-        ret = dict_set_uint64 (local->xattr_req, GLUSTERFS_INODELK_COUNT, 0);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "Unable to set dict value.");
-        }
+    if (local->xattr_req)
+        dict_unref(local->xattr_req);
 
-        ret = dict_set_uint64 (local->xattr_req, GLUSTERFS_ENTRYLK_COUNT, 0);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "Unable to set dict value.");
-        }
+    if (local->xattr_rsp)
+        dict_unref(local->xattr_rsp);
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND_COOKIE (frame, callback, (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->lookup,
-                                           loc, local->xattr_req);
-                        if (!--call_count)
-                                break;
-                }
-        }
+    if (local->dict)
+        dict_unref(local->dict);
 
-        ret = 0;
-out:
-        if (ret == -1)
-                AFR_STACK_UNWIND (lookup, frame, -1, op_errno,
-                                  NULL, NULL, NULL, NULL);
+    afr_local_replies_wipe(local, priv);
+    GF_FREE(local->replies);
 
-        return 0;
-}
+    GF_FREE(local->child_up);
 
+    GF_FREE(local->read_attempted);
+
+    GF_FREE(local->readable);
+    GF_FREE(local->readable2);
+
+    if (local->inode)
+        inode_unref(local->inode);
 
-/* {{{ open */
+    if (local->parent)
+        inode_unref(local->parent);
+
+    if (local->parent2)
+        inode_unref(local->parent2);
+
+    if (local->refreshinode)
+        inode_unref(local->refreshinode);
+
+    { /* getxattr */
+        GF_FREE(local->cont.getxattr.name);
+    }
+
+    { /* lk */
+        GF_FREE(local->cont.lk.locked_nodes);
+        GF_FREE(local->cont.lk.dom_locked_nodes);
+        GF_FREE(local->cont.lk.dom_lock_op_ret);
+        GF_FREE(local->cont.lk.dom_lock_op_errno);
+    }
+
+    { /* create */
+        if (local->cont.create.fd)
+            fd_unref(local->cont.create.fd);
+        if (local->cont.create.params)
+            dict_unref(local->cont.create.params);
+    }
+
+    { /* mknod */
+        if (local->cont.mknod.params)
+            dict_unref(local->cont.mknod.params);
+    }
+
+    { /* mkdir */
+        if (local->cont.mkdir.params)
+            dict_unref(local->cont.mkdir.params);
+    }
+
+    { /* symlink */
+        if (local->cont.symlink.params)
+            dict_unref(local->cont.symlink.params);
+    }
+
+    { /* writev */
+        GF_FREE(local->cont.writev.vector);
+        if (local->cont.writev.iobref)
+            iobref_unref(local->cont.writev.iobref);
+    }
+
+    { /* setxattr */
+        if (local->cont.setxattr.dict)
+            dict_unref(local->cont.setxattr.dict);
+    }
+
+    { /* fsetxattr */
+        if (local->cont.fsetxattr.dict)
+            dict_unref(local->cont.fsetxattr.dict);
+    }
+
+    { /* removexattr */
+        GF_FREE(local->cont.removexattr.name);
+    }
+    { /* xattrop */
+        if (local->cont.xattrop.xattr)
+            dict_unref(local->cont.xattrop.xattr);
+    }
+    { /* symlink */
+        GF_FREE(local->cont.symlink.linkpath);
+    }
+
+    { /* opendir */
+        GF_FREE(local->cont.opendir.checksum);
+    }
+
+    { /* open */
+        if (local->cont.open.fd)
+            fd_unref(local->cont.open.fd);
+    }
+
+    { /* readdirp */
+        if (local->cont.readdir.dict)
+            dict_unref(local->cont.readdir.dict);
+    }
+
+    { /* inodelk */
+        GF_FREE(local->cont.inodelk.volume);
+        if (local->cont.inodelk.xdata)
+            dict_unref(local->cont.inodelk.xdata);
+    }
+
+    { /* entrylk */
+        GF_FREE(local->cont.entrylk.volume);
+        GF_FREE(local->cont.entrylk.basename);
+        if (local->cont.entrylk.xdata)
+            dict_unref(local->cont.entrylk.xdata);
+    }
+
+    if (local->xdata_req)
+        dict_unref(local->xdata_req);
+
+    if (local->xdata_rsp)
+        dict_unref(local->xdata_rsp);
+}
 
 int
-afr_fd_ctx_set (xlator_t *this, fd_t *fd)
+afr_frame_return(call_frame_t *frame)
 {
-        afr_private_t * priv = NULL;
+    afr_local_t *local = NULL;
+    int call_count = 0;
 
-        int ret    = 0;
+    local = frame->local;
 
-        uint64_t       ctx;
-        afr_fd_ctx_t * fd_ctx = NULL;
+    LOCK(&frame->lock);
+    {
+        call_count = --local->call_count;
+    }
+    UNLOCK(&frame->lock);
 
-        VALIDATE_OR_GOTO (this->private, out);
-        VALIDATE_OR_GOTO (fd, out);
+    return call_count;
+}
 
-        priv = this->private;
+static char *afr_ignore_xattrs[] = {GF_SELINUX_XATTR_KEY, QUOTA_SIZE_KEY, NULL};
 
-        LOCK (&fd->lock);
-        {
-                ret = __fd_ctx_get (fd, this, &ctx);
+gf_boolean_t
+afr_is_xattr_ignorable(char *key)
+{
+    int i = 0;
+
+    if (!strncmp(key, AFR_XATTR_PREFIX, SLEN(AFR_XATTR_PREFIX)))
+        return _gf_true;
+    for (i = 0; afr_ignore_xattrs[i]; i++) {
+        if (!strcmp(key, afr_ignore_xattrs[i]))
+            return _gf_true;
+    }
+    return _gf_false;
+}
 
-                if (ret == 0)
-                        goto unlock;
+static gf_boolean_t
+afr_xattr_match_needed(dict_t *this, char *key1, data_t *value1, void *data)
+{
+    /* Ignore all non-disk (i.e. virtual) xattrs right away. */
+    if (!gf_is_valid_xattr_namespace(key1))
+        return _gf_false;
 
-                fd_ctx = GF_CALLOC (1, sizeof (afr_fd_ctx_t),
-                                    gf_afr_mt_afr_fd_ctx_t);
-                if (!fd_ctx) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Out of memory");
+    /* Ignore on-disk xattrs that AFR doesn't need to heal. */
+    if (!afr_is_xattr_ignorable(key1))
+        return _gf_true;
 
-                        ret = -ENOMEM;
-                        goto unlock;
-                }
+    return _gf_false;
+}
 
-                fd_ctx->pre_op_done = GF_CALLOC (sizeof (*fd_ctx->pre_op_done),
-                                                 priv->child_count,
-                                                 gf_afr_mt_char);
-                if (!fd_ctx->pre_op_done) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Out of memory");
-                        ret = -ENOMEM;
-                        goto unlock;
-                }
+gf_boolean_t
+afr_xattrs_are_equal(dict_t *dict1, dict_t *dict2)
+{
+    return are_dicts_equal(dict1, dict2, afr_xattr_match_needed, NULL);
+}
 
-                fd_ctx->pre_op_piggyback = GF_CALLOC (sizeof (*fd_ctx->pre_op_piggyback),
-                                                      priv->child_count,
-                                                      gf_afr_mt_char);
-                if (!fd_ctx->pre_op_piggyback) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Out of memory");
-                        ret = -ENOMEM;
-                        goto unlock;
-                }
+static int
+afr_get_parent_read_subvol(xlator_t *this, inode_t *parent,
+                           struct afr_reply *replies, unsigned char *readable)
+{
+    int i = 0;
+    int par_read_subvol = -1;
+    int par_read_subvol_iter = -1;
+    afr_private_t *priv = NULL;
 
-                fd_ctx->opened_on = GF_CALLOC (sizeof (*fd_ctx->opened_on),
-                                               priv->child_count,
-                                               gf_afr_mt_char);
-                if (!fd_ctx->opened_on) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Out of memory");
-                        ret = -ENOMEM;
-                        goto unlock;
-                }
+    priv = this->private;
 
-                fd_ctx->up_count   = priv->up_count;
-                fd_ctx->down_count = priv->down_count;
-
-                fd_ctx->locked_on = GF_CALLOC (sizeof (*fd_ctx->locked_on),
-                                               priv->child_count,
-                                               gf_afr_mt_char);
-                if (!fd_ctx->locked_on) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Out of memory");
-                        ret = -ENOMEM;
-                        goto unlock;
-                }
+    if (parent)
+        par_read_subvol = afr_data_subvol_get(parent, this, NULL, NULL, NULL,
+                                              NULL);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid)
+            continue;
 
-                ret = __fd_ctx_set (fd, this, (uint64_t)(long) fd_ctx);
+        if (replies[i].op_ret < 0)
+            continue;
 
-                INIT_LIST_HEAD (&fd_ctx->entries);
+        if (par_read_subvol_iter == -1) {
+            par_read_subvol_iter = i;
+            continue;
         }
-unlock:
-        UNLOCK (&fd->lock);
-out:
-        return ret;
-}
 
-/* {{{ flush */
+        if ((par_read_subvol_iter != par_read_subvol) && readable[i])
+            par_read_subvol_iter = i;
+
+        if (i == par_read_subvol)
+            par_read_subvol_iter = i;
+    }
+    /* At the end of the for-loop, the only reason why @par_read_subvol_iter
+     * could be -1 is when this LOOKUP has failed on all sub-volumes.
+     * So it is okay to send an arbitrary subvolume (0 in this case)
+     * as parent read subvol.
+     */
+    if (par_read_subvol_iter == -1)
+        par_read_subvol_iter = 0;
+
+    return par_read_subvol_iter;
+}
 
 int
-afr_flush_unwind (call_frame_t *frame, xlator_t *this)
+afr_read_subvol_decide(inode_t *inode, xlator_t *this,
+                       afr_read_subvol_args_t *args, unsigned char *readable)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
+    int event = 0;
+    afr_private_t *priv = NULL;
+    unsigned char *intersection = NULL;
 
-        local = frame->local;
+    priv = this->private;
+    intersection = alloca0(priv->child_count);
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
+    afr_readables_intersect_get(inode, this, &event, intersection);
 
-        if (main_frame) {
-                AFR_STACK_UNWIND (flush, main_frame,
-                                  local->op_ret, local->op_errno);
-        }
+    if (AFR_COUNT(intersection, priv->child_count) <= 0) {
+        /* TODO: If we have one brick with valid data_readable and
+         * another with metadata_readable, try to send an iatt with
+         * valid bits from both.*/
+        return -1;
+    }
 
-        return 0;
+    memcpy(readable, intersection, sizeof(*readable) * priv->child_count);
+
+    return afr_read_subvol_select_by_policy(inode, this, intersection, args);
 }
 
+static inline int
+afr_first_up_child(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i = 0;
 
-int
-afr_flush_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno)
+    local = frame->local;
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++)
+        if (local->replies[i].valid && local->replies[i].op_ret == 0)
+            return i;
+    return -1;
+}
+
+static void
+afr_attempt_readsubvol_set(call_frame_t *frame, xlator_t *this,
+                           unsigned char *success_replies,
+                           unsigned char *data_readable, int *read_subvol)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int spb_subvol = -1;
+    int child_count = -1;
 
-        int call_count  = -1;
-        int child_index = (long) cookie;
-        int need_unwind = 0;
+    if (*read_subvol != -1)
+        return;
 
-        local = frame->local;
-        priv  = this->private;
+    priv = this->private;
+    local = frame->local;
+    child_count = priv->child_count;
+
+    afr_split_brain_read_subvol_get(local->inode, this, frame, &spb_subvol);
+    if ((spb_subvol >= 0) &&
+        (AFR_COUNT(success_replies, child_count) == child_count)) {
+        *read_subvol = spb_subvol;
+    } else if (!priv->quorum_count ||
+               frame->root->pid == GF_CLIENT_PID_GLFS_HEAL) {
+        *read_subvol = afr_first_up_child(frame, this);
+    } else if (priv->quorum_count &&
+               afr_has_quorum(data_readable, this, NULL)) {
+        /* read_subvol is guaranteed to be valid if we hit this path. */
+        *read_subvol = afr_first_up_child(frame, this);
+    } else {
+        /* If quorum is enabled and we do not have a
+           readable yet, it means all good copies are down.
+        */
+        local->op_ret = -1;
+        local->op_errno = ENOTCONN;
+        gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_READ_SUBVOL_ERROR,
+               "no read "
+               "subvols for %s",
+               local->loc.path);
+    }
+    if (*read_subvol >= 0)
+        dict_del_sizen(local->replies[*read_subvol].xdata, GF_CONTENT_KEY);
+}
 
-        LOCK (&frame->lock);
-        {
-                if (afr_fop_failed (op_ret, op_errno))
-                        afr_transaction_fop_failed (frame, this, child_index);
-
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                        }
-                        local->success_count++;
-
-                        if (local->success_count == priv->wait_count) {
-                                need_unwind = 1;
-                        }
-                }
+static void
+afr_lookup_done(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i = -1;
+    int op_errno = 0;
+    int read_subvol = 0;
+    int par_read_subvol = 0;
+    int ret = -1;
+    unsigned char *readable = NULL;
+    unsigned char *success_replies = NULL;
+    int event = 0;
+    struct afr_reply *replies = NULL;
+    uuid_t read_gfid = {
+        0,
+    };
+    gf_boolean_t locked_entry = _gf_false;
+    gf_boolean_t in_flight_create = _gf_false;
+    gf_boolean_t can_interpret = _gf_true;
+    inode_t *parent = NULL;
+    ia_type_t ia_type = IA_INVAL;
+    afr_read_subvol_args_t args = {
+        0,
+    };
+    char *gfid_heal_msg = NULL;
+
+    priv = this->private;
+    local = frame->local;
+    replies = local->replies;
+    parent = local->loc.parent;
+
+    locked_entry = afr_is_possibly_under_txn(AFR_ENTRY_TRANSACTION, local,
+                                             this);
+
+    readable = alloca0(priv->child_count);
+    success_replies = alloca0(priv->child_count);
+
+    afr_inode_read_subvol_get(parent, this, readable, NULL, &event);
+    par_read_subvol = afr_get_parent_read_subvol(this, parent, replies,
+                                                 readable);
+
+    /* First, check if we have a gfid-change from somewhere,
+       If so, propagate that so that a fresh lookup can be
+       issued
+    */
+    if (local->cont.lookup.needs_fresh_lookup) {
+        local->op_ret = -1;
+        local->op_errno = ESTALE;
+        goto error;
+    }
 
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    op_errno = afr_final_errno(frame->local, this->private);
+    local->op_errno = op_errno;
 
-        if (need_unwind)
-                afr_flush_unwind (frame, this);
+    read_subvol = -1;
+    afr_fill_success_replies(local, priv, success_replies);
 
-        call_count = afr_frame_return (frame);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid)
+            continue;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
+        if (replies[i].op_ret == -1) {
+            if (locked_entry && replies[i].op_errno == ENOENT) {
+                in_flight_create = _gf_true;
+            }
+            continue;
         }
 
-        return 0;
-}
+        if (read_subvol == -1 || !readable[read_subvol]) {
+            read_subvol = i;
+            gf_uuid_copy(read_gfid, replies[i].poststat.ia_gfid);
+            ia_type = replies[i].poststat.ia_type;
+            local->op_ret = 0;
+        }
+    }
+
+    if (in_flight_create && !afr_has_quorum(success_replies, this, NULL)) {
+        local->op_ret = -1;
+        local->op_errno = ENOENT;
+        goto error;
+    }
+
+    if (read_subvol == -1)
+        goto error;
+    /* We now have a read_subvol, which is readable[] (if there
+       were any). Next we look for GFID mismatches. We don't
+       consider a GFID mismatch as an error if read_subvol is
+       readable[] but the mismatching GFID subvol is not.
+    */
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret == -1) {
+            continue;
+        }
 
+        if (!gf_uuid_compare(replies[i].poststat.ia_gfid, read_gfid))
+            continue;
 
-int
-afr_flush_wind (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
+        can_interpret = _gf_false;
 
-        int i = 0;
-        int call_count = -1;
+        if (locked_entry)
+            continue;
 
-        local = frame->local;
-        priv = this->private;
+        /* Now GFIDs mismatch. It's OK as long as this subvol
+           is not readable[] but read_subvol is */
+        if (readable[read_subvol] && !readable[i])
+            continue;
 
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
+        /* If we were called from glfsheal and there is still a gfid
+         * mismatch, succeed the lookup and let glfsheal print the
+         * response via gfid-heal-msg.*/
+        if (!dict_get_str_sizen(local->xattr_req, "gfid-heal-msg",
+                                &gfid_heal_msg))
+            goto cant_interpret;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
+        /* LOG ERROR */
+        local->op_ret = -1;
+        local->op_errno = EIO;
+        goto error;
+    }
+
+    /* Forth, for the finalized GFID, pick the best subvolume
+       to return stats from.
+    */
+    read_subvol = -1;
+    memset(readable, 0, sizeof(*readable) * priv->child_count);
+    if (can_interpret) {
+        if (!afr_has_quorum(success_replies, this, NULL))
+            goto cant_interpret;
+        /* It is safe to call afr_replies_interpret() because we have
+           a response from all the UP subvolumes and all of them resolved
+           to the same GFID
+        */
+        gf_uuid_copy(args.gfid, read_gfid);
+        args.ia_type = ia_type;
+        ret = afr_replies_interpret(frame, this, local->inode, NULL);
+        read_subvol = afr_read_subvol_decide(local->inode, this, &args,
+                                             readable);
+        if (read_subvol == -1)
+            goto cant_interpret;
+        if (ret) {
+            afr_inode_need_refresh_set(local->inode, this);
+            dict_del_sizen(local->replies[read_subvol].xdata, GF_CONTENT_KEY);
         }
+    } else {
+    cant_interpret:
+        afr_attempt_readsubvol_set(frame, this, success_replies, readable,
+                                   &read_subvol);
+        if (read_subvol == -1) {
+            goto error;
+        }
+    }
 
-        local->call_count = call_count;
+    afr_handle_quota_size(frame, this);
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND_COOKIE (frame, afr_flush_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->flush,
-                                           local->fd);
-
-                        if (!--call_count)
-                                break;
-                }
+    afr_set_need_heal(this, local);
+    if (AFR_IS_ARBITER_BRICK(priv, read_subvol) && local->op_ret == 0) {
+        local->op_ret = -1;
+        local->op_errno = ENOTCONN;
+        gf_msg_debug(this->name, 0,
+                     "Arbiter cannot be a read subvol "
+                     "for %s",
+                     local->loc.path);
+        goto error;
+    }
+
+    ret = dict_get_str_sizen(local->xattr_req, "gfid-heal-msg", &gfid_heal_msg);
+    if (!ret) {
+        ret = dict_set_str_sizen(local->replies[read_subvol].xdata,
+                                 "gfid-heal-msg", gfid_heal_msg);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_DICT_SET_FAILED,
+                   "Error setting gfid-heal-msg dict");
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
         }
+    }
 
-        return 0;
+    AFR_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno,
+                     local->inode, &local->replies[read_subvol].poststat,
+                     local->replies[read_subvol].xdata,
+                     &local->replies[par_read_subvol].postparent);
+    return;
+
+error:
+    AFR_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno, NULL, NULL,
+                     NULL, NULL);
 }
 
+/*
+ * During a lookup, some errors are more "important" than
+ * others in that they must be given higher priority while
+ * returning to the user.
+ *
+ * The hierarchy is ENODATA > ENOENT > ESTALE > ENOSPC others
+ */
 
 int
-afr_flush_done (call_frame_t *frame, xlator_t *this)
+afr_higher_errno(int32_t old_errno, int32_t new_errno)
 {
-        afr_local_t *local = NULL;
+    if (old_errno == ENODATA || new_errno == ENODATA)
+        return ENODATA;
+    if (old_errno == ENOENT || new_errno == ENOENT)
+        return ENOENT;
+    if (old_errno == ESTALE || new_errno == ESTALE)
+        return ESTALE;
+    if (old_errno == ENOSPC || new_errno == ENOSPC)
+        return ENOSPC;
+
+    return new_errno;
+}
 
-        local = frame->local;
+int
+afr_final_errno(afr_local_t *local, afr_private_t *priv)
+{
+    int i = 0;
+    int op_errno = 0;
+    int tmp_errno = 0;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
+        if (local->replies[i].op_ret >= 0)
+            continue;
+        tmp_errno = local->replies[i].op_errno;
+        op_errno = afr_higher_errno(op_errno, tmp_errno);
+    }
+
+    return op_errno;
+}
 
-        local->transaction.unwind (frame, this);
+static int32_t
+afr_local_discovery_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *dict,
+                        dict_t *xdata)
+{
+    int ret = 0;
+    char *pathinfo = NULL;
+    gf_boolean_t is_local = _gf_false;
+    afr_private_t *priv = NULL;
+    int32_t child_index = -1;
+
+    if (op_ret != 0) {
+        goto out;
+    }
+
+    priv = this->private;
+    child_index = (int32_t)(long)cookie;
+
+    ret = dict_get_str_sizen(dict, GF_XATTR_PATHINFO_KEY, &pathinfo);
+    if (ret != 0) {
+        goto out;
+    }
+
+    ret = glusterfs_is_local_pathinfo(pathinfo, &is_local);
+    if (ret) {
+        goto out;
+    }
+
+    /*
+     * Note that one local subvolume will override another here.  The only
+     * way to avoid that would be to retain extra information about whether
+     * the previous read_child is local, and it's just not worth it.  Even
+     * the slowest local subvolume is far preferable to a remote one.
+     */
+    if (is_local) {
+        priv->local[child_index] = 1;
+        /* Don't set arbiter as read child. */
+        if (AFR_IS_ARBITER_BRICK(priv, child_index))
+            goto out;
+        gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_LOCAL_CHILD,
+               "selecting local read_child %s",
+               priv->children[child_index]->name);
+
+        priv->read_child = child_index;
+    }
+out:
+    STACK_DESTROY(frame->root);
+    return 0;
+}
 
-        AFR_STACK_DESTROY (frame);
+static void
+afr_attempt_local_discovery(xlator_t *this, int32_t child_index)
+{
+    call_frame_t *newframe = NULL;
+    loc_t tmploc = {
+        0,
+    };
+    afr_private_t *priv = this->private;
+
+    newframe = create_frame(this, this->ctx->pool);
+    if (!newframe) {
+        return;
+    }
 
-        return 0;
+    tmploc.gfid[sizeof(tmploc.gfid) - 1] = 1;
+    STACK_WIND_COOKIE(newframe, afr_local_discovery_cbk,
+                      (void *)(long)child_index, priv->children[child_index],
+                      priv->children[child_index]->fops->getxattr, &tmploc,
+                      GF_XATTR_PATHINFO_KEY, NULL);
 }
 
-
 int
-afr_flush (call_frame_t *frame, xlator_t *this, fd_t *fd)
+afr_lookup_sh_metadata_wrap(void *opaque)
 {
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
+    call_frame_t *frame = opaque;
+    afr_local_t *local = NULL;
+    xlator_t *this = NULL;
+    inode_t *inode = NULL;
+    afr_private_t *priv = NULL;
+    struct afr_reply *replies = NULL;
+    int i = 0, first = -1;
+    int ret = -1;
+    dict_t *dict = NULL;
+
+    local = frame->local;
+    this = frame->this;
+    priv = this->private;
+    replies = local->replies;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret == -1)
+            continue;
+        first = i;
+        break;
+    }
+    if (first == -1)
+        goto out;
+
+    if (afr_selfheal_metadata_by_stbuf(this, &replies[first].poststat))
+        goto out;
+
+    afr_local_replies_wipe(local, this->private);
+
+    dict = dict_new();
+    if (!dict)
+        goto out;
+    if (local->xattr_req) {
+        dict_copy(local->xattr_req, dict);
+    }
+
+    ret = dict_set_sizen_str_sizen(dict, "link-count", GF_XATTROP_INDEX_COUNT);
+    if (ret) {
+        gf_msg_debug(this->name, -ret, "Unable to set link-count in dict ");
+    }
+
+    if (loc_is_nameless(&local->loc)) {
+        ret = afr_selfheal_unlocked_discover_on(frame, local->inode,
+                                                local->loc.gfid, local->replies,
+                                                local->child_up, dict);
+    } else {
+        inode = afr_selfheal_unlocked_lookup_on(frame, local->loc.parent,
+                                                local->loc.name, local->replies,
+                                                local->child_up, dict);
+    }
+    if (inode)
+        inode_unref(inode);
+out:
+    if (loc_is_nameless(&local->loc))
+        afr_discover_done(frame, this);
+    else
+        afr_lookup_done(frame, this);
+
+    if (dict)
+        dict_unref(dict);
+
+    return 0;
+}
 
-        call_frame_t  * transaction_frame = NULL;
+gf_boolean_t
+afr_is_pending_set(xlator_t *this, dict_t *xdata, int type)
+{
+    int idx = -1;
+    afr_private_t *priv = NULL;
+    void *pending_raw = NULL;
+    int *pending_int = NULL;
+    int i = 0;
 
-        int ret        = -1;
+    priv = this->private;
+    idx = afr_index_for_transaction_type(type);
 
-        int op_ret   = -1;
-        int op_errno = 0;
+    if (dict_get_ptr(xdata, AFR_DIRTY, &pending_raw) == 0) {
+        if (pending_raw) {
+            pending_int = pending_raw;
 
-        int call_count = 0;
+            if (ntoh32(pending_int[idx]))
+                return _gf_true;
+        }
+    }
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    for (i = 0; i < priv->child_count; i++) {
+        if (dict_get_ptr(xdata, priv->pending_key[i], &pending_raw))
+            continue;
+        if (!pending_raw)
+            continue;
+        pending_int = pending_raw;
 
-        priv = this->private;
+        if (ntoh32(pending_int[idx]))
+            return _gf_true;
+    }
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+    return _gf_false;
+}
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
+static gf_boolean_t
+afr_can_start_metadata_self_heal(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    struct afr_reply *replies = NULL;
+    int i = 0, first = -1;
+    gf_boolean_t start = _gf_false;
+    struct iatt stbuf = {
+        0,
+    };
+
+    local = frame->local;
+    replies = local->replies;
+    priv = this->private;
+
+    if (!priv->metadata_self_heal)
+        return _gf_false;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret == -1)
+            continue;
+        if (first == -1) {
+            first = i;
+            stbuf = replies[i].poststat;
+            continue;
         }
 
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
+        if (afr_is_pending_set(this, replies[i].xdata,
+                               AFR_METADATA_TRANSACTION)) {
+            /* Let shd do the heal so that lookup is not blocked
+             * on getting metadata lock/doing the heal */
+            start = _gf_false;
+            break;
+        }
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory.");
-                goto out;
+        if (gf_uuid_compare(stbuf.ia_gfid, replies[i].poststat.ia_gfid)) {
+            start = _gf_false;
+            break;
+        }
+        if (!IA_EQUAL(stbuf, replies[i].poststat, type)) {
+            start = _gf_false;
+            break;
         }
 
-        transaction_frame->local = local;
+        /*Check if iattrs need heal*/
+        if ((!IA_EQUAL(stbuf, replies[i].poststat, uid)) ||
+            (!IA_EQUAL(stbuf, replies[i].poststat, gid)) ||
+            (!IA_EQUAL(stbuf, replies[i].poststat, prot))) {
+            start = _gf_true;
+            continue;
+        }
 
-        local->op = GF_FOP_FLUSH;
+        /*Check if xattrs need heal*/
+        if (!afr_xattrs_are_equal(replies[first].xdata, replies[i].xdata))
+            start = _gf_true;
+    }
 
-        local->transaction.fop    = afr_flush_wind;
-        local->transaction.done   = afr_flush_done;
-        local->transaction.unwind = afr_flush_unwind;
+    return start;
+}
 
-        local->fd                 = fd_ref (fd);
+int
+afr_lookup_metadata_heal_check(call_frame_t *frame, xlator_t *this)
 
-        local->transaction.main_frame = frame;
-        local->transaction.start  = 0;
-        local->transaction.len    = 0;
+{
+    call_frame_t *heal = NULL;
+    afr_local_t *local = NULL;
+    int ret = 0;
+
+    local = frame->local;
+    if (!afr_can_start_metadata_self_heal(frame, this))
+        goto out;
+
+    heal = afr_frame_create(this, &ret);
+    if (!heal) {
+        ret = -ret;
+        goto out;
+    }
+
+    ret = synctask_new(this->ctx->env, afr_lookup_sh_metadata_wrap,
+                       afr_refresh_selfheal_done, heal, frame);
+    if (ret)
+        goto out;
+    return ret;
+out:
+    if (loc_is_nameless(&local->loc))
+        afr_discover_done(frame, this);
+    else
+        afr_lookup_done(frame, this);
+    if (heal)
+        AFR_STACK_DESTROY(heal);
+    return ret;
+}
 
-        afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+int
+afr_lookup_selfheal_wrap(void *opaque)
+{
+    int ret = 0;
+    call_frame_t *frame = opaque;
+    afr_local_t *local = NULL;
+    xlator_t *this = NULL;
+    inode_t *inode = NULL;
+    uuid_t pargfid = {
+        0,
+    };
+
+    local = frame->local;
+    this = frame->this;
+    loc_pargfid(&local->loc, pargfid);
+
+    ret = afr_selfheal_name(frame->this, pargfid, local->loc.name,
+                            &local->cont.lookup.gfid_req, local->xattr_req);
+    if (ret == -EIO)
+        goto unwind;
+
+    afr_local_replies_wipe(local, this->private);
+
+    inode = afr_selfheal_unlocked_lookup_on(frame, local->loc.parent,
+                                            local->loc.name, local->replies,
+                                            local->child_up, local->xattr_req);
+    if (inode)
+        inode_unref(inode);
+
+    afr_lookup_metadata_heal_check(frame, this);
+    return 0;
 
+unwind:
+    AFR_STACK_UNWIND(lookup, frame, -1, EIO, NULL, NULL, NULL, NULL);
+    return 0;
+}
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
+int
+afr_lookup_entry_heal(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    call_frame_t *heal = NULL;
+    int i = 0, first = -1;
+    gf_boolean_t name_state_mismatch = _gf_false;
+    struct afr_reply *replies = NULL;
+    int ret = 0;
+    unsigned char *par_readables = NULL;
+    unsigned char *success = NULL;
+    int32_t op_errno = 0;
+    uuid_t gfid = {0};
+
+    local = frame->local;
+    replies = local->replies;
+    priv = this->private;
+    par_readables = alloca0(priv->child_count);
+    success = alloca0(priv->child_count);
+
+    ret = afr_inode_read_subvol_get(local->loc.parent, this, par_readables,
+                                    NULL, NULL);
+    if (ret < 0 || AFR_COUNT(par_readables, priv->child_count) == 0) {
+        /* In this case set par_readables to all 1 so that name_heal
+         * need checks at the end of this function will flag missing
+         * entry when name state mismatches*/
+        memset(par_readables, 1, priv->child_count);
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid)
+            continue;
+
+        if (replies[i].op_ret == 0) {
+            if (gf_uuid_is_null(gfid)) {
+                gf_uuid_copy(gfid, replies[i].poststat.ia_gfid);
+            }
+            success[i] = 1;
+        } else {
+            if ((replies[i].op_errno != ENOTCONN) &&
+                (replies[i].op_errno != ENOENT) &&
+                (replies[i].op_errno != ESTALE)) {
+                op_errno = replies[i].op_errno;
+            }
+        }
 
-                AFR_STACK_UNWIND (flush, frame, op_ret, op_errno);
+        /*gfid is missing, needs heal*/
+        if ((replies[i].op_ret == -1) && (replies[i].op_errno == ENODATA)) {
+            goto name_heal;
         }
 
-        return 0;
-}
+        if (first == -1) {
+            first = i;
+            continue;
+        }
 
-/* }}} */
+        if (replies[i].op_ret != replies[first].op_ret) {
+            name_state_mismatch = _gf_true;
+        }
+
+        if (replies[i].op_ret == 0) {
+            /* Rename after this lookup may succeed if we don't do
+             * a name-heal and the destination may not have pending xattrs
+             * to indicate which name is good and which is bad so always do
+             * this heal*/
+            if (gf_uuid_compare(replies[i].poststat.ia_gfid, gfid)) {
+                goto name_heal;
+            }
+        }
+    }
+
+    if (name_state_mismatch) {
+        if (!priv->quorum_count)
+            goto name_heal;
+        if (!afr_has_quorum(success, this, NULL))
+            goto name_heal;
+        if (op_errno)
+            goto name_heal;
+        for (i = 0; i < priv->child_count; i++) {
+            if (!replies[i].valid)
+                continue;
+            if (par_readables[i] && replies[i].op_ret < 0 &&
+                replies[i].op_errno != ENOTCONN) {
+                goto name_heal;
+            }
+        }
+    }
 
+    goto metadata_heal;
+
+name_heal:
+    heal = afr_frame_create(this, NULL);
+    if (!heal)
+        goto metadata_heal;
+
+    ret = synctask_new(this->ctx->env, afr_lookup_selfheal_wrap,
+                       afr_refresh_selfheal_done, heal, frame);
+    if (ret) {
+        AFR_STACK_DESTROY(heal);
+        goto metadata_heal;
+    }
+    return ret;
+
+metadata_heal:
+    ret = afr_lookup_metadata_heal_check(frame, this);
+
+    return ret;
+}
 
 int
-afr_cleanup_fd_ctx (xlator_t *this, fd_t *fd)
+afr_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, inode_t *inode, struct iatt *buf, dict_t *xdata,
+               struct iatt *postparent)
 {
-        uint64_t        ctx = 0;
-        afr_fd_ctx_t    *fd_ctx = NULL;
-        int             ret = 0;
+    afr_local_t *local = NULL;
+    int call_count = -1;
+    int child_index = -1;
+    GF_UNUSED int ret = 0;
+    int8_t need_heal = 1;
+
+    child_index = (long)cookie;
+
+    local = frame->local;
+
+    local->replies[child_index].valid = 1;
+    local->replies[child_index].op_ret = op_ret;
+    local->replies[child_index].op_errno = op_errno;
+    /*
+     * On revalidate lookup if the gfid-changed, afr should unwind the fop
+     * with ESTALE so that a fresh lookup will be sent by the top xlator.
+     * So remember it.
+     */
+    if (xdata && dict_get_sizen(xdata, "gfid-changed"))
+        local->cont.lookup.needs_fresh_lookup = _gf_true;
+
+    if (xdata) {
+        ret = dict_get_int8(xdata, "link-count", &need_heal);
+        local->replies[child_index].need_heal = need_heal;
+    } else {
+        local->replies[child_index].need_heal = need_heal;
+    }
+    if (op_ret != -1) {
+        local->replies[child_index].poststat = *buf;
+        local->replies[child_index].postparent = *postparent;
+        if (xdata)
+            local->replies[child_index].xdata = dict_ref(xdata);
+    }
+
+    call_count = afr_frame_return(frame);
+    if (call_count == 0) {
+        afr_set_need_heal(this, local);
+        afr_lookup_entry_heal(frame, this);
+    }
+
+    return 0;
+}
 
-        ret = fd_ctx_get (fd, this, &ctx);
+static void
+afr_discover_unwind(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int read_subvol = -1;
+    int ret = 0;
+    unsigned char *data_readable = NULL;
+    unsigned char *success_replies = NULL;
+
+    priv = this->private;
+    local = frame->local;
+    data_readable = alloca0(priv->child_count);
+    success_replies = alloca0(priv->child_count);
+
+    afr_fill_success_replies(local, priv, success_replies);
+    if (AFR_COUNT(success_replies, priv->child_count) > 0)
+        local->op_ret = 0;
+
+    if (local->op_ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = afr_final_errno(frame->local, this->private);
+        goto error;
+    }
 
-        if (ret < 0)
-                goto out;
+    if (!afr_has_quorum(success_replies, this, frame))
+        goto unwind;
 
-        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+    ret = afr_replies_interpret(frame, this, local->inode, NULL);
+    if (ret) {
+        afr_inode_need_refresh_set(local->inode, this);
+    }
 
-        if (fd_ctx) {
-                if (fd_ctx->pre_op_done)
-                        GF_FREE (fd_ctx->pre_op_done);
+    read_subvol = afr_read_subvol_decide(local->inode, this, NULL,
+                                         data_readable);
 
-                if (fd_ctx->opened_on)
-                        GF_FREE (fd_ctx->opened_on);
+unwind:
+    afr_attempt_readsubvol_set(frame, this, success_replies, data_readable,
+                               &read_subvol);
+    if (read_subvol == -1)
+        goto error;
 
-                if (fd_ctx->locked_on)
-                        GF_FREE (fd_ctx->locked_on);
+    if (AFR_IS_ARBITER_BRICK(priv, read_subvol) && local->op_ret == 0) {
+        local->op_ret = -1;
+        local->op_errno = ENOTCONN;
+        gf_msg_debug(this->name, 0,
+                     "Arbiter cannot be a read subvol "
+                     "for %s",
+                     local->loc.path);
+    }
+
+    AFR_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno,
+                     local->inode, &local->replies[read_subvol].poststat,
+                     local->replies[read_subvol].xdata,
+                     &local->replies[read_subvol].postparent);
+    return;
+
+error:
+    AFR_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno, NULL, NULL,
+                     NULL, NULL);
+}
 
-                GF_FREE (fd_ctx);
-        }
+static int
+afr_ta_id_file_check(void *opaque)
+{
+    afr_private_t *priv = NULL;
+    xlator_t *this = NULL;
+    loc_t loc = {
+        0,
+    };
+    struct iatt stbuf = {
+        0,
+    };
+    dict_t *dict = NULL;
+    uuid_t gfid = {
+        0,
+    };
+    fd_t *fd = NULL;
+    int ret = 0;
+
+    this = opaque;
+    priv = this->private;
+
+    ret = afr_fill_ta_loc(this, &loc, _gf_false);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to populate thin-arbiter loc for: %s.", loc.name);
+        goto out;
+    }
+
+    ret = syncop_lookup(priv->children[THIN_ARBITER_BRICK_INDEX], &loc, &stbuf,
+                        0, 0, 0);
+    if (ret == 0) {
+        goto out;
+    } else if (ret == -ENOENT) {
+        fd = fd_create(loc.inode, getpid());
+        if (!fd)
+            goto out;
+        dict = dict_new();
+        if (!dict)
+            goto out;
+        gf_uuid_generate(gfid);
+        ret = dict_set_gfuuid(dict, "gfid-req", gfid, true);
+        ret = syncop_create(priv->children[THIN_ARBITER_BRICK_INDEX], &loc,
+                            O_RDWR, 0664, fd, &stbuf, dict, NULL);
+    }
 
 out:
-        return 0;
+    if (ret == 0) {
+        gf_uuid_copy(priv->ta_gfid, stbuf.ia_gfid);
+    } else {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to lookup/create thin-arbiter id file.");
+    }
+    if (dict)
+        dict_unref(dict);
+    if (fd)
+        fd_unref(fd);
+    loc_wipe(&loc);
+
+    return 0;
 }
 
+static int
+afr_ta_id_file_check_cbk(int ret, call_frame_t *ta_frame, void *opaque)
+{
+    return 0;
+}
+
+static void
+afr_discover_done(call_frame_t *frame, xlator_t *this)
+{
+    int ret = 0;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    if (!priv->thin_arbiter_count)
+        goto unwind;
+    if (!gf_uuid_is_null(priv->ta_gfid))
+        goto unwind;
+
+    ret = synctask_new(this->ctx->env, afr_ta_id_file_check,
+                       afr_ta_id_file_check_cbk, NULL, this);
+    if (ret)
+        goto unwind;
+unwind:
+    afr_discover_unwind(frame, this);
+}
 
 int
-afr_release (xlator_t *this, fd_t *fd)
+afr_discover_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, inode_t *inode, struct iatt *buf, dict_t *xdata,
+                 struct iatt *postparent)
 {
-        afr_locked_fd_t *locked_fd = NULL;
-        afr_locked_fd_t *tmp       = NULL;
-        afr_private_t   *priv      = NULL;
+    afr_local_t *local = NULL;
+    int call_count = -1;
+    int child_index = -1;
+    GF_UNUSED int ret = 0;
+    int8_t need_heal = 1;
+
+    child_index = (long)cookie;
+
+    local = frame->local;
+
+    local->replies[child_index].valid = 1;
+    local->replies[child_index].op_ret = op_ret;
+    local->replies[child_index].op_errno = op_errno;
+    if (op_ret != -1) {
+        local->replies[child_index].poststat = *buf;
+        local->replies[child_index].postparent = *postparent;
+        if (xdata)
+            local->replies[child_index].xdata = dict_ref(xdata);
+    }
+
+    if (local->do_discovery && (op_ret == 0))
+        afr_attempt_local_discovery(this, child_index);
+
+    if (xdata) {
+        ret = dict_get_int8(xdata, "link-count", &need_heal);
+        local->replies[child_index].need_heal = need_heal;
+    } else {
+        local->replies[child_index].need_heal = need_heal;
+    }
+
+    call_count = afr_frame_return(frame);
+    if (call_count == 0) {
+        afr_set_need_heal(this, local);
+        afr_lookup_metadata_heal_check(frame, this);
+    }
+
+    return 0;
+}
 
-        priv = this->private;
+int
+afr_discover_do(call_frame_t *frame, xlator_t *this, int err)
+{
+    int ret = 0;
+    int i = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int call_count = 0;
+
+    local = frame->local;
+    priv = this->private;
+
+    if (err) {
+        local->op_errno = err;
+        goto out;
+    }
+
+    call_count = local->call_count = AFR_COUNT(local->child_up,
+                                               priv->child_count);
+
+    ret = afr_lookup_xattr_req_prepare(local, this, local->xattr_req,
+                                       &local->loc);
+    if (ret) {
+        local->op_errno = -ret;
+        goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->child_up[i]) {
+            STACK_WIND_COOKIE(
+                frame, afr_discover_cbk, (void *)(long)i, priv->children[i],
+                priv->children[i]->fops->lookup, &local->loc, local->xattr_req);
+            if (!--call_count)
+                break;
+        }
+    }
+
+    return 0;
+out:
+    AFR_STACK_UNWIND(lookup, frame, -1, local->op_errno, 0, 0, 0, 0);
+    return 0;
+}
 
-        afr_cleanup_fd_ctx (this, fd);
+int
+afr_discover(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
+{
+    int op_errno = ENOMEM;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int event = 0;
+
+    priv = this->private;
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    if (!local->call_count) {
+        op_errno = ENOTCONN;
+        goto out;
+    }
+
+    if (__is_root_gfid(loc->inode->gfid)) {
+        if (!priv->root_inode)
+            priv->root_inode = inode_ref(loc->inode);
+
+        if (priv->choose_local && !priv->did_discovery) {
+            /* Logic to detect which subvolumes of AFR are
+               local, in order to prefer them for reads
+            */
+            local->do_discovery = _gf_true;
+            priv->did_discovery = _gf_true;
+        }
+    }
 
-        list_for_each_entry_safe (locked_fd, tmp, &priv->saved_fds,
-                                  list) {
+    local->op = GF_FOP_LOOKUP;
 
-                if (locked_fd->fd == fd) {
-                        list_del_init (&locked_fd->list);
-                        GF_FREE (locked_fd);
-                }
+    loc_copy(&local->loc, loc);
 
+    local->inode = inode_ref(loc->inode);
+
+    if (xattr_req) {
+        /* If xattr_req was null, afr_lookup_xattr_req_prepare() will
+           allocate one for us */
+        local->xattr_req = dict_copy_with_ref(xattr_req, NULL);
+        if (!local->xattr_req) {
+            op_errno = ENOMEM;
+            goto out;
         }
+    }
 
+    if (gf_uuid_is_null(loc->inode->gfid)) {
+        afr_discover_do(frame, this, 0);
         return 0;
-}
+    }
+
+    afr_read_subvol_get(loc->inode, this, NULL, NULL, &event,
+                        AFR_DATA_TRANSACTION, NULL);
 
+    afr_discover_do(frame, this, 0);
 
-/* {{{ fsync */
+    return 0;
+out:
+    AFR_STACK_UNWIND(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+    return 0;
+}
 
 int
-afr_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-               int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-               struct iatt *postbuf)
+afr_lookup_do(call_frame_t *frame, xlator_t *this, int err)
 {
-        afr_local_t *local = NULL;
-
-        int call_count = -1;
+    int ret = 0;
+    int i = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int call_count = 0;
+
+    local = frame->local;
+    priv = this->private;
+
+    if (err < 0) {
+        local->op_errno = err;
+        goto out;
+    }
+
+    call_count = local->call_count = AFR_COUNT(local->child_up,
+                                               priv->child_count);
+
+    ret = afr_lookup_xattr_req_prepare(local, this, local->xattr_req,
+                                       &local->loc);
+    if (ret) {
+        local->op_errno = -ret;
+        goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->child_up[i]) {
+            STACK_WIND_COOKIE(
+                frame, afr_lookup_cbk, (void *)(long)i, priv->children[i],
+                priv->children[i]->fops->lookup, &local->loc, local->xattr_req);
+            if (!--call_count)
+                break;
+        }
+    }
+    return 0;
+out:
+    AFR_STACK_UNWIND(lookup, frame, -1, local->op_errno, 0, 0, 0, 0);
+    return 0;
+}
 
-        int child_index = (long) cookie;
-        int read_child  = 0;
+/*
+ * afr_lookup()
+ *
+ * The goal here is to figure out what the element getting looked up is.
+ * i.e what is the GFID, inode type and a conservative estimate of the
+ * inode attributes are.
+ *
+ * As we lookup, operations may be underway on the entry name and the
+ * inode. In lookup() we are primarily concerned only with the entry
+ * operations. If the entry is getting unlinked or renamed, we detect
+ * what operation is underway by querying for on-going transactions and
+ * pending self-healing on the entry through xdata.
+ *
+ * If the entry is a file/dir, it may need self-heal and/or in a
+ * split-brain condition. Lookup is not the place to worry about these
+ * conditions. Outcast marking will naturally handle them in the read
+ * paths.
+ *
+ * Here is a brief goal of what we are trying to achieve:
+ *
+ * - LOOKUP on all subvolumes concurrently, querying on-going transaction
+ *   and pending self-heal info from the servers.
+ *
+ * - If all servers reply the same inode type and GFID, the overall call
+ *   MUST be a success.
+ *
+ * - If inode types or GFIDs mismatch, and there IS either an on-going
+ *   transaction or pending self-heal, inspect what the nature of the
+ *   transaction or pending heal is, and select the appropriate subvolume's
+ *   reply as the winner.
+ *
+ * - If inode types or GFIDs mismatch, and there are no on-going transactions
+ *   or pending self-heal on the entry name on any of the servers, fail the
+ *   lookup with EIO. Something has gone wrong beyond reasonable action.
+ */
 
-        local = frame->local;
+int
+afr_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
+{
+    afr_local_t *local = NULL;
+    int32_t op_errno = 0;
+    int event = 0;
+    int ret = 0;
+
+    if (loc_is_nameless(loc)) {
+        if (xattr_req)
+            dict_del_sizen(xattr_req, "gfid-req");
+        afr_discover(frame, this, loc, xattr_req);
+        return 0;
+    }
 
-        read_child = afr_read_child (this, local->fd->inode);
+    if (afr_is_private_directory(this->private, loc->parent->gfid, loc->name,
+                                 frame->root->pid)) {
+        op_errno = EPERM;
+        goto out;
+    }
 
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-                if (op_ret == 0) {
-                        local->op_ret = 0;
+    if (!local->call_count) {
+        op_errno = ENOTCONN;
+        goto out;
+    }
 
-                        if (local->success_count == 0) {
-                                local->cont.fsync.prebuf  = *prebuf;
-                                local->cont.fsync.postbuf = *postbuf;
-                        }
+    local->op = GF_FOP_LOOKUP;
 
-                        if (child_index == read_child) {
-                                local->cont.fsync.prebuf  = *prebuf;
-                                local->cont.fsync.postbuf = *postbuf;
-                        }
+    loc_copy(&local->loc, loc);
 
-                        local->success_count++;
-                }
+    local->inode = inode_ref(loc->inode);
 
-                local->op_errno = op_errno;
+    if (xattr_req) {
+        /* If xattr_req was null, afr_lookup_xattr_req_prepare() will
+           allocate one for us */
+        local->xattr_req = dict_copy_with_ref(xattr_req, NULL);
+        if (!local->xattr_req) {
+            op_errno = ENOMEM;
+            goto out;
+        }
+        ret = dict_get_gfuuid(local->xattr_req, "gfid-req",
+                              &local->cont.lookup.gfid_req);
+        if (ret == 0) {
+            dict_del_sizen(local->xattr_req, "gfid-req");
         }
-        UNLOCK (&frame->lock);
+    }
 
-        call_count = afr_frame_return (frame);
+    afr_read_subvol_get(loc->parent, this, NULL, NULL, &event,
+                        AFR_DATA_TRANSACTION, NULL);
 
-        if (call_count == 0) {
-                local->cont.fsync.prebuf.ia_ino  = local->cont.fsync.ino;
-                local->cont.fsync.postbuf.ia_ino = local->cont.fsync.ino;
+    afr_lookup_do(frame, this, 0);
 
-                AFR_STACK_UNWIND (fsync, frame, local->op_ret, local->op_errno,
-                                  &local->cont.fsync.prebuf,
-                                  &local->cont.fsync.postbuf);
-        }
+    return 0;
+out:
+    AFR_STACK_UNWIND(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
 
-        return 0;
+    return 0;
 }
 
+void
+_afr_cleanup_fd_ctx(xlator_t *this, afr_fd_ctx_t *fd_ctx)
+{
+    afr_private_t *priv = this->private;
+
+    if (fd_ctx->lk_heal_info) {
+        LOCK(&priv->lock);
+        {
+            list_del(&fd_ctx->lk_heal_info->pos);
+        }
+        afr_lk_heal_info_cleanup(fd_ctx->lk_heal_info);
+        fd_ctx->lk_heal_info = NULL;
+    }
+    GF_FREE(fd_ctx->opened_on);
+    GF_FREE(fd_ctx);
+    return;
+}
 
 int
-afr_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd,
-           int32_t datasync)
+afr_cleanup_fd_ctx(xlator_t *this, fd_t *fd)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local = NULL;
+    uint64_t ctx = 0;
+    afr_fd_ctx_t *fd_ctx = NULL;
+    int ret = 0;
 
-        int ret = -1;
+    ret = fd_ctx_get(fd, this, &ctx);
+    if (ret < 0)
+        goto out;
 
-        int i = 0;
-        int32_t call_count = 0;
-        int32_t op_ret   = -1;
-        int32_t op_errno = 0;
+    fd_ctx = (afr_fd_ctx_t *)(long)ctx;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    if (fd_ctx) {
+        _afr_cleanup_fd_ctx(this, fd_ctx);
+    }
 
-        priv = this->private;
+out:
+    return 0;
+}
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+int
+afr_release(xlator_t *this, fd_t *fd)
+{
+    afr_cleanup_fd_ctx(this, fd);
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+    return 0;
+}
 
-        call_count = local->call_count;
-        frame->local = local;
+afr_fd_ctx_t *
+__afr_fd_ctx_get(fd_t *fd, xlator_t *this)
+{
+    uint64_t ctx = 0;
+    int ret = 0;
+    afr_fd_ctx_t *fd_ctx = NULL;
 
-        local->fd             = fd_ref (fd);
-        local->cont.fsync.ino = fd->inode->ino;
+    ret = __fd_ctx_get(fd, this, &ctx);
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND_COOKIE (frame, afr_fsync_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->fsync,
-                                           fd, datasync);
-                        if (!--call_count)
-                                break;
-                }
-        }
+    if (ret < 0) {
+        ret = __afr_fd_ctx_set(this, fd);
+        if (ret < 0)
+            goto out;
 
-        op_ret = 0;
+        ret = __fd_ctx_get(fd, this, &ctx);
+        if (ret < 0)
+            goto out;
+    }
+
+    fd_ctx = (afr_fd_ctx_t *)(long)ctx;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (fsync, frame, op_ret, op_errno, NULL, NULL);
-        }
-        return 0;
+    return fd_ctx;
 }
 
-/* }}} */
+afr_fd_ctx_t *
+afr_fd_ctx_get(fd_t *fd, xlator_t *this)
+{
+    afr_fd_ctx_t *fd_ctx = NULL;
 
-/* {{{ fsync */
+    LOCK(&fd->lock);
+    {
+        fd_ctx = __afr_fd_ctx_get(fd, this);
+    }
+    UNLOCK(&fd->lock);
 
-int32_t
-afr_fsyncdir_cbk (call_frame_t *frame, void *cookie,
-                  xlator_t *this, int32_t op_ret, int32_t op_errno)
+    return fd_ctx;
+}
+
+int
+__afr_fd_ctx_set(xlator_t *this, fd_t *fd)
 {
-        afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int ret = -1;
+    uint64_t ctx = 0;
+    afr_fd_ctx_t *fd_ctx = NULL;
+    int i = 0;
+
+    VALIDATE_OR_GOTO(this->private, out);
+    VALIDATE_OR_GOTO(fd, out);
+
+    priv = this->private;
+
+    ret = __fd_ctx_get(fd, this, &ctx);
+
+    if (ret == 0)
+        goto out;
+
+    fd_ctx = GF_CALLOC(1, sizeof(afr_fd_ctx_t), gf_afr_mt_afr_fd_ctx_t);
+    if (!fd_ctx) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    fd_ctx->opened_on = GF_CALLOC(sizeof(*fd_ctx->opened_on), priv->child_count,
+                                  gf_afr_mt_int32_t);
+    if (!fd_ctx->opened_on) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (fd_is_anonymous(fd))
+            fd_ctx->opened_on[i] = AFR_FD_OPENED;
+        else
+            fd_ctx->opened_on[i] = AFR_FD_NOT_OPENED;
+    }
 
-        int call_count = -1;
+    fd_ctx->readdir_subvol = -1;
+    fd_ctx->lk_heal_info = NULL;
 
-        local = frame->local;
+    ret = __fd_ctx_set(fd, this, (uint64_t)(long)fd_ctx);
+    if (ret)
+        gf_msg_debug(this->name, 0, "failed to set fd ctx (%p)", fd);
+out:
+    if (ret && fd_ctx)
+        _afr_cleanup_fd_ctx(this, fd_ctx);
+    return ret;
+}
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == 0)
-                        local->op_ret = 0;
+/* {{{ flush */
+
+int
+afr_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int call_count = -1;
+
+    local = frame->local;
 
-                local->op_errno = op_errno;
+    LOCK(&frame->lock);
+    {
+        if (op_ret != -1) {
+            local->op_ret = op_ret;
+            if (!local->xdata_rsp && xdata)
+                local->xdata_rsp = dict_ref(xdata);
+        } else {
+            local->op_errno = op_errno;
         }
-        UNLOCK (&frame->lock);
+        call_count = --local->call_count;
+    }
+    UNLOCK(&frame->lock);
 
-        call_count = afr_frame_return (frame);
+    if (call_count == 0)
+        AFR_STACK_UNWIND(flush, frame, local->op_ret, local->op_errno,
+                         local->xdata_rsp);
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (fsyncdir, frame, local->op_ret,
-                                  local->op_errno);
+    return 0;
+}
 
-        return 0;
+static int
+afr_flush_wrapper(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    int i = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int call_count = -1;
+
+    priv = this->private;
+    local = frame->local;
+    call_count = local->call_count;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->child_up[i]) {
+            STACK_WIND_COOKIE(frame, afr_flush_cbk, (void *)(long)i,
+                              priv->children[i], priv->children[i]->fops->flush,
+                              local->fd, xdata);
+            if (!--call_count)
+                break;
+        }
+    }
+
+    return 0;
 }
 
+afr_local_t *
+afr_wakeup_same_fd_delayed_op(xlator_t *this, afr_lock_t *lock, fd_t *fd)
+{
+    afr_local_t *local = NULL;
+
+    if (lock->delay_timer) {
+        local = list_entry(lock->post_op.next, afr_local_t,
+                           transaction.owner_list);
+        if (fd == local->fd) {
+            if (gf_timer_call_cancel(this->ctx, lock->delay_timer)) {
+                local = NULL;
+            } else {
+                lock->delay_timer = NULL;
+            }
+        } else {
+            local = NULL;
+        }
+    }
 
-int32_t
-afr_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd,
-              int32_t datasync)
+    return local;
+}
+
+void
+afr_delayed_changelog_wake_resume(xlator_t *this, inode_t *inode,
+                                  call_stub_t *stub)
+{
+    afr_inode_ctx_t *ctx = NULL;
+    afr_lock_t *lock = NULL;
+    afr_local_t *metadata_local = NULL;
+    afr_local_t *data_local = NULL;
+    LOCK(&inode->lock);
+    {
+        (void)__afr_inode_ctx_get(this, inode, &ctx);
+        lock = &ctx->lock[AFR_DATA_TRANSACTION];
+        data_local = afr_wakeup_same_fd_delayed_op(this, lock, stub->args.fd);
+        lock = &ctx->lock[AFR_METADATA_TRANSACTION];
+        metadata_local = afr_wakeup_same_fd_delayed_op(this, lock,
+                                                       stub->args.fd);
+    }
+    UNLOCK(&inode->lock);
+
+    if (data_local) {
+        data_local->transaction.resume_stub = stub;
+    } else if (metadata_local) {
+        metadata_local->transaction.resume_stub = stub;
+    } else {
+        call_resume(stub);
+    }
+    if (data_local) {
+        afr_delayed_changelog_wake_up_cbk(data_local);
+    }
+    if (metadata_local) {
+        afr_delayed_changelog_wake_up_cbk(metadata_local);
+    }
+}
+
+int
+afr_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local = NULL;
+    afr_local_t *local = NULL;
+    call_stub_t *stub = NULL;
+    int op_errno = ENOMEM;
 
-        int ret = -1;
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-        int i = 0;
-        int32_t call_count = 0;
-        int32_t op_ret   = -1;
-        int32_t op_errno = 0;
+    local->op = GF_FOP_FLUSH;
+    if (!afr_is_consistent_io_possible(local, this->private, &op_errno))
+        goto out;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    local->fd = fd_ref(fd);
 
-        priv = this->private;
+    stub = fop_flush_stub(frame, afr_flush_wrapper, fd, xdata);
+    if (!stub)
+        goto out;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+    afr_delayed_changelog_wake_resume(this, fd->inode, stub);
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
+    return 0;
+out:
+    AFR_STACK_UNWIND(flush, frame, -1, op_errno, NULL);
+    return 0;
+}
+
+int
+afr_fsyncdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int call_count = -1;
+
+    local = frame->local;
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == 0) {
+            local->op_ret = 0;
+            if (!local->xdata_rsp && xdata)
+                local->xdata_rsp = dict_ref(xdata);
+        } else {
+            local->op_errno = op_errno;
         }
+        call_count = --local->call_count;
+    }
+    UNLOCK(&frame->lock);
 
-        call_count = local->call_count;
-        frame->local = local;
+    if (call_count == 0)
+        AFR_STACK_UNWIND(fsyncdir, frame, local->op_ret, local->op_errno,
+                         local->xdata_rsp);
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND (frame, afr_fsyncdir_cbk,
-                                    priv->children[i],
-                                    priv->children[i]->fops->fsyncdir,
-                                    fd, datasync);
-                        if (!--call_count)
-                                break;
-                }
+    return 0;
+}
+
+int
+afr_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+             dict_t *xdata)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i = 0;
+    int32_t call_count = 0;
+    int32_t op_errno = ENOMEM;
+
+    priv = this->private;
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    local->op = GF_FOP_FSYNCDIR;
+    if (!afr_is_consistent_io_possible(local, priv, &op_errno))
+        goto out;
+
+    call_count = local->call_count;
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->child_up[i]) {
+            STACK_WIND(frame, afr_fsyncdir_cbk, priv->children[i],
+                       priv->children[i]->fops->fsyncdir, fd, datasync, xdata);
+            if (!--call_count)
+                break;
         }
+    }
 
-        op_ret = 0;
+    return 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (fsyncdir, frame, op_ret, op_errno);
-        }
-        return 0;
+    AFR_STACK_UNWIND(fsyncdir, frame, -1, op_errno, NULL);
+
+    return 0;
 }
 
 /* }}} */
 
-/* {{{ xattrop */
+static int
+afr_serialized_lock_wind(call_frame_t *frame, xlator_t *this);
 
-int32_t
-afr_xattrop_cbk (call_frame_t *frame, void *cookie,
-                 xlator_t *this, int32_t op_ret, int32_t op_errno,
-                 dict_t *xattr)
+static gf_boolean_t
+afr_is_conflicting_lock_present(int32_t op_ret, int32_t op_errno)
 {
-        afr_local_t *local = NULL;
+    if (op_ret == -1 && op_errno == EAGAIN)
+        return _gf_true;
+    return _gf_false;
+}
+
+static void
+afr_fop_lock_unwind(call_frame_t *frame, glusterfs_fop_t op, int32_t op_ret,
+                    int32_t op_errno, dict_t *xdata)
+{
+    switch (op) {
+        case GF_FOP_INODELK:
+            AFR_STACK_UNWIND(inodelk, frame, op_ret, op_errno, xdata);
+            break;
+        case GF_FOP_FINODELK:
+            AFR_STACK_UNWIND(finodelk, frame, op_ret, op_errno, xdata);
+            break;
+        case GF_FOP_ENTRYLK:
+            AFR_STACK_UNWIND(entrylk, frame, op_ret, op_errno, xdata);
+            break;
+        case GF_FOP_FENTRYLK:
+            AFR_STACK_UNWIND(fentrylk, frame, op_ret, op_errno, xdata);
+            break;
+        default:
+            break;
+    }
+}
 
-        int call_count = -1;
+static void
+afr_fop_lock_wind(call_frame_t *frame, xlator_t *this, int child_index,
+                  int32_t (*lock_cbk)(call_frame_t *, void *, xlator_t *,
+                                      int32_t, int32_t, dict_t *))
+{
+    afr_local_t *local = frame->local;
+    afr_private_t *priv = this->private;
+    int i = child_index;
+
+    switch (local->op) {
+        case GF_FOP_INODELK:
+            STACK_WIND_COOKIE(
+                frame, lock_cbk, (void *)(long)i, priv->children[i],
+                priv->children[i]->fops->inodelk,
+                (const char *)local->cont.inodelk.volume, &local->loc,
+                local->cont.inodelk.cmd, &local->cont.inodelk.flock,
+                local->cont.inodelk.xdata);
+            break;
+        case GF_FOP_FINODELK:
+            STACK_WIND_COOKIE(
+                frame, lock_cbk, (void *)(long)i, priv->children[i],
+                priv->children[i]->fops->finodelk,
+                (const char *)local->cont.inodelk.volume, local->fd,
+                local->cont.inodelk.cmd, &local->cont.inodelk.flock,
+                local->cont.inodelk.xdata);
+            break;
+        case GF_FOP_ENTRYLK:
+            STACK_WIND_COOKIE(
+                frame, lock_cbk, (void *)(long)i, priv->children[i],
+                priv->children[i]->fops->entrylk, local->cont.entrylk.volume,
+                &local->loc, local->cont.entrylk.basename,
+                local->cont.entrylk.cmd, local->cont.entrylk.type,
+                local->cont.entrylk.xdata);
+            break;
+        case GF_FOP_FENTRYLK:
+            STACK_WIND_COOKIE(
+                frame, lock_cbk, (void *)(long)i, priv->children[i],
+                priv->children[i]->fops->fentrylk, local->cont.entrylk.volume,
+                local->fd, local->cont.entrylk.basename,
+                local->cont.entrylk.cmd, local->cont.entrylk.type,
+                local->cont.entrylk.xdata);
+            break;
+        default:
+            break;
+    }
+}
 
-        local = frame->local;
+void
+afr_fop_lock_proceed(call_frame_t *frame)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == 0)
-                        local->op_ret = 0;
+    local = frame->local;
+    priv = frame->this->private;
 
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    if (local->fop_lock_state != AFR_FOP_LOCK_PARALLEL) {
+        afr_fop_lock_unwind(frame, local->op, local->op_ret, local->op_errno,
+                            local->xdata_rsp);
+        return;
+    }
+    /* At least one child is up */
+    /*
+     * Non-blocking locks also need to be serialized.  Otherwise there is
+     * a chance that both the mounts which issued same non-blocking inodelk
+     * may endup not acquiring the lock on any-brick.
+     * Ex: Mount1 and Mount2
+     * request for full length lock on file f1.  Mount1 afr may acquire the
+     * partial lock on brick-1 and may not acquire the lock on brick-2
+     * because Mount2 already got the lock on brick-2, vice versa.  Since
+     * both the mounts only got partial locks, afr treats them as failure in
+     * gaining the locks and unwinds with EAGAIN errno.
+     */
+    local->op_ret = -1;
+    local->op_errno = EUCLEAN;
+    local->fop_lock_state = AFR_FOP_LOCK_SERIAL;
+    afr_local_replies_wipe(local, priv);
+    if (local->xdata_rsp)
+        dict_unref(local->xdata_rsp);
+    local->xdata_rsp = NULL;
+    switch (local->op) {
+        case GF_FOP_INODELK:
+        case GF_FOP_FINODELK:
+            local->cont.inodelk.cmd = local->cont.inodelk.in_cmd;
+            local->cont.inodelk.flock = local->cont.inodelk.in_flock;
+            if (local->cont.inodelk.xdata)
+                dict_unref(local->cont.inodelk.xdata);
+            local->cont.inodelk.xdata = NULL;
+            if (local->xdata_req)
+                local->cont.inodelk.xdata = dict_ref(local->xdata_req);
+            break;
+        case GF_FOP_ENTRYLK:
+        case GF_FOP_FENTRYLK:
+            local->cont.entrylk.cmd = local->cont.entrylk.in_cmd;
+            if (local->cont.entrylk.xdata)
+                dict_unref(local->cont.entrylk.xdata);
+            local->cont.entrylk.xdata = NULL;
+            if (local->xdata_req)
+                local->cont.entrylk.xdata = dict_ref(local->xdata_req);
+            break;
+        default:
+            break;
+    }
+    afr_serialized_lock_wind(frame, frame->this);
+}
 
-        call_count = afr_frame_return (frame);
+static int32_t
+afr_unlock_partial_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int32_t op_ret, int32_t op_errno, dict_t *xdata)
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (xattrop, frame, local->op_ret, local->op_errno,
-                                  xattr);
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int call_count = -1;
+    int child_index = (long)cookie;
+    uuid_t gfid = {0};
 
-        return 0;
+    local = frame->local;
+    priv = this->private;
+
+    if (op_ret < 0 && op_errno != ENOTCONN) {
+        if (local->fd)
+            gf_uuid_copy(gfid, local->fd->inode->gfid);
+        else
+            loc_gfid(&local->loc, gfid);
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, AFR_MSG_UNLOCK_FAIL,
+               "%s: Failed to unlock %s on %s "
+               "with lk_owner: %s",
+               uuid_utoa(gfid), gf_fop_list[local->op],
+               priv->children[child_index]->name,
+               lkowner_utoa(&frame->root->lk_owner));
+    }
+
+    call_count = afr_frame_return(frame);
+    if (call_count == 0)
+        afr_fop_lock_proceed(frame);
+
+    return 0;
 }
 
+static int32_t
+afr_unlock_locks_and_proceed(call_frame_t *frame, xlator_t *this,
+                             int call_count)
+{
+    int i = 0;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+
+    if (call_count == 0) {
+        afr_fop_lock_proceed(frame);
+        goto out;
+    }
+
+    local = frame->local;
+    priv = this->private;
+    local->call_count = call_count;
+    switch (local->op) {
+        case GF_FOP_INODELK:
+        case GF_FOP_FINODELK:
+            local->cont.inodelk.flock.l_type = F_UNLCK;
+            local->cont.inodelk.cmd = F_SETLK;
+            if (local->cont.inodelk.xdata)
+                dict_unref(local->cont.inodelk.xdata);
+            local->cont.inodelk.xdata = NULL;
+            break;
+        case GF_FOP_ENTRYLK:
+        case GF_FOP_FENTRYLK:
+            local->cont.entrylk.cmd = ENTRYLK_UNLOCK;
+            if (local->cont.entrylk.xdata)
+                dict_unref(local->cont.entrylk.xdata);
+            local->cont.entrylk.xdata = NULL;
+            break;
+        default:
+            break;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
+
+        if (local->replies[i].op_ret == -1)
+            continue;
+
+        afr_fop_lock_wind(frame, this, i, afr_unlock_partial_lock_cbk);
+
+        if (!--call_count)
+            break;
+    }
+
+out:
+    return 0;
+}
 
 int32_t
-afr_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc,
-             gf_xattrop_flags_t optype, dict_t *xattr)
+afr_fop_lock_done(call_frame_t *frame, xlator_t *this)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local  = NULL;
+    int i = 0;
+    int lock_count = 0;
+    unsigned char *success = NULL;
 
-        int ret = -1;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        int i = 0;
-        int32_t call_count = 0;
-        int32_t op_ret   = -1;
-        int32_t op_errno = 0;
+    local = frame->local;
+    priv = this->private;
+    success = alloca0(priv->child_count);
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
 
-        priv = this->private;
+        if (local->replies[i].op_ret == 0) {
+            lock_count++;
+            success[i] = 1;
+        }
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+        if (local->op_ret == -1 && local->op_errno == EAGAIN)
+            continue;
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
+        if ((local->replies[i].op_ret == -1) &&
+            (local->replies[i].op_errno == EAGAIN)) {
+            local->op_ret = -1;
+            local->op_errno = EAGAIN;
+            continue;
         }
 
-        call_count = local->call_count;
-        frame->local = local;
+        if (local->replies[i].op_ret == 0)
+            local->op_ret = 0;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND (frame, afr_xattrop_cbk,
-                                    priv->children[i],
-                                    priv->children[i]->fops->xattrop,
-                                    loc, optype, xattr);
-                        if (!--call_count)
-                                break;
-                }
+        local->op_errno = local->replies[i].op_errno;
+    }
+
+    if (afr_fop_lock_is_unlock(frame))
+        goto unwind;
+
+    if (afr_is_conflicting_lock_present(local->op_ret, local->op_errno)) {
+        afr_unlock_locks_and_proceed(frame, this, lock_count);
+    } else if (priv->quorum_count && !afr_has_quorum(success, this, NULL)) {
+        local->fop_lock_state = AFR_FOP_LOCK_QUORUM_FAILED;
+        local->op_ret = -1;
+        local->op_errno = afr_final_errno(local, priv);
+        if (local->op_errno == 0)
+            local->op_errno = afr_quorum_errno(priv);
+        afr_unlock_locks_and_proceed(frame, this, lock_count);
+    } else {
+        goto unwind;
+    }
+
+    return 0;
+unwind:
+    afr_fop_lock_unwind(frame, local->op, local->op_ret, local->op_errno,
+                        local->xdata_rsp);
+    return 0;
+}
+
+static int
+afr_common_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int child_index = (long)cookie;
+
+    local = frame->local;
+
+    local->replies[child_index].valid = 1;
+    local->replies[child_index].op_ret = op_ret;
+    local->replies[child_index].op_errno = op_errno;
+    if (op_ret == 0 && xdata) {
+        local->replies[child_index].xdata = dict_ref(xdata);
+        LOCK(&frame->lock);
+        {
+            if (!local->xdata_rsp)
+                local->xdata_rsp = dict_ref(xdata);
         }
+        UNLOCK(&frame->lock);
+    }
+    return 0;
+}
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (xattrop, frame, op_ret, op_errno, NULL);
+static int32_t
+afr_serialized_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int child_index = (long)cookie;
+    int next_child = 0;
+
+    local = frame->local;
+    priv = this->private;
+
+    afr_common_lock_cbk(frame, cookie, this, op_ret, op_errno, xdata);
+
+    for (next_child = child_index + 1; next_child < priv->child_count;
+         next_child++) {
+        if (local->child_up[next_child])
+            break;
+    }
+
+    if (afr_is_conflicting_lock_present(op_ret, op_errno) ||
+        (next_child == priv->child_count)) {
+        afr_fop_lock_done(frame, this);
+    } else {
+        afr_fop_lock_wind(frame, this, next_child, afr_serialized_lock_cbk);
+    }
+
+    return 0;
+}
+
+static int
+afr_serialized_lock_wind(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i = 0;
+
+    priv = this->private;
+    local = frame->local;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->child_up[i]) {
+            afr_fop_lock_wind(frame, this, i, afr_serialized_lock_cbk);
+            break;
         }
-        return 0;
+    }
+    return 0;
 }
 
-/* }}} */
+static int32_t
+afr_parallel_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
 
-/* {{{ fxattrop */
+{
+    int call_count = 0;
 
-int32_t
-afr_fxattrop_cbk (call_frame_t *frame, void *cookie,
-                  xlator_t *this, int32_t op_ret, int32_t op_errno,
-                  dict_t *xattr)
+    afr_common_lock_cbk(frame, cookie, this, op_ret, op_errno, xdata);
+
+    call_count = afr_frame_return(frame);
+    if (call_count == 0)
+        afr_fop_lock_done(frame, this);
+
+    return 0;
+}
+
+static int
+afr_parallel_lock_wind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int call_count = 0;
+    int i = 0;
+
+    priv = this->private;
+    local = frame->local;
+    call_count = local->call_count;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->child_up[i])
+            continue;
+        afr_fop_lock_wind(frame, this, i, afr_parallel_lock_cbk);
+        if (!--call_count)
+            break;
+    }
+    return 0;
+}
 
-        int call_count = -1;
+static int
+afr_fop_handle_lock(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = frame->local;
+    int op_errno = 0;
 
-        local = frame->local;
+    if (!afr_fop_lock_is_unlock(frame)) {
+        if (!afr_is_consistent_io_possible(local, this->private, &op_errno))
+            goto out;
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == 0)
-                        local->op_ret = 0;
+        switch (local->op) {
+            case GF_FOP_INODELK:
+            case GF_FOP_FINODELK:
+                local->cont.inodelk.cmd = F_SETLK;
+                break;
+            case GF_FOP_ENTRYLK:
+            case GF_FOP_FENTRYLK:
+                local->cont.entrylk.cmd = ENTRYLK_LOCK_NB;
+                break;
+            default:
+                break;
+        }
+    }
 
-                local->op_errno = op_errno;
+    if (local->xdata_req) {
+        switch (local->op) {
+            case GF_FOP_INODELK:
+            case GF_FOP_FINODELK:
+                local->cont.inodelk.xdata = dict_ref(local->xdata_req);
+                break;
+            case GF_FOP_ENTRYLK:
+            case GF_FOP_FENTRYLK:
+                local->cont.entrylk.xdata = dict_ref(local->xdata_req);
+                break;
+            default:
+                break;
         }
-        UNLOCK (&frame->lock);
+    }
 
-        call_count = afr_frame_return (frame);
+    local->fop_lock_state = AFR_FOP_LOCK_PARALLEL;
+    afr_parallel_lock_wind(frame, this);
+out:
+    return -op_errno;
+}
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (fxattrop, frame, local->op_ret, local->op_errno,
-                                  xattr);
+static int32_t
+afr_handle_inodelk(call_frame_t *frame, xlator_t *this, glusterfs_fop_t fop,
+                   const char *volume, loc_t *loc, fd_t *fd, int32_t cmd,
+                   struct gf_flock *flock, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int32_t op_errno = ENOMEM;
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    local->op = fop;
+    if (loc)
+        loc_copy(&local->loc, loc);
+    if (fd && (flock->l_type != F_UNLCK)) {
+        AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+        local->fd = fd_ref(fd);
+    }
+
+    local->cont.inodelk.volume = gf_strdup(volume);
+    if (!local->cont.inodelk.volume) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    local->cont.inodelk.in_cmd = cmd;
+    local->cont.inodelk.cmd = cmd;
+    local->cont.inodelk.in_flock = *flock;
+    local->cont.inodelk.flock = *flock;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
+
+    op_errno = -afr_fop_handle_lock(frame, frame->this);
+    if (op_errno)
+        goto out;
+    return 0;
+out:
+    afr_fop_lock_unwind(frame, fop, -1, op_errno, NULL);
 
-        return 0;
+    return 0;
 }
 
+int32_t
+afr_inodelk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+            int32_t cmd, struct gf_flock *flock, dict_t *xdata)
+{
+    afr_handle_inodelk(frame, this, GF_FOP_INODELK, volume, loc, NULL, cmd,
+                       flock, xdata);
+    return 0;
+}
 
 int32_t
-afr_fxattrop (call_frame_t *frame, xlator_t *this, fd_t *fd,
-              gf_xattrop_flags_t optype, dict_t *xattr)
+afr_finodelk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+             int32_t cmd, struct gf_flock *flock, dict_t *xdata)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local  = NULL;
+    afr_handle_inodelk(frame, this, GF_FOP_FINODELK, volume, NULL, fd, cmd,
+                       flock, xdata);
+    return 0;
+}
 
-        int ret = -1;
+static int
+afr_handle_entrylk(call_frame_t *frame, xlator_t *this, glusterfs_fop_t fop,
+                   const char *volume, loc_t *loc, fd_t *fd,
+                   const char *basename, entrylk_cmd cmd, entrylk_type type,
+                   dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int32_t op_errno = ENOMEM;
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    local->op = fop;
+    if (loc)
+        loc_copy(&local->loc, loc);
+    if (fd && (cmd != ENTRYLK_UNLOCK)) {
+        AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+        local->fd = fd_ref(fd);
+    }
+    local->cont.entrylk.cmd = cmd;
+    local->cont.entrylk.in_cmd = cmd;
+    local->cont.entrylk.type = type;
+    local->cont.entrylk.volume = gf_strdup(volume);
+    local->cont.entrylk.basename = gf_strdup(basename);
+    if (!local->cont.entrylk.volume || !local->cont.entrylk.basename) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
+    op_errno = -afr_fop_handle_lock(frame, frame->this);
+    if (op_errno)
+        goto out;
+
+    return 0;
+out:
+    afr_fop_lock_unwind(frame, fop, -1, op_errno, NULL);
+    return 0;
+}
 
-        int i = 0;
-        int32_t call_count = 0;
-        int32_t op_ret   = -1;
-        int32_t op_errno = 0;
+int
+afr_entrylk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+            const char *basename, entrylk_cmd cmd, entrylk_type type,
+            dict_t *xdata)
+{
+    afr_handle_entrylk(frame, this, GF_FOP_ENTRYLK, volume, loc, NULL, basename,
+                       cmd, type, xdata);
+    return 0;
+}
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+int
+afr_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+             const char *basename, entrylk_cmd cmd, entrylk_type type,
+             dict_t *xdata)
+{
+    afr_handle_entrylk(frame, this, GF_FOP_FENTRYLK, volume, NULL, fd, basename,
+                       cmd, type, xdata);
+    return 0;
+}
 
-        priv = this->private;
+int
+afr_statfs_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, struct statvfs *statvfs, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int call_count = 0;
+    struct statvfs *buf = NULL;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+    local = frame->local;
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
+    LOCK(&frame->lock);
+    {
+        if (op_ret != 0) {
+            local->op_errno = op_errno;
+            goto unlock;
         }
 
-        call_count = local->call_count;
-        frame->local = local;
+        local->op_ret = op_ret;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND (frame, afr_fxattrop_cbk,
-                                    priv->children[i],
-                                    priv->children[i]->fops->fxattrop,
-                                    fd, optype, xattr);
-                        if (!--call_count)
-                                break;
+        buf = &local->cont.statfs.buf;
+        if (local->cont.statfs.buf_set) {
+            if (statvfs->f_bavail < buf->f_bavail) {
+                *buf = *statvfs;
+                if (xdata) {
+                    if (local->xdata_rsp)
+                        dict_unref(local->xdata_rsp);
+                    local->xdata_rsp = dict_ref(xdata);
                 }
+            }
+        } else {
+            *buf = *statvfs;
+            local->cont.statfs.buf_set = 1;
+            if (xdata)
+                local->xdata_rsp = dict_ref(xdata);
         }
+    }
+unlock:
+    call_count = --local->call_count;
+    UNLOCK(&frame->lock);
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (fxattrop, frame, op_ret, op_errno, NULL);
-        }
-        return 0;
+    if (call_count == 0)
+        AFR_STACK_UNWIND(statfs, frame, local->op_ret, local->op_errno,
+                         &local->cont.statfs.buf, local->xdata_rsp);
+
+    return 0;
 }
 
-/* }}} */
+int
+afr_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int call_count = 0;
+    int32_t op_errno = ENOMEM;
+
+    priv = this->private;
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    local->op = GF_FOP_STATFS;
+    if (!afr_is_consistent_io_possible(local, priv, &op_errno))
+        goto out;
+
+    if (priv->arbiter_count == 1 && local->child_up[ARBITER_BRICK_INDEX])
+        local->call_count--;
+    call_count = local->call_count;
+    if (!call_count) {
+        op_errno = ENOTCONN;
+        goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->child_up[i]) {
+            if (AFR_IS_ARBITER_BRICK(priv, i))
+                continue;
+            STACK_WIND(frame, afr_statfs_cbk, priv->children[i],
+                       priv->children[i]->fops->statfs, loc, xdata);
+            if (!--call_count)
+                break;
+        }
+    }
 
+    return 0;
+out:
+    AFR_STACK_UNWIND(statfs, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
 
 int32_t
-afr_inodelk_cbk (call_frame_t *frame, void *cookie,
-                 xlator_t *this, int32_t op_ret, int32_t op_errno)
+afr_lk_unlock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct gf_flock *lock,
+                  dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = this->private;
+    int call_count = -1;
+    int child_index = (long)cookie;
+
+    local = frame->local;
+
+    if (op_ret < 0 && op_errno != ENOTCONN && op_errno != EBADFD) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, AFR_MSG_UNLOCK_FAIL,
+               "gfid=%s: unlock failed on subvolume %s "
+               "with lock owner %s",
+               uuid_utoa(local->fd->inode->gfid),
+               priv->children[child_index]->name,
+               lkowner_utoa(&frame->root->lk_owner));
+    }
+
+    call_count = afr_frame_return(frame);
+    if (call_count == 0) {
+        AFR_STACK_UNWIND(lk, frame, local->op_ret, local->op_errno, NULL,
+                         local->xdata_rsp);
+    }
+
+    return 0;
+}
 
+int32_t
+afr_lk_unlock(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *local = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int call_count = 0;
 
-        int call_count = -1;
+    local = frame->local;
+    priv = this->private;
 
-        local = frame->local;
+    call_count = afr_locked_nodes_count(local->cont.lk.locked_nodes,
+                                        priv->child_count);
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == 0)
-                        local->op_ret = 0;
+    if (call_count == 0) {
+        AFR_STACK_UNWIND(lk, frame, local->op_ret, local->op_errno, NULL,
+                         local->xdata_rsp);
+        return 0;
+    }
 
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    local->call_count = call_count;
 
-        call_count = afr_frame_return (frame);
+    local->cont.lk.user_flock.l_type = F_UNLCK;
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (inodelk, frame, local->op_ret,
-                                  local->op_errno);
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->cont.lk.locked_nodes[i]) {
+            STACK_WIND_COOKIE(frame, afr_lk_unlock_cbk, (void *)(long)i,
+                              priv->children[i], priv->children[i]->fops->lk,
+                              local->fd, F_SETLK, &local->cont.lk.user_flock,
+                              NULL);
 
-        return 0;
-}
+            if (!--call_count)
+                break;
+        }
+    }
 
+    return 0;
+}
 
 int32_t
-afr_inodelk (call_frame_t *frame, xlator_t *this,
-             const char *volume, loc_t *loc, int32_t cmd, struct gf_flock *flock)
+afr_lk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+           int32_t op_errno, struct gf_flock *lock, dict_t *xdata)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local  = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int child_index = -1;
 
-        int ret = -1;
+    local = frame->local;
+    priv = this->private;
 
-        int i = 0;
-        int32_t call_count = 0;
-        int32_t op_ret   = -1;
-        int32_t op_errno = 0;
+    child_index = (long)cookie;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    afr_common_lock_cbk(frame, cookie, this, op_ret, op_errno, xdata);
+    if (op_ret < 0 && op_errno == EAGAIN) {
+        local->op_ret = -1;
+        local->op_errno = EAGAIN;
 
-        priv = this->private;
+        afr_lk_unlock(frame, this);
+        return 0;
+    }
+
+    if (op_ret == 0) {
+        local->op_ret = 0;
+        local->op_errno = 0;
+        local->cont.lk.locked_nodes[child_index] = 1;
+        local->cont.lk.ret_flock = *lock;
+    }
+
+    child_index++;
+
+    if (child_index < priv->child_count) {
+        STACK_WIND_COOKIE(frame, afr_lk_cbk, (void *)(long)child_index,
+                          priv->children[child_index],
+                          priv->children[child_index]->fops->lk, local->fd,
+                          local->cont.lk.cmd, &local->cont.lk.user_flock,
+                          local->xdata_req);
+    } else if (priv->quorum_count &&
+               !afr_has_quorum(local->cont.lk.locked_nodes, this, NULL)) {
+        local->op_ret = -1;
+        local->op_errno = afr_final_errno(local, priv);
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+        afr_lk_unlock(frame, this);
+    } else {
+        if (local->op_ret < 0)
+            local->op_errno = afr_final_errno(local, priv);
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+        AFR_STACK_UNWIND(lk, frame, local->op_ret, local->op_errno,
+                         &local->cont.lk.ret_flock, local->xdata_rsp);
+    }
 
-        call_count = local->call_count;
-        frame->local = local;
+    return 0;
+}
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND (frame, afr_inodelk_cbk,
-                                    priv->children[i],
-                                    priv->children[i]->fops->inodelk,
-                                    volume, loc, cmd, flock);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+int
+afr_lk_transaction_cbk(int ret, call_frame_t *frame, void *opaque)
+{
+    return 0;
+}
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (inodelk, frame, op_ret, op_errno);
-        }
-        return 0;
+int
+afr_lk_txn_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct gf_flock *lock,
+                    dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int child_index = -1;
+
+    local = frame->local;
+    child_index = (long)cookie;
+    afr_common_lock_cbk(frame, cookie, this, op_ret, op_errno, xdata);
+    if (op_ret == 0) {
+        local->op_ret = 0;
+        local->op_errno = 0;
+        local->cont.lk.locked_nodes[child_index] = 1;
+        local->cont.lk.ret_flock = *lock;
+    }
+    syncbarrier_wake(&local->barrier);
+    return 0;
 }
 
+int
+afr_lk_txn_unlock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct gf_flock *lock,
+                      dict_t *xdata)
+{
+    afr_local_t *local = frame->local;
+    afr_private_t *priv = this->private;
+    int child_index = (long)cookie;
+
+    if (op_ret < 0 && op_errno != ENOTCONN && op_errno != EBADFD) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, AFR_MSG_UNLOCK_FAIL,
+               "gfid=%s: unlock failed on subvolume %s "
+               "with lock owner %s",
+               uuid_utoa(local->fd->inode->gfid),
+               priv->children[child_index]->name,
+               lkowner_utoa(&frame->root->lk_owner));
+    }
+    return 0;
+}
+int
+afr_lk_transaction(void *opaque)
+{
+    call_frame_t *frame = NULL;
+    xlator_t *this = NULL;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    char *wind_on = NULL;
+    int op_errno = 0;
+    int i = 0;
+    int ret = 0;
+
+    frame = (call_frame_t *)opaque;
+    local = frame->local;
+    this = frame->this;
+    priv = this->private;
+    wind_on = alloca0(priv->child_count);
+
+    if (priv->arbiter_count || priv->child_count != 3) {
+        op_errno = ENOTSUP;
+        gf_msg(frame->this->name, GF_LOG_ERROR, op_errno, AFR_MSG_LK_HEAL_DOM,
+               "%s: Lock healing supported only for replica 3 volumes.",
+               uuid_utoa(local->fd->inode->gfid));
+        goto err;
+    }
+
+    op_errno = -afr_dom_lock_acquire(frame);  // Released during
+                                              // AFR_STACK_UNWIND
+    if (op_errno != 0) {
+        goto err;
+    }
+    if (priv->quorum_count &&
+        !afr_has_quorum(local->cont.lk.dom_locked_nodes, this, NULL)) {
+        op_errno = afr_final_errno(local, priv);
+        goto err;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (priv->child_up[i] && local->cont.lk.dom_locked_nodes[i])
+            wind_on[i] = 1;
+    }
+    AFR_ONLIST(wind_on, frame, afr_lk_txn_wind_cbk, lk, local->fd,
+               local->cont.lk.cmd, &local->cont.lk.user_flock,
+               local->xdata_req);
+
+    if (priv->quorum_count &&
+        !afr_has_quorum(local->cont.lk.locked_nodes, this, NULL)) {
+        local->op_ret = -1;
+        local->op_errno = afr_final_errno(local, priv);
+        goto unlock;
+    } else {
+        if (local->cont.lk.user_flock.l_type == F_UNLCK)
+            ret = afr_remove_lock_from_saved_locks(local, this);
+        else
+            ret = afr_add_lock_to_saved_locks(frame, this);
+        if (ret) {
+            local->op_ret = -1;
+            local->op_errno = -ret;
+            goto unlock;
+        }
+        AFR_STACK_UNWIND(lk, frame, local->op_ret, local->op_errno,
+                         &local->cont.lk.ret_flock, local->xdata_rsp);
+    }
+
+    return 0;
 
-int32_t
-afr_finodelk_cbk (call_frame_t *frame, void *cookie,
-                  xlator_t *this, int32_t op_ret, int32_t op_errno)
+unlock:
+    local->cont.lk.user_flock.l_type = F_UNLCK;
+    AFR_ONLIST(local->cont.lk.locked_nodes, frame, afr_lk_txn_unlock_cbk, lk,
+               local->fd, F_SETLK, &local->cont.lk.user_flock, NULL);
+err:
+    AFR_STACK_UNWIND(lk, frame, -1, op_errno, NULL, NULL);
+    return -1;
+}
 
+int
+afr_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+       struct gf_flock *flock, dict_t *xdata)
 {
-        afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int ret = 0;
+    int i = 0;
+    int32_t op_errno = ENOMEM;
+
+    priv = this->private;
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    local->op = GF_FOP_LK;
+    if (!afr_lk_is_unlock(cmd, flock)) {
+        AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+        if (!afr_is_consistent_io_possible(local, priv, &op_errno))
+            goto out;
+    }
+
+    local->cont.lk.locked_nodes = GF_CALLOC(
+        priv->child_count, sizeof(*local->cont.lk.locked_nodes),
+        gf_afr_mt_char);
+
+    if (!local->cont.lk.locked_nodes) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    local->fd = fd_ref(fd);
+    local->cont.lk.cmd = cmd;
+    local->cont.lk.user_flock = *flock;
+    local->cont.lk.ret_flock = *flock;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
+
+    if (afr_is_lock_mode_mandatory(xdata)) {
+        ret = synctask_new(this->ctx->env, afr_lk_transaction,
+                           afr_lk_transaction_cbk, frame, frame);
+        if (ret) {
+            op_errno = ENOMEM;
+            goto out;
+        }
+        return 0;
+    }
 
-        int call_count = -1;
+    STACK_WIND_COOKIE(frame, afr_lk_cbk, (void *)(long)0, priv->children[i],
+                      priv->children[i]->fops->lk, fd, cmd, flock,
+                      local->xdata_req);
 
-        local = frame->local;
+    return 0;
+out:
+    AFR_STACK_UNWIND(lk, frame, -1, op_errno, NULL, NULL);
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == 0)
-                        local->op_ret = 0;
+    return 0;
+}
 
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+int32_t
+afr_lease_unlock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct gf_lease *lease,
+                     dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int call_count = -1;
 
-        call_count = afr_frame_return (frame);
+    local = frame->local;
+    call_count = afr_frame_return(frame);
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (finodelk, frame, local->op_ret,
-                                  local->op_errno);
+    if (call_count == 0)
+        AFR_STACK_UNWIND(lease, frame, local->op_ret, local->op_errno, lease,
+                         xdata);
 
-        return 0;
+    return 0;
 }
 
-
 int32_t
-afr_finodelk (call_frame_t *frame, xlator_t *this,
-              const char *volume, fd_t *fd, int32_t cmd, struct gf_flock *flock)
+afr_lease_unlock(call_frame_t *frame, xlator_t *this)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local  = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int call_count = 0;
 
-        int ret = -1;
+    local = frame->local;
+    priv = this->private;
 
-        int i = 0;
-        int32_t call_count = 0;
-        int32_t op_ret   = -1;
-        int32_t op_errno = 0;
+    call_count = afr_locked_nodes_count(local->cont.lease.locked_nodes,
+                                        priv->child_count);
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    if (call_count == 0) {
+        AFR_STACK_UNWIND(lease, frame, local->op_ret, local->op_errno,
+                         &local->cont.lease.ret_lease, NULL);
+        return 0;
+    }
 
-        priv = this->private;
+    local->call_count = call_count;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+    local->cont.lease.user_lease.cmd = GF_UNLK_LEASE;
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->cont.lease.locked_nodes[i]) {
+            STACK_WIND(frame, afr_lease_unlock_cbk, priv->children[i],
+                       priv->children[i]->fops->lease, &local->loc,
+                       &local->cont.lease.user_lease, NULL);
+
+            if (!--call_count)
+                break;
         }
+    }
 
-        call_count = local->call_count;
-        frame->local = local;
+    return 0;
+}
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND (frame, afr_finodelk_cbk,
-                                    priv->children[i],
-                                    priv->children[i]->fops->finodelk,
-                                    volume, fd, cmd, flock);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+int32_t
+afr_lease_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct gf_lease *lease, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int child_index = -1;
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (finodelk, frame, op_ret, op_errno);
-        }
+    local = frame->local;
+    priv = this->private;
+
+    child_index = (long)cookie;
+
+    afr_common_lock_cbk(frame, cookie, this, op_ret, op_errno, xdata);
+    if (op_ret < 0 && op_errno == EAGAIN) {
+        local->op_ret = -1;
+        local->op_errno = EAGAIN;
+
+        afr_lease_unlock(frame, this);
         return 0;
-}
+    }
+
+    if (op_ret == 0) {
+        local->op_ret = 0;
+        local->op_errno = 0;
+        local->cont.lease.locked_nodes[child_index] = 1;
+        local->cont.lease.ret_lease = *lease;
+    }
+
+    child_index++;
+    if (child_index < priv->child_count) {
+        STACK_WIND_COOKIE(frame, afr_lease_cbk, (void *)(long)child_index,
+                          priv->children[child_index],
+                          priv->children[child_index]->fops->lease, &local->loc,
+                          &local->cont.lease.user_lease, xdata);
+    } else if (priv->quorum_count &&
+               !afr_has_quorum(local->cont.lease.locked_nodes, this, NULL)) {
+        local->op_ret = -1;
+        local->op_errno = afr_final_errno(local, priv);
 
+        afr_lease_unlock(frame, this);
+    } else {
+        if (local->op_ret < 0)
+            local->op_errno = afr_final_errno(local, priv);
+        AFR_STACK_UNWIND(lease, frame, local->op_ret, local->op_errno,
+                         &local->cont.lease.ret_lease, NULL);
+    }
 
-int32_t
-afr_entrylk_cbk (call_frame_t *frame, void *cookie,
-                 xlator_t *this, int32_t op_ret, int32_t op_errno)
+    return 0;
+}
 
+int
+afr_lease(call_frame_t *frame, xlator_t *this, loc_t *loc,
+          struct gf_lease *lease, dict_t *xdata)
 {
-        afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int32_t op_errno = ENOMEM;
 
-        int call_count = -1;
+    priv = this->private;
 
-        local = frame->local;
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == 0)
-                        local->op_ret = 0;
+    local->op = GF_FOP_LEASE;
+    local->cont.lease.locked_nodes = GF_CALLOC(
+        priv->child_count, sizeof(*local->cont.lease.locked_nodes),
+        gf_afr_mt_char);
 
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    if (!local->cont.lease.locked_nodes) {
+        op_errno = ENOMEM;
+        goto out;
+    }
 
-        call_count = afr_frame_return (frame);
+    loc_copy(&local->loc, loc);
+    local->cont.lease.user_lease = *lease;
+    local->cont.lease.ret_lease = *lease;
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (entrylk, frame, local->op_ret,
-                                  local->op_errno);
+    STACK_WIND_COOKIE(frame, afr_lease_cbk, (void *)(long)0, priv->children[0],
+                      priv->children[0]->fops->lease, loc, lease, xdata);
 
-        return 0;
-}
+    return 0;
+out:
+    AFR_STACK_UNWIND(lease, frame, -1, op_errno, NULL, NULL);
 
+    return 0;
+}
 
-int32_t
-afr_entrylk (call_frame_t *frame, xlator_t *this,
-             const char *volume, loc_t *loc,
-             const char *basename, entrylk_cmd cmd, entrylk_type type)
+int
+afr_ipc_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, dict_t *xdata)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local  = NULL;
+    afr_local_t *local = NULL;
+    int child_index = (long)cookie;
+    int call_count = 0;
+    gf_boolean_t failed = _gf_false;
+    gf_boolean_t succeeded = _gf_false;
+    int i = 0;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    local->replies[child_index].valid = 1;
+    local->replies[child_index].op_ret = op_ret;
+    local->replies[child_index].op_errno = op_errno;
+    if (xdata)
+        local->replies[child_index].xdata = dict_ref(xdata);
+
+    call_count = afr_frame_return(frame);
+    if (call_count)
+        goto out;
+    /* If any of the subvolumes failed with other than ENOTCONN
+     * return error else return success unless all the subvolumes
+     * failed.
+     * TODO: In case of failure, we need to unregister the xattrs
+     * from the other subvolumes where it succeeded (once upcall
+     * fixes the Bz-1371622)*/
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
+        if (local->replies[i].op_ret < 0 &&
+            local->replies[i].op_errno != ENOTCONN) {
+            local->op_ret = local->replies[i].op_ret;
+            local->op_errno = local->replies[i].op_errno;
+            if (local->xdata_rsp)
+                dict_unref(local->xdata_rsp);
+            local->xdata_rsp = NULL;
+            if (local->replies[i].xdata) {
+                local->xdata_rsp = dict_ref(local->replies[i].xdata);
+            }
+            failed = _gf_true;
+            break;
+        }
+        if (local->replies[i].op_ret == 0) {
+            succeeded = _gf_true;
+            local->op_ret = 0;
+            local->op_errno = 0;
+            if (!local->xdata_rsp && local->replies[i].xdata) {
+                local->xdata_rsp = dict_ref(local->replies[i].xdata);
+            }
+        }
+    }
 
-        int ret = -1;
+    if (!succeeded && !failed) {
+        local->op_ret = -1;
+        local->op_errno = ENOTCONN;
+    }
 
-        int i = 0;
-        int32_t call_count = 0;
-        int32_t op_ret   = -1;
-        int32_t op_errno = 0;
+    AFR_STACK_UNWIND(ipc, frame, local->op_ret, local->op_errno,
+                     local->xdata_rsp);
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+out:
+    return 0;
+}
 
-        priv = this->private;
+int
+afr_ipc(call_frame_t *frame, xlator_t *this, int32_t op, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int32_t op_errno = -1;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int call_cnt = -1;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+    if (op != GF_IPC_TARGET_UPCALL)
+        goto wind_default;
 
-        call_count = local->call_count;
-        frame->local = local;
+    VALIDATE_OR_GOTO(this->private, err);
+    priv = this->private;
 
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto err;
+
+    call_cnt = local->call_count;
+
+    if (xdata) {
         for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND (frame, afr_entrylk_cbk,
-                                    priv->children[i],
-                                    priv->children[i]->fops->entrylk,
-                                    volume, loc, basename, cmd, type);
-
-                        if (!--call_count)
-                                break;
-                }
+            if (dict_set_int8(xdata, priv->pending_key[i], 0) < 0)
+                goto err;
         }
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->child_up[i])
+            continue;
+
+        STACK_WIND_COOKIE(frame, afr_ipc_cbk, (void *)(long)i,
+                          priv->children[i], priv->children[i]->fops->ipc, op,
+                          xdata);
+        if (!--call_cnt)
+            break;
+    }
+    return 0;
+
+err:
+    if (op_errno == -1)
+        op_errno = errno;
+    AFR_STACK_UNWIND(ipc, frame, -1, op_errno, NULL);
+
+    return 0;
+
+wind_default:
+    STACK_WIND(frame, default_ipc_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ipc, op, xdata);
+    return 0;
+}
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (entrylk, frame, op_ret, op_errno);
-        }
+int
+afr_forget(xlator_t *this, inode_t *inode)
+{
+    uint64_t ctx_int = 0;
+    afr_inode_ctx_t *ctx = NULL;
+
+    afr_spb_choice_timeout_cancel(this, inode);
+    inode_ctx_del(inode, this, &ctx_int);
+    if (!ctx_int)
         return 0;
+
+    ctx = (afr_inode_ctx_t *)(uintptr_t)ctx_int;
+    afr_inode_ctx_destroy(ctx);
+    return 0;
 }
 
+int
+afr_priv_dump(xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    char key_prefix[GF_DUMP_MAX_BUF_LEN];
+    char key[GF_DUMP_MAX_BUF_LEN];
+    int i = 0;
+
+    GF_ASSERT(this);
+    priv = this->private;
+
+    GF_ASSERT(priv);
+    snprintf(key_prefix, GF_DUMP_MAX_BUF_LEN, "%s.%s", this->type, this->name);
+    gf_proc_dump_add_section("%s", key_prefix);
+    gf_proc_dump_write("child_count", "%u", priv->child_count);
+    for (i = 0; i < priv->child_count; i++) {
+        sprintf(key, "child_up[%d]", i);
+        gf_proc_dump_write(key, "%d", priv->child_up[i]);
+        sprintf(key, "pending_key[%d]", i);
+        gf_proc_dump_write(key, "%s", priv->pending_key[i]);
+        sprintf(key, "pending_reads[%d]", i);
+        gf_proc_dump_write(key, "%" PRId64,
+                           GF_ATOMIC_GET(priv->pending_reads[i]));
+        sprintf(key, "child_latency[%d]", i);
+        gf_proc_dump_write(key, "%" PRId64, priv->child_latency[i]);
+        sprintf(key, "halo_child_up[%d]", i);
+        gf_proc_dump_write(key, "%d", priv->halo_child_up[i]);
+    }
+    gf_proc_dump_write("data_self_heal", "%d", priv->data_self_heal);
+    gf_proc_dump_write("metadata_self_heal", "%d", priv->metadata_self_heal);
+    gf_proc_dump_write("entry_self_heal", "%d", priv->entry_self_heal);
+    gf_proc_dump_write("read_child", "%d", priv->read_child);
+    gf_proc_dump_write("wait_count", "%u", priv->wait_count);
+    gf_proc_dump_write("heal-wait-queue-length", "%d", priv->heal_wait_qlen);
+    gf_proc_dump_write("heal-waiters", "%d", priv->heal_waiters);
+    gf_proc_dump_write("background-self-heal-count", "%d",
+                       priv->background_self_heal_count);
+    gf_proc_dump_write("healers", "%d", priv->healers);
+    gf_proc_dump_write("read-hash-mode", "%d", priv->hash_mode);
+    gf_proc_dump_write("use-anonymous-inode", "%d", priv->use_anon_inode);
+    if (priv->quorum_count == AFR_QUORUM_AUTO) {
+        gf_proc_dump_write("quorum-type", "auto");
+    } else if (priv->quorum_count == 0) {
+        gf_proc_dump_write("quorum-type", "none");
+    } else {
+        gf_proc_dump_write("quorum-type", "fixed");
+        gf_proc_dump_write("quorum-count", "%d", priv->quorum_count);
+    }
+    gf_proc_dump_write("up", "%u", afr_has_quorum(priv->child_up, this, NULL));
+    if (priv->thin_arbiter_count) {
+        gf_proc_dump_write("ta_child_up", "%d", priv->ta_child_up);
+        gf_proc_dump_write("ta_bad_child_index", "%d",
+                           priv->ta_bad_child_index);
+        gf_proc_dump_write("ta_notify_dom_lock_offset", "%" PRId64,
+                           priv->ta_notify_dom_lock_offset);
+    }
+
+    return 0;
+}
 
+/**
+ * find_child_index - find the child's index in the array of subvolumes
+ * @this: AFR
+ * @child: child
+ */
 
-int32_t
-afr_fentrylk_cbk (call_frame_t *frame, void *cookie,
-                 xlator_t *this, int32_t op_ret, int32_t op_errno)
+static int
+afr_find_child_index(xlator_t *this, xlator_t *child)
+{
+    afr_private_t *priv = NULL;
+    int child_count = -1;
+    int i = -1;
+
+    priv = this->private;
+    child_count = priv->child_count;
+    if (priv->thin_arbiter_count) {
+        child_count++;
+    }
+
+    for (i = 0; i < child_count; i++) {
+        if ((xlator_t *)child == priv->children[i])
+            break;
+    }
+
+    return i;
+}
 
+int
+__afr_get_up_children_count(afr_private_t *priv)
 {
-        afr_local_t *local = NULL;
+    int up_children = 0;
+    int i = 0;
 
-        int call_count = -1;
+    for (i = 0; i < priv->child_count; i++)
+        if (priv->child_up[i] == 1)
+            up_children++;
 
-        local = frame->local;
+    return up_children;
+}
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret == 0)
-                        local->op_ret = 0;
+static int
+__get_heard_from_all_status(xlator_t *this)
+{
+    afr_private_t *priv = this->private;
+    int i;
 
-                local->op_errno = op_errno;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!priv->last_event[i]) {
+            return 0;
         }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
+    }
+    if (priv->thin_arbiter_count && !priv->ta_child_up) {
+        return 0;
+    }
+    return 1;
+}
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (fentrylk, frame, local->op_ret,
-                                  local->op_errno);
+glusterfs_event_t
+__afr_transform_event_from_state(xlator_t *this)
+{
+    int i = 0;
+    int up_children = 0;
+    afr_private_t *priv = this->private;
+
+    if (__get_heard_from_all_status(this))
+        /* have_heard_from_all. Let afr_notify() do the propagation. */
+        return GF_EVENT_MAXVAL;
+
+    up_children = __afr_get_up_children_count(priv);
+    /* Treat the children with pending notification, as having sent a
+     * GF_EVENT_CHILD_DOWN. i.e. set the event as GF_EVENT_SOME_DESCENDENT_DOWN,
+     * as done in afr_notify() */
+    for (i = 0; i < priv->child_count; i++) {
+        if (priv->last_event[i])
+            continue;
+        priv->last_event[i] = GF_EVENT_SOME_DESCENDENT_DOWN;
+        priv->child_up[i] = 0;
+    }
+
+    if (up_children)
+        /* We received at least one child up */
+        return GF_EVENT_CHILD_UP;
+    else
+        return GF_EVENT_CHILD_DOWN;
+
+    return GF_EVENT_MAXVAL;
+}
 
-        return 0;
+static void
+afr_notify_cbk(void *data)
+{
+    xlator_t *this = data;
+    afr_private_t *priv = this->private;
+    glusterfs_event_t event = GF_EVENT_MAXVAL;
+    gf_boolean_t propagate = _gf_false;
+
+    LOCK(&priv->lock);
+    {
+        if (!priv->timer) {
+            /*
+             * Either child_up/child_down is already sent to parent.
+             * This is a spurious wake up.
+             */
+            goto unlock;
+        }
+        priv->timer = NULL;
+        event = __afr_transform_event_from_state(this);
+        if (event != GF_EVENT_MAXVAL)
+            propagate = _gf_true;
+    }
+unlock:
+    UNLOCK(&priv->lock);
+    if (propagate)
+        default_notify(this, event, NULL);
 }
 
+static void
+__afr_launch_notify_timer(xlator_t *this, afr_private_t *priv)
+{
+    struct timespec delay = {
+        0,
+    };
+
+    gf_msg_debug(this->name, 0, "Initiating child-down timer");
+    delay.tv_sec = 10;
+    delay.tv_nsec = 0;
+    priv->timer = gf_timer_call_after(this->ctx, delay, afr_notify_cbk, this);
+    if (priv->timer == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_TIMER_CREATE_FAIL,
+               "Cannot create timer for delayed initialization");
+    }
+}
 
-int32_t
-afr_fentrylk (call_frame_t *frame, xlator_t *this,
-              const char *volume, fd_t *fd,
-              const char *basename, entrylk_cmd cmd, entrylk_type type)
+static int
+find_best_down_child(xlator_t *this)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local  = NULL;
+    afr_private_t *priv = NULL;
+    int i = -1;
+    int32_t best_child = -1;
+    int64_t best_latency = INT64_MAX;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!priv->child_up[i] && priv->child_latency[i] >= 0 &&
+            priv->child_latency[i] < best_latency) {
+            best_child = i;
+            best_latency = priv->child_latency[i];
+        }
+    }
+    if (best_child >= 0) {
+        gf_msg_debug(this->name, 0,
+                     "Found best down child (%d) @ %" PRId64 " ms latency",
+                     best_child, best_latency);
+    }
+    return best_child;
+}
 
-        int ret = -1;
+int
+find_worst_up_child(xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    int i = -1;
+    int32_t worst_child = -1;
+    int64_t worst_latency = INT64_MIN;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (priv->child_up[i] && priv->child_latency[i] >= 0 &&
+            priv->child_latency[i] > worst_latency) {
+            worst_child = i;
+            worst_latency = priv->child_latency[i];
+        }
+    }
+    if (worst_child >= 0) {
+        gf_msg_debug(this->name, 0,
+                     "Found worst up child (%d) @ %" PRId64 " ms latency",
+                     worst_child, worst_latency);
+    }
+    return worst_child;
+}
 
-        int i = 0;
-        int32_t call_count = 0;
-        int32_t op_ret   = -1;
-        int32_t op_errno = 0;
+void
+__afr_handle_ping_event(xlator_t *this, xlator_t *child_xlator, const int idx,
+                        int64_t halo_max_latency_msec, int32_t *event,
+                        int64_t child_latency_msec)
+{
+    afr_private_t *priv = NULL;
+    int up_children = 0;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    priv = this->private;
 
-        priv = this->private;
+    priv->child_latency[idx] = child_latency_msec;
+    gf_msg_debug(child_xlator->name, 0, "Client ping @ %" PRId64 " ms",
+                 child_latency_msec);
+    if (priv->shd.iamshd)
+        return;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+    up_children = __afr_get_up_children_count(priv);
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
+    if (child_latency_msec > halo_max_latency_msec &&
+        priv->child_up[idx] == 1 && up_children > priv->halo_min_replicas) {
+        if ((up_children - 1) < priv->halo_min_replicas) {
+            gf_log(child_xlator->name, GF_LOG_INFO,
+                   "Overriding halo threshold, "
+                   "min replicas: %d",
+                   priv->halo_min_replicas);
+        } else {
+            gf_log(child_xlator->name, GF_LOG_INFO,
+                   "Child latency (%" PRId64
+                   " ms) "
+                   "exceeds halo threshold (%" PRId64
+                   "), "
+                   "marking child down.",
+                   child_latency_msec, halo_max_latency_msec);
+            if (priv->halo_child_up[idx]) {
+                *event = GF_EVENT_CHILD_DOWN;
+            }
+        }
+    } else if (child_latency_msec < halo_max_latency_msec &&
+               priv->child_up[idx] == 0) {
+        if (up_children < priv->halo_max_replicas) {
+            gf_log(child_xlator->name, GF_LOG_INFO,
+                   "Child latency (%" PRId64
+                   " ms) "
+                   "below halo threshold (%" PRId64
+                   "), "
+                   "marking child up.",
+                   child_latency_msec, halo_max_latency_msec);
+            if (priv->halo_child_up[idx]) {
+                *event = GF_EVENT_CHILD_UP;
+            }
+        } else {
+            gf_log(child_xlator->name, GF_LOG_INFO,
+                   "Not marking child %d up, "
+                   "max replicas (%d) reached.",
+                   idx, priv->halo_max_replicas);
         }
+    }
+}
 
-        call_count = local->call_count;
-        frame->local = local;
+static int64_t
+afr_get_halo_latency(xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    int64_t halo_max_latency_msec = 0;
+
+    priv = this->private;
+
+    if (priv->shd.iamshd) {
+        halo_max_latency_msec = priv->shd.halo_max_latency_msec;
+    } else if (priv->nfsd.iamnfsd) {
+        halo_max_latency_msec = priv->nfsd.halo_max_latency_msec;
+    } else {
+        halo_max_latency_msec = priv->halo_max_latency_msec;
+    }
+    gf_msg_debug(this->name, 0, "Using halo latency %" PRId64,
+                 halo_max_latency_msec);
+    return halo_max_latency_msec;
+}
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND (frame, afr_fentrylk_cbk,
-                                    priv->children[i],
-                                    priv->children[i]->fops->fentrylk,
-                                    volume, fd, basename, cmd, type);
-
-                        if (!--call_count)
-                                break;
-                }
+void
+__afr_handle_child_up_event(xlator_t *this, xlator_t *child_xlator,
+                            const int idx, int64_t child_latency_msec,
+                            int32_t *event, int32_t *call_psh,
+                            int32_t *up_child)
+{
+    afr_private_t *priv = NULL;
+    int up_children = 0;
+    int worst_up_child = -1;
+    int64_t halo_max_latency_msec = afr_get_halo_latency(this);
+
+    priv = this->private;
+
+    /*
+     * This only really counts if the child was never up
+     * (value = -1) or had been down (value = 0).  See
+     * comment at GF_EVENT_CHILD_DOWN for a more detailed
+     * explanation.
+     */
+    if (priv->child_up[idx] != 1) {
+        priv->event_generation++;
+    }
+    priv->child_up[idx] = 1;
+
+    *call_psh = 1;
+    *up_child = idx;
+    up_children = __afr_get_up_children_count(priv);
+    /*
+     * If this is an _actual_ CHILD_UP event, we
+     * want to set the child_latency to MAX to indicate
+     * the child needs ping data to be available before doing child-up
+     */
+    if (!priv->halo_enabled)
+        goto out;
+
+    if (child_latency_msec < 0) {
+        /*set to INT64_MAX-1 so that it is found for best_down_child*/
+        priv->halo_child_up[idx] = 1;
+        if (priv->child_latency[idx] < 0) {
+            priv->child_latency[idx] = AFR_HALO_MAX_LATENCY;
         }
+    }
+
+    /*
+     * Handle the edge case where we exceed
+     * halo_min_replicas and we've got a child which is
+     * marked up as it was helping to satisfy the
+     * halo_min_replicas even though it's latency exceeds
+     * halo_max_latency_msec.
+     */
+    if (up_children > priv->halo_min_replicas) {
+        worst_up_child = find_worst_up_child(this);
+        if (worst_up_child >= 0 &&
+            priv->child_latency[worst_up_child] > halo_max_latency_msec) {
+            gf_msg_debug(this->name, 0,
+                         "Marking child %d down, "
+                         "doesn't meet halo threshold (%" PRId64
+                         "), and > "
+                         "halo_min_replicas (%d)",
+                         worst_up_child, halo_max_latency_msec,
+                         priv->halo_min_replicas);
+            priv->child_up[worst_up_child] = 0;
+            up_children--;
+        }
+    }
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (fentrylk, frame, op_ret, op_errno);
+    if (up_children > priv->halo_max_replicas && !priv->shd.iamshd) {
+        worst_up_child = find_worst_up_child(this);
+        if (worst_up_child < 0) {
+            worst_up_child = idx;
         }
-        return 0;
+        priv->child_up[worst_up_child] = 0;
+        up_children--;
+        gf_msg_debug(this->name, 0,
+                     "Marking child %d down, "
+                     "up_children (%d) > halo_max_replicas (%d)",
+                     worst_up_child, up_children, priv->halo_max_replicas);
+    }
+out:
+    if (up_children == 1) {
+        gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_SUBVOL_UP,
+               "Subvolume '%s' came back up; "
+               "going online.",
+               child_xlator->name);
+        gf_event(EVENT_AFR_SUBVOL_UP, "client-pid=%d; subvol=%s",
+                 this->ctx->cmd_args.client_pid, this->name);
+    } else {
+        *event = GF_EVENT_SOME_DESCENDENT_UP;
+    }
+
+    priv->last_event[idx] = *event;
 }
 
-int32_t
-afr_statfs_cbk (call_frame_t *frame, void *cookie,
-                xlator_t *this, int32_t op_ret, int32_t op_errno,
-                struct statvfs *statvfs)
+void
+__afr_handle_child_down_event(xlator_t *this, xlator_t *child_xlator, int idx,
+                              int64_t child_latency_msec, int32_t *event,
+                              int32_t *call_psh, int32_t *up_child)
 {
-        afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int up_children = 0;
+    int down_children = 0;
+    int best_down_child = -1;
+
+    priv = this->private;
+
+    /*
+     * If a brick is down when we start, we'll get a
+     * CHILD_DOWN to indicate its initial state.  There
+     * was never a CHILD_UP in this case, so if we
+     * increment "down_count" the difference between than
+     * and "up_count" will no longer be the number of
+     * children that are currently up.  This has serious
+     * implications e.g. for quorum enforcement, so we
+     * don't increment these values unless the event
+     * represents an actual state transition between "up"
+     * (value = 1) and anything else.
+     */
+    if (priv->child_up[idx] == 1) {
+        priv->event_generation++;
+    }
+
+    /*
+     * If this is an _actual_ CHILD_DOWN event, we
+     * want to set the child_latency to < 0 to indicate
+     * the child is really disconnected.
+     */
+    if (child_latency_msec < 0) {
+        priv->child_latency[idx] = child_latency_msec;
+        priv->halo_child_up[idx] = 0;
+    }
+    priv->child_up[idx] = 0;
+
+    up_children = __afr_get_up_children_count(priv);
+    /*
+     * Handle the edge case where we need to find the
+     * next best child (to mark up) as marking this child
+     * down would cause us to fall below halo_min_replicas.
+     * We will also force the SHD to heal this child _now_
+     * as we want it to be up to date if we are going to
+     * begin using it synchronously.
+     */
+    if (priv->halo_enabled && up_children < priv->halo_min_replicas) {
+        best_down_child = find_best_down_child(this);
+        if (best_down_child >= 0) {
+            gf_msg_debug(this->name, 0,
+                         "Swapping out child %d for "
+                         "child %d to satisfy halo_min_replicas (%d).",
+                         idx, best_down_child, priv->halo_min_replicas);
+            priv->child_up[best_down_child] = 1;
+            *call_psh = 1;
+            *up_child = best_down_child;
+        }
+    }
+    for (i = 0; i < priv->child_count; i++)
+        if (priv->child_up[i] == 0)
+            down_children++;
+    if (down_children == priv->child_count) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SUBVOLS_DOWN,
+               "All subvolumes are down. Going "
+               "offline until at least one of them "
+               "comes back up.");
+        gf_event(EVENT_AFR_SUBVOLS_DOWN, "client-pid=%d; subvol=%s",
+                 this->ctx->cmd_args.client_pid, this->name);
+    } else {
+        *event = GF_EVENT_SOME_DESCENDENT_DOWN;
+    }
+    priv->last_event[idx] = *event;
+}
 
-        int call_count = 0;
+void
+afr_ta_lock_release_synctask(xlator_t *this)
+{
+    call_frame_t *ta_frame = NULL;
+    int ret = 0;
 
-        LOCK (&frame->lock);
-        {
-                local = frame->local;
-
-                if (op_ret == 0) {
-                        local->op_ret   = op_ret;
-
-                        if (local->cont.statfs.buf_set) {
-                                if (statvfs->f_bavail < local->cont.statfs.buf.f_bavail)
-                                        local->cont.statfs.buf = *statvfs;
-                        } else {
-                                local->cont.statfs.buf = *statvfs;
-                                local->cont.statfs.buf_set = 1;
-                        }
-                }
+    ta_frame = afr_ta_frame_create(this);
+    if (!ta_frame) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_THIN_ARB,
+               "Failed to create ta_frame");
+        return;
+    }
+
+    ret = synctask_new(this->ctx->env, afr_release_notify_lock_for_ta,
+                       afr_ta_lock_release_done, ta_frame, this);
+    if (ret) {
+        STACK_DESTROY(ta_frame->root);
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_THIN_ARB,
+               "Failed to release "
+               "AFR_TA_DOM_NOTIFY lock.");
+    }
+}
 
-                if (op_ret == -1)
-                        local->op_errno = op_errno;
+static void
+afr_handle_inodelk_contention(xlator_t *this, struct gf_upcall *upcall)
+{
+    struct gf_upcall_inodelk_contention *lc = NULL;
+    unsigned int inmem_count = 0;
+    unsigned int onwire_count = 0;
+    afr_private_t *priv = this->private;
 
-        }
-        UNLOCK (&frame->lock);
+    lc = upcall->data;
 
-        call_count = afr_frame_return (frame);
+    if (strcmp(lc->domain, AFR_TA_DOM_NOTIFY) != 0)
+        return;
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (statfs, frame, local->op_ret, local->op_errno,
-                                  &local->cont.statfs.buf);
+    if (priv->shd.iamshd) {
+        /* shd should ignore AFR_TA_DOM_NOTIFY release requests. */
+        return;
+    }
+    LOCK(&priv->lock);
+    {
+        if (priv->release_ta_notify_dom_lock == _gf_true) {
+            /* Ignore multiple release requests from shds.*/
+            UNLOCK(&priv->lock);
+            return;
+        }
+        priv->release_ta_notify_dom_lock = _gf_true;
+        inmem_count = priv->ta_in_mem_txn_count;
+        onwire_count = priv->ta_on_wire_txn_count;
+    }
+    UNLOCK(&priv->lock);
+    if (inmem_count || onwire_count)
+        /* lock release will happen in txn code path after
+         * in-memory or on-wire txns are over.*/
+        return;
 
-        return 0;
+    afr_ta_lock_release_synctask(this);
 }
 
+static void
+afr_handle_upcall_event(xlator_t *this, struct gf_upcall *upcall)
+{
+    struct gf_upcall_cache_invalidation *up_ci = NULL;
+    afr_private_t *priv = this->private;
+    inode_t *inode = NULL;
+    inode_table_t *itable = NULL;
+    int i = 0;
+
+    switch (upcall->event_type) {
+        case GF_UPCALL_INODELK_CONTENTION:
+            afr_handle_inodelk_contention(this, upcall);
+            break;
+        case GF_UPCALL_CACHE_INVALIDATION:
+            up_ci = (struct gf_upcall_cache_invalidation *)upcall->data;
+
+            /* Since md-cache will be aggressively filtering
+             * lookups, the stale read issue will be more
+             * pronounced. Hence when a pending xattr is set notify
+             * all the md-cache clients to invalidate the existing
+             * stat cache and send the lookup next time */
+            if (!up_ci->dict)
+                break;
+            for (i = 0; i < priv->child_count; i++) {
+                if (!dict_get(up_ci->dict, priv->pending_key[i]))
+                    continue;
+                up_ci->flags |= UP_INVAL_ATTR;
+                itable = ((xlator_t *)this->graph->top)->itable;
+                /*Internal processes may not have itable for
+                 *top xlator*/
+                if (itable)
+                    inode = inode_find(itable, upcall->gfid);
+                if (inode)
+                    afr_inode_need_refresh_set(inode, this);
+                break;
+            }
+            break;
+        default:
+            break;
+    }
+}
 
 int32_t
-afr_statfs (call_frame_t *frame, xlator_t *this,
-            loc_t *loc)
+afr_notify(xlator_t *this, int32_t event, void *data, void *data2)
 {
-        afr_private_t *  priv        = NULL;
-        int              child_count = 0;
-        afr_local_t   *  local       = NULL;
-        int              i           = 0;
-
-        int ret = -1;
-        int              call_count = 0;
-        int32_t          op_ret      = -1;
-        int32_t          op_errno    = 0;
+    afr_private_t *priv = NULL;
+    xlator_t *child_xlator = NULL;
+    int i = -1;
+    int propagate = 0;
+    int had_heard_from_all = 0;
+    int have_heard_from_all = 0;
+    int idx = -1;
+    int ret = -1;
+    int call_psh = 0;
+    int up_child = -1;
+    dict_t *input = NULL;
+    dict_t *output = NULL;
+    gf_boolean_t had_quorum = _gf_false;
+    gf_boolean_t has_quorum = _gf_false;
+    int64_t halo_max_latency_msec = 0;
+    int64_t child_latency_msec = -1;
+
+    child_xlator = (xlator_t *)data;
+
+    priv = this->private;
+
+    if (!priv)
+        return 0;
 
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-        VALIDATE_OR_GOTO (loc, out);
+    /*
+     * We need to reset this in case children come up in "staggered"
+     * fashion, so that we discover a late-arriving local subvolume.  Note
+     * that we could end up issuing N lookups to the first subvolume, and
+     * O(N^2) overall, but N is small for AFR so it shouldn't be an issue.
+     */
+    priv->did_discovery = _gf_false;
+
+    /* parent xlators don't need to know about every child_up, child_down
+     * because of afr ha. If all subvolumes go down, child_down has
+     * to be triggered. In that state when 1 subvolume comes up child_up
+     * needs to be triggered. dht optimizes revalidate lookup by sending
+     * it only to one of its subvolumes. When child up/down happens
+     * for afr's subvolumes dht should be notified by child_modified. The
+     * subsequent revalidate lookup happens on all the dht's subvolumes
+     * which triggers afr self-heals if any.
+     */
+    idx = afr_find_child_index(this, child_xlator);
+    if (idx < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_INVALID_CHILD_UP,
+               "Received child_up from invalid subvolume");
+        goto out;
+    }
+
+    had_quorum = priv->quorum_count &&
+                 afr_has_quorum(priv->child_up, this, NULL);
+    if (event == GF_EVENT_CHILD_PING) {
+        child_latency_msec = (int64_t)(uintptr_t)data2;
+        if (priv->halo_enabled) {
+            halo_max_latency_msec = afr_get_halo_latency(this);
+
+            /* Calculates the child latency and sets event
+             */
+            LOCK(&priv->lock);
+            {
+                __afr_handle_ping_event(this, child_xlator, idx,
+                                        halo_max_latency_msec, &event,
+                                        child_latency_msec);
+            }
+            UNLOCK(&priv->lock);
+        } else {
+            LOCK(&priv->lock);
+            {
+                priv->child_latency[idx] = child_latency_msec;
+            }
+            UNLOCK(&priv->lock);
+        }
+    }
 
-        priv = this->private;
-        child_count = priv->child_count;
+    if (event == GF_EVENT_CHILD_PING) {
+        /* This is the only xlator that handles PING, no reason to
+         * propagate.
+         */
+        goto out;
+    }
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+    if (event == GF_EVENT_TRANSLATOR_OP) {
+        LOCK(&priv->lock);
+        {
+            had_heard_from_all = __get_heard_from_all_status(this);
+        }
+        UNLOCK(&priv->lock);
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
+        if (!had_heard_from_all) {
+            ret = -1;
+        } else {
+            input = data;
+            output = data2;
+            ret = afr_xl_op(this, input, output);
         }
+        goto out;
+    }
+
+    if (event == GF_EVENT_UPCALL) {
+        afr_handle_upcall_event(this, data);
+    }
 
-        frame->local = local;
-        call_count = local->call_count;
+    LOCK(&priv->lock);
+    {
+        had_heard_from_all = __get_heard_from_all_status(this);
+        switch (event) {
+            case GF_EVENT_PARENT_UP:
+                __afr_launch_notify_timer(this, priv);
+                propagate = 1;
+                break;
+            case GF_EVENT_CHILD_UP:
+                if (priv->thin_arbiter_count &&
+                    (idx == AFR_CHILD_THIN_ARBITER)) {
+                    priv->ta_child_up = 1;
+                    priv->ta_event_gen++;
+                    break;
+                }
+                __afr_handle_child_up_event(this, child_xlator, idx,
+                                            child_latency_msec, &event,
+                                            &call_psh, &up_child);
+                __afr_lock_heal_synctask(this, priv, idx);
+                break;
 
-        for (i = 0; i < child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND (frame, afr_statfs_cbk,
-                                    priv->children[i],
-                                    priv->children[i]->fops->statfs,
-                                    loc);
-                        if (!--call_count)
-                                break;
+            case GF_EVENT_CHILD_DOWN:
+                if (priv->thin_arbiter_count &&
+                    (idx == AFR_CHILD_THIN_ARBITER)) {
+                    priv->ta_child_up = 0;
+                    priv->ta_event_gen++;
+                    afr_ta_locked_priv_invalidate(priv);
+                    break;
                 }
+                __afr_handle_child_down_event(this, child_xlator, idx,
+                                              child_latency_msec, &event,
+                                              &call_psh, &up_child);
+                __afr_mark_pending_lk_heal(this, priv, idx);
+                break;
+
+            case GF_EVENT_CHILD_CONNECTING:
+                priv->last_event[idx] = event;
+
+                break;
+
+            case GF_EVENT_SOME_DESCENDENT_DOWN:
+                priv->last_event[idx] = event;
+                break;
+            default:
+                propagate = 1;
+                break;
         }
+        have_heard_from_all = __get_heard_from_all_status(this);
+        if (!had_heard_from_all && have_heard_from_all) {
+            if (priv->timer) {
+                gf_timer_call_cancel(this->ctx, priv->timer);
+                priv->timer = NULL;
+            }
+            /* This is the first event which completes aggregation
+               of events from all subvolumes. If at least one subvol
+               had come up, propagate CHILD_UP, but only this time
+            */
+            event = GF_EVENT_CHILD_DOWN;
+            for (i = 0; i < priv->child_count; i++) {
+                if (priv->last_event[i] == GF_EVENT_CHILD_UP) {
+                    event = GF_EVENT_CHILD_UP;
+                    break;
+                }
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (statfs, frame, op_ret, op_errno, NULL);
+                if (priv->last_event[i] == GF_EVENT_CHILD_CONNECTING) {
+                    event = GF_EVENT_CHILD_CONNECTING;
+                    /* continue to check other events for CHILD_UP */
+                }
+            }
         }
-        return 0;
+    }
+    UNLOCK(&priv->lock);
+
+    if (priv->quorum_count) {
+        has_quorum = afr_has_quorum(priv->child_up, this, NULL);
+        if (!had_quorum && has_quorum) {
+            gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_QUORUM_MET,
+                   "Client-quorum is met");
+            gf_event(EVENT_AFR_QUORUM_MET, "client-pid=%d; subvol=%s",
+                     this->ctx->cmd_args.client_pid, this->name);
+        }
+        if (had_quorum && !has_quorum) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_QUORUM_FAIL,
+                   "Client-quorum is not met");
+            gf_event(EVENT_AFR_QUORUM_FAIL, "client-pid=%d; subvol=%s",
+                     this->ctx->cmd_args.client_pid, this->name);
+        }
+    }
+
+    /* if all subvols have reported status, no need to hide anything
+       or wait for anything else. Just propagate blindly */
+    if (have_heard_from_all)
+        propagate = 1;
+
+    ret = 0;
+    if (propagate)
+        ret = default_notify(this, event, data);
+
+    if ((!had_heard_from_all) || call_psh) {
+        /* Launch self-heal on all local subvolumes if:
+         * a) We have_heard_from_all for the first time
+         * b) Already heard from everyone, but we now got a child-up
+         *    event.
+         */
+        if (have_heard_from_all) {
+            afr_selfheal_childup(this, priv);
+        }
+    }
+out:
+    return ret;
 }
 
-
-int32_t
-afr_lk_unlock_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct gf_flock *lock)
+int
+afr_local_init(afr_local_t *local, afr_private_t *priv, int32_t *op_errno)
 {
-        afr_local_t * local = NULL;
+    int __ret = -1;
+    local->op_ret = -1;
+    local->op_errno = EUCLEAN;
+
+    __ret = syncbarrier_init(&local->barrier);
+    if (__ret) {
+        if (op_errno)
+            *op_errno = __ret;
+        goto out;
+    }
+
+    local->child_up = GF_MALLOC(priv->child_count * sizeof(*local->child_up),
+                                gf_afr_mt_char);
+    if (!local->child_up) {
+        if (op_errno)
+            *op_errno = ENOMEM;
+        goto out;
+    }
+
+    memcpy(local->child_up, priv->child_up,
+           sizeof(*local->child_up) * priv->child_count);
+    local->call_count = AFR_COUNT(local->child_up, priv->child_count);
+    if (local->call_count == 0) {
+        gf_msg(THIS->name, GF_LOG_INFO, 0, AFR_MSG_SUBVOLS_DOWN,
+               "no subvolumes up");
+        if (op_errno)
+            *op_errno = ENOTCONN;
+        goto out;
+    }
+
+    local->event_generation = priv->event_generation;
+
+    local->read_attempted = GF_CALLOC(priv->child_count, sizeof(char),
+                                      gf_afr_mt_char);
+    if (!local->read_attempted) {
+        if (op_errno)
+            *op_errno = ENOMEM;
+        goto out;
+    }
+
+    local->readable = GF_CALLOC(priv->child_count, sizeof(char),
+                                gf_afr_mt_char);
+    if (!local->readable) {
+        if (op_errno)
+            *op_errno = ENOMEM;
+        goto out;
+    }
+
+    local->readable2 = GF_CALLOC(priv->child_count, sizeof(char),
+                                 gf_afr_mt_char);
+    if (!local->readable2) {
+        if (op_errno)
+            *op_errno = ENOMEM;
+        goto out;
+    }
+
+    local->read_subvol = -1;
+
+    local->replies = GF_CALLOC(priv->child_count, sizeof(*local->replies),
+                               gf_afr_mt_reply_t);
+    if (!local->replies) {
+        if (op_errno)
+            *op_errno = ENOMEM;
+        goto out;
+    }
+
+    local->need_full_crawl = _gf_false;
+    if (priv->thin_arbiter_count) {
+        local->ta_child_up = priv->ta_child_up;
+        local->ta_failed_subvol = AFR_CHILD_UNKNOWN;
+        local->read_txn_query_child = AFR_CHILD_UNKNOWN;
+        local->ta_event_gen = priv->ta_event_gen;
+        local->fop_state = TA_SUCCESS;
+    }
+    local->is_new_entry = _gf_false;
+
+    INIT_LIST_HEAD(&local->healer);
+    return 0;
+out:
+    return -1;
+}
 
-        int call_count = -1;
+int
+afr_internal_lock_init(afr_internal_lock_t *lk, size_t child_count)
+{
+    int ret = -ENOMEM;
 
-        local = frame->local;
-        call_count = afr_frame_return (frame);
+    lk->lower_locked_nodes = GF_CALLOC(sizeof(*lk->lower_locked_nodes),
+                                       child_count, gf_afr_mt_char);
+    if (NULL == lk->lower_locked_nodes)
+        goto out;
 
-        if (call_count == 0)
-                AFR_STACK_UNWIND (lk, frame, local->op_ret, local->op_errno,
-                                  lock);
+    lk->lock_op_ret = -1;
+    lk->lock_op_errno = EUCLEAN;
 
-        return 0;
+    ret = 0;
+out:
+    return ret;
 }
 
-
-int32_t
-afr_lk_unlock (call_frame_t *frame, xlator_t *this)
+void
+afr_matrix_cleanup(int32_t **matrix, unsigned int m)
 {
-        afr_local_t   * local = NULL;
-        afr_private_t * priv  = NULL;
+    int i = 0;
 
-        int i;
-        int call_count = 0;
+    if (!matrix)
+        goto out;
+    for (i = 0; i < m; i++) {
+        GF_FREE(matrix[i]);
+    }
 
-        local = frame->local;
-        priv  = this->private;
+    GF_FREE(matrix);
+out:
+    return;
+}
 
-        call_count = afr_locked_nodes_count (local->cont.lk.locked_nodes,
-                                             priv->child_count);
+int32_t **
+afr_matrix_create(unsigned int m, unsigned int n)
+{
+    int32_t **matrix = NULL;
+    int i = 0;
+
+    matrix = GF_CALLOC(sizeof(*matrix), m, gf_afr_mt_int32_t);
+    if (!matrix)
+        goto out;
+
+    for (i = 0; i < m; i++) {
+        matrix[i] = GF_CALLOC(sizeof(*matrix[i]), n, gf_afr_mt_int32_t);
+        if (!matrix[i])
+            goto out;
+    }
+    return matrix;
+out:
+    afr_matrix_cleanup(matrix, m);
+    return NULL;
+}
 
-        if (call_count == 0) {
-                AFR_STACK_UNWIND (lk, frame, local->op_ret, local->op_errno,
-                                  &local->cont.lk.ret_flock);
-                return 0;
-        }
+int
+afr_transaction_local_init(afr_local_t *local, xlator_t *this)
+{
+    int ret = -ENOMEM;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    INIT_LIST_HEAD(&local->transaction.wait_list);
+    INIT_LIST_HEAD(&local->transaction.owner_list);
+    INIT_LIST_HEAD(&local->ta_waitq);
+    INIT_LIST_HEAD(&local->ta_onwireq);
+    ret = afr_internal_lock_init(&local->internal_lock, priv->child_count);
+    if (ret < 0)
+        goto out;
+
+    ret = -ENOMEM;
+    local->pre_op_compat = priv->pre_op_compat;
+
+    local->transaction.pre_op = GF_CALLOC(sizeof(*local->transaction.pre_op),
+                                          priv->child_count, gf_afr_mt_char);
+    if (!local->transaction.pre_op)
+        goto out;
+
+    local->transaction.changelog_xdata = GF_CALLOC(
+        sizeof(*local->transaction.changelog_xdata), priv->child_count,
+        gf_afr_mt_dict_t);
+    if (!local->transaction.changelog_xdata)
+        goto out;
+
+    if (priv->arbiter_count == 1) {
+        local->transaction.pre_op_sources = GF_CALLOC(
+            sizeof(*local->transaction.pre_op_sources), priv->child_count,
+            gf_afr_mt_char);
+        if (!local->transaction.pre_op_sources)
+            goto out;
+    }
+
+    local->transaction.failed_subvols = GF_CALLOC(
+        sizeof(*local->transaction.failed_subvols), priv->child_count,
+        gf_afr_mt_char);
+    if (!local->transaction.failed_subvols)
+        goto out;
+
+    local->pending = afr_matrix_create(priv->child_count, AFR_NUM_CHANGE_LOGS);
+    if (!local->pending)
+        goto out;
+
+    ret = 0;
+out:
+    return ret;
+}
+
+void
+afr_set_low_priority(call_frame_t *frame)
+{
+    frame->root->pid = LOW_PRIO_PROC_PID;
+}
 
-        local->call_count = call_count;
+void
+afr_priv_destroy(afr_private_t *priv)
+{
+    int i = 0;
+    int child_count = -1;
 
-        local->cont.lk.user_flock.l_type = F_UNLCK;
+    if (!priv)
+        goto out;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->cont.lk.locked_nodes[i]) {
-                        STACK_WIND (frame, afr_lk_unlock_cbk,
-                                    priv->children[i],
-                                    priv->children[i]->fops->lk,
-                                    local->fd, F_SETLK,
-                                    &local->cont.lk.user_flock);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+    GF_FREE(priv->sh_domain);
+    GF_FREE(priv->last_event);
 
-        return 0;
+    child_count = priv->child_count;
+    if (priv->thin_arbiter_count) {
+        child_count++;
+    }
+    if (priv->pending_key) {
+        for (i = 0; i < child_count; i++)
+            GF_FREE(priv->pending_key[i]);
+    }
+
+    GF_FREE(priv->pending_reads);
+    GF_FREE(priv->local);
+    GF_FREE(priv->pending_key);
+    GF_FREE(priv->children);
+    GF_FREE(priv->anon_inode);
+    GF_FREE(priv->child_up);
+    GF_FREE(priv->halo_child_up);
+    GF_FREE(priv->child_latency);
+    LOCK_DESTROY(&priv->lock);
+
+    GF_FREE(priv);
+out:
+    return;
 }
 
+int **
+afr_mark_pending_changelog(afr_private_t *priv, unsigned char *pending,
+                           dict_t *xattr, ia_type_t iat)
+{
+    int i = 0;
+    int **changelog = NULL;
+    int idx = -1;
+    int m_idx = 0;
+    int d_idx = 0;
+    int ret = 0;
+
+    m_idx = afr_index_for_transaction_type(AFR_METADATA_TRANSACTION);
+    d_idx = afr_index_for_transaction_type(AFR_DATA_TRANSACTION);
+
+    idx = afr_index_from_ia_type(iat);
+
+    changelog = afr_matrix_create(priv->child_count, AFR_NUM_CHANGE_LOGS);
+    if (!changelog)
+        goto out;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!pending[i])
+            continue;
+
+        changelog[i][m_idx] = hton32(1);
+        if (idx != -1)
+            changelog[i][idx] = hton32(1);
+        /* If the newentry marking is on a newly created directory,
+         * then mark it with the full-heal indicator.
+         */
+        if ((IA_ISDIR(iat)) && (priv->esh_granular))
+            changelog[i][d_idx] = hton32(1);
+    }
+    ret = afr_set_pending_dict(priv, xattr, changelog);
+    if (ret < 0) {
+        afr_matrix_cleanup(changelog, priv->child_count);
+        return NULL;
+    }
+out:
+    return changelog;
+}
 
-int32_t
-afr_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-            int32_t op_ret, int32_t op_errno, struct gf_flock *lock)
+static dict_t *
+afr_set_heal_info(char *status)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-        int            ret  = 0;
+    dict_t *dict = NULL;
+    int ret = -1;
+
+    dict = dict_new();
+    if (!dict) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ret = dict_set_dynstr_sizen(dict, "heal-info", status);
+    if (ret)
+        gf_msg("", GF_LOG_WARNING, -ret, AFR_MSG_DICT_SET_FAILED,
+               "Failed to set heal-info key to "
+               "%s",
+               status);
+out:
+    /* Any error other than EINVAL, dict_set_dynstr frees status */
+    if (ret == -ENOMEM || ret == -EINVAL) {
+        GF_FREE(status);
+    }
+
+    if (ret && dict) {
+        dict_unref(dict);
+        dict = NULL;
+    }
+    return dict;
+}
 
-        int child_index = -1;
+static gf_boolean_t
+afr_is_dirty_count_non_unary_for_txn(xlator_t *this, struct afr_reply *replies,
+                                     afr_transaction_type type)
+{
+    afr_private_t *priv = this->private;
+    int *dirty = alloca0(priv->child_count * sizeof(int));
+    int i = 0;
 
-        local = frame->local;
-        priv  = this->private;
+    afr_selfheal_extract_xattr(this, replies, type, dirty, NULL);
+    for (i = 0; i < priv->child_count; i++) {
+        if (dirty[i] > 1)
+            return _gf_true;
+    }
 
-        child_index = (long) cookie;
+    return _gf_false;
+}
 
-        if (!child_went_down (op_ret, op_errno) && (op_ret == -1)) {
-                local->op_ret   = -1;
-                local->op_errno = op_errno;
+static gf_boolean_t
+afr_is_dirty_count_non_unary(xlator_t *this, struct afr_reply *replies,
+                             ia_type_t ia_type)
+{
+    gf_boolean_t data_chk = _gf_false;
+    gf_boolean_t mdata_chk = _gf_false;
+    gf_boolean_t entry_chk = _gf_false;
+
+    switch (ia_type) {
+        case IA_IFDIR:
+            mdata_chk = _gf_true;
+            entry_chk = _gf_true;
+            break;
+        case IA_IFREG:
+            mdata_chk = _gf_true;
+            data_chk = _gf_true;
+            break;
+        default:
+            /*IA_IFBLK, IA_IFCHR, IA_IFLNK, IA_IFIFO, IA_IFSOCK*/
+            mdata_chk = _gf_true;
+            break;
+    }
+
+    if (data_chk && afr_is_dirty_count_non_unary_for_txn(
+                        this, replies, AFR_DATA_TRANSACTION)) {
+        return _gf_true;
+    } else if (mdata_chk && afr_is_dirty_count_non_unary_for_txn(
+                                this, replies, AFR_METADATA_TRANSACTION)) {
+        return _gf_true;
+    } else if (entry_chk && afr_is_dirty_count_non_unary_for_txn(
+                                this, replies, AFR_ENTRY_TRANSACTION)) {
+        return _gf_true;
+    }
+
+    return _gf_false;
+}
 
-                afr_lk_unlock (frame, this);
-                return 0;
+static int
+afr_update_heal_status(xlator_t *this, struct afr_reply *replies,
+                       ia_type_t ia_type, gf_boolean_t *esh, gf_boolean_t *dsh,
+                       gf_boolean_t *msh, unsigned char pending)
+{
+    int ret = -1;
+    GF_UNUSED int ret1 = 0;
+    int i = 0;
+    int io_domain_lk_count = 0;
+    int shd_domain_lk_count = 0;
+    afr_private_t *priv = NULL;
+    char *key1 = NULL;
+    char *key2 = NULL;
+
+    priv = this->private;
+    key1 = alloca0(strlen(GLUSTERFS_INODELK_DOM_PREFIX) + 2 +
+                   strlen(this->name));
+    key2 = alloca0(strlen(GLUSTERFS_INODELK_DOM_PREFIX) + 2 +
+                   strlen(priv->sh_domain));
+    sprintf(key1, "%s:%s", GLUSTERFS_INODELK_DOM_PREFIX, this->name);
+    sprintf(key2, "%s:%s", GLUSTERFS_INODELK_DOM_PREFIX, priv->sh_domain);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if ((replies[i].valid != 1) || (replies[i].op_ret != 0))
+            continue;
+        if (!io_domain_lk_count) {
+            ret1 = dict_get_int32(replies[i].xdata, key1, &io_domain_lk_count);
+        }
+        if (!shd_domain_lk_count) {
+            ret1 = dict_get_int32(replies[i].xdata, key2, &shd_domain_lk_count);
         }
+    }
 
-        if (op_ret == 0) {
-                local->op_ret        = 0;
-                local->op_errno      = 0;
-                local->cont.lk.locked_nodes[child_index] = 1;
-                local->cont.lk.ret_flock = *lock;
+    if (!pending) {
+        if ((afr_is_dirty_count_non_unary(this, replies, ia_type)) ||
+            (!io_domain_lk_count)) {
+            /* Needs heal. */
+            ret = 0;
+        } else {
+            /* No heal needed. */
+            *dsh = *esh = *msh = 0;
+        }
+    } else {
+        if (shd_domain_lk_count) {
+            ret = -EAGAIN; /*For 'possibly-healing'. */
+        } else {
+            ret = 0; /*needs heal. Just set a non -ve value so that it is
+                       assumed as the source index.*/
         }
+    }
+    return ret;
+}
 
-        child_index++;
+/*return EIO, EAGAIN or pending*/
+int
+afr_lockless_inspect(call_frame_t *frame, xlator_t *this, uuid_t gfid,
+                     inode_t **inode, gf_boolean_t *entry_selfheal,
+                     gf_boolean_t *data_selfheal,
+                     gf_boolean_t *metadata_selfheal, unsigned char *pending)
+{
+    int ret = -1;
+    int i = 0;
+    afr_private_t *priv = NULL;
+    struct afr_reply *replies = NULL;
+    gf_boolean_t dsh = _gf_false;
+    gf_boolean_t msh = _gf_false;
+    gf_boolean_t esh = _gf_false;
+    unsigned char *sources = NULL;
+    unsigned char *sinks = NULL;
+    unsigned char *valid_on = NULL;
+    uint64_t *witness = NULL;
+
+    priv = this->private;
+    replies = alloca0(sizeof(*replies) * priv->child_count);
+    sources = alloca0(sizeof(*sources) * priv->child_count);
+    sinks = alloca0(sizeof(*sinks) * priv->child_count);
+    witness = alloca0(sizeof(*witness) * priv->child_count);
+    valid_on = alloca0(sizeof(*valid_on) * priv->child_count);
+
+    ret = afr_selfheal_unlocked_inspect(frame, this, gfid, inode, &dsh, &msh,
+                                        &esh, replies);
+    if (ret)
+        goto out;
+    for (i = 0; i < priv->child_count; i++) {
+        if (replies[i].valid && replies[i].op_ret == 0) {
+            valid_on[i] = 1;
+        }
+    }
+    if (msh) {
+        ret = afr_selfheal_find_direction(frame, this, replies,
+                                          AFR_METADATA_TRANSACTION, valid_on,
+                                          sources, sinks, witness, pending);
+        if (*pending & PFLAG_SBRAIN)
+            ret = -EIO;
+        if (ret)
+            goto out;
+    }
+    if (dsh) {
+        ret = afr_selfheal_find_direction(frame, this, replies,
+                                          AFR_DATA_TRANSACTION, valid_on,
+                                          sources, sinks, witness, pending);
+        if (*pending & PFLAG_SBRAIN)
+            ret = -EIO;
+        if (ret)
+            goto out;
+    }
+    if (esh) {
+        ret = afr_selfheal_find_direction(frame, this, replies,
+                                          AFR_ENTRY_TRANSACTION, valid_on,
+                                          sources, sinks, witness, pending);
+        if (*pending & PFLAG_SBRAIN)
+            ret = -EIO;
+        if (ret)
+            goto out;
+    }
 
-        if (child_index < priv->child_count) {
-                STACK_WIND_COOKIE (frame, afr_lk_cbk, (void *) (long) child_index,
-                                   priv->children[child_index],
-                                   priv->children[child_index]->fops->lk,
-                                   local->fd, local->cont.lk.cmd,
-                                   &local->cont.lk.user_flock);
-        } else if (local->op_ret == -1) {
-                /* all nodes have gone down */
+    ret = afr_update_heal_status(this, replies, (*inode)->ia_type, &esh, &dsh,
+                                 &msh, *pending);
+out:
+    *data_selfheal = dsh;
+    *entry_selfheal = esh;
+    *metadata_selfheal = msh;
+    if (replies)
+        afr_replies_wipe(replies, priv->child_count);
+    return ret;
+}
 
-                AFR_STACK_UNWIND (lk, frame, -1, ENOTCONN,
-                                  &local->cont.lk.ret_flock);
+int
+afr_get_heal_info(call_frame_t *frame, xlator_t *this, loc_t *loc)
+{
+    gf_boolean_t data_selfheal = _gf_false;
+    gf_boolean_t metadata_selfheal = _gf_false;
+    gf_boolean_t entry_selfheal = _gf_false;
+    unsigned char pending = 0;
+    dict_t *dict = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
+    inode_t *inode = NULL;
+    char *substr = NULL;
+    char *status = NULL;
+    call_frame_t *heal_frame = NULL;
+    afr_local_t *heal_local = NULL;
+
+    /*Use frame with lk-owner set*/
+    heal_frame = afr_frame_create(frame->this, &op_errno);
+    if (!heal_frame) {
+        ret = -1;
+        goto out;
+    }
+    heal_local = heal_frame->local;
+    heal_frame->local = frame->local;
+
+    ret = afr_lockless_inspect(heal_frame, this, loc->gfid, &inode,
+                               &entry_selfheal, &data_selfheal,
+                               &metadata_selfheal, &pending);
+
+    if (ret == -ENOMEM) {
+        ret = -1;
+        goto out;
+    }
+
+    if (pending & PFLAG_PENDING) {
+        gf_asprintf(&substr, "-pending");
+        if (!substr)
+            goto out;
+    }
+
+    if (ret == -EIO) {
+        ret = gf_asprintf(&status, "split-brain%s", substr ? substr : "");
+        if (ret < 0) {
+            goto out;
+        }
+        dict = afr_set_heal_info(status);
+        if (!dict) {
+            ret = -1;
+            goto out;
+        }
+    } else if (ret == -EAGAIN) {
+        ret = gf_asprintf(&status, "possibly-healing%s", substr ? substr : "");
+        if (ret < 0) {
+            goto out;
+        }
+        dict = afr_set_heal_info(status);
+        if (!dict) {
+            ret = -1;
+            goto out;
+        }
+    } else if (ret >= 0) {
+        /* value of ret = source index
+         * so ret >= 0 and at least one of the 3 booleans set to
+         * true means a source is identified; heal is required.
+         */
+        if (!data_selfheal && !entry_selfheal && !metadata_selfheal) {
+            status = gf_strdup("no-heal");
+            if (!status) {
+                ret = -1;
+                goto out;
+            }
+            dict = afr_set_heal_info(status);
+            if (!dict) {
+                ret = -1;
+                goto out;
+            }
         } else {
-                /* locking has succeeded on all nodes that are up */
+            ret = gf_asprintf(&status, "heal%s", substr ? substr : "");
+            if (ret < 0) {
+                goto out;
+            }
+            dict = afr_set_heal_info(status);
+            if (!dict) {
+                ret = -1;
+                goto out;
+            }
+        }
+    } else if (ret < 0) {
+        /* Apart from above checked -ve ret values, there are
+         * other possible ret values like ENOTCONN
+         * (returned when number of valid replies received are
+         * less than 2)
+         * in which case heal is required when one of the
+         * selfheal booleans is set.
+         */
+        if (data_selfheal || entry_selfheal || metadata_selfheal) {
+            ret = gf_asprintf(&status, "heal%s", substr ? substr : "");
+            if (ret < 0) {
+                goto out;
+            }
+            dict = afr_set_heal_info(status);
+            if (!dict) {
+                ret = -1;
+                goto out;
+            }
+        }
+    }
 
-                ret = afr_mark_locked_nodes (this, local->fd,
-                                             local->cont.lk.locked_nodes);
-                if (ret)
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Could not save locked nodes info in fdctx");
+    ret = 0;
+    op_errno = 0;
 
-                ret = afr_save_locked_fd (this, local->fd);
-                if (ret)
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Could not save locked fd");
+out:
+    if (heal_frame) {
+        heal_frame->local = heal_local;
+        AFR_STACK_DESTROY(heal_frame);
+    }
+    AFR_STACK_UNWIND(getxattr, frame, ret, op_errno, dict, NULL);
+    if (dict)
+        dict_unref(dict);
+    if (inode)
+        inode_unref(inode);
+    GF_FREE(substr);
+    return ret;
+}
 
-		AFR_STACK_UNWIND (lk, frame, local->op_ret, local->op_errno,
-                                  &local->cont.lk.ret_flock);
-        }
+int
+_afr_is_split_brain(call_frame_t *frame, xlator_t *this,
+                    struct afr_reply *replies, afr_transaction_type type,
+                    gf_boolean_t *spb)
+{
+    afr_private_t *priv = NULL;
+    uint64_t *witness = NULL;
+    unsigned char *sources = NULL;
+    unsigned char *sinks = NULL;
+    int sources_count = 0;
+    int ret = 0;
+
+    priv = this->private;
+
+    sources = alloca0(priv->child_count);
+    sinks = alloca0(priv->child_count);
+    witness = alloca0(priv->child_count * sizeof(*witness));
+
+    ret = afr_selfheal_find_direction(frame, this, replies, type,
+                                      priv->child_up, sources, sinks, witness,
+                                      NULL);
+    if (ret)
+        return ret;
 
-        return 0;
-}
+    sources_count = AFR_COUNT(sources, priv->child_count);
+    if (!sources_count)
+        *spb = _gf_true;
 
+    return ret;
+}
 
 int
-afr_lk (call_frame_t *frame, xlator_t *this,
-        fd_t *fd, int32_t cmd,
-        struct gf_flock *flock)
+afr_is_split_brain(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                   uuid_t gfid, gf_boolean_t *d_spb, gf_boolean_t *m_spb)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t *local = NULL;
+    int ret = -1;
+    afr_private_t *priv = NULL;
+    struct afr_reply *replies = NULL;
 
-        int i = 0;
+    priv = this->private;
 
-        int32_t op_ret   = -1;
-        int32_t op_errno = 0;
+    replies = alloca0(sizeof(*replies) * priv->child_count);
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    ret = afr_selfheal_unlocked_discover(frame, inode, gfid, replies);
+    if (ret)
+        goto out;
 
-        priv = this->private;
+    if (!afr_can_decide_split_brain_source_sinks(replies, priv->child_count)) {
+        ret = -EAGAIN;
+        goto out;
+    }
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
-        AFR_LOCAL_INIT (local, priv);
+    ret = _afr_is_split_brain(frame, this, replies, AFR_DATA_TRANSACTION,
+                              d_spb);
+    if (ret)
+        goto out;
 
-        frame->local  = local;
+    ret = _afr_is_split_brain(frame, this, replies, AFR_METADATA_TRANSACTION,
+                              m_spb);
+out:
+    if (replies) {
+        afr_replies_wipe(replies, priv->child_count);
+        replies = NULL;
+    }
+    return ret;
+}
 
-        local->cont.lk.locked_nodes = GF_CALLOC (priv->child_count,
-                                              sizeof (*local->cont.lk.locked_nodes),
-                                              gf_afr_mt_char);
+int
+afr_get_split_brain_status_cbk(int ret, call_frame_t *frame, void *opaque)
+{
+    GF_FREE(opaque);
+    return 0;
+}
 
-        if (!local->cont.lk.locked_nodes) {
-                gf_log (this->name, GF_LOG_ERROR, "Out of memory");
-                op_errno = ENOMEM;
-                goto out;
+int
+afr_get_split_brain_status(void *opaque)
+{
+    gf_boolean_t d_spb = _gf_false;
+    gf_boolean_t m_spb = _gf_false;
+    int ret = -1;
+    int op_errno = 0;
+    int i = 0;
+    char *choices = NULL;
+    char *status = NULL;
+    dict_t *dict = NULL;
+    inode_t *inode = NULL;
+    afr_private_t *priv = NULL;
+    xlator_t **children = NULL;
+    call_frame_t *frame = NULL;
+    xlator_t *this = NULL;
+    loc_t *loc = NULL;
+    afr_spb_status_t *data = NULL;
+
+    data = opaque;
+    frame = data->frame;
+    this = frame->this;
+    loc = data->loc;
+    priv = this->private;
+    children = priv->children;
+
+    inode = afr_inode_find(this, loc->gfid);
+    if (!inode)
+        goto out;
+
+    dict = dict_new();
+    if (!dict) {
+        op_errno = ENOMEM;
+        ret = -1;
+        goto out;
+    }
+
+    /* Calculation for string length :
+     * (child_count X length of child-name) + SLEN("    Choices :")
+     * child-name consists of :
+     * a) 251 = max characters for volname according to GD_VOLUME_NAME_MAX
+     * b) strlen("-client-00,") assuming 16 replicas
+     */
+    choices = alloca0(priv->child_count * (256 + SLEN("-client-00,")) +
+                      SLEN("    Choices:"));
+
+    ret = afr_is_split_brain(frame, this, inode, loc->gfid, &d_spb, &m_spb);
+    if (ret) {
+        op_errno = -ret;
+        if (ret == -EAGAIN) {
+            ret = dict_set_sizen_str_sizen(dict, GF_AFR_SBRAIN_STATUS,
+                                           SBRAIN_HEAL_NO_GO_MSG);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_WARNING, -ret,
+                       AFR_MSG_DICT_SET_FAILED,
+                       "Failed to set GF_AFR_SBRAIN_STATUS in dict");
+            }
+        }
+        ret = -1;
+        goto out;
+    }
+
+    if (d_spb || m_spb) {
+        sprintf(choices, "    Choices:");
+        for (i = 0; i < priv->child_count; i++) {
+            strcat(choices, children[i]->name);
+            strcat(choices, ",");
         }
+        choices[strlen(choices) - 1] = '\0';
 
-        local->fd            = fd_ref (fd);
-        local->cont.lk.cmd   = cmd;
-        local->cont.lk.user_flock = *flock;
-        local->cont.lk.ret_flock = *flock;
+        ret = gf_asprintf(&status,
+                          "data-split-brain:%s    "
+                          "metadata-split-brain:%s%s",
+                          (d_spb) ? "yes" : "no", (m_spb) ? "yes" : "no",
+                          choices);
 
-        STACK_WIND_COOKIE (frame, afr_lk_cbk, (void *) (long) 0,
-                           priv->children[i],
-                           priv->children[i]->fops->lk,
-                           fd, cmd, flock);
+        if (-1 == ret) {
+            op_errno = ENOMEM;
+            goto out;
+        }
+        ret = dict_set_dynstr_sizen(dict, GF_AFR_SBRAIN_STATUS, status);
+        if (ret) {
+            op_errno = -ret;
+            ret = -1;
+            goto out;
+        }
+    } else {
+        ret = dict_set_sizen_str_sizen(dict, GF_AFR_SBRAIN_STATUS,
+                                       SFILE_NOT_UNDER_DATA);
+        if (ret) {
+            op_errno = -ret;
+            ret = -1;
+            goto out;
+        }
+    }
 
-        op_ret = 0;
+    ret = 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (lk, frame, op_ret, op_errno, NULL);
+    AFR_STACK_UNWIND(getxattr, frame, ret, op_errno, dict, NULL);
+    if (dict)
+        dict_unref(dict);
+    if (inode)
+        inode_unref(inode);
+    return ret;
+}
+
+int32_t
+afr_heal_splitbrain_file(call_frame_t *frame, xlator_t *this, loc_t *loc)
+{
+    int ret = 0;
+    int op_errno = 0;
+    dict_t *dict = NULL;
+    afr_local_t *local = NULL;
+    afr_local_t *heal_local = NULL;
+    call_frame_t *heal_frame = NULL;
+
+    local = frame->local;
+    dict = dict_new();
+    if (!dict) {
+        op_errno = ENOMEM;
+        ret = -1;
+        goto out;
+    }
+
+    heal_frame = afr_frame_create(this, &op_errno);
+    if (!heal_frame) {
+        ret = -1;
+        goto out;
+    }
+    heal_local = heal_frame->local;
+    heal_frame->local = frame->local;
+    /*Initiate heal with heal_frame with lk-owner set so that inodelk/entrylk
+     * work correctly*/
+    ret = afr_selfheal_do(heal_frame, this, loc->gfid);
+
+    if (ret == 1 || ret == 2) {
+        ret = dict_set_sizen_str_sizen(dict, "sh-fail-msg",
+                                       SFILE_NOT_IN_SPLIT_BRAIN);
+        if (ret)
+            gf_msg(this->name, GF_LOG_WARNING, -ret, AFR_MSG_DICT_SET_FAILED,
+                   "Failed to set sh-fail-msg in dict");
+        ret = 0;
+        goto out;
+    } else {
+        if (local->xdata_rsp) {
+            /* 'sh-fail-msg' has been set in the dict during self-heal.*/
+            dict_copy(local->xdata_rsp, dict);
+            ret = 0;
+        } else if (ret < 0) {
+            op_errno = -ret;
+            ret = -1;
         }
-        return 0;
+    }
+
+out:
+    if (heal_frame) {
+        heal_frame->local = heal_local;
+        AFR_STACK_DESTROY(heal_frame);
+    }
+    if (local->op == GF_FOP_GETXATTR)
+        AFR_STACK_UNWIND(getxattr, frame, ret, op_errno, dict, NULL);
+    else if (local->op == GF_FOP_SETXATTR)
+        AFR_STACK_UNWIND(setxattr, frame, ret, op_errno, NULL);
+    if (dict)
+        dict_unref(dict);
+    return ret;
 }
 
 int
-afr_priv_dump (xlator_t *this)
+afr_get_child_index_from_name(xlator_t *this, char *name)
 {
-        afr_private_t *priv = NULL;
-        char  key_prefix[GF_DUMP_MAX_BUF_LEN];
-        char  key[GF_DUMP_MAX_BUF_LEN];
-        int   i = 0;
+    afr_private_t *priv = this->private;
+    int index = -1;
+
+    for (index = 0; index < priv->child_count; index++) {
+        if (!strcmp(priv->children[index]->name, name))
+            goto out;
+    }
+    index = -1;
+out:
+    return index;
+}
 
+void
+afr_priv_need_heal_set(afr_private_t *priv, gf_boolean_t need_heal)
+{
+    LOCK(&priv->lock);
+    {
+        priv->need_heal = need_heal;
+    }
+    UNLOCK(&priv->lock);
+}
 
-        assert(this);
-        priv = this->private;
+void
+afr_set_need_heal(xlator_t *this, afr_local_t *local)
+{
+    int i = 0;
+    afr_private_t *priv = this->private;
+    gf_boolean_t need_heal = _gf_false;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->replies[i].valid && local->replies[i].need_heal) {
+            need_heal = _gf_true;
+            break;
+        }
+    }
+    afr_priv_need_heal_set(priv, need_heal);
+    return;
+}
 
-        assert(priv);
-        snprintf(key_prefix, GF_DUMP_MAX_BUF_LEN, "%s.%s", this->type, this->name);
-        gf_proc_dump_add_section(key_prefix);
-        gf_proc_dump_build_key(key, key_prefix, "child_count");
-        gf_proc_dump_write(key, "%u", priv->child_count);
-        gf_proc_dump_build_key(key, key_prefix, "read_child_rr");
-        gf_proc_dump_write(key, "%u", priv->read_child_rr);
-        for (i = 0; i < priv->child_count; i++) {
-                gf_proc_dump_build_key(key, key_prefix, "child_up[%d]", i);
-                gf_proc_dump_write(key, "%d", priv->child_up[i]);
-                gf_proc_dump_build_key(key, key_prefix,
-                                        "pending_key[%d]", i);
-                gf_proc_dump_write(key, "%s", priv->pending_key[i]);
-        }
-        gf_proc_dump_build_key(key, key_prefix, "data_self_heal");
-        gf_proc_dump_write(key, "%d", priv->data_self_heal);
-        gf_proc_dump_build_key(key, key_prefix, "metadata_self_heal");
-        gf_proc_dump_write(key, "%d", priv->metadata_self_heal);
-        gf_proc_dump_build_key(key, key_prefix, "entry_self_heal");
-        gf_proc_dump_write(key, "%d", priv->entry_self_heal);
-        gf_proc_dump_build_key(key, key_prefix, "data_change_log");
-        gf_proc_dump_write(key, "%d", priv->data_change_log);
-        gf_proc_dump_build_key(key, key_prefix, "metadata_change_log");
-        gf_proc_dump_write(key, "%d", priv->metadata_change_log);
-        gf_proc_dump_build_key(key, key_prefix, "entry_change_log");
-        gf_proc_dump_write(key, "%d", priv->entry_change_log);
-        gf_proc_dump_build_key(key, key_prefix, "read_child");
-        gf_proc_dump_write(key, "%d", priv->read_child);
-        gf_proc_dump_build_key(key, key_prefix, "favorite_child");
-        gf_proc_dump_write(key, "%u", priv->favorite_child);
-        gf_proc_dump_build_key(key, key_prefix, "data_lock_server_count");
-        gf_proc_dump_write(key, "%u", priv->data_lock_server_count);
-        gf_proc_dump_build_key(key, key_prefix, "metadata_lock_server_count");
-        gf_proc_dump_write(key, "%u", priv->metadata_lock_server_count);
-        gf_proc_dump_build_key(key, key_prefix, "entry_lock_server_count");
-        gf_proc_dump_write(key, "%u", priv->entry_lock_server_count);
-        gf_proc_dump_build_key(key, key_prefix, "wait_count");
-        gf_proc_dump_write(key, "%u", priv->wait_count);
+gf_boolean_t
+afr_get_need_heal(xlator_t *this)
+{
+    afr_private_t *priv = this->private;
+    gf_boolean_t need_heal = _gf_true;
+
+    LOCK(&priv->lock);
+    {
+        need_heal = priv->need_heal;
+    }
+    UNLOCK(&priv->lock);
+    return need_heal;
+}
 
-        return 0;
+int
+afr_get_msg_id(char *op_type)
+{
+    if (!strcmp(op_type, GF_AFR_REPLACE_BRICK))
+        return AFR_MSG_REPLACE_BRICK_STATUS;
+    else if (!strcmp(op_type, GF_AFR_ADD_BRICK))
+        return AFR_MSG_ADD_BRICK_STATUS;
+    return -1;
 }
 
+int
+afr_fav_child_reset_sink_xattrs_cbk(int ret, call_frame_t *heal_frame,
+                                    void *opaque)
+{
+    call_frame_t *txn_frame = NULL;
+    afr_local_t *local = NULL;
+    afr_local_t *heal_local = NULL;
+    xlator_t *this = NULL;
 
-/**
- * find_child_index - find the child's index in the array of subvolumes
- * @this: AFR
- * @child: child
+    heal_local = heal_frame->local;
+    txn_frame = heal_local->heal_frame;
+    local = txn_frame->local;
+    this = txn_frame->this;
+
+    /* Refresh the inode agan and proceed with the transaction.*/
+    afr_inode_refresh(txn_frame, this, local->inode, NULL, local->refreshfn);
+
+    AFR_STACK_DESTROY(heal_frame);
+
+    return 0;
+}
+
+int
+afr_fav_child_reset_sink_xattrs(void *opaque)
+{
+    call_frame_t *heal_frame = NULL;
+    call_frame_t *txn_frame = NULL;
+    xlator_t *this = NULL;
+    gf_boolean_t d_spb = _gf_false;
+    gf_boolean_t m_spb = _gf_false;
+    afr_local_t *heal_local = NULL;
+    afr_local_t *txn_local = NULL;
+    afr_private_t *priv = NULL;
+    inode_t *inode = NULL;
+    unsigned char *locked_on = NULL;
+    unsigned char *sources = NULL;
+    unsigned char *sinks = NULL;
+    unsigned char *healed_sinks = NULL;
+    unsigned char *undid_pending = NULL;
+    struct afr_reply *locked_replies = NULL;
+    int ret = 0;
+
+    heal_frame = (call_frame_t *)opaque;
+    heal_local = heal_frame->local;
+    txn_frame = heal_local->heal_frame;
+    txn_local = txn_frame->local;
+    this = txn_frame->this;
+    inode = txn_local->inode;
+    priv = this->private;
+    locked_on = alloca0(priv->child_count);
+    sources = alloca0(priv->child_count);
+    sinks = alloca0(priv->child_count);
+    healed_sinks = alloca0(priv->child_count);
+    undid_pending = alloca0(priv->child_count);
+    locked_replies = alloca0(sizeof(*locked_replies) * priv->child_count);
+
+    ret = _afr_is_split_brain(txn_frame, this, txn_local->replies,
+                              AFR_DATA_TRANSACTION, &d_spb);
+
+    ret = _afr_is_split_brain(txn_frame, this, txn_local->replies,
+                              AFR_METADATA_TRANSACTION, &m_spb);
+
+    /* Take appropriate locks and reset sink xattrs. */
+    if (d_spb) {
+        ret = afr_selfheal_inodelk(heal_frame, this, inode, this->name, 0, 0,
+                                   locked_on);
+        {
+            if (ret < priv->child_count)
+                goto data_unlock;
+            ret = __afr_selfheal_data_prepare(
+                heal_frame, this, inode, locked_on, sources, sinks,
+                healed_sinks, undid_pending, locked_replies, NULL);
+        }
+    data_unlock:
+        afr_selfheal_uninodelk(heal_frame, this, inode, this->name, 0, 0,
+                               locked_on);
+    }
+
+    if (m_spb) {
+        memset(locked_on, 0, sizeof(*locked_on) * priv->child_count);
+        memset(undid_pending, 0, sizeof(*undid_pending) * priv->child_count);
+        ret = afr_selfheal_inodelk(heal_frame, this, inode, this->name,
+                                   LLONG_MAX - 1, 0, locked_on);
+        {
+            if (ret < priv->child_count)
+                goto mdata_unlock;
+            ret = __afr_selfheal_metadata_prepare(
+                heal_frame, this, inode, locked_on, sources, sinks,
+                healed_sinks, undid_pending, locked_replies, NULL);
+        }
+    mdata_unlock:
+        afr_selfheal_uninodelk(heal_frame, this, inode, this->name,
+                               LLONG_MAX - 1, 0, locked_on);
+    }
+
+    return ret;
+}
+
+/*
+ * Concatenates the xattrs in local->replies separated by a delimiter.
  */
+int
+afr_serialize_xattrs_with_delimiter(call_frame_t *frame, xlator_t *this,
+                                    char *buf, const char *default_str,
+                                    int32_t *serz_len, char delimiter)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    char *xattr = NULL;
+    int i = 0;
+    int len = 0;
+    int keylen = 0;
+    size_t str_len = 0;
+    int ret = -1;
+
+    priv = this->private;
+    local = frame->local;
+
+    keylen = strlen(local->cont.getxattr.name);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid || local->replies[i].op_ret) {
+            str_len = strlen(default_str);
+            buf = strncat(buf, default_str, str_len);
+            len += str_len;
+            buf[len++] = delimiter;
+            buf[len] = '\0';
+        } else {
+            ret = dict_get_strn(local->replies[i].xattr,
+                                local->cont.getxattr.name, keylen, &xattr);
+            if (ret) {
+                gf_msg("TEST", GF_LOG_ERROR, -ret, AFR_MSG_DICT_GET_FAILED,
+                       "Failed to get the node_uuid of brick "
+                       "%d",
+                       i);
+                goto out;
+            }
+            str_len = strlen(xattr);
+            buf = strncat(buf, xattr, str_len);
+            len += str_len;
+            buf[len++] = delimiter;
+            buf[len] = '\0';
+        }
+    }
+    buf[--len] = '\0'; /*remove the last delimiter*/
+    if (serz_len)
+        *serz_len = ++len;
+    ret = 0;
 
-static int
-find_child_index (xlator_t *this, xlator_t *child)
+out:
+    return ret;
+}
+
+uint64_t
+afr_write_subvol_get(call_frame_t *frame, xlator_t *this)
 {
-        afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    uint64_t write_subvol = 0;
 
-        int i = -1;
+    local = frame->local;
+    LOCK(&local->inode->lock);
+    write_subvol = local->inode_ctx->write_subvol;
+    UNLOCK(&local->inode->lock);
 
-        priv = this->private;
+    return write_subvol;
+}
 
-        for (i = 0; i < priv->child_count; i++) {
-                if ((xlator_t *) child == priv->children[i])
-                        break;
+int
+afr_write_subvol_set(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    unsigned char *data_accused = NULL;
+    unsigned char *metadata_accused = NULL;
+    unsigned char *data_readable = NULL;
+    unsigned char *metadata_readable = NULL;
+    uint16_t datamap = 0;
+    uint16_t metadatamap = 0;
+    uint64_t val = 0;
+    int event = 0;
+    int i = 0;
+
+    local = frame->local;
+    priv = this->private;
+    data_accused = alloca0(priv->child_count);
+    metadata_accused = alloca0(priv->child_count);
+    data_readable = alloca0(priv->child_count);
+    metadata_readable = alloca0(priv->child_count);
+    event = local->event_generation;
+
+    afr_readables_fill(frame, this, local->inode, data_accused,
+                       metadata_accused, data_readable, metadata_readable,
+                       NULL);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (data_readable[i])
+            datamap |= (1 << i);
+        if (metadata_readable[i])
+            metadatamap |= (1 << i);
+    }
+
+    val = ((uint64_t)metadatamap) | (((uint64_t)datamap) << 16) |
+          (((uint64_t)event) << 32);
+
+    LOCK(&local->inode->lock);
+    {
+        if (local->inode_ctx->write_subvol == 0 &&
+            local->transaction.type == AFR_DATA_TRANSACTION) {
+            local->inode_ctx->write_subvol = val;
         }
+    }
+    UNLOCK(&local->inode->lock);
 
-        return i;
+    return 0;
 }
 
-int32_t
-afr_notify (xlator_t *this, int32_t event,
-            void *data, ...)
+int
+afr_write_subvol_reset(call_frame_t *frame, xlator_t *this)
 {
-        afr_private_t *     priv     = NULL;
-        unsigned char *     child_up = NULL;
+    afr_local_t *local = NULL;
 
-        int i           = -1;
-        int up_children = 0;
+    local = frame->local;
+    LOCK(&local->inode->lock);
+    {
+        GF_ASSERT(local->inode_ctx->lock_count > 0);
+        local->inode_ctx->lock_count--;
 
-        priv = this->private;
+        if (!local->inode_ctx->lock_count)
+            local->inode_ctx->write_subvol = 0;
+    }
+    UNLOCK(&local->inode->lock);
 
-        if (!priv)
-                return 0;
+    return 0;
+}
 
-        child_up = priv->child_up;
+int
+afr_set_inode_local(xlator_t *this, afr_local_t *local, inode_t *inode)
+{
+    int ret = 0;
+
+    local->inode = inode_ref(inode);
+    LOCK(&local->inode->lock);
+    {
+        ret = __afr_inode_ctx_get(this, local->inode, &local->inode_ctx);
+    }
+    UNLOCK(&local->inode->lock);
+    if (ret < 0) {
+        gf_msg_callingfn(
+            this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_INODE_CTX_GET_FAILED,
+            "Error getting inode ctx %s", uuid_utoa(local->inode->gfid));
+    }
+    return ret;
+}
 
-        switch (event) {
-        case GF_EVENT_CHILD_UP:
-                i = find_child_index (this, data);
+gf_boolean_t
+afr_ta_is_fop_called_from_synctask(xlator_t *this)
+{
+    struct synctask *task = NULL;
+    gf_lkowner_t tmp_owner = {
+        0,
+    };
 
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "child=%d up", i);
+    task = synctask_get();
+    if (!task)
+        return _gf_false;
 
-                afr_attempt_lock_recovery (this, i);
+    set_lk_owner_from_ptr(&tmp_owner, (void *)this);
 
-                child_up[i] = 1;
+    if (!is_same_lkowner(&tmp_owner, &task->frame->root->lk_owner))
+        return _gf_false;
 
-                LOCK (&priv->lock);
-                {
-                        priv->up_count++;
-                }
-                UNLOCK (&priv->lock);
+    return _gf_true;
+}
 
-                /*
-                   if all the children were down, and one child came up,
-                   send notify to parent
-                */
+int
+afr_ta_post_op_lock(xlator_t *this, loc_t *loc)
+{
+    int ret = 0;
+    uuid_t gfid = {
+        0,
+    };
+    afr_private_t *priv = this->private;
+    gf_boolean_t locked = _gf_false;
+    struct gf_flock flock1 = {
+        0,
+    };
+    struct gf_flock flock2 = {
+        0,
+    };
+    int32_t cmd = 0;
+
+    /* Clients must take AFR_TA_DOM_NOTIFY lock only when the previous lock
+     * has been released in afr_notify due to upcall notification from shd.
+     */
+    GF_ASSERT(priv->ta_notify_dom_lock_offset == 0);
+
+    if (!priv->shd.iamshd)
+        GF_ASSERT(afr_ta_is_fop_called_from_synctask(this));
+    flock1.l_type = F_WRLCK;
+
+    while (!locked) {
+        if (priv->shd.iamshd) {
+            cmd = F_SETLKW;
+            flock1.l_start = 0;
+            flock1.l_len = 0;
+        } else {
+            cmd = F_SETLK;
+            gf_uuid_generate(gfid);
+            flock1.l_start = gfid_to_ino(gfid);
+            if (flock1.l_start < 0)
+                flock1.l_start = -flock1.l_start;
+            flock1.l_len = 1;
+        }
+        ret = syncop_inodelk(priv->children[THIN_ARBITER_BRICK_INDEX],
+                             AFR_TA_DOM_NOTIFY, loc, cmd, &flock1, NULL, NULL);
+        if (!ret) {
+            locked = _gf_true;
+            priv->ta_notify_dom_lock_offset = flock1.l_start;
+        } else if (ret == -EAGAIN) {
+            continue;
+        } else {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+                   "Failed to get "
+                   "AFR_TA_DOM_NOTIFY lock on %s.",
+                   loc->name);
+            goto out;
+        }
+    }
+
+    flock2.l_type = F_WRLCK;
+    flock2.l_start = 0;
+    flock2.l_len = 0;
+    ret = syncop_inodelk(priv->children[THIN_ARBITER_BRICK_INDEX],
+                         AFR_TA_DOM_MODIFY, loc, F_SETLKW, &flock2, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to get AFR_TA_DOM_MODIFY lock on %s.", loc->name);
+        flock1.l_type = F_UNLCK;
+        ret = syncop_inodelk(priv->children[THIN_ARBITER_BRICK_INDEX],
+                             AFR_TA_DOM_NOTIFY, loc, F_SETLK, &flock1, NULL,
+                             NULL);
+    }
+out:
+    return ret;
+}
 
-                for (i = 0; i < priv->child_count; i++)
-                        if (child_up[i])
-                                up_children++;
+int
+afr_ta_post_op_unlock(xlator_t *this, loc_t *loc)
+{
+    afr_private_t *priv = this->private;
+    struct gf_flock flock = {
+        0,
+    };
+    int ret = 0;
+
+    if (!priv->shd.iamshd)
+        GF_ASSERT(afr_ta_is_fop_called_from_synctask(this));
+    flock.l_type = F_UNLCK;
+    flock.l_start = 0;
+    flock.l_len = 0;
+
+    ret = syncop_inodelk(priv->children[THIN_ARBITER_BRICK_INDEX],
+                         AFR_TA_DOM_MODIFY, loc, F_SETLK, &flock, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to unlock AFR_TA_DOM_MODIFY lock.");
+        goto out;
+    }
+
+    if (!priv->shd.iamshd)
+        /* Mounts (clients) will not release the AFR_TA_DOM_NOTIFY lock
+         * in post-op as they use it as a notification mechanism. When
+         * shd sends a lock request on TA during heal, the clients will
+         * receive a lock-contention upcall notification upon which they
+         * will release the AFR_TA_DOM_NOTIFY lock after completing the
+         * in flight I/O.*/
+        goto out;
+
+    ret = syncop_inodelk(priv->children[THIN_ARBITER_BRICK_INDEX],
+                         AFR_TA_DOM_NOTIFY, loc, F_SETLK, &flock, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to unlock AFR_TA_DOM_NOTIFY lock.");
+    }
+out:
+    return ret;
+}
 
-                if (up_children == 1) {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "Subvolume '%s' came back up; "
-                                "going online.", ((xlator_t *)data)->name);
+call_frame_t *
+afr_ta_frame_create(xlator_t *this)
+{
+    call_frame_t *frame = NULL;
+    void *lk_owner = NULL;
+
+    frame = create_frame(this, this->ctx->pool);
+    if (!frame)
+        return NULL;
+    lk_owner = (void *)this;
+    afr_set_lk_owner(frame, this, lk_owner);
+    return frame;
+}
 
-                        default_notify (this, event, data);
-                }
+gf_boolean_t
+afr_ta_has_quorum(afr_private_t *priv, afr_local_t *local)
+{
+    int data_count = 0;
 
-                break;
+    data_count = AFR_COUNT(local->child_up, priv->child_count);
+    if (data_count == 2) {
+        return _gf_true;
+    } else if (data_count == 1 && local->ta_child_up) {
+        return _gf_true;
+    }
 
-        case GF_EVENT_CHILD_DOWN:
-                i = find_child_index (this, data);
+    return _gf_false;
+}
 
-                child_up[i] = 0;
+static gf_boolean_t
+afr_is_add_replica_mount_lookup_on_root(call_frame_t *frame)
+{
+    afr_local_t *local = NULL;
 
-                LOCK (&priv->lock);
-                {
-                        priv->down_count++;
-                }
-                UNLOCK (&priv->lock);
+    if (frame->root->pid != GF_CLIENT_PID_ADD_REPLICA_MOUNT)
+        return _gf_false;
 
-                /*
-                   if all children are down, and this was the last to go down,
-                   send notify to parent
-                */
+    local = frame->local;
 
-                for (i = 0; i < priv->child_count; i++)
-                        if (child_up[i])
-                                up_children++;
+    if (local->op != GF_FOP_LOOKUP)
+        /* TODO:If the replica count is being increased on a plain distribute
+         * volume that was never mounted, we need to allow setxattr on '/' with
+         * GF_CLIENT_PID_NO_ROOT_SQUASH to accomodate for DHT layout setting */
+        return _gf_false;
 
-                if (up_children == 0) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "All subvolumes are down. Going offline "
-                                "until atleast one of them comes back up.");
+    if (local->inode == NULL)
+        return _gf_false;
 
-                        default_notify (this, event, data);
-                }
+    if (!__is_root_gfid(local->inode->gfid))
+        return _gf_false;
 
-                break;
+    return _gf_true;
+}
 
-        default:
-                default_notify (this, event, data);
-        }
+gf_boolean_t
+afr_lookup_has_quorum(call_frame_t *frame, const unsigned int up_children_count)
+{
+    if (frame && (up_children_count > 0) &&
+        afr_is_add_replica_mount_lookup_on_root(frame))
+        return _gf_true;
 
-        return 0;
+    return _gf_false;
+}
+
+void
+afr_handle_replies_quorum(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = frame->local;
+    afr_private_t *priv = this->private;
+    unsigned char *success_replies = NULL;
+
+    success_replies = alloca0(priv->child_count);
+    afr_fill_success_replies(local, priv, success_replies);
+
+    if (priv->quorum_count && !afr_has_quorum(success_replies, this, NULL)) {
+        local->op_errno = afr_final_errno(local, priv);
+        if (!local->op_errno)
+            local->op_errno = afr_quorum_errno(priv);
+        local->op_ret = -1;
+    }
+}
+
+gf_boolean_t
+afr_ta_dict_contains_pending_xattr(dict_t *dict, afr_private_t *priv, int child)
+{
+    int *pending = NULL;
+    int ret = 0;
+    int i = 0;
+
+    ret = dict_get_ptr(dict, priv->pending_key[child], (void *)&pending);
+    if (ret == 0) {
+        for (i = 0; i < AFR_NUM_CHANGE_LOGS; i++) {
+            /* Not doing a ntoh32(pending) as we just want to check
+             * if it is non-zero or not. */
+            if (pending[i]) {
+                return _gf_true;
+            }
+        }
+    }
 
+    return _gf_false;
 }
diff --git a/xlators/cluster/afr/src/afr-dir-read.c b/xlators/cluster/afr/src/afr-dir-read.c
index 005f62c5719..f8bf8340dab 100644
--- a/xlators/cluster/afr/src/afr-dir-read.c
+++ b/xlators/cluster/afr/src/afr-dir-read.c
@@ -1,753 +1,346 @@
 /*
-   Copyright (c) 2007-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
 
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
 
 #include <libgen.h>
 #include <unistd.h>
-#include <fnmatch.h>
 #include <sys/time.h>
 #include <stdlib.h>
 #include <signal.h>
 #include <string.h>
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-#include "checksum.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/dict.h>
+#include <glusterfs/list.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/compat.h>
 
 #include "afr.h"
-#include "afr-self-heal.h"
-#include "afr-self-heal-common.h"
-
-
-int
-afr_examine_dir_sh_unwind (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local  = NULL;
-
-        local = frame->local;
-
-        afr_set_opendir_done (this, local->fd->inode);
-
-        AFR_STACK_UNWIND (opendir, frame, local->op_ret,
-                          local->op_errno, local->fd);
-
-        return 0;
-}
-
-
-gf_boolean_t
-__checksums_differ (uint32_t *checksum, int child_count,
-                    unsigned char *child_up)
-{
-        int ret = _gf_false;
-        int i = 0;
-
-        uint32_t cksum;
-
-        cksum = checksum[0];
-
-        for (i = 0; i < child_count; i++) {
-                if (!child_up[i])
-                        continue;
-
-                if (cksum != checksum[i]) {
-                        ret = _gf_true;
-                        break;
-                }
-
-                cksum = checksum[i];
-        }
-
-        return ret;
-}
-
+#include "afr-transaction.h"
 
 int32_t
-afr_examine_dir_readdir_cbk (call_frame_t *frame, void *cookie,
-                             xlator_t *this, int32_t op_ret, int32_t op_errno,
-                             gf_dirent_t *entries)
+afr_opendir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
 {
-	afr_private_t *    priv  = NULL;
-	afr_local_t *      local = NULL;
-        afr_self_heal_t *  sh    = NULL;
-
-        gf_dirent_t * entry = NULL;
-        gf_dirent_t * tmp   = NULL;
-
-        int child_index = 0;
-
-        uint32_t entry_cksum;
-
-        int   call_count    = 0;
-        off_t last_offset = 0;
-        char  sh_type_str[256] = {0,};
+    afr_local_t *local = NULL;
+    int call_count = -1;
+    int32_t child_index = 0;
+    afr_fd_ctx_t *fd_ctx = NULL;
 
-        priv  = this->private;
-        local = frame->local;
-        sh    = &local->self_heal;
+    local = frame->local;
+    fd_ctx = local->fd_ctx;
+    child_index = (long)cookie;
 
-        child_index = (long) cookie;
+    local->replies[child_index].valid = 1;
+    local->replies[child_index].op_ret = op_ret;
+    local->replies[child_index].op_errno = op_errno;
 
+    LOCK(&frame->lock);
+    {
         if (op_ret == -1) {
-                local->op_ret = -1;
-                local->op_ret = op_errno;
-                goto out;
+            local->op_errno = op_errno;
+            fd_ctx->opened_on[child_index] = AFR_FD_NOT_OPENED;
+        } else {
+            local->op_ret = op_ret;
+            fd_ctx->opened_on[child_index] = AFR_FD_OPENED;
+            if (!local->xdata_rsp && xdata)
+                local->xdata_rsp = dict_ref(xdata);
         }
+        call_count = --local->call_count;
+    }
+    UNLOCK(&frame->lock);
 
-        if (op_ret == 0)
-                goto out;
+    if (call_count == 0) {
+        afr_handle_replies_quorum(frame, this);
+        AFR_STACK_UNWIND(opendir, frame, local->op_ret, local->op_errno,
+                         local->fd, NULL);
+    }
 
-        list_for_each_entry_safe (entry, tmp, &entries->list, list) {
-                entry_cksum = gf_rsync_weak_checksum (entry->d_name,
-                                                      strlen (entry->d_name));
-                local->cont.opendir.checksum[child_index] ^= entry_cksum;
-        }
-
-        list_for_each_entry (entry, &entries->list, list) {
-                last_offset = entry->d_off;
-        }
-
-        /* read more entries */
-
-        STACK_WIND_COOKIE (frame, afr_examine_dir_readdir_cbk,
-                           (void *) (long) child_index,
-                           priv->children[child_index],
-                           priv->children[child_index]->fops->readdir,
-                           local->fd, 131072, last_offset);
-
-out:
-        if ((op_ret == 0) || (op_ret == -1)) {
-                call_count = afr_frame_return (frame);
-
-                if (call_count == 0) {
-                        if (__checksums_differ (local->cont.opendir.checksum,
-                                                priv->child_count,
-                                                local->child_up)) {
-
-                                sh->need_entry_self_heal  = _gf_true;
-                                sh->forced_merge          = _gf_true;
-                                sh->type                  = local->fd->inode->ia_type;
-                                sh->background            = _gf_false;
-                                sh->unwind                = afr_examine_dir_sh_unwind;
-
-                                afr_self_heal_type_str_get(&local->self_heal,
-                                                           sh_type_str,
-                                                           sizeof(sh_type_str));
-                                gf_log (this->name, GF_LOG_NORMAL,
-                                        "%s self-heal triggered. path: %s, "
-                                         "reason: checksums of directory differ,"
-                                        " forced merge option set",
-                                        sh_type_str, local->loc.path);
-
-                                afr_self_heal (frame, this);
-                        } else {
-                                afr_set_opendir_done (this, local->fd->inode);
-
-                                AFR_STACK_UNWIND (opendir, frame, local->op_ret,
-                                                  local->op_errno, local->fd);
-                        }
-                }
-        }
-
-        return 0;
+    return 0;
 }
 
-
 int
-afr_examine_dir (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t * priv  = NULL;
-        afr_local_t *   local = NULL;
-
-        int i;
-        int call_count = 0;
-
-        local = frame->local;
-        priv  = this->private;
-
-        local->cont.opendir.checksum = GF_CALLOC (priv->child_count,
-                                        sizeof (*local->cont.opendir.checksum),
-                                        gf_afr_mt_int32_t);
-
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-        local->call_count = call_count;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND_COOKIE (frame, afr_examine_dir_readdir_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->readdir,
-                                           local->fd, 131072, 0);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-        return 0;
-}
-
-
-int32_t
-afr_opendir_cbk (call_frame_t *frame, void *cookie,
-		 xlator_t *this, int32_t op_ret, int32_t op_errno,
-		 fd_t *fd)
-{
-        afr_private_t *priv              = NULL;
-	afr_local_t   *local             = NULL;
-        int32_t        up_children_count = 0;
-        int           ret                = -1;
-
-	int call_count = -1;
-
-        priv  = this->private;
-        local = frame->local;
-
-        up_children_count = afr_up_children_count (priv->child_count,
-                                                   local->child_up);
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret >= 0)
-			local->op_ret = op_ret;
-
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-                if (local->op_ret == 0) {
-
-                        ret = afr_fd_ctx_set (this, local->fd);
-
-                        if (ret) {
-                                local->op_ret = -1;
-                                local->op_errno = -1;
-                                gf_log (this->name, GF_LOG_ERROR, " failed to "
-                                        "set fd ctx for fd %d", local->fd);
-                                goto out;
-                        }
-                        if (!afr_is_opendir_done (this, local->fd->inode) &&
-                            up_children_count > 1) {
-
-                                /*
-                                 * This is the first opendir on this inode. We need
-                                 * to check if the directory's entries are the same
-                                 * on all subvolumes. This is needed in addition
-                                 * to regular entry self-heal because the readdir
-                                 * call is sent only to the first subvolume, and
-                                 * thus files that exist only there will never be healed
-                                 * otherwise (assuming changelog shows no anamolies).
-                                 */
-
-                                gf_log (this->name, GF_LOG_TRACE,
-                                        "reading contents of directory %s looking for mismatch",
-                                        local->loc.path);
-
-                                afr_examine_dir (frame, this);
-
-                        } else {
-                                AFR_STACK_UNWIND (opendir, frame, local->op_ret,
-                                                  local->op_errno, local->fd);
-                        }
-                } else {
-out:
-                        AFR_STACK_UNWIND (opendir, frame, local->op_ret,
-                                          local->op_errno, local->fd);
-                }
-        }
-
-	return 0;
-}
-
-
-int32_t 
-afr_opendir (call_frame_t *frame, xlator_t *this,
-	     loc_t *loc, fd_t *fd)
+afr_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+            dict_t *xdata)
 {
-	afr_private_t * priv        = NULL;
-	afr_local_t   * local       = NULL;
-
-	int             child_count = 0;
-	int             i           = 0;
-
-	int ret = -1;
-	int call_count = -1;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i = 0;
+    int call_count = -1;
+    int32_t op_errno = ENOMEM;
+    afr_fd_ctx_t *fd_ctx = NULL;
 
-	int32_t         op_ret   = -1;
-	int32_t         op_errno = 0;
+    priv = this->private;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-	priv = this->private;
+    local->op = GF_FOP_OPENDIR;
 
-	child_count = priv->child_count;
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-        loc_copy (&local->loc, loc);
-
-	frame->local = local;
-	local->fd    = fd_ref (fd);
-
-	call_count = local->call_count;
-	
-	for (i = 0; i < child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND (frame, afr_opendir_cbk, 
-				    priv->children[i],
-				    priv->children[i]->fops->opendir,
-				    loc, fd);
-
-			if (!--call_count)
-				break;
-		}
-	}
-
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (opendir, frame, op_ret, op_errno, fd);
-	}
+    if (priv->quorum_count && !afr_has_quorum(local->child_up, this, NULL)) {
+        op_errno = afr_quorum_errno(priv);
+        goto out;
+    }
 
-	return 0;
-}
+    if (!afr_is_consistent_io_possible(local, priv, &op_errno))
+        goto out;
 
+    fd_ctx = afr_fd_ctx_get(fd, this);
+    if (!fd_ctx)
+        goto out;
 
-/**
- * Common algorithm for directory read calls:
- * 
- * - Try the fop on the first child that is up
- * - if we have failed due to ENOTCONN:
- *     try the next child
- *
- * Applicable to: readdir
- */
+    loc_copy(&local->loc, loc);
 
+    local->fd = fd_ref(fd);
+    local->fd_ctx = fd_ctx;
 
-struct entry_name {
-        char *name;
-        struct list_head list;
-};
+    call_count = local->call_count;
 
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->child_up[i]) {
+            STACK_WIND_COOKIE(frame, afr_opendir_cbk, (void *)(long)i,
+                              priv->children[i],
+                              priv->children[i]->fops->opendir, loc, fd, NULL);
 
-static gf_boolean_t
-remembered_name (const char *name, struct list_head *entries)
-{
-        struct entry_name *e;
-        gf_boolean_t ret = _gf_false;
-
-        list_for_each_entry (e, entries, list) {
-                if (!strcmp (name, e->name)) {
-                        ret = _gf_true;
-                        goto out;
-                }
+            if (!--call_count)
+                break;
         }
+    }
 
+    return 0;
 out:
-        return ret;
+    AFR_STACK_UNWIND(opendir, frame, -1, op_errno, fd, NULL);
+    return 0;
 }
 
-
-static void
-afr_remember_entries (gf_dirent_t *entries, fd_t *fd)
+static int
+afr_validate_read_subvol(inode_t *inode, xlator_t *this, int par_read_subvol)
 {
-	struct entry_name *n       = NULL;
-	gf_dirent_t *      entry   = NULL;
-
-	int ret = 0;
-
-	uint64_t      ctx;
-	afr_fd_ctx_t *fd_ctx;
-
-	ret = fd_ctx_get (fd, THIS, &ctx);
-	if (ret < 0) {
-		gf_log (THIS->name, GF_LOG_DEBUG,
-			"could not get fd ctx for fd=%p", fd);
-		return;
-	}
-
-        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-	list_for_each_entry (entry, &entries->list, list) {
-		n = GF_CALLOC (1, sizeof (*n), gf_afr_mt_entry_name);
-		n->name = gf_strdup (entry->d_name);
-		INIT_LIST_HEAD (&n->list);
+    int gen = 0;
+    int entry_read_subvol = 0;
+    unsigned char *data_readable = NULL;
+    unsigned char *metadata_readable = NULL;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    data_readable = alloca0(priv->child_count);
+    metadata_readable = alloca0(priv->child_count);
+
+    afr_inode_read_subvol_get(inode, this, data_readable, metadata_readable,
+                              &gen);
+
+    if (gen != priv->event_generation || !data_readable[par_read_subvol] ||
+        !metadata_readable[par_read_subvol])
+        return -1;
+
+    /* Once the control reaches the following statement, it means that the
+     * parent's read subvol is perfectly readable. So calling
+     * either afr_data_subvol_get() or afr_metadata_subvol_get() would
+     * yield the same result. Hence, choosing afr_data_subvol_get() below.
+     */
+
+    if (!priv->consistent_metadata)
+        return 0;
 
-		list_add (&n->list, &fd_ctx->entries);
-	}
+    /* For an inode fetched through readdirp which is yet to be linked,
+     * inode ctx would not be initialised (yet). So this function returns
+     * -1 above due to gen being 0, which is why it is OK to pass NULL for
+     *  read_subvol_args here.
+     */
+    entry_read_subvol = afr_data_subvol_get(inode, this, NULL, NULL, NULL,
+                                            NULL);
+    if (entry_read_subvol != par_read_subvol)
+        return -1;
+
+    return 0;
 }
 
-
-static off_t
-afr_filter_entries (gf_dirent_t *entries, fd_t *fd)
+static void
+afr_readdir_transform_entries(call_frame_t *frame, gf_dirent_t *subvol_entries,
+                              int subvol, gf_dirent_t *entries, fd_t *fd)
 {
-	gf_dirent_t *entry, *tmp;
-	int ret = 0;
-
-	uint64_t      ctx;
-	afr_fd_ctx_t *fd_ctx;
-
-	off_t offset = 0;
-
-	ret = fd_ctx_get (fd, THIS, &ctx);
-	if (ret < 0) {
-		gf_log (THIS->name, GF_LOG_DEBUG,
-			"could not get fd ctx for fd=%p", fd);
-		return -1;
-	}
-
-        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-	list_for_each_entry_safe (entry, tmp, &entries->list, list) {
-		offset = entry->d_off;
-
-		if (remembered_name (entry->d_name, &fd_ctx->entries)) {
-			list_del (&entry->list);
-			GF_FREE (entry);
-		}
-	}
+    int ret = -1;
+    gf_dirent_t *entry = NULL;
+    gf_dirent_t *tmp = NULL;
+    xlator_t *this = NULL;
+    afr_private_t *priv = NULL;
+    gf_boolean_t need_heal = _gf_false;
+    gf_boolean_t validate_subvol = _gf_false;
+
+    this = THIS;
+    priv = this->private;
+
+    need_heal = afr_get_need_heal(this);
+    validate_subvol = need_heal | priv->consistent_metadata;
+
+    list_for_each_entry_safe(entry, tmp, &subvol_entries->list, list)
+    {
+        if (afr_is_private_directory(priv, fd->inode->gfid, entry->d_name,
+                                     frame->root->pid)) {
+            continue;
+        }
 
-	return offset;
-}
+        list_del_init(&entry->list);
+        list_add_tail(&entry->list, &entries->list);
 
+        if (!validate_subvol)
+            continue;
 
-static void
-afr_forget_entries (fd_t *fd)
-{
-	struct entry_name *entry, *tmp;
-	int ret = 0;
-
-	uint64_t      ctx;
-	afr_fd_ctx_t *fd_ctx;
-
-	ret = fd_ctx_get (fd, THIS, &ctx);
-	if (ret < 0) {
-		gf_log (THIS->name, GF_LOG_DEBUG,
-			"could not get fd ctx for fd=%p", fd);
-		return;
-	}
-
-        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-	list_for_each_entry_safe (entry, tmp, &fd_ctx->entries, list) {
-		GF_FREE (entry->name);
-		list_del (&entry->list);
-		GF_FREE (entry);
-	}
+        if (entry->inode) {
+            ret = afr_validate_read_subvol(entry->inode, this, subvol);
+            if (ret == -1) {
+                inode_unref(entry->inode);
+                entry->inode = NULL;
+                continue;
+            }
+        }
+    }
 }
 
-
 int32_t
-afr_readdir_cbk (call_frame_t *frame, void *cookie,
-		 xlator_t *this, int32_t op_ret, int32_t op_errno,
-		 gf_dirent_t *entries)
+afr_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, gf_dirent_t *subvol_entries,
+                dict_t *xdata)
 {
-	afr_private_t * priv     = NULL;
-	afr_local_t *   local    = NULL;
+    afr_local_t *local = NULL;
+    gf_dirent_t entries;
 
-        gf_dirent_t * entry = NULL;
-        gf_dirent_t * tmp   = NULL;
+    INIT_LIST_HEAD(&entries.list);
 
-        int child_index = -1;
+    local = frame->local;
 
-	priv     = this->private;
-	local = frame->local;
-        child_index = (long) cookie;
+    if (op_ret < 0 && !local->cont.readdir.offset) {
+        /* failover only if this was first readdir, detected
+           by offset == 0 */
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
 
-	if (op_ret != -1) {
-                list_for_each_entry_safe (entry, tmp, &entries->list, list) {
-                        entry->d_ino = afr_itransform (entry->d_ino,
-                                                       priv->child_count,
-                                                       child_index);
+        afr_read_txn_continue(frame, this, (long)cookie);
+        return 0;
+    }
 
-                        if ((local->fd->inode == local->fd->inode->table->root)
-                            && !strcmp (entry->d_name, GF_REPLICATE_TRASH_DIR)) {
-                                list_del_init (&entry->list);
-                                GF_FREE (entry);
-                        }
-                }
-    	}
+    if (op_ret >= 0)
+        afr_readdir_transform_entries(frame, subvol_entries, (long)cookie,
+                                      &entries, local->fd);
 
-        AFR_STACK_UNWIND (readdir, frame, op_ret, op_errno, entries);
+    AFR_STACK_UNWIND(readdir, frame, op_ret, op_errno, &entries, xdata);
 
-	return 0;
-}
+    gf_dirent_free(&entries);
 
+    return 0;
+}
 
-int32_t
-afr_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, gf_dirent_t *entries)
+int
+afr_readdir_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_private_t * priv     = NULL;
-        afr_local_t *   local    = NULL;
-        xlator_t **     children = NULL;
-        ino_t           inum = 0;
-
-        int call_child = 0;
-        int ret        = 0;
-
-        gf_dirent_t * entry = NULL;
-        gf_dirent_t * tmp   = NULL;
-
-        int child_index = -1;
-
-        uint64_t      ctx    = 0;
-        afr_fd_ctx_t *fd_ctx = NULL;
-
-	off_t offset = 0;
-
-        priv     = this->private;
-        children = priv->children;
-
-        local = frame->local;
-
-        child_index = (long) cookie;
-
-	if (priv->strict_readdir) {
-		ret = fd_ctx_get (local->fd, this, &ctx);
-		if (ret < 0) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"could not get fd ctx for fd=%p", local->fd);
-			op_ret   = -1;
-			op_errno = -ret;
-			goto out;
-		}
-
-		fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-		if (child_went_down (op_ret, op_errno)) {
-			if (all_tried (child_index, priv->child_count)) {
-				goto out;
-			}
-
-			call_child = ++child_index;
-
-			gf_log (this->name, GF_LOG_TRACE,
-				"starting readdir afresh on child %d, offset %"PRId64,
-				call_child, (uint64_t) 0);
-
-			fd_ctx->failed_over = _gf_true;
-
-			STACK_WIND_COOKIE (frame, afr_readdirp_cbk,
-					   (void *) (long) call_child,
-					   children[call_child],
-					   children[call_child]->fops->readdirp, local->fd,
-					   local->cont.readdir.size, 0);
-			return 0;
-		}
-	}
-
-	if (op_ret != -1) {
-		list_for_each_entry_safe (entry, tmp, &entries->list, list) {
-			inum = afr_itransform (entry->d_ino, priv->child_count,
-					       child_index);
-			entry->d_ino = inum;
-			inum  = afr_itransform (entry->d_stat.ia_ino,
-						priv->child_count, child_index);
-			entry->d_stat.ia_ino = inum;
-
-			if ((local->fd->inode == local->fd->inode->table->root)
-			    && !strcmp (entry->d_name, GF_REPLICATE_TRASH_DIR)) {
-				list_del_init (&entry->list);
-				GF_FREE (entry);
-			}
-		}
-	}
-
-	if (priv->strict_readdir) {
-		if (fd_ctx->failed_over) {
-			if (list_empty (&entries->list)) {
-				goto out;
-			}
-
-			offset = afr_filter_entries (entries, local->fd);
-
-			afr_remember_entries (entries, local->fd);
-
-			if (list_empty (&entries->list)) {
-				/* All the entries we got were duplicate. We
-				   shouldn't send an empty list now, because
-				   that'll make the application stop reading. So
-				   try to get more entries */
-
-				gf_log (this->name, GF_LOG_TRACE,
-					"trying to fetch non-duplicate entries from offset %"PRId64", child %s",
-					offset, children[child_index]->name);
-
-				STACK_WIND_COOKIE (frame, afr_readdirp_cbk,
-						   (void *) (long) child_index,
-						   children[child_index],
-						   children[child_index]->fops->readdirp,
-						   local->fd, local->cont.readdir.size, offset);
-				return 0;
-			}
-		} else {
-			afr_remember_entries (entries, local->fd);
-		}
-	}
-
-out:
-        AFR_STACK_UNWIND (readdirp, frame, op_ret, op_errno, entries);
-
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    afr_fd_ctx_t *fd_ctx = NULL;
+
+    priv = this->private;
+    local = frame->local;
+    fd_ctx = afr_fd_ctx_get(local->fd, this);
+    if (!fd_ctx) {
+        local->op_errno = EINVAL;
+        local->op_ret = -1;
+    }
+
+    if (subvol == -1 || !fd_ctx) {
+        AFR_STACK_UNWIND(readdir, frame, local->op_ret, local->op_errno, 0, 0);
         return 0;
+    }
+
+    fd_ctx->readdir_subvol = subvol;
+
+    if (local->op == GF_FOP_READDIR)
+        STACK_WIND_COOKIE(frame, afr_readdir_cbk, (void *)(long)subvol,
+                          priv->children[subvol],
+                          priv->children[subvol]->fops->readdir, local->fd,
+                          local->cont.readdir.size, local->cont.readdir.offset,
+                          local->xdata_req);
+    else
+        STACK_WIND_COOKIE(frame, afr_readdir_cbk, (void *)(long)subvol,
+                          priv->children[subvol],
+                          priv->children[subvol]->fops->readdirp, local->fd,
+                          local->cont.readdir.size, local->cont.readdir.offset,
+                          local->xdata_req);
+    return 0;
 }
 
-
-int32_t
-afr_do_readdir (call_frame_t *frame, xlator_t *this,
-	        fd_t *fd, size_t size, off_t offset, int whichop)
+int
+afr_do_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+               off_t offset, int whichop, dict_t *dict)
 {
-	afr_private_t * priv       = NULL;
-	xlator_t **     children   = NULL;
-	int             call_child = 0;
-	afr_local_t     *local     = NULL;
-
-        uint64_t      ctx;
-        afr_fd_ctx_t *fd_ctx;
-
-	int ret = -1;
-
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv     = this->private;
-	children = priv->children;
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-	frame->local = local;
-
-	call_child = afr_first_up_child (priv);
-	if (call_child == -1) {
-		op_errno = ENOTCONN;
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no child is up");
-		goto out;
-	}
-
-        local->fd                  = fd_ref (fd);
-        local->cont.readdir.size   = size;
-
-	if (priv->strict_readdir) {
-		ret = fd_ctx_get (fd, this, &ctx);
-		if (ret < 0) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"could not get fd ctx for fd=%p", fd);
-			op_errno = -ret;
-			goto out;
-		}
-
-		fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-		if (fd_ctx->last_tried != call_child) {
-			gf_log (this->name, GF_LOG_TRACE,
-				"first up child has changed from %d to %d, restarting readdir from offset 0",
-				fd_ctx->last_tried, call_child);
-
-			fd_ctx->failed_over = _gf_true;
-			offset = 0;
-		}
-
-		fd_ctx->last_tried = call_child;
-	}
-
-        if (whichop == GF_FOP_READDIR)
-                STACK_WIND_COOKIE (frame, afr_readdir_cbk,
-                                   (void *) (long) call_child,
-                                   children[call_child],
-                                   children[call_child]->fops->readdir, fd,
-                                   size, offset);
-        else
-                STACK_WIND_COOKIE (frame, afr_readdirp_cbk,
-                                   (void *) (long) call_child,
-                                   children[call_child],
-                                   children[call_child]->fops->readdirp, fd,
-                                   size, offset);
-
-	op_ret = 0;
+    afr_local_t *local = NULL;
+    int32_t op_errno = 0;
+    int subvol = -1;
+    afr_fd_ctx_t *fd_ctx = NULL;
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    fd_ctx = afr_fd_ctx_get(fd, this);
+    if (!fd_ctx) {
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    local->op = whichop;
+    local->fd = fd_ref(fd);
+    local->cont.readdir.size = size;
+    local->cont.readdir.offset = offset;
+    local->xdata_req = (dict) ? dict_ref(dict) : NULL;
+
+    subvol = fd_ctx->readdir_subvol;
+
+    if (offset == 0 || subvol == -1) {
+        /* First readdir has option of failing over and selecting
+           an appropriate read subvolume */
+        afr_read_txn(frame, this, fd->inode, afr_readdir_wind,
+                     AFR_DATA_TRANSACTION);
+    } else {
+        /* But continued readdirs MUST stick to the same subvolume
+           without an option to failover */
+        afr_readdir_wind(frame, this, subvol);
+    }
+
+    return 0;
 out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (readdir, frame, op_ret, op_errno, NULL);
-	}
-	return 0;
+    AFR_STACK_UNWIND(readdir, frame, -1, op_errno, NULL, NULL);
+    return 0;
 }
 
-
 int32_t
-afr_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-             off_t offset)
+afr_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t offset, dict_t *xdata)
 {
-        afr_do_readdir (frame, this, fd, size, offset, GF_FOP_READDIR);
-        return 0;
-}
+    afr_do_readdir(frame, this, fd, size, offset, GF_FOP_READDIR, xdata);
 
+    return 0;
+}
 
 int32_t
-afr_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-              off_t offset)
+afr_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+             off_t offset, dict_t *dict)
 {
-        afr_do_readdir (frame, this, fd, size, offset, GF_FOP_READDIRP);
-        return 0;
-}
+    afr_do_readdir(frame, this, fd, size, offset, GF_FOP_READDIRP, dict);
 
+    return 0;
+}
 
 int32_t
-afr_releasedir (xlator_t *this, fd_t *fd)
+afr_releasedir(xlator_t *this, fd_t *fd)
 {
-	afr_forget_entries (fd);
-        afr_cleanup_fd_ctx (this, fd);
+    afr_cleanup_fd_ctx(this, fd);
 
-	return 0;
+    return 0;
 }
diff --git a/xlators/cluster/afr/src/afr-dir-read.h b/xlators/cluster/afr/src/afr-dir-read.h
index e071107e62f..773e925ec6c 100644
--- a/xlators/cluster/afr/src/afr-dir-read.h
+++ b/xlators/cluster/afr/src/afr-dir-read.h
@@ -1,50 +1,33 @@
 /*
-   Copyright (c) 2007-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef __DIR_READ_H__
 #define __DIR_READ_H__
 
-
 int32_t
-afr_opendir (call_frame_t *frame, xlator_t *this,
-	     loc_t *loc, fd_t *fd);
+afr_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+            dict_t *xdata);
 
 int32_t
-afr_releasedir (xlator_t *this, fd_t *fd);
+afr_releasedir(xlator_t *this, fd_t *fd);
 
 int32_t
-afr_readdir (call_frame_t *frame, xlator_t *this,
-	     fd_t *fd, size_t size, off_t offset);
-
+afr_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t offset, dict_t *xdata);
 
 int32_t
-afr_readdirp (call_frame_t *frame, xlator_t *this,
-	     fd_t *fd, size_t size, off_t offset);
+afr_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+             off_t offset, dict_t *dict);
 
 int32_t
-afr_getdents (call_frame_t *frame, xlator_t *this,
-	      fd_t *fd, size_t size, off_t offset, int32_t flag);
-
-
-int32_t
-afr_checksum (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, int32_t flags);
-
+afr_checksum(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+             dict_t *xdata);
 
 #endif /* __DIR_READ_H__ */
diff --git a/xlators/cluster/afr/src/afr-dir-write.c b/xlators/cluster/afr/src/afr-dir-write.c
index 9f77324a32f..b7cceb79158 100644
--- a/xlators/cluster/afr/src/afr-dir-write.c
+++ b/xlators/cluster/afr/src/afr-dir-write.c
@@ -1,1349 +1,937 @@
 /*
-  Copyright (c) 2007-2009 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
-
 #include <libgen.h>
 #include <unistd.h>
-#include <fnmatch.h>
 #include <sys/time.h>
 #include <stdlib.h>
 #include <signal.h>
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
+#include <glusterfs/glusterfs.h>
 #include "afr.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
+#include <glusterfs/dict.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/list.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/byte-order.h>
 
 #include "afr.h"
 #include "afr-transaction.h"
 
-
 void
-afr_build_parent_loc (loc_t *parent, loc_t *child)
+afr_mark_entry_pending_changelog(call_frame_t *frame, xlator_t *this);
+
+int
+afr_build_parent_loc(loc_t *parent, loc_t *child, int32_t *op_errno)
 {
-	char *tmp = NULL;
+    int ret = -1;
+    char *child_path = NULL;
+
+    if (!child->parent) {
+        if (op_errno)
+            *op_errno = EINVAL;
+        goto out;
+    }
+
+    child_path = gf_strdup(child->path);
+    if (!child_path) {
+        if (op_errno)
+            *op_errno = ENOMEM;
+        goto out;
+    }
+
+    parent->path = gf_strdup(dirname(child_path));
+    if (!parent->path) {
+        if (op_errno)
+            *op_errno = ENOMEM;
+        goto out;
+    }
+
+    parent->inode = inode_ref(child->parent);
+    gf_uuid_copy(parent->gfid, child->pargfid);
+
+    ret = 0;
+out:
+    GF_FREE(child_path);
 
-	if (!child->parent) {
-		loc_copy (parent, child);
-		return;
-	}
+    return ret;
+}
 
-	tmp = gf_strdup (child->path);
-	parent->path   = gf_strdup (dirname (tmp));
-	GF_FREE (tmp);
+static void
+__afr_dir_write_finalize(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int inode_read_subvol = -1;
+    int parent_read_subvol = -1;
+    int parent2_read_subvol = -1;
+    int i = 0;
+    afr_read_subvol_args_t args = {
+        0,
+    };
+
+    local = frame->local;
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
+        if (local->replies[i].op_ret == -1)
+            continue;
+        gf_uuid_copy(args.gfid, local->replies[i].poststat.ia_gfid);
+        args.ia_type = local->replies[i].poststat.ia_type;
+        break;
+    }
+
+    if (local->inode) {
+        if (local->op != GF_FOP_RENAME && local->op != GF_FOP_LINK)
+            afr_replies_interpret(frame, this, local->inode, NULL);
+
+        inode_read_subvol = afr_data_subvol_get(local->inode, this, NULL, NULL,
+                                                NULL, &args);
+    }
+
+    if (local->parent)
+        parent_read_subvol = afr_data_subvol_get(local->parent, this, NULL,
+                                                 local->readable, NULL, NULL);
+
+    if (local->parent2)
+        parent2_read_subvol = afr_data_subvol_get(local->parent2, this, NULL,
+                                                  local->readable2, NULL, NULL);
+
+    local->op_ret = -1;
+    local->op_errno = afr_final_errno(local, priv);
+    afr_pick_error_xdata(local, priv, local->parent, local->readable,
+                         local->parent2, local->readable2);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
+        if (local->replies[i].op_ret < 0) {
+            if (local->inode)
+                afr_inode_need_refresh_set(local->inode, this);
+            if (local->parent)
+                afr_inode_need_refresh_set(local->parent, this);
+            if (local->parent2)
+                afr_inode_need_refresh_set(local->parent2, this);
+            continue;
+        }
 
-        parent->name   = strrchr (parent->path, '/');
-	if (parent->name)
-		parent->name++;
+        if (local->op_ret == -1) {
+            local->op_ret = local->replies[i].op_ret;
+            local->op_errno = local->replies[i].op_errno;
+
+            local->cont.dir_fop.buf = local->replies[i].poststat;
+            local->cont.dir_fop.preparent = local->replies[i].preparent;
+            local->cont.dir_fop.postparent = local->replies[i].postparent;
+            local->cont.dir_fop.prenewparent = local->replies[i].preparent2;
+            local->cont.dir_fop.postnewparent = local->replies[i].postparent2;
+            if (local->xdata_rsp) {
+                dict_unref(local->xdata_rsp);
+                local->xdata_rsp = NULL;
+            }
+
+            if (local->replies[i].xdata)
+                local->xdata_rsp = dict_ref(local->replies[i].xdata);
+            continue;
+        }
 
-	parent->inode  = inode_ref (child->parent);
-	parent->parent = inode_parent (parent->inode, 0, NULL);
-	parent->ino    = parent->inode->ino;
-}
+        if (i == inode_read_subvol) {
+            local->cont.dir_fop.buf = local->replies[i].poststat;
+            if (local->replies[i].xdata) {
+                if (local->xdata_rsp)
+                    dict_unref(local->xdata_rsp);
+                local->xdata_rsp = dict_ref(local->replies[i].xdata);
+            }
+        }
 
-/* {{{ create */
+        if (i == parent_read_subvol) {
+            local->cont.dir_fop.preparent = local->replies[i].preparent;
+            local->cont.dir_fop.postparent = local->replies[i].postparent;
+        }
 
-int
-afr_create_unwind (call_frame_t *frame, xlator_t *this)
-{
-	call_frame_t *main_frame = NULL;
-	afr_local_t  *local = NULL;
-        struct iatt  *unwind_buf = NULL;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame) {
-			main_frame = local->transaction.main_frame;
-		}
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
-
-	if (main_frame) {
-                if (local->cont.create.read_child_buf.ia_ino) {
-                        unwind_buf = &local->cont.create.read_child_buf;
-                } else {
-                        unwind_buf = &local->cont.create.buf;
-                }
-
-                unwind_buf->ia_ino = local->cont.create.ino;
-
-                local->cont.create.preparent.ia_ino  = local->cont.create.parent_ino;
-                local->cont.create.postparent.ia_ino = local->cont.create.parent_ino;
-
-		AFR_STACK_UNWIND (create, main_frame,
-                                  local->op_ret, local->op_errno,
-				  local->cont.create.fd,
-				  local->cont.create.inode,
-				  unwind_buf, &local->cont.create.preparent,
-                                  &local->cont.create.postparent);
+        if (i == parent2_read_subvol) {
+            local->cont.dir_fop.prenewparent = local->replies[i].preparent2;
+            local->cont.dir_fop.postnewparent = local->replies[i].postparent2;
         }
-        
-	return 0;
+    }
 }
 
-
-int
-afr_create_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		     int32_t op_ret, int32_t op_errno, 
-		     fd_t *fd, inode_t *inode, struct iatt *buf,
-                     struct iatt *preparent, struct iatt *postparent)
+static void
+__afr_dir_write_fill(call_frame_t *frame, xlator_t *this, int child_index,
+                     int op_ret, int op_errno, struct iatt *poststat,
+                     struct iatt *preparent, struct iatt *postparent,
+                     struct iatt *preparent2, struct iatt *postparent2,
+                     dict_t *xdata)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-
-        uint64_t      ctx;
-        afr_fd_ctx_t *fd_ctx;
-
-        int ret = 0;
-
-	int call_count = -1;
-	int child_index = -1;
-
-	local = frame->local;
-	priv  = this->private;
-
-	child_index = (long) cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (afr_fop_failed (op_ret, op_errno))
-			afr_transaction_fop_failed (frame, this, child_index);
-                
-		if (op_ret != -1) {
-			local->op_ret = op_ret;
-
-                        ret = afr_fd_ctx_set (this, fd);
-
-                        if (ret < 0) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "could not set ctx on fd=%p", fd);
-
-                                local->op_ret   = -1;
-                                local->op_errno = -ret;
-                                goto unlock;
-                        }
-
-                        ret = fd_ctx_get (fd, this, &ctx);
-
-                        if (ret < 0) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "could not get fd ctx for fd=%p", fd);
-                                local->op_ret   = -1;
-                                local->op_errno = -ret;
-                                goto unlock;
-                        }
-
-                        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-                        fd_ctx->opened_on[child_index] = 1;
-                        fd_ctx->flags                  = local->cont.create.flags;
-
-                        if (local->success_count == 0) {
-				local->cont.create.buf        = *buf;
-
-                                local->cont.create.ino =
-                                        afr_itransform (buf->ia_ino,
-                                                        priv->child_count,
-                                                        child_index);
-
-                                if (priv->read_child >= 0) {
-                                        afr_set_read_child (this, inode, 
-                                                            priv->read_child);
-                                } else {
-                                        afr_set_read_child (this, inode, 
-                                                            local->read_child_index);
-                                }
-			}
-                        
-                        if (child_index == local->first_up_child) {
-                                local->cont.create.ino =
-                                        afr_itransform (buf->ia_ino,
-                                                        priv->child_count,
-                                                        local->first_up_child);
-                        }
-                        
-                        if (child_index == local->read_child_index) {
-                                local->cont.create.read_child_buf = *buf;
-                                local->cont.create.preparent      = *preparent;
-                                local->cont.create.postparent     = *postparent;
-                        }
-
-			local->cont.create.inode = inode;
-
-			local->success_count++;
-		}
-
-		local->op_errno = op_errno;
-	}
-
-unlock:
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		local->transaction.unwind (frame, this);
-
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+    afr_local_t *local = NULL;
+    afr_fd_ctx_t *fd_ctx = NULL;
+
+    local = frame->local;
+    fd_ctx = local->fd_ctx;
+
+    local->replies[child_index].valid = 1;
+    local->replies[child_index].op_ret = op_ret;
+    local->replies[child_index].op_errno = op_errno;
+    if (xdata)
+        local->replies[child_index].xdata = dict_ref(xdata);
+
+    if (op_ret >= 0) {
+        if (poststat)
+            local->replies[child_index].poststat = *poststat;
+        if (preparent)
+            local->replies[child_index].preparent = *preparent;
+        if (postparent)
+            local->replies[child_index].postparent = *postparent;
+        if (preparent2)
+            local->replies[child_index].preparent2 = *preparent2;
+        if (postparent2)
+            local->replies[child_index].postparent2 = *postparent2;
+        if (fd_ctx)
+            fd_ctx->opened_on[child_index] = AFR_FD_OPENED;
+    } else {
+        if (op_errno != ENOTEMPTY)
+            afr_transaction_fop_failed(frame, this, child_index);
+        if (fd_ctx)
+            fd_ctx->opened_on[child_index] = AFR_FD_NOT_OPENED;
+    }
+
+    return;
 }
 
-
-int
-afr_create_wind (call_frame_t *frame, xlator_t *this)
+static int
+__afr_dir_write_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int op_ret, int op_errno, struct iatt *buf,
+                    struct iatt *preparent, struct iatt *postparent,
+                    struct iatt *preparent2, struct iatt *postparent2,
+                    dict_t *xdata)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-
-	int call_count = -1;
-	int i = 0;
-
-	local = frame->local;
-	priv = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_create_wind_cbk,
-					   (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->create,
-					   &local->loc, 
-					   local->cont.create.flags, 
-					   local->cont.create.mode, 
-					   local->cont.create.fd,
-                                           local->cont.create.params);
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
-}
+    afr_local_t *local = NULL;
+    int child_index = (long)cookie;
+    int call_count = -1;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    LOCK(&frame->lock);
+    {
+        __afr_dir_write_fill(frame, this, child_index, op_ret, op_errno, buf,
+                             preparent, postparent, preparent2, postparent2,
+                             xdata);
+        call_count = --local->call_count;
+    }
+    UNLOCK(&frame->lock);
+
+    if (call_count == 0) {
+        __afr_dir_write_finalize(frame, this);
+
+        if (afr_txn_nothing_failed(frame, this)) {
+            /*if it did pre-op, it will do post-op changing ctime*/
+            if (priv->consistent_metadata && afr_needs_changelog_update(local))
+                afr_zero_fill_stat(local);
+            local->transaction.unwind(frame, this);
+        }
+
+        afr_mark_entry_pending_changelog(frame, this);
 
+        afr_transaction_resume(frame, this);
+    }
+
+    return 0;
+}
 
 int
-afr_create_done (call_frame_t *frame, xlator_t *this)
+afr_mark_new_entry_changelog_cbk(call_frame_t *frame, void *cookie,
+                                 xlator_t *this, int op_ret, int op_errno,
+                                 dict_t *xattr, dict_t *xdata)
 {
-	afr_local_t * local = NULL;
+    int call_count = 0;
 
-	local = frame->local;
+    call_count = afr_frame_return(frame);
 
-	local->transaction.unwind (frame, this);
+    if (call_count == 0)
+        AFR_STACK_DESTROY(frame);
 
-	AFR_STACK_DESTROY (frame);
-
-	return 0;
+    return 0;
 }
 
-
-int
-afr_create (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc, int32_t flags, mode_t mode,
-            fd_t *fd, dict_t *params)
+void
+afr_mark_new_entry_changelog(call_frame_t *frame, xlator_t *this)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t  * transaction_frame = NULL;
-
-	int ret = -1;
+    call_frame_t *new_frame = NULL;
+    afr_local_t *local = NULL;
+    afr_local_t *new_local = NULL;
+    afr_private_t *priv = NULL;
+    dict_t *xattr = NULL;
+    int32_t **changelog = NULL;
+    int i = 0;
+    int op_errno = ENOMEM;
+    unsigned char *pending = NULL;
+    int call_count = 0;
+
+    local = frame->local;
+    priv = this->private;
+
+    new_frame = copy_frame(frame);
+    if (!new_frame)
+        goto out;
+
+    new_local = AFR_FRAME_INIT(new_frame, op_errno);
+    if (!new_local)
+        goto out;
+
+    xattr = dict_new();
+    if (!xattr)
+        goto out;
+
+    pending = alloca0(priv->child_count);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.pre_op[i] &&
+            !local->transaction.failed_subvols[i]) {
+            call_count++;
+            continue;
+        }
+        pending[i] = 1;
+    }
 
-	int op_ret   = -1;
-	int op_errno = 0;
+    changelog = afr_mark_pending_changelog(priv, pending, xattr,
+                                           local->cont.dir_fop.buf.ia_type);
+    if (!changelog)
+        goto out;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+    new_local->pending = changelog;
+    gf_uuid_copy(new_local->loc.gfid, local->cont.dir_fop.buf.ia_gfid);
+    new_local->loc.inode = inode_ref(local->inode);
 
-	priv = this->private;
+    new_local->call_count = call_count;
 
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		goto out;
-	}
+    for (i = 0; i < priv->child_count; i++) {
+        if (pending[i])
+            continue;
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+        STACK_WIND_COOKIE(new_frame, afr_mark_new_entry_changelog_cbk,
+                          (void *)(long)i, priv->children[i],
+                          priv->children[i]->fops->xattrop, &new_local->loc,
+                          GF_XATTROP_ADD_ARRAY, xattr, NULL);
+        if (!--call_count)
+            break;
+    }
 
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
+    new_frame = NULL;
+out:
+    if (new_frame)
+        AFR_STACK_DESTROY(new_frame);
+    if (xattr)
+        dict_unref(xattr);
+    return;
+}
 
-	transaction_frame->local = local;
+void
+afr_mark_entry_pending_changelog(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int pre_op_count = 0;
+    int failed_count = 0;
+    unsigned char *success_replies = NULL;
 
-	loc_copy (&local->loc, loc);
+    local = frame->local;
+    priv = this->private;
 
-        LOCK (&priv->read_child_lock);
-        {
-                local->read_child_index = (++priv->read_child_rr) 
-                        % (priv->child_count);
-        }
-        UNLOCK (&priv->read_child_lock);
+    if (local->op_ret < 0)
+        return;
 
-	local->cont.create.flags = flags;
-	local->cont.create.mode  = mode;
-	local->cont.create.fd    = fd_ref (fd);
-        if (params)
-                local->cont.create.params = dict_ref (params);
+    if (local->op != GF_FOP_CREATE && local->op != GF_FOP_MKNOD &&
+        local->op != GF_FOP_MKDIR)
+        return;
 
-        if (loc->parent)
-                local->cont.create.parent_ino = loc->parent->ino;
+    pre_op_count = AFR_COUNT(local->transaction.pre_op, priv->child_count);
+    failed_count = AFR_COUNT(local->transaction.failed_subvols,
+                             priv->child_count);
 
-	local->transaction.fop    = afr_create_wind;
-	local->transaction.done   = afr_create_done;
-	local->transaction.unwind = afr_create_unwind;
+    /* FOP succeeded on all bricks. */
+    if (pre_op_count == priv->child_count && !failed_count)
+        return;
 
-	afr_build_parent_loc (&local->transaction.parent_loc, loc);
+    /* FOP did not suceed on quorum no. of bricks. */
+    success_replies = alloca0(priv->child_count);
+    afr_fill_success_replies(local, priv, success_replies);
+    if (!afr_has_quorum(success_replies, this, NULL))
+        return;
 
-	local->transaction.main_frame = frame;
-	local->transaction.basename = AFR_BASENAME (loc->path);
+    if (priv->thin_arbiter_count) {
+        /*Mark new entry using ta file*/
+        local->is_new_entry = _gf_true;
+        return;
+    }
 
-	afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+    afr_mark_new_entry_changelog(frame, this);
 
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (create, frame, op_ret, op_errno,
-                                  NULL, NULL, NULL, NULL, NULL);
-	}
-
-	return 0;
+    return;
 }
 
-/* }}} */
-
-/* {{{ mknod */
+/* {{{ create */
 
 int
-afr_mknod_unwind (call_frame_t *frame, xlator_t *this)
+afr_create_unwind(call_frame_t *frame, xlator_t *this)
 {
-	call_frame_t *main_frame = NULL;
-	afr_local_t  *local = NULL;
-
-        struct iatt *unwind_buf = NULL;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame) {
-			main_frame = local->transaction.main_frame;
-		}
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
-
-	if (main_frame) {
-                if (local->cont.mknod.read_child_buf.ia_ino) {
-                        unwind_buf = &local->cont.mknod.read_child_buf;
-                } else {
-                        unwind_buf = &local->cont.mknod.buf;
-                }
-
-                unwind_buf->ia_ino = local->cont.mknod.ino;
-
-                local->cont.mknod.preparent.ia_ino  = local->cont.mknod.parent_ino;
-                local->cont.mknod.postparent.ia_ino = local->cont.mknod.parent_ino;
-
-		AFR_STACK_UNWIND (mknod, main_frame,
-                                  local->op_ret, local->op_errno,
-				  local->cont.mknod.inode,
-				  unwind_buf, &local->cont.mknod.preparent,
-                                  &local->cont.mknod.postparent);
-        }
+    call_frame_t *main_frame = NULL;
+    afr_local_t *local = NULL;
 
-	return 0;
-}
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+
+    if (!main_frame)
+        return 0;
+
+    AFR_STACK_UNWIND(create, main_frame, local->op_ret, local->op_errno,
+                     local->cont.create.fd, local->inode,
+                     &local->cont.dir_fop.buf, &local->cont.dir_fop.preparent,
+                     &local->cont.dir_fop.postparent, local->xdata_rsp);
+    return 0;
+}
 
 int
-afr_mknod_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, inode_t *inode,
+afr_create_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, fd_t *fd, inode_t *inode,
                     struct iatt *buf, struct iatt *preparent,
-                    struct iatt *postparent)
+                    struct iatt *postparent, dict_t *xdata)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-
-	int call_count = -1;
-	int child_index = -1;
-
-	local = frame->local;
-	priv = this->private;
-
-	child_index = (long) cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (afr_fop_failed (op_ret, op_errno))
-			afr_transaction_fop_failed (frame, this, child_index);
-		
-		if (op_ret != -1) {
-			local->op_ret = op_ret;
-
-			if (local->success_count == 0){
-				local->cont.mknod.buf   = *buf;
-				local->cont.mknod.ino   = 
-					afr_itransform (buf->ia_ino,
-							priv->child_count,
-							child_index);
-
-                                if (priv->read_child >= 0) {
-                                        afr_set_read_child (this, inode,
-                                                            priv->read_child);
-                                } else {
-                                        afr_set_read_child (this, inode,
-                                                            local->read_child_index);
-                                }
-			}
-
-                        if (child_index == local->first_up_child) {
-                                local->cont.mknod.ino =
-                                        afr_itransform (buf->ia_ino,
-                                                        priv->child_count,
-                                                        local->first_up_child);
-                        }
-
-                        if (child_index == local->read_child_index) {
-                                local->cont.mknod.read_child_buf = *buf;
-                                local->cont.mknod.preparent      = *preparent;
-                                local->cont.mknod.postparent     = *postparent;
-                        }
-                        
-			local->cont.mknod.inode = inode;
-
-			local->success_count++;
-		}
-
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		local->transaction.unwind (frame, this);
-
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+    return __afr_dir_write_cbk(frame, cookie, this, op_ret, op_errno, buf,
+                               preparent, postparent, NULL, NULL, xdata);
 }
 
-
-int32_t
-afr_mknod_wind (call_frame_t *frame, xlator_t *this)
+int
+afr_create_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-
-	int call_count = -1;
-	int i = 0;
-
-	local = frame->local;
-	priv  = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_mknod_wind_cbk, (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->mknod,
-					   &local->loc, local->cont.mknod.mode,
-					   local->cont.mknod.dev,
-                                           local->cont.mknod.params);
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    STACK_WIND_COOKIE(frame, afr_create_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->create, &local->loc,
+                      local->cont.create.flags, local->cont.create.mode,
+                      local->umask, local->cont.create.fd, local->xdata_req);
+    return 0;
 }
 
-
 int
-afr_mknod_done (call_frame_t *frame, xlator_t *this)
+afr_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+           mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
 {
-	afr_local_t * local = NULL;
-
-	local = frame->local;
-
-	local->transaction.unwind (frame, this);
-	AFR_STACK_DESTROY (frame);
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
+
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
+
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
+
+    loc_copy(&local->loc, loc);
+
+    local->fd_ctx = afr_fd_ctx_get(fd, this);
+    if (!local->fd_ctx)
+        goto out;
+
+    local->inode = inode_ref(loc->inode);
+    local->parent = inode_ref(loc->parent);
+
+    local->op = GF_FOP_CREATE;
+    local->cont.create.flags = flags;
+    local->fd_ctx->flags = flags;
+    local->cont.create.mode = mode;
+    local->cont.create.fd = fd_ref(fd);
+    local->umask = umask;
+
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
+
+    if (!local->xdata_req)
+        goto out;
+
+    local->transaction.wind = afr_create_wind;
+    local->transaction.unwind = afr_create_unwind;
+
+    ret = afr_build_parent_loc(&local->transaction.parent_loc, loc, &op_errno);
+    if (ret)
+        goto out;
+
+    local->transaction.main_frame = frame;
+    local->transaction.basename = AFR_BASENAME(loc->path);
+    ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
+out:
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-	return 0;
+    AFR_STACK_UNWIND(create, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL,
+                     NULL);
+    return 0;
 }
 
+/* }}} */
+
+/* {{{ mknod */
 
 int
-afr_mknod (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, mode_t mode, dev_t dev, dict_t *params)
+afr_mknod_unwind(call_frame_t *frame, xlator_t *this)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t  * transaction_frame = NULL;
-
-	int ret = -1;
-
-	int op_ret   = -1;
-	int op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv = this->private;
-
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		goto out;
-	}
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+    call_frame_t *main_frame = NULL;
+    afr_local_t *local = NULL;
 
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
+    local = frame->local;
 
-	transaction_frame->local = local;
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-	loc_copy (&local->loc, loc);
-
-        LOCK (&priv->read_child_lock);
-        {
-                local->read_child_index = (++priv->read_child_rr) 
-                        % (priv->child_count);
-        }
-        UNLOCK (&priv->read_child_lock);
-
-	local->cont.mknod.mode  = mode;
-	local->cont.mknod.dev   = dev;
-        if (params)
-                local->cont.mknod.params = dict_ref (params);
-
-        if (loc->parent)
-                local->cont.mknod.parent_ino = loc->parent->ino;
-
-	local->transaction.fop    = afr_mknod_wind;
-	local->transaction.done   = afr_mknod_done;
-	local->transaction.unwind = afr_mknod_unwind;
-
-	afr_build_parent_loc (&local->transaction.parent_loc, loc);
+    AFR_STACK_UNWIND(mknod, main_frame, local->op_ret, local->op_errno,
+                     local->inode, &local->cont.dir_fop.buf,
+                     &local->cont.dir_fop.preparent,
+                     &local->cont.dir_fop.postparent, local->xdata_rsp);
+    return 0;
+}
 
-	local->transaction.main_frame = frame;
-	local->transaction.basename = AFR_BASENAME (loc->path);
+int
+afr_mknod_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, inode_t *inode,
+                   struct iatt *buf, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
+{
+    return __afr_dir_write_cbk(frame, cookie, this, op_ret, op_errno, buf,
+                               preparent, postparent, NULL, NULL, xdata);
+}
 
-	afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
+int
+afr_mknod_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    STACK_WIND_COOKIE(frame, afr_mknod_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->mknod, &local->loc,
+                      local->cont.mknod.mode, local->cont.mknod.dev,
+                      local->umask, local->xdata_req);
+    return 0;
+}
 
-	op_ret = 0;
+int
+afr_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+          dev_t dev, mode_t umask, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
+
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
+
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
+
+    loc_copy(&local->loc, loc);
+    local->inode = inode_ref(loc->inode);
+    local->parent = inode_ref(loc->parent);
+
+    local->op = GF_FOP_MKNOD;
+    local->cont.mknod.mode = mode;
+    local->cont.mknod.dev = dev;
+    local->umask = umask;
+
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
+
+    if (!local->xdata_req)
+        goto out;
+
+    local->transaction.wind = afr_mknod_wind;
+    local->transaction.unwind = afr_mknod_unwind;
+
+    ret = afr_build_parent_loc(&local->transaction.parent_loc, loc, &op_errno);
+    if (ret)
+        goto out;
+
+    local->transaction.main_frame = frame;
+    local->transaction.basename = AFR_BASENAME(loc->path);
+    ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (mknod, frame, op_ret, op_errno,
-                                  NULL, NULL, NULL, NULL);
-	}
-
-	return 0;
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
+
+    AFR_STACK_UNWIND(mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+    return 0;
 }
 
 /* }}} */
 
 /* {{{ mkdir */
 
-
 int
-afr_mkdir_unwind (call_frame_t *frame, xlator_t *this)
+afr_mkdir_unwind(call_frame_t *frame, xlator_t *this)
 {
-	call_frame_t *main_frame = NULL;
-	afr_local_t  *local = NULL;
-
-        struct iatt *unwind_buf = NULL;
-        
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame) {
-			main_frame = local->transaction.main_frame;
-		}
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
-
-	if (main_frame) {
-                if (local->cont.mkdir.read_child_buf.ia_ino) {
-                        unwind_buf = &local->cont.mkdir.read_child_buf;
-                } else {
-                        unwind_buf = &local->cont.mkdir.buf;
-                }
-
-                unwind_buf->ia_ino = local->cont.mkdir.ino;
-
-                local->cont.mkdir.preparent.ia_ino  = local->cont.mkdir.parent_ino;
-                local->cont.mkdir.postparent.ia_ino = local->cont.mkdir.parent_ino;
-
-		AFR_STACK_UNWIND (mkdir, main_frame,
-                                  local->op_ret, local->op_errno,
-				  local->cont.mkdir.inode,
-				  unwind_buf, &local->cont.mkdir.preparent,
-                        &local->cont.mkdir.postparent);
-        }
+    call_frame_t *main_frame = NULL;
+    afr_local_t *local = NULL;
 
-	return 0;
-}
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-int
-afr_mkdir_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, inode_t *inode,
-                    struct iatt *buf, struct iatt *preparent,
-                    struct iatt *postparent)
-{
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-
-	int call_count = -1;
-	int child_index = -1;
-
-	local = frame->local;
-	priv = this->private;
-
-	child_index = (long) cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (afr_fop_failed (op_ret, op_errno))
-			afr_transaction_fop_failed (frame, this, child_index);
-
-		if (op_ret != -1) {
-			local->op_ret           = op_ret;
-
-			if (local->success_count == 0) {
-				local->cont.mkdir.buf   = *buf;
-
-                                local->cont.mkdir.ino =
-                                        afr_itransform (buf->ia_ino,
-                                                        priv->child_count,
-                                                        child_index);
-
-                                if (priv->read_child >= 0) {
-                                        afr_set_read_child (this, inode,
-                                                            priv->read_child);
-                                } else {
-                                        afr_set_read_child (this, inode,
-                                                            local->read_child_index);
-                                }
-			}
-
-                        if (child_index == local->first_up_child) {
-                                local->cont.mkdir.ino =
-                                        afr_itransform (buf->ia_ino,
-                                                        priv->child_count,
-                                                        local->first_up_child);
-                        }
-                        
-                        if (child_index == local->read_child_index) {
-                                local->cont.mkdir.read_child_buf = *buf;
-                                local->cont.mkdir.preparent      = *preparent;
-                                local->cont.mkdir.postparent     = *postparent;
-                        }
-
-			local->cont.mkdir.inode = inode;
-
-			local->success_count++;
-		}
-
-		local->op_errno         = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		local->transaction.unwind (frame, this);
-
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+    AFR_STACK_UNWIND(mkdir, main_frame, local->op_ret, local->op_errno,
+                     local->inode, &local->cont.dir_fop.buf,
+                     &local->cont.dir_fop.preparent,
+                     &local->cont.dir_fop.postparent, local->xdata_rsp);
+    return 0;
 }
 
-
 int
-afr_mkdir_wind (call_frame_t *frame, xlator_t *this)
+afr_mkdir_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, inode_t *inode,
+                   struct iatt *buf, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-	
-	int call_count = -1;
-	int i = 0;
-
-	local = frame->local;
-	priv  = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_mkdir_wind_cbk,
-					   (void *) (long) i,	
-					   priv->children[i], 
-					   priv->children[i]->fops->mkdir,
-					   &local->loc, local->cont.mkdir.mode,
-                                           local->cont.mkdir.params);
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
+    return __afr_dir_write_cbk(frame, cookie, this, op_ret, op_errno, buf,
+                               preparent, postparent, NULL, NULL, xdata);
 }
 
-
 int
-afr_mkdir_done (call_frame_t *frame, xlator_t *this)
+afr_mkdir_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-	afr_local_t * local = NULL;
-
-	local = frame->local;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-	local->transaction.unwind (frame, this);
+    local = frame->local;
+    priv = this->private;
 
-	AFR_STACK_DESTROY (frame);
-
-	return 0;
+    STACK_WIND_COOKIE(frame, afr_mkdir_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->mkdir, &local->loc,
+                      local->cont.mkdir.mode, local->umask, local->xdata_req);
+    return 0;
 }
 
-
 int
-afr_mkdir (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, mode_t mode, dict_t *params)
+afr_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+          mode_t umask, dict_t *xdata)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t  * transaction_frame = NULL;
-
-	int ret = -1;
-
-	int op_ret   = -1;
-	int op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv = this->private;
-
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		goto out;
-	}
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-	transaction_frame->local = local;
-
-	loc_copy (&local->loc, loc);
-
-        LOCK (&priv->read_child_lock);
-        {
-                local->read_child_index = (++priv->read_child_rr) 
-                        % (priv->child_count);
-        }
-        UNLOCK (&priv->read_child_lock);
-
-	local->cont.mkdir.mode  = mode;
-        if (params)
-                local->cont.mkdir.params = dict_ref (params);
-
-        if (loc->parent)
-                local->cont.mkdir.parent_ino = loc->parent->ino;
-
-	local->transaction.fop    = afr_mkdir_wind;
-	local->transaction.done   = afr_mkdir_done;
-	local->transaction.unwind = afr_mkdir_unwind;
-
-	afr_build_parent_loc (&local->transaction.parent_loc, loc);
-
-	local->transaction.main_frame = frame;
-	local->transaction.basename = AFR_BASENAME (loc->path);
-
-	afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
-
-	op_ret = 0;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
+
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
+
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
+
+    loc_copy(&local->loc, loc);
+    local->inode = inode_ref(loc->inode);
+    local->parent = inode_ref(loc->parent);
+
+    local->cont.mkdir.mode = mode;
+    local->umask = umask;
+
+    if (!xdata || !dict_get_sizen(xdata, "gfid-req")) {
+        op_errno = EPERM;
+        gf_msg_callingfn(this->name, GF_LOG_WARNING, op_errno,
+                         AFR_MSG_GFID_NULL,
+                         "mkdir: %s is received "
+                         "without gfid-req %p",
+                         loc->path, xdata);
+        goto out;
+    }
+
+    local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    if (!local->xdata_req) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    local->op = GF_FOP_MKDIR;
+    local->transaction.wind = afr_mkdir_wind;
+    local->transaction.unwind = afr_mkdir_unwind;
+
+    ret = afr_build_parent_loc(&local->transaction.parent_loc, loc, &op_errno);
+    if (ret)
+        goto out;
+
+    local->transaction.main_frame = frame;
+    local->transaction.basename = AFR_BASENAME(loc->path);
+    ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-
-		AFR_STACK_UNWIND (mkdir, frame, op_ret, op_errno,
-                                  NULL, NULL, NULL, NULL);
-	}
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-	return 0;
+    AFR_STACK_UNWIND(mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+    return 0;
 }
 
 /* }}} */
 
 /* {{{ link */
 
-
-int
-afr_link_unwind (call_frame_t *frame, xlator_t *this)
-{
-	call_frame_t *main_frame = NULL;
-	afr_local_t  *local = NULL;
-
-        struct iatt *unwind_buf = NULL;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame) {
-			main_frame = local->transaction.main_frame;
-		}
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
-
-	if (main_frame) {
-                if (local->cont.link.read_child_buf.ia_ino) {
-                        unwind_buf = &local->cont.link.read_child_buf;
-                } else {
-                        unwind_buf = &local->cont.link.buf;
-                }
-
-		unwind_buf->ia_ino = local->cont.link.ino;
-
-                local->cont.link.preparent.ia_ino  = local->cont.link.parent_ino;
-                local->cont.link.postparent.ia_ino = local->cont.link.parent_ino;
-
-		AFR_STACK_UNWIND (link, main_frame,
-                                  local->op_ret, local->op_errno, 
-				  local->cont.link.inode,
-				  unwind_buf, &local->cont.link.preparent,
-                                  &local->cont.link.postparent);
-	}
-
-	return 0;
-}
-
-
 int
-afr_link_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		   int32_t op_ret, int32_t op_errno, inode_t *inode,
-                   struct iatt *buf, struct iatt *preparent,
-                   struct iatt *postparent)
+afr_link_unwind(call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-
-	int call_count = -1;
-	int child_index = -1;
-
-	local = frame->local;
-	priv = this->private;
-
-	child_index = (long) cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (afr_fop_failed (op_ret, op_errno))
-			afr_transaction_fop_failed (frame, this, child_index);
-
-		if (op_ret != -1) {
-			local->op_ret   = op_ret;
-
-			if (local->success_count == 0) {
-				local->cont.link.buf        = *buf;
-
-                                if (priv->read_child >= 0) {
-                                        afr_set_read_child (this, inode,
-                                                            priv->read_child);
-                                } else {
-                                        afr_set_read_child (this, inode,
-                                                            local->read_child_index);
-                                }
-			}
-
-                        if (child_index == local->read_child_index) {
-                                local->cont.link.read_child_buf = *buf;
-                                local->cont.link.preparent      = *preparent;
-                                local->cont.link.postparent     = *postparent;
-                        }
+    call_frame_t *main_frame = NULL;
+    afr_local_t *local = NULL;
 
-			local->cont.link.inode    = inode;
+    local = frame->local;
 
-			local->success_count++;
-		}
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-		local->op_errno = op_errno;		
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		local->transaction.unwind (frame, this);
-
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+    AFR_STACK_UNWIND(link, main_frame, local->op_ret, local->op_errno,
+                     local->inode, &local->cont.dir_fop.buf,
+                     &local->cont.dir_fop.preparent,
+                     &local->cont.dir_fop.postparent, local->xdata_rsp);
+    return 0;
 }
 
-
 int
-afr_link_wind (call_frame_t *frame, xlator_t *this)
+afr_link_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, inode_t *inode,
+                  struct iatt *buf, struct iatt *preparent,
+                  struct iatt *postparent, dict_t *xdata)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-
-	int call_count = -1;
-	int i = 0;
-
-	local = frame->local;
-	priv  = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_link_wind_cbk, (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->link,
-					   &local->loc,
-					   &local->newloc);
-			
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
+    return __afr_dir_write_cbk(frame, cookie, this, op_ret, op_errno, buf,
+                               preparent, postparent, NULL, NULL, xdata);
 }
 
-
 int
-afr_link_done (call_frame_t *frame, xlator_t *this)
+afr_link_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-	afr_local_t * local = frame->local;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-	local->transaction.unwind (frame, this);
+    local = frame->local;
+    priv = this->private;
 
-	AFR_STACK_DESTROY (frame);
-
-	return 0;
+    STACK_WIND_COOKIE(frame, afr_link_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->link, &local->loc,
+                      &local->newloc, local->xdata_req);
+    return 0;
 }
 
-
 int
-afr_link (call_frame_t *frame, xlator_t *this,
-	  loc_t *oldloc, loc_t *newloc)
+afr_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+         dict_t *xdata)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t  * transaction_frame = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-	int ret = -1;
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-	int op_ret   = -1;
-	int op_errno = 0;
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+    loc_copy(&local->loc, oldloc);
+    loc_copy(&local->newloc, newloc);
 
-	priv = this->private;
+    local->inode = inode_ref(oldloc->inode);
+    local->parent = inode_ref(newloc->parent);
 
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		goto out;
-	}
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+    if (!local->xdata_req)
+        goto out;
 
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
+    local->op = GF_FOP_LINK;
 
-	transaction_frame->local = local;
+    local->transaction.wind = afr_link_wind;
+    local->transaction.unwind = afr_link_unwind;
 
-	loc_copy (&local->loc,    oldloc);
-	loc_copy (&local->newloc, newloc);
-
-        LOCK (&priv->read_child_lock);
-        {
-                local->read_child_index = (++priv->read_child_rr) 
-                        % (priv->child_count);
-        }
-        UNLOCK (&priv->read_child_lock);
+    ret = afr_build_parent_loc(&local->transaction.parent_loc, newloc,
+                               &op_errno);
+    if (ret)
+        goto out;
 
-	local->cont.link.ino = oldloc->inode->ino;
+    local->transaction.main_frame = frame;
+    local->transaction.basename = AFR_BASENAME(newloc->path);
+    ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        if (oldloc->parent)
-                local->cont.link.parent_ino = newloc->parent->ino;
-
-	local->transaction.fop    = afr_link_wind;
-	local->transaction.done   = afr_link_done;
-	local->transaction.unwind = afr_link_unwind;
-
-	afr_build_parent_loc (&local->transaction.parent_loc, oldloc);
-
-	local->transaction.main_frame   = frame;
-	local->transaction.basename     = AFR_BASENAME (oldloc->path);
-	local->transaction.new_basename = AFR_BASENAME (newloc->path);
-
-	afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
-
-	op_ret = 0;
+    return 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (link, frame, op_ret, op_errno,
-                                  NULL, NULL, NULL, NULL);
-	}
-
-	return 0;
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
+
+    AFR_STACK_UNWIND(link, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+    return 0;
 }
 
 /* }}} */
 
 /* {{{ symlink */
 
-
 int
-afr_symlink_unwind (call_frame_t *frame, xlator_t *this)
+afr_symlink_unwind(call_frame_t *frame, xlator_t *this)
 {
-	call_frame_t *main_frame = NULL;
-	afr_local_t  *local = NULL;
-        
-        struct iatt *unwind_buf = NULL;
-        
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame) {
-			main_frame = local->transaction.main_frame;
-		}
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
-
-	if (main_frame) {
-                if (local->cont.symlink.read_child_buf.ia_ino) {
-                        unwind_buf = &local->cont.symlink.read_child_buf;
-                } else {
-                        unwind_buf = &local->cont.symlink.buf;
-                }
-
-                unwind_buf->ia_ino = local->cont.symlink.ino;
-
-                local->cont.symlink.preparent.ia_ino  = local->cont.symlink.parent_ino;
-                local->cont.symlink.postparent.ia_ino = local->cont.symlink.parent_ino;
-
-		AFR_STACK_UNWIND (symlink, main_frame,
-                                  local->op_ret, local->op_errno,
-				  local->cont.symlink.inode,
-				  unwind_buf, &local->cont.symlink.preparent,
-                                  &local->cont.symlink.postparent);
-        }
+    call_frame_t *main_frame = NULL;
+    afr_local_t *local = NULL;
 
-	return 0;
-}
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-int
-afr_symlink_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		      int32_t op_ret, int32_t op_errno, inode_t *inode,
-                      struct iatt *buf, struct iatt *preparent,
-                      struct iatt *postparent)
-{
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-
-	int call_count = -1;
-	int child_index = -1;
-
-	local = frame->local;
-	priv = this->private;
-
-	child_index = (long) cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (afr_fop_failed (op_ret, op_errno))
-			afr_transaction_fop_failed (frame, this, child_index);
-		
-		if (op_ret != -1) {
-			local->op_ret   = op_ret;
-
-			if (local->success_count == 0) {
-				local->cont.symlink.buf        = *buf;
-				local->cont.symlink.ino = 
-					afr_itransform (buf->ia_ino, priv->child_count,
-							child_index);
-
-                                if (priv->read_child >= 0) {
-                                        afr_set_read_child (this, inode,
-                                                            priv->read_child);
-                                } else {
-                                        afr_set_read_child (this, inode,
-                                                            local->read_child_index);
-                                }
-			}
-
-                        if (child_index == local->first_up_child) {
-                                local->cont.symlink.ino =
-                                        afr_itransform (buf->ia_ino,
-                                                        priv->child_count,
-                                                        local->first_up_child);
-                        }
-
-                        if (child_index == local->read_child_index) {
-                                local->cont.symlink.read_child_buf = *buf;
-                                local->cont.symlink.preparent      = *preparent;
-                                local->cont.symlink.postparent     = *postparent;
-                        }
-
-			local->cont.symlink.inode    = inode;
-
-			local->success_count++;
-		}
-
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		local->transaction.unwind (frame, this);
-
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+    AFR_STACK_UNWIND(symlink, main_frame, local->op_ret, local->op_errno,
+                     local->inode, &local->cont.dir_fop.buf,
+                     &local->cont.dir_fop.preparent,
+                     &local->cont.dir_fop.postparent, local->xdata_rsp);
+    return 0;
 }
 
-
 int
-afr_symlink_wind (call_frame_t *frame, xlator_t *this)
+afr_symlink_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, inode_t *inode,
+                     struct iatt *buf, struct iatt *preparent,
+                     struct iatt *postparent, dict_t *xdata)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-
-	int call_count = -1;
-	int i = 0;
-
-	local = frame->local;
-	priv = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_symlink_wind_cbk,
-					   (void *) (long) i,	
-					   priv->children[i], 
-					   priv->children[i]->fops->symlink,
-					   local->cont.symlink.linkpath,
-					   &local->loc,
-                                           local->cont.symlink.params);
-
-			if (!--call_count)
-				break;
-
-		}
-	}
-	
-	return 0;
+    return __afr_dir_write_cbk(frame, cookie, this, op_ret, op_errno, buf,
+                               preparent, postparent, NULL, NULL, xdata);
 }
 
-
 int
-afr_symlink_done (call_frame_t *frame, xlator_t *this)
+afr_symlink_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-	afr_local_t * local = frame->local;
-
-	local->transaction.unwind (frame, this);
-
-	AFR_STACK_DESTROY (frame);
-	
-	return 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    STACK_WIND_COOKIE(frame, afr_symlink_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->symlink,
+                      local->cont.symlink.linkpath, &local->loc, local->umask,
+                      local->xdata_req);
+    return 0;
 }
 
-
 int
-afr_symlink (call_frame_t *frame, xlator_t *this,
-	     const char *linkpath, loc_t *loc, dict_t *params)
+afr_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+            loc_t *loc, mode_t umask, dict_t *xdata)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t  * transaction_frame = NULL;
-
-	int ret = -1;
-
-	int op_ret   = -1;
-	int op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv = this->private;
-
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		goto out;
-	}
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-	transaction_frame->local = local;
-	
-	loc_copy (&local->loc, loc);
-
-        LOCK (&priv->read_child_lock);
-        {
-                local->read_child_index = (++priv->read_child_rr) 
-                        % (priv->child_count);
-        }
-        UNLOCK (&priv->read_child_lock);
-
-	local->cont.symlink.linkpath = gf_strdup (linkpath);
-        if (params)
-                local->cont.symlink.params = dict_ref (params);
-
-        if (loc->parent)
-                local->cont.symlink.parent_ino = loc->parent->ino;
-
-	local->transaction.fop    = afr_symlink_wind;
-	local->transaction.done   = afr_symlink_done;
-	local->transaction.unwind = afr_symlink_unwind;
-
-	afr_build_parent_loc (&local->transaction.parent_loc, loc);
-
-	local->transaction.main_frame   = frame;
-	local->transaction.basename     = AFR_BASENAME (loc->path);
-
-	afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
-
-	op_ret = 0;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
+
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
+
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
+
+    loc_copy(&local->loc, loc);
+    local->inode = inode_ref(loc->inode);
+    local->parent = inode_ref(loc->parent);
+
+    local->cont.symlink.linkpath = gf_strdup(linkpath);
+    local->umask = umask;
+
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
+
+    if (!local->xdata_req)
+        goto out;
+
+    local->op = GF_FOP_SYMLINK;
+    local->transaction.wind = afr_symlink_wind;
+    local->transaction.unwind = afr_symlink_unwind;
+
+    ret = afr_build_parent_loc(&local->transaction.parent_loc, loc, &op_errno);
+    if (ret)
+        goto out;
+
+    local->transaction.main_frame = frame;
+    local->transaction.basename = AFR_BASENAME(loc->path);
+    ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (symlink, frame, op_ret, op_errno,
-                                  NULL, NULL, NULL, NULL);
-	}
-
-	return 0;
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
+
+    AFR_STACK_UNWIND(symlink, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                     NULL);
+    return 0;
 }
 
 /* }}} */
@@ -1351,230 +939,118 @@ out:
 /* {{{ rename */
 
 int
-afr_rename_unwind (call_frame_t *frame, xlator_t *this)
+afr_rename_unwind(call_frame_t *frame, xlator_t *this)
 {
-	call_frame_t *main_frame = NULL;
-	afr_local_t  *local = NULL;
-
-        struct iatt *unwind_buf = NULL;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame) {
-			main_frame = local->transaction.main_frame;
-		}
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
-
-	if (main_frame) {
-                if (local->cont.rename.read_child_buf.ia_ino) {
-                        unwind_buf = &local->cont.rename.read_child_buf;
-                } else {
-                        unwind_buf = &local->cont.rename.buf;
-                }
-
-		unwind_buf->ia_ino = local->cont.rename.ino;
-
-                local->cont.rename.preoldparent.ia_ino  = local->cont.rename.oldparent_ino;
-                local->cont.rename.postoldparent.ia_ino = local->cont.rename.oldparent_ino;
-                local->cont.rename.prenewparent.ia_ino  = local->cont.rename.newparent_ino;
-                local->cont.rename.postnewparent.ia_ino = local->cont.rename.newparent_ino;
-
-		AFR_STACK_UNWIND (rename, main_frame,
-                                  local->op_ret, local->op_errno, 
-				  unwind_buf,
-                                  &local->cont.rename.preoldparent,
-                                  &local->cont.rename.postoldparent,
-                                  &local->cont.rename.prenewparent,
-                                  &local->cont.rename.postnewparent);
-	}
-
-	return 0;
-}
+    call_frame_t *main_frame = NULL;
+    afr_local_t *local = NULL;
 
+    local = frame->local;
 
-int
-afr_rename_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		     int32_t op_ret, int32_t op_errno, struct iatt *buf,
-                     struct iatt *preoldparent, struct iatt *postoldparent,
-                     struct iatt *prenewparent, struct iatt *postnewparent)
-{
-	afr_local_t *   local = NULL;
-
-	int call_count = -1;
-	int child_index = -1;
-
-	local = frame->local;
-
-	child_index = (long) cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (afr_fop_failed (op_ret, op_errno) && op_errno != ENOTEMPTY)
-			afr_transaction_fop_failed (frame, this, child_index);
-
-		if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-		
-                                if (buf) {
-                                        local->cont.rename.buf = *buf;
-                                }
-
-                                local->success_count++;
-                        }
-
-                        if (child_index == local->read_child_index) {
-                                local->cont.rename.read_child_buf = *buf;
-
-                                local->cont.rename.preoldparent  = *preoldparent;
-                                local->cont.rename.postoldparent = *postoldparent;
-                                local->cont.rename.prenewparent  = *prenewparent;
-                                local->cont.rename.postnewparent = *postnewparent;
-                        }
-                }
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		local->transaction.unwind (frame, this);
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+    AFR_STACK_UNWIND(rename, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.dir_fop.buf, &local->cont.dir_fop.preparent,
+                     &local->cont.dir_fop.postparent,
+                     &local->cont.dir_fop.prenewparent,
+                     &local->cont.dir_fop.postnewparent, local->xdata_rsp);
+    return 0;
 }
 
-
-int32_t
-afr_rename_wind (call_frame_t *frame, xlator_t *this)
+int
+afr_rename_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                    struct iatt *preoldparent, struct iatt *postoldparent,
+                    struct iatt *prenewparent, struct iatt *postnewparent,
+                    dict_t *xdata)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-
-	int call_count = -1;
-	int i = 0;
-
-	local = frame->local;
-	priv = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_rename_wind_cbk, 
-					   (void *) (long) i,	
-					   priv->children[i], 
-					   priv->children[i]->fops->rename,
-					   &local->loc,
-					   &local->newloc);
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
+    return __afr_dir_write_cbk(frame, cookie, this, op_ret, op_errno, buf,
+                               preoldparent, postoldparent, prenewparent,
+                               postnewparent, xdata);
 }
 
-
 int
-afr_rename_done (call_frame_t *frame, xlator_t *this)
+afr_rename_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-	afr_local_t * local = frame->local;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-	local->transaction.unwind (frame, this);
+    local = frame->local;
+    priv = this->private;
 
-	AFR_STACK_DESTROY (frame);
-	
-	return 0;
+    STACK_WIND_COOKIE(frame, afr_rename_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->rename, &local->loc,
+                      &local->newloc, local->xdata_req);
+    return 0;
 }
 
-
 int
-afr_rename (call_frame_t *frame, xlator_t *this,
-	    loc_t *oldloc, loc_t *newloc)
+afr_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+           dict_t *xdata)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t  * transaction_frame = NULL;
-
-	int ret = -1;
-
-	int op_ret   = -1;
-	int op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv = this->private;
-
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		goto out;
-	}
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-	transaction_frame->local = local;
-
-	loc_copy (&local->loc,    oldloc);
-	loc_copy (&local->newloc, newloc);
-
-        local->read_child_index = afr_read_child (this, oldloc->inode);
-
-	local->cont.rename.ino = oldloc->inode->ino;
-
-        if (oldloc->parent)
-                local->cont.rename.oldparent_ino = oldloc->parent->ino;
-        if (newloc->parent)
-                local->cont.rename.newparent_ino = newloc->parent->ino;
-
-	local->transaction.fop    = afr_rename_wind;
-	local->transaction.done   = afr_rename_done;
-	local->transaction.unwind = afr_rename_unwind;
-
-	afr_build_parent_loc (&local->transaction.parent_loc, oldloc);
-	afr_build_parent_loc (&local->transaction.new_parent_loc, newloc);
-
-	local->transaction.main_frame   = frame;
-	local->transaction.basename     = AFR_BASENAME (oldloc->path);
-	local->transaction.new_basename = AFR_BASENAME (newloc->path);
-
-	afr_transaction (transaction_frame, this, AFR_ENTRY_RENAME_TRANSACTION);
-
-	op_ret = 0;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
+
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
+
+    loc_copy(&local->loc, oldloc);
+    loc_copy(&local->newloc, newloc);
+
+    local->inode = inode_ref(oldloc->inode);
+    local->parent = inode_ref(oldloc->parent);
+    local->parent2 = inode_ref(newloc->parent);
+
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
+
+    if (!local->xdata_req)
+        goto out;
+
+    local->op = GF_FOP_RENAME;
+    local->transaction.wind = afr_rename_wind;
+    local->transaction.unwind = afr_rename_unwind;
+
+    ret = afr_build_parent_loc(&local->transaction.parent_loc, oldloc,
+                               &op_errno);
+    if (ret)
+        goto out;
+    ret = afr_build_parent_loc(&local->transaction.new_parent_loc, newloc,
+                               &op_errno);
+    if (ret)
+        goto out;
+
+    local->transaction.main_frame = frame;
+    local->transaction.basename = AFR_BASENAME(oldloc->path);
+    local->transaction.new_basename = AFR_BASENAME(newloc->path);
+    ret = afr_transaction(transaction_frame, this,
+                          AFR_ENTRY_RENAME_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-
-		AFR_STACK_UNWIND (rename, frame, op_ret, op_errno,
-                                  NULL, NULL, NULL, NULL, NULL);
-	}
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-	return 0;
+    AFR_STACK_UNWIND(rename, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL,
+                     NULL);
+    return 0;
 }
 
 /* }}} */
@@ -1582,412 +1058,205 @@ out:
 /* {{{ unlink */
 
 int
-afr_unlink_unwind (call_frame_t *frame, xlator_t *this)
-{
-	call_frame_t *main_frame = NULL;
-	afr_local_t  *local = NULL;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame) {
-			main_frame = local->transaction.main_frame;
-		}
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
-
-	if (main_frame) {
-                local->cont.unlink.preparent.ia_ino  = local->cont.unlink.parent_ino;
-                local->cont.unlink.postparent.ia_ino = local->cont.unlink.parent_ino;
-
-		AFR_STACK_UNWIND (unlink, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  &local->cont.unlink.preparent,
-                                  &local->cont.unlink.postparent);
-        }
-
-	return 0;
-}
-
-
-int
-afr_unlink_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		     int32_t op_ret, int32_t op_errno, struct iatt *preparent,
-                     struct iatt *postparent)
+afr_unlink_unwind(call_frame_t *frame, xlator_t *this)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-
-	int call_count  = -1;
-	int child_index = (long) cookie;
-
-	local = frame->local;
-	priv  = this->private;
-
-	LOCK (&frame->lock);
-	{
-                if (child_index == local->read_child_index) {
-                        local->read_child_returned = _gf_true;
-                }
-
-		if (afr_fop_failed (op_ret, op_errno))
-			afr_transaction_fop_failed (frame, this, child_index);
+    call_frame_t *main_frame = NULL;
+    afr_local_t *local = NULL;
 
-		if (op_ret != -1) {
-			if (local->success_count == 0) {
-				local->op_ret   = op_ret;
-                                local->cont.unlink.preparent  = *preparent;
-                                local->cont.unlink.postparent = *postparent;
-			}
+    local = frame->local;
 
-                        if (child_index == local->read_child_index) {
-                                local->cont.unlink.preparent  = *preparent;
-                                local->cont.unlink.postparent = *postparent;
-                        }
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-			local->success_count++;
-		}
-
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		local->transaction.unwind (frame, this);
-
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+    AFR_STACK_UNWIND(unlink, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.dir_fop.preparent,
+                     &local->cont.dir_fop.postparent, local->xdata_rsp);
+    return 0;
 }
 
-
-int32_t
-afr_unlink_wind (call_frame_t *frame, xlator_t *this)
+int
+afr_unlink_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-
-	int call_count = -1;
-	int i = 0;
-
-	local = frame->local;
-	priv  = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_unlink_wind_cbk,	
-					   (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->unlink,
-					   &local->loc);
-			
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
+    return __afr_dir_write_cbk(frame, cookie, this, op_ret, op_errno, NULL,
+                               preparent, postparent, NULL, NULL, xdata);
 }
 
-
-int32_t
-afr_unlink_done (call_frame_t *frame, xlator_t *this)
+int
+afr_unlink_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-	afr_local_t * local = frame->local;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-	local->transaction.unwind (frame, this);
+    local = frame->local;
+    priv = this->private;
 
-	AFR_STACK_DESTROY (frame);
-	
-	return 0;
+    STACK_WIND_COOKIE(frame, afr_unlink_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->unlink, &local->loc,
+                      local->xflag, local->xdata_req);
+    return 0;
 }
 
-
-int32_t
-afr_unlink (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc)
+int
+afr_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+           dict_t *xdata)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t  * transaction_frame = NULL;
-
-	int ret = -1;
-
-	int op_ret   = -1;
-	int op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv = this->private;
-
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		goto out;
-	}
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-	transaction_frame->local = local;
-
-	loc_copy (&local->loc, loc);
-
-        if (loc->parent)
-                local->cont.unlink.parent_ino = loc->parent->ino;
-
-	local->transaction.fop    = afr_unlink_wind;
-	local->transaction.done   = afr_unlink_done;
-	local->transaction.unwind = afr_unlink_unwind;
-
-	afr_build_parent_loc (&local->transaction.parent_loc, loc);
-
-	local->transaction.main_frame = frame;
-	local->transaction.basename = AFR_BASENAME (loc->path);
-
-	afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
-
-	op_ret = 0;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
+
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
+
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
+
+    loc_copy(&local->loc, loc);
+    local->xflag = xflag;
+
+    local->inode = inode_ref(loc->inode);
+    local->parent = inode_ref(loc->parent);
+
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
+
+    if (!local->xdata_req)
+        goto out;
+
+    local->op = GF_FOP_UNLINK;
+    local->transaction.wind = afr_unlink_wind;
+    local->transaction.unwind = afr_unlink_unwind;
+
+    ret = afr_build_parent_loc(&local->transaction.parent_loc, loc, &op_errno);
+    if (ret)
+        goto out;
+
+    local->transaction.main_frame = frame;
+    local->transaction.basename = AFR_BASENAME(loc->path);
+    ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (unlink, frame, op_ret, op_errno,
-                                  NULL, NULL);
-	}
-
-	return 0;
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
+
+    AFR_STACK_UNWIND(unlink, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
 /* }}} */
 
 /* {{{ rmdir */
 
-
-
 int
-afr_rmdir_unwind (call_frame_t *frame, xlator_t *this)
+afr_rmdir_unwind(call_frame_t *frame, xlator_t *this)
 {
-	call_frame_t *main_frame = NULL;
-	afr_local_t  *local = NULL;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame) {
-			main_frame = local->transaction.main_frame;
-		}
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
-
-	if (main_frame) {
-                local->cont.rmdir.preparent.ia_ino  = local->cont.rmdir.parent_ino;
-                local->cont.rmdir.postparent.ia_ino = local->cont.rmdir.parent_ino;
-
-		AFR_STACK_UNWIND (rmdir, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  &local->cont.rmdir.preparent,
-                                  &local->cont.rmdir.postparent);
-        }
-
-	return 0;
-}
-
-
-int
-afr_rmdir_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-		    int32_t op_ret, int32_t op_errno, struct iatt *preparent,
-                    struct iatt *postparent)
-{
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-
-	int call_count  = -1;
-	int child_index = (long) cookie;
-        int read_child  = 0;
+    call_frame_t *main_frame = NULL;
+    afr_local_t *local = NULL;
 
-	local = frame->local;
-	priv = this->private;
+    local = frame->local;
 
-	LOCK (&frame->lock);
-	{
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-		if (afr_fop_failed (op_ret, op_errno) && (op_errno != ENOTEMPTY))
-			afr_transaction_fop_failed (frame, this, child_index);
-
-		if (op_ret != -1) {
-			if (local->success_count == 0) {
-				local->op_ret = op_ret;
-                                local->cont.rmdir.preparent  = *preparent;
-                                local->cont.rmdir.postparent = *postparent;
-
-			}
-
-                        if (child_index == read_child) {
-                                local->cont.rmdir.preparent  = *preparent;
-                                local->cont.rmdir.postparent = *postparent;
-                        }
-
-			local->success_count++;
-		}
-
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		local->transaction.unwind (frame, this);
-		local->transaction.resume (frame, this);
-	}
-	
-	return 0;
+    AFR_STACK_UNWIND(rmdir, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.dir_fop.preparent,
+                     &local->cont.dir_fop.postparent, local->xdata_rsp);
+    return 0;
 }
 
-
 int
-afr_rmdir_wind (call_frame_t *frame, xlator_t *this)
+afr_rmdir_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
 {
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-
-	int call_count = -1;
-	int i = 0;
-
-	local = frame->local;
-	priv  = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {				
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_rmdir_wind_cbk,	
-					   (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->rmdir,
-					   &local->loc, local->cont.rmdir.flags);
-
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	return 0;
+    return __afr_dir_write_cbk(frame, cookie, this, op_ret, op_errno, NULL,
+                               preparent, postparent, NULL, NULL, xdata);
 }
 
-
 int
-afr_rmdir_done (call_frame_t *frame, xlator_t *this)
+afr_rmdir_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-	afr_local_t * local = frame->local;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-	local->transaction.unwind (frame, this);
+    local = frame->local;
+    priv = this->private;
 
-	AFR_STACK_DESTROY (frame);
-	
-	return 0;
+    STACK_WIND_COOKIE(frame, afr_rmdir_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->rmdir, &local->loc,
+                      local->cont.rmdir.flags, local->xdata_req);
+    return 0;
 }
 
-
 int
-afr_rmdir (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, int flags)
+afr_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+          dict_t *xdata)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t  * transaction_frame = NULL;
-	
-	int ret = -1;
-
-	int op_ret   = -1;
-	int op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv = this->private;
-
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		goto out;
-	}
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-	transaction_frame->local = local;
-
-        local->cont.rmdir.flags = flags;
-	loc_copy (&local->loc, loc);
-
-        if (loc->parent)
-                local->cont.rmdir.parent_ino = loc->parent->ino;
-
-	local->transaction.fop    = afr_rmdir_wind;
-	local->transaction.done   = afr_rmdir_done;
-	local->transaction.unwind = afr_rmdir_unwind;
-
-	afr_build_parent_loc (&local->transaction.parent_loc, loc);
-
-	local->transaction.main_frame = frame;
-	local->transaction.basename = AFR_BASENAME (loc->path);
-
-	afr_transaction (transaction_frame, this, AFR_ENTRY_TRANSACTION);
-
-	op_ret = 0;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
+
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
+
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
+
+    loc_copy(&local->loc, loc);
+    local->inode = inode_ref(loc->inode);
+    local->parent = inode_ref(loc->parent);
+
+    local->cont.rmdir.flags = flags;
+
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
+
+    if (!local->xdata_req)
+        goto out;
+
+    local->op = GF_FOP_RMDIR;
+    local->transaction.wind = afr_rmdir_wind;
+    local->transaction.unwind = afr_rmdir_unwind;
+
+    ret = afr_build_parent_loc(&local->transaction.parent_loc, loc, &op_errno);
+    if (ret)
+        goto out;
+
+    local->transaction.main_frame = frame;
+    local->transaction.basename = AFR_BASENAME(loc->path);
+    ret = afr_transaction(transaction_frame, this, AFR_ENTRY_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
 out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (rmdir, frame, op_ret, op_errno,
-                                  NULL, NULL);
-	}
-
-	return 0;
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
+
+    AFR_STACK_UNWIND(rmdir, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
 /* }}} */
-
diff --git a/xlators/cluster/afr/src/afr-dir-write.h b/xlators/cluster/afr/src/afr-dir-write.h
index 4f180857e2e..1d88c3b9b26 100644
--- a/xlators/cluster/afr/src/afr-dir-write.h
+++ b/xlators/cluster/afr/src/afr-dir-write.h
@@ -1,60 +1,46 @@
 /*
-   Copyright (c) 2007-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef __DIR_WRITE_H__
 #define __DIR_WRITE_H__
 
 int32_t
-afr_create (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, int32_t flags, mode_t mode,
-            fd_t *fd, dict_t *params);
+afr_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+           mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata);
 
 int32_t
-afr_mknod (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, mode_t mode, dev_t dev, dict_t *params);
+afr_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+          dev_t dev, mode_t umask, dict_t *xdata);
 
 int32_t
-afr_mkdir (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, mode_t mode, dict_t *params);
+afr_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+          mode_t umask, dict_t *xdata);
 
 int32_t
-afr_unlink (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc);
+afr_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+           dict_t *xdata);
 
 int32_t
-afr_rmdir (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, int flags);
+afr_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+          dict_t *xdata);
 
 int32_t
-afr_link (call_frame_t *frame, xlator_t *this,
-	  loc_t *oldloc, loc_t *newloc);
+afr_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+         dict_t *xdata);
 
 int32_t
-afr_rename (call_frame_t *frame, xlator_t *this,
-	    loc_t *oldloc, loc_t *newloc);
+afr_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+           dict_t *xdata);
 
 int
-afr_symlink (call_frame_t *frame, xlator_t *this,
-	     const char *linkpath, loc_t *oldloc, dict_t *params);
-
-int32_t
-afr_setdents (call_frame_t *frame, xlator_t *this,
-	      fd_t *fd, int32_t flags, dir_entry_t *entries, int32_t count);
+afr_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+            loc_t *oldloc, mode_t umask, dict_t *params);
 
 #endif /* __DIR_WRITE_H__ */
diff --git a/xlators/cluster/afr/src/afr-inode-read.c b/xlators/cluster/afr/src/afr-inode-read.c
index 68ac78731e8..c5521704de2 100644
--- a/xlators/cluster/afr/src/afr-inode-read.c
+++ b/xlators/cluster/afr/src/afr-inode-read.c
@@ -1,22 +1,12 @@
 /*
-   Copyright (c) 2007-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
 
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
 
 #include <libgen.h>
 #include <unistd.h>
@@ -25,863 +15,1880 @@
 #include <stdlib.h>
 #include <signal.h>
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "afr.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-
+#include <glusterfs/glusterfs.h>
 #include "afr.h"
+#include <glusterfs/dict.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/list.h>
+#include <glusterfs/byte-order.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/quota-common-utils.h>
+
+#include "afr-transaction.h"
+#include "afr-messages.h"
 
-
-/**
- * Common algorithm for inode read calls:
- *
- * - Try the fop on the first child that is up
- * - if we have failed due to ENOTCONN:
- *     try the next child
- *
- * Applicable to: access, stat, fstat, readlink, getxattr
- */
+/*
+ * Quota size xattrs are not maintained by afr. There is a
+ * possibility that they differ even when both the directory changelog xattrs
+ * suggest everything is fine. So if there is at least one 'source' check among
+ * the sources which has the maximum quota size. Otherwise check among all the
+ * available ones for maximum quota size. This way if there is a source and
+ * stale copies it always votes for the 'source'.
+ * */
+
+int
+afr_handle_quota_size(call_frame_t *frame, xlator_t *this)
+{
+    unsigned char *readable = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    struct afr_reply *replies = NULL;
+    int i = 0;
+    int ret = 0;
+    quota_meta_t size = {
+        0,
+    };
+    quota_meta_t max_size = {
+        0,
+    };
+    int readable_cnt = 0;
+    int read_subvol = -1;
+
+    local = frame->local;
+    priv = this->private;
+    replies = local->replies;
+
+    readable = alloca0(priv->child_count);
+
+    afr_inode_read_subvol_get(local->inode, this, readable, 0, 0);
+
+    readable_cnt = AFR_COUNT(readable, priv->child_count);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret == -1)
+            continue;
+        if (readable_cnt && !readable[i])
+            continue;
+        if (!replies[i].xdata)
+            continue;
+        ret = quota_dict_get_meta(replies[i].xdata, QUOTA_SIZE_KEY,
+                                  SLEN(QUOTA_SIZE_KEY), &size);
+        if (ret == -1)
+            continue;
+        if (read_subvol == -1)
+            read_subvol = i;
+        if (size.size > max_size.size ||
+            (size.file_count + size.dir_count) >
+                (max_size.file_count + max_size.dir_count))
+            read_subvol = i;
+
+        if (size.size > max_size.size)
+            max_size.size = size.size;
+        if (size.file_count > max_size.file_count)
+            max_size.file_count = size.file_count;
+        if (size.dir_count > max_size.dir_count)
+            max_size.dir_count = size.dir_count;
+    }
+
+    if (max_size.size == 0 && max_size.file_count == 0 &&
+        max_size.dir_count == 0)
+        return read_subvol;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret == -1)
+            continue;
+        if (readable_cnt && !readable[i])
+            continue;
+        if (!replies[i].xdata)
+            continue;
+        quota_dict_set_meta(replies[i].xdata, QUOTA_SIZE_KEY, &max_size,
+                            IA_IFDIR);
+    }
+
+    return read_subvol;
+}
 
 /* {{{ access */
 
-int32_t
-afr_access_cbk (call_frame_t *frame, void *cookie,
-		xlator_t *this, int32_t op_ret, int32_t op_errno)
+int
+afr_access_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, dict_t *xdata)
 {
-	afr_private_t * priv     = NULL;
-	afr_local_t *   local    = NULL;
-	xlator_t **     children = NULL;
+    afr_local_t *local = NULL;
 
-	int unwind     = 1;
-	int last_tried = -1;
-	int this_try = -1;
-        int read_child = -1;
+    local = frame->local;
 
-	priv     = this->private;
-	children = priv->children;
+    if (op_ret < 0) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
 
-	local = frame->local;
+        afr_read_txn_continue(frame, this, (long)cookie);
+        return 0;
+    }
 
-        read_child = (long) cookie;
+    AFR_STACK_UNWIND(access, frame, op_ret, op_errno, xdata);
 
-	if (op_ret == -1) {
-        retry:
-		last_tried = local->cont.access.last_tried;
+    return 0;
+}
 
-		if (all_tried (last_tried, priv->child_count)) {
-			goto out;
-		}
-		this_try    = ++local->cont.access.last_tried;
+int
+afr_access_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    if (subvol == -1) {
+        AFR_STACK_UNWIND(access, frame, local->op_ret, local->op_errno, 0);
+        return 0;
+    }
+
+    STACK_WIND_COOKIE(frame, afr_access_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->access, &local->loc,
+                      local->cont.access.mask, local->xdata_req);
+    return 0;
+}
 
-                if (this_try == read_child) {
-                        goto retry;
-                }
+int
+afr_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int mask,
+           dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int op_errno = 0;
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-		unwind = 0;
+    local->op = GF_FOP_ACCESS;
+    loc_copy(&local->loc, loc);
+    local->cont.access.mask = mask;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
 
-		STACK_WIND_COOKIE (frame, afr_access_cbk,
-				   (void *) (long) read_child,
-				   children[this_try],
-				   children[this_try]->fops->access,
-				   &local->loc, local->cont.access.mask);
-	}
+    afr_read_txn(frame, this, loc->inode, afr_access_wind,
+                 AFR_METADATA_TRANSACTION);
 
+    return 0;
 out:
-	if (unwind) {
-		AFR_STACK_UNWIND (access, frame, op_ret, op_errno);
-	}
+    AFR_STACK_UNWIND(access, frame, -1, op_errno, NULL);
 
-	return 0;
+    return 0;
 }
 
+/* }}} */
 
-int32_t
-afr_access (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc, int32_t mask)
-{
-	afr_private_t * priv       = NULL;
-	xlator_t **     children   = NULL;
-	int             call_child = 0;
-	afr_local_t     *local     = NULL;
+/* {{{ stat */
 
-        int32_t         read_child = -1;
+int
+afr_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, struct iatt *buf, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
 
+    local = frame->local;
 
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
+    if (op_ret < 0) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        afr_read_txn_continue(frame, this, (long)cookie);
+        return 0;
+    }
 
-	priv     = this->private;
-	VALIDATE_OR_GOTO (priv->children, out);
+    AFR_STACK_UNWIND(stat, frame, op_ret, op_errno, buf, xdata);
 
-	children = priv->children;
+    return 0;
+}
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+int
+afr_stat_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
 
-        read_child = afr_read_child (this, loc->inode);
+    priv = this->private;
+    local = frame->local;
 
-        if ((read_child >= 0) && (priv->child_up[read_child])) {
-                call_child = read_child;
+    if (subvol == -1) {
+        AFR_STACK_UNWIND(stat, frame, local->op_ret, local->op_errno, 0, 0);
+        return 0;
+    }
 
-                local->cont.access.last_tried = -1;
+    STACK_WIND_COOKIE(
+        frame, afr_stat_cbk, (void *)(long)subvol, priv->children[subvol],
+        priv->children[subvol]->fops->stat, &local->loc, local->xdata_req);
+    return 0;
+}
 
-        } else {
-                call_child = afr_first_up_child (priv);
-                if (call_child == -1) {
-                        op_errno = ENOTCONN;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "no child is up");
-                        goto out;
-                }
+int
+afr_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int op_errno = 0;
 
-                local->cont.access.last_tried = call_child;
-        }
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-	loc_copy (&local->loc, loc);
-	local->cont.access.mask       = mask;
+    local->op = GF_FOP_STAT;
+    loc_copy(&local->loc, loc);
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
 
-	STACK_WIND_COOKIE (frame, afr_access_cbk,
-			   (void *) (long) call_child,
-			   children[call_child], children[call_child]->fops->access,
-			   loc, mask);
+    afr_read_txn(frame, this, loc->inode, afr_stat_wind, AFR_DATA_TRANSACTION);
 
-	op_ret = 0;
+    return 0;
 out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (access, frame, op_ret, op_errno);
-	}
-	return 0;
-}
+    AFR_STACK_UNWIND(stat, frame, -1, op_errno, NULL, NULL);
 
+    return 0;
+}
 
 /* }}} */
 
-/* {{{ stat */
+/* {{{ fstat */
 
-int32_t
-afr_stat_cbk (call_frame_t *frame, void *cookie,
-	      xlator_t *this, int32_t op_ret, int32_t op_errno,
-	      struct iatt *buf)
+int
+afr_fstat_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct iatt *buf, dict_t *xdata)
 {
-	afr_private_t * priv     = NULL;
-	afr_local_t *   local    = NULL;
-	xlator_t **     children = NULL;
+    afr_local_t *local = NULL;
 
-	int unwind     = 1;
-	int last_tried = -1;
-	int this_try = -1;
-	int read_child = -1;
+    local = frame->local;
 
-	priv     = this->private;
-	children = priv->children;
+    if (op_ret < 0) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
 
-	read_child = (long) cookie;
+        afr_read_txn_continue(frame, this, (long)cookie);
+        return 0;
+    }
 
-	local = frame->local;
+    AFR_STACK_UNWIND(fstat, frame, op_ret, op_errno, buf, xdata);
 
-	if (op_ret == -1) {
-	retry:
-		last_tried = local->cont.stat.last_tried;
-
-		if (all_tried (last_tried, priv->child_count)) {
-			goto out;
-		}
-		this_try = ++local->cont.stat.last_tried;
-
-		if (this_try == read_child) {
-			goto retry;
-		}
-
-		unwind = 0;
+    return 0;
+}
 
-		STACK_WIND_COOKIE (frame, afr_stat_cbk,
-				   (void *) (long) read_child,
-				   children[this_try],
-				   children[this_try]->fops->stat,
-				   &local->loc);
-	}
+int
+afr_fstat_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
 
-out:
-	if (unwind) {
-		if (buf)
-			buf->ia_ino = local->cont.stat.ino;
+    priv = this->private;
+    local = frame->local;
 
-		AFR_STACK_UNWIND (stat, frame, op_ret, op_errno, buf);
-	}
+    if (subvol == -1) {
+        AFR_STACK_UNWIND(fstat, frame, local->op_ret, local->op_errno, 0, 0);
+        return 0;
+    }
 
-	return 0;
+    STACK_WIND_COOKIE(
+        frame, afr_fstat_cbk, (void *)(long)subvol, priv->children[subvol],
+        priv->children[subvol]->fops->fstat, local->fd, local->xdata_req);
+    return 0;
 }
 
-
 int32_t
-afr_stat (call_frame_t *frame, xlator_t *this,
-	  loc_t *loc)
+afr_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
-	afr_private_t * priv       = NULL;
-	afr_local_t   * local      = NULL;
-	xlator_t **     children   = NULL;
+    afr_local_t *local = NULL;
+    int op_errno = 0;
 
-        int32_t         read_child = -1;
-	int             call_child = 0;
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-	int32_t         op_ret     = -1;
-	int32_t         op_errno   = 0;
+    local->op = GF_FOP_FSTAT;
+    local->fd = fd_ref(fd);
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+    afr_fix_open(fd, this);
 
-	priv     = this->private;
-	VALIDATE_OR_GOTO (priv->children, out);
+    afr_read_txn(frame, this, fd->inode, afr_fstat_wind, AFR_DATA_TRANSACTION);
 
-	children = priv->children;
+    return 0;
+out:
+    AFR_STACK_UNWIND(fstat, frame, -1, op_errno, NULL, NULL);
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+    return 0;
+}
 
-	frame->local = local;
+/* }}} */
 
-        read_child = afr_read_child (this, loc->inode);
+/* {{{ readlink */
 
-        if ((read_child >= 0) && (priv->child_up[read_child])) {
-                call_child = read_child;
+int
+afr_readlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, const char *buf,
+                 struct iatt *sbuf, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
 
-                local->cont.stat.last_tried = -1;
+    local = frame->local;
 
-        } else {
-		call_child = afr_first_up_child (priv);
-		if (call_child == -1) {
-			op_errno = ENOTCONN;
-			gf_log (this->name, GF_LOG_DEBUG,
-				"no child is up");
-			goto out;
-		}
+    if (op_ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = op_errno;
 
-		local->cont.stat.last_tried = call_child;
-	}
+        afr_read_txn_continue(frame, this, (long)cookie);
+        return 0;
+    }
 
-	loc_copy (&local->loc, loc);
+    AFR_STACK_UNWIND(readlink, frame, op_ret, op_errno, buf, sbuf, xdata);
+    return 0;
+}
 
-	local->cont.stat.ino = loc->inode->ino;
+int
+afr_readlink_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    if (subvol == -1) {
+        AFR_STACK_UNWIND(readlink, frame, local->op_ret, local->op_errno, 0, 0,
+                         0);
+        return 0;
+    }
+
+    STACK_WIND_COOKIE(frame, afr_readlink_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->readlink, &local->loc,
+                      local->cont.readlink.size, local->xdata_req);
+    return 0;
+}
 
-	STACK_WIND_COOKIE (frame, afr_stat_cbk, (void *) (long) call_child,
-			   children[call_child],
-			   children[call_child]->fops->stat,
-			   loc);
+int
+afr_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+             dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int32_t op_errno = 0;
 
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (stat, frame, op_ret, op_errno, NULL);
-	}
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-	return 0;
-}
+    local->op = GF_FOP_READLINK;
+    loc_copy(&local->loc, loc);
+    local->cont.readlink.size = size;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
 
+    afr_read_txn(frame, this, loc->inode, afr_readlink_wind,
+                 AFR_DATA_TRANSACTION);
 
-/* }}} */
+    return 0;
+out:
+    AFR_STACK_UNWIND(readlink, frame, -1, op_errno, 0, 0, 0);
 
-/* {{{ fstat */
+    return 0;
+}
 
-int32_t
-afr_fstat_cbk (call_frame_t *frame, void *cookie,
-	       xlator_t *this, int32_t op_ret, int32_t op_errno,
-	       struct iatt *buf)
-{
-	afr_private_t * priv     = NULL;
-	afr_local_t *   local    = NULL;
-	xlator_t **     children = NULL;
+/* }}} */
 
-	int unwind     = 1;
-	int last_tried = -1;
-	int this_try = -1;
-        int read_child = -1;
+/* {{{ getxattr */
 
-	priv     = this->private;
-	children = priv->children;
+struct _xattr_key {
+    char *key;
+    struct list_head list;
+};
 
-	local = frame->local;
+int
+__gather_xattr_keys(dict_t *dict, char *key, data_t *value, void *data)
+{
+    struct list_head *list = data;
+    struct _xattr_key *xkey = NULL;
 
-	read_child = (long) cookie;
+    if (!strncmp(key, AFR_XATTR_PREFIX, SLEN(AFR_XATTR_PREFIX))) {
+        xkey = GF_MALLOC(sizeof(*xkey), gf_afr_mt_xattr_key);
+        if (!xkey)
+            return -1;
 
-	if (op_ret == -1) {
-	retry:
-		last_tried = local->cont.fstat.last_tried;
+        xkey->key = key;
+        INIT_LIST_HEAD(&xkey->list);
 
-		if (all_tried (last_tried, priv->child_count)) {
-			goto out;
-		}
-		this_try   = ++local->cont.fstat.last_tried;
+        list_add_tail(&xkey->list, list);
+    }
+    return 0;
+}
 
-		if (this_try == read_child) {
-			goto retry;
-		}
+void
+afr_filter_xattrs(dict_t *dict)
+{
+    struct list_head keys = {
+        0,
+    };
+    struct _xattr_key *key = NULL;
+    struct _xattr_key *tmp = NULL;
 
-		unwind = 0;
+    INIT_LIST_HEAD(&keys);
 
-		STACK_WIND_COOKIE (frame, afr_fstat_cbk,
-				   (void *) (long) read_child,
-				   children[this_try],
-				   children[this_try]->fops->fstat,
-				   local->fd);
-	}
+    dict_foreach(dict, __gather_xattr_keys, (void *)&keys);
 
-out:
-	if (unwind) {
-		if (buf)
-			buf->ia_ino = local->cont.fstat.ino;
+    list_for_each_entry_safe(key, tmp, &keys, list)
+    {
+        dict_del(dict, key->key);
 
-		AFR_STACK_UNWIND (fstat, frame, op_ret, op_errno, buf);
-	}
+        list_del_init(&key->list);
 
-	return 0;
+        GF_FREE(key);
+    }
 }
 
+static gf_boolean_t
+afr_getxattr_ignorable_errnos(int32_t op_errno)
+{
+    if (op_errno == ENODATA || op_errno == ENOTSUP || op_errno == ERANGE ||
+        op_errno == ENAMETOOLONG)
+        return _gf_true;
 
-int32_t
-afr_fstat (call_frame_t *frame, xlator_t *this,
-	   fd_t *fd)
+    return _gf_false;
+}
+int
+afr_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
 {
-	afr_private_t * priv       = NULL;
-	afr_local_t   * local      = NULL;
-	xlator_t **     children   = NULL;
+    afr_local_t *local = NULL;
 
-	int             call_child = 0;
-        int32_t         read_child = -1;
+    local = frame->local;
 
-	int32_t         op_ret     = -1;
-	int32_t         op_errno   = 0;
+    if (op_ret < 0 && !afr_getxattr_ignorable_errnos(op_errno)) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (fd, out);
-	VALIDATE_OR_GOTO (this->private, out);
+        afr_read_txn_continue(frame, this, (long)cookie);
+        return 0;
+    }
 
-	priv     = this->private;
-	VALIDATE_OR_GOTO (priv->children, out);
+    if (dict)
+        afr_filter_xattrs(dict);
 
-	children = priv->children;
+    AFR_STACK_UNWIND(getxattr, frame, op_ret, op_errno, dict, xdata);
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+    return 0;
+}
 
-	frame->local = local;
+int
+afr_getxattr_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    if (subvol == -1) {
+        AFR_STACK_UNWIND(getxattr, frame, local->op_ret, local->op_errno, NULL,
+                         NULL);
+        return 0;
+    }
+
+    STACK_WIND_COOKIE(frame, afr_getxattr_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->getxattr, &local->loc,
+                      local->cont.getxattr.name, local->xdata_req);
+    return 0;
+}
 
-	VALIDATE_OR_GOTO (fd->inode, out);
+int32_t
+afr_getxattr_unwind(call_frame_t *frame, int op_ret, int op_errno, dict_t *dict,
+                    dict_t *xdata)
 
-        read_child = afr_read_child (this, fd->inode);
+{
+    AFR_STACK_UNWIND(getxattr, frame, op_ret, op_errno, dict, xdata);
+    return 0;
+}
 
-        if ((read_child >= 0) && (priv->child_up[read_child])) {
-                call_child = read_child;
+int32_t
+afr_fgetxattr_clrlk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *dict,
+                        dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    xlator_t **children = NULL;
+    dict_t *xattr = NULL;
+    char *tmp_report = NULL;
+    char lk_summary[1024] = {
+        0,
+    };
+    int serz_len = 0;
+    int32_t callcnt = 0;
+    long int cky = 0;
+    int ret = 0;
+    int keylen = 0;
+    int children_keylen = 0;
+
+    priv = this->private;
+    children = priv->children;
+
+    local = frame->local;
+    cky = (long)cookie;
+    keylen = strlen(local->cont.getxattr.name);
+    children_keylen = strlen(children[cky]->name);
+
+    LOCK(&frame->lock);
+    {
+        callcnt = --local->call_count;
+        if (op_ret == -1)
+            local->replies[cky].op_errno = op_errno;
+
+        if (!local->dict)
+            local->dict = dict_new();
+        if (local->dict) {
+            ret = dict_get_strn(dict, local->cont.getxattr.name, keylen,
+                                &tmp_report);
+            if (ret)
+                goto unlock;
+            ret = dict_set_dynstrn(local->dict, children[cky]->name,
+                                   children_keylen, gf_strdup(tmp_report));
+            if (ret)
+                goto unlock;
+        }
+    }
+unlock:
+    UNLOCK(&frame->lock);
+
+    if (!callcnt) {
+        xattr = dict_new();
+        if (!xattr) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto unwind;
+        }
+        ret = dict_serialize_value_with_delim(local->dict, lk_summary,
+                                              &serz_len, '\n');
+        if (ret) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto unwind;
+        }
+        if (serz_len == -1)
+            snprintf(lk_summary, sizeof(lk_summary), "No locks cleared.");
+        ret = dict_set_dynstrn(xattr, local->cont.getxattr.name, keylen,
+                               gf_strdup(lk_summary));
+        if (ret) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_DICT_SET_FAILED,
+                   "Error setting dictionary");
+            goto unwind;
+        }
 
-                local->cont.fstat.last_tried = -1;
-        } else {
-                call_child = afr_first_up_child (priv);
+        op_errno = afr_final_errno(local, priv);
 
-                if (call_child == -1) {
-                        op_errno = ENOTCONN;
-                        gf_log (this->name, GF_LOG_DEBUG,
-				"no child is up");
-			goto out;
-		}
+    unwind:
+        AFR_STACK_UNWIND(fgetxattr, frame, op_ret, op_errno, xattr, xdata);
+        if (xattr)
+            dict_unref(xattr);
+    }
 
-                local->cont.fstat.last_tried = call_child;
+    return ret;
+}
+
+int32_t
+afr_getxattr_clrlk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *dict,
+                       dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    xlator_t **children = NULL;
+    dict_t *xattr = NULL;
+    char *tmp_report = NULL;
+    char lk_summary[1024] = {
+        0,
+    };
+    int serz_len = 0;
+    int32_t callcnt = 0;
+    long int cky = 0;
+    int ret = 0;
+    int keylen = 0;
+    int children_keylen = 0;
+
+    priv = this->private;
+    children = priv->children;
+
+    local = frame->local;
+    cky = (long)cookie;
+
+    keylen = strlen(local->cont.getxattr.name);
+    children_keylen = strlen(children[cky]->name);
+
+    LOCK(&frame->lock);
+    {
+        callcnt = --local->call_count;
+        if (op_ret == -1)
+            local->replies[cky].op_errno = op_errno;
+
+        if (!local->dict)
+            local->dict = dict_new();
+        if (local->dict) {
+            ret = dict_get_strn(dict, local->cont.getxattr.name, keylen,
+                                &tmp_report);
+            if (ret)
+                goto unlock;
+            ret = dict_set_dynstrn(local->dict, children[cky]->name,
+                                   children_keylen, gf_strdup(tmp_report));
+            if (ret)
+                goto unlock;
+        }
+    }
+unlock:
+    UNLOCK(&frame->lock);
+
+    if (!callcnt) {
+        xattr = dict_new();
+        if (!xattr) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto unwind;
+        }
+        ret = dict_serialize_value_with_delim(local->dict, lk_summary,
+                                              &serz_len, '\n');
+        if (ret) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto unwind;
+        }
+        if (serz_len == -1)
+            snprintf(lk_summary, sizeof(lk_summary), "No locks cleared.");
+        ret = dict_set_dynstrn(xattr, local->cont.getxattr.name, keylen,
+                               gf_strdup(lk_summary));
+        if (ret) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_DICT_SET_FAILED,
+                   "Error setting dictionary");
+            goto unwind;
         }
 
-	local->cont.fstat.ino = fd->inode->ino;
-	local->fd = fd_ref (fd);
+        op_errno = afr_final_errno(local, priv);
 
-	STACK_WIND_COOKIE (frame, afr_fstat_cbk, (void *) (long) call_child,
-			   children[call_child],
-			   children[call_child]->fops->fstat,
-			   fd);
+    unwind:
+        AFR_STACK_UNWIND(getxattr, frame, op_ret, op_errno, xattr, xdata);
 
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (fstat, frame, op_ret, op_errno, NULL);
-	}
+        if (xattr)
+            dict_unref(xattr);
+    }
 
-	return 0;
+    return ret;
 }
 
-/* }}} */
-
-/* {{{ readlink */
+/**
+ * node-uuid cbk uses next child querying mechanism
+ */
+int32_t
+afr_getxattr_node_uuid_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno, dict_t *dict,
+                           dict_t *xdata)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    xlator_t **children = NULL;
+    int unwind = 1;
+    int curr_call_child = 0;
+
+    priv = this->private;
+    children = priv->children;
+
+    local = frame->local;
+
+    if (op_ret == -1) { /** query the _next_ child */
+
+        /**
+         * _current_ becomes _next_
+         * If done with all children and yet no success; give up !
+         */
+        curr_call_child = (int)((long)cookie);
+        if (++curr_call_child == priv->child_count)
+            goto unwind;
+
+        gf_msg_debug(this->name, op_errno,
+                     "op_ret (-1): Re-querying afr-child (%d/%d)",
+                     curr_call_child, priv->child_count);
+
+        unwind = 0;
+        STACK_WIND_COOKIE(
+            frame, afr_getxattr_node_uuid_cbk, (void *)(long)curr_call_child,
+            children[curr_call_child],
+            children[curr_call_child]->fops->getxattr, &local->loc,
+            local->cont.getxattr.name, local->xdata_req);
+    }
+
+unwind:
+    if (unwind)
+        AFR_STACK_UNWIND(getxattr, frame, op_ret, op_errno, dict, xdata);
+
+    return 0;
+}
 
+/**
+ * list-node-uuids cbk returns the list of node_uuids for the subvolume.
+ */
 int32_t
-afr_readlink_cbk (call_frame_t *frame, void *cookie,
-		  xlator_t *this, int32_t op_ret, int32_t op_errno,
-		  const char *buf, struct iatt *sbuf)
+afr_getxattr_list_node_uuids_cbk(call_frame_t *frame, void *cookie,
+                                 xlator_t *this, int32_t op_ret,
+                                 int32_t op_errno, dict_t *dict, dict_t *xdata)
 {
-	afr_private_t * priv     = NULL;
-	afr_local_t *   local    = NULL;
-	xlator_t **     children = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int32_t callcnt = 0;
+    int ret = 0;
+    char *xattr_serz = NULL;
+    long cky = 0;
+    int32_t tlen = 0;
+
+    local = frame->local;
+    priv = this->private;
+    cky = (long)cookie;
+
+    LOCK(&frame->lock);
+    {
+        callcnt = --local->call_count;
+        local->replies[cky].valid = 1;
+        local->replies[cky].op_ret = op_ret;
+        local->replies[cky].op_errno = op_errno;
+
+        if (op_ret < 0)
+            goto unlock;
+
+        local->op_ret = 0;
+
+        if (!local->xdata_rsp && xdata)
+            local->xdata_rsp = dict_ref(xdata);
+        local->replies[cky].xattr = dict_ref(dict);
+    }
+
+unlock:
+    UNLOCK(&frame->lock);
+
+    if (!callcnt) {
+        if (local->op_ret != 0) {
+            /* All bricks gave an error. */
+            local->op_errno = afr_final_errno(local, priv);
+            goto unwind;
+        }
 
-	int unwind     = 1;
-	int last_tried = -1;
-	int this_try = -1;
-        int read_child = -1;
+        /*Since we store the UUID0_STR as node uuid for down bricks and
+         *for non zero op_ret, assigning length to  priv->child_count
+         *number of uuids*/
+        local->cont.getxattr.xattr_len = (SLEN(UUID0_STR) + 2) *
+                                         priv->child_count;
+
+        if (!local->dict)
+            local->dict = dict_new();
+        if (!local->dict) {
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            goto unwind;
+        }
 
-	priv     = this->private;
-	children = priv->children;
+        xattr_serz = GF_CALLOC(local->cont.getxattr.xattr_len, sizeof(char),
+                               gf_common_mt_char);
 
-	local = frame->local;
+        if (!xattr_serz) {
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            goto unwind;
+        }
 
-        read_child = (long) cookie;
+        ret = afr_serialize_xattrs_with_delimiter(frame, this, xattr_serz,
+                                                  UUID0_STR, &tlen, ' ');
+        if (ret) {
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            GF_FREE(xattr_serz);
+            goto unwind;
+        }
+        ret = dict_set_dynstr_sizen(local->dict, GF_XATTR_LIST_NODE_UUIDS_KEY,
+                                    xattr_serz);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+                   "Cannot set node_uuid key in dict");
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            if (ret == -EINVAL)
+                GF_FREE(xattr_serz);
+        } else {
+            local->op_ret = local->cont.getxattr.xattr_len - 1;
+            local->op_errno = 0;
+        }
 
-	if (op_ret == -1) {
-        retry:
-		last_tried = local->cont.readlink.last_tried;
+    unwind:
+        AFR_STACK_UNWIND(getxattr, frame, local->op_ret, local->op_errno,
+                         local->dict, local->xdata_rsp);
+    }
 
-		if (all_tried (last_tried, priv->child_count)) {
-			goto out;
-		}
-		this_try = ++local->cont.readlink.last_tried;
+    return ret;
+}
 
-                if (this_try == read_child) {
-                        goto retry;
-                }
+int32_t
+afr_getxattr_quota_size_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int32_t op_ret, int32_t op_errno, dict_t *dict,
+                            dict_t *xdata)
+{
+    int idx = (long)cookie;
+    int call_count = 0;
+    afr_local_t *local = frame->local;
+    int read_subvol = -1;
+
+    local->replies[idx].valid = 1;
+    local->replies[idx].op_ret = op_ret;
+    local->replies[idx].op_errno = op_errno;
+    if (dict)
+        local->replies[idx].xdata = dict_ref(dict);
+    call_count = afr_frame_return(frame);
+    if (call_count == 0) {
+        local->inode = inode_ref(local->loc.inode);
+        read_subvol = afr_handle_quota_size(frame, this);
+        if (read_subvol != -1) {
+            op_ret = local->replies[read_subvol].op_ret;
+            op_errno = local->replies[read_subvol].op_errno;
+            dict = local->replies[read_subvol].xdata;
+        }
+        AFR_STACK_UNWIND(getxattr, frame, op_ret, op_errno, dict, xdata);
+    }
 
-		unwind = 0;
-		STACK_WIND_COOKIE (frame, afr_readlink_cbk,
-				   (void *) (long) read_child,
-				   children[this_try],
-				   children[this_try]->fops->readlink,
-				   &local->loc,
-				   local->cont.readlink.size);
-	}
+    return 0;
+}
 
-out:
-	if (unwind) {
-                if (sbuf)
-                        sbuf->ia_ino = local->cont.readlink.ino;
+int32_t
+afr_getxattr_lockinfo_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, dict_t *dict,
+                          dict_t *xdata)
+{
+    int call_cnt = 0, len = 0;
+    char *lockinfo_buf = NULL;
+    dict_t *lockinfo = NULL, *newdict = NULL;
+    afr_local_t *local = NULL;
 
-		AFR_STACK_UNWIND (readlink, frame, op_ret, op_errno, buf, sbuf);
-	}
+    LOCK(&frame->lock);
+    {
+        local = frame->local;
 
-	return 0;
-}
+        call_cnt = --local->call_count;
 
+        if ((op_ret < 0) || (!dict && !xdata)) {
+            goto unlock;
+        }
 
-int32_t
-afr_readlink (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, size_t size)
-{
-	afr_private_t * priv       = NULL;
-	xlator_t **     children   = NULL;
-	int             call_child = 0;
-	afr_local_t     *local     = NULL;
+        if (xdata) {
+            if (!local->xdata_rsp) {
+                local->xdata_rsp = dict_new();
+                if (!local->xdata_rsp) {
+                    local->op_ret = -1;
+                    local->op_errno = ENOMEM;
+                    goto unlock;
+                }
+            }
+        }
+
+        if (!dict) {
+            goto unlock;
+        }
+
+        op_ret = dict_get_ptr_and_len(dict, GF_XATTR_LOCKINFO_KEY,
+                                      (void **)&lockinfo_buf, &len);
 
-        int32_t         read_child = -1;
+        if (!lockinfo_buf) {
+            goto unlock;
+        }
 
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
+        if (!local->dict) {
+            local->dict = dict_new();
+            if (!local->dict) {
+                local->op_ret = -1;
+                local->op_errno = ENOMEM;
+                goto unlock;
+            }
+        }
+    }
+unlock:
+    UNLOCK(&frame->lock);
+
+    if (lockinfo_buf != NULL) {
+        lockinfo = dict_new();
+        if (lockinfo == NULL) {
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+        } else {
+            op_ret = dict_unserialize(lockinfo_buf, len, &lockinfo);
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+            if (lockinfo && local->dict) {
+                dict_copy(lockinfo, local->dict);
+            }
+        }
+    }
+
+    if (xdata && local->xdata_rsp) {
+        dict_copy(xdata, local->xdata_rsp);
+    }
+
+    if (!call_cnt) {
+        newdict = dict_new();
+        if (!newdict) {
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            goto unwind;
+        }
 
-	priv     = this->private;
-	VALIDATE_OR_GOTO (priv->children, out);
+        op_ret = dict_allocate_and_serialize(
+            local->dict, (char **)&lockinfo_buf, (unsigned int *)&len);
+        if (op_ret != 0) {
+            local->op_ret = -1;
+            goto unwind;
+        }
 
-	children = priv->children;
+        op_ret = dict_set_dynptr(newdict, GF_XATTR_LOCKINFO_KEY,
+                                 (void *)lockinfo_buf, len);
+        if (op_ret < 0) {
+            local->op_ret = -1;
+            local->op_errno = -op_ret;
+            goto unwind;
+        }
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+    unwind:
+        AFR_STACK_UNWIND(getxattr, frame, op_ret, op_errno, newdict,
+                         local->xdata_rsp);
+    }
 
-	frame->local = local;
+    dict_unref(lockinfo);
 
-        read_child = afr_read_child (this, loc->inode);
+    return 0;
+}
 
-        if ((read_child >= 0) && (priv->child_up[read_child])) {
-                call_child = read_child;
+int32_t
+afr_fgetxattr_lockinfo_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno, dict_t *dict,
+                           dict_t *xdata)
+{
+    int call_cnt = 0, len = 0;
+    char *lockinfo_buf = NULL;
+    dict_t *lockinfo = NULL, *newdict = NULL;
+    afr_local_t *local = NULL;
 
-                local->cont.readlink.last_tried = -1;
+    LOCK(&frame->lock);
+    {
+        local = frame->local;
 
-        } else {
-                call_child = afr_first_up_child (priv);
+        call_cnt = --local->call_count;
 
-                if (call_child == -1) {
-                        op_errno = ENOTCONN;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "no child is up");
-                        goto out;
+        if ((op_ret < 0) || (!dict && !xdata)) {
+            goto unlock;
+        }
+
+        if (xdata) {
+            if (!local->xdata_rsp) {
+                local->xdata_rsp = dict_new();
+                if (!local->xdata_rsp) {
+                    local->op_ret = -1;
+                    local->op_errno = ENOMEM;
+                    goto unlock;
                 }
+            }
+        }
 
-                local->cont.readlink.last_tried = call_child;
+        if (!dict) {
+            goto unlock;
         }
 
-	loc_copy (&local->loc, loc);
+        op_ret = dict_get_ptr_and_len(dict, GF_XATTR_LOCKINFO_KEY,
+                                      (void **)&lockinfo_buf, &len);
 
-	local->cont.readlink.size       = size;
-        local->cont.readlink.ino        = loc->inode->ino;
+        if (!lockinfo_buf) {
+            goto unlock;
+        }
 
-	STACK_WIND_COOKIE (frame, afr_readlink_cbk,
-			   (void *) (long) call_child,
-			   children[call_child], children[call_child]->fops->readlink,
-			   loc, size);
+        if (!local->dict) {
+            local->dict = dict_new();
+            if (!local->dict) {
+                local->op_ret = -1;
+                local->op_errno = ENOMEM;
+                goto unlock;
+            }
+        }
+    }
+unlock:
+    UNLOCK(&frame->lock);
+
+    if (lockinfo_buf != NULL) {
+        lockinfo = dict_new();
+        if (lockinfo == NULL) {
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+        } else {
+            op_ret = dict_unserialize(lockinfo_buf, len, &lockinfo);
 
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (readlink, frame, op_ret, op_errno, NULL, NULL);
-	}
-	return 0;
-}
+            if (lockinfo && local->dict) {
+                dict_copy(lockinfo, local->dict);
+            }
+        }
+    }
+
+    if (xdata && local->xdata_rsp) {
+        dict_copy(xdata, local->xdata_rsp);
+    }
+
+    if (!call_cnt) {
+        newdict = dict_new();
+        if (!newdict) {
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            goto unwind;
+        }
 
+        op_ret = dict_allocate_and_serialize(
+            local->dict, (char **)&lockinfo_buf, (unsigned int *)&len);
+        if (op_ret != 0) {
+            local->op_ret = -1;
+            goto unwind;
+        }
 
-/* }}} */
+        op_ret = dict_set_dynptr(newdict, GF_XATTR_LOCKINFO_KEY,
+                                 (void *)lockinfo_buf, len);
+        if (op_ret < 0) {
+            local->op_ret = -1;
+            local->op_errno = -op_ret;
+            goto unwind;
+        }
 
-/* {{{ getxattr */
+    unwind:
+        AFR_STACK_UNWIND(fgetxattr, frame, op_ret, op_errno, newdict,
+                         local->xdata_rsp);
+    }
 
-struct _xattr_key {
-        char *key;
-        struct list_head list;
-};
+    dict_unref(lockinfo);
 
+    return 0;
+}
 
-void
-__gather_xattr_keys (dict_t *dict, char *key, data_t *value,
-                     void *data)
+int32_t
+afr_fgetxattr_pathinfo_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno, dict_t *dict,
+                           dict_t *xdata)
 {
-        struct list_head *  list  = data;
-        struct _xattr_key * xkey  = NULL;
+    afr_local_t *local = NULL;
+    int32_t callcnt = 0;
+    int ret = 0;
+    char *xattr = NULL;
+    char *xattr_serz = NULL;
+    int keylen = 0;
+    char xattr_cky[1024] = {
+        0,
+    };
+    int xattr_cky_len = 0;
+    dict_t *nxattr = NULL;
+    long cky = 0;
+    int32_t padding = 0;
+    int32_t tlen = 0;
+
+    if (!frame || !frame->local || !this) {
+        gf_msg("", GF_LOG_ERROR, 0, AFR_MSG_INVALID_ARG, "possible NULL deref");
+        goto out;
+    }
+
+    local = frame->local;
+    cky = (long)cookie;
+    keylen = strlen(local->cont.getxattr.name);
+    xattr_cky_len = snprintf(xattr_cky, sizeof(xattr_cky), "%s-%ld",
+                             local->cont.getxattr.name, cky);
+    LOCK(&frame->lock);
+    {
+        callcnt = --local->call_count;
+
+        if (op_ret < 0) {
+            local->op_errno = op_errno;
+        } else {
+            local->op_ret = op_ret;
+            if (!local->xdata_rsp && xdata)
+                local->xdata_rsp = dict_ref(xdata);
+        }
 
-        if (!strncmp (key, AFR_XATTR_PREFIX,
-                      strlen (AFR_XATTR_PREFIX))) {
+        if (!dict || (op_ret < 0))
+            goto unlock;
 
-                xkey = GF_CALLOC (1, sizeof (*xkey), gf_afr_mt_xattr_key);
-                if (!xkey)
-                        return;
+        if (!local->dict) {
+            local->dict = dict_new();
+            if (!local->dict)
+                goto unlock;
+        }
+        ret = dict_get_strn(dict, local->cont.getxattr.name, keylen, &xattr);
+        if (ret)
+            goto unlock;
+
+        xattr = gf_strdup(xattr);
+
+        ret = dict_set_dynstrn(local->dict, xattr_cky, xattr_cky_len, xattr);
+        if (ret) {
+            UNLOCK(&frame->lock);
+            gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+                   "Cannot set xattr cookie key");
+            goto post_unlock;
+        }
 
-                xkey->key = key;
-                INIT_LIST_HEAD (&xkey->list);
+        local->cont.getxattr.xattr_len += strlen(xattr) + 1;
+    }
+unlock:
+    UNLOCK(&frame->lock);
+post_unlock:
+    if (!callcnt) {
+        if (!local->cont.getxattr.xattr_len)
+            goto unwind;
+
+        nxattr = dict_new();
+        if (!nxattr)
+            goto unwind;
+
+        /* extra bytes for decorations (brackets and <>'s) */
+        padding += strlen(this->name) + SLEN(AFR_PATHINFO_HEADER) + 4;
+        local->cont.getxattr.xattr_len += (padding + 2);
+
+        xattr_serz = GF_MALLOC(local->cont.getxattr.xattr_len,
+                               gf_common_mt_char);
+
+        if (!xattr_serz)
+            goto unwind;
+
+        /* the xlator info */
+        int xattr_serz_len = sprintf(
+            xattr_serz, "(<" AFR_PATHINFO_HEADER "%s> ", this->name);
+
+        /* actual series of pathinfo */
+        ret = dict_serialize_value_with_delim(
+            local->dict, xattr_serz + xattr_serz_len, &tlen, ' ');
+        if (ret) {
+            GF_FREE(xattr_serz);
+            goto unwind;
+        }
 
-                list_add_tail (&xkey->list, list);
+        /* closing part */
+        *(xattr_serz + padding + tlen) = ')';
+        *(xattr_serz + padding + tlen + 1) = '\0';
+
+        ret = dict_set_dynstrn(nxattr, local->cont.getxattr.name, keylen,
+                               xattr_serz);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+                   "Cannot set pathinfo key in dict");
+            if (ret == -EINVAL)
+                GF_FREE(xattr_serz);
         }
-}
 
+    unwind:
+        AFR_STACK_UNWIND(fgetxattr, frame, local->op_ret, local->op_errno,
+                         nxattr, local->xdata_rsp);
 
-void
-__filter_xattrs (dict_t *dict)
-{
-        struct list_head keys;
+        if (nxattr)
+            dict_unref(nxattr);
+    }
 
-        struct _xattr_key *key;
-        struct _xattr_key *tmp;
+out:
+    return ret;
+}
 
-        INIT_LIST_HEAD (&keys);
+int32_t
+afr_getxattr_pathinfo_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, dict_t *dict,
+                          dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int32_t callcnt = 0;
+    int ret = 0;
+    char *xattr = NULL;
+    char *xattr_serz = NULL;
+    char xattr_cky[1024] = {
+        0,
+    };
+    int keylen = 0;
+    int xattr_cky_len = 0;
+    dict_t *nxattr = NULL;
+    long cky = 0;
+    int32_t padding = 0;
+    int32_t tlen = 0;
+
+    if (!frame || !frame->local || !this) {
+        gf_msg("", GF_LOG_ERROR, 0, AFR_MSG_INVALID_ARG, "possible NULL deref");
+        goto out;
+    }
+
+    local = frame->local;
+    cky = (long)cookie;
+    keylen = strlen(local->cont.getxattr.name);
+    xattr_cky_len = snprintf(xattr_cky, sizeof(xattr_cky), "%s-%ld",
+                             local->cont.getxattr.name, cky);
+    LOCK(&frame->lock);
+    {
+        callcnt = --local->call_count;
+
+        if (op_ret < 0) {
+            local->op_errno = op_errno;
+        } else {
+            local->op_ret = op_ret;
+            if (!local->xdata_rsp && xdata)
+                local->xdata_rsp = dict_ref(xdata);
+        }
 
-        dict_foreach (dict, __gather_xattr_keys,
-                      (void *) &keys);
+        if (!dict || (op_ret < 0))
+            goto unlock;
 
-        list_for_each_entry_safe (key, tmp, &keys, list) {
-                dict_del (dict, key->key);
+        if (!local->dict) {
+            local->dict = dict_new();
+            if (!local->dict)
+                goto unlock;
+        }
+        ret = dict_get_strn(dict, local->cont.getxattr.name, keylen, &xattr);
+        if (ret)
+            goto unlock;
+
+        xattr = gf_strdup(xattr);
+
+        ret = dict_set_dynstrn(local->dict, xattr_cky, xattr_cky_len, xattr);
+        if (ret) {
+            UNLOCK(&frame->lock);
+            gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+                   "Cannot set xattr cookie key");
+            goto post_unlock;
+        }
 
-                list_del_init (&key->list);
+        local->cont.getxattr.xattr_len += strlen(xattr) + 1;
+    }
+unlock:
+    UNLOCK(&frame->lock);
+post_unlock:
+    if (!callcnt) {
+        if (!local->cont.getxattr.xattr_len)
+            goto unwind;
+
+        nxattr = dict_new();
+        if (!nxattr)
+            goto unwind;
+
+        /* extra bytes for decorations (brackets and <>'s) */
+        padding += strlen(this->name) + SLEN(AFR_PATHINFO_HEADER) + 4;
+        local->cont.getxattr.xattr_len += (padding + 2);
+
+        xattr_serz = GF_MALLOC(local->cont.getxattr.xattr_len,
+                               gf_common_mt_char);
+
+        if (!xattr_serz)
+            goto unwind;
+
+        /* the xlator info */
+        int xattr_serz_len = sprintf(
+            xattr_serz, "(<" AFR_PATHINFO_HEADER "%s> ", this->name);
+
+        /* actual series of pathinfo */
+        ret = dict_serialize_value_with_delim(
+            local->dict, xattr_serz + xattr_serz_len, &tlen, ' ');
+        if (ret) {
+            GF_FREE(xattr_serz);
+            goto unwind;
+        }
 
-                GF_FREE (key);
+        /* closing part */
+        *(xattr_serz + padding + tlen) = ')';
+        *(xattr_serz + padding + tlen + 1) = '\0';
+
+        ret = dict_set_dynstrn(nxattr, local->cont.getxattr.name, keylen,
+                               xattr_serz);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+                   "Cannot set pathinfo key in dict");
+            if (ret == -EINVAL)
+                GF_FREE(xattr_serz);
         }
+
+    unwind:
+        AFR_STACK_UNWIND(getxattr, frame, local->op_ret, local->op_errno,
+                         nxattr, local->xdata_rsp);
+
+        if (nxattr)
+            dict_unref(nxattr);
+    }
+
+out:
+    return ret;
 }
 
+static int
+afr_aggregate_stime_xattr(dict_t *this, char *key, data_t *value, void *data)
+{
+    int ret = 0;
+
+    if (fnmatch(GF_XATTR_STIME_PATTERN, key, FNM_NOESCAPE) == 0)
+        ret = gf_get_max_stime(THIS, data, key, value);
 
+    return ret;
+}
 
 int32_t
-afr_getxattr_cbk (call_frame_t *frame, void *cookie,
-		  xlator_t *this, int32_t op_ret, int32_t op_errno,
-		  dict_t *dict)
+afr_common_getxattr_stime_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                              int32_t op_ret, int32_t op_errno, dict_t *dict,
+                              dict_t *xdata)
 {
-	afr_private_t * priv     = NULL;
-	afr_local_t *   local    = NULL;
-	xlator_t **     children = NULL;
+    afr_local_t *local = NULL;
+    int32_t callcnt = 0;
 
-	int unwind     = 1;
-	int last_tried = -1;
-	int this_try = -1;
-        int read_child = -1;
+    if (!frame || !frame->local || !this) {
+        gf_msg("", GF_LOG_ERROR, 0, AFR_MSG_INVALID_ARG, "possible NULL deref");
+        goto out;
+    }
 
-	priv     = this->private;
-	children = priv->children;
+    local = frame->local;
 
-	local = frame->local;
+    LOCK(&frame->lock);
+    {
+        callcnt = --local->call_count;
 
-        read_child = (long) cookie;
-
-	if (op_ret == -1) {
-        retry:
-		last_tried = local->cont.getxattr.last_tried;
+        if (!dict || (op_ret < 0)) {
+            local->op_errno = op_errno;
+            goto cleanup;
+        }
 
-		if (all_tried (last_tried, priv->child_count)) {
-			goto out;
-		}
-		this_try = ++local->cont.getxattr.last_tried;
+        if (!local->dict)
+            local->dict = dict_copy_with_ref(dict, NULL);
+        else
+            dict_foreach(dict, afr_aggregate_stime_xattr, local->dict);
+        local->op_ret = 0;
+    }
 
-                if (this_try == read_child) {
-                        goto retry;
-                }
+cleanup:
+    UNLOCK(&frame->lock);
 
-		unwind = 0;
-		STACK_WIND_COOKIE (frame, afr_getxattr_cbk,
-				   (void *) (long) read_child,
-				   children[this_try],
-				   children[this_try]->fops->getxattr,
-				   &local->loc,
-				   local->cont.getxattr.name);
-	}
+    if (!callcnt) {
+        AFR_STACK_UNWIND(getxattr, frame, local->op_ret, local->op_errno,
+                         local->dict, xdata);
+    }
 
 out:
-	if (unwind) {
-                if (op_ret >= 0 && dict)
-                        __filter_xattrs (dict);
+    return 0;
+}
 
-		AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict);
-	}
+static gf_boolean_t
+afr_is_special_xattr(const char *name, fop_getxattr_cbk_t *cbk,
+                     gf_boolean_t is_fgetxattr)
+{
+    gf_boolean_t is_spl = _gf_true;
+
+    GF_ASSERT(cbk);
+    if (!cbk || !name) {
+        is_spl = _gf_false;
+        goto out;
+    }
+
+    if (!strcmp(name, GF_XATTR_PATHINFO_KEY) ||
+        !strcmp(name, GF_XATTR_USER_PATHINFO_KEY)) {
+        if (is_fgetxattr) {
+            *cbk = afr_fgetxattr_pathinfo_cbk;
+        } else {
+            *cbk = afr_getxattr_pathinfo_cbk;
+        }
+    } else if (!strncmp(name, GF_XATTR_CLRLK_CMD, SLEN(GF_XATTR_CLRLK_CMD))) {
+        if (is_fgetxattr) {
+            *cbk = afr_fgetxattr_clrlk_cbk;
+        } else {
+            *cbk = afr_getxattr_clrlk_cbk;
+        }
+    } else if (!strncmp(name, GF_XATTR_LOCKINFO_KEY,
+                        SLEN(GF_XATTR_LOCKINFO_KEY))) {
+        if (is_fgetxattr) {
+            *cbk = afr_fgetxattr_lockinfo_cbk;
+        } else {
+            *cbk = afr_getxattr_lockinfo_cbk;
+        }
+    } else if (fnmatch(GF_XATTR_STIME_PATTERN, name, FNM_NOESCAPE) == 0) {
+        *cbk = afr_common_getxattr_stime_cbk;
+    } else if (strcmp(name, QUOTA_SIZE_KEY) == 0) {
+        *cbk = afr_getxattr_quota_size_cbk;
+    } else if (!strcmp(name, GF_XATTR_LIST_NODE_UUIDS_KEY)) {
+        *cbk = afr_getxattr_list_node_uuids_cbk;
+    } else {
+        is_spl = _gf_false;
+    }
 
-	return 0;
+out:
+    return is_spl;
 }
 
-int32_t
-afr_getxattr (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, const char *name)
+static void
+afr_getxattr_all_subvols(xlator_t *this, call_frame_t *frame, const char *name,
+                         loc_t *loc, fop_getxattr_cbk_t cbk)
 {
-	afr_private_t *   priv       = NULL;
-	xlator_t **       children   = NULL;
-	int               call_child = 0;
-	afr_local_t     * local      = NULL;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i = 0;
+    int call_count = 0;
+
+    priv = this->private;
+
+    local = frame->local;
+    // local->call_count set in afr_local_init
+    call_count = local->call_count;
+
+    if (!strcmp(name, GF_XATTR_LIST_NODE_UUIDS_KEY)) {
+        GF_FREE(local->cont.getxattr.name);
+        local->cont.getxattr.name = gf_strdup(GF_XATTR_NODE_UUID_KEY);
+    }
+
+    // If up-children count is 0, afr_local_init would have failed already
+    // and the call would have unwound so not handling it here.
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->child_up[i]) {
+            STACK_WIND_COOKIE(frame, cbk, (void *)(long)i, priv->children[i],
+                              priv->children[i]->fops->getxattr, loc,
+                              local->cont.getxattr.name, NULL);
+            if (!--call_count)
+                break;
+        }
+    }
+    return;
+}
 
-        int               read_child = -1;
+int
+afr_marker_populate_args(call_frame_t *frame, int type, int *gauge,
+                         xlator_t **subvols)
+{
+    xlator_t *this = frame->this;
+    afr_private_t *priv = this->private;
 
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
+    memcpy(subvols, priv->children, sizeof(*subvols) * priv->child_count);
 
+    if (type == MARKER_XTIME_TYPE) {
+        /*Don't error out on ENOENT/ENOTCONN */
+        gauge[MCNT_NOTFOUND] = 0;
+        gauge[MCNT_ENOTCONN] = 0;
+    }
+    return priv->child_count;
+}
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+static int
+afr_handle_heal_xattrs(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                       const char *heal_op)
+{
+    int ret = -1;
+    afr_spb_status_t *data = NULL;
+
+    if (!strcmp(heal_op, GF_HEAL_INFO)) {
+        afr_get_heal_info(frame, this, loc);
+        ret = 0;
+        goto out;
+    }
+
+    if (!strcmp(heal_op, GF_AFR_HEAL_SBRAIN)) {
+        afr_heal_splitbrain_file(frame, this, loc);
+        ret = 0;
+        goto out;
+    }
+
+    if (!strcmp(heal_op, GF_AFR_SBRAIN_STATUS)) {
+        data = GF_CALLOC(1, sizeof(*data), gf_afr_mt_spb_status_t);
+        if (!data) {
+            ret = 1;
+            goto out;
+        }
+        data->frame = frame;
+        data->loc = loc;
+        ret = synctask_new(this->ctx->env, afr_get_split_brain_status,
+                           afr_get_split_brain_status_cbk, NULL, data);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN_STATUS,
+                   "Failed to create"
+                   " synctask. Unable to fetch split-brain status"
+                   " for %s.",
+                   loc->name);
+            ret = 1;
+            goto out;
+        }
+        goto out;
+    }
 
-	priv     = this->private;
-	VALIDATE_OR_GOTO (priv->children, out);
+out:
+    if (ret == 1) {
+        AFR_STACK_UNWIND(getxattr, frame, -1, ENOMEM, NULL, NULL);
+        if (data)
+            GF_FREE(data);
+        ret = 0;
+    }
+    return ret;
+}
 
-	children = priv->children;
+int32_t
+afr_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, const char *name,
+             dict_t *xdata)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    xlator_t **children = NULL;
+    int i = 0;
+    int32_t op_errno = 0;
+    int ret = -1;
+    fop_getxattr_cbk_t cbk = NULL;
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    priv = this->private;
+
+    children = priv->children;
+
+    loc_copy(&local->loc, loc);
+
+    local->op = GF_FOP_GETXATTR;
+
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
+
+    if (!name)
+        goto no_name;
+
+    local->cont.getxattr.name = gf_strdup(name);
+
+    if (!local->cont.getxattr.name) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    if (!strncmp(name, AFR_XATTR_PREFIX, SLEN(AFR_XATTR_PREFIX))) {
+        op_errno = ENODATA;
+        goto out;
+    }
+
+    if (cluster_handle_marker_getxattr(frame, loc, name, priv->vol_uuid,
+                                       afr_getxattr_unwind,
+                                       afr_marker_populate_args) == 0)
+        return 0;
+
+    ret = afr_handle_heal_xattrs(frame, this, &local->loc, name);
+    if (ret == 0)
+        return 0;
+
+    /*
+     * Heal daemons don't have IO threads ... and as a result they
+     * send this getxattr down and eventually crash :(
+     */
+    op_errno = -1;
+    GF_CHECK_XATTR_KEY_AND_GOTO(name, IO_THREADS_QUEUE_SIZE_KEY, op_errno, out);
+
+    /*
+     * Special xattrs which need responses from all subvols
+     */
+    if (afr_is_special_xattr(name, &cbk, 0)) {
+        afr_getxattr_all_subvols(this, frame, name, loc, cbk);
+        return 0;
+    }
+
+    if (XATTR_IS_NODE_UUID(name)) {
+        i = 0;
+        STACK_WIND_COOKIE(frame, afr_getxattr_node_uuid_cbk, (void *)(long)i,
+                          children[i], children[i]->fops->getxattr, loc, name,
+                          xdata);
+        return 0;
+    }
+
+no_name:
+
+    afr_read_txn(frame, this, local->loc.inode, afr_getxattr_wind,
+                 AFR_METADATA_TRANSACTION);
+
+    ret = 0;
+out:
+    if (ret < 0)
+        AFR_STACK_UNWIND(getxattr, frame, -1, op_errno, NULL, NULL);
+    return 0;
+}
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-	frame->local = local;
+/* {{{ fgetxattr */
 
-        if (name) {
-                if (!strncmp (name, AFR_XATTR_PREFIX,
-                              strlen (AFR_XATTR_PREFIX))) {
+int32_t
+afr_fgetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
 
-                        op_errno = ENODATA;
-                        goto out;
-                }
+    local = frame->local;
 
-        }
+    if (op_ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = op_errno;
 
-        read_child = afr_read_child (this, loc->inode);
+        afr_read_txn_continue(frame, this, (long)cookie);
+        return 0;
+    }
 
-        if ((read_child >= 0) && (priv->child_up[read_child])) {
-                call_child = read_child;
+    if (dict)
+        afr_filter_xattrs(dict);
 
-                local->cont.getxattr.last_tried = -1;
-        } else {
-                call_child = afr_first_up_child (priv);
+    AFR_STACK_UNWIND(fgetxattr, frame, op_ret, op_errno, dict, xdata);
 
-                if (call_child == -1) {
-                        op_errno = ENOTCONN;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "no child is up");
-                        goto out;
-                }
+    return 0;
+}
 
-                local->cont.getxattr.last_tried = call_child;
-        }
+int
+afr_fgetxattr_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    if (subvol == -1) {
+        AFR_STACK_UNWIND(fgetxattr, frame, local->op_ret, local->op_errno, NULL,
+                         NULL);
+        return 0;
+    }
+
+    STACK_WIND_COOKIE(frame, afr_fgetxattr_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->fgetxattr, local->fd,
+                      local->cont.getxattr.name, local->xdata_req);
+    return 0;
+}
 
-	loc_copy (&local->loc, loc);
-	if (name)
-	  local->cont.getxattr.name       = gf_strdup (name);
+static void
+afr_fgetxattr_all_subvols(xlator_t *this, call_frame_t *frame,
+                          fop_fgetxattr_cbk_t cbk)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i = 0;
+    int call_count = 0;
+
+    priv = this->private;
+
+    local = frame->local;
+    // local->call_count set in afr_local_init
+    call_count = local->call_count;
+
+    // If up-children count is 0, afr_local_init would have failed already
+    // and the call would have unwound so not handling it here.
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->child_up[i]) {
+            STACK_WIND_COOKIE(frame, cbk, (void *)(long)i, priv->children[i],
+                              priv->children[i]->fops->fgetxattr, local->fd,
+                              local->cont.getxattr.name, NULL);
+            if (!--call_count)
+                break;
+        }
+    }
 
-	STACK_WIND_COOKIE (frame, afr_getxattr_cbk,
-			   (void *) (long) call_child,
-			   children[call_child], children[call_child]->fops->getxattr,
-			   loc, name);
+    return;
+}
 
-	op_ret = 0;
+int
+afr_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+              dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int32_t op_errno = 0;
+    fop_fgetxattr_cbk_t cbk = NULL;
+
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    local->op = GF_FOP_FGETXATTR;
+    local->fd = fd_ref(fd);
+    if (name) {
+        local->cont.getxattr.name = gf_strdup(name);
+        if (!local->cont.getxattr.name) {
+            op_errno = ENOMEM;
+            goto out;
+        }
+    }
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
+
+    /* pathinfo gets handled only in getxattr(), but we need to handle
+     * lockinfo.
+     * If we are doing fgetxattr with lockinfo as the key then we
+     * collect information from all children.
+     */
+    if (afr_is_special_xattr(name, &cbk, 1)) {
+        afr_fgetxattr_all_subvols(this, frame, cbk);
+        return 0;
+    }
+
+    afr_fix_open(fd, this);
+
+    afr_read_txn(frame, this, fd->inode, afr_fgetxattr_wind,
+                 AFR_METADATA_TRANSACTION);
+
+    return 0;
 out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, NULL);
-	}
-	return 0;
-}
+    AFR_STACK_UNWIND(fgetxattr, frame, -1, op_errno, NULL, NULL);
 
+    return 0;
+}
 
 /* }}} */
 
 /* {{{ readv */
 
-/**
- * read algorithm:
- *
- * if the user has specified a read subvolume, use it
- * otherwise -
- *   use the inode number to hash it to one of the subvolumes, and
- *   read from there (to balance read load)
- *
- * if any of the above read's fail, try the children in sequence
- * beginning at the beginning
- */
-
-int32_t
-afr_readv_cbk (call_frame_t *frame, void *cookie,
-	       xlator_t *this, int32_t op_ret, int32_t op_errno,
-	       struct iovec *vector, int32_t count, struct iatt *buf,
-               struct iobref *iobref)
+int
+afr_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct iovec *vector, int32_t count,
+              struct iatt *buf, struct iobref *iobref, dict_t *xdata)
 {
-	afr_private_t * priv     = NULL;
-	afr_local_t *   local    = NULL;
-	xlator_t **     children = NULL;
-
-	int unwind     = 1;
-	int last_tried = -1;
-	int this_try   = -1;
-        int read_child = -1;
+    afr_local_t *local = NULL;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
+    local = frame->local;
 
-	priv     = this->private;
-	VALIDATE_OR_GOTO (priv->children, out);
+    if (op_ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = op_errno;
 
-	children = priv->children;
+        afr_read_txn_continue(frame, this, (long)cookie);
+        return 0;
+    }
 
-	local = frame->local;
+    AFR_STACK_UNWIND(readv, frame, op_ret, op_errno, vector, count, buf, iobref,
+                     xdata);
+    return 0;
+}
 
-        read_child = (long) cookie;
+int
+afr_readv_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    if (subvol == -1) {
+        AFR_STACK_UNWIND(readv, frame, local->op_ret, local->op_errno, 0, 0, 0,
+                         0, 0);
+        return 0;
+    }
+
+    STACK_WIND_COOKIE(
+        frame, afr_readv_cbk, (void *)(long)subvol, priv->children[subvol],
+        priv->children[subvol]->fops->readv, local->fd, local->cont.readv.size,
+        local->cont.readv.offset, local->cont.readv.flags, local->xdata_req);
+    return 0;
+}
 
-	if (op_ret == -1) {
-	retry:
-		last_tried = local->cont.readv.last_tried;
+int
+afr_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+          off_t offset, uint32_t flags, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int32_t op_errno = 0;
 
-		if (all_tried (last_tried, priv->child_count)) {
-			goto out;
-		}
-		this_try = ++local->cont.readv.last_tried;
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-		if (this_try == read_child) {
-			/*
-			   skip the read child since if we are here
-			   we must have already tried that child
-			*/
-			goto retry;
-		}
+    local->op = GF_FOP_READ;
+    local->fd = fd_ref(fd);
+    local->cont.readv.size = size;
+    local->cont.readv.offset = offset;
+    local->cont.readv.flags = flags;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
 
-		unwind = 0;
+    afr_fix_open(fd, this);
 
-		STACK_WIND_COOKIE (frame, afr_readv_cbk,
-				   (void *) (long) read_child,
-				   children[this_try],
-				   children[this_try]->fops->readv,
-				   local->fd, local->cont.readv.size,
-				   local->cont.readv.offset);
-	}
+    afr_read_txn(frame, this, fd->inode, afr_readv_wind, AFR_DATA_TRANSACTION);
 
+    return 0;
 out:
-	if (unwind) {
-                if (buf && local)
-                        buf->ia_ino = local->cont.readv.ino;
-
-		AFR_STACK_UNWIND (readv, frame, op_ret, op_errno,
-                                  vector, count, buf, iobref);
-	}
+    AFR_STACK_UNWIND(readv, frame, -1, op_errno, 0, 0, 0, 0, 0);
 
-	return 0;
+    return 0;
 }
 
+/* }}} */
 
-int32_t
-afr_readv (call_frame_t *frame, xlator_t *this,
-	   fd_t *fd, size_t size, off_t offset)
+/* {{{ seek */
+
+int
+afr_seek_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, off_t offset, dict_t *xdata)
 {
-	afr_private_t * priv       = NULL;
-	afr_local_t   * local      = NULL;
-	xlator_t **     children   = NULL;
+    afr_local_t *local = NULL;
 
-        int32_t         read_child = -1;
-	int             call_child = 0;
+    local = frame->local;
 
-	int32_t         op_ret     = -1;
-	int32_t         op_errno   = 0;
+    if (op_ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = op_errno;
 
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-	VALIDATE_OR_GOTO (fd, out);
+        afr_read_txn_continue(frame, this, (long)cookie);
+        return 0;
+    }
 
-	priv     = this->private;
-	children = priv->children;
+    AFR_STACK_UNWIND(seek, frame, op_ret, op_errno, offset, xdata);
+    return 0;
+}
 
-	ALLOC_OR_GOTO (local, afr_local_t, out);
+int
+afr_seek_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    if (subvol == -1) {
+        AFR_STACK_UNWIND(seek, frame, local->op_ret, local->op_errno, 0, NULL);
+        return 0;
+    }
+
+    STACK_WIND_COOKIE(
+        frame, afr_seek_cbk, (void *)(long)subvol, priv->children[subvol],
+        priv->children[subvol]->fops->seek, local->fd, local->cont.seek.offset,
+        local->cont.seek.what, local->xdata_req);
+    return 0;
+}
 
-	frame->local = local;
+int
+afr_seek(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+         gf_seek_what_t what, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int32_t op_errno = 0;
 
-        read_child = afr_read_child (this, fd->inode);
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
 
-        if ((read_child >= 0) && (priv->child_up[read_child])) {
-                call_child = read_child;
+    local->op = GF_FOP_SEEK;
+    local->fd = fd_ref(fd);
+    local->cont.seek.offset = offset;
+    local->cont.seek.what = what;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
 
-		/*
-		   if read fails from the read child, we try
-		   all children starting with the first one
-		*/
-                local->cont.readv.last_tried = -1;
+    afr_fix_open(fd, this);
 
-        } else {
-		call_child = afr_first_up_child (priv);
-		if (call_child == -1) {
-			op_errno = ENOTCONN;
-			gf_log (this->name, GF_LOG_DEBUG,
-				"no child is up");
-			goto out;
-		}
-
-		local->cont.readv.last_tried = call_child;
-	}
-
-	local->fd                    = fd_ref (fd);
-
-        local->cont.readv.ino        = fd->inode->ino;
-	local->cont.readv.size       = size;
-	local->cont.readv.offset     = offset;
-
-	STACK_WIND_COOKIE (frame, afr_readv_cbk,
-			   (void *) (long) call_child,
-			   children[call_child],
-			   children[call_child]->fops->readv,
-			   fd, size, offset);
-
-	op_ret = 0;
+    afr_read_txn(frame, this, fd->inode, afr_seek_wind, AFR_DATA_TRANSACTION);
+
+    return 0;
 out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (readv, frame, op_ret, op_errno, NULL, 0, NULL,
-                                  NULL);
-	}
-	return 0;
-}
+    AFR_STACK_UNWIND(seek, frame, -1, op_errno, 0, NULL);
 
+    return 0;
+}
 /* }}} */
diff --git a/xlators/cluster/afr/src/afr-inode-read.h b/xlators/cluster/afr/src/afr-inode-read.h
index 9b29d5055aa..8c982bc7e6f 100644
--- a/xlators/cluster/afr/src/afr-inode-read.h
+++ b/xlators/cluster/afr/src/afr-inode-read.h
@@ -1,47 +1,45 @@
 /*
-   Copyright (c) 2007-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef __INODE_READ_H__
 #define __INODE_READ_H__
 
 int32_t
-afr_access (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc, int32_t mask);
+afr_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+           dict_t *xdata);
 
 int32_t
-afr_stat (call_frame_t *frame, xlator_t *this,
-	  loc_t *loc);
+afr_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata);
 
 int32_t
-afr_fstat (call_frame_t *frame, xlator_t *this,
-	   fd_t *fd);
+afr_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata);
 
 int32_t
-afr_readlink (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, size_t size);
+afr_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+             dict_t *xdata);
 
 int32_t
-afr_readv (call_frame_t *frame, xlator_t *this,
-	   fd_t *fd, size_t size, off_t offset);
+afr_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+          off_t offset, uint32_t flags, dict_t *xdata);
 
 int32_t
-afr_getxattr (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, const char *name);
+afr_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, const char *name,
+             dict_t *xdata);
 
+int32_t
+afr_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+              dict_t *xdata);
+
+int
+afr_seek(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+         gf_seek_what_t what, dict_t *xdata);
+int
+afr_handle_quota_size(call_frame_t *frame, xlator_t *this);
 #endif /* __INODE_READ_H__ */
diff --git a/xlators/cluster/afr/src/afr-inode-write.c b/xlators/cluster/afr/src/afr-inode-write.c
index 2eaf99c55d3..1d6e4f3570a 100644
--- a/xlators/cluster/afr/src/afr-inode-write.c
+++ b/xlators/cluster/afr/src/afr-inode-write.c
@@ -1,1594 +1,2565 @@
 /*
-   Copyright (c) 2007-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
 
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
 
-#include <libgen.h>
 #include <unistd.h>
-#include <fnmatch.h>
 #include <sys/time.h>
 #include <stdlib.h>
 #include <signal.h>
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "afr.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-
+#include <glusterfs/glusterfs.h>
 #include "afr.h"
+#include <glusterfs/dict.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/compat.h>
+#include "protocol-common.h"
+#include <glusterfs/byte-order.h>
 #include "afr-transaction.h"
+#include "afr-self-heal.h"
+#include "afr-messages.h"
 
-/* {{{ writev */
+static void
+__afr_inode_write_finalize(call_frame_t *frame, xlator_t *this)
+{
+    int i = 0;
+    int ret = 0;
+    int read_subvol = 0;
+    struct iatt *stbuf = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    afr_read_subvol_args_t args = {
+        0,
+    };
+
+    local = frame->local;
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, local->inode, out);
+
+    /*This code needs to stay till DHT sends fops on linked
+     * inodes*/
+    if (!inode_is_linked(local->inode)) {
+        for (i = 0; i < priv->child_count; i++) {
+            if (!local->replies[i].valid)
+                continue;
+            if (local->replies[i].op_ret == -1)
+                continue;
+            if (!gf_uuid_is_null(local->replies[i].poststat.ia_gfid)) {
+                gf_uuid_copy(args.gfid, local->replies[i].poststat.ia_gfid);
+                args.ia_type = local->replies[i].poststat.ia_type;
+                break;
+            } else {
+                ret = dict_get_bin(local->replies[i].xdata,
+                                   DHT_IATT_IN_XDATA_KEY, (void **)&stbuf);
+                if (ret)
+                    continue;
+                gf_uuid_copy(args.gfid, stbuf->ia_gfid);
+                args.ia_type = stbuf->ia_type;
+                break;
+            }
+        }
+    }
+
+    if (local->transaction.type == AFR_METADATA_TRANSACTION) {
+        read_subvol = afr_metadata_subvol_get(local->inode, this, NULL,
+                                              local->readable, NULL, &args);
+    } else {
+        read_subvol = afr_data_subvol_get(local->inode, this, NULL,
+                                          local->readable, NULL, &args);
+    }
+
+    local->op_ret = -1;
+    local->op_errno = afr_final_errno(local, priv);
+    afr_pick_error_xdata(local, priv, local->inode, local->readable, NULL,
+                         NULL);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
+        if (local->replies[i].op_ret < 0)
+            continue;
+
+        /* Order of checks in the compound conditional
+           below is important.
+
+           - Highest precedence: largest op_ret
+           - Next precedence: if all op_rets are equal, read subvol
+           - Least precedence: any succeeded subvol
+        */
+        if ((local->op_ret < local->replies[i].op_ret) ||
+            ((local->op_ret == local->replies[i].op_ret) &&
+             (i == read_subvol))) {
+            local->op_ret = local->replies[i].op_ret;
+            local->op_errno = local->replies[i].op_errno;
+
+            local->cont.inode_wfop.prebuf = local->replies[i].prestat;
+            local->cont.inode_wfop.postbuf = local->replies[i].poststat;
+
+            if (local->replies[i].xdata) {
+                if (local->xdata_rsp)
+                    dict_unref(local->xdata_rsp);
+                local->xdata_rsp = dict_ref(local->replies[i].xdata);
+            }
+            if (local->replies[i].xattr) {
+                if (local->xattr_rsp)
+                    dict_unref(local->xattr_rsp);
+                local->xattr_rsp = dict_ref(local->replies[i].xattr);
+            }
+        }
+    }
+
+    afr_set_in_flight_sb_status(this, frame, local->inode);
+out:
+    return;
+}
 
-int
-afr_writev_unwind (call_frame_t *frame, xlator_t *this)
+static void
+__afr_inode_write_fill(call_frame_t *frame, xlator_t *this, int child_index,
+                       int op_ret, int op_errno, struct iatt *prebuf,
+                       struct iatt *postbuf, dict_t *xattr, dict_t *xdata)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    local->replies[child_index].valid = 1;
+
+    if (AFR_IS_ARBITER_BRICK(priv, child_index) && op_ret == 1)
+        op_ret = iov_length(local->cont.writev.vector,
+                            local->cont.writev.count);
+
+    local->replies[child_index].op_ret = op_ret;
+    local->replies[child_index].op_errno = op_errno;
+    if (xdata)
+        local->replies[child_index].xdata = dict_ref(xdata);
+
+    if (op_ret >= 0) {
+        if (prebuf)
+            local->replies[child_index].prestat = *prebuf;
+        if (postbuf)
+            local->replies[child_index].poststat = *postbuf;
+        if (xattr)
+            local->replies[child_index].xattr = dict_ref(xattr);
+    } else {
+        afr_transaction_fop_failed(frame, this, child_index);
+    }
+
+    return;
+}
 
-        local = frame->local;
+static int
+__afr_inode_write_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                      struct iatt *postbuf, dict_t *xattr, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int child_index = (long)cookie;
+    int call_count = -1;
+    afr_private_t *priv = NULL;
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
+    priv = this->private;
+    local = frame->local;
 
-        if (main_frame) {
-                local->cont.writev.prebuf.ia_ino  = local->cont.writev.ino;
-                local->cont.writev.postbuf.ia_ino = local->cont.writev.ino;
+    LOCK(&frame->lock);
+    {
+        __afr_inode_write_fill(frame, this, child_index, op_ret, op_errno,
+                               prebuf, postbuf, xattr, xdata);
+        call_count = --local->call_count;
+    }
+    UNLOCK(&frame->lock);
 
-                AFR_STACK_UNWIND (writev, main_frame,
-                                  local->op_ret, local->op_errno,
-                                  &local->cont.writev.prebuf,
-                                  &local->cont.writev.postbuf);
+    if (call_count == 0) {
+        __afr_inode_write_finalize(frame, this);
+
+        if (afr_txn_nothing_failed(frame, this)) {
+            /*if it did pre-op, it will do post-op changing ctime*/
+            if (priv->consistent_metadata && afr_needs_changelog_update(local))
+                afr_zero_fill_stat(local);
+            local->transaction.unwind(frame, this);
         }
-        return 0;
+
+        afr_transaction_resume(frame, this);
+    }
+
+    return 0;
 }
 
+/* {{{ writev */
 
-int
-afr_writev_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                     struct iatt *postbuf)
+void
+afr_writev_copy_outvars(call_frame_t *src_frame, call_frame_t *dst_frame)
 {
-        afr_local_t *   local = NULL;
-
-        int child_index = (long) cookie;
-        int call_count  = -1;
-        int read_child  = 0;
+    afr_local_t *src_local = NULL;
+    afr_local_t *dst_local = NULL;
+
+    src_local = src_frame->local;
+    dst_local = dst_frame->local;
+
+    dst_local->op_ret = src_local->op_ret;
+    dst_local->op_errno = src_local->op_errno;
+    dst_local->cont.inode_wfop.prebuf = src_local->cont.inode_wfop.prebuf;
+    dst_local->cont.inode_wfop.postbuf = src_local->cont.inode_wfop.postbuf;
+    if (src_local->xdata_rsp)
+        dst_local->xdata_rsp = dict_ref(src_local->xdata_rsp);
+}
 
-        local = frame->local;
+void
+afr_writev_unwind(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = this->private;
 
-        read_child = afr_read_child (this, local->fd->inode);
+    local = frame->local;
 
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
+    if (priv->consistent_metadata)
+        afr_zero_fill_stat(local);
 
-                if (afr_fop_failed (op_ret, op_errno))
-                        afr_transaction_fop_failed (frame, this, child_index);
+    AFR_STACK_UNWIND(writev, frame, local->op_ret, local->op_errno,
+                     &local->cont.inode_wfop.prebuf,
+                     &local->cont.inode_wfop.postbuf, local->xdata_rsp);
+}
 
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret              = op_ret;
-                                local->cont.writev.prebuf  = *prebuf;
-                                local->cont.writev.postbuf = *postbuf;
-                        }
+int
+afr_transaction_writev_unwind(call_frame_t *frame, xlator_t *this)
+{
+    call_frame_t *fop_frame = NULL;
 
-                        if (child_index == read_child) {
-                                local->cont.writev.prebuf  = *prebuf;
-                                local->cont.writev.postbuf = *postbuf;
-                        }
-                }
+    fop_frame = afr_transaction_detach_fop_frame(frame);
 
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
+    if (fop_frame) {
+        afr_writev_copy_outvars(frame, fop_frame);
+        afr_writev_unwind(fop_frame, this);
+    }
+    return 0;
+}
 
-        if (call_count == 0) {
-                local->transaction.unwind (frame, this);
+static void
+afr_writev_handle_short_writes(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+
+    local = frame->local;
+    priv = this->private;
+    /*
+     * We already have the best case result of the writev calls staged
+     * as the return value. Any writev that returns some value less
+     * than the best case is now out of sync, so mark the fop as
+     * failed. Note that fops that have returned with errors have
+     * already been marked as failed.
+     */
+    for (i = 0; i < priv->child_count; i++) {
+        if ((!local->replies[i].valid) || (local->replies[i].op_ret == -1))
+            continue;
+
+        if (local->replies[i].op_ret < local->op_ret)
+            afr_transaction_fop_failed(frame, this, i);
+    }
+}
 
-                local->transaction.resume (frame, this);
-        }
-        return 0;
+void
+afr_inode_write_fill(call_frame_t *frame, xlator_t *this, int child_index,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+    int ret = 0;
+    afr_local_t *local = frame->local;
+    uint32_t open_fd_count = 0;
+    uint32_t write_is_append = 0;
+    int32_t num_inodelks = 0;
+
+    LOCK(&frame->lock);
+    {
+        __afr_inode_write_fill(frame, this, child_index, op_ret, op_errno,
+                               prebuf, postbuf, NULL, xdata);
+        if (op_ret == -1 || !xdata)
+            goto unlock;
+
+        write_is_append = 0;
+        ret = dict_get_uint32(xdata, GLUSTERFS_WRITE_IS_APPEND,
+                              &write_is_append);
+        if (ret || !write_is_append)
+            local->append_write = _gf_false;
+
+        ret = dict_get_uint32(xdata, GLUSTERFS_ACTIVE_FD_COUNT, &open_fd_count);
+        if (ret < 0)
+            goto unlock;
+        if (open_fd_count > local->open_fd_count) {
+            local->open_fd_count = open_fd_count;
+            local->update_open_fd_count = _gf_true;
+        }
+
+        ret = dict_get_int32_sizen(xdata, GLUSTERFS_INODELK_COUNT,
+                                   &num_inodelks);
+        if (ret < 0)
+            goto unlock;
+        if (num_inodelks > local->num_inodelks) {
+            local->num_inodelks = num_inodelks;
+            local->update_num_inodelks = _gf_true;
+        }
+    }
+unlock:
+    UNLOCK(&frame->lock);
 }
 
+void
+afr_process_post_writev(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_lock_t *lock = NULL;
+
+    local = frame->local;
+
+    if (!local->stable_write && !local->append_write)
+        /* An appended write removes the necessity to
+           fsync() the file. This is because self-heal
+           has the logic to check for larger file when
+           the xattrs are not reliably pointing at
+           a stale file.
+        */
+        afr_fd_report_unstable_write(this, local);
+
+    __afr_inode_write_finalize(frame, this);
+
+    afr_writev_handle_short_writes(frame, this);
+
+    if (local->update_open_fd_count)
+        local->inode_ctx->open_fd_count = local->open_fd_count;
+    if (local->update_num_inodelks &&
+        local->transaction.type == AFR_DATA_TRANSACTION) {
+        lock = &local->inode_ctx->lock[local->transaction.type];
+        lock->num_inodelks = local->num_inodelks;
+    }
+}
 
 int
-afr_writev_wind (call_frame_t *frame, xlator_t *this)
+afr_writev_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                    struct iatt *postbuf, dict_t *xdata)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
+    call_frame_t *fop_frame = NULL;
+    int child_index = (long)cookie;
+    int call_count = -1;
 
-        int i = 0;
-        int call_count = -1;
+    afr_inode_write_fill(frame, this, child_index, op_ret, op_errno, prebuf,
+                         postbuf, xdata);
 
-        local = frame->local;
-        priv = this->private;
+    call_count = afr_frame_return(frame);
 
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
+    if (call_count == 0) {
+        afr_process_post_writev(frame, this);
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
-
-        local->call_count = call_count;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND_COOKIE (frame, afr_writev_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->writev,
-                                           local->fd,
-                                           local->cont.writev.vector,
-                                           local->cont.writev.count,
-                                           local->cont.writev.offset,
-                                           local->cont.writev.iobref);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-        return 0;
+        if (!afr_txn_nothing_failed(frame, this)) {
+            // Don't unwind until post-op is complete
+            afr_transaction_resume(frame, this);
+        } else {
+            /*
+             * Generally inode-write fops do transaction.unwind then
+             * transaction.resume, but writev needs to make sure that
+             * delayed post-op frame is placed in fdctx before unwind
+             * happens. This prevents the race of flush doing the
+             * changelog wakeup first in fuse thread and then this
+             * writev placing its delayed post-op frame in fdctx.
+             * This helps flush make sure all the delayed post-ops are
+             * completed.
+             */
+
+            fop_frame = afr_transaction_detach_fop_frame(frame);
+            afr_writev_copy_outvars(frame, fop_frame);
+            afr_transaction_resume(frame, this);
+            afr_writev_unwind(fop_frame, this);
+        }
+    }
+    return 0;
 }
 
+static int
+afr_arbiter_writev_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = frame->local;
+    afr_private_t *priv = this->private;
+    static char byte = 0xFF;
+    static struct iovec vector = {&byte, 1};
+    int32_t count = 1;
+
+    STACK_WIND_COOKIE(
+        frame, afr_writev_wind_cbk, (void *)(long)subvol,
+        priv->children[subvol], priv->children[subvol]->fops->writev, local->fd,
+        &vector, count, local->cont.writev.offset, local->cont.writev.flags,
+        local->cont.writev.iobref, local->xdata_req);
+
+    return 0;
+}
 
 int
-afr_writev_done (call_frame_t *frame, xlator_t *this)
+afr_writev_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_local_t *local = NULL;
-
-        local = frame->local;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        iobref_unref (local->cont.writev.iobref);
-        local->cont.writev.iobref = NULL;
-
-        local->transaction.unwind (frame, this);
-
-        AFR_STACK_DESTROY (frame);
+    local = frame->local;
+    priv = this->private;
 
+    if (AFR_IS_ARBITER_BRICK(priv, subvol)) {
+        afr_arbiter_writev_wind(frame, this, subvol);
         return 0;
+    }
+
+    STACK_WIND_COOKIE(frame, afr_writev_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->writev, local->fd,
+                      local->cont.writev.vector, local->cont.writev.count,
+                      local->cont.writev.offset, local->cont.writev.flags,
+                      local->cont.writev.iobref, local->xdata_req);
+    return 0;
 }
 
+int
+afr_do_writev(call_frame_t *frame, xlator_t *this)
+{
+    call_frame_t *transaction_frame = NULL;
+    afr_local_t *local = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
+
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
+
+    local = frame->local;
+    transaction_frame->local = local;
+    frame->local = NULL;
+
+    if (!AFR_FRAME_INIT(frame, op_errno))
+        goto out;
+
+    local->op = GF_FOP_WRITE;
+
+    local->transaction.wind = afr_writev_wind;
+    local->transaction.unwind = afr_transaction_writev_unwind;
+
+    local->transaction.main_frame = frame;
+
+    if (local->fd->flags & O_APPEND) {
+        /*
+         * Backend vfs ignores the 'offset' for append mode fd so
+         * locking just the region provided for the writev does not
+         * give consistency guarantee. The actual write may happen at a
+         * completely different range than the one provided by the
+         * offset, len in the fop. So lock the entire file.
+         */
+        local->transaction.start = 0;
+        local->transaction.len = 0;
+    } else {
+        local->transaction.start = local->cont.writev.offset;
+        local->transaction.len = iov_length(local->cont.writev.vector,
+                                            local->cont.writev.count);
+    }
+
+    ret = afr_transaction(transaction_frame, this, AFR_DATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
+out:
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
+
+    AFR_STACK_UNWIND(writev, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
+}
 
 int
-afr_do_writev (call_frame_t *frame, xlator_t *this)
+afr_writev(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
+           int32_t count, off_t offset, uint32_t flags, struct iobref *iobref,
+           dict_t *xdata)
 {
-        call_frame_t * transaction_frame = NULL;
-        afr_local_t *  local             = NULL;
+    afr_local_t *local = NULL;
+    int op_errno = ENOMEM;
+    int ret = -1;
+
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    local->cont.writev.vector = iov_dup(vector, count);
+    if (!local->cont.writev.vector)
+        goto out;
+    local->cont.writev.count = count;
+    local->cont.writev.offset = offset;
+    local->cont.writev.flags = flags;
+    local->cont.writev.iobref = iobref_ref(iobref);
+
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
+
+    if (!local->xdata_req)
+        goto out;
+
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
+
+    if (dict_set_uint32(local->xdata_req, GLUSTERFS_ACTIVE_FD_COUNT, 4)) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    if (dict_set_str_sizen(local->xdata_req, GLUSTERFS_INODELK_DOM_COUNT,
+                           this->name)) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    if (dict_set_uint32(local->xdata_req, GLUSTERFS_WRITE_IS_APPEND, 4)) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    /* Set append_write to be true speculatively. If on any
+       server it turns not be true, we unset it in the
+       callback.
+    */
+    local->append_write = _gf_true;
+
+    /* detect here, but set it in writev_wind_cbk *after* the unstable
+       write is performed
+    */
+    local->stable_write = !!((fd->flags | flags) & (O_SYNC | O_DSYNC));
+
+    afr_fix_open(fd, this);
+
+    afr_do_writev(frame, this);
+
+    return 0;
+out:
+    AFR_STACK_UNWIND(writev, frame, -1, op_errno, NULL, NULL, NULL);
 
-        int op_ret   = -1;
-        int op_errno = 0;
+    return 0;
+}
 
-        local = frame->local;
+/* }}} */
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory.");
-                op_errno = ENOMEM;
-                goto out;
-        }
+/* {{{ truncate */
 
-        transaction_frame->local = local;
-        frame->local = NULL;
+int
+afr_truncate_unwind(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        local->op = GF_FOP_WRITE;
+    local = frame->local;
 
-        local->success_count      = 0;
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-        local->transaction.fop    = afr_writev_wind;
-        local->transaction.done   = afr_writev_done;
-        local->transaction.unwind = afr_writev_unwind;
+    AFR_STACK_UNWIND(truncate, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.inode_wfop.prebuf,
+                     &local->cont.inode_wfop.postbuf, local->xdata_rsp);
+    return 0;
+}
 
-        local->transaction.main_frame = frame;
-        if (local->fd->flags & O_APPEND) {
-                local->transaction.start   = 0;
-                local->transaction.len     = 0;
-        } else {
-                local->transaction.start   = local->cont.writev.offset;
-                local->transaction.len     = iov_length (local->cont.writev.vector,
-                                                         local->cont.writev.count);
-        }
+int
+afr_truncate_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                      struct iatt *postbuf, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
 
-        afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+    local = frame->local;
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (writev, frame, op_ret, op_errno, NULL, NULL);
-        }
+    if (op_ret == 0 && prebuf->ia_size != postbuf->ia_size)
+        local->stable_write = _gf_false;
 
-        return 0;
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, prebuf,
+                                 postbuf, NULL, xdata);
 }
 
-
 int
-afr_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
-            struct iovec *vector, int32_t count, off_t offset,
-            struct iobref *iobref)
+afr_truncate_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        int ret = -1;
+    local = frame->local;
+    priv = this->private;
 
-        int op_ret   = -1;
-        int op_errno = 0;
+    STACK_WIND_COOKIE(frame, afr_truncate_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->truncate, &local->loc,
+                      local->cont.truncate.offset, local->xdata_req);
+    return 0;
+}
 
-        uint64_t ctx;
-        afr_fd_ctx_t *fd_ctx = NULL;
+int
+afr_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+             dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        priv = this->private;
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+    local->cont.truncate.offset = offset;
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+    if (!local->xdata_req)
+        goto out;
 
-        frame->local = local;
+    local->transaction.wind = afr_truncate_wind;
+    local->transaction.unwind = afr_truncate_unwind;
 
-        local->cont.writev.vector     = iov_dup (vector, count);
-        local->cont.writev.count      = count;
-        local->cont.writev.offset     = offset;
-        local->cont.writev.ino        = fd->inode->ino;
-        local->cont.writev.iobref     = iobref_ref (iobref);
+    loc_copy(&local->loc, loc);
+    ret = afr_set_inode_local(this, local, loc->inode);
+    if (ret)
+        goto out;
 
-        local->fd                = fd_ref (fd);
+    local->op = GF_FOP_TRUNCATE;
 
-        ret = fd_ctx_get (fd, this, &ctx);
-        if (ret < 0) {
-                goto out;
-        }
+    local->transaction.main_frame = frame;
+    local->transaction.start = offset;
+    local->transaction.len = 0;
 
-        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+    /* Set it true speculatively, will get reset in afr_truncate_wind_cbk
+       if truncate was not a NOP */
+    local->stable_write = _gf_true;
 
-        if (fd_ctx->up_count < priv->up_count) {
-                local->openfd_flush_cbk = afr_do_writev;
-                afr_openfd_flush (frame, this, fd);
-        } else {
-                afr_do_writev (frame, this);
-        }
+    ret = afr_transaction(transaction_frame, this, AFR_DATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        op_ret = 0;
+    return 0;
 out:
-        if (op_ret == -1) {
-                AFR_STACK_UNWIND (writev, frame, op_ret, op_errno, NULL, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(truncate, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
-
 /* }}} */
 
-/* {{{ truncate */
+/* {{{ ftruncate */
 
 int
-afr_truncate_unwind (call_frame_t *frame, xlator_t *this)
+afr_ftruncate_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
-
-        local = frame->local;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
-
-        if (main_frame) {
-                local->cont.truncate.prebuf.ia_ino  = local->cont.truncate.ino;
-                local->cont.truncate.postbuf.ia_ino = local->cont.truncate.ino;
-
-                AFR_STACK_UNWIND (truncate, main_frame, local->op_ret,
-                                  local->op_errno,
-                                  &local->cont.truncate.prebuf,
-                                  &local->cont.truncate.postbuf);
-        }
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
 
+    AFR_STACK_UNWIND(ftruncate, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.inode_wfop.prebuf,
+                     &local->cont.inode_wfop.postbuf, local->xdata_rsp);
+    return 0;
+}
 
 int
-afr_truncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+afr_ftruncate_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
                        int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                       struct iatt *postbuf)
-{
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-
-        int child_index = (long) cookie;
-        int read_child  = 0;
-        int call_count  = -1;
-        int need_unwind = 0;
-
-        local = frame->local;
-        priv  = this->private;
-
-        read_child = afr_read_child (this, local->loc.inode);
-
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
-
-                if (afr_fop_failed (op_ret, op_errno) && op_errno != EFBIG)
-                        afr_transaction_fop_failed (frame, this, child_index);
-
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                                local->cont.truncate.prebuf  = *prebuf;
-                                local->cont.truncate.postbuf = *postbuf;
-                        }
-
-                        if (child_index == read_child) {
-                                local->cont.truncate.prebuf  = *prebuf;
-                                local->cont.truncate.postbuf = *postbuf;
-                        }
-
-                        local->success_count++;
-
-                        if ((local->success_count >= priv->wait_count)
-                            && local->read_child_returned) {
-                                need_unwind = 1;
-                        }
-                }
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
-
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
+                       struct iatt *postbuf, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
 
-        call_count = afr_frame_return (frame);
+    local = frame->local;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+    if (op_ret == 0 && prebuf->ia_size != postbuf->ia_size)
+        local->stable_write = _gf_false;
 
-        return 0;
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, prebuf,
+                                 postbuf, NULL, xdata);
 }
 
-
-int32_t
-afr_truncate_wind (call_frame_t *frame, xlator_t *this)
+int
+afr_ftruncate_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-
-        int call_count = -1;
-        int i = 0;
-
-        local = frame->local;
-        priv = this->private;
-
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
-
-        local->call_count = call_count;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND_COOKIE (frame, afr_truncate_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->truncate,
-                                           &local->loc,
-                                           local->cont.truncate.offset);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+    local = frame->local;
+    priv = this->private;
 
-        return 0;
+    STACK_WIND_COOKIE(frame, afr_ftruncate_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->ftruncate, local->fd,
+                      local->cont.ftruncate.offset, local->xdata_req);
+    return 0;
 }
 
-
 int
-afr_truncate_done (call_frame_t *frame, xlator_t *this)
+afr_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+              dict_t *xdata)
 {
-        afr_local_t *local = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        local = frame->local;
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        local->transaction.unwind (frame, this);
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        AFR_STACK_DESTROY (frame);
-
-        return 0;
-}
+    local->cont.ftruncate.offset = offset;
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
+    if (!local->xdata_req)
+        goto out;
 
-int
-afr_truncate (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, off_t offset)
-{
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
-        call_frame_t   *transaction_frame = NULL;
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
 
-        int ret = -1;
+    local->op = GF_FOP_FTRUNCATE;
 
-        int op_ret   = -1;
-        int op_errno = 0;
+    local->transaction.wind = afr_ftruncate_wind;
+    local->transaction.unwind = afr_ftruncate_unwind;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    local->transaction.main_frame = frame;
 
-        priv = this->private;
+    local->transaction.start = local->cont.ftruncate.offset;
+    local->transaction.len = 0;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory.");
-                goto out;
-        }
+    afr_fix_open(fd, this);
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+    /* Set it true speculatively, will get reset in afr_ftruncate_wind_cbk
+       if truncate was not a NOP */
+    local->stable_write = _gf_true;
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+    ret = afr_transaction(transaction_frame, this, AFR_DATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        transaction_frame->local = local;
+    return 0;
+out:
+    AFR_STACK_UNWIND(ftruncate, frame, -1, op_errno, NULL, NULL, NULL);
 
-        local->op_ret = -1;
+    return 0;
+}
 
-        local->cont.truncate.offset  = offset;
-        local->cont.truncate.ino     = loc->inode->ino;
+/* }}} */
 
-        local->transaction.fop    = afr_truncate_wind;
-        local->transaction.done   = afr_truncate_done;
-        local->transaction.unwind = afr_truncate_unwind;
+/* {{{ setattr */
 
-        loc_copy (&local->loc, loc);
+int
+afr_setattr_unwind(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        local->transaction.main_frame = frame;
-        local->transaction.start   = 0;
-        local->transaction.len     = offset;
+    local = frame->local;
 
-        afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (truncate, frame, op_ret, op_errno, NULL, NULL);
-        }
+    AFR_STACK_UNWIND(setattr, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.inode_wfop.prebuf,
+                     &local->cont.inode_wfop.postbuf, local->xdata_rsp);
+    return 0;
+}
 
-        return 0;
+int
+afr_setattr_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, struct iatt *preop,
+                     struct iatt *postop, dict_t *xdata)
+{
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, preop,
+                                 postop, NULL, xdata);
 }
 
+int
+afr_setattr_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    STACK_WIND_COOKIE(frame, afr_setattr_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->setattr, &local->loc,
+                      &local->cont.setattr.in_buf, local->cont.setattr.valid,
+                      local->xdata_req);
+    return 0;
+}
 
-/* }}} */
+int
+afr_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *buf,
+            int32_t valid, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
+
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
+
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
+
+    local->cont.setattr.in_buf = *buf;
+    local->cont.setattr.valid = valid;
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
+
+    if (!local->xdata_req)
+        goto out;
+
+    local->transaction.wind = afr_setattr_wind;
+    local->transaction.unwind = afr_setattr_unwind;
+
+    loc_copy(&local->loc, loc);
+    ret = afr_set_inode_local(this, local, loc->inode);
+    if (ret)
+        goto out;
+
+    local->op = GF_FOP_SETATTR;
+
+    local->transaction.main_frame = frame;
+    local->transaction.start = LLONG_MAX - 1;
+    local->transaction.len = 0;
+
+    ret = afr_transaction(transaction_frame, this, AFR_METADATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
+out:
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-/* {{{ ftruncate */
+    AFR_STACK_UNWIND(setattr, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
+}
 
+/* {{{ fsetattr */
 
 int
-afr_ftruncate_unwind (call_frame_t *frame, xlator_t *this)
+afr_fsetattr_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        local = frame->local;
+    local = frame->local;
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-        if (main_frame) {
-                local->cont.ftruncate.prebuf.ia_ino  = local->cont.ftruncate.ino;
-                local->cont.ftruncate.postbuf.ia_ino = local->cont.ftruncate.ino;
+    AFR_STACK_UNWIND(fsetattr, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.inode_wfop.prebuf,
+                     &local->cont.inode_wfop.postbuf, local->xdata_rsp);
+    return 0;
+}
 
-                AFR_STACK_UNWIND (ftruncate, main_frame, local->op_ret,
-                                  local->op_errno,
-                                  &local->cont.ftruncate.prebuf,
-                                  &local->cont.ftruncate.postbuf);
-        }
-        return 0;
+int
+afr_fsetattr_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *preop,
+                      struct iatt *postop, dict_t *xdata)
+{
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, preop,
+                                 postop, NULL, xdata);
 }
 
+int
+afr_fsetattr_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    STACK_WIND_COOKIE(frame, afr_fsetattr_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->fsetattr, local->fd,
+                      &local->cont.fsetattr.in_buf, local->cont.fsetattr.valid,
+                      local->xdata_req);
+    return 0;
+}
 
 int
-afr_ftruncate_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                        struct iatt *postbuf)
+afr_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *buf,
+             int32_t valid, dict_t *xdata)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        int child_index = (long) cookie;
-        int call_count  = -1;
-        int need_unwind = 0;
-        int read_child  = 0;
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        local = frame->local;
-        priv  = this->private;
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        read_child = afr_read_child (this, local->fd->inode);
+    local->cont.fsetattr.in_buf = *buf;
+    local->cont.fsetattr.valid = valid;
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
+    if (!local->xdata_req)
+        goto out;
 
-                if (afr_fop_failed (op_ret, op_errno))
-                        afr_transaction_fop_failed (frame, this, child_index);
+    local->transaction.wind = afr_fsetattr_wind;
+    local->transaction.unwind = afr_fsetattr_unwind;
 
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                                local->cont.ftruncate.prebuf  = *prebuf;
-                                local->cont.ftruncate.postbuf = *postbuf;
-                        }
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
 
-                        if (child_index == read_child) {
-                                local->cont.ftruncate.prebuf  = *prebuf;
-                                local->cont.ftruncate.postbuf = *postbuf;
-                        }
+    local->op = GF_FOP_FSETATTR;
 
-                        local->success_count++;
+    afr_fix_open(fd, this);
 
-                        if ((local->success_count >= priv->wait_count)
-                            && local->read_child_returned) {
-                                need_unwind = 1;
-                        }
-                }
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    local->transaction.main_frame = frame;
+    local->transaction.start = LLONG_MAX - 1;
+    local->transaction.len = 0;
 
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
+    ret = afr_transaction(transaction_frame, this, AFR_METADATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        call_count = afr_frame_return (frame);
+    return 0;
+out:
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+    AFR_STACK_UNWIND(fsetattr, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
+}
+
+/* {{{ setxattr */
+
+int
+afr_setxattr_unwind(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
+    local = frame->local;
+
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
+
+    AFR_STACK_UNWIND(setxattr, main_frame, local->op_ret, local->op_errno,
+                     local->xdata_rsp);
+    return 0;
 }
 
+int
+afr_setxattr_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, NULL,
+                                 NULL, NULL, xdata);
+}
 
 int
-afr_ftruncate_wind (call_frame_t *frame, xlator_t *this)
+afr_setxattr_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    STACK_WIND_COOKIE(frame, afr_setxattr_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->setxattr, &local->loc,
+                      local->cont.setxattr.dict, local->cont.setxattr.flags,
+                      local->xdata_req);
+    return 0;
+}
 
-        int call_count = -1;
-        int i = 0;
+int
+afr_emptyb_set_pending_changelog_cbk(call_frame_t *frame, void *cookie,
+                                     xlator_t *this, int op_ret, int op_errno,
+                                     dict_t *xattr, dict_t *xdata)
 
-        local = frame->local;
-        priv = this->private;
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i, ret = 0;
+    char *op_type = NULL;
 
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
+    local = frame->local;
+    priv = this->private;
+    i = (long)cookie;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+    local->replies[i].valid = 1;
+    local->replies[i].op_ret = op_ret;
+    local->replies[i].op_errno = op_errno;
 
-        local->call_count = call_count;
+    ret = dict_get_str_sizen(local->xdata_req, "replicate-brick-op", &op_type);
+    if (ret)
+        goto out;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND_COOKIE (frame, afr_ftruncate_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->ftruncate,
-                                           local->fd, local->cont.ftruncate.offset);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+    gf_smsg(this->name, op_ret ? GF_LOG_ERROR : GF_LOG_INFO,
+            op_ret ? op_errno : 0, AFR_MSG_SET_PEND_XATTR, "name=%s",
+            priv->children[i]->name, "op_ret=%s",
+            op_ret ? "failed" : "succeeded", NULL);
 
-        return 0;
+out:
+    syncbarrier_wake(&local->barrier);
+    return 0;
 }
 
-
 int
-afr_ftruncate_done (call_frame_t *frame, xlator_t *this)
+afr_emptyb_set_pending_changelog(call_frame_t *frame, xlator_t *this,
+                                 unsigned char *locked_nodes)
 {
-        afr_local_t *local = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int ret = 0, i = 0;
 
-        local = frame->local;
+    local = frame->local;
+    priv = this->private;
 
-        local->transaction.unwind (frame, this);
+    AFR_ONLIST(locked_nodes, frame, afr_emptyb_set_pending_changelog_cbk,
+               xattrop, &local->loc, GF_XATTROP_ADD_ARRAY, local->xattr_req,
+               NULL);
 
-        AFR_STACK_DESTROY (frame);
+    /* It is sufficient if xattrop was successful on one child */
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
 
-        return 0;
+        if (local->replies[i].op_ret == 0) {
+            ret = 0;
+            goto out;
+        } else {
+            ret = afr_higher_errno(ret, local->replies[i].op_errno);
+        }
+    }
+out:
+    return -ret;
 }
 
+static int
+_afr_handle_empty_brick_type(xlator_t *this, call_frame_t *frame, loc_t *loc,
+                             int empty_index, afr_transaction_type type,
+                             char *op_type, const int op_type_len)
+{
+    int count = 0;
+    int ret = -ENOMEM;
+    int idx = -1;
+    int d_idx = -1;
+    unsigned char *locked_nodes = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    locked_nodes = alloca0(priv->child_count);
+
+    idx = afr_index_for_transaction_type(type);
+    d_idx = afr_index_for_transaction_type(AFR_DATA_TRANSACTION);
+
+    local->pending = afr_matrix_create(priv->child_count, AFR_NUM_CHANGE_LOGS);
+    if (!local->pending)
+        goto out;
+
+    local->pending[empty_index][idx] = hton32(1);
+
+    if ((priv->esh_granular) && (type == AFR_ENTRY_TRANSACTION))
+        local->pending[empty_index][d_idx] = hton32(1);
+
+    local->xdata_req = dict_new();
+    if (!local->xdata_req)
+        goto out;
+
+    ret = dict_set_nstrn(local->xdata_req, "replicate-brick-op",
+                         SLEN("replicate-brick-op"), op_type, op_type_len);
+    if (ret)
+        goto out;
+
+    local->xattr_req = dict_new();
+    if (!local->xattr_req)
+        goto out;
+
+    ret = afr_set_pending_dict(priv, local->xattr_req, local->pending);
+    if (ret < 0)
+        goto out;
+
+    if (AFR_ENTRY_TRANSACTION == type) {
+        count = afr_selfheal_entrylk(frame, this, loc->inode, this->name, NULL,
+                                     locked_nodes);
+    } else {
+        count = afr_selfheal_inodelk(frame, this, loc->inode, this->name,
+                                     LLONG_MAX - 1, 0, locked_nodes);
+    }
+
+    if (!count) {
+        gf_smsg(this->name, GF_LOG_ERROR, EAGAIN, AFR_MSG_REPLACE_BRICK_STATUS,
+                NULL);
+        ret = -EAGAIN;
+        goto unlock;
+    }
+
+    ret = afr_emptyb_set_pending_changelog(frame, this, locked_nodes);
+    if (ret)
+        goto unlock;
+    ret = 0;
+unlock:
+    if (AFR_ENTRY_TRANSACTION == type) {
+        afr_selfheal_unentrylk(frame, this, loc->inode, this->name, NULL,
+                               locked_nodes, NULL);
+    } else {
+        afr_selfheal_uninodelk(frame, this, loc->inode, this->name,
+                               LLONG_MAX - 1, 0, locked_nodes);
+    }
+out:
+    return ret;
+}
 
-int
-afr_do_ftruncate (call_frame_t *frame, xlator_t *this)
+void
+afr_brick_args_cleanup(void *opaque)
 {
-        call_frame_t * transaction_frame = NULL;
-        afr_local_t *  local             = NULL;
+    afr_empty_brick_args_t *data = NULL;
 
-        int op_ret   = -1;
-        int op_errno = 0;
+    data = opaque;
+    loc_wipe(&data->loc);
+    GF_FREE(data);
+}
 
-        local = frame->local;
+int
+_afr_handle_empty_brick_cbk(int ret, call_frame_t *frame, void *opaque)
+{
+    afr_brick_args_cleanup(opaque);
+    return 0;
+}
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory.");
-                goto out;
-        }
+int
+_afr_handle_empty_brick(void *opaque)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int empty_index = -1;
+    int ret = -1;
+    int op_errno = ENOMEM;
+    call_frame_t *frame = NULL;
+    xlator_t *this = NULL;
+    char *op_type = NULL;
+    int op_type_len = 0;
+    afr_empty_brick_args_t *data = NULL;
+    call_frame_t *op_frame = NULL;
+
+    data = opaque;
+    frame = data->frame;
+    empty_index = data->empty_index;
+    if (!data->op_type)
+        goto out;
+
+    op_frame = copy_frame(frame);
+    if (!op_frame) {
+        ret = -1;
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    op_type = data->op_type;
+    op_type_len = strlen(op_type);
+    this = op_frame->this;
+    priv = this->private;
+
+    afr_set_lk_owner(op_frame, this, op_frame->root);
+    local = AFR_FRAME_INIT(op_frame, op_errno);
+    if (!local)
+        goto out;
+
+    loc_copy(&local->loc, &data->loc);
+
+    gf_smsg(this->name, GF_LOG_INFO, 0, AFR_MSG_NEW_BRICK, "name=%s",
+            priv->children[empty_index]->name, NULL);
+
+    ret = _afr_handle_empty_brick_type(this, op_frame, &local->loc, empty_index,
+                                       AFR_METADATA_TRANSACTION, op_type,
+                                       op_type_len);
+    if (ret) {
+        op_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    dict_unref(local->xdata_req);
+    dict_unref(local->xattr_req);
+    afr_matrix_cleanup(local->pending, priv->child_count);
+    local->pending = NULL;
+    local->xattr_req = NULL;
+    local->xdata_req = NULL;
+
+    ret = _afr_handle_empty_brick_type(this, op_frame, &local->loc, empty_index,
+                                       AFR_ENTRY_TRANSACTION, op_type,
+                                       op_type_len);
+    if (ret) {
+        op_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+    ret = 0;
+out:
+    if (op_frame) {
+        AFR_STACK_DESTROY(op_frame);
+    }
+    AFR_STACK_UNWIND(setxattr, frame, ret, op_errno, NULL);
+    return 0;
+}
 
-        transaction_frame->local = local;
-        frame->local = NULL;
+int
+afr_split_brain_resolve_do(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                           char *data)
+{
+    afr_local_t *local = NULL;
+    int ret = -1;
+    int op_errno = EINVAL;
+
+    local = frame->local;
+    local->xdata_req = dict_new();
+
+    if (!local->xdata_req) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    ret = dict_set_int32_sizen(local->xdata_req, "heal-op",
+                               GF_SHD_OP_SBRAIN_HEAL_FROM_BRICK);
+    if (ret) {
+        op_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+    ret = dict_set_str_sizen(local->xdata_req, "child-name", data);
+    if (ret) {
+        op_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+    /* set spb choice to -1 whether heal succeeds or not:
+     * If heal succeeds : spb-choice should be set to -1 as
+     *                    it is no longer valid; file is not
+     *                    in split-brain anymore.
+     * If heal doesn't succeed:
+     *                    spb-choice should be set to -1
+     *                    otherwise reads will be served
+     *                    from spb-choice which is misleading.
+     */
+    ret = afr_inode_split_brain_choice_set(loc->inode, this, -1);
+    if (ret)
+        gf_smsg(this->name, GF_LOG_WARNING, 0, AFR_MSG_SPLIT_BRAIN_SET_FAILED,
+                NULL);
+    afr_heal_splitbrain_file(frame, this, loc);
+    ret = 0;
+out:
+    if (ret < 0)
+        AFR_STACK_UNWIND(setxattr, frame, -1, op_errno, NULL);
+    return 0;
+}
 
-        local->op = GF_FOP_FTRUNCATE;
+int
+afr_get_split_brain_child_index(xlator_t *this, void *value, size_t len)
+{
+    int spb_child_index = -1;
+    char *spb_child_str = NULL;
 
-        local->transaction.fop    = afr_ftruncate_wind;
-        local->transaction.done   = afr_ftruncate_done;
-        local->transaction.unwind = afr_ftruncate_unwind;
+    spb_child_str = alloca0(len + 1);
+    memcpy(spb_child_str, value, len);
 
-        local->transaction.main_frame = frame;
+    if (!strcmp(spb_child_str, "none"))
+        return -2;
 
-        local->transaction.start   = 0;
-        local->transaction.len     = local->cont.ftruncate.offset;
+    spb_child_index = afr_get_child_index_from_name(this, spb_child_str);
+    if (spb_child_index < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, AFR_MSG_INVALID_SUBVOL,
+                "subvol=%s", spb_child_str, NULL);
+    }
+    return spb_child_index;
+}
 
-        afr_transaction (transaction_frame, this, AFR_DATA_TRANSACTION);
+int
+afr_can_set_split_brain_choice(void *opaque)
+{
+    afr_spbc_timeout_t *data = opaque;
+    call_frame_t *frame = NULL;
+    xlator_t *this = NULL;
+    loc_t *loc = NULL;
+    int ret = -1;
+
+    frame = data->frame;
+    loc = data->loc;
+    this = frame->this;
+
+    ret = afr_is_split_brain(frame, this, loc->inode, loc->gfid, &data->d_spb,
+                             &data->m_spb);
+
+    if (ret)
+        gf_smsg(this->name, GF_LOG_ERROR, 0,
+                AFR_MSG_SPLIT_BRAIN_DETERMINE_FAILED, "gfid=%s",
+                uuid_utoa(loc->gfid), NULL);
+    return ret;
+}
 
-        op_ret = 0;
+int
+afr_handle_split_brain_commands(xlator_t *this, call_frame_t *frame, loc_t *loc,
+                                dict_t *dict)
+{
+    void *choice_value = NULL;
+    void *resolve_value = NULL;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    afr_spbc_timeout_t *data = NULL;
+    int len = 0;
+    int spb_child_index = -1;
+    int ret = -1;
+    int op_errno = EINVAL;
+
+    priv = this->private;
+
+    ret = dict_get_ptr_and_len(dict, GF_AFR_SBRAIN_CHOICE, &choice_value, &len);
+    ret = dict_get_ptr_and_len(dict, GF_AFR_SBRAIN_RESOLVE, &resolve_value,
+                               &len);
+    if (!choice_value && !resolve_value) {
+        ret = -1;
+        goto out;
+    }
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local) {
+        ret = 1;
+        goto out;
+    }
+
+    local->op = GF_FOP_SETXATTR;
+
+    if (choice_value) {
+        spb_child_index = afr_get_split_brain_child_index(this, choice_value,
+                                                          len);
+        if (spb_child_index < 0) {
+            /* Case where value was "none" */
+            if (spb_child_index == -2)
+                spb_child_index = -1;
+            else {
+                ret = 1;
+                op_errno = EINVAL;
+                goto out;
+            }
+        }
+
+        data = GF_CALLOC(1, sizeof(*data), gf_afr_mt_spbc_timeout_t);
+        if (!data) {
+            ret = 1;
+            goto out;
+        }
+        data->spb_child_index = spb_child_index;
+        data->frame = frame;
+        loc_copy(&local->loc, loc);
+        data->loc = &local->loc;
+        ret = synctask_new(this->ctx->env, afr_can_set_split_brain_choice,
+                           afr_set_split_brain_choice, NULL, data);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN_STATUS,
+                    "name=%s", loc->name, NULL);
+            ret = 1;
+            op_errno = ENOMEM;
+            goto out;
+        }
+        ret = 0;
+        goto out;
+    }
+
+    if (resolve_value) {
+        spb_child_index = afr_get_split_brain_child_index(this, resolve_value,
+                                                          len);
+        if (spb_child_index < 0) {
+            ret = 1;
+            goto out;
+        }
+
+        afr_split_brain_resolve_do(frame, this, loc,
+                                   priv->children[spb_child_index]->name);
+        ret = 0;
+    }
 out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (ftruncate, frame, op_ret, op_errno, NULL, NULL);
-        }
-
-        return 0;
+    /* key was correct but value was invalid when ret == 1 */
+    if (ret == 1) {
+        AFR_STACK_UNWIND(setxattr, frame, -1, op_errno, NULL);
+        if (data)
+            GF_FREE(data);
+        ret = 0;
+    }
+    return ret;
 }
 
-
 int
-afr_ftruncate (call_frame_t *frame, xlator_t *this,
-               fd_t *fd, off_t offset)
+afr_handle_spb_choice_timeout(xlator_t *this, call_frame_t *frame, dict_t *dict)
 {
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
-        call_frame_t   *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = 0;
+    uint64_t timeout = 0;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
 
-        int ret = -1;
+    ret = dict_get_uint64(dict, GF_AFR_SPB_CHOICE_TIMEOUT, &timeout);
+    if (!ret) {
+        priv->spb_choice_timeout = timeout * 60;
+        AFR_STACK_UNWIND(setxattr, frame, ret, op_errno, NULL);
+    }
 
-        int op_ret   = -1;
-        int op_errno = 0;
+    return ret;
+}
 
-        uint64_t ctx;
-        afr_fd_ctx_t *fd_ctx = NULL;
+int
+afr_handle_empty_brick(xlator_t *this, call_frame_t *frame, loc_t *loc,
+                       dict_t *dict)
+{
+    int ret = -1;
+    int ab_ret = -1;
+    int empty_index = -1;
+    int op_errno = EPERM;
+    char *empty_brick = NULL;
+    char *op_type = NULL;
+    afr_empty_brick_args_t *data = NULL;
+
+    ret = dict_get_str_sizen(dict, GF_AFR_REPLACE_BRICK, &empty_brick);
+    if (!ret)
+        op_type = GF_AFR_REPLACE_BRICK;
+
+    ab_ret = dict_get_str_sizen(dict, GF_AFR_ADD_BRICK, &empty_brick);
+    if (!ab_ret)
+        op_type = GF_AFR_ADD_BRICK;
+
+    if (ret && ab_ret)
+        goto out;
+
+    if (frame->root->pid != GF_CLIENT_PID_ADD_REPLICA_MOUNT) {
+        gf_smsg(this->name, GF_LOG_ERROR, EPERM, AFR_MSG_INTERNAL_ATTR,
+                "op_type=%s", op_type, NULL);
+        ret = 1;
+        goto out;
+    }
+    empty_index = afr_get_child_index_from_name(this, empty_brick);
+
+    if (empty_index < 0) {
+        /* Didn't belong to this replica pair
+         * Just do a no-op
+         */
+        AFR_STACK_UNWIND(setxattr, frame, 0, 0, NULL);
+        return 0;
+    } else {
+        data = GF_CALLOC(1, sizeof(*data), gf_afr_mt_empty_brick_t);
+        if (!data) {
+            ret = 1;
+            op_errno = ENOMEM;
+            goto out;
+        }
+        data->frame = frame;
+        loc_copy(&data->loc, loc);
+        data->empty_index = empty_index;
+        data->op_type = op_type;
+        ret = synctask_new(this->ctx->env, _afr_handle_empty_brick,
+                           _afr_handle_empty_brick_cbk, NULL, data);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN_STATUS,
+                    NULL);
+            ret = 1;
+            op_errno = ENOMEM;
+            afr_brick_args_cleanup(data);
+            goto out;
+        }
+    }
+    ret = 0;
+out:
+    if (ret == 1) {
+        AFR_STACK_UNWIND(setxattr, frame, -1, op_errno, NULL);
+        ret = 0;
+    }
+    return ret;
+}
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+static int
+afr_handle_special_xattr(xlator_t *this, call_frame_t *frame, loc_t *loc,
+                         dict_t *dict)
+{
+    int ret = -1;
 
-        priv = this->private;
+    ret = afr_handle_split_brain_commands(this, frame, loc, dict);
+    if (ret == 0)
+        goto out;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+    ret = afr_handle_spb_choice_timeout(this, frame, dict);
+    if (ret == 0)
+        goto out;
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+    /* Applicable for replace-brick and add-brick commands */
+    ret = afr_handle_empty_brick(this, frame, loc, dict);
+out:
+    return ret;
+}
 
-        frame->local = local;
+int
+afr_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+             int32_t flags, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = EINVAL;
 
-        local->cont.ftruncate.offset  = offset;
-        local->cont.ftruncate.ino     = fd->inode->ino;
+    GF_IF_INTERNAL_XATTR_GOTO("trusted.afr.*", dict, op_errno, out);
 
-        local->fd = fd_ref (fd);
+    GF_IF_INTERNAL_XATTR_GOTO("trusted.glusterfs.afr.*", dict, op_errno, out);
 
-        ret = fd_ctx_get (fd, this, &ctx);
-        if (ret < 0) {
-                goto out;
-        }
+    ret = afr_handle_special_xattr(this, frame, loc, dict);
+    if (ret == 0)
+        return 0;
 
-        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        if (fd_ctx->up_count < priv->up_count) {
-                local->openfd_flush_cbk = afr_do_ftruncate;
-                afr_openfd_flush (frame, this, fd);
-        } else {
-                afr_do_ftruncate (frame, this);
-        }
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        op_ret = 0;
-out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (ftruncate, frame, op_ret, op_errno, NULL, NULL);
-        }
+    local->cont.setxattr.dict = dict_ref(dict);
+    local->cont.setxattr.flags = flags;
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-        return 0;
-}
+    if (!local->xdata_req)
+        goto out;
 
-/* }}} */
+    local->transaction.wind = afr_setxattr_wind;
+    local->transaction.unwind = afr_setxattr_unwind;
 
-/* {{{ setattr */
+    loc_copy(&local->loc, loc);
+    ret = afr_set_inode_local(this, local, loc->inode);
+    if (ret)
+        goto out;
 
-int
-afr_setattr_unwind (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
+    local->transaction.main_frame = frame;
+    local->transaction.start = LLONG_MAX - 1;
+    local->transaction.len = 0;
 
-        local = frame->local;
+    local->op = GF_FOP_SETXATTR;
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
+    ret = afr_transaction(transaction_frame, this, AFR_METADATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        if (main_frame) {
-                local->cont.setattr.preop_buf.ia_ino  = local->cont.setattr.ino;
-                local->cont.setattr.postop_buf.ia_ino = local->cont.setattr.ino;
+    return 0;
+out:
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-                AFR_STACK_UNWIND (setattr, main_frame, local->op_ret,
-                                  local->op_errno,
-                                  &local->cont.setattr.preop_buf,
-                                  &local->cont.setattr.postop_buf);
-        }
+    AFR_STACK_UNWIND(setxattr, frame, -1, op_errno, NULL);
 
-        return 0;
+    return 0;
 }
 
+/* {{{ fsetxattr */
 
 int
-afr_setattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno,
-                      struct iatt *preop, struct iatt *postop)
+afr_fsetxattr_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-
-        int child_index = (long) cookie;
-        int read_child  = 0;
-        int call_count  = -1;
-        int need_unwind = 0;
-
-        local = frame->local;
-        priv  = this->private;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        read_child = afr_read_child (this, local->loc.inode);
+    local = frame->local;
 
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-                if (afr_fop_failed (op_ret, op_errno))
-                        afr_transaction_fop_failed (frame, this, child_index);
+    AFR_STACK_UNWIND(fsetxattr, main_frame, local->op_ret, local->op_errno,
+                     local->xdata_rsp);
+    return 0;
+}
 
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                                local->cont.setattr.preop_buf  = *preop;
-                                local->cont.setattr.postop_buf = *postop;
-                        }
+int
+afr_fsetxattr_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, NULL,
+                                 NULL, NULL, xdata);
+}
 
-                        if (child_index == read_child) {
-                                local->cont.setattr.preop_buf  = *preop;
-                                local->cont.setattr.postop_buf = *postop;
-                        }
+int
+afr_fsetxattr_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    STACK_WIND_COOKIE(frame, afr_fsetxattr_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->fsetxattr, local->fd,
+                      local->cont.fsetxattr.dict, local->cont.fsetxattr.flags,
+                      local->xdata_req);
+    return 0;
+}
 
-                        local->success_count++;
+int
+afr_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+              int32_t flags, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-                        if ((local->success_count >= priv->wait_count)
-                            && local->read_child_returned) {
-                                need_unwind = 1;
-                        }
-                }
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    GF_IF_INTERNAL_XATTR_GOTO("trusted.afr.*", dict, op_errno, out);
 
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
+    GF_IF_INTERNAL_XATTR_GOTO("trusted.glusterfs.afr.*", dict, op_errno, out);
 
-        call_count = afr_frame_return (frame);
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        return 0;
-}
+    local->cont.fsetxattr.dict = dict_ref(dict);
+    local->cont.fsetxattr.flags = flags;
 
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-int32_t
-afr_setattr_wind (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
+    if (!local->xdata_req)
+        goto out;
 
-        int call_count = -1;
-        int i = 0;
+    local->transaction.wind = afr_fsetxattr_wind;
+    local->transaction.unwind = afr_fsetxattr_unwind;
 
-        local = frame->local;
-        priv = this->private;
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
 
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
+    local->op = GF_FOP_FSETXATTR;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+    local->transaction.main_frame = frame;
+    local->transaction.start = LLONG_MAX - 1;
+    local->transaction.len = 0;
 
-        local->call_count = call_count;
+    ret = afr_transaction(transaction_frame, this, AFR_METADATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND_COOKIE (frame, afr_setattr_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->setattr,
-                                           &local->loc,
-                                           &local->cont.setattr.in_buf,
-                                           local->cont.setattr.valid);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+    return 0;
+out:
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(fsetxattr, frame, -1, op_errno, NULL);
+    return 0;
 }
 
+/* }}} */
+
+/* {{{ removexattr */
 
 int
-afr_setattr_done (call_frame_t *frame, xlator_t *this)
+afr_removexattr_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *local = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        local = frame->local;
-
-        local->transaction.unwind (frame, this);
-
-        AFR_STACK_DESTROY (frame);
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
+
+    AFR_STACK_UNWIND(removexattr, main_frame, local->op_ret, local->op_errno,
+                     local->xdata_rsp);
+    return 0;
 }
 
+int
+afr_removexattr_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, NULL,
+                                 NULL, NULL, xdata);
+}
 
 int
-afr_setattr (call_frame_t *frame, xlator_t *this,
-             loc_t *loc, struct iatt *buf, int32_t valid)
+afr_removexattr_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
-        call_frame_t   *transaction_frame = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        int ret = -1;
+    local = frame->local;
+    priv = this->private;
 
-        int op_ret   = -1;
-        int op_errno = 0;
+    STACK_WIND_COOKIE(frame, afr_removexattr_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->removexattr, &local->loc,
+                      local->cont.removexattr.name, local->xdata_req);
+    return 0;
+}
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+int
+afr_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                const char *name, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        priv = this->private;
+    GF_IF_NATIVE_XATTR_GOTO("trusted.afr.*", name, op_errno, out);
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory.");
-                goto out;
-        }
+    GF_IF_NATIVE_XATTR_GOTO("trusted.glusterfs.afr.*", name, op_errno, out);
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        transaction_frame->local = local;
+    local->cont.removexattr.name = gf_strdup(name);
 
-        local->op_ret = -1;
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-        local->cont.setattr.ino     = loc->inode->ino;
+    if (!local->xdata_req)
+        goto out;
 
-        local->cont.setattr.in_buf = *buf;
-        local->cont.setattr.valid  = valid;
+    local->transaction.wind = afr_removexattr_wind;
+    local->transaction.unwind = afr_removexattr_unwind;
 
-        local->transaction.fop    = afr_setattr_wind;
-        local->transaction.done   = afr_setattr_done;
-        local->transaction.unwind = afr_setattr_unwind;
+    loc_copy(&local->loc, loc);
+    ret = afr_set_inode_local(this, local, loc->inode);
+    if (ret)
+        goto out;
 
-        loc_copy (&local->loc, loc);
+    local->op = GF_FOP_REMOVEXATTR;
 
-        local->transaction.main_frame = frame;
-        local->transaction.start   = LLONG_MAX - 1;
-        local->transaction.len     = 0;
+    local->transaction.main_frame = frame;
+    local->transaction.start = LLONG_MAX - 1;
+    local->transaction.len = 0;
 
-        afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+    ret = afr_transaction(transaction_frame, this, AFR_METADATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        op_ret = 0;
+    return 0;
 out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (setattr, frame, op_ret, op_errno, NULL, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(removexattr, frame, -1, op_errno, NULL);
+    return 0;
 }
 
-/* {{{ fsetattr */
-
+/* ffremovexattr */
 int
-afr_fsetattr_unwind (call_frame_t *frame, xlator_t *this)
+afr_fremovexattr_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
-
-        if (main_frame) {
-                local->cont.fsetattr.preop_buf.ia_ino  =
-                        local->cont.fsetattr.ino;
-                local->cont.fsetattr.postop_buf.ia_ino =
-                        local->cont.fsetattr.ino;
-
-                AFR_STACK_UNWIND (fsetattr, main_frame, local->op_ret,
-                                  local->op_errno,
-                                  &local->cont.fsetattr.preop_buf,
-                                  &local->cont.fsetattr.postop_buf);
-        }
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
 
+    AFR_STACK_UNWIND(fremovexattr, main_frame, local->op_ret, local->op_errno,
+                     local->xdata_rsp);
+    return 0;
+}
 
 int
-afr_fsetattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno,
-                      struct iatt *preop, struct iatt *postop)
+afr_fremovexattr_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-
-        int child_index = (long) cookie;
-        int read_child  = 0;
-        int call_count  = -1;
-        int need_unwind = 0;
-
-        local = frame->local;
-        priv  = this->private;
-
-        read_child = afr_read_child (this, local->fd->inode);
-
-        LOCK (&frame->lock);
-        {
-                if (child_index == read_child) {
-                        local->read_child_returned = _gf_true;
-                }
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, NULL,
+                                 NULL, NULL, xdata);
+}
 
-                if (afr_fop_failed (op_ret, op_errno))
-                        afr_transaction_fop_failed (frame, this, child_index);
+int
+afr_fremovexattr_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                                local->cont.fsetattr.preop_buf  = *preop;
-                                local->cont.fsetattr.postop_buf = *postop;
-                        }
+    local = frame->local;
+    priv = this->private;
 
-                        if (child_index == read_child) {
-                                local->cont.fsetattr.preop_buf  = *preop;
-                                local->cont.fsetattr.postop_buf = *postop;
-                        }
+    STACK_WIND_COOKIE(frame, afr_fremovexattr_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->fremovexattr, local->fd,
+                      local->cont.removexattr.name, local->xdata_req);
+    return 0;
+}
 
-                        local->success_count++;
+int
+afr_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                 const char *name, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-                        if ((local->success_count >= priv->wait_count)
-                            && local->read_child_returned) {
-                                need_unwind = 1;
-                        }
-                }
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    GF_IF_NATIVE_XATTR_GOTO("trusted.afr.*", name, op_errno, out);
 
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
+    GF_IF_NATIVE_XATTR_GOTO("trusted.glusterfs.afr.*", name, op_errno, out);
 
-        call_count = afr_frame_return (frame);
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        return 0;
-}
+    local->cont.removexattr.name = gf_strdup(name);
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
+    if (!local->xdata_req)
+        goto out;
 
-int32_t
-afr_fsetattr_wind (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
+    local->transaction.wind = afr_fremovexattr_wind;
+    local->transaction.unwind = afr_fremovexattr_unwind;
 
-        int call_count = -1;
-        int i = 0;
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
 
-        local = frame->local;
-        priv = this->private;
+    local->op = GF_FOP_FREMOVEXATTR;
 
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
+    local->transaction.main_frame = frame;
+    local->transaction.start = LLONG_MAX - 1;
+    local->transaction.len = 0;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+    ret = afr_transaction(transaction_frame, this, AFR_METADATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        local->call_count = call_count;
+    return 0;
+out:
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND_COOKIE (frame, afr_fsetattr_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->fsetattr,
-                                           local->fd,
-                                           &local->cont.fsetattr.in_buf,
-                                           local->cont.fsetattr.valid);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+    AFR_STACK_UNWIND(fremovexattr, frame, -1, op_errno, NULL);
 
-        return 0;
+    return 0;
 }
 
-
 int
-afr_fsetattr_done (call_frame_t *frame, xlator_t *this)
+afr_fallocate_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *local = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        local = frame->local;
-
-        local->transaction.unwind (frame, this);
-
-        AFR_STACK_DESTROY (frame);
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
 
+    AFR_STACK_UNWIND(fallocate, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.inode_wfop.prebuf,
+                     &local->cont.inode_wfop.postbuf, local->xdata_rsp);
+    return 0;
+}
 
 int
-afr_fsetattr (call_frame_t *frame, xlator_t *this,
-              fd_t *fd, struct iatt *buf, int32_t valid)
+afr_fallocate_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                       struct iatt *postbuf, dict_t *xdata)
 {
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
-        call_frame_t   *transaction_frame = NULL;
-
-        int ret = -1;
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, prebuf,
+                                 postbuf, NULL, xdata);
+}
 
-        int op_ret   = -1;
-        int op_errno = 0;
+int
+afr_fallocate_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    STACK_WIND_COOKIE(frame, afr_fallocate_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->fallocate, local->fd,
+                      local->cont.fallocate.mode, local->cont.fallocate.offset,
+                      local->cont.fallocate.len, local->xdata_req);
+    return 0;
+}
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+int
+afr_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+              off_t offset, size_t len, dict_t *xdata)
+{
+    call_frame_t *transaction_frame = NULL;
+    afr_local_t *local = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        priv = this->private;
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory.");
-                goto out;
-        }
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+    local->cont.fallocate.mode = mode;
+    local->cont.fallocate.offset = offset;
+    local->cont.fallocate.len = len;
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
 
-        transaction_frame->local = local;
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-        local->op_ret = -1;
+    if (!local->xdata_req)
+        goto out;
 
-        local->cont.fsetattr.ino     = fd->inode->ino;
+    local->op = GF_FOP_FALLOCATE;
 
-        local->cont.fsetattr.in_buf = *buf;
-        local->cont.fsetattr.valid  = valid;
+    local->transaction.wind = afr_fallocate_wind;
+    local->transaction.unwind = afr_fallocate_unwind;
 
-        local->transaction.fop    = afr_fsetattr_wind;
-        local->transaction.done   = afr_fsetattr_done;
-        local->transaction.unwind = afr_fsetattr_unwind;
+    local->transaction.main_frame = frame;
 
-        local->fd                 = fd_ref (fd);
+    local->transaction.start = local->cont.fallocate.offset;
+    local->transaction.len = 0;
 
-        local->transaction.main_frame = frame;
-        local->transaction.start   = LLONG_MAX - 1;
-        local->transaction.len     = 0;
+    afr_fix_open(fd, this);
 
-        afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+    ret = afr_transaction(transaction_frame, this, AFR_DATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        op_ret = 0;
+    return 0;
 out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (fsetattr, frame, op_ret, op_errno, NULL, NULL);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
+/* }}} */
 
-/* {{{ setxattr */
-
+/* {{{ discard */
 
 int
-afr_setxattr_unwind (call_frame_t *frame, xlator_t *this)
+afr_discard_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
-
-        local = frame->local;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
+    local = frame->local;
 
-        if (main_frame) {
-                AFR_STACK_UNWIND (setxattr, main_frame,
-                                  local->op_ret, local->op_errno)
-        }
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
 
+    AFR_STACK_UNWIND(discard, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.inode_wfop.prebuf,
+                     &local->cont.inode_wfop.postbuf, local->xdata_rsp);
+    return 0;
+}
 
 int
-afr_setxattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno)
+afr_discard_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
-
-        int call_count  = -1;
-        int need_unwind = 0;
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, prebuf,
+                                 postbuf, NULL, xdata);
+}
 
-        local = frame->local;
-        priv = this->private;
+int
+afr_discard_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    STACK_WIND_COOKIE(frame, afr_discard_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->discard, local->fd,
+                      local->cont.discard.offset, local->cont.discard.len,
+                      local->xdata_req);
+    return 0;
+}
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                        }
-                        local->success_count++;
+int
+afr_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+            size_t len, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-                        if (local->success_count == priv->child_count) {
-                                need_unwind = 1;
-                        }
-                }
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
+    local->cont.discard.offset = offset;
+    local->cont.discard.len = len;
 
-        call_count = afr_frame_return (frame);
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-        return 0;
-}
+    if (!local->xdata_req)
+        goto out;
 
+    local->op = GF_FOP_DISCARD;
 
-int
-afr_setxattr_wind (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
+    local->transaction.wind = afr_discard_wind;
+    local->transaction.unwind = afr_discard_unwind;
 
-        int call_count = -1;
-        int i = 0;
+    local->transaction.main_frame = frame;
 
-        local = frame->local;
-        priv = this->private;
+    local->transaction.start = local->cont.discard.offset;
+    local->transaction.len = 0;
 
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
+    afr_fix_open(fd, this);
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
+    ret = afr_transaction(transaction_frame, this, AFR_DATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        local->call_count = call_count;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND_COOKIE (frame, afr_setxattr_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->setxattr,
-                                           &local->loc,
-                                           local->cont.setxattr.dict,
-                                           local->cont.setxattr.flags);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+    return 0;
+out:
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(discard, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
+/* {{{ zerofill */
 
 int
-afr_setxattr_done (call_frame_t *frame, xlator_t *this)
+afr_zerofill_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t * local = frame->local;
-
-        local->transaction.unwind (frame, this);
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        AFR_STACK_DESTROY (frame);
+    local = frame->local;
 
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
+
+    AFR_STACK_UNWIND(discard, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.inode_wfop.prebuf,
+                     &local->cont.inode_wfop.postbuf, local->xdata_rsp);
+    return 0;
 }
 
 int
-afr_setxattr (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, dict_t *dict, int32_t flags)
+afr_zerofill_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                      struct iatt *postbuf, dict_t *xdata)
 {
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
-        call_frame_t   *transaction_frame = NULL;
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, prebuf,
+                                 postbuf, NULL, xdata);
+}
 
-        int ret = -1;
+int
+afr_zerofill_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    STACK_WIND_COOKIE(frame, afr_zerofill_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->zerofill, local->fd,
+                      local->cont.zerofill.offset, local->cont.zerofill.len,
+                      local->xdata_req);
+    return 0;
+}
 
-        int op_ret   = -1;
-        int op_errno = 0;
+int
+afr_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             size_t len, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        priv = this->private;
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+    local->cont.zerofill.offset = offset;
+    local->cont.zerofill.len = len;
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory.");
-                goto out;
-        }
+    if (xdata)
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+    else
+        local->xdata_req = dict_new();
 
-        transaction_frame->local = local;
+    if (!local->xdata_req)
+        goto out;
 
-        local->op_ret = -1;
+    local->op = GF_FOP_ZEROFILL;
 
-        local->cont.setxattr.dict  = dict_ref (dict);
-        local->cont.setxattr.flags = flags;
+    local->transaction.wind = afr_zerofill_wind;
+    local->transaction.unwind = afr_zerofill_unwind;
 
-        local->transaction.fop    = afr_setxattr_wind;
-        local->transaction.done   = afr_setxattr_done;
-        local->transaction.unwind = afr_setxattr_unwind;
+    local->transaction.main_frame = frame;
 
-        loc_copy (&local->loc, loc);
+    local->transaction.start = local->cont.zerofill.offset;
+    local->transaction.len = len;
 
-        local->transaction.main_frame = frame;
-        local->transaction.start   = LLONG_MAX - 1;
-        local->transaction.len     = 0;
+    afr_fix_open(fd, this);
 
-        afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+    ret = afr_transaction(transaction_frame, this, AFR_DATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        op_ret = 0;
+    return 0;
 out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (setxattr, frame, op_ret, op_errno);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(zerofill, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
 /* }}} */
 
-/* {{{ removexattr */
+int32_t
+afr_xattrop_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, dict_t *xattr,
+                     dict_t *xdata)
+{
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, NULL,
+                                 NULL, xattr, xdata);
+}
 
+int
+afr_xattrop_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    STACK_WIND_COOKIE(frame, afr_xattrop_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->xattrop, &local->loc,
+                      local->cont.xattrop.optype, local->cont.xattrop.xattr,
+                      local->xdata_req);
+    return 0;
+}
 
 int
-afr_removexattr_unwind (call_frame_t *frame, xlator_t *this)
+afr_xattrop_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t *   local = NULL;
-        call_frame_t   *main_frame = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        local = frame->local;
+    local = frame->local;
 
-        LOCK (&frame->lock);
-        {
-                if (local->transaction.main_frame)
-                        main_frame = local->transaction.main_frame;
-                local->transaction.main_frame = NULL;
-        }
-        UNLOCK (&frame->lock);
-
-        if (main_frame) {
-                AFR_STACK_UNWIND (removexattr, main_frame,
-                                  local->op_ret, local->op_errno)
-        }
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
         return 0;
-}
 
+    AFR_STACK_UNWIND(xattrop, main_frame, local->op_ret, local->op_errno,
+                     local->xattr_rsp, local->xdata_rsp);
+    return 0;
+}
 
-int
-afr_removexattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t op_ret, int32_t op_errno)
+int32_t
+afr_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+            gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
 {
-        afr_local_t *   local = NULL;
-        afr_private_t * priv  = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
 
-        int call_count  = -1;
-        int need_unwind = 0;
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
 
-        local = frame->local;
-        priv = this->private;
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
 
-        LOCK (&frame->lock);
-        {
-                if (op_ret != -1) {
-                        if (local->success_count == 0) {
-                                local->op_ret = op_ret;
-                        }
-                        local->success_count++;
+    local->cont.xattrop.xattr = dict_ref(xattr);
+    local->cont.xattrop.optype = optype;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
 
-                        if (local->success_count == priv->wait_count) {
-                                need_unwind = 1;
-                        }
-                }
+    local->transaction.wind = afr_xattrop_wind;
+    local->transaction.unwind = afr_xattrop_unwind;
 
-                local->op_errno = op_errno;
-        }
-        UNLOCK (&frame->lock);
+    loc_copy(&local->loc, loc);
+    ret = afr_set_inode_local(this, local, loc->inode);
+    if (ret)
+        goto out;
 
-        if (need_unwind)
-                local->transaction.unwind (frame, this);
+    local->op = GF_FOP_XATTROP;
 
-        call_count = afr_frame_return (frame);
+    local->transaction.main_frame = frame;
+    local->transaction.start = LLONG_MAX - 1;
+    local->transaction.len = 0;
 
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-        }
+    ret = afr_transaction(transaction_frame, this, AFR_METADATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
 
-        return 0;
-}
+    return 0;
+out:
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
+    AFR_STACK_UNWIND(xattrop, frame, -1, op_errno, NULL, NULL);
+    return 0;
+}
 
 int32_t
-afr_removexattr_wind (call_frame_t *frame, xlator_t *this)
+afr_fxattrop_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xattr,
+                      dict_t *xdata)
 {
-        afr_local_t *local = NULL;
-        afr_private_t *priv = NULL;
-
-        int call_count = -1;
-        int i = 0;
-
-        local = frame->local;
-        priv = this->private;
-
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-        if (call_count == 0) {
-                local->transaction.resume (frame, this);
-                return 0;
-        }
-
-        local->call_count = call_count;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (local->child_up[i]) {
-                        STACK_WIND_COOKIE (frame, afr_removexattr_wind_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->removexattr,
-                                           &local->loc,
-                                           local->cont.removexattr.name);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-        return 0;
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, NULL,
+                                 NULL, xattr, xdata);
 }
 
-
 int
-afr_removexattr_done (call_frame_t *frame, xlator_t *this)
+afr_fxattrop_wind(call_frame_t *frame, xlator_t *this, int subvol)
 {
-        afr_local_t * local = frame->local;
-
-        local->transaction.unwind (frame, this);
-
-        AFR_STACK_DESTROY (frame);
-
-        return 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    STACK_WIND_COOKIE(frame, afr_fxattrop_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->fxattrop, local->fd,
+                      local->cont.xattrop.optype, local->cont.xattrop.xattr,
+                      local->xdata_req);
+    return 0;
 }
 
-
 int
-afr_removexattr (call_frame_t *frame, xlator_t *this,
-                 loc_t *loc, const char *name)
+afr_fxattrop_unwind(call_frame_t *frame, xlator_t *this)
 {
-        afr_private_t * priv  = NULL;
-        afr_local_t   * local = NULL;
-        call_frame_t   *transaction_frame = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        int ret = -1;
+    local = frame->local;
 
-        int op_ret   = -1;
-        int op_errno = 0;
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (this->private, out);
-        VALIDATE_OR_GOTO (loc, out);
+    AFR_STACK_UNWIND(fxattrop, main_frame, local->op_ret, local->op_errno,
+                     local->xattr_rsp, local->xdata_rsp);
+    return 0;
+}
 
-        priv = this->private;
+int32_t
+afr_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+             gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int op_errno = ENOMEM;
+
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
+
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
+
+    local->cont.xattrop.xattr = dict_ref(xattr);
+    local->cont.xattrop.optype = optype;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
+
+    local->transaction.wind = afr_fxattrop_wind;
+    local->transaction.unwind = afr_fxattrop_unwind;
+
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
+
+    local->op = GF_FOP_FXATTROP;
+
+    local->transaction.main_frame = frame;
+    local->transaction.start = LLONG_MAX - 1;
+    local->transaction.len = 0;
+
+    ret = afr_transaction(transaction_frame, this, AFR_METADATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
+out:
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        transaction_frame = copy_frame (frame);
-        if (!transaction_frame) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory.");
-                goto out;
-        }
+    AFR_STACK_UNWIND(fxattrop, frame, -1, op_errno, NULL, NULL);
+    return 0;
+}
 
-        ALLOC_OR_GOTO (local, afr_local_t, out);
+int
+afr_fsync_unwind(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
 
-        ret = AFR_LOCAL_INIT (local, priv);
-        if (ret < 0) {
-                op_errno = -ret;
-                goto out;
-        }
+    local = frame->local;
 
-        transaction_frame->local = local;
+    main_frame = afr_transaction_detach_fop_frame(frame);
+    if (!main_frame)
+        return 0;
 
-        local->op_ret = -1;
+    AFR_STACK_UNWIND(fsync, main_frame, local->op_ret, local->op_errno,
+                     &local->cont.inode_wfop.prebuf,
+                     &local->cont.inode_wfop.postbuf, local->xdata_rsp);
 
-        local->cont.removexattr.name = gf_strdup (name);
+    return 0;
+}
 
-        local->transaction.fop    = afr_removexattr_wind;
-        local->transaction.done   = afr_removexattr_done;
-        local->transaction.unwind = afr_removexattr_unwind;
+int
+afr_fsync_wind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                   struct iatt *postbuf, dict_t *xdata)
+{
+    return __afr_inode_write_cbk(frame, cookie, this, op_ret, op_errno, prebuf,
+                                 postbuf, NULL, xdata);
+}
 
-        loc_copy (&local->loc, loc);
+int
+afr_fsync_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        local->transaction.main_frame = frame;
-        local->transaction.start   = LLONG_MAX - 1;
-        local->transaction.len     = 0;
+    local = frame->local;
+    priv = this->private;
 
-        afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
+    STACK_WIND_COOKIE(frame, afr_fsync_wind_cbk, (void *)(long)subvol,
+                      priv->children[subvol],
+                      priv->children[subvol]->fops->fsync, local->fd,
+                      local->cont.fsync.datasync, local->xdata_req);
+    return 0;
+}
 
-        op_ret = 0;
+int
+afr_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+          dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    call_frame_t *transaction_frame = NULL;
+    int ret = -1;
+    int32_t op_errno = ENOMEM;
+    int8_t last_fsync = 0;
+
+    AFR_ERROR_OUT_IF_FDCTX_INVALID(fd, this, op_errno, out);
+    transaction_frame = copy_frame(frame);
+    if (!transaction_frame)
+        goto out;
+
+    local = AFR_FRAME_INIT(transaction_frame, op_errno);
+    if (!local)
+        goto out;
+
+    if (xdata) {
+        local->xdata_req = dict_copy_with_ref(xdata, NULL);
+        if (dict_get_int8(xdata, "last-fsync", &last_fsync) == 0) {
+            if (last_fsync) {
+                local->transaction.disable_delayed_post_op = _gf_true;
+            }
+        }
+    } else {
+        local->xdata_req = dict_new();
+    }
+
+    if (!local->xdata_req)
+        goto out;
+
+    local->fd = fd_ref(fd);
+    ret = afr_set_inode_local(this, local, fd->inode);
+    if (ret)
+        goto out;
+
+    local->op = GF_FOP_FSYNC;
+    local->cont.fsync.datasync = datasync;
+
+    if (afr_fd_has_witnessed_unstable_write(this, fd->inode)) {
+        /* don't care. we only wanted to CLEAR the bit */
+    }
+
+    local->transaction.wind = afr_fsync_wind;
+    local->transaction.unwind = afr_fsync_unwind;
+
+    local->transaction.main_frame = frame;
+
+    ret = afr_transaction(transaction_frame, this, AFR_DATA_TRANSACTION);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto out;
+    }
+
+    return 0;
 out:
-        if (op_ret == -1) {
-                if (transaction_frame)
-                        AFR_STACK_DESTROY (transaction_frame);
-                AFR_STACK_UNWIND (removexattr, frame, op_ret, op_errno);
-        }
+    if (transaction_frame)
+        AFR_STACK_DESTROY(transaction_frame);
 
-        return 0;
+    AFR_STACK_UNWIND(fsync, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
 }
diff --git a/xlators/cluster/afr/src/afr-inode-write.h b/xlators/cluster/afr/src/afr-inode-write.h
index 1c6b87a97af..a787069b7a1 100644
--- a/xlators/cluster/afr/src/afr-inode-write.h
+++ b/xlators/cluster/afr/src/afr-inode-write.h
@@ -1,72 +1,94 @@
 /*
-   Copyright (c) 2007-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef __INODE_WRITE_H__
 #define __INODE_WRITE_H__
 
 int32_t
-afr_chmod (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, mode_t mode);
+afr_chmod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+          dict_t *xdata);
 
 int32_t
-afr_chown (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, uid_t uid, gid_t gid);
+afr_chown(call_frame_t *frame, xlator_t *this, loc_t *loc, uid_t uid, gid_t gid,
+          dict_t *xdata);
 
 int
-afr_fchown (call_frame_t *frame, xlator_t *this,
-	    fd_t *fd, uid_t uid, gid_t gid);
+afr_fchown(call_frame_t *frame, xlator_t *this, fd_t *fd, uid_t uid, gid_t gid,
+           dict_t *xdata);
+
+int32_t
+afr_fchmod(call_frame_t *frame, xlator_t *this, fd_t *fd, mode_t mode,
+           dict_t *xdata);
+
+int32_t
+afr_writev(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
+           int32_t count, off_t offset, uint32_t flags, struct iobref *iobref,
+           dict_t *xdata);
+
+int32_t
+afr_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+             dict_t *xdata);
+
+int32_t
+afr_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+              dict_t *xdata);
 
 int32_t
-afr_fchmod (call_frame_t *frame, xlator_t *this,
-	    fd_t *fd, mode_t mode);
+afr_utimens(call_frame_t *frame, xlator_t *this, loc_t *loc,
+            struct timespec tv[2], dict_t *xdata);
+
+int
+afr_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *buf,
+            int32_t valid, dict_t *xdata);
+
+int
+afr_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *buf,
+             int32_t valid, dict_t *xdata);
 
 int32_t
-afr_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, 
-	    struct iovec *vector, int32_t count, off_t offset,
-            struct iobref *iobref);
+afr_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+             int32_t flags, dict_t *xdata);
 
 int32_t
-afr_truncate (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, off_t offset);
+afr_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+              int32_t flags, dict_t *xdata);
 
 int32_t
-afr_ftruncate (call_frame_t *frame, xlator_t *this,
-	       fd_t *fd, off_t offset);
+afr_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                const char *name, dict_t *xdata);
 
 int32_t
-afr_utimens (call_frame_t *frame, xlator_t *this,
-	     loc_t *loc, struct timespec tv[2]);
+afr_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                 const char *name, dict_t *xdata);
+
+int
+afr_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+            size_t len, dict_t *xdata);
 
 int
-afr_setattr (call_frame_t *frame, xlator_t *this,
-             loc_t *loc, struct iatt *buf, int32_t valid);
+afr_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+              off_t offset, size_t len, dict_t *xdata);
 
 int
-afr_fsetattr (call_frame_t *frame, xlator_t *this,
-              fd_t *fd, struct iatt *buf, int32_t valid);
+afr_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             off_t len, dict_t *xdata);
 
 int32_t
-afr_setxattr (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, dict_t *dict, int32_t flags);
+afr_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+            gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata);
 
 int32_t
-afr_removexattr (call_frame_t *frame, xlator_t *this,
-		 loc_t *loc, const char *name);
+afr_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+             gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata);
 
+int
+afr_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+          dict_t *xdata);
 #endif /* __INODE_WRITE_H__ */
diff --git a/xlators/cluster/afr/src/afr-lk-common.c b/xlators/cluster/afr/src/afr-lk-common.c
index 69160d72ff2..bc8eabe0f43 100644
--- a/xlators/cluster/afr/src/afr-lk-common.c
+++ b/xlators/cluster/afr/src/afr-lk-common.c
@@ -1,2169 +1,791 @@
 /*
-  Copyright (c) 2007-2009 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
-#include "dict.h"
-#include "byte-order.h"
-#include "common-utils.h"
+#include <glusterfs/dict.h>
+#include <glusterfs/byte-order.h>
+#include <glusterfs/common-utils.h>
 
 #include "afr.h"
 #include "afr-transaction.h"
+#include "afr-messages.h"
 
 #include <signal.h>
 
-
-#define LOCKED_NO       0x0        /* no lock held */
-#define LOCKED_YES      0x1        /* for DATA, METADATA, ENTRY and higher_path */
-#define LOCKED_LOWER    0x2        /* for lower path */
-
-int
-afr_lock_blocking (call_frame_t *frame, xlator_t *this, int child_index);
-
-static uint64_t afr_lock_number = 1;
-
-static uint64_t
-get_afr_lock_number ()
-{
-        return (++afr_lock_number);
-}
-
-int
-afr_set_lock_number (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t         *local    = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        int_lock->lock_number = get_afr_lock_number ();
-
-        return 0;
-}
+#define LOCKED_NO 0x0    /* no lock held */
+#define LOCKED_YES 0x1   /* for DATA, METADATA, ENTRY and higher_path */
+#define LOCKED_LOWER 0x2 /* for lower path */
 
 void
-afr_set_lk_owner (call_frame_t *frame, xlator_t *this)
-{
-        if (!frame->root->lk_owner) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "Setting lk-owner=%llu",
-                        (unsigned long long) (unsigned long)frame->root);
-                frame->root->lk_owner = (uint64_t) (unsigned long)frame->root;
-        }
-}
-
-static int
-is_afr_lock_selfheal (afr_local_t *local)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        int                  ret      = -1;
-
-        int_lock = &local->internal_lock;
-
-        switch (int_lock->selfheal_lk_type) {
-        case AFR_DATA_SELF_HEAL_LK:
-        case AFR_METADATA_SELF_HEAL_LK:
-                ret = 1;
-                break;
-        case AFR_ENTRY_SELF_HEAL_LK:
-                ret = 0;
-                break;
-        }
-
-        return ret;
-
-}
-
-int32_t
-internal_lock_count (call_frame_t *frame, xlator_t *this,
-                     afr_fd_ctx_t *fd_ctx)
-{
-        afr_local_t   *local = NULL;
-        afr_private_t *priv  = NULL;
-
-        int32_t call_count = 0;
-        int i = 0;
-
-        local = frame->local;
-        priv  = this->private;
-
-        if (fd_ctx) {
-                GF_ASSERT (local->fd);
-                for (i = 0; i < priv->child_count; i++) {
-                        if (local->child_up[i] && fd_ctx->opened_on[i])
-                                ++call_count;
-                }
-        } else {
-                for (i = 0; i < priv->child_count; i++) {
-                        if (local->child_up[i])
-                                ++call_count;
-                }
-        }
-
-        return call_count;
-}
-
-static void
-afr_print_inodelk (char *str, int size, int cmd,
-                   struct gf_flock *flock, uint64_t owner)
+afr_lockee_cleanup(afr_lockee_t *lockee)
 {
-        char *cmd_str = NULL;
-        char *type_str = NULL;
-
-        switch (cmd) {
-#if F_GETLK != F_GETLK64
-        case F_GETLK64:
-#endif
-        case F_GETLK:
-                cmd_str = "GETLK";
-                break;
+    if (lockee->fd) {
+        fd_unref(lockee->fd);
+        lockee->fd = NULL;
+    } else {
+        loc_wipe(&lockee->loc);
+    }
 
-#if F_SETLK != F_SETLK64
-        case F_SETLK64:
-#endif
-        case F_SETLK:
-                cmd_str = "SETLK";
-                break;
+    GF_FREE(lockee->basename);
+    lockee->basename = NULL;
+    GF_FREE(lockee->locked_nodes);
+    lockee->locked_nodes = NULL;
 
-#if F_SETLKW != F_SETLKW64
-        case F_SETLKW64:
-#endif
-        case F_SETLKW:
-                cmd_str = "SETLKW";
-                break;
-
-        default:
-                cmd_str = "<null>";
-                break;
-        }
-
-        switch (flock->l_type) {
-        case F_RDLCK:
-                type_str = "READ";
-                break;
-        case F_WRLCK:
-                type_str = "WRITE";
-                break;
-        case F_UNLCK:
-                type_str = "UNLOCK";
-                break;
-        default:
-                type_str = "UNKNOWN";
-                break;
-        }
-
-        snprintf (str, size, "lock=INODELK, cmd=%s, type=%s, "
-                  "start=%llu, len=%llu, pid=%llu, lk-owner=%llu",
-                  cmd_str, type_str, (unsigned long long) flock->l_start,
-                  (unsigned long long) flock->l_len,
-                  (unsigned long long) flock->l_pid,
-                  (unsigned long long) owner);
-
-}
-
-static void
-afr_print_lockee (char *str, int size, loc_t *loc, fd_t *fd,
-                  int child_index)
-{
-        snprintf (str, size, "path=%s, fd=%p, child=%d",
-                  loc->path ? loc->path : "<nul>",
-                  fd ? fd : NULL,
-                  child_index);
+    return;
 }
 
 void
-afr_print_entrylk (char *str, int size, const char *basename,
-                   uint64_t owner)
+afr_lockees_cleanup(afr_internal_lock_t *int_lock)
 {
-        snprintf (str, size, "Basename=%s, lk-owner=%llu",
-                  basename ? basename : "<nul>",
-                  (unsigned long long)owner);
-}
+    int i = 0;
 
-static void
-afr_print_verdict (int op_ret, int op_errno, char *str)
-{
-        if (op_ret < 0) {
-                if (op_errno == EAGAIN)
-                        strcpy (str, "EAGAIN");
-                else
-                        strcpy (str, "FAILED");
-        }
-        else
-                strcpy (str, "GRANTED");
-}
-
-static void
-afr_set_lock_call_type (afr_lock_call_type_t lock_call_type,
-                        char *lock_call_type_str,
-                        afr_internal_lock_t *int_lock)
-{
-        switch (lock_call_type) {
-        case AFR_INODELK_TRANSACTION:
-                if (int_lock->transaction_lk_type == AFR_TRANSACTION_LK)
-                        strcpy (lock_call_type_str, "AFR_INODELK_TRANSACTION");
-                else
-                        strcpy (lock_call_type_str, "AFR_INODELK_SELFHEAL");
-                break;
-        case AFR_INODELK_NB_TRANSACTION:
-                if (int_lock->transaction_lk_type == AFR_TRANSACTION_LK)
-                        strcpy (lock_call_type_str, "AFR_INODELK_NB_TRANSACTION");
-                else
-                        strcpy (lock_call_type_str, "AFR_INODELK_NB_SELFHEAL");
-                break;
-        case AFR_ENTRYLK_TRANSACTION:
-                if (int_lock->transaction_lk_type == AFR_TRANSACTION_LK)
-                        strcpy (lock_call_type_str, "AFR_ENTRYLK_TRANSACTION");
-                else
-                        strcpy (lock_call_type_str, "AFR_ENTRYLK_SELFHEAL");
-                break;
-        case AFR_ENTRYLK_NB_TRANSACTION:
-                if (int_lock->transaction_lk_type == AFR_TRANSACTION_LK)
-                        strcpy (lock_call_type_str, "AFR_ENTRYLK_NB_TRANSACTION");
-                else
-                        strcpy (lock_call_type_str, "AFR_ENTRYLK_NB_SELFHEAL");
-                break;
-        default:
-                strcpy (lock_call_type_str, "UNKNOWN");
-                break;
-        }
-
-}
-
-static void
-afr_trace_inodelk_out (call_frame_t *frame, afr_lock_call_type_t lock_call_type,
-                       afr_lock_op_type_t lk_op_type, struct gf_flock *flock,
-                       int op_ret, int op_errno, int32_t child_index)
-{
-        xlator_t            *this     = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = NULL;
-
-        char lockee[256];
-        char lock_call_type_str[256];
-        char verdict[16];
-
-        this     = THIS;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        priv     = this->private;
-
-        if (!priv->inodelk_trace) {
-                return;
-        }
-
-        afr_print_lockee (lockee, 256, &local->loc, local->fd, child_index);
-
-        afr_set_lock_call_type (lock_call_type, lock_call_type_str, int_lock);
-
-        afr_print_verdict (op_ret, op_errno, verdict);
-
-        gf_log (this->name, GF_LOG_NORMAL,
-                "[%s %s] [%s] Lockee={%s} Number={%llu}",
-                lock_call_type_str,
-                lk_op_type == AFR_LOCK_OP ? "LOCK REPLY" : "UNLOCK REPLY",
-                verdict,
-                lockee,
-                (unsigned long long) int_lock->lock_number);
+    for (i = 0; i < int_lock->lockee_count; i++) {
+        afr_lockee_cleanup(&int_lock->lockee[i]);
+    }
 
+    return;
 }
-
-static void
-afr_trace_inodelk_in (call_frame_t *frame, afr_lock_call_type_t lock_call_type,
-                      afr_lock_op_type_t lk_op_type, struct gf_flock *flock,
-                      int32_t cmd, int32_t child_index)
-{
-        xlator_t            *this     = NULL;
-        afr_local_t         *local    = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-        afr_private_t       *priv     = NULL;
-
-        char lock[256];
-        char lockee[256];
-        char lock_call_type_str[256];
-
-        this     = THIS;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        priv     = this->private;
-
-        if (!priv->inodelk_trace) {
-                return;
-        }
-
-        afr_print_inodelk (lock, 256, cmd, flock, frame->root->lk_owner);
-        afr_print_lockee (lockee, 256, &local->loc, local->fd, child_index);
-
-        afr_set_lock_call_type (lock_call_type, lock_call_type_str, int_lock);
-
-        gf_log (this->name, GF_LOG_NORMAL,
-                "[%s %s] Lock={%s} Lockee={%s} Number={%llu}",
-                lock_call_type_str,
-                lk_op_type == AFR_LOCK_OP ? "LOCK REQUEST" : "UNLOCK REQUEST",
-                lock, lockee,
-                (unsigned long long) int_lock->lock_number);
-
+int
+afr_entry_lockee_cmp(const void *l1, const void *l2)
+{
+    const afr_lockee_t *r1 = l1;
+    const afr_lockee_t *r2 = l2;
+    int ret = 0;
+    uuid_t gfid1 = {0};
+    uuid_t gfid2 = {0};
+
+    loc_gfid((loc_t *)&r1->loc, gfid1);
+    loc_gfid((loc_t *)&r2->loc, gfid2);
+    ret = gf_uuid_compare(gfid1, gfid2);
+    /*Entrylks with NULL basename are the 'smallest'*/
+    if (ret == 0) {
+        if (!r1->basename)
+            return -1;
+        if (!r2->basename)
+            return 1;
+        ret = strcmp(r1->basename, r2->basename);
+    }
+
+    if (ret <= 0)
+        return -1;
+    else
+        return 1;
 }
 
-static void
-afr_trace_entrylk_in (call_frame_t *frame, afr_lock_call_type_t lock_call_type,
-                      afr_lock_op_type_t lk_op_type, const char *basename,
-                      int32_t child_index)
-{
-        xlator_t            *this     = NULL;
-        afr_local_t         *local    = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-        afr_private_t       *priv     = NULL;
-
-        char lock[256];
-        char lockee[256];
-        char lock_call_type_str[256];
-
-        this     = THIS;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        priv     = this->private;
-
-        if (!priv->entrylk_trace) {
-                return;
-        }
-
-        afr_print_entrylk (lock, 256, basename, frame->root->lk_owner);
-        afr_print_lockee (lockee, 256, &local->loc, local->fd, child_index);
-
-        afr_set_lock_call_type (lock_call_type, lock_call_type_str, int_lock);
-
-        gf_log (this->name, GF_LOG_NORMAL,
-                "[%s %s] Lock={%s} Lockee={%s} Number={%llu}",
-                lock_call_type_str,
-                lk_op_type == AFR_LOCK_OP ? "LOCK REQUEST" : "UNLOCK REQUEST",
-                lock, lockee,
-                (unsigned long long) int_lock->lock_number);
-}
+int
+afr_lock_blocking(call_frame_t *frame, xlator_t *this, int child_index);
 
-static void
-afr_trace_entrylk_out (call_frame_t *frame, afr_lock_call_type_t lock_call_type,
-                       afr_lock_op_type_t lk_op_type, const char *basename, int op_ret,
-                       int op_errno, int32_t child_index)
+void
+afr_set_lk_owner(call_frame_t *frame, xlator_t *this, void *lk_owner)
 {
-        xlator_t            *this     = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = NULL;
-
-        char lock[256];
-        char lockee[256];
-        char lock_call_type_str[256];
-        char verdict[16];
-
-        this     = THIS;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        priv     = this->private;
-
-        if (!priv->entrylk_trace) {
-                return;
-        }
-
-        afr_print_lockee (lockee, 256, &local->loc, local->fd, child_index);
-
-        afr_set_lock_call_type (lock_call_type, lock_call_type_str, int_lock);
-
-        afr_print_verdict (op_ret, op_errno, verdict);
-
-        gf_log (this->name, GF_LOG_NORMAL,
-                "[%s %s] [%s] Lock={%s} Lockee={%s} Number={%llu}",
-                lock_call_type_str,
-                lk_op_type == AFR_LOCK_OP ? "LOCK REPLY" : "UNLOCK REPLY",
-                verdict,
-                lock, lockee,
-                (unsigned long long) int_lock->lock_number);
+    gf_msg_trace(this->name, 0, "Setting lk-owner=%llu",
+                 (unsigned long long)(unsigned long)lk_owner);
 
+    set_lk_owner_from_ptr(&frame->root->lk_owner, lk_owner);
 }
 
-static int
-transaction_lk_op (afr_local_t *local)
+int32_t
+internal_lock_count(call_frame_t *frame, xlator_t *this)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        int ret = -1;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int32_t call_count = 0;
+    int i = 0;
 
-        int_lock = &local->internal_lock;
+    local = frame->local;
+    priv = this->private;
 
-        if (int_lock->transaction_lk_type == AFR_TRANSACTION_LK) {
-                gf_log (THIS->name, GF_LOG_DEBUG,
-                        "lk op is for a transaction");
-                ret = 1;
-        }
-        else if (int_lock->transaction_lk_type == AFR_SELFHEAL_LK) {
-                gf_log (THIS->name, GF_LOG_DEBUG,
-                        "lk op is for a self heal");
-
-                ret = 0;
-        }
-
-        if (ret == -1)
-                gf_log (THIS->name, GF_LOG_DEBUG,
-                        "lk op is not set");
-
-        return ret;
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->child_up[i])
+            ++call_count;
+    }
 
+    return call_count;
 }
 
-static int
-is_afr_lock_transaction (afr_local_t *local)
+int
+afr_add_entry_lockee(afr_local_t *local, loc_t *loc, char *basename,
+                     int child_count)
 {
-        int ret = 0;
+    int ret = -ENOMEM;
+    afr_internal_lock_t *int_lock = &local->internal_lock;
+    afr_lockee_t *lockee = &int_lock->lockee[int_lock->lockee_count];
 
-        switch (local->transaction.type) {
-	case AFR_DATA_TRANSACTION:
-	case AFR_METADATA_TRANSACTION:
-                ret = 1;
-                break;
+    GF_ASSERT(int_lock->lockee_count < AFR_LOCKEE_COUNT_MAX);
+    loc_copy(&lockee->loc, loc);
+    lockee->basename = (basename) ? gf_strdup(basename) : NULL;
+    if (basename && !lockee->basename)
+        goto out;
 
-        case AFR_ENTRY_RENAME_TRANSACTION:
-        case AFR_ENTRY_TRANSACTION:
-                ret = 0;
-                break;
+    lockee->locked_count = 0;
+    lockee->locked_nodes = GF_CALLOC(child_count, sizeof(*lockee->locked_nodes),
+                                     gf_afr_mt_afr_node_character);
 
-        }
+    if (!lockee->locked_nodes)
+        goto out;
 
-        return ret;
+    ret = 0;
+    int_lock->lockee_count++;
+out:
+    if (ret) {
+        afr_lockee_cleanup(lockee);
+    }
+    return ret;
 }
 
-static int
-initialize_entrylk_variables (call_frame_t *frame, xlator_t *this)
+int
+afr_add_inode_lockee(afr_local_t *local, int child_count)
 {
-        afr_local_t         *local    = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-        afr_private_t       *priv     = NULL;
-
-        int i = 0;
+    int ret = -ENOMEM;
+    afr_internal_lock_t *int_lock = &local->internal_lock;
+    afr_lockee_t *lockee = &int_lock->lockee[int_lock->lockee_count];
 
-        priv     = this->private;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+    if (local->fd) {
+        lockee->fd = fd_ref(local->fd);
+    } else {
+        loc_copy(&lockee->loc, &local->loc);
+    }
 
-        int_lock->entrylk_lock_count = 0;
-        int_lock->lock_op_ret        = -1;
-        int_lock->lock_op_errno      = 0;
+    lockee->locked_count = 0;
+    lockee->locked_nodes = GF_CALLOC(child_count, sizeof(*lockee->locked_nodes),
+                                     gf_afr_mt_afr_node_character);
 
-        for (i = 0; i < priv->child_count; i++) {
-                int_lock->entry_locked_nodes[i] = 0;
-        }
+    if (!lockee->locked_nodes)
+        goto out;
 
-        return 0;
+    ret = 0;
+    int_lock->lockee_count++;
+out:
+    if (ret) {
+        afr_lockee_cleanup(lockee);
+    }
+    return ret;
 }
 
 static int
-initialize_inodelk_variables (call_frame_t *frame, xlator_t *this)
+initialize_internal_lock_variables(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t         *local    = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-        afr_private_t       *priv     = NULL;
-
-        int i = 0;
-
-        priv     = this->private;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        int_lock->inodelk_lock_count = 0;
-        int_lock->lock_op_ret        = -1;
-        int_lock->lock_op_errno      = 0;
+    afr_local_t *local = NULL;
+    afr_internal_lock_t *int_lock = NULL;
+    afr_private_t *priv = NULL;
 
-        for (i = 0; i < priv->child_count; i++) {
-                int_lock->inode_locked_nodes[i] = 0;
-        }
-
-        return 0;
-}
+    int i = 0;
 
-loc_t *
-lower_path (loc_t *l1, const char *b1, loc_t *l2, const char *b2)
-{
-	int ret = 0;
+    priv = this->private;
+    local = frame->local;
+    int_lock = &local->internal_lock;
 
-	ret = strcmp (l1->path, l2->path);
+    int_lock->lock_count = 0;
+    int_lock->lock_op_ret = -1;
+    int_lock->lock_op_errno = 0;
+    int_lock->lk_attempted_count = 0;
 
-	if (ret == 0)
-		ret = strcmp (b1, b2);
+    for (i = 0; i < AFR_LOCKEE_COUNT_MAX; i++) {
+        if (!int_lock->lockee[i].locked_nodes)
+            break;
+        int_lock->lockee[i].locked_count = 0;
+        memset(int_lock->lockee[i].locked_nodes, 0,
+               sizeof(*int_lock->lockee[i].locked_nodes) * priv->child_count);
+    }
 
-	if (ret <= 0)
-		return l1;
-	else
-		return l2;
+    return 0;
 }
 
 int
-afr_locked_nodes_count (unsigned char *locked_nodes, int child_count)
-
+afr_lockee_locked_nodes_count(afr_internal_lock_t *int_lock)
 {
-        int i;
-        int call_count = 0;
+    int call_count = 0;
+    int i = 0;
 
-        for (i = 0; i < child_count; i++) {
-                if (locked_nodes[i] & LOCKED_YES)
-                        call_count++;
-        }
+    for (i = 0; i < int_lock->lockee_count; i++)
+        call_count += int_lock->lockee[i].locked_count;
 
-        return call_count;
+    return call_count;
 }
 
-/* FIXME: What if UNLOCK fails */
-static int32_t
-afr_unlock_common_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno)
-{
-	afr_local_t         *local    = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-	int call_count = 0;
-
-	local    = frame->local;
-        int_lock = &local->internal_lock;
-
-	LOCK (&frame->lock);
-	{
-		call_count = --int_lock->lk_call_count;
-	}
-	UNLOCK (&frame->lock);
-
-	if (call_count == 0) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "All internal locks unlocked");
-                int_lock->lock_cbk (frame, this);
-        }
-
-	return 0;
-}
+int
+afr_locked_nodes_count(unsigned char *locked_nodes, int child_count)
 
-static int32_t
-afr_unlock_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno)
 {
-        afr_trace_inodelk_out (frame, AFR_INODELK_TRANSACTION,
-                               AFR_UNLOCK_OP, NULL, op_ret,
-                               op_errno, (long) cookie);
-
-        if (op_ret < 0 && op_errno != ENOTCONN && op_errno != EBADFD) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "Unlock failed for some reason");
-        }
+    int i = 0;
+    int call_count = 0;
 
-        afr_unlock_common_cbk (frame, cookie, this, op_ret, op_errno);
-
-        return 0;
+    for (i = 0; i < child_count; i++) {
+        if (locked_nodes[i] & LOCKED_YES)
+            call_count++;
+    }
 
+    return call_count;
 }
 
-static int
-afr_unlock_inodelk (call_frame_t *frame, xlator_t *this)
+static void
+afr_log_locks_failure(call_frame_t *frame, char *where, char *what,
+                      int op_errno)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = NULL;
-
-        struct gf_flock flock;
-        int call_count = 0;
-        int i = 0;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        priv     = this->private;
+    xlator_t *this = frame->this;
+    gf_lkowner_t *lk_owner = &frame->root->lk_owner;
+    afr_local_t *local = frame->local;
+    const char *fop = NULL;
+    char *gfid = NULL;
+    const char *name = NULL;
 
-        flock.l_start = int_lock->lk_flock.l_start;
-        flock.l_len   = int_lock->lk_flock.l_len;
-        flock.l_type  = F_UNLCK;
+    fop = gf_fop_list[local->op];
 
-        call_count = afr_locked_nodes_count (int_lock->inode_locked_nodes,
-                                             priv->child_count);
-
-        int_lock->lk_call_count = call_count;
-
-        if (!call_count) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "No internal locks unlocked");
-                int_lock->lock_cbk (frame, this);
-                goto out;
-        }
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (int_lock->inode_locked_nodes[i] & LOCKED_YES) {
-                        if (local->fd) {
-                                afr_trace_inodelk_in (frame, AFR_INODELK_TRANSACTION,
-                                                      AFR_UNLOCK_OP, &flock, F_SETLK, i);
-
-                                STACK_WIND_COOKIE (frame, afr_unlock_inodelk_cbk,
-                                                   (void *) (long)i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->finodelk,
-                                                   this->name, local->fd,
-                                                   F_SETLK, &flock);
-
-                                if (!--call_count)
-                                        break;
-
-                        } else {
-                                afr_trace_inodelk_in (frame, AFR_INODELK_TRANSACTION,
-                                                      AFR_UNLOCK_OP, &flock, F_SETLK, i);
-
-                                STACK_WIND_COOKIE (frame, afr_unlock_inodelk_cbk,
-                                                   (void *) (long)i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->inodelk,
-                                                   this->name, &local->loc,
-                                                   F_SETLK, &flock);
-
-                                if (!--call_count)
-                                        break;
-
-                        }
-
-                }
-
-        }
-
-out:
-        return 0;
+    switch (local->transaction.type) {
+        case AFR_ENTRY_RENAME_TRANSACTION:
+        case AFR_ENTRY_TRANSACTION:
+            switch (local->op) {
+                case GF_FOP_LINK:
+                    gfid = uuid_utoa(local->newloc.pargfid);
+                    name = local->newloc.name;
+                    break;
+                default:
+                    gfid = uuid_utoa(local->loc.pargfid);
+                    name = local->loc.name;
+                    break;
+            }
+            gf_msg(this->name, GF_LOG_WARNING, op_errno,
+                   AFR_MSG_INTERNAL_LKS_FAILED,
+                   "Unable to do entry %s with lk-owner:%s on %s "
+                   "while attempting %s on {pgfid:%s, name:%s}.",
+                   what, lkowner_utoa(lk_owner), where, fop, gfid, name);
+            break;
+        case AFR_DATA_TRANSACTION:
+        case AFR_METADATA_TRANSACTION:
+            gfid = uuid_utoa(local->inode->gfid);
+            gf_msg(this->name, GF_LOG_WARNING, op_errno,
+                   AFR_MSG_INTERNAL_LKS_FAILED,
+                   "Unable to do inode %s with lk-owner:%s on %s "
+                   "while attempting %s on gfid:%s.",
+                   what, lkowner_utoa(lk_owner), where, fop, gfid);
+            break;
+    }
 }
 
 static int32_t
-afr_unlock_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno)
+afr_unlock_common_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        afr_trace_entrylk_out (frame, AFR_ENTRYLK_TRANSACTION,
-                              AFR_UNLOCK_OP, NULL, op_ret,
-                               op_errno, (long) cookie);
-
-        afr_unlock_common_cbk (frame, cookie, this, op_ret, op_errno);
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    afr_internal_lock_t *int_lock = NULL;
+    int lockee_num = 0;
+    int call_count = 0;
+    int child_index = 0;
+    int ret = 0;
 
-        return 0;
-}
-
-static int
-afr_unlock_entrylk (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = NULL;
-        const char          *basename = NULL;
-        loc_t               *loc      = NULL;
-
-        int call_count = 0;
-        int i = -1;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        priv     = this->private;
-
-        basename = int_lock->lk_basename;
-        if (int_lock->lk_loc)
-                loc = int_lock->lk_loc;
-
-        call_count = afr_locked_nodes_count (int_lock->entry_locked_nodes,
-                                             priv->child_count);
-        int_lock->lk_call_count = call_count;
-
-        if (!call_count){
-                gf_log (this->name, GF_LOG_TRACE,
-                        "No internal locks unlocked");
-                int_lock->lock_cbk (frame, this);
-                goto out;
-        }
+    local = frame->local;
+    int_lock = &local->internal_lock;
+    priv = this->private;
+    lockee_num = (int)((long)cookie) / priv->child_count;
+    child_index = (int)((long)cookie) % priv->child_count;
 
-        for (i = 0; i < priv->child_count; i++) {
-                if (int_lock->entry_locked_nodes[i] & LOCKED_YES) {
-                        afr_trace_entrylk_in (frame, AFR_ENTRYLK_NB_TRANSACTION,
-                                              AFR_UNLOCK_OP, basename, i);
-
-                        STACK_WIND_COOKIE (frame, afr_unlock_entrylk_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->entrylk,
-                                           this->name,
-                                           loc, basename,
-                                           ENTRYLK_UNLOCK, ENTRYLK_WRLCK);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
+    if (op_ret < 0 && op_errno != ENOTCONN && op_errno != EBADFD) {
+        afr_log_locks_failure(frame, priv->children[child_index]->name,
+                              "unlock", op_errno);
+    }
 
-out:
-        return 0;
+    int_lock->lockee[lockee_num].locked_nodes[child_index] &= LOCKED_NO;
+    if (local->transaction.type == AFR_DATA_TRANSACTION && op_ret != 1)
+        ret = afr_write_subvol_reset(frame, this);
 
-}
+    LOCK(&frame->lock);
+    {
+        call_count = --int_lock->lk_call_count;
+    }
+    UNLOCK(&frame->lock);
 
-static int32_t
-afr_lock_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-	      int32_t op_ret, int32_t op_errno)
-{
-        afr_internal_lock_t *int_lock = NULL;
-	afr_local_t         *local    = NULL;
-	afr_private_t       *priv     = NULL;
-
-	int done = 0;
-	int child_index = (long) cookie;
-
-	local    = frame->local;
-        int_lock = &local->internal_lock;
-	priv     = this->private;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			if (op_errno == ENOSYS) {
-				/* return ENOTSUP */
-				gf_log (this->name, GF_LOG_ERROR,
-					"subvolume does not support locking. "
-					"please load features/posix-locks xlator on server");
-				local->op_ret = op_ret;
-                                int_lock->lock_op_ret = op_ret;
-				done = 1;
-			}
-
-			local->child_up[child_index] = 0;
-			local->op_errno              = op_errno;
-                        int_lock->lock_op_errno      = op_errno;
-		}
-	}
-	UNLOCK (&frame->lock);
-
-        if ((op_ret == -1) &&
-            (op_errno == ENOSYS)) {
-                afr_unlock (frame, this);
-        } else {
-                if (op_ret == 0) {
-                        int_lock->locked_nodes[child_index]
-                                |= LOCKED_YES;
-                        int_lock->lock_count++;
-                }
-                afr_lock_blocking (frame, this, child_index + 1);
-        }
+    if (call_count == 0) {
+        int_lock->lock_cbk(frame, this);
+    }
 
-	return 0;
+    return ret;
 }
 
-static int32_t
-afr_blocking_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t op_ret, int32_t op_errno)
-{
-        afr_trace_inodelk_out (frame, AFR_INODELK_TRANSACTION,
-                               AFR_LOCK_OP, NULL, op_ret,
-                               op_errno, (long) cookie);
-
-        afr_lock_cbk (frame, cookie, this, op_ret, op_errno);
-        return 0;
-
-}
-
-static int32_t
-afr_lock_lower_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno)
-{
-        afr_internal_lock_t *int_lock    = NULL;
-        afr_private_t       *priv        = NULL;
-        afr_local_t         *local       = NULL;
-        loc_t               *lower       = NULL;
-        loc_t               *higher      = NULL;
-        const char          *lower_name  = NULL;
-        const char          *higher_name = NULL;
-
-        int child_index = (long) cookie;
-
-        priv     = this->private;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret == -1) {
-                        if (op_errno == ENOSYS) {
-                                /* return ENOTSUP */
-
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "subvolume does not support locking. "
-                                        "please load features/posix-locks xlator on server");
-
-                                local->op_ret   = op_ret;
-                        }
-
-                        local->child_up[child_index] = 0;
-                        local->op_errno = op_errno;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (op_ret != 0) {
-                afr_unlock (frame, this);
-                goto out;
-        } else {
-                int_lock->lower_locked_nodes[child_index] |= LOCKED_LOWER;
-                int_lock->lock_count++;
-        }
-
-        /* The lower path has been locked. Now lock the higher path */
-
-        lower = lower_path (&local->transaction.parent_loc,
-                            local->transaction.basename,
-                            &local->transaction.new_parent_loc,
-                            local->transaction.new_basename);
-
-        lower_name = (lower == &local->transaction.parent_loc ?
-                      local->transaction.basename :
-                      local->transaction.new_basename);
-
-        higher = (lower == &local->transaction.parent_loc ?
-                  &local->transaction.new_parent_loc :
-                  &local->transaction.parent_loc);
-
-        higher_name = (higher == &local->transaction.parent_loc ?
-                       local->transaction.basename :
-                       local->transaction.new_basename);
-
-        afr_trace_entrylk_in (frame, AFR_ENTRYLK_TRANSACTION,
-                              AFR_LOCK_OP, higher_name, child_index);
-
-
-        STACK_WIND_COOKIE (frame, afr_lock_cbk,
-                           (void *) (long) child_index,
-                           priv->children[child_index],
-                           priv->children[child_index]->fops->entrylk,
-                           this->name, higher, higher_name,
-                           ENTRYLK_LOCK, ENTRYLK_WRLCK);
-
-out:
-        return 0;
-}
-
-static int32_t
-afr_blocking_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t op_ret, int32_t op_errno)
-{
-        afr_trace_entrylk_out (frame, AFR_ENTRYLK_TRANSACTION,
-                               AFR_LOCK_OP, NULL, op_ret,
-                               op_errno, (long)cookie);
-
-        afr_lock_cbk (frame, cookie, this, op_ret, op_errno);
-        return 0;
-}
-
-static int
-afr_copy_locked_nodes (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = NULL;
-
-        priv     = this->private;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        switch (local->transaction.type) {
-	case AFR_DATA_TRANSACTION:
-	case AFR_METADATA_TRANSACTION:
-                memcpy (int_lock->inode_locked_nodes,
-                        int_lock->locked_nodes,
-                        priv->child_count);
-                int_lock->inodelk_lock_count = int_lock->lock_count;
-                break;
-
-        case AFR_ENTRY_RENAME_TRANSACTION:
+void
+afr_internal_lock_wind(call_frame_t *frame,
+                       int32_t (*cbk)(call_frame_t *, void *, xlator_t *,
+                                      int32_t, int32_t, dict_t *),
+                       void *cookie, int child, int lockee_num,
+                       gf_boolean_t blocking, gf_boolean_t unlock)
+{
+    afr_local_t *local = frame->local;
+    xlator_t *this = frame->this;
+    afr_private_t *priv = this->private;
+    afr_internal_lock_t *int_lock = &local->internal_lock;
+    entrylk_cmd cmd = ENTRYLK_LOCK_NB;
+    int32_t cmd1 = F_SETLK;
+    struct gf_flock flock = {
+        0,
+    };
+
+    switch (local->transaction.type) {
         case AFR_ENTRY_TRANSACTION:
-                memcpy (int_lock->entry_locked_nodes,
-                        int_lock->locked_nodes,
-                        priv->child_count);
-                int_lock->entrylk_lock_count = int_lock->lock_count;
-                break;
-        }
-
-        return 0;
-
-}
-
-int
-afr_lock_blocking (call_frame_t *frame, xlator_t *this, int child_index)
-{
-        afr_internal_lock_t *int_lock    = NULL;
-        afr_local_t         *local       = NULL;
-	afr_private_t       *priv        = NULL;
-        afr_fd_ctx_t        *fd_ctx      = NULL;
-	loc_t               *lower       = NULL;
-	loc_t               *higher      = NULL;
-	const char          *lower_name  = NULL;
-	const char          *higher_name = NULL;
-
-	struct gf_flock flock;
-        uint64_t ctx;
-        int ret = 0;
-
-	local    = frame->local;
-        int_lock = &local->internal_lock;
-	priv     = this->private;
-
-	flock.l_start = int_lock->lk_flock.l_start;
-	flock.l_len   = int_lock->lk_flock.l_len;
-	flock.l_type  = int_lock->lk_flock.l_type;
-
-        if (local->fd) {
-                ret = fd_ctx_get (local->fd, this, &ctx);
-
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "unable to get fd ctx for fd=%p",
-                                local->fd);
-
-                        local->op_ret           = -1;
-                        int_lock->lock_op_ret   = -1;
-                        local->op_errno         = EINVAL;
-                        int_lock->lock_op_errno = EINVAL;
-
-                        afr_copy_locked_nodes (frame, this);
-
-                        afr_unlock (frame, this);
-
-                        return 0;
-                }
-
-                fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-                /* skip over children that or down
-                   or don't have the fd open */
-
-                while ((child_index < priv->child_count)
-                       && (!local->child_up[child_index]
-                           || !fd_ctx->opened_on[child_index]))
-
-                        child_index++;
-        } else {
-                /* skip over children that are down */
-                while ((child_index < priv->child_count)
-                       && !local->child_up[child_index])
-                        child_index++;
-        }
-
-	if ((child_index == priv->child_count) &&
-	    int_lock->lock_count == 0) {
-
-		gf_log (this->name, GF_LOG_DEBUG,
-			"unable to lock on even one child");
-
-		local->op_ret           = -1;
-                int_lock->lock_op_ret   = -1;
-		local->op_errno         = EAGAIN;
-                int_lock->lock_op_errno = EAGAIN;
-
-                afr_copy_locked_nodes (frame, this);
-
-                afr_unlock(frame, this);
-
-		return 0;
-
-	}
-
-	if ((child_index == priv->child_count)
-	    || (int_lock->lock_count ==
-		afr_up_children_count (priv->child_count,
-                                       local->child_up))) {
-
-		/* we're done locking */
-
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "we're done locking");
-
-                afr_copy_locked_nodes (frame, this);
-
-                int_lock->lock_op_ret = 0;
-                int_lock->lock_cbk (frame, this);
-		return 0;
-	}
-
-	switch (local->transaction.type) {
-	case AFR_DATA_TRANSACTION:
-	case AFR_METADATA_TRANSACTION:
-
-		if (local->fd) {
-                        afr_trace_inodelk_in (frame, AFR_INODELK_TRANSACTION,
-                                              AFR_LOCK_OP, &flock, F_SETLKW,
-                                              child_index);
-
-			STACK_WIND_COOKIE (frame, afr_blocking_inodelk_cbk,
-					   (void *) (long) child_index,
-					   priv->children[child_index],
-					   priv->children[child_index]->fops->finodelk,
-					   this->name, local->fd,
-                                           F_SETLKW, &flock);
-
-		} else {
-                        afr_trace_inodelk_in (frame, AFR_INODELK_TRANSACTION,
-                                              AFR_LOCK_OP, &flock, F_SETLKW,
-                                              child_index);
-
-			STACK_WIND_COOKIE (frame, afr_blocking_inodelk_cbk,
-					   (void *) (long) child_index,
-					   priv->children[child_index],
-					   priv->children[child_index]->fops->inodelk,
-					   this->name, &local->loc,
-                                           F_SETLKW, &flock);
-		}
-
-		break;
-
-	case AFR_ENTRY_RENAME_TRANSACTION:
-	{
-		lower = lower_path (&local->transaction.parent_loc,
-				    local->transaction.basename,
-				    &local->transaction.new_parent_loc,
-				    local->transaction.new_basename);
-
-		lower_name = (lower == &local->transaction.parent_loc ?
-			      local->transaction.basename :
-			      local->transaction.new_basename);
-
-		higher = (lower == &local->transaction.parent_loc ?
-			  &local->transaction.new_parent_loc :
-			  &local->transaction.parent_loc);
-
-		higher_name = (higher == &local->transaction.parent_loc ?
-			       local->transaction.basename :
-			       local->transaction.new_basename);
-
-                afr_trace_entrylk_in (frame, AFR_ENTRYLK_TRANSACTION,
-                                      AFR_LOCK_OP, lower_name, child_index);
-
-
-		STACK_WIND_COOKIE (frame, afr_lock_lower_cbk,
-				   (void *) (long) child_index,
-				   priv->children[child_index],
-				   priv->children[child_index]->fops->entrylk,
-				   this->name, lower, lower_name,
-				   ENTRYLK_LOCK, ENTRYLK_WRLCK);
-
-		break;
-	}
-
-	case AFR_ENTRY_TRANSACTION:
-		if (local->fd) {
-                        afr_trace_entrylk_in (frame, AFR_ENTRYLK_TRANSACTION,
-                                              AFR_LOCK_OP, local->transaction.basename,
-                                              child_index);
-
-			STACK_WIND_COOKIE (frame, afr_blocking_entrylk_cbk,
-					   (void *) (long) child_index,
-					   priv->children[child_index],
-					   priv->children[child_index]->fops->fentrylk,
-					   this->name, local->fd,
-					   local->transaction.basename,
-					   ENTRYLK_LOCK, ENTRYLK_WRLCK);
-		} else {
-                        afr_trace_entrylk_in (frame, AFR_ENTRYLK_TRANSACTION,
-                                              AFR_LOCK_OP, local->transaction.basename,
-                                              child_index);
-
-			STACK_WIND_COOKIE (frame, afr_blocking_entrylk_cbk,
-					   (void *) (long) child_index,
-					   priv->children[child_index],
-					   priv->children[child_index]->fops->entrylk,
-					   this->name,
-                                           &local->transaction.parent_loc,
-					   local->transaction.basename,
-					   ENTRYLK_LOCK, ENTRYLK_WRLCK);
-		}
-
-		break;
-	}
-
-	return 0;
-
-
+        case AFR_ENTRY_RENAME_TRANSACTION:
+            if (unlock) {
+                cmd = ENTRYLK_UNLOCK;
+            } else if (blocking) { /*Doesn't make sense to have blocking
+                                      unlock*/
+                cmd = ENTRYLK_LOCK;
+            }
+
+            if (local->fd) {
+                STACK_WIND_COOKIE(frame, cbk, cookie, priv->children[child],
+                                  priv->children[child]->fops->fentrylk,
+                                  int_lock->domain,
+                                  int_lock->lockee[lockee_num].fd,
+                                  int_lock->lockee[lockee_num].basename, cmd,
+                                  ENTRYLK_WRLCK, NULL);
+            } else {
+                STACK_WIND_COOKIE(frame, cbk, cookie, priv->children[child],
+                                  priv->children[child]->fops->entrylk,
+                                  int_lock->domain,
+                                  &int_lock->lockee[lockee_num].loc,
+                                  int_lock->lockee[lockee_num].basename, cmd,
+                                  ENTRYLK_WRLCK, NULL);
+            }
+            break;
+
+        case AFR_DATA_TRANSACTION:
+        case AFR_METADATA_TRANSACTION:
+            flock = int_lock->lockee[lockee_num].flock;
+            if (unlock) {
+                flock.l_type = F_UNLCK;
+            } else if (blocking) { /*Doesn't make sense to have blocking
+                                      unlock*/
+                cmd1 = F_SETLKW;
+            }
+
+            if (local->fd) {
+                STACK_WIND_COOKIE(
+                    frame, cbk, cookie, priv->children[child],
+                    priv->children[child]->fops->finodelk, int_lock->domain,
+                    int_lock->lockee[lockee_num].fd, cmd1, &flock, NULL);
+            } else {
+                STACK_WIND_COOKIE(
+                    frame, cbk, cookie, priv->children[child],
+                    priv->children[child]->fops->inodelk, int_lock->domain,
+                    &int_lock->lockee[lockee_num].loc, cmd1, &flock, NULL);
+            }
+            break;
+    }
 }
 
-int32_t
-afr_blocking_lock (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = NULL;
-
-        priv     = this->private;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        switch (local->transaction.type) {
-	case AFR_DATA_TRANSACTION:
-	case AFR_METADATA_TRANSACTION:
-                initialize_inodelk_variables (frame, this);
-                break;
-
-        case AFR_ENTRY_RENAME_TRANSACTION:
-        case AFR_ENTRY_TRANSACTION:
-                initialize_entrylk_variables (frame, this);
+static int
+afr_unlock_now(call_frame_t *frame, xlator_t *this)
+{
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int call_count = 0;
+    int child_index = 0;
+    int lockee_num = 0;
+    int i = -1;
+
+    local = frame->local;
+    int_lock = &local->internal_lock;
+    priv = this->private;
+
+    call_count = afr_lockee_locked_nodes_count(int_lock);
+
+    int_lock->lk_call_count = call_count;
+
+    if (!call_count) {
+        gf_msg_trace(this->name, 0, "No internal locks unlocked");
+        int_lock->lock_cbk(frame, this);
+        goto out;
+    }
+
+    for (i = 0; i < int_lock->lockee_count * priv->child_count; i++) {
+        lockee_num = i / priv->child_count;
+        child_index = i % priv->child_count;
+        if (int_lock->lockee[lockee_num].locked_nodes[child_index] &
+            LOCKED_YES) {
+            afr_internal_lock_wind(frame, afr_unlock_common_cbk,
+                                   (void *)(long)i, child_index, lockee_num,
+                                   _gf_false, _gf_true);
+            if (!--call_count)
                 break;
         }
+    }
 
-        afr_lock_blocking (frame, this, 0);
-
-        return 0;
+out:
+    return 0;
 }
 
 static int32_t
-afr_nonblocking_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                             int32_t op_ret, int32_t op_errno)
-{
-        afr_internal_lock_t *int_lock = NULL;
-	afr_local_t         *local    = NULL;
-	afr_private_t       *priv     = NULL;
-
-        int call_count          = 0;
-	int child_index         = (long) cookie;
-
-	local    = frame->local;
-        int_lock = &local->internal_lock;
-	priv     = this->private;
-
-        afr_trace_entrylk_out (frame, AFR_ENTRYLK_TRANSACTION,
-                               AFR_LOCK_OP, NULL, op_ret,
-                               op_errno, (long) cookie);
-
-        LOCK (&frame->lock);
-        {
-                call_count = --int_lock->lk_call_count;
-        }
-        UNLOCK (&frame->lock);
-
-        if (op_ret < 0 ) {
-			if (op_errno == ENOSYS) {
-				/* return ENOTSUP */
-				gf_log (this->name, GF_LOG_ERROR,
-					"subvolume does not support locking. "
-					"please load features/posix-locks xlator on server");
-				local->op_ret         = op_ret;
-                                int_lock->lock_op_ret = op_ret;
-
-                                local->child_up[child_index] = 0;
-                                int_lock->lock_op_errno      = op_errno;
-                                local->op_errno              = op_errno;
-                        }
-        } else if (op_ret == 0) {
-                        int_lock->entry_locked_nodes[child_index]
-                                |= LOCKED_YES;
-                        int_lock->entrylk_lock_count++;
-        }
-
-        if (call_count == 0) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "Last locking reply received");
-                /* all locks successfull. Proceed to call FOP */
-                if (int_lock->entrylk_lock_count ==
-                    afr_up_children_count (priv->child_count, local->child_up)) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "All servers locked. Calling the cbk");
-                        int_lock->lock_op_ret = 0;
-                        int_lock->lock_cbk (frame, this);
-                }
-                /* Not all locks were successfull. Unlock and try locking
-                   again, this time with serially blocking locks */
-                else {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "%d servers locked. Trying again with blocking calls",
-                                int_lock->lock_count);
-
-                        afr_unlock(frame, this);
+afr_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, dict_t *xdata)
+{
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int cky = (long)cookie;
+    int child_index = 0;
+    int lockee_num = 0;
+
+    priv = this->private;
+    local = frame->local;
+    int_lock = &local->internal_lock;
+
+    child_index = ((int)cky) % priv->child_count;
+    lockee_num = ((int)cky) / priv->child_count;
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            if (op_errno == ENOSYS) {
+                /* return ENOTSUP */
+                gf_msg(this->name, GF_LOG_ERROR, ENOSYS,
+                       AFR_MSG_LOCK_XLATOR_NOT_LOADED,
+                       "subvolume does not support locking. "
+                       "please load features/locks xlator on server");
+                local->op_ret = op_ret;
+                int_lock->lock_op_ret = op_ret;
+            }
+
+            local->op_errno = op_errno;
+            int_lock->lock_op_errno = op_errno;
+        }
+
+        int_lock->lk_attempted_count++;
+    }
+    UNLOCK(&frame->lock);
+
+    if ((op_ret == -1) && (op_errno == ENOSYS)) {
+        afr_unlock_now(frame, this);
+    } else {
+        if (op_ret == 0) {
+            int_lock->lockee[lockee_num]
+                .locked_nodes[child_index] |= LOCKED_YES;
+            int_lock->lockee[lockee_num].locked_count++;
+            int_lock->lock_count++;
+            if (local->transaction.type == AFR_DATA_TRANSACTION) {
+                LOCK(&local->inode->lock);
+                {
+                    local->inode_ctx->lock_count++;
                 }
+                UNLOCK(&local->inode->lock);
+            }
         }
+        afr_lock_blocking(frame, this, cky + 1);
+    }
 
-        return 0;
+    return 0;
 }
 
-int
-afr_nonblocking_entrylk (call_frame_t *frame, xlator_t *this)
+static gf_boolean_t
+_is_lock_wind_needed(afr_local_t *local, int child_index)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = NULL;
-        afr_fd_ctx_t        *fd_ctx   = NULL;
-        const char          *basename = NULL;
-        loc_t               *loc      = NULL;
-
-        int32_t call_count = 0;
-        int i = 0;
-        uint64_t  ctx;
-        int ret = 0;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        priv     = this->private;
-
-        initialize_entrylk_variables (frame, this);
-
-        basename = int_lock->lk_basename;
-        if (int_lock->lk_loc)
-                loc = int_lock->lk_loc;
-
-        if (local->fd) {
-                ret = fd_ctx_get (local->fd, this, &ctx);
-
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "unable to get fd ctx for fd=%p",
-                                local->fd);
-
-                        local->op_ret           = -1;
-                        int_lock->lock_op_ret   = -1;
-                        local->op_errno         = EINVAL;
-                        int_lock->lock_op_errno = EINVAL;
-
-                        return -1;
-                }
-
-                fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-                call_count = internal_lock_count (frame, this, fd_ctx);
-                int_lock->lk_call_count = call_count;
-
-                /* Send non-blocking entrylk calls only on up children
-                   and where the fd has been opened */
-                for (i = 0; i < priv->child_count; i++) {
-                        if (local->child_up[i] && fd_ctx->opened_on[i]) {
-                                afr_trace_entrylk_in (frame, AFR_ENTRYLK_NB_TRANSACTION,
-                                                      AFR_LOCK_OP, basename, i);
-
-                                STACK_WIND_COOKIE (frame, afr_nonblocking_entrylk_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->fentrylk,
-                                                   this->name, local->fd,
-                                                   basename,
-                                                   ENTRYLK_LOCK_NB, ENTRYLK_WRLCK);
-                        }
-                }
-        } else {
-                GF_ASSERT (loc);
-
-                call_count = internal_lock_count (frame, this, NULL);
-                int_lock->lk_call_count = call_count;
-
-                for (i = 0; i < priv->child_count; i++) {
-                        if (local->child_up[i]) {
-                                afr_trace_entrylk_in (frame, AFR_ENTRYLK_NB_TRANSACTION,
-                                                      AFR_LOCK_OP, basename, i);
-
-                                STACK_WIND_COOKIE (frame, afr_nonblocking_entrylk_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->entrylk,
-                                                   this->name, loc, basename,
-                                                   ENTRYLK_LOCK, ENTRYLK_WRLCK);
-
-                                if (!--call_count)
-                                        break;
+    if (!local->child_up[child_index])
+        return _gf_false;
 
-                        }
-                }
-        }
-
-        return 0;
+    return _gf_true;
 }
 
-int32_t
-afr_nonblocking_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                             int32_t op_ret, int32_t op_errno)
+static gf_boolean_t
+is_blocking_locks_count_sufficient(call_frame_t *frame, xlator_t *this)
 {
-        afr_internal_lock_t *int_lock = NULL;
-	afr_local_t         *local    = NULL;
-	afr_private_t       *priv     = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    afr_internal_lock_t *int_lock = NULL;
+    int child = 0;
+    int nlockee = 0;
+    int lockee_count = 0;
+    gf_boolean_t ret = _gf_true;
 
-        int call_count  = 0;
-	int child_index = (long) cookie;
-
-	local    = frame->local;
-        int_lock = &local->internal_lock;
-	priv     = this->private;
-
-        afr_trace_inodelk_out (frame, AFR_INODELK_NB_TRANSACTION,
-                               AFR_LOCK_OP, NULL, op_ret,
-                               op_errno, (long) cookie);
-
-        LOCK (&frame->lock);
-        {
-                call_count = --int_lock->lk_call_count;
-        }
-        UNLOCK (&frame->lock);
-
-        if (op_ret < 0 ) {
-			if (op_errno == ENOSYS) {
-				/* return ENOTSUP */
-				gf_log (this->name, GF_LOG_ERROR,
-					"subvolume does not support locking. "
-					"please load features/posix-locks xlator on server");
-				local->op_ret                = op_ret;
-                                int_lock->lock_op_ret        = op_ret;
-                                local->child_up[child_index] = 0;
-                                int_lock->lock_op_errno      = op_errno;
-                                local->op_errno              = op_errno;
-                        }
-        } else if (op_ret == 0) {
-                        int_lock->inode_locked_nodes[child_index]
-                                |= LOCKED_YES;
-                        int_lock->inodelk_lock_count++;
-        }
+    local = frame->local;
+    priv = this->private;
+    int_lock = &local->internal_lock;
+    lockee_count = int_lock->lockee_count;
 
-        if (call_count == 0) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "Last inode locking reply received");
-                /* all locks successfull. Proceed to call FOP */
-                if (int_lock->inodelk_lock_count ==
-                    afr_up_children_count (priv->child_count, local->child_up)) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "All servers locked. Calling the cbk");
-                        int_lock->lock_op_ret = 0;
-                        int_lock->lock_cbk (frame, this);
-                }
-                /* Not all locks were successfull. Unlock and try locking
-                   again, this time with serially blocking locks */
-                else {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "%d servers locked. Trying again with blocking calls",
-                                int_lock->lock_count);
-
-                        afr_unlock(frame, this);
-                }
+    if (int_lock->lock_count == 0) {
+        afr_log_locks_failure(frame, "any subvolume", "lock",
+                              int_lock->lock_op_errno);
+        return _gf_false;
+    }
+    /* For FOPS that take multiple sets of locks (mkdir, rename),
+     * there must be at least one brick on which the locks from
+     * all lock sets were successful. */
+    for (child = 0; child < priv->child_count; child++) {
+        ret = _gf_true;
+        for (nlockee = 0; nlockee < lockee_count; nlockee++) {
+            if (!(int_lock->lockee[nlockee].locked_nodes[child] & LOCKED_YES))
+                ret = _gf_false;
         }
+        if (ret)
+            return ret;
+    }
+    if (!ret)
+        afr_log_locks_failure(frame, "all", "lock", int_lock->lock_op_errno);
 
-        return 0;
+    return ret;
 }
 
 int
-afr_nonblocking_inodelk (call_frame_t *frame, xlator_t *this)
+afr_lock_blocking(call_frame_t *frame, xlator_t *this, int cookie)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = NULL;
-        afr_fd_ctx_t        *fd_ctx   = NULL;
-
-        int32_t  call_count = 0;
-        uint64_t ctx        = 0;
-        int      i          = 0;
-        int      ret        = 0;
-        struct gf_flock flock;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        priv     = this->private;
-
-	flock.l_start = int_lock->lk_flock.l_start;
-	flock.l_len   = int_lock->lk_flock.l_len;
-	flock.l_type  = int_lock->lk_flock.l_type;
-
-        initialize_inodelk_variables (frame, this);
-
-        if (local->fd) {
-                ret = fd_ctx_get (local->fd, this, &ctx);
-
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "unable to get fd ctx for fd=%p",
-                                local->fd);
-
-                        local->op_ret           = -1;
-                        int_lock->lock_op_ret   = -1;
-                        local->op_errno         = EINVAL;
-                        int_lock->lock_op_errno = EINVAL;
-
-                        ret = -1;
-                        goto out;
-                }
-
-                fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    uint64_t ctx = 0;
+    int ret = 0;
+    int child_index = 0;
+    int lockee_num = 0;
 
-                call_count = internal_lock_count (frame, this, fd_ctx);
-                int_lock->lk_call_count = call_count;
+    local = frame->local;
+    int_lock = &local->internal_lock;
+    priv = this->private;
+    child_index = cookie % priv->child_count;
+    lockee_num = cookie / priv->child_count;
 
-                /* Send non-blocking inodelk calls only on up children
-                   and where the fd has been opened */
-                for (i = 0; i < priv->child_count; i++) {
-                        if (local->child_up[i] && fd_ctx->opened_on[i]) {
-                                afr_trace_inodelk_in (frame, AFR_INODELK_NB_TRANSACTION,
-                                                      AFR_LOCK_OP, &flock, F_SETLK, i);
+    if (local->fd) {
+        ret = fd_ctx_get(local->fd, this, &ctx);
 
-                                STACK_WIND_COOKIE (frame, afr_nonblocking_inodelk_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->finodelk,
-                                                   this->name, local->fd,
-                                                   F_SETLK, &flock);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_FD_CTX_GET_FAILED,
+                   "unable to get fd ctx for fd=%p", local->fd);
 
-                                if (!--call_count)
-                                        break;
+            local->op_ret = -1;
+            int_lock->lock_op_ret = -1;
 
-                        }
+            afr_unlock_now(frame, this);
 
-                }
-        } else {
-                call_count = internal_lock_count (frame, this, NULL);
-                int_lock->lk_call_count = call_count;
-
-                for (i = 0; i < priv->child_count; i++) {
-                        if (local->child_up[i]) {
-                                afr_trace_inodelk_in (frame, AFR_INODELK_NB_TRANSACTION,
-                                                      AFR_LOCK_OP, &flock, F_SETLK, i);
-
-                                STACK_WIND_COOKIE (frame, afr_nonblocking_inodelk_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->inodelk,
-                                                   this->name, &local->loc,
-                                                   F_SETLK, &flock);
-
-                                if (!--call_count)
-                                        break;
-
-                        }
-                }
+            return 0;
         }
+    }
 
-out:
-        return ret;
-}
+    if (int_lock->lk_expected_count == int_lock->lk_attempted_count) {
+        if (!is_blocking_locks_count_sufficient(frame, this)) {
+            local->op_ret = -1;
+            int_lock->lock_op_ret = -1;
 
-static int
-__is_lower_locked (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_private_t       *priv     = NULL;
-        afr_local_t         *local    = NULL;
-
-        int count = 0;
-        int i     = 0;
+            afr_unlock_now(frame, this);
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        priv     = this->private;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (int_lock->lower_locked_nodes[i] & LOCKED_LOWER)
-                        count++;
+            return 0;
         }
+    }
 
-        return count;
-
-}
-
-static int
-__is_higher_locked (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_private_t       *priv     = NULL;
-        afr_local_t         *local    = NULL;
-
-        int count = 0;
-        int i = 0;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        priv     = this->private;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (int_lock->locked_nodes[i] & LOCKED_YES)
-                        count++;
-        }
+    if (int_lock->lk_expected_count == int_lock->lk_attempted_count) {
+        /* we're done locking */
 
-        return count;
+        gf_msg_debug(this->name, 0, "we're done locking");
 
-}
-
-static int
-afr_unlock_lower_entrylk (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = NULL;
-        const char          *basename = NULL;
-        loc_t               *loc      = NULL;
-
-        int call_count = 0;
-        int i = -1;
-
-        local = frame->local;
-        int_lock = &local->internal_lock;
-        priv  = this->private;
-
-        basename = int_lock->lk_basename;
-        if (int_lock->lk_loc)
-                loc = int_lock->lk_loc;
-
-        call_count = __is_lower_locked (frame, this);
-        int_lock->lk_call_count = call_count;
-
-        if (!call_count){
-                gf_log (this->name, GF_LOG_TRACE,
-                        "No internal locks unlocked");
-                int_lock->lock_cbk (frame, this);
-                goto out;
-        }
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (int_lock->lower_locked_nodes[i] & LOCKED_LOWER) {
-                        afr_trace_entrylk_in (frame, AFR_ENTRYLK_NB_TRANSACTION,
-                                              AFR_UNLOCK_OP, basename, i);
-
-                        STACK_WIND_COOKIE (frame, afr_unlock_entrylk_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->entrylk,
-                                           this->name,
-                                           loc, basename,
-                                           ENTRYLK_UNLOCK, ENTRYLK_WRLCK);
-
-                        if (!--call_count)
-                                break;
-
-                }
-        }
-
-out:
+        int_lock->lock_op_ret = 0;
+        int_lock->lock_cbk(frame, this);
         return 0;
+    }
 
-}
-
-
-static int
-afr_post_unlock_higher_cbk (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
-
-        local = frame->local;
-
-        local->transaction.done (frame, this);
+    if (!_is_lock_wind_needed(local, child_index)) {
+        afr_lock_blocking(frame, this, cookie + 1);
         return 0;
-}
+    }
 
-static int
-afr_post_unlock_lower_cbk (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock    = NULL;
-        afr_local_t         *local       = NULL;
-	loc_t               *lower       = NULL;
-	loc_t               *higher      = NULL;
-	const char          *lower_name  = NULL;
-	const char          *higher_name = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        lower = lower_path (&local->transaction.parent_loc,
-                            local->transaction.basename,
-                            &local->transaction.new_parent_loc,
-                            local->transaction.new_basename);
-
-        lower_name = (lower == &local->transaction.parent_loc ?
-                      local->transaction.basename :
-                      local->transaction.new_basename);
-
-        higher = (lower == &local->transaction.parent_loc ?
-                  &local->transaction.new_parent_loc :
-                  &local->transaction.parent_loc);
-
-        higher_name = (higher == &local->transaction.parent_loc ?
-                       local->transaction.basename :
-                       local->transaction.new_basename);
-
-        if (__is_higher_locked (frame, this)) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "unlocking higher");
-                int_lock->lk_basename = higher_name;
-                int_lock->lk_loc      = higher;
-                int_lock->lock_cbk    = afr_post_unlock_higher_cbk;
-
-                afr_unlock_entrylk (frame, this);
-        } else
-                local->transaction.done (frame, this);
-
-        return 0;
-}
-
-static int
-afr_rename_unlock (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock    = NULL;
-        afr_local_t         *local       = NULL;
-	loc_t               *lower       = NULL;
-	loc_t               *higher      = NULL;
-	const char          *lower_name  = NULL;
-	const char          *higher_name = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        lower = lower_path (&local->transaction.parent_loc,
-                            local->transaction.basename,
-                            &local->transaction.new_parent_loc,
-                            local->transaction.new_basename);
-
-        lower_name = (lower == &local->transaction.parent_loc ?
-                      local->transaction.basename :
-                      local->transaction.new_basename);
-
-        higher = (lower == &local->transaction.parent_loc ?
-                  &local->transaction.new_parent_loc :
-                  &local->transaction.parent_loc);
-
-        higher_name = (higher == &local->transaction.parent_loc ?
-                       local->transaction.basename :
-                       local->transaction.new_basename);
-
-
-        if (__is_lower_locked (frame, this)) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "unlocking lower");
-                int_lock->lk_basename = lower_name;
-                int_lock->lk_loc      = lower;
-                int_lock->lock_cbk    = afr_post_unlock_lower_cbk;
-
-                afr_unlock_lower_entrylk (frame, this);
-        } else
-                afr_post_unlock_lower_cbk (frame, this);
-
-        return 0;
-}
-
-static int
-afr_rename_transaction (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
-
-        local = frame->local;
-
-        return (local->transaction.type ==
-                AFR_ENTRY_RENAME_TRANSACTION);
+    afr_internal_lock_wind(frame, afr_lock_cbk, (void *)(long)cookie,
+                           child_index, lockee_num, _gf_true, _gf_false);
 
+    return 0;
 }
 
 int32_t
-afr_unlock (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
-
-        local = frame->local;
-
-        if (transaction_lk_op (local)) {
-                if (is_afr_lock_transaction (local))
-                        afr_unlock_inodelk (frame, this);
-                else
-                        if (!afr_rename_transaction (frame, this))
-                                afr_unlock_entrylk (frame, this);
-                        else
-                                afr_rename_unlock (frame, this);
-        } else {
-                if (is_afr_lock_selfheal (local))
-                        afr_unlock_inodelk (frame, this);
-                else
-                        afr_unlock_entrylk (frame, this);
-        }
-
-        return 0;
-}
-
-int
-afr_mark_locked_nodes (xlator_t *this, fd_t *fd,
-                       unsigned char *locked_nodes)
+afr_blocking_lock(call_frame_t *frame, xlator_t *this)
 {
-        afr_private_t *priv  = NULL;
-        afr_fd_ctx_t  *fdctx = NULL;
-        uint64_t       tmp   = 0;
-        int            ret   = 0;
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int up_count = 0;
 
-        priv = this->private;
+    priv = this->private;
+    local = frame->local;
+    int_lock = &local->internal_lock;
 
-        afr_fd_ctx_set (this, fd);
-        if (ret < 0)
-                goto out;
+    up_count = AFR_COUNT(local->child_up, priv->child_count);
+    int_lock->lk_call_count = int_lock->lk_expected_count =
+        (int_lock->lockee_count * up_count);
+    initialize_internal_lock_variables(frame, this);
 
-        ret = fd_ctx_get (fd, this, &tmp);
-        fdctx = (afr_fd_ctx_t *) (long) tmp;
+    afr_lock_blocking(frame, this, 0);
 
-        GF_ASSERT (fdctx->locked_on);
-
-        memcpy (fdctx->locked_on, locked_nodes,
-                priv->child_count);
-
-out:
-        return ret;
+    return 0;
 }
 
-static int
-__is_fd_saved (xlator_t *this, fd_t *fd)
-{
-        afr_locked_fd_t *locked_fd = NULL;
-        afr_private_t   *priv      = NULL;
-        int              found     = 0;
-
-        priv = this->private;
-
-        list_for_each_entry (locked_fd, &priv->saved_fds, list) {
-                if (locked_fd->fd == fd) {
-                        found = 1;
-                        break;
-                }
-        }
-
-        return found;
-}
-
-static int
-__afr_save_locked_fd (xlator_t *this, fd_t *fd)
+static int32_t
+afr_nb_internal_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        afr_private_t   *priv      = NULL;
-        afr_locked_fd_t *locked_fd = NULL;
-        int              ret       = 0;
-
-        priv = this->private;
-
-        locked_fd = GF_CALLOC (1, sizeof (*locked_fd),
-                               gf_afr_mt_locked_fd);
-        if (!locked_fd) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                ret = -1;
-                goto out;
-        }
-
-        locked_fd->fd = fd;
-        INIT_LIST_HEAD (&locked_fd->list);
-
-        list_add_tail (&locked_fd->list, &priv->saved_fds);
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
+    int call_count = 0;
+    int child_index = 0;
+    int lockee_num = 0;
+    afr_private_t *priv = NULL;
 
-out:
-        return ret;
-}
+    priv = this->private;
 
-int
-afr_save_locked_fd (xlator_t *this, fd_t *fd)
-{
-        afr_private_t   *priv      = NULL;
-        int              ret       = 0;
+    child_index = ((long)cookie) % priv->child_count;
+    lockee_num = ((long)cookie) / priv->child_count;
 
-        priv = this->private;
+    local = frame->local;
+    int_lock = &local->internal_lock;
 
-        pthread_mutex_lock (&priv->mutex);
+    if (op_ret == 0 && local->transaction.type == AFR_DATA_TRANSACTION) {
+        LOCK(&local->inode->lock);
         {
-                if (__is_fd_saved (this, fd)) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "fd=%p already saved", fd);
-                        goto unlock;
-                }
-
-                ret = __afr_save_locked_fd (this, fd);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "fd=%p could not be saved");
-                        goto unlock;
-                }
+            local->inode_ctx->lock_count++;
         }
-unlock:
-        pthread_mutex_unlock (&priv->mutex);
-
-        return ret;
-}
-
-static int
-afr_lock_recovery_cleanup (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t     *local     = NULL;
-        afr_locked_fd_t *locked_fd = NULL;
-
-        local = frame->local;
-
-        locked_fd = local->locked_fd;
-
-        STACK_DESTROY (frame->root);
-        afr_local_cleanup (local, this);
-
-        afr_save_locked_fd (this, locked_fd->fd);
-
-        return 0;
-
-}
-
-static int
-afr_get_source_lock_recovery (xlator_t *this, fd_t *fd)
-{
-        afr_fd_ctx_t  *fdctx        = NULL;
-        afr_private_t *priv         = NULL;
-        uint64_t      tmp           = 0;
-        int           i             = 0;
-        int           source_child  = -1;
-        int           ret           = 0;
-
-        priv = this->private;
-
-        ret = fd_ctx_get (fd, this, &tmp);
-        if (ret)
-                goto out;
-
-        fdctx = (afr_fd_ctx_t *) (long) tmp;
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (fdctx->locked_on[i]) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Found lock recovery source=%d",
-                                i);
-                        source_child = i;
-                        break;
-                }
-
-        }
-
-out:
-        return source_child;
-
-}
+        UNLOCK(&local->inode->lock);
+    }
 
-int32_t
-afr_get_locks_fd_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, struct gf_flock *lock);
-int32_t
-afr_recover_lock_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, struct gf_flock *lock)
-{
-        afr_local_t   *local = NULL;
-        afr_private_t *priv  = NULL;
-        int32_t source_child = 0;
-        struct gf_flock flock   = {0,};
-
-        local = frame->local;
-        priv  = this->private;
-
-        if (op_ret) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "lock recovery failed");
-                goto cleanup;
+    LOCK(&frame->lock);
+    {
+        if (op_ret < 0) {
+            if (op_errno == ENOSYS) {
+                /* return ENOTSUP */
+                gf_msg(this->name, GF_LOG_ERROR, ENOSYS,
+                       AFR_MSG_LOCK_XLATOR_NOT_LOADED,
+                       "subvolume does not support "
+                       "locking. please load features/locks"
+                       " xlator on server");
+                local->op_ret = op_ret;
+                int_lock->lock_op_ret = op_ret;
+
+                int_lock->lock_op_errno = op_errno;
+                local->op_errno = op_errno;
+            }
+        } else if (op_ret == 0) {
+            int_lock->lockee[lockee_num]
+                .locked_nodes[child_index] |= LOCKED_YES;
+            int_lock->lockee[lockee_num].locked_count++;
+            int_lock->lock_count++;
         }
 
-        source_child = local->source_child;
-
-        memcpy (&flock, lock, sizeof (*lock));
-
-        STACK_WIND_COOKIE (frame, afr_get_locks_fd_cbk,
-                           (void *) (long) source_child,
-                           priv->children[source_child],
-                           priv->children[source_child]->fops->lk,
-                           local->fd, F_GETLK_FD, &flock);
-
-        return 0;
-
-cleanup:
-        afr_lock_recovery_cleanup (frame, this);
-        return 0;
-}
-
-int
-afr_recover_lock (call_frame_t *frame, xlator_t *this,
-                  struct gf_flock *flock)
-{
-        afr_local_t   *local             = NULL;
-        afr_private_t *priv              = NULL;
-        int32_t      lock_recovery_child = 0;
-
-        priv  = this->private;
-        local = frame->local;
-
-        lock_recovery_child = local->lock_recovery_child;
-
-        frame->root->lk_owner = flock->l_owner;
+        call_count = --int_lock->lk_call_count;
+    }
+    UNLOCK(&frame->lock);
 
-        STACK_WIND_COOKIE (frame, afr_recover_lock_cbk,
-                           (void *) (long) lock_recovery_child,
-                           priv->children[lock_recovery_child],
-                           priv->children[lock_recovery_child]->fops->lk,
-                           local->fd, F_SETLK, flock);
-
-        return 0;
-}
-
-static int
-is_afr_lock_eol (struct gf_flock *lock)
-{
-        int ret = 0;
-
-        if ((lock->l_type == GF_LK_EOL))
-                ret = 1;
-
-        return ret;
-}
-
-int32_t
-afr_get_locks_fd_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, struct gf_flock *lock)
-{
-        if (op_ret) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Failed to get locks on fd");
-                goto cleanup;
+    if (call_count == 0) {
+        gf_msg_trace(this->name, 0, "Last locking reply received");
+        /* all locks successful. Proceed to call FOP */
+        if (int_lock->lock_count == int_lock->lk_expected_count) {
+            gf_msg_trace(this->name, 0, "All servers locked. Calling the cbk");
+            int_lock->lock_op_ret = 0;
+            int_lock->lock_cbk(frame, this);
         }
+        /* Not all locks were successful. Unlock and try locking
+           again, this time with serially blocking locks */
+        else {
+            gf_msg_trace(this->name, 0,
+                         "%d servers locked. Trying again "
+                         "with blocking calls",
+                         int_lock->lock_count);
 
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Got a lock on fd");
-
-        if (is_afr_lock_eol (lock)) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Reached EOL on locks on fd");
-                goto cleanup;
+            afr_unlock_now(frame, this);
         }
+    }
 
-        afr_recover_lock (frame, this, lock);
-
-        return 0;
-
-cleanup:
-        afr_lock_recovery_cleanup (frame, this);
-
-        return 0;
+    return 0;
 }
 
-static int
-afr_lock_recovery (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t   *local        = NULL;
-        afr_private_t *priv         = NULL;
-        fd_t          *fd           = NULL;
-        int            ret          = 0;
-        int32_t        source_child = 0;
-        struct gf_flock   flock        = {0,};
-
-        priv  = this->private;
-        local = frame->local;
-
-        fd = local->fd;
-
-        source_child = afr_get_source_lock_recovery (this, fd);
-        if (source_child < 0) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Could not recover locks due to lock "
-                        "split brain");
-                ret = -1;
-                goto out;
+int
+afr_lock_nonblocking(call_frame_t *frame, xlator_t *this)
+{
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    afr_fd_ctx_t *fd_ctx = NULL;
+    int child = 0;
+    int lockee_num = 0;
+    int32_t call_count = 0;
+    int i = 0;
+    int ret = 0;
+
+    local = frame->local;
+    int_lock = &local->internal_lock;
+    priv = this->private;
+
+    initialize_internal_lock_variables(frame, this);
+
+    if (local->fd) {
+        fd_ctx = afr_fd_ctx_get(local->fd, this);
+        if (!fd_ctx) {
+            gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_FD_CTX_GET_FAILED,
+                   "unable to get fd ctx for fd=%p", local->fd);
+
+            local->op_ret = -1;
+            int_lock->lock_op_ret = -1;
+            local->op_errno = EINVAL;
+            int_lock->lock_op_errno = EINVAL;
+
+            afr_unlock_now(frame, this);
+            ret = -1;
+            goto out;
+        }
+    }
+
+    call_count = int_lock->lockee_count * internal_lock_count(frame, this);
+    int_lock->lk_call_count = call_count;
+    int_lock->lk_expected_count = call_count;
+
+    if (!call_count) {
+        gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_INFO_COMMON,
+               "fd not open on any subvolumes. aborting.");
+        afr_unlock_now(frame, this);
+        goto out;
+    }
+
+    /* Send non-blocking lock calls only on up children
+       and where the fd has been opened */
+    for (i = 0; i < int_lock->lockee_count * priv->child_count; i++) {
+        child = i % priv->child_count;
+        lockee_num = i / priv->child_count;
+        if (local->child_up[child]) {
+            afr_internal_lock_wind(frame, afr_nb_internal_lock_cbk,
+                                   (void *)(long)i, child, lockee_num,
+                                   _gf_false, _gf_false);
+            if (!--call_count)
+                break;
         }
-
-        local->source_child = source_child;
-
-        /* the flock can be zero filled as we're querying incrementally
-           the locks held on the fd.
-        */
-        STACK_WIND_COOKIE (frame, afr_get_locks_fd_cbk,
-                           (void *) (long) source_child,
-                           priv->children[source_child],
-                           priv->children[source_child]->fops->lk,
-                           local->fd, F_GETLK_FD, &flock);
-
+    }
 out:
-        return ret;
-}
-
-
-static int
-afr_mark_fd_opened (xlator_t *this, fd_t *fd, int32_t child_index)
-{
-        afr_fd_ctx_t *fdctx = NULL;
-        uint64_t      tmp   = 0;
-        int           ret   = 0;
-
-        ret = fd_ctx_get (fd, this, &tmp);
-        if (ret)
-                goto out;
-
-        fdctx = (afr_fd_ctx_t *) (long) tmp;
-
-        fdctx->opened_on[child_index] = 1;
-
-out:
-        return ret;
+    return ret;
 }
 
 int32_t
-afr_lock_recovery_preopen_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                               int32_t op_ret, int32_t op_errno, fd_t *fd)
-{
-        int32_t child_index = (long )cookie;
-        int ret = 0;
-
-        if (op_ret) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Reopen during lock-recovery failed");
-                goto cleanup;
-        }
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Open succeeded => proceed to recover locks");
-
-        ret = afr_lock_recovery (frame, this);
-        if (ret) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Lock recovery failed");
-                goto cleanup;
-        }
-
-        ret = afr_mark_fd_opened (this, fd, child_index);
-        if (ret) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Marking fd open failed");
-                goto cleanup;
-        }
-
-        return 0;
-
-cleanup:
-        afr_lock_recovery_cleanup (frame, this);
+afr_unlock(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_lock_t *lock = NULL;
+
+    local = frame->local;
+
+    if (!local->transaction.eager_lock_on)
+        goto out;
+    lock = &local->inode_ctx->lock[local->transaction.type];
+    LOCK(&local->inode->lock);
+    {
+        list_del_init(&local->transaction.owner_list);
+        if (list_empty(&lock->owners) && list_empty(&lock->post_op)) {
+            local->transaction.do_eager_unlock = _gf_true;
+            /*TODO: Need to get metadata use on_disk and inherit/uninherit
+             *GF_ASSERT (!local->inode_ctx->on_disk[local->transaction.type]);
+             *GF_ASSERT (!local->inode_ctx->inherited[local->transaction.type]);
+             */
+            GF_ASSERT(lock->release);
+        }
+    }
+    UNLOCK(&local->inode->lock);
+    if (!local->transaction.do_eager_unlock) {
+        local->internal_lock.lock_cbk(frame, this);
         return 0;
-}
-
-static int
-afr_lock_recovery_preopen (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t *priv  = NULL;
-        afr_local_t   *local = NULL;
-        uint64_t       tmp   = 0;
-        afr_fd_ctx_t  *fdctx = NULL;
-        loc_t          loc   = {0,};
-        int32_t        child_index = 0;
-        int            ret   = 0;
-
-        priv  = this->private;
-        local = frame->local;
-
-        GF_ASSERT (local && local->fd);
-
-        ret = fd_ctx_get (local->fd, this, &tmp);
-        fdctx = (afr_fd_ctx_t *) (long) tmp;
-        GF_ASSERT (fdctx);
-
-        child_index = local->lock_recovery_child;
-
-        inode_path (local->fd->inode, NULL, (char **)&loc.path);
-        loc.name   = strrchr (loc.path, '/');
-        loc.inode  = inode_ref (local->fd->inode);
-        loc.parent = inode_parent (local->fd->inode, 0, NULL);
-
-
-        STACK_WIND_COOKIE (frame, afr_lock_recovery_preopen_cbk,
-                           (void *)(long) child_index,
-                           priv->children[child_index],
-                           priv->children[child_index]->fops->open,
-                           &loc, fdctx->flags, local->fd,
-                           fdctx->wbflags);
-
-        return 0;
-}
-
-static int
-is_fd_opened (fd_t *fd, int32_t child_index)
-{
-        afr_fd_ctx_t *fdctx = NULL;
-        uint64_t      tmp = 0;
-        int           ret = 0;
-
-        ret = fd_ctx_get (fd, THIS, &tmp);
-        if (ret)
-                goto out;
-
-        fdctx = (afr_fd_ctx_t *) (long) tmp;
-
-        if (fdctx->opened_on[child_index])
-                ret = 1;
-
-out:
-        return ret;
-}
-
-int
-afr_attempt_lock_recovery (xlator_t *this, int32_t child_index)
-{
-        call_frame_t    *frame     = NULL;
-        afr_private_t   *priv      = NULL;
-        afr_local_t     *local     = NULL;
-        afr_locked_fd_t *locked_fd = NULL;
-       afr_locked_fd_t  *tmp       = NULL;
-        int              ret       = 0;
-        struct list_head locks_list;
-
-
-        priv = this->private;
-
-        if (list_empty (&priv->saved_fds))
-                goto out;
-
-        frame = create_frame (this, this->ctx->pool);
-        if (!frame) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                ret = -1;
-                goto out;
-        }
-
-        local = GF_CALLOC (1, sizeof (*local),
-                           gf_afr_mt_afr_local_t);
-        if (!local) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Out of memory");
-                ret = -1;
-                goto out;
-        }
-
-        AFR_LOCAL_INIT (local, priv);
-        if (!local) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                ret = -1;
-                goto out;
-        }
-
-        frame->local = local;
-
-        INIT_LIST_HEAD (&locks_list);
-
-        pthread_mutex_lock (&priv->mutex);
-        {
-                list_splice_init (&priv->saved_fds, &locks_list);
-        }
-        pthread_mutex_unlock (&priv->mutex);
-
-        list_for_each_entry_safe (locked_fd, tmp,
-                                  &locks_list, list) {
-
-                list_del_init (&locked_fd->list);
-
-                local->fd                  = fd_ref (locked_fd->fd);
-                local->lock_recovery_child = child_index;
-                local->locked_fd           = locked_fd;
-
-                if (!is_fd_opened (locked_fd->fd, child_index)) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "attempting open before lock "
-                                "recovery");
-                        afr_lock_recovery_preopen (frame, this);
-                } else {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "attempting lock recovery "
-                                "without a preopen");
-                        afr_lock_recovery (frame, this);
-                }
-        }
+    }
 
 out:
-        return ret;
+    afr_unlock_now(frame, this);
+    return 0;
 }
diff --git a/xlators/cluster/afr/src/afr-mem-types.h b/xlators/cluster/afr/src/afr-mem-types.h
index b4937efa25a..816065fb57a 100644
--- a/xlators/cluster/afr/src/afr-mem-types.h
+++ b/xlators/cluster/afr/src/afr-mem-types.h
@@ -1,48 +1,38 @@
 /*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
 
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
-
 #ifndef __AFR_MEM_TYPES_H__
 #define __AFR_MEM_TYPES_H__
 
-#include "mem-types.h"
+#include <glusterfs/mem-types.h>
 
 enum gf_afr_mem_types_ {
-        gf_afr_mt_iovec  = gf_common_mt_end + 1,
-        gf_afr_mt_afr_fd_ctx_t,
-        gf_afr_mt_afr_local_t,
-        gf_afr_mt_afr_private_t,
-        gf_afr_mt_int32_t,
-        gf_afr_mt_char,
-        gf_afr_mt_xattr_key,
-        gf_afr_mt_dict_t,
-        gf_afr_mt_xlator_t,
-        gf_afr_mt_iatt,
-        gf_afr_mt_int,
-        gf_afr_mt_afr_node_character,
-        gf_afr_mt_sh_diff_loop_state,
-        gf_afr_mt_uint8_t,
-        gf_afr_mt_loc_t,
-        gf_afr_mt_entry_name,
-        gf_afr_mt_pump_priv,
-        gf_afr_mt_locked_fd,
-        gf_afr_mt_end
+    gf_afr_mt_afr_fd_ctx_t = gf_common_mt_end + 1,
+    gf_afr_mt_afr_private_t,
+    gf_afr_mt_int32_t,
+    gf_afr_mt_char,
+    gf_afr_mt_xattr_key,
+    gf_afr_mt_dict_t,
+    gf_afr_mt_xlator_t,
+    gf_afr_mt_afr_node_character,
+    gf_afr_mt_inode_ctx_t,
+    gf_afr_mt_shd_event_t,
+    gf_afr_mt_reply_t,
+    gf_afr_mt_subvol_healer_t,
+    gf_afr_mt_spbc_timeout_t,
+    gf_afr_mt_spb_status_t,
+    gf_afr_mt_empty_brick_t,
+    gf_afr_mt_child_latency_t,
+    gf_afr_mt_atomic_t,
+    gf_afr_mt_lk_heal_info_t,
+    gf_afr_mt_gf_lock,
+    gf_afr_mt_end
 };
 #endif
-
diff --git a/xlators/cluster/afr/src/afr-messages.h b/xlators/cluster/afr/src/afr-messages.h
new file mode 100644
index 00000000000..e73fd997765
--- /dev/null
+++ b/xlators/cluster/afr/src/afr-messages.h
@@ -0,0 +1,167 @@
+/*
+ Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+ This file is part of GlusterFS.
+
+ This file is licensed to you under your choice of the GNU Lesser
+ General Public License, version 3 or any later version (LGPLv3 or
+ later), or the GNU General Public License, version 2 (GPLv2), in all
+ cases as published by the Free Software Foundation.
+ */
+
+#ifndef _AFR_MESSAGES_H_
+#define _AFR_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(
+    AFR, AFR_MSG_QUORUM_FAIL, AFR_MSG_QUORUM_MET, AFR_MSG_QUORUM_OVERRIDE,
+    AFR_MSG_INVALID_CHILD_UP, AFR_MSG_SUBVOL_UP, AFR_MSG_SUBVOLS_DOWN,
+    AFR_MSG_ENTRY_UNLOCK_FAIL, AFR_MSG_SPLIT_BRAIN, AFR_MSG_OPEN_FAIL,
+    AFR_MSG_UNLOCK_FAIL, AFR_MSG_REPLACE_BRICK_STATUS, AFR_MSG_GFID_NULL,
+    AFR_MSG_FD_CREATE_FAILED, AFR_MSG_DICT_SET_FAILED,
+    AFR_MSG_EXPUNGING_FILE_OR_DIR, AFR_MSG_MIGRATION_IN_PROGRESS,
+    AFR_MSG_CHILD_MISCONFIGURED, AFR_MSG_VOL_MISCONFIGURED,
+    AFR_MSG_INTERNAL_LKS_FAILED, AFR_MSG_INVALID_FD, AFR_MSG_LOCK_INFO,
+    AFR_MSG_LOCK_XLATOR_NOT_LOADED, AFR_MSG_FD_CTX_GET_FAILED,
+    AFR_MSG_INVALID_SUBVOL, AFR_MSG_PUMP_XLATOR_ERROR, AFR_MSG_SELF_HEAL_INFO,
+    AFR_MSG_READ_SUBVOL_ERROR, AFR_MSG_DICT_GET_FAILED, AFR_MSG_INFO_COMMON,
+    AFR_MSG_SPLIT_BRAIN_CHOICE_ERROR, AFR_MSG_LOCAL_CHILD, AFR_MSG_INVALID_DATA,
+    AFR_MSG_INVALID_ARG, AFR_MSG_INDEX_DIR_GET_FAILED, AFR_MSG_FSYNC_FAILED,
+    AFR_MSG_FAVORITE_CHILD, AFR_MSG_SELF_HEAL_FAILED,
+    AFR_MSG_SPLIT_BRAIN_STATUS, AFR_MSG_ADD_BRICK_STATUS, AFR_MSG_NO_CHANGELOG,
+    AFR_MSG_TIMER_CREATE_FAIL, AFR_MSG_SBRAIN_FAV_CHILD_POLICY,
+    AFR_MSG_INODE_CTX_GET_FAILED, AFR_MSG_THIN_ARB,
+    AFR_MSG_THIN_ARB_XATTROP_FAILED, AFR_MSG_THIN_ARB_LOC_POP_FAILED,
+    AFR_MSG_GET_PEND_VAL, AFR_MSG_THIN_ARB_SKIP_SHD, AFR_MSG_UNKNOWN_SET,
+    AFR_MSG_NO_XL_ID, AFR_MSG_SELF_HEAL_INFO_START,
+    AFR_MSG_SELF_HEAL_INFO_FINISH, AFR_MSG_INCRE_COUNT,
+    AFR_MSG_ADD_TO_OUTPUT_FAILED, AFR_MSG_SET_TIME_FAILED,
+    AFR_MSG_GFID_MISMATCH_DETECTED, AFR_MSG_GFID_HEAL_MSG,
+    AFR_MSG_THIN_ARB_LOOKUP_FAILED, AFR_MSG_DICT_CREATE_FAILED,
+    AFR_MSG_NO_MAJORITY_TO_RESOLVE, AFR_MSG_TYPE_MISMATCH,
+    AFR_MSG_SIZE_POLICY_NOT_APPLICABLE, AFR_MSG_NO_CHILD_SELECTED,
+    AFR_MSG_INVALID_CHILD, AFR_MSG_RESOLVE_CONFLICTING_DATA,
+    SERROR_GETTING_SRC_BRICK, SNO_DIFF_IN_MTIME, SNO_BIGGER_FILE,
+    SALL_BRICKS_UP_TO_RESOLVE, AFR_MSG_UNLOCK_FAILED, AFR_MSG_POST_OP_FAILED,
+    AFR_MSG_TA_FRAME_CREATE_FAILED, AFR_MSG_SET_KEY_XATTROP_FAILED,
+    AFR_MSG_BLOCKING_ENTRYLKS_FAILED, AFR_MSG_FOP_FAILED,
+    AFR_MSG_CLEAN_UP_FAILED, AFR_MSG_UNABLE_TO_FETCH, AFR_MSG_XATTR_SET_FAILED,
+    AFR_MSG_SPLIT_BRAIN_REPLICA, AFR_MSG_INODE_CTX_FAILED,
+    AFR_MSG_LOOKUP_FAILED, AFR_MSG_ALL_SUBVOLS_DOWN,
+    AFR_MSG_RELEASE_LOCK_FAILED, AFR_MSG_CLEAR_TIME_SPLIT_BRAIN,
+    AFR_MSG_READ_FAILED, AFR_MSG_LAUNCH_FAILED, AFR_MSG_READ_SUBVOL_NOT_UP,
+    AFR_MSG_LK_HEAL_DOM, AFR_MSG_NEW_BRICK, AFR_MSG_SPLIT_BRAIN_SET_FAILED,
+    AFR_MSG_SPLIT_BRAIN_DETERMINE_FAILED, AFR_MSG_HEALER_SPAWN_FAILED,
+    AFR_MSG_ADD_CRAWL_EVENT_FAILED, AFR_MSG_NULL_DEREF, AFR_MSG_SET_PEND_XATTR,
+    AFR_MSG_INTERNAL_ATTR);
+
+#define AFR_MSG_DICT_GET_FAILED_STR "Dict get failed"
+#define AFR_MSG_DICT_SET_FAILED_STR "Dict set failed"
+#define AFR_MSG_HEALER_SPAWN_FAILED_STR "Healer spawn failed"
+#define AFR_MSG_ADD_CRAWL_EVENT_FAILED_STR "Adding crawl event failed"
+#define AFR_MSG_INVALID_ARG_STR "Invalid argument"
+#define AFR_MSG_INDEX_DIR_GET_FAILED_STR "unable to get index-dir on "
+#define AFR_MSG_THIN_ARB_LOOKUP_FAILED_STR "Failed lookup on file"
+#define AFR_MSG_DICT_CREATE_FAILED_STR "Failed to create dict."
+#define AFR_MSG_THIN_ARB_XATTROP_FAILED_STR "Xattrop failed."
+#define AFR_MSG_THIN_ARB_LOC_POP_FAILED_STR                                    \
+    "Failed to populate loc for thin-arbiter"
+#define AFR_MSG_GET_PEND_VAL_STR "Error getting value of pending"
+#define AFR_MSG_THIN_ARB_SKIP_SHD_STR "I am not the god shd. skipping."
+#define AFR_MSG_UNKNOWN_SET_STR "Unknown set"
+#define AFR_MSG_NO_XL_ID_STR "xl does not have id"
+#define AFR_MSG_SELF_HEAL_INFO_START_STR "starting full sweep on"
+#define AFR_MSG_SELF_HEAL_INFO_FINISH_STR "finished full sweep on"
+#define AFR_MSG_INCRE_COUNT_STR "Could not increment the counter."
+#define AFR_MSG_ADD_TO_OUTPUT_FAILED_STR "Could not add to output"
+#define AFR_MSG_SET_TIME_FAILED_STR "Could not set time"
+#define AFR_MSG_GFID_HEAL_MSG_STR "Error setting gfid-heal-msg dict"
+#define AFR_MSG_NO_MAJORITY_TO_RESOLVE_STR                                     \
+    "No majority to resolve gfid split brain"
+#define AFR_MSG_GFID_MISMATCH_DETECTED_STR "Gfid mismatch dectected"
+#define AFR_MSG_SELF_HEAL_INFO_STR "performing selfheal"
+#define AFR_MSG_TYPE_MISMATCH_STR "TYPE mismatch"
+#define AFR_MSG_SIZE_POLICY_NOT_APPLICABLE_STR                                 \
+    "Size policy is not applicable to directories."
+#define AFR_MSG_NO_CHILD_SELECTED_STR                                          \
+    "No child selected by favorite-child policy"
+#define AFR_MSG_INVALID_CHILD_STR "Invalid child"
+#define AFR_MSG_RESOLVE_CONFLICTING_DATA_STR                                   \
+    "selected as authentic to resolve conflicting data"
+#define SERROR_GETTING_SRC_BRICK_STR "Error getting the source brick"
+#define SNO_DIFF_IN_MTIME_STR "No difference in mtime"
+#define SNO_BIGGER_FILE_STR "No bigger file"
+#define SALL_BRICKS_UP_TO_RESOLVE_STR                                          \
+    "All the bricks should be up to resolve the gfid split brain"
+#define AFR_MSG_UNLOCK_FAILED_STR "Failed to unlock"
+#define AFR_MSG_POST_OP_FAILED_STR "Post-op on thin-arbiter failed"
+#define AFR_MSG_TA_FRAME_CREATE_FAILED_STR "Failed to create ta_frame"
+#define AFR_MSG_SET_KEY_XATTROP_FAILED_STR "Could not set key during xattrop"
+#define AFR_MSG_BLOCKING_ENTRYLKS_FAILED_STR "Blocking entrylks failed"
+#define AFR_MSG_FSYNC_FAILED_STR "fsync failed"
+#define AFR_MSG_QUORUM_FAIL_STR "quorum is not met"
+#define AFR_MSG_FOP_FAILED_STR "Failing Fop"
+#define AFR_MSG_INVALID_SUBVOL_STR "not a subvolume"
+#define AFR_MSG_VOL_MISCONFIGURED_STR "Volume is dangling"
+#define AFR_MSG_CHILD_MISCONFIGURED_STR                                        \
+    "replicate translator needs more than one subvolume defined"
+#define AFR_MSG_CLEAN_UP_FAILED_STR "Failed to clean up healer threads"
+#define AFR_MSG_QUORUM_OVERRIDE_STR "overriding quorum-count"
+#define AFR_MSG_UNABLE_TO_FETCH_STR                                            \
+    "Unable to fetch afr-pending-xattr option from volfile. Falling back to "  \
+    "using client translator names"
+#define AFR_MSG_NULL_DEREF_STR "possible NULL deref"
+#define AFR_MSG_XATTR_SET_FAILED_STR "Cannot set xattr cookie key"
+#define AFR_MSG_SPLIT_BRAIN_STATUS_STR "Failed to create synctask"
+#define AFR_MSG_SUBVOLS_DOWN_STR "All subvolumes are not up"
+#define AFR_MSG_SPLIT_BRAIN_CHOICE_ERROR_STR                                   \
+    "Failed to cancel split-brain choice"
+#define AFR_MSG_SPLIT_BRAIN_REPLICA_STR                                        \
+    "Cannot set replica. File is not in data/metadata split-brain"
+#define AFR_MSG_INODE_CTX_FAILED_STR "Failed to get inode_ctx"
+#define AFR_MSG_READ_SUBVOL_ERROR_STR "no read subvols"
+#define AFR_MSG_LOCAL_CHILD_STR "selecting local read-child"
+#define AFR_MSG_LOOKUP_FAILED_STR "Failed to lookup/create thin-arbiter id file"
+#define AFR_MSG_TIMER_CREATE_FAIL_STR                                          \
+    "Cannot create timer for delayed initialization"
+#define AFR_MSG_SUBVOL_UP_STR "Subvolume came back up; going online"
+#define AFR_MSG_ALL_SUBVOLS_DOWN_STR                                           \
+    "All subvolumes are down. Going offline until atleast one of them is up"
+#define AFR_MSG_RELEASE_LOCK_FAILED_STR "Failed to release lock"
+#define AFR_MSG_INVALID_CHILD_UP_STR "Received child_up from invalid subvolume"
+#define AFR_MSG_QUORUM_MET_STR "Client-quorum is met"
+#define AFR_MSG_EXPUNGING_FILE_OR_DIR_STR "expunging file or dir"
+#define AFR_MSG_SELF_HEAL_FAILED_STR "Invalid"
+#define AFR_MSG_SPLIT_BRAIN_STR "Skipping conservative mergeon the file"
+#define AFR_MSG_CLEAR_TIME_SPLIT_BRAIN_STR "clear time split brain"
+#define AFR_MSG_READ_FAILED_STR "Failing read since good brick is down"
+#define AFR_MSG_LAUNCH_FAILED_STR "Failed to launch synctask"
+#define AFR_MSG_READ_SUBVOL_NOT_UP_STR                                         \
+    "read subvolume in this generation is not up"
+#define AFR_MSG_INTERNAL_LKS_FAILED_STR                                        \
+    "Unable to work with lk-owner while attempting fop"
+#define AFR_MSG_LOCK_XLATOR_NOT_LOADED_STR                                     \
+    "subvolume does not support locking. please load features/locks xlator "   \
+    "on server."
+#define AFR_MSG_FD_CTX_GET_FAILED_STR "unable to get fd ctx"
+#define AFR_MSG_INFO_COMMON_STR "fd not open on any subvolumes, aborting."
+#define AFR_MSG_REPLACE_BRICK_STATUS_STR "Couldn't acquire lock on any child."
+#define AFR_MSG_NEW_BRICK_STR "New brick"
+#define AFR_MSG_SPLIT_BRAIN_SET_FAILED_STR                                     \
+    "Failed to set split-brain choice to -1"
+#define AFR_MSG_SPLIT_BRAIN_DETERMINE_FAILED_STR                               \
+    "Failed to determine split-brain. Aborting split-brain-choice set"
+#define AFR_MSG_OPEN_FAIL_STR "Failed to open subvolume"
+#define AFR_MSG_SET_PEND_XATTR_STR "Set of pending xattr"
+#define AFR_MSG_INTERNAL_ATTR_STR "is an internal extended attribute"
+#endif /* !_AFR_MESSAGES_H_ */
diff --git a/xlators/cluster/afr/src/afr-open.c b/xlators/cluster/afr/src/afr-open.c
index 5569dd1e228..64856042b65 100644
--- a/xlators/cluster/afr/src/afr-open.c
+++ b/xlators/cluster/afr/src/afr-open.c
@@ -1,473 +1,353 @@
 /*
-   Copyright (c) 2007-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
-#include <libgen.h>
 #include <unistd.h>
-#include <fnmatch.h>
 #include <sys/time.h>
 #include <stdlib.h>
 #include <signal.h>
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
+#include <glusterfs/glusterfs.h>
 #include "afr.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-#include "byte-order.h"
-#include "statedump.h"
-
-#include "fd.h"
-
-#include "afr-inode-read.h"
-#include "afr-inode-write.h"
-#include "afr-dir-read.h"
-#include "afr-dir-write.h"
-#include "afr-transaction.h"
-#include "afr-self-heal.h"
-#include "afr-self-heal-common.h"
+#include <glusterfs/dict.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/byte-order.h>
+#include <glusterfs/statedump.h>
 
+#include "afr-transaction.h"
 
-int
-afr_open_ftruncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this, 
-			int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                        struct iatt *postbuf)
+gf_boolean_t
+afr_is_fd_fixable(fd_t *fd)
 {
-	afr_local_t * local = frame->local;
-
-	AFR_STACK_UNWIND (open, frame, local->op_ret, local->op_errno,
-			  local->fd);
-	return 0;
+    if (!fd || !fd->inode)
+        return _gf_false;
+    else if (fd_is_anonymous(fd))
+        return _gf_false;
+    else if (gf_uuid_is_null(fd->inode->gfid))
+        return _gf_false;
+
+    return _gf_true;
 }
 
-
 int
-afr_open_cbk (call_frame_t *frame, void *cookie,
-	      xlator_t *this, int32_t op_ret, int32_t op_errno,
-	      fd_t *fd)
+afr_open_ftruncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                       struct iatt *postbuf, dict_t *xdata)
 {
-	afr_local_t *  local = NULL;
-
-        int child_index = (long) cookie;
-
-        uint64_t      ctx;
-        afr_fd_ctx_t *fd_ctx;
-
-        int ret = 0;
-
-	int call_count = -1;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-		}
-
-		if (op_ret >= 0) {
-			local->op_ret = op_ret;
-                        local->success_count++;
-
-                        ret = afr_fd_ctx_set (this, fd);
+    afr_local_t *local = frame->local;
 
-                        if (ret < 0) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "could not set fd ctx for fd=%p",
-                                        fd);
-
-                                local->op_ret   = -1;
-                                local->op_errno = -ret;
-                                goto unlock;
-                        }
-
-                        ret = fd_ctx_get (fd, this, &ctx);
-
-                        if (ret < 0) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "could not get fd ctx for fd=%p", fd);
-                                local->op_ret   = -1;
-                                local->op_errno = -ret;
-                                goto unlock;
-                        }
-
-                        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-                        fd_ctx->opened_on[child_index] = 1;
-                        fd_ctx->flags                  = local->cont.open.flags;
-                        fd_ctx->wbflags                = local->cont.open.wbflags;
-		}
-	}
-unlock:
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		if ((local->cont.open.flags & O_TRUNC)
-		    && (local->op_ret >= 0)) {
-			STACK_WIND (frame, afr_open_ftruncate_cbk,
-				    this, this->fops->ftruncate,
-				    fd, 0);
-		} else {
-                        AFR_STACK_UNWIND (open, frame, local->op_ret,
-                                          local->op_errno, local->fd);
-		}
-	}
-
-	return 0;
+    AFR_STACK_UNWIND(open, frame, local->op_ret, local->op_errno,
+                     local->cont.open.fd, xdata);
+    return 0;
 }
 
-
 int
-afr_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
-          fd_t *fd, int32_t wbflags)
+afr_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, fd_t *fd, dict_t *xdata)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t *   local = NULL;
-
-	int     i = 0;
-	int   ret = -1;
-
-	int32_t call_count = 0;	
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
-	int32_t wind_flags = flags & (~O_TRUNC);
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-	VALIDATE_OR_GOTO (loc, out);
-
-	priv = this->private;
-
-        if (afr_is_split_brain (this, loc->inode)) {
-		/* self-heal failed */
-		op_errno = EIO;
-		goto out;
-	}
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-	frame->local = local;
-	call_count   = local->call_count;
-
-        loc_copy (&local->loc, loc);
-
-	local->cont.open.flags   = flags;
-        local->cont.open.wbflags = wbflags;
-
-	local->fd = fd_ref (fd);
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_open_cbk, (void *) (long) i,
-					   priv->children[i],
-					   priv->children[i]->fops->open,
-					   loc, wind_flags, fd, wbflags);
-
-			if (!--call_count)
-				break;
-		}
-	}
-
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (open, frame, op_ret, op_errno, fd);
-	}
-
-	return 0;
-}
-
-
-int
-afr_openfd_sh_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, fd_t *fd)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_private_t       *priv     = NULL;
-
-        int ret = 0;
-
-        uint64_t      ctx;
-        afr_fd_ctx_t *fd_ctx;
-
-        int call_count  = 0;
-        int child_index = (long) cookie;
-
-        priv     = this->private;
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        LOCK (&frame->lock);
-        {
-                if (op_ret >= 0) {
-                        ret = fd_ctx_get (fd, this, &ctx);
-
-                        if (ret < 0) {
-                                goto out;
-                        }
-
-                        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-                        fd_ctx->opened_on[child_index] = 1;
-
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "fd for %s opened successfully on subvolume %s",
-                                local->loc.path, priv->children[child_index]->name);
-                }
-        }
-out:
-        UNLOCK (&frame->lock);
-
-        call_count = afr_frame_return (frame);
-
-        if (call_count == 0) {
-                int_lock->lock_cbk = local->transaction.done;
-                local->transaction.resume (frame, this);
+    afr_local_t *local = NULL;
+    int call_count = -1;
+    int child_index = (long)cookie;
+    afr_fd_ctx_t *fd_ctx = NULL;
+
+    local = frame->local;
+    fd_ctx = local->fd_ctx;
+
+    local->replies[child_index].valid = 1;
+    local->replies[child_index].op_ret = op_ret;
+    local->replies[child_index].op_errno = op_errno;
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+            fd_ctx->opened_on[child_index] = AFR_FD_NOT_OPENED;
+        } else {
+            local->op_ret = op_ret;
+            fd_ctx->opened_on[child_index] = AFR_FD_OPENED;
+            if (!local->xdata_rsp && xdata)
+                local->xdata_rsp = dict_ref(xdata);
         }
-
-        return 0;
-}
-
-
-static int
-__unopened_count (int child_count, unsigned int *opened_on, unsigned char *child_up)
-{
-        int i;
-        int count = 0;
-
-        for (i = 0; i < child_count; i++) {
-                if (!opened_on[i] && child_up[i])
-                        count++;
+        call_count = --local->call_count;
+    }
+    UNLOCK(&frame->lock);
+
+    if (call_count == 0) {
+        afr_handle_replies_quorum(frame, this);
+        if (local->op_ret == -1) {
+            AFR_STACK_UNWIND(open, frame, local->op_ret, local->op_errno, NULL,
+                             NULL);
+        } else if (fd_ctx->flags & O_TRUNC) {
+            STACK_WIND(frame, afr_open_ftruncate_cbk, this,
+                       this->fops->ftruncate, fd, 0, NULL);
+        } else {
+            AFR_STACK_UNWIND(open, frame, local->op_ret, local->op_errno,
+                             local->cont.open.fd, local->xdata_rsp);
         }
+    }
 
-        return count;
+    return 0;
 }
 
-
 int
-afr_openfd_sh_unwind (call_frame_t *frame, xlator_t *this)
+afr_open_continue(call_frame_t *frame, xlator_t *this, int err)
 {
-        afr_local_t *local  = NULL;
-        afr_private_t *priv = NULL;
-
-        uint64_t      ctx;
-        afr_fd_ctx_t *fd_ctx;
-
-        int abandon = 0;
-        int ret = 0;
-        int i;
-        int call_count = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int call_count = 0;
+    int i = 0;
 
-        priv  = this->private;
-        local = frame->local;
+    local = frame->local;
+    priv = this->private;
 
-        /*
-         * Some subvolumes might have come up on which we never
-         * opened this fd in the first place. Re-open fd's on those
-         * subvolumes now.
-         */
-
-        ret = fd_ctx_get (local->fd, this, &ctx);
-
-        if (ret < 0) {
-                abandon = 1;
-                goto out;
-        }
-
-        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-        LOCK (&local->fd->lock);
-        {
-                call_count = __unopened_count (priv->child_count,
-                                               fd_ctx->opened_on,
-                                               local->child_up);
-                for (i = 0; i < priv->child_count; i++) {
-                        fd_ctx->pre_op_done[i] = 0;
-                        fd_ctx->pre_op_piggyback[i] = 0;
-                }
-        }
-        UNLOCK (&local->fd->lock);
-
-        if (call_count == 0) {
-                abandon = 1;
-                goto out;
-        }
-
-        local->call_count = call_count;
+    if (err) {
+        AFR_STACK_UNWIND(open, frame, -1, err, NULL, NULL);
+    } else {
+        local->call_count = AFR_COUNT(local->child_up, priv->child_count);
+        call_count = local->call_count;
 
         for (i = 0; i < priv->child_count; i++) {
-                if (!fd_ctx->opened_on[i] && local->child_up[i]) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "opening fd for %s on subvolume %s",
-                                local->loc.path, priv->children[i]->name);
-
-                        STACK_WIND_COOKIE (frame, afr_openfd_sh_open_cbk,
-                                           (void *)(long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->open,
-                                           &local->loc, fd_ctx->flags, local->fd,
-                                           fd_ctx->wbflags);
-
-                        if (!--call_count)
-                                break;
-                }
+            if (local->child_up[i]) {
+                STACK_WIND_COOKIE(frame, afr_open_cbk, (void *)(long)i,
+                                  priv->children[i],
+                                  priv->children[i]->fops->open, &local->loc,
+                                  (local->cont.open.flags & ~O_TRUNC),
+                                  local->cont.open.fd, local->xdata_req);
+                if (!--call_count)
+                    break;
+            }
         }
-
-out:
-        if (abandon)
-                local->transaction.resume (frame, this);
-
-        return 0;
+    }
+    return 0;
 }
 
-
 int
-afr_openfd_sh (call_frame_t *frame, xlator_t *this)
+afr_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+         fd_t *fd, dict_t *xdata)
 {
-        afr_local_t     *local  = NULL;
-        afr_self_heal_t *sh = NULL;
-        char            sh_type_str[256] = {0,};
-
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        inode_path (local->fd->inode, NULL, (char **)&local->loc.path);
-        local->loc.name   = strrchr (local->loc.path, '/');
-        local->loc.inode  = inode_ref (local->fd->inode);
-        local->loc.parent = inode_parent (local->fd->inode, 0, NULL);
-
-        /* forcibly trigger missing-entries self-heal */
-
-        local->success_count    = 1;
-        local->enoent_count     = 1;
-
-        sh->data_lock_held      = _gf_true;
-        sh->need_data_self_heal = _gf_true;
-        sh->type                = local->fd->inode->ia_type;
-        sh->background          = _gf_false;
-        sh->unwind              = afr_openfd_sh_unwind;
-
-        afr_self_heal_type_str_get(&local->self_heal,
-                                   sh_type_str,
-                                   sizeof(sh_type_str));
-        gf_log (this->name, GF_LOG_NORMAL, "%s self-heal triggered. "
-                "path: %s, reason: Replicate up down flush, data lock is held",
-                sh_type_str, local->loc.path);
-
-        afr_self_heal (frame, this);
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int spb_subvol = 0;
+    int event_generation = 0;
+    int ret = 0;
+    int32_t op_errno = 0;
+    afr_fd_ctx_t *fd_ctx = NULL;
+
+    // We can't let truncation to happen outside transaction.
+
+    priv = this->private;
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    local->op = GF_FOP_OPEN;
+    fd_ctx = afr_fd_ctx_get(fd, this);
+    if (!fd_ctx) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    if (priv->quorum_count && !afr_has_quorum(local->child_up, this, NULL)) {
+        op_errno = afr_quorum_errno(priv);
+        goto out;
+    }
+
+    if (!afr_is_consistent_io_possible(local, priv, &op_errno))
+        goto out;
+
+    local->inode = inode_ref(loc->inode);
+    loc_copy(&local->loc, loc);
+    local->fd_ctx = fd_ctx;
+    fd_ctx->flags = flags;
+    if (xdata)
+        local->xdata_req = dict_ref(xdata);
+
+    local->cont.open.flags = flags;
+    local->cont.open.fd = fd_ref(fd);
+
+    ret = afr_inode_get_readable(frame, local->inode, this, NULL,
+                                 &event_generation, AFR_DATA_TRANSACTION);
+    if ((ret < 0) &&
+        (afr_split_brain_read_subvol_get(local->inode, this, NULL,
+                                         &spb_subvol) == 0) &&
+        spb_subvol < 0) {
+        afr_inode_refresh(frame, this, local->inode, local->inode->gfid,
+                          afr_open_continue);
+    } else {
+        afr_open_continue(frame, this, 0);
+    }
+
+    return 0;
+out:
+    AFR_STACK_UNWIND(open, frame, -1, op_errno, fd, NULL);
 
-        return 0;
+    return 0;
 }
 
-
 int
-afr_openfd_flush_done (call_frame_t *frame, xlator_t *this)
+afr_openfd_fix_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, fd_t *fd,
+                        dict_t *xdata)
 {
-        afr_private_t *priv = NULL;
-	afr_local_t *local  = NULL;
-
-        uint64_t       ctx;
-        afr_fd_ctx_t * fd_ctx = NULL;
-
-        int _ret = -1;
-
-        priv  = this->private;
-	local = frame->local;
-
-        LOCK (&local->fd->lock);
-        {
-                _ret = __fd_ctx_get (local->fd, this, &ctx);
-
-                if (_ret < 0) {
-                        goto out;
-                }
-
-                fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
-                fd_ctx->down_count = priv->down_count;
-                fd_ctx->up_count   = priv->up_count;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    afr_fd_ctx_t *fd_ctx = NULL;
+    int call_count = 0;
+    int child_index = (long)cookie;
+
+    priv = this->private;
+    local = frame->local;
+
+    if (op_ret >= 0) {
+        gf_msg_debug(this->name, 0,
+                     "fd for %s opened "
+                     "successfully on subvolume %s",
+                     local->loc.path, priv->children[child_index]->name);
+    } else {
+        gf_smsg(this->name, fop_log_level(GF_FOP_OPEN, op_errno), op_errno,
+                AFR_MSG_OPEN_FAIL, "path=%s", local->loc.path, "subvolume=%s",
+                priv->children[child_index]->name, NULL);
+    }
+
+    fd_ctx = local->fd_ctx;
+
+    LOCK(&local->fd->lock);
+    {
+        if (op_ret >= 0) {
+            fd_ctx->opened_on[child_index] = AFR_FD_OPENED;
+        } else {
+            fd_ctx->opened_on[child_index] = AFR_FD_NOT_OPENED;
         }
-out:
-        UNLOCK (&local->fd->lock);
+    }
+    UNLOCK(&local->fd->lock);
 
-        afr_local_transaction_cleanup (local, this);
+    call_count = afr_frame_return(frame);
+    if (call_count == 0)
+        AFR_STACK_DESTROY(frame);
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "The up/down flush is over");
-
-        fd_unref (local->fd);
-        local->openfd_flush_cbk (frame, this);
-
-	return 0;
+    return 0;
 }
 
-
-int
-afr_openfd_flush (call_frame_t *frame, xlator_t *this, fd_t *fd)
+static int
+afr_fd_ctx_need_open(fd_t *fd, xlator_t *this, unsigned char *need_open)
 {
-	afr_local_t   * local = NULL;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-        local = frame->local;
+    afr_fd_ctx_t *fd_ctx = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int count = 0;
 
-        local->op = GF_FOP_FLUSH;
+    priv = this->private;
 
-        local->fd = fd_ref (fd);
+    fd_ctx = afr_fd_ctx_get(fd, this);
+    if (!fd_ctx)
+        return 0;
 
-        local->transaction.fop          = afr_openfd_sh;
-        local->transaction.done         = afr_openfd_flush_done;
+    LOCK(&fd->lock);
+    {
+        for (i = 0; i < priv->child_count; i++) {
+            if (fd_ctx->opened_on[i] == AFR_FD_NOT_OPENED &&
+                priv->child_up[i]) {
+                fd_ctx->opened_on[i] = AFR_FD_OPENING;
+                need_open[i] = 1;
+                count++;
+            } else {
+                need_open[i] = 0;
+            }
+        }
+    }
+    UNLOCK(&fd->lock);
 
-        local->transaction.start  = 0;
-        local->transaction.len    = 0;
+    return count;
+}
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "doing up/down flush on fd=%p",
-                fd);
+void
+afr_fix_open(fd_t *fd, xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    int i = 0;
+    call_frame_t *frame = NULL;
+    afr_local_t *local = NULL;
+    int ret = -1;
+    int32_t op_errno = 0;
+    afr_fd_ctx_t *fd_ctx = NULL;
+    unsigned char *need_open = NULL;
+    int call_count = 0;
+
+    priv = this->private;
+
+    if (!afr_is_fd_fixable(fd))
+        goto out;
+
+    fd_ctx = afr_fd_ctx_get(fd, this);
+    if (!fd_ctx)
+        goto out;
+
+    need_open = alloca0(priv->child_count);
+
+    call_count = afr_fd_ctx_need_open(fd, this, need_open);
+    if (!call_count)
+        goto out;
+
+    frame = create_frame(this, this->ctx->pool);
+    if (!frame)
+        goto out;
+
+    local = AFR_FRAME_INIT(frame, op_errno);
+    if (!local)
+        goto out;
+
+    local->loc.inode = inode_ref(fd->inode);
+    ret = loc_path(&local->loc, NULL);
+    if (ret < 0)
+        goto out;
+
+    local->fd = fd_ref(fd);
+    local->fd_ctx = fd_ctx;
+
+    local->call_count = call_count;
+
+    gf_msg_debug(this->name, 0, "need open count: %d", call_count);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!need_open[i])
+            continue;
+
+        if (IA_IFDIR == fd->inode->ia_type) {
+            gf_msg_debug(this->name, 0, "opening fd for dir %s on subvolume %s",
+                         local->loc.path, priv->children[i]->name);
+
+            STACK_WIND_COOKIE(frame, afr_openfd_fix_open_cbk, (void *)(long)i,
+                              priv->children[i],
+                              priv->children[i]->fops->opendir, &local->loc,
+                              local->fd, NULL);
+        } else {
+            gf_msg_debug(this->name, 0,
+                         "opening fd for file %s on subvolume %s",
+                         local->loc.path, priv->children[i]->name);
+
+            STACK_WIND_COOKIE(frame, afr_openfd_fix_open_cbk, (void *)(long)i,
+                              priv->children[i], priv->children[i]->fops->open,
+                              &local->loc, fd_ctx->flags & (~O_TRUNC),
+                              local->fd, NULL);
+        }
 
-        afr_transaction (frame, this, AFR_DATA_TRANSACTION);
+        if (!--call_count)
+            break;
+    }
 
+    return;
 out:
-	return 0;
+    if (frame)
+        AFR_STACK_DESTROY(frame);
 }
-
diff --git a/xlators/cluster/afr/src/afr-read-txn.c b/xlators/cluster/afr/src/afr-read-txn.c
new file mode 100644
index 00000000000..6fc2c75145c
--- /dev/null
+++ b/xlators/cluster/afr/src/afr-read-txn.c
@@ -0,0 +1,494 @@
+/*
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "afr.h"
+#include "afr-transaction.h"
+#include "afr-messages.h"
+
+void
+afr_pending_read_increment(afr_private_t *priv, int child_index)
+{
+    if (child_index < 0 || child_index > priv->child_count)
+        return;
+
+    GF_ATOMIC_INC(priv->pending_reads[child_index]);
+}
+
+void
+afr_pending_read_decrement(afr_private_t *priv, int child_index)
+{
+    if (child_index < 0 || child_index > priv->child_count)
+        return;
+
+    GF_ATOMIC_DEC(priv->pending_reads[child_index]);
+}
+
+void
+afr_read_txn_wind(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    afr_pending_read_decrement(priv, local->read_subvol);
+    local->read_subvol = subvol;
+    afr_pending_read_increment(priv, subvol);
+    local->readfn(frame, this, subvol);
+}
+
+int
+afr_read_txn_next_subvol(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int subvol = -1;
+
+    local = frame->local;
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->readable[i]) {
+            /* don't even bother trying here.
+               just mark as attempted and move on. */
+            local->read_attempted[i] = 1;
+            continue;
+        }
+
+        if (!local->read_attempted[i]) {
+            subvol = i;
+            break;
+        }
+    }
+
+    /* If no more subvols were available for reading, we leave
+       @subvol as -1, which is an indication we have run out of
+       readable subvols. */
+    if (subvol != -1)
+        local->read_attempted[subvol] = 1;
+    afr_read_txn_wind(frame, this, subvol);
+
+    return 0;
+}
+
+static int
+afr_ta_read_txn_done(int ret, call_frame_t *ta_frame, void *opaque)
+{
+    STACK_DESTROY(ta_frame->root);
+    return 0;
+}
+
+static int
+afr_ta_read_txn(void *opaque)
+{
+    call_frame_t *frame = NULL;
+    xlator_t *this = NULL;
+    int read_subvol = -1;
+    int query_child = AFR_CHILD_UNKNOWN;
+    int possible_bad_child = AFR_CHILD_UNKNOWN;
+    int ret = 0;
+    int op_errno = ENOMEM;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    struct gf_flock flock = {
+        0,
+    };
+    dict_t *xdata_req = NULL;
+    dict_t *xdata_rsp = NULL;
+    int **pending = NULL;
+    loc_t loc = {
+        0,
+    };
+
+    frame = (call_frame_t *)opaque;
+    this = frame->this;
+    local = frame->local;
+    priv = this->private;
+    query_child = local->read_txn_query_child;
+
+    if (query_child == AFR_CHILD_ZERO) {
+        possible_bad_child = AFR_CHILD_ONE;
+    } else if (query_child == AFR_CHILD_ONE) {
+        possible_bad_child = AFR_CHILD_ZERO;
+    } else {
+        /*read_txn_query_child is AFR_CHILD_UNKNOWN*/
+        goto out;
+    }
+
+    /* Ask the query_child to see if it blames the possibly bad one. */
+    xdata_req = dict_new();
+    if (!xdata_req)
+        goto out;
+
+    pending = afr_matrix_create(priv->child_count, AFR_NUM_CHANGE_LOGS);
+    if (!pending)
+        goto out;
+
+    ret = afr_set_pending_dict(priv, xdata_req, pending);
+    if (ret < 0)
+        goto out;
+
+    if (local->fd) {
+        ret = syncop_fxattrop(priv->children[query_child], local->fd,
+                              GF_XATTROP_ADD_ARRAY, xdata_req, NULL, &xdata_rsp,
+                              NULL);
+    } else {
+        ret = syncop_xattrop(priv->children[query_child], &local->loc,
+                             GF_XATTROP_ADD_ARRAY, xdata_req, NULL, &xdata_rsp,
+                             NULL);
+    }
+    if (ret || !xdata_rsp) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed xattrop for gfid %s on %s",
+               uuid_utoa(local->inode->gfid),
+               priv->children[query_child]->name);
+        op_errno = -ret;
+        goto out;
+    }
+
+    if (afr_ta_dict_contains_pending_xattr(xdata_rsp, priv,
+                                           possible_bad_child)) {
+        read_subvol = query_child;
+        goto out;
+    }
+    dict_unref(xdata_rsp);
+    xdata_rsp = NULL;
+
+    /* It doesn't. So query thin-arbiter to see if it blames any data brick. */
+    ret = afr_fill_ta_loc(this, &loc, _gf_true);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to populate thin-arbiter loc for: %s.", loc.name);
+        goto out;
+    }
+    flock.l_type = F_WRLCK; /*start and length are already zero. */
+    ret = syncop_inodelk(priv->children[THIN_ARBITER_BRICK_INDEX],
+                         AFR_TA_DOM_MODIFY, &loc, F_SETLKW, &flock, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "gfid:%s: Failed to get AFR_TA_DOM_MODIFY lock on %s.",
+               uuid_utoa(local->inode->gfid),
+               priv->pending_key[THIN_ARBITER_BRICK_INDEX]);
+        op_errno = -ret;
+        goto out;
+    }
+
+    ret = syncop_xattrop(priv->children[THIN_ARBITER_BRICK_INDEX], &loc,
+                         GF_XATTROP_ADD_ARRAY, xdata_req, NULL, &xdata_rsp,
+                         NULL);
+    if (ret || !xdata_rsp) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "gfid:%s: Failed xattrop on %s.", uuid_utoa(local->inode->gfid),
+               priv->pending_key[THIN_ARBITER_BRICK_INDEX]);
+        op_errno = -ret;
+        goto unlock;
+    }
+
+    if (!afr_ta_dict_contains_pending_xattr(xdata_rsp, priv, query_child)) {
+        read_subvol = query_child;
+    } else {
+        gf_msg(this->name, GF_LOG_ERROR, EIO, AFR_MSG_THIN_ARB,
+               "Failing read for gfid %s since good brick %s is down",
+               uuid_utoa(local->inode->gfid),
+               priv->children[possible_bad_child]->name);
+        op_errno = EIO;
+    }
+
+unlock:
+    flock.l_type = F_UNLCK;
+    ret = syncop_inodelk(priv->children[THIN_ARBITER_BRICK_INDEX],
+                         AFR_TA_DOM_MODIFY, &loc, F_SETLK, &flock, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "gfid:%s: Failed to unlock AFR_TA_DOM_MODIFY lock on "
+               "%s.",
+               uuid_utoa(local->inode->gfid),
+               priv->pending_key[THIN_ARBITER_BRICK_INDEX]);
+    }
+out:
+    if (xdata_req)
+        dict_unref(xdata_req);
+    if (xdata_rsp)
+        dict_unref(xdata_rsp);
+    if (pending)
+        afr_matrix_cleanup(pending, priv->child_count);
+    loc_wipe(&loc);
+
+    if (read_subvol == -1) {
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+    }
+    afr_read_txn_wind(frame, this, read_subvol);
+    return ret;
+}
+
+void
+afr_ta_read_txn_synctask(call_frame_t *frame, xlator_t *this)
+{
+    call_frame_t *ta_frame = NULL;
+    afr_local_t *local = NULL;
+    int ret = 0;
+
+    local = frame->local;
+    ta_frame = afr_ta_frame_create(this);
+    if (!ta_frame) {
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_THIN_ARB,
+               "Failed to create ta_frame");
+        goto out;
+    }
+    ret = synctask_new(this->ctx->env, afr_ta_read_txn, afr_ta_read_txn_done,
+                       ta_frame, frame);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_THIN_ARB,
+               "Failed to launch "
+               "afr_ta_read_txn synctask for gfid %s.",
+               uuid_utoa(local->inode->gfid));
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        STACK_DESTROY(ta_frame->root);
+        goto out;
+    }
+    return;
+out:
+    afr_read_txn_wind(frame, this, -1);
+}
+
+int
+afr_read_txn_refresh_done(call_frame_t *frame, xlator_t *this, int err)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int read_subvol = -1;
+    inode_t *inode = NULL;
+    int ret = -1;
+    int spb_subvol = -1;
+
+    local = frame->local;
+    inode = local->inode;
+    priv = this->private;
+
+    if (err) {
+        if (!priv->thin_arbiter_count)
+            goto readfn;
+        if (err != EINVAL)
+            goto readfn;
+        /* We need to query the good bricks and/or thin-arbiter.*/
+        afr_ta_read_txn_synctask(frame, this);
+        return 0;
+    }
+
+    read_subvol = afr_read_subvol_select_by_policy(inode, this, local->readable,
+                                                   NULL);
+    if (read_subvol == -1) {
+        err = EIO;
+        goto readfn;
+    }
+
+    if (local->read_attempted[read_subvol]) {
+        afr_read_txn_next_subvol(frame, this);
+        return 0;
+    }
+
+    local->read_attempted[read_subvol] = 1;
+readfn:
+    if (read_subvol == -1) {
+        ret = afr_split_brain_read_subvol_get(inode, this, frame, &spb_subvol);
+        if ((ret == 0) && spb_subvol >= 0)
+            read_subvol = spb_subvol;
+    }
+
+    if (read_subvol == -1) {
+        AFR_SET_ERROR_AND_CHECK_SPLIT_BRAIN(-1, err);
+    }
+    afr_read_txn_wind(frame, this, read_subvol);
+
+    return 0;
+}
+
+int
+afr_read_txn_continue(call_frame_t *frame, xlator_t *this, int subvol)
+{
+    afr_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (!local->refreshed) {
+        local->refreshed = _gf_true;
+        afr_inode_refresh(frame, this, local->inode, NULL,
+                          afr_read_txn_refresh_done);
+    } else {
+        afr_read_txn_next_subvol(frame, this);
+    }
+
+    return 0;
+}
+
+/* afr_read_txn_wipe:
+
+   clean internal variables in @local in order to make
+   it possible to call afr_read_txn() multiple times from
+   the same frame
+*/
+
+void
+afr_read_txn_wipe(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+
+    local = frame->local;
+    priv = this->private;
+
+    local->readfn = NULL;
+
+    if (local->inode)
+        inode_unref(local->inode);
+
+    for (i = 0; i < priv->child_count; i++) {
+        local->read_attempted[i] = 0;
+        local->readable[i] = 0;
+    }
+}
+
+/*
+  afr_read_txn:
+
+  This is the read transaction function. The way it works:
+
+  - Determine read-subvolume from inode ctx.
+
+  - If read-subvolume's generation was stale, refresh ctx once by
+    calling afr_inode_refresh()
+
+    Else make an attempt to read on read-subvolume.
+
+  - If attempted read on read-subvolume fails, refresh ctx once
+    by calling afr_inode_refresh()
+
+  - After ctx refresh, query read-subvolume freshly and attempt
+    read once.
+
+  - If read fails, try every other readable[] subvolume before
+    finally giving up. readable[] elements are set by afr_inode_refresh()
+    based on dirty and pending flags.
+
+  - If file is in split brain in the backend, generation will be
+    kept 0 by afr_inode_refresh() and readable[] will be set 0 for
+    all elements. Therefore reads always fail.
+*/
+
+int
+afr_read_txn(call_frame_t *frame, xlator_t *this, inode_t *inode,
+             afr_read_txn_wind_t readfn, afr_transaction_type type)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    unsigned char *data = NULL;
+    unsigned char *metadata = NULL;
+    int read_subvol = -1;
+    int event_generation = 0;
+    int ret = -1;
+
+    priv = this->private;
+    local = frame->local;
+    data = alloca0(priv->child_count);
+    metadata = alloca0(priv->child_count);
+
+    afr_read_txn_wipe(frame, this);
+
+    local->readfn = readfn;
+    local->inode = inode_ref(inode);
+    local->is_read_txn = _gf_true;
+    local->transaction.type = type;
+
+    if (priv->quorum_count && !afr_has_quorum(local->child_up, this, NULL)) {
+        local->op_ret = -1;
+        local->op_errno = afr_quorum_errno(priv);
+        goto read;
+    }
+
+    if (!afr_is_consistent_io_possible(local, priv, &local->op_errno)) {
+        local->op_ret = -1;
+        goto read;
+    }
+
+    if (priv->thin_arbiter_count && !afr_ta_has_quorum(priv, local)) {
+        local->op_ret = -1;
+        local->op_errno = -afr_quorum_errno(priv);
+        goto read;
+    }
+
+    if (priv->thin_arbiter_count &&
+        AFR_COUNT(local->child_up, priv->child_count) != priv->child_count) {
+        if (local->child_up[0]) {
+            local->read_txn_query_child = AFR_CHILD_ZERO;
+        } else if (local->child_up[1]) {
+            local->read_txn_query_child = AFR_CHILD_ONE;
+        }
+        afr_ta_read_txn_synctask(frame, this);
+        return 0;
+    }
+
+    ret = afr_inode_read_subvol_get(inode, this, data, metadata,
+                                    &event_generation);
+    if (ret == -1)
+        /* very first transaction on this inode */
+        goto refresh;
+    AFR_INTERSECT(local->readable, data, metadata, priv->child_count);
+
+    gf_msg_debug(this->name, 0,
+                 "%s: generation now vs cached: %d, "
+                 "%d",
+                 uuid_utoa(inode->gfid), local->event_generation,
+                 event_generation);
+    if (afr_is_inode_refresh_reqd(inode, this, local->event_generation,
+                                  event_generation))
+        /* servers have disconnected / reconnected, and possibly
+           rebooted, very likely changing the state of freshness
+           of copies */
+        goto refresh;
+
+    read_subvol = afr_read_subvol_select_by_policy(inode, this, local->readable,
+                                                   NULL);
+
+    if (read_subvol < 0 || read_subvol > priv->child_count) {
+        gf_msg_debug(this->name, 0,
+                     "Unreadable subvolume %d found "
+                     "with event generation %d for gfid %s.",
+                     read_subvol, event_generation, uuid_utoa(inode->gfid));
+        goto refresh;
+    }
+
+    if (!local->child_up[read_subvol]) {
+        /* should never happen, just in case */
+        gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_READ_SUBVOL_ERROR,
+               "subvolume %d is the "
+               "read subvolume in this generation, but is not up",
+               read_subvol);
+        goto refresh;
+    }
+
+    local->read_attempted[read_subvol] = 1;
+
+read:
+    afr_read_txn_wind(frame, this, read_subvol);
+
+    return 0;
+
+refresh:
+    afr_inode_refresh(frame, this, inode, NULL, afr_read_txn_refresh_done);
+
+    return 0;
+}
diff --git a/xlators/cluster/afr/src/afr-self-heal-algorithm.c b/xlators/cluster/afr/src/afr-self-heal-algorithm.c
deleted file mode 100644
index 80478320173..00000000000
--- a/xlators/cluster/afr/src/afr-self-heal-algorithm.c
+++ /dev/null
@@ -1,1072 +0,0 @@
-/*
-   Copyright (c) 2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-
-#include "glusterfs.h"
-#include "afr.h"
-#include "xlator.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-#include "byte-order.h"
-#include "md5.h"
-
-#include "afr-transaction.h"
-#include "afr-self-heal.h"
-#include "afr-self-heal-common.h"
-#include "afr-self-heal-algorithm.h"
-
-/*
-  This file contains the various self-heal algorithms
-*/
-
-
-/*
-  The "full" algorithm. Copies the entire file from
-  source to sinks.
-*/
-
-
-static void
-sh_full_private_cleanup (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *               local   = NULL;
-        afr_self_heal_t *           sh      = NULL;
-        afr_sh_algo_full_private_t *sh_priv = NULL;
-
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        sh_priv = sh->private;
-
-        if (sh_priv)
-                GF_FREE (sh_priv);
-}
-
-
-static int
-sh_full_loop_driver (call_frame_t *frame, xlator_t *this);
-
-static int
-sh_full_loop_return (call_frame_t *rw_frame, xlator_t *this, off_t offset)
-{
-        afr_local_t *               rw_local   = NULL;
-        afr_self_heal_t *           rw_sh      = NULL;
-
-        call_frame_t *sh_frame              = NULL;
-	afr_local_t * sh_local              = NULL;
-	afr_self_heal_t *sh                 = NULL;
-        afr_sh_algo_full_private_t *sh_priv = NULL;
-
-        rw_local = rw_frame->local;
-        rw_sh    = &rw_local->self_heal;
-
-        sh_frame = rw_sh->sh_frame;
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-        sh_priv  = sh->private;
-
-        LOCK (&sh_priv->lock);
-        {
-                sh_priv->loops_running--;
-        }
-        UNLOCK (&sh_priv->lock);
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "loop for offset %"PRId64" returned", offset);
-
-        AFR_STACK_DESTROY (rw_frame);
-
-        sh_full_loop_driver (sh_frame, this);
-
-        return 0;
-}
-
-
-static int
-sh_full_write_cbk (call_frame_t *rw_frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                   struct iatt *postbuf)
-{
-	afr_private_t * priv    = NULL;
-	afr_local_t * rw_local  = NULL;
-	afr_self_heal_t *rw_sh  = NULL;
-
-        call_frame_t *sh_frame  = NULL;
-	afr_local_t * sh_local  = NULL;
-	afr_self_heal_t *sh     = NULL;
-
-	int child_index = (long) cookie;
-	int call_count = 0;
-
-	priv = this->private;
-
-	rw_local = rw_frame->local;
-	rw_sh    = &rw_local->self_heal;
-
-        sh_frame = rw_sh->sh_frame;
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-
-	gf_log (this->name, GF_LOG_TRACE,
-		"wrote %d bytes of data from %s to child %d, offset %"PRId64"",
-		op_ret, sh_local->loc.path, child_index,
-                rw_sh->offset - op_ret);
-
-	LOCK (&sh_frame->lock);
-	{
-		if (op_ret == -1) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"write to %s failed on subvolume %s (%s)",
-				sh_local->loc.path,
-				priv->children[child_index]->name,
-				strerror (op_errno));
-
-			sh->op_failed = 1;
-		}
-	}
-	UNLOCK (&sh_frame->lock);
-
-	call_count = afr_frame_return (rw_frame);
-
-	if (call_count == 0) {
-		sh_full_loop_return (rw_frame, this, rw_sh->offset - op_ret);
-	}
-
-	return 0;
-}
-
-
-static int
-sh_full_read_cbk (call_frame_t *rw_frame, void *cookie,
-                  xlator_t *this, int32_t op_ret, int32_t op_errno,
-                  struct iovec *vector, int32_t count, struct iatt *buf,
-                  struct iobref *iobref)
-{
-	afr_private_t * priv    = NULL;
-	afr_local_t * rw_local  = NULL;
-	afr_self_heal_t *rw_sh  = NULL;
-
-        call_frame_t *sh_frame  = NULL;
-	afr_local_t * sh_local  = NULL;
-	afr_self_heal_t *sh     = NULL;
-
-	int i = 0;
-	int call_count = 0;
-
-	off_t offset = (long) cookie;
-
-	priv = this->private;
-	rw_local = rw_frame->local;
-	rw_sh    = &rw_local->self_heal;
-
-        sh_frame = rw_sh->sh_frame;
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-
-	call_count = sh->active_sinks;
-
-        rw_local->call_count = call_count;
-
-	gf_log (this->name, GF_LOG_TRACE,
-		"read %d bytes of data from %s, offset %"PRId64"",
-		op_ret, sh_local->loc.path, offset);
-
-	if (op_ret <= 0) {
-                sh->op_failed = 1;
-
-                sh_full_loop_return (rw_frame, this, offset);
-		return 0;
-	}
-
-	rw_sh->offset += op_ret;
-
-	if (sh->file_has_holes) {
-		if (iov_0filled (vector, count) == 0) {
-			/* the iter function depends on the
-			   sh->offset already being updated
-			   above
-			*/
-
-                        sh_full_loop_return (rw_frame, this, offset);
-			goto out;
-		}
-	}
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->sources[i] || !sh_local->child_up[i])
-			continue;
-
-		/* this is a sink, so write to it */
-
-		STACK_WIND_COOKIE (rw_frame, sh_full_write_cbk,
-				   (void *) (long) i,
-				   priv->children[i],
-				   priv->children[i]->fops->writev,
-				   sh->healing_fd, vector, count, offset,
-                                   iobref);
-
-		if (!--call_count)
-			break;
-	}
-
-out:
-	return 0;
-}
-
-
-static int
-sh_full_read_write (call_frame_t *frame, xlator_t *this, off_t offset)
-{
-	afr_private_t * priv    = NULL;
-	afr_local_t * local     = NULL;
-	afr_local_t * rw_local  = NULL;
-	afr_self_heal_t *rw_sh  = NULL;
-	afr_self_heal_t *sh     = NULL;
-
-        call_frame_t *rw_frame = NULL;
-
-        int32_t op_errno = 0;
-
-	priv  = this->private;
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        rw_frame = copy_frame (frame);
-        if (!rw_frame)
-                goto out;
-
-        ALLOC_OR_GOTO (rw_local, afr_local_t, out);
-
-        rw_frame->local = rw_local;
-	rw_sh           = &rw_local->self_heal;
-
-        rw_sh->offset       = offset;
-        rw_sh->sh_frame     = frame;
-
-	STACK_WIND_COOKIE (rw_frame, sh_full_read_cbk,
-			   (void *) (long) offset,
-			   priv->children[sh->source],
-			   priv->children[sh->source]->fops->readv,
-			   sh->healing_fd, sh->block_size,
-			   offset);
-        return 0;
-
-out:
-        sh->op_failed = 1;
-
-        sh_full_loop_driver (frame, this);
-
-	return 0;
-}
-
-
-static int
-sh_full_loop_driver (call_frame_t *frame, xlator_t *this)
-{
-	afr_private_t * priv = NULL;
-	afr_local_t * local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-        afr_sh_algo_full_private_t *sh_priv = NULL;
-
-        int   loop    = 0;
-        int   recurse = 0;
-
-        off_t offset  = 0;
-
-	priv    = this->private;
-	local   = frame->local;
-	sh      = &local->self_heal;
-        sh_priv = sh->private;
-
-	if (sh->op_failed) {
-                if (sh_priv->loops_running == 0) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "full self-heal aborting on %s",
-                                local->loc.path);
-
-                        sh_full_private_cleanup (frame, this);
-                        local->self_heal.algo_abort_cbk (frame, this);
-                }
-
-		goto out;
-	}
-
-	if (sh_priv->offset >= sh->file_size) {
-                if (sh_priv->loops_running == 0) {
-
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "full self-heal completed on %s",
-                                local->loc.path);
-
-                        sh_full_private_cleanup (frame, this);
-                        local->self_heal.algo_completion_cbk (frame, this);
-                }
-
-		goto out;
-	}
-
-spawn:
-        loop    = 0;
-        recurse = 0;
-
-        LOCK (&sh_priv->lock);
-        {
-                if ((sh_priv->loops_running < priv->data_self_heal_window_size)
-                    && (sh_priv->offset < sh->file_size)) {
-
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "spawning a loop for offset %"PRId64,
-                                sh_priv->offset);
-
-                        offset           = sh_priv->offset;
-                        sh_priv->offset += sh->block_size;
-
-                        sh_priv->loops_running++;
-
-                        loop = 1;
-
-                        if (sh_priv->offset < sh->file_size)
-                                recurse = 1;
-                }
-        }
-        UNLOCK (&sh_priv->lock);
-
-        if (loop) {
-                sh_full_read_write (frame, this, offset);
-                if (recurse)
-                        goto spawn;
-        }
-
-out:
-	return 0;
-}
-
-
-int
-afr_sh_algo_full (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *               local   = NULL;
-        afr_self_heal_t *           sh      = NULL;
-        afr_sh_algo_full_private_t *sh_priv = NULL;
-
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        sh_priv = GF_CALLOC (1, sizeof (*sh_priv),
-                             gf_afr_mt_afr_private_t);
-
-        LOCK_INIT (&sh_priv->lock);
-
-        sh->private = sh_priv;
-
-        local->call_count = 0;
-
-        sh_full_loop_driver (frame, this);
-        return 0;
-}
-
-
-/*
- * The "diff" algorithm. Copies only those blocks whose checksums
- * don't match with those of source.
- */
-
-
-static void
-sh_diff_private_cleanup (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t *             priv    = NULL;
-        afr_local_t *               local   = NULL;
-        afr_self_heal_t *           sh      = NULL;
-        afr_sh_algo_diff_private_t *sh_priv = NULL;
-
-        int i;
-
-        priv  = this->private;
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        sh_priv = sh->private;
-
-        for (i = 0; i < priv->data_self_heal_window_size; i++) {
-                if (sh_priv->loops[i]) {
-                        if (sh_priv->loops[i]->write_needed)
-                                GF_FREE (sh_priv->loops[i]->write_needed);
-
-                        if (sh_priv->loops[i]->checksum)
-                                GF_FREE (sh_priv->loops[i]->checksum);
-
-                        GF_FREE (sh_priv->loops[i]);
-                }
-        }
-
-        if (sh_priv) {
-                if (sh_priv->loops)
-                        GF_FREE (sh_priv->loops);
-
-                GF_FREE (sh_priv);
-        }
-
-
-}
-
-
-static uint32_t
-__make_cookie (int loop_index, int child_index)
-{
-        uint32_t ret = (loop_index << 16) | child_index;
-        return ret;
-}
-
-
-static int
-__loop_index (uint32_t cookie)
-{
-        return (cookie & 0xFFFF0000) >> 16;
-}
-
-
-static int
-__child_index (uint32_t cookie)
-{
-        return (cookie & 0x0000FFFF);
-}
-
-
-static void
-sh_diff_loop_state_reset (struct sh_diff_loop_state *loop_state, int child_count)
-{
-        loop_state->active = _gf_false;
-//        loop_state->offset = 0;
-
-        memset (loop_state->write_needed,
-                0, sizeof (*loop_state->write_needed) * child_count);
-
-        memset (loop_state->checksum,
-                0, MD5_DIGEST_LEN * child_count);
-}
-
-
-static int
-sh_diff_number_of_writes_needed (unsigned char *write_needed, int child_count)
-{
-        int writes = 0;
-        int i;
-
-        for (i = 0; i < child_count; i++) {
-                if (write_needed[i])
-                        writes++;
-        }
-
-        return writes;
-}
-
-
-static int
-sh_diff_loop_driver (call_frame_t *frame, xlator_t *this);
-
-
-static int
-sh_diff_loop_return (call_frame_t *rw_frame, xlator_t *this,
-                     struct sh_diff_loop_state *loop_state)
-{
-        afr_private_t *             priv       = NULL;
-        afr_local_t *               rw_local   = NULL;
-        afr_self_heal_t *           rw_sh      = NULL;
-
-        call_frame_t *sh_frame              = NULL;
-	afr_local_t * sh_local              = NULL;
-	afr_self_heal_t *sh                 = NULL;
-        afr_sh_algo_diff_private_t *sh_priv = NULL;
-
-        priv  = this->private;
-
-        rw_local = rw_frame->local;
-        rw_sh    = &rw_local->self_heal;
-
-        sh_frame = rw_sh->sh_frame;
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-        sh_priv  = sh->private;
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "loop for offset %"PRId64" returned", loop_state->offset);
-
-        LOCK (&sh_priv->lock);
-        {
-                sh_priv->loops_running--;
-                sh_diff_loop_state_reset (loop_state, priv->child_count);
-        }
-        UNLOCK (&sh_priv->lock);
-
-        AFR_STACK_DESTROY (rw_frame);
-
-        sh_diff_loop_driver (sh_frame, this);
-
-        return 0;
-}
-
-
-static int
-sh_diff_write_cbk (call_frame_t *rw_frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct iatt *buf,
-                   struct iatt *postbuf)
-{
-	afr_private_t *   priv     = NULL;
-	afr_local_t *     rw_local = NULL;
-	afr_self_heal_t * rw_sh    = NULL;
-
-        call_frame_t *sh_frame  = NULL;
-	afr_local_t * sh_local  = NULL;
-	afr_self_heal_t *sh     = NULL;
-
-        afr_sh_algo_diff_private_t *sh_priv;
-        struct sh_diff_loop_state *loop_state;
-
-	int call_count  = 0;
-        int child_index = 0;
-        int loop_index  = 0;
-
-	priv     = this->private;
-	rw_local = rw_frame->local;
-	rw_sh    = &rw_local->self_heal;
-
-        sh_frame = rw_sh->sh_frame;
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-        sh_priv  = sh->private;
-
-        child_index = __child_index ((uint32_t) (long) cookie);
-        loop_index  = __loop_index ((uint32_t) (long) cookie);
-        loop_state  = sh_priv->loops[loop_index];
-
-	gf_log (this->name, GF_LOG_TRACE,
-		"wrote %d bytes of data from %s to child %d, offset %"PRId64"",
-		op_ret, sh_local->loc.path, child_index,
-                loop_state->offset);
-
-	LOCK (&sh_frame->lock);
-	{
-		if (op_ret == -1) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"write to %s failed on subvolume %s (%s)",
-				sh_local->loc.path,
-				priv->children[child_index]->name,
-				strerror (op_errno));
-
-			sh->op_failed = 1;
-		}
-	}
-	UNLOCK (&sh_frame->lock);
-
-	call_count = afr_frame_return (rw_frame);
-
-	if (call_count == 0) {
-		sh_diff_loop_return (rw_frame, this, loop_state);
-	}
-
-	return 0;
-}
-
-
-static int
-sh_diff_read_cbk (call_frame_t *rw_frame, void *cookie,
-                  xlator_t *this, int32_t op_ret, int32_t op_errno,
-                  struct iovec *vector, int32_t count, struct iatt *buf,
-                  struct iobref *iobref)
-{
-	afr_private_t *   priv     = NULL;
-	afr_local_t *     rw_local = NULL;
-	afr_self_heal_t * rw_sh    = NULL;
-
-        afr_sh_algo_diff_private_t * sh_priv = NULL;
-
-        call_frame_t *sh_frame  = NULL;
-	afr_local_t * sh_local  = NULL;
-	afr_self_heal_t *sh     = NULL;
-
-        int loop_index;
-        struct sh_diff_loop_state *loop_state;
-
-        uint32_t wcookie;
-
-	int i = 0;
-	int call_count = 0;
-
-	priv     = this->private;
-	rw_local = rw_frame->local;
-	rw_sh    = &rw_local->self_heal;
-
-        sh_frame = rw_sh->sh_frame;
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-        sh_priv  = sh->private;
-
-        loop_index = __loop_index ((uint32_t) (long) cookie);
-        loop_state = sh_priv->loops[loop_index];
-
-	call_count = sh_diff_number_of_writes_needed (loop_state->write_needed,
-                                                      priv->child_count);
-
-	rw_local->call_count = call_count;
-
-	gf_log (this->name, GF_LOG_TRACE,
-		"read %d bytes of data from %s, offset %"PRId64"",
-		op_ret, sh_local->loc.path, loop_state->offset);
-
-	if ((op_ret <= 0) ||
-            (call_count == 0)) {
-                sh_diff_loop_return (rw_frame, this, loop_state);
-
-		return 0;
-	}
-
-	if (sh->file_has_holes) {
-		if (iov_0filled (vector, count) == 0) {
-
-                        sh_diff_loop_return (rw_frame, this, loop_state);
-			goto out;
-		}
-	}
-
-	for (i = 0; i < priv->child_count; i++) {
-                if (loop_state->write_needed[i]) {
-                        wcookie = __make_cookie (loop_index, i);
-
-                        STACK_WIND_COOKIE (rw_frame, sh_diff_write_cbk,
-                                           (void *) (long) wcookie,
-                                           priv->children[i],
-                                           priv->children[i]->fops->writev,
-                                           sh->healing_fd, vector, count,
-                                           loop_state->offset, iobref);
-
-                        if (!--call_count)
-                                break;
-                }
-        }
-
-out:
-	return 0;
-}
-
-
-static int
-sh_diff_read (call_frame_t *rw_frame, xlator_t *this,
-              int loop_index)
-{
-	afr_private_t *   priv     = NULL;
-	afr_local_t *     rw_local = NULL;
-	afr_self_heal_t * rw_sh    = NULL;
-
-        afr_sh_algo_diff_private_t * sh_priv = NULL;
-        struct sh_diff_loop_state *loop_state;
-
-        call_frame_t *sh_frame  = NULL;
-	afr_local_t * sh_local  = NULL;
-	afr_self_heal_t *sh     = NULL;
-
-        uint32_t cookie;
-
-	priv     = this->private;
-	rw_local = rw_frame->local;
-	rw_sh    = &rw_local->self_heal;
-
-        sh_frame = rw_sh->sh_frame;
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-        sh_priv  = sh->private;
-
-        loop_state = sh_priv->loops[loop_index];
-
-        cookie = __make_cookie (loop_index, sh->source);
-
-	STACK_WIND_COOKIE (rw_frame, sh_diff_read_cbk,
-			   (void *) (long) cookie,
-			   priv->children[sh->source],
-			   priv->children[sh->source]->fops->readv,
-			   sh->healing_fd, sh_priv->block_size,
-			   loop_state->offset);
-
-	return 0;
-}
-
-
-static int
-sh_diff_checksum_cbk (call_frame_t *rw_frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno,
-                      uint32_t weak_checksum, uint8_t *strong_checksum)
-{
-	afr_private_t * priv    = NULL;
-	afr_local_t * rw_local  = NULL;
-	afr_self_heal_t *rw_sh  = NULL;
-
-        call_frame_t *sh_frame  = NULL;
-	afr_local_t * sh_local  = NULL;
-	afr_self_heal_t *sh     = NULL;
-
-        afr_sh_algo_diff_private_t * sh_priv = NULL;
-
-        int loop_index = 0;
-        int child_index = 0;
-        struct sh_diff_loop_state *loop_state;
-
-        int call_count   = 0;
-        int i            = 0;
-        int write_needed = 0;
-
-	priv  = this->private;
-
-	rw_local = rw_frame->local;
-	rw_sh    = &rw_local->self_heal;
-
-        sh_frame = rw_sh->sh_frame;
-        sh_local = sh_frame->local;
-        sh       = &sh_local->self_heal;
-
-        sh_priv = sh->private;
-
-        child_index = __child_index ((uint32_t) (long) cookie);
-        loop_index  = __loop_index ((uint32_t) (long) cookie);
-
-        loop_state  = sh_priv->loops[loop_index];
-
-        if (op_ret < 0) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "checksum on %s failed on subvolume %s (%s)",
-                        sh_local->loc.path, priv->children[child_index]->name,
-                        strerror (op_errno));
-
-                sh->op_failed = 1;
-        } else {
-                memcpy (loop_state->checksum + child_index * MD5_DIGEST_LEN,
-                        strong_checksum,
-                        MD5_DIGEST_LEN);
-        }
-
-        call_count = afr_frame_return (rw_frame);
-
-        if (call_count == 0) {
-                for (i = 0; i < priv->child_count; i++) {
-                        if (sh->sources[i] || !sh_local->child_up[i])
-                                continue;
-
-                        if (memcmp (loop_state->checksum + (i * MD5_DIGEST_LEN),
-                                    loop_state->checksum + (sh->source * MD5_DIGEST_LEN),
-                                    MD5_DIGEST_LEN)) {
-                                /*
-                                   Checksums differ, so this block
-                                   must be written to this sink
-                                */
-
-                                gf_log (this->name, GF_LOG_TRACE,
-                                        "checksum on subvolume %s at offset %"
-                                        PRId64" differs from that on source",
-                                        priv->children[i]->name, loop_state->offset);
-
-                                write_needed = loop_state->write_needed[i] = 1;
-                        }
-                }
-
-                LOCK (&sh_priv->lock);
-                {
-                        sh_priv->total_blocks++;
-                        if (write_needed)
-                                sh_priv->diff_blocks++;
-                }
-                UNLOCK (&sh_priv->lock);
-
-                if (write_needed && !sh->op_failed) {
-                        sh_diff_read (rw_frame, this, loop_index);
-                } else {
-                        sh->offset += sh_priv->block_size;
-
-                        sh_diff_loop_return (rw_frame, this, loop_state);
-                }
-        }
-
-        return 0;
-}
-
-
-static int
-sh_diff_find_unused_loop (afr_sh_algo_diff_private_t *sh_priv, int max)
-{
-        int i;
-
-        LOCK (&sh_priv->lock);
-        {
-                for (i = 0; i < max; i++) {
-                        if (sh_priv->loops[i]->active == _gf_false) {
-                                sh_priv->loops[i]->active = _gf_true;
-                                break;
-                        }
-                }
-        }
-        UNLOCK (&sh_priv->lock);
-
-        if (i == max) {
-                gf_log ("[sh-diff]", GF_LOG_ERROR,
-                        "no free loops found! This shouldn't happen. Please"
-                        " report this to gluster-devel@nongnu.org");
-        }
-
-        return i;
-}
-
-
-static int
-sh_diff_checksum (call_frame_t *frame, xlator_t *this, off_t offset)
-{
-	afr_private_t *   priv     = NULL;
-	afr_local_t *     local    = NULL;
-	afr_local_t *     rw_local = NULL;
-	afr_self_heal_t * sh       = NULL;
-	afr_self_heal_t * rw_sh    = NULL;
-
-        afr_sh_algo_diff_private_t * sh_priv = NULL;
-
-        call_frame_t *rw_frame = NULL;
-
-        uint32_t cookie;
-        int loop_index = 0;
-        struct sh_diff_loop_state *loop_state = NULL;
-
-        int32_t op_errno = 0;
-
-        int call_count = 0;
-        int i          = 0;
-
-	priv    = this->private;
-	local   = frame->local;
-	sh      = &local->self_heal;
-
-        sh_priv = sh->private;
-
-        rw_frame = copy_frame (frame);
-        if (!rw_frame)
-                goto out;
-
-        ALLOC_OR_GOTO (rw_local, afr_local_t, out);
-
-        rw_frame->local = rw_local;
-	rw_sh           = &rw_local->self_heal;
-
-        rw_sh->offset       = sh->offset;
-        rw_sh->sh_frame     = frame;
-
-        call_count = sh->active_sinks + 1;  /* sinks and source */
-
-        rw_local->call_count = call_count;
-
-        loop_index = sh_diff_find_unused_loop (sh_priv, priv->data_self_heal_window_size);
-
-        loop_state = sh_priv->loops[loop_index];
-        loop_state->offset       = offset;
-
-        /* we need to send both the loop index and child index,
-           so squeeze them both into a 32-bit number */
-
-        cookie = __make_cookie (loop_index, sh->source);
-
-        STACK_WIND_COOKIE (rw_frame, sh_diff_checksum_cbk,
-                           (void *) (long) cookie,
-                           priv->children[sh->source],
-                           priv->children[sh->source]->fops->rchecksum,
-                           sh->healing_fd,
-                           offset, sh_priv->block_size);
-
-        for (i = 0; i < priv->child_count; i++) {
-                if (sh->sources[i] || !local->child_up[i])
-                        continue;
-
-                cookie = __make_cookie (loop_index, i);
-
-                STACK_WIND_COOKIE (rw_frame, sh_diff_checksum_cbk,
-                                   (void *) (long) cookie,
-                                   priv->children[i],
-                                   priv->children[i]->fops->rchecksum,
-                                   sh->healing_fd,
-                                   offset, sh_priv->block_size);
-
-                if (!--call_count)
-                        break;
-        }
-
-        return 0;
-
-out:
-        sh->op_failed = 1;
-
-        sh_diff_loop_driver (frame, this);
-
-        return 0;
-}
-
-
-static int
-sh_diff_loop_driver (call_frame_t *frame, xlator_t *this)
-{
-	afr_private_t *   priv  = NULL;
-	afr_local_t *     local = NULL;
-	afr_self_heal_t * sh    = NULL;
-        afr_sh_algo_diff_private_t *sh_priv = NULL;
-
-        int   loop    = 0;
-        int   recurse = 0;
-
-        off_t offset = 0;
-        char  sh_type_str[256] = {0,};
- 
-	priv    = this->private;
-	local   = frame->local;
-	sh      = &local->self_heal;
-        sh_priv = sh->private;
-
-        afr_self_heal_type_str_get(sh, sh_type_str, sizeof(sh_type_str));
-
-	if (sh->op_failed) {
-                if (sh_priv->loops_running == 0) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "diff %s self-heal aborting on %s",
-                                sh_type_str, local->loc.path);
-
-                        sh_diff_private_cleanup (frame, this);
-                        local->self_heal.algo_abort_cbk (frame, this);
-                }
-
-		goto out;
-	}
-
-	if (sh_priv->offset >= sh->file_size) {
-                if (sh_priv->loops_running == 0) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "diff %s self-heal completed on %s",
-                                sh_type_str, local->loc.path);
-
-
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "diff %s self-heal on %s: %d blocks of %d were different (%.2f%%)",
-                                sh_type_str, local->loc.path,
-                                sh_priv->diff_blocks, sh_priv->total_blocks,
-                                ((sh_priv->diff_blocks * 1.0)/sh_priv->total_blocks) * 100);
-
-                        sh_diff_private_cleanup (frame, this);
-                        local->self_heal.algo_completion_cbk (frame, this);
-                }
-
-		goto out;
-	}
-
-spawn:
-        loop    = 0;
-        recurse = 0;
-
-        LOCK (&sh_priv->lock);
-        {
-                if ((sh_priv->loops_running < priv->data_self_heal_window_size)
-                    && (sh_priv->offset < sh->file_size)) {
-
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "spawning a loop for offset %"PRId64,
-                                sh_priv->offset);
-
-                        offset           = sh_priv->offset;
-                        sh_priv->offset += sh_priv->block_size;
-
-                        sh_priv->loops_running++;
-
-                        loop = 1;
-
-                        if (sh_priv->offset < sh->file_size)
-                                recurse = 1;
-                }
-        }
-        UNLOCK (&sh_priv->lock);
-
-        if (loop) {
-                sh_diff_checksum (frame, this, offset);
-                if (recurse)
-                    goto spawn;
-        }
-
-out:
-        return 0;
-}
-
-
-int
-afr_sh_algo_diff (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t *             priv    = NULL;
-        afr_local_t *               local   = NULL;
-        afr_self_heal_t *           sh      = NULL;
-        afr_sh_algo_diff_private_t *sh_priv = NULL;
-
-        int i;
-
-        priv  = this->private;
-        local = frame->local;
-        sh    = &local->self_heal;
-
-        sh_priv = GF_CALLOC (1, sizeof (*sh_priv),
-                             gf_afr_mt_afr_private_t);
-
-        sh_priv->block_size = this->ctx->page_size;
-
-        sh->private = sh_priv;
-
-        LOCK_INIT (&sh_priv->lock);
-
-        local->call_count = 0;
-
-        sh_priv->loops = GF_CALLOC (priv->data_self_heal_window_size,
-                                    sizeof (*sh_priv->loops),
-                                    gf_afr_mt_sh_diff_loop_state);
-
-        for (i = 0; i < priv->data_self_heal_window_size; i++) {
-                sh_priv->loops[i]               = GF_CALLOC (1, sizeof (*sh_priv->loops[i]),
-                                                             gf_afr_mt_sh_diff_loop_state);
-
-                sh_priv->loops[i]->checksum     = GF_CALLOC (priv->child_count,
-                                                             MD5_DIGEST_LEN, gf_afr_mt_uint8_t);
-                sh_priv->loops[i]->write_needed = GF_CALLOC (priv->child_count,
-                                                             sizeof (*sh_priv->loops[i]->write_needed),
-                                                             gf_afr_mt_char);
-        }
-
-        sh_diff_loop_driver (frame, this);
-
-        return 0;
-}
-
-
-struct afr_sh_algorithm afr_self_heal_algorithms[] = {
-        {.name = "full",  .fn = afr_sh_algo_full},
-        {.name = "diff",  .fn = afr_sh_algo_diff},
-        {0, 0},
-};
diff --git a/xlators/cluster/afr/src/afr-self-heal-algorithm.h b/xlators/cluster/afr/src/afr-self-heal-algorithm.h
deleted file mode 100644
index 4b27d2afe40..00000000000
--- a/xlators/cluster/afr/src/afr-self-heal-algorithm.h
+++ /dev/null
@@ -1,60 +0,0 @@
-/*
-   Copyright (c) 2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef __AFR_SELF_HEAL_ALGORITHM_H__
-#define __AFR_SELF_HEAL_ALGORITHM_H__
-
-
-typedef int (*afr_sh_algo_fn) (call_frame_t *frame,
-                               xlator_t *this);
-
-struct afr_sh_algorithm {
-        const char *name;
-        afr_sh_algo_fn fn;
-};
-
-extern struct afr_sh_algorithm afr_self_heal_algorithms[3];
-
-typedef struct {
-        gf_lock_t lock;
-        unsigned int loops_running;
-        off_t offset;
-} afr_sh_algo_full_private_t;
-
-struct sh_diff_loop_state {
-        off_t   offset;
-        unsigned char *write_needed;
-        uint8_t *checksum;
-        gf_boolean_t active;
-};
-
-typedef struct {
-        size_t block_size;
-
-        gf_lock_t lock;
-        unsigned int loops_running;
-        off_t offset;
-
-        int32_t total_blocks;
-        int32_t diff_blocks;
-
-        struct sh_diff_loop_state **loops;
-} afr_sh_algo_diff_private_t;
-
-#endif /* __AFR_SELF_HEAL_ALGORITHM_H__ */
diff --git a/xlators/cluster/afr/src/afr-self-heal-common.c b/xlators/cluster/afr/src/afr-self-heal-common.c
index 83be5ed8c41..a580a1584cc 100644
--- a/xlators/cluster/afr/src/afr-self-heal-common.c
+++ b/xlators/cluster/afr/src/afr-self-heal-common.c
@@ -1,1657 +1,2934 @@
 /*
-  Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
-#include "glusterfs.h"
-#include "xlator.h"
-#include "byte-order.h"
-
 #include "afr.h"
-#include "afr-transaction.h"
-#include "afr-self-heal-common.h"
 #include "afr-self-heal.h"
-#include "pump.h"
+#include <glusterfs/byte-order.h>
+#include "protocol-common.h"
+#include "afr-messages.h"
+#include <glusterfs/events.h>
 
-/**
- * select_source - select a source and return it
- */
+void
+afr_heal_synctask(xlator_t *this, afr_local_t *local);
 
 int
-afr_sh_select_source (int sources[], int child_count)
+afr_lookup_and_heal_gfid(xlator_t *this, inode_t *parent, const char *name,
+                         inode_t *inode, struct afr_reply *replies, int source,
+                         unsigned char *sources, void *gfid, int *gfid_idx)
 {
-	int i;
-	for (i = 0; i < child_count; i++)
-		if (sources[i])
-			return i;
+    afr_private_t *priv = NULL;
+    call_frame_t *frame = NULL;
+    afr_local_t *local = NULL;
+    unsigned char *wind_on = NULL;
+    ia_type_t ia_type = IA_INVAL;
+    dict_t *xdata = NULL;
+    loc_t loc = {
+        0,
+    };
+    int ret = 0;
+    int i = 0;
+
+    priv = this->private;
+    wind_on = alloca0(priv->child_count);
+    if (source >= 0 && replies[source].valid && replies[source].op_ret == 0)
+        ia_type = replies[source].poststat.ia_type;
+
+    if (ia_type != IA_INVAL)
+        goto heal;
+
+    /* If ia_type is still invalid, it means either
+     * (a)'source' was -1, i.e. parent dir pending xattrs are in split-brain
+     * (or) (b) The parent dir pending xattrs are all zeroes (i.e. all bricks
+     * are sources) and the 'source' we selected earlier might be the one where
+     * the file is not actually present.
+     *
+     * In both cases, let us pick a brick with a successful reply and use its
+     * ia_type.
+     * */
+    for (i = 0; i < priv->child_count; i++) {
+        if (source == -1) {
+            /* case (a) above. */
+            if (replies[i].valid && replies[i].op_ret == 0 &&
+                replies[i].poststat.ia_type != IA_INVAL) {
+                ia_type = replies[i].poststat.ia_type;
+                break;
+            }
+        } else {
+            /* case (b) above. */
+            if (i == source)
+                continue;
+            if (sources[i] && replies[i].valid && replies[i].op_ret == 0 &&
+                replies[i].poststat.ia_type != IA_INVAL) {
+                ia_type = replies[i].poststat.ia_type;
+                break;
+            }
+        }
+    }
 
-	return -1;
-}
+heal:
+    /* gfid heal on those subvolumes that do not have gfid associated
+     * with the inode and update those replies.
+     */
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret != 0)
+            continue;
 
+        if (gf_uuid_is_null(gfid) &&
+            !gf_uuid_is_null(replies[i].poststat.ia_gfid) &&
+            replies[i].poststat.ia_type == ia_type)
+            gfid = replies[i].poststat.ia_gfid;
 
-/**
- * sink_count - return number of sinks in sources array
- */
+        if (!gf_uuid_is_null(replies[i].poststat.ia_gfid) ||
+            replies[i].poststat.ia_type != ia_type)
+            continue;
+
+        wind_on[i] = 1;
+    }
+
+    if (AFR_COUNT(wind_on, priv->child_count) == 0)
+        return 0;
+
+    xdata = dict_new();
+    if (!xdata) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ret = dict_set_gfuuid(xdata, "gfid-req", gfid, true);
+    if (ret) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    frame = afr_frame_create(this, &ret);
+    if (!frame) {
+        ret = -ret;
+        goto out;
+    }
+
+    local = frame->local;
+    loc.parent = inode_ref(parent);
+    gf_uuid_copy(loc.pargfid, parent->gfid);
+    loc.name = name;
+    loc.inode = inode_ref(inode);
+
+    AFR_ONLIST(wind_on, frame, afr_selfheal_discover_cbk, lookup, &loc, xdata);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!wind_on[i])
+            continue;
+        afr_reply_wipe(&replies[i]);
+        afr_reply_copy(&replies[i], &local->replies[i]);
+    }
+    if (gfid_idx && (*gfid_idx == -1)) {
+        /*Pick a brick where the gifd heal was successful.*/
+        for (i = 0; i < priv->child_count; i++) {
+            if (!wind_on[i])
+                continue;
+            if (replies[i].valid && replies[i].op_ret == 0 &&
+                !gf_uuid_is_null(replies[i].poststat.ia_gfid)) {
+                *gfid_idx = i;
+                break;
+            }
+        }
+    }
+out:
+    if (gfid_idx && (*gfid_idx == -1) && (ret == 0) && local) {
+        ret = -afr_final_errno(local, priv);
+    }
+    loc_wipe(&loc);
+    if (frame)
+        AFR_STACK_DESTROY(frame);
+    if (xdata)
+        dict_unref(xdata);
+
+    return ret;
+}
 
 int
-afr_sh_sink_count (int sources[], int child_count)
+afr_gfid_sbrain_source_from_src_brick(xlator_t *this, struct afr_reply *replies,
+                                      char *src_brick)
 {
-	int i;
-	int sinks = 0;
-	for (i = 0; i < child_count; i++)
-		if (!sources[i])
-			sinks++;
-	return sinks;
+    int i = 0;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret == -1)
+            continue;
+        if (strcmp(priv->children[i]->name, src_brick) == 0)
+            return i;
+    }
+    return -1;
 }
 
 int
-afr_sh_source_count (int sources[], int child_count)
+afr_selfheal_gfid_mismatch_by_majority(struct afr_reply *replies,
+                                       int child_count)
 {
-	int i;
-	int nsource = 0;
+    int j = 0;
+    int i = 0;
+    int votes;
+
+    for (i = 0; i < child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret == -1)
+            continue;
+
+        votes = 1;
+        for (j = i + 1; j < child_count; j++) {
+            if ((!gf_uuid_compare(replies[i].poststat.ia_gfid,
+                                  replies[j].poststat.ia_gfid)))
+                votes++;
+            if (votes > child_count / 2)
+                return i;
+        }
+    }
 
-	for (i = 0; i < child_count; i++)
-		if (sources[i])
-			nsource++;
-	return nsource;
+    return -1;
 }
 
-
 int
-afr_sh_supress_errenous_children (int sources[], int child_errno[],
-				  int child_count)
+afr_gfid_sbrain_source_from_bigger_file(struct afr_reply *replies,
+                                        int child_count)
 {
-	int i = 0;
-
-	for (i = 0; i < child_count; i++) {
-		if (child_errno[i] && sources[i]) {
-			sources[i] = 0;
-		}
-	}
-
-	return 0;
+    int i = 0;
+    int src = -1;
+    uint64_t size = 0;
+
+    for (i = 0; i < child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret == -1)
+            continue;
+        if (size < replies[i].poststat.ia_size) {
+            src = i;
+            size = replies[i].poststat.ia_size;
+        } else if (replies[i].poststat.ia_size == size) {
+            src = -1;
+        }
+    }
+    return src;
 }
 
-
-void
-afr_sh_print_pending_matrix (int32_t *pending_matrix[], xlator_t *this)
+int
+afr_gfid_sbrain_source_from_latest_mtime(struct afr_reply *replies,
+                                         int child_count)
 {
-	afr_private_t * priv = this->private;
-
-	char *buf = NULL;
-	char *ptr = NULL;
-
-	int i, j;
+    int i = 0;
+    int src = -1;
+    uint32_t mtime = 0;
+    uint32_t mtime_nsec = 0;
+
+    for (i = 0; i < child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret != 0)
+            continue;
+        if ((mtime < replies[i].poststat.ia_mtime) ||
+            ((mtime == replies[i].poststat.ia_mtime) &&
+             (mtime_nsec < replies[i].poststat.ia_mtime_nsec))) {
+            src = i;
+            mtime = replies[i].poststat.ia_mtime;
+            mtime_nsec = replies[i].poststat.ia_mtime_nsec;
+        } else if ((mtime == replies[i].poststat.ia_mtime) &&
+                   (mtime_nsec == replies[i].poststat.ia_mtime_nsec)) {
+            src = -1;
+        }
+    }
+    return src;
+}
 
-        /* 10 digits per entry + 1 space + '[' and ']' */
-	buf = GF_MALLOC (priv->child_count * 11 + 8, gf_afr_mt_char);
+int
+afr_gfid_split_brain_source(xlator_t *this, struct afr_reply *replies,
+                            inode_t *inode, uuid_t pargfid, const char *bname,
+                            int src_idx, int child_idx,
+                            unsigned char *locked_on, int *src, dict_t *xdata)
+{
+    afr_private_t *priv = NULL;
+    char g1[64] = {
+        0,
+    };
+    char g2[64] = {
+        0,
+    };
+    int up_count = 0;
+    int heal_op = -1;
+    int ret = -1;
+    char *src_brick = NULL;
+
+    *src = -1;
+    priv = this->private;
+    up_count = AFR_COUNT(locked_on, priv->child_count);
+    if (up_count != priv->child_count) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+               "All the bricks should be up to resolve the gfid split "
+               "barin");
+        if (xdata) {
+            ret = dict_set_sizen_str_sizen(xdata, "gfid-heal-msg",
+                                           SALL_BRICKS_UP_TO_RESOLVE);
+            if (ret)
+                gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_DICT_SET_FAILED,
+                       "Error setting"
+                       " gfid-heal-msg dict");
+        }
+        goto out;
+    }
 
-	for (i = 0; i < priv->child_count; i++) {
-		ptr = buf;
-		ptr += sprintf (ptr, "[ ");
-		for (j = 0; j < priv->child_count; j++) {
-			ptr += sprintf (ptr, "%d ", pending_matrix[i][j]);
-		}
-		sprintf (ptr, "]");
-		gf_log (this->name, GF_LOG_TRACE,
-			"pending_matrix: %s", buf);
-	}
+    if (xdata) {
+        ret = dict_get_int32_sizen(xdata, "heal-op", &heal_op);
+        if (ret)
+            goto fav_child;
+    } else {
+        goto fav_child;
+    }
+
+    switch (heal_op) {
+        case GF_SHD_OP_SBRAIN_HEAL_FROM_BIGGER_FILE:
+            *src = afr_gfid_sbrain_source_from_bigger_file(replies,
+                                                           priv->child_count);
+            if (*src == -1) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+                       SNO_BIGGER_FILE);
+                if (xdata) {
+                    ret = dict_set_sizen_str_sizen(xdata, "gfid-heal-msg",
+                                                   SNO_BIGGER_FILE);
+                    if (ret)
+                        gf_msg(this->name, GF_LOG_ERROR, 0,
+                               AFR_MSG_DICT_SET_FAILED,
+                               "Error"
+                               " setting gfid-heal-msg dict");
+                }
+            }
+            break;
+
+        case GF_SHD_OP_SBRAIN_HEAL_FROM_LATEST_MTIME:
+            *src = afr_gfid_sbrain_source_from_latest_mtime(replies,
+                                                            priv->child_count);
+            if (*src == -1) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+                       SNO_DIFF_IN_MTIME);
+                if (xdata) {
+                    ret = dict_set_sizen_str_sizen(xdata, "gfid-heal-msg",
+                                                   SNO_DIFF_IN_MTIME);
+                    if (ret)
+                        gf_msg(this->name, GF_LOG_ERROR, 0,
+                               AFR_MSG_DICT_SET_FAILED,
+                               "Error"
+                               "setting gfid-heal-msg dict");
+                }
+            }
+            break;
+
+        case GF_SHD_OP_SBRAIN_HEAL_FROM_BRICK:
+            ret = dict_get_str_sizen(xdata, "child-name", &src_brick);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+                       "Error getting the source "
+                       "brick");
+                break;
+            }
+            *src = afr_gfid_sbrain_source_from_src_brick(this, replies,
+                                                         src_brick);
+            if (*src == -1) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+                       SERROR_GETTING_SRC_BRICK);
+                if (xdata) {
+                    ret = dict_set_sizen_str_sizen(xdata, "gfid-heal-msg",
+                                                   SERROR_GETTING_SRC_BRICK);
+                    if (ret)
+                        gf_msg(this->name, GF_LOG_ERROR, 0,
+                               AFR_MSG_DICT_SET_FAILED,
+                               "Error"
+                               " setting gfid-heal-msg dict");
+                }
+            }
+            break;
+
+        default:
+            break;
+    }
+    goto out;
+
+fav_child:
+    switch (priv->fav_child_policy) {
+        case AFR_FAV_CHILD_BY_SIZE:
+            *src = afr_sh_fav_by_size(this, replies, inode);
+            break;
+        case AFR_FAV_CHILD_BY_MTIME:
+            *src = afr_sh_fav_by_mtime(this, replies, inode);
+            break;
+        case AFR_FAV_CHILD_BY_CTIME:
+            *src = afr_sh_fav_by_ctime(this, replies, inode);
+            break;
+        case AFR_FAV_CHILD_BY_MAJORITY:
+            if (priv->child_count != 2)
+                *src = afr_selfheal_gfid_mismatch_by_majority(
+                    replies, priv->child_count);
+            else
+                *src = -1;
+
+            if (*src == -1) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+                       "No majority to resolve "
+                       "gfid split brain");
+            }
+            break;
+        default:
+            break;
+    }
 
-	GF_FREE (buf);
+out:
+    if (*src == -1) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+               "Gfid mismatch detected for <gfid:%s>/%s>, %s on %s and"
+               " %s on %s.",
+               uuid_utoa(pargfid), bname,
+               uuid_utoa_r(replies[child_idx].poststat.ia_gfid, g1),
+               priv->children[child_idx]->name,
+               uuid_utoa_r(replies[src_idx].poststat.ia_gfid, g2),
+               priv->children[src_idx]->name);
+        gf_event(EVENT_AFR_SPLIT_BRAIN,
+                 "client-pid=%d;"
+                 "subvol=%s;type=gfid;file="
+                 "<gfid:%s>/%s>;count=2;child-%d=%s;gfid-%d=%s;"
+                 "child-%d=%s;gfid-%d=%s",
+                 this->ctx->cmd_args.client_pid, this->name, uuid_utoa(pargfid),
+                 bname, child_idx, priv->children[child_idx]->name, child_idx,
+                 uuid_utoa_r(replies[child_idx].poststat.ia_gfid, g1), src_idx,
+                 priv->children[src_idx]->name, src_idx,
+                 uuid_utoa_r(replies[src_idx].poststat.ia_gfid, g2));
+        return -1;
+    }
+    return 0;
 }
 
+int
+afr_selfheal_post_op_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
 
-void
-afr_sh_build_pending_matrix (afr_private_t *priv,
-                             int32_t *pending_matrix[], dict_t *xattr[],
-			     int child_count, afr_transaction_type type)
-{
-	int i, j, k;
-
-	/* Indexable by result of afr_index_for_transaction_type(): 0 -- 2. */
-	int32_t pending[3];
-	void *pending_raw = NULL;
-	int ret = -1;
-
-        unsigned char *ignorant_subvols = NULL;
-
-        ignorant_subvols = GF_CALLOC (sizeof (*ignorant_subvols), child_count,
-                                      gf_afr_mt_char);
-
-	/* start clean */
-	for (i = 0; i < child_count; i++) {
-		for (j = 0; j < child_count; j++) {
-			pending_matrix[i][j] = 0;
-		}
-	}
-
-	for (i = 0; i < child_count; i++) {
-		pending_raw = NULL;
-
-                for (j = 0; j < child_count; j++) {
-                        ret = dict_get_ptr (xattr[i], priv->pending_key[j],
-                                            &pending_raw);
-                        
-                        if (ret != 0) {
-                                /*
-                                 * There is no xattr present. This means this
-                                 * subvolume should be considered an 'ignorant'
-                                 * subvolume.
-                                 */
-
-                                ignorant_subvols[i] = 1;
-                                continue;
-                        }
-
-			memcpy (pending, pending_raw, sizeof(pending));
-                        k = afr_index_for_transaction_type (type);
-                        
-                        pending_matrix[i][j] = ntoh32 (pending[k]);
-                }
-	}
-
-        /*
-         * Make all non-ignorant subvols point towards the ignorant
-         * subvolumes.
-         */
+    local = frame->local;
 
-        for (i = 0; i < child_count; i++) {
-                if (ignorant_subvols[i]) {
-                        for (j = 0; j < child_count; j++) {
-                                if (!ignorant_subvols[j])
-                                        pending_matrix[j][i] += 1;
-                        }
-                }
-        }
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
+    syncbarrier_wake(&local->barrier);
 
-        GF_FREE (ignorant_subvols);
+    return 0;
 }
 
+int
+afr_selfheal_post_op(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                     int subvol, dict_t *xattr, dict_t *xdata)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    loc_t loc = {
+        0,
+    };
+    int ret = 0;
 
-/**
- * mark_sources: Mark all 'source' nodes and return number of source
- * nodes found
- *
- * A node (a row in the pending matrix) belongs to one of
- * three categories:
- *
- * M is the pending matrix.
- *
- * 'innocent' - M[i] is all zeroes
- * 'fool'     - M[i] has i'th element = 1 (self-reference)
- * 'wise'     - M[i] has i'th element = 0, others are 1 or 0.
- *
- * All 'innocent' nodes are sinks. If all nodes are innocent, no self-heal is
- * needed.
- *
- * A 'wise' node can be a source. If two 'wise' nodes conflict, it is 
- * a split-brain. If one wise node refers to the other but the other doesn't
- * refer back, the referrer is a source.
- *
- * All fools are sinks, unless there are no 'wise' nodes. In that case,
- * one of the fools is made a source.
- */
+    priv = this->private;
+    local = frame->local;
 
-typedef enum {
-        AFR_NODE_INNOCENT,
-        AFR_NODE_FOOL,
-        AFR_NODE_WISE
-} afr_node_type;
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-typedef struct {
-        afr_node_type type;
-        int           wisdom;
-} afr_node_character;
+    local->op_ret = 0;
 
+    STACK_WIND(frame, afr_selfheal_post_op_cbk, priv->children[subvol],
+               priv->children[subvol]->fops->xattrop, &loc,
+               GF_XATTROP_ADD_ARRAY, xattr, xdata);
 
-static int
-afr_sh_is_innocent (int32_t *array, int child_count)
-{
-        int i   = 0;
-        int ret = 1;   /* innocent until proven guilty */
+    syncbarrier_wait(&local->barrier, 1);
+    if (local->op_ret < 0)
+        ret = -local->op_errno;
 
-        for (i = 0; i < child_count; i++) {
-                if (array[i]) {
-                        ret = 0;
-                        break;
-                }
-        }
+    loc_wipe(&loc);
+    local->op_ret = 0;
 
-        return ret;
+    return ret;
 }
 
-
-static int
-afr_sh_is_fool (int32_t *array, int i, int child_count)
+int
+afr_check_stale_error(struct afr_reply *replies, afr_private_t *priv)
 {
-        return array[i];   /* fool if accuses itself */
+    int i = 0;
+    int op_errno = 0;
+    int tmp_errno = 0;
+    int stale_count = 0;
+
+    for (i = 0; i < priv->child_count; i++) {
+        tmp_errno = replies[i].op_errno;
+        if (tmp_errno == ENOENT || tmp_errno == ESTALE) {
+            op_errno = afr_higher_errno(op_errno, tmp_errno);
+            stale_count++;
+        }
+    }
+    if (stale_count != priv->child_count)
+        return -ENOTCONN;
+    else
+        return -op_errno;
 }
 
-
-static int
-afr_sh_is_wise (int32_t *array, int i, int child_count)
+int
+afr_sh_generic_fop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, struct iatt *pre,
+                       struct iatt *post, dict_t *xdata)
 {
-        return !array[i];  /* wise if does not accuse itself */
-}
+    int i = (long)cookie;
+    afr_local_t *local = NULL;
 
+    local = frame->local;
 
-static int
-afr_sh_all_nodes_innocent (afr_node_character *characters, 
-                           int child_count)
-{
-        int i   = 0;
-        int ret = 1;
+    local->replies[i].valid = 1;
+    local->replies[i].op_ret = op_ret;
+    local->replies[i].op_errno = op_errno;
+    if (pre)
+        local->replies[i].prestat = *pre;
+    if (post)
+        local->replies[i].poststat = *post;
+    if (xdata)
+        local->replies[i].xdata = dict_ref(xdata);
 
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].type != AFR_NODE_INNOCENT) {
-                        ret = 0;
-                        break;
-                }
-        }
+    syncbarrier_wake(&local->barrier);
 
-        return ret;
+    return 0;
 }
 
-
-static int
-afr_sh_wise_nodes_exist (afr_node_character *characters, int child_count)
+int
+afr_selfheal_restore_time(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                          int source, unsigned char *healed_sinks,
+                          struct afr_reply *replies)
 {
-        int i   = 0;
-        int ret = 0;
+    loc_t loc = {
+        0,
+    };
 
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].type == AFR_NODE_WISE) {
-                        ret = 1;
-                        break;
-                }
-        }
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-        return ret;
-}
+    AFR_ONLIST(healed_sinks, frame, afr_sh_generic_fop_cbk, setattr, &loc,
+               &replies[source].poststat,
+               (GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME | GF_SET_ATTR_CTIME),
+               NULL);
 
+    loc_wipe(&loc);
 
-/*
- * The 'wisdom' of a wise node is 0 if any other wise node accuses it.
- * It is 1 if no other wise node accuses it. 
- * Only wise nodes with wisdom 1 are sources.
- *
- * If no nodes with wisdom 1 exist, a split-brain has occured.
- */
-
-static void
-afr_sh_compute_wisdom (int32_t *pending_matrix[],
-                       afr_node_character characters[], int child_count)
-{
-        int i = 0;
-        int j = 0;
-
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].type == AFR_NODE_WISE) {
-                        characters[i].wisdom = 1;
-
-                        for (j = 0; j < child_count; j++) {
-                                if ((characters[j].type == AFR_NODE_WISE)
-                                    && pending_matrix[j][i]) {
-                                        
-                                        characters[i].wisdom = 0;
-                                }
-                        }
-                }
-        }
+    return 0;
 }
 
-
-static int
-afr_sh_wise_nodes_conflict (afr_node_character *characters, 
-                            int child_count)
+dict_t *
+afr_selfheal_output_xattr(xlator_t *this, gf_boolean_t is_full_crawl,
+                          afr_transaction_type type, int *output_dirty,
+                          int **output_matrix, int subvol,
+                          int **full_heal_mtx_out)
 {
-        int i   = 0;
-        int ret = 1;
-
-        for (i = 0; i < child_count; i++) {
-                if ((characters[i].type == AFR_NODE_WISE)
-                    && characters[i].wisdom == 1) {
-
-                        /* There is atleast one bona-fide wise node */
-                        ret = 0;
-                        break;
-                }
+    int j = 0;
+    int idx = 0;
+    int d_idx = 0;
+    int ret = 0;
+    int *raw = 0;
+    dict_t *xattr = NULL;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    idx = afr_index_for_transaction_type(type);
+    d_idx = afr_index_for_transaction_type(AFR_DATA_TRANSACTION);
+
+    xattr = dict_new();
+    if (!xattr)
+        return NULL;
+
+    /* clear dirty */
+    raw = GF_CALLOC(sizeof(int), AFR_NUM_CHANGE_LOGS, gf_afr_mt_int32_t);
+    if (!raw)
+        goto err;
+
+    raw[idx] = hton32(output_dirty[subvol]);
+    ret = dict_set_bin(xattr, AFR_DIRTY, raw,
+                       sizeof(int) * AFR_NUM_CHANGE_LOGS);
+    if (ret) {
+        GF_FREE(raw);
+        goto err;
+    }
+
+    /* clear/set pending */
+    for (j = 0; j < priv->child_count; j++) {
+        raw = GF_CALLOC(sizeof(int), AFR_NUM_CHANGE_LOGS, gf_afr_mt_int32_t);
+        if (!raw)
+            goto err;
+
+        raw[idx] = hton32(output_matrix[subvol][j]);
+        if (is_full_crawl)
+            raw[d_idx] = hton32(full_heal_mtx_out[subvol][j]);
+
+        ret = dict_set_bin(xattr, priv->pending_key[j], raw,
+                           sizeof(int) * AFR_NUM_CHANGE_LOGS);
+        if (ret) {
+            GF_FREE(raw);
+            goto err;
         }
+    }
 
-        return ret;
+    return xattr;
+err:
+    if (xattr)
+        dict_unref(xattr);
+    return NULL;
 }
 
+int
+afr_selfheal_undo_pending(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                          unsigned char *sources, unsigned char *sinks,
+                          unsigned char *healed_sinks,
+                          unsigned char *undid_pending,
+                          afr_transaction_type type, struct afr_reply *replies,
+                          unsigned char *locked_on)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i = 0;
+    int j = 0;
+    unsigned char *pending = NULL;
+    int *input_dirty = NULL;
+    int **input_matrix = NULL;
+    int **full_heal_mtx_in = NULL;
+    int **full_heal_mtx_out = NULL;
+    int *output_dirty = NULL;
+    int **output_matrix = NULL;
+    dict_t *xattr = NULL;
+    dict_t *xdata = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    pending = alloca0(priv->child_count);
+
+    input_dirty = alloca0(priv->child_count * sizeof(int));
+    input_matrix = ALLOC_MATRIX(priv->child_count, int);
+    full_heal_mtx_in = ALLOC_MATRIX(priv->child_count, int);
+    full_heal_mtx_out = ALLOC_MATRIX(priv->child_count, int);
+    output_dirty = alloca0(priv->child_count * sizeof(int));
+    output_matrix = ALLOC_MATRIX(priv->child_count, int);
+
+    xdata = dict_new();
+    if (!xdata)
+        return -1;
+
+    afr_selfheal_extract_xattr(this, replies, type, input_dirty, input_matrix);
+
+    if (local->need_full_crawl)
+        afr_selfheal_extract_xattr(this, replies, AFR_DATA_TRANSACTION, NULL,
+                                   full_heal_mtx_in);
+
+    for (i = 0; i < priv->child_count; i++)
+        if (sinks[i] && !healed_sinks[i])
+            pending[i] = 1;
+
+    for (i = 0; i < priv->child_count; i++) {
+        for (j = 0; j < priv->child_count; j++) {
+            if (pending[j]) {
+                output_matrix[i][j] = 1;
+                if (type == AFR_ENTRY_TRANSACTION)
+                    full_heal_mtx_out[i][j] = 1;
+            } else if (locked_on[j]) {
+                output_matrix[i][j] = -input_matrix[i][j];
+                if (type == AFR_ENTRY_TRANSACTION)
+                    full_heal_mtx_out[i][j] = -full_heal_mtx_in[i][j];
+            }
+        }
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!pending[i])
+            output_dirty[i] = -input_dirty[i];
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!locked_on[i])
+            /* perform post-op only on subvols we had locked
+               and inspected on.
+            */
+            continue;
+        if (undid_pending[i])
+            /* We already unset the pending xattrs in
+             * _afr_fav_child_reset_sink_xattrs(). */
+            continue;
+
+        xattr = afr_selfheal_output_xattr(this, local->need_full_crawl, type,
+                                          output_dirty, output_matrix, i,
+                                          full_heal_mtx_out);
+        if (!xattr) {
+            continue;
+        }
 
-static int
-afr_sh_mark_wisest_as_sources (int sources[], 
-                               afr_node_character *characters, 
-                               int child_count)
-{
-        int nsources = 0;
-        
-        int i = 0;
-
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].wisdom == 1) {
-                        sources[i] = 1;
-                        nsources++;
-                }
+        if ((type == AFR_ENTRY_TRANSACTION) && (priv->esh_granular)) {
+            if (xdata && dict_set_int8(xdata, GF_XATTROP_PURGE_INDEX, 1))
+                gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_DICT_SET_FAILED,
+                       "Failed to set"
+                       " dict value for %s",
+                       GF_XATTROP_PURGE_INDEX);
         }
 
-        return nsources;
+        afr_selfheal_post_op(frame, this, inode, i, xattr, xdata);
+        dict_unref(xattr);
+    }
+
+    if (xdata)
+        dict_unref(xdata);
+
+    return 0;
 }
 
+void
+afr_reply_copy(struct afr_reply *dst, struct afr_reply *src)
+{
+    dict_t *xdata = NULL;
+
+    dst->valid = src->valid;
+    dst->op_ret = src->op_ret;
+    dst->op_errno = src->op_errno;
+    dst->prestat = src->prestat;
+    dst->poststat = src->poststat;
+    dst->preparent = src->preparent;
+    dst->postparent = src->postparent;
+    dst->preparent2 = src->preparent2;
+    dst->postparent2 = src->postparent2;
+    if (src->xdata)
+        xdata = dict_ref(src->xdata);
+    else
+        xdata = NULL;
+    if (dst->xdata)
+        dict_unref(dst->xdata);
+    dst->xdata = xdata;
+    if (xdata && dict_get_str_boolean(xdata, "fips-mode-rchecksum",
+                                      _gf_false) == _gf_true) {
+        memcpy(dst->checksum, src->checksum, SHA256_DIGEST_LENGTH);
+    } else {
+        memcpy(dst->checksum, src->checksum, MD5_DIGEST_LENGTH);
+    }
+    dst->fips_mode_rchecksum = src->fips_mode_rchecksum;
+}
 
-static int
-afr_sh_mark_if_size_differs (afr_self_heal_t *sh, int child_count)
+void
+afr_replies_copy(struct afr_reply *dst, struct afr_reply *src, int count)
 {
-        int32_t ** pending_matrix;
-        int i, j;
+    int i = 0;
 
-        int size_differs = 0;
+    if (dst == src)
+        return;
 
-        pending_matrix = sh->pending_matrix;
+    for (i = 0; i < count; i++) {
+        afr_reply_copy(&dst[i], &src[i]);
+    }
+}
 
-        for (i = 0; i < child_count; i++) {
-                for (j = 0; j < child_count; j++) {
-                        if (!sh->buf)
-                                break;
+int
+afr_selfheal_fill_dirty(xlator_t *this, int *dirty, int subvol, int idx,
+                        dict_t *xdata)
+{
+    void *pending_raw = NULL;
+    int pending[3] = {
+        0,
+    };
 
-                        if (SIZE_DIFFERS (&sh->buf[i], &sh->buf[j])
-                            && (pending_matrix[i][j] == 0)
-                            && (pending_matrix[j][i] == 0)) {
+    if (!dirty)
+        return 0;
 
-                                pending_matrix[i][j] = 1;
-                                pending_matrix[j][i] = 1;
+    if (dict_get_ptr(xdata, AFR_DIRTY, &pending_raw))
+        return -1;
 
-                                size_differs = 1;
-                        }
-                }
-        }
+    if (!pending_raw)
+        return -1;
 
-        return size_differs;
+    memcpy(pending, pending_raw, sizeof(pending));
+
+    dirty[subvol] = ntoh32(pending[idx]);
+
+    return 0;
 }
 
-        
-static int
-afr_sh_mark_biggest_fool_as_source (afr_self_heal_t *sh,
-                                    afr_node_character *characters, 
-                                    int child_count)
+int
+afr_selfheal_fill_matrix(xlator_t *this, int **matrix, int subvol, int idx,
+                         dict_t *xdata)
 {
-        int i = 0;
-        int biggest = 0;
+    int i = 0;
+    void *pending_raw = NULL;
+    int pending[3] = {
+        0,
+    };
+    afr_private_t *priv = NULL;
 
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].type == AFR_NODE_FOOL) {
-                        biggest = i;
-                        break;
-                }
-        }
+    priv = this->private;
 
-        for (i = 0; i < child_count; i++) {
-                if (characters[i].type != AFR_NODE_FOOL)
-                        continue;
+    if (!matrix)
+        return 0;
 
-                if (!sh->buf)
-                        break;
+    for (i = 0; i < priv->child_count; i++) {
+        if (dict_get_ptr(xdata, priv->pending_key[i], &pending_raw))
+            continue;
 
-                if (SIZE_GREATER (&sh->buf[i], &sh->buf[biggest])) {
-                        biggest = i;
-                }
-        }
+        if (!pending_raw)
+            continue;
 
-        sh->sources[biggest] = 1;
+        memcpy(pending, pending_raw, sizeof(pending));
 
-        return 1;
-}
+        matrix[subvol][i] = ntoh32(pending[idx]);
+    }
 
+    return 0;
+}
 
-static int
-afr_sh_mark_biggest_as_source (afr_self_heal_t *sh, int child_count)
+int
+afr_selfheal_extract_xattr(xlator_t *this, struct afr_reply *replies,
+                           afr_transaction_type type, int *dirty, int **matrix)
 {
-        int biggest = 0;
-        int i;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    dict_t *xdata = NULL;
+    int idx = -1;
 
-        for (i = 0; i < child_count; i++) {
-                if (!sh->buf)
-                        break;
+    idx = afr_index_for_transaction_type(type);
 
-                if (SIZE_GREATER (&sh->buf[i], &sh->buf[biggest])) {
-                        biggest = i;
-                }
-        }
+    priv = this->private;
 
-        sh->sources[biggest] = 1;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret != 0)
+            continue;
 
-        return 1;
-}
+        if (!replies[i].xdata)
+            continue;
 
+        xdata = replies[i].xdata;
 
-static int
-afr_sh_mark_loweia_uid_as_source (afr_self_heal_t *sh, int child_count)
-{
-        uid_t smallest = 0;
-        int i;
+        afr_selfheal_fill_dirty(this, dirty, i, idx, xdata);
+        afr_selfheal_fill_matrix(this, matrix, i, idx, xdata);
+    }
 
-        for (i = 0; i < child_count; i++) {
-                if (!sh->buf)
-                        break;
+    return 0;
+}
 
-                if (sh->buf[i].ia_uid < sh->buf[smallest].ia_uid) {
-                        smallest = i;
-                }
+/*
+ * If by chance there are multiple sources with differing sizes, select
+ * the largest file as the source.
+ *
+ * This can happen if data was directly modified in the backend or for snapshots
+ */
+void
+afr_mark_largest_file_as_source(xlator_t *this, unsigned char *sources,
+                                struct afr_reply *replies)
+{
+    int i = 0;
+    afr_private_t *priv = NULL;
+    uint64_t size = 0;
+
+    /* Find source with biggest file size */
+    priv = this->private;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+        if (!replies[i].valid || replies[i].op_ret != 0) {
+            sources[i] = 0;
+            continue;
         }
-
-        sh->sources[smallest] = 1;
-
-        return 1;
+        if (size <= replies[i].poststat.ia_size) {
+            size = replies[i].poststat.ia_size;
+        }
+    }
+
+    /* Mark sources with less size as not source */
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+        if (size > replies[i].poststat.ia_size)
+            sources[i] = 0;
+    }
 }
 
-
-int
-afr_sh_mark_sources (afr_self_heal_t *sh, int child_count,
-                     afr_self_heal_type type)
+void
+afr_mark_latest_mtime_file_as_source(xlator_t *this, unsigned char *sources,
+                                     struct afr_reply *replies)
 {
-	int i = 0;
+    int i = 0;
+    afr_private_t *priv = NULL;
+    uint32_t mtime = 0;
+    uint32_t mtime_nsec = 0;
+
+    priv = this->private;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+        if (!replies[i].valid || replies[i].op_ret != 0) {
+            sources[i] = 0;
+            continue;
+        }
+        if ((mtime < replies[i].poststat.ia_mtime) ||
+            ((mtime == replies[i].poststat.ia_mtime) &&
+             (mtime_nsec < replies[i].poststat.ia_mtime_nsec))) {
+            mtime = replies[i].poststat.ia_mtime;
+            mtime_nsec = replies[i].poststat.ia_mtime_nsec;
+        }
+    }
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+        if ((mtime > replies[i].poststat.ia_mtime) ||
+            ((mtime == replies[i].poststat.ia_mtime) &&
+             (mtime_nsec > replies[i].poststat.ia_mtime_nsec))) {
+            sources[i] = 0;
+        }
+    }
+}
 
-        int32_t ** pending_matrix;
-        int *      sources;
+void
+afr_mark_active_sinks(xlator_t *this, unsigned char *sources,
+                      unsigned char *locked_on, unsigned char *sinks)
+{
+    int i = 0;
+    afr_private_t *priv = NULL;
 
-        int size_differs = 0;
+    priv = this->private;
 
-        pending_matrix = sh->pending_matrix;
-        sources        = sh->sources;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i] && locked_on[i])
+            sinks[i] = 1;
+        else
+            sinks[i] = 0;
+    }
+}
 
-	int nsources = 0;
+gf_boolean_t
+afr_dict_contains_heal_op(call_frame_t *frame)
+{
+    afr_local_t *local = NULL;
+    dict_t *xdata_req = NULL;
+    int ret = 0;
+    int heal_op = -1;
+
+    local = frame->local;
+    xdata_req = local->xdata_req;
+    ret = dict_get_int32_sizen(xdata_req, "heal-op", &heal_op);
+    if (ret)
+        return _gf_false;
+    if (local->xdata_rsp == NULL) {
+        local->xdata_rsp = dict_new();
+        if (!local->xdata_rsp)
+            return _gf_true;
+    }
+    ret = dict_set_sizen_str_sizen(local->xdata_rsp, "sh-fail-msg",
+                                   SFILE_NOT_IN_SPLIT_BRAIN);
+
+    return _gf_true;
+}
 
-        /* stores the 'characters' (innocent, fool, wise) of the nodes */
-        afr_node_character *
-                characters = GF_CALLOC (sizeof (afr_node_character), 
-                                        child_count,
-                                        gf_afr_mt_afr_node_character) ;
+gf_boolean_t
+afr_can_decide_split_brain_source_sinks(struct afr_reply *replies,
+                                        int child_count)
+{
+    int i = 0;
 
-	/* start clean */
-	for (i = 0; i < child_count; i++) {
-		sources[i] = 0;
-	}
-        
-        for (i = 0; i < child_count; i++) {
-                if (afr_sh_is_innocent (pending_matrix[i], child_count)) {
-                        characters[i].type = AFR_NODE_INNOCENT;
+    for (i = 0; i < child_count; i++)
+        if (replies[i].valid != 1 || replies[i].op_ret != 0)
+            return _gf_false;
 
-                } else if (afr_sh_is_fool (pending_matrix[i], i, child_count)) {
-                        characters[i].type = AFR_NODE_FOOL;
+    return _gf_true;
+}
 
-                } else if (afr_sh_is_wise (pending_matrix[i], i, child_count)) {
-                        characters[i].type = AFR_NODE_WISE;
+int
+afr_mark_split_brain_source_sinks_by_heal_op(
+    call_frame_t *frame, xlator_t *this, unsigned char *sources,
+    unsigned char *sinks, unsigned char *healed_sinks, unsigned char *locked_on,
+    struct afr_reply *replies, afr_transaction_type type, int heal_op)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    dict_t *xdata_req = NULL;
+    dict_t *xdata_rsp = NULL;
+    int ret = 0;
+    int i = 0;
+    char *name = NULL;
+    int source = -1;
+
+    local = frame->local;
+    priv = this->private;
+    xdata_req = local->xdata_req;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (locked_on[i])
+            if (sources[i] || !sinks[i] || !healed_sinks[i]) {
+                ret = -1;
+                goto out;
+            }
+    }
+    if (local->xdata_rsp == NULL) {
+        local->xdata_rsp = dict_new();
+        if (!local->xdata_rsp) {
+            ret = -1;
+            goto out;
+        }
+    }
+    xdata_rsp = local->xdata_rsp;
+
+    if (!afr_can_decide_split_brain_source_sinks(replies, priv->child_count)) {
+        ret = dict_set_sizen_str_sizen(xdata_rsp, "sh-fail-msg",
+                                       SBRAIN_HEAL_NO_GO_MSG);
+        ret = -1;
+        goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++)
+        if (locked_on[i])
+            sources[i] = 1;
+    switch (heal_op) {
+        case GF_SHD_OP_SBRAIN_HEAL_FROM_BIGGER_FILE:
+            if (type == AFR_METADATA_TRANSACTION) {
+                ret = dict_set_sizen_str_sizen(xdata_rsp, "sh-fail-msg",
+                                               SUSE_SOURCE_BRICK_TO_HEAL);
+                if (!ret)
+                    ret = -1;
+                goto out;
+            }
+            afr_mark_largest_file_as_source(this, sources, replies);
+            if (AFR_COUNT(sources, priv->child_count) != 1) {
+                ret = dict_set_sizen_str_sizen(xdata_rsp, "sh-fail-msg",
+                                               SNO_BIGGER_FILE);
+                if (!ret)
+                    ret = -1;
+                goto out;
+            }
+            break;
+        case GF_SHD_OP_SBRAIN_HEAL_FROM_LATEST_MTIME:
+            if (type == AFR_METADATA_TRANSACTION) {
+                ret = dict_set_sizen_str_sizen(xdata_rsp, "sh-fail-msg",
+                                               SUSE_SOURCE_BRICK_TO_HEAL);
+                if (!ret)
+                    ret = -1;
+                goto out;
+            }
+            afr_mark_latest_mtime_file_as_source(this, sources, replies);
+            if (AFR_COUNT(sources, priv->child_count) != 1) {
+                ret = dict_set_sizen_str_sizen(xdata_rsp, "sh-fail-msg",
+                                               SNO_DIFF_IN_MTIME);
+                if (!ret)
+                    ret = -1;
+                goto out;
+            }
+            break;
+        case GF_SHD_OP_SBRAIN_HEAL_FROM_BRICK:
+            ret = dict_get_str_sizen(xdata_req, "child-name", &name);
+            if (ret)
+                goto out;
+            source = afr_get_child_index_from_name(this, name);
+            if (source < 0) {
+                ret = dict_set_sizen_str_sizen(xdata_rsp, "sh-fail-msg",
+                                               SINVALID_BRICK_NAME);
+                if (!ret)
+                    ret = -1;
+                goto out;
+            }
+            if (locked_on[source] != 1) {
+                ret = dict_set_sizen_str_sizen(xdata_rsp, "sh-fail-msg",
+                                               SBRICK_IS_NOT_UP);
+                if (!ret)
+                    ret = -1;
+                goto out;
+            }
+            memset(sources, 0, sizeof(*sources) * priv->child_count);
+            sources[source] = 1;
+            break;
+        default:
+            ret = -1;
+            goto out;
+    }
+    for (i = 0; i < priv->child_count; i++) {
+        if (sources[i]) {
+            source = i;
+            break;
+        }
+    }
+    sinks[source] = 0;
+    healed_sinks[source] = 0;
+    ret = source;
+out:
+    if (ret < 0)
+        memset(sources, 0, sizeof(*sources) * priv->child_count);
+    return ret;
+}
 
-                } else {
-                        gf_log ("[module:replicate]", GF_LOG_ERROR,
-                                "Could not determine the state of subvolume %d!"
-                                " (This message should never appear."
-                                " Please file a bug report to "
-                                "<gluster-devel@nongnu.org>.)", i);
+int
+afr_sh_fav_by_majority(xlator_t *this, struct afr_reply *replies,
+                       inode_t *inode)
+{
+    afr_private_t *priv;
+    int vote_count = -1;
+    int fav_child = -1;
+    int i = 0;
+    int k = 0;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (replies[i].valid == 1) {
+            gf_msg_debug(this->name, 0,
+                         "Child:%s mtime_sec = %" PRId64 ", size = %" PRIu64
+                         " for gfid %s",
+                         priv->children[i]->name, replies[i].poststat.ia_mtime,
+                         replies[i].poststat.ia_size, uuid_utoa(inode->gfid));
+            vote_count = 0;
+            for (k = 0; k < priv->child_count; k++) {
+                if ((replies[k].poststat.ia_mtime ==
+                     replies[i].poststat.ia_mtime) &&
+                    (replies[k].poststat.ia_size ==
+                     replies[i].poststat.ia_size)) {
+                    vote_count++;
                 }
+            }
+            if (vote_count > priv->child_count / 2) {
+                fav_child = i;
+                break;
+            }
         }
+    }
+    return fav_child;
+}
 
-        if (type == AFR_SELF_HEAL_DATA) {
-                size_differs = afr_sh_mark_if_size_differs (sh, child_count);
+/*
+ * afr_sh_fav_by_mtime: Choose favorite child by mtime.
+ */
+int
+afr_sh_fav_by_mtime(xlator_t *this, struct afr_reply *replies, inode_t *inode)
+{
+    afr_private_t *priv;
+    int fav_child = -1;
+    int i = 0;
+    uint32_t cmp_mtime = 0;
+    uint32_t cmp_mtime_nsec = 0;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (replies[i].valid == 1) {
+            gf_msg_debug(this->name, 0,
+                         "Child:%s mtime = %" PRId64
+                         ", mtime_nsec = %d for "
+                         "gfid %s",
+                         priv->children[i]->name, replies[i].poststat.ia_mtime,
+                         replies[i].poststat.ia_mtime_nsec,
+                         uuid_utoa(inode->gfid));
+            if (replies[i].poststat.ia_mtime > cmp_mtime) {
+                cmp_mtime = replies[i].poststat.ia_mtime;
+                cmp_mtime_nsec = replies[i].poststat.ia_mtime_nsec;
+                fav_child = i;
+            } else if ((replies[i].poststat.ia_mtime == cmp_mtime) &&
+                       (replies[i].poststat.ia_mtime_nsec > cmp_mtime_nsec)) {
+                cmp_mtime = replies[i].poststat.ia_mtime;
+                cmp_mtime_nsec = replies[i].poststat.ia_mtime_nsec;
+                fav_child = i;
+            }
         }
+    }
+    return fav_child;
+}
 
-        if ((type == AFR_SELF_HEAL_METADATA)
-            && afr_sh_all_nodes_innocent (characters, child_count)) {
+/*
+ * afr_sh_fav_by_ctime: Choose favorite child by ctime.
+ */
+int
+afr_sh_fav_by_ctime(xlator_t *this, struct afr_reply *replies, inode_t *inode)
+{
+    afr_private_t *priv;
+    int fav_child = -1;
+    int i = 0;
+    uint32_t cmp_ctime = 0;
+    uint32_t cmp_ctime_nsec = 0;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (replies[i].valid == 1) {
+            gf_msg_debug(this->name, 0,
+                         "Child:%s ctime = %" PRId64
+                         ", ctime_nsec = %d for "
+                         "gfid %s",
+                         priv->children[i]->name, replies[i].poststat.ia_ctime,
+                         replies[i].poststat.ia_ctime_nsec,
+                         uuid_utoa(inode->gfid));
+            if (replies[i].poststat.ia_ctime > cmp_ctime) {
+                cmp_ctime = replies[i].poststat.ia_ctime;
+                cmp_ctime_nsec = replies[i].poststat.ia_ctime_nsec;
+                fav_child = i;
+            } else if ((replies[i].poststat.ia_ctime == cmp_ctime) &&
+                       (replies[i].poststat.ia_ctime_nsec > cmp_ctime_nsec)) {
+                cmp_ctime = replies[i].poststat.ia_ctime;
+                cmp_ctime_nsec = replies[i].poststat.ia_ctime_nsec;
+                fav_child = i;
+            }
+        }
+    }
+    return fav_child;
+}
 
-                nsources = afr_sh_mark_loweia_uid_as_source (sh, child_count);
-                goto out;
+/*
+ * afr_sh_fav_by_size: Choose favorite child by size
+ * when not all files are of zero size.
+ */
+int
+afr_sh_fav_by_size(xlator_t *this, struct afr_reply *replies, inode_t *inode)
+{
+    afr_private_t *priv;
+    int fav_child = -1;
+    int i = 0;
+    uint64_t cmp_sz = 0;
+
+    priv = this->private;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid) {
+            continue;
         }
+        gf_msg_debug(this->name, 0,
+                     "Child:%s file size = %" PRIu64 " for gfid %s",
+                     priv->children[i]->name, replies[i].poststat.ia_size,
+                     uuid_utoa(inode->gfid));
+        if (replies[i].poststat.ia_type == IA_IFDIR) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SBRAIN_FAV_CHILD_POLICY,
+                   "Cannot perform selfheal on %s. "
+                   "Size policy is not applicable to directories.",
+                   uuid_utoa(inode->gfid));
+            break;
+        }
+        if (replies[i].poststat.ia_size > cmp_sz) {
+            cmp_sz = replies[i].poststat.ia_size;
+            fav_child = i;
+        } else if (replies[i].poststat.ia_size == cmp_sz) {
+            fav_child = -1;
+        }
+    }
+    if (fav_child == -1) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+               "No bigger file");
+    }
+    return fav_child;
+}
 
-        if (afr_sh_all_nodes_innocent (characters, child_count)) {
-                if (size_differs) {
-                        nsources = afr_sh_mark_biggest_as_source (sh,
-                                                                  child_count);
-                }
+int
+afr_sh_get_fav_by_policy(xlator_t *this, struct afr_reply *replies,
+                         inode_t *inode, char **policy_str)
+{
+    afr_private_t *priv = NULL;
+    int fav_child = -1;
+
+    priv = this->private;
+    if (!afr_can_decide_split_brain_source_sinks(replies, priv->child_count)) {
+        return -1;
+    }
+
+    switch (priv->fav_child_policy) {
+        case AFR_FAV_CHILD_BY_SIZE:
+            fav_child = afr_sh_fav_by_size(this, replies, inode);
+            if (policy_str && fav_child >= 0) {
+                *policy_str = "SIZE";
+            }
+            break;
+        case AFR_FAV_CHILD_BY_CTIME:
+            fav_child = afr_sh_fav_by_ctime(this, replies, inode);
+            if (policy_str && fav_child >= 0) {
+                *policy_str = "CTIME";
+            }
+            break;
+        case AFR_FAV_CHILD_BY_MTIME:
+            fav_child = afr_sh_fav_by_mtime(this, replies, inode);
+            if (policy_str && fav_child >= 0) {
+                *policy_str = "MTIME";
+            }
+            break;
+        case AFR_FAV_CHILD_BY_MAJORITY:
+            fav_child = afr_sh_fav_by_majority(this, replies, inode);
+            if (policy_str && fav_child >= 0) {
+                *policy_str = "MAJORITY";
+            }
+            break;
+        case AFR_FAV_CHILD_NONE:
+        default:
+            break;
+    }
+
+    return fav_child;
+}
 
-        } else if (afr_sh_wise_nodes_exist (characters, child_count)) {
-                afr_sh_compute_wisdom (pending_matrix, characters, child_count);
+int
+afr_mark_split_brain_source_sinks_by_policy(
+    call_frame_t *frame, xlator_t *this, inode_t *inode, unsigned char *sources,
+    unsigned char *sinks, unsigned char *healed_sinks, unsigned char *locked_on,
+    struct afr_reply *replies, afr_transaction_type type)
+{
+    afr_private_t *priv = NULL;
+    int fav_child = -1;
+    char mtime_str[256];
+    char ctime_str[256];
+    char *policy_str = NULL;
+    struct tm *tm_ptr;
+    time_t time;
+
+    priv = this->private;
+
+    fav_child = afr_sh_get_fav_by_policy(this, replies, inode, &policy_str);
+    if (fav_child == -1) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SBRAIN_FAV_CHILD_POLICY,
+               "No child selected by favorite-child policy.");
+    } else if (fav_child > priv->child_count - 1) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SBRAIN_FAV_CHILD_POLICY,
+               "Invalid child (%d) "
+               "selected by policy %s.",
+               fav_child, policy_str);
+    } else if (fav_child >= 0) {
+        time = replies[fav_child].poststat.ia_mtime;
+        tm_ptr = localtime(&time);
+        strftime(mtime_str, sizeof(mtime_str), "%Y-%m-%d %H:%M:%S", tm_ptr);
+        time = replies[fav_child].poststat.ia_ctime;
+        tm_ptr = localtime(&time);
+        strftime(ctime_str, sizeof(ctime_str), "%Y-%m-%d %H:%M:%S", tm_ptr);
+
+        gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_SBRAIN_FAV_CHILD_POLICY,
+               "Source %s selected as authentic to resolve conflicting data "
+               "in file (gfid:%s) by %s (%" PRIu64
+               " bytes @ %s mtime, %s "
+               "ctime).",
+               priv->children[fav_child]->name, uuid_utoa(inode->gfid),
+               policy_str, replies[fav_child].poststat.ia_size, mtime_str,
+               ctime_str);
+
+        sources[fav_child] = 1;
+        sinks[fav_child] = 0;
+        healed_sinks[fav_child] = 0;
+    }
+    return fav_child;
+}
 
-                if (afr_sh_wise_nodes_conflict (characters, child_count)) {
-                        /* split-brain */
+gf_boolean_t
+afr_is_file_empty_on_all_children(afr_private_t *priv,
+                                  struct afr_reply *replies)
+{
+    int i = 0;
 
-                        nsources = -1;
-                        goto out;
+    for (i = 0; i < priv->child_count; i++) {
+        if ((!replies[i].valid) || (replies[i].op_ret != 0) ||
+            (replies[i].poststat.ia_size != 0))
+            return _gf_false;
+    }
 
-                } else {
-                        nsources = afr_sh_mark_wisest_as_sources (sources, 
-                                                                  characters,
-                                                                  child_count);
-                }
-        } else {
-                nsources = afr_sh_mark_biggest_fool_as_source (sh, characters,
-                                                               child_count);
-        }
+    return _gf_true;
+}
 
-out:
-        GF_FREE (characters);
+int
+afr_mark_source_sinks_if_file_empty(xlator_t *this, unsigned char *sources,
+                                    unsigned char *sinks,
+                                    unsigned char *healed_sinks,
+                                    unsigned char *locked_on,
+                                    struct afr_reply *replies,
+                                    afr_transaction_type type)
+{
+    int source = -1;
+    int i = 0;
+    afr_private_t *priv = this->private;
+    struct iatt stbuf = {
+        0,
+    };
+
+    if ((AFR_COUNT(locked_on, priv->child_count) < priv->child_count) ||
+        (afr_success_count(replies, priv->child_count) < priv->child_count))
+        return -1;
+
+    if (type == AFR_DATA_TRANSACTION) {
+        if (!afr_is_file_empty_on_all_children(priv, replies))
+            return -1;
+        goto mark;
+    }
+
+    /*For AFR_METADATA_TRANSACTION, metadata must be same on all bricks.*/
+    stbuf = replies[0].poststat;
+    for (i = 1; i < priv->child_count; i++) {
+        if ((!IA_EQUAL(stbuf, replies[i].poststat, type)) ||
+            (!IA_EQUAL(stbuf, replies[i].poststat, uid)) ||
+            (!IA_EQUAL(stbuf, replies[i].poststat, gid)) ||
+            (!IA_EQUAL(stbuf, replies[i].poststat, prot)))
+            return -1;
+    }
+    for (i = 1; i < priv->child_count; i++) {
+        if (!afr_xattrs_are_equal(replies[0].xdata, replies[i].xdata))
+            return -1;
+    }
+
+mark:
+    /* data/metadata is same on all bricks. Pick one of them as source. Rest
+     * are sinks.*/
+    for (i = 0; i < priv->child_count; i++) {
+        if (source == -1) {
+            source = i;
+            sources[i] = 1;
+            sinks[i] = 0;
+            healed_sinks[i] = 0;
+            continue;
+        }
+        sources[i] = 0;
+        sinks[i] = 1;
+        healed_sinks[i] = 1;
+    }
 
-	return nsources;
+    return source;
 }
 
+/* Return a source depending on the type of heal_op, and set sources[source],
+ * sinks[source] and healed_sinks[source] to 1, 0 and 0 respectively. Do so
+ * only if the following condition is met:
+ * ∀i((i ∈ locked_on[] ∧ i=1)==>(sources[i]=0 ∧ sinks[i]=1 ∧ healed_sinks[i]=1))
+ * i.e. for each locked node, sources[node] is 0; healed_sinks[node] and
+ * sinks[node] are 1. This should be the case if the file is in split-brain.
+ */
+int
+afr_mark_split_brain_source_sinks(
+    call_frame_t *frame, xlator_t *this, inode_t *inode, unsigned char *sources,
+    unsigned char *sinks, unsigned char *healed_sinks, unsigned char *locked_on,
+    struct afr_reply *replies, afr_transaction_type type)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    dict_t *xdata_req = NULL;
+    int heal_op = -1;
+    int ret = -1;
+    int source = -1;
+
+    local = frame->local;
+    priv = this->private;
+    xdata_req = local->xdata_req;
+
+    source = afr_mark_source_sinks_if_file_empty(
+        this, sources, sinks, healed_sinks, locked_on, replies, type);
+    if (source >= 0)
+        return source;
+
+    ret = dict_get_int32_sizen(xdata_req, "heal-op", &heal_op);
+    if (ret)
+        goto autoheal;
+
+    source = afr_mark_split_brain_source_sinks_by_heal_op(
+        frame, this, sources, sinks, healed_sinks, locked_on, replies, type,
+        heal_op);
+    return source;
+
+autoheal:
+    /* Automatically heal if fav_child_policy is set. */
+    if (priv->fav_child_policy != AFR_FAV_CHILD_NONE) {
+        source = afr_mark_split_brain_source_sinks_by_policy(
+            frame, this, inode, sources, sinks, healed_sinks, locked_on,
+            replies, type);
+        if (source != -1) {
+            ret = dict_set_int32_sizen(xdata_req, "fav-child-policy", 1);
+            if (ret)
+                return -1;
+        }
+    }
 
-void
-afr_sh_pending_to_delta (afr_private_t *priv, dict_t **xattr,
-                         int32_t *delta_matrix[], int success[],
-                         int child_count, afr_transaction_type type)
-{
-	int i = 0;
-	int j = 0;
-        int k = 0;
-
-	/* Indexable by result of afr_index_for_transaction_type(): 0 -- 2. */
-	int32_t   pending[3];
-	void    * pending_raw = NULL;
-        int       ret         = 0;
-
-	/* start clean */
-	for (i = 0; i < child_count; i++) {
-		for (j = 0; j < child_count; j++) {
-			delta_matrix[i][j] = 0;
-		}
-	}
-
-	for (i = 0; i < child_count; i++) {
-                pending_raw = NULL;
-
-                for (j = 0; j < child_count; j++) {
-                        ret = dict_get_ptr (xattr[i], priv->pending_key[j],
-                                            &pending_raw);
-                        if (ret < 0)
-                                gf_log ("afr_sh_pending_to_delta",
-                                        GF_LOG_WARNING,
-                                        "Unable to get dict value.");
-
-                        if (!success[j])
-                                continue;
-
-                        k = afr_index_for_transaction_type (type);
-
-                        if (pending_raw) {
-				memcpy (pending, pending_raw, sizeof(pending));
-                                delta_matrix[i][j] = -(ntoh32 (pending[k]));
-                        } else {
-                                delta_matrix[i][j]  = 0;
-                        }
-
-                }
-	}
+    return source;
 }
 
-
 int
-afr_sh_delta_to_xattr (afr_private_t *priv,
-                       int32_t *delta_matrix[], dict_t *xattr[],
-		       int child_count, afr_transaction_type type)
+_afr_fav_child_reset_sink_xattrs(call_frame_t *frame, xlator_t *this,
+                                 inode_t *inode, int source,
+                                 unsigned char *healed_sinks,
+                                 unsigned char *undid_pending,
+                                 afr_transaction_type type,
+                                 unsigned char *locked_on,
+                                 struct afr_reply *replies)
 {
-	int i = 0;
-	int j = 0;
-        int k = 0;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int *input_dirty = NULL;
+    int **input_matrix = NULL;
+    int *output_dirty = NULL;
+    int **output_matrix = NULL;
+    dict_t *xattr = NULL;
+    dict_t *xdata = NULL;
+    int i = 0;
+
+    priv = this->private;
+    local = frame->local;
+
+    if (!dict_get_sizen(local->xdata_req, "fav-child-policy"))
+        return 0;
 
-	int ret = 0;
+    xdata = dict_new();
+    if (!xdata)
+        return -1;
 
-	int32_t *pending = 0;
+    input_dirty = alloca0(priv->child_count * sizeof(int));
+    input_matrix = ALLOC_MATRIX(priv->child_count, int);
+    output_dirty = alloca0(priv->child_count * sizeof(int));
+    output_matrix = ALLOC_MATRIX(priv->child_count, int);
 
-	for (i = 0; i < child_count; i++) {
-		if (!xattr[i])
-			continue;
+    afr_selfheal_extract_xattr(this, replies, type, input_dirty, input_matrix);
 
-		for (j = 0; j < child_count; j++) {
-                        pending = GF_CALLOC (sizeof (int32_t), 3,
-                                             gf_afr_mt_int32_t);
-                        /* 3 = data+metadata+entry */
+    for (i = 0; i < priv->child_count; i++) {
+        if (i == source || !healed_sinks[i])
+            continue;
+        output_dirty[i] = -input_dirty[i];
+        output_matrix[i][source] = -input_matrix[i][source];
+    }
 
-                        k = afr_index_for_transaction_type (type);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!healed_sinks[i] || !locked_on[i])
+            continue;
+        xattr = afr_selfheal_output_xattr(this, _gf_false, type, output_dirty,
+                                          output_matrix, i, NULL);
 
-			pending[k] = hton32 (delta_matrix[i][j]);
+        afr_selfheal_post_op(frame, this, inode, i, xattr, xdata);
 
-                        ret = dict_set_bin (xattr[i], priv->pending_key[j],
-                                            pending,
-                                            3 * sizeof (int32_t));
-                        if (ret < 0)
-                                gf_log ("afr_sh_delta_to_xattr",
-                                        GF_LOG_WARNING,
-                                        "Unable to set dict value.");
-		}
-	}
-	return 0;
-}
+        undid_pending[i] = 1;
+        dict_unref(xattr);
+    }
 
+    if (xdata)
+        dict_unref(xdata);
 
-int
-afr_sh_has_metadata_pending (dict_t *xattr, int child_count, xlator_t *this)
+    return 0;
+}
+
+gf_boolean_t
+afr_does_witness_exist(xlator_t *this, uint64_t *witness)
 {
-	afr_private_t *priv = NULL;
-	/* Indexable by result of afr_index_for_transaction_type(): 0 -- 2. */
-	int32_t       pending[3];
-	void          *pending_raw = NULL;
+    int i = 0;
+    afr_private_t *priv = NULL;
 
-	int           ret = -1;
-	int            i  = 0;
-        int            j  = 0;
+    priv = this->private;
 
-	priv = this->private;
+    for (i = 0; i < priv->child_count; i++) {
+        if (witness[i])
+            return _gf_true;
+    }
+    return _gf_false;
+}
 
-        for (i = 0; i < priv->child_count; i++) {
-                ret = dict_get_ptr (xattr, priv->pending_key[i],
-                                    &pending_raw);
+unsigned int
+afr_get_quorum_count(afr_private_t *priv)
+{
+    if (priv->quorum_count == AFR_QUORUM_AUTO) {
+        return priv->child_count / 2 + 1;
+    } else {
+        return priv->quorum_count;
+    }
+}
 
-                if (ret != 0)
-                        return 0;
+void
+afr_selfheal_post_op_failure_accounting(afr_private_t *priv, char *accused,
+                                        unsigned char *sources,
+                                        unsigned char *locked_on)
+{
+    int i = 0;
+    unsigned int quorum_count = 0;
 
-		memcpy (pending, pending_raw, sizeof(pending));
-                j = afr_index_for_transaction_type (AFR_METADATA_TRANSACTION);
+    if (AFR_COUNT(sources, priv->child_count) != 0)
+        return;
 
-                if (pending[j])
-                        return 1;
+    quorum_count = afr_get_quorum_count(priv);
+    for (i = 0; i < priv->child_count; i++) {
+        if ((accused[i] < quorum_count) && locked_on[i]) {
+            sources[i] = 1;
         }
-
-	return 0;
+    }
+    return;
 }
 
+/*
+ * This function determines if a self-heal is required for a given inode,
+ * and if needed, in what direction.
+ *
+ * locked_on[] is the array representing servers which have been locked and
+ * from which xattrs have been fetched for analysis.
+ *
+ * The output of the function is by filling the arrays sources[] and sinks[].
+ *
+ * sources[i] is set if i'th server is an eligible source for a selfheal.
+ *
+ * sinks[i] is set if i'th server needs to be healed.
+ *
+ * if sources[0..N] are all set, there is no need for a selfheal.
+ *
+ * if sinks[0..N] are all set, the inode is in split brain.
+ *
+ */
 
 int
-afr_sh_has_data_pending (dict_t *xattr, int child_count, xlator_t *this)
+afr_selfheal_find_direction(call_frame_t *frame, xlator_t *this,
+                            struct afr_reply *replies,
+                            afr_transaction_type type, unsigned char *locked_on,
+                            unsigned char *sources, unsigned char *sinks,
+                            uint64_t *witness, unsigned char *pflag)
 {
-	afr_private_t *priv = NULL;
-	/* Indexable by result of afr_index_for_transaction_type(): 0 -- 2. */
-	int32_t       pending[3];
-	void          *pending_raw = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int j = 0;
+    int *dirty = NULL;         /* Denotes if dirty xattr is set */
+    int **matrix = NULL;       /* Changelog matrix */
+    char *accused = NULL;      /* Accused others without any self-accusal */
+    char *pending = NULL;      /* Have pending operations on others */
+    char *self_accused = NULL; /* Accused itself */
+
+    priv = this->private;
+
+    dirty = alloca0(priv->child_count * sizeof(int));
+    accused = alloca0(priv->child_count);
+    pending = alloca0(priv->child_count);
+    self_accused = alloca0(priv->child_count);
+    matrix = ALLOC_MATRIX(priv->child_count, int);
+    memset(witness, 0, sizeof(*witness) * priv->child_count);
+
+    /* First construct the pending matrix for further analysis */
+    afr_selfheal_extract_xattr(this, replies, type, dirty, matrix);
+
+    if (pflag) {
+        for (i = 0; i < priv->child_count; i++) {
+            for (j = 0; j < priv->child_count; j++)
+                if (matrix[i][j])
+                    *pflag |= PFLAG_PENDING;
+            if (*pflag)
+                break;
+        }
+    }
+
+    if (afr_success_count(replies, priv->child_count) < priv->child_count) {
+        /* Treat this just like locks not being acquired */
+        return -ENOTCONN;
+    }
+
+    /* short list all self-accused */
+    for (i = 0; i < priv->child_count; i++) {
+        if (matrix[i][i])
+            self_accused[i] = 1;
+    }
+
+    /* Next short list all accused to exclude them from being sources */
+    /* Self-accused can't accuse others as they are FOOLs */
+    for (i = 0; i < priv->child_count; i++) {
+        for (j = 0; j < priv->child_count; j++) {
+            if (matrix[i][j]) {
+                if (!self_accused[i])
+                    accused[j] += 1;
+                if (i != j)
+                    pending[i] += 1;
+            }
+        }
+    }
 
-	int           ret = -1;
-	int            i  = 0;
-        int            j  = 0;
+    /* Short list all non-accused as sources */
+    for (i = 0; i < priv->child_count; i++) {
+        if (!accused[i] && locked_on[i])
+            sources[i] = 1;
+        else
+            sources[i] = 0;
+    }
+
+    /* Everyone accused by non-self-accused sources are sinks */
+    memset(sinks, 0, priv->child_count);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+        if (self_accused[i])
+            continue;
+        for (j = 0; j < priv->child_count; j++) {
+            if (matrix[i][j])
+                sinks[j] = 1;
+        }
+    }
+
+    /* For breaking ties provide with number of fops they witnessed */
+
+    /*
+     * count the pending fops witnessed from itself to others when it is
+     * self-accused
+     */
+    for (i = 0; i < priv->child_count; i++) {
+        if (!self_accused[i])
+            continue;
+        for (j = 0; j < priv->child_count; j++) {
+            if (i == j)
+                continue;
+            witness[i] += matrix[i][j];
+        }
+    }
 
-	priv = this->private;
+    if (type == AFR_DATA_TRANSACTION || type == AFR_METADATA_TRANSACTION)
+        afr_selfheal_post_op_failure_accounting(priv, accused, sources,
+                                                locked_on);
 
+    /* If no sources, all locked nodes are sinks - split brain */
+    if (AFR_COUNT(sources, priv->child_count) == 0) {
         for (i = 0; i < priv->child_count; i++) {
-                ret = dict_get_ptr (xattr, priv->pending_key[i],
-                                    &pending_raw);
-
-                if (ret != 0)
-                        return 0;
-                
-		memcpy (pending, pending_raw, sizeof(pending));
-                j = afr_index_for_transaction_type (AFR_DATA_TRANSACTION);
-
-                if (pending[j])
-                        return 1;
+            if (locked_on[i])
+                sinks[i] = 1;
         }
+        if (pflag)
+            *pflag |= PFLAG_SBRAIN;
+    }
+
+    /* One more class of witness similar to dirty in v2 is where no pending
+     * exists but we have self-accusing markers. This can happen in afr-v1
+     * if the brick crashes just after doing xattrop on self but
+     * before xattrop on the other xattrs on the brick in pre-op. */
+    if (AFR_COUNT(pending, priv->child_count) == 0) {
+        for (i = 0; i < priv->child_count; i++) {
+            if (self_accused[i])
+                witness[i] += matrix[i][i];
+        }
+    } else {
+        /* In afr-v1 if a file is self-accused and has pending
+         * operations on others then it is similar to 'dirty' in afr-v2.
+         * Consider such cases as witness.
+         */
+        for (i = 0; i < priv->child_count; i++) {
+            if (self_accused[i] && pending[i])
+                witness[i] += matrix[i][i];
+        }
+    }
+
+    /* count the number of dirty fops witnessed */
+    for (i = 0; i < priv->child_count; i++)
+        witness[i] += dirty[i];
 
-	return 0;
+    return 0;
 }
 
+void
+afr_log_selfheal(uuid_t gfid, xlator_t *this, int ret, char *type, int source,
+                 unsigned char *sources, unsigned char *healed_sinks)
+{
+    char *status = NULL;
+    char *sinks_str = NULL;
+    char *p = NULL;
+    char *sources_str = NULL;
+    char *q = NULL;
+    afr_private_t *priv = NULL;
+    gf_loglevel_t loglevel = GF_LOG_NONE;
+    int i = 0;
+
+    priv = this->private;
+    sinks_str = alloca0(priv->child_count * 8);
+    p = sinks_str;
+    sources_str = alloca0(priv->child_count * 8);
+    q = sources_str;
+    for (i = 0; i < priv->child_count; i++) {
+        if (healed_sinks[i])
+            p += sprintf(p, "%d ", i);
+        if (sources[i]) {
+            if (source == i) {
+                q += sprintf(q, "[%d] ", i);
+            } else {
+                q += sprintf(q, "%d ", i);
+            }
+        }
+    }
+
+    if (ret < 0) {
+        status = "Failed";
+        loglevel = GF_LOG_DEBUG;
+    } else {
+        status = "Completed";
+        loglevel = GF_LOG_INFO;
+    }
+
+    gf_msg(this->name, loglevel, 0, AFR_MSG_SELF_HEAL_INFO,
+           "%s %s selfheal on %s. "
+           "sources=%s sinks=%s",
+           status, type, uuid_utoa(gfid), sources_str, sinks_str);
+}
 
 int
-afr_sh_has_entry_pending (dict_t *xattr, int child_count, xlator_t *this)
+afr_selfheal_discover_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int op_ret, int op_errno, inode_t *inode,
+                          struct iatt *buf, dict_t *xdata, struct iatt *parbuf)
 {
-        afr_private_t *priv = NULL;
-	/* Indexable by result of afr_index_for_transaction_type(): 0 -- 2. */
-	int32_t       pending[3];
-	void          *pending_raw = NULL;
+    afr_local_t *local = NULL;
+    int i = -1;
+    GF_UNUSED int ret = -1;
+    int8_t need_heal = 1;
+
+    local = frame->local;
+    i = (long)cookie;
+
+    local->replies[i].valid = 1;
+    local->replies[i].op_ret = op_ret;
+    local->replies[i].op_errno = op_errno;
+    if (buf)
+        local->replies[i].poststat = *buf;
+    if (parbuf)
+        local->replies[i].postparent = *parbuf;
+    if (xdata) {
+        local->replies[i].xdata = dict_ref(xdata);
+        ret = dict_get_int8(xdata, "link-count", &need_heal);
+    }
+
+    local->replies[i].need_heal = need_heal;
+    syncbarrier_wake(&local->barrier);
+
+    return 0;
+}
 
-	int           ret = -1;
-	int            i  = 0;
-        int            j  = 0;
+inode_t *
+afr_selfheal_unlocked_lookup_on(call_frame_t *frame, inode_t *parent,
+                                const char *name, struct afr_reply *replies,
+                                unsigned char *lookup_on, dict_t *xattr)
+{
+    loc_t loc = {
+        0,
+    };
+    dict_t *xattr_req = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    inode_t *inode = NULL;
 
-	priv = this->private;
+    local = frame->local;
+    priv = frame->this->private;
 
-        for (i = 0; i < priv->child_count; i++) {
-                ret = dict_get_ptr (xattr, priv->pending_key[i],
-                                    &pending_raw);
+    xattr_req = dict_new();
+    if (!xattr_req)
+        return NULL;
 
-                if (ret != 0)
-                        return 0;
-                
-		memcpy (pending, pending_raw, sizeof(pending));
-                j = afr_index_for_transaction_type (AFR_ENTRY_TRANSACTION);
+    if (xattr)
+        dict_copy(xattr, xattr_req);
 
-                if (pending[j])
-                        return 1;
-        }
+    if (afr_xattr_req_prepare(frame->this, xattr_req) != 0) {
+        dict_unref(xattr_req);
+        return NULL;
+    }
 
-	return 0;
-}
+    inode = inode_new(parent->table);
+    if (!inode) {
+        dict_unref(xattr_req);
+        return NULL;
+    }
 
+    loc.parent = inode_ref(parent);
+    gf_uuid_copy(loc.pargfid, parent->gfid);
+    loc.name = name;
+    loc.inode = inode_ref(inode);
 
-/**
- * is_matrix_zero - return true if pending matrix is all zeroes
- */
+    AFR_ONLIST(lookup_on, frame, afr_selfheal_discover_cbk, lookup, &loc,
+               xattr_req);
 
-int
-afr_sh_is_matrix_zero (int32_t *pending_matrix[], int child_count)
-{
-	int i, j;
+    afr_replies_copy(replies, local->replies, priv->child_count);
+
+    loc_wipe(&loc);
+    dict_unref(xattr_req);
 
-	for (i = 0; i < child_count; i++) 
-		for (j = 0; j < child_count; j++) 
-			if (pending_matrix[i][j]) 
-				return 0;
-	return 1;
+    return inode;
 }
 
+static int
+afr_set_multi_dom_lock_count_request(xlator_t *this, dict_t *dict)
+{
+    int ret = 0;
+    afr_private_t *priv = NULL;
+    char *key1 = NULL;
+    char *key2 = NULL;
+
+    priv = this->private;
+    key1 = alloca0(strlen(GLUSTERFS_INODELK_DOM_PREFIX) + 2 +
+                   strlen(this->name));
+    key2 = alloca0(strlen(GLUSTERFS_INODELK_DOM_PREFIX) + 2 +
+                   strlen(priv->sh_domain));
+
+    ret = dict_set_uint32(dict, GLUSTERFS_MULTIPLE_DOM_LK_CNT_REQUESTS, 1);
+    if (ret)
+        return ret;
+
+    sprintf(key1, "%s:%s", GLUSTERFS_INODELK_DOM_PREFIX, this->name);
+    ret = dict_set_uint32(dict, key1, 1);
+    if (ret)
+        return ret;
+
+    sprintf(key2, "%s:%s", GLUSTERFS_INODELK_DOM_PREFIX, priv->sh_domain);
+    ret = dict_set_uint32(dict, key2, 1);
+    if (ret)
+        return ret;
+
+    return 0;
+}
 
 int
-afr_sh_missing_entries_done (call_frame_t *frame, xlator_t *this)
+afr_selfheal_unlocked_discover_on(call_frame_t *frame, inode_t *inode,
+                                  uuid_t gfid, struct afr_reply *replies,
+                                  unsigned char *discover_on, dict_t *dict)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              i = 0;
+    loc_t loc = {
+        0,
+    };
+    dict_t *xattr_req = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
+    local = frame->local;
+    priv = frame->this->private;
 
-//	memset (sh->child_errno, 0, sizeof (int) * priv->child_count);
-	memset (sh->buf, 0, sizeof (struct iatt) * priv->child_count);
+    xattr_req = dict_new();
+    if (!xattr_req)
+        return -ENOMEM;
+    if (dict)
+        dict_copy(dict, xattr_req);
 
-        for (i = 0; i < priv->child_count; i++) {
-                sh->locked_nodes[i] = 0;
-        }
+    if (afr_xattr_req_prepare(frame->this, xattr_req) != 0) {
+        dict_unref(xattr_req);
+        return -ENOMEM;
+    }
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->xattr[i])
-			dict_unref (sh->xattr[i]);
-		sh->xattr[i] = NULL;
-	}
+    if (afr_set_multi_dom_lock_count_request(frame->this, xattr_req)) {
+        dict_unref(xattr_req);
+        return -1;
+    }
 
-	if (local->govinda_gOvinda) {
-		gf_log (this->name, GF_LOG_TRACE,
-			"aborting selfheal of %s",
-			local->loc.path);
-		sh->completion_cbk (frame, this);
-	} else {
-		gf_log (this->name, GF_LOG_TRACE,
-			"proceeding to metadata check on %s",
-			local->loc.path);
-		afr_self_heal_metadata (frame, this);
-	}
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, gfid);
 
-	return 0;
-}
+    AFR_ONLIST(discover_on, frame, afr_selfheal_discover_cbk, lookup, &loc,
+               xattr_req);
 
+    afr_replies_copy(replies, local->replies, priv->child_count);
 
-static int
-sh_missing_entries_finish (call_frame_t *frame, xlator_t *this)
+    loc_wipe(&loc);
+    dict_unref(xattr_req);
+
+    return 0;
+}
+
+int
+afr_selfheal_unlocked_discover(call_frame_t *frame, inode_t *inode, uuid_t gfid,
+                               struct afr_reply *replies)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
+    afr_local_t *local = NULL;
+    dict_t *dict = NULL;
 
-	local = frame->local;
-        int_lock = &local->internal_lock;
+    local = frame->local;
 
-        int_lock->lock_cbk = afr_sh_missing_entries_done;
-        afr_unlock (frame, this);
+    if (local->xattr_req)
+        dict = local->xattr_req;
 
-	return 0;
+    return afr_selfheal_unlocked_discover_on(frame, inode, gfid, replies,
+                                             local->child_up, dict);
 }
 
-
-static int
-sh_destroy_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		int32_t op_ret, int op_errno,
-                struct iatt *preop, struct iatt *postop)
+unsigned int
+afr_success_count(struct afr_reply *replies, unsigned int count)
 {
-        afr_local_t *local = NULL;
+    int i = 0;
+    unsigned int success = 0;
 
-        loc_t *parent_loc = cookie;
+    for (i = 0; i < count; i++)
+        if (replies[i].valid && replies[i].op_ret == 0)
+            success++;
+    return success;
+}
 
-        int call_count = 0;
+int
+afr_selfheal_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, dict_t *xdata)
+{
+    afr_local_t *local = NULL;
+    int i = 0;
 
-        local = frame->local;
+    local = frame->local;
+    i = (long)cookie;
 
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "setattr on %s failed: %s",
-                        local->loc.path, strerror (op_errno));
-        }
+    local->replies[i].valid = 1;
+    local->replies[i].op_ret = op_ret;
+    local->replies[i].op_errno = op_errno;
 
-        if (parent_loc) {
-                loc_wipe (parent_loc);
-                GF_FREE (parent_loc);
-        }
+    syncbarrier_wake(&local->barrier);
 
-        call_count = afr_frame_return (frame);
-        
-        if (call_count == 0) {
-                STACK_DESTROY (frame->root);
-        }
-        
-	return 0;
+    return 0;
 }
 
+int
+afr_locked_fill(call_frame_t *frame, xlator_t *this, unsigned char *locked_on)
+{
+    int i = 0;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int count = 0;
+
+    local = frame->local;
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->replies[i].valid && local->replies[i].op_ret == 0) {
+            locked_on[i] = 1;
+            count++;
+        } else {
+            locked_on[i] = 0;
+        }
+    }
 
-static int
-sh_missing_entries_newentry_cbk (call_frame_t *frame, void *cookie,
-				 xlator_t *this,
-				 int32_t op_ret, int32_t op_errno,
-				 inode_t *inode, struct iatt *buf,
-                                 struct iatt *preparent,
-                                 struct iatt *postparent)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	call_frame_t    *setattr_frame = NULL;
-	int              call_count = 0;
-	int              child_index = 0;
-
-        loc_t *parent_loc = NULL;
-
-	struct iatt     stbuf;
-        int32_t valid;
-
-	local = frame->local;
-	sh    = &local->self_heal;
-	priv  = this->private;
-
-	child_index = (long) cookie;
-
-	stbuf.ia_atime = sh->buf[sh->source].ia_atime;
-	stbuf.ia_atime_nsec = sh->buf[sh->source].ia_atime_nsec;
-	stbuf.ia_mtime = sh->buf[sh->source].ia_mtime;
-	stbuf.ia_mtime_nsec = sh->buf[sh->source].ia_mtime_nsec;
-
-        stbuf.ia_uid = sh->buf[sh->source].ia_uid;
-        stbuf.ia_gid = sh->buf[sh->source].ia_gid;
-        
-        valid = GF_SET_ATTR_UID   | GF_SET_ATTR_GID |
-                GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME;
-        
-	if (op_ret == 0) {
-		setattr_frame = copy_frame (frame);
-                
-                setattr_frame->local = GF_CALLOC (1, sizeof (afr_local_t),
-                                                  gf_afr_mt_afr_local_t);
-
-                ((afr_local_t *)setattr_frame->local)->call_count = 2;
-
-		gf_log (this->name, GF_LOG_TRACE,
-			"setattr (%s) on subvolume %s",
-			local->loc.path, priv->children[child_index]->name);
-
-		STACK_WIND_COOKIE (setattr_frame, sh_destroy_cbk,
-                                   (void *) (long) 0,
-                                   priv->children[child_index],
-                                   priv->children[child_index]->fops->setattr,
-                                   &local->loc, &stbuf, valid);
-
-                valid      = GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME;
-                parent_loc = GF_CALLOC (1, sizeof (*parent_loc), 
-                                        gf_afr_mt_loc_t);
-                afr_build_parent_loc (parent_loc, &local->loc);
-
-                STACK_WIND_COOKIE (setattr_frame, sh_destroy_cbk,
-                                   (void *) (long) parent_loc,
-                                   priv->children[child_index],
-                                   priv->children[child_index]->fops->setattr,
-                                   parent_loc, &sh->parentbuf, valid);
-	}
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		sh_missing_entries_finish (frame, this);
-	}
-
-	return 0;
+    return count;
 }
 
+int
+afr_selfheal_tryinodelk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                        char *dom, off_t off, size_t size,
+                        unsigned char *locked_on)
+{
+    loc_t loc = {
+        0,
+    };
+    struct gf_flock flock = {
+        0,
+    };
 
-static int
-sh_missing_entries_mknod (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              i = 0;
-        int              ret = 0;
-	int              enoent_count = 0;
-	int              call_count = 0;
-	mode_t           st_mode = 0;
-	dev_t            ia_dev = 0;
-        dict_t          *dict = NULL;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	for (i = 0; i < priv->child_count; i++)
-		if (sh->child_errno[i] == ENOENT)
-			enoent_count++;
-
-	call_count = enoent_count;
-	local->call_count = call_count;
-
-	st_mode = st_mode_from_ia (sh->buf[sh->source].ia_prot,
-                                   sh->buf[sh->source].ia_type);
-	ia_dev  = sh->buf[sh->source].ia_dev;
-
-	gf_log (this->name, GF_LOG_TRACE,
-		"mknod %s mode 0%o on %d subvolumes",
-		local->loc.path, st_mode, enoent_count);
-
-        dict = dict_new ();
-        if (!dict)
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-
-        ret = afr_set_dict_gfid (dict, sh->buf[sh->source].ia_gfid);
-        if (ret)
-                gf_log (this->name, GF_LOG_DEBUG, "gfid set failed");
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    flock.l_type = F_WRLCK;
+    flock.l_start = off;
+    flock.l_len = size;
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->child_errno[i] == ENOENT) {
-			STACK_WIND_COOKIE (frame,
-					   sh_missing_entries_newentry_cbk,
-					   (void *) (long) i,
-					   priv->children[i],
-					   priv->children[i]->fops->mknod,
-					   &local->loc, st_mode, ia_dev, dict);
-			if (!--call_count)
-				break;
-		}
-	}
+    AFR_ONALL(frame, afr_selfheal_lock_cbk, inodelk, dom, &loc, F_SETLK, &flock,
+              NULL);
 
-        if (dict)
-                dict_unref (dict);
+    loc_wipe(&loc);
 
-	return 0;
+    return afr_locked_fill(frame, this, locked_on);
 }
 
+int
+afr_selfheal_inodelk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                     char *dom, off_t off, size_t size,
+                     unsigned char *locked_on)
+{
+    loc_t loc = {
+        0,
+    };
+    struct gf_flock flock = {
+        0,
+    };
+    afr_local_t *local = NULL;
+    int i = 0;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    flock.l_type = F_WRLCK;
+    flock.l_start = off;
+    flock.l_len = size;
+
+    AFR_ONALL(frame, afr_selfheal_lock_cbk, inodelk, dom, &loc, F_SETLK, &flock,
+              NULL);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->replies[i].op_ret == -1 &&
+            local->replies[i].op_errno == EAGAIN) {
+            afr_locked_fill(frame, this, locked_on);
+            afr_selfheal_uninodelk(frame, this, inode, dom, off, size,
+                                   locked_on);
+
+            AFR_SEQ(frame, afr_selfheal_lock_cbk, inodelk, dom, &loc, F_SETLKW,
+                    &flock, NULL);
+            break;
+        }
+    }
 
-static int
-sh_missing_entries_mkdir (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-        dict_t          *dict = NULL;
-	int              i = 0;
-        int              ret = 0;
-	int              enoent_count = 0;
-	int              call_count = 0;
-	mode_t           st_mode = 0;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	for (i = 0; i < priv->child_count; i++)
-		if (sh->child_errno[i] == ENOENT)
-			enoent_count++;
-
-	call_count = enoent_count;
-	local->call_count = call_count;
-
-	st_mode = st_mode_from_ia (sh->buf[sh->source].ia_prot,
-                                   sh->buf[sh->source].ia_type);
-
-        dict = dict_new ();
-        if (!dict) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                sh_missing_entries_finish (frame, this);
-                return 0;
+    loc_wipe(&loc);
+
+    return afr_locked_fill(frame, this, locked_on);
+}
+
+static void
+afr_get_lock_and_eagain_counts(afr_private_t *priv, struct afr_reply *replies,
+                               int *lock_count, int *eagain_count)
+{
+    int i = 0;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid)
+            continue;
+        if (replies[i].op_ret == 0) {
+            (*lock_count)++;
+        } else if (replies[i].op_ret == -1 && replies[i].op_errno == EAGAIN) {
+            (*eagain_count)++;
         }
+    }
+}
 
-        ret = afr_set_dict_gfid (dict, sh->buf[sh->source].ia_gfid);
-        if (ret)
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "inode gfid set failed");
+/*Do blocking locks if number of locks acquired is majority and there were some
+ * EAGAINs. Useful for odd-way replication*/
+int
+afr_selfheal_tie_breaker_inodelk(call_frame_t *frame, xlator_t *this,
+                                 inode_t *inode, char *dom, off_t off,
+                                 size_t size, unsigned char *locked_on)
+{
+    loc_t loc = {
+        0,
+    };
+    struct gf_flock flock = {
+        0,
+    };
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int lock_count = 0;
+    int eagain_count = 0;
 
+    priv = this->private;
+    local = frame->local;
 
-	gf_log (this->name, GF_LOG_TRACE,
-		"mkdir %s mode 0%o on %d subvolumes",
-		local->loc.path, st_mode, enoent_count);
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->child_errno[i] == ENOENT) {
-                        if (!strcmp (local->loc.path, "/")) {
-                                /* We shouldn't try to create "/" */
+    flock.l_type = F_WRLCK;
+    flock.l_start = off;
+    flock.l_len = size;
 
-                                sh_missing_entries_finish (frame, this);
+    AFR_ONALL(frame, afr_selfheal_lock_cbk, inodelk, dom, &loc, F_SETLK, &flock,
+              NULL);
 
-                                return 0;
-                        } else {
-                                STACK_WIND_COOKIE (frame,
-                                                   sh_missing_entries_newentry_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->mkdir,
-                                                   &local->loc, st_mode, dict);
-                                if (!--call_count)
-                                        break;
-                        }
-		}
-	}
+    afr_get_lock_and_eagain_counts(priv, local->replies, &lock_count,
+                                   &eagain_count);
 
-        if (dict)
-                dict_unref (dict);
+    if (lock_count > priv->child_count / 2 && eagain_count) {
+        afr_locked_fill(frame, this, locked_on);
+        afr_selfheal_uninodelk(frame, this, inode, dom, off, size, locked_on);
 
-	return 0;
+        AFR_SEQ(frame, afr_selfheal_lock_cbk, inodelk, dom, &loc, F_SETLKW,
+                &flock, NULL);
+    }
+
+    loc_wipe(&loc);
+
+    return afr_locked_fill(frame, this, locked_on);
 }
 
+int
+afr_selfheal_uninodelk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                       char *dom, off_t off, size_t size,
+                       const unsigned char *locked_on)
+{
+    loc_t loc = {
+        0,
+    };
+    struct gf_flock flock = {
+        0,
+    };
 
-static int
-sh_missing_entries_symlink (call_frame_t *frame, xlator_t *this,
-			    const char *link, struct iatt *buf)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-        dict_t          *dict = NULL;
-	int              i = 0;
-        int              ret = 0;
-	int              enoent_count = 0;
-	int              call_count = 0;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	for (i = 0; i < priv->child_count; i++)
-		if (sh->child_errno[i] == ENOENT)
-			enoent_count++;
-
-	call_count = enoent_count;
-	local->call_count = call_count;
-
-        dict = dict_new ();
-        if (!dict) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                sh_missing_entries_finish (frame, this);
-                return 0;
-        }
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-        ret = afr_set_dict_gfid (dict, buf->ia_gfid);
-        if (ret)
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "dict gfid set failed");
+    flock.l_type = F_UNLCK;
+    flock.l_start = off;
+    flock.l_len = size;
 
-	gf_log (this->name, GF_LOG_TRACE,
-		"symlink %s -> %s on %d subvolumes",
-		local->loc.path, link, enoent_count);
+    AFR_ONLIST(locked_on, frame, afr_selfheal_lock_cbk, inodelk, dom, &loc,
+               F_SETLK, &flock, NULL);
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->child_errno[i] == ENOENT) {
-			STACK_WIND_COOKIE (frame,
-					   sh_missing_entries_newentry_cbk,
-					   (void *) (long) i,
-					   priv->children[i],
-					   priv->children[i]->fops->symlink,
-					   link, &local->loc, dict);
-			if (!--call_count)
-				break;
-		}
-	}
+    loc_wipe(&loc);
 
-	return 0;
+    return 0;
 }
 
-
-static int
-sh_missing_entries_readlink_cbk (call_frame_t *frame, void *cookie,
-				 xlator_t *this,
-				 int32_t op_ret, int32_t op_errno,
-				 const char *link, struct iatt *sbuf)
+int
+afr_selfheal_tryentrylk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                        char *dom, const char *name, unsigned char *locked_on)
 {
-	if (op_ret > 0)
-		sh_missing_entries_symlink (frame, this, link, sbuf);
-	else
-		sh_missing_entries_finish (frame, this);
+    loc_t loc = {
+        0,
+    };
 
-	return 0;
-}
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
+    AFR_ONALL(frame, afr_selfheal_lock_cbk, entrylk, dom, &loc, name,
+              ENTRYLK_LOCK_NB, ENTRYLK_WRLCK, NULL);
 
-static int
-sh_missing_entries_readlink (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
+    loc_wipe(&loc);
 
+    return afr_locked_fill(frame, this, locked_on);
+}
 
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
+int
+afr_selfheal_entrylk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                     char *dom, const char *name, unsigned char *locked_on)
+{
+    loc_t loc = {
+        0,
+    };
+    afr_local_t *local = NULL;
+    int i = 0;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    AFR_ONALL(frame, afr_selfheal_lock_cbk, entrylk, dom, &loc, name,
+              ENTRYLK_LOCK_NB, ENTRYLK_WRLCK, NULL);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->replies[i].op_ret == -1 &&
+            local->replies[i].op_errno == EAGAIN) {
+            afr_locked_fill(frame, this, locked_on);
+            afr_selfheal_unentrylk(frame, this, inode, dom, name, locked_on,
+                                   NULL);
+
+            AFR_SEQ(frame, afr_selfheal_lock_cbk, entrylk, dom, &loc, name,
+                    ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL);
+            break;
+        }
+    }
 
-	STACK_WIND (frame, sh_missing_entries_readlink_cbk,
-		    priv->children[sh->source],
-		    priv->children[sh->source]->fops->readlink,
-		    &local->loc, 4096);
+    loc_wipe(&loc);
 
-	return 0;
+    return afr_locked_fill(frame, this, locked_on);
 }
 
+int
+afr_selfheal_tie_breaker_entrylk(call_frame_t *frame, xlator_t *this,
+                                 inode_t *inode, char *dom, const char *name,
+                                 unsigned char *locked_on)
+{
+    loc_t loc = {
+        0,
+    };
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int lock_count = 0;
+    int eagain_count = 0;
 
-static int
-sh_missing_entries_create (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	int              type = 0;
-	int              i = 0;
-	afr_private_t   *priv = NULL;
-	int              enoent_count = 0;
-	int              govinda_gOvinda = 0;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	for (i = 0; i < priv->child_count; i++) {
-                if (!local->child_up[i])
-                        continue;
-
-		if (sh->child_errno[i]) {
-			if (sh->child_errno[i] == ENOENT)
-				enoent_count++;
-		} else {
-			if (type) {
-				if (type != sh->buf[i].ia_type) {
-                                        gf_log (this->name, GF_LOG_TRACE,
-                                                "file %s is govinda!",
-                                                local->loc.path);
-
-					govinda_gOvinda = 1;
-                                }
-			} else {
-				sh->source = i;
-				type = sh->buf[i].ia_type;
-			}
-		}
-	}
-
-	if (govinda_gOvinda) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"conflicting filetypes exist for path %s. returning.",
-			local->loc.path);
-
-		local->govinda_gOvinda = 1;
-		sh_missing_entries_finish (frame, this);
-		return 0;
-	}
-
-	if (!type) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no source found for %s. all nodes down?. returning.",
-			local->loc.path);
-		/* subvolumes down and/or file does not exist */
-		sh_missing_entries_finish (frame, this);
-		return 0;
-	}
-
-	if (enoent_count == 0) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"no missing files - %s. proceeding to metadata check",
-			local->loc.path);
-		/* proceed to next step - metadata self-heal */
-		sh_missing_entries_finish (frame, this);
-		return 0;
-	}
-
-	switch (type) {
-	case IA_IFSOCK:
-	case IA_IFREG:
-	case IA_IFBLK:
-	case IA_IFCHR:
-	case IA_IFIFO:
-		sh_missing_entries_mknod (frame, this);
-		break;
-	case IA_IFLNK:
-		sh_missing_entries_readlink (frame, this);
-		break;
-	case IA_IFDIR:
-		sh_missing_entries_mkdir (frame, this);
-		break;
-	default:
-		gf_log (this->name, GF_LOG_ERROR,
-			"unknown file type: 0%o", type);
-		local->govinda_gOvinda = 1;
-		sh_missing_entries_finish (frame, this);
-	}
-
-	return 0;
-}
+    priv = this->private;
+    local = frame->local;
 
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-static int
-sh_missing_entries_lookup_cbk (call_frame_t *frame, void *cookie,
-			       xlator_t *this,
-			       int32_t op_ret, int32_t op_errno,
-                               inode_t *inode, struct iatt *buf, dict_t *xattr,
-                               struct iatt *postparent)
-{
-	int              child_index = 0;
-	afr_local_t     *local = NULL;
-	int              call_count = 0;
-	afr_private_t   *priv = NULL;
+    AFR_ONALL(frame, afr_selfheal_lock_cbk, entrylk, dom, &loc, name,
+              ENTRYLK_LOCK_NB, ENTRYLK_WRLCK, NULL);
 
+    afr_get_lock_and_eagain_counts(priv, local->replies, &lock_count,
+                                   &eagain_count);
 
-	local = frame->local;
-	priv = this->private;
+    if (lock_count > priv->child_count / 2 && eagain_count) {
+        afr_locked_fill(frame, this, locked_on);
+        afr_selfheal_unentrylk(frame, this, inode, dom, name, locked_on, NULL);
 
-	child_index = (long) cookie;
+        AFR_SEQ(frame, afr_selfheal_lock_cbk, entrylk, dom, &loc, name,
+                ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL);
+    }
 
-	LOCK (&frame->lock);
-	{
-		if (op_ret == 0) {
-			gf_log (this->name, GF_LOG_TRACE,
-				"path %s on subvolume %s is of mode 0%o",
-				local->loc.path,
-				priv->children[child_index]->name,
-				buf->ia_type);
+    loc_wipe(&loc);
 
-			local->self_heal.buf[child_index] = *buf;
-                        local->self_heal.parentbuf        = *postparent;
-		} else {
-			gf_log (this->name, GF_LOG_TRACE,
-				"path %s on subvolume %s => -1 (%s)",
-				local->loc.path,
-				priv->children[child_index]->name,
-				strerror (op_errno));
+    return afr_locked_fill(frame, this, locked_on);
+}
 
-			local->self_heal.child_errno[child_index] = op_errno;
-		}
+int
+afr_selfheal_unentrylk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                       char *dom, const char *name, unsigned char *locked_on,
+                       dict_t *xdata)
+{
+    loc_t loc = {
+        0,
+    };
 
-	}
-	UNLOCK (&frame->lock);
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
 
-	call_count = afr_frame_return (frame);
+    AFR_ONLIST(locked_on, frame, afr_selfheal_lock_cbk, entrylk, dom, &loc,
+               name, ENTRYLK_UNLOCK, ENTRYLK_WRLCK, xdata);
 
-	if (call_count == 0) {
-		sh_missing_entries_create (frame, this);
-	}
+    loc_wipe(&loc);
 
-	return 0;
+    return 0;
 }
 
+gf_boolean_t
+afr_is_data_set(xlator_t *this, dict_t *xdata)
+{
+    return afr_is_pending_set(this, xdata, AFR_DATA_TRANSACTION);
+}
 
-static int
-sh_missing_entries_lookup (call_frame_t *frame, xlator_t *this)
+gf_boolean_t
+afr_is_metadata_set(xlator_t *this, dict_t *xdata)
 {
-	afr_local_t    *local = NULL;
-	int             i = 0;
-	int             call_count = 0;
-	afr_private_t  *priv = NULL;
-	dict_t         *xattr_req = NULL;
-	int             ret = -1;
+    return afr_is_pending_set(this, xdata, AFR_METADATA_TRANSACTION);
+}
 
-	local = frame->local;
-	priv  = this->private;
+gf_boolean_t
+afr_is_entry_set(xlator_t *this, dict_t *xdata)
+{
+    return afr_is_pending_set(this, xdata, AFR_ENTRY_TRANSACTION);
+}
 
-	call_count = afr_up_children_count (priv->child_count,
-                                            local->child_up);
+/*
+ * This function inspects the looked up replies (in an unlocked manner)
+ * and decides whether a locked verification and possible healing is
+ * required or not. It updates the three booleans for each type
+ * of healing. If the boolean flag gets set to FALSE, then we are sure
+ * no healing is required. If the boolean flag gets set to TRUE then
+ * we have to proceed with locked reinspection.
+ */
 
-	local->call_count = call_count;
+int
+afr_selfheal_unlocked_inspect(call_frame_t *frame, xlator_t *this, uuid_t gfid,
+                              inode_t **link_inode, gf_boolean_t *data_selfheal,
+                              gf_boolean_t *metadata_selfheal,
+                              gf_boolean_t *entry_selfheal,
+                              struct afr_reply *replies_dst)
+{
+    afr_private_t *priv = NULL;
+    inode_t *inode = NULL;
+    int i = 0;
+    int valid_cnt = 0;
+    struct iatt first = {
+        0,
+    };
+    int first_idx = 0;
+    struct afr_reply *replies = NULL;
+    int ret = -1;
+
+    priv = this->private;
+
+    inode = afr_inode_find(this, gfid);
+    if (!inode)
+        goto out;
+
+    replies = alloca0(sizeof(*replies) * priv->child_count);
+
+    ret = afr_selfheal_unlocked_discover(frame, inode, gfid, replies);
+    if (ret)
+        goto out;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid)
+            continue;
+        if (replies[i].op_ret == -1)
+            continue;
+
+        /* The data segment of the changelog can be non-zero to indicate
+         * the directory needs a full heal. So the check below ensures
+         * it's not a directory before setting the data_selfheal boolean.
+         */
+        if (data_selfheal && !IA_ISDIR(replies[i].poststat.ia_type) &&
+            afr_is_data_set(this, replies[i].xdata))
+            *data_selfheal = _gf_true;
 
-	xattr_req = dict_new();
+        if (metadata_selfheal && afr_is_metadata_set(this, replies[i].xdata))
+            *metadata_selfheal = _gf_true;
 
-	if (xattr_req) {
-                for (i = 0; i < priv->child_count; i++) {
-                        ret = dict_set_uint64 (xattr_req,
-                                               priv->pending_key[i],
-                                               3 * sizeof(int32_t));
-                        if (ret < 0)
-                                gf_log (this->name, GF_LOG_WARNING,
-                                        "Unable to set dict value.");
-                }
+        if (entry_selfheal && afr_is_entry_set(this, replies[i].xdata))
+            *entry_selfheal = _gf_true;
+
+        valid_cnt++;
+        if (valid_cnt == 1) {
+            first = replies[i].poststat;
+            first_idx = i;
+            continue;
         }
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			gf_log (this->name, GF_LOG_TRACE,
-				"looking up %s on subvolume %s",
-				local->loc.path, priv->children[i]->name);
+        if (!IA_EQUAL(first, replies[i].poststat, type)) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+                   "TYPE mismatch %d vs %d on %s for gfid:%s",
+                   (int)first.ia_type, (int)replies[i].poststat.ia_type,
+                   priv->children[i]->name,
+                   uuid_utoa(replies[i].poststat.ia_gfid));
+            gf_event(EVENT_AFR_SPLIT_BRAIN,
+                     "client-pid=%d;"
+                     "subvol=%s;"
+                     "type=file;gfid=%s;"
+                     "ia_type-%d=%s;ia_type-%d=%s",
+                     this->ctx->cmd_args.client_pid, this->name,
+                     uuid_utoa(replies[i].poststat.ia_gfid), first_idx,
+                     gf_inode_type_to_str(first.ia_type), i,
+                     gf_inode_type_to_str(replies[i].poststat.ia_type));
+            ret = -EIO;
+            goto out;
+        }
 
-			STACK_WIND_COOKIE (frame,
-					   sh_missing_entries_lookup_cbk,
-					   (void *) (long) i,
-					   priv->children[i],
-					   priv->children[i]->fops->lookup,
-					   &local->loc, xattr_req);
+        if (!IA_EQUAL(first, replies[i].poststat, uid)) {
+            gf_msg_debug(this->name, 0,
+                         "UID mismatch "
+                         "%d vs %d on %s for gfid:%s",
+                         (int)first.ia_uid, (int)replies[i].poststat.ia_uid,
+                         priv->children[i]->name,
+                         uuid_utoa(replies[i].poststat.ia_gfid));
 
-			if (!--call_count)
-				break;
-		}
-	}
+            if (metadata_selfheal)
+                *metadata_selfheal = _gf_true;
+        }
 
-	if (xattr_req)
-		dict_unref (xattr_req);
+        if (!IA_EQUAL(first, replies[i].poststat, gid)) {
+            gf_msg_debug(this->name, 0,
+                         "GID mismatch "
+                         "%d vs %d on %s for gfid:%s",
+                         (int)first.ia_uid, (int)replies[i].poststat.ia_uid,
+                         priv->children[i]->name,
+                         uuid_utoa(replies[i].poststat.ia_gfid));
 
-	return 0;
-}
+            if (metadata_selfheal)
+                *metadata_selfheal = _gf_true;
+        }
 
+        if (!IA_EQUAL(first, replies[i].poststat, prot)) {
+            gf_msg_debug(this->name, 0,
+                         "MODE mismatch "
+                         "%d vs %d on %s for gfid:%s",
+                         (int)st_mode_from_ia(first.ia_prot, 0),
+                         (int)st_mode_from_ia(replies[i].poststat.ia_prot, 0),
+                         priv->children[i]->name,
+                         uuid_utoa(replies[i].poststat.ia_gfid));
+
+            if (metadata_selfheal)
+                *metadata_selfheal = _gf_true;
+        }
 
+        if (IA_ISREG(first.ia_type) &&
+            !IA_EQUAL(first, replies[i].poststat, size)) {
+            gf_msg_debug(this->name, 0,
+                         "SIZE mismatch "
+                         "%lld vs %lld on %s for gfid:%s",
+                         (long long)first.ia_size,
+                         (long long)replies[i].poststat.ia_size,
+                         priv->children[i]->name,
+                         uuid_utoa(replies[i].poststat.ia_gfid));
+
+            if (data_selfheal)
+                *data_selfheal = _gf_true;
+        }
+    }
 
-int
-afr_sh_post_nonblocking_entrylk_cbk (call_frame_t *frame, xlator_t *this)
+    if (valid_cnt > 0 && link_inode) {
+        *link_inode = inode_link(inode, NULL, NULL, &first);
+        if (!*link_inode) {
+            ret = -EINVAL;
+            goto out;
+        }
+    } else if (valid_cnt < 2) {
+        ret = afr_check_stale_error(replies, priv);
+        goto out;
+    }
+
+    ret = 0;
+out:
+    if (replies && replies_dst)
+        afr_replies_copy(replies_dst, replies, priv->child_count);
+    if (inode)
+        inode_unref(inode);
+    if (replies)
+        afr_replies_wipe(replies, priv->child_count);
+
+    return ret;
+}
+
+inode_t *
+afr_inode_find(xlator_t *this, uuid_t gfid)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
+    inode_table_t *table = NULL;
+    inode_t *inode = NULL;
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+    table = this->itable;
+    if (!table)
+        return NULL;
 
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Non blocking entrylks failed.");
-                afr_sh_missing_entries_done (frame, this);
-        } else {
+    inode = inode_find(table, gfid);
+    if (inode)
+        return inode;
 
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Non blocking entrylks done. Proceeding to FOP");
-                sh_missing_entries_lookup (frame, this);
-        }
+    inode = inode_new(table);
+    if (!inode)
+        return NULL;
 
-        return 0;
+    gf_uuid_copy(inode->gfid, gfid);
+
+    return inode;
 }
 
-static int
-afr_sh_entrylk (call_frame_t *frame, xlator_t *this)
+call_frame_t *
+afr_frame_create(xlator_t *this, int32_t *op_errno)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_self_heal_t     *sh       = NULL;
+    call_frame_t *frame = NULL;
+    afr_local_t *local = NULL;
+    pid_t pid = GF_CLIENT_PID_SELF_HEALD;
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        sh       = &local->self_heal;
+    frame = create_frame(this, this->ctx->pool);
+    if (!frame) {
+        if (op_errno)
+            *op_errno = ENOMEM;
+        return NULL;
+    }
 
-        int_lock->transaction_lk_type = AFR_SELFHEAL_LK;
-        int_lock->selfheal_lk_type    = AFR_ENTRY_SELF_HEAL_LK;
+    local = AFR_FRAME_INIT(frame, (*op_errno));
+    if (!local) {
+        STACK_DESTROY(frame->root);
+        return NULL;
+    }
 
-        afr_set_lock_number (frame, this);
+    syncopctx_setfspid(&pid);
 
-        int_lock->lk_basename = local->loc.name;
-        int_lock->lk_loc      = &sh->parent_loc;
-        int_lock->lock_cbk    = afr_sh_post_nonblocking_entrylk_cbk;
+    frame->root->pid = pid;
 
-        afr_nonblocking_entrylk (frame, this);
+    afr_set_lk_owner(frame, this, frame->root);
 
-        return 0;
+    return frame;
 }
 
-static int
-afr_self_heal_missing_entries (call_frame_t *frame, xlator_t *this)
+int
+afr_selfheal_newentry_mark(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                           int source, struct afr_reply *replies,
+                           unsigned char *sources, unsigned char *newentry)
 {
-        afr_internal_lock_t *int_lock = NULL;
-	afr_local_t         *local    = NULL;
-	afr_self_heal_t     *sh       = NULL;
-	afr_private_t       *priv     = NULL;
+    int ret = 0;
+    int i = 0;
+    afr_private_t *priv = NULL;
+    dict_t *xattr = NULL;
+    int **changelog = NULL;
 
-	local    = frame->local;
-        int_lock = &local->internal_lock;
-	sh       = &local->self_heal;
-	priv     = this->private;
+    priv = this->private;
 
-	gf_log (this->name, GF_LOG_TRACE,
-		"attempting to recreate missing entries for path=%s",
-		local->loc.path);
+    gf_uuid_copy(inode->gfid, replies[source].poststat.ia_gfid);
 
-	afr_build_parent_loc (&sh->parent_loc, &local->loc);
+    xattr = dict_new();
+    if (!xattr)
+        return -ENOMEM;
 
-        afr_sh_entrylk (frame, this);
-	return 0;
+    changelog = afr_mark_pending_changelog(priv, newentry, xattr,
+                                           replies[source].poststat.ia_type);
+
+    if (!changelog) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+        ret |= afr_selfheal_post_op(frame, this, inode, i, xattr, NULL);
+    }
+out:
+    if (changelog)
+        afr_matrix_cleanup(changelog, priv->child_count);
+    if (xattr)
+        dict_unref(xattr);
+    return ret;
 }
 
-afr_local_t *afr_local_copy (afr_local_t *l, xlator_t *this)
+int
+afr_selfheal_do(call_frame_t *frame, xlator_t *this, uuid_t gfid)
 {
-        afr_private_t *priv = NULL;
-        afr_local_t   *lc     = NULL;
-        afr_self_heal_t *sh = NULL;
-        afr_self_heal_t *shc = NULL;
+    int ret = -1;
+    int entry_ret = 1;
+    int metadata_ret = 1;
+    int data_ret = 1;
+    int or_ret = 0;
+    inode_t *inode = NULL;
+    fd_t *fd = NULL;
+    gf_boolean_t data_selfheal = _gf_false;
+    gf_boolean_t metadata_selfheal = _gf_false;
+    gf_boolean_t entry_selfheal = _gf_false;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+
+    ret = afr_selfheal_unlocked_inspect(frame, this, gfid, &inode,
+                                        &data_selfheal, &metadata_selfheal,
+                                        &entry_selfheal, NULL);
+    if (ret)
+        goto out;
+
+    if (!(data_selfheal || metadata_selfheal || entry_selfheal)) {
+        ret = 2;
+        goto out;
+    }
+
+    if (inode->ia_type == IA_IFREG) {
+        ret = afr_selfheal_data_open(this, inode, &fd);
+        if (!fd) {
+            ret = -EIO;
+            goto out;
+        }
+    }
 
+    if (data_selfheal && priv->data_self_heal)
+        data_ret = afr_selfheal_data(frame, this, fd);
 
-        priv = this->private;
+    if (metadata_selfheal && priv->metadata_self_heal)
+        metadata_ret = afr_selfheal_metadata(frame, this, inode);
 
-        sh = &l->self_heal;
+    if (entry_selfheal && priv->entry_self_heal)
+        entry_ret = afr_selfheal_entry(frame, this, inode);
 
-        lc = GF_CALLOC (1, sizeof (afr_local_t),
-                        gf_afr_mt_afr_local_t);
+    or_ret = (data_ret | metadata_ret | entry_ret);
 
-        shc = &lc->self_heal;
+    if (data_ret == -EIO || metadata_ret == -EIO || entry_ret == -EIO)
+        ret = -EIO;
+    else if (data_ret == 1 && metadata_ret == 1 && entry_ret == 1)
+        ret = 1;
+    else if (or_ret < 0)
+        ret = or_ret;
+    else
+        ret = 0;
 
-        shc->unwind = sh->unwind;
-        shc->need_data_self_heal = sh->need_data_self_heal;
-        shc->need_metadata_self_heal = sh->need_metadata_self_heal;
-        shc->need_entry_self_heal = sh->need_entry_self_heal;
-        shc->forced_merge = sh->forced_merge;
-        shc->healing_fd_opened = sh->healing_fd_opened;
-        shc->data_lock_held = sh->data_lock_held;
-        if (sh->healing_fd && !sh->healing_fd_opened)
-                shc->healing_fd = fd_ref (sh->healing_fd);
-        else
-                shc->healing_fd = sh->healing_fd;
-        shc->background = sh->background;
-        shc->type = sh->type;
-
-        if (l->loc.path)
-                loc_copy (&lc->loc, &l->loc);
-
-        lc->child_up  = memdup (l->child_up, priv->child_count);
-        if (l->xattr_req)
-                lc->xattr_req = dict_ref (l->xattr_req);
-
-        if (l->cont.lookup.inode)
-                lc->cont.lookup.inode = inode_ref (l->cont.lookup.inode);
-        if (l->cont.lookup.xattr)
-                lc->cont.lookup.xattr = dict_ref (l->cont.lookup.xattr);
-        if (l->internal_lock.inode_locked_nodes)
-                lc->internal_lock.inode_locked_nodes =
-                        memdup (l->internal_lock.inode_locked_nodes,
-                                priv->child_count);
-        else
-                lc->internal_lock.inode_locked_nodes =
-                        GF_CALLOC (sizeof (*l->internal_lock.inode_locked_nodes),
-                                   priv->child_count,
-                                   gf_afr_mt_char);
-        if (l->internal_lock.entry_locked_nodes)
-                lc->internal_lock.entry_locked_nodes =
-                        memdup (l->internal_lock.entry_locked_nodes,
-                                priv->child_count);
-        else
-                lc->internal_lock.entry_locked_nodes =
-                        GF_CALLOC (sizeof (*l->internal_lock.entry_locked_nodes),
-                                   priv->child_count,
-                                   gf_afr_mt_char);
-        if (l->internal_lock.locked_nodes)
-                lc->internal_lock.locked_nodes =
-                        memdup (l->internal_lock.locked_nodes,
-                                priv->child_count);
-        else
-                lc->internal_lock.locked_nodes =
-                        GF_CALLOC (sizeof (*l->internal_lock.locked_nodes),
-                                   priv->child_count,
-                                   gf_afr_mt_char);
+out:
+    if (inode)
+        inode_unref(inode);
+    if (fd)
+        fd_unref(fd);
+    return ret;
+}
+/*
+ * This is the entry point for healing a given GFID. The return values for this
+ * function are as follows:
+ * '0' if the self-heal is successful
+ * '1' if the afr-xattrs are non-zero (due to on-going IO) and no heal is needed
+ * '2' if the afr-xattrs are all-zero and no heal is needed
+ * $errno if the heal on the gfid failed.
+ */
 
-        lc->internal_lock.inodelk_lock_count =
-                l->internal_lock.inodelk_lock_count;
-        lc->internal_lock.entrylk_lock_count =
-                l->internal_lock.entrylk_lock_count;
+int
+afr_selfheal(xlator_t *this, uuid_t gfid)
+{
+    int ret = -1;
+    call_frame_t *frame = NULL;
+    afr_local_t *local = NULL;
 
-        return lc;
-}
+    frame = afr_frame_create(this, NULL);
+    if (!frame)
+        return ret;
 
+    local = frame->local;
+    local->xdata_req = dict_new();
 
-int
-afr_self_heal_completion_cbk (call_frame_t *bgsh_frame, xlator_t *this)
-{
-        afr_private_t *   priv  = NULL;
-	afr_local_t *     local = NULL;
-        afr_self_heal_t * sh    = NULL;
-        char              sh_type_str[256] = {0,};
-
-        priv  = this->private;
-	local = bgsh_frame->local;
-        sh    = &local->self_heal;
-
-	if (local->govinda_gOvinda) {
-                afr_set_split_brain (this, local->cont.lookup.inode, _gf_true);
-	} else {
-                afr_set_split_brain (this, local->cont.lookup.inode, _gf_false);
-	}
-
-        afr_self_heal_type_str_get(sh, sh_type_str,
-                                   sizeof(sh_type_str));
-        gf_log (this->name, GF_LOG_NORMAL,
-                "background %s self-heal completed on %s", sh_type_str,
-                local->loc.path);
-
-        if (!sh->unwound) {
-                sh->unwind (sh->orig_frame, this);
-        }
+    ret = afr_selfheal_do(frame, this, gfid);
 
-        if (sh->background) {
-                LOCK (&priv->lock);
-                {
-                        priv->background_self_heals_started--;
-                }
-                UNLOCK (&priv->lock);
-        }
+    if (frame)
+        AFR_STACK_DESTROY(frame);
 
-        AFR_STACK_DESTROY (bgsh_frame);
+    return ret;
+}
 
-	return 0;
+afr_local_t *
+__afr_dequeue_heals(afr_private_t *priv)
+{
+    afr_local_t *local = NULL;
+
+    if (list_empty(&priv->heal_waiting))
+        goto none;
+    if ((priv->background_self_heal_count > 0) &&
+        (priv->healers >= priv->background_self_heal_count))
+        goto none;
+
+    local = list_entry(priv->heal_waiting.next, afr_local_t, healer);
+    priv->heal_waiters--;
+    GF_ASSERT(priv->heal_waiters >= 0);
+    list_del_init(&local->healer);
+    list_add(&local->healer, &priv->healing);
+    priv->healers++;
+    return local;
+none:
+    gf_msg_debug(THIS->name, 0,
+                 "Nothing dequeued. "
+                 "Num healers: %d, Num Waiters: %d",
+                 priv->healers, priv->heal_waiters);
+    return NULL;
 }
 
 int
-afr_self_heal (call_frame_t *frame, xlator_t *this)
+afr_refresh_selfheal_wrap(void *opaque)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              i = 0;
+    call_frame_t *heal_frame = opaque;
+    afr_local_t *local = heal_frame->local;
+    int ret = 0;
 
-        call_frame_t *sh_frame = NULL;
-        afr_local_t  *sh_local = NULL;
+    ret = afr_selfheal(heal_frame->this, local->refreshinode->gfid);
+    return ret;
+}
 
+int
+afr_refresh_heal_done(int ret, call_frame_t *frame, void *opaque)
+{
+    call_frame_t *heal_frame = opaque;
+    xlator_t *this = heal_frame->this;
+    afr_private_t *priv = this->private;
+    afr_local_t *local = heal_frame->local;
+
+    LOCK(&priv->lock);
+    {
+        list_del_init(&local->healer);
+        priv->healers--;
+        GF_ASSERT(priv->healers >= 0);
+        local = __afr_dequeue_heals(priv);
+    }
+    UNLOCK(&priv->lock);
+
+    AFR_STACK_DESTROY(heal_frame);
+
+    if (local)
+        afr_heal_synctask(this, local);
+    return 0;
+}
 
-	local = frame->local;
-	priv  = this->private;
+void
+afr_heal_synctask(xlator_t *this, afr_local_t *local)
+{
+    int ret = 0;
+    call_frame_t *heal_frame = NULL;
+
+    heal_frame = local->heal_frame;
+    ret = synctask_new(this->ctx->env, afr_refresh_selfheal_wrap,
+                       afr_refresh_heal_done, heal_frame, heal_frame);
+    if (ret < 0)
+        /* Heal not launched. Will be queued when the next inode
+         * refresh happens and shd hasn't healed it yet. */
+        afr_refresh_heal_done(ret, heal_frame, heal_frame);
+}
 
-        afr_set_lk_owner (frame, this);
+gf_boolean_t
+afr_throttled_selfheal(call_frame_t *frame, xlator_t *this)
+{
+    gf_boolean_t can_heal = _gf_true;
+    afr_private_t *priv = this->private;
+    afr_local_t *local = frame->local;
+
+    LOCK(&priv->lock);
+    {
+        if ((priv->background_self_heal_count > 0) &&
+            (priv->heal_wait_qlen + priv->background_self_heal_count) >
+                (priv->heal_waiters + priv->healers)) {
+            list_add_tail(&local->healer, &priv->heal_waiting);
+            priv->heal_waiters++;
+            local = __afr_dequeue_heals(priv);
+        } else {
+            can_heal = _gf_false;
+        }
+    }
+    UNLOCK(&priv->lock);
 
-        if (local->self_heal.background) {
-                LOCK (&priv->lock);
-                {
-                        if (priv->background_self_heals_started
-                            > priv->background_self_heal_count) {
+    if (can_heal) {
+        if (local)
+            afr_heal_synctask(this, local);
+        else
+            gf_msg_debug(this->name, 0,
+                         "Max number of heals are "
+                         "pending, background self-heal rejected.");
+    }
 
-                                local->self_heal.background = _gf_false;
+    return can_heal;
+}
 
-                        } else {
-                                priv->background_self_heals_started++;
-                        }
-                }
-                UNLOCK (&priv->lock);
+int
+afr_choose_source_by_policy(afr_private_t *priv, unsigned char *sources,
+                            afr_transaction_type type)
+{
+    int source = -1;
+    int i = 0;
+
+    /* Give preference to local child to save on bandwidth */
+    for (i = 0; i < priv->child_count; i++) {
+        if (priv->local[i] && sources[i]) {
+            if ((type == AFR_DATA_TRANSACTION) && AFR_IS_ARBITER_BRICK(priv, i))
+                continue;
+
+            source = i;
+            goto out;
         }
+    }
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "performing self heal on %s (metadata=%d data=%d entry=%d)",
-                local->loc.path,
-                local->self_heal.need_metadata_self_heal,
-                local->self_heal.need_data_self_heal,
-                local->self_heal.need_entry_self_heal);
-
-        sh_frame        = copy_frame (frame);
-        sh_local        = afr_local_copy (local, this);
-        sh_frame->local = sh_local;
-        sh              = &sh_local->self_heal;
-
-        sh->orig_frame  = frame;
-
-        sh->completion_cbk = afr_self_heal_completion_cbk;
-
-
-	sh->buf = GF_CALLOC (priv->child_count, sizeof (struct iatt),
-                             gf_afr_mt_iatt);
-	sh->child_errno = GF_CALLOC (priv->child_count, sizeof (int),
-                                     gf_afr_mt_int);
-	sh->success = GF_CALLOC (priv->child_count, sizeof (int),
-                                gf_afr_mt_int);
-	sh->xattr = GF_CALLOC (priv->child_count, sizeof (dict_t *),
-                                gf_afr_mt_dict_t);
-	sh->sources = GF_CALLOC (sizeof (*sh->sources), priv->child_count,
-                                gf_afr_mt_int);
-	sh->locked_nodes = GF_CALLOC (sizeof (*sh->locked_nodes), 
-                                      priv->child_count,
-                                      gf_afr_mt_int);
-
-	sh->pending_matrix = GF_CALLOC (sizeof (int32_t *), priv->child_count,
-                                        gf_afr_mt_int32_t);
-
-	for (i = 0; i < priv->child_count; i++) {
-		sh->pending_matrix[i] = GF_CALLOC (sizeof (int32_t),
-					  	   priv->child_count,
-                                                   gf_afr_mt_int32_t);
-	}
-
-	sh->delta_matrix = GF_CALLOC (sizeof (int32_t *), priv->child_count,
-                                      gf_afr_mt_int32_t);
-	for (i = 0; i < priv->child_count; i++) {
-		sh->delta_matrix[i] = GF_CALLOC (sizeof (int32_t),
-					         priv->child_count,
-                                                 gf_afr_mt_int32_t);
-	}
-
-	if (local->success_count && local->enoent_count) {
-		afr_self_heal_missing_entries (sh_frame, this);
-	} else {
-		gf_log (this->name, GF_LOG_TRACE,
-			"proceeding to metadata check on %s",
-			local->loc.path);
-
-		afr_sh_missing_entries_done (sh_frame, this);
-	}
-
-	return 0;
+    for (i = 0; i < priv->child_count; i++) {
+        if (sources[i]) {
+            source = i;
+            goto out;
+        }
+    }
+out:
+    return source;
 }
 
-void
-afr_self_heal_type_str_get (afr_self_heal_t *self_heal_p, char *str,
-                            size_t size)
+static int
+afr_anon_inode_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, inode_t *inode,
+                         struct iatt *buf, struct iatt *preparent,
+                         struct iatt *postparent, dict_t *xdata)
 {
-        assert(str && (size > 0));
+    afr_local_t *local = frame->local;
+    int i = (long)cookie;
+
+    local->replies[i].valid = 1;
+    local->replies[i].op_ret = op_ret;
+    local->replies[i].op_errno = op_errno;
+    if (op_ret == 0) {
+        local->op_ret = 0;
+        local->replies[i].poststat = *buf;
+        local->replies[i].preparent = *preparent;
+        local->replies[i].postparent = *postparent;
+    }
+    if (xdata) {
+        local->replies[i].xdata = dict_ref(xdata);
+    }
+
+    syncbarrier_wake(&local->barrier);
+    return 0;
+}
 
-        if (self_heal_p->need_metadata_self_heal) {
-                snprintf(str, size, " meta-data");
+int
+afr_anon_inode_create(xlator_t *this, int child, inode_t **linked_inode)
+{
+    call_frame_t *frame = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = this->private;
+    unsigned char *mkdir_on = alloca0(priv->child_count);
+    unsigned char *lookup_on = alloca0(priv->child_count);
+    loc_t loc = {0};
+    int32_t op_errno = 0;
+    int32_t child_op_errno = 0;
+    struct iatt iatt = {0};
+    dict_t *xdata = NULL;
+    uuid_t anon_inode_gfid = {0};
+    int mkdir_count = 0;
+    int i = 0;
+
+    /*Try to mkdir everywhere and return success if the dir exists on 'child'
+     */
+
+    if (!priv->use_anon_inode) {
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    frame = afr_frame_create(this, &op_errno);
+    if (op_errno) {
+        goto out;
+    }
+    local = frame->local;
+    if (!local->child_up[child]) {
+        /*Other bricks may need mkdir so don't error out yet*/
+        child_op_errno = ENOTCONN;
+    }
+    gf_uuid_parse(priv->anon_gfid_str, anon_inode_gfid);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->child_up[i])
+            continue;
+
+        if (priv->anon_inode[i]) {
+            mkdir_on[i] = 0;
+        } else {
+            mkdir_on[i] = 1;
+            mkdir_count++;
         }
+    }
 
-        if (self_heal_p->need_data_self_heal) {
-                snprintf(str + strlen(str), size - strlen(str),
-                         " data");
+    if (mkdir_count == 0) {
+        *linked_inode = inode_find(this->itable, anon_inode_gfid);
+        if (*linked_inode) {
+            op_errno = 0;
+            goto out;
+        }
+    }
+
+    loc.parent = inode_ref(this->itable->root);
+    loc.name = priv->anon_inode_name;
+    loc.inode = inode_new(this->itable);
+    if (!loc.inode) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    xdata = dict_new();
+    if (!xdata) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    op_errno = -dict_set_gfuuid(xdata, "gfid-req", anon_inode_gfid, _gf_true);
+    if (op_errno) {
+        goto out;
+    }
+
+    if (mkdir_count == 0) {
+        memcpy(lookup_on, local->child_up, priv->child_count);
+        goto lookup;
+    }
+
+    AFR_ONLIST(mkdir_on, frame, afr_anon_inode_mkdir_cbk, mkdir, &loc, 0755, 0,
+               xdata);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!mkdir_on[i]) {
+            continue;
         }
 
-        if (self_heal_p->need_entry_self_heal) {
-                snprintf(str + strlen(str), size - strlen(str),
-                         " entry");
+        if (local->replies[i].op_ret == 0) {
+            priv->anon_inode[i] = 1;
+            iatt = local->replies[i].poststat;
+        } else if (local->replies[i].op_ret < 0 &&
+                   local->replies[i].op_errno == EEXIST) {
+            lookup_on[i] = 1;
+        } else if (i == child) {
+            child_op_errno = local->replies[i].op_errno;
+        }
+    }
+
+    if (AFR_COUNT(lookup_on, priv->child_count) == 0) {
+        goto link;
+    }
+
+lookup:
+    AFR_ONLIST(lookup_on, frame, afr_selfheal_discover_cbk, lookup, &loc,
+               xdata);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!lookup_on[i]) {
+            continue;
         }
+
+        if (local->replies[i].op_ret == 0) {
+            if (gf_uuid_compare(anon_inode_gfid,
+                                local->replies[i].poststat.ia_gfid) == 0) {
+                priv->anon_inode[i] = 1;
+                iatt = local->replies[i].poststat;
+            } else {
+                if (i == child)
+                    child_op_errno = EINVAL;
+                gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_INVALID_DATA,
+                       "%s has gfid: %s", priv->anon_inode_name,
+                       uuid_utoa(local->replies[i].poststat.ia_gfid));
+            }
+        } else if (i == child) {
+            child_op_errno = local->replies[i].op_errno;
+        }
+    }
+link:
+    if (!gf_uuid_is_null(iatt.ia_gfid)) {
+        *linked_inode = inode_link(loc.inode, loc.parent, loc.name, &iatt);
+        if (*linked_inode) {
+            op_errno = 0;
+            inode_lookup(*linked_inode);
+        } else {
+            op_errno = ENOMEM;
+        }
+        goto out;
+    }
+
+out:
+    if (xdata)
+        dict_unref(xdata);
+    loc_wipe(&loc);
+    /*child_op_errno takes precedence*/
+    if (child_op_errno == 0) {
+        child_op_errno = op_errno;
+    }
+
+    if (child_op_errno && *linked_inode) {
+        inode_unref(*linked_inode);
+        *linked_inode = NULL;
+    }
+    if (frame)
+        AFR_STACK_DESTROY(frame);
+    return -child_op_errno;
 }
diff --git a/xlators/cluster/afr/src/afr-self-heal-common.h b/xlators/cluster/afr/src/afr-self-heal-common.h
deleted file mode 100644
index ecec8a5ae58..00000000000
--- a/xlators/cluster/afr/src/afr-self-heal-common.h
+++ /dev/null
@@ -1,73 +0,0 @@
-/*
-  Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef __AFR_SELF_HEAL_COMMON_H__
-#define __AFR_SELF_HEAL_COMMON_H__
-
-#define FILE_HAS_HOLES(buf) (((buf)->ia_size) > ((buf)->ia_blocks * 512))
-
-typedef enum {
-        AFR_SELF_HEAL_ENTRY,
-        AFR_SELF_HEAL_METADATA,
-        AFR_SELF_HEAL_DATA,
-} afr_self_heal_type;
-
-int
-afr_sh_select_source (int sources[], int child_count);
-
-int
-afr_sh_sink_count (int sources[], int child_count);
-
-int
-afr_sh_source_count (int sources[], int child_count);
-
-int
-afr_sh_supress_errenous_children (int sources[], int child_errno[],
-				  int child_count);
-
-void
-afr_sh_print_pending_matrix (int32_t *pending_matrix[], xlator_t *this);
-
-void
-afr_sh_build_pending_matrix (afr_private_t *priv,
-                             int32_t *pending_matrix[], dict_t *xattr[],
-			     int child_count, afr_transaction_type type);
-
-void
-afr_sh_pending_to_delta (afr_private_t *priv, dict_t **xattr,
-                         int32_t *delta_matrix[], int success[],
-                         int child_count, afr_transaction_type type);
-
-int
-afr_sh_mark_sources (afr_self_heal_t *sh, int child_count,
-                     afr_self_heal_type type);
-
-int
-afr_sh_delta_to_xattr (afr_private_t *priv,
-                       int32_t *delta_matrix[], dict_t *xattr[],
-		       int child_count, afr_transaction_type type);
-
-int
-afr_sh_is_matrix_zero (int32_t *pending_matrix[], int child_count);
-
-void
-afr_self_heal_type_str_get (afr_self_heal_t *self_heal_p, char *str,
-                            size_t size);
-
-#endif /* __AFR_SELF_HEAL_COMMON_H__ */
diff --git a/xlators/cluster/afr/src/afr-self-heal-data.c b/xlators/cluster/afr/src/afr-self-heal-data.c
index 07773816969..37bcc2b3f9e 100644
--- a/xlators/cluster/afr/src/afr-self-heal-data.c
+++ b/xlators/cluster/afr/src/afr-self-heal-data.c
@@ -1,1113 +1,891 @@
 /*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-#include <libgen.h>
-#include <unistd.h>
-#include <fnmatch.h>
-#include <sys/time.h>
-#include <stdlib.h>
-#include <signal.h>
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
 
-#include "glusterfs.h"
 #include "afr.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-#include "byte-order.h"
-
-#include "afr-transaction.h"
 #include "afr-self-heal.h"
-#include "afr-self-heal-common.h"
-#include "afr-self-heal-algorithm.h"
-
+#include <glusterfs/byte-order.h>
+#include "protocol-common.h"
+#include "afr-messages.h"
+#include <glusterfs/events.h>
 
-int
-afr_sh_data_done (call_frame_t *frame, xlator_t *this)
+#define HAS_HOLES(i) ((i->ia_blocks * 512) < (i->ia_size))
+static int
+__checksum_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, uint32_t weak, uint8_t *strong, dict_t *xdata)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	/* 
-	   TODO: cleanup sh->* 
-	 */
-
-        if (sh->healing_fd && !sh->healing_fd_opened) {
-                /* unref only if we created the fd ourselves */
-
-		fd_unref (sh->healing_fd);
-		sh->healing_fd = NULL;
+    afr_local_t *local = NULL;
+    struct afr_reply *replies = NULL;
+    int i = (long)cookie;
+
+    local = frame->local;
+    replies = local->replies;
+
+    replies[i].valid = 1;
+    replies[i].op_ret = op_ret;
+    replies[i].op_errno = op_errno;
+    if (xdata) {
+        replies[i].buf_has_zeroes = dict_get_str_boolean(
+            xdata, "buf-has-zeroes", _gf_false);
+        replies[i].fips_mode_rchecksum = dict_get_str_boolean(
+            xdata, "fips-mode-rchecksum", _gf_false);
+    }
+    if (strong) {
+        if (replies[i].fips_mode_rchecksum) {
+            memcpy(local->replies[i].checksum, strong, SHA256_DIGEST_LENGTH);
+        } else {
+            memcpy(local->replies[i].checksum, strong, MD5_DIGEST_LENGTH);
         }
+    }
 
-/*         for (i = 0; i < priv->child_count; i++) */
-/*                 sh->locked_nodes[i] = 0; */
-
-	gf_log (this->name, GF_LOG_TRACE,
-		"self heal of %s completed",
-		local->loc.path);
-
-	sh->completion_cbk (frame, this);
-
-	return 0;
+    syncbarrier_wake(&local->barrier);
+    return 0;
 }
 
-
-int
-afr_sh_data_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno)
+static gf_boolean_t
+__afr_can_skip_data_block_heal(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                               int source, unsigned char *healed_sinks,
+                               off_t offset, size_t size, struct iatt *poststat)
 {
-	afr_local_t     *local = NULL;
-	afr_private_t   *priv  = NULL;
-	int              call_count = 0;
-
-        int child_index = (long) cookie;
-
-	local = frame->local;
-	priv = this->private;
-
-	LOCK (&frame->lock);
-	{
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "flush or setattr failed on %s on subvolume %s: %s",
-                                local->loc.path, priv->children[child_index]->name,
-                                strerror (op_errno));
-                }
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    unsigned char *wind_subvols = NULL;
+    gf_boolean_t checksum_match = _gf_true;
+    struct afr_reply *replies = NULL;
+    dict_t *xdata = NULL;
+    int i = 0;
+
+    priv = this->private;
+    local = frame->local;
+    replies = local->replies;
+
+    xdata = dict_new();
+    if (!xdata)
+        goto out;
+    if (dict_set_int32_sizen(xdata, "check-zero-filled", 1)) {
+        dict_unref(xdata);
+        goto out;
+    }
+
+    wind_subvols = alloca0(priv->child_count);
+    for (i = 0; i < priv->child_count; i++) {
+        if (i == source || healed_sinks[i])
+            wind_subvols[i] = 1;
+    }
+
+    AFR_ONLIST(wind_subvols, frame, __checksum_cbk, rchecksum, fd, offset, size,
+               xdata);
+    if (xdata)
+        dict_unref(xdata);
+
+    if (!replies[source].valid || replies[source].op_ret != 0)
+        return _gf_false;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (i == source)
+            continue;
+        if (replies[i].valid) {
+            if (memcmp(replies[source].checksum, replies[i].checksum,
+                       replies[source].fips_mode_rchecksum
+                           ? SHA256_DIGEST_LENGTH
+                           : MD5_DIGEST_LENGTH)) {
+                checksum_match = _gf_false;
+                break;
+            }
+        }
+    }
 
-	if (call_count == 0) {
-		afr_sh_data_done (frame, this);
-	}
+    if (checksum_match) {
+        if (HAS_HOLES(poststat))
+            return _gf_true;
 
-	return 0;
+        /* For non-sparse files, we might be better off writing the
+         * zeroes to sinks to avoid mismatch of disk-usage in bricks. */
+        if (local->replies[source].buf_has_zeroes)
+            return _gf_false;
+        else
+            return _gf_true;
+    }
+out:
+    return _gf_false;
 }
 
-
-int
-afr_sh_data_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, struct iatt *statpre, struct iatt *statpost)
+static gf_boolean_t
+__afr_is_sink_zero_filled(xlator_t *this, fd_t *fd, size_t size, off_t offset,
+                          int sink)
 {
-        afr_sh_data_flush_cbk (frame, cookie, this, op_ret, op_errno);
-
-        return 0;
+    afr_private_t *priv = NULL;
+    struct iobref *iobref = NULL;
+    struct iovec *iovec = NULL;
+    int count = 0;
+    int ret = 0;
+    gf_boolean_t zero_filled = _gf_false;
+
+    priv = this->private;
+    ret = syncop_readv(priv->children[sink], fd, size, offset, 0, &iovec,
+                       &count, &iobref, NULL, NULL, NULL);
+    if (ret < 0)
+        goto out;
+    ret = iov_0filled(iovec, count);
+    if (!ret)
+        zero_filled = _gf_true;
+out:
+    if (iovec)
+        GF_FREE(iovec);
+    if (iobref)
+        iobref_unref(iobref);
+    return zero_filled;
 }
 
-
-int
-afr_sh_data_close (call_frame_t *frame, xlator_t *this)
+static int
+__afr_selfheal_data_read_write(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                               int source, unsigned char *healed_sinks,
+                               off_t offset, size_t size,
+                               struct afr_reply *replies, int type)
 {
-	afr_local_t     *local = NULL;
-	afr_private_t   *priv  = NULL;
-	afr_self_heal_t *sh    = NULL;
-
-	int  i            = 0;
-	int  call_count   = 0;
-        int  source       = 0;
-        int32_t valid     = 0;
-
-        struct iatt stbuf = {0,};
-
-        local = frame->local;
-        sh    = &local->self_heal;
-        priv  = this->private;
-
-        source       = sh->source;
-
-        valid |= (GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME);
-
-	stbuf.ia_atime = sh->buf[source].ia_atime;
-	stbuf.ia_atime_nsec = sh->buf[source].ia_atime_nsec;
-	stbuf.ia_mtime = sh->buf[source].ia_mtime;
-	stbuf.ia_mtime_nsec = sh->buf[source].ia_mtime_nsec;
-
-        if (sh->healing_fd_opened) {
-                /* not our job to close the fd */
+    struct iovec *iovec = NULL;
+    int count = 0;
+    struct iobref *iobref = NULL;
+    int ret = 0;
+    int i = 0;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+
+    ret = syncop_readv(priv->children[source], fd, size, offset, 0, &iovec,
+                       &count, &iobref, NULL, NULL, NULL);
+    if (ret <= 0)
+        return ret;
 
-                afr_sh_data_done (frame, this);
-                return 0;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!healed_sinks[i])
+            continue;
+
+            /*
+             * TODO: Use fiemap() and discard() to heal holes
+             * in the future.
+             *
+             * For now,
+             *
+             * - if the source had any holes at all,
+             * AND
+             * - if we are writing past the original file size
+             *   of the sink
+             * AND
+             * - is NOT the last block of the source file. if
+             *   the block contains EOF, it has to be written
+             *   in order to set the file size even if the
+             *   last block is 0-filled.
+             * AND
+             * - if the read buffer is filled with only 0's
+             *
+             * then, skip writing to this source. We don't depend
+             * on the write to happen to update the size as we
+             * have performed an ftruncate() upfront anyways.
+             */
+#define is_last_block(o, b, s) ((s >= o) && (s <= (o + b)))
+        if (HAS_HOLES((&replies[source].poststat)) &&
+            offset >= replies[i].poststat.ia_size &&
+            !is_last_block(offset, size, replies[source].poststat.ia_size) &&
+            (iov_0filled(iovec, count) == 0))
+            continue;
+
+        /* Avoid filling up sparse regions of the sink with 0-filled
+         * writes.*/
+        if (type == AFR_SELFHEAL_DATA_FULL &&
+            HAS_HOLES((&replies[source].poststat)) &&
+            ((offset + size) <= replies[i].poststat.ia_size) &&
+            (iov_0filled(iovec, count) == 0) &&
+            __afr_is_sink_zero_filled(this, fd, size, offset, i)) {
+            continue;
         }
 
-	if (!sh->healing_fd) {
-		afr_sh_data_done (frame, this);
-		return 0;
-	}
-
-	call_count        = (sh->active_sinks + 1) * 2;
-	local->call_count = call_count;
-
-	/* closed source */
-	gf_log (this->name, GF_LOG_TRACE,
-		"closing fd of %s on %s",
-		local->loc.path, priv->children[sh->source]->name);
-
-	STACK_WIND_COOKIE (frame, afr_sh_data_flush_cbk,
-			   (void *) (long) sh->source,
-			   priv->children[sh->source],
-			   priv->children[sh->source]->fops->flush,
-			   sh->healing_fd);
-	call_count--;
-
-        STACK_WIND_COOKIE (frame, afr_sh_data_setattr_cbk,
-                           (void *) (long) sh->source,
-                           priv->children[sh->source],
-                           priv->children[sh->source]->fops->setattr,
-                           &local->loc, &stbuf, valid);
-
-        call_count--;
-
-        if (call_count == 0)
-                return 0;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->sources[i] || !local->child_up[i])
-			continue;
-
-		gf_log (this->name, GF_LOG_TRACE,
-			"closing fd of %s on %s",
-			local->loc.path, priv->children[i]->name);
-
-		STACK_WIND_COOKIE (frame, afr_sh_data_flush_cbk,
-				   (void *) (long) i,
-				   priv->children[i],
-				   priv->children[i]->fops->flush,
-				   sh->healing_fd);
-
-                call_count--;
-
-                STACK_WIND_COOKIE (frame, afr_sh_data_setattr_cbk,
-				   (void *) (long) i,
-				   priv->children[i],
-				   priv->children[i]->fops->setattr,
-				   &local->loc, &stbuf, valid);
-
-		if (!--call_count)
-			break;
-	}
-
-	return 0;
-}
-
+        ret = syncop_writev(priv->children[i], fd, iovec, count, offset, iobref,
+                            0, NULL, NULL, NULL, NULL);
+        if (ret != iov_length(iovec, count)) {
+            /* write() failed on this sink. unset the corresponding
+               member in sinks[] (which is healed_sinks[] in the
+               caller) so that this server does NOT get considered
+               as successfully healed.
+            */
+            healed_sinks[i] = 0;
+        }
+    }
+    if (iovec)
+        GF_FREE(iovec);
+    if (iobref)
+        iobref_unref(iobref);
 
-int
-afr_sh_data_unlck_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno)
-{
-	afr_local_t * local = NULL;
-	int           call_count = 0;
-	int           child_index = (long) cookie;
-
-	
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"locking inode of %s on child %d failed: %s",
-				local->loc.path, child_index,
-				strerror (op_errno));
-		} else {
-			gf_log (this->name, GF_LOG_TRACE,
-				"inode of %s on child %d locked",
-				local->loc.path, child_index);
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		afr_sh_data_close (frame, this);
-	}
-
-	return 0;
+    return ret;
 }
 
-
-int
-afr_sh_data_unlock (call_frame_t *frame, xlator_t *this)
+static gf_boolean_t
+afr_source_sinks_locked(xlator_t *this, unsigned char *locked_on, int source,
+                        unsigned char *healed_sinks)
 {
-        afr_local_t         *local    = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-	afr_self_heal_t     *sh       = NULL;
+    afr_private_t *priv = this->private;
+    int i = 0;
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-	sh       = &local->self_heal;
+    if (!locked_on[source])
+        return _gf_false;
 
-        GF_ASSERT (!sh->data_lock_held);
+    for (i = 0; i < priv->child_count; i++) {
+        if (healed_sinks[i] && locked_on[i])
+            return _gf_true;
+    }
 
-        int_lock->lock_cbk = afr_sh_data_close;
-        afr_unlock (frame, this);
-
-	return 0;
+    return _gf_false;
 }
 
-
-int
-afr_sh_data_finish (call_frame_t *frame, xlator_t *this)
+static int
+afr_selfheal_data_block(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                        int source, unsigned char *healed_sinks, off_t offset,
+                        size_t size, int type, struct afr_reply *replies)
 {
-	afr_local_t   *local = NULL;
-	afr_self_heal_t *sh = NULL;
-
-	local = frame->local;
-	sh = &local->self_heal;
-
-	gf_log (this->name, GF_LOG_TRACE,
-		"finishing data selfheal of %s", local->loc.path);
+    int ret = -1;
+    afr_private_t *priv = NULL;
+    unsigned char *data_lock = NULL;
+
+    priv = this->private;
+    data_lock = alloca0(priv->child_count);
+
+    ret = afr_selfheal_inodelk(frame, this, fd->inode, this->name, offset, size,
+                               data_lock);
+    {
+        if (!afr_source_sinks_locked(this, data_lock, source, healed_sinks)) {
+            ret = -ENOTCONN;
+            goto unlock;
+        }
 
-        if (!sh->data_lock_held)
-                afr_sh_data_unlock (frame, this);
-        else
-                afr_sh_data_close (frame, this);
+        if (type == AFR_SELFHEAL_DATA_DIFF &&
+            __afr_can_skip_data_block_heal(frame, this, fd, source,
+                                           healed_sinks, offset, size,
+                                           &replies[source].poststat)) {
+            ret = 0;
+            goto unlock;
+        }
 
-	return 0;
+        ret = __afr_selfheal_data_read_write(
+            frame, this, fd, source, healed_sinks, offset, size, replies, type);
+    }
+unlock:
+    afr_selfheal_uninodelk(frame, this, fd->inode, this->name, offset, size,
+                           data_lock);
+    return ret;
 }
 
-
-int
-afr_sh_data_erase_pending_cbk (call_frame_t *frame, void *cookie,
-			       xlator_t *this, int32_t op_ret,
-			       int32_t op_errno, dict_t *xattr)
+static int
+afr_selfheal_data_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                        unsigned char *healed_sinks)
 {
-	int             call_count = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
 
-	call_count = afr_frame_return (frame);
+    local = frame->local;
+    priv = this->private;
 
-	if (call_count == 0)
-		afr_sh_data_finish (frame, this);
-
-	return 0;
-}
+    if (!priv->ensure_durability)
+        return 0;
 
+    AFR_ONLIST(healed_sinks, frame, afr_sh_generic_fop_cbk, fsync, fd, 0, NULL);
 
-int
-afr_sh_data_erase_pending (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              call_count = 0;
-	int              i = 0;
-	dict_t          **erase_xattr = NULL;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	afr_sh_pending_to_delta (priv, sh->xattr, sh->delta_matrix, sh->success,
-                                 priv->child_count, AFR_DATA_TRANSACTION);
-
-	erase_xattr = GF_CALLOC (sizeof (*erase_xattr), priv->child_count,
-                              gf_afr_mt_dict_t);
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->xattr[i]) {
-			call_count++;
-
-			erase_xattr[i] = get_new_dict();
-			dict_ref (erase_xattr[i]);
-		}
-	}
-
-	afr_sh_delta_to_xattr (priv, sh->delta_matrix, erase_xattr,
-			       priv->child_count, AFR_DATA_TRANSACTION);
-
-	local->call_count = call_count;
-	for (i = 0; i < priv->child_count; i++) {
-		if (!erase_xattr[i])
-			continue;
-
-		gf_log (this->name, GF_LOG_TRACE,
-			"erasing pending flags from %s on %s",
-			local->loc.path, priv->children[i]->name);
-
-		STACK_WIND_COOKIE (frame, afr_sh_data_erase_pending_cbk,
-				   (void *) (long) i,
-				   priv->children[i],
-				   priv->children[i]->fops->fxattrop,
-				   sh->healing_fd,
-				   GF_XATTROP_ADD_ARRAY, erase_xattr[i]);
-		if (!--call_count)
-			break;
-	}
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (erase_xattr[i]) {
-			dict_unref (erase_xattr[i]);
-		}
-	}
-	GF_FREE (erase_xattr);
-
-	return 0;
+    for (i = 0; i < priv->child_count; i++)
+        if (healed_sinks[i] && local->replies[i].op_ret != 0)
+            /* fsync() failed. Do NOT consider this server
+               as successfully healed. Mark it so.
+            */
+            healed_sinks[i] = 0;
+    return 0;
 }
 
-
-int
-afr_sh_data_trim_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                      struct iatt *postbuf)
+static int
+afr_data_self_heal_type_get(afr_private_t *priv, unsigned char *healed_sinks,
+                            int source, struct afr_reply *replies)
 {
-	afr_private_t * priv = NULL;
-	afr_local_t * local  = NULL;
-	int              call_count = 0;
-	int              child_index = 0;
-
-	priv = this->private;
-	local = frame->local;
-
-	child_index = (long) cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1)
-			gf_log (this->name, GF_LOG_DEBUG,
-				"ftruncate of %s on subvolume %s failed (%s)",
-				local->loc.path,
-				priv->children[child_index]->name,
-				strerror (op_errno));
-		else
-			gf_log (this->name, GF_LOG_TRACE,
-				"ftruncate of %s on subvolume %s completed",
-				local->loc.path,
-				priv->children[child_index]->name);
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		afr_sh_data_erase_pending (frame, this);
-	}
-
-	return 0;
+    int type = AFR_SELFHEAL_DATA_FULL;
+    int i = 0;
+
+    if (priv->data_self_heal_algorithm == AFR_SELFHEAL_DATA_DYNAMIC) {
+        type = AFR_SELFHEAL_DATA_FULL;
+        for (i = 0; i < priv->child_count; i++) {
+            if (!healed_sinks[i] && i != source)
+                continue;
+            if (replies[i].poststat.ia_size) {
+                type = AFR_SELFHEAL_DATA_DIFF;
+                break;
+            }
+        }
+    } else {
+        type = priv->data_self_heal_algorithm;
+    }
+    return type;
 }
 
-
-int
-afr_sh_data_trim_sinks (call_frame_t *frame, xlator_t *this)
+static int
+afr_selfheal_data_do(call_frame_t *frame, xlator_t *this, fd_t *fd, int source,
+                     unsigned char *healed_sinks, struct afr_reply *replies)
 {
-	afr_private_t * priv = NULL;
-	afr_local_t * local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	int             *sources = NULL;
-	int              call_count = 0;
-	int              i = 0;
-
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	sources = sh->sources;
-	call_count = sh->active_sinks;
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (sources[i] || !local->child_up[i])
-			continue;
-
-		STACK_WIND_COOKIE (frame, afr_sh_data_trim_cbk,
-				   (void *) (long) i,
-				   priv->children[i], 
-				   priv->children[i]->fops->ftruncate,
-				   sh->healing_fd, sh->file_size); 
-
-		if (!--call_count)
-			break;
-	}
-
-	return 0;
-}
+    afr_private_t *priv = NULL;
+    off_t off = 0;
+    size_t block = 0;
+    int type = AFR_SELFHEAL_DATA_FULL;
+    int ret = -1;
+    call_frame_t *iter_frame = NULL;
+    unsigned char arbiter_sink_status = 0;
+
+    gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_SELF_HEAL_INFO,
+           "performing data selfheal on %s", uuid_utoa(fd->inode->gfid));
+
+    priv = this->private;
+    if (priv->arbiter_count) {
+        arbiter_sink_status = healed_sinks[ARBITER_BRICK_INDEX];
+        healed_sinks[ARBITER_BRICK_INDEX] = 0;
+    }
+
+    block = 128 * 1024 * priv->data_self_heal_window_size;
+
+    type = afr_data_self_heal_type_get(priv, healed_sinks, source, replies);
+
+    iter_frame = afr_copy_frame(frame);
+    if (!iter_frame) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    for (off = 0; off < replies[source].poststat.ia_size; off += block) {
+        if (AFR_COUNT(healed_sinks, priv->child_count) == 0) {
+            ret = -ENOTCONN;
+            goto out;
+        }
 
+        ret = afr_selfheal_data_block(iter_frame, this, fd, source,
+                                      healed_sinks, off, block, type, replies);
+        if (ret < 0)
+            goto out;
 
-static struct afr_sh_algorithm *
-sh_algo_from_name (xlator_t *this, char *name)
-{
-        int i = 0;
+        AFR_STACK_RESET(iter_frame);
+        if (iter_frame->local == NULL) {
+            ret = -ENOTCONN;
+            goto out;
+        }
+    }
 
-        while (afr_self_heal_algorithms[i].name) {
-                if (!strcmp (name, afr_self_heal_algorithms[i].name)) {
-                        return &afr_self_heal_algorithms[i];
-                }
+    ret = afr_selfheal_data_fsync(frame, this, fd, healed_sinks);
 
-                i++;
-        }
+out:
+    if (arbiter_sink_status)
+        healed_sinks[ARBITER_BRICK_INDEX] = arbiter_sink_status;
 
-        return NULL;
+    if (iter_frame)
+        AFR_STACK_DESTROY(iter_frame);
+    return ret;
 }
 
-
 static int
-sh_zero_byte_files_exist (afr_self_heal_t *sh, int child_count)
+__afr_selfheal_truncate_sinks(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                              unsigned char *healed_sinks, uint64_t size)
 {
-        int i;
-        int ret = 0;
-
-        for (i = 0; i < child_count; i++) {
-                if (sh->buf[i].ia_size == 0) {
-                        ret = 1;
-                        break;
-                }
-        }
-
-        return ret;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+
+    local = frame->local;
+    priv = this->private;
+
+    /* This will send truncate on the arbiter brick as well if it is marked as
+     * sink. If changelog is enabled on the volume it captures truncate as a
+     * data transactions on the arbiter brick. This will help geo-rep to
+     * properly sync the data from master to slave if arbiter is the ACTIVE
+     * brick during syncing and which had got some entries healed for data as
+     * part of self heal.
+     */
+    AFR_ONLIST(healed_sinks, frame, afr_sh_generic_fop_cbk, ftruncate, fd, size,
+               NULL);
+
+    for (i = 0; i < priv->child_count; i++)
+        if (healed_sinks[i] && local->replies[i].op_ret == -1)
+            /* truncate() failed. Do NOT consider this server
+               as successfully healed. Mark it so.
+            */
+            healed_sinks[i] = 0;
+
+    return 0;
 }
 
-
-struct afr_sh_algorithm *
-afr_sh_data_pick_algo (call_frame_t *frame, xlator_t *this)
+gf_boolean_t
+afr_has_source_witnesses(xlator_t *this, unsigned char *sources,
+                         uint64_t *witness)
 {
-        afr_private_t *           priv  = NULL;
-        struct afr_sh_algorithm * algo  = NULL;
-        afr_local_t *             local = NULL;
-        afr_self_heal_t *         sh    = NULL;
-
-        priv  = this->private;
-        local = frame->local;
-        sh    = &local->self_heal;
-        algo  = sh_algo_from_name (this, priv->data_self_heal_algorithm);
-
-        if (algo == NULL) {
-                /* option not set, so fall back on heuristics */
-
-                if ((local->enoent_count != 0)
-                    || sh_zero_byte_files_exist (sh, priv->child_count)
-                    || (sh->file_size <= (priv->data_self_heal_window_size * this->ctx->page_size))) {
-
-                        /*
-                         * If the file does not exist on one of the subvolumes,
-                         * or a zero-byte file exists (created by entry self-heal)
-                         * the entire content has to be copied anyway, so there
-                         * is no benefit from using the "diff" algorithm.
-                         *
-                         * If the file size is about the same as page size,
-                         * the entire file can be read and written with a few
-                         * (pipelined) STACK_WINDs, which will be faster
-                         * than "diff" which has to read checksums and then
-                         * read and write.
-                         */
-
-                        algo = sh_algo_from_name (this, "full");
-
-                } else {
-                        algo = sh_algo_from_name (this, "diff");
-                }
-        }
-
-        return algo;
-}
+    int i = 0;
+    afr_private_t *priv = NULL;
 
+    priv = this->private;
 
-int
-afr_sh_data_sync_prepare (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              active_sinks = 0;
-	int              source = 0;
-	int              i = 0;
-
-        struct afr_sh_algorithm *sh_algo = NULL;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	source = sh->source;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->sources[i] == 0 && local->child_up[i] == 1) {
-			active_sinks++;
-			sh->success[i] = 1;
-		}
-	}
-	sh->success[source] = 1;
-
-	if (active_sinks == 0) {
-		gf_log (this->name, GF_LOG_TRACE,
-			"no active sinks for performing self-heal on file %s",
-			local->loc.path);
-		afr_sh_data_finish (frame, this);
-		return 0;
-	}
-	sh->active_sinks = active_sinks;
-
-	gf_log (this->name, GF_LOG_DEBUG,
-		"self-healing file %s from subvolume %s to %d other",
-		local->loc.path, priv->children[source]->name, active_sinks);
-
-        sh->algo_completion_cbk = afr_sh_data_trim_sinks;
-        sh->algo_abort_cbk      = afr_sh_data_finish;
-
-        sh_algo = afr_sh_data_pick_algo (frame, this);
-
-        sh_algo->fn (frame, this);
-
-	return 0;
+    for (i = 0; i < priv->child_count; i++) {
+        if (sources[i] && witness[i])
+            return _gf_true;
+    }
+    return _gf_false;
 }
 
-
-int
-afr_sh_data_fix (call_frame_t *frame, xlator_t *this)
+static gf_boolean_t
+afr_does_size_mismatch(xlator_t *this, unsigned char *sources,
+                       struct afr_reply *replies)
 {
-	afr_local_t     *local      = NULL;
-        afr_local_t *    orig_local = NULL;
+    int i = 0;
+    afr_private_t *priv = NULL;
+    struct iatt *min = NULL;
+    struct iatt *max = NULL;
 
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              nsources = 0;
-	int              source = 0;
-	int              i = 0;
+    priv = this->private;
 
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid)
+            continue;
 
-	afr_sh_build_pending_matrix (priv, sh->pending_matrix, sh->xattr, 
-				     priv->child_count, AFR_DATA_TRANSACTION);
+        if (replies[i].op_ret < 0)
+            continue;
 
-	afr_sh_print_pending_matrix (sh->pending_matrix, this);
+        if (!sources[i])
+            continue;
 
-	nsources = afr_sh_mark_sources (sh, priv->child_count,
-                                        AFR_SELF_HEAL_DATA);
+        if (AFR_IS_ARBITER_BRICK(priv, i) && (replies[i].poststat.ia_size == 0))
+            continue;
 
-	afr_sh_supress_errenous_children (sh->sources, sh->child_errno,
-					  priv->child_count);
+        if (!min)
+            min = &replies[i].poststat;
 
-        if (nsources == 0) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "No self-heal needed for %s",
-                        local->loc.path);
+        if (!max)
+            max = &replies[i].poststat;
 
-                afr_sh_data_finish (frame, this);
-                return 0;
-        }
-
-	if ((nsources == -1)
-	    && (priv->favorite_child != -1)
-	    && (sh->child_errno[priv->favorite_child] == 0)) {
-
-		gf_log (this->name, GF_LOG_DEBUG,
-			"Picking favorite child %s as authentic source to resolve conflicting data of %s",
-			priv->children[priv->favorite_child]->name,
-			local->loc.path);
-
-		sh->sources[priv->favorite_child] = 1;
-
-		nsources = afr_sh_source_count (sh->sources,
-						priv->child_count);
-	}
-
-	if (nsources == -1) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Unable to self-heal contents of '%s' (possible split-brain). "
-                        "Please delete the file from all but the preferred "
-                        "subvolume.", local->loc.path);
-
-		local->govinda_gOvinda = 1;
-
-		afr_sh_data_finish (frame, this);
-		return 0;
-	}
-
-	source = afr_sh_select_source (sh->sources, priv->child_count);
-        
-        if (source == -1) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "No active sources found.");
-
-                afr_sh_data_finish (frame, this);
-                return 0;
-        }
-
-	sh->source     = source;
-	sh->block_size = 65536;
-	sh->file_size  = sh->buf[source].ia_size;
-
-	if (FILE_HAS_HOLES (&sh->buf[source]))
-		sh->file_has_holes = 1;
-
-        orig_local = sh->orig_frame->local;
-        orig_local->cont.lookup.buf.ia_size = sh->buf[source].ia_size;
+        if (min->ia_size > replies[i].poststat.ia_size)
+            min = &replies[i].poststat;
 
-	/* detect changes not visible through pending flags -- JIC */
-	for (i = 0; i < priv->child_count; i++) {
-		if (i == source || sh->child_errno[i])
-			continue;
+        if (max->ia_size < replies[i].poststat.ia_size)
+            max = &replies[i].poststat;
+    }
 
-		if (SIZE_DIFFERS (&sh->buf[i], &sh->buf[source]))
-			sh->sources[i] = 0;
-	}
+    if (min && max) {
+        if (min->ia_size != max->ia_size)
+            return _gf_true;
+    }
 
-        afr_set_read_child (this, local->loc.inode, sh->source);
-
-	/*
-	   quick-read might have read the file, so send xattr from
-	   the source subvolume (http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=815)
-	*/
-
-	dict_unref (orig_local->cont.lookup.xattr);
-        if (orig_local->cont.lookup.xattrs)
-                orig_local->cont.lookup.xattr = dict_ref (orig_local->cont.lookup.xattrs[sh->source]);
-
-        if (sh->background) {
-                sh->unwind (sh->orig_frame, this);
-                sh->unwound = _gf_true;
-        }
-
-	afr_sh_data_sync_prepare (frame, this);
-
-	return 0;
+    return _gf_false;
 }
 
-
-int
-afr_self_heal_get_source (xlator_t *this, afr_local_t *local, dict_t **xattr)
+static void
+afr_mark_biggest_witness_as_source(xlator_t *this, unsigned char *sources,
+                                   uint64_t *witness)
 {
-	afr_self_heal_t *sh   = NULL;
-	afr_private_t   *priv = NULL;
-
-	int              source = 0;
-	int              i = 0;
-
-	sh   = &local->self_heal;
-	priv = this->private;
-
-	sh->pending_matrix = GF_CALLOC (sizeof (int32_t *), priv->child_count,
-                                        gf_afr_mt_int32_t);
-	for (i = 0; i < priv->child_count; i++) {
-		sh->pending_matrix[i] = GF_CALLOC (sizeof (int32_t),
-						priv->child_count,
-                                                gf_afr_mt_int32_t);
-	}
-
-	sh->sources = GF_CALLOC (priv->child_count, sizeof (*sh->sources),
-                                 gf_afr_mt_int32_t);
-
-	afr_sh_build_pending_matrix (priv, sh->pending_matrix, xattr,
-				     priv->child_count, AFR_DATA_TRANSACTION);
-
-	(void)afr_sh_mark_sources (sh, priv->child_count, AFR_SELF_HEAL_DATA);
-
-	source = afr_sh_select_source (sh->sources, priv->child_count);
-
-	return source;
+    int i = 0;
+    afr_private_t *priv = NULL;
+    uint64_t biggest_witness = 0;
+
+    priv = this->private;
+    /* Find source with biggest witness count */
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+        if (biggest_witness < witness[i])
+            biggest_witness = witness[i];
+    }
+
+    /* Mark files with less witness count as not source */
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+        if (witness[i] < biggest_witness)
+            sources[i] = 0;
+    }
+
+    return;
 }
 
-
-int
-afr_sh_data_fstat_cbk (call_frame_t *frame, void *cookie,
-                       xlator_t *this, int32_t op_ret, int32_t op_errno,
-                       struct iatt *buf)
+/* This is a tie breaker function. Only one source be assigned here */
+static void
+afr_mark_newest_file_as_source(xlator_t *this, unsigned char *sources,
+                               struct afr_reply *replies)
 {
-	afr_private_t   *priv  = NULL;
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-
-	int call_count  = -1;
-	int child_index = (long) cookie;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret != -1) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "fstat of %s on %s succeeded",
-                                local->loc.path,
-                                priv->children[child_index]->name);
-
-			sh->buf[child_index] = *buf;
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		afr_sh_data_fix (frame, this);
-	}
+    int i = 0;
+    afr_private_t *priv = NULL;
+    int source = -1;
+    uint32_t max_ctime = 0;
+
+    priv = this->private;
+    /* Find source with latest ctime */
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+
+        if (max_ctime <= replies[i].poststat.ia_ctime) {
+            source = i;
+            max_ctime = replies[i].poststat.ia_ctime;
+        }
+    }
 
-	return 0;
+    /* Only mark one of the files as source to break ties */
+    memset(sources, 0, sizeof(*sources) * priv->child_count);
+    sources[source] = 1;
 }
 
-
-int
-afr_sh_data_fstat (call_frame_t *frame, xlator_t *this)
+static int
+__afr_selfheal_data_finalize_source(
+    call_frame_t *frame, xlator_t *this, inode_t *inode, unsigned char *sources,
+    unsigned char *sinks, unsigned char *healed_sinks, unsigned char *locked_on,
+    unsigned char *undid_pending, struct afr_reply *replies, uint64_t *witness)
 {
-	afr_self_heal_t *sh    = NULL; 
-	afr_local_t     *local = NULL;
-	afr_private_t   *priv  = NULL;
-
-	int call_count = 0;
-	int i = 0;
-
-	priv  = this->private;
-	local = frame->local;
-	sh    = &local->self_heal;
-
-	call_count = afr_up_children_count (priv->child_count,
-                                            local->child_up);
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_sh_data_fstat_cbk,
-					   (void *) (long) i,
-					   priv->children[i], 
-					   priv->children[i]->fops->fstat,
-					   sh->healing_fd);
-
-			if (!--call_count)
-				break;
-		}
-	}
+    afr_private_t *priv = NULL;
+    int source = -1;
+    int sources_count = 0;
+    priv = this->private;
+
+    sources_count = AFR_COUNT(sources, priv->child_count);
+
+    if ((AFR_CMP(locked_on, healed_sinks, priv->child_count) == 0) ||
+        !sources_count) {
+        /* split brain */
+        source = afr_mark_split_brain_source_sinks(
+            frame, this, inode, sources, sinks, healed_sinks, locked_on,
+            replies, AFR_DATA_TRANSACTION);
+        if (source < 0) {
+            gf_event(EVENT_AFR_SPLIT_BRAIN,
+                     "client-pid=%d;"
+                     "subvol=%s;type=data;"
+                     "file=%s",
+                     this->ctx->cmd_args.client_pid, this->name,
+                     uuid_utoa(inode->gfid));
+            return -EIO;
+        }
 
-	return 0;
+        _afr_fav_child_reset_sink_xattrs(
+            frame, this, inode, source, healed_sinks, undid_pending,
+            AFR_DATA_TRANSACTION, locked_on, replies);
+        goto out;
+    }
+
+    /* No split brain at this point. If we were called from
+     * afr_heal_splitbrain_file(), abort.*/
+    if (afr_dict_contains_heal_op(frame))
+        return -EIO;
+
+    /* If there are no witnesses/size-mismatches on sources we are done*/
+    if (!afr_does_size_mismatch(this, sources, replies) &&
+        !afr_has_source_witnesses(this, sources, witness))
+        goto out;
+
+    afr_mark_largest_file_as_source(this, sources, replies);
+    afr_mark_biggest_witness_as_source(this, sources, witness);
+    afr_mark_newest_file_as_source(this, sources, replies);
+    if (priv->arbiter_count)
+        /* Choose non-arbiter brick as source for empty files. */
+        afr_mark_source_sinks_if_file_empty(this, sources, sinks, healed_sinks,
+                                            locked_on, replies,
+                                            AFR_DATA_TRANSACTION);
+
+out:
+    afr_mark_active_sinks(this, sources, locked_on, healed_sinks);
+    source = afr_choose_source_by_policy(priv, sources, AFR_DATA_TRANSACTION);
+
+    return source;
 }
 
-
+/*
+ * __afr_selfheal_data_prepare:
+ *
+ * This function inspects the on-disk xattrs and determines which subvols
+ * are sources and sinks.
+ *
+ * The return value is the index of the subvolume to be used as the source
+ * for self-healing, or -1 if no healing is necessary/split brain.
+ */
 int
-afr_sh_data_fxattrop_cbk (call_frame_t *frame, void *cookie,
-                          xlator_t *this, int32_t op_ret, int32_t op_errno,
-                          dict_t *xattr)
+__afr_selfheal_data_prepare(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                            unsigned char *locked_on, unsigned char *sources,
+                            unsigned char *sinks, unsigned char *healed_sinks,
+                            unsigned char *undid_pending,
+                            struct afr_reply *replies, unsigned char *pflag)
 {
-	afr_private_t   *priv  = NULL;
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-
-	int call_count  = -1;
-	int child_index = (long) cookie;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
+    int ret = -1;
+    int source = -1;
+    afr_private_t *priv = NULL;
+    uint64_t *witness = NULL;
 
-	LOCK (&frame->lock);
-	{
-		if (op_ret != -1) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "fxattrop of %s on %s succeeded",
-                                local->loc.path,
-                                priv->children[child_index]->name);
+    priv = this->private;
 
-			sh->xattr[child_index] = dict_ref (xattr);
-		}
-	}
-	UNLOCK (&frame->lock);
+    ret = afr_selfheal_unlocked_discover(frame, inode, inode->gfid, replies);
 
-	call_count = afr_frame_return (frame);
+    if (ret)
+        return ret;
 
-	if (call_count == 0) {
-		afr_sh_data_fstat (frame, this);
-	}
+    witness = alloca0(priv->child_count * sizeof(*witness));
+    ret = afr_selfheal_find_direction(frame, this, replies,
+                                      AFR_DATA_TRANSACTION, locked_on, sources,
+                                      sinks, witness, pflag);
+    if (ret)
+        return ret;
 
-	return 0;
+    /* Initialize the healed_sinks[] array optimistically to
+       the intersection of to-be-healed (i.e sinks[]) and
+       the list of servers which are up (i.e locked_on[]).
+       As we encounter failures in the healing process, we
+       will unmark the respective servers in the healed_sinks[]
+       array.
+    */
+    AFR_INTERSECT(healed_sinks, sinks, locked_on, priv->child_count);
+
+    source = __afr_selfheal_data_finalize_source(
+        frame, this, inode, sources, sinks, healed_sinks, locked_on,
+        undid_pending, replies, witness);
+    if (source < 0)
+        return -EIO;
+
+    return source;
 }
 
-
-int
-afr_sh_data_fxattrop (call_frame_t *frame, xlator_t *this)
+static int
+__afr_selfheal_data(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                    unsigned char *locked_on)
 {
-	afr_self_heal_t *sh    = NULL;
-	afr_local_t     *local = NULL;
-	afr_private_t   *priv  = NULL;
-	dict_t          *xattr_req = NULL;
-
-        int32_t zero_pending[3] = {0, 0, 0};
-
-	int call_count = 0;
-	int i = 0;
-	int ret = 0;
-
-	priv  = this->private;
-	local = frame->local;
-	sh    = &local->self_heal;
-
-	call_count = afr_up_children_count (priv->child_count,
-                                            local->child_up);
-
-	local->call_count = call_count;
-
-	xattr_req = dict_new();
-	if (xattr_req) {
-                for (i = 0; i < priv->child_count; i++) {
-                        ret = dict_set_static_bin (xattr_req, priv->pending_key[i],
-                                                   zero_pending, 3 * sizeof(int32_t));
-                        if (ret < 0)
-				gf_log (this->name, GF_LOG_WARNING,
-					"Unable to set dict value");
-                }
+    afr_private_t *priv = NULL;
+    int ret = -1;
+    unsigned char *sources = NULL;
+    unsigned char *sinks = NULL;
+    unsigned char *data_lock = NULL;
+    unsigned char *healed_sinks = NULL;
+    unsigned char *undid_pending = NULL;
+    struct afr_reply *locked_replies = NULL;
+    int source = -1;
+    gf_boolean_t did_sh = _gf_true;
+    gf_boolean_t is_arbiter_the_only_sink = _gf_false;
+    gf_boolean_t empty_file = _gf_false;
+
+    priv = this->private;
+
+    sources = alloca0(priv->child_count);
+    sinks = alloca0(priv->child_count);
+    healed_sinks = alloca0(priv->child_count);
+    data_lock = alloca0(priv->child_count);
+    undid_pending = alloca0(priv->child_count);
+
+    locked_replies = alloca0(sizeof(*locked_replies) * priv->child_count);
+
+    ret = afr_selfheal_inodelk(frame, this, fd->inode, this->name, 0, 0,
+                               data_lock);
+    {
+        if (ret < priv->child_count) {
+            gf_msg_debug(this->name, 0,
+                         "%s: Skipping "
+                         "self-heal as only %d number "
+                         "of subvolumes "
+                         "could be locked",
+                         uuid_utoa(fd->inode->gfid), ret);
+            ret = -ENOTCONN;
+            goto unlock;
         }
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_sh_data_fxattrop_cbk,
-					   (void *) (long) i,
-					   priv->children[i],
-					   priv->children[i]->fops->fxattrop,
-					   sh->healing_fd, GF_XATTROP_ADD_ARRAY,
-                                           xattr_req);
-
-			if (!--call_count)
-				break;
-		}
-	}
-
-	if (xattr_req)
-		dict_unref (xattr_req);
+        ret = __afr_selfheal_data_prepare(frame, this, fd->inode, data_lock,
+                                          sources, sinks, healed_sinks,
+                                          undid_pending, locked_replies, NULL);
+        if (ret < 0)
+            goto unlock;
 
-	return 0;
-}
-
-
-int
-afr_sh_data_lock_rec (call_frame_t *frame, xlator_t *this);
-
-int
-afr_sh_data_post_nonblocking_inodelk_cbk (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
+        if (AFR_COUNT(healed_sinks, priv->child_count) == 0) {
+            did_sh = _gf_false;
+            goto unlock;
+        }
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+        source = ret;
 
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Non Blocking inodelks failed.");
-                afr_sh_data_done (frame, this);
-        } else {
+        if (AFR_IS_ARBITER_BRICK(priv, source)) {
+            empty_file = afr_is_file_empty_on_all_children(priv,
+                                                           locked_replies);
+            if (empty_file)
+                goto restore_time;
 
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Non Blocking inodelks done. Proceeding to FOP");
-                afr_sh_data_fxattrop (frame, this);
+            did_sh = _gf_false;
+            goto unlock;
         }
 
-        return 0;
+        ret = __afr_selfheal_truncate_sinks(
+            frame, this, fd, healed_sinks,
+            locked_replies[source].poststat.ia_size);
+        if (ret < 0)
+            goto unlock;
+
+        if (priv->arbiter_count &&
+            AFR_COUNT(healed_sinks, priv->child_count) == 1 &&
+            healed_sinks[ARBITER_BRICK_INDEX]) {
+            is_arbiter_the_only_sink = _gf_true;
+            goto restore_time;
+        }
+        ret = 0;
+    }
+unlock:
+    afr_selfheal_uninodelk(frame, this, fd->inode, this->name, 0, 0, data_lock);
+    if (ret < 0)
+        goto out;
+
+    if (!did_sh)
+        goto out;
+
+    ret = afr_selfheal_data_do(frame, this, fd, source, healed_sinks,
+                               locked_replies);
+    if (ret)
+        goto out;
+restore_time:
+    afr_selfheal_restore_time(frame, this, fd->inode, source, healed_sinks,
+                              locked_replies);
+
+    if (!is_arbiter_the_only_sink && !empty_file) {
+        ret = afr_selfheal_inodelk(frame, this, fd->inode, this->name, 0, 0,
+                                   data_lock);
+        if (ret < priv->child_count) {
+            ret = -ENOTCONN;
+            did_sh = _gf_false;
+            goto skip_undo_pending;
+        }
+    }
+    ret = afr_selfheal_undo_pending(
+        frame, this, fd->inode, sources, sinks, healed_sinks, undid_pending,
+        AFR_DATA_TRANSACTION, locked_replies, data_lock);
+skip_undo_pending:
+    afr_selfheal_uninodelk(frame, this, fd->inode, this->name, 0, 0, data_lock);
+out:
+
+    if (did_sh)
+        afr_log_selfheal(fd->inode->gfid, this, ret, "data", source, sources,
+                         healed_sinks);
+    else
+        ret = 1;
+
+    if (locked_replies)
+        afr_replies_wipe(locked_replies, priv->child_count);
+
+    return ret;
 }
 
 int
-afr_sh_data_lock_rec (call_frame_t *frame, xlator_t *this)
+afr_selfheal_data_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno, fd_t *fd,
+                           dict_t *xdata)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-        afr_self_heal_t     *sh       = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-        sh       = &local->self_heal;
+    afr_local_t *local = NULL;
+    int i = (long)cookie;
 
-        int_lock->transaction_lk_type = AFR_SELFHEAL_LK;
-        int_lock->selfheal_lk_type    = AFR_DATA_SELF_HEAL_LK;
+    local = frame->local;
 
-        afr_set_lock_number (frame, this);
+    local->replies[i].valid = 1;
+    local->replies[i].op_ret = op_ret;
+    local->replies[i].op_errno = op_errno;
 
-        int_lock->lk_flock.l_start = 0;
-        int_lock->lk_flock.l_len   = 0;
-        int_lock->lk_flock.l_type  = F_WRLCK;
-        int_lock->lock_cbk         = afr_sh_data_post_nonblocking_inodelk_cbk;
+    syncbarrier_wake(&local->barrier);
 
-        afr_nonblocking_inodelk (frame, this);
-
-
-	return 0;
+    return 0;
 }
 
-
 int
-afr_sh_data_lock (call_frame_t *frame, xlator_t *this)
+afr_selfheal_data_open(xlator_t *this, inode_t *inode, fd_t **fd)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	afr_self_heal_t * sh  = NULL;
-
-
-	local = frame->local;
-	sh    = &local->self_heal;
-	priv  = this->private;
-
-        if (sh->data_lock_held) {
-                /* caller has held the lock already,
-                   so skip locking */
-
-                afr_sh_data_fxattrop (frame, this);
-                return 0;
+    int ret = 0;
+    fd_t *fd_tmp = NULL;
+    loc_t loc = {
+        0,
+    };
+    call_frame_t *frame = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+
+    priv = this->private;
+
+    fd_tmp = fd_create(inode, 0);
+    if (!fd_tmp)
+        return -ENOMEM;
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    frame = afr_frame_create(this, &ret);
+    if (!frame) {
+        ret = -ret;
+        fd_unref(fd_tmp);
+        goto out;
+    }
+    local = frame->local;
+
+    AFR_ONLIST(local->child_up, frame, afr_selfheal_data_open_cbk, open, &loc,
+               O_RDWR | O_LARGEFILE, fd_tmp, NULL);
+
+    ret = -ENOTCONN;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
+
+        if (local->replies[i].op_ret < 0) {
+            ret = -local->replies[i].op_errno;
+            continue;
         }
 
-        return afr_sh_data_lock_rec (frame, this);
+        ret = 0;
+        break;
+    }
+
+    if (ret < 0) {
+        fd_unref(fd_tmp);
+        goto out;
+    } else {
+        fd_bind(fd_tmp);
+    }
+
+    *fd = fd_tmp;
+out:
+    loc_wipe(&loc);
+    if (frame)
+        AFR_STACK_DESTROY(frame);
+    return ret;
 }
 
-
-int
-afr_sh_data_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno, fd_t *fd)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              call_count = 0;
-	int              child_index = 0;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	child_index = (long) cookie;
-
-	/* TODO: some of the open's might fail.
-	   In that case, modify cleanup fn to send flush on those 
-	   fd's which are already open */
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			gf_log (this->name, GF_LOG_TRACE,
-				"open of %s failed on child %s (%s)",
-				local->loc.path,
-				priv->children[child_index]->name,
-				strerror (op_errno));
-			sh->op_failed = 1;
-		}
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "open of %s succeeded on child %s",
-                        local->loc.path,
-                        priv->children[child_index]->name);
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		if (sh->op_failed) {
-			afr_sh_data_finish (frame, this);
-			return 0;
-		}
-
-		gf_log (this->name, GF_LOG_TRACE,
-			"fd for %s opened, commencing sync",
-			local->loc.path);
-
-                afr_sh_data_lock (frame, this);
-	}
-
-	return 0;
-}
-
-
 int
-afr_sh_data_open (call_frame_t *frame, xlator_t *this)
+afr_selfheal_data(call_frame_t *frame, xlator_t *this, fd_t *fd)
 {
-	int i = 0;
-	int call_count = 0;
-
-	fd_t *fd = NULL;
-
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	afr_self_heal_t *sh = NULL;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-        if (sh->healing_fd_opened) {
-                /* caller has opened the fd for us already, so skip open */
-
-                afr_sh_data_lock (frame, this);
-                return 0;
+    afr_private_t *priv = NULL;
+    unsigned char *locked_on = NULL;
+    int ret = 0;
+    inode_t *inode = fd->inode;
+
+    priv = this->private;
+
+    locked_on = alloca0(priv->child_count);
+
+    ret = afr_selfheal_tie_breaker_inodelk(frame, this, inode, priv->sh_domain,
+                                           0, 0, locked_on);
+    {
+        if (ret < priv->child_count) {
+            gf_msg_debug(this->name, 0,
+                         "%s: Skipping "
+                         "self-heal as only %d number of "
+                         "subvolumes could be locked",
+                         uuid_utoa(fd->inode->gfid), ret);
+            /* Either less than two subvols available, or another
+               selfheal (from another server) is in progress. Skip
+               for now in any case there isn't anything to do.
+            */
+            ret = -ENOTCONN;
+            goto unlock;
         }
 
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-	local->call_count = call_count;
-
-	fd = fd_create (local->loc.inode, frame->root->pid);
-	sh->healing_fd = fd;
+        ret = __afr_selfheal_data(frame, this, fd, locked_on);
+    }
+unlock:
+    afr_selfheal_uninodelk(frame, this, inode, priv->sh_domain, 0, 0,
+                           locked_on);
 
-	/* open sinks */
-	for (i = 0; i < priv->child_count; i++) {
-		if(!local->child_up[i])
-			continue;
-
-		STACK_WIND_COOKIE (frame, afr_sh_data_open_cbk,
-				   (void *) (long) i,
-				   priv->children[i], 
-				   priv->children[i]->fops->open,
-				   &local->loc, 
-				   O_RDWR|O_LARGEFILE, fd, 0); 
-
-		if (!--call_count)
-			break;
-	}
-
-	return 0;
+    return ret;
 }
-
-
-int
-afr_self_heal_data (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t   *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t *priv = this->private;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-
-	if (sh->need_data_self_heal && priv->data_self_heal) {
-		afr_sh_data_open (frame, this);
-	} else {
-		gf_log (this->name, GF_LOG_TRACE,
-			"not doing data self heal on %s",
-			local->loc.path);
-		afr_sh_data_done (frame, this);
-	}
-
-	return 0;
-}
-
diff --git a/xlators/cluster/afr/src/afr-self-heal-entry.c b/xlators/cluster/afr/src/afr-self-heal-entry.c
index db12c3b8346..64893f441e3 100644
--- a/xlators/cluster/afr/src/afr-self-heal-entry.c
+++ b/xlators/cluster/afr/src/afr-self-heal-entry.c
@@ -1,2273 +1,1276 @@
 /*
-  Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
-#include <libgen.h>
-#include <unistd.h>
-#include <fnmatch.h>
-#include <sys/time.h>
-#include <stdlib.h>
-#include <signal.h>
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "inode.h"
 #include "afr.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-#include "byte-order.h"
-
-#include "afr-transaction.h"
 #include "afr-self-heal.h"
-#include "afr-self-heal-common.h"
-
-int
-afr_sh_post_nonblocking_entrylk_cbk (call_frame_t *frame, xlator_t *this);
-
-int
-afr_sh_entry_done (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	/* 
-	   TODO: cleanup sh->* 
-	*/
-
-        if (sh->healing_fd)
-                fd_unref (sh->healing_fd);
-        sh->healing_fd = NULL;
-
-/*         for (i = 0; i < priv->child_count; i++) { */
-/*                 sh->locked_nodes[i] = 0; */
-/*         } */
-
-	gf_log (this->name, GF_LOG_TRACE,
-		"self heal of %s completed",
-		local->loc.path);
-
-	sh->completion_cbk (frame, this);
-
-	return 0;
-}
-
-
-int
-afr_sh_entry_unlock (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t         *local    = NULL;
-        afr_internal_lock_t *int_lock = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        int_lock->lock_cbk = afr_sh_entry_done;
-        afr_unlock (frame, this);
-
-        return 0;
-}
-
+#include <glusterfs/byte-order.h>
+#include "afr-transaction.h"
+#include "afr-messages.h"
+#include <glusterfs/syncop-utils.h>
+#include <glusterfs/events.h>
 
 int
-afr_sh_entry_finish (call_frame_t *frame, xlator_t *this)
+afr_selfheal_entry_anon_inode(xlator_t *this, inode_t *dir, const char *name,
+                              inode_t *inode, int child,
+                              struct afr_reply *replies,
+                              gf_boolean_t *anon_inode)
 {
-	afr_local_t   *local = NULL;
-
-	local = frame->local;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    xlator_t *subvol = NULL;
+    int ret = 0;
+    int i = 0;
+    char g[64] = {0};
+    unsigned char *lookup_success = NULL;
+    call_frame_t *frame = NULL;
+    loc_t loc2 = {
+        0,
+    };
+    loc_t loc = {
+        0,
+    };
+
+    priv = this->private;
+    subvol = priv->children[child];
+    lookup_success = alloca0(priv->child_count);
+    uuid_utoa_r(replies[child].poststat.ia_gfid, g);
+    loc.inode = inode_new(inode->table);
+    if (!loc.inode) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    if (replies[child].poststat.ia_type == IA_IFDIR) {
+        /* This directory may have sub-directory hierarchy which may need to
+         * be preserved for subsequent heals. So unconditionally move the
+         * directory to anonymous-inode directory*/
+        *anon_inode = _gf_true;
+        goto anon_inode;
+    }
+
+    frame = afr_frame_create(this, &ret);
+    if (!frame) {
+        ret = -ret;
+        goto out;
+    }
+    local = frame->local;
+    gf_uuid_copy(loc.gfid, replies[child].poststat.ia_gfid);
+    AFR_ONLIST(local->child_up, frame, afr_selfheal_discover_cbk, lookup, &loc,
+               NULL);
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->replies[i].op_ret == 0) {
+            lookup_success[i] = 1;
+        } else if (local->replies[i].op_errno != ENOENT &&
+                   local->replies[i].op_errno != ESTALE) {
+            ret = -local->replies[i].op_errno;
+        }
+    }
 
-	gf_log (this->name, GF_LOG_TRACE,
-		"finishing entry selfheal of %s", local->loc.path);
+    if (priv->quorum_count) {
+        if (afr_has_quorum(lookup_success, this, NULL)) {
+            *anon_inode = _gf_true;
+        }
+    } else if (AFR_COUNT(lookup_success, priv->child_count) > 1) {
+        *anon_inode = _gf_true;
+    } else if (ret) {
+        goto out;
+    }
+
+anon_inode:
+    if (!*anon_inode) {
+        ret = 0;
+        goto out;
+    }
+
+    loc.parent = inode_ref(dir);
+    gf_uuid_copy(loc.pargfid, dir->gfid);
+    loc.name = name;
+
+    ret = afr_anon_inode_create(this, child, &loc2.parent);
+    if (ret < 0)
+        goto out;
+
+    loc2.name = g;
+    ret = syncop_rename(subvol, &loc, &loc2, NULL, NULL);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, AFR_MSG_EXPUNGING_FILE_OR_DIR,
+               "Rename to %s dir %s/%s (%s) on %s failed",
+               priv->anon_inode_name, uuid_utoa(dir->gfid), name, g,
+               subvol->name);
+    } else {
+        gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_EXPUNGING_FILE_OR_DIR,
+               "Rename to %s dir %s/%s (%s) on %s successful",
+               priv->anon_inode_name, uuid_utoa(dir->gfid), name, g,
+               subvol->name);
+    }
 
-	afr_sh_entry_unlock (frame, this);
+out:
+    loc_wipe(&loc);
+    loc_wipe(&loc2);
+    if (frame) {
+        AFR_STACK_DESTROY(frame);
+    }
 
-	return 0;
+    return ret;
 }
 
-
 int
-afr_sh_entry_erase_pending_cbk (call_frame_t *frame, void *cookie,
-				xlator_t *this, int32_t op_ret,
-				int32_t op_errno, dict_t *xattr)
+afr_selfheal_entry_delete(xlator_t *this, inode_t *dir, const char *name,
+                          inode_t *inode, int child, struct afr_reply *replies)
 {
-	int             call_count = 0;
+    char g[64] = {0};
+    afr_private_t *priv = NULL;
+    xlator_t *subvol = NULL;
+    int ret = 0;
+    loc_t loc = {
+        0,
+    };
+    gf_boolean_t anon_inode = _gf_false;
+
+    priv = this->private;
+    subvol = priv->children[child];
+
+    if ((!replies[child].valid) || (replies[child].op_ret < 0)) {
+        /*Nothing to do*/
+        ret = 0;
+        goto out;
+    }
+
+    if (priv->use_anon_inode) {
+        ret = afr_selfheal_entry_anon_inode(this, dir, name, inode, child,
+                                            replies, &anon_inode);
+        if (ret < 0 || anon_inode)
+            goto out;
+    }
+
+    loc.parent = inode_ref(dir);
+    loc.inode = inode_new(inode->table);
+    if (!loc.inode) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    loc.name = name;
+    switch (replies[child].poststat.ia_type) {
+        case IA_IFDIR:
+            gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_EXPUNGING_FILE_OR_DIR,
+                   "expunging dir %s/%s (%s) on %s", uuid_utoa(dir->gfid), name,
+                   uuid_utoa_r(replies[child].poststat.ia_gfid, g),
+                   subvol->name);
+            ret = syncop_rmdir(subvol, &loc, 1, NULL, NULL);
+            break;
+        default:
+            gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_EXPUNGING_FILE_OR_DIR,
+                   "expunging file %s/%s (%s) on %s", uuid_utoa(dir->gfid),
+                   name, uuid_utoa_r(replies[child].poststat.ia_gfid, g),
+                   subvol->name);
+            ret = syncop_unlink(subvol, &loc, NULL, NULL);
+            break;
+    }
 
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0)
-		afr_sh_entry_finish (frame, this);
-
-	return 0;
+out:
+    loc_wipe(&loc);
+    return ret;
 }
 
-
 int
-afr_sh_entry_erase_pending (call_frame_t *frame, xlator_t *this)
+afr_selfheal_recreate_entry(call_frame_t *frame, int dst, int source,
+                            unsigned char *sources, inode_t *dir,
+                            const char *name, inode_t *inode,
+                            struct afr_reply *replies)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              call_count = 0;
-	int              i = 0;
-	dict_t          **erase_xattr = NULL;
-        int              need_unwind = 0;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	afr_sh_pending_to_delta (priv, sh->xattr, sh->delta_matrix, sh->success,
-                                 priv->child_count, AFR_ENTRY_TRANSACTION);
-
-	erase_xattr = GF_CALLOC (sizeof (*erase_xattr), priv->child_count,
-                                 gf_afr_mt_dict_t);
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->xattr[i]) {
-			call_count++;
-
-			erase_xattr[i] = get_new_dict();
-			dict_ref (erase_xattr[i]);
-		}
-	}
-
-        if (call_count == 0)
-                need_unwind = 1;
-
-	afr_sh_delta_to_xattr (priv, sh->delta_matrix, erase_xattr,
-			       priv->child_count, AFR_ENTRY_TRANSACTION);
-
-	local->call_count = call_count;
-	for (i = 0; i < priv->child_count; i++) {
-		if (!erase_xattr[i])
-			continue;
-
-		gf_log (this->name, GF_LOG_TRACE,
-			"erasing pending flags from %s on %s",
-			local->loc.path, priv->children[i]->name);
-
-		STACK_WIND_COOKIE (frame, afr_sh_entry_erase_pending_cbk,
-				   (void *) (long) i,
-				   priv->children[i],
-				   priv->children[i]->fops->xattrop,
-				   &local->loc,
-				   GF_XATTROP_ADD_ARRAY, erase_xattr[i]);
-		if (!--call_count)
-			break;
-	}
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (erase_xattr[i]) {
-			dict_unref (erase_xattr[i]);
-		}
-	}
-	GF_FREE (erase_xattr);
-
-        if (need_unwind)
-                afr_sh_entry_finish (frame, this);
-
-	return 0;
-}
-
-
+    int ret = 0;
+    loc_t loc = {
+        0,
+    };
+    loc_t srcloc = {
+        0,
+    };
+    loc_t anonloc = {
+        0,
+    };
+    xlator_t *this = frame->this;
+    afr_private_t *priv = NULL;
+    dict_t *xdata = NULL;
+    struct iatt *iatt = NULL;
+    char *linkname = NULL;
+    mode_t mode = 0;
+    struct iatt newent = {
+        0,
+    };
+    unsigned char *newentry = NULL;
+    char iatt_uuid_str[64] = {0};
+    char dir_uuid_str[64] = {0};
+
+    priv = this->private;
+    iatt = &replies[source].poststat;
+    uuid_utoa_r(iatt->ia_gfid, iatt_uuid_str);
+    if (iatt->ia_type == IA_INVAL || gf_uuid_is_null(iatt->ia_gfid)) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SELF_HEAL_FAILED,
+               "Invalid ia_type (%d) or gfid(%s). source brick=%d, "
+               "pargfid=%s, name=%s",
+               iatt->ia_type, iatt_uuid_str, source,
+               uuid_utoa_r(dir->gfid, dir_uuid_str), name);
+        ret = -EINVAL;
+        goto out;
+    }
+
+    xdata = dict_new();
+    if (!xdata)
+        return -ENOMEM;
+    newentry = alloca0(priv->child_count);
+    loc.parent = inode_ref(dir);
+    gf_uuid_copy(loc.pargfid, dir->gfid);
+    loc.name = name;
+    loc.inode = inode_ref(inode);
+
+    ret = afr_selfheal_entry_delete(this, dir, name, inode, dst, replies);
+    if (ret)
+        goto out;
+
+    ret = dict_set_gfuuid(xdata, "gfid-req", replies[source].poststat.ia_gfid,
+                          true);
+    if (ret)
+        goto out;
+
+    srcloc.inode = inode_ref(inode);
+    gf_uuid_copy(srcloc.gfid, iatt->ia_gfid);
+    ret = syncop_lookup(priv->children[dst], &srcloc, 0, 0, 0, 0);
+    if (ret == -ENOENT || ret == -ESTALE) {
+        newentry[dst] = 1;
+        ret = afr_selfheal_newentry_mark(frame, this, inode, source, replies,
+                                         sources, newentry);
+        if (ret)
+            goto out;
+    } else if (ret == 0 && iatt->ia_type == IA_IFDIR && priv->use_anon_inode) {
+        // Try rename from hidden directory
+        ret = afr_anon_inode_create(this, dst, &anonloc.parent);
+        if (ret < 0)
+            goto out;
+        anonloc.inode = inode_ref(inode);
+        anonloc.name = iatt_uuid_str;
+        ret = syncop_rename(priv->children[dst], &anonloc, &loc, NULL, NULL);
+        if (ret == -ENOENT || ret == -ESTALE)
+            ret = -1; /*This sets 'mismatch' to true*/
+        goto out;
+    }
+
+    mode = st_mode_from_ia(iatt->ia_prot, iatt->ia_type);
+
+    switch (iatt->ia_type) {
+        case IA_IFDIR:
+            ret = syncop_mkdir(priv->children[dst], &loc, mode, 0, xdata, NULL);
+            break;
+        case IA_IFLNK:
+            if (!newentry[dst]) {
+                ret = syncop_link(priv->children[dst], &srcloc, &loc, &newent,
+                                  NULL, NULL);
+            } else {
+                ret = syncop_readlink(priv->children[source], &srcloc,
+                                      &linkname, 4096, NULL, NULL);
+                if (ret <= 0)
+                    goto out;
+                ret = syncop_symlink(priv->children[dst], &loc, linkname, NULL,
+                                     xdata, NULL);
+            }
+            break;
+        default:
+            ret = dict_set_int32_sizen(xdata, GLUSTERFS_INTERNAL_FOP_KEY, 1);
+            if (ret)
+                goto out;
+            ret = syncop_mknod(
+                priv->children[dst], &loc, mode,
+                makedev(ia_major(iatt->ia_rdev), ia_minor(iatt->ia_rdev)),
+                &newent, xdata, NULL);
+            break;
+    }
 
-static int
-next_active_source (call_frame_t *frame, xlator_t *this,
-		    int current_active_source)
-{
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	int              source = -1;
-	int              next_active_source = -1;
-	int              i = 0;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	source = sh->source;
-
-	if (source != -1) {
-		if (current_active_source != source)
-			next_active_source = source;
-		goto out;
-	}
-
-	/*
-	  the next active sink becomes the source for the
-	  'conservative decision' of merging all entries
-	*/
-
-	for (i = 0; i < priv->child_count; i++) {
-		if ((sh->sources[i] == 0)
-		    && (local->child_up[i] == 1)
-		    && (i > current_active_source)) {
-
-			next_active_source = i;
-			break;
-		}
-	}
 out:
-	return next_active_source;
+    if (xdata)
+        dict_unref(xdata);
+    GF_FREE(linkname);
+    loc_wipe(&loc);
+    loc_wipe(&srcloc);
+    loc_wipe(&anonloc);
+    return ret;
 }
 
-
-
 static int
-next_active_sink (call_frame_t *frame, xlator_t *this,
-		  int current_active_sink)
+__afr_selfheal_heal_dirent(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                           char *name, inode_t *inode, int source,
+                           unsigned char *sources, unsigned char *healed_sinks,
+                           unsigned char *locked_on, struct afr_reply *replies)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	int              next_active_sink = -1;
-	int              i = 0;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	/*
-	  the next active sink becomes the source for the
-	  'conservative decision' of merging all entries
-	*/
-
-	for (i = 0; i < priv->child_count; i++) {
-		if ((sh->sources[i] == 0)
-		    && (local->child_up[i] == 1)
-		    && (i > current_active_sink)) {
-
-			next_active_sink = i;
-			break;
-		}
-	}
-
-	return next_active_sink;
-}
+    int ret = 0;
+    afr_private_t *priv = NULL;
+    int i = 0;
 
+    priv = this->private;
 
-int
-build_child_loc (xlator_t *this, loc_t *child, loc_t *parent, char *name)
-{
-	int   ret = -1;
-
-	if (!child) {
-		goto out;
-	}
-
-	if (strcmp (parent->path, "/") == 0)
-		ret = gf_asprintf ((char **)&child->path, "/%s", name);
-	else
-		ret = gf_asprintf ((char **)&child->path, "%s/%s", parent->path, 
-                                name);
-
-        if (-1 == ret) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "asprintf failed while setting child path");
-        }
-
-	if (!child->path) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		goto out;
-	}
+    if (!replies[source].valid)
+        return -EIO;
 
-	child->name = strrchr (child->path, '/');
-	if (child->name)
-		child->name++;
-
-	child->parent = inode_ref (parent->inode);
-	child->inode = inode_new (parent->inode->table);
-
-	if (!child->inode) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		goto out;
-	}
-
-	ret = 0;
-out:
-	if (ret == -1)
-		loc_wipe (child);
-
-	return ret;
-}
+    /* Skip healing this entry if the last lookup on it failed for reasons
+     * other than ENOENT.
+     */
+    if ((replies[source].op_ret < 0) && (replies[source].op_errno != ENOENT))
+        return -replies[source].op_errno;
 
+    if (replies[source].op_ret == 0) {
+        ret = afr_lookup_and_heal_gfid(this, fd->inode, name, inode, replies,
+                                       source, sources,
+                                       &replies[source].poststat.ia_gfid, NULL);
+        if (ret)
+            return ret;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!healed_sinks[i])
+            continue;
+        if (replies[source].op_ret == -1 &&
+            replies[source].op_errno == ENOENT) {
+            ret = afr_selfheal_entry_delete(this, fd->inode, name, inode, i,
+                                            replies);
+        } else {
+            if (!gf_uuid_compare(replies[i].poststat.ia_gfid,
+                                 replies[source].poststat.ia_gfid))
+                continue;
 
-int
-afr_sh_entry_expunge_all (call_frame_t *frame, xlator_t *this);
-
-int
-afr_sh_entry_expunge_subvol (call_frame_t *frame, xlator_t *this,
-			     int active_src);
-
-int
-afr_sh_entry_expunge_entry_done (call_frame_t *frame, xlator_t *this,
-				 int active_src)
-{
-	int              call_count = 0;
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0)
-		afr_sh_entry_expunge_subvol (frame, this, active_src);
-
-	return 0;
-}
-
-int
-afr_sh_entry_expunge_parent_setattr_cbk (call_frame_t *expunge_frame,
-                                         void *cookie, xlator_t *this,
-                                         int32_t op_ret, int32_t op_errno,
-                                         struct iatt *preop, struct iatt *postop)
-{
-	afr_private_t   *priv          = NULL;
-	afr_local_t     *expunge_local = NULL;
-	afr_self_heal_t *expunge_sh    = NULL;
-        call_frame_t    *frame         = NULL;
-
-        int active_src = (long) cookie;
-
-        priv          = this->private;
-        expunge_local = expunge_frame->local;
-	expunge_sh    = &expunge_local->self_heal;
-	frame         = expunge_sh->sh_frame;
-
-        if (op_ret != 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "setattr on parent directory of %s on subvolume %s failed: %s",
-                        expunge_local->loc.path,
-                        priv->children[active_src]->name, strerror (op_errno));
+            ret = afr_selfheal_recreate_entry(frame, i, source, sources,
+                                              fd->inode, name, inode, replies);
         }
+        if (ret < 0)
+            break;
+    }
 
-	AFR_STACK_DESTROY (expunge_frame);
-	afr_sh_entry_expunge_entry_done (frame, this, active_src);
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_expunge_remove_cbk (call_frame_t *expunge_frame, void *cookie,
-				 xlator_t *this,
-				 int32_t op_ret, int32_t op_errno,
-                                 struct iatt *preparent,
-                                 struct iatt *postparent)
-{
-	afr_private_t   *priv = NULL;
-	afr_local_t     *expunge_local = NULL;
-	afr_self_heal_t *expunge_sh = NULL;
-	int              active_src = 0;
-	call_frame_t    *frame = NULL;
-
-        int32_t valid = 0;
-
-	priv = this->private;
-	expunge_local = expunge_frame->local;
-	expunge_sh = &expunge_local->self_heal;
-	frame = expunge_sh->sh_frame;
-
-	active_src = (long) cookie;
-
-	if (op_ret == 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"removed %s on %s",
-			expunge_local->loc.path,
-			priv->children[active_src]->name);
-	} else {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"removing %s on %s failed (%s)",
-			expunge_local->loc.path,
-			priv->children[active_src]->name,
-			strerror (op_errno));
-	}
-
-        valid = GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME;
-        afr_build_parent_loc (&expunge_sh->parent_loc, &expunge_local->loc);
-
-        STACK_WIND_COOKIE (expunge_frame, afr_sh_entry_expunge_parent_setattr_cbk,
-                           (void *) (long) active_src,
-                           priv->children[active_src],
-                           priv->children[active_src]->fops->setattr,
-                           &expunge_sh->parent_loc,
-                           &expunge_sh->parentbuf,
-                           valid);
-
-        return 0;
-}
-
-
-int
-afr_sh_entry_expunge_unlink (call_frame_t *expunge_frame, xlator_t *this,
-                             int active_src)
-{
-	afr_private_t   *priv = NULL;
-	afr_local_t     *expunge_local = NULL;
-
-	priv          = this->private;
-	expunge_local = expunge_frame->local;
-
-	gf_log (this->name, GF_LOG_TRACE,
-		"expunging file %s on %s",
-		expunge_local->loc.path, priv->children[active_src]->name);
-
-	STACK_WIND_COOKIE (expunge_frame, afr_sh_entry_expunge_remove_cbk,
-			   (void *) (long) active_src,
-			   priv->children[active_src],
-			   priv->children[active_src]->fops->unlink,
-			   &expunge_local->loc);
-
-	return 0;
+    return ret;
 }
 
-
-
-int
-afr_sh_entry_expunge_rmdir (call_frame_t *expunge_frame, xlator_t *this,
-                            int active_src)
+static int
+afr_selfheal_detect_gfid_and_type_mismatch(xlator_t *this,
+                                           struct afr_reply *replies,
+                                           inode_t *inode, uuid_t pargfid,
+                                           char *bname, int src_idx,
+                                           unsigned char *locked_on, int *src)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *expunge_local = NULL;
+    int i = 0;
+    int ret = -1;
+    afr_private_t *priv = NULL;
+    void *gfid = NULL;
+    ia_type_t ia_type = IA_INVAL;
 
-	priv          = this->private;
-	expunge_local = expunge_frame->local;
+    priv = this->private;
+    gfid = &replies[src_idx].poststat.ia_gfid;
+    ia_type = replies[src_idx].poststat.ia_type;
 
-	gf_log (this->name, GF_LOG_DEBUG,
-		"expunging directory %s on %s",
-		expunge_local->loc.path, priv->children[active_src]->name);
+    for (i = 0; i < priv->child_count; i++) {
+        if (i == src_idx)
+            continue;
 
-	STACK_WIND_COOKIE (expunge_frame, afr_sh_entry_expunge_remove_cbk,
-			   (void *) (long) active_src,
-			   priv->children[active_src],
-			   priv->children[active_src]->fops->rmdir,
-			   &expunge_local->loc, 1);
+        if (!replies[i].valid)
+            continue;
 
-	return 0;
-}
+        if (replies[i].op_ret != 0)
+            continue;
 
+        if (gf_uuid_is_null(replies[i].poststat.ia_gfid))
+            continue;
 
-int
-afr_sh_entry_expunge_remove (call_frame_t *expunge_frame, xlator_t *this,
-			     int active_src, struct iatt *buf)
-{
-	afr_private_t   *priv = NULL;
-	afr_local_t     *expunge_local = NULL;
-	afr_self_heal_t *expunge_sh = NULL;
-	int              source = 0;
-	call_frame_t    *frame = NULL;
-	int              type = 0;
-
-	priv = this->private;
-	expunge_local = expunge_frame->local;
-	expunge_sh = &expunge_local->self_heal;
-	frame = expunge_sh->sh_frame;
-	source = expunge_sh->source;
-
-	type = buf->ia_type;
-
-	switch (type) {
-	case IA_IFSOCK:
-	case IA_IFREG:
-	case IA_IFBLK:
-	case IA_IFCHR:
-	case IA_IFIFO:
-	case IA_IFLNK:
-		afr_sh_entry_expunge_unlink (expunge_frame, this, active_src);
-                break;
-	case IA_IFDIR:
-		afr_sh_entry_expunge_rmdir (expunge_frame, this, active_src);
-		break;
-	default:
-		gf_log (this->name, GF_LOG_ERROR,
-			"%s has unknown file type on %s: 0%o",
-			expunge_local->loc.path,
-			priv->children[source]->name, type);
-		goto out;
-		break;
-	}
-
-	return 0;
-out:
-	AFR_STACK_DESTROY (expunge_frame);
-	afr_sh_entry_expunge_entry_done (frame, this, active_src);
+        if (replies[i].poststat.ia_type == IA_INVAL)
+            continue;
 
-	return 0;
-}
+        if (ia_type == IA_INVAL || gf_uuid_is_null(gfid)) {
+            src_idx = i;
+            ia_type = replies[src_idx].poststat.ia_type;
+            gfid = &replies[src_idx].poststat.ia_gfid;
+            continue;
+        }
 
+        if (gf_uuid_compare(gfid, replies[i].poststat.ia_gfid) &&
+            (ia_type == replies[i].poststat.ia_type)) {
+            ret = afr_gfid_split_brain_source(this, replies, inode, pargfid,
+                                              bname, src_idx, i, locked_on, src,
+                                              NULL);
+            if (ret)
+                gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+                       "Skipping conservative merge on the "
+                       "file.");
+            return ret;
+        }
 
-int
-afr_sh_entry_expunge_lookup_cbk (call_frame_t *expunge_frame, void *cookie,
-				xlator_t *this,
-				int32_t op_ret,	int32_t op_errno,
-                                inode_t *inode, struct iatt *buf, dict_t *x,
-                                struct iatt *postparent)
-{
-	afr_private_t   *priv = NULL;
-	afr_local_t     *expunge_local = NULL;
-	afr_self_heal_t *expunge_sh = NULL;
-	call_frame_t    *frame = NULL;
-	int              active_src = 0;
-
-	priv = this->private;
-	expunge_local = expunge_frame->local;
-	expunge_sh = &expunge_local->self_heal;
-	frame = expunge_sh->sh_frame;
-	active_src = (long) cookie;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_TRACE,
-			"lookup of %s on %s failed (%s)",
-			expunge_local->loc.path,
-			priv->children[active_src]->name,
-			strerror (op_errno));
-		goto out;
-	}
-
-	afr_sh_entry_expunge_remove (expunge_frame, this, active_src, buf);
-
-	return 0;
-out:
-	AFR_STACK_DESTROY (expunge_frame);
-	afr_sh_entry_expunge_entry_done (frame, this, active_src);
+        if (ia_type != replies[i].poststat.ia_type) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SPLIT_BRAIN,
+                   "Type mismatch detected "
+                   "for <gfid:%s>/%s>, %s on %s and %s on %s. "
+                   "Skipping conservative merge on the file.",
+                   uuid_utoa(pargfid), bname,
+                   gf_inode_type_to_str(replies[i].poststat.ia_type),
+                   priv->children[i]->name,
+                   gf_inode_type_to_str(replies[src_idx].poststat.ia_type),
+                   priv->children[src_idx]->name);
+            gf_event(EVENT_AFR_SPLIT_BRAIN,
+                     "client-pid=%d;"
+                     "subvol=%s;type=file;"
+                     "file=<gfid:%s>/%s>;count=2;child-%d=%s;type-"
+                     "%d=%s;child-%d=%s;type-%d=%s",
+                     this->ctx->cmd_args.client_pid, this->name,
+                     uuid_utoa(pargfid), bname, i, priv->children[i]->name, i,
+                     gf_inode_type_to_str(replies[i].poststat.ia_type), src_idx,
+                     priv->children[src_idx]->name, src_idx,
+                     gf_inode_type_to_str(replies[src_idx].poststat.ia_type));
+            return -1;
+        }
+    }
 
-	return 0;
+    return 0;
 }
 
-
-int
-afr_sh_entry_expunge_purge (call_frame_t *expunge_frame, xlator_t *this,
-			    int active_src)
+static int
+__afr_selfheal_merge_dirent(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                            char *name, inode_t *inode, unsigned char *sources,
+                            unsigned char *healed_sinks,
+                            unsigned char *locked_on, struct afr_reply *replies)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *expunge_local = NULL;
-
-	priv = this->private;
-	expunge_local = expunge_frame->local;
-
-	gf_log (this->name, GF_LOG_TRACE,
-		"looking up %s on %s",
-		expunge_local->loc.path, priv->children[active_src]->name);
-	
-	STACK_WIND_COOKIE (expunge_frame, afr_sh_entry_expunge_lookup_cbk,
-			   (void *) (long) active_src,
-			   priv->children[active_src],
-			   priv->children[active_src]->fops->lookup,
-			   &expunge_local->loc, 0);
-
-	return 0;
-}
-
+    int ret = 0;
+    int i = 0;
+    int source = -1;
+    int src = -1;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (replies[i].valid && replies[i].op_ret == 0) {
+            source = i;
+            break;
+        }
+    }
 
-int
-afr_sh_entry_expunge_entry_cbk (call_frame_t *expunge_frame, void *cookie,
-				xlator_t *this,
-				int32_t op_ret,	int32_t op_errno,
-                                inode_t *inode, struct iatt *buf, dict_t *x,
-                                struct iatt *postparent)
-{
-	afr_private_t   *priv = NULL;
-	afr_local_t     *expunge_local = NULL;
-	afr_self_heal_t *expunge_sh = NULL;
-	int              source = 0;
-	call_frame_t    *frame = NULL;
-	int              active_src = 0;
-
-
-	priv = this->private;
-	expunge_local = expunge_frame->local;
-	expunge_sh = &expunge_local->self_heal;
-	frame = expunge_sh->sh_frame;
-	active_src = expunge_sh->active_source;
-	source = (long) cookie;
-
-	if (op_ret == -1 && op_errno == ENOENT && postparent) {
-
-		gf_log (this->name, GF_LOG_TRACE,
-			"missing entry %s on %s",
-			expunge_local->loc.path,
-			priv->children[source]->name);
-
-                expunge_sh->parentbuf = *postparent;
-
-		afr_sh_entry_expunge_purge (expunge_frame, this, active_src);
-
-		return 0;
-	}
-
-	if (op_ret == 0) {
-		gf_log (this->name, GF_LOG_TRACE,
-			"%s exists under %s",
-			expunge_local->loc.path,
-			priv->children[source]->name);
-	} else {
-		gf_log (this->name, GF_LOG_TRACE,
-			"looking up %s under %s failed (%s)",
-			expunge_local->loc.path,
-			priv->children[source]->name,
-			strerror (op_errno));
-	}
-
-	AFR_STACK_DESTROY (expunge_frame);
-	afr_sh_entry_expunge_entry_done (frame, this, active_src);
-
-	return 0;
-}
+    if (source == -1) {
+        /* entry got deleted in the mean time? */
+        return 0;
+    }
 
+    /* Set all the sources as 1, otheriwse newentry_mark won't be set */
+    for (i = 0; i < priv->child_count; i++) {
+        if (replies[i].valid && replies[i].op_ret == 0) {
+            sources[i] = 1;
+        }
+    }
+
+    ret = afr_lookup_and_heal_gfid(this, fd->inode, name, inode, replies,
+                                   source, sources,
+                                   &replies[source].poststat.ia_gfid, NULL);
+    if (ret)
+        return ret;
+
+    /* In case of type mismatch / unable to resolve gfid mismatch on the
+     * entry, return -1.*/
+    ret = afr_selfheal_detect_gfid_and_type_mismatch(
+        this, replies, inode, fd->inode->gfid, name, source, locked_on, &src);
+
+    if (ret < 0)
+        return ret;
+    if (src != -1) {
+        source = src;
+        for (i = 0; i < priv->child_count; i++) {
+            if (i != src && replies[i].valid &&
+                gf_uuid_compare(replies[src].poststat.ia_gfid,
+                                replies[i].poststat.ia_gfid)) {
+                sources[i] = 0;
+            }
+        }
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (i == source || !healed_sinks[i])
+            continue;
+
+        if (src != -1) {
+            if (!gf_uuid_compare(replies[src].poststat.ia_gfid,
+                                 replies[i].poststat.ia_gfid))
+                continue;
+        } else if (replies[i].op_errno != ENOENT) {
+            continue;
+        }
 
-int
-afr_sh_entry_expunge_entry (call_frame_t *frame, xlator_t *this,
-			    char *name)
-{
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	int              ret = -1;
-	call_frame_t    *expunge_frame = NULL;
-	afr_local_t     *expunge_local = NULL;
-	afr_self_heal_t *expunge_sh = NULL;
-	int              active_src = 0;
-	int              source = 0;
-	int              op_errno = 0;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	active_src = sh->active_source;
-	source = sh->source;
-
-	if ((strcmp (name, ".") == 0)
-	    || (strcmp (name, "..") == 0)
-            || ((strcmp (local->loc.path, "/") == 0)
-                && (strcmp (name, GF_REPLICATE_TRASH_DIR) == 0))) {
-
-                gf_log (this->name, GF_LOG_TRACE,
-			"skipping inspection of %s under %s",
-			name, local->loc.path);
-		goto out;
-	}
-
-	gf_log (this->name, GF_LOG_TRACE,
-		"inspecting existance of %s under %s",
-		name, local->loc.path);
-
-	expunge_frame = copy_frame (frame);
-	if (!expunge_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		goto out;
-	}
-
-	ALLOC_OR_GOTO (expunge_local, afr_local_t, out);
-
-	expunge_frame->local = expunge_local;
-	expunge_sh = &expunge_local->self_heal;
-	expunge_sh->sh_frame = frame;
-	expunge_sh->active_source = active_src;
-
-	ret = build_child_loc (this, &expunge_local->loc, &local->loc, name);
-	if (ret != 0) {
-		goto out;
-	}
-
-	gf_log (this->name, GF_LOG_TRACE,
-		"looking up %s on %s", expunge_local->loc.path,
-		priv->children[source]->name);
-
-	STACK_WIND_COOKIE (expunge_frame,
-			   afr_sh_entry_expunge_entry_cbk,
-			   (void *) (long) source,
-			   priv->children[source],
-			   priv->children[source]->fops->lookup,
-			   &expunge_local->loc, 0);
-
-	ret = 0;
-out:
-	if (ret == -1)
-		afr_sh_entry_expunge_entry_done (frame, this, active_src);
+        ret |= afr_selfheal_recreate_entry(frame, i, source, sources, fd->inode,
+                                           name, inode, replies);
+    }
 
-	return 0;
+    return ret;
 }
 
-
-int
-afr_sh_entry_expunge_readdir_cbk (call_frame_t *frame, void *cookie,
-				  xlator_t *this,
-				  int32_t op_ret, int32_t op_errno,
-				  gf_dirent_t *entries)
+static int
+__afr_selfheal_entry_dirent(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                            char *name, inode_t *inode, int source,
+                            unsigned char *sources, unsigned char *healed_sinks,
+                            unsigned char *locked_on, struct afr_reply *replies)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	gf_dirent_t     *entry = NULL;
-	off_t            last_offset = 0;
-	int              active_src = 0;
-	int              entry_count = 0;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	active_src = sh->active_source;
-
-	if (op_ret <= 0) {
-		if (op_ret < 0) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"readdir of %s on subvolume %s failed (%s)",
-				local->loc.path,
-				priv->children[active_src]->name,
-				strerror (op_errno));
-		} else {
-			gf_log (this->name, GF_LOG_TRACE,
-				"readdir of %s on subvolume %s complete",
-				local->loc.path,
-				priv->children[active_src]->name);
-		}
-
-		afr_sh_entry_expunge_all (frame, this);
-		return 0;
-	}
-
-	list_for_each_entry (entry, &entries->list, list) {
-		last_offset = entry->d_off;
-                entry_count++;
-	}
-
-	gf_log (this->name, GF_LOG_TRACE,
-		"readdir'ed %d entries from %s",
-		entry_count, priv->children[active_src]->name);
-
-	sh->offset = last_offset;
-	local->call_count = entry_count;
-
-	list_for_each_entry (entry, &entries->list, list) {
-                afr_sh_entry_expunge_entry (frame, this, entry->d_name);
-	}
-
-	return 0;
+    int ret = -1;
+
+    if (source < 0)
+        ret = __afr_selfheal_merge_dirent(frame, this, fd, name, inode, sources,
+                                          healed_sinks, locked_on, replies);
+    else
+        ret = __afr_selfheal_heal_dirent(frame, this, fd, name, inode, source,
+                                         sources, healed_sinks, locked_on,
+                                         replies);
+    return ret;
 }
 
-int
-afr_sh_entry_expunge_subvol (call_frame_t *frame, xlator_t *this,
-			     int active_src)
+static gf_boolean_t
+is_full_heal_marker_present(xlator_t *this, dict_t *xdata, int idx)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	STACK_WIND (frame, afr_sh_entry_expunge_readdir_cbk,
-		    priv->children[active_src],
-		    priv->children[active_src]->fops->readdir,
-		    sh->healing_fd, sh->block_size, sh->offset);
-
-	return 0;
+    int i = 0;
+    int pending[3] = {
+        0,
+    };
+    void *pending_raw = NULL;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+
+    if (!xdata)
+        return _gf_false;
+
+    /* Iterate over each of the priv->pending_keys[] elements and then
+     * see if any of them have data segment non-zero. If they do, return
+     * true. Else return false.
+     */
+    for (i = 0; i < priv->child_count; i++) {
+        if (dict_get_ptr(xdata, priv->pending_key[i], &pending_raw))
+            continue;
+
+        if (!pending_raw)
+            continue;
+
+        memcpy(pending, pending_raw, sizeof(pending));
+        if (ntoh32(pending[idx]))
+            return _gf_true;
+    }
+
+    return _gf_false;
 }
 
-
-int
-afr_sh_entry_expunge_all (call_frame_t *frame, xlator_t *this)
+static gf_boolean_t
+afr_need_full_heal(xlator_t *this, struct afr_reply *replies, int source,
+                   unsigned char *healed_sinks, afr_transaction_type type)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	int              active_src = -1;
+    int i = 0;
+    int idx = 0;
+    afr_private_t *priv = NULL;
 
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
+    priv = this->private;
 
-	sh->offset = 0;
+    if (!priv->esh_granular)
+        return _gf_true;
 
-	if (sh->source == -1) {
-		gf_log (this->name, GF_LOG_TRACE,
-			"no active sources for %s to expunge entries",
-			local->loc.path);
-		goto out;
-	}
+    if (type != AFR_ENTRY_TRANSACTION)
+        return _gf_true;
 
-	active_src = next_active_sink (frame, this, sh->active_source);
-	sh->active_source = active_src;
+    priv = this->private;
+    idx = afr_index_for_transaction_type(AFR_DATA_TRANSACTION);
 
-	if (sh->op_failed) {
-		goto out;
-	}
+    /* If there is a clear source, check whether the full-heal-indicator
+     * is present in its xdata. Otherwise, we need to examine all the
+     * participating bricks and then figure if *even* one of them has a
+     * full-heal-indicator.
+     */
 
-	if (active_src == -1) {
-		/* completed creating missing files on all subvolumes */
-		goto out;
-	}
+    if (source != -1) {
+        if (is_full_heal_marker_present(this, replies[source].xdata, idx))
+            return _gf_true;
+    }
 
-	gf_log (this->name, GF_LOG_TRACE,
-		"expunging entries of %s on %s to other sinks",
-		local->loc.path, priv->children[active_src]->name);
+    /* else ..*/
 
-	afr_sh_entry_expunge_subvol (frame, this, active_src);
+    for (i = 0; i < priv->child_count; i++) {
+        if (!healed_sinks[i])
+            continue;
 
-	return 0;
-out:
-	afr_sh_entry_erase_pending (frame, this);
-	return 0;
+        if (is_full_heal_marker_present(this, replies[i].xdata, idx))
+            return _gf_true;
+    }
 
+    return _gf_false;
 }
 
-
-int
-afr_sh_entry_impunge_all (call_frame_t *frame, xlator_t *this);
-
-int
-afr_sh_entry_impunge_subvol (call_frame_t *frame, xlator_t *this,
-			     int active_src);
-
-int
-afr_sh_entry_impunge_entry_done (call_frame_t *frame, xlator_t *this,
-				 int active_src)
+static int
+__afr_selfheal_entry_finalize_source(xlator_t *this, unsigned char *sources,
+                                     unsigned char *healed_sinks,
+                                     unsigned char *locked_on,
+                                     struct afr_reply *replies,
+                                     uint64_t *witness)
 {
-	int              call_count = 0;
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0)
-		afr_sh_entry_impunge_subvol (frame, this, active_src);
+    afr_private_t *priv = NULL;
+    int source = -1;
+    int sources_count = 0;
+    int i = 0;
+
+    priv = this->private;
+
+    sources_count = AFR_COUNT(sources, priv->child_count);
+
+    if ((AFR_CMP(locked_on, healed_sinks, priv->child_count) == 0) ||
+        !sources_count || afr_does_witness_exist(this, witness)) {
+        memset(sources, 0, sizeof(*sources) * priv->child_count);
+        afr_mark_active_sinks(this, sources, locked_on, healed_sinks);
+        return -1;
+    }
+
+    source = afr_choose_source_by_policy(priv, sources, AFR_ENTRY_TRANSACTION);
+
+    /*If the selected source does not blame any other brick, then mark
+     * everything as sink to trigger conservative merge.
+     */
+    if (source != -1 && !AFR_COUNT(healed_sinks, priv->child_count)) {
+        for (i = 0; i < priv->child_count; i++) {
+            if (locked_on[i]) {
+                sources[i] = 0;
+                healed_sinks[i] = 1;
+            }
+        }
+        return -1;
+    }
 
-	return 0;
+    return source;
 }
 
-
 int
-afr_sh_entry_impunge_setattr_cbk (call_frame_t *impunge_frame, void *cookie,
-				  xlator_t *this,
-                                  int32_t op_ret, int32_t op_errno,
-                                  struct iatt *preop, struct iatt *postop)
+__afr_selfheal_entry_prepare(call_frame_t *frame, xlator_t *this,
+                             inode_t *inode, unsigned char *locked_on,
+                             unsigned char *sources, unsigned char *sinks,
+                             unsigned char *healed_sinks,
+                             struct afr_reply *replies, int *source_p,
+                             unsigned char *pflag)
 {
-	int              call_count = 0;
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-	call_frame_t    *frame = NULL;
-	int              active_src = 0;
-	int              child_index = 0;
-
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh = &impunge_local->self_heal;
-	frame = impunge_sh->sh_frame;
-        local = frame->local;
-        sh    = &local->self_heal;
-        active_src = sh->active_source;
-	child_index = (long) cookie;
-
-	if (op_ret == 0) {
-		gf_log (this->name, GF_LOG_TRACE,
-			"setattr done for %s on %s",
-			impunge_local->loc.path,
-			priv->children[child_index]->name);
-	} else {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"setattr (%s) on %s failed (%s)",
-			impunge_local->loc.path,
-			priv->children[child_index]->name,
-			strerror (op_errno));
-	}
-
-	LOCK (&impunge_frame->lock);
-	{
-		call_count = --impunge_local->call_count;
-	}
-	UNLOCK (&impunge_frame->lock);
-
-	if (call_count == 0) {
-		AFR_STACK_DESTROY (impunge_frame);
-		afr_sh_entry_impunge_entry_done (frame, this, active_src);
-	}
-
-	return 0;
+    int ret = -1;
+    int source = -1;
+    afr_private_t *priv = NULL;
+    uint64_t *witness = NULL;
+
+    priv = this->private;
+
+    ret = afr_selfheal_unlocked_discover(frame, inode, inode->gfid, replies);
+    if (ret)
+        return ret;
+
+    witness = alloca0(sizeof(*witness) * priv->child_count);
+    ret = afr_selfheal_find_direction(frame, this, replies,
+                                      AFR_ENTRY_TRANSACTION, locked_on, sources,
+                                      sinks, witness, pflag);
+    if (ret)
+        return ret;
+
+    /* Initialize the healed_sinks[] array optimistically to
+       the intersection of to-be-healed (i.e sinks[]) and
+       the list of servers which are up (i.e locked_on[]).
+
+       As we encounter failures in the healing process, we
+       will unmark the respective servers in the healed_sinks[]
+       array.
+    */
+    AFR_INTERSECT(healed_sinks, sinks, locked_on, priv->child_count);
+
+    source = __afr_selfheal_entry_finalize_source(this, sources, healed_sinks,
+                                                  locked_on, replies, witness);
+
+    if (source < 0) {
+        /* If source is < 0 (typically split-brain), we perform a
+           conservative merge of entries rather than erroring out */
+    }
+    *source_p = source;
+
+    return ret;
 }
 
-
-int
-afr_sh_entry_impunge_xattrop_cbk (call_frame_t *impunge_frame, void *cookie,
-                                  xlator_t *this,
-                                  int32_t op_ret, int32_t op_errno,
-                                  dict_t *xattr)
+static int
+afr_selfheal_entry_dirent(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                          char *name, inode_t *parent_idx_inode,
+                          xlator_t *subvol, gf_boolean_t full_crawl)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-	int              child_index = 0;
-
-        struct iatt stbuf;
-        int32_t     valid = 0;
-
-	priv          = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh    = &impunge_local->self_heal;
-
-	child_index = (long) cookie;
-
-	gf_log (this->name, GF_LOG_TRACE,
-		"setting ownership of %s on %s to %d/%d",
-		impunge_local->loc.path,
-		priv->children[child_index]->name,
-		impunge_local->cont.lookup.buf.ia_uid,
-		impunge_local->cont.lookup.buf.ia_gid);
-
-	stbuf.ia_atime = impunge_local->cont.lookup.buf.ia_atime;
-	stbuf.ia_atime_nsec = impunge_local->cont.lookup.buf.ia_atime_nsec;
-	stbuf.ia_mtime = impunge_local->cont.lookup.buf.ia_mtime;
-	stbuf.ia_mtime_nsec = impunge_local->cont.lookup.buf.ia_mtime_nsec;
-
-        stbuf.ia_uid = impunge_local->cont.lookup.buf.ia_uid;
-        stbuf.ia_gid = impunge_local->cont.lookup.buf.ia_gid;
-
-        valid = GF_SET_ATTR_UID   | GF_SET_ATTR_GID |
-                GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME;
-
-	STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_setattr_cbk,
-			   (void *) (long) child_index,
-			   priv->children[child_index],
-			   priv->children[child_index]->fops->setattr,
-			   &impunge_local->loc,
-                           &stbuf, valid);
+    int ret = 0;
+    int source = -1;
+    unsigned char *locked_on = NULL;
+    unsigned char *sources = NULL;
+    unsigned char *sinks = NULL;
+    unsigned char *healed_sinks = NULL;
+    inode_t *inode = NULL;
+    struct afr_reply *replies = NULL;
+    struct afr_reply *par_replies = NULL;
+    afr_private_t *priv = NULL;
+    dict_t *xattr = NULL;
+
+    priv = this->private;
+
+    if (afr_is_private_directory(priv, fd->inode->gfid, name,
+                                 GF_CLIENT_PID_SELF_HEALD)) {
         return 0;
-}
-
-
-int
-afr_sh_entry_impunge_parent_setattr_cbk (call_frame_t *setattr_frame,
-                                         void *cookie, xlator_t *this,
-                                         int32_t op_ret, int32_t op_errno,
-                                         struct iatt *preop, struct iatt *postop)
-{
-        loc_t *parent_loc = cookie;
-
-        if (op_ret != 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "setattr on parent directory failed: %s",
-                        strerror (op_errno));
+    }
+
+    xattr = dict_new();
+    if (!xattr)
+        return -ENOMEM;
+    ret = dict_set_int32_sizen(xattr, GF_GFIDLESS_LOOKUP, 1);
+    if (ret) {
+        dict_unref(xattr);
+        return -1;
+    }
+
+    sources = alloca0(priv->child_count);
+    sinks = alloca0(priv->child_count);
+    healed_sinks = alloca0(priv->child_count);
+    locked_on = alloca0(priv->child_count);
+
+    replies = alloca0(priv->child_count * sizeof(*replies));
+    par_replies = alloca0(priv->child_count * sizeof(*par_replies));
+
+    ret = afr_selfheal_entrylk(frame, this, fd->inode, this->name, NULL,
+                               locked_on);
+    {
+        if (ret < priv->child_count) {
+            gf_msg_debug(this->name, 0,
+                         "%s: Skipping "
+                         "entry self-heal as only %d sub-volumes "
+                         " could be locked in %s domain",
+                         uuid_utoa(fd->inode->gfid), ret, this->name);
+            ret = -ENOTCONN;
+            goto unlock;
         }
 
-        loc_wipe (parent_loc);
-
-        GF_FREE (parent_loc);
-
-        AFR_STACK_DESTROY (setattr_frame);
-        return 0;
-}
-
-
-int
-afr_sh_entry_impunge_newfile_cbk (call_frame_t *impunge_frame, void *cookie,
-				  xlator_t *this,
-				  int32_t op_ret, int32_t op_errno,
-                                  inode_t *inode, struct iatt *stbuf,
-                                  struct iatt *preparent,
-                                  struct iatt *postparent)
-{
-	int              call_count = 0;
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-	call_frame_t    *frame = NULL;
-	int              active_src = 0;
-	int              child_index = 0;
-        int              pending_array[3] = {0, };
-        dict_t          *xattr = NULL;
-        int              ret = 0;
-        int              idx = 0;
-        afr_local_t     *local = NULL;
-        afr_self_heal_t *sh = NULL;
-
-        call_frame_t *setattr_frame = NULL;
-        int32_t valid = 0;
-        loc_t *parent_loc = NULL;
-        struct iatt parentbuf;
-
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh = &impunge_local->self_heal;
-	frame = impunge_sh->sh_frame;
-        local = frame->local;
-        sh    = &local->self_heal;
-        active_src = sh->active_source;
-
-	child_index = (long) cookie;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"creation of %s on %s failed (%s)",
-			impunge_local->loc.path,
-			priv->children[child_index]->name,
-			strerror (op_errno));
-		goto out;
-	}
-
-	inode->ia_type = stbuf->ia_type;
-
-        xattr = get_new_dict ();
-        dict_ref (xattr);
-
-        idx = afr_index_for_transaction_type (AFR_METADATA_TRANSACTION);
-        pending_array[idx] = hton32 (1);
-        if (IA_ISDIR (stbuf->ia_type))
-                idx = afr_index_for_transaction_type (AFR_ENTRY_TRANSACTION);
-        else
-                idx = afr_index_for_transaction_type (AFR_DATA_TRANSACTION);
-        pending_array[idx] = hton32 (1);
-
-        ret = dict_set_static_bin (xattr, priv->pending_key[child_index],
-                                   pending_array, sizeof (pending_array));
+        ret = __afr_selfheal_entry_prepare(frame, this, fd->inode, locked_on,
+                                           sources, sinks, healed_sinks,
+                                           par_replies, &source, NULL);
         if (ret < 0)
-                gf_log (this->name, GF_LOG_WARNING,
-                        "Unable to set dict value.");
-
-        valid         = GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME;
-        parentbuf     = impunge_sh->parentbuf;
-        setattr_frame = copy_frame (impunge_frame);
-
-        parent_loc = GF_CALLOC (1, sizeof (*parent_loc),
-                                gf_afr_mt_loc_t);
-        afr_build_parent_loc (parent_loc, &impunge_local->loc);
-
-	STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_xattrop_cbk,
-			   (void *) (long) child_index,
-			   priv->children[active_src],
-			   priv->children[active_src]->fops->xattrop,
-			   &impunge_local->loc, GF_XATTROP_ADD_ARRAY, xattr);
-
-        STACK_WIND_COOKIE (setattr_frame, afr_sh_entry_impunge_parent_setattr_cbk,
-                           (void *) (long) parent_loc,
-                           priv->children[child_index],
-                           priv->children[child_index]->fops->setattr,
-                           parent_loc, &parentbuf, valid);
-
-        dict_unref (xattr);
-
-	return 0;
-
-out:
-	LOCK (&impunge_frame->lock);
-	{
-		call_count = --impunge_local->call_count;
-	}
-	UNLOCK (&impunge_frame->lock);
-
-	if (call_count == 0) {
-		AFR_STACK_DESTROY (impunge_frame);
-		afr_sh_entry_impunge_entry_done (frame, this, active_src);
-	}
-
-	return 0;
-}
-
-
-int
-afr_sh_entry_impunge_mknod (call_frame_t *impunge_frame, xlator_t *this,
-			    int child_index, struct iatt *stbuf)
-{
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-        dict_t          *dict = NULL;
-
-        int ret = 0;
-
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-
-	gf_log (this->name, GF_LOG_DEBUG,
-		"creating missing file %s on %s",
-		impunge_local->loc.path,
-		priv->children[child_index]->name);
-
-        dict = dict_new ();
-        if (!dict)
-                gf_log (this->name, GF_LOG_ERROR, "Out of memory");
-
-        ret = afr_set_dict_gfid (dict, stbuf->ia_gfid);
-        if (ret)
-                gf_log (this->name, GF_LOG_DEBUG, "gfid set failed");
-
-	STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_newfile_cbk,
-			   (void *) (long) child_index,
-			   priv->children[child_index],
-			   priv->children[child_index]->fops->mknod,
-			   &impunge_local->loc,
-			   st_mode_from_ia (stbuf->ia_prot, stbuf->ia_type),
-                           stbuf->ia_rdev, dict);
-
-        if (dict)
-                dict_unref (dict);
-
-	return 0;
-}
-
-
-
-int
-afr_sh_entry_impunge_mkdir (call_frame_t *impunge_frame, xlator_t *this,
-			    int child_index, struct iatt *stbuf)
-{
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-        dict_t          *dict = NULL;
-
-        int ret = 0;
+            goto unlock;
 
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-
-        dict = dict_new ();
-        if (!dict) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                return 0;
+        inode = afr_selfheal_unlocked_lookup_on(frame, fd->inode, name, replies,
+                                                locked_on, xattr);
+        if (!inode) {
+            ret = -ENOMEM;
+            goto unlock;
         }
 
-        ret = afr_set_dict_gfid (dict, stbuf->ia_gfid);
-        if (ret)
-                gf_log (this->name, GF_LOG_DEBUG, "gfid set failed");
-
-	gf_log (this->name, GF_LOG_DEBUG,
-		"creating missing directory %s on %s",
-		impunge_local->loc.path,
-		priv->children[child_index]->name);
-
-	STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_newfile_cbk,
-			   (void *) (long) child_index,
-			   priv->children[child_index],
-			   priv->children[child_index]->fops->mkdir,
-			   &impunge_local->loc,
-                           st_mode_from_ia (stbuf->ia_prot, stbuf->ia_type),
-                           dict);
-
-        if (dict)
-                dict_unref (dict);
-
-	return 0;
-}
-
-
-int
-afr_sh_entry_impunge_symlink (call_frame_t *impunge_frame, xlator_t *this,
-			      int child_index, const char *linkname)
-{
-	afr_private_t   *priv          = NULL;
-	afr_local_t     *impunge_local = NULL;
-        dict_t          *dict          = NULL;
-        struct iatt     *buf           = NULL;
-
-        int ret = 0;
-
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-
-        buf = &impunge_local->cont.symlink.buf;
-
-        dict = dict_new ();
-        if (!dict) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                afr_sh_entry_impunge_entry_done (impunge_frame, this, 0);
+        ret = __afr_selfheal_entry_dirent(frame, this, fd, name, inode, source,
+                                          sources, healed_sinks, locked_on,
+                                          replies);
+
+        if ((ret == 0) && (priv->esh_granular) && parent_idx_inode) {
+            ret = afr_shd_entry_purge(subvol, parent_idx_inode, name,
+                                      inode->ia_type);
+            /* Why is ret force-set to 0? We do not care about
+             * index purge failing for full heal as it is quite
+             * possible during replace-brick that not all files
+             * and directories have their name indices present in
+             * entry-changes/.
+             */
+            ret = 0;
         }
-
-        ret = afr_set_dict_gfid (dict, buf->ia_gfid);
-        if (ret)
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "dict set gfid failed");
-
-	gf_log (this->name, GF_LOG_DEBUG,
-		"creating missing symlink %s -> %s on %s",
-		impunge_local->loc.path, linkname,
-		priv->children[child_index]->name);
-
-	STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_newfile_cbk,
-			   (void *) (long) child_index,
-			   priv->children[child_index],
-			   priv->children[child_index]->fops->symlink,
-			   linkname, &impunge_local->loc, dict);
-
-        if (dict)
-                dict_unref (dict);
-
-	return 0;
+    }
+
+unlock:
+    afr_selfheal_unentrylk(frame, this, fd->inode, this->name, NULL, locked_on,
+                           NULL);
+    if (inode)
+        inode_unref(inode);
+    if (replies)
+        afr_replies_wipe(replies, priv->child_count);
+    if (par_replies)
+        afr_replies_wipe(par_replies, priv->child_count);
+    if (xattr)
+        dict_unref(xattr);
+
+    return ret;
 }
 
-
-int
-afr_sh_entry_impunge_symlink_unlink_cbk (call_frame_t *impunge_frame,
-                                         void *cookie, xlator_t *this,
-                                         int32_t op_ret, int32_t op_errno,
-                                         struct iatt *preparent,
-                                         struct iatt *postparent)
+static inode_t *
+afr_shd_entry_changes_index_inode(xlator_t *this, xlator_t *subvol,
+                                  uuid_t pargfid)
 {
-        afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-	int              child_index = -1;
-	call_frame_t    *frame = NULL;
-	int              call_count = -1;
-	int              active_src = -1;
-
-	priv          = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh    = &impunge_local->self_heal;
-	frame         = impunge_sh->sh_frame;
-	active_src    = impunge_sh->active_source;
-
-	child_index = (long) cookie;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"unlink of %s on %s failed (%s)",
-			impunge_local->loc.path,
-			priv->children[child_index]->name,
-			strerror (op_errno));
-		goto out;
-	}
-
-        afr_sh_entry_impunge_symlink (impunge_frame, this, child_index,
-                                      impunge_sh->linkname);
+    int ret = -1;
+    void *index_gfid = NULL;
+    loc_t rootloc = {
+        0,
+    };
+    loc_t loc = {
+        0,
+    };
+    dict_t *xattr = NULL;
+    inode_t *inode = NULL;
+    struct iatt iatt = {
+        0,
+    };
+
+    rootloc.inode = inode_ref(this->itable->root);
+    gf_uuid_copy(rootloc.gfid, rootloc.inode->gfid);
+
+    ret = syncop_getxattr(subvol, &rootloc, &xattr,
+                          GF_XATTROP_ENTRY_CHANGES_GFID, NULL, NULL);
+    if (ret || !xattr) {
+        errno = -ret;
+        goto out;
+    }
+
+    ret = dict_get_ptr(xattr, GF_XATTROP_ENTRY_CHANGES_GFID, &index_gfid);
+    if (ret) {
+        errno = EINVAL;
+        goto out;
+    }
+
+    loc.inode = inode_new(this->itable);
+    if (!loc.inode) {
+        errno = ENOMEM;
+        goto out;
+    }
+
+    gf_uuid_copy(loc.pargfid, index_gfid);
+    loc.name = gf_strdup(uuid_utoa(pargfid));
+
+    ret = syncop_lookup(subvol, &loc, &iatt, NULL, NULL, NULL);
+    if (ret < 0) {
+        errno = -ret;
+        goto out;
+    }
+
+    inode = inode_link(loc.inode, NULL, NULL, &iatt);
 
-        return 0;
 out:
-	LOCK (&impunge_frame->lock);
-	{
-		call_count = --impunge_local->call_count;
-	}
-	UNLOCK (&impunge_frame->lock);
-
-	if (call_count == 0) {
-		AFR_STACK_DESTROY (impunge_frame);
-		afr_sh_entry_impunge_entry_done (frame, this, active_src);
-	}
-
-	return 0;
-}
+    if (xattr)
+        dict_unref(xattr);
+    loc_wipe(&rootloc);
+    GF_FREE((char *)loc.name);
+    loc_wipe(&loc);
 
-
-int
-afr_sh_entry_impunge_symlink_unlink (call_frame_t *impunge_frame, xlator_t *this,
-                                     int child_index)
-{
-	afr_private_t   *priv          = NULL;
-	afr_local_t     *impunge_local = NULL;
-
-	priv          = this->private;
-	impunge_local = impunge_frame->local;
-
-	gf_log (this->name, GF_LOG_DEBUG,
-		"unlinking symlink %s with wrong target on %s",
-		impunge_local->loc.path,
-		priv->children[child_index]->name);
-
-        STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_symlink_unlink_cbk,
-                           (void *) (long) child_index,
-                           priv->children[child_index],
-                           priv->children[child_index]->fops->unlink,
-                           &impunge_local->loc);
-
-        return 0;
+    return inode;
 }
 
-
-int
-afr_sh_entry_impunge_readlink_sink_cbk (call_frame_t *impunge_frame, void *cookie,
-                                        xlator_t *this,
-                                        int32_t op_ret, int32_t op_errno,
-                                        const char *linkname, struct iatt *sbuf)
+static int
+afr_selfheal_entry_do_subvol(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                             int child)
 {
-        afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-	int              child_index = -1;
-	call_frame_t    *frame = NULL;
-	int              call_count = -1;
-	int              active_src = -1;
-
-	priv          = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh    = &impunge_local->self_heal;
-	frame         = impunge_sh->sh_frame;
-	active_src    = impunge_sh->active_source;
-
-	child_index = (long) cookie;
-
-	if ((op_ret == -1) && (op_errno != ENOENT)) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"readlink of %s on %s failed (%s)",
-			impunge_local->loc.path,
-			priv->children[active_src]->name,
-			strerror (op_errno));
-		goto out;
-	}
-
-        /* symlink doesn't exist on the sink */
-
-        if ((op_ret == -1) && (op_errno == ENOENT)) {
-                afr_sh_entry_impunge_symlink (impunge_frame, this,
-                                              child_index, impunge_sh->linkname);
-                return 0;
-        }
-
-
-        /* symlink exists on the sink, so check if targets match */
-
-        if (strcmp (linkname, impunge_sh->linkname) == 0) {
-                /* targets match, nothing to do */
-
-                goto out;
-        } else {
-                /*
-                 * Hah! Sneaky wolf in sheep's clothing!
-                 */
-                afr_sh_entry_impunge_symlink_unlink (impunge_frame, this,
-                                                     child_index);
-                return 0;
+    int ret = 0;
+    gf_dirent_t entries;
+    gf_dirent_t *entry = NULL;
+    off_t offset = 0;
+    call_frame_t *iter_frame = NULL;
+    xlator_t *subvol = NULL;
+    afr_private_t *priv = NULL;
+    gf_boolean_t mismatch = _gf_false;
+    afr_local_t *local = NULL;
+    loc_t loc = {
+        0,
+    };
+
+    priv = this->private;
+    subvol = priv->children[child];
+
+    INIT_LIST_HEAD(&entries.list);
+
+    local = frame->local;
+
+    iter_frame = afr_copy_frame(frame);
+    if (!iter_frame)
+        return -ENOMEM;
+
+    loc.inode = afr_shd_entry_changes_index_inode(this, subvol,
+                                                  fd->inode->gfid);
+
+    while ((ret = syncop_readdir(subvol, fd, 131072, offset, &entries, NULL,
+                                 NULL))) {
+        if (ret > 0)
+            ret = 0;
+        list_for_each_entry(entry, &entries.list, list)
+        {
+            offset = entry->d_off;
+
+            if (!strcmp(entry->d_name, ".") || !strcmp(entry->d_name, ".."))
+                continue;
+
+            ret = afr_selfheal_entry_dirent(iter_frame, this, fd, entry->d_name,
+                                            loc.inode, subvol,
+                                            local->need_full_crawl);
+            AFR_STACK_RESET(iter_frame);
+            if (iter_frame->local == NULL) {
+                ret = -ENOTCONN;
+                break;
+            }
+
+            if (ret == -1) {
+                /* gfid or type mismatch. */
+                mismatch = _gf_true;
+                ret = 0;
+            }
+            if (ret)
+                break;
         }
 
-out:
-	LOCK (&impunge_frame->lock);
-	{
-		call_count = --impunge_local->call_count;
-	}
-	UNLOCK (&impunge_frame->lock);
-
-	if (call_count == 0) {
-		AFR_STACK_DESTROY (impunge_frame);
-		afr_sh_entry_impunge_entry_done (frame, this, active_src);
-	}
-
-	return 0;
-}
-
-
-int
-afr_sh_entry_impunge_readlink_sink (call_frame_t *impunge_frame, xlator_t *this,
-                                    int child_index)
-{
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-
-	gf_log (this->name, GF_LOG_DEBUG,
-		"checking symlink target of %s on %s",
-		impunge_local->loc.path, priv->children[child_index]->name);
-
-	STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_readlink_sink_cbk,
-			   (void *) (long) child_index,
-			   priv->children[child_index],
-			   priv->children[child_index]->fops->readlink,
-			   &impunge_local->loc, 4096);
-
-	return 0;
-}
-
-
-int
-afr_sh_entry_impunge_readlink_cbk (call_frame_t *impunge_frame, void *cookie,
-				   xlator_t *this,
-				   int32_t op_ret, int32_t op_errno,
-				   const char *linkname, struct iatt *sbuf)
-{
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-	int              child_index = -1;
-	call_frame_t    *frame = NULL;
-	int              call_count = -1;
-	int              active_src = -1;
-
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh = &impunge_local->self_heal;
-	frame = impunge_sh->sh_frame;
-	active_src = impunge_sh->active_source;
-
-	child_index = (long) cookie;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"readlink of %s on %s failed (%s)",
-			impunge_local->loc.path,
-			priv->children[active_src]->name,
-			strerror (op_errno));
-		goto out;
-	}
-
-        impunge_sh->linkname = gf_strdup (linkname);
-	afr_sh_entry_impunge_readlink_sink (impunge_frame, this, child_index);
-
-	return 0;
-
-out:
-	LOCK (&impunge_frame->lock);
-	{
-		call_count = --impunge_local->call_count;
-	}
-	UNLOCK (&impunge_frame->lock);
-
-	if (call_count == 0) {
-		AFR_STACK_DESTROY (impunge_frame);
-		afr_sh_entry_impunge_entry_done (frame, this, active_src);
-	}
-
-	return 0;
-}
+        gf_dirent_free(&entries);
+        if (ret)
+            break;
+    }
 
+    loc_wipe(&loc);
 
-int
-afr_sh_entry_impunge_readlink (call_frame_t *impunge_frame, xlator_t *this,
-			       int child_index, struct iatt *stbuf)
-{
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-	int              active_src = -1;
-
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh = &impunge_local->self_heal;
-	active_src = impunge_sh->active_source;
-        impunge_local->cont.symlink.buf = *stbuf;
-
-	STACK_WIND_COOKIE (impunge_frame, afr_sh_entry_impunge_readlink_cbk,
-			   (void *) (long) child_index,
-			   priv->children[active_src],
-			   priv->children[active_src]->fops->readlink,
-			   &impunge_local->loc, 4096);
-
-	return 0;
+    AFR_STACK_DESTROY(iter_frame);
+    if (mismatch == _gf_true)
+        /* undo pending will be skipped */
+        ret = -1;
+    return ret;
 }
 
-
-int
-afr_sh_entry_impunge_recreate_lookup_cbk (call_frame_t *impunge_frame,
-					  void *cookie, xlator_t *this,
-					  int32_t op_ret, int32_t op_errno,
-					  inode_t *inode, struct iatt *buf,
-					  dict_t *xattr,struct iatt *postparent)
+static int
+afr_selfheal_entry_granular_dirent(xlator_t *subvol, gf_dirent_t *entry,
+                                   loc_t *parent, void *data)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-	int              active_src = 0;
-	int              type = 0;
-	int              child_index = 0;
-	call_frame_t    *frame = NULL;
-	int              call_count = 0;
-
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh = &impunge_local->self_heal;
-	frame = impunge_sh->sh_frame;
-
-	child_index = (long) cookie;
-
-	active_src = impunge_sh->active_source;
-
-	if (op_ret != 0) {
-		gf_log (this->name, GF_LOG_TRACE,
-			"looking up %s on %s (for %s) failed (%s)",
-			impunge_local->loc.path,
-			priv->children[active_src]->name,
-			priv->children[child_index]->name,
-			strerror (op_errno));
-		goto out;
-	}
-
-        impunge_sh->parentbuf = *postparent;
-
-	impunge_local->cont.lookup.buf = *buf;
-	type = buf->ia_type;
-
-	switch (type) {
-	case IA_IFSOCK:
-	case IA_IFREG:
-	case IA_IFBLK:
-	case IA_IFCHR:
-	case IA_IFIFO:
-		afr_sh_entry_impunge_mknod (impunge_frame, this,
-					    child_index, buf);
-		break;
-	case IA_IFLNK:
-		afr_sh_entry_impunge_readlink (impunge_frame, this,
-					       child_index, buf);
-		break;
-	case IA_IFDIR:
-		afr_sh_entry_impunge_mkdir (impunge_frame, this,
-					    child_index, buf);
-		break;
-	default:
-		gf_log (this->name, GF_LOG_ERROR,
-			"%s has unknown file type on %s: 0%o",
-			impunge_local->loc.path,
-			priv->children[active_src]->name, type);
-		goto out;
-		break;
-	}
-
-	return 0;
+    int ret = 0;
+    loc_t loc = {
+        0,
+    };
+    struct iatt iatt = {
+        0,
+    };
+    afr_granular_esh_args_t *args = data;
+
+    /* Look up the actual inode associated with entry. If the lookup returns
+     * ESTALE or ENOENT, then it means we have a stale index. Remove it.
+     * This is analogous to the check in afr_shd_index_heal() except that
+     * here it is achieved through LOOKUP and in afr_shd_index_heal() through
+     * a GETXATTR.
+     */
+
+    loc.inode = inode_new(args->xl->itable);
+    loc.parent = inode_ref(args->heal_fd->inode);
+    gf_uuid_copy(loc.pargfid, loc.parent->gfid);
+    loc.name = entry->d_name;
+
+    ret = syncop_lookup(args->xl, &loc, &iatt, NULL, NULL, NULL);
+    if ((ret == -ENOENT) || (ret == -ESTALE)) {
+        /* The name indices under the pgfid index dir are guaranteed
+         * to be regular files. Hence the hardcoding.
+         */
+        afr_shd_entry_purge(subvol, parent->inode, entry->d_name, IA_IFREG);
+        ret = 0;
+        goto out;
+    }
+    /* TBD: afr_shd_zero_xattrop? */
+
+    ret = afr_selfheal_entry_dirent(args->frame, args->xl, args->heal_fd,
+                                    entry->d_name, parent->inode, subvol,
+                                    _gf_false);
+    AFR_STACK_RESET(args->frame);
+    if (args->frame->local == NULL)
+        ret = -ENOTCONN;
+
+    if (ret == -1)
+        args->mismatch = _gf_true;
 
 out:
-	LOCK (&impunge_frame->lock);
-	{
-		call_count = --impunge_local->call_count;
-	}
-	UNLOCK (&impunge_frame->lock);
-
-	if (call_count == 0) {
-		AFR_STACK_DESTROY (impunge_frame);
-		afr_sh_entry_impunge_entry_done (frame, this, active_src);
-	}
-
-	return 0;
+    loc_wipe(&loc);
+    return 0;
 }
 
-
-int
-afr_sh_entry_impunge_recreate (call_frame_t *impunge_frame, xlator_t *this,
-			       int child_index)
+static int
+afr_selfheal_entry_granular(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                            int subvol_idx, gf_boolean_t is_src)
 {
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-	int              active_src = 0;
-
-
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh = &impunge_local->self_heal;
-
-	active_src = impunge_sh->active_source;
-
-	STACK_WIND_COOKIE (impunge_frame,
-			   afr_sh_entry_impunge_recreate_lookup_cbk,
-			   (void *) (long) child_index,
-			   priv->children[active_src],
-			   priv->children[active_src]->fops->lookup,
-			   &impunge_local->loc, 0);
-
-	return 0;
-}
-
+    int ret = 0;
+    loc_t loc = {
+        0,
+    };
+    xlator_t *subvol = NULL;
+    afr_private_t *priv = NULL;
+    afr_granular_esh_args_t args = {
+        0,
+    };
+
+    priv = this->private;
+    subvol = priv->children[subvol_idx];
+
+    args.frame = afr_copy_frame(frame);
+    if (!args.frame)
+        goto out;
+    args.xl = this;
+    /* args.heal_fd represents the fd associated with the original directory
+     * on which entry heal is being attempted.
+     */
+    args.heal_fd = fd;
+
+    /* @subvol here represents the subvolume of AFR where
+     * indices/entry-changes/<pargfid> will be processed
+     */
+    loc.inode = afr_shd_entry_changes_index_inode(this, subvol,
+                                                  fd->inode->gfid);
+    if (!loc.inode) {
+        /* If granular heal failed on the sink (as it might sometimes
+         * because it is the src that would mostly contain the granular
+         * changelogs and the sink's entry-changes would be empty),
+         * do not treat heal as failure.
+         */
+        if (is_src)
+            ret = -errno;
+        else
+            ret = 0;
+        goto out;
+    }
 
-int
-afr_sh_entry_impunge_entry_cbk (call_frame_t *impunge_frame, void *cookie,
-				xlator_t *this,
-				int32_t op_ret,	int32_t op_errno,
-                                inode_t *inode, struct iatt *buf, dict_t *x,
-                                struct iatt *postparent)
-{
-	afr_private_t   *priv = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-	int              call_count = 0;
-	int              child_index = 0;
-	call_frame_t    *frame = NULL;
-	int              active_src = 0;
-
-	priv = this->private;
-	impunge_local = impunge_frame->local;
-	impunge_sh = &impunge_local->self_heal;
-	frame = impunge_sh->sh_frame;
-	child_index = (long) cookie;
-	active_src = impunge_sh->active_source;
-
-	if ((op_ret == -1 && op_errno == ENOENT)
-            || (IA_ISLNK (impunge_sh->impunging_entry_mode))) {
-
-                /*
-                 * A symlink's target might have changed, so
-                 * always go down the recreate path for them.
-                 */
-
-		/* decrease call_count in recreate-callback */
-
-		gf_log (this->name, GF_LOG_TRACE,
-			"missing entry %s on %s",
-			impunge_local->loc.path,
-			priv->children[child_index]->name);
-
-		afr_sh_entry_impunge_recreate (impunge_frame, this,
-					       child_index);
-		return 0;
-	}
-
-	if (op_ret == 0) {
-		gf_log (this->name, GF_LOG_TRACE,
-			"%s exists under %s",
-			impunge_local->loc.path,
-			priv->children[child_index]->name);
-
-                impunge_sh->parentbuf = *postparent;
-	} else {
-		gf_log (this->name, GF_LOG_TRACE,
-			"looking up %s under %s failed (%s)",
-			impunge_local->loc.path,
-			priv->children[child_index]->name,
-			strerror (op_errno));
-	}
-
-	LOCK (&impunge_frame->lock);
-	{
-		call_count = --impunge_local->call_count;
-	}
-	UNLOCK (&impunge_frame->lock);
-
-	if (call_count == 0) {
-		AFR_STACK_DESTROY (impunge_frame);
-		afr_sh_entry_impunge_entry_done (frame, this, active_src);
-	}
-
-	return 0;
-}
+    ret = syncop_dir_scan(subvol, &loc, GF_CLIENT_PID_SELF_HEALD, &args,
+                          afr_selfheal_entry_granular_dirent);
 
+    loc_wipe(&loc);
 
-int
-afr_sh_entry_impunge_entry (call_frame_t *frame, xlator_t *this,
-			    gf_dirent_t *entry)
-{
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	int              ret = -1;
-	call_frame_t    *impunge_frame = NULL;
-	afr_local_t     *impunge_local = NULL;
-	afr_self_heal_t *impunge_sh = NULL;
-	int              active_src = 0;
-	int              i = 0;
-	int              call_count = 0;
-	int              op_errno = 0;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	active_src = sh->active_source;
-
-	if ((strcmp (entry->d_name, ".") == 0)
-	    || (strcmp (entry->d_name, "..") == 0)
-            || ((strcmp (local->loc.path, "/") == 0)
-                && (strcmp (entry->d_name, GF_REPLICATE_TRASH_DIR) == 0))) {
-
-		gf_log (this->name, GF_LOG_TRACE,
-			"skipping inspection of %s under %s",
-			entry->d_name, local->loc.path);
-		goto out;
-	}
-
-	gf_log (this->name, GF_LOG_TRACE,
-		"inspecting existance of %s under %s",
-		entry->d_name, local->loc.path);
-
-	impunge_frame = copy_frame (frame);
-	if (!impunge_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		goto out;
-	}
-
-	ALLOC_OR_GOTO (impunge_local, afr_local_t, out);
-
-	impunge_frame->local = impunge_local;
-	impunge_sh = &impunge_local->self_heal;
-	impunge_sh->sh_frame = frame;
-	impunge_sh->active_source = active_src;
-
-        impunge_sh->impunging_entry_mode =
-                st_mode_from_ia (entry->d_stat.ia_prot, entry->d_stat.ia_type);
-
-	ret = build_child_loc (this, &impunge_local->loc, &local->loc, entry->d_name);
-	if (ret != 0) {
-		goto out;
-	}
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (i == active_src)
-			continue;
-		if (local->child_up[i] == 0)
-			continue;
-		if (sh->sources[i] == 1)
-			continue;
-		call_count++;
-	}
-
-	impunge_local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (i == active_src)
-			continue;
-		if (local->child_up[i] == 0)
-			continue;
-		if (sh->sources[i] == 1)
-			continue;
-
-		gf_log (this->name, GF_LOG_TRACE,
-			"looking up %s on %s", impunge_local->loc.path,
-			priv->children[i]->name);
-
-		STACK_WIND_COOKIE (impunge_frame,
-				   afr_sh_entry_impunge_entry_cbk,
-				   (void *) (long) i,
-				   priv->children[i],
-				   priv->children[i]->fops->lookup,
-				   &impunge_local->loc, 0);
-
-		if (!--call_count)
-			break;
-	}
-
-	ret = 0;
+    if (args.mismatch == _gf_true)
+        ret = -1;
 out:
-	if (ret == -1)
-		afr_sh_entry_impunge_entry_done (frame, this, active_src);
-
-	return 0;
-}
-
-
-int
-afr_sh_entry_impunge_readdir_cbk (call_frame_t *frame, void *cookie,
-				  xlator_t *this,
-				  int32_t op_ret, int32_t op_errno,
-				  gf_dirent_t *entries)
-{
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	gf_dirent_t     *entry = NULL;
-	off_t            last_offset = 0;
-	int              active_src = 0;
-	int              entry_count = 0;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	active_src = sh->active_source;
-
-	if (op_ret <= 0) {
-		if (op_ret < 0) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"readdir of %s on subvolume %s failed (%s)",
-				local->loc.path,
-				priv->children[active_src]->name,
-				strerror (op_errno));
-		} else {
-			gf_log (this->name, GF_LOG_TRACE,
-				"readdir of %s on subvolume %s complete",
-				local->loc.path,
-				priv->children[active_src]->name);
-		}
-
-		afr_sh_entry_impunge_all (frame, this);
-		return 0;
-	}
-
-	list_for_each_entry (entry, &entries->list, list) {
-		last_offset = entry->d_off;
-                entry_count++;
-	}
-
-	gf_log (this->name, GF_LOG_TRACE,
-		"readdir'ed %d entries from %s",
-		entry_count, priv->children[active_src]->name);
-
-	sh->offset = last_offset;
-	local->call_count = entry_count;
-
-	list_for_each_entry (entry, &entries->list, list) {
-                afr_sh_entry_impunge_entry (frame, this, entry);
-	}
-
-	return 0;
-}
-				  
-
-int
-afr_sh_entry_impunge_subvol (call_frame_t *frame, xlator_t *this,
-			     int active_src)
-{
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	STACK_WIND (frame, afr_sh_entry_impunge_readdir_cbk,
-		    priv->children[active_src],
-		    priv->children[active_src]->fops->readdirp,
-		    sh->healing_fd, sh->block_size, sh->offset);
-
-	return 0;
-}
-
-
-int
-afr_sh_entry_impunge_all (call_frame_t *frame, xlator_t *this)
-{
-	afr_private_t   *priv = NULL;
-	afr_local_t     *local  = NULL;
-	afr_self_heal_t *sh  = NULL;
-	int              active_src = -1;
-
-	priv = this->private;
-	local = frame->local;
-	sh = &local->self_heal;
-
-	sh->offset = 0;
-
-	active_src = next_active_source (frame, this, sh->active_source);
-	sh->active_source = active_src;
-
-	if (sh->op_failed) {
-		afr_sh_entry_finish (frame, this);
-		return 0;
-	}
-
-	if (active_src == -1) {
-		/* completed creating missing files on all subvolumes */
-		afr_sh_entry_expunge_all (frame, this);
-		return 0;
-	}
-
-	gf_log (this->name, GF_LOG_TRACE,
-		"impunging entries of %s on %s to other sinks",
-		local->loc.path, priv->children[active_src]->name);
-
-	afr_sh_entry_impunge_subvol (frame, this, active_src);
-
-	return 0;
-}
-
-
-int
-afr_sh_entry_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			  int32_t op_ret, int32_t op_errno, fd_t *fd)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              call_count = 0;
-	int              child_index = 0;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	child_index = (long) cookie;
-
-	/* TODO: some of the open's might fail.
-	   In that case, modify cleanup fn to send flush on those 
-	   fd's which are already open */
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"opendir of %s failed on child %s (%s)",
-				local->loc.path,
-				priv->children[child_index]->name,
-				strerror (op_errno));
-			sh->op_failed = 1;
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		if (sh->op_failed) {
-			afr_sh_entry_finish (frame, this);
-			return 0;
-		}
-		gf_log (this->name, GF_LOG_TRACE,
-			"fd for %s opened, commencing sync",
-			local->loc.path);
-
-		sh->active_source = -1;
-		afr_sh_entry_impunge_all (frame, this);
-	}
-
-	return 0;
-}
-
-
-int
-afr_sh_entry_open (call_frame_t *frame, xlator_t *this)
-{
-	int i = 0;				
-	int call_count = 0;		     
-
-	int source = -1;
-	int *sources = NULL;
-
-	fd_t *fd = NULL;
-
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	afr_self_heal_t *sh = NULL;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	source  = local->self_heal.source;
-	sources = local->self_heal.sources;
-
-	sh->block_size = 65536; //131072
-	sh->offset = 0;
-
-	call_count = sh->active_sinks;
-	if (source != -1)
-		call_count++;
-
-	local->call_count = call_count;
-
-	fd = fd_create (local->loc.inode, frame->root->pid);
-	sh->healing_fd = fd;
-
-	if (source != -1) {
-		gf_log (this->name, GF_LOG_TRACE,
-			"opening directory %s on subvolume %s (source)",
-			local->loc.path, priv->children[source]->name);
-
-		/* open source */
-		STACK_WIND_COOKIE (frame, afr_sh_entry_opendir_cbk,
-				   (void *) (long) source,
-				   priv->children[source],
-				   priv->children[source]->fops->opendir,
-				   &local->loc, fd);
-		call_count--;
-	}
-
-	/* open sinks */
-	for (i = 0; i < priv->child_count; i++) {
-		if (sources[i] || !local->child_up[i])
-			continue;
-
-		gf_log (this->name, GF_LOG_TRACE,
-			"opening directory %s on subvolume %s (sink)",
-			local->loc.path, priv->children[i]->name);
-
-		STACK_WIND_COOKIE (frame, afr_sh_entry_opendir_cbk,
-				   (void *) (long) i,
-				   priv->children[i], 
-				   priv->children[i]->fops->opendir,
-				   &local->loc, fd);
-
-		if (!--call_count)
-			break;
-	}
-
-	return 0;
+    if (args.frame)
+        AFR_STACK_DESTROY(args.frame);
+    return ret;
 }
 
-
-int
-afr_sh_entry_sync_prepare (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              active_sinks = 0;
-	int              source = 0;
-	int              i = 0;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	source = sh->source;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->sources[i] == 0 && local->child_up[i] == 1) {
-			active_sinks++;
-			sh->success[i] = 1;
-		}
-	}
-	if (source != -1)
-		sh->success[source] = 1;
-
-	if (active_sinks == 0) {
-		gf_log (this->name, GF_LOG_TRACE,
-			"no active sinks for self-heal on dir %s",
-			local->loc.path);
-		afr_sh_entry_finish (frame, this);
-		return 0;
-	}
-	if (source == -1 && active_sinks < 2) {
-		gf_log (this->name, GF_LOG_TRACE,
-			"cannot sync with 0 sources and 1 sink on dir %s",
-			local->loc.path);
-		afr_sh_entry_finish (frame, this);
-		return 0;
-	}
-	sh->active_sinks = active_sinks;
-
-	if (source != -1)
-		gf_log (this->name, GF_LOG_DEBUG,
-			"self-healing directory %s from subvolume %s to "
-                        "%d other",
-			local->loc.path, priv->children[source]->name,
-			active_sinks);
-	else
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no active sources for %s found. "
-			"merging all entries as a conservative decision",
-			local->loc.path);
-
-	afr_sh_entry_open (frame, this);
-
-	return 0;
-}
-
-
-int
-afr_sh_entry_fix (call_frame_t *frame, xlator_t *this)
+static int
+afr_selfheal_entry_do(call_frame_t *frame, xlator_t *this, fd_t *fd, int source,
+                      unsigned char *sources, unsigned char *healed_sinks)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              source = 0;
-
-        int nsources = 0;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-        if (sh->forced_merge) {
-                sh->source = -1;
-                goto heal;
-        }
-
-	afr_sh_build_pending_matrix (priv, sh->pending_matrix, sh->xattr, 
-				     priv->child_count, AFR_ENTRY_TRANSACTION);
-
-	afr_sh_print_pending_matrix (sh->pending_matrix, this);
-
-	nsources = afr_sh_mark_sources (sh, priv->child_count,
-                                        AFR_SELF_HEAL_ENTRY);
-
-        if (nsources == 0) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "No self-heal needed for %s",
-                        local->loc.path);
+    int i = 0;
+    int ret = 0;
+    gf_boolean_t mismatch = _gf_false;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_SELF_HEAL_INFO,
+           "performing entry selfheal on %s", uuid_utoa(fd->inode->gfid));
+
+    for (i = 0; i < priv->child_count; i++) {
+        /* Expunge */
+        if (!healed_sinks[i])
+            continue;
+
+        if (!local->need_full_crawl)
+            /* Why call afr_selfheal_entry_granular() on a "healed sink",
+             * given that it is the source that contains the granular
+             * indices?
+             * If the index for this directory is non-existent or empty on
+             * this subvol (=> clear sink), then it will return early
+             * without failure status.
+             * If the index is non-empty and it is yet a 'healed sink', then
+             * it is due to a split-brain in which case we anyway need to
+             * crawl the indices/entry-changes/pargfid directory.
+             */
+            ret = afr_selfheal_entry_granular(frame, this, fd, i, _gf_false);
+        else
+            ret = afr_selfheal_entry_do_subvol(frame, this, fd, i);
 
-                afr_sh_entry_finish (frame, this);
-                return 0;
+        if (ret == -1) {
+            /* gfid or type mismatch. */
+            mismatch = _gf_true;
+            ret = 0;
         }
+        if (ret)
+            break;
+    }
 
-	afr_sh_supress_errenous_children (sh->sources, sh->child_errno,
-					  priv->child_count);
-
-	source = afr_sh_select_source (sh->sources, priv->child_count);
-
-        sh->source = source;
-
-heal:
-	afr_sh_entry_sync_prepare (frame, this);
-
-	return 0;
-}
-
-
-
-int
-afr_sh_entry_lookup_cbk (call_frame_t *frame, void *cookie,
-			 xlator_t *this, int32_t op_ret, int32_t op_errno,
-                         inode_t *inode, struct iatt *buf, dict_t *xattr,
-                         struct iatt *postparent)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-
-	int call_count  = -1;
-	int child_index = (long) cookie;
-
-	local = frame->local;
-	sh = &local->self_heal;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret != -1) {
-			sh->xattr[child_index] = dict_ref (xattr);
-			sh->buf[child_index] = *buf;
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		afr_sh_entry_fix (frame, this);
-	}
-
-	return 0;
+    if (!ret && source != -1) {
+        /* Impunge */
+        if (local->need_full_crawl)
+            ret = afr_selfheal_entry_do_subvol(frame, this, fd, source);
+        else
+            ret = afr_selfheal_entry_granular(frame, this, fd, source,
+                                              _gf_true);
+    }
+
+    if (mismatch == _gf_true)
+        /* undo pending will be skipped */
+        ret = -1;
+    return ret;
 }
 
-
-
-int
-afr_sh_entry_lookup (call_frame_t *frame, xlator_t *this)
+static int
+__afr_selfheal_entry(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                     unsigned char *locked_on)
 {
-	afr_local_t    *  local = NULL;
-	afr_private_t  *  priv  = NULL;
-	dict_t         *xattr_req = NULL;
-	int ret = 0;
-	int call_count = 0;
-	int i = 0;
-
-	priv  = this->private;
-	local = frame->local;
-
-	call_count = afr_up_children_count (priv->child_count,
-                                            local->child_up);
-
-	local->call_count = call_count;
-
-	xattr_req = dict_new();
-	if (xattr_req) {
-                for (i = 0; i < priv->child_count; i++) {
-                        ret = dict_set_uint64 (xattr_req,
-                                               priv->pending_key[i],
-                                               3 * sizeof(int32_t));
-                        if (ret < 0)
-                                gf_log (this->name, GF_LOG_WARNING,
-                                        "Unable to set dict value.");
-                }
+    int ret = -1;
+    int source = -1;
+    unsigned char *sources = NULL;
+    unsigned char *sinks = NULL;
+    unsigned char *data_lock = NULL;
+    unsigned char *postop_lock = NULL;
+    unsigned char *healed_sinks = NULL;
+    unsigned char *undid_pending = NULL;
+    struct afr_reply *locked_replies = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    gf_boolean_t did_sh = _gf_true;
+
+    priv = this->private;
+    local = frame->local;
+
+    sources = alloca0(priv->child_count);
+    sinks = alloca0(priv->child_count);
+    healed_sinks = alloca0(priv->child_count);
+    undid_pending = alloca0(priv->child_count);
+    data_lock = alloca0(priv->child_count);
+    postop_lock = alloca0(priv->child_count);
+
+    locked_replies = alloca0(sizeof(*locked_replies) * priv->child_count);
+
+    ret = afr_selfheal_entrylk(frame, this, fd->inode, this->name, NULL,
+                               data_lock);
+    {
+        if (ret < priv->child_count) {
+            gf_msg_debug(this->name, 0,
+                         "%s: Skipping "
+                         "entry self-heal as only %d sub-volumes could "
+                         "be locked in %s domain",
+                         uuid_utoa(fd->inode->gfid), ret, this->name);
+            ret = -ENOTCONN;
+            goto unlock;
         }
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame,
-					   afr_sh_entry_lookup_cbk,
-					   (void *) (long) i,
-					   priv->children[i],
-					   priv->children[i]->fops->lookup,
-					   &local->loc, xattr_req);
-			if (!--call_count)
-				break;
-		}
-	}
-
-	if (xattr_req)
-		dict_unref (xattr_req);
-
-	return 0;
-}
-
-int
-afr_sh_post_nonblocking_entry_cbk (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Non Blocking entrylks failed.");
-                afr_sh_entry_done (frame, this);
-        } else {
+        ret = __afr_selfheal_entry_prepare(frame, this, fd->inode, data_lock,
+                                           sources, sinks, healed_sinks,
+                                           locked_replies, &source, NULL);
+        if (AFR_COUNT(healed_sinks, priv->child_count) == 0) {
+            did_sh = _gf_false;
+            goto unlock;
+        }
 
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Non Blocking entrylks done. Proceeding to FOP");
-                afr_sh_entry_lookup(frame, this);
+        local->need_full_crawl = afr_need_full_heal(
+            this, locked_replies, source, healed_sinks, AFR_ENTRY_TRANSACTION);
+    }
+unlock:
+    afr_selfheal_unentrylk(frame, this, fd->inode, this->name, NULL, data_lock,
+                           NULL);
+    if (ret < 0)
+        goto out;
+
+    if (!did_sh)
+        goto out;
+
+    ret = afr_selfheal_entry_do(frame, this, fd, source, sources, healed_sinks);
+    if (ret)
+        goto out;
+
+    /* Take entrylks in xlator domain before doing post-op (undo-pending) in
+     * entry self-heal. This is to prevent a parallel name self-heal on
+     * an entry under @fd->inode from reading pending xattrs while it is
+     * being modified by SHD after entry sh below, given that
+     * name self-heal takes locks ONLY in xlator domain and is free to read
+     * pending changelog in the absence of the following locking.
+     */
+    ret = afr_selfheal_entrylk(frame, this, fd->inode, this->name, NULL,
+                               postop_lock);
+    {
+        if (AFR_CMP(data_lock, postop_lock, priv->child_count) != 0) {
+            gf_msg_debug(this->name, 0,
+                         "%s: Skipping "
+                         "post-op after entry self-heal as %d "
+                         "sub-volumes, as opposed to %d, "
+                         "could be locked in %s domain",
+                         uuid_utoa(fd->inode->gfid), ret,
+                         AFR_COUNT(data_lock, priv->child_count), this->name);
+            ret = -ENOTCONN;
+            goto postop_unlock;
         }
 
-        return 0;
+        afr_selfheal_restore_time(frame, this, fd->inode, source, healed_sinks,
+                                  locked_replies);
+        ret = afr_selfheal_undo_pending(
+            frame, this, fd->inode, sources, sinks, healed_sinks, undid_pending,
+            AFR_ENTRY_TRANSACTION, locked_replies, postop_lock);
+    }
+postop_unlock:
+    afr_selfheal_unentrylk(frame, this, fd->inode, this->name, NULL,
+                           postop_lock, NULL);
+out:
+    if (did_sh)
+        afr_log_selfheal(fd->inode->gfid, this, ret, "entry", source, sources,
+                         healed_sinks);
+    else
+        ret = 1;
+
+    if (locked_replies)
+        afr_replies_wipe(locked_replies, priv->child_count);
+    return ret;
 }
 
-int
-afr_sh_entry_lock (call_frame_t *frame, xlator_t *this)
+static fd_t *
+afr_selfheal_data_opendir(xlator_t *this, inode_t *inode)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        int_lock->transaction_lk_type = AFR_SELFHEAL_LK;
-        int_lock->selfheal_lk_type    = AFR_ENTRY_SELF_HEAL_LK;
-
-        afr_set_lock_number (frame, this);
-
-        int_lock->lk_basename = NULL;
-        int_lock->lk_loc      = &local->loc;
-        int_lock->lock_cbk    = afr_sh_post_nonblocking_entry_cbk;
-
-        afr_nonblocking_entrylk (frame, this);
-
-
-	return 0;
+    loc_t loc = {
+        0,
+    };
+    int ret = 0;
+    fd_t *fd = NULL;
+
+    fd = fd_create(inode, 0);
+    if (!fd)
+        return NULL;
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    ret = syncop_opendir(this, &loc, fd, NULL, NULL);
+    if (ret) {
+        fd_unref(fd);
+        fd = NULL;
+    } else {
+        fd_bind(fd);
+    }
+
+    loc_wipe(&loc);
+    return fd;
 }
 
-
 int
-afr_self_heal_entry (call_frame_t *frame, xlator_t *this)
+afr_selfheal_entry(call_frame_t *frame, xlator_t *this, inode_t *inode)
 {
-	afr_local_t   *local = NULL;
-	afr_private_t   *priv = NULL;
-
+    afr_private_t *priv = NULL;
+    unsigned char *locked_on = NULL;
+    fd_t *fd = NULL;
+    int ret = 0;
+
+    priv = this->private;
+
+    fd = afr_selfheal_data_opendir(this, inode);
+    if (!fd)
+        return -EIO;
+
+    locked_on = alloca0(priv->child_count);
+
+    ret = afr_selfheal_tie_breaker_entrylk(frame, this, inode, priv->sh_domain,
+                                           NULL, locked_on);
+    {
+        if (ret < priv->child_count) {
+            gf_msg_debug(this->name, 0,
+                         "%s: Skipping "
+                         "entry self-heal as only %d sub-volumes could "
+                         "be locked in %s domain",
+                         uuid_utoa(fd->inode->gfid), ret, priv->sh_domain);
+            /* Either less than two subvols available, or another
+               selfheal (from another server) is in progress. Skip
+               for now in any case there isn't anything to do.
+            */
+            ret = -ENOTCONN;
+            goto unlock;
+        }
 
-	priv = this->private;
-	local = frame->local;
+        ret = __afr_selfheal_entry(frame, this, fd, locked_on);
+    }
+unlock:
+    afr_selfheal_unentrylk(frame, this, inode, priv->sh_domain, NULL, locked_on,
+                           NULL);
 
-	if (local->self_heal.need_entry_self_heal && priv->entry_self_heal) {
-		afr_sh_entry_lock (frame, this);
-	} else {
-		gf_log (this->name, GF_LOG_TRACE,
-			"proceeding to completion on %s",
-			local->loc.path);
-		afr_sh_entry_done (frame, this);
-	}
+    if (fd)
+        fd_unref(fd);
 
-	return 0;
+    return ret;
 }
-
diff --git a/xlators/cluster/afr/src/afr-self-heal-metadata.c b/xlators/cluster/afr/src/afr-self-heal-metadata.c
index 124bcdb1037..03f43bad16e 100644
--- a/xlators/cluster/afr/src/afr-self-heal-metadata.c
+++ b/xlators/cluster/afr/src/afr-self-heal-metadata.c
@@ -1,731 +1,546 @@
 /*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-#include <libgen.h>
-#include <unistd.h>
-#include <fnmatch.h>
-#include <sys/time.h>
-#include <stdlib.h>
-#include <signal.h>
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
 
-#include "glusterfs.h"
 #include "afr.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-#include "byte-order.h"
-
-#include "afr-transaction.h"
 #include "afr-self-heal.h"
-#include "afr-self-heal-common.h"
+#include <glusterfs/byte-order.h>
+#include "protocol-common.h"
+#include <glusterfs/events.h>
 
+#define AFR_HEAL_ATTR (GF_SET_ATTR_UID | GF_SET_ATTR_GID | GF_SET_ATTR_MODE)
 
-int
-afr_sh_metadata_done (call_frame_t *frame, xlator_t *this)
+static gf_boolean_t
+_afr_ignorable_key_match(dict_t *d, char *k, data_t *val, void *mdata)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              i = 0;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-//	memset (sh->child_errno, 0, sizeof (int) * priv->child_count);
-	memset (sh->buf, 0, sizeof (struct iatt) * priv->child_count);
-	memset (sh->success, 0, sizeof (int) * priv->child_count);
-
-/*         for (i = 0; i < priv->child_count; i++) { */
-/*                 sh->locked_nodes[i] = 1; */
-/*         } */
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->xattr[i])
-			dict_unref (sh->xattr[i]);
-		sh->xattr[i] = NULL;
-	}
-
-	if (local->govinda_gOvinda) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"aborting selfheal of %s",
-			local->loc.path);
-		sh->completion_cbk (frame, this);
-	} else {
-		if (IA_ISREG (sh->type)) {
-			gf_log (this->name, GF_LOG_TRACE,
-				"proceeding to data check on %s",
-				local->loc.path);
-			afr_self_heal_data (frame, this);
-			return 0;
-		}
-
-		if (IA_ISDIR (sh->type)) {
-			gf_log (this->name, GF_LOG_TRACE,
-				"proceeding to entry check on %s",
-				local->loc.path);
-			afr_self_heal_entry (frame, this);
-			return 0;
-		}
-		gf_log (this->name, GF_LOG_DEBUG,
-			"completed self heal of %s",
-			local->loc.path);
-
-		sh->completion_cbk (frame, this);
-	}
-
-	return 0;
+    return afr_is_xattr_ignorable(k);
 }
 
-
-int
-afr_sh_metadata_unlck_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			   int32_t op_ret, int32_t op_errno)
+void
+afr_delete_ignorable_xattrs(dict_t *xattr)
 {
-	int               call_count = 0;
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0)
-		afr_sh_metadata_done (frame, this);
-
-	return 0;
+    dict_foreach_match(xattr, _afr_ignorable_key_match, NULL,
+                       dict_remove_foreach_fn, NULL);
 }
 
 int
-afr_sh_inode_unlock (call_frame_t *frame, xlator_t *this)
+__afr_selfheal_metadata_do(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                           int source, unsigned char *healed_sinks,
+                           struct afr_reply *locked_replies)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        int_lock->lock_cbk = afr_sh_metadata_done;
-        afr_unlock (frame, this);
-
-        return 0;
-}
-
-int
-afr_sh_metadata_finish (call_frame_t *frame, xlator_t *this)
-{
-        afr_sh_inode_unlock (frame, this);
-
-	return 0;
-}
-
-
-int
-afr_sh_metadata_erase_pending_cbk (call_frame_t *frame, void *cookie,
-				   xlator_t *this, int32_t op_ret,
-				   int32_t op_errno, dict_t *xattr)
-{
-	afr_local_t     *local = NULL;
-	int             call_count = 0;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
+    int ret = -1;
+    loc_t loc = {
+        0,
+    };
+    dict_t *xattr = NULL;
+    dict_t *old_xattr = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+
+    priv = this->private;
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_SELF_HEAL_INFO,
+           "performing metadata selfheal on %s", uuid_utoa(inode->gfid));
+
+    ret = syncop_getxattr(priv->children[source], &loc, &xattr, NULL, NULL,
+                          NULL);
+    if (ret < 0) {
+        ret = -EIO;
+        goto out;
+    }
+
+    afr_delete_ignorable_xattrs(xattr);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (old_xattr) {
+            dict_unref(old_xattr);
+            old_xattr = NULL;
+        }
 
-	if (call_count == 0)
-		afr_sh_metadata_finish (frame, this);
+        if (!healed_sinks[i])
+            continue;
+
+        ret = syncop_setattr(priv->children[i], &loc,
+                             &locked_replies[source].poststat, AFR_HEAL_ATTR,
+                             NULL, NULL, NULL, NULL);
+        if (ret)
+            healed_sinks[i] = 0;
+
+        ret = syncop_getxattr(priv->children[i], &loc, &old_xattr, 0, NULL,
+                              NULL);
+        if (old_xattr) {
+            afr_delete_ignorable_xattrs(old_xattr);
+            ret = syncop_removexattr(priv->children[i], &loc, "", old_xattr,
+                                     NULL);
+            if (ret)
+                healed_sinks[i] = 0;
+        }
 
-	return 0;
-}
+        ret = syncop_setxattr(priv->children[i], &loc, xattr, 0, NULL, NULL);
+        if (ret)
+            healed_sinks[i] = 0;
+    }
+    ret = 0;
 
+out:
+    loc_wipe(&loc);
+    if (xattr)
+        dict_unref(xattr);
+    if (old_xattr)
+        dict_unref(old_xattr);
 
-int
-afr_sh_metadata_erase_pending (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              call_count = 0;
-	int              i = 0;
-	dict_t          **erase_xattr = NULL;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	afr_sh_pending_to_delta (priv, sh->xattr, sh->delta_matrix, 
-                                 sh->success, priv->child_count,
-                                 AFR_METADATA_TRANSACTION);
-
-	erase_xattr = GF_CALLOC (sizeof (*erase_xattr), priv->child_count,
-                                 gf_afr_mt_dict_t);
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->xattr[i]) {
-			call_count++;
-
-			erase_xattr[i] = get_new_dict();
-			dict_ref (erase_xattr[i]);
-		}
-	}
-
-	afr_sh_delta_to_xattr (priv, sh->delta_matrix, erase_xattr,
-			       priv->child_count, AFR_METADATA_TRANSACTION);
-
-	local->call_count = call_count;
-
-	if (call_count == 0) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"metadata of %s not healed on any subvolume",
-			local->loc.path);
-
-		afr_sh_metadata_finish (frame, this);
-	}
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (!erase_xattr[i])
-			continue;
-
-		gf_log (this->name, GF_LOG_TRACE,
-			"erasing pending flags from %s on %s",
-			local->loc.path, priv->children[i]->name);
-
-		STACK_WIND_COOKIE (frame, afr_sh_metadata_erase_pending_cbk,
-				   (void *) (long) i,
-				   priv->children[i],
-				   priv->children[i]->fops->xattrop,
-				   &local->loc,
-				   GF_XATTROP_ADD_ARRAY, erase_xattr[i]);
-		if (!--call_count)
-			break;
-	}
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (erase_xattr[i]) {
-			dict_unref (erase_xattr[i]);
-		}
-	}
-	GF_FREE (erase_xattr);
-
-	return 0;
+    return ret;
 }
 
-
-int
-afr_sh_metadata_sync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			  int32_t op_ret, int32_t op_errno)
+static uint64_t
+mtime_ns(struct iatt *ia)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              call_count = 0;
-	int              child_index = 0;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	child_index = (long) cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"setting attributes failed for %s on %s (%s)",
-				local->loc.path,
-				priv->children[child_index]->name,
-				strerror (op_errno));
-
-			sh->success[child_index] = 0;
-		}
-	}
-	UNLOCK (&frame->lock);
+    uint64_t ret;
 
-	call_count = afr_frame_return (frame);
+    ret = (((uint64_t)(ia->ia_mtime)) * 1000000000) +
+          (uint64_t)(ia->ia_mtime_nsec);
 
-	if (call_count == 0)
-		afr_sh_metadata_erase_pending (frame, this);
-
-	return 0;
+    return ret;
 }
 
-
-int
-afr_sh_metadata_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                             int32_t op_ret, int32_t op_errno,
-                             struct iatt *preop, struct iatt *postop)
+/*
+ * When directory content is modified, [mc]time is updated. On
+ * Linux, the filesystem does it, while at least on NetBSD, the
+ * kernel file-system independent code does it. This means that
+ * when entries are added while bricks are down, the kernel sends
+ * a SETATTR [mc]time which will cause metadata split brain for
+ * the directory. In this case, clear the split brain by finding
+ * the source with the most recent modification date.
+ */
+static int
+afr_dirtime_splitbrain_source(call_frame_t *frame, xlator_t *this,
+                              struct afr_reply *replies,
+                              unsigned char *locked_on)
 {
-	afr_sh_metadata_sync_cbk (frame, cookie, this, op_ret, op_errno);
-
-	return 0;
+    afr_private_t *priv = NULL;
+    int source = -1;
+    struct iatt source_ia;
+    struct iatt child_ia;
+    uint64_t mtime = 0;
+    int i;
+    int ret = -1;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!locked_on[i])
+            continue;
+
+        if (!replies[i].valid)
+            continue;
+
+        if (replies[i].op_ret != 0)
+            continue;
+
+        if (mtime_ns(&replies[i].poststat) <= mtime)
+            continue;
+
+        mtime = mtime_ns(&replies[i].poststat);
+        source = i;
+    }
+
+    if (source == -1)
+        goto out;
+
+    source_ia = replies[source].poststat;
+    if (source_ia.ia_type != IA_IFDIR)
+        goto out;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (i == source)
+            continue;
+
+        if (!replies[i].valid)
+            continue;
+
+        if (replies[i].op_ret != 0)
+            continue;
+
+        child_ia = replies[i].poststat;
+
+        if (!IA_EQUAL(source_ia, child_ia, gfid) ||
+            !IA_EQUAL(source_ia, child_ia, type) ||
+            !IA_EQUAL(source_ia, child_ia, prot) ||
+            !IA_EQUAL(source_ia, child_ia, uid) ||
+            !IA_EQUAL(source_ia, child_ia, gid) ||
+            !afr_xattrs_are_equal(replies[source].xdata, replies[i].xdata))
+            goto out;
+    }
+
+    /*
+     * Metadata split brain is just about [amc]time
+     * We return our source.
+     */
+    ret = source;
+out:
+    return ret;
 }
 
-
-int
-afr_sh_metadata_xattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			   int32_t op_ret, int32_t op_errno)
+static int
+__afr_selfheal_metadata_mark_pending_xattrs(call_frame_t *frame, xlator_t *this,
+                                            inode_t *inode,
+                                            struct afr_reply *replies,
+                                            unsigned char *sources)
 {
-	afr_sh_metadata_sync_cbk (frame, cookie, this, op_ret, op_errno);
-
-	return 0;
-}
+    int ret = 0;
+    int i = 0;
+    int m_idx = 0;
+    afr_private_t *priv = NULL;
+    int raw[AFR_NUM_CHANGE_LOGS] = {0};
+    dict_t *xattr = NULL;
+
+    priv = this->private;
+    m_idx = afr_index_for_transaction_type(AFR_METADATA_TRANSACTION);
+    raw[m_idx] = 1;
+
+    xattr = dict_new();
+    if (!xattr)
+        return -ENOMEM;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (sources[i])
+            continue;
+        ret = dict_set_static_bin(xattr, priv->pending_key[i], raw,
+                                  sizeof(int) * AFR_NUM_CHANGE_LOGS);
+        if (ret) {
+            ret = -1;
+            goto out;
+        }
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+        ret = afr_selfheal_post_op(frame, this, inode, i, xattr, NULL);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_INFO, -ret, AFR_MSG_SELF_HEAL_INFO,
+                   "Failed to set pending metadata xattr on child %d for %s", i,
+                   uuid_utoa(inode->gfid));
+            goto out;
+        }
+    }
 
+    afr_replies_wipe(replies, priv->child_count);
+    ret = afr_selfheal_unlocked_discover(frame, inode, inode->gfid, replies);
 
-int
-afr_sh_metadata_sync (call_frame_t *frame, xlator_t *this, dict_t *xattr)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              source = 0;
-	int              active_sinks = 0;
-	int              call_count = 0;
-	int              i = 0;
-
-        struct iatt      stbuf;
-        int32_t          valid = 0;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	source = sh->source;
-	active_sinks = sh->active_sinks;
-
-	/*
-	 * 2 calls per sink - setattr, setxattr
-	 */
-	if (xattr)
-		call_count = active_sinks * 2;
-	else
-		call_count = active_sinks;
-
-	local->call_count = call_count;
-
-	stbuf.ia_atime = sh->buf[source].ia_atime;
-	stbuf.ia_atime_nsec = sh->buf[source].ia_atime_nsec;
-	stbuf.ia_mtime = sh->buf[source].ia_mtime;
-	stbuf.ia_mtime_nsec = sh->buf[source].ia_mtime_nsec;
-
-        stbuf.ia_uid = sh->buf[source].ia_uid;
-        stbuf.ia_gid = sh->buf[source].ia_gid;
-
-        stbuf.ia_type = sh->buf[source].ia_type;
-        stbuf.ia_prot = sh->buf[source].ia_prot;
-        
-        valid = GF_SET_ATTR_MODE  | 
-                GF_SET_ATTR_UID   | GF_SET_ATTR_GID |
-                GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (call_count == 0) {
-			break;
-		}
-		if (sh->sources[i] || !local->child_up[i])
-			continue;
-
-		gf_log (this->name, GF_LOG_DEBUG,
-			"self-healing metadata of %s from %s to %s",
-			local->loc.path, priv->children[source]->name,
-			priv->children[i]->name);
-
-		STACK_WIND_COOKIE (frame, afr_sh_metadata_setattr_cbk,
-				   (void *) (long) i,
-				   priv->children[i],
-				   priv->children[i]->fops->setattr,
-				   &local->loc, &stbuf, valid);
-
-		call_count--;
-
-		if (!xattr)
-			continue;
-
-		STACK_WIND_COOKIE (frame, afr_sh_metadata_xattr_cbk,
-				   (void *) (long) i,
-				   priv->children[i],
-				   priv->children[i]->fops->setxattr,
-				   &local->loc, xattr, 0);
-		call_count--;
-	}
-
-	return 0;
+out:
+    if (xattr)
+        dict_unref(xattr);
+    return ret;
 }
 
-
-int
-afr_sh_metadata_getxattr_cbk (call_frame_t *frame, void *cookie,
-			      xlator_t *this,
-			      int32_t op_ret, int32_t op_errno, dict_t *xattr)
+/*
+ * Look for mismatching uid/gid or mode or user xattrs even if
+ * AFR xattrs don't say so, and pick one arbitrarily as winner. */
+
+static int
+__afr_selfheal_metadata_finalize_source(call_frame_t *frame, xlator_t *this,
+                                        inode_t *inode, unsigned char *sources,
+                                        unsigned char *sinks,
+                                        unsigned char *healed_sinks,
+                                        unsigned char *undid_pending,
+                                        unsigned char *locked_on,
+                                        struct afr_reply *replies)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              source = 0;
-
-        int i;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	source = sh->source;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"getxattr of %s failed on subvolume %s (%s). proceeding without xattr",
-			local->loc.path, priv->children[source]->name,
-			strerror (op_errno));
-
-		afr_sh_metadata_sync (frame, this, NULL);
-	} else {
-                for (i = 0; i < priv->child_count; i++) {
-                        dict_del (xattr, priv->pending_key[i]);
-                }
-
-		afr_sh_metadata_sync (frame, this, xattr);
-	}
+    int i = 0;
+    afr_private_t *priv = NULL;
+    struct iatt srcstat = {
+        0,
+    };
+    int source = -1;
+    int sources_count = 0;
+    int ret = 0;
+
+    priv = this->private;
+
+    sources_count = AFR_COUNT(sources, priv->child_count);
+
+    if ((AFR_CMP(locked_on, healed_sinks, priv->child_count) == 0) ||
+        !sources_count) {
+        source = afr_mark_split_brain_source_sinks(
+            frame, this, inode, sources, sinks, healed_sinks, locked_on,
+            replies, AFR_METADATA_TRANSACTION);
+        if (source >= 0) {
+            _afr_fav_child_reset_sink_xattrs(
+                frame, this, inode, source, healed_sinks, undid_pending,
+                AFR_METADATA_TRANSACTION, locked_on, replies);
+            goto out;
+        }
 
-	return 0;
-}
+        /* If this is a directory mtime/ctime only split brain
+           use the most recent */
+        source = afr_dirtime_splitbrain_source(frame, this, replies, locked_on);
+        if (source != -1) {
+            gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_SPLIT_BRAIN,
+                   "clear time "
+                   "split brain on %s",
+                   uuid_utoa(replies[source].poststat.ia_gfid));
+            sources[source] = 1;
+            healed_sinks[source] = 0;
+            goto out;
+        }
 
+        if (!priv->metadata_splitbrain_forced_heal) {
+            gf_event(EVENT_AFR_SPLIT_BRAIN,
+                     "client-pid=%d;"
+                     "subvol=%s;"
+                     "type=metadata;file=%s",
+                     this->ctx->cmd_args.client_pid, this->name,
+                     uuid_utoa(inode->gfid));
+            return -EIO;
+        }
 
-int
-afr_sh_metadata_sync_prepare (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              active_sinks = 0;
-	int              source = 0;
-	int              i = 0;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	source = sh->source;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (sh->sources[i] == 0 && local->child_up[i] == 1) {
-			active_sinks++;
-			sh->success[i] = 1;
-		}
-	}
-	sh->success[source] = 1;
-
-	if (active_sinks == 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no active sinks for performing self-heal on file %s",
-			local->loc.path);
-		afr_sh_metadata_finish (frame, this);
-		return 0;
-	}
-	sh->active_sinks = active_sinks;
-
-	gf_log (this->name, GF_LOG_TRACE,
-		"syncing metadata of %s from subvolume %s to %d active sinks",
-		local->loc.path, priv->children[source]->name, active_sinks);
-
-	STACK_WIND (frame, afr_sh_metadata_getxattr_cbk,
-		    priv->children[source],
-		    priv->children[source]->fops->getxattr,
-		    &local->loc, NULL);
-
-	return 0;
+        /* Metadata split brain, select one subvol
+           arbitrarily */
+        for (i = 0; i < priv->child_count; i++) {
+            if (locked_on[i] && healed_sinks[i]) {
+                sources[i] = 1;
+                healed_sinks[i] = 0;
+                break;
+            }
+        }
+    }
+
+    /* No split brain at this point. If we were called from
+     * afr_heal_splitbrain_file(), abort.*/
+    if (afr_dict_contains_heal_op(frame))
+        return -EIO;
+
+    source = afr_choose_source_by_policy(priv, sources,
+                                         AFR_METADATA_TRANSACTION);
+    srcstat = replies[source].poststat;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i] || i == source)
+            continue;
+        if (!IA_EQUAL(srcstat, replies[i].poststat, type) ||
+            !IA_EQUAL(srcstat, replies[i].poststat, uid) ||
+            !IA_EQUAL(srcstat, replies[i].poststat, gid) ||
+            !IA_EQUAL(srcstat, replies[i].poststat, prot)) {
+            gf_msg_debug(this->name, 0,
+                         "%s: iatt mismatch "
+                         "for source(%d) vs (%d)",
+                         uuid_utoa(replies[source].poststat.ia_gfid), source,
+                         i);
+            sources[i] = 0;
+            healed_sinks[i] = 1;
+        }
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i] || i == source)
+            continue;
+        if (!afr_xattrs_are_equal(replies[source].xdata, replies[i].xdata)) {
+            gf_msg_debug(this->name, 0,
+                         "%s: xattr mismatch "
+                         "for source(%d) vs (%d)",
+                         uuid_utoa(replies[source].poststat.ia_gfid), source,
+                         i);
+            sources[i] = 0;
+            healed_sinks[i] = 1;
+        }
+    }
+    if ((sources_count == priv->child_count) && (source > -1) &&
+        (AFR_COUNT(healed_sinks, priv->child_count) != 0)) {
+        ret = __afr_selfheal_metadata_mark_pending_xattrs(frame, this, inode,
+                                                          replies, sources);
+        if (ret < 0)
+            return ret;
+    }
+out:
+    afr_mark_active_sinks(this, sources, locked_on, healed_sinks);
+    return source;
 }
 
-
 int
-afr_sh_metadata_fix (call_frame_t *frame, xlator_t *this)
+__afr_selfheal_metadata_prepare(call_frame_t *frame, xlator_t *this,
+                                inode_t *inode, unsigned char *locked_on,
+                                unsigned char *sources, unsigned char *sinks,
+                                unsigned char *healed_sinks,
+                                unsigned char *undid_pending,
+                                struct afr_reply *replies, unsigned char *pflag)
 {
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              nsources = 0;
-	int              source = 0;
-	int              i = 0;
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	afr_sh_build_pending_matrix (priv, sh->pending_matrix, sh->xattr, 
-				     priv->child_count, 
-                                     AFR_METADATA_TRANSACTION);
-
-	afr_sh_print_pending_matrix (sh->pending_matrix, this);
-
-	nsources = afr_sh_mark_sources (sh, priv->child_count,
-                                        AFR_SELF_HEAL_METADATA);
-
-	afr_sh_supress_errenous_children (sh->sources, sh->child_errno,
-					  priv->child_count);
-
-        if (nsources == 0) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "No self-heal needed for %s",
-                        local->loc.path);
-
-                afr_sh_metadata_finish (frame, this);
-                return 0;
+    int ret = -1;
+    int source = -1;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    uint64_t *witness = NULL;
+
+    priv = this->private;
+
+    ret = afr_selfheal_unlocked_discover(frame, inode, inode->gfid, replies);
+    if (ret)
+        return ret;
+
+    witness = alloca0(sizeof(*witness) * priv->child_count);
+    ret = afr_selfheal_find_direction(frame, this, replies,
+                                      AFR_METADATA_TRANSACTION, locked_on,
+                                      sources, sinks, witness, pflag);
+    if (ret)
+        return ret;
+
+    /* Initialize the healed_sinks[] array optimistically to
+       the intersection of to-be-healed (i.e sinks[]) and
+       the list of servers which are up (i.e locked_on[]).
+
+       As we encounter failures in the healing process, we
+       will unmark the respective servers in the healed_sinks[]
+       array.
+    */
+    AFR_INTERSECT(healed_sinks, sinks, locked_on, priv->child_count);
+
+    /* If any source has witness, pick first
+     * witness source and make everybody else sinks */
+    for (i = 0; i < priv->child_count; i++) {
+        if (sources[i] && witness[i]) {
+            source = i;
+            break;
         }
-
-	if ((nsources == -1)
-	    && (priv->favorite_child != -1)
-	    && (sh->child_errno[priv->favorite_child] == 0)) {
-
-		gf_log (this->name, GF_LOG_WARNING,
-			"Picking favorite child %s as authentic source to resolve conflicting metadata of %s",
-			priv->children[priv->favorite_child]->name,
-			local->loc.path);
-
-		sh->sources[priv->favorite_child] = 1;
-
-		nsources = afr_sh_source_count (sh->sources,
-						priv->child_count);
-	}
-
-	if (nsources == -1) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Unable to self-heal permissions/ownership of '%s' "
-                        "(possible split-brain). Please fix the file on "
-                        "all backend volumes", local->loc.path);
-
-		local->govinda_gOvinda = 1;
-
-		afr_sh_metadata_finish (frame, this);
-		return 0;
-	}
-
-	source = afr_sh_select_source (sh->sources, priv->child_count);
-
-        if (source == -1) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "No active sources found.");
-
-                afr_sh_metadata_finish (frame, this);
-                return 0;
+    }
+
+    if (source != -1) {
+        for (i = 0; i < priv->child_count; i++) {
+            if (i != source && sources[i]) {
+                sources[i] = 0;
+                healed_sinks[i] = 1;
+            }
         }
+    }
 
-	sh->source = source;
-
-	/* detect changes not visible through pending flags -- JIC */
-	for (i = 0; i < priv->child_count; i++) {
-		if (i == source || sh->child_errno[i])
-			continue;
+    source = __afr_selfheal_metadata_finalize_source(
+        frame, this, inode, sources, sinks, healed_sinks, undid_pending,
+        locked_on, replies);
 
-		if (PERMISSION_DIFFERS (&sh->buf[i], &sh->buf[source]))
-			sh->sources[i] = 0;
+    if (source < 0)
+        return -EIO;
 
-		if (OWNERSHIP_DIFFERS (&sh->buf[i], &sh->buf[source]))
-			sh->sources[i] = 0;
-	}
-
-	afr_sh_metadata_sync_prepare (frame, this);
-
-	return 0;
-}
-
-
-int
-afr_sh_metadata_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			    int32_t op_ret, int32_t op_errno,
-                            inode_t *inode, struct iatt *buf, dict_t *xattr,
-                            struct iatt *postparent)
-{
-	afr_local_t     *local = NULL;
-	afr_self_heal_t *sh = NULL;
-	afr_private_t   *priv = NULL;
-	int              call_count = 0;
-	int              child_index = 0;
-
-
-	local = frame->local;
-	sh = &local->self_heal;
-	priv = this->private;
-
-	child_index = (long) cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == 0) {
-			gf_log (this->name, GF_LOG_TRACE,
-				"path %s on subvolume %s is of mode 0%o",
-				local->loc.path,
-				priv->children[child_index]->name,
-				buf->ia_type);
-
-			sh->buf[child_index] = *buf;
-			if (xattr)
-				sh->xattr[child_index] = dict_ref (xattr);
-		} else {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"path %s on subvolume %s => -1 (%s)",
-				local->loc.path,
-				priv->children[child_index]->name,
-				strerror (op_errno));
-
-			sh->child_errno[child_index] = op_errno;
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0)
-		afr_sh_metadata_fix (frame, this);
-
-	return 0;
+    return source;
 }
 
-
 int
-afr_sh_metadata_lookup (call_frame_t *frame, xlator_t *this)
+afr_selfheal_metadata(call_frame_t *frame, xlator_t *this, inode_t *inode)
 {
-	afr_local_t     *local = NULL;
-	afr_private_t   *priv = NULL;
-	int              i = 0;
-	int              call_count = 0;
-	dict_t          *xattr_req = NULL;
-	int              ret = 0;
-
-	local = frame->local;
-	priv = this->private;
-
-	call_count = afr_up_children_count (priv->child_count,
-                                            local->child_up);
-	local->call_count = call_count;
-
-	xattr_req = dict_new();
-
-	if (xattr_req) {
-                for (i = 0; i < priv->child_count; i++) {
-                        ret = dict_set_uint64 (xattr_req,
-                                               priv->pending_key[i],
-                                               3 * sizeof(int32_t));
-                        if (ret < 0)
-                                gf_log (this->name, GF_LOG_WARNING,
-                                        "Unable to set dict value.");
-                }
+    afr_private_t *priv = NULL;
+    int ret = -1;
+    unsigned char *sources = NULL;
+    unsigned char *sinks = NULL;
+    unsigned char *data_lock = NULL;
+    unsigned char *healed_sinks = NULL;
+    unsigned char *undid_pending = NULL;
+    struct afr_reply *locked_replies = NULL;
+    gf_boolean_t did_sh = _gf_true;
+    int source = -1;
+
+    priv = this->private;
+
+    sources = alloca0(priv->child_count);
+    sinks = alloca0(priv->child_count);
+    healed_sinks = alloca0(priv->child_count);
+    undid_pending = alloca0(priv->child_count);
+    data_lock = alloca0(priv->child_count);
+
+    locked_replies = alloca0(sizeof(*locked_replies) * priv->child_count);
+
+    ret = afr_selfheal_inodelk(frame, this, inode, this->name, LLONG_MAX - 1, 0,
+                               data_lock);
+    {
+        if (ret < priv->child_count) {
+            ret = -ENOTCONN;
+            goto unlock;
         }
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			gf_log (this->name, GF_LOG_TRACE,
-				"looking up %s on %s",
-				local->loc.path, priv->children[i]->name);
-
-			STACK_WIND_COOKIE (frame, afr_sh_metadata_lookup_cbk,
-					   (void *) (long) i,
-					   priv->children[i],
-					   priv->children[i]->fops->lookup,
-					   &local->loc, xattr_req);
-			if (!--call_count)
-				break;
-		}
-	}
-	
-	if (xattr_req)
-		dict_unref (xattr_req);
-
-	return 0;
-}
-
-int
-afr_sh_post_nonblocking_inodelk_cbk (call_frame_t *frame, xlator_t *this)
-{
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
+        ret = __afr_selfheal_metadata_prepare(
+            frame, this, inode, data_lock, sources, sinks, healed_sinks,
+            undid_pending, locked_replies, NULL);
+        if (ret < 0)
+            goto unlock;
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+        source = ret;
 
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Non Blocking inodelks failed.");
-                afr_sh_metadata_done (frame, this);
-        } else {
-
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Non Blocking inodelks done. Proceeding to FOP");
-                afr_sh_metadata_lookup (frame, this);
+        if (AFR_COUNT(healed_sinks, priv->child_count) == 0) {
+            did_sh = _gf_false;
+            goto unlock;
         }
 
-        return 0;
+        ret = __afr_selfheal_metadata_do(frame, this, inode, source,
+                                         healed_sinks, locked_replies);
+        if (ret)
+            goto unlock;
+
+        afr_selfheal_restore_time(frame, this, inode, source, healed_sinks,
+                                  locked_replies);
+
+        ret = afr_selfheal_undo_pending(
+            frame, this, inode, sources, sinks, healed_sinks, undid_pending,
+            AFR_METADATA_TRANSACTION, locked_replies, data_lock);
+    }
+unlock:
+    afr_selfheal_uninodelk(frame, this, inode, this->name, LLONG_MAX - 1, 0,
+                           data_lock);
+
+    if (did_sh)
+        afr_log_selfheal(inode->gfid, this, ret, "metadata", source, sources,
+                         healed_sinks);
+    else
+        ret = 1;
+
+    if (locked_replies)
+        afr_replies_wipe(locked_replies, priv->child_count);
+    return ret;
 }
 
 int
-afr_sh_metadata_lock (call_frame_t *frame, xlator_t *this)
+afr_selfheal_metadata_by_stbuf(xlator_t *this, struct iatt *stbuf)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
-
-        int_lock->transaction_lk_type = AFR_SELFHEAL_LK;
-        int_lock->selfheal_lk_type    = AFR_METADATA_SELF_HEAL_LK;
-
-        afr_set_lock_number (frame, this);
-
-        int_lock->lk_flock.l_start = 0;
-        int_lock->lk_flock.l_len   = 0;
-        int_lock->lk_flock.l_type  = F_WRLCK;
-        int_lock->lock_cbk         = afr_sh_post_nonblocking_inodelk_cbk;
-
-        afr_nonblocking_inodelk (frame, this);
-
-	return 0;
-}
-
-
-int
-afr_self_heal_metadata (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t   *local = NULL;
-	afr_private_t *priv = this->private;
-
-
-	local = frame->local;
-
-	if (local->self_heal.need_metadata_self_heal && priv->metadata_self_heal) {
-		afr_sh_metadata_lock (frame, this);
-	} else {
-		afr_sh_metadata_done (frame, this);
-	}
-
-	return 0;
+    inode_t *inode = NULL;
+    inode_t *link_inode = NULL;
+    call_frame_t *frame = NULL;
+    int ret = 0;
+
+    if (gf_uuid_is_null(stbuf->ia_gfid)) {
+        ret = -EINVAL;
+        goto out;
+    }
+
+    inode = inode_new(this->itable);
+    if (!inode) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    link_inode = inode_link(inode, NULL, NULL, stbuf);
+    if (!link_inode) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    frame = afr_frame_create(this, &ret);
+    if (!frame) {
+        ret = -ret;
+        goto out;
+    }
+
+    ret = afr_selfheal_metadata(frame, this, link_inode);
+out:
+    if (inode)
+        inode_unref(inode);
+    if (link_inode)
+        inode_unref(link_inode);
+    if (frame)
+        AFR_STACK_DESTROY(frame);
+    return ret;
 }
-
diff --git a/xlators/cluster/afr/src/afr-self-heal-name.c b/xlators/cluster/afr/src/afr-self-heal-name.c
new file mode 100644
index 00000000000..834aac86d48
--- /dev/null
+++ b/xlators/cluster/afr/src/afr-self-heal-name.c
@@ -0,0 +1,616 @@
+/*
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/events.h>
+#include "afr.h"
+#include "afr-self-heal.h"
+#include "afr-messages.h"
+
+int
+__afr_selfheal_assign_gfid(xlator_t *this, inode_t *parent, uuid_t pargfid,
+                           const char *bname, inode_t *inode,
+                           struct afr_reply *replies, void *gfid,
+                           unsigned char *locked_on, int source,
+                           unsigned char *sources, gf_boolean_t is_gfid_absent,
+                           int *gfid_idx)
+{
+    int ret = 0;
+    int up_count = 0;
+    int locked_count = 0;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+
+    gf_uuid_copy(parent->gfid, pargfid);
+
+    if (is_gfid_absent) {
+        /* Ensure all children of AFR are up before performing gfid heal, to
+         * guard against the possibility of gfid split brain. */
+
+        up_count = AFR_COUNT(priv->child_up, priv->child_count);
+        if (up_count != priv->child_count) {
+            ret = -EIO;
+            goto out;
+        }
+
+        locked_count = AFR_COUNT(locked_on, priv->child_count);
+        if (locked_count != priv->child_count) {
+            ret = -EIO;
+            goto out;
+        }
+    }
+
+    ret = afr_lookup_and_heal_gfid(this, parent, bname, inode, replies, source,
+                                   sources, gfid, gfid_idx);
+
+out:
+    return ret;
+}
+
+int
+__afr_selfheal_name_impunge(call_frame_t *frame, xlator_t *this,
+                            inode_t *parent, uuid_t pargfid, const char *bname,
+                            inode_t *inode, struct afr_reply *replies,
+                            int gfid_idx)
+{
+    int i = 0;
+    afr_private_t *priv = NULL;
+    int ret = 0;
+    unsigned char *sources = NULL;
+
+    priv = this->private;
+
+    sources = alloca0(priv->child_count);
+
+    gf_uuid_copy(parent->gfid, pargfid);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret != 0)
+            continue;
+
+        if (gf_uuid_compare(replies[i].poststat.ia_gfid,
+                            replies[gfid_idx].poststat.ia_gfid) == 0) {
+            sources[i] = 1;
+            continue;
+        }
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (sources[i])
+            continue;
+
+        ret |= afr_selfheal_recreate_entry(frame, i, gfid_idx, sources, parent,
+                                           bname, inode, replies);
+    }
+
+    return ret;
+}
+
+int
+__afr_selfheal_name_expunge(xlator_t *this, inode_t *parent, uuid_t pargfid,
+                            const char *bname, inode_t *inode,
+                            struct afr_reply *replies)
+{
+    int i = 0;
+    afr_private_t *priv = NULL;
+    int ret = 0;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid)
+            continue;
+
+        if (replies[i].op_ret)
+            continue;
+
+        ret |= afr_selfheal_entry_delete(this, parent, bname, inode, i,
+                                         replies);
+    }
+
+    return ret;
+}
+
+static gf_boolean_t
+afr_selfheal_name_need_heal_check(xlator_t *this, struct afr_reply *replies)
+{
+    int i = 0;
+    int first_idx = -1;
+    gf_boolean_t need_heal = _gf_false;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid)
+            continue;
+
+        if ((replies[i].op_ret == -1) && (replies[i].op_errno == ENODATA))
+            need_heal = _gf_true;
+
+        if (first_idx == -1) {
+            first_idx = i;
+            continue;
+        }
+
+        if (replies[i].op_ret != replies[first_idx].op_ret)
+            need_heal = _gf_true;
+
+        if (gf_uuid_compare(replies[i].poststat.ia_gfid,
+                            replies[first_idx].poststat.ia_gfid))
+            need_heal = _gf_true;
+
+        if ((replies[i].op_ret == 0) &&
+            (gf_uuid_is_null(replies[i].poststat.ia_gfid)))
+            need_heal = _gf_true;
+    }
+
+    return need_heal;
+}
+
+static int
+afr_selfheal_name_type_mismatch_check(xlator_t *this, struct afr_reply *replies,
+                                      int source, unsigned char *sources,
+                                      uuid_t pargfid, const char *bname)
+{
+    int i = 0;
+    int type_idx = -1;
+    ia_type_t inode_type = IA_INVAL;
+    ia_type_t inode_type1 = IA_INVAL;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret != 0)
+            continue;
+
+        if (replies[i].poststat.ia_type == IA_INVAL)
+            continue;
+
+        if (inode_type == IA_INVAL) {
+            inode_type = replies[i].poststat.ia_type;
+            type_idx = i;
+            continue;
+        }
+        inode_type1 = replies[i].poststat.ia_type;
+        if (sources[i] || source == -1) {
+            if ((sources[type_idx] || source == -1) &&
+                (inode_type != inode_type1)) {
+                gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_SPLIT_BRAIN,
+                       "Type mismatch for <gfid:%s>/%s: "
+                       "%s on %s and %s on %s",
+                       uuid_utoa(pargfid), bname,
+                       gf_inode_type_to_str(inode_type1),
+                       priv->children[i]->name,
+                       gf_inode_type_to_str(inode_type),
+                       priv->children[type_idx]->name);
+                gf_event(EVENT_AFR_SPLIT_BRAIN,
+                         "client-pid=%d;"
+                         "subvol=%s;type=file;"
+                         "file=<gfid:%s>/%s;count=2;"
+                         "child-%d=%s;type-%d=%s;child-%d=%s;"
+                         "type-%d=%s",
+                         this->ctx->cmd_args.client_pid, this->name,
+                         uuid_utoa(pargfid), bname, i, priv->children[i]->name,
+                         i, gf_inode_type_to_str(inode_type1), type_idx,
+                         priv->children[type_idx]->name, type_idx,
+                         gf_inode_type_to_str(inode_type));
+                return -EIO;
+            }
+            inode_type = replies[i].poststat.ia_type;
+            type_idx = i;
+        }
+    }
+    return 0;
+}
+
+static int
+afr_selfheal_name_gfid_mismatch_check(xlator_t *this, struct afr_reply *replies,
+                                      int source, unsigned char *sources,
+                                      int *gfid_idx, uuid_t pargfid,
+                                      const char *bname, inode_t *inode,
+                                      unsigned char *locked_on, dict_t *xdata)
+{
+    int i = 0;
+    int gfid_idx_iter = -1;
+    int ret = -1;
+    void *gfid = NULL;
+    void *gfid1 = NULL;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid || replies[i].op_ret != 0)
+            continue;
+
+        if (gf_uuid_is_null(replies[i].poststat.ia_gfid))
+            continue;
+
+        if (!gfid) {
+            gfid = &replies[i].poststat.ia_gfid;
+            gfid_idx_iter = i;
+            continue;
+        }
+
+        gfid1 = &replies[i].poststat.ia_gfid;
+        if (sources[i] || source == -1) {
+            if ((sources[gfid_idx_iter] || source == -1) &&
+                gf_uuid_compare(gfid, gfid1)) {
+                ret = afr_gfid_split_brain_source(this, replies, inode, pargfid,
+                                                  bname, gfid_idx_iter, i,
+                                                  locked_on, gfid_idx, xdata);
+                if (!ret && *gfid_idx >= 0) {
+                    ret = dict_set_sizen_str_sizen(xdata, "gfid-heal-msg",
+                                                   "GFID split-brain resolved");
+                    if (ret)
+                        gf_msg(this->name, GF_LOG_ERROR, 0,
+                               AFR_MSG_DICT_SET_FAILED,
+                               "Error setting gfid-"
+                               "heal-msg dict");
+                }
+                return ret;
+            }
+            gfid = &replies[i].poststat.ia_gfid;
+            gfid_idx_iter = i;
+        }
+    }
+
+    *gfid_idx = gfid_idx_iter;
+    return 0;
+}
+
+static gf_boolean_t
+afr_selfheal_name_source_empty_check(xlator_t *this, struct afr_reply *replies,
+                                     unsigned char *sources, int source)
+{
+    int i = 0;
+    afr_private_t *priv = NULL;
+    gf_boolean_t source_is_empty = _gf_true;
+
+    priv = this->private;
+
+    if (source == -1) {
+        source_is_empty = _gf_false;
+        goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!sources[i])
+            continue;
+
+        if (replies[i].op_ret == -1 && replies[i].op_errno == ENOENT)
+            continue;
+
+        source_is_empty = _gf_false;
+        break;
+    }
+out:
+    return source_is_empty;
+}
+
+int
+__afr_selfheal_name_do(call_frame_t *frame, xlator_t *this, inode_t *parent,
+                       uuid_t pargfid, const char *bname, inode_t *inode,
+                       unsigned char *sources, unsigned char *sinks,
+                       unsigned char *healed_sinks, int source,
+                       unsigned char *locked_on, struct afr_reply *replies,
+                       void *gfid_req, dict_t *xdata)
+{
+    int gfid_idx = -1;
+    int ret = -1;
+    void *gfid = NULL;
+    gf_boolean_t source_is_empty = _gf_true;
+    gf_boolean_t need_heal = _gf_false;
+    gf_boolean_t is_gfid_absent = _gf_false;
+
+    need_heal = afr_selfheal_name_need_heal_check(this, replies);
+    if (!need_heal)
+        return 0;
+
+    source_is_empty = afr_selfheal_name_source_empty_check(this, replies,
+                                                           sources, source);
+    if (source_is_empty) {
+        ret = __afr_selfheal_name_expunge(this, parent, pargfid, bname, inode,
+                                          replies);
+        if (ret == -EIO)
+            ret = -1;
+        return ret;
+    }
+
+    ret = afr_selfheal_name_type_mismatch_check(this, replies, source, sources,
+                                                pargfid, bname);
+    if (ret)
+        return ret;
+
+    ret = afr_selfheal_name_gfid_mismatch_check(this, replies, source, sources,
+                                                &gfid_idx, pargfid, bname,
+                                                inode, locked_on, xdata);
+    if (ret)
+        return ret;
+
+    if (gfid_idx == -1) {
+        if (!gfid_req || gf_uuid_is_null(gfid_req))
+            return -1;
+        gfid = gfid_req;
+    } else {
+        gfid = &replies[gfid_idx].poststat.ia_gfid;
+        if (source == -1)
+            /* Either entry split-brain or dirty xattrs are present on parent.*/
+            source = gfid_idx;
+    }
+
+    is_gfid_absent = (gfid_idx == -1) ? _gf_true : _gf_false;
+    ret = __afr_selfheal_assign_gfid(this, parent, pargfid, bname, inode,
+                                     replies, gfid, locked_on, source, sources,
+                                     is_gfid_absent, &gfid_idx);
+    if (ret || (gfid_idx < 0))
+        return ret;
+
+    ret = __afr_selfheal_name_impunge(frame, this, parent, pargfid, bname,
+                                      inode, replies, gfid_idx);
+    if (ret == -EIO)
+        ret = -1;
+
+    return ret;
+}
+
+int
+__afr_selfheal_name_finalize_source(xlator_t *this, unsigned char *sources,
+                                    unsigned char *healed_sinks,
+                                    unsigned char *locked_on, uint64_t *witness)
+{
+    int i = 0;
+    afr_private_t *priv = NULL;
+    int source = -1;
+    int sources_count = 0;
+
+    priv = this->private;
+
+    sources_count = AFR_COUNT(sources, priv->child_count);
+
+    if ((AFR_CMP(locked_on, healed_sinks, priv->child_count) == 0) ||
+        !sources_count || afr_does_witness_exist(this, witness)) {
+        memset(sources, 0, sizeof(*sources) * priv->child_count);
+        afr_mark_active_sinks(this, sources, locked_on, healed_sinks);
+        return -1;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (sources[i]) {
+            source = i;
+            break;
+        }
+    }
+
+    return source;
+}
+
+int
+__afr_selfheal_name_prepare(call_frame_t *frame, xlator_t *this,
+                            inode_t *parent, uuid_t pargfid,
+                            unsigned char *locked_on, unsigned char *sources,
+                            unsigned char *sinks, unsigned char *healed_sinks,
+                            int *source_p)
+{
+    int ret = -1;
+    int source = -1;
+    afr_private_t *priv = NULL;
+    struct afr_reply *replies = NULL;
+    uint64_t *witness = NULL;
+
+    priv = this->private;
+
+    replies = alloca0(priv->child_count * sizeof(*replies));
+
+    ret = afr_selfheal_unlocked_discover(frame, parent, pargfid, replies);
+    if (ret)
+        goto out;
+
+    witness = alloca0(sizeof(*witness) * priv->child_count);
+    ret = afr_selfheal_find_direction(frame, this, replies,
+                                      AFR_ENTRY_TRANSACTION, locked_on, sources,
+                                      sinks, witness, NULL);
+    if (ret)
+        goto out;
+
+    /* Initialize the healed_sinks[] array optimistically to
+       the intersection of to-be-healed (i.e sinks[]) and
+       the list of servers which are up (i.e locked_on[]).
+
+       As we encounter failures in the healing process, we
+       will unmark the respective servers in the healed_sinks[]
+       array.
+    */
+    AFR_INTERSECT(healed_sinks, sinks, locked_on, priv->child_count);
+
+    source = __afr_selfheal_name_finalize_source(this, sources, healed_sinks,
+                                                 locked_on, witness);
+    if (source < 0) {
+        /* If source is < 0 (typically split-brain), we perform a
+           conservative merge of entries rather than erroring out */
+    }
+    *source_p = source;
+
+out:
+    if (replies)
+        afr_replies_wipe(replies, priv->child_count);
+
+    return ret;
+}
+
+int
+afr_selfheal_name_do(call_frame_t *frame, xlator_t *this, inode_t *parent,
+                     uuid_t pargfid, const char *bname, void *gfid_req,
+                     dict_t *xdata)
+{
+    afr_private_t *priv = NULL;
+    unsigned char *sources = NULL;
+    unsigned char *sinks = NULL;
+    unsigned char *healed_sinks = NULL;
+    unsigned char *locked_on = NULL;
+    int source = -1;
+    struct afr_reply *replies = NULL;
+    int ret = -1;
+    inode_t *inode = NULL;
+    dict_t *xattr = NULL;
+
+    xattr = dict_new();
+    if (!xattr)
+        return -ENOMEM;
+
+    ret = dict_set_int32_sizen(xattr, GF_GFIDLESS_LOOKUP, 1);
+    if (ret) {
+        dict_unref(xattr);
+        return -1;
+    }
+
+    priv = this->private;
+
+    locked_on = alloca0(priv->child_count);
+    sources = alloca0(priv->child_count);
+    sinks = alloca0(priv->child_count);
+    healed_sinks = alloca0(priv->child_count);
+
+    replies = alloca0(priv->child_count * sizeof(*replies));
+
+    ret = afr_selfheal_entrylk(frame, this, parent, this->name, bname,
+                               locked_on);
+    {
+        if (ret < priv->child_count) {
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+
+        ret = __afr_selfheal_name_prepare(frame, this, parent, pargfid,
+                                          locked_on, sources, sinks,
+                                          healed_sinks, &source);
+        if (ret)
+            goto unlock;
+
+        inode = afr_selfheal_unlocked_lookup_on(frame, parent, bname, replies,
+                                                locked_on, xattr);
+        if (!inode) {
+            ret = -ENOMEM;
+            goto unlock;
+        }
+
+        ret = __afr_selfheal_name_do(frame, this, parent, pargfid, bname, inode,
+                                     sources, sinks, healed_sinks, source,
+                                     locked_on, replies, gfid_req, xdata);
+    }
+unlock:
+    afr_selfheal_unentrylk(frame, this, parent, this->name, bname, locked_on,
+                           NULL);
+    if (inode)
+        inode_unref(inode);
+
+    if (replies)
+        afr_replies_wipe(replies, priv->child_count);
+    if (xattr)
+        dict_unref(xattr);
+
+    return ret;
+}
+
+int
+afr_selfheal_name_unlocked_inspect(call_frame_t *frame, xlator_t *this,
+                                   inode_t *parent, uuid_t pargfid,
+                                   const char *bname, gf_boolean_t *need_heal)
+{
+    afr_private_t *priv = NULL;
+    int i = 0;
+    struct afr_reply *replies = NULL;
+    inode_t *inode = NULL;
+    int first_idx = -1;
+    afr_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    replies = alloca0(sizeof(*replies) * priv->child_count);
+
+    inode = afr_selfheal_unlocked_lookup_on(frame, parent, bname, replies,
+                                            local->child_up, NULL);
+    if (!inode)
+        return -ENOMEM;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!replies[i].valid)
+            continue;
+
+        if ((replies[i].op_ret == -1) && (replies[i].op_errno == ENODATA)) {
+            *need_heal = _gf_true;
+            break;
+        }
+
+        if (first_idx == -1) {
+            first_idx = i;
+            continue;
+        }
+
+        if (replies[i].op_ret != replies[first_idx].op_ret) {
+            *need_heal = _gf_true;
+            break;
+        }
+
+        if (gf_uuid_compare(replies[i].poststat.ia_gfid,
+                            replies[first_idx].poststat.ia_gfid)) {
+            *need_heal = _gf_true;
+            break;
+        }
+    }
+
+    if (inode)
+        inode_unref(inode);
+    if (replies)
+        afr_replies_wipe(replies, priv->child_count);
+    return 0;
+}
+
+int
+afr_selfheal_name(xlator_t *this, uuid_t pargfid, const char *bname,
+                  void *gfid_req, dict_t *xdata)
+{
+    inode_t *parent = NULL;
+    call_frame_t *frame = NULL;
+    int ret = -1;
+    gf_boolean_t need_heal = _gf_false;
+
+    parent = afr_inode_find(this, pargfid);
+    if (!parent)
+        goto out;
+
+    frame = afr_frame_create(this, NULL);
+    if (!frame)
+        goto out;
+
+    ret = afr_selfheal_name_unlocked_inspect(frame, this, parent, pargfid,
+                                             bname, &need_heal);
+    if (ret)
+        goto out;
+
+    if (need_heal) {
+        ret = afr_selfheal_name_do(frame, this, parent, pargfid, bname,
+                                   gfid_req, xdata);
+        if (ret)
+            goto out;
+    }
+
+    ret = 0;
+out:
+    if (parent)
+        inode_unref(parent);
+    if (frame)
+        AFR_STACK_DESTROY(frame);
+
+    return ret;
+}
diff --git a/xlators/cluster/afr/src/afr-self-heal.h b/xlators/cluster/afr/src/afr-self-heal.h
index 63354926b9f..48e6dbcfb18 100644
--- a/xlators/cluster/afr/src/afr-self-heal.h
+++ b/xlators/cluster/afr/src/afr-self-heal.h
@@ -1,54 +1,377 @@
 /*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
-#ifndef __AFR_SELF_HEAL_H__
-#define __AFR_SELF_HEAL_H__
+#ifndef _AFR_SELFHEAL_H
+#define _AFR_SELFHEAL_H
+
+/* Perform fop on all UP subvolumes and wait for all callbacks to return */
+
+#define AFR_ONALL(frame, rfn, fop, args...)                                    \
+    do {                                                                       \
+        afr_local_t *__local = frame->local;                                   \
+        afr_private_t *__priv = frame->this->private;                          \
+        int __i = 0, __count = 0;                                              \
+        unsigned char *__child_up = alloca(__priv->child_count);               \
+                                                                               \
+        memcpy(__child_up, __priv->child_up,                                   \
+               sizeof(*__child_up) * __priv->child_count);                     \
+        __count = AFR_COUNT(__child_up, __priv->child_count);                  \
+                                                                               \
+        __local->barrier.waitfor = __count;                                    \
+        afr_local_replies_wipe(__local, __priv);                               \
+                                                                               \
+        for (__i = 0; __i < __priv->child_count; __i++) {                      \
+            if (!__child_up[__i])                                              \
+                continue;                                                      \
+            STACK_WIND_COOKIE(frame, rfn, (void *)(long)__i,                   \
+                              __priv->children[__i],                           \
+                              __priv->children[__i]->fops->fop, args);         \
+        }                                                                      \
+        syncbarrier_wait(&__local->barrier, __count);                          \
+    } while (0)
+
+/* Perform fop on all subvolumes represented by list[] array and wait
+   for all callbacks to return */
+
+#define AFR_ONLIST(list, frame, rfn, fop, args...)                             \
+    do {                                                                       \
+        afr_local_t *__local = frame->local;                                   \
+        afr_private_t *__priv = frame->this->private;                          \
+        int __i = 0;                                                           \
+        int __count = 0;                                                       \
+        unsigned char *__list = alloca(__priv->child_count);                   \
+                                                                               \
+        memcpy(__list, list, sizeof(*__list) * __priv->child_count);           \
+        __count = AFR_COUNT(__list, __priv->child_count);                      \
+        __local->barrier.waitfor = __count;                                    \
+        afr_local_replies_wipe(__local, __priv);                               \
+                                                                               \
+        for (__i = 0; __i < __priv->child_count; __i++) {                      \
+            if (!__list[__i])                                                  \
+                continue;                                                      \
+            STACK_WIND_COOKIE(frame, rfn, (void *)(long)__i,                   \
+                              __priv->children[__i],                           \
+                              __priv->children[__i]->fops->fop, args);         \
+        }                                                                      \
+        syncbarrier_wait(&__local->barrier, __count);                          \
+    } while (0)
+
+#define AFR_SEQ(frame, rfn, fop, args...)                                      \
+    do {                                                                       \
+        afr_local_t *__local = frame->local;                                   \
+        afr_private_t *__priv = frame->this->private;                          \
+        int __i = 0;                                                           \
+                                                                               \
+        afr_local_replies_wipe(__local, __priv);                               \
+                                                                               \
+        for (__i = 0; __i < __priv->child_count; __i++) {                      \
+            if (!__priv->child_up[__i])                                        \
+                continue;                                                      \
+            STACK_WIND_COOKIE(frame, rfn, (void *)(long)__i,                   \
+                              __priv->children[__i],                           \
+                              __priv->children[__i]->fops->fop, args);         \
+            syncbarrier_wait(&__local->barrier, 1);                            \
+        }                                                                      \
+    } while (0)
+
+#define ALLOC_MATRIX(n, type)                                                  \
+    ({                                                                         \
+        int __i;                                                               \
+        type **__ptr = alloca(n * sizeof(type *));                             \
+                                                                               \
+        for (__i = 0; __i < n; __i++)                                          \
+            __ptr[__i] = alloca0(n * sizeof(type));                            \
+        __ptr;                                                                 \
+    })
+
+#define IA_EQUAL(f, s, field)                                                  \
+    (memcmp(&(f.ia_##field), &(s.ia_##field), sizeof(s.ia_##field)) == 0)
+
+#define SBRAIN_HEAL_NO_GO_MSG                                                  \
+    "Failed to obtain replies from all bricks of "                             \
+    "the replica (are they up?). Cannot resolve split-brain."
+#define SFILE_NOT_IN_SPLIT_BRAIN "File not in split-brain"
+#define SNO_BIGGER_FILE "No bigger file"
+#define SNO_DIFF_IN_MTIME "No difference in mtime"
+#define SUSE_SOURCE_BRICK_TO_HEAL                                              \
+    "Use source-brick option to heal metadata"                                 \
+    " split-brain"
+#define SINVALID_BRICK_NAME "Invalid brick name"
+#define SBRICK_IS_NOT_UP "Brick is not up"
+#define SBRICK_NOT_CONNECTED "Brick is not connected"
+#define SLESS_THAN2_BRICKS_in_REP "< 2 bricks in replica are up"
+#define SBRICK_IS_REMOTE "Brick is remote"
+#define SSTARTED_SELF_HEAL "Started self-heal"
+#define SOP_NOT_SUPPORTED "Operation Not Supported"
+#define SFILE_NOT_UNDER_DATA                                                   \
+    "The file is not under data or metadata "                                  \
+    "split-brain"
+#define SFILE_NOT_IN_SPLIT_BRAIN "File not in split-brain"
+#define SALL_BRICKS_UP_TO_RESOLVE                                              \
+    "All the bricks should be up to resolve the"                               \
+    " gfid split brain"
+#define SERROR_GETTING_SRC_BRICK "Error getting the source brick"
+int
+afr_selfheal(xlator_t *this, uuid_t gfid);
+
+gf_boolean_t
+afr_throttled_selfheal(call_frame_t *frame, xlator_t *this);
+
+int
+afr_selfheal_name(xlator_t *this, uuid_t gfid, const char *name, void *gfid_req,
+                  dict_t *xdata);
+
+int
+afr_selfheal_data(call_frame_t *frame, xlator_t *this, fd_t *fd);
+
+int
+afr_selfheal_metadata(call_frame_t *frame, xlator_t *this, inode_t *inode);
+
+int
+afr_selfheal_entry(call_frame_t *frame, xlator_t *this, inode_t *inode);
+
+int
+afr_lookup_and_heal_gfid(xlator_t *this, inode_t *parent, const char *name,
+                         inode_t *inode, struct afr_reply *replies, int source,
+                         unsigned char *sources, void *gfid, int *gfid_idx);
+
+int
+afr_selfheal_inodelk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                     char *dom, off_t off, size_t size,
+                     unsigned char *locked_on);
+
+int
+afr_selfheal_tryinodelk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                        char *dom, off_t off, size_t size,
+                        unsigned char *locked_on);
+
+int
+afr_selfheal_tie_breaker_inodelk(call_frame_t *frame, xlator_t *this,
+                                 inode_t *inode, char *dom, off_t off,
+                                 size_t size, unsigned char *locked_on);
+
+int
+afr_selfheal_uninodelk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                       char *dom, off_t off, size_t size,
+                       const unsigned char *locked_on);
+
+int
+afr_selfheal_entrylk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                     char *dom, const char *name, unsigned char *locked_on);
+
+int
+afr_selfheal_tryentrylk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                        char *dom, const char *name, unsigned char *locked_on);
+
+int
+afr_selfheal_tie_breaker_entrylk(call_frame_t *frame, xlator_t *this,
+                                 inode_t *inode, char *dom, const char *name,
+                                 unsigned char *locked_on);
+
+int
+afr_selfheal_unentrylk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                       char *dom, const char *name, unsigned char *locked_on,
+                       dict_t *xdata);
+
+int
+afr_selfheal_unlocked_discover(call_frame_t *frame, inode_t *inode, uuid_t gfid,
+                               struct afr_reply *replies);
+
+int
+afr_selfheal_unlocked_discover_on(call_frame_t *frame, inode_t *inode,
+                                  uuid_t gfid, struct afr_reply *replies,
+                                  unsigned char *discover_on, dict_t *dict);
+inode_t *
+afr_selfheal_unlocked_lookup_on(call_frame_t *frame, inode_t *parent,
+                                const char *name, struct afr_reply *replies,
+                                unsigned char *lookup_on, dict_t *xattr);
+
+int
+afr_selfheal_find_direction(call_frame_t *frame, xlator_t *this,
+                            struct afr_reply *replies,
+                            afr_transaction_type type, unsigned char *locked_on,
+                            unsigned char *sources, unsigned char *sinks,
+                            uint64_t *witness, unsigned char *flag);
+int
+afr_selfheal_fill_matrix(xlator_t *this, int **matrix, int subvol, int idx,
+                         dict_t *xdata);
+
+int
+afr_selfheal_extract_xattr(xlator_t *this, struct afr_reply *replies,
+                           afr_transaction_type type, int *dirty, int **matrix);
+
+int
+afr_sh_generic_fop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, struct iatt *pre,
+                       struct iatt *post, dict_t *xdata);
+
+int
+afr_selfheal_restore_time(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                          int source, unsigned char *healed_sinks,
+                          struct afr_reply *replies);
+int
+afr_selfheal_undo_pending(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                          unsigned char *sources, unsigned char *sinks,
+                          unsigned char *healed_sinks,
+                          unsigned char *undid_pending,
+                          afr_transaction_type type, struct afr_reply *replies,
+                          unsigned char *locked_on);
+
+int
+afr_selfheal_recreate_entry(call_frame_t *frame, int dst, int source,
+                            unsigned char *sources, inode_t *dir,
+                            const char *name, inode_t *inode,
+                            struct afr_reply *replies);
+
+int
+afr_selfheal_post_op(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                     int subvol, dict_t *xattr, dict_t *xdata);
+
+call_frame_t *
+afr_frame_create(xlator_t *this, int32_t *op_errno);
+
+inode_t *
+afr_inode_find(xlator_t *this, uuid_t gfid);
+
+int
+afr_selfheal_discover_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int op_ret, int op_errno, inode_t *inode,
+                          struct iatt *buf, dict_t *xdata, struct iatt *parbuf);
+void
+afr_reply_copy(struct afr_reply *dst, struct afr_reply *src);
+
+void
+afr_replies_copy(struct afr_reply *dst, struct afr_reply *src, int count);
+
+int
+afr_selfheal_newentry_mark(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                           int source, struct afr_reply *replies,
+                           unsigned char *sources, unsigned char *newentry);
+
+unsigned int
+afr_success_count(struct afr_reply *replies, unsigned int count);
 
-#include <sys/stat.h>
+void
+afr_log_selfheal(uuid_t gfid, xlator_t *this, int ret, char *type, int source,
+                 unsigned char *sources, unsigned char *healed_sinks);
 
-#define FILETYPE_DIFFERS(buf1,buf2) ((buf1)->ia_type != (buf2)->ia_type)
-#define PERMISSION_DIFFERS(buf1,buf2) (st_mode_from_ia ((buf1)->ia_prot, (buf1)->ia_type) != st_mode_from_ia ((buf2)->ia_prot, (buf2)->ia_type))
-#define OWNERSHIP_DIFFERS(buf1,buf2) (((buf1)->ia_uid != (buf2)->ia_uid) || ((buf1)->ia_gid != (buf2)->ia_gid))
-#define SIZE_DIFFERS(buf1,buf2) ((buf1)->ia_size != (buf2)->ia_size)
+void
+afr_mark_largest_file_as_source(xlator_t *this, unsigned char *sources,
+                                struct afr_reply *replies);
+void
+afr_mark_active_sinks(xlator_t *this, unsigned char *sources,
+                      unsigned char *locked_on, unsigned char *sinks);
 
-#define SIZE_GREATER(buf1,buf2) ((buf1)->ia_size > (buf2)->ia_size)
+gf_boolean_t
+afr_dict_contains_heal_op(call_frame_t *frame);
 
+gf_boolean_t
+afr_can_decide_split_brain_source_sinks(struct afr_reply *replies,
+                                        int child_count);
 int
-afr_sh_has_metadata_pending (dict_t *xattr, int child_count, xlator_t *this);
+afr_mark_split_brain_source_sinks(
+    call_frame_t *frame, xlator_t *this, inode_t *inode, unsigned char *sources,
+    unsigned char *sinks, unsigned char *healed_sinks, unsigned char *locked_on,
+    struct afr_reply *replies, afr_transaction_type type);
+
+int
+afr_sh_get_fav_by_policy(xlator_t *this, struct afr_reply *replies,
+                         inode_t *inode, char **policy_str);
+
 int
-afr_sh_has_entry_pending (dict_t *xattr, int child_count, xlator_t *this);
+_afr_fav_child_reset_sink_xattrs(call_frame_t *frame, xlator_t *this,
+                                 inode_t *inode, int source,
+                                 unsigned char *healed_sinks,
+                                 unsigned char *undid_pending,
+                                 afr_transaction_type type,
+                                 unsigned char *locked_on,
+                                 struct afr_reply *replies);
+
 int
-afr_sh_has_data_pending (dict_t *xattr, int child_count, xlator_t *this);
+afr_get_child_index_from_name(xlator_t *this, char *name);
+
+gf_boolean_t
+afr_does_witness_exist(xlator_t *this, uint64_t *witness);
 
 int
-afr_self_heal_entry (call_frame_t *frame, xlator_t *this);
+__afr_selfheal_data_prepare(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                            unsigned char *locked_on, unsigned char *sources,
+                            unsigned char *sinks, unsigned char *healed_sinks,
+                            unsigned char *undid_pending,
+                            struct afr_reply *replies, unsigned char *flag);
 
 int
-afr_self_heal_data (call_frame_t *frame, xlator_t *this);
+__afr_selfheal_metadata_prepare(call_frame_t *frame, xlator_t *this,
+                                inode_t *inode, unsigned char *locked_on,
+                                unsigned char *sources, unsigned char *sinks,
+                                unsigned char *healed_sinks,
+                                unsigned char *undid_pending,
+                                struct afr_reply *replies, unsigned char *flag);
+int
+__afr_selfheal_entry_prepare(call_frame_t *frame, xlator_t *this,
+                             inode_t *inode, unsigned char *locked_on,
+                             unsigned char *sources, unsigned char *sinks,
+                             unsigned char *healed_sinks,
+                             struct afr_reply *replies, int *source_p,
+                             unsigned char *flag);
 
 int
-afr_self_heal_metadata (call_frame_t *frame, xlator_t *this);
+afr_selfheal_unlocked_inspect(call_frame_t *frame, xlator_t *this, uuid_t gfid,
+                              inode_t **link_inode, gf_boolean_t *data_selfheal,
+                              gf_boolean_t *metadata_selfheal,
+                              gf_boolean_t *entry_selfheal,
+                              struct afr_reply *replies);
 
 int
-afr_self_heal_get_source (xlator_t *this, afr_local_t *local, dict_t **xattr);
+afr_selfheal_do(call_frame_t *frame, xlator_t *this, uuid_t gfid);
+
+int
+afr_selfheal_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, dict_t *xdata);
+
+int
+afr_locked_fill(call_frame_t *frame, xlator_t *this, unsigned char *locked_on);
+int
+afr_choose_source_by_policy(afr_private_t *priv, unsigned char *sources,
+                            afr_transaction_type type);
 
 int
-afr_self_heal (call_frame_t *frame, xlator_t *this);
+afr_selfheal_metadata_by_stbuf(xlator_t *this, struct iatt *stbuf);
 
-#endif /* __AFR_SELF_HEAL_H__ */
+int
+afr_sh_fav_by_size(xlator_t *this, struct afr_reply *replies, inode_t *inode);
+int
+afr_sh_fav_by_mtime(xlator_t *this, struct afr_reply *replies, inode_t *inode);
+int
+afr_sh_fav_by_ctime(xlator_t *this, struct afr_reply *replies, inode_t *inode);
+
+int
+afr_gfid_split_brain_source(xlator_t *this, struct afr_reply *replies,
+                            inode_t *inode, uuid_t pargfid, const char *bname,
+                            int src_idx, int child_idx,
+                            unsigned char *locked_on, int *src, dict_t *xdata);
+int
+afr_mark_source_sinks_if_file_empty(xlator_t *this, unsigned char *sources,
+                                    unsigned char *sinks,
+                                    unsigned char *healed_sinks,
+                                    unsigned char *locked_on,
+                                    struct afr_reply *replies,
+                                    afr_transaction_type type);
+
+gf_boolean_t
+afr_is_file_empty_on_all_children(afr_private_t *priv,
+                                  struct afr_reply *replies);
+
+int
+afr_selfheal_entry_delete(xlator_t *this, inode_t *dir, const char *name,
+                          inode_t *inode, int child, struct afr_reply *replies);
+int
+afr_anon_inode_create(xlator_t *this, int child, inode_t **linked_inode);
+#endif /* !_AFR_SELFHEAL_H */
diff --git a/xlators/cluster/afr/src/afr-self-heald.c b/xlators/cluster/afr/src/afr-self-heald.c
new file mode 100644
index 00000000000..109fd4b7421
--- /dev/null
+++ b/xlators/cluster/afr/src/afr-self-heald.c
@@ -0,0 +1,1716 @@
+/*
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "afr.h"
+#include "afr-self-heal.h"
+#include "afr-self-heald.h"
+#include "protocol-common.h"
+#include <glusterfs/syncop-utils.h>
+#include "afr-messages.h"
+#include <glusterfs/byte-order.h>
+
+#define AFR_EH_SPLIT_BRAIN_LIMIT 1024
+#define AFR_STATISTICS_HISTORY_SIZE 50
+
+#define ASSERT_LOCAL(this, healer)                                             \
+    if (!afr_shd_is_subvol_local(this, healer->subvol)) {                      \
+        healer->local = _gf_false;                                             \
+        if (safe_break(healer)) {                                              \
+            break;                                                             \
+        } else {                                                               \
+            continue;                                                          \
+        }                                                                      \
+    } else {                                                                   \
+        healer->local = _gf_true;                                              \
+    }
+
+#define NTH_INDEX_HEALER(this, n)                                              \
+    &((((afr_private_t *)this->private))->shd.index_healers[n])
+#define NTH_FULL_HEALER(this, n)                                               \
+    &((((afr_private_t *)this->private))->shd.full_healers[n])
+
+char *
+afr_subvol_name(xlator_t *this, int subvol)
+{
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    if (subvol < 0 || subvol > priv->child_count)
+        return NULL;
+
+    return priv->children[subvol]->name;
+}
+
+void
+afr_destroy_crawl_event_data(void *data)
+{
+    return;
+}
+
+void
+afr_destroy_shd_event_data(void *data)
+{
+    shd_event_t *shd_event = data;
+
+    if (!shd_event)
+        return;
+    GF_FREE(shd_event->path);
+
+    return;
+}
+
+gf_boolean_t
+afr_shd_is_subvol_local(xlator_t *this, int subvol)
+{
+    afr_private_t *priv = NULL;
+    gf_boolean_t is_local = _gf_false;
+    loc_t loc = {
+        0,
+    };
+
+    loc.inode = this->itable->root;
+    gf_uuid_copy(loc.gfid, loc.inode->gfid);
+    priv = this->private;
+    syncop_is_subvol_local(priv->children[subvol], &loc, &is_local);
+    return is_local;
+}
+
+int
+__afr_shd_healer_wait(struct subvol_healer *healer)
+{
+    afr_private_t *priv = NULL;
+    struct timespec wait_till = {
+        0,
+    };
+    int ret = 0;
+
+    priv = healer->this->private;
+
+disabled_loop:
+    wait_till.tv_sec = gf_time() + priv->shd.timeout;
+
+    while (!healer->rerun) {
+        ret = pthread_cond_timedwait(&healer->cond, &healer->mutex, &wait_till);
+        if (ret == ETIMEDOUT)
+            break;
+    }
+
+    ret = healer->rerun;
+    healer->rerun = 0;
+
+    if (!priv->shd.enabled)
+        goto disabled_loop;
+
+    return ret;
+}
+
+int
+afr_shd_healer_wait(struct subvol_healer *healer)
+{
+    int ret = 0;
+
+    pthread_mutex_lock(&healer->mutex);
+    {
+        ret = __afr_shd_healer_wait(healer);
+    }
+    pthread_mutex_unlock(&healer->mutex);
+
+    return ret;
+}
+
+gf_boolean_t
+safe_break(struct subvol_healer *healer)
+{
+    gf_boolean_t ret = _gf_false;
+
+    pthread_mutex_lock(&healer->mutex);
+    {
+        if (healer->rerun)
+            goto unlock;
+
+        healer->running = _gf_false;
+        ret = _gf_true;
+    }
+unlock:
+    pthread_mutex_unlock(&healer->mutex);
+
+    return ret;
+}
+
+inode_t *
+afr_shd_inode_find(xlator_t *this, xlator_t *subvol, uuid_t gfid)
+{
+    int ret = 0;
+    uint64_t val = IA_INVAL;
+    dict_t *xdata = NULL;
+    dict_t *rsp_dict = NULL;
+    inode_t *inode = NULL;
+
+    xdata = dict_new();
+    if (!xdata)
+        goto out;
+
+    ret = dict_set_int8(xdata, GF_INDEX_IA_TYPE_GET_REQ, 1);
+    if (ret)
+        goto out;
+
+    ret = syncop_inode_find(this, subvol, gfid, &inode, xdata, &rsp_dict);
+    if (ret < 0)
+        goto out;
+
+    if (rsp_dict) {
+        ret = dict_get_uint64(rsp_dict, GF_INDEX_IA_TYPE_GET_RSP, &val);
+        if (ret)
+            goto out;
+    }
+    ret = inode_ctx_set2(inode, subvol, 0, &val);
+out:
+    if (ret && inode) {
+        inode_unref(inode);
+        inode = NULL;
+    }
+    if (xdata)
+        dict_unref(xdata);
+    if (rsp_dict)
+        dict_unref(rsp_dict);
+    return inode;
+}
+
+inode_t *
+afr_shd_index_inode(xlator_t *this, xlator_t *subvol, char *vgfid)
+{
+    loc_t rootloc = {
+        0,
+    };
+    inode_t *inode = NULL;
+    int ret = 0;
+    dict_t *xattr = NULL;
+    void *index_gfid = NULL;
+
+    rootloc.inode = inode_ref(this->itable->root);
+    gf_uuid_copy(rootloc.gfid, rootloc.inode->gfid);
+
+    ret = syncop_getxattr(subvol, &rootloc, &xattr, vgfid, NULL, NULL);
+    if (ret || !xattr) {
+        errno = -ret;
+        goto out;
+    }
+
+    ret = dict_get_ptr(xattr, vgfid, &index_gfid);
+    if (ret)
+        goto out;
+
+    gf_msg_debug(this->name, 0, "%s dir gfid for %s: %s", vgfid, subvol->name,
+                 uuid_utoa(index_gfid));
+
+    inode = afr_shd_inode_find(this, subvol, index_gfid);
+
+out:
+    loc_wipe(&rootloc);
+
+    if (xattr)
+        dict_unref(xattr);
+
+    return inode;
+}
+
+int
+afr_shd_entry_purge(xlator_t *subvol, inode_t *inode, char *name,
+                    ia_type_t type)
+{
+    int ret = 0;
+    loc_t loc = {
+        0,
+    };
+
+    loc.parent = inode_ref(inode);
+    loc.name = name;
+
+    if (IA_ISDIR(type))
+        ret = syncop_rmdir(subvol, &loc, 1, NULL, NULL);
+    else
+        ret = syncop_unlink(subvol, &loc, NULL, NULL);
+
+    loc_wipe(&loc);
+    return ret;
+}
+
+void
+afr_shd_zero_xattrop(xlator_t *this, uuid_t gfid)
+{
+    call_frame_t *frame = NULL;
+    inode_t *inode = NULL;
+    afr_private_t *priv = NULL;
+    dict_t *xattr = NULL;
+    int ret = 0;
+    int i = 0;
+    int raw[AFR_NUM_CHANGE_LOGS] = {0};
+
+    priv = this->private;
+    frame = afr_frame_create(this, NULL);
+    if (!frame)
+        goto out;
+    inode = afr_inode_find(this, gfid);
+    if (!inode)
+        goto out;
+    xattr = dict_new();
+    if (!xattr)
+        goto out;
+    ret = dict_set_static_bin(xattr, AFR_DIRTY, raw,
+                              sizeof(int) * AFR_NUM_CHANGE_LOGS);
+    if (ret)
+        goto out;
+    for (i = 0; i < priv->child_count; i++) {
+        ret = dict_set_static_bin(xattr, priv->pending_key[i], raw,
+                                  sizeof(int) * AFR_NUM_CHANGE_LOGS);
+        if (ret)
+            goto out;
+    }
+
+    /*Send xattrop to all bricks. Doing a lookup to see if bricks are up or
+     * has valid repies for this gfid seems a bit of an overkill.*/
+    for (i = 0; i < priv->child_count; i++)
+        afr_selfheal_post_op(frame, this, inode, i, xattr, NULL);
+
+out:
+    if (frame)
+        AFR_STACK_DESTROY(frame);
+    if (inode)
+        inode_unref(inode);
+    if (xattr)
+        dict_unref(xattr);
+    return;
+}
+
+int
+afr_shd_selfheal_name(struct subvol_healer *healer, int child, uuid_t parent,
+                      const char *bname)
+{
+    int ret = -1;
+
+    ret = afr_selfheal_name(THIS, parent, bname, NULL, NULL);
+
+    return ret;
+}
+
+int
+afr_shd_selfheal(struct subvol_healer *healer, int child, uuid_t gfid)
+{
+    int ret = 0;
+    eh_t *eh = NULL;
+    afr_private_t *priv = NULL;
+    afr_self_heald_t *shd = NULL;
+    shd_event_t *shd_event = NULL;
+    char *path = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *this = NULL;
+    crawl_event_t *crawl_event = NULL;
+
+    this = healer->this;
+    priv = this->private;
+    shd = &priv->shd;
+    crawl_event = &healer->crawl_event;
+
+    subvol = priv->children[child];
+
+    // If this fails with ENOENT/ESTALE index is stale
+    ret = syncop_gfid_to_path(this->itable, subvol, gfid, &path);
+    if (ret < 0)
+        return ret;
+
+    ret = afr_selfheal(this, gfid);
+
+    LOCK(&priv->lock);
+    {
+        if (ret == -EIO) {
+            eh = shd->split_brain;
+            crawl_event->split_brain_count++;
+        } else if (ret < 0) {
+            crawl_event->heal_failed_count++;
+        } else if (ret == 0) {
+            crawl_event->healed_count++;
+        }
+    }
+    UNLOCK(&priv->lock);
+
+    if (eh) {
+        shd_event = GF_CALLOC(1, sizeof(*shd_event), gf_afr_mt_shd_event_t);
+        if (!shd_event)
+            goto out;
+
+        shd_event->child = child;
+        shd_event->path = path;
+
+        if (eh_save_history(eh, shd_event) < 0)
+            goto out;
+
+        shd_event = NULL;
+        path = NULL;
+    }
+out:
+    GF_FREE(shd_event);
+    GF_FREE(path);
+    return ret;
+}
+
+void
+afr_shd_sweep_prepare(struct subvol_healer *healer)
+{
+    crawl_event_t *event = NULL;
+
+    event = &healer->crawl_event;
+
+    event->healed_count = 0;
+    event->split_brain_count = 0;
+    event->heal_failed_count = 0;
+
+    event->start_time = gf_time();
+    event->end_time = 0;
+    _mask_cancellation();
+}
+
+void
+afr_shd_sweep_done(struct subvol_healer *healer)
+{
+    crawl_event_t *event = NULL;
+    crawl_event_t *history = NULL;
+    afr_self_heald_t *shd = NULL;
+
+    event = &healer->crawl_event;
+    shd = &(((afr_private_t *)healer->this->private)->shd);
+
+    event->end_time = gf_time();
+    history = gf_memdup(event, sizeof(*event));
+    event->start_time = 0;
+
+    if (!history)
+        return;
+
+    if (eh_save_history(shd->statistics[healer->subvol], history) < 0)
+        GF_FREE(history);
+    _unmask_cancellation();
+}
+
+int
+afr_shd_index_heal(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent,
+                   void *data)
+{
+    struct subvol_healer *healer = data;
+    afr_private_t *priv = NULL;
+    uuid_t gfid = {0};
+    int ret = 0;
+    uint64_t val = IA_INVAL;
+
+    priv = healer->this->private;
+    if (!priv->shd.enabled)
+        return -EBUSY;
+
+    gf_msg_debug(healer->this->name, 0, "got entry: %s from %s", entry->d_name,
+                 priv->children[healer->subvol]->name);
+
+    ret = gf_uuid_parse(entry->d_name, gfid);
+    if (ret)
+        return 0;
+
+    inode_ctx_get2(parent->inode, subvol, NULL, &val);
+
+    ret = afr_shd_selfheal(healer, healer->subvol, gfid);
+
+    if (ret == -ENOENT || ret == -ESTALE)
+        afr_shd_entry_purge(subvol, parent->inode, entry->d_name, val);
+
+    if (ret == 2)
+        /* If bricks crashed in pre-op after creating indices/xattrop
+         * link but before setting afr changelogs, we end up with stale
+         * xattrop links but zero changelogs. Remove such entries by
+         * sending a post-op with zero changelogs.
+         */
+        afr_shd_zero_xattrop(healer->this, gfid);
+
+    return 0;
+}
+
+int
+afr_shd_index_sweep(struct subvol_healer *healer, char *vgfid)
+{
+    loc_t loc = {0};
+    afr_private_t *priv = NULL;
+    int ret = 0;
+    xlator_t *subvol = NULL;
+    dict_t *xdata = NULL;
+    call_frame_t *frame = NULL;
+
+    priv = healer->this->private;
+    subvol = priv->children[healer->subvol];
+
+    frame = afr_frame_create(healer->this, &ret);
+    if (!frame) {
+        ret = -ret;
+        goto out;
+    }
+
+    loc.inode = afr_shd_index_inode(healer->this, subvol, vgfid);
+    if (!loc.inode) {
+        gf_msg(healer->this->name, GF_LOG_WARNING, 0,
+               AFR_MSG_INDEX_DIR_GET_FAILED, "unable to get index-dir on %s",
+               subvol->name);
+        ret = -errno;
+        goto out;
+    }
+
+    xdata = dict_new();
+    if (!xdata || dict_set_int32_sizen(xdata, "get-gfid-type", 1)) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ret = syncop_mt_dir_scan(frame, subvol, &loc, GF_CLIENT_PID_SELF_HEALD,
+                             healer, afr_shd_index_heal, xdata,
+                             priv->shd.max_threads, priv->shd.wait_qlength);
+
+    if (ret == 0)
+        ret = healer->crawl_event.healed_count;
+
+out:
+    loc_wipe(&loc);
+
+    if (xdata)
+        dict_unref(xdata);
+    if (frame)
+        AFR_STACK_DESTROY(frame);
+    return ret;
+}
+
+int
+afr_shd_index_sweep_all(struct subvol_healer *healer)
+{
+    int ret = 0;
+    int count = 0;
+
+    ret = afr_shd_index_sweep(healer, GF_XATTROP_INDEX_GFID);
+    if (ret < 0)
+        goto out;
+    count = ret;
+
+    ret = afr_shd_index_sweep(healer, GF_XATTROP_DIRTY_GFID);
+    if (ret < 0)
+        goto out;
+    count += ret;
+
+    ret = afr_shd_index_sweep(healer, GF_XATTROP_ENTRY_CHANGES_GFID);
+    if (ret < 0)
+        goto out;
+    count += ret;
+out:
+    if (ret < 0)
+        return ret;
+    else
+        return count;
+}
+
+int
+afr_shd_full_heal(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent,
+                  void *data)
+{
+    struct subvol_healer *healer = data;
+    xlator_t *this = healer->this;
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+
+    if (this->cleanup_starting) {
+        return -ENOTCONN;
+    }
+
+    if (!priv->shd.enabled)
+        return -EBUSY;
+
+    afr_shd_selfheal_name(healer, healer->subvol, parent->inode->gfid,
+                          entry->d_name);
+
+    afr_shd_selfheal(healer, healer->subvol, entry->d_stat.ia_gfid);
+
+    return 0;
+}
+
+int
+afr_shd_full_sweep(struct subvol_healer *healer, inode_t *inode)
+{
+    afr_private_t *priv = NULL;
+    loc_t loc = {0};
+
+    priv = healer->this->private;
+    loc.inode = inode;
+    return syncop_ftw(priv->children[healer->subvol], &loc,
+                      GF_CLIENT_PID_SELF_HEALD, healer, afr_shd_full_heal);
+}
+
+int
+afr_shd_fill_ta_loc(xlator_t *this, loc_t *loc)
+{
+    afr_private_t *priv = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+    int ret = -1;
+
+    priv = this->private;
+    loc->parent = inode_ref(this->itable->root);
+    gf_uuid_copy(loc->pargfid, loc->parent->gfid);
+    loc->name = priv->pending_key[THIN_ARBITER_BRICK_INDEX];
+    loc->inode = inode_new(loc->parent->table);
+    GF_CHECK_ALLOC(loc->inode, ret, out);
+
+    if (!gf_uuid_is_null(priv->ta_gfid))
+        goto assign_gfid;
+
+    ret = syncop_lookup(priv->children[THIN_ARBITER_BRICK_INDEX], loc, &stbuf,
+                        0, 0, 0);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed lookup on file %s.", loc->name);
+        goto out;
+    }
+
+    gf_uuid_copy(priv->ta_gfid, stbuf.ia_gfid);
+
+assign_gfid:
+    gf_uuid_copy(loc->gfid, priv->ta_gfid);
+    ret = 0;
+
+out:
+    if (ret)
+        loc_wipe(loc);
+
+    return ret;
+}
+
+int
+_afr_shd_ta_get_xattrs(xlator_t *this, loc_t *loc, dict_t **xdata)
+{
+    afr_private_t *priv = NULL;
+    dict_t *xattr = NULL;
+    int raw[AFR_NUM_CHANGE_LOGS] = {
+        0,
+    };
+    int ret = -1;
+    int i = 0;
+
+    priv = this->private;
+
+    xattr = dict_new();
+    if (!xattr) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_DICT_GET_FAILED,
+               "Failed to create dict.");
+        goto out;
+    }
+    for (i = 0; i < priv->child_count; i++) {
+        ret = dict_set_static_bin(xattr, priv->pending_key[i], &raw,
+                                  AFR_NUM_CHANGE_LOGS * sizeof(int));
+        if (ret)
+            goto out;
+    }
+
+    ret = syncop_xattrop(priv->children[THIN_ARBITER_BRICK_INDEX], loc,
+                         GF_XATTROP_ADD_ARRAY, xattr, NULL, xdata, NULL);
+    if (ret || !(*xdata)) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Xattrop failed on %s.", loc->name);
+    }
+
+out:
+    if (xattr)
+        dict_unref(xattr);
+
+    return ret;
+}
+
+void
+afr_shd_ta_get_xattrs(xlator_t *this, loc_t *loc, struct subvol_healer *healer,
+                      dict_t **xdata)
+{
+    int ret = 0;
+
+    loc_wipe(loc);
+    if (afr_shd_fill_ta_loc(this, loc)) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to populate thin-arbiter loc for: %s.", loc->name);
+        ret = -1;
+        goto out;
+    }
+
+    ret = afr_ta_post_op_lock(this, loc);
+    if (ret)
+        goto out;
+
+    ret = _afr_shd_ta_get_xattrs(this, loc, xdata);
+    if (ret) {
+        if (*xdata) {
+            dict_unref(*xdata);
+            *xdata = NULL;
+        }
+    }
+
+    afr_ta_post_op_unlock(this, loc);
+
+out:
+    if (ret)
+        healer->rerun = 1;
+}
+
+int
+afr_shd_ta_unset_xattrs(xlator_t *this, loc_t *loc, dict_t **xdata, int healer)
+{
+    afr_private_t *priv = NULL;
+    dict_t *xattr = NULL;
+    gf_boolean_t need_xattrop = _gf_false;
+    void *pending_raw = NULL;
+    int *raw = NULL;
+    int pending[AFR_NUM_CHANGE_LOGS] = {
+        0,
+    };
+    int i = 0;
+    int j = 0;
+    int val = 0;
+    int ret = -1;
+
+    priv = this->private;
+
+    xattr = dict_new();
+    if (!xattr) {
+        goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        raw = GF_CALLOC(AFR_NUM_CHANGE_LOGS, sizeof(int), gf_afr_mt_int32_t);
+        if (!raw) {
+            goto out;
+        }
+
+        ret = dict_get_ptr(*xdata, priv->pending_key[i], &pending_raw);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_GET_FAILED,
+                   "Error getting value "
+                   "of pending key %s",
+                   priv->pending_key[i]);
+            GF_FREE(raw);
+            goto out;
+        }
+
+        memcpy(pending, pending_raw, sizeof(pending));
+        for (j = 0; j < AFR_NUM_CHANGE_LOGS; j++) {
+            val = ntoh32(pending[j]);
+            if (val) {
+                if (i == healer) {
+                    gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_THIN_ARB,
+                           "I am "
+                           "not the good shd. Skipping. "
+                           "SHD = %d.",
+                           healer);
+                    ret = 0;
+                    GF_FREE(raw);
+                    goto out;
+                }
+                need_xattrop = _gf_true;
+                raw[j] = hton32(-val);
+            }
+        }
+
+        ret = dict_set_bin(xattr, priv->pending_key[i], raw,
+                           AFR_NUM_CHANGE_LOGS * sizeof(int));
+        if (ret) {
+            GF_FREE(raw);
+            goto out;
+        }
+
+        if (need_xattrop)
+            break;
+    }
+
+    if (!need_xattrop) {
+        ret = 0;
+        goto out;
+    }
+
+    ret = syncop_xattrop(priv->children[THIN_ARBITER_BRICK_INDEX], loc,
+                         GF_XATTROP_ADD_ARRAY, xattr, NULL, NULL, NULL);
+    if (ret)
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Xattrop failed.");
+
+out:
+    if (xattr)
+        dict_unref(xattr);
+
+    return ret;
+}
+
+void
+afr_shd_ta_check_and_unset_xattrs(xlator_t *this, loc_t *loc,
+                                  struct subvol_healer *healer,
+                                  dict_t *pre_crawl_xdata)
+{
+    int ret_lock = 0;
+    int ret = 0;
+    dict_t *post_crawl_xdata = NULL;
+
+    ret_lock = afr_ta_post_op_lock(this, loc);
+    if (ret_lock)
+        goto unref;
+
+    ret = _afr_shd_ta_get_xattrs(this, loc, &post_crawl_xdata);
+    if (ret)
+        goto unref;
+
+    if (!are_dicts_equal(pre_crawl_xdata, post_crawl_xdata, NULL, NULL)) {
+        ret = -1;
+        goto unref;
+    }
+
+    ret = afr_shd_ta_unset_xattrs(this, loc, &post_crawl_xdata, healer->subvol);
+
+unref:
+    if (post_crawl_xdata) {
+        dict_unref(post_crawl_xdata);
+        post_crawl_xdata = NULL;
+    }
+
+    if (ret || ret_lock)
+        healer->rerun = 1;
+
+    if (!ret_lock)
+        afr_ta_post_op_unlock(this, loc);
+}
+
+gf_boolean_t
+afr_bricks_available_for_heal(afr_private_t *priv)
+{
+    int up_children = 0;
+
+    up_children = __afr_get_up_children_count(priv);
+    if (up_children < 2) {
+        return _gf_false;
+    }
+    return _gf_true;
+}
+
+static gf_boolean_t
+afr_shd_ta_needs_heal(xlator_t *this, struct subvol_healer *healer)
+{
+    dict_t *xdata = NULL;
+    afr_private_t *priv = NULL;
+    loc_t loc = {
+        0,
+    };
+    int ret = -1;
+    int i = 0;
+    gf_boolean_t need_heal = _gf_false;
+
+    priv = this->private;
+
+    ret = afr_shd_fill_ta_loc(this, &loc);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to populate thin-arbiter loc for: %s.", loc.name);
+        healer->rerun = 1;
+        goto out;
+    }
+
+    if (_afr_shd_ta_get_xattrs(this, &loc, &xdata)) {
+        healer->rerun = 1;
+        goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (afr_ta_dict_contains_pending_xattr(xdata, priv, i)) {
+            need_heal = _gf_true;
+            break;
+        }
+    }
+
+out:
+    if (xdata)
+        dict_unref(xdata);
+    loc_wipe(&loc);
+
+    return need_heal;
+}
+
+static int
+afr_shd_anon_inode_cleaner(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent,
+                           void *data)
+{
+    struct subvol_healer *healer = data;
+    afr_private_t *priv = healer->this->private;
+    call_frame_t *frame = NULL;
+    afr_local_t *local = NULL;
+    int ret = 0;
+    loc_t loc = {0};
+    int count = 0;
+    int i = 0;
+    int op_errno = 0;
+    struct iatt *iatt = NULL;
+    gf_boolean_t multiple_links = _gf_false;
+    unsigned char *gfid_present = alloca0(priv->child_count);
+    unsigned char *entry_present = alloca0(priv->child_count);
+    char *type = "file";
+
+    frame = afr_frame_create(healer->this, &ret);
+    if (!frame) {
+        ret = -ret;
+        goto out;
+    }
+    local = frame->local;
+    if (AFR_COUNT(local->child_up, priv->child_count) != priv->child_count) {
+        gf_msg_debug(healer->this->name, 0,
+                     "Not all bricks are up. Skipping "
+                     "cleanup of %s on %s",
+                     entry->d_name, subvol->name);
+        ret = 0;
+        goto out;
+    }
+
+    loc.inode = inode_new(parent->inode->table);
+    if (!loc.inode) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    ret = gf_uuid_parse(entry->d_name, loc.gfid);
+    if (ret) {
+        ret = 0;
+        goto out;
+    }
+    AFR_ONLIST(local->child_up, frame, afr_selfheal_discover_cbk, lookup, &loc,
+               NULL);
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->replies[i].op_ret == 0) {
+            count++;
+            gfid_present[i] = 1;
+            iatt = &local->replies[i].poststat;
+            if (iatt->ia_type == IA_IFDIR) {
+                type = "dir";
+            }
+
+            if (i == healer->subvol) {
+                if (local->replies[i].poststat.ia_nlink > 1) {
+                    multiple_links = _gf_true;
+                }
+            }
+        } else if (local->replies[i].op_errno != ENOENT &&
+                   local->replies[i].op_errno != ESTALE) {
+            /*We don't have complete view. Skip the entry*/
+            gf_msg_debug(healer->this->name, local->replies[i].op_errno,
+                         "Skipping cleanup of %s on %s", entry->d_name,
+                         subvol->name);
+            ret = 0;
+            goto out;
+        }
+    }
+
+    /*Inode is deleted from subvol*/
+    if (count == 1 || (iatt->ia_type != IA_IFDIR && multiple_links)) {
+        gf_msg(healer->this->name, GF_LOG_WARNING, 0,
+               AFR_MSG_EXPUNGING_FILE_OR_DIR, "expunging %s %s/%s on %s", type,
+               priv->anon_inode_name, entry->d_name, subvol->name);
+        ret = afr_shd_entry_purge(subvol, parent->inode, entry->d_name,
+                                  iatt->ia_type);
+        if (ret == -ENOENT || ret == -ESTALE)
+            ret = 0;
+    } else if (count > 1) {
+        loc_wipe(&loc);
+        loc.parent = inode_ref(parent->inode);
+        loc.name = entry->d_name;
+        loc.inode = inode_new(parent->inode->table);
+        if (!loc.inode) {
+            ret = -ENOMEM;
+            goto out;
+        }
+        AFR_ONLIST(local->child_up, frame, afr_selfheal_discover_cbk, lookup,
+                   &loc, NULL);
+        count = 0;
+        for (i = 0; i < priv->child_count; i++) {
+            if (local->replies[i].op_ret == 0) {
+                count++;
+                entry_present[i] = 1;
+                iatt = &local->replies[i].poststat;
+            } else if (local->replies[i].op_errno != ENOENT &&
+                       local->replies[i].op_errno != ESTALE) {
+                /*We don't have complete view. Skip the entry*/
+                gf_msg_debug(healer->this->name, local->replies[i].op_errno,
+                             "Skipping cleanup of %s on %s", entry->d_name,
+                             subvol->name);
+                ret = 0;
+                goto out;
+            }
+        }
+        for (i = 0; i < priv->child_count; i++) {
+            if (gfid_present[i] && !entry_present[i]) {
+                /*Entry is not anonymous on at least one subvol*/
+                gf_msg_debug(healer->this->name, 0,
+                             "Valid entry present on %s "
+                             "Skipping cleanup of %s on %s",
+                             priv->children[i]->name, entry->d_name,
+                             subvol->name);
+                ret = 0;
+                goto out;
+            }
+        }
+
+        gf_msg(healer->this->name, GF_LOG_WARNING, 0,
+               AFR_MSG_EXPUNGING_FILE_OR_DIR,
+               "expunging %s %s/%s on all subvols", type, priv->anon_inode_name,
+               entry->d_name);
+        ret = 0;
+        for (i = 0; i < priv->child_count; i++) {
+            op_errno = -afr_shd_entry_purge(priv->children[i], loc.parent,
+                                            entry->d_name, iatt->ia_type);
+            if (op_errno != ENOENT && op_errno != ESTALE) {
+                ret |= -op_errno;
+            }
+        }
+    }
+
+out:
+    if (frame)
+        AFR_STACK_DESTROY(frame);
+    loc_wipe(&loc);
+    return ret;
+}
+
+static void
+afr_cleanup_anon_inode_dir(struct subvol_healer *healer)
+{
+    int ret = 0;
+    call_frame_t *frame = NULL;
+    afr_private_t *priv = healer->this->private;
+    loc_t loc = {0};
+
+    ret = afr_anon_inode_create(healer->this, healer->subvol, &loc.inode);
+    if (ret)
+        goto out;
+
+    frame = afr_frame_create(healer->this, &ret);
+    if (!frame) {
+        ret = -ret;
+        goto out;
+    }
+
+    ret = syncop_mt_dir_scan(frame, priv->children[healer->subvol], &loc,
+                             GF_CLIENT_PID_SELF_HEALD, healer,
+                             afr_shd_anon_inode_cleaner, NULL,
+                             priv->shd.max_threads, priv->shd.wait_qlength);
+out:
+    if (frame)
+        AFR_STACK_DESTROY(frame);
+    loc_wipe(&loc);
+    return;
+}
+
+void *
+afr_shd_index_healer(void *data)
+{
+    struct subvol_healer *healer = NULL;
+    xlator_t *this = NULL;
+    int ret = 0;
+    afr_private_t *priv = NULL;
+    dict_t *pre_crawl_xdata = NULL;
+    loc_t loc = {
+        0,
+    };
+
+    healer = data;
+    THIS = this = healer->this;
+    priv = this->private;
+
+    for (;;) {
+        afr_shd_healer_wait(healer);
+
+        if (!afr_bricks_available_for_heal(priv))
+            continue;
+
+        ASSERT_LOCAL(this, healer);
+        priv->local[healer->subvol] = healer->local;
+
+        if (priv->thin_arbiter_count) {
+            if (afr_shd_ta_needs_heal(this, healer))
+                afr_shd_ta_get_xattrs(this, &loc, healer, &pre_crawl_xdata);
+        }
+
+        do {
+            gf_msg_debug(this->name, 0, "starting index sweep on subvol %s",
+                         afr_subvol_name(this, healer->subvol));
+
+            afr_shd_sweep_prepare(healer);
+
+            ret = afr_shd_index_sweep_all(healer);
+
+            afr_shd_sweep_done(healer);
+            /*
+              As long as at least one gfid was
+              healed, keep retrying. We may have
+              just healed a directory and thereby
+              created entries for other gfids which
+              could not be healed thus far.
+            */
+
+            gf_msg_debug(this->name, 0, "finished index sweep on subvol %s",
+                         afr_subvol_name(this, healer->subvol));
+            /*
+              Give a pause before retrying to avoid a busy loop
+              in case the only entry in index is because of
+              an ongoing I/O.
+            */
+            sleep(1);
+        } while (ret > 0);
+
+        if (ret == 0) {
+            afr_cleanup_anon_inode_dir(healer);
+        }
+
+        if (ret == 0 && pre_crawl_xdata &&
+            !healer->crawl_event.heal_failed_count) {
+            afr_shd_ta_check_and_unset_xattrs(this, &loc, healer,
+                                              pre_crawl_xdata);
+        }
+
+        if (pre_crawl_xdata) {
+            dict_unref(pre_crawl_xdata);
+            pre_crawl_xdata = NULL;
+        }
+    }
+
+    return NULL;
+}
+
+void *
+afr_shd_full_healer(void *data)
+{
+    struct subvol_healer *healer = NULL;
+    xlator_t *this = NULL;
+    int run = 0;
+
+    healer = data;
+    THIS = this = healer->this;
+
+    for (;;) {
+        pthread_mutex_lock(&healer->mutex);
+        {
+            run = __afr_shd_healer_wait(healer);
+            if (!run)
+                healer->running = _gf_false;
+        }
+        pthread_mutex_unlock(&healer->mutex);
+
+        if (!run)
+            break;
+
+        ASSERT_LOCAL(this, healer);
+
+        gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_SELF_HEAL_INFO,
+               "starting full sweep on subvol %s",
+               afr_subvol_name(this, healer->subvol));
+
+        afr_shd_sweep_prepare(healer);
+
+        afr_shd_full_sweep(healer, this->itable->root);
+
+        afr_shd_sweep_done(healer);
+
+        gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_SELF_HEAL_INFO,
+               "finished full sweep on subvol %s",
+               afr_subvol_name(this, healer->subvol));
+    }
+
+    return NULL;
+}
+
+int
+afr_shd_healer_init(xlator_t *this, struct subvol_healer *healer)
+{
+    int ret = 0;
+
+    ret = pthread_mutex_init(&healer->mutex, NULL);
+    if (ret)
+        goto out;
+
+    ret = pthread_cond_init(&healer->cond, NULL);
+    if (ret)
+        goto out;
+
+    healer->this = this;
+    healer->running = _gf_false;
+    healer->rerun = _gf_false;
+    healer->local = _gf_false;
+out:
+    return ret;
+}
+
+int
+afr_shd_healer_spawn(xlator_t *this, struct subvol_healer *healer,
+                     void *(threadfn)(void *))
+{
+    int ret = 0;
+
+    pthread_mutex_lock(&healer->mutex);
+    {
+        if (healer->running) {
+            pthread_cond_signal(&healer->cond);
+        } else {
+            ret = gf_thread_create(&healer->thread, NULL, threadfn, healer,
+                                   "shdheal");
+            if (ret)
+                goto unlock;
+            healer->running = 1;
+        }
+
+        healer->rerun = 1;
+    }
+unlock:
+    pthread_mutex_unlock(&healer->mutex);
+
+    return ret;
+}
+
+int
+afr_shd_full_healer_spawn(xlator_t *this, int subvol)
+{
+    return afr_shd_healer_spawn(this, NTH_FULL_HEALER(this, subvol),
+                                afr_shd_full_healer);
+}
+
+int
+afr_shd_index_healer_spawn(xlator_t *this, int subvol)
+{
+    return afr_shd_healer_spawn(this, NTH_INDEX_HEALER(this, subvol),
+                                afr_shd_index_healer);
+}
+
+int
+afr_shd_dict_add_crawl_event(xlator_t *this, dict_t *output,
+                             crawl_event_t *crawl_event)
+{
+    int ret = 0;
+    uint64_t count = 0;
+    char key[128] = {0};
+    int keylen = 0;
+    char suffix[64] = {0};
+    int xl_id = 0;
+    uint64_t healed_count = 0;
+    uint64_t split_brain_count = 0;
+    uint64_t heal_failed_count = 0;
+    char *start_time_str = 0;
+    char *end_time_str = NULL;
+    char *crawl_type = NULL;
+    int progress = -1;
+    int child = -1;
+
+    child = crawl_event->child;
+    healed_count = crawl_event->healed_count;
+    split_brain_count = crawl_event->split_brain_count;
+    heal_failed_count = crawl_event->heal_failed_count;
+    crawl_type = crawl_event->crawl_type;
+
+    if (!crawl_event->start_time)
+        goto out;
+
+    start_time_str = gf_strdup(ctime(&crawl_event->start_time));
+
+    if (crawl_event->end_time)
+        end_time_str = gf_strdup(ctime(&crawl_event->end_time));
+
+    ret = dict_get_int32(output, this->name, &xl_id);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_GET_FAILED,
+               "xl does not have id");
+        goto out;
+    }
+
+    snprintf(key, sizeof(key), "statistics-%d-%d-count", xl_id, child);
+    ret = dict_get_uint64(output, key, &count);
+
+    snprintf(suffix, sizeof(suffix), "%d-%d-%" PRIu64, xl_id, child, count);
+    snprintf(key, sizeof(key), "statistics_healed_cnt-%s", suffix);
+    ret = dict_set_uint64(output, key, healed_count);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+               "Could not add statistics_healed_count to output");
+        goto out;
+    }
+
+    snprintf(key, sizeof(key), "statistics_sb_cnt-%s", suffix);
+    ret = dict_set_uint64(output, key, split_brain_count);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+               "Could not add statistics_split_brain_count to output");
+        goto out;
+    }
+
+    keylen = snprintf(key, sizeof(key), "statistics_crawl_type-%s", suffix);
+    ret = dict_set_strn(output, key, keylen, crawl_type);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+               "Could not add statistics_crawl_type to output");
+        goto out;
+    }
+
+    snprintf(key, sizeof(key), "statistics_heal_failed_cnt-%s", suffix);
+    ret = dict_set_uint64(output, key, heal_failed_count);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+               "Could not add statistics_healed_failed_count to output");
+        goto out;
+    }
+
+    keylen = snprintf(key, sizeof(key), "statistics_strt_time-%s", suffix);
+    ret = dict_set_dynstrn(output, key, keylen, start_time_str);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+               "Could not add statistics_crawl_start_time to output");
+        goto out;
+    } else {
+        start_time_str = NULL;
+    }
+
+    if (!end_time_str)
+        progress = 1;
+    else
+        progress = 0;
+
+    keylen = snprintf(key, sizeof(key), "statistics_end_time-%s", suffix);
+    if (!end_time_str)
+        end_time_str = gf_strdup("Could not determine the end time");
+    ret = dict_set_dynstrn(output, key, keylen, end_time_str);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+               "Could not add statistics_crawl_end_time to output");
+        goto out;
+    } else {
+        end_time_str = NULL;
+    }
+
+    keylen = snprintf(key, sizeof(key), "statistics_inprogress-%s", suffix);
+
+    ret = dict_set_int32n(output, key, keylen, progress);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+               "Could not add statistics_inprogress to output");
+        goto out;
+    }
+
+    snprintf(key, sizeof(key), "statistics-%d-%d-count", xl_id, child);
+    ret = dict_set_uint64(output, key, count + 1);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+               "Could not increment the counter.");
+        goto out;
+    }
+out:
+    GF_FREE(start_time_str);
+    GF_FREE(end_time_str);
+    return ret;
+}
+
+int
+afr_shd_dict_add_path(xlator_t *this, dict_t *output, int child, char *path,
+                      struct timeval *tv)
+{
+    int ret = -1;
+    uint64_t count = 0;
+    char key[64] = {0};
+    int keylen = 0;
+    char xl_id_child_str[32] = {0};
+    int xl_id = 0;
+
+    ret = dict_get_int32(output, this->name, &xl_id);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_GET_FAILED,
+               "xl does not have id");
+        goto out;
+    }
+
+    snprintf(xl_id_child_str, sizeof(xl_id_child_str), "%d-%d", xl_id, child);
+    snprintf(key, sizeof(key), "%s-count", xl_id_child_str);
+    ret = dict_get_uint64(output, key, &count);
+
+    keylen = snprintf(key, sizeof(key), "%s-%" PRIu64, xl_id_child_str, count);
+    ret = dict_set_dynstrn(output, key, keylen, path);
+
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+               "%s: Could not add to output", path);
+        goto out;
+    }
+
+    if (tv) {
+        snprintf(key, sizeof(key), "%s-%" PRIu64 "-time", xl_id_child_str,
+                 count);
+        ret = dict_set_uint32(output, key, tv->tv_sec);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+                   "%s: Could not set time", path);
+            goto out;
+        }
+    }
+
+    snprintf(key, sizeof(key), "%s-count", xl_id_child_str);
+
+    ret = dict_set_uint64(output, key, count + 1);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+               "Could not increment count");
+        goto out;
+    }
+
+    ret = 0;
+out:
+    return ret;
+}
+
+int
+afr_add_shd_event(circular_buffer_t *cb, void *data)
+{
+    dict_t *output = NULL;
+    xlator_t *this = THIS;
+    afr_private_t *priv = NULL;
+    afr_self_heald_t *shd = NULL;
+    shd_event_t *shd_event = NULL;
+    char *path = NULL;
+
+    output = data;
+    priv = this->private;
+    shd = &priv->shd;
+    shd_event = cb->data;
+
+    if (!shd->index_healers[shd_event->child].local)
+        return 0;
+
+    path = gf_strdup(shd_event->path);
+    if (!path)
+        return -ENOMEM;
+
+    afr_shd_dict_add_path(this, output, shd_event->child, path, &cb->tv);
+    return 0;
+}
+
+int
+afr_add_crawl_event(circular_buffer_t *cb, void *data)
+{
+    dict_t *output = NULL;
+    xlator_t *this = THIS;
+    afr_private_t *priv = NULL;
+    afr_self_heald_t *shd = NULL;
+    crawl_event_t *crawl_event = NULL;
+
+    output = data;
+    priv = this->private;
+    shd = &priv->shd;
+    crawl_event = cb->data;
+
+    if (!shd->index_healers[crawl_event->child].local)
+        return 0;
+
+    afr_shd_dict_add_crawl_event(this, output, crawl_event);
+
+    return 0;
+}
+
+int
+afr_selfheal_daemon_init(xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    afr_self_heald_t *shd = NULL;
+    int ret = -1;
+    int i = 0;
+
+    priv = this->private;
+    shd = &priv->shd;
+
+    shd->index_healers = GF_CALLOC(sizeof(*shd->index_healers),
+                                   priv->child_count,
+                                   gf_afr_mt_subvol_healer_t);
+    if (!shd->index_healers)
+        goto out;
+
+    for (i = 0; i < priv->child_count; i++) {
+        shd->index_healers[i].subvol = i;
+        ret = afr_shd_healer_init(this, &shd->index_healers[i]);
+        if (ret)
+            goto out;
+    }
+
+    shd->full_healers = GF_CALLOC(sizeof(*shd->full_healers), priv->child_count,
+                                  gf_afr_mt_subvol_healer_t);
+    if (!shd->full_healers)
+        goto out;
+    for (i = 0; i < priv->child_count; i++) {
+        shd->full_healers[i].subvol = i;
+        ret = afr_shd_healer_init(this, &shd->full_healers[i]);
+        if (ret)
+            goto out;
+    }
+
+    shd->split_brain = eh_new(AFR_EH_SPLIT_BRAIN_LIMIT, _gf_false,
+                              afr_destroy_shd_event_data);
+    if (!shd->split_brain)
+        goto out;
+
+    shd->statistics = GF_CALLOC(sizeof(eh_t *), priv->child_count,
+                                gf_common_mt_eh_t);
+    if (!shd->statistics)
+        goto out;
+
+    for (i = 0; i < priv->child_count; i++) {
+        shd->statistics[i] = eh_new(AFR_STATISTICS_HISTORY_SIZE, _gf_false,
+                                    afr_destroy_crawl_event_data);
+        if (!shd->statistics[i])
+            goto out;
+        shd->full_healers[i].crawl_event.child = i;
+        shd->full_healers[i].crawl_event.crawl_type = "FULL";
+        shd->index_healers[i].crawl_event.child = i;
+        shd->index_healers[i].crawl_event.crawl_type = "INDEX";
+    }
+
+    ret = 0;
+out:
+    return ret;
+}
+
+void
+afr_selfheal_childup(xlator_t *this, afr_private_t *priv)
+{
+    int subvol = 0;
+
+    if (!priv->shd.iamshd)
+        return;
+    for (subvol = 0; subvol < priv->child_count; subvol++)
+        if (priv->child_up[subvol])
+            afr_shd_index_healer_spawn(this, subvol);
+
+    return;
+}
+
+int
+afr_shd_get_index_count(xlator_t *this, int i, uint64_t *count)
+{
+    afr_private_t *priv = NULL;
+    xlator_t *subvol = NULL;
+    loc_t rootloc = {
+        0,
+    };
+    dict_t *xattr = NULL;
+    int ret = -1;
+
+    priv = this->private;
+    subvol = priv->children[i];
+
+    rootloc.inode = inode_ref(this->itable->root);
+    gf_uuid_copy(rootloc.gfid, rootloc.inode->gfid);
+
+    ret = syncop_getxattr(subvol, &rootloc, &xattr, GF_XATTROP_INDEX_COUNT,
+                          NULL, NULL);
+    if (ret < 0)
+        goto out;
+
+    ret = dict_get_uint64(xattr, GF_XATTROP_INDEX_COUNT, count);
+    if (ret)
+        goto out;
+
+    ret = 0;
+
+out:
+    if (xattr)
+        dict_unref(xattr);
+    loc_wipe(&rootloc);
+
+    return ret;
+}
+
+int
+afr_xl_op(xlator_t *this, dict_t *input, dict_t *output)
+{
+    gf_xl_afr_op_t op = GF_SHD_OP_INVALID;
+    int ret = 0;
+    int xl_id = 0;
+    afr_private_t *priv = NULL;
+    afr_self_heald_t *shd = NULL;
+    struct subvol_healer *healer = NULL;
+    int i = 0;
+    char key[64];
+    int keylen = 0;
+    int this_name_len = 0;
+    int op_ret = 0;
+    uint64_t cnt = 0;
+
+#define AFR_SET_DICT_AND_LOG(name, output, key, keylen, dict_str,              \
+                             dict_str_len)                                     \
+    {                                                                          \
+        int ret;                                                               \
+                                                                               \
+        ret = dict_set_nstrn(output, key, keylen, dict_str, dict_str_len);     \
+        if (ret) {                                                             \
+            gf_smsg(name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,         \
+                    "key=%s", key, "value=%s", dict_str, NULL);                \
+        }                                                                      \
+    }
+
+    priv = this->private;
+    shd = &priv->shd;
+
+    ret = dict_get_int32_sizen(input, "xl-op", (int32_t *)&op);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_GET_FAILED,
+                "key=xl-op", NULL);
+        goto out;
+    }
+    this_name_len = strlen(this->name);
+    ret = dict_get_int32n(input, this->name, this_name_len, &xl_id);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_GET_FAILED,
+                "key=%s", this->name, NULL);
+        goto out;
+    }
+    ret = dict_set_int32n(output, this->name, this_name_len, xl_id);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_DICT_SET_FAILED,
+                "key=%s", this->name, NULL);
+        goto out;
+    }
+    switch (op) {
+        case GF_SHD_OP_HEAL_INDEX:
+            op_ret = 0;
+
+            for (i = 0; i < priv->child_count; i++) {
+                healer = &shd->index_healers[i];
+                keylen = snprintf(key, sizeof(key), "%d-%d-status", xl_id, i);
+
+                if (!priv->child_up[i]) {
+                    AFR_SET_DICT_AND_LOG(this->name, output, key, keylen,
+                                         SBRICK_NOT_CONNECTED,
+                                         SLEN(SBRICK_NOT_CONNECTED));
+                    op_ret = -1;
+                } else if (AFR_COUNT(priv->child_up, priv->child_count) < 2) {
+                    AFR_SET_DICT_AND_LOG(this->name, output, key, keylen,
+                                         SLESS_THAN2_BRICKS_in_REP,
+                                         SLEN(SLESS_THAN2_BRICKS_in_REP));
+                    op_ret = -1;
+                } else if (!afr_shd_is_subvol_local(this, healer->subvol)) {
+                    AFR_SET_DICT_AND_LOG(this->name, output, key, keylen,
+                                         SBRICK_IS_REMOTE,
+                                         SLEN(SBRICK_IS_REMOTE));
+                } else {
+                    AFR_SET_DICT_AND_LOG(this->name, output, key, keylen,
+                                         SSTARTED_SELF_HEAL,
+                                         SLEN(SSTARTED_SELF_HEAL));
+
+                    ret = afr_shd_index_healer_spawn(this, i);
+
+                    if (ret) {
+                        gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                                AFR_MSG_HEALER_SPAWN_FAILED, NULL);
+                    }
+                }
+            }
+            break;
+        case GF_SHD_OP_HEAL_FULL:
+            op_ret = -1;
+
+            for (i = 0; i < priv->child_count; i++) {
+                healer = &shd->full_healers[i];
+                keylen = snprintf(key, sizeof(key), "%d-%d-status", xl_id, i);
+
+                if (!priv->child_up[i]) {
+                    AFR_SET_DICT_AND_LOG(this->name, output, key, keylen,
+                                         SBRICK_NOT_CONNECTED,
+                                         SLEN(SBRICK_NOT_CONNECTED));
+                } else if (AFR_COUNT(priv->child_up, priv->child_count) < 2) {
+                    AFR_SET_DICT_AND_LOG(this->name, output, key, keylen,
+                                         SLESS_THAN2_BRICKS_in_REP,
+                                         SLEN(SLESS_THAN2_BRICKS_in_REP));
+                } else if (!afr_shd_is_subvol_local(this, healer->subvol)) {
+                    AFR_SET_DICT_AND_LOG(this->name, output, key, keylen,
+                                         SBRICK_IS_REMOTE,
+                                         SLEN(SBRICK_IS_REMOTE));
+                } else {
+                    AFR_SET_DICT_AND_LOG(this->name, output, key, keylen,
+                                         SSTARTED_SELF_HEAL,
+                                         SLEN(SSTARTED_SELF_HEAL));
+
+                    ret = afr_shd_full_healer_spawn(this, i);
+
+                    if (ret) {
+                        gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                                AFR_MSG_HEALER_SPAWN_FAILED, NULL);
+                    }
+                    op_ret = 0;
+                }
+            }
+            break;
+        case GF_SHD_OP_INDEX_SUMMARY:
+            /* this case has been handled in glfs-heal.c */
+            break;
+        case GF_SHD_OP_SPLIT_BRAIN_FILES:
+            eh_dump(shd->split_brain, output, afr_add_shd_event);
+            break;
+        case GF_SHD_OP_STATISTICS:
+            for (i = 0; i < priv->child_count; i++) {
+                eh_dump(shd->statistics[i], output, afr_add_crawl_event);
+                ret = afr_shd_dict_add_crawl_event(
+                    this, output, &shd->index_healers[i].crawl_event);
+                if (ret) {
+                    gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                            AFR_MSG_ADD_CRAWL_EVENT_FAILED, NULL);
+                }
+
+                ret = afr_shd_dict_add_crawl_event(
+                    this, output, &shd->full_healers[i].crawl_event);
+                if (ret) {
+                    gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                            AFR_MSG_ADD_CRAWL_EVENT_FAILED, NULL);
+                }
+            }
+            break;
+        case GF_SHD_OP_STATISTICS_HEAL_COUNT:
+        case GF_SHD_OP_STATISTICS_HEAL_COUNT_PER_REPLICA:
+            op_ret = -1;
+
+            for (i = 0; i < priv->child_count; i++) {
+                if (!priv->child_up[i]) {
+                    keylen = snprintf(key, sizeof(key), "%d-%d-status", xl_id,
+                                      i);
+                    AFR_SET_DICT_AND_LOG(this->name, output, key, keylen,
+                                         SBRICK_NOT_CONNECTED,
+                                         SLEN(SBRICK_NOT_CONNECTED));
+                } else {
+                    snprintf(key, sizeof(key), "%d-%d-hardlinks", xl_id, i);
+                    ret = afr_shd_get_index_count(this, i, &cnt);
+                    if (ret == 0) {
+                        ret = dict_set_uint64(output, key, cnt);
+                    }
+                    if (ret) {
+                        gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                                AFR_MSG_DICT_SET_FAILED, NULL);
+                    }
+                    op_ret = 0;
+                }
+            }
+
+            break;
+
+        default:
+            gf_smsg(this->name, GF_LOG_ERROR, 0, AFR_MSG_INVALID_ARG, "op=%d",
+                    op, NULL);
+            break;
+    }
+out:
+    dict_deln(output, this->name, this_name_len);
+    return op_ret;
+
+#undef AFR_SET_DICT_AND_LOG
+}
diff --git a/xlators/cluster/afr/src/afr-self-heald.h b/xlators/cluster/afr/src/afr-self-heald.h
new file mode 100644
index 00000000000..18db728ea7b
--- /dev/null
+++ b/xlators/cluster/afr/src/afr-self-heald.h
@@ -0,0 +1,75 @@
+/*
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _AFR_SELF_HEALD_H
+#define _AFR_SELF_HEALD_H
+
+#include <pthread.h>
+
+typedef struct {
+    char *path;
+    int child;
+} shd_event_t;
+
+typedef struct {
+    uint64_t healed_count;
+    uint64_t split_brain_count;
+    uint64_t heal_failed_count;
+
+    /* If start_time is 0, it means crawler is not in progress
+       and stats are not valid */
+    time_t start_time;
+    /* If start_time is NOT 0 and end_time is 0, it means
+       cralwer is in progress */
+    time_t end_time;
+    char *crawl_type;
+    int child;
+} crawl_event_t;
+
+struct subvol_healer {
+    xlator_t *this;
+    crawl_event_t crawl_event;
+    pthread_mutex_t mutex;
+    pthread_cond_t cond;
+    pthread_t thread;
+    int subvol;
+    gf_boolean_t local;
+    gf_boolean_t running;
+    gf_boolean_t rerun;
+};
+
+typedef struct {
+    struct subvol_healer *index_healers;
+    struct subvol_healer *full_healers;
+
+    eh_t *split_brain;
+    eh_t **statistics;
+    int timeout;
+    uint32_t max_threads;
+    uint32_t wait_qlength;
+    uint32_t halo_max_latency_msec;
+    gf_boolean_t iamshd;
+    gf_boolean_t enabled;
+} afr_self_heald_t;
+
+int
+afr_selfheal_daemon_init(xlator_t *this);
+
+int
+afr_xl_op(xlator_t *this, dict_t *input, dict_t *output);
+
+int
+afr_shd_gfid_to_path(xlator_t *this, xlator_t *subvol, uuid_t gfid,
+                     char **path_p);
+
+int
+afr_shd_entry_purge(xlator_t *subvol, inode_t *inode, char *name,
+                    ia_type_t type);
+#endif /* !_AFR_SELF_HEALD_H */
diff --git a/xlators/cluster/afr/src/afr-transaction.c b/xlators/cluster/afr/src/afr-transaction.c
index 130f5bd2763..a51f79b1f43 100644
--- a/xlators/cluster/afr/src/afr-transaction.c
+++ b/xlators/cluster/afr/src/afr-transaction.c
@@ -1,1199 +1,2927 @@
 /*
-  Copyright (c) 2007-2009 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
-#include "dict.h"
-#include "byte-order.h"
-#include "common-utils.h"
+#include <glusterfs/dict.h>
+#include <glusterfs/byte-order.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/timer.h>
 
 #include "afr.h"
 #include "afr-transaction.h"
+#include "afr-self-heal.h"
+#include "afr-messages.h"
 
 #include <signal.h>
 
+typedef enum {
+    AFR_TRANSACTION_PRE_OP,
+    AFR_TRANSACTION_POST_OP,
+} afr_xattrop_type_t;
 
-#define LOCKED_NO       0x0        /* no lock held */
-#define LOCKED_YES      0x1        /* for DATA, METADATA, ENTRY and higher_path
-                                      of RENAME */
-#define LOCKED_LOWER    0x2        /* for lower_path of RENAME */
+static void
+afr_lock_resume_shared(struct list_head *list);
 
+static void
+afr_post_op_handle_success(call_frame_t *frame, xlator_t *this);
 
-afr_fd_ctx_t *
-afr_fd_ctx_get (fd_t *fd, xlator_t *this)
-{
-        uint64_t       ctx = 0;
-        afr_fd_ctx_t  *fd_ctx = NULL;
-        int            ret = 0;
+static void
+afr_post_op_handle_failure(call_frame_t *frame, xlator_t *this, int op_errno);
 
-        ret = fd_ctx_get (fd, this, &ctx);
+void
+__afr_transaction_wake_shared(afr_local_t *local, struct list_head *shared);
 
-        if (ret < 0)
-                goto out;
+void
+afr_changelog_post_op_do(call_frame_t *frame, xlator_t *this);
 
-        fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+int
+afr_changelog_post_op_safe(call_frame_t *frame, xlator_t *this);
 
-out:
-        return fd_ctx;
-}
+gf_boolean_t
+afr_changelog_pre_op_uninherit(call_frame_t *frame, xlator_t *this);
 
+gf_boolean_t
+afr_changelog_pre_op_update(call_frame_t *frame, xlator_t *this);
+
+int
+afr_changelog_call_count(afr_transaction_type type,
+                         unsigned char *pre_op_subvols,
+                         unsigned char *failed_subvols,
+                         unsigned int child_count);
+int
+afr_changelog_do(call_frame_t *frame, xlator_t *this, dict_t *xattr,
+                 afr_changelog_resume_t changelog_resume,
+                 afr_xattrop_type_t op);
 
 static void
-afr_pid_save (call_frame_t *frame)
-{
-        afr_local_t * local = NULL;
+afr_ta_decide_post_op_state(call_frame_t *frame, xlator_t *this);
 
-        local = frame->local;
+static int
+afr_ta_post_op_do(void *opaque);
 
-        local->saved_pid = frame->root->pid;
-}
+static int
+afr_ta_post_op_synctask(xlator_t *this, afr_local_t *local);
 
+static int
+afr_changelog_post_op_done(call_frame_t *frame, xlator_t *this);
 
 static void
-afr_pid_restore (call_frame_t *frame)
+afr_changelog_post_op_fail(call_frame_t *frame, xlator_t *this, int op_errno);
+
+void
+afr_ta_locked_priv_invalidate(afr_private_t *priv)
+{
+    priv->ta_bad_child_index = AFR_CHILD_UNKNOWN;
+    priv->release_ta_notify_dom_lock = _gf_false;
+    priv->ta_notify_dom_lock_offset = 0;
+}
+
+static void
+afr_ta_process_waitq(xlator_t *this)
 {
-        afr_local_t * local = NULL;
+    afr_local_t *entry = NULL;
+    afr_private_t *priv = this->private;
+    struct list_head waitq = {
+        0,
+    };
+
+    INIT_LIST_HEAD(&waitq);
+    LOCK(&priv->lock);
+    list_splice_init(&priv->ta_waitq, &waitq);
+    UNLOCK(&priv->lock);
+    list_for_each_entry(entry, &waitq, ta_waitq)
+    {
+        afr_ta_decide_post_op_state(entry->transaction.frame, this);
+    }
+}
 
-        local = frame->local;
+int
+afr_ta_lock_release_done(int ret, call_frame_t *ta_frame, void *opaque)
+{
+    afr_ta_process_waitq(ta_frame->this);
+    STACK_DESTROY(ta_frame->root);
+    return 0;
+}
 
-        frame->root->pid = local->saved_pid;
+int
+afr_release_notify_lock_for_ta(void *opaque)
+{
+    xlator_t *this = NULL;
+    afr_private_t *priv = NULL;
+    loc_t loc = {
+        0,
+    };
+    struct gf_flock flock = {
+        0,
+    };
+    int ret = -1;
+
+    this = (xlator_t *)opaque;
+    priv = this->private;
+    ret = afr_fill_ta_loc(this, &loc, _gf_true);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to populate loc for thin-arbiter.");
+        goto out;
+    }
+    flock.l_type = F_UNLCK;
+    flock.l_start = priv->ta_notify_dom_lock_offset;
+    flock.l_len = 1;
+    ret = syncop_inodelk(priv->children[THIN_ARBITER_BRICK_INDEX],
+                         AFR_TA_DOM_NOTIFY, &loc, F_SETLK, &flock, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to unlock AFR_TA_DOM_NOTIFY lock.");
+    }
+
+    LOCK(&priv->lock);
+    {
+        afr_ta_locked_priv_invalidate(priv);
+    }
+    UNLOCK(&priv->lock);
+out:
+    loc_wipe(&loc);
+    return ret;
 }
 
+void
+afr_zero_fill_stat(afr_local_t *local)
+{
+    if (!local)
+        return;
+    if (local->transaction.type == AFR_DATA_TRANSACTION ||
+        local->transaction.type == AFR_METADATA_TRANSACTION) {
+        gf_zero_fill_stat(&local->cont.inode_wfop.prebuf);
+        gf_zero_fill_stat(&local->cont.inode_wfop.postbuf);
+    } else if (local->transaction.type == AFR_ENTRY_TRANSACTION ||
+               local->transaction.type == AFR_ENTRY_RENAME_TRANSACTION) {
+        gf_zero_fill_stat(&local->cont.dir_fop.buf);
+        gf_zero_fill_stat(&local->cont.dir_fop.preparent);
+        gf_zero_fill_stat(&local->cont.dir_fop.postparent);
+        if (local->transaction.type == AFR_ENTRY_TRANSACTION)
+            return;
+        gf_zero_fill_stat(&local->cont.dir_fop.prenewparent);
+        gf_zero_fill_stat(&local->cont.dir_fop.postnewparent);
+    }
+}
 
-static void
-__mark_all_pending (int32_t *pending[], int child_count,
-                    afr_transaction_type type)
+/* In case of errors afr needs to choose which xdata from lower xlators it needs
+ * to unwind with. The way it is done is by checking if there are
+ * any good subvols which failed. Give preference to errnos other than
+ * ENOTCONN even if the child is source */
+void
+afr_pick_error_xdata(afr_local_t *local, afr_private_t *priv, inode_t *inode1,
+                     unsigned char *readable1, inode_t *inode2,
+                     unsigned char *readable2)
 {
-	int i;
-        int j;
+    int s = -1; /*selection*/
+    int i = 0;
+    unsigned char *readable = NULL;
+
+    if (local->xdata_rsp) {
+        dict_unref(local->xdata_rsp);
+        local->xdata_rsp = NULL;
+    }
+
+    readable = alloca0(priv->child_count * sizeof(*readable));
+    if (inode2 && readable2) { /*rename fop*/
+        AFR_INTERSECT(readable, readable1, readable2, priv->child_count);
+    } else {
+        memcpy(readable, readable1, sizeof(*readable) * priv->child_count);
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->replies[i].valid)
+            continue;
+
+        if (local->replies[i].op_ret >= 0)
+            continue;
+
+        if (local->replies[i].op_errno == ENOTCONN)
+            continue;
+
+        /*Order is important in the following condition*/
+        if ((s < 0) || (!readable[s] && readable[i]))
+            s = i;
+    }
+
+    if (s != -1 && local->replies[s].xdata) {
+        local->xdata_rsp = dict_ref(local->replies[s].xdata);
+    } else if (s == -1) {
+        for (i = 0; i < priv->child_count; i++) {
+            if (!local->replies[i].valid)
+                continue;
+
+            if (local->replies[i].op_ret >= 0)
+                continue;
 
-        for (i = 0; i < child_count; i++) {
-                j = afr_index_for_transaction_type (type);
-		pending[i][j] = hton32 (1);
+            if (!local->replies[i].xdata)
+                continue;
+            local->xdata_rsp = dict_ref(local->replies[i].xdata);
+            break;
         }
+    }
 }
 
+gf_boolean_t
+afr_needs_changelog_update(afr_local_t *local)
+{
+    if (local->transaction.type == AFR_DATA_TRANSACTION)
+        return _gf_true;
+    if (!local->optimistic_change_log)
+        return _gf_true;
+    return _gf_false;
+}
 
-static void
-__mark_child_dead (int32_t *pending[], int child_count, int child,
-                   afr_transaction_type type)
+gf_boolean_t
+afr_changelog_has_quorum(afr_local_t *local, xlator_t *this)
 {
-        int j;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    unsigned char *success_children = NULL;
 
-        j = afr_index_for_transaction_type (type);
+    priv = this->private;
+    success_children = alloca0(priv->child_count);
 
-	pending[child][j] = 0;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->transaction.failed_subvols[i]) {
+            success_children[i] = 1;
+        }
+    }
+
+    if (afr_has_quorum(success_children, this, NULL)) {
+        return _gf_true;
+    }
+
+    return _gf_false;
 }
 
+gf_boolean_t
+afr_is_write_subvol_valid(call_frame_t *frame, xlator_t *this)
+{
+    int i = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    uint64_t write_subvol = 0;
+    unsigned char *writable = NULL;
+    uint16_t datamap = 0;
+
+    local = frame->local;
+    priv = this->private;
+    writable = alloca0(priv->child_count);
+
+    write_subvol = afr_write_subvol_get(frame, this);
+    datamap = (write_subvol & 0x00000000ffff0000) >> 16;
+    for (i = 0; i < priv->child_count; i++) {
+        if (datamap & (1 << i))
+            writable[i] = 1;
+
+        if (writable[i] && !local->transaction.failed_subvols[i])
+            return _gf_true;
+    }
+
+    return _gf_false;
+}
 
-static void
-__mark_pre_op_done_on_fd (call_frame_t *frame, xlator_t *this, int child_index)
+int
+afr_transaction_fop(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t   *local = NULL;
-        afr_fd_ctx_t  *fd_ctx = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int call_count = -1;
+    unsigned char *failed_subvols = NULL;
+    int i = 0;
+
+    local = frame->local;
+    priv = this->private;
+
+    failed_subvols = local->transaction.failed_subvols;
+    call_count = priv->child_count -
+                 AFR_COUNT(failed_subvols, priv->child_count);
+    /* Fail if pre-op did not succeed on quorum no. of bricks. */
+    if (!afr_changelog_has_quorum(local, this) || !call_count) {
+        local->op_ret = -1;
+        /* local->op_errno is already captured in changelog cbk. */
+        afr_transaction_resume(frame, this);
+        return 0;
+    }
+
+    /* Fail if at least one writeable brick isn't up.*/
+    if (local->transaction.type == AFR_DATA_TRANSACTION &&
+        !afr_is_write_subvol_valid(frame, this)) {
+        local->op_ret = -1;
+        local->op_errno = EIO;
+        afr_transaction_resume(frame, this);
+        return 0;
+    }
 
-        local = frame->local;
+    local->call_count = call_count;
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.pre_op[i] && !failed_subvols[i]) {
+            local->transaction.wind(frame, this, i);
 
-        if (!local->fd)
-                return;
+            if (!--call_count)
+                break;
+        }
+    }
 
-        fd_ctx = afr_fd_ctx_get (local->fd, this);
+    return 0;
+}
 
-        if (!fd_ctx)
-                goto out;
+int
+afr_transaction_done(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    gf_boolean_t unwind = _gf_false;
+    afr_lock_t *lock = NULL;
+    afr_local_t *lock_local = NULL;
+
+    priv = this->private;
+    local = frame->local;
 
-        LOCK (&local->fd->lock);
+    if (priv->consistent_metadata) {
+        LOCK(&frame->lock);
         {
-                if (local->transaction.type == AFR_DATA_TRANSACTION)
-                        fd_ctx->pre_op_done[child_index]++;
+            unwind = (local->transaction.main_frame != NULL);
         }
-        UNLOCK (&local->fd->lock);
-out:
-        return;
+        UNLOCK(&frame->lock);
+        if (unwind) /*It definitely did post-op*/
+            afr_zero_fill_stat(local);
+    }
+
+    if (local->transaction.do_eager_unlock) {
+        lock = &local->inode_ctx->lock[local->transaction.type];
+        LOCK(&local->inode->lock);
+        {
+            lock->acquired = _gf_false;
+            lock->release = _gf_false;
+            list_splice_init(&lock->frozen, &lock->waiting);
+            if (list_empty(&lock->waiting))
+                goto unlock;
+            lock_local = list_entry(lock->waiting.next, afr_local_t,
+                                    transaction.wait_list);
+            list_del_init(&lock_local->transaction.wait_list);
+            list_add(&lock_local->transaction.owner_list, &lock->owners);
+        }
+    unlock:
+        UNLOCK(&local->inode->lock);
+    }
+    if (lock_local) {
+        afr_lock(lock_local->transaction.frame,
+                 lock_local->transaction.frame->this);
+    }
+    local->transaction.unwind(frame, this);
+
+    GF_ASSERT(list_empty(&local->transaction.owner_list));
+    GF_ASSERT(list_empty(&local->transaction.wait_list));
+    AFR_STACK_DESTROY(frame);
+
+    return 0;
 }
 
+static void
+afr_lock_fail_shared(afr_local_t *local, struct list_head *list)
+{
+    afr_local_t *each = NULL;
+
+    while (!list_empty(list)) {
+        each = list_entry(list->next, afr_local_t, transaction.wait_list);
+        list_del_init(&each->transaction.wait_list);
+        each->op_ret = -1;
+        each->op_errno = local->op_errno;
+        afr_transaction_done(each->transaction.frame,
+                             each->transaction.frame->this);
+    }
+}
 
 static void
-__mark_pre_op_undone_on_fd (call_frame_t *frame, xlator_t *this, int child_index)
+afr_handle_lock_acquire_failure(afr_local_t *local)
 {
-        afr_local_t   *local = NULL;
-        afr_fd_ctx_t  *fd_ctx = NULL;
+    struct list_head shared;
+    afr_lock_t *lock = NULL;
 
-        local = frame->local;
+    if (!local->transaction.eager_lock_on)
+        goto out;
 
-        if (!local->fd)
-                return;
+    lock = &local->inode_ctx->lock[local->transaction.type];
 
-        fd_ctx = afr_fd_ctx_get (local->fd, this);
+    INIT_LIST_HEAD(&shared);
+    LOCK(&local->inode->lock);
+    {
+        lock->release = _gf_true;
+        list_splice_init(&lock->waiting, &shared);
+    }
+    UNLOCK(&local->inode->lock);
 
-        if (!fd_ctx)
-                goto out;
-
-        LOCK (&local->fd->lock);
-        {
-                if (local->transaction.type == AFR_DATA_TRANSACTION)
-                        fd_ctx->pre_op_done[child_index]--;
-        }
-        UNLOCK (&local->fd->lock);
+    afr_lock_fail_shared(local, &shared);
+    local->transaction.do_eager_unlock = _gf_true;
 out:
-        return;
+    local->internal_lock.lock_cbk = afr_transaction_done;
+    afr_unlock(local->transaction.frame, local->transaction.frame->this);
 }
 
-
-static void
-__mark_down_children (int32_t *pending[], int child_count,
-                      unsigned char *child_up, afr_transaction_type type)
+call_frame_t *
+afr_transaction_detach_fop_frame(call_frame_t *frame)
 {
-	int i;
-	int j;
+    afr_local_t *local = NULL;
+    call_frame_t *fop_frame = NULL;
 
-	for (i = 0; i < child_count; i++) {
-                j = afr_index_for_transaction_type (type);
+    local = frame->local;
 
-		if (!child_up[i])
-			pending[i][j] = 0;
-        }
-}
+    afr_handle_inconsistent_fop(frame, &local->op_ret, &local->op_errno);
+    LOCK(&frame->lock);
+    {
+        fop_frame = local->transaction.main_frame;
+        local->transaction.main_frame = NULL;
+    }
+    UNLOCK(&frame->lock);
 
+    return fop_frame;
+}
 
 static void
-__mark_all_success (int32_t *pending[], int child_count,
-                    afr_transaction_type type)
+afr_save_lk_owner(call_frame_t *frame)
 {
-	int i;
-        int j;
+    afr_local_t *local = NULL;
 
-	for (i = 0; i < child_count; i++) {
-                j = afr_index_for_transaction_type (type);
-		pending[i][j] = hton32 (-1);
-        }
-}
+    local = frame->local;
 
+    local->saved_lk_owner = frame->root->lk_owner;
+}
 
-static int
-__changelog_enabled (afr_private_t *priv, afr_transaction_type type)
+static void
+afr_restore_lk_owner(call_frame_t *frame)
 {
-	int ret = 0;
-
-	switch (type) {
-	case AFR_DATA_TRANSACTION:
-		if (priv->data_change_log)
-			ret = 1;
+    afr_local_t *local = NULL;
 
-		break;
+    local = frame->local;
 
-	case AFR_METADATA_TRANSACTION:
-		if (priv->metadata_change_log)
-			ret = 1;
+    frame->root->lk_owner = local->saved_lk_owner;
+}
 
-		break;
+void
+__mark_all_success(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i;
 
-	case AFR_ENTRY_TRANSACTION:
-	case AFR_ENTRY_RENAME_TRANSACTION:
-		if (priv->entry_change_log)
-			ret = 1;
+    local = frame->local;
+    priv = this->private;
 
-		break;
-	}
+    for (i = 0; i < priv->child_count; i++) {
+        local->transaction.failed_subvols[i] = 0;
+    }
+}
 
-	return ret;
+void
+afr_compute_pre_op_sources(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    afr_transaction_type type = -1;
+    dict_t *xdata = NULL;
+    int **matrix = NULL;
+    int idx = -1;
+    int i = 0;
+    int j = 0;
+
+    priv = this->private;
+    local = frame->local;
+    type = local->transaction.type;
+    idx = afr_index_for_transaction_type(type);
+    matrix = ALLOC_MATRIX(priv->child_count, int);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->transaction.changelog_xdata[i])
+            continue;
+        xdata = local->transaction.changelog_xdata[i];
+        afr_selfheal_fill_matrix(this, matrix, i, idx, xdata);
+    }
+
+    memset(local->transaction.pre_op_sources, 1, priv->child_count);
+
+    /*If lock or pre-op failed on a brick, it is not a source. */
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.failed_subvols[i])
+            local->transaction.pre_op_sources[i] = 0;
+    }
+
+    /* If brick is blamed by others, it is not a source. */
+    for (i = 0; i < priv->child_count; i++)
+        for (j = 0; j < priv->child_count; j++)
+            if (matrix[i][j] != 0)
+                local->transaction.pre_op_sources[j] = 0;
 }
 
+void
+afr_txn_arbitrate_fop(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int pre_op_sources_count = 0;
+    int i = 0;
+
+    priv = this->private;
+    local = frame->local;
+
+    afr_compute_pre_op_sources(frame, this);
+    pre_op_sources_count = AFR_COUNT(local->transaction.pre_op_sources,
+                                     priv->child_count);
+
+    /* If arbiter is the only source, do not proceed. */
+    if (pre_op_sources_count < 2 &&
+        local->transaction.pre_op_sources[ARBITER_BRICK_INDEX]) {
+        local->op_ret = -1;
+        local->op_errno = ENOTCONN;
+        for (i = 0; i < priv->child_count; i++)
+            local->transaction.failed_subvols[i] = 1;
+    }
+
+    afr_transaction_fop(frame, this);
+
+    return;
+}
 
-static int
-__changelog_needed_pre_op (call_frame_t *frame, xlator_t *this)
+int
+afr_transaction_perform_fop(call_frame_t *frame, xlator_t *this)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    int ret = 0;
+    int failure_count = 0;
+    struct list_head shared;
+    afr_lock_t *lock = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    INIT_LIST_HEAD(&shared);
+    if (local->transaction.type == AFR_DATA_TRANSACTION &&
+        !local->transaction.inherited) {
+        ret = afr_write_subvol_set(frame, this);
+        if (ret) {
+            /*act as if operation failed on all subvols*/
+            local->op_ret = -1;
+            local->op_errno = -ret;
+            for (i = 0; i < priv->child_count; i++)
+                local->transaction.failed_subvols[i] = 1;
+        }
+    }
+
+    if (local->pre_op_compat)
+        /* old mode, pre-op was done as afr_changelog_do()
+           just now, before OP */
+        afr_changelog_pre_op_update(frame, this);
+
+    if (!local->transaction.eager_lock_on || local->transaction.inherited)
+        goto fop;
+    failure_count = AFR_COUNT(local->transaction.failed_subvols,
+                              priv->child_count);
+    if (failure_count == priv->child_count) {
+        afr_handle_lock_acquire_failure(local);
+        return 0;
+    } else {
+        lock = &local->inode_ctx->lock[local->transaction.type];
+        LOCK(&local->inode->lock);
+        {
+            lock->acquired = _gf_true;
+            __afr_transaction_wake_shared(local, &shared);
+        }
+        UNLOCK(&local->inode->lock);
+    }
+
+fop:
+    /*  Perform fops with the lk-owner from top xlator.
+     *  Eg: lk-owner of posix-lk and flush should be same,
+     *  flush cant clear the  posix-lks without that lk-owner.
+     */
+    afr_save_lk_owner(frame);
+    frame->root->lk_owner = local->transaction.main_frame->root->lk_owner;
+
+    if (priv->arbiter_count == 1) {
+        afr_txn_arbitrate_fop(frame, this);
+    } else {
+        afr_transaction_fop(frame, this);
+    }
+
+    afr_lock_resume_shared(&shared);
+    return 0;
+}
 
-	int op_ret   = 0;
+int
+afr_set_pending_dict(afr_private_t *priv, dict_t *xattr, int **pending)
+{
+    int i = 0;
+    int ret = 0;
 
-	priv  = this->private;
-	local = frame->local;
+    for (i = 0; i < priv->child_count; i++) {
+        ret = dict_set_static_bin(xattr, priv->pending_key[i], pending[i],
+                                  AFR_NUM_CHANGE_LOGS * sizeof(int));
+        /* 3 = data+metadata+entry */
 
-	if (__changelog_enabled (priv, local->transaction.type)) {
-		switch (local->op) {
+        if (ret)
+            break;
+    }
 
-		case GF_FOP_WRITE:
-		case GF_FOP_FTRUNCATE:
-                        op_ret = 1;
-			break;
+    return ret;
+}
 
-		case GF_FOP_FLUSH:
-			op_ret = 0;
-			break;
+static void
+afr_ta_dom_lock_check_and_release(afr_ta_fop_state_t fop_state, xlator_t *this)
+{
+    afr_private_t *priv = this->private;
+    unsigned int inmem_count = 0;
+    unsigned int onwire_count = 0;
+    gf_boolean_t release = _gf_false;
+
+    LOCK(&priv->lock);
+    {
+        /*Once we get notify lock release upcall notification,
+         if any of the fop state counters are non-zero, we will
+         not release the lock.
+         */
+        onwire_count = priv->ta_on_wire_txn_count;
+        inmem_count = priv->ta_in_mem_txn_count;
+        switch (fop_state) {
+            case TA_GET_INFO_FROM_TA_FILE:
+                onwire_count = --priv->ta_on_wire_txn_count;
+                break;
+            case TA_INFO_IN_MEMORY_SUCCESS:
+            case TA_INFO_IN_MEMORY_FAILED:
+                inmem_count = --priv->ta_in_mem_txn_count;
+                break;
+            case TA_WAIT_FOR_NOTIFY_LOCK_REL:
+                GF_ASSERT(0);
+                break;
+            case TA_SUCCESS:
+                break;
+        }
+        release = priv->release_ta_notify_dom_lock;
+    }
+    UNLOCK(&priv->lock);
 
-		default:
-			op_ret = 1;
-		}
-	}
+    if (inmem_count != 0 || release == _gf_false || onwire_count != 0)
+        return;
 
-	return op_ret;
+    afr_ta_lock_release_synctask(this);
 }
 
+static void
+afr_ta_process_onwireq(afr_ta_fop_state_t fop_state, xlator_t *this)
+{
+    afr_private_t *priv = this->private;
+    afr_local_t *entry = NULL;
+    int bad_child = AFR_CHILD_UNKNOWN;
+
+    struct list_head onwireq = {
+        0,
+    };
+    INIT_LIST_HEAD(&onwireq);
+
+    LOCK(&priv->lock);
+    {
+        bad_child = priv->ta_bad_child_index;
+        if (bad_child == AFR_CHILD_UNKNOWN) {
+            /*The previous on-wire ta_post_op was a failure. Just dequeue
+             *one element to wind on-wire again. */
+            entry = list_entry(priv->ta_onwireq.next, afr_local_t, ta_onwireq);
+            list_del_init(&entry->ta_onwireq);
+        } else {
+            /* Prepare to process all fops based on bad_child_index. */
+            list_splice_init(&priv->ta_onwireq, &onwireq);
+        }
+    }
+    UNLOCK(&priv->lock);
+
+    if (entry) {
+        afr_ta_post_op_synctask(this, entry);
+        return;
+    } else {
+        while (!list_empty(&onwireq)) {
+            entry = list_entry(onwireq.next, afr_local_t, ta_onwireq);
+            list_del_init(&entry->ta_onwireq);
+            if (entry->ta_failed_subvol == bad_child) {
+                afr_post_op_handle_success(entry->transaction.frame, this);
+            } else {
+                afr_post_op_handle_failure(entry->transaction.frame, this, EIO);
+            }
+        }
+    }
+}
 
-static int
-__changelog_needed_post_op (call_frame_t *frame, xlator_t *this)
+int
+afr_changelog_post_op_done(call_frame_t *frame, xlator_t *this)
 {
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
+    afr_local_t *local = NULL;
+    afr_internal_lock_t *int_lock = NULL;
+    afr_private_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+    int_lock = &local->internal_lock;
+
+    if (priv->thin_arbiter_count) {
+        /*fop should not come here with TA_WAIT_FOR_NOTIFY_LOCK_REL state */
+        afr_ta_dom_lock_check_and_release(local->fop_state, this);
+    }
+
+    /* Fail the FOP if post-op did not succeed on quorum no. of bricks. */
+    if (!afr_changelog_has_quorum(local, this)) {
+        local->op_ret = -1;
+        /*local->op_errno is already captured in changelog cbk*/
+    }
+
+    if (local->transaction.resume_stub) {
+        call_resume(local->transaction.resume_stub);
+        local->transaction.resume_stub = NULL;
+    }
+
+    int_lock->lock_cbk = afr_transaction_done;
+    afr_unlock(frame, this);
+
+    return 0;
+}
 
-	int op_ret = 0;
-	afr_transaction_type type = -1;
+static void
+afr_changelog_post_op_fail(call_frame_t *frame, xlator_t *this, int op_errno)
+{
+    afr_local_t *local = frame->local;
+    local->op_ret = -1;
+    local->op_errno = op_errno;
 
-	priv  = this->private;
-	local = frame->local;
-	type  = local->transaction.type;
+    gf_msg(this->name, GF_LOG_ERROR, op_errno, AFR_MSG_THIN_ARB,
+           "Failing %s for gfid %s. Fop state is:%d", gf_fop_list[local->op],
+           uuid_utoa(local->inode->gfid), local->fop_state);
 
-	if (__changelog_enabled (priv, type)) {
-                switch (local->op) {
+    afr_changelog_post_op_done(frame, this);
+}
 
-                case GF_FOP_WRITE:
-                case GF_FOP_FTRUNCATE:
-                        op_ret = 1;
-                        break;
+unsigned char *
+afr_locked_nodes_get(afr_transaction_type type, afr_internal_lock_t *int_lock)
+{
+    /*Because same set of subvols participate in all lockee
+     * entities*/
+    return int_lock->lockee[0].locked_nodes;
+}
 
-                case GF_FOP_FLUSH:
-                        op_ret = 0;
-                        break;
+int
+afr_changelog_call_count(afr_transaction_type type,
+                         unsigned char *pre_op_subvols,
+                         unsigned char *failed_subvols,
+                         unsigned int child_count)
+{
+    int i = 0;
+    int call_count = 0;
 
-                default:
-                        op_ret = 1;
-                }
+    for (i = 0; i < child_count; i++) {
+        if (pre_op_subvols[i] && !failed_subvols[i]) {
+            call_count++;
         }
+    }
 
-	return op_ret;
-}
+    if (type == AFR_ENTRY_RENAME_TRANSACTION)
+        call_count *= 2;
 
+    return call_count;
+}
 
-static int
-afr_set_pending_dict (afr_private_t *priv, dict_t *xattr, int32_t **pending)
+gf_boolean_t
+afr_txn_nothing_failed(call_frame_t *frame, xlator_t *this)
 {
-        int i;
-        int ret = 0;
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int i = 0;
+
+    local = frame->local;
+    priv = this->private;
+
+    if (priv->thin_arbiter_count) {
+        /* We need to perform post-op even if 1 data brick was down
+         * before the txn started.*/
+        if (AFR_COUNT(local->transaction.failed_subvols, priv->child_count))
+            return _gf_false;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.pre_op[i] &&
+            local->transaction.failed_subvols[i])
+            return _gf_false;
+    }
+
+    return _gf_true;
+}
 
-        for (i = 0; i < priv->child_count; i++) {
-                ret = dict_set_static_bin (xattr, priv->pending_key[i],
-                                           pending[i], 3 * sizeof (int32_t));
-                /* 3 = data+metadata+entry */
+void
+afr_handle_symmetric_errors(call_frame_t *frame, xlator_t *this)
+{
+    if (afr_is_symmetric_error(frame, this))
+        __mark_all_success(frame, this);
+}
 
-                if (ret < 0)
-                        goto out;
+gf_boolean_t
+afr_has_quorum(unsigned char *subvols, xlator_t *this, call_frame_t *frame)
+{
+    unsigned int quorum_count = 0;
+    afr_private_t *priv = NULL;
+    unsigned int up_children_count = 0;
+
+    priv = this->private;
+    up_children_count = AFR_COUNT(subvols, priv->child_count);
+
+    if (afr_lookup_has_quorum(frame, up_children_count))
+        return _gf_true;
+
+    if (priv->quorum_count == AFR_QUORUM_AUTO) {
+        /*
+         * Special case for auto-quorum with an even number of nodes.
+         *
+         * A replica set with even count N can only handle the same
+         * number of failures as odd N-1 before losing "vanilla"
+         * quorum, and the probability of more simultaneous failures is
+         * actually higher.  For example, with a 1% chance of failure
+         * we'd have a 0.03% chance of two simultaneous failures with
+         * N=3 but a 0.06% chance with N=4.  However, the special case
+         * is necessary for N=2 because there's no real quorum in that
+         * case (i.e. can't normally survive *any* failures).  In that
+         * case, we treat the first node as a tie-breaker, allowing
+         * quorum to be retained in some cases while still honoring the
+         * all-important constraint that there can not simultaneously
+         * be two partitioned sets of nodes each believing they have
+         * quorum.  Of two equally sized sets, the one without that
+         * first node will lose.
+         *
+         * It turns out that the special case is beneficial for higher
+         * values of N as well.  Continuing the example above, the
+         * probability of losing quorum with N=4 and this type of
+         * quorum is (very) slightly lower than with N=3 and vanilla
+         * quorum.  The difference becomes even more pronounced with
+         * higher N.  Therefore, even though such replica counts are
+         * unlikely to be seen in practice, we might as well use the
+         * "special" quorum then as well.
+         */
+        if ((up_children_count * 2) == priv->child_count) {
+            return subvols[0];
         }
+    }
 
-out:
-        return ret;
+    if (priv->quorum_count == AFR_QUORUM_AUTO) {
+        quorum_count = priv->child_count / 2 + 1;
+    } else {
+        quorum_count = priv->quorum_count;
+    }
+
+    if (up_children_count >= quorum_count)
+        return _gf_true;
+
+    return _gf_false;
 }
 
+static gf_boolean_t
+afr_has_fop_quorum(call_frame_t *frame)
+{
+    xlator_t *this = frame->this;
+    afr_local_t *local = frame->local;
+    unsigned char *locked_nodes = NULL;
 
-static int
-afr_set_piggyback_dict (afr_private_t *priv, dict_t *xattr, int32_t **pending,
-                        afr_transaction_type type)
+    locked_nodes = afr_locked_nodes_get(local->transaction.type,
+                                        &local->internal_lock);
+    return afr_has_quorum(locked_nodes, this, NULL);
+}
+
+static gf_boolean_t
+afr_has_fop_cbk_quorum(call_frame_t *frame)
 {
-        int i;
-        int ret = 0;
-        int *arr = NULL;
-        int index = 0;
+    afr_local_t *local = frame->local;
+    xlator_t *this = frame->this;
+    afr_private_t *priv = this->private;
+    unsigned char *success = alloca0(priv->child_count);
+    int i = 0;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.pre_op[i])
+            if (!local->transaction.failed_subvols[i])
+                success[i] = 1;
+    }
+
+    return afr_has_quorum(success, this, NULL);
+}
 
-        index = afr_index_for_transaction_type (type);
+gf_boolean_t
+afr_need_dirty_marking(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = this->private;
+    afr_local_t *local = NULL;
+    gf_boolean_t need_dirty = _gf_false;
 
-        for (i = 0; i < priv->child_count; i++) {
-                arr = GF_CALLOC (3 * sizeof (int32_t), priv->child_count,
-                                 gf_afr_mt_char);
-                if (!arr) {
-                        ret = -1;
-                        goto out;
-                }
+    local = frame->local;
 
-                memcpy (arr, pending[i], 3 * sizeof (int32_t));
+    if (!priv->quorum_count || !local->optimistic_change_log)
+        return _gf_false;
 
-                arr[index]++;
+    if (local->transaction.type == AFR_DATA_TRANSACTION ||
+        local->transaction.type == AFR_METADATA_TRANSACTION)
+        return _gf_false;
 
-                ret = dict_set_bin (xattr, priv->pending_key[i],
-                                    arr, 3 * sizeof (int32_t));
-                /* 3 = data+metadata+entry */
+    if (AFR_COUNT(local->transaction.failed_subvols, priv->child_count) ==
+        priv->child_count)
+        return _gf_false;
 
-                if (ret < 0)
-                        goto out;
-        }
+    if (!afr_has_fop_cbk_quorum(frame))
+        need_dirty = _gf_true;
 
-out:
-        return ret;
+    return need_dirty;
 }
 
-
-int
-afr_lock_server_count (afr_private_t *priv, afr_transaction_type type)
+void
+afr_handle_quorum(call_frame_t *frame, xlator_t *this)
 {
-	int ret = 0;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    const char *file = NULL;
+    uuid_t gfid = {0};
+
+    local = frame->local;
+    priv = frame->this->private;
 
-	switch (type) {
-	case AFR_DATA_TRANSACTION:
-		ret = priv->child_count;
-		break;
+    if (priv->quorum_count == 0)
+        return;
 
-	case AFR_METADATA_TRANSACTION:
-		ret = priv->child_count;
-		break;
+    /* If the fop already failed return right away to preserve errno */
+    if (local->op_ret == -1)
+        return;
 
-	case AFR_ENTRY_TRANSACTION:
-	case AFR_ENTRY_RENAME_TRANSACTION:
-		ret = priv->child_count;
-		break;
-	}
+    /*
+     * Network split may happen just after the fops are unwound, so check
+     * if the fop succeeded in a way it still follows quorum. If it doesn't,
+     * mark the fop as failure, mark the changelogs so it reflects that
+     * failure.
+     *
+     * Scenario:
+     * There are 3 mounts on 3 machines(node1, node2, node3) all writing to
+     * single file. Network split happened in a way that node1 can't see
+     * node2, node3. Node2, node3 both of them can't see node1. Now at the
+     * time of sending write all the bricks are up. Just after write fop is
+     * wound on node1, network split happens. Node1 thinks write fop failed
+     * on node2, node3 so marks pending changelog for those 2 extended
+     * attributes on node1. Node2, node3 thinks writes failed on node1 so
+     * they mark pending changelog for node1. When the network is stable
+     * again the file already is in split-brain. These checks prevent
+     * marking pending changelog on other subvolumes if the fop doesn't
+     * succeed in a way it is still following quorum. So with this fix what
+     * is happening is, node1 will have all pending changelog(FOOL) because
+     * the write succeeded only on node1 but failed on node2, node3 so
+     * instead of marking pending changelogs on node2, node3 it just treats
+     * the fop as failure and goes into DIRTY state. Where as node2, node3
+     * say they are sources and have pending changelog to node1 so there is
+     * no split-brain with the fix. The problem is eliminated completely.
+     */
+
+    if (afr_has_fop_cbk_quorum(frame))
+        return;
 
-	return ret;
+    if (afr_need_dirty_marking(frame, this))
+        goto set_response;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.pre_op[i])
+            afr_transaction_fop_failed(frame, frame->this, i);
+    }
+
+set_response:
+    local->op_ret = -1;
+    local->op_errno = afr_final_errno(local, priv);
+    if (local->op_errno == 0)
+        local->op_errno = afr_quorum_errno(priv);
+
+    if (local->fd) {
+        gf_uuid_copy(gfid, local->fd->inode->gfid);
+        file = uuid_utoa(gfid);
+    } else {
+        loc_path(&local->loc, local->loc.name);
+        file = local->loc.path;
+    }
+
+    gf_msg(frame->this->name, GF_LOG_WARNING, local->op_errno,
+           AFR_MSG_QUORUM_FAIL, "%s: Failing %s as quorum is not met", file,
+           gf_fop_list[local->op]);
+
+    switch (local->transaction.type) {
+        case AFR_ENTRY_TRANSACTION:
+        case AFR_ENTRY_RENAME_TRANSACTION:
+            afr_pick_error_xdata(local, priv, local->parent, local->readable,
+                                 local->parent2, local->readable2);
+            break;
+        default:
+            afr_pick_error_xdata(local, priv, local->inode, local->readable,
+                                 NULL, NULL);
+            break;
+    }
 }
 
-/* {{{ pending */
+int
+afr_fill_ta_loc(xlator_t *this, loc_t *loc, gf_boolean_t is_gfid_based_fop)
+{
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+    loc->parent = inode_ref(priv->root_inode);
+    gf_uuid_copy(loc->pargfid, loc->parent->gfid);
+    loc->name = priv->pending_key[THIN_ARBITER_BRICK_INDEX];
+    if (is_gfid_based_fop && gf_uuid_is_null(priv->ta_gfid)) {
+        /* Except afr_ta_id_file_check() which is path based, all other gluster
+         * FOPS need gfid.*/
+        return -EINVAL;
+    }
+    gf_uuid_copy(loc->gfid, priv->ta_gfid);
+    loc->inode = inode_new(loc->parent->table);
+    if (!loc->inode) {
+        loc_wipe(loc);
+        return -ENOMEM;
+    }
+    return 0;
+}
 
-int32_t
-afr_changelog_post_op_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			   int32_t op_ret, int32_t op_errno, dict_t *xattr)
+static int
+afr_ta_post_op_done(int ret, call_frame_t *frame, void *opaque)
 {
-        afr_internal_lock_t *int_lock = NULL;
-	afr_private_t       *priv     = NULL;
-	afr_local_t         *local    = NULL;
-        int                  child_index = 0;
+    xlator_t *this = NULL;
+    afr_local_t *local = NULL;
+    call_frame_t *txn_frame = NULL;
+    afr_ta_fop_state_t fop_state;
+
+    local = (afr_local_t *)opaque;
+    fop_state = local->fop_state;
+    txn_frame = local->transaction.frame;
+    this = frame->this;
+
+    if (ret == 0) {
+        /*Mark pending xattrs on the up data brick.*/
+        afr_post_op_handle_success(txn_frame, this);
+    } else {
+        afr_post_op_handle_failure(txn_frame, this, -ret);
+    }
+
+    STACK_DESTROY(frame->root);
+    afr_ta_process_onwireq(fop_state, this);
+
+    return 0;
+}
 
-	int call_count = -1;
+int **
+afr_set_changelog_xattr(afr_private_t *priv, unsigned char *pending,
+                        dict_t *xattr, afr_local_t *local)
+{
+    int **changelog = NULL;
+    int idx = 0;
+    int ret = 0;
+    int i;
+
+    if (local->is_new_entry == _gf_true) {
+        changelog = afr_mark_pending_changelog(priv, pending, xattr,
+                                               local->cont.dir_fop.buf.ia_type);
+    } else {
+        idx = afr_index_for_transaction_type(local->transaction.type);
+        changelog = afr_matrix_create(priv->child_count, AFR_NUM_CHANGE_LOGS);
+        if (!changelog) {
+            goto out;
+        }
+        for (i = 0; i < priv->child_count; i++) {
+            if (local->transaction.failed_subvols[i])
+                changelog[i][idx] = hton32(1);
+        }
+        ret = afr_set_pending_dict(priv, xattr, changelog);
+        if (ret < 0) {
+            afr_matrix_cleanup(changelog, priv->child_count);
+            return NULL;
+        }
+    }
 
-	priv     = this->private;
-	local    = frame->local;
-        int_lock = &local->internal_lock;
+out:
+    return changelog;
+}
 
-        child_index = (long) cookie;
+static void
+afr_ta_locked_xattrop_validate(afr_private_t *priv, afr_local_t *local,
+                               gf_boolean_t *valid)
+{
+    if (priv->ta_event_gen > local->ta_event_gen) {
+        /* We can't trust the ta's response anymore.*/
+        afr_ta_locked_priv_invalidate(priv);
+        *valid = _gf_false;
+        return;
+    }
+    return;
+}
 
-        if (op_ret == 1) {
+static int
+afr_ta_post_op_do(void *opaque)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    xlator_t *this = NULL;
+    dict_t *xattr = NULL;
+    unsigned char *pending = NULL;
+    int **changelog = NULL;
+    int failed_subvol = -1;
+    int success_subvol = -1;
+    loc_t loc = {
+        0,
+    };
+    int i = 0;
+    int ret = 0;
+    gf_boolean_t valid = _gf_true;
+
+    local = (afr_local_t *)opaque;
+    this = local->transaction.frame->this;
+    priv = this->private;
+
+    ret = afr_fill_ta_loc(this, &loc, _gf_true);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Failed to populate loc for thin-arbiter.");
+        goto out;
+    }
+
+    xattr = dict_new();
+    if (!xattr) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    pending = alloca0(priv->child_count);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.failed_subvols[i]) {
+            pending[i] = 1;
+            failed_subvol = i;
+        } else {
+            success_subvol = i;
         }
-
-        if (op_ret == 0) {
-                __mark_pre_op_undone_on_fd (frame, this, child_index);
+    }
+
+    changelog = afr_set_changelog_xattr(priv, pending, xattr, local);
+
+    if (!changelog) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ret = afr_ta_post_op_lock(this, &loc);
+    if (ret)
+        goto out;
+
+    ret = syncop_xattrop(priv->children[THIN_ARBITER_BRICK_INDEX], &loc,
+                         GF_XATTROP_ADD_ARRAY, xattr, NULL, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, AFR_MSG_THIN_ARB,
+               "Post-op on thin-arbiter id file %s failed for gfid %s.",
+               priv->pending_key[THIN_ARBITER_BRICK_INDEX],
+               uuid_utoa(local->inode->gfid));
+    }
+    LOCK(&priv->lock);
+    {
+        if (ret == 0) {
+            priv->ta_bad_child_index = failed_subvol;
+        } else if (ret == -EINVAL) {
+            priv->ta_bad_child_index = success_subvol;
+            ret = -EIO; /* TA failed the fop. Return EIO to application. */
         }
 
-	LOCK (&frame->lock);
-	{
-		call_count = --local->call_count;
-	}
-	UNLOCK (&frame->lock);
+        afr_ta_locked_xattrop_validate(priv, local, &valid);
+    }
+    UNLOCK(&priv->lock);
+    if (valid == _gf_false) {
+        gf_msg(this->name, GF_LOG_ERROR, EIO, AFR_MSG_THIN_ARB,
+               "Post-op on thin-arbiter id file %s for gfid %s invalidated due "
+               "to event-gen mismatch.",
+               priv->pending_key[THIN_ARBITER_BRICK_INDEX],
+               uuid_utoa(local->inode->gfid));
+        ret = -EIO;
+    }
+
+    afr_ta_post_op_unlock(this, &loc);
+out:
+    if (xattr)
+        dict_unref(xattr);
 
-	if (call_count == 0) {
-                if (afr_lock_server_count (priv, local->transaction.type) == 0) {
-                        local->transaction.done (frame, this);
-                } else {
-                        int_lock->lock_cbk = local->transaction.done;
-                        afr_unlock (frame, this);
-                }
-	}
+    if (changelog)
+        afr_matrix_cleanup(changelog, priv->child_count);
 
-	return 0;
-}
+    loc_wipe(&loc);
 
+    return ret;
+}
 
-void
-afr_update_read_child (call_frame_t *frame, xlator_t *this, inode_t *inode,
-                       afr_transaction_type type)
+static int
+afr_ta_post_op_synctask(xlator_t *this, afr_local_t *local)
 {
-        int           curr_read_child = -1;
-        int           new_read_child = -1;
-        afr_private_t   *priv = NULL;
-        afr_local_t  *local = NULL;
-        int         **pending = NULL;
-        int           idx = 0;
+    call_frame_t *ta_frame = NULL;
+    int ret = 0;
+
+    ta_frame = afr_ta_frame_create(this);
+    if (!ta_frame) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_THIN_ARB,
+               "Failed to create ta_frame");
+        goto err;
+    }
+    ret = synctask_new(this->ctx->env, afr_ta_post_op_do, afr_ta_post_op_done,
+                       ta_frame, local);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, AFR_MSG_THIN_ARB,
+               "Failed to launch post-op on thin arbiter for gfid %s",
+               uuid_utoa(local->inode->gfid));
+        STACK_DESTROY(ta_frame->root);
+        goto err;
+    }
+
+    return ret;
+err:
+    afr_changelog_post_op_fail(local->transaction.frame, this, ENOMEM);
+    return ret;
+}
 
-        idx = afr_index_for_transaction_type (type);
+static void
+afr_ta_set_fop_state(afr_private_t *priv, afr_local_t *local,
+                     int *on_wire_count)
+{
+    LOCK(&priv->lock);
+    {
+        if (priv->release_ta_notify_dom_lock == _gf_true) {
+            /* Put the fop in waitq until notify dom lock is released.*/
+            local->fop_state = TA_WAIT_FOR_NOTIFY_LOCK_REL;
+            list_add_tail(&local->ta_waitq, &priv->ta_waitq);
+        } else if (priv->ta_bad_child_index == AFR_CHILD_UNKNOWN) {
+            /* Post-op on thin-arbiter to decide success/failure. */
+            local->fop_state = TA_GET_INFO_FROM_TA_FILE;
+            *on_wire_count = ++priv->ta_on_wire_txn_count;
+            if (*on_wire_count > 1) {
+                /*Avoid sending multiple on-wire post-ops on TA*/
+                list_add_tail(&local->ta_onwireq, &priv->ta_onwireq);
+            }
+        } else if (local->ta_failed_subvol == priv->ta_bad_child_index) {
+            /* Post-op on TA not needed as the fop failed on the in-memory bad
+             * brick. Just mark pending xattrs on the good data brick.*/
+            local->fop_state = TA_INFO_IN_MEMORY_SUCCESS;
+            priv->ta_in_mem_txn_count++;
+        } else {
+            /* Post-op on TA not needed as the fop succeeded only on the
+             * in-memory bad data brick and not the good one. Fail the fop.*/
+            local->fop_state = TA_INFO_IN_MEMORY_FAILED;
+            priv->ta_in_mem_txn_count++;
+        }
+    }
+    UNLOCK(&priv->lock);
+}
 
-        priv = this->private;
-        local = frame->local;
-        curr_read_child = afr_read_child (this, inode);
-        pending = local->pending;
+static void
+afr_ta_fill_failed_subvol(afr_private_t *priv, afr_local_t *local)
+{
+    int i = 0;
 
-        if (pending[curr_read_child][idx] != 0)
-                return;
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.failed_subvols[i]) {
+            local->ta_failed_subvol = i;
+            break;
+        }
+    }
+}
 
-        /* need to set new read_child */
-        for (new_read_child = 0; new_read_child < priv->child_count;
-             new_read_child++) {
+static void
+afr_post_op_handle_success(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
 
-                if (!priv->child_up[new_read_child])
-                        /* child is down */
-                        continue;
+    local = frame->local;
+    if (local->is_new_entry == _gf_true) {
+        afr_mark_new_entry_changelog(frame, this);
+    }
+    afr_changelog_post_op_do(frame, this);
 
-                if (pending[new_read_child][idx] == 0)
-                        /* op just failed */
-                        continue;
+    return;
+}
 
-                break;
-        }
+static void
+afr_post_op_handle_failure(call_frame_t *frame, xlator_t *this, int op_errno)
+{
+    afr_changelog_post_op_fail(frame, this, op_errno);
 
-        if (new_read_child == priv->child_count)
-                /* all children uneligible. leave as-is */
-                return;
+    return;
+}
 
-        afr_set_read_child (this, inode, new_read_child);
+static void
+afr_ta_decide_post_op_state(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int on_wire_count = 0;
+
+    priv = this->private;
+    local = frame->local;
+
+    afr_ta_set_fop_state(priv, local, &on_wire_count);
+
+    switch (local->fop_state) {
+        case TA_GET_INFO_FROM_TA_FILE:
+            if (on_wire_count == 1)
+                afr_ta_post_op_synctask(this, local);
+            /*else, fop is queued in ta_onwireq.*/
+            break;
+        case TA_WAIT_FOR_NOTIFY_LOCK_REL:
+            /*Post releasing the notify lock, we will act on this queue*/
+            break;
+        case TA_INFO_IN_MEMORY_SUCCESS:
+            afr_post_op_handle_success(frame, this);
+            break;
+        case TA_INFO_IN_MEMORY_FAILED:
+            afr_post_op_handle_failure(frame, this, EIO);
+            break;
+        default:
+            break;
+    }
+    return;
 }
 
+static void
+afr_handle_failure_using_thin_arbiter(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = this->private;
+    afr_local_t *local = frame->local;
+
+    afr_ta_fill_failed_subvol(priv, local);
+    gf_msg_debug(this->name, 0,
+                 "Fop failed on data brick (%s) for gfid=%s. "
+                 "ta info needed to decide fop result.",
+                 priv->children[local->ta_failed_subvol]->name,
+                 uuid_utoa(local->inode->gfid));
+    afr_ta_decide_post_op_state(frame, this);
+}
 
-int 
-afr_changelog_post_op (call_frame_t *frame, xlator_t *this)
+void
+afr_changelog_post_op_do(call_frame_t *frame, xlator_t *this)
 {
-	afr_private_t * priv = this->private;
-        afr_internal_lock_t *int_lock = NULL;
-	int ret        = 0;
-	int i          = 0;
-	int call_count = 0;
+    afr_private_t *priv = this->private;
+    afr_local_t *local = NULL;
+    dict_t *xattr = NULL;
+    int i = 0;
+    int ret = 0;
+    int idx = 0;
+    int nothing_failed = 1;
+    gf_boolean_t need_undirty = _gf_false;
+
+    afr_handle_quorum(frame, this);
+    local = frame->local;
+    idx = afr_index_for_transaction_type(local->transaction.type);
+
+    xattr = dict_new();
+    if (!xattr) {
+        afr_changelog_post_op_fail(frame, this, ENOMEM);
+        goto out;
+    }
+
+    nothing_failed = afr_txn_nothing_failed(frame, this);
+
+    if (afr_changelog_pre_op_uninherit(frame, this))
+        need_undirty = _gf_false;
+    else
+        need_undirty = _gf_true;
+
+    if (local->op_ret < 0 && !nothing_failed) {
+        if (afr_need_dirty_marking(frame, this)) {
+            local->dirty[idx] = hton32(1);
+            goto set_dirty;
+        }
 
-	afr_local_t *  local = NULL;
-        afr_fd_ctx_t  *fdctx = NULL;
-	dict_t        **xattr = NULL;
-        int            piggyback = 0;
-        int            index = 0;
-        int            nothing_failed = 1;
+        afr_changelog_post_op_done(frame, this);
+        goto out;
+    }
+
+    if (nothing_failed && !need_undirty) {
+        afr_changelog_post_op_done(frame, this);
+        goto out;
+    }
+
+    if (local->transaction.in_flight_sb) {
+        afr_changelog_post_op_fail(frame, this,
+                                   local->transaction.in_flight_sb_errno);
+        goto out;
+    }
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.failed_subvols[i])
+            local->pending[i][idx] = hton32(1);
+    }
+
+    ret = afr_set_pending_dict(priv, xattr, local->pending);
+    if (ret < 0) {
+        afr_changelog_post_op_fail(frame, this, ENOMEM);
+        goto out;
+    }
+
+    if (need_undirty)
+        local->dirty[idx] = hton32(-1);
+    else
+        local->dirty[idx] = hton32(0);
+
+set_dirty:
+    ret = dict_set_static_bin(xattr, AFR_DIRTY, local->dirty,
+                              sizeof(int) * AFR_NUM_CHANGE_LOGS);
+    if (ret) {
+        afr_changelog_post_op_fail(frame, this, ENOMEM);
+        goto out;
+    }
+
+    afr_changelog_do(frame, this, xattr, afr_changelog_post_op_done,
+                     AFR_TRANSACTION_POST_OP);
+out:
+    if (xattr)
+        dict_unref(xattr);
 
-	local    = frame->local;
-        int_lock = &local->internal_lock;
+    return;
+}
 
-	__mark_down_children (local->pending, priv->child_count,
-                              local->child_up, local->transaction.type);
+static int
+afr_changelog_post_op_now(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
+    int failed_count = 0;
+
+    priv = this->private;
+    local = frame->local;
+
+    if (priv->thin_arbiter_count) {
+        failed_count = AFR_COUNT(local->transaction.failed_subvols,
+                                 priv->child_count);
+        if (failed_count == 1) {
+            afr_handle_failure_using_thin_arbiter(frame, this);
+            return 0;
+        } else {
+            /* Txn either succeeded or failed on both data bricks. Let
+             * post_op_do handle it as the case might be. */
+        }
+    }
 
-        if (local->fd)
-                afr_update_read_child (frame, this, local->fd->inode,
-                                       local->transaction.type);
+    afr_changelog_post_op_do(frame, this);
+    return 0;
+}
 
-        xattr = alloca (priv->child_count * sizeof (*xattr));
-        memset (xattr, 0, (priv->child_count * sizeof (*xattr)));
-	for (i = 0; i < priv->child_count; i++) {
-                xattr[i] = get_new_dict ();
-                dict_ref (xattr[i]);
+gf_boolean_t
+afr_changelog_pre_op_uninherit(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    afr_inode_ctx_t *ctx = NULL;
+    int i = 0;
+    gf_boolean_t ret = _gf_false;
+    int type = 0;
+
+    local = frame->local;
+    priv = this->private;
+    ctx = local->inode_ctx;
+
+    type = afr_index_for_transaction_type(local->transaction.type);
+    if (type != AFR_DATA_TRANSACTION)
+        return !local->transaction.dirtied;
+
+    if (local->transaction.no_uninherit)
+        return _gf_false;
+
+    /* This function must be idempotent. So check if we
+       were called before and return the same answer again.
+
+       It is important to keep this function idempotent for
+       the call in afr_changelog_post_op_safe() to not have
+       side effects on the call from afr_changelog_post_op_now()
+    */
+    if (local->transaction.uninherit_done)
+        return local->transaction.uninherit_value;
+
+    LOCK(&local->inode->lock);
+    {
+        for (i = 0; i < priv->child_count; i++) {
+            if (local->transaction.pre_op[i] != ctx->pre_op_done[type][i]) {
+                ret = !local->transaction.dirtied;
+                goto unlock;
+            }
         }
 
-        call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-        if (local->transaction.type == AFR_ENTRY_RENAME_TRANSACTION) {
-                call_count *= 2;
+        if (ctx->inherited[type]) {
+            ret = _gf_true;
+            ctx->inherited[type]--;
+        } else if (ctx->on_disk[type]) {
+            ret = _gf_false;
+            ctx->on_disk[type]--;
+        } else {
+            /* ASSERT */
+            ret = _gf_false;
         }
 
-	local->call_count = call_count;
+        if (!ctx->inherited[type] && !ctx->on_disk[type]) {
+            for (i = 0; i < priv->child_count; i++)
+                ctx->pre_op_done[type][i] = 0;
+        }
+    }
+unlock:
+    UNLOCK(&local->inode->lock);
 
-        if (local->fd)
-                fdctx = afr_fd_ctx_get (local->fd, this);
+    local->transaction.uninherit_done = _gf_true;
+    local->transaction.uninherit_value = ret;
 
-	if (call_count == 0) {
-		/* no child is up */
-                for (i = 0; i < priv->child_count; i++) {
-                        dict_unref (xattr[i]);
-                }
+    return ret;
+}
 
-                int_lock->lock_cbk = local->transaction.done;
-		afr_unlock (frame, this);
-		return 0;
-	}
+gf_boolean_t
+afr_changelog_pre_op_inherit(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    gf_boolean_t ret = _gf_false;
+    int type = 0;
+
+    local = frame->local;
+    priv = this->private;
+
+    if (local->transaction.type != AFR_DATA_TRANSACTION)
+        return _gf_false;
+
+    type = afr_index_for_transaction_type(local->transaction.type);
+
+    LOCK(&local->inode->lock);
+    {
+        if (!local->inode_ctx->on_disk[type]) {
+            /* nothing to inherit yet */
+            ret = _gf_false;
+            goto unlock;
+        }
 
-        /* check if something has failed, to handle piggybacking */
-        nothing_failed = 1;
-        index = afr_index_for_transaction_type (local->transaction.type);
         for (i = 0; i < priv->child_count; i++) {
-                if (local->pending[i][index] == 0) {
-                        nothing_failed = 0;
-                        break;
-                }
+            if (local->transaction.pre_op[i] !=
+                local->inode_ctx->pre_op_done[type][i]) {
+                /* either inherit exactly, or don't */
+                ret = _gf_false;
+                goto unlock;
+            }
         }
 
-	for (i = 0; i < priv->child_count; i++) {
-		if (!local->child_up[i])
-                        continue;
+        local->inode_ctx->inherited[type]++;
 
-                ret = afr_set_pending_dict (priv, xattr[i],
-                                            local->pending);
-
-                if (ret < 0)
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "failed to set pending entry");
-
-
-                switch (local->transaction.type) {
-                case AFR_DATA_TRANSACTION:
-                {
-                        if (!fdctx) {
-                                STACK_WIND (frame, afr_changelog_post_op_cbk,
-                                            priv->children[i],
-                                            priv->children[i]->fops->xattrop,
-                                            &local->loc,
-                                            GF_XATTROP_ADD_ARRAY, xattr[i]);
-                                break;
-                        }
-
-                        LOCK (&local->fd->lock);
-                        {
-                                piggyback = 0;
-                                if (fdctx->pre_op_piggyback[i]) {
-                                        fdctx->pre_op_piggyback[i]--;
-                                        piggyback = 1;
-                                }
-                        }
-                        UNLOCK (&local->fd->lock);
-
-                        if (piggyback && !nothing_failed)
-                                ret = afr_set_piggyback_dict (priv, xattr[i],
-                                                              local->pending,
-                                                              local->transaction.type);
-
-                        if (nothing_failed && piggyback) {
-                                afr_changelog_post_op_cbk (frame, (void *)(long)i,
-                                                          this, 1, 0, xattr[i]);
-                        } else {
-                                STACK_WIND_COOKIE (frame,
-                                                   afr_changelog_post_op_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->fxattrop,
-                                                   local->fd,
-                                                   GF_XATTROP_ADD_ARRAY, xattr[i]);
-                        }
-                }
-                break;
-                case AFR_METADATA_TRANSACTION:
-                {
-                        if (local->fd)
-                                STACK_WIND (frame, afr_changelog_post_op_cbk,
-                                            priv->children[i],
-                                            priv->children[i]->fops->fxattrop,
-                                            local->fd,
-                                            GF_XATTROP_ADD_ARRAY, xattr[i]);
-                        else
-                                STACK_WIND (frame, afr_changelog_post_op_cbk,
-                                            priv->children[i],
-                                            priv->children[i]->fops->xattrop,
-                                            &local->loc,
-                                            GF_XATTROP_ADD_ARRAY, xattr[i]);
-                }
-                break;
+        ret = _gf_true;
 
-                case AFR_ENTRY_RENAME_TRANSACTION:
-                {
-                        STACK_WIND_COOKIE (frame, afr_changelog_post_op_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->xattrop,
-                                           &local->transaction.new_parent_loc,
-                                           GF_XATTROP_ADD_ARRAY, xattr[i]);
-                        call_count--;
-                }
+        local->transaction.inherited = _gf_true;
+    }
+unlock:
+    UNLOCK(&local->inode->lock);
 
-                /*
-                  set it again because previous stack_wind
-                  might have already returned (think of case
-                  where subvolume is posix) and would have
-                  used the dict as placeholder for return
-                  value
-                */
+    return ret;
+}
 
-                ret = afr_set_pending_dict (priv, xattr[i],
-                                            local->pending);
+gf_boolean_t
+afr_changelog_pre_op_update(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+    gf_boolean_t ret = _gf_false;
+    int type = 0;
 
-                if (ret < 0)
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "failed to set pending entry");
+    local = frame->local;
+    priv = this->private;
 
-                /* fall through */
+    if (local->transaction.type == AFR_ENTRY_TRANSACTION ||
+        local->transaction.type == AFR_ENTRY_RENAME_TRANSACTION)
+        return _gf_false;
 
-                case AFR_ENTRY_TRANSACTION:
-                {
-                        if (local->fd)
-                                STACK_WIND (frame, afr_changelog_post_op_cbk,
-                                            priv->children[i],
-                                            priv->children[i]->fops->fxattrop,
-                                            local->fd,
-                                            GF_XATTROP_ADD_ARRAY, xattr[i]);
-                        else
-                                STACK_WIND (frame, afr_changelog_post_op_cbk,
-                                            priv->children[i],
-                                            priv->children[i]->fops->xattrop,
-                                            &local->transaction.parent_loc,
-                                            GF_XATTROP_ADD_ARRAY, xattr[i]);
-                }
-                break;
-                }
+    if (local->transaction.inherited)
+        /* was already inherited in afr_changelog_pre_op */
+        return _gf_false;
 
-                if (!--call_count)
-                        break;
-	}
+    if (!local->transaction.dirtied)
+        return _gf_false;
 
-        for (i = 0; i < priv->child_count; i++) {
-                dict_unref (xattr[i]);
+    if (!afr_txn_nothing_failed(frame, this))
+        return _gf_false;
+
+    type = afr_index_for_transaction_type(local->transaction.type);
+
+    ret = _gf_false;
+
+    LOCK(&local->inode->lock);
+    {
+        if (!local->inode_ctx->on_disk[type]) {
+            for (i = 0; i < priv->child_count; i++)
+                local->inode_ctx->pre_op_done[type][i] =
+                    (!local->transaction.failed_subvols[i]);
+        } else {
+            for (i = 0; i < priv->child_count; i++)
+                if (local->inode_ctx->pre_op_done[type][i] !=
+                    (!local->transaction.failed_subvols[i])) {
+                    local->transaction.no_uninherit = 1;
+                    goto unlock;
+                }
         }
+        local->inode_ctx->on_disk[type]++;
 
-	return 0;
-}
+        ret = _gf_true;
+    }
+unlock:
+    UNLOCK(&local->inode->lock);
 
+    return ret;
+}
 
-int32_t
-afr_changelog_pre_op_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t op_ret, int32_t op_errno, dict_t *xattr)
+int
+afr_changelog_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                  int op_errno, dict_t *xattr, dict_t *xdata)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = this->private;
-	loc_t       *   loc   = NULL;
+    afr_local_t *local = NULL;
+    int call_count = -1;
+    int child_index = -1;
 
-	int call_count  = -1;
-	int child_index = (long) cookie;
+    local = frame->local;
+    child_index = (long)cookie;
 
-	local = frame->local;
-	loc   = &local->loc;
+    if (op_ret == -1) {
+        local->op_errno = op_errno;
+        afr_transaction_fop_failed(frame, this, child_index);
+    }
 
-	LOCK (&frame->lock);
-	{
-                if (op_ret == 1) {
-                        /* special op_ret for piggyback */
-                }
+    if (xattr)
+        local->transaction.changelog_xdata[child_index] = dict_ref(xattr);
+
+    call_count = afr_frame_return(frame);
+
+    if (call_count == 0) {
+        local->transaction.changelog_resume(frame, this);
+    }
 
-                if (op_ret == 0) {
-                        __mark_pre_op_done_on_fd (frame, this, child_index);
+    return 0;
+}
+
+void
+afr_changelog_populate_xdata(call_frame_t *frame, afr_xattrop_type_t op,
+                             dict_t **xdata, dict_t **newloc_xdata)
+{
+    int i = 0;
+    int ret = 0;
+    char *key = NULL;
+    int keylen = 0;
+    const char *name = NULL;
+    dict_t *xdata1 = NULL;
+    dict_t *xdata2 = NULL;
+    xlator_t *this = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    gf_boolean_t need_entry_key_set = _gf_true;
+
+    local = frame->local;
+    this = THIS;
+    priv = this->private;
+
+    if (local->transaction.type == AFR_DATA_TRANSACTION ||
+        local->transaction.type == AFR_METADATA_TRANSACTION)
+        goto out;
+
+    if (!priv->esh_granular)
+        goto out;
+
+    xdata1 = dict_new();
+    if (!xdata1)
+        goto out;
+
+    name = local->loc.name;
+    if (local->op == GF_FOP_LINK)
+        name = local->newloc.name;
+
+    switch (op) {
+        case AFR_TRANSACTION_PRE_OP:
+            key = GF_XATTROP_ENTRY_IN_KEY;
+            break;
+        case AFR_TRANSACTION_POST_OP:
+            if (afr_txn_nothing_failed(frame, this)) {
+                key = GF_XATTROP_ENTRY_OUT_KEY;
+                for (i = 0; i < priv->child_count; i++) {
+                    if (!local->transaction.failed_subvols[i])
+                        continue;
+                    need_entry_key_set = _gf_false;
+                    break;
                 }
+                /* If the transaction itself did not fail and there
+                 * are no failed subvolumes, check whether the fop
+                 * failed due to a symmetric error. If it did, do
+                 * not set the ENTRY_OUT xattr which would end up
+                 * deleting a name index which was created possibly by
+                 * an earlier entry txn that may have failed on some
+                 * of the sub-volumes.
+                 */
+                if (local->op_ret)
+                    need_entry_key_set = _gf_false;
+            } else {
+                key = GF_XATTROP_ENTRY_IN_KEY;
+            }
+            break;
+    }
+
+    if (need_entry_key_set) {
+        keylen = strlen(key);
+        ret = dict_set_strn(xdata1, key, keylen, (char *)name);
+        if (ret)
+            gf_msg(THIS->name, GF_LOG_ERROR, 0, AFR_MSG_DICT_SET_FAILED,
+                   "%s/%s: Could not set %s key during xattrop",
+                   uuid_utoa(local->loc.pargfid), local->loc.name, key);
+        if (local->transaction.type == AFR_ENTRY_RENAME_TRANSACTION) {
+            xdata2 = dict_new();
+            if (!xdata2)
+                goto out;
 
-		if (op_ret == -1) {
-			local->child_up[child_index] = 0;
+            ret = dict_set_strn(xdata2, key, keylen,
+                                (char *)local->newloc.name);
+            if (ret)
+                gf_msg(THIS->name, GF_LOG_ERROR, 0, AFR_MSG_DICT_SET_FAILED,
+                       "%s/%s: Could not set %s key during "
+                       "xattrop",
+                       uuid_utoa(local->newloc.pargfid), local->newloc.name,
+                       key);
+        }
+    }
 
-			if (op_errno == ENOTSUP) {
-				gf_log (this->name, GF_LOG_ERROR,
-					"xattrop not supported by %s",
-					priv->children[child_index]->name);
-				local->op_ret = -1;
+    *xdata = xdata1;
+    *newloc_xdata = xdata2;
+    xdata1 = xdata2 = NULL;
+out:
+    if (xdata1)
+        dict_unref(xdata1);
+    return;
+}
 
-			} else if (!child_went_down (op_ret, op_errno)) {
-				gf_log (this->name, GF_LOG_ERROR,
-					"xattrop failed on child %s: %s",
-					priv->children[child_index]->name,
-					strerror (op_errno));
-			}
-			local->op_errno = op_errno;
-		}
+int
+afr_changelog_prepare(xlator_t *this, call_frame_t *frame, int *call_count,
+                      afr_changelog_resume_t changelog_resume,
+                      afr_xattrop_type_t op, dict_t **xdata,
+                      dict_t **newloc_xdata)
+{
+    afr_private_t *priv = NULL;
+    afr_local_t *local = NULL;
 
-		call_count = --local->call_count;
-	}
-	UNLOCK (&frame->lock);
+    local = frame->local;
+    priv = this->private;
 
-	if (call_count == 0) {
-		if ((local->op_ret == -1) &&
-		    (local->op_errno == ENOTSUP)) {
-			local->transaction.resume (frame, this);
-		} else {
-                        __mark_all_success (local->pending, priv->child_count,
-                                            local->transaction.type);
+    *call_count = afr_changelog_call_count(
+        local->transaction.type, local->transaction.pre_op,
+        local->transaction.failed_subvols, priv->child_count);
 
-                        afr_pid_restore (frame);
+    if (*call_count == 0) {
+        changelog_resume(frame, this);
+        return -1;
+    }
 
-			local->transaction.fop (frame, this);
-		}
-	}
+    afr_changelog_populate_xdata(frame, op, xdata, newloc_xdata);
+    local->call_count = *call_count;
 
-	return 0;
+    local->transaction.changelog_resume = changelog_resume;
+    return 0;
 }
 
-
 int
-afr_changelog_pre_op (call_frame_t *frame, xlator_t *this)
+afr_changelog_do(call_frame_t *frame, xlator_t *this, dict_t *xattr,
+                 afr_changelog_resume_t changelog_resume, afr_xattrop_type_t op)
 {
-	afr_private_t * priv = this->private;
-	int i = 0;
-	int ret = 0;
-	int call_count = 0;
-	dict_t **xattr = NULL;
-        afr_fd_ctx_t *fdctx = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    dict_t *xdata = NULL;
+    dict_t *newloc_xdata = NULL;
+    int i = 0;
+    int call_count = 0;
+    int ret = 0;
+
+    local = frame->local;
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (local->transaction.changelog_xdata[i]) {
+            dict_unref(local->transaction.changelog_xdata[i]);
+            local->transaction.changelog_xdata[i] = NULL;
+        }
+    }
+
+    ret = afr_changelog_prepare(this, frame, &call_count, changelog_resume, op,
+                                &xdata, &newloc_xdata);
 
-	afr_local_t *local = NULL;
-        int          piggyback = 0;
+    if (ret)
+        return 0;
 
-	local = frame->local;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->transaction.pre_op[i] ||
+            local->transaction.failed_subvols[i])
+            continue;
+
+        switch (local->transaction.type) {
+            case AFR_DATA_TRANSACTION:
+            case AFR_METADATA_TRANSACTION:
+                if (!local->fd) {
+                    STACK_WIND_COOKIE(
+                        frame, afr_changelog_cbk, (void *)(long)i,
+                        priv->children[i], priv->children[i]->fops->xattrop,
+                        &local->loc, GF_XATTROP_ADD_ARRAY, xattr, xdata);
+                } else {
+                    STACK_WIND_COOKIE(
+                        frame, afr_changelog_cbk, (void *)(long)i,
+                        priv->children[i], priv->children[i]->fops->fxattrop,
+                        local->fd, GF_XATTROP_ADD_ARRAY, xattr, xdata);
+                }
+                break;
+            case AFR_ENTRY_RENAME_TRANSACTION:
 
-        xattr = alloca (priv->child_count * sizeof (*xattr));
-        memset (xattr, 0, (priv->child_count * sizeof (*xattr)));
+                STACK_WIND_COOKIE(frame, afr_changelog_cbk, (void *)(long)i,
+                                  priv->children[i],
+                                  priv->children[i]->fops->xattrop,
+                                  &local->transaction.new_parent_loc,
+                                  GF_XATTROP_ADD_ARRAY, xattr, newloc_xdata);
+                call_count--;
 
-	for (i = 0; i < priv->child_count; i++) {
-                xattr[i] = get_new_dict ();
-                dict_ref (xattr[i]);
+                /* fall through */
+
+            case AFR_ENTRY_TRANSACTION:
+                if (local->fd)
+                    STACK_WIND_COOKIE(
+                        frame, afr_changelog_cbk, (void *)(long)i,
+                        priv->children[i], priv->children[i]->fops->fxattrop,
+                        local->fd, GF_XATTROP_ADD_ARRAY, xattr, xdata);
+                else
+                    STACK_WIND_COOKIE(frame, afr_changelog_cbk, (void *)(long)i,
+                                      priv->children[i],
+                                      priv->children[i]->fops->xattrop,
+                                      &local->transaction.parent_loc,
+                                      GF_XATTROP_ADD_ARRAY, xattr, xdata);
+                break;
         }
 
-	call_count = afr_up_children_count (priv->child_count,
-					    local->child_up);
+        if (!--call_count)
+            break;
+    }
 
-	if (local->transaction.type == AFR_ENTRY_RENAME_TRANSACTION) {
-		call_count *= 2;
-	}
+    if (xdata)
+        dict_unref(xdata);
+    if (newloc_xdata)
+        dict_unref(newloc_xdata);
+    return 0;
+}
 
-	if (call_count == 0) {
-		/* no child is up */
-                for (i = 0; i < priv->child_count; i++) {
-                        dict_unref (xattr[i]);
-                }
+static void
+afr_init_optimistic_changelog_for_txn(xlator_t *this, afr_local_t *local)
+{
+    int locked_count = 0;
+    afr_private_t *priv = NULL;
 
-                local->internal_lock.lock_cbk =
-                        local->transaction.done;
-		afr_unlock (frame, this);
-		return 0;
-	}
+    priv = this->private;
 
-	local->call_count = call_count;
+    locked_count = AFR_COUNT(local->transaction.pre_op, priv->child_count);
+    if (priv->optimistic_change_log && locked_count == priv->child_count)
+        local->optimistic_change_log = 1;
 
-	__mark_all_pending (local->pending, priv->child_count,
-                            local->transaction.type);
+    return;
+}
 
-        if (local->fd)
-                fdctx = afr_fd_ctx_get (local->fd, this);
+int
+afr_changelog_pre_op(call_frame_t *frame, xlator_t *this)
+{
+    afr_private_t *priv = this->private;
+    int i = 0;
+    int ret = 0;
+    int call_count = 0;
+    int op_errno = 0;
+    afr_local_t *local = NULL;
+    afr_internal_lock_t *int_lock = NULL;
+    unsigned char *locked_nodes = NULL;
+    int idx = -1;
+    gf_boolean_t pre_nop = _gf_true;
+    dict_t *xdata_req = NULL;
+
+    local = frame->local;
+    int_lock = &local->internal_lock;
+    idx = afr_index_for_transaction_type(local->transaction.type);
+
+    locked_nodes = afr_locked_nodes_get(local->transaction.type, int_lock);
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (locked_nodes[i]) {
+            local->transaction.pre_op[i] = 1;
+            call_count++;
+        } else {
+            local->transaction.failed_subvols[i] = 1;
+        }
+    }
+
+    afr_init_optimistic_changelog_for_txn(this, local);
+
+    if (afr_changelog_pre_op_inherit(frame, this))
+        goto next;
+
+    /* This condition should not be met with present code, as
+     * transaction.done will be called if locks are not acquired on even a
+     * single node.
+     */
+    if (call_count == 0) {
+        op_errno = ENOTCONN;
+        goto err;
+    }
+
+    /* Check if the fop can be performed on at least
+     * quorum number of nodes.
+     */
+    if (priv->quorum_count && !afr_has_fop_quorum(frame)) {
+        op_errno = int_lock->lock_op_errno;
+        if (op_errno == 0)
+            op_errno = afr_quorum_errno(priv);
+        goto err;
+    }
+
+    xdata_req = dict_new();
+    if (!xdata_req) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    if (call_count < priv->child_count)
+        pre_nop = _gf_false;
+
+    /* Set an all-zero pending changelog so that in the cbk, we can get the
+     * current on-disk values. In a replica 3 volume with arbiter enabled,
+     * these values are needed to arrive at a go/ no-go of the fop phase to
+     * avoid ending up in split-brain.*/
+
+    ret = afr_set_pending_dict(priv, xdata_req, local->pending);
+    if (ret < 0) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    if (afr_needs_changelog_update(local)) {
+        local->dirty[idx] = hton32(1);
+
+        ret = dict_set_static_bin(xdata_req, AFR_DIRTY, local->dirty,
+                                  sizeof(int) * AFR_NUM_CHANGE_LOGS);
+        if (ret) {
+            op_errno = ENOMEM;
+            goto err;
+        }
 
-	for (i = 0; i < priv->child_count; i++) {
-                if (!local->child_up[i])
-                        continue;
-                ret = afr_set_pending_dict (priv, xattr[i],
-                                            local->pending);
-
-                if (ret < 0)
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "failed to set pending entry");
-
-
-                switch (local->transaction.type) {
-                case AFR_DATA_TRANSACTION:
-                {
-                        if (!fdctx) {
-                                STACK_WIND_COOKIE (frame,
-                                                   afr_changelog_pre_op_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->xattrop,
-                                                   &(local->loc),
-                                                   GF_XATTROP_ADD_ARRAY, xattr[i]);
-                                break;
-                        }
-
-                        LOCK (&local->fd->lock);
-                        {
-                                piggyback = 0;
-                                if (fdctx->pre_op_done[i]) {
-                                        fdctx->pre_op_piggyback[i]++;
-                                        piggyback = 1;
-                                        fdctx->hit++;
-                                } else {
-                                        fdctx->miss++;
-                                }
-                        }
-                        UNLOCK (&local->fd->lock);
-
-                        if (piggyback)
-                                afr_changelog_pre_op_cbk (frame, (void *)(long)i,
-                                                          this, 1, 0, xattr[i]);
-                        else
-                                STACK_WIND_COOKIE (frame,
-                                                   afr_changelog_pre_op_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->fxattrop,
-                                                   local->fd,
-                                                   GF_XATTROP_ADD_ARRAY, xattr[i]);
-                }
-                break;
-                case AFR_METADATA_TRANSACTION:
-                {
-                        if (local->fd)
-                                STACK_WIND_COOKIE (frame,
-                                                   afr_changelog_pre_op_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->fxattrop,
-                                                   local->fd,
-                                                   GF_XATTROP_ADD_ARRAY, xattr[i]);
-                        else
-                                STACK_WIND_COOKIE (frame,
-                                                   afr_changelog_pre_op_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->xattrop,
-                                                   &(local->loc),
-                                                   GF_XATTROP_ADD_ARRAY, xattr[i]);
-                }
-                break;
+        pre_nop = _gf_false;
+        local->transaction.dirtied = 1;
+    }
 
-                case AFR_ENTRY_RENAME_TRANSACTION:
-                {
-                        STACK_WIND_COOKIE (frame,
-                                           afr_changelog_pre_op_cbk,
-                                           (void *) (long) i,
-                                           priv->children[i],
-                                           priv->children[i]->fops->xattrop,
-                                           &local->transaction.new_parent_loc,
-                                           GF_XATTROP_ADD_ARRAY, xattr[i]);
-
-                        call_count--;
-                }
+    if (pre_nop)
+        goto next;
 
+    if (!local->pre_op_compat) {
+        dict_copy(xdata_req, local->xdata_req);
+        goto next;
+    }
 
-                /*
-                  set it again because previous stack_wind
-                  might have already returned (think of case
-                  where subvolume is posix) and would have
-                  used the dict as placeholder for return
-                  value
-                */
+    afr_changelog_do(frame, this, xdata_req, afr_transaction_perform_fop,
+                     AFR_TRANSACTION_PRE_OP);
 
-                ret = afr_set_pending_dict (priv, xattr[i],
-                                            local->pending);
+    if (xdata_req)
+        dict_unref(xdata_req);
 
-                if (ret < 0)
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "failed to set pending entry");
+    return 0;
+next:
+    afr_transaction_perform_fop(frame, this);
 
-                /* fall through */
+    if (xdata_req)
+        dict_unref(xdata_req);
 
-                case AFR_ENTRY_TRANSACTION:
-                {
-                        if (local->fd)
-                                STACK_WIND_COOKIE (frame,
-                                                   afr_changelog_pre_op_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->fxattrop,
-                                                   local->fd,
-                                                   GF_XATTROP_ADD_ARRAY, xattr[i]);
-                        else
-                                STACK_WIND_COOKIE (frame,
-                                                   afr_changelog_pre_op_cbk,
-                                                   (void *) (long) i,
-                                                   priv->children[i],
-                                                   priv->children[i]->fops->xattrop,
-                                                   &local->transaction.parent_loc,
-                                                   GF_XATTROP_ADD_ARRAY, xattr[i]);
-                }
-                break;
-                }
+    return 0;
+err:
+    local->internal_lock.lock_cbk = afr_transaction_done;
+    local->op_ret = -1;
+    local->op_errno = op_errno;
 
-                if (!--call_count)
-                        break;
-	}
+    afr_handle_lock_acquire_failure(local);
 
-        for (i = 0; i < priv->child_count; i++) {
-                dict_unref (xattr[i]);
-        }
+    if (xdata_req)
+        dict_unref(xdata_req);
 
-	return 0;
+    return 0;
 }
 
+int
+afr_post_nonblocking_lock_cbk(call_frame_t *frame, xlator_t *this)
+{
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
+
+    local = frame->local;
+    int_lock = &local->internal_lock;
+
+    /* Initiate blocking locks if non-blocking has failed */
+    if (int_lock->lock_op_ret < 0) {
+        gf_msg_debug(this->name, 0,
+                     "Non blocking locks failed. Proceeding to blocking");
+        int_lock->lock_cbk = afr_internal_lock_finish;
+        afr_blocking_lock(frame, this);
+    } else {
+        gf_msg_debug(this->name, 0,
+                     "Non blocking locks done. Proceeding to FOP");
+
+        afr_internal_lock_finish(frame, this);
+    }
+
+    return 0;
+}
 
 int
-afr_post_blocking_inodelk_cbk (call_frame_t *frame, xlator_t *this)
+afr_post_blocking_rename_cbk(call_frame_t *frame, xlator_t *this)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+    local = frame->local;
+    int_lock = &local->internal_lock;
 
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Blocking inodelks failed.");
-		local->transaction.done (frame, this);
-        } else {
+    if (int_lock->lock_op_ret < 0) {
+        gf_msg(this->name, GF_LOG_INFO, 0, AFR_MSG_INTERNAL_LKS_FAILED,
+               "Blocking entrylks failed.");
 
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Blocking inodelks done. Proceeding to FOP");
-                afr_internal_lock_finish (frame, this);
-        }
+        afr_transaction_done(frame, this);
+    } else {
+        gf_msg_debug(this->name, 0,
+                     "Blocking entrylks done. Proceeding to FOP");
 
-        return 0;
+        afr_internal_lock_finish(frame, this);
+    }
+    return 0;
 }
 
-
 int
-afr_post_nonblocking_inodelk_cbk (call_frame_t *frame, xlator_t *this)
+afr_post_lower_unlock_cbk(call_frame_t *frame, xlator_t *this)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+    local = frame->local;
+    int_lock = &local->internal_lock;
 
-        /* Initiate blocking locks if non-blocking has failed */
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Non blocking inodelks failed. Proceeding to blocking");
-                int_lock->lock_cbk = afr_post_blocking_inodelk_cbk;
-                afr_blocking_lock (frame, this);
-        } else {
+    GF_ASSERT(!int_lock->higher_locked);
 
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Non blocking inodelks done. Proceeding to FOP");
-                afr_internal_lock_finish (frame, this);
-        }
+    int_lock->lock_cbk = afr_post_blocking_rename_cbk;
+    afr_blocking_lock(frame, this);
 
-        return 0;
+    return 0;
 }
 
+int
+afr_set_transaction_flock(xlator_t *this, afr_local_t *local,
+                          afr_lockee_t *lockee)
+{
+    afr_private_t *priv = NULL;
+    struct gf_flock *flock = NULL;
+
+    priv = this->private;
+    flock = &lockee->flock;
+
+    if ((priv->arbiter_count || local->transaction.eager_lock_on ||
+         priv->full_lock) &&
+        local->transaction.type == AFR_DATA_TRANSACTION) {
+        /*Lock entire file to avoid network split brains.*/
+        flock->l_len = 0;
+        flock->l_start = 0;
+    } else {
+        flock->l_len = local->transaction.len;
+        flock->l_start = local->transaction.start;
+    }
+    flock->l_type = F_WRLCK;
+
+    return 0;
+}
 
 int
-afr_post_blocking_entrylk_cbk (call_frame_t *frame, xlator_t *this)
+afr_lock(call_frame_t *frame, xlator_t *this)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
+    afr_internal_lock_t *int_lock = NULL;
+    afr_local_t *local = NULL;
+    int i = 0;
+
+    local = frame->local;
+    int_lock = &local->internal_lock;
+
+    int_lock->lock_cbk = afr_post_nonblocking_lock_cbk;
+    int_lock->domain = this->name;
+
+    switch (local->transaction.type) {
+        case AFR_DATA_TRANSACTION:
+        case AFR_METADATA_TRANSACTION:
+            for (i = 0; i < int_lock->lockee_count; i++) {
+                afr_set_transaction_flock(this, local, &int_lock->lockee[i]);
+            }
+
+            break;
+
+        case AFR_ENTRY_TRANSACTION:
+            int_lock->lk_basename = local->transaction.basename;
+            if (local->transaction.parent_loc.path)
+                int_lock->lk_loc = &local->transaction.parent_loc;
+            else
+                GF_ASSERT(local->fd);
+            break;
+        case AFR_ENTRY_RENAME_TRANSACTION:
+            break;
+    }
+    afr_lock_nonblocking(frame, this);
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+    return 0;
+}
 
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Blocking entrylks failed.");
-		local->transaction.done (frame, this);
-        } else {
+static gf_boolean_t
+afr_locals_overlap(afr_local_t *local1, afr_local_t *local2)
+{
+    uint64_t start1 = local1->transaction.start;
+    uint64_t start2 = local2->transaction.start;
+    uint64_t end1 = 0;
+    uint64_t end2 = 0;
+
+    if (local1->transaction.len)
+        end1 = start1 + local1->transaction.len - 1;
+    else
+        end1 = ULLONG_MAX;
+
+    if (local2->transaction.len)
+        end2 = start2 + local2->transaction.len - 1;
+    else
+        end2 = ULLONG_MAX;
+
+    return ((end1 >= start2) && (end2 >= start1));
+}
 
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Blocking entrylks done. Proceeding to FOP");
-                afr_internal_lock_finish (frame, this);
+gf_boolean_t
+afr_has_lock_conflict(afr_local_t *local, gf_boolean_t waitlist_check)
+{
+    afr_local_t *each = NULL;
+    afr_lock_t *lock = NULL;
+
+    lock = &local->inode_ctx->lock[local->transaction.type];
+    /*
+     * Once full file lock is acquired in eager-lock phase, overlapping
+     * writes do not compete for inode-locks, instead are transferred to the
+     * next writes. Because of this overlapping writes are not ordered.
+     * This can cause inconsistencies in replication.
+     * Example:
+     * Two overlapping writes w1, w2 are sent in parallel on same fd
+     * in two threads t1, t2.
+     * Both threads can execute afr_writev_wind in the following manner.
+     * t1 winds w1 on brick-0
+     * t2 winds w2 on brick-0
+     * t2 winds w2 on brick-1
+     * t1 winds w1 on brick-1
+     *
+     * This check makes sure the locks are not transferred for
+     * overlapping writes.
+     */
+    list_for_each_entry(each, &lock->owners, transaction.owner_list)
+    {
+        if (afr_locals_overlap(each, local)) {
+            return _gf_true;
         }
-
-        return 0;
+    }
+
+    if (!waitlist_check)
+        return _gf_false;
+    list_for_each_entry(each, &lock->waiting, transaction.wait_list)
+    {
+        if (afr_locals_overlap(each, local)) {
+            return _gf_true;
+        }
+    }
+    return _gf_false;
 }
 
+/* }}} */
+static void
+afr_copy_inodelk_vars(afr_internal_lock_t *dst, afr_internal_lock_t *src,
+                      xlator_t *this, int lockee_num)
+{
+    afr_private_t *priv = this->private;
+    afr_lockee_t *sl = &src->lockee[lockee_num];
+    afr_lockee_t *dl = &dst->lockee[lockee_num];
+
+    dst->domain = src->domain;
+    dl->flock.l_len = sl->flock.l_len;
+    dl->flock.l_start = sl->flock.l_start;
+    dl->flock.l_type = sl->flock.l_type;
+    dl->locked_count = sl->locked_count;
+    memcpy(dl->locked_nodes, sl->locked_nodes,
+           priv->child_count * sizeof(*dl->locked_nodes));
+}
 
-int
-afr_post_nonblocking_entrylk_cbk (call_frame_t *frame, xlator_t *this)
+void
+__afr_transaction_wake_shared(afr_local_t *local, struct list_head *shared)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
+    gf_boolean_t conflict = _gf_false;
+    afr_local_t *each = NULL;
+    afr_lock_t *lock = &local->inode_ctx->lock[local->transaction.type];
+
+    while (!conflict) {
+        if (list_empty(&lock->waiting))
+            return;
+        each = list_entry(lock->waiting.next, afr_local_t,
+                          transaction.wait_list);
+        if (afr_has_lock_conflict(each, _gf_false)) {
+            conflict = _gf_true;
+        }
+        if (conflict && !list_empty(&lock->owners))
+            return;
+        afr_copy_inodelk_vars(&each->internal_lock, &local->internal_lock,
+                              each->transaction.frame->this, 0);
+        list_move_tail(&each->transaction.wait_list, shared);
+        list_add_tail(&each->transaction.owner_list, &lock->owners);
+    }
+}
 
-        local = frame->local;
-        int_lock = &local->internal_lock;
+static void
+afr_lock_resume_shared(struct list_head *list)
+{
+    afr_local_t *each = NULL;
+
+    while (!list_empty(list)) {
+        each = list_entry(list->next, afr_local_t, transaction.wait_list);
+        list_del_init(&each->transaction.wait_list);
+        afr_changelog_pre_op(each->transaction.frame,
+                             each->transaction.frame->this);
+    }
+}
 
-        /* Initiate blocking locks if non-blocking has failed */
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Non blocking entrylks failed. Proceeding to blocking");
-                int_lock->lock_cbk = afr_post_blocking_entrylk_cbk;
-                afr_blocking_lock (frame, this);
+int
+afr_internal_lock_finish(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = frame->local;
+    afr_lock_t *lock = NULL;
+
+    local->internal_lock.lock_cbk = NULL;
+    if (!local->transaction.eager_lock_on) {
+        if (local->internal_lock.lock_op_ret < 0) {
+            afr_transaction_done(frame, this);
+            return 0;
+        }
+        afr_changelog_pre_op(frame, this);
+    } else {
+        lock = &local->inode_ctx->lock[local->transaction.type];
+        if (local->internal_lock.lock_op_ret < 0) {
+            afr_handle_lock_acquire_failure(local);
         } else {
-
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Non blocking entrylks done. Proceeding to FOP");
-                afr_internal_lock_finish (frame, this);
+            lock->event_generation = local->event_generation;
+            afr_changelog_pre_op(frame, this);
         }
+    }
 
-        return 0;
+    return 0;
 }
 
-
-int
-afr_post_blocking_rename_cbk (call_frame_t *frame, xlator_t *this)
+gf_boolean_t
+afr_are_conflicting_ops_waiting(afr_local_t *local, xlator_t *this)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
+    afr_lock_t *lock = NULL;
+    lock = &local->inode_ctx->lock[local->transaction.type];
+
+    /* Lets say mount1 has eager-lock(full-lock) and after the eager-lock
+     * is taken mount2 opened the same file, it won't be able to
+     * perform any {meta,}data operations until mount1 releases eager-lock.
+     * To avoid such scenario do not enable eager-lock for this transaction
+     * if open-fd-count is > 1 for metadata transactions and if num-inodelks > 1
+     * for data transactions
+     */
+
+    if (local->transaction.type == AFR_METADATA_TRANSACTION) {
+        if (local->inode_ctx->open_fd_count > 1) {
+            return _gf_true;
+        }
+    } else if (local->transaction.type == AFR_DATA_TRANSACTION) {
+        if (lock->num_inodelks > 1) {
+            return _gf_true;
+        }
+    }
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+    return _gf_false;
+}
 
-        if (int_lock->lock_op_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Blocking entrylks failed.");
-		local->transaction.done (frame, this);
-        } else {
+gf_boolean_t
+afr_is_delayed_changelog_post_op_needed(call_frame_t *frame, xlator_t *this,
+                                        int delay)
+{
+    afr_local_t *local = NULL;
+    afr_lock_t *lock = NULL;
+    gf_boolean_t res = _gf_false;
+
+    local = frame->local;
+    lock = &local->inode_ctx->lock[local->transaction.type];
+
+    if (!afr_txn_nothing_failed(frame, this)) {
+        lock->release = _gf_true;
+        goto out;
+    }
+
+    if (afr_are_conflicting_ops_waiting(local, this)) {
+        lock->release = _gf_true;
+        goto out;
+    }
+
+    if (!list_empty(&lock->owners))
+        goto out;
+    else
+        GF_ASSERT(list_empty(&lock->waiting));
+
+    if (lock->release) {
+        goto out;
+    }
+
+    if (!delay) {
+        goto out;
+    }
+
+    if (local->transaction.disable_delayed_post_op) {
+        goto out;
+    }
+
+    if ((local->op != GF_FOP_WRITE) && (local->op != GF_FOP_FXATTROP) &&
+        (local->op != GF_FOP_FSYNC)) {
+        /*Only allow writes/fsyncs but shard does [f]xattrops on writes, so
+         * they are fine too*/
+        goto out;
+    }
+
+    res = _gf_true;
+out:
+    return res;
+}
 
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Blocking entrylks done. Proceeding to FOP");
-                afr_internal_lock_finish (frame, this);
+void
+afr_delayed_changelog_wake_up_cbk(void *data)
+{
+    afr_lock_t *lock = NULL;
+    afr_local_t *local = data;
+    afr_local_t *timer_local = NULL;
+    struct list_head shared;
+
+    INIT_LIST_HEAD(&shared);
+    lock = &local->inode_ctx->lock[local->transaction.type];
+    LOCK(&local->inode->lock);
+    {
+        timer_local = list_entry(lock->post_op.next, afr_local_t,
+                                 transaction.owner_list);
+        if (list_empty(&lock->owners) && (local == timer_local)) {
+            GF_ASSERT(list_empty(&lock->waiting));
+            /*Last owner*/
+            lock->release = _gf_true;
+            lock->delay_timer = NULL;
         }
-        return 0;
+    }
+    UNLOCK(&local->inode->lock);
+    afr_changelog_post_op_now(local->transaction.frame,
+                              local->transaction.frame->this);
 }
 
-
+/* SET operation */
 int
-afr_post_lower_unlock_cbk (call_frame_t *frame, xlator_t *this)
+afr_fd_report_unstable_write(xlator_t *this, afr_local_t *local)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
-
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+    LOCK(&local->inode->lock);
+    {
+        local->inode_ctx->witnessed_unstable_write = _gf_true;
+    }
+    UNLOCK(&local->inode->lock);
 
-        GF_ASSERT (!int_lock->higher_locked);
-
-        int_lock->lock_cbk = afr_post_blocking_rename_cbk;
-        afr_blocking_lock (frame, this);
-
-        return 0;
+    return 0;
 }
 
-
-int
-afr_set_transaction_flock (afr_local_t *local)
+/* TEST and CLEAR operation */
+gf_boolean_t
+afr_fd_has_witnessed_unstable_write(xlator_t *this, inode_t *inode)
 {
-        afr_internal_lock_t *int_lock = NULL;
+    afr_inode_ctx_t *ctx = NULL;
+    gf_boolean_t witness = _gf_false;
 
-        int_lock = &local->internal_lock;
+    LOCK(&inode->lock);
+    {
+        (void)__afr_inode_ctx_get(this, inode, &ctx);
 
-        int_lock->lk_flock.l_len   = local->transaction.len;
-        int_lock->lk_flock.l_start = local->transaction.start;
-        int_lock->lk_flock.l_type  = F_WRLCK;
+        if (ctx->witnessed_unstable_write) {
+            witness = _gf_true;
+            ctx->witnessed_unstable_write = _gf_false;
+        }
+    }
+    UNLOCK(&inode->lock);
 
-        return 0;
+    return witness;
 }
 
 int
-afr_lock_rec (call_frame_t *frame, xlator_t *this)
+afr_changelog_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, struct iatt *pre,
+                        struct iatt *post, dict_t *xdata)
 {
-        afr_internal_lock_t *int_lock = NULL;
-        afr_local_t         *local    = NULL;
+    afr_private_t *priv = NULL;
+    int child_index = (long)cookie;
+    int call_count = -1;
+    afr_local_t *local = NULL;
 
-        local    = frame->local;
-        int_lock = &local->internal_lock;
+    priv = this->private;
+    local = frame->local;
 
-        int_lock->transaction_lk_type = AFR_TRANSACTION_LK;
+    if (op_ret != 0) {
+        /* Failure of fsync() is as good as failure of previous
+           write(). So treat it like one.
+        */
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, AFR_MSG_FSYNC_FAILED,
+               "fsync(%s) failed on subvolume %s. Transaction was %s",
+               uuid_utoa(local->fd->inode->gfid),
+               priv->children[child_index]->name, gf_fop_list[local->op]);
 
-	switch (local->transaction.type) {
-	case AFR_DATA_TRANSACTION:
-	case AFR_METADATA_TRANSACTION:
-                afr_set_transaction_flock (local);
+        afr_transaction_fop_failed(frame, this, child_index);
+    }
 
-                int_lock->lock_cbk = afr_post_nonblocking_inodelk_cbk;
+    call_count = afr_frame_return(frame);
 
-                afr_nonblocking_inodelk (frame, this);
-		break;
+    if (call_count == 0)
+        afr_changelog_post_op_now(frame, this);
 
-        case AFR_ENTRY_RENAME_TRANSACTION:
+    return 0;
+}
 
-                int_lock->lock_cbk = afr_post_blocking_rename_cbk;
-                afr_blocking_lock (frame, this);
-                break;
+int
+afr_changelog_fsync(call_frame_t *frame, xlator_t *this)
+{
+    afr_local_t *local = NULL;
+    int i = 0;
+    int call_count = 0;
+    afr_private_t *priv = NULL;
+    dict_t *xdata = NULL;
+    GF_UNUSED int ret = -1;
 
-	case AFR_ENTRY_TRANSACTION:
-                int_lock->lk_basename = local->transaction.basename;
-                if (&local->transaction.parent_loc)
-                        int_lock->lk_loc = &local->transaction.parent_loc;
-                else
-                        GF_ASSERT (local->fd);
+    local = frame->local;
+    priv = this->private;
 
-                int_lock->lock_cbk = afr_post_nonblocking_entrylk_cbk;
-                afr_nonblocking_entrylk (frame, this);
-                break;
-        }
+    call_count = AFR_COUNT(local->transaction.pre_op, priv->child_count);
 
+    if (!call_count) {
+        /* will go straight to unlock */
+        afr_changelog_post_op_now(frame, this);
         return 0;
-}
+    }
 
+    local->call_count = call_count;
 
-int
-afr_lock (call_frame_t *frame, xlator_t *this)
-{
-        afr_pid_save (frame);
+    xdata = dict_new();
+    if (xdata) {
+        ret = dict_set_int32_sizen(xdata, "batch-fsync", 1);
+        ret = dict_set_str(xdata, GLUSTERFS_INTERNAL_FOP_KEY, "yes");
+    }
 
-        frame->root->pid = (long) frame->root;
+    for (i = 0; i < priv->child_count; i++) {
+        if (!local->transaction.pre_op[i])
+            continue;
 
-        afr_set_lk_owner (frame, this);
+        STACK_WIND_COOKIE(frame, afr_changelog_fsync_cbk, (void *)(long)i,
+                          priv->children[i], priv->children[i]->fops->fsync,
+                          local->fd, 1, xdata);
+        if (!--call_count)
+            break;
+    }
 
-        afr_set_lock_number (frame, this);
+    if (xdata)
+        dict_unref(xdata);
 
-	return afr_lock_rec (frame, this);
+    return 0;
 }
 
-
-/* }}} */
-
 int
-afr_internal_lock_finish (call_frame_t *frame, xlator_t *this)
+afr_changelog_post_op_safe(call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t   *local = NULL;
-        afr_private_t *priv  = NULL;
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
 
-        priv  = this->private;
-        local = frame->local;
+    local = frame->local;
+    priv = this->private;
 
-        if (__changelog_needed_pre_op (frame, this)) {
-                afr_changelog_pre_op (frame, this);
-        } else {
-                __mark_all_success (local->pending, priv->child_count,
-                                    local->transaction.type);
+    if (!local->fd || local->transaction.type != AFR_DATA_TRANSACTION) {
+        afr_changelog_post_op_now(frame, this);
+        return 0;
+    }
+
+    if (afr_changelog_pre_op_uninherit(frame, this) &&
+        afr_txn_nothing_failed(frame, this)) {
+        /* just detected that this post-op is about to
+           be optimized away as a new write() has
+           already piggybacked on this frame's changelog.
+           */
+        afr_changelog_post_op_now(frame, this);
+        return 0;
+    }
+
+    /* Calling afr_changelog_post_op_now() now will result in
+       issuing ->[f]xattrop().
+
+       Performing a hard POST-OP (->[f]xattrop() FOP) is a more
+       responsible operation that what it might appear on the surface.
+
+       The changelog of a file (in the xattr of the file on the server)
+       stores information (pending count) about the state of the file
+       on the OTHER server. This changelog is blindly trusted, and must
+       therefore be updated in such a way it remains trustworthy. This
+       implies that decrementing the pending count (essentially "clearing
+       the dirty flag") must be done STRICTLY after we are sure that the
+       operation on the other server has reached stable storage.
+
+       While the backend filesystem on that server will eventually flush
+       it to stable storage, we (being in userspace) have no mechanism
+       to get notified when the write became "stable".
+
+       This means we need take matter into our own hands and issue an
+       fsync() EVEN IF THE APPLICATION WAS PERFORMING UNSTABLE WRITES,
+       and get an acknowledgement for it. And we need to wait for the
+       fsync() acknowledgement before initiating the hard POST-OP.
+
+       However if the FD itself was opened in O_SYNC or O_DSYNC then
+       we are already guaranteed that the writes were made stable as
+       part of the FOP itself. The same holds true for NFS stable
+       writes which happen on an anonymous FD with O_DSYNC or O_SYNC
+       flag set in the writev() @flags param. For all other write types,
+       mark a flag in the fdctx whenever an unstable write is witnessed.
+       */
+
+    if (!afr_fd_has_witnessed_unstable_write(this, local->inode)) {
+        afr_changelog_post_op_now(frame, this);
+        return 0;
+    }
+
+    /* Check whether users want durability and perform fsync/post-op
+     * accordingly.
+     */
+    if (priv->ensure_durability) {
+        /* Time to fsync() */
+        afr_changelog_fsync(frame, this);
+    } else {
+        afr_changelog_post_op_now(frame, this);
+    }
+
+    return 0;
+}
 
-                afr_pid_restore (frame);
+void
+afr_changelog_post_op(call_frame_t *frame, xlator_t *this)
+{
+    struct timespec delta = {
+        0,
+    };
+    afr_private_t *priv = NULL;
+    afr_local_t *local = frame->local;
+    afr_lock_t *lock = NULL;
+    gf_boolean_t post_op = _gf_true;
+    struct list_head shared;
+
+    priv = this->private;
+    delta.tv_sec = priv->post_op_delay_secs;
+    delta.tv_nsec = 0;
+
+    INIT_LIST_HEAD(&shared);
+    if (!local->transaction.eager_lock_on)
+        goto out;
+
+    lock = &local->inode_ctx->lock[local->transaction.type];
+    LOCK(&local->inode->lock);
+    {
+        list_del_init(&local->transaction.owner_list);
+        list_add(&local->transaction.owner_list, &lock->post_op);
+        __afr_transaction_wake_shared(local, &shared);
+
+        if (!afr_is_delayed_changelog_post_op_needed(frame, this,
+                                                     delta.tv_sec)) {
+            if (list_empty(&lock->owners))
+                lock->release = _gf_true;
+            goto unlock;
+        }
 
-                local->transaction.fop (frame, this);
+        GF_ASSERT(lock->delay_timer == NULL);
+        lock->delay_timer = gf_timer_call_after(
+            this->ctx, delta, afr_delayed_changelog_wake_up_cbk, local);
+        if (!lock->delay_timer) {
+            lock->release = _gf_true;
+        } else {
+            post_op = _gf_false;
         }
+    }
+unlock:
+    UNLOCK(&local->inode->lock);
 
-        return 0;
-}
+    if (!list_empty(&shared)) {
+        afr_lock_resume_shared(&shared);
+    }
 
+out:
+    if (post_op) {
+        if (!local->transaction.eager_lock_on || lock->release) {
+            afr_changelog_post_op_safe(frame, this);
+        } else {
+            afr_changelog_post_op_now(frame, this);
+        }
+    }
+}
 
 int
-afr_transaction_resume (call_frame_t *frame, xlator_t *this)
+afr_transaction_resume(call_frame_t *frame, xlator_t *this)
 {
-        afr_internal_lock_t *int_lock = NULL;
-	afr_local_t         *local    = NULL;
-	afr_private_t       *priv     = NULL;
+    afr_local_t *local = NULL;
 
-	local    = frame->local;
-        int_lock = &local->internal_lock;
-	priv     = this->private;
+    local = frame->local;
 
-	if (__changelog_needed_post_op (frame, this)) {
-		afr_changelog_post_op (frame, this);
-	} else {
-		if (afr_lock_server_count (priv, local->transaction.type) == 0) {
-			local->transaction.done (frame, this);
-		} else {
-                        int_lock->lock_cbk = local->transaction.done;
-			afr_unlock (frame, this);
-		}
-	}
+    afr_restore_lk_owner(frame);
 
-	return 0;
-}
+    afr_handle_symmetric_errors(frame, this);
 
+    if (!local->pre_op_compat)
+        /* new mode, pre-op was done along
+           with OP */
+        afr_changelog_pre_op_update(frame, this);
+
+    afr_changelog_post_op(frame, this);
+
+    return 0;
+}
 
 /**
  * afr_transaction_fop_failed - inform that an fop failed
  */
 
 void
-afr_transaction_fop_failed (call_frame_t *frame, xlator_t *this, int child_index)
+afr_transaction_fop_failed(call_frame_t *frame, xlator_t *this, int child_index)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
+    afr_local_t *local = NULL;
 
-	local = frame->local;
-	priv  = this->private;
+    local = frame->local;
 
-        __mark_child_dead (local->pending, priv->child_count,
-                           child_index, local->transaction.type);
+    local->transaction.failed_subvols[child_index] = 1;
 }
 
+static gf_boolean_t
+__need_previous_lock_unlocked(afr_local_t *local)
+{
+    afr_lock_t *lock = NULL;
+
+    lock = &local->inode_ctx->lock[local->transaction.type];
+    if (!lock->acquired)
+        return _gf_false;
+    if (lock->acquired && lock->event_generation != local->event_generation)
+        return _gf_true;
+    return _gf_false;
+}
 
-int
-afr_transaction (call_frame_t *frame, xlator_t *this, afr_transaction_type type)
+void
+__afr_eager_lock_handle(afr_local_t *local, gf_boolean_t *take_lock,
+                        gf_boolean_t *do_pre_op, afr_local_t **timer_local)
 {
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
+    afr_lock_t *lock = NULL;
+    afr_local_t *owner_local = NULL;
+    xlator_t *this = local->transaction.frame->this;
+
+    local->transaction.eager_lock_on = _gf_true;
+    afr_set_lk_owner(local->transaction.frame, this, local->inode);
+
+    lock = &local->inode_ctx->lock[local->transaction.type];
+    if (__need_previous_lock_unlocked(local)) {
+        if (!list_empty(&lock->owners)) {
+            lock->release = _gf_true;
+        } else if (lock->delay_timer) {
+            lock->release = _gf_true;
+            if (gf_timer_call_cancel(this->ctx, lock->delay_timer)) {
+                /* It will be put in frozen list
+                 * in the code flow below*/
+            } else {
+                *timer_local = list_entry(lock->post_op.next, afr_local_t,
+                                          transaction.owner_list);
+                lock->delay_timer = NULL;
+            }
+        }
+    }
+
+    if (lock->release) {
+        list_add_tail(&local->transaction.wait_list, &lock->frozen);
+        *take_lock = _gf_false;
+        goto out;
+    }
+
+    if (lock->delay_timer) {
+        *take_lock = _gf_false;
+        if (gf_timer_call_cancel(this->ctx, lock->delay_timer)) {
+            list_add_tail(&local->transaction.wait_list, &lock->frozen);
+        } else {
+            *timer_local = list_entry(lock->post_op.next, afr_local_t,
+                                      transaction.owner_list);
+            afr_copy_inodelk_vars(&local->internal_lock,
+                                  &(*timer_local)->internal_lock, this, 0);
+            lock->delay_timer = NULL;
+            *do_pre_op = _gf_true;
+            list_add_tail(&local->transaction.owner_list, &lock->owners);
+        }
+        goto out;
+    }
+
+    if (!list_empty(&lock->owners)) {
+        if (!lock->acquired || afr_has_lock_conflict(local, _gf_true)) {
+            list_add_tail(&local->transaction.wait_list, &lock->waiting);
+            *take_lock = _gf_false;
+            goto out;
+        }
+        owner_local = list_entry(lock->owners.next, afr_local_t,
+                                 transaction.owner_list);
+        afr_copy_inodelk_vars(&local->internal_lock,
+                              &owner_local->internal_lock, this, 0);
+        *take_lock = _gf_false;
+        *do_pre_op = _gf_true;
+    }
+
+    if (lock->acquired)
+        GF_ASSERT(!(*take_lock));
+    list_add_tail(&local->transaction.owner_list, &lock->owners);
+out:
+    return;
+}
 
-	local = frame->local;
-	priv  = this->private;
+void
+afr_transaction_start(afr_local_t *local, xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+    gf_boolean_t take_lock = _gf_true;
+    gf_boolean_t do_pre_op = _gf_false;
+    afr_local_t *timer_local = NULL;
+
+    priv = this->private;
+
+    if (local->transaction.type != AFR_DATA_TRANSACTION &&
+        local->transaction.type != AFR_METADATA_TRANSACTION)
+        goto lock_phase;
+
+    if (!priv->eager_lock)
+        goto lock_phase;
+
+    LOCK(&local->inode->lock);
+    {
+        __afr_eager_lock_handle(local, &take_lock, &do_pre_op, &timer_local);
+    }
+    UNLOCK(&local->inode->lock);
+lock_phase:
+    if (!local->transaction.eager_lock_on) {
+        afr_set_lk_owner(local->transaction.frame, this,
+                         local->transaction.frame->root);
+    }
+
+    if (take_lock) {
+        afr_lock(local->transaction.frame, this);
+    } else if (do_pre_op) {
+        afr_changelog_pre_op(local->transaction.frame, this);
+    }
+    /*Always call delayed_changelog_wake_up_cbk after calling pre-op above
+     * so that any inheriting can happen*/
+    if (timer_local)
+        afr_delayed_changelog_wake_up_cbk(timer_local);
+}
 
-	afr_transaction_local_init (local, priv);
+int
+afr_write_txn_refresh_done(call_frame_t *frame, xlator_t *this, int err)
+{
+    afr_local_t *local = frame->local;
+
+    if (err) {
+        AFR_SET_ERROR_AND_CHECK_SPLIT_BRAIN(-1, err);
+        goto fail;
+    }
+
+    afr_transaction_start(local, this);
+    return 0;
+fail:
+    local->transaction.unwind(frame, this);
+    AFR_STACK_DESTROY(frame);
+    return 0;
+}
 
-	local->transaction.resume = afr_transaction_resume;
-	local->transaction.type   = type;
+int
+afr_transaction_lockee_init(call_frame_t *frame)
+{
+    afr_local_t *local = frame->local;
+    afr_internal_lock_t *int_lock = &local->internal_lock;
+    afr_private_t *priv = frame->this->private;
+    int ret = 0;
+
+    switch (local->transaction.type) {
+        case AFR_DATA_TRANSACTION:
+        case AFR_METADATA_TRANSACTION:
+            ret = afr_add_inode_lockee(local, priv->child_count);
+            break;
+
+        case AFR_ENTRY_TRANSACTION:
+        case AFR_ENTRY_RENAME_TRANSACTION:
+            ret = afr_add_entry_lockee(local, &local->transaction.parent_loc,
+                                       local->transaction.basename,
+                                       priv->child_count);
+            if (ret) {
+                goto out;
+            }
+            if (local->op == GF_FOP_RENAME) {
+                ret = afr_add_entry_lockee(
+                    local, &local->transaction.new_parent_loc,
+                    local->transaction.new_basename, priv->child_count);
+                if (ret) {
+                    goto out;
+                }
 
-	if (afr_lock_server_count (priv, local->transaction.type) == 0) {
-                afr_internal_lock_finish (frame, this);
-	} else {
-		afr_lock (frame, this);
-	}
+                if (local->newloc.inode &&
+                    IA_ISDIR(local->newloc.inode->ia_type)) {
+                    ret = afr_add_entry_lockee(local, &local->newloc, NULL,
+                                               priv->child_count);
+                    if (ret) {
+                        goto out;
+                    }
+                }
+            } else if (local->op == GF_FOP_RMDIR) {
+                ret = afr_add_entry_lockee(local, &local->loc, NULL,
+                                           priv->child_count);
+                if (ret) {
+                    goto out;
+                }
+            }
+
+            if (int_lock->lockee_count > 1) {
+                qsort(int_lock->lockee, int_lock->lockee_count,
+                      sizeof(*int_lock->lockee), afr_entry_lockee_cmp);
+            }
+            break;
+    }
+out:
+    return ret;
+}
 
-	return 0;
+int
+afr_transaction(call_frame_t *frame, xlator_t *this, afr_transaction_type type)
+{
+    afr_local_t *local = NULL;
+    afr_private_t *priv = NULL;
+    int ret = -1;
+    int event_generation = 0;
+
+    local = frame->local;
+    priv = this->private;
+    local->transaction.frame = frame;
+
+    local->transaction.type = type;
+
+    if (priv->quorum_count && !afr_has_quorum(local->child_up, this, NULL)) {
+        ret = -afr_quorum_errno(priv);
+        goto out;
+    }
+
+    if (!afr_is_consistent_io_possible(local, priv, &ret)) {
+        ret = -ret; /*op_errno to ret conversion*/
+        goto out;
+    }
+
+    if (priv->thin_arbiter_count && !afr_ta_has_quorum(priv, local)) {
+        ret = -afr_quorum_errno(priv);
+        goto out;
+    }
+
+    ret = afr_transaction_local_init(local, this);
+    if (ret < 0)
+        goto out;
+
+    ret = afr_transaction_lockee_init(frame);
+    if (ret)
+        goto out;
+
+    if (type != AFR_METADATA_TRANSACTION) {
+        goto txn_start;
+    }
+
+    ret = afr_inode_get_readable(frame, local->inode, this, local->readable,
+                                 &event_generation, type);
+    if (ret < 0 ||
+        afr_is_inode_refresh_reqd(local->inode, this, priv->event_generation,
+                                  event_generation)) {
+        afr_inode_refresh(frame, this, local->inode, local->loc.gfid,
+                          afr_write_txn_refresh_done);
+        ret = 0;
+        goto out;
+    }
+
+txn_start:
+    ret = 0;
+    afr_transaction_start(local, this);
+out:
+    return ret;
 }
diff --git a/xlators/cluster/afr/src/afr-transaction.h b/xlators/cluster/afr/src/afr-transaction.h
index 5b45d5a5d9f..beefa26f4a6 100644
--- a/xlators/cluster/afr/src/afr-transaction.h
+++ b/xlators/cluster/afr/src/afr-transaction.h
@@ -1,33 +1,75 @@
 /*
-   Copyright (c) 2007-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #ifndef __TRANSACTION_H__
 #define __TRANSACTION_H__
 
+#include "afr.h"
+
 void
-afr_transaction_fop_failed (call_frame_t *frame, xlator_t *this,
-			    int child_index);
+afr_transaction_fop_failed(call_frame_t *frame, xlator_t *this,
+                           int child_index);
+
+int32_t
+afr_transaction(call_frame_t *frame, xlator_t *this, afr_transaction_type type);
 
 int
-afr_lock_server_count (afr_private_t *priv, afr_transaction_type type);
+afr_set_pending_dict(afr_private_t *priv, dict_t *xattr, int32_t **pending);
 
-int32_t
-afr_transaction (call_frame_t *frame, xlator_t *this, afr_transaction_type type);
+void
+afr_delayed_changelog_wake_up(xlator_t *this, fd_t *fd);
+
+void
+__mark_all_success(call_frame_t *frame, xlator_t *this);
+
+gf_boolean_t
+afr_txn_nothing_failed(call_frame_t *frame, xlator_t *this);
+
+int
+afr_read_txn(call_frame_t *frame, xlator_t *this, inode_t *inode,
+             afr_read_txn_wind_t readfn, afr_transaction_type type);
+
+int
+afr_read_txn_continue(call_frame_t *frame, xlator_t *this, int subvol);
+
+void
+afr_pending_read_increment(afr_private_t *priv, int child_index);
 
+void
+afr_pending_read_decrement(afr_private_t *priv, int child_index);
+
+call_frame_t *
+afr_transaction_detach_fop_frame(call_frame_t *frame);
+gf_boolean_t
+afr_has_quorum(unsigned char *subvols, xlator_t *this, call_frame_t *frame);
+gf_boolean_t
+afr_needs_changelog_update(afr_local_t *local);
+void
+afr_zero_fill_stat(afr_local_t *local);
+
+void
+afr_pick_error_xdata(afr_local_t *local, afr_private_t *priv, inode_t *inode1,
+                     unsigned char *readable1, inode_t *inode2,
+                     unsigned char *readable2);
+int
+afr_transaction_resume(call_frame_t *frame, xlator_t *this);
+
+int
+afr_lock(call_frame_t *frame, xlator_t *this);
+
+void
+afr_delayed_changelog_wake_up_cbk(void *data);
+
+int
+afr_release_notify_lock_for_ta(void *opaque);
+
+int
+afr_ta_lock_release_done(int ret, call_frame_t *ta_frame, void *opaque);
 #endif /* __TRANSACTION_H__ */
diff --git a/xlators/cluster/afr/src/afr.c b/xlators/cluster/afr/src/afr.c
index b026db594ca..df7366f0a65 100644
--- a/xlators/cluster/afr/src/afr.c
+++ b/xlators/cluster/afr/src/afr.c
@@ -1,20 +1,11 @@
 /*
-   Copyright (c) 2007-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 #include <libgen.h>
@@ -24,990 +15,1330 @@
 #include <stdlib.h>
 #include <signal.h>
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
 #include "afr-common.c"
+#include "afr-messages.h"
+
+struct volume_options options[];
+
+static char *afr_favorite_child_policies[AFR_FAV_CHILD_POLICY_MAX + 1] = {
+    [AFR_FAV_CHILD_NONE] = "none",
+    [AFR_FAV_CHILD_BY_SIZE] = "size",
+    [AFR_FAV_CHILD_BY_CTIME] = "ctime",
+    [AFR_FAV_CHILD_BY_MTIME] = "mtime",
+    [AFR_FAV_CHILD_BY_MAJORITY] = "majority",
+    [AFR_FAV_CHILD_POLICY_MAX] = NULL,
+};
 
 int32_t
-notify (xlator_t *this, int32_t event,
-	void *data, ...)
+notify(xlator_t *this, int32_t event, void *data, ...)
 {
-        int ret = -1;
+    int ret = -1;
+    va_list ap;
+    void *data2 = NULL;
 
-        ret = afr_notify (this, event, data);
+    va_start(ap, data);
+    data2 = va_arg(ap, dict_t *);
+    va_end(ap);
+    ret = afr_notify(this, event, data, data2);
 
-        return ret;
+    return ret;
 }
 
 int32_t
-mem_acct_init (xlator_t *this)
+mem_acct_init(xlator_t *this)
 {
-        int     ret = -1;
-
-        if (!this)
-                return ret;
+    int ret = -1;
 
-        ret = xlator_mem_acct_init (this, gf_afr_mt_end + 1);
+    if (!this)
+        return ret;
 
-        if (ret != 0) {
-                gf_log(this->name, GF_LOG_ERROR, "Memory accounting init"
-                                "failed");
-                return ret;
-        }
+    ret = xlator_mem_acct_init(this, gf_afr_mt_end + 1);
 
+    if (ret != 0) {
         return ret;
+    }
+
+    return ret;
 }
+
 int
-validate_options (xlator_t *this, dict_t *options, char **op_errstr)
+xlator_subvolume_index(xlator_t *this, xlator_t *subvol)
 {
+    int index = -1;
+    int i = 0;
+    xlator_list_t *list = NULL;
 
- 
-        gf_boolean_t metadata_self_heal;   
-        gf_boolean_t entry_self_heal;
-        gf_boolean_t data_self_heal;
-        gf_boolean_t data_change_log;       
-        gf_boolean_t metadata_change_log; 
-        gf_boolean_t entry_change_log;      
-        gf_boolean_t strict_readdir;
-
-        xlator_list_t * trav        = NULL;
-        
-        char * read_subvol     = NULL;
-        char * self_heal       = NULL;
-        char * change_log      = NULL;
-        char * str_readdir     = NULL;
-
-        int32_t background_count  = 0;
-        int32_t window_size       = 0;
-
-        int    read_ret      = -1;
-        int    dict_ret      = -1;
-        int    flag          = 1;
-        int    ret           = 0;
-        int    temp_ret      = -1;
-        
-
-        
-        dict_ret = dict_get_int32 (options, "background-self-heal-count",
-                                   &background_count);
-        if (dict_ret == 0) {
-                if (background_count < 0) {
-                        *op_errstr = gf_strdup ("Error, option should be >= 0");
-                        ret = -1;
-                        goto out;
-                }
-
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "validated background self-heal count to %d",
-                        background_count);
-        }
+    list = this->children;
 
-        dict_ret = dict_get_str (options, "metadata-self-heal",
-                                 &self_heal);
-        if (dict_ret == 0) {
-                temp_ret = gf_string2boolean (self_heal, &metadata_self_heal);
-                if (temp_ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "validation failed 'option metadata"
-                                                "-self-heal %s'.not correct.", 
-                                                self_heal);
-                        *op_errstr = gf_strdup ("Error, option should be boolean");
-                        ret = -1;
-                        goto out;
-                } 
-        
-
-                
+    while (list) {
+        if (subvol == list->xlator ||
+            strcmp(subvol->name, list->xlator->name) == 0) {
+            index = i;
+            break;
         }
+        list = list->next;
+        i++;
+    }
 
-        dict_ret = dict_get_str (options, "data-self-heal",
-                                 &self_heal);
-        if (dict_ret == 0) {
-                temp_ret = gf_string2boolean (self_heal, &data_self_heal);
-                if (temp_ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Validation failed for data self heal", 
-                                self_heal);
-                        *op_errstr = gf_strdup ("Error, option should be boolean");
-                        ret = -1;
-                        goto out;
-                } 
-        
-
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Reconfiguring 'option data"
-                                        "-self-heal %s'.", self_heal);
-        }
+    return index;
+}
 
-        dict_ret = dict_get_str (options, "entry-self-heal",
-                                 &self_heal);
-        if (dict_ret == 0) {
-                temp_ret = gf_string2boolean (self_heal, &entry_self_heal);
-                if (temp_ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Validation faled for entry-self-heal", 
-                                self_heal);
-                        *op_errstr = gf_strdup ("Error, option should be boolean");
-                        ret = -1;
-                        goto out;
-                } 
-        
-
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Validated 'option entry"
-                                        "-self-heal %s'.", self_heal);
-        }
+static void
+fix_quorum_options(xlator_t *this, afr_private_t *priv, char *qtype,
+                   dict_t *options)
+{
+    if (dict_get_sizen(options, "quorum-type") == NULL) {
+        /* If user doesn't configure anything enable auto-quorum if the
+         * replica has more than two subvolumes */
+        if (priv->child_count > 2)
+            qtype = "auto";
+    }
+
+    if (priv->quorum_count && strcmp(qtype, "fixed")) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_QUORUM_OVERRIDE,
+               "quorum-type %s overriding quorum-count %u", qtype,
+               priv->quorum_count);
+    }
+
+    if (!strcmp(qtype, "none")) {
+        priv->quorum_count = 0;
+    } else if (!strcmp(qtype, "auto")) {
+        priv->quorum_count = AFR_QUORUM_AUTO;
+    }
+}
 
+int
+afr_set_favorite_child_policy(afr_private_t *priv, char *policy)
+{
+    int index = -1;
 
-        dict_ret = dict_get_str (options, "strict-readdir",
-                                 &str_readdir);
-        if (dict_ret == 0) {
-                temp_ret = gf_string2boolean (str_readdir, &strict_readdir);
-                if (temp_ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Validation faled for strict_readdir",
-                                str_readdir);
-                        *op_errstr = gf_strdup ("Error, option should be boolean");
-                        ret = -1;
-                        goto out;
-                }
-                
-
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Validated 'option strict"
-                                        "-readdir %s'.", str_readdir);
-        }
+    index = gf_get_index_by_elem(afr_favorite_child_policies, policy);
+    if (index < 0 || index >= AFR_FAV_CHILD_POLICY_MAX)
+        return -1;
 
-        dict_ret = dict_get_int32 (options, "data-self-heal-window-size",
-                                   &window_size);
-        if (dict_ret == 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "validated data self-heal window size to %d",
-                        window_size);
+    priv->fav_child_policy = index;
 
-                if (window_size < 0) {
-                        *op_errstr = gf_strdup ("Error, option should be >= 0");
-                        ret = -1;
-                        goto out;
-                }
+    return 0;
+}
 
-                if (window_size > 1024) {
-                        *op_errstr = gf_strdup ("Error, option should be <= 1024");
-                        ret = -1;
-                        goto out;
-                }
+static void
+set_data_self_heal_algorithm(afr_private_t *priv, char *algo)
+{
+    if (!algo) {
+        priv->data_self_heal_algorithm = AFR_SELFHEAL_DATA_DYNAMIC;
+    } else if (strcmp(algo, "full") == 0) {
+        priv->data_self_heal_algorithm = AFR_SELFHEAL_DATA_FULL;
+    } else if (strcmp(algo, "diff") == 0) {
+        priv->data_self_heal_algorithm = AFR_SELFHEAL_DATA_DIFF;
+    } else {
+        priv->data_self_heal_algorithm = AFR_SELFHEAL_DATA_DYNAMIC;
+    }
+}
 
+void
+afr_handle_anon_inode_options(afr_private_t *priv, dict_t *options)
+{
+    char *volfile_id_str = NULL;
+    uuid_t anon_inode_gfid = {0};
+
+    /*If volume id is not present don't enable anything*/
+    if (dict_get_str(options, "volume-id", &volfile_id_str))
+        return;
+    GF_ASSERT(strlen(AFR_ANON_DIR_PREFIX) + strlen(volfile_id_str) <= NAME_MAX);
+    /*anon_inode_name is not supposed to change once assigned*/
+    if (!priv->anon_inode_name[0]) {
+        snprintf(priv->anon_inode_name, sizeof(priv->anon_inode_name), "%s-%s",
+                 AFR_ANON_DIR_PREFIX, volfile_id_str);
+        gf_uuid_parse(volfile_id_str, anon_inode_gfid);
+        /*Flip a bit to make sure volfile-id and anon-gfid are not same*/
+        anon_inode_gfid[0] ^= 1;
+        uuid_utoa_r(anon_inode_gfid, priv->anon_gfid_str);
+    }
+}
 
-        }
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    afr_private_t *priv = NULL;
+    xlator_t *read_subvol = NULL;
+    int read_subvol_index = -1;
+    int timeout_old = 0;
+    int ret = -1;
+    int index = -1;
+    char *qtype = NULL;
+    char *fav_child_policy = NULL;
+    char *data_self_heal = NULL;
+    char *data_self_heal_algorithm = NULL;
+    char *locking_scheme = NULL;
+    gf_boolean_t consistent_io = _gf_false;
+    gf_boolean_t choose_local_old = _gf_false;
+    gf_boolean_t enabled_old = _gf_false;
 
-        dict_ret = dict_get_str (options, "data-change-log",
-                                 &change_log);
-        if (dict_ret == 0) {
-                temp_ret = gf_string2boolean (change_log, &data_change_log);
-                if (temp_ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Validation faled for data-change-log");
-                        *op_errstr = gf_strdup ("Error, option should be boolean");
-                        ret = -1;
-                        goto out;
-                }
- 
-
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Validated 'option data-"
-                                        "change-log %s'.", change_log);
-        }
+    priv = this->private;
 
-        dict_ret = dict_get_str (options, "metadata-change-log",
-                                 &change_log);
-        if (dict_ret == 0) {
-                temp_ret = gf_string2boolean (change_log,
-                                              &metadata_change_log);
-                if (temp_ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Validation faild for metadata-change-log");
-                        *op_errstr = gf_strdup ("Error, option should be boolean");
-                        ret = -1;
-                        goto out;
-                } 
-                
-
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Validated 'option metadata-"
-                                        "change-log %s'.", change_log);
-        }
+    GF_OPTION_RECONF("metadata-splitbrain-forced-heal",
+                     priv->metadata_splitbrain_forced_heal, options, bool, out);
 
-        dict_ret = dict_get_str (options, "entry-change-log",
-                                 &change_log);
-        if (dict_ret == 0) {
-                temp_ret = gf_string2boolean (change_log, &entry_change_log);
-                if (temp_ret < 0) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Validation faild for entr-change-log");
-                        *op_errstr = gf_strdup ("Error, option should be boolean");
-                        ret = -1;
-                        goto out;
-                } 
-
-
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Validated 'option entry-"
-                                        "change-log %s'.", change_log);
-        }
+    GF_OPTION_RECONF("background-self-heal-count",
+                     priv->background_self_heal_count, options, uint32, out);
 
-        read_ret = dict_get_str (options, "read-subvolume", &read_subvol);
-
-        if (read_ret) 
-                goto next;// No need to traverse, hence set the next option
-        
-        trav = this->children;
-        flag = 0;
-        while (trav) {
-                if (!read_ret && !strcmp (read_subvol, trav->xlator->name)) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Validated Subvolume '%s'  as read child.",
-                                trav->xlator->name);
-
-                        flag = 1;
-                        ret = 0; 
-                        goto out;
-                }
-
-                
-                trav = trav->next;
-        }
+    GF_OPTION_RECONF("heal-wait-queue-length", priv->heal_wait_qlen, options,
+                     uint32, out);
 
-        if (flag == 0 ) {
+    GF_OPTION_RECONF("metadata-self-heal", priv->metadata_self_heal, options,
+                     bool, out);
 
-                gf_log (this->name, GF_LOG_WARNING,
-                        "Invalid 'option read-subvolume %s', no such subvolume"
-                                        , read_subvol);
-                *op_errstr = gf_strdup ("Error, the sub-volume is not right");
-                ret = -1;
-                goto out;
-        }
+    GF_OPTION_RECONF("data-self-heal", data_self_heal, options, str, out);
+    if (gf_string2boolean(data_self_heal, &priv->data_self_heal) == -1)
+        goto out;
 
-next:
-out:
-                return ret;
+    GF_OPTION_RECONF("entry-self-heal", priv->entry_self_heal, options, bool,
+                     out);
 
+    GF_OPTION_RECONF("data-self-heal-window-size",
+                     priv->data_self_heal_window_size, options, uint32, out);
 
-}
+    GF_OPTION_RECONF("data-self-heal-algorithm", data_self_heal_algorithm,
+                     options, str, out);
+    set_data_self_heal_algorithm(priv, data_self_heal_algorithm);
 
+    GF_OPTION_RECONF("halo-enabled", priv->halo_enabled, options, bool, out);
 
-int
-reconfigure (xlator_t *this, dict_t *options)
-{
+    GF_OPTION_RECONF("halo-shd-max-latency", priv->shd.halo_max_latency_msec,
+                     options, uint32, out);
+
+    GF_OPTION_RECONF("halo-nfsd-max-latency", priv->nfsd.halo_max_latency_msec,
+                     options, uint32, out);
+
+    GF_OPTION_RECONF("halo-max-latency", priv->halo_max_latency_msec, options,
+                     uint32, out);
+
+    GF_OPTION_RECONF("halo-max-replicas", priv->halo_max_replicas, options,
+                     uint32, out);
+
+    GF_OPTION_RECONF("halo-min-replicas", priv->halo_min_replicas, options,
+                     uint32, out);
 
-	gf_boolean_t metadata_self_heal;   /* on/off */
-	gf_boolean_t entry_self_heal;
-	gf_boolean_t data_self_heal;
-	gf_boolean_t data_change_log;       /* on/off */
-	gf_boolean_t metadata_change_log;   /* on/off */
-	gf_boolean_t entry_change_log;      /* on/off */
-	gf_boolean_t strict_readdir;
-
-	afr_private_t * priv        = NULL;
-	xlator_list_t * trav        = NULL;
-	
-	char * read_subvol     = NULL;
-	char * self_heal       = NULL;
-	char * change_log      = NULL;
-	char * str_readdir     = NULL;
-
-        int32_t background_count  = 0;
-        int32_t window_size       = 0;
-
-	int    read_ret      = -1;
-	int    dict_ret      = -1;
-	int    flag	     = 1;
-	int    ret           = 0;
-	int    temp_ret	     = -1;
-	
-	priv = this->private;
-	
-	dict_ret = dict_get_int32 (options, "background-self-heal-count",
-				   &background_count);
-	if (dict_ret == 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"Reconfiguring background self-heal count to %d",
-			background_count);
-
-		priv->background_self_heal_count = background_count;
-	}
-
-	dict_ret = dict_get_str (options, "metadata-self-heal",
-				 &self_heal);
-	if (dict_ret == 0) {
-		temp_ret = gf_string2boolean (self_heal, &metadata_self_heal);
-		if (temp_ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"Reconfiguration Invalid 'option metadata"
-				"-self-heal %s'. Defaulting to old value.", 
-				self_heal);
-			ret = -1;
-			goto out;
-		} 
-	
-		priv->metadata_self_heal = metadata_self_heal;
-		gf_log (this->name, GF_LOG_DEBUG,
-			"Reconfiguring 'option metadata"
-			"-self-heal %s'.", 
-			self_heal);
-	}
-
-	dict_ret = dict_get_str (options, "data-self-heal",
-				 &self_heal);
-	if (dict_ret == 0) {
-		temp_ret = gf_string2boolean (self_heal, &data_self_heal);
-		if (temp_ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"Reconfiguration Invalid 'option data"
-				"-self-heal %s'. Defaulting to old value.", 
-				self_heal);
-			ret = -1;
-			goto out;
-		} 
-	
-		priv->data_self_heal = data_self_heal;
-		gf_log (this->name, GF_LOG_DEBUG,
-			"Reconfiguring 'option data"
-			"-self-heal %s'.", self_heal);
-	}
-
-	dict_ret = dict_get_str (options, "entry-self-heal",
-				 &self_heal);
-	if (dict_ret == 0) {
-		temp_ret = gf_string2boolean (self_heal, &entry_self_heal);
-		if (temp_ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"Reconfiguration Invalid 'option data"
-				"-self-heal %s'. Defaulting to old value.", 
-				self_heal);
-			ret = -1;
-			goto out;
-		} 
-	
-		priv->entry_self_heal = entry_self_heal;
-		gf_log (this->name, GF_LOG_DEBUG,
-			"Reconfiguring 'option entry"
-			"-self-heal %s'.", self_heal);
-	}
-
-
-	dict_ret = dict_get_str (options, "strict-readdir",
-				 &str_readdir);
-	if (dict_ret == 0) {
-		temp_ret = gf_string2boolean (str_readdir, &strict_readdir);
-		if (temp_ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"Invalid 'option strict-readdir %s'. "
-				"Defaulting to old value.",
-				str_readdir);
-			ret = -1;
-			goto out;
-		}
-		
-		priv->strict_readdir = strict_readdir;
-		gf_log (this->name, GF_LOG_DEBUG,
-			"Reconfiguring 'option strict"
-			"-readdir %s'.", str_readdir);
-	}
-
-	dict_ret = dict_get_int32 (options, "data-self-heal-window-size",
-				   &window_size);
-	if (dict_ret == 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"Reconfiguring, Setting data self-heal window size to %d",
-			window_size);
-
-		priv->data_self_heal_window_size = window_size;
-	}
-        else {
-                priv->data_self_heal_window_size = 16;
+    GF_OPTION_RECONF("read-subvolume", read_subvol, options, xlator, out);
+
+    choose_local_old = priv->choose_local;
+    GF_OPTION_RECONF("choose-local", priv->choose_local, options, bool, out);
+
+    if (choose_local_old != priv->choose_local) {
+        priv->read_child = -1;
+        if (choose_local_old == _gf_false)
+            priv->did_discovery = _gf_false;
+    }
+
+    GF_OPTION_RECONF("read-hash-mode", priv->hash_mode, options, uint32, out);
+
+    if (read_subvol) {
+        index = xlator_subvolume_index(this, read_subvol);
+        if (index == -1) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_INVALID_SUBVOL,
+                   "%s not a subvolume", read_subvol->name);
+            goto out;
         }
-        
-
-	dict_ret = dict_get_str (options, "data-change-log",
-				 &change_log);
-	if (dict_ret == 0) {
-		temp_ret = gf_string2boolean (change_log, &data_change_log);
-		if (temp_ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"Reconfiguration Invalid 'option data-"
-				"change-log %s'. Defaulting to old value.", 
-				change_log);
-			ret = -1;
-			goto out;
-		}
- 
-		priv->data_change_log = data_change_log;
-		gf_log (this->name, GF_LOG_DEBUG,
-			"Reconfiguring 'option data-"
-			"change-log %s'.", change_log);
-	}
-
-	dict_ret = dict_get_str (options, "metadata-change-log",
-				 &change_log);
-	if (dict_ret == 0) {
-		temp_ret = gf_string2boolean (change_log,
-					 &metadata_change_log);
-		if (temp_ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"Invalid 'option metadata-change-log %s'. "
-				"Defaulting to metadata-change-log as 'off'.",
-				change_log);
-			ret = -1;
-			goto out;
-		} 
-		
-		priv->metadata_change_log = metadata_change_log;
-		gf_log (this->name, GF_LOG_DEBUG,
-			"Reconfiguring 'option metadata-"
-			"change-log %s'.", change_log);
-	}
-
-	dict_ret = dict_get_str (options, "entry-change-log",
-				 &change_log);
-	if (dict_ret == 0) {
-		temp_ret = gf_string2boolean (change_log, &entry_change_log);
-		if (temp_ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"Invalid 'option entry-change-log %s'. "
-				"Defaulting to entry-change-log as 'on'.", 
-				change_log);
-			ret = -1;
-			goto out;
-		} 
-
-		priv->entry_change_log = entry_change_log;
-		gf_log (this->name, GF_LOG_DEBUG,
-			"Reconfiguring 'option entry-"
-			"change-log %s'.", change_log);
-	}
-
-	read_ret = dict_get_str (options, "read-subvolume", &read_subvol);
-
-	if (read_ret < 0) 
-		goto next;// No need to traverse, hence set the next option
-	
-	trav = this->children;
-	flag = 0;
-	while (trav) {
-		if (!read_ret && !strcmp (read_subvol, trav->xlator->name)) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"Subvolume '%s' specified as read child.",
-				trav->xlator->name);
-
-			flag = 1;
-			ret = -1; 
-			goto out;
-		}
-
-		
-		trav = trav->next;
-	}
-
-	if (flag == 0 ) {
-
-		gf_log (this->name, GF_LOG_ERROR,
-			"Invalid 'option read-subvolume %s', no such subvolume"
-			, read_subvol);
-		ret = -1;
-		goto out;
-	}
-
-next:
+        priv->read_child = index;
+    }
+
+    GF_OPTION_RECONF("read-subvolume-index", read_subvol_index, options, int32,
+                     out);
+
+    if (read_subvol_index > -1) {
+        index = read_subvol_index;
+        if (index >= priv->child_count) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_INVALID_SUBVOL,
+                   "%d not a subvolume-index", index);
+            goto out;
+        }
+        priv->read_child = index;
+    }
+
+    GF_OPTION_RECONF("pre-op-compat", priv->pre_op_compat, options, bool, out);
+    GF_OPTION_RECONF("locking-scheme", locking_scheme, options, str, out);
+    priv->granular_locks = (strcmp(locking_scheme, "granular") == 0);
+    GF_OPTION_RECONF("full-lock", priv->full_lock, options, bool, out);
+    GF_OPTION_RECONF("granular-entry-heal", priv->esh_granular, options, bool,
+                     out);
+
+    GF_OPTION_RECONF("eager-lock", priv->eager_lock, options, bool, out);
+    GF_OPTION_RECONF("optimistic-change-log", priv->optimistic_change_log,
+                     options, bool, out);
+    GF_OPTION_RECONF("quorum-type", qtype, options, str, out);
+    GF_OPTION_RECONF("quorum-count", priv->quorum_count, options, uint32, out);
+    fix_quorum_options(this, priv, qtype, options);
+    if (priv->quorum_count && !afr_has_quorum(priv->child_up, this, NULL))
+        gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_QUORUM_FAIL,
+               "Client-quorum is not met");
+
+    GF_OPTION_RECONF("post-op-delay-secs", priv->post_op_delay_secs, options,
+                     uint32, out);
+
+    GF_OPTION_RECONF(AFR_SH_READDIR_SIZE_KEY, priv->sh_readdir_size, options,
+                     size_uint64, out);
+    /* Reset this so we re-discover in case the topology changed.  */
+    GF_OPTION_RECONF("ensure-durability", priv->ensure_durability, options,
+                     bool, out);
+
+    enabled_old = priv->shd.enabled;
+    GF_OPTION_RECONF("self-heal-daemon", priv->shd.enabled, options, bool, out);
+
+    GF_OPTION_RECONF("iam-self-heal-daemon", priv->shd.iamshd, options, bool,
+                     out);
+
+    timeout_old = priv->shd.timeout;
+    GF_OPTION_RECONF("heal-timeout", priv->shd.timeout, options, int32, out);
+
+    GF_OPTION_RECONF("consistent-metadata", priv->consistent_metadata, options,
+                     bool, out);
+
+    GF_OPTION_RECONF("shd-max-threads", priv->shd.max_threads, options, uint32,
+                     out);
+
+    GF_OPTION_RECONF("shd-wait-qlength", priv->shd.wait_qlength, options,
+                     uint32, out);
+
+    GF_OPTION_RECONF("favorite-child-policy", fav_child_policy, options, str,
+                     out);
+    if (afr_set_favorite_child_policy(priv, fav_child_policy) == -1)
+        goto out;
+
+    priv->did_discovery = _gf_false;
+
+    GF_OPTION_RECONF("consistent-io", consistent_io, options, bool, out);
+    if (priv->quorum_count != 0)
+        consistent_io = _gf_false;
+    priv->consistent_io = consistent_io;
+
+    afr_handle_anon_inode_options(priv, options);
+
+    GF_OPTION_RECONF("use-anonymous-inode", priv->use_anon_inode, options, bool,
+                     out);
+    if (priv->shd.enabled) {
+        if ((priv->shd.enabled != enabled_old) ||
+            (timeout_old != priv->shd.timeout))
+            afr_selfheal_childup(this, priv);
+    }
+
+    ret = 0;
 out:
-	return ret;
+    return ret;
+}
 
+static int
+afr_pending_xattrs_init(afr_private_t *priv, xlator_t *this)
+{
+    int ret = -1;
+    int i = 0;
+    char *ptr = NULL;
+    char *ptr1 = NULL;
+    char *xattrs_list = NULL;
+    xlator_list_t *trav = NULL;
+    int child_count = -1;
+
+    trav = this->children;
+    child_count = priv->child_count;
+    if (priv->thin_arbiter_count) {
+        /* priv->pending_key[THIN_ARBITER_BRICK_INDEX] is used as the
+         * name of the thin arbiter file for persistence across add/
+         * removal of DHT subvols.*/
+        child_count++;
+    }
+
+    GF_OPTION_INIT("afr-pending-xattr", xattrs_list, str, out);
+    priv->pending_key = GF_CALLOC(sizeof(*priv->pending_key), child_count,
+                                  gf_afr_mt_char);
+    if (!priv->pending_key) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    if (!xattrs_list) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_NO_CHANGELOG,
+               "Unable to fetch afr-pending-xattr option from volfile."
+               " Falling back to using client translator names. ");
+
+        while (i < child_count) {
+            ret = gf_asprintf(&priv->pending_key[i], "%s.%s", AFR_XATTR_PREFIX,
+                              trav->xlator->name);
+            if (ret == -1) {
+                ret = -ENOMEM;
+                goto out;
+            }
+            trav = trav->next;
+            i++;
+        }
+        ret = 0;
+        goto out;
+    }
+
+    ptr = ptr1 = gf_strdup(xattrs_list);
+    if (!ptr) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    for (i = 0, ptr = strtok(ptr, ","); ptr; ptr = strtok(NULL, ",")) {
+        ret = gf_asprintf(&priv->pending_key[i], "%s.%s", AFR_XATTR_PREFIX,
+                          ptr);
+        if (ret == -1) {
+            ret = -ENOMEM;
+            goto out;
+        }
+        i++;
+    }
+    ret = 0;
+
+out:
+    GF_FREE(ptr1);
+    return ret;
 }
 
+void
+afr_ta_init(afr_private_t *priv)
+{
+    priv->thin_arbiter_count = 1;
+    priv->child_count--;
+    priv->ta_child_up = 0;
+    priv->ta_bad_child_index = AFR_CHILD_UNKNOWN;
+    priv->ta_notify_dom_lock_offset = 0;
+    priv->ta_in_mem_txn_count = 0;
+    priv->ta_on_wire_txn_count = 0;
+    priv->release_ta_notify_dom_lock = _gf_false;
+    INIT_LIST_HEAD(&priv->ta_waitq);
+    INIT_LIST_HEAD(&priv->ta_onwireq);
+    gf_uuid_clear(priv->ta_gfid);
+}
 
-static const char *favorite_child_warning_str = "You have specified subvolume '%s' "
-	"as the 'favorite child'. This means that if a discrepancy in the content "
-	"or attributes (ownership, permission, etc.) of a file is detected among "
-	"the subvolumes, the file on '%s' will be considered the definitive "
-	"version and its contents will OVERWRITE the contents of the file on other "
-	"subvolumes. All versions of the file except that on '%s' "
-	"WILL BE LOST.";
-
-static const char *no_lock_servers_warning_str = "You have set lock-server-count = 0. "
-	"This means correctness is NO LONGER GUARANTEED in all cases. If two or more "
-	"applications write to the same region of a file, there is a possibility that "
-	"its copies will be INCONSISTENT. Set it to a value greater than 0 unless you "
-	"are ABSOLUTELY SURE of what you are doing and WILL NOT HOLD GlusterFS "
-	"RESPONSIBLE for inconsistent data. If you are in doubt, set it to a value "
-	"greater than 0.";
-
-int32_t 
-init (xlator_t *this)
+int32_t
+init(xlator_t *this)
 {
-	afr_private_t * priv        = NULL;
-	int             child_count = 0;
-	xlator_list_t * trav        = NULL;
-	int             i           = 0;
-	int             ret         = -1;
-	int             op_errno    = 0;
-
-	char * read_subvol     = NULL;
-	char * fav_child       = NULL;
-	char * self_heal       = NULL;
-        char * algo            = NULL;
-	char * change_log      = NULL;
-	char * strict_readdir  = NULL;
-        char * inodelk_trace   = NULL;
-        char * entrylk_trace   = NULL;
-
-        int32_t background_count  = 0;
-	int32_t lock_server_count = 1;
-        int32_t window_size       = 0;
-
-	int    fav_ret       = -1;
-	int    read_ret      = -1;
-	int    dict_ret      = -1;
-
-
-	if (!this->children) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"replicate translator needs more than one "
-                        "subvolume defined.");
-		return -1;
-	}
-  
-	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"Volume is dangling.");
-	}
-
-
-	ALLOC_OR_GOTO (this->private, afr_private_t, out);
-
-	priv = this->private;
-
-	read_ret = dict_get_str (this->options, "read-subvolume", &read_subvol);
-	priv->read_child = -1;
-
-	fav_ret = dict_get_str (this->options, "favorite-child", &fav_child);
-	priv->favorite_child = -1;
-
-        priv->background_self_heal_count = 16;
-
-	dict_ret = dict_get_int32 (this->options, "background-self-heal-count",
-				   &background_count);
-	if (dict_ret == 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"Setting background self-heal count to %d",
-			background_count);
-
-		priv->background_self_heal_count = background_count;
-	}
-
-	/* Default values */
-
-	priv->data_self_heal     = 1;
-	priv->metadata_self_heal = 1;
-	priv->entry_self_heal    = 1;
-
-	dict_ret = dict_get_str (this->options, "data-self-heal", &self_heal);
-	if (dict_ret == 0) {
-		ret = gf_string2boolean (self_heal, &priv->data_self_heal);
-		if (ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"Invalid 'option data-self-heal %s'. "
-				"Defaulting to data-self-heal as 'on'",
-				self_heal);
-			priv->data_self_heal = 1;
-		}
-	}
-
-        priv->data_self_heal_algorithm = "";
-
-        dict_ret = dict_get_str (this->options, "data-self-heal-algorithm",
-                                 &algo);
-        if (dict_ret == 0) {
-                priv->data_self_heal_algorithm = gf_strdup (algo);
+    afr_private_t *priv = NULL;
+    int child_count = 0;
+    xlator_list_t *trav = NULL;
+    int i = 0;
+    int ret = -1;
+    GF_UNUSED int op_errno = 0;
+    xlator_t *read_subvol = NULL;
+    int read_subvol_index = -1;
+    char *qtype = NULL;
+    char *fav_child_policy = NULL;
+    char *thin_arbiter = NULL;
+    char *data_self_heal = NULL;
+    char *locking_scheme = NULL;
+    char *data_self_heal_algorithm = NULL;
+
+    if (!this->children) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_CHILD_MISCONFIGURED,
+               "replicate translator needs more than one "
+               "subvolume defined.");
+        return -1;
+    }
+
+    if (!this->parents) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_VOL_MISCONFIGURED,
+               "Volume is dangling.");
+    }
+
+    this->private = GF_CALLOC(1, sizeof(afr_private_t),
+                              gf_afr_mt_afr_private_t);
+    if (!this->private)
+        goto out;
+
+    priv = this->private;
+    INIT_LIST_HEAD(&priv->saved_locks);
+    INIT_LIST_HEAD(&priv->lk_healq);
+    LOCK_INIT(&priv->lock);
+
+    child_count = xlator_subvolume_count(this);
+
+    priv->child_count = child_count;
+
+    priv->read_child = -1;
+
+    GF_OPTION_INIT("arbiter-count", priv->arbiter_count, uint32, out);
+    GF_OPTION_INIT("thin-arbiter", thin_arbiter, str, out);
+    if (thin_arbiter && strlen(thin_arbiter) > 0) {
+        afr_ta_init(priv);
+    }
+    INIT_LIST_HEAD(&priv->healing);
+    INIT_LIST_HEAD(&priv->heal_waiting);
+
+    priv->spb_choice_timeout = AFR_DEFAULT_SPB_CHOICE_TIMEOUT;
+
+    GF_OPTION_INIT("afr-dirty-xattr", priv->afr_dirty, str, out);
+
+    GF_OPTION_INIT("metadata-splitbrain-forced-heal",
+                   priv->metadata_splitbrain_forced_heal, bool, out);
+
+    GF_OPTION_INIT("read-subvolume", read_subvol, xlator, out);
+    if (read_subvol) {
+        priv->read_child = xlator_subvolume_index(this, read_subvol);
+        if (priv->read_child == -1) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_INVALID_SUBVOL,
+                   "%s not a subvolume", read_subvol->name);
+            goto out;
+        }
+    }
+    GF_OPTION_INIT("read-subvolume-index", read_subvol_index, int32, out);
+    if (read_subvol_index > -1) {
+        if (read_subvol_index >= priv->child_count) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_INVALID_SUBVOL,
+                   "%d not a subvolume-index", read_subvol_index);
+            goto out;
         }
+        priv->read_child = read_subvol_index;
+    }
+    GF_OPTION_INIT("choose-local", priv->choose_local, bool, out);
+
+    priv->pending_reads = GF_CALLOC(sizeof(*priv->pending_reads),
+                                    priv->child_count, gf_afr_mt_atomic_t);
+
+    GF_OPTION_INIT("read-hash-mode", priv->hash_mode, uint32, out);
+
+    priv->favorite_child = -1;
+
+    GF_OPTION_INIT("favorite-child-policy", fav_child_policy, str, out);
+    if (afr_set_favorite_child_policy(priv, fav_child_policy) == -1)
+        goto out;
+
+    GF_OPTION_INIT("shd-max-threads", priv->shd.max_threads, uint32, out);
+
+    GF_OPTION_INIT("shd-wait-qlength", priv->shd.wait_qlength, uint32, out);
+
+    GF_OPTION_INIT("background-self-heal-count",
+                   priv->background_self_heal_count, uint32, out);
+
+    GF_OPTION_INIT("heal-wait-queue-length", priv->heal_wait_qlen, uint32, out);
+
+    GF_OPTION_INIT("data-self-heal", data_self_heal, str, out);
+    if (gf_string2boolean(data_self_heal, &priv->data_self_heal) == -1)
+        goto out;
+
+    GF_OPTION_INIT("data-self-heal-algorithm", data_self_heal_algorithm, str,
+                   out);
+    set_data_self_heal_algorithm(priv, data_self_heal_algorithm);
+
+    GF_OPTION_INIT("data-self-heal-window-size",
+                   priv->data_self_heal_window_size, uint32, out);
+
+    GF_OPTION_INIT("metadata-self-heal", priv->metadata_self_heal, bool, out);
+
+    GF_OPTION_INIT("entry-self-heal", priv->entry_self_heal, bool, out);
+
+    GF_OPTION_INIT("halo-shd-max-latency", priv->shd.halo_max_latency_msec,
+                   uint32, out);
+
+    GF_OPTION_INIT("halo-max-latency", priv->halo_max_latency_msec, uint32,
+                   out);
+    GF_OPTION_INIT("halo-max-replicas", priv->halo_max_replicas, uint32, out);
+    GF_OPTION_INIT("halo-min-replicas", priv->halo_min_replicas, uint32, out);
+
+    GF_OPTION_INIT("halo-enabled", priv->halo_enabled, bool, out);
+
+    GF_OPTION_INIT("halo-nfsd-max-latency", priv->nfsd.halo_max_latency_msec,
+                   uint32, out);
+
+    GF_OPTION_INIT("iam-nfs-daemon", priv->nfsd.iamnfsd, bool, out);
+
+    GF_OPTION_INIT("optimistic-change-log", priv->optimistic_change_log, bool,
+                   out);
+
+    GF_OPTION_INIT("pre-op-compat", priv->pre_op_compat, bool, out);
+    GF_OPTION_INIT("locking-scheme", locking_scheme, str, out);
+    priv->granular_locks = (strcmp(locking_scheme, "granular") == 0);
+    GF_OPTION_INIT("full-lock", priv->full_lock, bool, out);
+    GF_OPTION_INIT("granular-entry-heal", priv->esh_granular, bool, out);
+
+    GF_OPTION_INIT("eager-lock", priv->eager_lock, bool, out);
+    GF_OPTION_INIT("quorum-type", qtype, str, out);
+    GF_OPTION_INIT("quorum-count", priv->quorum_count, uint32, out);
+    GF_OPTION_INIT(AFR_SH_READDIR_SIZE_KEY, priv->sh_readdir_size, size_uint64,
+                   out);
+    fix_quorum_options(this, priv, qtype, this->options);
+
+    GF_OPTION_INIT("post-op-delay-secs", priv->post_op_delay_secs, uint32, out);
+    GF_OPTION_INIT("ensure-durability", priv->ensure_durability, bool, out);
+
+    GF_OPTION_INIT("self-heal-daemon", priv->shd.enabled, bool, out);
 
+    GF_OPTION_INIT("iam-self-heal-daemon", priv->shd.iamshd, bool, out);
+    GF_OPTION_INIT("heal-timeout", priv->shd.timeout, int32, out);
 
-        priv->data_self_heal_window_size = 16;
-
-	dict_ret = dict_get_int32 (this->options, "data-self-heal-window-size",
-				   &window_size);
-	if (dict_ret == 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"Setting data self-heal window size to %d",
-			window_size);
-
-		priv->data_self_heal_window_size = window_size;
-	}
-
-	dict_ret = dict_get_str (this->options, "metadata-self-heal",
-				 &self_heal);
-	if (dict_ret == 0) {
-		ret = gf_string2boolean (self_heal, &priv->metadata_self_heal);
-		if (ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"Invalid 'option metadata-self-heal %s'. "
-				"Defaulting to metadata-self-heal as 'on'.", 
-				self_heal);
-			priv->metadata_self_heal = 1;
-		} 
-	}
-
-	dict_ret = dict_get_str (this->options, "entry-self-heal", &self_heal);
-	if (dict_ret == 0) {
-		ret = gf_string2boolean (self_heal, &priv->entry_self_heal);
-		if (ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"Invalid 'option entry-self-heal %s'. "
-				"Defaulting to entry-self-heal as 'on'.", 
-				self_heal);
-			priv->entry_self_heal = 1;
-		} 
-	}
-
-	/* Change log options */
-
-	priv->data_change_log     = 1;
-	priv->metadata_change_log = 1;
-	priv->entry_change_log    = 1;
-
-	dict_ret = dict_get_str (this->options, "data-change-log",
-				 &change_log);
-	if (dict_ret == 0) {
-		ret = gf_string2boolean (change_log, &priv->data_change_log);
-		if (ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"Invalid 'option data-change-log %s'. "
-				"Defaulting to data-change-log as 'on'.", 
-				change_log);
-			priv->data_change_log = 1;
-		} 
-	}
-
-	dict_ret = dict_get_str (this->options, "metadata-change-log",
-				 &change_log);
-	if (dict_ret == 0) {
-		ret = gf_string2boolean (change_log,
-					 &priv->metadata_change_log);
-		if (ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"Invalid 'option metadata-change-log %s'. "
-				"Defaulting to metadata-change-log as 'off'.",
-				change_log);
-			priv->metadata_change_log = 0;
-		} 
-	}
-
-	dict_ret = dict_get_str (this->options, "entry-change-log",
-				 &change_log);
-	if (dict_ret == 0) {
-		ret = gf_string2boolean (change_log, &priv->entry_change_log);
-		if (ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"Invalid 'option entry-change-log %s'. "
-				"Defaulting to entry-change-log as 'on'.", 
-				change_log);
-			priv->entry_change_log = 1;
-		} 
-	}
-
-	/* Locking options */
-
-        priv->inodelk_trace = 0;
-        priv->entrylk_trace = 0;
-
-	dict_ret = dict_get_str (this->options, "inodelk-trace",
-				 &inodelk_trace);
-	if (dict_ret == 0) {
-		ret = gf_string2boolean (inodelk_trace, &priv->inodelk_trace);
-		if (ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"Invalid 'option inodelk-trace %s' ",
-                                inodelk_trace);
-
-			priv->inodelk_trace = 0;
-		}
-	}
-
-
-	dict_ret = dict_get_str (this->options, "entrylk-trace",
-				 &entrylk_trace);
-	if (dict_ret == 0) {
-		ret = gf_string2boolean (entrylk_trace, &priv->entrylk_trace);
-		if (ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"Invalid 'option entrylk-trace %s' ",
-                                inodelk_trace);
-
-			priv->entrylk_trace = 0;
-		}
-	}
-
-
-	priv->data_lock_server_count = 1;
-	priv->metadata_lock_server_count = 0;
-	priv->entry_lock_server_count = 1;
-
-	dict_ret = dict_get_int32 (this->options, "data-lock-server-count", 
-				   &lock_server_count);
-	if (dict_ret == 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"Setting data lock server count to %d.",
-			lock_server_count);
-
-		if (lock_server_count == 0) 
-			gf_log (this->name, GF_LOG_WARNING, "%s",
-                                no_lock_servers_warning_str);
-
-		priv->data_lock_server_count = lock_server_count;
-	}
-
-
-	dict_ret = dict_get_int32 (this->options,
-				   "metadata-lock-server-count", 
-				   &lock_server_count);
-	if (dict_ret == 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"Setting metadata lock server count to %d.",
-			lock_server_count);
-		priv->metadata_lock_server_count = lock_server_count;
-	}
-
-
-	dict_ret = dict_get_int32 (this->options, "entry-lock-server-count", 
-				   &lock_server_count);
-	if (dict_ret == 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"Setting entry lock server count to %d.",
-			lock_server_count);
-
-		priv->entry_lock_server_count = lock_server_count;
-	}
-
-	priv->strict_readdir = _gf_false;
-
-	dict_ret = dict_get_str (this->options, "strict-readdir",
-				 &strict_readdir);
-	if (dict_ret == 0) {
-		ret = gf_string2boolean (strict_readdir, &priv->strict_readdir);
-		if (ret < 0) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"Invalid 'option strict-readdir %s'. "
-				"Defaulting to strict-readdir as 'off'.",
-				strict_readdir);
-		}
-	}
-
-	trav = this->children;
-	while (trav) {
-		if (!read_ret && !strcmp (read_subvol, trav->xlator->name)) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"Subvolume '%s' specified as read child.",
-				trav->xlator->name);
-
-			priv->read_child = child_count;
-		}
-
-		if (fav_ret == 0 && !strcmp (fav_child, trav->xlator->name)) {
-			gf_log (this->name, GF_LOG_WARNING,
-				favorite_child_warning_str, trav->xlator->name,
-				trav->xlator->name, trav->xlator->name);
-			priv->favorite_child = child_count;
-		}
-
-		child_count++;
-		trav = trav->next;
-	}
-
-	priv->wait_count = 1;
-
-	priv->child_count = child_count;
-
-	LOCK_INIT (&priv->lock);
-        LOCK_INIT (&priv->read_child_lock);
-
-	priv->child_up = GF_CALLOC (sizeof (unsigned char), child_count,
+    GF_OPTION_INIT("consistent-metadata", priv->consistent_metadata, bool, out);
+    GF_OPTION_INIT("consistent-io", priv->consistent_io, bool, out);
+    afr_handle_anon_inode_options(priv, this->options);
+
+    GF_OPTION_INIT("use-anonymous-inode", priv->use_anon_inode, bool, out);
+    if (priv->quorum_count != 0)
+        priv->consistent_io = _gf_false;
+
+    priv->wait_count = 1;
+
+    priv->local = GF_CALLOC(sizeof(unsigned char), child_count, gf_afr_mt_char);
+    if (!priv->local) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    priv->anon_inode = GF_CALLOC(sizeof(unsigned char), child_count,
+                                 gf_afr_mt_char);
+
+    priv->child_up = GF_CALLOC(sizeof(unsigned char), child_count,
+                               gf_afr_mt_char);
+
+    priv->child_latency = GF_MALLOC(sizeof(*priv->child_latency) * child_count,
+                                    gf_afr_mt_child_latency_t);
+    priv->halo_child_up = GF_CALLOC(sizeof(unsigned char), child_count,
                                     gf_afr_mt_char);
-	if (!priv->child_up) {
-		gf_log (this->name, GF_LOG_ERROR,	
-			"Out of memory.");		
-		ret = -ENOMEM;
-		goto out;
-	}
-
-	priv->children = GF_CALLOC (sizeof (xlator_t *), child_count,
-                                    gf_afr_mt_xlator_t);
-	if (!priv->children) {
-		gf_log (this->name, GF_LOG_ERROR,	
-			"Out of memory.");		
-		ret = -ENOMEM;
-		goto out;
-	}
-
-        priv->pending_key = GF_CALLOC (sizeof (*priv->pending_key), 
-                                        child_count,
-                                        gf_afr_mt_char);
-        if (!priv->pending_key) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory.");
-                ret = -ENOMEM;
-                goto out;
+
+    if (!priv->child_up || !priv->child_latency || !priv->halo_child_up ||
+        !priv->anon_inode) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    /*Initialize to -ve ping timeout so that they are not considered
+     * in child-up events until ping-event comes*/
+    for (i = 0; i < child_count; i++)
+        priv->child_latency[i] = -1;
+
+    priv->children = GF_CALLOC(sizeof(xlator_t *), child_count,
+                               gf_afr_mt_xlator_t);
+    if (!priv->children) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ret = afr_pending_xattrs_init(priv, this);
+    if (ret)
+        goto out;
+
+    trav = this->children;
+    i = 0;
+    while (i < child_count) {
+        priv->children[i] = trav->xlator;
+        trav = trav->next;
+        i++;
+    }
+
+    ret = gf_asprintf(&priv->sh_domain, AFR_SH_DATA_DOMAIN_FMT, this->name);
+    if (-1 == ret) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    priv->last_event = GF_CALLOC(child_count, sizeof(*priv->last_event),
+                                 gf_afr_mt_int32_t);
+    if (!priv->last_event) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    this->itable = inode_table_new(SHD_INODE_LRU_LIMIT, this);
+    if (!this->itable) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    if (priv->shd.iamshd) {
+        ret = afr_selfheal_daemon_init(this);
+        if (ret) {
+            ret = -ENOMEM;
+            goto out;
         }
+    }
+
+    /* keep more local here as we may need them for self-heal etc */
+    this->local_pool = mem_pool_new(afr_local_t, 512);
+    if (!this->local_pool) {
+        ret = -1;
+        goto out;
+    }
+
+    priv->root_inode = NULL;
 
-	trav = this->children;
-	i = 0;
-	while (i < child_count) {
-		priv->children[i] = trav->xlator;
-
-                ret = gf_asprintf (&priv->pending_key[i], "%s.%s", 
-                                   AFR_XATTR_PREFIX,
-                                   trav->xlator->name);
-                if (-1 == ret) {
-                        gf_log (this->name, GF_LOG_ERROR, 
-                                "asprintf failed to set pending key");
-                        ret = -ENOMEM;
-                        goto out;
-                }
-
-		trav = trav->next;
-		i++;
-	}
-
-        LOCK_INIT (&priv->root_inode_lk);
-        priv->first_lookup = 1;
-        priv->root_inode = NULL;
-
-        pthread_mutex_init (&priv->mutex, NULL);
-        INIT_LIST_HEAD (&priv->saved_fds);
-
-	ret = 0;
+    ret = 0;
 out:
-	return ret;
+    return ret;
+}
+void
+afr_destroy_healer_object(xlator_t *this, struct subvol_healer *healer)
+{
+    int ret = -1;
+
+    if (!healer)
+        return;
+
+    if (healer->running) {
+        /*
+         * If there are any resources to cleanup, We need
+         * to do that gracefully using pthread_cleanup_push
+         */
+        ret = gf_thread_cleanup_xint(healer->thread);
+        if (ret)
+            gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_SELF_HEAL_FAILED,
+                   "Failed to clean up healer threads.");
+        healer->thread = 0;
+    }
+    pthread_cond_destroy(&healer->cond);
+    pthread_mutex_destroy(&healer->mutex);
 }
 
-
-int
-fini (xlator_t *this)
+void
+afr_selfheal_daemon_fini(xlator_t *this)
 {
-	return 0;
+    struct subvol_healer *healer = NULL;
+    afr_self_heald_t *shd = NULL;
+    afr_private_t *priv = NULL;
+    int i = 0;
+
+    priv = this->private;
+    if (!priv)
+        return;
+
+    shd = &priv->shd;
+    if (!shd->iamshd)
+        return;
+
+    for (i = 0; i < priv->child_count; i++) {
+        healer = &shd->index_healers[i];
+        afr_destroy_healer_object(this, healer);
+
+        healer = &shd->full_healers[i];
+        afr_destroy_healer_object(this, healer);
+
+        if (shd->statistics[i])
+            eh_destroy(shd->statistics[i]);
+    }
+    GF_FREE(shd->index_healers);
+    GF_FREE(shd->full_healers);
+    GF_FREE(shd->statistics);
+    if (shd->split_brain)
+        eh_destroy(shd->split_brain);
 }
+void
+fini(xlator_t *this)
+{
+    afr_private_t *priv = NULL;
+
+    priv = this->private;
+
+    afr_selfheal_daemon_fini(this);
+    GF_ASSERT(list_empty(&priv->saved_locks));
 
+    LOCK(&priv->lock);
+    if (priv->timer != NULL) {
+        gf_timer_call_cancel(this->ctx, priv->timer);
+        priv->timer = NULL;
+    }
+    UNLOCK(&priv->lock);
+
+    if (this->local_pool != NULL) {
+        mem_pool_destroy(this->local_pool);
+        this->local_pool = NULL;
+    }
+
+    this->private = NULL;
+    afr_priv_destroy(priv);
+    if (this->itable) {
+        inode_table_destroy(this->itable);
+        this->itable = NULL;
+    }
+
+    return;
+}
 
 struct xlator_fops fops = {
-	.lookup      = afr_lookup,
-	.open        = afr_open,
-	.lk          = afr_lk,
-	.flush       = afr_flush,
-	.statfs      = afr_statfs,
-	.fsync       = afr_fsync,
-	.fsyncdir    = afr_fsyncdir,
-	.xattrop     = afr_xattrop,
-	.fxattrop    = afr_fxattrop,
-	.inodelk     = afr_inodelk,
-	.finodelk    = afr_finodelk,
-	.entrylk     = afr_entrylk,
-	.fentrylk    = afr_fentrylk,
-
-	/* inode read */
-	.access      = afr_access,
-	.stat        = afr_stat,
-	.fstat       = afr_fstat,
-	.readlink    = afr_readlink,
-	.getxattr    = afr_getxattr,
-	.readv       = afr_readv,
-
-	/* inode write */
-	.writev      = afr_writev,
-	.truncate    = afr_truncate,
-	.ftruncate   = afr_ftruncate,
-	.setxattr    = afr_setxattr,
-        .setattr     = afr_setattr,
-	.fsetattr    = afr_fsetattr,
-	.removexattr = afr_removexattr,
-
-	/* dir read */
-	.opendir     = afr_opendir,
-	.readdir     = afr_readdir,
-	.readdirp    = afr_readdirp,
-
-	/* dir write */
-	.create      = afr_create,
-	.mknod       = afr_mknod,
-	.mkdir       = afr_mkdir,
-	.unlink      = afr_unlink,
-	.rmdir       = afr_rmdir,
-	.link        = afr_link,
-	.symlink     = afr_symlink,
-	.rename      = afr_rename,
+    .lookup = afr_lookup,
+    .lk = afr_lk,
+    .flush = afr_flush,
+    .statfs = afr_statfs,
+    .fsyncdir = afr_fsyncdir,
+    .inodelk = afr_inodelk,
+    .finodelk = afr_finodelk,
+    .entrylk = afr_entrylk,
+    .fentrylk = afr_fentrylk,
+    .ipc = afr_ipc,
+    .lease = afr_lease,
+
+    /* inode read */
+    .access = afr_access,
+    .stat = afr_stat,
+    .fstat = afr_fstat,
+    .readlink = afr_readlink,
+    .getxattr = afr_getxattr,
+    .fgetxattr = afr_fgetxattr,
+    .readv = afr_readv,
+    .seek = afr_seek,
+
+    /* inode write */
+    .writev = afr_writev,
+    .truncate = afr_truncate,
+    .ftruncate = afr_ftruncate,
+    .setxattr = afr_setxattr,
+    .fsetxattr = afr_fsetxattr,
+    .setattr = afr_setattr,
+    .fsetattr = afr_fsetattr,
+    .removexattr = afr_removexattr,
+    .fremovexattr = afr_fremovexattr,
+    .fallocate = afr_fallocate,
+    .discard = afr_discard,
+    .zerofill = afr_zerofill,
+    .xattrop = afr_xattrop,
+    .fxattrop = afr_fxattrop,
+    .fsync = afr_fsync,
+
+    /*inode open*/
+    .opendir = afr_opendir,
+    .open = afr_open,
+
+    /* dir read */
+    .readdir = afr_readdir,
+    .readdirp = afr_readdirp,
+
+    /* dir write */
+    .create = afr_create,
+    .mknod = afr_mknod,
+    .mkdir = afr_mkdir,
+    .unlink = afr_unlink,
+    .rmdir = afr_rmdir,
+    .link = afr_link,
+    .symlink = afr_symlink,
+    .rename = afr_rename,
 };
 
-
 struct xlator_dumpops dumpops = {
-        .priv       = afr_priv_dump,
+    .priv = afr_priv_dump,
 };
 
-
 struct xlator_cbks cbks = {
-	.release     = afr_release,
-	.releasedir  = afr_releasedir,
+    .release = afr_release,
+    .releasedir = afr_releasedir,
+    .forget = afr_forget,
 };
 
-
 struct volume_options options[] = {
-	{ .key  = {"read-subvolume" }, 
-	  .type = GF_OPTION_TYPE_XLATOR
-	},
-	{ .key  = {"favorite-child"}, 
-	  .type = GF_OPTION_TYPE_XLATOR
-	},
-        { .key  = {"background-self-heal-count"},
-          .type = GF_OPTION_TYPE_INT,
-          .min  = 0
-        },
-	{ .key  = {"data-self-heal"},  
-	  .type = GF_OPTION_TYPE_BOOL 
-	},
-        { .key  = {"data-self-heal-algorithm"},
-          .type = GF_OPTION_TYPE_STR
-        },
-        { .key  = {"data-self-heal-window-size"},
-          .type = GF_OPTION_TYPE_INT,
-          .min  = 1,
-          .max  = 1024
-        },
-	{ .key  = {"metadata-self-heal"},  
-	  .type = GF_OPTION_TYPE_BOOL
-	},
-	{ .key  = {"entry-self-heal"},  
-	  .type = GF_OPTION_TYPE_BOOL 
-	},
-	{ .key  = {"data-change-log"},  
-	  .type = GF_OPTION_TYPE_BOOL 
-	},
-	{ .key  = {"metadata-change-log"},  
-	  .type = GF_OPTION_TYPE_BOOL
-	},
-	{ .key  = {"entry-change-log"},  
-	  .type = GF_OPTION_TYPE_BOOL
-	},
-	{ .key  = {"data-lock-server-count"},  
-	  .type = GF_OPTION_TYPE_INT, 
-	  .min  = 0
-	},
-	{ .key  = {"metadata-lock-server-count"},  
-	  .type = GF_OPTION_TYPE_INT, 
-	  .min  = 0
-	},
-	{ .key  = {"entry-lock-server-count"},  
-	  .type = GF_OPTION_TYPE_INT,
-	  .min  = 0
-	},
-	{ .key  = {"strict-readdir"},
-	  .type = GF_OPTION_TYPE_BOOL,
-	},
-	{ .key  = {NULL} },
+    {.key = {"read-subvolume"},
+     .type = GF_OPTION_TYPE_XLATOR,
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "inode-read fops happen only on one of the bricks in "
+                    "replicate. Afr will prefer the one specified using "
+                    "this option if it is not stale. Option value must be "
+                    "one of the xlator names of the children. "
+                    "Ex: <volname>-client-0 till "
+                    "<volname>-client-<number-of-bricks - 1>"},
+    {.key = {"read-subvolume-index"},
+     .type = GF_OPTION_TYPE_INT,
+     .default_value = "-1",
+     .op_version = {2},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "inode-read fops happen only on one of the bricks in "
+                    "replicate. AFR will prefer the one specified using "
+                    "this option if it is not stale. allowed options"
+                    " include -1 till replica-count - 1"},
+    {.key = {"read-hash-mode"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 0,
+     .max = 5,
+     .default_value = "1",
+     .op_version = {2},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description =
+         "inode-read fops happen only on one of the bricks in "
+         "replicate. AFR will prefer the one computed using "
+         "the method specified using this option.\n"
+         "0 = first readable child of AFR, starting from 1st child.\n"
+         "1 = hash by GFID of file (all clients use "
+         "same subvolume).\n"
+         "2 = hash by GFID of file and client PID.\n"
+         "3 = brick having the least outstanding read requests.\n"
+         "4 = brick having the least network ping latency.\n"
+         "5 = Hybrid mode between 3 and 4, ie least value among "
+         "network-latency multiplied by outstanding-read-requests."},
+    {
+        .key = {"choose-local"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "true",
+        .op_version = {2},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        .description = "Choose a local subvolume (i.e. Brick) to read from"
+                       " if read-subvolume is not explicitly set.",
+    },
+    {.key = {"background-self-heal-count"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 0,
+     .max = 256,
+     .default_value = "8",
+     .validate = GF_OPT_VALIDATE_MIN,
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "This specifies the number of per client self-heal "
+                    "jobs that can perform parallel heals in the "
+                    "background."},
+    {.key = {"halo-shd-max-latency"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 99999,
+     .default_value = "99999",
+     .op_version = {GD_OP_VERSION_3_11_0},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate", "halo"},
+     .description = "Maximum latency for shd halo replication in msec."},
+    {.key = {"halo-enabled"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "False",
+     .op_version = {GD_OP_VERSION_3_11_0},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate", "halo"},
+     .description = "Enable Halo (geo) replication mode."},
+    {.key = {"halo-nfsd-max-latency"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 99999,
+     .default_value = "5",
+     .op_version = {GD_OP_VERSION_3_11_0},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate", "halo"},
+     .description = "Maximum latency for nfsd halo replication in msec."},
+    {.key = {"halo-max-latency"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = AFR_HALO_MAX_LATENCY,
+     .default_value = "5",
+     .op_version = {GD_OP_VERSION_3_11_0},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate", "halo"},
+     .description = "Maximum latency for halo replication in msec."},
+    {.key = {"halo-max-replicas"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 99999,
+     .default_value = "99999",
+     .op_version = {GD_OP_VERSION_3_11_0},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate", "halo"},
+     .description = "The maximum number of halo replicas; replicas"
+                    " beyond this value will be written asynchronously"
+                    "via the SHD."},
+    {.key = {"halo-min-replicas"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 99999,
+     .default_value = "2",
+     .op_version = {GD_OP_VERSION_3_11_0},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate", "halo"},
+     .description = "The minimmum number of halo replicas, before adding "
+                    "out of region replicas."},
+    {.key = {"heal-wait-queue-length"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 0,
+     .max = 10000, /*Around 100MB with sizeof(afr_local_t)= 10496 bytes*/
+     .default_value = "128",
+     .validate = GF_OPT_VALIDATE_MIN,
+     .op_version = {GD_OP_VERSION_3_7_10},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "This specifies the number of heals that can be queued"
+                    " for the parallel background self heal jobs."},
+    {.key = {"data-self-heal"},
+     .type = GF_OPTION_TYPE_STR,
+     .value = {"1", "on", "yes", "true", "enable", "0", "off", "no", "false",
+               "disable", "open"},
+     .default_value = "off",
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "Using this option we can enable/disable data "
+                    "self-heal on the file. \"open\" means data "
+                    "self-heal action will only be triggered by file "
+                    "open operations."},
+    {.key = {"data-self-heal-algorithm"},
+     .type = GF_OPTION_TYPE_STR,
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "Select between \"full\", \"diff\". The "
+                    "\"full\" algorithm copies the entire file from "
+                    "source to sink. The \"diff\" algorithm copies to "
+                    "sink only those blocks whose checksums don't match "
+                    "with those of source. If no option is configured "
+                    "the option is chosen dynamically as follows: "
+                    "If the file does not exist on one of the sinks "
+                    "or empty file exists or if the source file size is "
+                    "about the same as page size the entire file will "
+                    "be read and written i.e \"full\" algo, "
+                    "otherwise \"diff\" algo is chosen.",
+     .value = {"diff", "full"}},
+    {.key = {"data-self-heal-window-size"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 1024,
+     .default_value = "1",
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "Maximum number blocks per file for which self-heal "
+                    "process would be applied simultaneously."},
+    {.key = {"metadata-self-heal"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     /*.validate_fn = validate_replica*/
+     .description = "Using this option we can enable/disable metadata "
+                    "i.e. Permissions, ownerships, xattrs self-heal on "
+                    "the file/directory."},
+    {.key = {"entry-self-heal"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     /*.validate_fn = validate_replica*/
+     .description = "Using this option we can enable/disable entry "
+                    "self-heal on the directory."},
+    {.key = {"data-change-log"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "This option exists only for backward compatibility "
+                    "and configuring it doesn't have any effect"},
+    {.key = {"metadata-change-log"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "This option exists only for backward compatibility "
+                    "and configuring it doesn't have any effect"},
+    {.key = {"entry-change-log"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "This option exists only for backward compatibility "
+                    "and configuring it doesn't have any effect"},
+    {.key = {"optimistic-change-log"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .description = "Entry/Metadata fops will not perform "
+                    "pre fop changelog operations in afr transaction "
+                    "if this option is enabled."},
+    {.key = {"inodelk-trace"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = "Enabling this option logs inode lock/unlocks"},
+    {.key = {"entrylk-trace"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = "Enabling this option logs entry lock/unlocks"},
+    {.key = {"pre-op-compat"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .description = "Use separate pre-op xattrop() FOP rather than "
+                    "overloading xdata of the OP"},
+    {.key = {"eager-lock"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description =
+         "Enable/Disable eager lock for replica volume. "
+         "Lock phase of a transaction has two sub-phases. "
+         "First is an attempt to acquire locks in parallel by "
+         "broadcasting non-blocking lock requests. If lock "
+         "acquisition fails on any server, then the held locks "
+         "are unlocked and we revert to a blocking locks mode "
+         "sequentially on one server after another.  If this "
+         "option is enabled the initial broadcasting lock "
+         "request attempts to acquire a full lock on the entire file. "
+         "If this fails, we revert back to the sequential "
+         "\"regional\" blocking locks as before. In the case "
+         "where such an \"eager\" lock is granted in the "
+         "non-blocking phase, it gives rise to an opportunity "
+         "for optimization. i.e, if the next write transaction "
+         "on the same FD arrives before the unlock phase of "
+         "the first transaction, it \"takes over\" the full "
+         "file lock. Similarly if yet another data transaction "
+         "arrives before the unlock phase of the \"optimized\" "
+         "transaction, that in turn \"takes over\" the lock as "
+         "well. The actual unlock now happens at the end of "
+         "the last \"optimized\" transaction."
+
+    },
+    {.key = {"self-heal-daemon"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .op_version = {1},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE,
+     .tags = {"replicate"},
+     /*.validate_fn   = validate_replica_heal_enable_disable*/
+     .description = "This option applies to only self-heal-daemon. "
+                    "Index directory crawl and automatic healing of files "
+                    "will not be performed if this option is turned off."},
+    {.key = {"iam-self-heal-daemon"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = "This option differentiates if the replicate "
+                    "translator is running as part of self-heal-daemon "
+                    "or not."},
+    {.key = {"iam-nfs-daemon"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = "This option differentiates if the replicate "
+                    "translator is running as part of an NFS daemon "
+                    "or not."},
+    {
+        .key = {"quorum-type"},
+        .type = GF_OPTION_TYPE_STR,
+        .value = {"none", "auto", "fixed"},
+        .default_value = "none",
+        .op_version = {1},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        /*.option = quorum-type*/
+        .description = "If value is \"fixed\" only allow writes if "
+                       "quorum-count bricks are present.  If value is "
+                       "\"auto\" only allow writes if more than half of "
+                       "bricks, or exactly half including the first, are "
+                       "present.",
+    },
+    {
+        .key = {"quorum-count"},
+        .type = GF_OPTION_TYPE_INT,
+        .min = 1,
+        .max = INT_MAX,
+        .default_value = 0,
+        .op_version = {1},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        /*.option = quorum-count*/
+        /*.validate_fn = validate_quorum_count*/
+        .description = "If quorum-type is \"fixed\" only allow writes if "
+                       "this many bricks are present.  Other quorum types "
+                       "will OVERWRITE this value.",
+    },
+    {
+        .key = {"quorum-reads"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "no",
+        .op_version = {GD_OP_VERSION_3_7_0},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        .description = "This option has been removed. Reads are not allowed "
+                       "if quorum is not met.",
+    },
+    {
+        .key = {"node-uuid"},
+        .type = GF_OPTION_TYPE_STR,
+        .description = "Local glusterd uuid string, used in starting "
+                       "self-heal-daemon so that it can crawl only on "
+                       "local index directories.",
+    },
+    {
+        .key = {"post-op-delay-secs"},
+        .type = GF_OPTION_TYPE_INT,
+        .min = 0,
+        .max = INT_MAX,
+        .default_value = "1",
+        .op_version = {2},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        .description = "Time interval induced artificially before "
+                       "post-operation phase of the transaction to "
+                       "enhance overlap of adjacent write operations.",
+    },
+    {
+        .key = {AFR_SH_READDIR_SIZE_KEY},
+        .type = GF_OPTION_TYPE_SIZET,
+        .description = "readdirp size for performing entry self-heal",
+        .min = 1024,
+        .max = 131072,
+        .op_version = {2},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE,
+        .tags = {"replicate"},
+        .default_value = "1KB",
+    },
+    {
+        .key = {"ensure-durability"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .op_version = {3},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        .description = "Afr performs fsyncs for transactions if this "
+                       "option is on to make sure the changelogs/data is "
+                       "written to the disk",
+        .default_value = "on",
+    },
+    {
+        .key = {"afr-dirty-xattr"},
+        .type = GF_OPTION_TYPE_STR,
+        .default_value = AFR_DIRTY_DEFAULT,
+    },
+    {.key = {"afr-pending-xattr"},
+     .type = GF_OPTION_TYPE_STR,
+     .description = "Comma separated list of xattrs that are used to  "
+                    "capture information on pending heals."},
+    {
+        .key = {"metadata-splitbrain-forced-heal"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "off",
+    },
+    {.key = {"heal-timeout"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 5,
+     .max = INT_MAX,
+     .default_value = "600",
+     .op_version = {2},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "time interval for checking the need to self-heal "
+                    "in self-heal-daemon"},
+    {
+        .key = {"consistent-metadata"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "no",
+        .op_version = {GD_OP_VERSION_3_7_0},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        .description = "If this option is enabled, readdirp will force "
+                       "lookups on those entries read whose read child is "
+                       "not the same as that of the parent. This will "
+                       "guarantee that all read operations on a file serve "
+                       "attributes from the same subvol as long as it holds "
+                       " a good copy of the file/dir.",
+    },
+    {.key = {"arbiter-count"},
+     .type = GF_OPTION_TYPE_INT,
+     .description = "subset of child_count. Has to be 0 or 1."},
+    {
+        .key = {"thin-arbiter"},
+        .type = GF_OPTION_TYPE_STR,
+        .op_version = {GD_OP_VERSION_4_1_0},
+        .flags = OPT_FLAG_SETTABLE,
+        .tags = {"replicate"},
+        .description = "contains host:path of thin abriter brick",
+    },
+    {.key = {"shd-max-threads"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 64,
+     .default_value = "1",
+     .op_version = {GD_OP_VERSION_3_7_12},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "Maximum number of parallel heals SHD can do per "
+                    "local brick. This can substantially lower heal times"
+                    ", but can also crush your bricks if you don't have "
+                    "the storage hardware to support this."},
+    {
+        .key = {"shd-wait-qlength"},
+        .type = GF_OPTION_TYPE_INT,
+        .min = 1,
+        .max = 655536,
+        .default_value = "1024",
+        .op_version = {GD_OP_VERSION_3_7_12},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        .description = "This option can be used to control number of heals"
+                       " that can wait in SHD per subvolume",
+    },
+    {
+        .key = {"locking-scheme"},
+        .type = GF_OPTION_TYPE_STR,
+        .value = {"full", "granular"},
+        .default_value = "full",
+        .op_version = {GD_OP_VERSION_3_7_12},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        .description = "If this option is set to granular, self-heal will "
+                       "stop being compatible with afr-v1, which helps afr "
+                       "be more granular while self-healing",
+    },
+    {.key = {"full-lock"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "yes",
+     .op_version = {GD_OP_VERSION_3_13_2},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE,
+     .tags = {"replicate"},
+     .description = "If this option is disabled, then the IOs will take "
+                    "range locks same as versions till 3.13.1."},
+    {
+        .key = {"granular-entry-heal"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "no",
+        .op_version = {GD_OP_VERSION_3_8_0},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        .description = "If this option is enabled, self-heal will resort to "
+                       "granular way of recording changelogs and doing entry "
+                       "self-heal.",
+    },
+    {
+        .key = {"favorite-child-policy"},
+        .type = GF_OPTION_TYPE_STR,
+        .value = {"none", "size", "ctime", "mtime", "majority"},
+        .default_value = "none",
+        .op_version = {GD_OP_VERSION_3_7_12},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"replicate"},
+        .description = "This option can be used to automatically resolve "
+                       "split-brains using various policies without user "
+                       "intervention. \"size\" picks the file with the "
+                       "biggest size as the source. \"ctime\" and \"mtime\" "
+                       "pick the file with the latest ctime and mtime "
+                       "respectively as the source. \"majority\" picks a file"
+                       " with identical mtime and size in more than half the "
+                       "number of bricks in the replica.",
+    },
+    {
+        .key = {"consistent-io"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "no",
+        .description = "If this option is enabled, i/o will fail even if "
+                       "one of the bricks is down in the replicas",
+    },
+    {.key = {"use-compound-fops"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "no",
+     .op_version = {GD_OP_VERSION_3_8_4},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"replicate"},
+     .description = "This option exists only for backward compatibility "
+                    "and configuring it doesn't have any effect"},
+    {.key = {"use-anonymous-inode"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "no",
+     .op_version = {GD_OP_VERSION_8_0},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE,
+     .tags = {"replicate"},
+     .description = "Setting this option heals directory renames efficiently"},
+
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .dumpops = &dumpops,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "replicate",
+    .category = GF_MAINTAINED,
 };
diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h
index 9593fec3c3f..d62f9a9caf2 100644
--- a/xlators/cluster/afr/src/afr.h
+++ b/xlators/cluster/afr/src/afr.h
@@ -1,922 +1,1423 @@
 /*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
 
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
 
 #ifndef __AFR_H__
 #define __AFR_H__
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "call-stub.h"
-#include "compat-errno.h"
+#include <glusterfs/call-stub.h>
+#include <glusterfs/compat-errno.h>
 #include "afr-mem-types.h"
 
-#define AFR_XATTR_PREFIX "trusted.afr"
+#include "libxlator.h"
+#include <glusterfs/timer.h>
+#include <glusterfs/syncop.h>
+
+#include "afr-self-heald.h"
+#include "afr-messages.h"
+
+#define SHD_INODE_LRU_LIMIT 1
+#define AFR_PATHINFO_HEADER "REPLICATE:"
+#define AFR_SH_READDIR_SIZE_KEY "self-heal-readdir-size"
+#define AFR_SH_DATA_DOMAIN_FMT "%s:self-heal"
+#define AFR_DIRTY_DEFAULT AFR_XATTR_PREFIX ".dirty"
+#define AFR_DIRTY (((afr_private_t *)(THIS->private))->afr_dirty)
+
+#define AFR_LOCKEE_COUNT_MAX 3
+#define AFR_DOM_COUNT_MAX 3
+#define AFR_NUM_CHANGE_LOGS 3              /*data + metadata + entry*/
+#define AFR_DEFAULT_SPB_CHOICE_TIMEOUT 300 /*in seconds*/
+
+#define ARBITER_BRICK_INDEX 2
+#define THIN_ARBITER_BRICK_INDEX 2
+#define AFR_TA_DOM_NOTIFY "afr.ta.dom-notify"
+#define AFR_TA_DOM_MODIFY "afr.ta.dom-modify"
+
+#define AFR_LK_HEAL_DOM "afr.lock-heal.domain"
+
+#define AFR_HALO_MAX_LATENCY 99999
+#define AFR_ANON_DIR_PREFIX ".glusterfs-anonymous-inode"
+
+#define PFLAG_PENDING (1 << 0)
+#define PFLAG_SBRAIN (1 << 1)
+
+typedef int (*afr_lock_cbk_t)(call_frame_t *frame, xlator_t *this);
+
+typedef int (*afr_read_txn_wind_t)(call_frame_t *frame, xlator_t *this,
+                                   int subvol);
+
+typedef int (*afr_inode_refresh_cbk_t)(call_frame_t *frame, xlator_t *this,
+                                       int err);
+
+typedef int (*afr_changelog_resume_t)(call_frame_t *frame, xlator_t *this);
+
+#define AFR_COUNT(array, max)                                                  \
+    ({                                                                         \
+        int __i;                                                               \
+        int __res = 0;                                                         \
+        for (__i = 0; __i < max; __i++)                                        \
+            if (array[__i])                                                    \
+                __res++;                                                       \
+        __res;                                                                 \
+    })
+#define AFR_INTERSECT(dst, src1, src2, max)                                    \
+    ({                                                                         \
+        int __i;                                                               \
+        for (__i = 0; __i < max; __i++)                                        \
+            dst[__i] = src1[__i] && src2[__i];                                 \
+    })
+#define AFR_CMP(a1, a2, len)                                                   \
+    ({                                                                         \
+        int __cmp = 0;                                                         \
+        int __i;                                                               \
+        for (__i = 0; __i < len; __i++)                                        \
+            if (a1[__i] != a2[__i]) {                                          \
+                __cmp = 1;                                                     \
+                break;                                                         \
+            }                                                                  \
+        __cmp;                                                                 \
+    })
+#define AFR_IS_ARBITER_BRICK(priv, index)                                      \
+    ((priv->arbiter_count == 1) && (index == ARBITER_BRICK_INDEX))
+
+#define AFR_SET_ERROR_AND_CHECK_SPLIT_BRAIN(ret, errnum)                       \
+    do {                                                                       \
+        local->op_ret = ret;                                                   \
+        local->op_errno = errnum;                                              \
+        if (local->op_errno == EIO)                                            \
+            gf_msg(this->name, GF_LOG_ERROR, local->op_errno,                  \
+                   AFR_MSG_SPLIT_BRAIN,                                        \
+                   "Failing %s on gfid %s: "                                   \
+                   "split-brain observed.",                                    \
+                   gf_fop_list[local->op], uuid_utoa(local->inode->gfid));     \
+    } while (0)
+
+#define AFR_ERROR_OUT_IF_FDCTX_INVALID(__fd, __this, __error, __label)         \
+    do {                                                                       \
+        afr_fd_ctx_t *__fd_ctx = NULL;                                         \
+        __fd_ctx = afr_fd_ctx_get(__fd, __this);                               \
+        if (__fd_ctx && __fd_ctx->is_fd_bad) {                                 \
+            __error = EBADF;                                                   \
+            goto __label;                                                      \
+        }                                                                      \
+    } while (0)
 
-struct _pump_private;
+typedef enum {
+    AFR_READ_POLICY_FIRST_UP,
+    AFR_READ_POLICY_GFID_HASH,
+    AFR_READ_POLICY_GFID_PID_HASH,
+    AFR_READ_POLICY_LESS_LOAD,
+    AFR_READ_POLICY_LEAST_LATENCY,
+    AFR_READ_POLICY_LOAD_LATENCY_HYBRID,
+} afr_read_hash_mode_t;
 
-typedef struct _afr_private {
-	gf_lock_t lock;               /* to guard access to child_count, etc */
-	unsigned int child_count;     /* total number of children   */
+typedef enum {
+    AFR_FAV_CHILD_NONE,
+    AFR_FAV_CHILD_BY_SIZE,
+    AFR_FAV_CHILD_BY_CTIME,
+    AFR_FAV_CHILD_BY_MTIME,
+    AFR_FAV_CHILD_BY_MAJORITY,
+    AFR_FAV_CHILD_POLICY_MAX,
+} afr_favorite_child_policy;
 
-        unsigned int read_child_rr;   /* round-robin index of the read_child */
-        gf_lock_t read_child_lock;    /* lock to protect above */
+typedef enum {
+    AFR_SELFHEAL_DATA_FULL = 0,
+    AFR_SELFHEAL_DATA_DIFF,
+    AFR_SELFHEAL_DATA_DYNAMIC,
+} afr_data_self_heal_type_t;
 
-	xlator_t **children;
+typedef enum {
+    AFR_CHILD_UNKNOWN = -1,
+    AFR_CHILD_ZERO,
+    AFR_CHILD_ONE,
+    AFR_CHILD_THIN_ARBITER,
+} afr_child_index;
 
-        gf_lock_t root_inode_lk;
-        int first_lookup;
-        inode_t *root_inode;
+typedef enum {
+    TA_WAIT_FOR_NOTIFY_LOCK_REL, /*FOP came after notify domain lock upcall
+                                   notification and waiting for its release.*/
+    TA_GET_INFO_FROM_TA_FILE,    /*FOP needs post-op on ta file to get
+                                  *info about which brick is bad.*/
+    TA_INFO_IN_MEMORY_SUCCESS,   /*Bad brick info is in memory and fop failed
+                                  *on BAD brick - Success*/
+    TA_INFO_IN_MEMORY_FAILED,    /*Bad brick info is in memory and fop failed
+                                  *on GOOD brick - Failed*/
+    TA_SUCCESS,                  /*FOP succeeded on both data bricks.*/
+} afr_ta_fop_state_t;
+
+struct afr_nfsd {
+    uint32_t halo_max_latency_msec;
+    gf_boolean_t iamnfsd;
+};
+
+typedef struct _afr_lk_heal_info {
+    fd_t *fd;
+    int32_t cmd;
+    struct gf_flock flock;
+    dict_t *xdata_req;
+    unsigned char *locked_nodes;
+    struct list_head pos;
+    gf_lkowner_t lk_owner;
+    pid_t pid;
+    int32_t *child_up_event_gen;
+    int32_t *child_down_event_gen;
+} afr_lk_heal_info_t;
 
-	unsigned char *child_up;
+typedef struct _afr_private {
+    gf_lock_t lock;             /* to guard access to child_count, etc */
+    unsigned int child_count;   /* total number of children   */
+    unsigned int arbiter_count; /*subset of child_count.
+                                  Has to be 0 or 1.*/
+
+    xlator_t **children;
+
+    inode_t *root_inode;
+
+    int favorite_child; /* subvolume to be preferred in resolving
+                                    split-brain cases */
+    /* For thin-arbiter. */
+    uuid_t ta_gfid;
+    unsigned int thin_arbiter_count; /* 0 or 1 at the moment.*/
+    int ta_bad_child_index;
+    int ta_event_gen;
+    unsigned int ta_in_mem_txn_count;
+    unsigned int ta_on_wire_txn_count;
+    struct list_head ta_waitq;
+    struct list_head ta_onwireq;
+
+    unsigned char *anon_inode;
+    unsigned char *child_up;
+    unsigned char *halo_child_up;
+    int64_t *child_latency;
+    unsigned char *local;
+
+    char **pending_key;
+
+    afr_data_self_heal_type_t data_self_heal_algorithm;
+    unsigned int data_self_heal_window_size; /* max number of pipelined
+                                                read/writes */
+
+    struct list_head heal_waiting; /*queue for files that need heal*/
+    uint32_t heal_wait_qlen; /*configurable queue length for heal_waiting*/
+    int32_t heal_waiters;    /* No. of elements currently in wait queue.*/
+
+    struct list_head healing;            /* queue for files that are undergoing
+                                            background heal*/
+    uint32_t background_self_heal_count; /*configurable queue length for
+                                           healing queue*/
+    int32_t healers; /* No. of elements currently undergoing background
+                      heal*/
+
+    gf_boolean_t release_ta_notify_dom_lock;
+
+    gf_boolean_t metadata_self_heal; /* on/off */
+    gf_boolean_t entry_self_heal;    /* on/off */
+
+    gf_boolean_t metadata_splitbrain_forced_heal; /* on/off */
+    int read_child;                               /* read-subvolume */
+    gf_atomic_t *pending_reads; /*No. of pending read cbks per child.*/
+
+    gf_timer_t *timer; /* launched when parent up is received */
+
+    unsigned int wait_count; /* # of servers to wait for success */
+
+    unsigned char ta_child_up;
+    gf_boolean_t optimistic_change_log;
+    gf_boolean_t eager_lock;
+    gf_boolean_t pre_op_compat; /* on/off */
+    uint32_t post_op_delay_secs;
+    unsigned int quorum_count;
+
+    off_t ta_notify_dom_lock_offset;
+    afr_favorite_child_policy fav_child_policy; /*Policy to use for automatic
+                                              resolution of split-brains.*/
+    afr_read_hash_mode_t hash_mode; /* for when read_child is not set */
+
+    int32_t *last_event;
+
+    /* @event_generation: Keeps count of number of events received which can
+       potentially impact consistency decisions. The events are CHILD_UP
+       and CHILD_DOWN, when we have to recalculate the freshness/staleness
+       of copies to detect if changes had happened while the other server
+       was down. CHILD_DOWN and CHILD_UP can also be received on network
+       disconnect/reconnects and not necessarily server going down/up.
+       Recalculating freshness/staleness on network events is equally
+       important as we might have had a network split brain.
+    */
+    uint32_t event_generation;
+    char vol_uuid[UUID_SIZE + 1];
+
+    gf_boolean_t choose_local;
+    gf_boolean_t did_discovery;
+    gf_boolean_t ensure_durability;
+    gf_boolean_t halo_enabled;
+    gf_boolean_t consistent_metadata;
+    gf_boolean_t need_heal;
+    gf_boolean_t granular_locks;
+    uint64_t sh_readdir_size;
+    char *sh_domain;
+    char *afr_dirty;
+
+    uint64_t spb_choice_timeout;
+
+    afr_self_heald_t shd;
+    struct afr_nfsd nfsd;
+
+    uint32_t halo_max_latency_msec;
+    uint32_t halo_max_replicas;
+    uint32_t halo_min_replicas;
+
+    gf_boolean_t full_lock;
+    gf_boolean_t esh_granular;
+    gf_boolean_t consistent_io;
+    gf_boolean_t data_self_heal; /* on/off */
+    gf_boolean_t use_anon_inode;
+
+    /*For lock healing.*/
+    struct list_head saved_locks;
+    struct list_head lk_healq;
+
+    /*For anon-inode handling */
+    char anon_inode_name[NAME_MAX + 1];
+    char anon_gfid_str[UUID_SIZE + 1];
+} afr_private_t;
 
-        char **pending_key;
+typedef enum {
+    AFR_DATA_TRANSACTION,         /* truncate, write, ... */
+    AFR_METADATA_TRANSACTION,     /* chmod, chown, ... */
+    AFR_ENTRY_TRANSACTION,        /* create, rmdir, ... */
+    AFR_ENTRY_RENAME_TRANSACTION, /* rename */
+} afr_transaction_type;
 
-	gf_boolean_t data_self_heal;              /* on/off */
-        char *       data_self_heal_algorithm;    /* name of algorithm */
-        unsigned int data_self_heal_window_size;  /* max number of pipelined
-                                                     read/writes */
+/*
+  xattr format: trusted.afr.volume = [x y z]
+  x - data pending
+  y - metadata pending
+  z - entry pending
+*/
 
-        unsigned int background_self_heal_count;
-        unsigned int background_self_heals_started;
-	gf_boolean_t metadata_self_heal;   /* on/off */
-	gf_boolean_t entry_self_heal;      /* on/off */
+static inline int
+afr_index_for_transaction_type(afr_transaction_type type)
+{
+    switch (type) {
+        case AFR_DATA_TRANSACTION:
+            return 0;
 
-	gf_boolean_t data_change_log;       /* on/off */
-	gf_boolean_t metadata_change_log;   /* on/off */
-	gf_boolean_t entry_change_log;      /* on/off */
+        case AFR_METADATA_TRANSACTION:
+            return 1;
 
-	int read_child;               /* read-subvolume */
-	unsigned int favorite_child;  /* subvolume to be preferred in resolving
-					 split-brain cases */
+        case AFR_ENTRY_TRANSACTION:
+        case AFR_ENTRY_RENAME_TRANSACTION:
+            return 2;
+    }
 
-	unsigned int data_lock_server_count;
-	unsigned int metadata_lock_server_count;
-	unsigned int entry_lock_server_count;
+    return -1; /* make gcc happy */
+}
 
-	gf_boolean_t inodelk_trace;
-	gf_boolean_t entrylk_trace;
+static inline int
+afr_index_from_ia_type(ia_type_t type)
+{
+    switch (type) {
+        case IA_IFDIR:
+            return afr_index_for_transaction_type(AFR_ENTRY_TRANSACTION);
+        case IA_IFREG:
+            return afr_index_for_transaction_type(AFR_DATA_TRANSACTION);
+        default:
+            return -1;
+    }
+}
 
-	gf_boolean_t strict_readdir;
+typedef struct {
+    struct gf_flock flock;
+    loc_t loc;
+    fd_t *fd;
+    char *basename;
+    unsigned char *locked_nodes;
+    int locked_count;
 
-	unsigned int wait_count;      /* # of servers to wait for success */
+} afr_lockee_t;
 
-        uint64_t up_count;      /* number of CHILD_UPs we have seen */
-        uint64_t down_count;    /* number of CHILD_DOWNs we have seen */
+int
+afr_entry_lockee_cmp(const void *l1, const void *l2);
 
-	struct _pump_private *pump_private; /* Set if we are loaded as pump */
-        int                   use_afr_in_pump;
+typedef struct {
+    loc_t *lk_loc;
 
-        pthread_mutex_t  mutex;
-        struct list_head saved_fds;   /* list of fds on which locks have succeeded */
-} afr_private_t;
+    afr_lockee_t lockee[AFR_LOCKEE_COUNT_MAX];
 
-typedef struct {
-        /* External interface: These are variables (some optional) that
-           are set by whoever has triggered self-heal */
+    const char *lk_basename;
+    const char *lower_basename;
+    const char *higher_basename;
 
-        gf_boolean_t need_data_self_heal;
-        gf_boolean_t need_metadata_self_heal;
-        gf_boolean_t need_entry_self_heal;
+    unsigned char *lower_locked_nodes;
 
-        gf_boolean_t forced_merge;        /* Is this a self-heal triggered to
-                                             forcibly merge the directories? */
+    afr_lock_cbk_t lock_cbk;
 
-        gf_boolean_t healing_fd_opened;   /* true if caller has already
-                                             opened fd */
+    int lockee_count;
 
-        gf_boolean_t data_lock_held;      /* true if caller has already
-                                             acquired 0-0 lock */
+    int32_t lk_call_count;
+    int32_t lk_expected_count;
+    int32_t lk_attempted_count;
 
-	fd_t *healing_fd;                 /* set if callers has opened fd */
+    int32_t lock_op_ret;
+    int32_t lock_op_errno;
+    char *domain; /* Domain on which inode/entry lock/unlock in progress.*/
+    int32_t lock_count;
+    char lower_locked;
+    char higher_locked;
+} afr_internal_lock_t;
 
-        gf_boolean_t background;          /* do self-heal in background
-                                             if possible */
+struct afr_reply {
+    int valid;
+    int32_t op_ret;
+    dict_t *xattr; /*For xattrop*/
+    dict_t *xdata;
+    struct iatt poststat;
+    struct iatt postparent;
+    struct iatt prestat;
+    struct iatt preparent;
+    struct iatt preparent2;
+    struct iatt postparent2;
+    int32_t op_errno;
+    /* For rchecksum */
+    uint8_t checksum[SHA256_DIGEST_LENGTH];
+    gf_boolean_t buf_has_zeroes;
+    gf_boolean_t fips_mode_rchecksum;
+    /* For lookup */
+    int8_t need_heal;
+};
 
-        ia_type_t type;                   /* st_mode of the entry we're doing
-                                             self-heal on */
+typedef enum {
+    AFR_FD_NOT_OPENED,
+    AFR_FD_OPENED,
+    AFR_FD_OPENING
+} afr_fd_open_status_t;
 
-        /* Function to call to unwind. If self-heal is being done in the
-           background, this function will be called as soon as possible. */
+typedef struct {
+    afr_fd_open_status_t *opened_on; /* which subvolumes the fd is open on */
+    int flags;
+
+    /* the subvolume on which the latest sequence of readdirs (starting
+       at offset 0) has begun. Till the next readdir request with 0 offset
+       arrives, we continue to read off this subvol.
+    */
+    int readdir_subvol;
+    /* lock-healing related members. */
+    gf_boolean_t is_fd_bad;
+    afr_lk_heal_info_t *lk_heal_info;
 
-        int (*unwind) (call_frame_t *frame, xlator_t *this);
+} afr_fd_ctx_t;
 
-        /* End of external interface members */
+typedef enum {
+    AFR_FOP_LOCK_PARALLEL,
+    AFR_FOP_LOCK_SERIAL,
+    AFR_FOP_LOCK_QUORUM_FAILED,
+} afr_fop_lock_state_t;
+
+typedef struct _afr_inode_lock_t {
+    /* @num_inodelks:
+       Number of inodelks queried from the server, as queried through
+       xdata in FOPs. Currently, used to decide if eager-locking must be
+       temporarily disabled.
+    */
+    int32_t num_inodelks;
+    unsigned int event_generation;
+    gf_timer_t *delay_timer;
+    struct list_head owners;  /*Transactions that are performing fop*/
+    struct list_head post_op; /*Transactions that are done with the fop
+                               *So can not conflict with the fops*/
+    struct list_head waiting; /*Transaction that are waiting for
+                               *conflicting transactions to complete*/
+    struct list_head frozen;  /*Transactions that need to go as part of
+                               * next batch of eager-lock*/
+    gf_boolean_t release;
+    gf_boolean_t acquired;
+} afr_lock_t;
+
+typedef struct _afr_inode_ctx {
+    uint64_t read_subvol;
+    uint64_t write_subvol;
+    int lock_count;
+    int spb_choice;
+    gf_timer_t *timer;
+    unsigned int *pre_op_done[AFR_NUM_CHANGE_LOGS];
+    int inherited[AFR_NUM_CHANGE_LOGS];
+    int on_disk[AFR_NUM_CHANGE_LOGS];
+    /*Only 2 types of transactions support eager-locks now. DATA/METADATA*/
+    afr_lock_t lock[2];
+
+    /* @open_fd_count:
+       Number of open FDs queried from the server, as queried through
+       xdata in FOPs. Currently, used to decide if eager-locking must be
+       temporarily disabled.
+    */
+    uint32_t open_fd_count;
+    gf_boolean_t need_refresh;
+
+    /* set if any write on this fd was a non stable write
+       (i.e, without O_SYNC or O_DSYNC)
+    */
+    gf_boolean_t witnessed_unstable_write;
+} afr_inode_ctx_t;
 
+typedef struct _afr_local {
+    glusterfs_fop_t op;
+    unsigned int call_count;
 
-	/* array of stat's, one for each child */
-	struct iatt *buf;
-        struct iatt parentbuf;
+    /* @event_generation: copy of priv->event_generation taken at the
+       time of starting the transaction. The copy is made so that we
+       have a stable value through the various phases of the transaction.
+    */
+    unsigned int event_generation;
 
-	/* array of xattr's, one for each child */
-	dict_t **xattr;
+    uint32_t open_fd_count;
+    int32_t num_inodelks;
 
-	/* array of errno's, one for each child */
-	int *child_errno;
+    int32_t op_ret;
+    int32_t op_errno;
 
-	int32_t **pending_matrix;
-	int32_t **delta_matrix;
+    int dirty[AFR_NUM_CHANGE_LOGS];
 
-	int *sources;
-	int source;
-	int active_source;
-	int active_sinks;
-	int *success;
-	unsigned char *locked_nodes;
-        int lock_count;
+    int32_t **pending;
 
-        mode_t impunging_entry_mode;
-        const char *linkname;
+    loc_t loc;
+    loc_t newloc;
 
-	int   op_failed;
+    fd_t *fd;
+    afr_fd_ctx_t *fd_ctx;
 
-	int   file_has_holes;
-	blksize_t block_size;
-	off_t file_size;
-	off_t offset;
+    /* @child_up: copy of priv->child_up taken at the time of transaction
+       start. The copy is taken so that we have a stable child_up array
+       through the phases of the transaction as priv->child_up[i] can keep
+       changing through time.
+    */
+    unsigned char *child_up;
 
-	loc_t parent_loc;
+    /* @read_attempted:
+       array of flags representing subvolumes where read operations of
+       the read transaction have already been attempted. The array is
+       first pre-filled with down subvolumes, and as reads are performed
+       on other subvolumes, those are set as well. This way if the read
+       operation fails we do not retry on that subvolume again.
+    */
+    unsigned char *read_attempted;
 
-        call_frame_t *orig_frame;
-        gf_boolean_t unwound;
+    /* @readfn:
 
-        /* private data for the particular self-heal algorithm */
-        void *private;
+       pointer to function which will perform the read operation on a given
+       subvolume. Used in read transactions.
+    */
 
-        int (*flush_self_heal_cbk) (call_frame_t *frame, xlator_t *this);
+    afr_read_txn_wind_t readfn;
 
-	int (*completion_cbk) (call_frame_t *frame, xlator_t *this);
-        int (*algo_completion_cbk) (call_frame_t *frame, xlator_t *this);
-        int (*algo_abort_cbk) (call_frame_t *frame, xlator_t *this);
+    /* @inode:
 
-	call_frame_t *sh_frame;
-} afr_self_heal_t;
+       the inode on which the read txn is performed on. ref'ed and copied
+       from either fd->inode or loc.inode
+    */
 
+    inode_t *inode;
 
-typedef enum {
-	AFR_DATA_TRANSACTION,          /* truncate, write, ... */
-	AFR_METADATA_TRANSACTION,      /* chmod, chown, ... */
-	AFR_ENTRY_TRANSACTION,         /* create, rmdir, ... */
-	AFR_ENTRY_RENAME_TRANSACTION,  /* rename */
-} afr_transaction_type;
+    /* @parent[2]:
 
-typedef enum {
-        AFR_TRANSACTION_LK,
-        AFR_SELFHEAL_LK,
-} transaction_lk_type_t;
+       parent inode[s] on which directory transactions are performed.
+    */
 
-typedef enum {
-        AFR_LOCK_OP,
-        AFR_UNLOCK_OP,
-} afr_lock_op_type_t;
+    inode_t *parent;
+    inode_t *parent2;
 
-typedef enum {
-        AFR_DATA_SELF_HEAL_LK,
-        AFR_METADATA_SELF_HEAL_LK,
-        AFR_ENTRY_SELF_HEAL_LK,
-}selfheal_lk_type_t;
+    /* @readable:
 
-typedef enum {
-        AFR_INODELK_TRANSACTION,
-        AFR_INODELK_NB_TRANSACTION,
-        AFR_ENTRYLK_TRANSACTION,
-        AFR_ENTRYLK_NB_TRANSACTION,
-        AFR_INODELK_SELFHEAL,
-        AFR_INODELK_NB_SELFHEAL,
-        AFR_ENTRYLK_SELFHEAL,
-        AFR_ENTRYLK_NB_SELFHEAL,
-} afr_lock_call_type_t;
+       array of flags representing servers from which a read can be
+       performed. This is the output of afr_inode_refresh()
+    */
+    unsigned char *readable;
+    unsigned char *readable2; /*For rename transaction*/
 
-/*
-  xattr format: trusted.afr.volume = [x y z]
-  x - data pending
-  y - metadata pending
-  z - entry pending
-*/
+    afr_inode_refresh_cbk_t refreshfn;
 
-static inline int
-afr_index_for_transaction_type (afr_transaction_type type)
-{
-        switch (type) {
+    /* @refreshinode:
 
-        case AFR_DATA_TRANSACTION:
-                return 0;
+       Inode currently getting refreshed.
+    */
+    inode_t *refreshinode;
 
-        case AFR_METADATA_TRANSACTION:
-                return 1;
+    dict_t *xattr_req;
 
-        case AFR_ENTRY_TRANSACTION:
-        case AFR_ENTRY_RENAME_TRANSACTION:
-                return 2;
-        }
+    dict_t *dict;
 
-        return -1;  /* make gcc happy */
-}
+    int read_subvol; /* Current read subvolume */
 
+    int optimistic_change_log;
 
-typedef struct {
-        loc_t *lk_loc;
-        struct gf_flock lk_flock;
+    afr_internal_lock_t internal_lock;
 
-        const char *lk_basename;
-        const char *lower_basename;
-        const char *higher_basename;
-        char lower_locked;
-        char higher_locked;
+    /*To handle setattr/setxattr on yet to be linked inode from dht*/
+    uuid_t refreshgfid;
 
-        unsigned char *locked_nodes;
-        unsigned char *lower_locked_nodes;
-        unsigned char *inode_locked_nodes;
-        unsigned char *entry_locked_nodes;
+    /* @refreshed:
 
-        selfheal_lk_type_t selfheal_lk_type;
-        transaction_lk_type_t transaction_lk_type;
+       the inode was "refreshed" (i.e, pending xattrs from all subvols
+       freshly inspected and inode ctx updated accordingly) as part of
+       this transaction already.
+    */
+    gf_boolean_t refreshed;
 
-        int32_t lock_count;
-        int32_t inodelk_lock_count;
-        int32_t entrylk_lock_count;
+    gf_boolean_t update_num_inodelks;
+    gf_boolean_t update_open_fd_count;
 
-        uint64_t lock_number;
-        int32_t lk_call_count;
+    /*
+      @pre_op_compat:
 
-        int32_t lock_op_ret;
-        int32_t lock_op_errno;
+      compatibility mode of pre-op. send a separate pre-op and
+      op operations as part of transaction, rather than combining
+    */
 
-        int (*lock_cbk) (call_frame_t *, xlator_t *);
+    gf_boolean_t pre_op_compat;
 
-} afr_internal_lock_t;
+    /* Is the current writev() going to perform a stable write?
+       i.e, is fd->flags or @flags writev param have O_SYNC or
+       O_DSYNC?
+    */
+    gf_boolean_t stable_write;
+
+    /* This write appended to the file. Nnot necessarily O_APPEND,
+       just means the offset of write was at the end of file.
+    */
+    gf_boolean_t append_write;
+
+    /*
+      This struct contains the arguments for the "continuation"
+      (scheme-like) of fops
+    */
+
+    struct {
+        struct {
+            struct statvfs buf;
+            unsigned char buf_set;
+        } statfs;
+
+        struct {
+            fd_t *fd;
+            int32_t flags;
+        } open;
+
+        struct {
+            struct gf_flock user_flock;
+            struct gf_flock ret_flock;
+            unsigned char *locked_nodes;
+            int32_t cmd;
+            /*For lock healing only.*/
+            unsigned char *dom_locked_nodes;
+            int32_t *dom_lock_op_ret;
+            int32_t *dom_lock_op_errno;
+            struct gf_flock *getlk_rsp;
+        } lk;
+
+        /* inode read */
+
+        struct {
+            int32_t mask;
+            int last_index; /* index of the child we tried previously */
+        } access;
+
+        struct {
+            int last_index;
+        } stat;
+
+        struct {
+            int last_index;
+        } fstat;
+
+        struct {
+            size_t size;
+            int last_index;
+        } readlink;
+
+        struct {
+            char *name;
+            long xattr_len;
+            int last_index;
+        } getxattr;
+
+        struct {
+            size_t size;
+            off_t offset;
+            int last_index;
+            uint32_t flags;
+        } readv;
+
+        /* dir read */
+
+        struct {
+            uint32_t *checksum;
+            int success_count;
+            int32_t op_ret;
+            int32_t op_errno;
+        } opendir;
+
+        struct {
+            int32_t op_ret;
+            int32_t op_errno;
+            size_t size;
+            off_t offset;
+            dict_t *dict;
+            int last_index;
+            gf_boolean_t failed;
+        } readdir;
+        /* inode write */
+
+        struct {
+            struct iatt prebuf;
+            struct iatt postbuf;
+        } inode_wfop;  // common structure for all inode-write-fops
+
+        struct {
+            struct iovec *vector;
+            struct iobref *iobref;
+            off_t offset;
+            int32_t op_ret;
+            int32_t count;
+            uint32_t flags;
+        } writev;
+
+        struct {
+            off_t offset;
+        } truncate;
+
+        struct {
+            off_t offset;
+        } ftruncate;
+
+        struct {
+            struct iatt in_buf;
+            int32_t valid;
+        } setattr;
+
+        struct {
+            struct iatt in_buf;
+            int32_t valid;
+        } fsetattr;
+
+        struct {
+            dict_t *dict;
+            int32_t flags;
+        } setxattr;
+
+        struct {
+            dict_t *dict;
+            int32_t flags;
+        } fsetxattr;
+
+        struct {
+            char *name;
+        } removexattr;
+
+        struct {
+            dict_t *xattr;
+            gf_xattrop_flags_t optype;
+        } xattrop;
+
+        /* dir write */
+
+        struct {
+            inode_t *inode;
+            struct iatt buf;
+            struct iatt preparent;
+            struct iatt postparent;
+            struct iatt prenewparent;
+            struct iatt postnewparent;
+        } dir_fop;  // common structure for all dir fops
+
+        struct {
+            fd_t *fd;
+            dict_t *params;
+            int32_t flags;
+            mode_t mode;
+        } create;
+
+        struct {
+            dict_t *params;
+            dev_t dev;
+            mode_t mode;
+        } mknod;
+
+        struct {
+            dict_t *params;
+            int32_t mode;
+        } mkdir;
+
+        struct {
+            dict_t *params;
+            char *linkpath;
+        } symlink;
+
+        struct {
+            off_t offset;
+            size_t len;
+            int32_t mode;
+        } fallocate;
+
+        struct {
+            off_t offset;
+            size_t len;
+        } discard;
+
+        struct {
+            off_t offset;
+            off_t len;
+            struct iatt prebuf;
+            struct iatt postbuf;
+        } zerofill;
+
+        struct {
+            char *volume;
+            int32_t cmd;
+            int32_t in_cmd;
+            struct gf_flock in_flock;
+            struct gf_flock flock;
+            void *xdata;
+        } inodelk;
+
+        struct {
+            char *volume;
+            char *basename;
+            void *xdata;
+            entrylk_cmd in_cmd;
+            entrylk_cmd cmd;
+            entrylk_type type;
+        } entrylk;
+
+        struct {
+            off_t offset;
+            gf_seek_what_t what;
+        } seek;
+
+        struct {
+            struct gf_lease user_lease;
+            struct gf_lease ret_lease;
+            unsigned char *locked_nodes;
+        } lease;
+
+        struct {
+            int flags;
+        } rmdir;
+
+        struct {
+            int32_t datasync;
+        } fsync;
+
+        struct {
+            uuid_t gfid_req;
+            gf_boolean_t needs_fresh_lookup;
+        } lookup;
+
+    } cont;
+
+    struct {
+        char *basename;
+        char *new_basename;
+
+        loc_t parent_loc;
+        loc_t new_parent_loc;
+
+        /* stub to resume on destruction
+           of the transaction frame */
+        call_stub_t *resume_stub;
+
+        struct list_head owner_list;
+        struct list_head wait_list;
+
+        unsigned char *pre_op;
+
+        /* Changelog xattr dict for [f]xattrop*/
+        dict_t **changelog_xdata;
+        unsigned char *pre_op_sources;
+
+        /* @failed_subvols: subvolumes on which a pre-op or a
+            FOP failed. */
+        unsigned char *failed_subvols;
+
+        call_frame_t *main_frame; /*Fop frame*/
+        call_frame_t *frame;      /*Transaction frame*/
+
+        int (*wind)(call_frame_t *frame, xlator_t *this, int subvol);
+
+        int (*unwind)(call_frame_t *frame, xlator_t *this);
+
+        off_t start, len;
+
+        afr_transaction_type type;
+
+        int32_t in_flight_sb_errno; /* This is where the cause of the
+                                       failure on the last good copy of
+                                       the file is stored.
+                                       */
+
+        /* @changelog_resume: function to be called after changlogging
+           (either pre-op or post-op) is done
+        */
+        afr_changelog_resume_t changelog_resume;
+
+        gf_boolean_t eager_lock_on;
+        gf_boolean_t do_eager_unlock;
+
+        /* @dirtied: flag which indicates whether we set dirty flag
+           in the OP. Typically true when we are performing operation
+           on more than one subvol and optimistic changelog is disabled
+
+           A 'true' value set in @dirtied flag means an 'undirtying'
+           has to be done in POST-OP phase.
+        */
+        gf_boolean_t dirtied;
+
+        /* @inherited: flag which indicates that the dirty flags
+           of the previous transaction were inherited
+        */
+        gf_boolean_t inherited;
+
+        /*
+          @no_uninherit: flag which indicates that a pre_op_uninherit()
+          must _not_ be attempted (and returned as failure) always. This
+          flag is set when a hard pre-op is performed, but not accounted
+          for it in fd_ctx->on_disk[]. Such transactions are "isolated"
+          from the pre-op piggybacking entirely and therefore uninherit
+          must not be attempted.
+        */
+        gf_boolean_t no_uninherit;
+
+        gf_boolean_t in_flight_sb; /* Indicator for occurrence of
+                                      split-brain while in the middle of
+                                      a txn. */
+
+        /* @uninherit_done:
+           @uninherit_value:
+
+           The above pair variables make pre_op_uninherit() idempotent.
+           Both are FALSE initially. The first call to pre_op_uninherit
+           sets @uninherit_done to TRUE and the return value to
+           @uninherit_value. Further calls will check for @uninherit_done
+           to be TRUE and if so will simply return @uninherit_value.
+        */
+        gf_boolean_t uninherit_done;
+        gf_boolean_t uninherit_value;
+
+        gf_boolean_t disable_delayed_post_op;
+    } transaction;
+
+    syncbarrier_t barrier;
 
-typedef struct _afr_locked_fd {
-        fd_t  *fd;
-        struct list_head list;
-} afr_locked_fd_t;
+    /* extra data for fops */
+    dict_t *xdata_req;
+    dict_t *xdata_rsp;
 
-typedef struct _afr_local {
-	unsigned int call_count;
-	unsigned int success_count;
-	unsigned int enoent_count;
+    dict_t *xattr_rsp; /*for [f]xattrop*/
+
+    mode_t umask;
+    int xflag;
+    struct afr_reply *replies;
+
+    /* For  client side background heals. */
+    struct list_head healer;
+    call_frame_t *heal_frame;
+
+    afr_inode_ctx_t *inode_ctx;
+
+    /*For thin-arbiter transactions.*/
+    int ta_failed_subvol;
+    int ta_event_gen;
+    struct list_head ta_waitq;
+    struct list_head ta_onwireq;
+    afr_ta_fop_state_t fop_state;
+    afr_fop_lock_state_t fop_lock_state;
+    gf_lkowner_t saved_lk_owner;
+    unsigned char read_txn_query_child;
+    unsigned char ta_child_up;
+    gf_boolean_t do_discovery;
+    gf_boolean_t need_full_crawl;
+    gf_boolean_t is_read_txn;
+    gf_boolean_t is_new_entry;
+} afr_local_t;
 
+typedef struct afr_spbc_timeout {
+    call_frame_t *frame;
+    loc_t *loc;
+    int spb_child_index;
+    gf_boolean_t d_spb;
+    gf_boolean_t m_spb;
+} afr_spbc_timeout_t;
+
+typedef struct afr_spb_status {
+    call_frame_t *frame;
+    loc_t *loc;
+} afr_spb_status_t;
+
+typedef struct afr_empty_brick_args {
+    call_frame_t *frame;
+    char *op_type;
+    loc_t loc;
+    int empty_index;
+} afr_empty_brick_args_t;
+
+typedef struct afr_read_subvol_args {
+    ia_type_t ia_type;
+    uuid_t gfid;
+} afr_read_subvol_args_t;
+
+typedef struct afr_granular_esh_args {
+    fd_t *heal_fd;
+    xlator_t *xl;
+    call_frame_t *frame;
+    gf_boolean_t mismatch; /* flag to represent occurrence of type/gfid
+                              mismatch */
+} afr_granular_esh_args_t;
 
-	unsigned int govinda_gOvinda;
+int
+afr_inode_get_readable(call_frame_t *frame, inode_t *inode, xlator_t *this,
+                       unsigned char *readable, int *event_p, int type);
+int
+afr_inode_read_subvol_get(inode_t *inode, xlator_t *this,
+                          unsigned char *data_subvols,
+                          unsigned char *metadata_subvols,
+                          int *event_generation);
+int
+__afr_inode_read_subvol_get(inode_t *inode, xlator_t *this,
+                            unsigned char *data_subvols,
+                            unsigned char *metadata_subvols,
+                            int *event_generation);
 
-	unsigned int read_child_index;
-        unsigned char read_child_returned;
-        unsigned int first_up_child;
+int
+__afr_inode_read_subvol_set(inode_t *inode, xlator_t *this,
+                            unsigned char *data_subvols,
+                            unsigned char *metadata_subvol,
+                            int event_generation);
+int
+afr_inode_read_subvol_set(inode_t *inode, xlator_t *this,
+                          unsigned char *data_subvols,
+                          unsigned char *metadata_subvols,
+                          int event_generation);
 
-        pid_t saved_pid;
+int
+__afr_inode_need_refresh_set(inode_t *inode, xlator_t *this);
 
-	int32_t op_ret;
-	int32_t op_errno;
+int
+afr_inode_need_refresh_set(inode_t *inode, xlator_t *this);
 
-	int32_t **pending;
+int
+afr_read_subvol_select_by_policy(inode_t *inode, xlator_t *this,
+                                 unsigned char *readable,
+                                 afr_read_subvol_args_t *args);
 
-	loc_t loc;
-	loc_t newloc;
+int
+afr_inode_read_subvol_type_get(inode_t *inode, xlator_t *this,
+                               unsigned char *readable, int *event_p, int type);
+int
+afr_read_subvol_get(inode_t *inode, xlator_t *this, int *subvol_p,
+                    unsigned char *readables, int *event_p,
+                    afr_transaction_type type, afr_read_subvol_args_t *args);
 
-	fd_t *fd;
+#define afr_data_subvol_get(i, t, s, r, e, a)                                  \
+    afr_read_subvol_get(i, t, s, r, e, AFR_DATA_TRANSACTION, a)
 
-	glusterfs_fop_t fop;
+#define afr_metadata_subvol_get(i, t, s, r, e, a)                              \
+    afr_read_subvol_get(i, t, s, r, e, AFR_METADATA_TRANSACTION, a)
 
-	unsigned char *child_up;
+int
+afr_inode_refresh(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                  uuid_t gfid, afr_inode_refresh_cbk_t cbk);
 
-	int32_t *child_errno;
+int32_t
+afr_notify(xlator_t *this, int32_t event, void *data, void *data2);
 
-	dict_t  *xattr_req;
+int
+xattr_is_equal(dict_t *this, char *key1, data_t *value1, void *data);
 
-	int32_t  inodelk_count;
-	int32_t  entrylk_count;
+int
+afr_add_entry_lockee(afr_local_t *local, loc_t *loc, char *basename,
+                     int child_count);
 
-        afr_internal_lock_t internal_lock;
+int
+afr_add_inode_lockee(afr_local_t *local, int child_count);
 
-        afr_locked_fd_t *locked_fd;
-        int32_t          source_child;
-        int32_t          lock_recovery_child;
+void
+afr_lockees_cleanup(afr_internal_lock_t *int_lock);
 
-        dict_t  *dict;
+int
+afr_attempt_lock_recovery(xlator_t *this, int32_t child_index);
 
-        int (*openfd_flush_cbk) (call_frame_t *frame, xlator_t *this);
+int
+afr_mark_locked_nodes(xlator_t *this, fd_t *fd, unsigned char *locked_nodes);
 
-	/*
-	   This struct contains the arguments for the "continuation"
-	   (scheme-like) of fops
-	*/
+void
+afr_set_lk_owner(call_frame_t *frame, xlator_t *this, void *lk_owner);
 
-	int   op;
-	struct {
-		struct {
-			unsigned char buf_set;
-			struct statvfs buf;
-		} statfs;
+int
+afr_set_lock_number(call_frame_t *frame, xlator_t *this);
 
-		struct {
-			inode_t *inode;
-			struct iatt buf;
-                        struct iatt read_child_buf;
-                        struct iatt postparent;
-                        ino_t ino;
-                        uint64_t gen;
-                        ino_t parent_ino;
-			dict_t *xattr;
-			dict_t **xattrs;
-                        gf_boolean_t is_revalidate;
-		} lookup;
+int32_t
+afr_unlock(call_frame_t *frame, xlator_t *this);
 
-		struct {
-			int32_t flags;
-                        int32_t wbflags;
-		} open;
+int
+afr_lock_nonblocking(call_frame_t *frame, xlator_t *this);
 
-		struct {
-			int32_t cmd;
-                        struct gf_flock user_flock;
-                        struct gf_flock ret_flock;
-			unsigned char *locked_nodes;
-		} lk;
-
-		/* inode read */
+int
+afr_blocking_lock(call_frame_t *frame, xlator_t *this);
 
-		struct {
-			int32_t mask;
-			int last_tried;  /* index of the child we tried previously */
-		} access;
-
-		struct {
-			int last_tried;
-			ino_t ino;
-		} stat;
-
-		struct {
-			int last_tried;
-			ino_t ino;
-		} fstat;
-
-		struct {
-			size_t size;
-			int last_tried;
-                        ino_t ino;
-		} readlink;
-
-		struct {
-			char *name;
-			int last_tried;
-		} getxattr;
-
-		struct {
-                        ino_t ino;
-			size_t size;
-			off_t offset;
-			int last_tried;
-		} readv;
-
-		/* dir read */
-
-		struct {
-			int success_count;
-			int32_t op_ret;
-			int32_t op_errno;
-
-                        uint32_t *checksum;
-		} opendir;
-
-		struct {
-			int32_t op_ret;
-			int32_t op_errno;
-			size_t size;
-			off_t offset;
-
-                        gf_boolean_t failed;
-			int last_tried;
-		} readdir;
-
-		struct {
-			int32_t op_ret;
-			int32_t op_errno;
-
-			size_t size;
-			off_t offset;
-			int32_t flag;
-
-			int last_tried;
-		} getdents;
-
-		/* inode write */
-
-		struct {
-			ino_t ino;
-			struct iatt prebuf;
-			struct iatt postbuf;
-
-			int32_t op_ret;
-
-			struct iovec *vector;
-			struct iobref *iobref;
-			int32_t count;
-			off_t offset;
-		} writev;
-
-                struct {
-                        ino_t ino;
-                        struct iatt prebuf;
-                        struct iatt postbuf;
-                } fsync;
-
-		struct {
-			ino_t ino;
-			off_t offset;
-			struct iatt prebuf;
-                        struct iatt postbuf;
-		} truncate;
-
-		struct {
-			ino_t ino;
-			off_t offset;
-			struct iatt prebuf;
-                        struct iatt postbuf;
-		} ftruncate;
-
-		struct {
-			ino_t ino;
-			struct iatt in_buf;
-                        int32_t valid;
-                        struct iatt preop_buf;
-                        struct iatt postop_buf;
-		} setattr;
-
-		struct {
-			ino_t ino;
-			struct iatt in_buf;
-                        int32_t valid;
-                        struct iatt preop_buf;
-                        struct iatt postop_buf;
-		} fsetattr;
-
-		struct {
-			dict_t *dict;
-			int32_t flags;
-		} setxattr;
-
-		struct {
-			char *name;
-		} removexattr;
-
-		/* dir write */
-
-		struct {
-			ino_t ino;
-                        uint64_t gen;
-                        ino_t parent_ino;
-			fd_t *fd;
-                        dict_t *params;
-			int32_t flags;
-			mode_t mode;
-			inode_t *inode;
-			struct iatt buf;
-                        struct iatt preparent;
-                        struct iatt postparent;
-                        struct iatt read_child_buf;
-		} create;
-
-		struct {
-			ino_t ino;
-                        uint64_t gen;
-                        ino_t parent_ino;
-			dev_t dev;
-			mode_t mode;
-                        dict_t *params;
-			inode_t *inode;
-			struct iatt buf;
-                        struct iatt preparent;
-                        struct iatt postparent;
-                        struct iatt read_child_buf;
-		} mknod;
-
-		struct {
-			ino_t ino;
-                        uint64_t gen;
-                        ino_t parent_ino;
-			int32_t mode;
-                        dict_t *params;
-			inode_t *inode;
-			struct iatt buf;
-                        struct iatt read_child_buf;
-                        struct iatt preparent;
-                        struct iatt postparent;
-		} mkdir;
-
-		struct {
-                        ino_t parent_ino;
-			int32_t op_ret;
-			int32_t op_errno;
-                        struct iatt preparent;
-                        struct iatt postparent;
-		} unlink;
-
-		struct {
-                        int   flags;
-                        ino_t parent_ino;
-			int32_t op_ret;
-			int32_t op_errno;
-                        struct iatt preparent;
-                        struct iatt postparent;
-		} rmdir;
-
-		struct {
-                        ino_t oldparent_ino;
-                        ino_t newparent_ino;
-			ino_t ino;
-			struct iatt buf;
-                        struct iatt read_child_buf;
-                        struct iatt preoldparent;
-                        struct iatt prenewparent;
-                        struct iatt postoldparent;
-                        struct iatt postnewparent;
-		} rename;
-
-		struct {
-			ino_t ino;
-                        uint64_t gen;
-                        ino_t parent_ino;
-			inode_t *inode;
-			struct iatt buf;
-                        struct iatt read_child_buf;
-                        struct iatt preparent;
-                        struct iatt postparent;
-		} link;
-
-		struct {
-			ino_t ino;
-                        uint64_t gen;
-                        ino_t parent_ino;
-			inode_t *inode;
-                        dict_t *params;
-			struct iatt buf;
-                        struct iatt read_child_buf;
-			char *linkpath;
-                        struct iatt preparent;
-                        struct iatt postparent;
-		} symlink;
-
-		struct {
-			int32_t flags;
-			dir_entry_t *entries;
-			int32_t count;
-		} setdents;
-	} cont;
-
-	struct {
-		off_t start, len;
-
-		char *basename;
-		char *new_basename;
-
-		loc_t parent_loc;
-		loc_t new_parent_loc;
-
-		afr_transaction_type type;
-
-		int success_count;
-		int erase_pending;
-		int failure_count;
-
-		int last_tried;
-		int32_t *child_errno;
-
-		call_frame_t *main_frame;
-
-		int (*fop) (call_frame_t *frame, xlator_t *this);
-
-		int (*done) (call_frame_t *frame, xlator_t *this);
-
-		int (*resume) (call_frame_t *frame, xlator_t *this);
-
-		int (*unwind) (call_frame_t *frame, xlator_t *this);
-
-                /* post-op hook */
-	} transaction;
-
-	afr_self_heal_t self_heal;
-} afr_local_t;
+int
+afr_internal_lock_finish(call_frame_t *frame, xlator_t *this);
 
+int
+__afr_fd_ctx_set(xlator_t *this, fd_t *fd);
 
-typedef struct {
-        unsigned int *pre_op_done;
-        unsigned int *opened_on;     /* which subvolumes the fd is open on */
-        unsigned int *pre_op_piggyback;
+afr_fd_ctx_t *
+afr_fd_ctx_get(fd_t *fd, xlator_t *this);
 
-        int flags;
-        int32_t wbflags;
-        uint64_t up_count;   /* number of CHILD_UPs this fd has seen */
-        uint64_t down_count; /* number of CHILD_DOWNs this fd has seen */
+int
+afr_build_parent_loc(loc_t *parent, loc_t *child, int32_t *op_errno);
 
-        int32_t last_tried;
+int
+afr_locked_nodes_count(unsigned char *locked_nodes, int child_count);
 
-        int  hit, miss;
-        gf_boolean_t failed_over;
-        struct list_head entries; /* needed for readdir failover */
+int
+afr_replies_interpret(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                      gf_boolean_t *start_heal);
 
-        unsigned char *locked_on; /* which subvolumes locks have been successful */
-} afr_fd_ctx_t;
+void
+afr_local_replies_wipe(afr_local_t *local, afr_private_t *priv);
 
+void
+afr_local_cleanup(afr_local_t *local, xlator_t *this);
 
-/* try alloc and if it fails, goto label */
-#define ALLOC_OR_GOTO(var, type, label) do {                     \
-		var = GF_CALLOC (sizeof (type), 1,               \
-                                 gf_afr_mt_##type);              \
-		if (!var) {                                      \
-			gf_log (this->name, GF_LOG_ERROR,        \
-				"out of memory :(");             \
-			op_errno = ENOMEM;                       \
-			goto label;                              \
-		}                                                \
-	} while (0);
+int
+afr_frame_return(call_frame_t *frame);
 
+int
+afr_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+         fd_t *fd, dict_t *xdata);
 
-/* did a call fail due to a child failing? */
-#define child_went_down(op_ret, op_errno) (((op_ret) < 0) &&	      \
-					   ((op_errno == ENOTCONN) || \
-					    (op_errno == EBADFD)))
+void
+afr_local_transaction_cleanup(afr_local_t *local, xlator_t *this);
 
-#define afr_fop_failed(op_ret, op_errno) ((op_ret) == -1)
+int
+afr_cleanup_fd_ctx(xlator_t *this, fd_t *fd);
+
+#define AFR_STACK_UNWIND(fop, frame, op_ret, op_errno, params...)              \
+    do {                                                                       \
+        afr_local_t *__local = NULL;                                           \
+        xlator_t *__this = NULL;                                               \
+        int32_t __op_ret = 0;                                                  \
+        int32_t __op_errno = 0;                                                \
+                                                                               \
+        __op_ret = op_ret;                                                     \
+        __op_errno = op_errno;                                                 \
+        if (frame) {                                                           \
+            __local = frame->local;                                            \
+            __this = frame->this;                                              \
+            afr_handle_inconsistent_fop(frame, &__op_ret, &__op_errno);        \
+            if (__local && __local->is_read_txn)                               \
+                afr_pending_read_decrement(__this->private,                    \
+                                           __local->read_subvol);              \
+            if (__local && __local->xdata_req &&                               \
+                afr_is_lock_mode_mandatory(__local->xdata_req))                \
+                afr_dom_lock_release(frame);                                   \
+            frame->local = NULL;                                               \
+        }                                                                      \
+                                                                               \
+        STACK_UNWIND_STRICT(fop, frame, __op_ret, __op_errno, params);         \
+        if (__local) {                                                         \
+            afr_local_cleanup(__local, __this);                                \
+            mem_put(__local);                                                  \
+        }                                                                      \
+    } while (0)
+
+#define AFR_STACK_DESTROY(frame)                                               \
+    do {                                                                       \
+        afr_local_t *__local = NULL;                                           \
+        xlator_t *__this = NULL;                                               \
+        __local = frame->local;                                                \
+        __this = frame->this;                                                  \
+        frame->local = NULL;                                                   \
+        STACK_DESTROY(frame->root);                                            \
+        if (__local) {                                                         \
+            afr_local_cleanup(__local, __this);                                \
+            mem_put(__local);                                                  \
+        }                                                                      \
+    } while (0);
+
+#define AFR_FRAME_INIT(frame, op_errno)                                        \
+    ({                                                                         \
+        frame->local = mem_get0(THIS->local_pool);                             \
+        if (afr_local_init(frame->local, frame->this->private, &op_errno)) {   \
+            afr_local_cleanup(frame->local, frame->this);                      \
+            mem_put(frame->local);                                             \
+            frame->local = NULL;                                               \
+        };                                                                     \
+        frame->local;                                                          \
+    })
+
+#define AFR_STACK_RESET(frame)                                                 \
+    do {                                                                       \
+        afr_local_t *__local = NULL;                                           \
+        xlator_t *__this = NULL;                                               \
+        __local = frame->local;                                                \
+        __this = frame->this;                                                  \
+        frame->local = NULL;                                                   \
+        int __opr;                                                             \
+        STACK_RESET(frame->root);                                              \
+        if (__local) {                                                         \
+            afr_local_cleanup(__local, __this);                                \
+            mem_put(__local);                                                  \
+        }                                                                      \
+        AFR_FRAME_INIT(frame, __opr);                                          \
+    } while (0)
 
-/* have we tried all children? */
-#define all_tried(i, count)  ((i) == (count) - 1)
+/* allocate and return a string that is the basename of argument */
+static inline char *
+AFR_BASENAME(const char *str)
+{
+    char *__tmp_str = NULL;
+    char *__basename_str = NULL;
+    __tmp_str = gf_strdup(str);
+    __basename_str = gf_strdup(basename(__tmp_str));
+    GF_FREE(__tmp_str);
+    return __basename_str;
+}
 
-int32_t
-afr_set_dict_gfid (dict_t *dict, uuid_t gfid);
+call_frame_t *
+afr_copy_frame(call_frame_t *base);
 
 int
-pump_command_reply (call_frame_t *frame, xlator_t *this);
+afr_transaction_local_init(afr_local_t *local, xlator_t *this);
 
 int32_t
-afr_notify (xlator_t *this, int32_t event,
-            void *data, ...);
+afr_marker_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                    const char *name, afr_local_t *local, afr_private_t *priv);
 
 int
-afr_attempt_lock_recovery (xlator_t *this, int32_t child_index);
+afr_local_init(afr_local_t *local, afr_private_t *priv, int32_t *op_errno);
 
 int
-afr_save_locked_fd (xlator_t *this, fd_t *fd);
+afr_internal_lock_init(afr_internal_lock_t *lk, size_t child_count);
 
 int
-afr_mark_locked_nodes (xlator_t *this, fd_t *fd,
-                       unsigned char *locked_nodes);
+afr_higher_errno(int32_t old_errno, int32_t new_errno);
+
+int
+afr_final_errno(afr_local_t *local, afr_private_t *priv);
+
+int
+afr_xattr_req_prepare(xlator_t *this, dict_t *xattr_req);
 
 void
-afr_set_lk_owner (call_frame_t *frame, xlator_t *this);
+afr_fix_open(fd_t *fd, xlator_t *this);
+
+afr_fd_ctx_t *
+afr_fd_ctx_get(fd_t *fd, xlator_t *this);
 
+void
+afr_set_low_priority(call_frame_t *frame);
 int
-afr_set_lock_number (call_frame_t *frame, xlator_t *this);
+afr_child_fd_ctx_set(xlator_t *this, fd_t *fd, int32_t child, int flags);
 
+void
+afr_matrix_cleanup(int32_t **pending, unsigned int m);
 
-loc_t *
-lower_path (loc_t *l1, const char *b1, loc_t *l2, const char *b2);
+int32_t **
+afr_matrix_create(unsigned int m, unsigned int n);
 
-int32_t
-afr_unlock (call_frame_t *frame, xlator_t *this);
+int **
+afr_mark_pending_changelog(afr_private_t *priv, unsigned char *pending,
+                           dict_t *xattr, ia_type_t iat);
+
+void
+afr_filter_xattrs(dict_t *xattr);
+
+/*
+ * Special value indicating we should use the "auto" quorum method instead of
+ * a fixed value (including zero to turn off quorum enforcement).
+ */
+#define AFR_QUORUM_AUTO INT_MAX
 
 int
-afr_nonblocking_entrylk (call_frame_t *frame, xlator_t *this);
+afr_fd_report_unstable_write(xlator_t *this, afr_local_t *local);
+
+gf_boolean_t
+afr_fd_has_witnessed_unstable_write(xlator_t *this, inode_t *inode);
+
+void
+afr_reply_wipe(struct afr_reply *reply);
+
+void
+afr_replies_wipe(struct afr_reply *replies, int count);
+
+gf_boolean_t
+afr_xattrs_are_equal(dict_t *dict1, dict_t *dict2);
+
+gf_boolean_t
+afr_is_xattr_ignorable(char *key);
 
 int
-afr_nonblocking_inodelk (call_frame_t *frame, xlator_t *this);
+afr_get_heal_info(call_frame_t *frame, xlator_t *this, loc_t *loc);
 
 int
-afr_blocking_lock (call_frame_t *frame, xlator_t *this);
+afr_heal_splitbrain_file(call_frame_t *frame, xlator_t *this, loc_t *loc);
 
 int
-afr_internal_lock_finish (call_frame_t *frame, xlator_t *this);
+afr_get_split_brain_status(void *opaque);
 
+int
+afr_get_split_brain_status_cbk(int ret, call_frame_t *frame, void *opaque);
 
-int pump_start (call_frame_t *frame, xlator_t *this);
+int
+afr_inode_split_brain_choice_set(inode_t *inode, xlator_t *this,
+                                 int spb_choice);
+int
+afr_split_brain_read_subvol_get(inode_t *inode, xlator_t *this,
+                                call_frame_t *frame, int *spb_subvol);
+int
+afr_get_child_index_from_name(xlator_t *this, char *name);
 
 int
-afr_fd_ctx_set (xlator_t *this, fd_t *fd);
+afr_is_split_brain(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                   uuid_t gfid, gf_boolean_t *d_spb, gf_boolean_t *m_spb);
+int
+afr_spb_choice_timeout_cancel(xlator_t *this, inode_t *inode);
 
-uint64_t
-afr_read_child (xlator_t *this, inode_t *inode);
+int
+afr_set_split_brain_choice(int ret, call_frame_t *frame, void *opaque);
 
-void
-afr_set_read_child (xlator_t *this, inode_t *inode, int32_t read_child);
+gf_boolean_t
+afr_get_need_heal(xlator_t *this);
 
 void
-afr_build_parent_loc (loc_t *parent, loc_t *child);
+afr_set_need_heal(xlator_t *this, afr_local_t *local);
 
 int
-afr_up_children_count (int child_count, unsigned char *child_up);
+afr_selfheal_data_open(xlator_t *this, inode_t *inode, fd_t **fd);
 
 int
-afr_locked_nodes_count (unsigned char *locked_nodes, int child_count);
-
-ino64_t
-afr_itransform (ino64_t ino, int child_count, int child_index);
+afr_get_msg_id(char *op_type);
 
 int
-afr_deitransform (ino64_t ino, int child_count);
+afr_set_in_flight_sb_status(xlator_t *this, call_frame_t *frame,
+                            inode_t *inode);
 
+int32_t
+afr_quorum_errno(afr_private_t *priv);
+
+gf_boolean_t
+afr_is_consistent_io_possible(afr_local_t *local, afr_private_t *priv,
+                              int32_t *op_errno);
 void
-afr_local_cleanup (afr_local_t *local, xlator_t *this);
+afr_handle_inconsistent_fop(call_frame_t *frame, int32_t *op_ret,
+                            int32_t *op_errno);
 
-int
-afr_frame_return (call_frame_t *frame);
+void
+afr_inode_write_fill(call_frame_t *frame, xlator_t *this, int child_index,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata);
+void
+afr_process_post_writev(call_frame_t *frame, xlator_t *this);
 
-uint64_t
-afr_is_split_brain (xlator_t *this, inode_t *inode);
+void
+afr_writev_unwind(call_frame_t *frame, xlator_t *this);
 
 void
-afr_set_split_brain (xlator_t *this, inode_t *inode, gf_boolean_t set);
+afr_writev_copy_outvars(call_frame_t *src_frame, call_frame_t *dst_frame);
+
+void
+afr_update_uninodelk(afr_local_t *local, afr_internal_lock_t *int_lock,
+                     int32_t child_index);
+afr_fd_ctx_t *
+__afr_fd_ctx_get(fd_t *fd, xlator_t *this);
+
+gf_boolean_t
+afr_is_inode_refresh_reqd(inode_t *inode, xlator_t *this, int event_gen1,
+                          int event_gen2);
 
 int
-afr_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
-          fd_t *fd, int32_t wbflags);
+afr_serialize_xattrs_with_delimiter(call_frame_t *frame, xlator_t *this,
+                                    char *buf, const char *default_str,
+                                    int32_t *serz_len, char delimiter);
+gf_boolean_t
+afr_is_symmetric_error(call_frame_t *frame, xlator_t *this);
 
-void
-afr_set_opendir_done (xlator_t *this, inode_t *inode);
+int
+__afr_inode_ctx_get(xlator_t *this, inode_t *inode, afr_inode_ctx_t **ctx);
 
 uint64_t
-afr_is_opendir_done (xlator_t *this, inode_t *inode);
+afr_write_subvol_get(call_frame_t *frame, xlator_t *this);
 
-void
-afr_local_transaction_cleanup (afr_local_t *local, xlator_t *this);
-
-int
-afr_cleanup_fd_ctx (xlator_t *this, fd_t *fd);
-
-int
-afr_openfd_flush (call_frame_t *frame, xlator_t *this, fd_t *fd);
-
-#define AFR_STACK_UNWIND(fop, frame, params ...)        \
-	do {						\
-		afr_local_t *__local = NULL;		\
-		xlator_t    *__this = NULL;		\
-                if (frame) {                            \
-                        __local = frame->local;		\
-                        __this = frame->this;           \
-                        frame->local = NULL;            \
-                }                                               \
-		STACK_UNWIND_STRICT (fop, frame, params);       \
-		afr_local_cleanup (__local, __this);            \
-		GF_FREE (__local);				\
-        } while (0);
-
-#define AFR_STACK_DESTROY(frame)			\
-	do {						\
-		afr_local_t *__local = NULL;		\
-		xlator_t    *__this = NULL;		\
-		__local = frame->local;			\
-		__this = frame->this;			\
-		frame->local = NULL;                    \
-		STACK_DESTROY (frame->root);		\
-		afr_local_cleanup (__local, __this);	\
-		GF_FREE (__local);			\
-        } while (0);
+int
+afr_write_subvol_set(call_frame_t *frame, xlator_t *this);
 
-/* allocate and return a string that is the basename of argument */
-static inline char *
-AFR_BASENAME (const char *str)
-{
-	char *__tmp_str = NULL;
-	char *__basename_str = NULL;
-	__tmp_str = gf_strdup (str);
-	__basename_str = gf_strdup (basename (__tmp_str));
-	GF_FREE (__tmp_str);
-	return __basename_str;
-}
+int
+afr_write_subvol_reset(call_frame_t *frame, xlator_t *this);
 
-/* initialize local_t */
-static inline int
-AFR_LOCAL_INIT (afr_local_t *local, afr_private_t *priv)
-{
-	local->child_up = GF_CALLOC (sizeof (*local->child_up),
-                                     priv->child_count,
-                                     gf_afr_mt_char);
-	if (!local->child_up) {
-		return -ENOMEM;
-	}
+int
+afr_set_inode_local(xlator_t *this, afr_local_t *local, inode_t *inode);
+
+int
+afr_fill_ta_loc(xlator_t *this, loc_t *loc, gf_boolean_t is_gfid_based_fop);
 
-	memcpy (local->child_up, priv->child_up,
-		sizeof (*local->child_up) * priv->child_count);
+int
+afr_ta_post_op_lock(xlator_t *this, loc_t *loc);
 
+int
+afr_ta_post_op_unlock(xlator_t *this, loc_t *loc);
 
-	local->call_count = afr_up_children_count (priv->child_count, local->child_up);
-	if (local->call_count == 0)
-		return -ENOTCONN;
+gf_boolean_t
+afr_is_pending_set(xlator_t *this, dict_t *xdata, int type);
 
-	local->transaction.erase_pending = 1;
+int
+__afr_get_up_children_count(afr_private_t *priv);
 
-	local->op_ret = -1;
-	local->op_errno = EUCLEAN;
+call_frame_t *
+afr_ta_frame_create(xlator_t *this);
 
-        local->internal_lock.lock_op_ret   = -1;
-        local->internal_lock.lock_op_errno = EUCLEAN;
+gf_boolean_t
+afr_ta_has_quorum(afr_private_t *priv, afr_local_t *local);
 
+void
+afr_ta_lock_release_synctask(xlator_t *this);
 
-	return 0;
-}
+void
+afr_ta_locked_priv_invalidate(afr_private_t *priv);
 
+gf_boolean_t
+afr_lookup_has_quorum(call_frame_t *frame,
+                      const unsigned int up_children_count);
 
-/**
- * first_up_child - return the index of the first child that is up
- */
+void
+afr_mark_new_entry_changelog(call_frame_t *frame, xlator_t *this);
 
-static inline int
-afr_first_up_child (afr_private_t *priv)
-{
-	xlator_t ** children = NULL;
-	int         ret      = -1;
-	int         i        = 0;
-
-	LOCK (&priv->lock);
-	{
-		children = priv->children;
-		for (i = 0; i < priv->child_count; i++) {
-			if (priv->child_up[i]) {
-				ret = i;
-				break;
-			}
-		}
-	}
-	UNLOCK (&priv->lock);
-
-	return ret;
-}
+void
+afr_handle_replies_quorum(call_frame_t *frame, xlator_t *this);
 
+gf_boolean_t
+afr_ta_dict_contains_pending_xattr(dict_t *dict, afr_private_t *priv,
+                                   int child);
 
-static inline int
-afr_transaction_local_init (afr_local_t *local, afr_private_t *priv)
-{
-        int i;
-
-        local->first_up_child = afr_first_up_child (priv);
-
-	local->child_errno = GF_CALLOC (sizeof (*local->child_errno),
-				        priv->child_count,
-                                        gf_afr_mt_int32_t);
-	if (!local->child_errno) {
-		return -ENOMEM;
-	}
-
-	local->pending = GF_CALLOC (sizeof (*local->pending),
-                                    priv->child_count,
-                                    gf_afr_mt_int32_t);
-
-	if (!local->pending) {
-		return -ENOMEM;
-	}
-
-        for (i = 0; i < priv->child_count; i++) {
-                local->pending[i] = GF_CALLOC (sizeof (*local->pending[i]),
-                                               3, /* data + metadata + entry */
-                                               gf_afr_mt_int32_t);
-                if (!local->pending[i])
-                        return -ENOMEM;
-        }
-
-	local->internal_lock.inode_locked_nodes =
-                GF_CALLOC (sizeof (*local->internal_lock.inode_locked_nodes),
-                           priv->child_count,
-                           gf_afr_mt_char);
-
-	local->internal_lock.entry_locked_nodes =
-                GF_CALLOC (sizeof (*local->internal_lock.entry_locked_nodes),
-                           priv->child_count,
-                           gf_afr_mt_char);
-
-	local->internal_lock.locked_nodes =
-                GF_CALLOC (sizeof (*local->internal_lock.locked_nodes),
-                           priv->child_count,
-                           gf_afr_mt_char);
-
-	local->internal_lock.lower_locked_nodes
-                = GF_CALLOC (sizeof (*local->internal_lock.lower_locked_nodes),
-                             priv->child_count,
-                             gf_afr_mt_char);
-
-	local->transaction.child_errno = GF_CALLOC (sizeof (*local->transaction.child_errno),
-					            priv->child_count,
-                                                    gf_afr_mt_int32_t);
-
-        local->internal_lock.transaction_lk_type = AFR_TRANSACTION_LK;
-
-	return 0;
-}
+void
+afr_selfheal_childup(xlator_t *this, afr_private_t *priv);
+
+gf_boolean_t
+afr_is_lock_mode_mandatory(dict_t *xdata);
+
+void
+afr_dom_lock_release(call_frame_t *frame);
+
+void
+afr_fill_success_replies(afr_local_t *local, afr_private_t *priv,
+                         unsigned char *replies);
 
+gf_boolean_t
+afr_is_private_directory(afr_private_t *priv, uuid_t pargfid, const char *name,
+                         pid_t pid);
 #endif /* __AFR_H__ */
diff --git a/xlators/cluster/afr/src/pump.c b/xlators/cluster/afr/src/pump.c
deleted file mode 100644
index 9a3a25af39a..00000000000
--- a/xlators/cluster/afr/src/pump.c
+++ /dev/null
@@ -1,2519 +0,0 @@
-/*
-   Copyright (c) 2007-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-#include <unistd.h>
-#include <sys/time.h>
-#include <stdlib.h>
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "afr-common.c"
-#include "defaults.c"
-
-static int
-pump_mark_start_pending (xlator_t *this)
-{
-        afr_private_t  *priv      = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        priv      = this->private;
-        pump_priv = priv->pump_private;
-
-        pump_priv->pump_start_pending = 1;
-
-        return 0;
-}
-
-static int
-is_pump_start_pending (xlator_t *this)
-{
-        afr_private_t  *priv      = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        priv      = this->private;
-        pump_priv = priv->pump_private;
-
-        return (pump_priv->pump_start_pending);
-}
-
-static int
-pump_remove_start_pending (xlator_t *this)
-{
-        afr_private_t  *priv      = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        priv      = this->private;
-        pump_priv = priv->pump_private;
-
-        pump_priv->pump_start_pending = 0;
-
-        return 0;
-}
-
-static pump_state_t
-pump_get_state ()
-{
-        xlator_t *this = NULL;
-        afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        pump_state_t ret;
-
-        this = THIS;
-        priv = this->private;
-        pump_priv = priv->pump_private;
-
-        LOCK (&pump_priv->pump_state_lock);
-        {
-                ret = pump_priv->pump_state;
-        }
-        UNLOCK (&pump_priv->pump_state_lock);
-
-        return ret;
-}
-
-int
-pump_change_state (xlator_t *this, pump_state_t state)
-{
-        afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        pump_state_t state_old;
-        pump_state_t state_new;
-
-
-        priv = this->private;
-        pump_priv = priv->pump_private;
-
-        GF_ASSERT (pump_priv);
-
-        LOCK (&pump_priv->pump_state_lock);
-        {
-                state_old = pump_priv->pump_state;
-                state_new = state;
-
-                pump_priv->pump_state = state;
-
-        }
-        UNLOCK (&pump_priv->pump_state_lock);
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Pump changing state from %d to %d",
-                state_old,
-                state_new);
-
-        return  0;
-}
-
-static int
-pump_set_resume_path (xlator_t *this, const char *path)
-{
-        int ret = 0;
-
-        afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        priv = this->private;
-        pump_priv = priv->pump_private;
-
-        assert (pump_priv);
-
-        LOCK (&pump_priv->resume_path_lock);
-        {
-                pump_priv->resume_path = strdup (path);
-                if (!pump_priv->resume_path)
-                        ret = -1;
-        }
-        UNLOCK (&pump_priv->resume_path_lock);
-
-        return ret;
-}
-
-static void
-build_child_loc (loc_t *parent, loc_t *child, char *path, char *name)
-{
-        child->path = path;
-        child->name = name;
-
-        child->parent = inode_ref (parent->inode);
-        child->inode = inode_new (parent->inode->table);
-}
-
-static char *
-build_file_path (loc_t *loc, gf_dirent_t *entry)
-{
-        xlator_t *this = NULL;
-        char *file_path = NULL;
-        int pathlen = 0;
-        int total_size = 0;
-
-        this = THIS;
-
-        pathlen = STRLEN_0 (loc->path);
-
-        if (IS_ROOT_PATH (loc->path)) {
-                total_size = pathlen + entry->d_len;
-                file_path = GF_CALLOC (1, total_size, gf_afr_mt_char);
-                if (!file_path) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Out of memory");
-                        return NULL;
-                }
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "constructing file path of size=%d"
-                        "pathlen=%d, d_len=%d",
-                        total_size, pathlen,
-                        entry->d_len);
-
-                snprintf(file_path, total_size, "%s%s", loc->path, entry->d_name);
-
-        } else {
-                total_size = pathlen + entry->d_len + 1; /* for the extra '/' in the path */
-                file_path = GF_CALLOC (1, total_size + 1, gf_afr_mt_char);
-                if (!file_path) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Out of memory");
-                        return NULL;
-                }
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "constructing file path of size=%d"
-                        "pathlen=%d, d_len=%d",
-                        total_size, pathlen,
-                        entry->d_len);
-
-                snprintf(file_path, total_size, "%s/%s", loc->path, entry->d_name);
-        }
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "path=%s and d_name=%s", loc->path, entry->d_name);
-        gf_log (this->name, GF_LOG_TRACE,
-                "constructed file_path=%s of size=%d", file_path, total_size);
-
-        return file_path;
-}
-
-static int
-pump_save_path (xlator_t *this, const char *path)
-{
-        afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
-        pump_state_t state;
-        dict_t *dict = NULL;
-        loc_t  loc;
-        int dict_ret = 0;
-        int ret = -1;
-
-        state = pump_get_state ();
-        if (state == PUMP_STATE_RESUME)
-                return 0;
-
-        priv = this->private;
-        pump_priv = priv->pump_private;
-
-        assert (priv->root_inode);
-
-        build_root_loc (priv->root_inode, &loc);
-
-        dict = dict_new ();
-        dict_ret = dict_set_str (dict, PUMP_PATH, (char *)path);
-
-        ret = syncop_setxattr (PUMP_SOURCE_CHILD (this), &loc, dict, 0);
-
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "setxattr failed - could not save path=%s", path);
-        } else {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "setxattr succeeded - saved path=%s", path);
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Saving path for status info");
-        }
-
-        dict_unref (dict);
-
-        return 0;
-}
-
-static int
-pump_check_and_update_status (xlator_t *this)
-{
-        pump_state_t state;
-        int ret = -1;
-
-        state = pump_get_state ();
-
-        switch (state) {
-
-        case PUMP_STATE_RESUME:
-        case PUMP_STATE_RUNNING:
-        {
-                ret = 0;
-                break;
-        }
-        case PUMP_STATE_PAUSE:
-        {
-                ret = -1;
-                break;
-        }
-        case PUMP_STATE_ABORT:
-        {
-                pump_save_path (this, "/");
-                ret = -1;
-                break;
-        }
-        default:
-        {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Unknown pump state");
-                ret = -1;
-                break;
-        }
-
-        }
-
-        return ret;
-}
-
-static const char *
-pump_get_resume_path (xlator_t *this)
-{
-        afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        const char *resume_path = NULL;
-
-        priv = this->private;
-        pump_priv = priv->pump_private;
-
-        resume_path = pump_priv->resume_path;
-
-        return resume_path;
-}
-
-static int
-pump_update_resume_state (xlator_t *this, const char *path)
-{
-        afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        pump_state_t state;
-        const char *resume_path = NULL;
-
-        priv = this->private;
-        pump_priv = priv->pump_private;
-
-        state = pump_get_state ();
-
-        if (state == PUMP_STATE_RESUME) {
-                resume_path = pump_get_resume_path (this);
-                if (strcmp (resume_path, "/") == 0) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Reached the resume path (/). Proceeding to change state"
-                                " to running");
-                        pump_change_state (this, PUMP_STATE_RUNNING);
-                } else if (strcmp (resume_path, path) == 0) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Reached the resume path. Proceeding to change state"
-                                " to running");
-                        pump_change_state (this, PUMP_STATE_RUNNING);
-                } else {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Not yet hit the resume path:res-path=%s,path=%s",
-                                resume_path, path);
-                }
-        }
-
-        return 0;
-}
-
-static gf_boolean_t
-is_pump_traversal_allowed (xlator_t *this, const char *path)
-{
-        afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        pump_state_t state;
-        const char *resume_path = NULL;
-        gf_boolean_t ret = _gf_true;
-
-        priv = this->private;
-        pump_priv = priv->pump_private;
-
-        state = pump_get_state ();
-
-        if (state == PUMP_STATE_RESUME) {
-                resume_path = pump_get_resume_path (this);
-                if (strstr (resume_path, path)) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "On the right path to resumption path");
-                        ret = _gf_true;
-                } else {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Not the right path to resuming=> ignoring traverse");
-                        ret = _gf_false;
-                }
-        }
-
-        return ret;
-}
-
-static int
-pump_save_file_stats (xlator_t *this, const char *path)
-{
-        afr_private_t  *priv        = NULL;
-        pump_private_t *pump_priv   = NULL;
-
-        priv      = this->private;
-        pump_priv = priv->pump_private;
-
-        LOCK (&pump_priv->resume_path_lock);
-        {
-                pump_priv->number_files_pumped++;
-
-                strncpy (pump_priv->current_file, path,
-                         PATH_MAX);
-        }
-        UNLOCK (&pump_priv->resume_path_lock);
-
-        return 0;
-}
-
-static int
-gf_pump_traverse_directory (loc_t *loc)
-{
-        xlator_t *this = NULL;
-        afr_private_t *priv = NULL;
-        fd_t     *fd   = NULL;
-
-        off_t       offset   = 0;
-        loc_t       entry_loc;
-        gf_dirent_t *entry = NULL;
-        gf_dirent_t *tmp = NULL;
-        gf_dirent_t entries;
-
-	struct iatt iatt, parent;
-	dict_t *xattr_rsp;
-
-        int source = 0;
-
-        char *file_path = NULL;
-        int ret = 0;
-
-        INIT_LIST_HEAD (&entries.list);
-        this = THIS;
-        priv = this->private;
-
-        assert (loc->inode);
-
-	fd = fd_create (loc->inode, PUMP_PID);
-        if (!fd) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Failed to create fd for %s", loc->path);
-                goto out;
-        }
-
-        ret = syncop_opendir (priv->children[source], loc, fd);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "opendir failed on %s", loc->path);
-                goto out;
-        }
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "pump opendir on %s returned=%d",
-                loc->path, ret);
-
-        while (syncop_readdirp (priv->children[source], fd, 131072, offset, &entries)) {
-
-                if (list_empty (&entries.list)) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "no more entries in directory");
-                        goto out;
-                }
-
-                list_for_each_entry_safe (entry, tmp, &entries.list, list) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "found readdir entry=%s", entry->d_name);
-
-                        file_path = build_file_path (loc, entry);
-                        if (!file_path) {
-                                gf_log (this->name, GF_LOG_DEBUG,
-                                        "file path construction failed");
-                                goto out;
-                        }
-
-                        build_child_loc (loc, &entry_loc, file_path, entry->d_name);
-
-                        if (!IS_ENTRY_CWD (entry->d_name) &&
-                                           !IS_ENTRY_PARENT (entry->d_name)) {
-
-                                    ret = syncop_lookup (this, &entry_loc, NULL,
-                                                         &iatt, &xattr_rsp, &parent);
-
-                                    entry_loc.ino = iatt.ia_ino;
-                                    entry_loc.inode->ino = iatt.ia_ino;
-                                    memcpy (entry_loc.inode->gfid, iatt.ia_gfid, 16);
-
-                                    gf_log (this->name, GF_LOG_DEBUG,
-                                            "lookup %s => %"PRId64,
-                                            entry_loc.path,
-                                            iatt.ia_ino);
-
-                                    ret = syncop_lookup (this, &entry_loc, NULL,
-                                                         &iatt, &xattr_rsp, &parent);
-
-
-                                    gf_log (this->name, GF_LOG_DEBUG,
-                                            "second lookup ret=%d: %s => %"PRId64,
-                                            ret,
-                                            entry_loc.path,
-                                            iatt.ia_ino);
-
-                                    pump_update_resume_state (this, entry_loc.path);
-
-                                    pump_save_path (this, entry_loc.path);
-                                    pump_save_file_stats (this, entry_loc.path);
-
-                                    ret = pump_check_and_update_status (this);
-                                    if (ret < 0) {
-                                            gf_log (this->name, GF_LOG_DEBUG,
-                                                    "Pump beginning to exit out");
-                                            goto out;
-                                    }
-
-                                    gf_log (this->name, GF_LOG_TRACE,
-                                            "type of file=%d, IFDIR=%d",
-                                            iatt.ia_type, IA_IFDIR);
-
-                                    if (IA_ISDIR (iatt.ia_type)) {
-                                            if (is_pump_traversal_allowed (this, entry_loc.path)) {
-                                                    gf_log (this->name, GF_LOG_TRACE,
-                                                            "entering dir=%s",
-                                                            entry->d_name);
-                                                    gf_pump_traverse_directory (&entry_loc);
-                                            }
-                                    }
-                            }
-                        offset = entry->d_off;
-                        loc_wipe (&entry_loc);
-                }
-
-                gf_dirent_free (&entries);
-                gf_log (this->name, GF_LOG_TRACE,
-                        "offset incremented to %d",
-                        (int32_t ) offset);
-
-        }
-
-out:
-        return 0;
-
-}
-
-void
-build_root_loc (inode_t *inode, loc_t *loc)
-{
-        loc->path = "/";
-        loc->name = "";
-        loc->inode = inode;
-        loc->ino = 1;
-        loc->inode->ino = 1;
-        memset (loc->inode->gfid, 0, 16);
-        loc->inode->gfid[15] = 1;
-
-}
-
-static int
-pump_update_resume_path (xlator_t *this)
-{
-        afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        const char *resume_path = NULL;
-
-        priv = this->private;
-        pump_priv = priv->pump_private;
-
-        resume_path = pump_get_resume_path (this);
-
-        if (resume_path) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Found a path to resume from: %s",
-                        resume_path);
-
-        }else {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Did not find a path=> setting to '/'");
-                pump_set_resume_path (this, "/");
-        }
-
-        pump_change_state (this, PUMP_STATE_RESUME);
-
-        return 0;
-}
-
-static int
-pump_complete_migration (xlator_t *this)
-{
-        afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
-        dict_t *dict = NULL;
-        pump_state_t state;
-        loc_t  loc;
-        int dict_ret = 0;
-        int ret = -1;
-
-        priv = this->private;
-        pump_priv = priv->pump_private;
-
-        assert (priv->root_inode);
-
-        build_root_loc (priv->root_inode, &loc);
-
-        dict = dict_new ();
-
-        state = pump_get_state ();
-        if (state == PUMP_STATE_RUNNING) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Pump finished pumping");
-
-                pump_priv->pump_finished = _gf_true;
-
-                dict_ret = dict_set_str (dict, PUMP_SOURCE_COMPLETE, "jargon");
-
-                ret = syncop_setxattr (PUMP_SOURCE_CHILD (this), &loc, dict, 0);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "setxattr failed - while  notifying source complete");
-                }
-                dict_ret = dict_set_str (dict, PUMP_SINK_COMPLETE, "jargon");
-
-                ret = syncop_setxattr (PUMP_SINK_CHILD (this), &loc, dict, 0);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "setxattr failed - while notifying sink complete");
-                }
-        }
-
-        return 0;
-}
-
-static int
-pump_set_root_gfid (dict_t *dict)
-{
-        uuid_t gfid;
-        int ret = 0;
-
-        memset (gfid, 0, 16);
-        gfid[15] = 1;
-
-        ret = afr_set_dict_gfid (dict, gfid);
-
-        return ret;
-}
-
-static int
-pump_lookup_sink (loc_t *loc)
-{
-        xlator_t *this = NULL;
-	struct iatt iatt, parent;
-	dict_t *xattr_rsp;
-        dict_t *xattr_req = NULL;
-        int ret = 0;
-
-        this = THIS;
-
-        xattr_req = dict_new ();
-
-        ret = pump_set_root_gfid (xattr_req);
-        if (ret)
-                goto out;
-
-        ret = syncop_lookup (PUMP_SINK_CHILD (this), loc,
-                             xattr_req, &iatt, &xattr_rsp, &parent);
-
-        if (ret) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Lookup on sink child failed");
-                goto out;
-        }
-
-out:
-        if (xattr_req)
-                dict_unref (xattr_req);
-
-        return ret;
-}
-
-static int
-pump_task (void *data)
-{
-	xlator_t *this = NULL;
-        afr_private_t *priv = NULL;
-
-
-        loc_t loc;
-	struct iatt iatt, parent;
-	dict_t *xattr_rsp = NULL;
-        dict_t *xattr_req = NULL;
-
-        int ret = -1;
-
-        this = THIS;
-        priv = this->private;
-
-        assert (priv->root_inode);
-
-        build_root_loc (priv->root_inode, &loc);
-        xattr_req = dict_new ();
-        if (!xattr_req) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Out of memory");
-                ret = -1;
-                goto out;
-        }
-
-        pump_set_root_gfid (xattr_req);
-        ret = syncop_lookup (this, &loc, xattr_req,
-                             &iatt, &xattr_rsp, &parent);
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "lookup: ino=%"PRId64", path=%s",
-                loc.ino,
-                loc.path);
-
-        ret = pump_check_and_update_status (this);
-        if (ret < 0) {
-                goto out;
-        }
-
-        pump_update_resume_path (this);
-
-        pump_set_root_gfid (xattr_req);
-        ret = pump_lookup_sink (&loc);
-        if (ret) {
-                pump_update_resume_path (this);
-                goto out;
-        }
-
-        gf_pump_traverse_directory (&loc);
-
-        pump_complete_migration (this);
-out:
-        if (xattr_req)
-                dict_unref (xattr_req);
-
-	return 0;
-}
-
-
-static int
-pump_task_completion (int ret, void *data)
-{
-        xlator_t *this = NULL;
-        call_frame_t *frame = NULL;
-        afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        this = THIS;
-
-        frame = (call_frame_t *) data;
-
-        priv = this->private;
-        pump_priv = priv->pump_private;
-
-        inode_unref (priv->root_inode);
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Pump xlator exiting");
-	return 0;
-}
-
-int
-pump_start (call_frame_t *pump_frame, xlator_t *this)
-{
-	afr_private_t *priv = NULL;
-	pump_private_t *pump_priv = NULL;
-
-	int ret = -1;
-
-	priv = this->private;
-        pump_priv = priv->pump_private;
-
-        if (!pump_frame->root->lk_owner)
-                pump_frame->root->lk_owner = PUMP_LK_OWNER;
-
-	ret = synctask_new (pump_priv->env, pump_task,
-                            pump_task_completion,
-                            pump_frame);
-        if (ret == -1) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "starting pump failed");
-                pump_change_state (this, PUMP_STATE_ABORT);
-                goto out;
-        }
-
-        gf_log (this->name, GF_LOG_TRACE,
-                "setting pump as started");
-
-        priv->use_afr_in_pump = 1;
-out:
-	return ret;
-}
-
-static int
-pump_start_synctask (xlator_t *this)
-{
-        call_frame_t *frame = NULL;
-        int ret = 0;
-
-        frame = create_frame (this, this->ctx->pool);
-        if (!frame) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                ret = -1;
-                goto out;
-        }
-
-        pump_change_state (this, PUMP_STATE_RUNNING);
-
-        ret = pump_start (frame, this);
-
-out:
-        return ret;
-}
-
-int32_t
-pump_cmd_start_setxattr_cbk (call_frame_t *frame,
-                             void *cookie,
-                             xlator_t *this,
-                             int32_t op_ret,
-                             int32_t op_errno)
-
-{
-        afr_local_t *local = NULL;
-        int ret = 0;
-
-        local = frame->local;
-
-        if (op_ret < 0) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Could not initiate destination "
-                        "brick connect");
-                ret = op_ret;
-                goto out;
-        }
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Successfully initiated destination "
-                "brick connect");
-
-        pump_mark_start_pending (this);
-
-out:
-        local->op_ret = ret;
-        pump_command_reply (frame, this);
-
-        return 0;
-}
-
-static int
-pump_initiate_sink_connect (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t   *local     = NULL;
-        afr_private_t *priv      = NULL;
-        dict_t        *dict      = NULL;
-        char          *dst_brick = NULL;
-        loc_t loc;
-
-        int ret = 0;
-
-        priv  = this->private;
-        local = frame->local;
-
-        GF_ASSERT (priv->root_inode);
-
-        build_root_loc (priv->root_inode, &loc);
-
-        ret = dict_get_str (local->dict, PUMP_CMD_START, &dst_brick);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Could not get destination brick value");
-                goto out;
-        }
-
-        dict = dict_new ();
-        if (!dict) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                ret = -1;
-                goto out;
-        }
-
-        GF_ASSERT (dst_brick);
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Got destination brick as %s", dst_brick);
-
-        ret = dict_set_str (dict, CLIENT_CMD_CONNECT, dst_brick);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Could not inititiate destination brick "
-                        "connect");
-                goto out;
-        }
-
-	STACK_WIND (frame,
-		    pump_cmd_start_setxattr_cbk,
-		    PUMP_SINK_CHILD(this),
-		    PUMP_SINK_CHILD(this)->fops->setxattr,
-		    &loc,
-		    dict,
-		    0);
-
-        ret = 0;
-
-        dict_unref (dict);
-out:
-        return ret;
-}
-
-static int
-is_pump_aborted (xlator_t *this)
-{
-        pump_state_t state;
-
-        state = pump_get_state ();
-
-        return ((state == PUMP_STATE_ABORT));
-}
-
-int32_t
-pump_cmd_start_getxattr_cbk (call_frame_t *frame,
-                             void *cookie,
-                             xlator_t *this,
-                             int32_t op_ret,
-                             int32_t op_errno,
-                             dict_t *dict)
-{
-        afr_local_t *local = NULL;
-        char *path = NULL;
-
-        pump_state_t state;
-        int ret = 0;
-        int need_unwind = 0;
-        int dict_ret = -1;
-
-        local = frame->local;
-
-        if (op_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "getxattr failed - changing pump "
-                        "state to RUNNING with '/'");
-                path = "/";
-                ret = op_ret;
-        } else {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "getxattr succeeded");
-
-                dict_ret =  dict_get_str (dict, PUMP_PATH, &path);
-                if (dict_ret < 0)
-                        path = "/";
-        }
-
-        state = pump_get_state ();
-        if ((state == PUMP_STATE_RUNNING) ||
-            (state == PUMP_STATE_RESUME)) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Pump is already started");
-                ret = -1;
-                goto out;
-        }
-
-        pump_set_resume_path (this, path);
-
-        if (is_pump_aborted (this))
-                /* We're re-starting pump afresh */
-                ret = pump_initiate_sink_connect (frame, this);
-        else {
-                /* We're re-starting pump from a previous
-                   pause */
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "about to start synctask");
-                ret = pump_start_synctask (this);
-                need_unwind = 1;
-        }
-
-out:
-        if ((ret < 0) || (need_unwind == 1)) {
-                local->op_ret = ret;
-                pump_command_reply (frame, this);
-        }
-	return 0;
-}
-
-int
-pump_execute_status (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t *priv = NULL;
-        pump_private_t *pump_priv = NULL;
-
-        uint64_t number_files = 0;
-
-        char filename[PATH_MAX];
-        char *dict_str = NULL;
-
-        int32_t op_ret = 0;
-        int32_t op_errno = 0;
-
-        dict_t *dict = NULL;
-        int ret = -1;
-
-        priv = this->private;
-        pump_priv = priv->pump_private;
-
-        LOCK (&pump_priv->resume_path_lock);
-        {
-                number_files  = pump_priv->number_files_pumped;
-                strncpy (filename, pump_priv->current_file, PATH_MAX);
-        }
-        UNLOCK (&pump_priv->resume_path_lock);
-
-        dict_str     = GF_CALLOC (1, PATH_MAX + 256, gf_afr_mt_char);
-        if (!dict_str) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                op_ret = -1;
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-        if (pump_priv->pump_finished) {
-        snprintf (dict_str, PATH_MAX + 256, "Number of files migrated = %"PRIu64"        Migration complete ",
-                  number_files);
-        } else {
-        snprintf (dict_str, PATH_MAX + 256, "Number of files migrated = %"PRIu64"       Current file= %s ",
-                  number_files, filename);
-        }
-
-        dict = dict_new ();
-
-        ret = dict_set_str (dict, PUMP_CMD_STATUS, dict_str);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "dict_set_str returned negative value");
-        }
-
-        op_ret = 0;
-
-out:
-
-        AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict);
-
-        dict_unref (dict);
-        GF_FREE (dict_str);
-
-        return 0;
-}
-
-int
-pump_execute_pause (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
-
-        local = frame->local;
-
-        pump_change_state (this, PUMP_STATE_PAUSE);
-
-        local->op_ret = 0;
-        pump_command_reply (frame, this);
-
-        return 0;
-}
-
-int
-pump_execute_start (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t *priv = NULL;
-        afr_local_t   *local = NULL;
-
-        int ret = 0;
-        loc_t loc;
-
-        priv = this->private;
-        local = frame->local;
-
-        if (!priv->root_inode) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Pump xlator cannot be started without an initial "
-                        "lookup");
-                ret = -1;
-                goto out;
-        }
-
-        GF_ASSERT (priv->root_inode);
-
-        build_root_loc (priv->root_inode, &loc);
-
-	STACK_WIND (frame,
-		    pump_cmd_start_getxattr_cbk,
-		    PUMP_SOURCE_CHILD(this),
-		    PUMP_SOURCE_CHILD(this)->fops->getxattr,
-		    &loc,
-		    PUMP_PATH);
-
-        ret = 0;
-
-out:
-        if (ret < 0) {
-                local->op_ret = ret;
-                pump_command_reply (frame, this);
-        }
-
-	return 0;
-}
-
-int
-pump_execute_abort (call_frame_t *frame, xlator_t *this)
-{
-        afr_private_t *priv = NULL;
-        afr_local_t   *local = NULL;
-
-        priv = this->private;
-        local = frame->local;
-
-        pump_change_state (this, PUMP_STATE_ABORT);
-
-        local->op_ret = 0;
-        pump_command_reply (frame, this);
-
-        return 0;
-}
-
-gf_boolean_t
-pump_command_status (xlator_t *this, dict_t *dict)
-{
-        char *cmd = NULL;
-        int dict_ret = -1;
-        int ret = _gf_true;
-
-        dict_ret = dict_get_str (dict, PUMP_CMD_STATUS, &cmd);
-        if (dict_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Not a pump status command");
-                ret = _gf_false;
-                goto out;
-        }
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Hit a pump command - status");
-        ret = _gf_true;
-
-out:
-        return ret;
-
-}
-
-gf_boolean_t
-pump_command_pause (xlator_t *this, dict_t *dict)
-{
-        char *cmd = NULL;
-        int dict_ret = -1;
-        int ret = _gf_true;
-
-        dict_ret = dict_get_str (dict, PUMP_CMD_PAUSE, &cmd);
-        if (dict_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Not a pump pause command");
-                ret = _gf_false;
-                goto out;
-        }
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Hit a pump command - pause");
-        ret = _gf_true;
-
-out:
-        return ret;
-
-}
-
-gf_boolean_t
-pump_command_abort (xlator_t *this, dict_t *dict)
-{
-        char *cmd = NULL;
-        int dict_ret = -1;
-        int ret = _gf_true;
-
-        dict_ret = dict_get_str (dict, PUMP_CMD_ABORT, &cmd);
-        if (dict_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Not a pump abort command");
-                ret = _gf_false;
-                goto out;
-        }
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Hit a pump command - abort");
-        ret = _gf_true;
-
-out:
-        return ret;
-
-}
-
-gf_boolean_t
-pump_command_start (xlator_t *this, dict_t *dict)
-{
-        char *cmd = NULL;
-        int dict_ret = -1;
-        int ret = _gf_true;
-
-        dict_ret = dict_get_str (dict, PUMP_CMD_START, &cmd);
-        if (dict_ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Not a pump start command");
-                ret = _gf_false;
-                goto out;
-        }
-
-        gf_log (this->name, GF_LOG_DEBUG,
-                "Hit a pump command - start");
-        ret = _gf_true;
-
-out:
-        return ret;
-
-}
-
-struct _xattr_key {
-        char *key;
-        struct list_head list;
-};
-
-static void
-__gather_xattr_keys (dict_t *dict, char *key, data_t *value,
-                     void *data)
-{
-        struct list_head *  list  = data;
-        struct _xattr_key * xkey  = NULL;
-
-        if (!strncmp (key, AFR_XATTR_PREFIX,
-                      strlen (AFR_XATTR_PREFIX))) {
-
-                xkey = GF_CALLOC (1, sizeof (*xkey), gf_afr_mt_xattr_key);
-                if (!xkey)
-                        return;
-
-                xkey->key = key;
-                INIT_LIST_HEAD (&xkey->list);
-
-                list_add_tail (&xkey->list, list);
-        }
-}
-
-static void
-__filter_xattrs (dict_t *dict)
-{
-        struct list_head keys;
-
-        struct _xattr_key *key;
-        struct _xattr_key *tmp;
-
-        INIT_LIST_HEAD (&keys);
-
-        dict_foreach (dict, __gather_xattr_keys,
-                      (void *) &keys);
-
-        list_for_each_entry_safe (key, tmp, &keys, list) {
-                dict_del (dict, key->key);
-
-                list_del_init (&key->list);
-
-                GF_FREE (key);
-        }
-}
-
-int32_t
-pump_getxattr_cbk (call_frame_t *frame, void *cookie,
-		  xlator_t *this, int32_t op_ret, int32_t op_errno,
-		  dict_t *dict)
-{
-	afr_private_t * priv     = NULL;
-	afr_local_t *   local    = NULL;
-	xlator_t **     children = NULL;
-
-	int unwind     = 1;
-	int last_tried = -1;
-	int this_try = -1;
-        int read_child = -1;
-
-	priv     = this->private;
-	children = priv->children;
-
-	local = frame->local;
-
-        read_child = (long) cookie;
-
-	if (op_ret == -1) {
-        retry:
-		last_tried = local->cont.getxattr.last_tried;
-
-		if (all_tried (last_tried, priv->child_count)) {
-			goto out;
-		}
-		this_try = ++local->cont.getxattr.last_tried;
-
-                if (this_try == read_child) {
-                        goto retry;
-                }
-
-		unwind = 0;
-		STACK_WIND_COOKIE (frame, pump_getxattr_cbk,
-				   (void *) (long) read_child,
-				   children[this_try],
-				   children[this_try]->fops->getxattr,
-				   &local->loc,
-				   local->cont.getxattr.name);
-	}
-
-out:
-	if (unwind) {
-                if (op_ret >= 0 && dict)
-                        __filter_xattrs (dict);
-
-		AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict);
-	}
-
-	return 0;
-}
-
-int32_t
-pump_getxattr (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, const char *name)
-{
-	afr_private_t *   priv       = NULL;
-	xlator_t **       children   = NULL;
-	int               call_child = 0;
-	afr_local_t     * local      = NULL;
-
-        int               read_child = -1;
-
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
-
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv     = this->private;
-	VALIDATE_OR_GOTO (priv->children, out);
-
-	children = priv->children;
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-	frame->local = local;
-
-        if (name) {
-                if (!strncmp (name, AFR_XATTR_PREFIX,
-                              strlen (AFR_XATTR_PREFIX))) {
-
-                        op_errno = ENODATA;
-                        goto out;
-                }
-
-                if (!strcmp (name, PUMP_CMD_STATUS)) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Hit pump command - status");
-                        pump_execute_status (frame, this);
-                        op_ret = 0;
-                        goto out;
-                }
-        }
-
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame, default_getxattr_cbk,
-                            FIRST_CHILD (this),
-                            (FIRST_CHILD (this))->fops->getxattr,
-                            loc, name);
-                return 0;
-        }
-
-        read_child = afr_read_child (this, loc->inode);
-
-        if (read_child >= 0) {
-                call_child = read_child;
-
-                local->cont.getxattr.last_tried = -1;
-        } else {
-                call_child = afr_first_up_child (priv);
-
-                if (call_child == -1) {
-                        op_errno = ENOTCONN;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "no child is up");
-                        goto out;
-                }
-
-                local->cont.getxattr.last_tried = call_child;
-        }
-
-	loc_copy (&local->loc, loc);
-	if (name)
-	  local->cont.getxattr.name       = gf_strdup (name);
-
-	STACK_WIND_COOKIE (frame, pump_getxattr_cbk,
-			   (void *) (long) call_child,
-			   children[call_child], children[call_child]->fops->getxattr,
-			   loc, name);
-
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		AFR_STACK_UNWIND (getxattr, frame, op_ret, op_errno, NULL);
-	}
-	return 0;
-}
-
-static int
-afr_setxattr_unwind (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-	call_frame_t   *main_frame = NULL;
-
-	local = frame->local;
-	priv  = this->private;
-
-	LOCK (&frame->lock);
-	{
-		if (local->transaction.main_frame)
-			main_frame = local->transaction.main_frame;
-		local->transaction.main_frame = NULL;
-	}
-	UNLOCK (&frame->lock);
-
-	if (main_frame) {
-		AFR_STACK_UNWIND (setxattr, main_frame,
-                                  local->op_ret, local->op_errno)
-	}
-	return 0;
-}
-
-static int
-afr_setxattr_wind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno)
-{
-	afr_local_t *   local = NULL;
-	afr_private_t * priv  = NULL;
-
-	int call_count  = -1;
-	int need_unwind = 0;
-
-	local = frame->local;
-	priv = this->private;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret != -1) {
-			if (local->success_count == 0) {
-				local->op_ret = op_ret;
-			}
-			local->success_count++;
-
-			if (local->success_count == priv->child_count) {
-				need_unwind = 1;
-			}
-		}
-
-		local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	if (need_unwind)
-		local->transaction.unwind (frame, this);
-
-	call_count = afr_frame_return (frame);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-	}
-
-	return 0;
-}
-
-static int
-afr_setxattr_wind (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t *local = NULL;
-	afr_private_t *priv = NULL;
-
-	int call_count = -1;
-	int i = 0;
-
-	local = frame->local;
-	priv = this->private;
-
-	call_count = afr_up_children_count (priv->child_count, local->child_up);
-
-	if (call_count == 0) {
-		local->transaction.resume (frame, this);
-		return 0;
-	}
-
-	local->call_count = call_count;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (local->child_up[i]) {
-			STACK_WIND_COOKIE (frame, afr_setxattr_wind_cbk,
-					   (void *) (long) i,
-					   priv->children[i],
-					   priv->children[i]->fops->setxattr,
-					   &local->loc,
-					   local->cont.setxattr.dict,
-					   local->cont.setxattr.flags);
-
-			if (!--call_count)
-				break;
-		}
-	}
-
-	return 0;
-}
-
-
-static int
-afr_setxattr_done (call_frame_t *frame, xlator_t *this)
-{
-	afr_local_t * local = frame->local;
-
-	local->transaction.unwind (frame, this);
-
-	AFR_STACK_DESTROY (frame);
-
-	return 0;
-}
-
-int32_t
-pump_setxattr_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno)
-{
-	STACK_UNWIND (frame,
-		      op_ret,
-		      op_errno);
-	return 0;
-}
-
-int
-pump_command_reply (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t *local = NULL;
-
-        local = frame->local;
-
-        if (local->op_ret < 0)
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "Command failed");
-        else
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "Command succeeded");
-
-        dict_unref (local->dict);
-
-        AFR_STACK_UNWIND (setxattr,
-                          frame,
-                          local->op_ret,
-                          local->op_errno);
-
-        return 0;
-}
-
-int
-pump_parse_command (call_frame_t *frame, xlator_t *this,
-                    afr_local_t *local, dict_t *dict)
-{
-
-        int ret = -1;
-
-        if (pump_command_start (this, dict)) {
-                frame->local = local;
-                local->dict = dict_ref (dict);
-                ret = pump_execute_start (frame, this);
-
-        } else if (pump_command_pause (this, dict)) {
-                frame->local = local;
-                local->dict = dict_ref (dict);
-                ret = pump_execute_pause (frame, this);
-
-        } else if (pump_command_abort (this, dict)) {
-                frame->local = local;
-                local->dict = dict_ref (dict);
-                ret = pump_execute_abort (frame, this);
-        }
-        return ret;
-}
-
-int
-pump_setxattr (call_frame_t *frame, xlator_t *this,
-               loc_t *loc, dict_t *dict, int32_t flags)
-{
-	afr_private_t * priv  = NULL;
-	afr_local_t   * local = NULL;
-	call_frame_t   *transaction_frame = NULL;
-
-	int ret = -1;
-
-	int op_ret   = -1;
-	int op_errno = 0;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (this, out);
-	VALIDATE_OR_GOTO (this->private, out);
-
-	priv = this->private;
-
-	ALLOC_OR_GOTO (local, afr_local_t, out);
-
-	ret = AFR_LOCAL_INIT (local, priv);
-	if (ret < 0) {
-		op_errno = -ret;
-		goto out;
-	}
-
-        ret = pump_parse_command (frame, this,
-                                  local, dict);
-        if (ret >= 0) {
-                op_ret = 0;
-                goto out;
-        }
-
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame, default_setxattr_cbk,
-                            FIRST_CHILD (this),
-                            (FIRST_CHILD (this))->fops->setxattr,
-                            loc, dict, flags);
-                return 0;
-        }
-
-	transaction_frame = copy_frame (frame);
-	if (!transaction_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		goto out;
-	}
-
-	transaction_frame->local = local;
-
-	local->op_ret = -1;
-
-	local->cont.setxattr.dict  = dict_ref (dict);
-	local->cont.setxattr.flags = flags;
-
-	local->transaction.fop    = afr_setxattr_wind;
-	local->transaction.done   = afr_setxattr_done;
-	local->transaction.unwind = afr_setxattr_unwind;
-
-	loc_copy (&local->loc, loc);
-
-	local->transaction.main_frame = frame;
-	local->transaction.start   = LLONG_MAX - 1;
-	local->transaction.len     = 0;
-
-	afr_transaction (transaction_frame, this, AFR_METADATA_TRANSACTION);
-
-	op_ret = 0;
-out:
-	if (op_ret == -1) {
-		if (transaction_frame)
-			AFR_STACK_DESTROY (transaction_frame);
-		AFR_STACK_UNWIND (setxattr, frame, op_ret, op_errno);
-	}
-
-	return 0;
-}
-
-/* Defaults */
-static int32_t
-pump_lookup (call_frame_t *frame,
-             xlator_t *this,
-             loc_t *loc,
-             dict_t *xattr_req)
-{
-	afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_lookup_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->lookup,
-                            loc,
-                            xattr_req);
-                return 0;
-        }
-
-        afr_lookup (frame, this, loc, xattr_req);
-        return 0;
-}
-
-
-static int32_t
-pump_truncate (call_frame_t *frame,
-               xlator_t *this,
-               loc_t *loc,
-               off_t offset)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_truncate_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->truncate,
-                            loc,
-                            offset);
-                return 0;
-        }
-
-        afr_truncate (frame, this, loc, offset);
-        return 0;
-}
-
-
-static int32_t
-pump_ftruncate (call_frame_t *frame,
-                xlator_t *this,
-                fd_t *fd,
-                off_t offset)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_ftruncate_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->ftruncate,
-                            fd,
-                            offset);
-                return 0;
-        }
-
-        afr_ftruncate (frame, this, fd, offset);
-        return 0;
-}
-
-
-
-
-int
-pump_mknod (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, mode_t mode, dev_t rdev, dict_t *parms)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame, default_mknod_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->mknod,
-                            loc, mode, rdev, parms);
-                return 0;
-        }
-        afr_mknod (frame, this, loc, mode, rdev, parms);
-        return 0;
-
-}
-
-
-
-int
-pump_mkdir (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, mode_t mode, dict_t *params)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame, default_mkdir_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->mkdir,
-                            loc, mode, params);
-                return 0;
-        }
-        afr_mkdir (frame, this, loc, mode, params);
-        return 0;
-
-}
-
-
-static int32_t
-pump_unlink (call_frame_t *frame,
-             xlator_t *this,
-             loc_t *loc)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_unlink_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->unlink,
-                            loc);
-                return 0;
-        }
-        afr_unlink (frame, this, loc);
-        return 0;
-
-}
-
-
-static int
-pump_rmdir (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, int flags)
-{
-        afr_private_t *priv  = NULL;
-
-	priv = this->private;
-
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame, default_rmdir_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->rmdir,
-                            loc, flags);
-                return 0;
-        }
-
-        afr_rmdir (frame, this, loc, flags);
-        return 0;
-
-}
-
-
-
-int
-pump_symlink (call_frame_t *frame, xlator_t *this,
-              const char *linkpath, loc_t *loc, dict_t *params)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame, default_symlink_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->symlink,
-                            linkpath, loc, params);
-                return 0;
-        }
-        afr_symlink (frame, this, linkpath, loc, params);
-        return 0;
-
-}
-
-
-static int32_t
-pump_rename (call_frame_t *frame,
-             xlator_t *this,
-             loc_t *oldloc,
-             loc_t *newloc)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_rename_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->rename,
-                            oldloc, newloc);
-                return 0;
-        }
-        afr_rename (frame, this, oldloc, newloc);
-        return 0;
-
-}
-
-
-static int32_t
-pump_link (call_frame_t *frame,
-           xlator_t *this,
-           loc_t *oldloc,
-           loc_t *newloc)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_link_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->link,
-                            oldloc, newloc);
-                return 0;
-        }
-        afr_link (frame, this, oldloc, newloc);
-        return 0;
-
-}
-
-
-static int32_t
-pump_create (call_frame_t *frame, xlator_t *this,
-             loc_t *loc, int32_t flags, mode_t mode,
-             fd_t *fd, dict_t *params)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame, default_create_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->create,
-                            loc, flags, mode, fd, params);
-                return 0;
-        }
-        afr_create (frame, this, loc, flags, mode, fd, params);
-        return 0;
-
-}
-
-
-static int32_t
-pump_open (call_frame_t *frame,
-           xlator_t *this,
-           loc_t *loc,
-           int32_t flags, fd_t *fd,
-           int32_t wbflags)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_open_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->open,
-                            loc, flags, fd, wbflags);
-                return 0;
-        }
-        afr_open (frame, this, loc, flags, fd, wbflags);
-        return 0;
-
-}
-
-
-static int32_t
-pump_writev (call_frame_t *frame,
-             xlator_t *this,
-             fd_t *fd,
-             struct iovec *vector,
-             int32_t count,
-             off_t off,
-             struct iobref *iobref)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_writev_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->writev,
-                            fd,
-                            vector,
-                            count,
-                            off,
-                            iobref);
-                return 0;
-        }
-        afr_writev (frame, this, fd, vector, count, off, iobref);
-        return 0;
-
-}
-
-
-static int32_t
-pump_flush (call_frame_t *frame,
-            xlator_t *this,
-            fd_t *fd)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_flush_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->flush,
-                            fd);
-                return 0;
-        }
-        afr_flush (frame, this, fd);
-        return 0;
-
-}
-
-
-static int32_t
-pump_fsync (call_frame_t *frame,
-            xlator_t *this,
-            fd_t *fd,
-            int32_t flags)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_fsync_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->fsync,
-                            fd,
-                            flags);
-                return 0;
-        }
-        afr_fsync (frame, this, fd, flags);
-        return 0;
-
-}
-
-
-static int32_t
-pump_opendir (call_frame_t *frame,
-              xlator_t *this,
-              loc_t *loc, fd_t *fd)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_opendir_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->opendir,
-                            loc, fd);
-                return 0;
-        }
-        afr_opendir (frame, this, loc, fd);
-        return 0;
-
-}
-
-
-static int32_t
-pump_fsyncdir (call_frame_t *frame,
-               xlator_t *this,
-               fd_t *fd,
-               int32_t flags)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_fsyncdir_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->fsyncdir,
-                            fd,
-                            flags);
-                return 0;
-        }
-        afr_fsyncdir (frame, this, fd, flags);
-        return 0;
-
-}
-
-
-static int32_t
-pump_xattrop (call_frame_t *frame,
-              xlator_t *this,
-              loc_t *loc,
-              gf_xattrop_flags_t flags,
-              dict_t *dict)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_xattrop_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->xattrop,
-                            loc,
-                            flags,
-                            dict);
-                return 0;
-        }
-        afr_xattrop (frame, this, loc, flags, dict);
-        return 0;
-
-}
-
-static int32_t
-pump_fxattrop (call_frame_t *frame,
-               xlator_t *this,
-               fd_t *fd,
-               gf_xattrop_flags_t flags,
-               dict_t *dict)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_fxattrop_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->fxattrop,
-                            fd,
-                            flags,
-                            dict);
-                return 0;
-        }
-        afr_fxattrop (frame, this, fd, flags, dict);
-        return 0;
-
-}
-
-
-static int32_t
-pump_removexattr (call_frame_t *frame,
-                  xlator_t *this,
-                  loc_t *loc,
-                  const char *name)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_removexattr_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->removexattr,
-                            loc,
-                            name);
-                return 0;
-        }
-        afr_removexattr (frame, this, loc, name);
-        return 0;
-
-}
-
-
-
-static int32_t
-pump_readdir (call_frame_t *frame,
-              xlator_t *this,
-              fd_t *fd,
-              size_t size,
-              off_t off)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_readdir_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->readdir,
-                            fd, size, off);
-                return 0;
-        }
-        afr_readdir (frame, this, fd, size, off);
-        return 0;
-
-}
-
-
-static int32_t
-pump_readdirp (call_frame_t *frame,
-               xlator_t *this,
-               fd_t *fd,
-               size_t size,
-               off_t off)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_readdirp_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->readdirp,
-                            fd, size, off);
-                return 0;
-        }
-        afr_readdirp (frame, this, fd, size, off);
-        return 0;
-
-}
-
-
-
-static int32_t
-pump_releasedir (xlator_t *this,
-                 fd_t *fd)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (priv->use_afr_in_pump)
-                afr_releasedir (this, fd);
-	return 0;
-
-}
-
-static int32_t
-pump_release (xlator_t *this,
-              fd_t *fd)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (priv->use_afr_in_pump)
-                afr_release (this, fd);
-	return 0;
-
-}
-
-
-static int32_t
-pump_setattr (call_frame_t *frame,
-              xlator_t *this,
-              loc_t *loc,
-              struct iatt *stbuf,
-              int32_t valid)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_setattr_cbk,
-                            FIRST_CHILD (this),
-                            FIRST_CHILD (this)->fops->setattr,
-                            loc, stbuf, valid);
-                return 0;
-        }
-        afr_setattr (frame, this, loc, stbuf, valid);
-        return 0;
-
-}
-
-
-static int32_t
-pump_fsetattr (call_frame_t *frame,
-               xlator_t *this,
-               fd_t *fd,
-               struct iatt *stbuf,
-               int32_t valid)
-{
-        afr_private_t *priv  = NULL;
-	priv = this->private;
-        if (!priv->use_afr_in_pump) {
-                STACK_WIND (frame,
-                            default_fsetattr_cbk,
-                            FIRST_CHILD (this),
-                            FIRST_CHILD (this)->fops->fsetattr,
-                            fd, stbuf, valid);
-                return 0;
-        }
-        afr_fsetattr (frame, this, fd, stbuf, valid);
-        return 0;
-
-}
-
-
-/* End of defaults */
-
-
-int32_t
-mem_acct_init (xlator_t *this)
-{
-        int     ret = -1;
-
-        if (!this)
-                return ret;
-
-        ret = xlator_mem_acct_init (this, gf_afr_mt_end + 1);
-
-        if (ret != 0) {
-                gf_log(this->name, GF_LOG_ERROR, "Memory accounting init"
-                                "failed");
-                return ret;
-        }
-
-        return ret;
-}
-
-static int
-is_xlator_pump_sink (xlator_t *child)
-{
-        return (child == PUMP_SINK_CHILD(THIS));
-}
-
-static int
-is_xlator_pump_source (xlator_t *child)
-{
-        return (child == PUMP_SOURCE_CHILD(THIS));
-}
-
-int32_t
-notify (xlator_t *this, int32_t event,
-	void *data, ...)
-{
-        int ret = -1;
-        xlator_t *child_xl = NULL;
-
-        child_xl = (xlator_t *) data;
-
-        ret = afr_notify (this, event, data);
-
-	switch (event) {
-	case GF_EVENT_CHILD_DOWN:
-                if (is_xlator_pump_source (child_xl))
-                        pump_change_state (this, PUMP_STATE_ABORT);
-                break;
-
-        case GF_EVENT_CHILD_UP:
-                if (is_xlator_pump_sink (child_xl))
-                        if (is_pump_start_pending (this)) {
-                                gf_log (this->name, GF_LOG_DEBUG,
-                                        "about to start synctask");
-                                ret = pump_start_synctask (this);
-                                if (ret < 0)
-                                        gf_log (this->name, GF_LOG_DEBUG,
-                                                "Could not start pump "
-                                                "synctask");
-                                else
-                                        pump_remove_start_pending (this);
-                        }
-        }
-
-        return ret;
-}
-
-int32_t
-init (xlator_t *this)
-{
-	afr_private_t * priv        = NULL;
-        pump_private_t *pump_priv = NULL;
-	int             child_count = 0;
-	xlator_list_t * trav        = NULL;
-	int             i           = 0;
-	int             ret         = -1;
-	int             op_errno    = 0;
-
-        int source_child = 0;
-
-	if (!this->children) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"pump translator needs a source and sink"
-                        "subvolumes defined.");
-		return -1;
-	}
-
-	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"Volume is dangling.");
-	}
-
-	ALLOC_OR_GOTO (this->private, afr_private_t, out);
-
-	priv = this->private;
-
-        priv->read_child = source_child;
-        priv->favorite_child = source_child;
-        priv->background_self_heal_count = 0;
-
-	priv->data_self_heal     = 1;
-	priv->metadata_self_heal = 1;
-	priv->entry_self_heal    = 1;
-
-        priv->data_self_heal_algorithm = "";
-
-        priv->data_self_heal_window_size = 16;
-
-	priv->data_change_log     = 1;
-	priv->metadata_change_log = 1;
-	priv->entry_change_log    = 1;
-        priv->use_afr_in_pump = 1;
-
-	/* Locking options */
-
-        /* Lock server count infact does not matter. Locks are held
-           on all subvolumes, in this case being the source
-           and the sink.
-        */
-
-	priv->data_lock_server_count = 2;
-	priv->metadata_lock_server_count = 2;
-	priv->entry_lock_server_count = 2;
-
-	priv->strict_readdir = _gf_false;
-
-        trav = this->children;
-        while (trav) {
-                child_count++;
-                trav = trav->next;
-        }
-
-	priv->wait_count = 1;
-
-        if (child_count != 2) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "There should be exactly 2 children - one source "
-                        "and one sink");
-                return -1;
-        }
-	priv->child_count = child_count;
-
-	LOCK_INIT (&priv->lock);
-        LOCK_INIT (&priv->read_child_lock);
-
-	priv->child_up = GF_CALLOC (sizeof (unsigned char), child_count,
-                                 gf_afr_mt_char);
-	if (!priv->child_up) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		op_errno = ENOMEM;
-		goto out;
-	}
-
-	priv->children = GF_CALLOC (sizeof (xlator_t *), child_count,
-                                 gf_afr_mt_xlator_t);
-	if (!priv->children) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		op_errno = ENOMEM;
-		goto out;
-	}
-
-        priv->pending_key = GF_CALLOC (sizeof (*priv->pending_key),
-                                       child_count,
-                                       gf_afr_mt_char);
-        if (!priv->pending_key) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory.");
-                op_errno = ENOMEM;
-                goto out;
-        }
-
-	trav = this->children;
-	i = 0;
-	while (i < child_count) {
-		priv->children[i] = trav->xlator;
-
-                ret = asprintf (&priv->pending_key[i], "%s.%s", AFR_XATTR_PREFIX,
-                                trav->xlator->name);
-                if (-1 == ret) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "asprintf failed to set pending key");
-                        op_errno = ENOMEM;
-                        goto out;
-                }
-
-		trav = trav->next;
-		i++;
-	}
-
-        priv->first_lookup = 1;
-        priv->root_inode = NULL;
-
-	pump_priv = GF_CALLOC (1, sizeof (*pump_priv),
-                            gf_afr_mt_pump_priv);
-	if (!pump_priv) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-                op_errno = ENOMEM;
-		goto out;
-	}
-
-        LOCK_INIT (&pump_priv->resume_path_lock);
-        LOCK_INIT (&pump_priv->pump_state_lock);
-
-        pump_priv->resume_path = GF_CALLOC (1, PATH_MAX,
-                                            gf_afr_mt_char);
-        if (!pump_priv->resume_path) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                ret = -1;
-                goto out;
-        }
-
-	pump_priv->env = syncenv_new (0);
-        if (!pump_priv->env) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Could not create new sync-environment");
-                ret = -1;
-                goto out;
-        }
-
-	priv->pump_private = pump_priv;
-
-        pthread_mutex_init (&priv->mutex, NULL);
-        INIT_LIST_HEAD (&priv->saved_fds);
-
-        pump_change_state (this, PUMP_STATE_ABORT);
-
-	ret = 0;
-out:
-	return ret;
-}
-
-int
-fini (xlator_t *this)
-{
-	return 0;
-}
-
-
-struct xlator_fops fops = {
-	.lookup      = pump_lookup,
-	.open        = pump_open,
-	.flush       = pump_flush,
-	.fsync       = pump_fsync,
-	.fsyncdir    = pump_fsyncdir,
-	.xattrop     = pump_xattrop,
-	.fxattrop    = pump_fxattrop,
-        .getxattr    = pump_getxattr,
-
-	/* inode write */
-	.writev      = pump_writev,
-	.truncate    = pump_truncate,
-	.ftruncate   = pump_ftruncate,
-	.setxattr    = pump_setxattr,
-        .setattr     = pump_setattr,
-	.fsetattr    = pump_fsetattr,
-	.removexattr = pump_removexattr,
-
-	/* dir read */
-	.opendir     = pump_opendir,
-	.readdir     = pump_readdir,
-	.readdirp    = pump_readdirp,
-
-	/* dir write */
-	.create      = pump_create,
-	.mknod       = pump_mknod,
-	.mkdir       = pump_mkdir,
-	.unlink      = pump_unlink,
-	.rmdir       = pump_rmdir,
-	.link        = pump_link,
-	.symlink     = pump_symlink,
-	.rename      = pump_rename,
-};
-
-struct xlator_dumpops dumpops = {
-        .priv       = afr_priv_dump,
-};
-
-
-struct xlator_cbks cbks = {
-	.release     = pump_release,
-	.releasedir  = pump_releasedir,
-};
-
-struct volume_options options[] = {
-	{ .key  = {NULL} },
-};
diff --git a/xlators/cluster/afr/src/pump.h b/xlators/cluster/afr/src/pump.h
deleted file mode 100644
index 0512290f2d6..00000000000
--- a/xlators/cluster/afr/src/pump.h
+++ /dev/null
@@ -1,97 +0,0 @@
-/*
-   Copyright (c) 2007-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef __PUMP_H__
-#define __PUMP_H__
-
-#include "syncop.h"
-
-/* FIXME: Needs to be defined in a common file */
-#define CLIENT_CMD_CONNECT "trusted.glusterfs.client-connect"
-#define CLIENT_CMD_DISCONNECT "trusted.glusterfs.client-disconnect"
-
-#define PUMP_PID 696969
-#define PUMP_LK_OWNER 696969
-
-#define IS_ROOT_PATH(path) (!strcmp (path, "/"))
-#define IS_ENTRY_CWD(entry) (!strcmp (entry, "."))
-#define IS_ENTRY_PARENT(entry) (!strcmp (entry, ".."))
-
-#define PUMP_CMD_START  "trusted.glusterfs.pump.start"
-#define PUMP_CMD_ABORT  "trusted.glusterfs.pump.abort"
-#define PUMP_CMD_PAUSE  "trusted.glusterfs.pump.pause"
-#define PUMP_CMD_STATUS "trusted.glusterfs.pump.status"
-
-#define PUMP_SOURCE_COMPLETE "trusted.glusterfs.pump-source-complete"
-#define PUMP_SINK_COMPLETE "trusted.glusterfs.pump-sink-complete"
-
-#define PUMP_PATH "trusted.glusterfs.pump-path"
-
-#define PUMP_SOURCE_CHILD(xl) (xl->children->xlator)
-#define PUMP_SINK_CHILD(xl) (xl->children->next->xlator)
-
-typedef enum {
-        PUMP_STATE_RUNNING,             /* Pump is running and migrating files */
-        PUMP_STATE_RESUME,              /* Pump is resuming from a previous pause */
-        PUMP_STATE_PAUSE,               /* Pump is paused */
-        PUMP_STATE_ABORT,               /* Pump is aborted */
-} pump_state_t;
-
-typedef struct _pump_private {
-	struct syncenv *env;            /* The env pointer to the pump synctask */
-        const char *resume_path;        /* path to resume from the last pause */
-        gf_lock_t resume_path_lock;     /* Synchronize resume_path changes */
-        gf_lock_t pump_state_lock;      /* Synchronize pump_state changes */
-        pump_state_t pump_state;        /* State of pump */
-        char current_file[PATH_MAX];    /* Current file being pumped */
-        uint64_t number_files_pumped;   /* Number of files pumped */
-        gf_boolean_t pump_finished;     /* Boolean to indicate pump termination */
-        char pump_start_pending;        /* Boolean to mark start pending until
-                                           CHILD_UP */
-} pump_private_t;
-
-void
-build_root_loc (inode_t *inode, loc_t *loc);
-int pump_start (call_frame_t *frame, xlator_t *this);
-
-gf_boolean_t
-pump_command_start (xlator_t *this, dict_t *dict);
-
-int
-pump_execute_start (call_frame_t *frame, xlator_t *this);
-
-gf_boolean_t
-pump_command_pause (xlator_t *this, dict_t *dict);
-
-int
-pump_execute_pause (call_frame_t *frame, xlator_t *this);
-
-gf_boolean_t
-pump_command_abort (xlator_t *this, dict_t *dict);
-
-int
-pump_execute_abort (call_frame_t *frame, xlator_t *this);
-
-gf_boolean_t
-pump_command_status (xlator_t *this, dict_t *dict);
-
-int
-pump_execute_status (call_frame_t *frame, xlator_t *this);
-
-#endif /* __PUMP_H__ */
diff --git a/xlators/cluster/dht/src/Makefile.am b/xlators/cluster/dht/src/Makefile.am
index 4b69aa07100..56f1f2ad7c8 100644
--- a/xlators/cluster/dht/src/Makefile.am
+++ b/xlators/cluster/dht/src/Makefile.am
@@ -1,34 +1,48 @@
-
 xlator_LTLIBRARIES = dht.la nufa.la switch.la
-xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/cluster
 
+AM_CFLAGS = -Wall $(GF_CFLAGS)
 
-dht_common_source = dht-layout.c dht-helper.c dht-linkfile.c \
-		dht-selfheal.c dht-rename.c dht-hashfn.c dht-diskusage.c
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/cluster
 
-dht_la_SOURCES = $(dht_common_source) dht.c 
+dht_common_source = dht-layout.c dht-helper.c dht-linkfile.c dht-rebalance.c \
+	dht-selfheal.c dht-rename.c dht-hashfn.c dht-diskusage.c \
+	dht-common.c dht-inode-write.c dht-inode-read.c dht-shared.c \
+	dht-lock.c $(top_builddir)/xlators/lib/src/libxlator.c
+
+dht_la_SOURCES = $(dht_common_source) dht.c
 
 nufa_la_SOURCES = $(dht_common_source) nufa.c
 switch_la_SOURCES = $(dht_common_source) switch.c
 
-dht_la_LDFLAGS = -module -avoidversion
+dht_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
 dht_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
-nufa_la_LDFLAGS = -module -avoidversion
+nufa_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
 nufa_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
-switch_la_LDFLAGS = -module -avoidversion
+switch_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
 switch_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
-noinst_HEADERS = dht-common.h dht-common.c dht-mem-types.h
+noinst_HEADERS = dht-common.h dht-mem-types.h dht-messages.h \
+	dht-lock.h $(top_builddir)/xlators/lib/src/libxlator.h
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS) \
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src \
+	-I$(top_srcdir)/rpc/rpc-lib/src \
+	-I$(top_srcdir)/xlators/lib/src \
+	-DDATADIR=\"$(localstatedir)\" \
+	-DLIBDIR=\"$(libdir)\"
 
-CLEANFILES = 
+CLEANFILES =
 
 uninstall-local:
 	rm -f $(DESTDIR)$(xlatordir)/distribute.so
 
 install-data-hook:
-	ln -sf dht.so $(DESTDIR)$(xlatordir)/distribute.so
\ No newline at end of file
+	ln -sf dht.so $(DESTDIR)$(xlatordir)/distribute.so
+
+if UNITTEST
+CLEANFILES += *.gcda *.gcno *_xunit.xml
+noinst_PROGRAMS =
+TESTS =
+endif
diff --git a/xlators/cluster/dht/src/dht-common.c b/xlators/cluster/dht/src/dht-common.c
index d537c1178bb..8ba0cc4c732 100644
--- a/xlators/cluster/dht/src/dht-common.c
+++ b/xlators/cluster/dht/src/dht-common.c
@@ -1,4838 +1,11391 @@
 /*
-   Copyright (c) 2009-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
 
 /* TODO: add NS locking */
 
-#include "glusterfs.h"
-#include "xlator.h"
+#include "libxlator.h"
 #include "dht-common.h"
-#include "defaults.h"
+#include "dht-lock.h"
+#include <glusterfs/byte-order.h>
+#include <glusterfs/quota-common-utils.h>
+#include <glusterfs/upcall-utils.h>
+#include "glusterfs/compat-errno.h"  // for ENODATA on BSD
+#include <glusterfs/common-utils.h>
 
 #include <sys/time.h>
 #include <libgen.h>
+#include <signal.h>
 
-/* TODO:
-   - use volumename in xattr instead of "dht"
-   - use NS locks
-   - handle all cases in self heal layout reconstruction
-   - complete linkfile selfheal
-*/
+static int
+dht_rmdir_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, gf_dirent_t *entries,
+                       dict_t *xdata);
 
+static int
+dht_link2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret);
 
-int
-dht_lookup_selfheal_cbk (call_frame_t *frame, void *cookie,
-			 xlator_t *this,
-			 int op_ret, int op_errno)
-{
-	dht_local_t  *local = NULL;
-	dht_layout_t *layout = NULL;
-	int           ret = 0;
+static int
+dht_set_dir_xattr_req(xlator_t *this, loc_t *loc, dict_t *xattr_req);
 
-	local = frame->local;
-	ret = op_ret;
+static int
+dht_lookup_everywhere_done(call_frame_t *frame, xlator_t *this);
 
-        dht_frame_su_undo (frame);
+static int
+dht_common_mark_mdsxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                             int op_ret, int op_errno, dict_t *xdata);
 
-	if (ret == 0) {
-		layout = local->selfheal.layout;
-		ret = dht_layout_set (this, local->inode, layout);
+static int
+dht_rmdir_unlock(call_frame_t *frame, xlator_t *this);
 
-		if (local->ia_ino) {
-			local->stbuf.ia_ino = local->ia_ino;
-		} else {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"could not find hashed subvolume for %s",
-				local->loc.path);
-		}
+static const char *dht_dbg_vxattrs[] = {DHT_DBG_HASHED_SUBVOL_PATTERN, NULL};
 
-                if (local->loc.parent)
-                        local->postparent.ia_ino = local->loc.parent->ino;
-	}
+/* Check the xdata to make sure EBADF has been set by client xlator */
+int32_t
+dht_check_remote_fd_failed_error(dht_local_t *local, int op_ret, int op_errno)
+{
+    if (op_ret == -1 && (op_errno == EBADF || op_errno == EBADFD) &&
+        !(local->fd_checked)) {
+        return 1;
+    }
+    return 0;
+}
 
-        WIPE (&local->postparent);
+/* Sets the blocks and size values to fixed values. This is to be called
+ * only for dirs. The caller is responsible for checking the type
+ */
+int32_t
+dht_set_fixed_dir_stat(struct iatt *stat)
+{
+    if (stat) {
+        stat->ia_blocks = DHT_DIR_STAT_BLOCKS;
+        stat->ia_size = DHT_DIR_STAT_SIZE;
+        return 0;
+    }
+    return -1;
+}
 
-	DHT_STACK_UNWIND (lookup, frame, ret, local->op_errno, local->inode,
-			  &local->stbuf, local->xattr, &local->postparent);
+/* Return true if key exists in array
+ */
+static gf_boolean_t
+dht_match_xattr(const char *key)
+{
+    char **xattrs_to_heal = get_xattrs_to_heal();
 
-	return 0;
+    return gf_get_index_by_elem(xattrs_to_heal, (char *)key) >= 0;
 }
 
+static int
+dht_aggregate_quota_xattr(dict_t *dst, char *key, data_t *value)
+{
+    int ret = -1;
+    quota_meta_t *meta_dst = NULL;
+    quota_meta_t *meta_src = NULL;
+    int64_t *size = NULL;
+    int64_t dst_dir_count = 0;
+    int64_t src_dir_count = 0;
+
+    if (value == NULL) {
+        gf_msg("dht", GF_LOG_WARNING, 0, DHT_MSG_DATA_NULL,
+               "data value is NULL");
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_get_bin(dst, key, (void **)&meta_dst);
+    if (ret < 0) {
+        meta_dst = GF_CALLOC(1, sizeof(quota_meta_t), gf_common_quota_meta_t);
+        if (meta_dst == NULL) {
+            gf_msg("dht", GF_LOG_WARNING, ENOMEM, DHT_MSG_NO_MEMORY,
+                   "Memory allocation failed");
+            ret = -1;
+            goto out;
+        }
+        ret = dict_set_bin(dst, key, meta_dst, sizeof(quota_meta_t));
+        if (ret < 0) {
+            gf_msg("dht", GF_LOG_WARNING, EINVAL, DHT_MSG_DICT_SET_FAILED,
+                   "dht aggregate dict set failed");
+            GF_FREE(meta_dst);
+            ret = -1;
+            goto out;
+        }
+    }
 
-int
-dht_lookup_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int op_ret, int op_errno,
-                    inode_t *inode, struct iatt *stbuf, dict_t *xattr,
-                    struct iatt *postparent)
-{
-	dht_conf_t   *conf                    = NULL;
-        dht_local_t  *local                   = NULL;
-        int           this_call_cnt           = 0;
-        call_frame_t *prev                    = NULL;
-	dht_layout_t *layout                  = NULL;
-	int           ret                     = 0;
-	int           is_dir                  = 0;
-
-	conf  = this->private;
-        local = frame->local;
-        prev  = cookie;
-
-	layout = local->layout;
-
-        if (!op_ret && uuid_is_null (local->gfid))
-                memcpy (local->gfid, stbuf->ia_gfid, 16);
-
-        LOCK (&frame->lock);
-        {
-                /* TODO: assert equal mode on stbuf->st_mode and
-		   local->stbuf->st_mode
-
-		   else mkdir/chmod/chown and fix
-		*/
-		ret = dht_layout_merge (this, layout, prev->this,
-					op_ret, op_errno, xattr);
-
-		if (op_ret == -1) {
-			local->op_errno = ENOENT;
-			gf_log (this->name, GF_LOG_DEBUG,
-				"lookup of %s on %s returned error (%s)",
-				local->loc.path, prev->this->name,
-				strerror (op_errno));
-
-			goto unlock;
-		}
-
- 		is_dir = check_is_dir (inode, stbuf, xattr);
- 		if (!is_dir) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "lookup of %s on %s returned non dir 0%o",
-                                local->loc.path, prev->this->name,
-                                stbuf->ia_type);
-                        local->need_selfheal = 1;
- 			goto unlock;
-                }
-
- 		local->op_ret = 0;
- 		if (local->xattr == NULL)
- 			local->xattr = dict_ref (xattr);
- 		if (local->inode == NULL)
- 			local->inode = inode_ref (inode);
+    if (value->len > sizeof(int64_t)) {
+        meta_src = data_to_bin(value);
 
-		dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
-                dht_iatt_merge (this, &local->postparent, postparent,
-                                prev->this);
+        meta_dst->size = hton64(ntoh64(meta_dst->size) +
+                                ntoh64(meta_src->size));
+        meta_dst->file_count = hton64(ntoh64(meta_dst->file_count) +
+                                      ntoh64(meta_src->file_count));
 
-                if (prev->this == dht_first_up_subvol (this)) {
-			local->ia_ino = local->stbuf.ia_ino;
-                }
+        if (value->len > (2 * sizeof(int64_t))) {
+            dst_dir_count = ntoh64(meta_dst->dir_count);
+            src_dir_count = ntoh64(meta_src->dir_count);
 
+            if (src_dir_count > dst_dir_count)
+                meta_dst->dir_count = meta_src->dir_count;
+        } else {
+            meta_dst->dir_count = 0;
         }
-unlock:
-        UNLOCK (&frame->lock);
-
+    } else {
+        size = data_to_bin(value);
+        meta_dst->size = hton64(ntoh64(meta_dst->size) + ntoh64(*size));
+    }
 
-        this_call_cnt = dht_frame_return (frame);
+    ret = 0;
+out:
+    return ret;
+}
 
-        if (is_last_call (this_call_cnt)) {
-                if (local->need_selfheal) {
-                        local->need_selfheal = 0;
-                        dht_lookup_everywhere (frame, this, &local->loc);
-                        return 0;
-                }
+static int
+add_opt(char **optsp, const char *opt)
+{
+    char *newopts = NULL;
+    unsigned oldsize = 0;
+    unsigned newsize = 0;
+
+    if (*optsp == NULL)
+        newopts = gf_strdup(opt);
+    else {
+        oldsize = strlen(*optsp);
+        newsize = oldsize + 1 + strlen(opt) + 1;
+        newopts = GF_REALLOC(*optsp, newsize);
+        if (newopts)
+            sprintf(newopts + oldsize, ",%s", opt);
+    }
+    if (newopts == NULL) {
+        gf_msg("dht", GF_LOG_WARNING, 0, DHT_MSG_NO_MEMORY,
+               "Error to add choices in buffer in add_opt");
+        return -1;
+    }
+    *optsp = newopts;
+    return 0;
+}
 
-		if (local->op_ret == 0) {
-			ret = dht_layout_normalize (this, &local->loc, layout);
+/* Return Choice list from Split brain status */
+static char *
+getChoices(const char *value)
+{
+    int i = 0;
+    char *ptr = NULL;
+    char *tok = NULL;
+    char *result = NULL;
+    char *newval = NULL;
+
+    ptr = strstr(value, "Choices:");
+    if (!ptr) {
+        result = ptr;
+        goto out;
+    }
+
+    newval = gf_strdup(ptr);
+    if (!newval) {
+        result = newval;
+        goto out;
+    }
+
+    tok = strtok(newval, ":");
+    if (!tok) {
+        result = tok;
+        goto out;
+    }
+
+    while (tok) {
+        i++;
+        if (i == 2)
+            break;
+        tok = strtok(NULL, ":");
+    }
+
+    result = gf_strdup(tok);
 
-			if (ret != 0) {
-				gf_log (this->name, GF_LOG_DEBUG,
-					"fixing assignment on %s",
-					local->loc.path);
-				goto selfheal;
-			}
+out:
+    if (newval)
+        GF_FREE(newval);
 
-			dht_layout_set (this, local->inode, layout);
+    return result;
+}
 
-			if (local->ia_ino) {
-				local->stbuf.ia_ino = local->ia_ino;
-			} else {
-				gf_log (this->name, GF_LOG_DEBUG,
-					"could not find hashed subvol for %s",
-					local->loc.path);
-			}
+/* This function prepare a list of choices for key
+   (replica.split-brain-status) in   case of metadata split brain
+   only on the basis of key-value passed to this function.
+   After prepare the list of choices it update the same key in dict
+   with this value to reflect the same in
+   replica.split-brain-status attr for file.
 
-                        if (local->loc.parent)
-                                local->postparent.ia_ino =
-                                        local->loc.parent->ino;
-		}
+*/
 
-		DHT_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno,
-				  local->inode, &local->stbuf, local->xattr,
-                                  &local->postparent);
+static int
+dht_aggregate_split_brain_xattr(dict_t *dst, char *key, data_t *value)
+{
+    int ret = 0;
+    char *oldvalue = NULL;
+    char *old_choice = NULL;
+    char *new_choice = NULL;
+    char *full_choice = NULL;
+    char *status = NULL;
+
+    if (value == NULL) {
+        gf_msg("dht", GF_LOG_WARNING, 0, DHT_MSG_DATA_NULL,
+               "GF_AFR_SBRAIN_STATUS value is NULL");
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_get_str(dst, key, &oldvalue);
+    if (ret)
+        goto out;
+
+    /* skip code that is irrelevant if !oldvalue */
+    if (!oldvalue)
+        goto out;
+
+    if (strstr(oldvalue, "not")) {
+        gf_msg_debug("dht", 0, "Need to update split-brain status in dict");
+        ret = -1;
+        goto out;
+    }
+    if (strstr(oldvalue, "metadata-split-brain:yes") &&
+        (strstr(oldvalue, "data-split-brain:no"))) {
+        if (strstr(value->data, "not")) {
+            gf_msg_debug("dht", 0, "No need to update split-brain status");
+            ret = 0;
+            goto out;
         }
+        if (strstr(value->data, "yes") &&
+            (strncmp(oldvalue, value->data, strlen(oldvalue)))) {
+            old_choice = getChoices(oldvalue);
+            if (!old_choice) {
+                gf_msg("dht", GF_LOG_WARNING, 0, DHT_MSG_NO_MEMORY,
+                       "Error to get choices");
+                ret = -1;
+                goto out;
+            }
 
-	return 0;
+            ret = add_opt(&full_choice, old_choice);
+            if (ret) {
+                gf_msg("dht", GF_LOG_WARNING, 0, DHT_MSG_NO_MEMORY,
+                       "Error to add choices");
+                ret = -1;
+                goto out;
+            }
 
-selfheal:
-        dht_frame_su_do (frame);
-	ret = dht_selfheal_directory (frame, dht_lookup_selfheal_cbk,
-				      &local->loc, layout);
+            new_choice = getChoices(value->data);
+            if (!new_choice) {
+                gf_msg("dht", GF_LOG_WARNING, 0, DHT_MSG_NO_MEMORY,
+                       "Error to get choices");
+                ret = -1;
+                goto out;
+            }
 
-	return 0;
+            ret = add_opt(&full_choice, new_choice);
+            if (ret) {
+                gf_msg("dht", GF_LOG_WARNING, 0, DHT_MSG_NO_MEMORY,
+                       "Error to add choices ");
+                ret = -1;
+                goto out;
+            }
+            ret = gf_asprintf(&status,
+                              "data-split-brain:%s    "
+                              "metadata-split-brain:%s   Choices:%s",
+                              "no", "yes", full_choice);
+
+            if (-1 == ret) {
+                gf_msg("dht", GF_LOG_WARNING, 0, DHT_MSG_NO_MEMORY,
+                       "Error to prepare status ");
+                goto out;
+            }
+            ret = dict_set_dynstr(dst, key, status);
+            if (ret) {
+                gf_msg("dht", GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+                       "Failed to set full choice");
+            }
+        }
+    }
+
+out:
+    if (old_choice)
+        GF_FREE(old_choice);
+    if (new_choice)
+        GF_FREE(new_choice);
+    if (full_choice)
+        GF_FREE(full_choice);
+
+    return ret;
 }
 
-int
-dht_lookup_root_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int op_ret, int op_errno,
-                         inode_t *inode, struct iatt *stbuf, dict_t *xattr,
-                         struct iatt *postparent)
-{
-	dht_conf_t   *conf                    = NULL;
-        dht_local_t  *local                   = NULL;
-        int           this_call_cnt           = 0;
-        call_frame_t *prev                    = NULL;
-	dht_layout_t *layout                  = NULL;
-	int           ret                     = 0;
-	int           is_dir                  = 0;
-
-	conf  = this->private;
-        local = frame->local;
-        prev  = cookie;
-
-	layout = local->layout;
-
-        LOCK (&frame->lock);
-        {
-		ret = dht_layout_merge (this, layout, prev->this,
-					op_ret, op_errno, xattr);
-
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			gf_log (this->name, GF_LOG_ERROR,
-				"lookup of %s on %s returned error (%s)",
-				local->loc.path, prev->this->name,
-				strerror (op_errno));
-			goto unlock;
-		}
-
-                is_dir = check_is_dir (inode, stbuf, xattr);
-                if (!is_dir) {
-                        gf_log (this->name, GF_LOG_CRITICAL,
-                                "lookup of %s on %s returned non dir 0%o",
-                                local->loc.path, prev->this->name,
-                                stbuf->ia_type);
-                        goto unlock;
-                }
+static int
+dht_aggregate(dict_t *this, char *key, data_t *value, void *data)
+{
+    dict_t *dst = NULL;
+    int32_t ret = -1;
+    data_t *dict_data = NULL;
+
+    dst = data;
+
+    /* compare split brain xattr only */
+    if (strcmp(key, GF_AFR_SBRAIN_STATUS) == 0) {
+        ret = dht_aggregate_split_brain_xattr(dst, key, value);
+        if (!ret)
+            goto out;
+    } else if (strcmp(key, QUOTA_SIZE_KEY) == 0) {
+        ret = dht_aggregate_quota_xattr(dst, key, value);
+        if (ret) {
+            gf_msg("dht", GF_LOG_WARNING, 0,
+                   DHT_MSG_AGGREGATE_QUOTA_XATTR_FAILED,
+                   "Failed to aggregate quota xattr");
+        }
+        goto out;
+    } else if (fnmatch(GF_XATTR_STIME_PATTERN, key, FNM_NOESCAPE) == 0) {
+        ret = gf_get_min_stime(THIS, dst, key, value);
+        goto out;
+    } else {
+        /* compare user xattrs only */
+        if (!strncmp(key, "user.", SLEN("user."))) {
+            ret = dict_lookup(dst, key, &dict_data);
+            if (!ret && dict_data && value) {
+                ret = is_data_equal(dict_data, value);
+                if (!ret)
+                    gf_msg_debug("dht", 0, "xattr mismatch for %s", key);
+            }
+        }
+    }
 
-                local->op_ret = 0;
-                if (local->xattr == NULL)
-                        local->xattr = dict_ref (xattr);
-                if (local->inode == NULL)
-                        local->inode = inode_ref (inode);
+    ret = dict_set(dst, key, value);
+    if (ret) {
+        gf_msg("dht", GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+               "Failed to set dictionary value: key = %s", key);
+    }
 
-		dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
+out:
+    return ret;
+}
 
-                if (prev->this == dht_first_up_subvol (this)) {
-			local->ia_ino = local->stbuf.ia_ino;
-                }
+static void
+dht_aggregate_xattr(dict_t *dst, dict_t *src)
+{
+    if ((dst == NULL) || (src == NULL)) {
+        goto out;
+    }
+
+    dict_foreach(src, dht_aggregate, dst);
+out:
+    return;
+}
 
+/* Code to save hashed subvol on inode ctx as a mds subvol
+ */
+int
+dht_inode_ctx_mdsvol_set(inode_t *inode, xlator_t *this, xlator_t *mds_subvol)
+{
+    dht_inode_ctx_t *ctx = NULL;
+    int ret = -1;
+    uint64_t ctx_int = 0;
+    gf_boolean_t ctx_free = _gf_false;
+
+    LOCK(&inode->lock);
+    {
+        ret = __inode_ctx_get(inode, this, &ctx_int);
+        if (ctx_int) {
+            ctx = (dht_inode_ctx_t *)(uintptr_t)ctx_int;
+            ctx->mds_subvol = mds_subvol;
+        } else {
+            ctx = GF_CALLOC(1, sizeof(*ctx), gf_dht_mt_inode_ctx_t);
+            if (!ctx)
+                goto unlock;
+            ctx->mds_subvol = mds_subvol;
+            ctx_free = _gf_true;
+            ctx_int = (long)ctx;
+            ret = __inode_ctx_set(inode, this, &ctx_int);
         }
+    }
 unlock:
-        UNLOCK (&frame->lock);
+    UNLOCK(&inode->lock);
+    if (ret && ctx_free)
+        GF_FREE(ctx);
+    return ret;
+}
 
+/*Code to get mds subvol from inode ctx */
 
-        this_call_cnt = dht_frame_return (frame);
+int
+dht_inode_ctx_mdsvol_get(inode_t *inode, xlator_t *this, xlator_t **mdsvol)
+{
+    dht_inode_ctx_t *ctx = NULL;
+    int ret = -1;
 
-        if (is_last_call (this_call_cnt)) {
-		if (local->op_ret == 0) {
-			ret = dht_layout_normalize (this, &local->loc, layout);
-			if (ret != 0) {
-				gf_log (this->name, GF_LOG_INFO,
-					"fixing assignment on %s",
-					local->loc.path);
-			}
+    if (!mdsvol)
+        return ret;
 
-			dht_layout_set (this, local->inode, layout);
-		}
+    if (__is_root_gfid(inode->gfid)) {
+        (*mdsvol) = FIRST_CHILD(this);
+        return 0;
+    }
+
+    ret = dht_inode_ctx_get(inode, this, &ctx);
 
-		DHT_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno,
-				  local->inode, &local->stbuf, local->xattr,
-                                  &local->postparent);
+    if (!ret && ctx) {
+        if (ctx->mds_subvol) {
+            *mdsvol = ctx->mds_subvol;
+            ret = 0;
+        } else {
+            ret = -1;
         }
+    }
 
-	return 0;
+    return ret;
 }
 
+/* TODO:
+   - use volumename in xattr instead of "dht"
+   - use NS locks
+   - handle all cases in self heal layout reconstruction
+   - complete linkfile selfheal
+*/
+
 static int
-dht_do_fresh_lookup_on_root (xlator_t *this, call_frame_t *frame)
+dht_lookup_selfheal_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, dict_t *xdata)
 {
-        dht_local_t  *local    = NULL;
-	dht_conf_t   *conf     = NULL;
-	int           ret      = -1;
-        int           call_cnt = 0;
-        int           i        = 0;
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
 
-        local = frame->local;
-        conf = this->private;
-        if (!conf)
-                goto err;
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+
+    local = frame->local;
+    conf = this->private;
+    ret = op_ret;
+
+    FRAME_SU_UNDO(frame, dht_local_t);
+
+    if (ret == 0) {
+        layout = local->selfheal.layout;
+        ret = dht_layout_set(this, local->inode, layout);
+    }
+
+    dht_inode_ctx_time_update(local->inode, this, &local->stbuf, 1);
+    if (local->loc.parent) {
+        dht_inode_ctx_time_update(local->loc.parent, this, &local->postparent,
+                                  1);
+    }
+
+    DHT_STRIP_PHASE1_FLAGS(&local->stbuf);
+    dht_set_fixed_dir_stat(&local->postparent);
+    /* Delete mds xattr at the time of STACK UNWIND */
+    GF_REMOVE_INTERNAL_XATTR(conf->mds_xattr_key, local->xattr);
 
-        if (local->layout) {
-                dht_layout_unref (this, local->layout);
-                local->layout = NULL;
+    DHT_STACK_UNWIND(lookup, frame, ret, local->op_errno, local->inode,
+                     &local->stbuf, local->xattr, &local->postparent);
+
+out:
+    return ret;
+}
+
+static int
+dht_discover_complete(xlator_t *this, call_frame_t *discover_frame)
+{
+    dht_local_t *local = NULL;
+    dht_local_t *heal_local = NULL;
+    call_frame_t *main_frame = NULL;
+    call_frame_t *heal_frame = NULL;
+    int op_errno = 0;
+    int ret = -1;
+    dht_layout_t *layout = NULL;
+    dht_conf_t *conf = NULL;
+    uint32_t vol_commit_hash = 0;
+    xlator_t *source = NULL;
+    int heal_path = 0;
+    int error_while_marking_mds = 0;
+    int i = 0;
+    loc_t loc = {0};
+    int8_t is_read_only = 0, layout_anomalies = 0;
+    char gfid_local[GF_UUID_BUF_SIZE] = {0};
+
+    local = discover_frame->local;
+    layout = local->layout;
+    conf = this->private;
+    gf_uuid_unparse(local->gfid, gfid_local);
+
+    LOCK(&discover_frame->lock);
+    {
+        main_frame = local->main_frame;
+        local->main_frame = NULL;
+    }
+    UNLOCK(&discover_frame->lock);
+
+    if (!main_frame)
+        return 0;
+
+    /* Code to update all extended attributed from
+       subvol to local->xattr on that internal xattr has found
+    */
+    if (conf->subvolume_cnt == 1)
+        local->need_xattr_heal = 0;
+    if (local->need_xattr_heal && (local->mds_xattr)) {
+        dht_dir_set_heal_xattr(this, local, local->xattr, local->mds_xattr,
+                               NULL, NULL);
+        dict_unref(local->mds_xattr);
+        local->mds_xattr = NULL;
+    }
+
+    ret = dict_get_int8(local->xattr_req, QUOTA_READ_ONLY_KEY, &is_read_only);
+    if (ret < 0)
+        gf_msg_debug(this->name, 0, "key = %s not present in dict",
+                     QUOTA_READ_ONLY_KEY);
+
+    if (local->file_count && local->dir_count) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_FILE_TYPE_MISMATCH,
+               "path %s exists as a file on one subvolume "
+               "and directory on another. "
+               "Please fix it manually",
+               local->loc.path);
+        op_errno = EIO;
+        goto out;
+    }
+
+    if (local->cached_subvol) {
+        ret = dht_layout_preset(this, local->cached_subvol, local->inode);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_SET_FAILED,
+                   "failed to set layout for subvolume %s",
+                   local->cached_subvol ? local->cached_subvol->name : "<nil>");
+            op_errno = EINVAL;
+            goto out;
         }
+    } else {
+        ret = dht_layout_normalize(this, &local->loc, layout);
+        if ((ret < 0) || ((ret > 0) && (local->op_ret != 0))) {
+            /* either the layout is incorrect or the directory is
+             * not found even in one subvolume.
+             */
+            gf_msg_debug(this->name, 0,
+                         "normalizing failed on %s "
+                         "(overlaps/holes present: %s, "
+                         "ENOENT errors: %d)",
+                         local->loc.path, (ret < 0) ? "yes" : "no",
+                         (ret > 0) ? ret : 0);
+            layout_anomalies = 1;
+        } else if (local->inode) {
+            dht_layout_set(this, local->inode, layout);
+        }
+    }
 
-        ret = dict_set_uint32 (local->xattr_req,
-                               "trusted.glusterfs.dht", 4 * 4);
-        if (ret)
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "failed to set the dict entry for dht");
+    if (!conf->vch_forced) {
+        ret = dict_get_uint32(local->xattr, conf->commithash_xattr_name,
+                              &vol_commit_hash);
+        if (ret == 0) {
+            conf->vol_commit_hash = vol_commit_hash;
+        }
+    }
 
-        call_cnt = local->call_cnt = conf->subvolume_cnt;
+    if (IA_ISDIR(local->stbuf.ia_type) && !is_read_only) {
+        for (i = 0; i < layout->cnt; i++) {
+            if (!source && !layout->list[i].err)
+                source = layout->list[i].xlator;
+            if (layout->list[i].err == ENOENT ||
+                layout->list[i].err == ESTALE) {
+                heal_path = 1;
+            }
+
+            if (source && heal_path)
+                break;
+        }
+    }
+
+    if (IA_ISDIR(local->stbuf.ia_type)) {
+        /* Call function to save hashed subvol on inode ctx if
+           internal mds xattr is not present and all subvols are up
+        */
+        if (!local->op_ret && !__is_root_gfid(local->stbuf.ia_gfid))
+            (void)dht_common_mark_mdsxattr(discover_frame,
+                                           &error_while_marking_mds, 1);
+
+        if (local->need_xattr_heal && !heal_path) {
+            local->need_xattr_heal = 0;
+            ret = dht_dir_xattr_heal(this, local, &op_errno);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                       DHT_MSG_DIR_XATTR_HEAL_FAILED,
+                       "xattr heal failed for "
+                       "directory  gfid is %s ",
+                       gfid_local);
+            }
+        }
+    }
 
-        local->layout = dht_layout_new (this,
-                                        conf->subvolume_cnt);
-        if (!local->layout) {
-                local->op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                goto err;
+    if (source && (heal_path || layout_anomalies || error_while_marking_mds)) {
+        gf_uuid_copy(loc.gfid, local->gfid);
+        if (gf_uuid_is_null(loc.gfid)) {
+            goto done;
         }
 
-        for (i = 0; i < call_cnt; i++) {
-                STACK_WIND (frame, dht_lookup_root_dir_cbk,
-                            conf->subvolumes[i],
-                            conf->subvolumes[i]->fops->lookup,
-                            &local->loc, local->xattr_req);
+        if (local->inode)
+            loc.inode = inode_ref(local->inode);
+        else
+            goto done;
+
+        heal_frame = create_frame(this, this->ctx->pool);
+        if (heal_frame) {
+            heal_local = dht_local_init(heal_frame, &loc, NULL, 0);
+            if (!heal_local)
+                goto cleanup;
+
+            gf_uuid_copy(heal_local->gfid, local->gfid);
+            heal_frame->cookie = source;
+            heal_local->xattr = dict_ref(local->xattr);
+            heal_local->stbuf = local->stbuf;
+            heal_local->postparent = local->postparent;
+            heal_local->inode = inode_ref(loc.inode);
+            heal_local->main_frame = main_frame;
+            FRAME_SU_DO(heal_frame, dht_local_t);
+            ret = synctask_new(this->ctx->env, dht_heal_full_path,
+                               dht_heal_full_path_done, heal_frame, heal_frame);
+            if (!ret) {
+                loc_wipe(&loc);
+                return 0;
+            }
+            /*
+             * Failed to spawn the synctask. Returning
+             * with out doing heal.
+             */
+        cleanup:
+            loc_wipe(&loc);
+            DHT_STACK_DESTROY(heal_frame);
         }
+    }
+done:
+    dht_set_fixed_dir_stat(&local->postparent);
+    /* Delete mds xattr at the time of STACK UNWIND */
+    if (local->xattr)
+        GF_REMOVE_INTERNAL_XATTR(conf->mds_xattr_key, local->xattr);
 
-        return 0;
-err:
-        DHT_STACK_UNWIND (lookup, frame, -1, local->op_errno,
-                          local->inode, &local->stbuf, local->xattr,
-                          &local->postparent);
-        return 0;
+    DHT_STACK_UNWIND(lookup, main_frame, local->op_ret, local->op_errno,
+                     local->inode, &local->stbuf, local->xattr,
+                     &local->postparent);
+    return 0;
+
+out:
+    DHT_STACK_UNWIND(lookup, main_frame, -1, op_errno, NULL, NULL, NULL, NULL);
+
+    return ret;
 }
 
-int
-dht_revalidate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int op_ret, int op_errno,
-                    inode_t *inode, struct iatt *stbuf, dict_t *xattr,
-                    struct iatt *postparent)
-{
-        dht_local_t  *local         = NULL;
-        int           this_call_cnt = 0;
-        call_frame_t *prev          = NULL;
-	dht_layout_t *layout        = NULL;
-	dht_conf_t   *conf          = NULL;
-	int           ret  = -1;
-	int           is_dir = 0;
-	int           is_linkfile = 0;
-        unsigned char root_gfid[16] = {0,};
+static int
+dht_common_mark_mdsxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                             int op_ret, int op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = cookie;
+    int ret = -1;
+    dht_conf_t *conf = 0;
+    dht_layout_t *layout = NULL;
+    int32_t mds_heal_fresh_lookup = 0;
+
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+
+    local = frame->local;
+    conf = this->private;
+    layout = local->selfheal.layout;
+    mds_heal_fresh_lookup = local->mds_heal_fresh_lookup;
+
+    if (op_ret) {
+        gf_msg_debug(this->name, op_ret,
+                     "Failed to set %s on the MDS %s for path %s. ",
+                     conf->mds_xattr_key, prev->name, local->loc.path);
+    } else {
+        /* Save mds subvol on inode ctx */
+        ret = dht_inode_ctx_mdsvol_set(local->inode, this, prev);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SET_INODE_CTX_FAILED,
+                   "Failed to set mds subvol on inode ctx"
+                   " %s for %s ",
+                   prev->name, local->loc.path);
+        }
+    }
+    if (!local->mds_heal_fresh_lookup && layout) {
+        dht_selfheal_dir_setattr(frame, &local->loc, &local->stbuf, 0xffffffff,
+                                 layout);
+    }
+out:
+    if (mds_heal_fresh_lookup)
+        DHT_STACK_DESTROY(frame);
+    return 0;
+}
 
-        local = frame->local;
-        prev  = cookie;
-	conf = this->private;
-        if (!conf)
+static xlator_t *
+dht_inode_get_hashed_subvol(inode_t *inode, xlator_t *this, loc_t *loc)
+{
+    char *path = NULL;
+    loc_t populate_loc = {
+        0,
+    };
+    char *name = NULL;
+    xlator_t *hash_subvol = NULL;
+
+    if (!inode)
+        return hash_subvol;
+
+    if (loc && loc->parent && loc->path) {
+        if (!loc->name) {
+            name = strrchr(loc->path, '/');
+            if (name) {
+                loc->name = name + 1;
+            } else {
                 goto out;
+            }
+        }
+        hash_subvol = dht_subvol_get_hashed(this, loc);
+        goto out;
+    }
 
-        LOCK (&frame->lock);
-        {
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-
-			if ((op_errno != ENOTCONN) 
-                            && (op_errno != ENOENT)
-                            && (op_errno != ESTALE)) {
-				gf_log (this->name, GF_LOG_INFO,
-					"subvolume %s returned -1 (%s)",
-					prev->this->name, strerror (op_errno));
-			}
-                        
-                        if (op_errno == ESTALE) {
-                                /* propogate the ESTALE to parent. 
-                                 * setting local->layout_mismatch would send
-                                 * ESTALE to parent. */
-                                local->layout_mismatch = 1;
-                        }
+    if (!gf_uuid_is_null(inode->gfid)) {
+        populate_loc.inode = inode_ref(inode);
+        populate_loc.parent = inode_parent(populate_loc.inode, NULL, NULL);
+        inode_path(populate_loc.inode, NULL, &path);
 
-			goto unlock;
-		}
-
-		if (stbuf->ia_type != local->inode->ia_type) {
-			gf_log (this->name, GF_LOG_INFO,
-				"mismatching filetypes 0%o v/s 0%o for %s",
-				(stbuf->ia_type), (local->inode->ia_type),
-				local->loc.path);
-
-			local->op_ret = -1;
-			local->op_errno = EINVAL;
-
-			goto unlock;
-		}
-
-		layout = local->layout;
-		
-		is_dir = check_is_dir (inode, stbuf, xattr);
-		is_linkfile = check_is_linkfile (inode, stbuf, xattr);
-		
-		if (is_linkfile) {
-			gf_log (this->name, GF_LOG_INFO,
-				"linkfile found in revalidate for %s",
-				local->loc.path);
-			local->layout_mismatch = 1;
-
-			goto unlock;
-		}
-
-		if (is_dir) {
-			ret = dht_layout_dir_mismatch (this, layout,
-						       prev->this, &local->loc,
-						       xattr);
-			if (ret != 0) {
-				gf_log (this->name, GF_LOG_INFO,
-					"mismatching layouts for %s", 
-					local->loc.path);
-			
-				local->layout_mismatch = 1;
-
-				goto unlock;
-			}
-		} 
-		
-		dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
-                dht_iatt_merge (this, &local->postparent, postparent,
-                                prev->this);
-		
-		local->op_ret = 0;
-		local->stbuf.ia_ino = local->ia_ino;
-
-                if (local->loc.parent)
-                        local->postparent.ia_ino = local->loc.parent->ino;
-
-		if (!local->xattr)
-			local->xattr = dict_ref (xattr);
-	}
-unlock:
-	UNLOCK (&frame->lock);
+        if (!path)
+            goto out;
+
+        populate_loc.path = path;
+        if (!populate_loc.name && populate_loc.path) {
+            name = strrchr(populate_loc.path, '/');
+            if (name) {
+                populate_loc.name = name + 1;
+
+            } else {
+                goto out;
+            }
+        }
+        hash_subvol = dht_subvol_get_hashed(this, &populate_loc);
+    }
 out:
-        this_call_cnt = dht_frame_return (frame);
-
-        if (is_last_call (this_call_cnt)) {
-		if (!IA_ISDIR (local->stbuf.ia_type)
-		    && (local->hashed_subvol != local->cached_subvol)
-		    && (local->stbuf.ia_nlink == 1)
-                    && (conf && conf->unhashed_sticky_bit)) {
-			local->stbuf.ia_prot.sticky = 1;
-		}
-
-		if (local->layout_mismatch) {
-			local->op_ret = -1;
-			local->op_errno = ESTALE;
-
-                        /* Because for 'root' inode, there is no FRESH lookup
-                         * sent from FUSE layer upon ESTALE, we need to handle
-                         * that one case here */
-                        root_gfid[15] = 1;
-                        if (!local->loc.parent &&
-                            !uuid_compare (local->loc.inode->gfid, root_gfid)) {
-                                dht_do_fresh_lookup_on_root (this, frame);
-                                return 0;
-                        }
-		}
+    if (populate_loc.inode)
+        loc_wipe(&populate_loc);
+    return hash_subvol;
+}
 
-                WIPE (&local->postparent);
+/* Common function call by revalidate/selfheal code path to populate
+   internal xattr if it is not present, mark_during_fresh_lookup value
+   determines either function is call by revalidate_cbk(discover_complete)
+   or call by selfheal code path while fresh lookup.
+   Here we do wind a call serially in case of fresh lookup and
+   for other lookup code path we do wind a call parallel.The reason
+   to wind a call serially is at the time of fresh lookup directory is not
+   discovered and at the time of revalidate_lookup directory is
+   already discovered. So, revalidate codepath can race with setxattr
+   codepath and can get into spurious heals because of an ongoing setxattr.
+   This can slow down revalidates, if healing happens in foreground.
+   However, if healing happens in background, there is no direct performance
+   penalty.
+*/
+int
+dht_common_mark_mdsxattr(call_frame_t *frame, int *errst,
+                         int mark_during_fresh_lookup)
+{
+    dht_local_t *local = NULL;
+    xlator_t *this = NULL;
+    xlator_t *hashed_subvol = NULL;
+    int ret = 0;
+    int i = 0;
+    dict_t *xattrs = NULL;
+    char gfid_local[GF_UUID_BUF_SIZE] = {
+        0,
+    };
+    int32_t zero[1] = {0};
+    dht_conf_t *conf = 0;
+    dht_layout_t *layout = NULL;
+    dht_local_t *copy_local = NULL;
+    call_frame_t *xattr_frame = NULL;
+    gf_boolean_t vol_down = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    this = frame->this;
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    local = frame->local;
+    conf = this->private;
+    layout = local->selfheal.layout;
+    local->mds_heal_fresh_lookup = mark_during_fresh_lookup;
+
+    gf_uuid_unparse(local->gfid, gfid_local);
+
+    /* Code to update hashed subvol consider as a mds subvol
+       and wind a setxattr call on hashed subvol to update
+       internal xattr
+    */
+    if (!local->xattr || !dict_get(local->xattr, conf->mds_xattr_key)) {
+        /* It means no internal MDS xattr has been set yet
+         */
+        /* Check the status of all subvol are up while call
+           this function call by lookup code path
+        */
+        if (mark_during_fresh_lookup) {
+            for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (!conf->subvolume_status[i]) {
+                    vol_down = _gf_true;
+                    break;
+                }
+            }
+            if (vol_down) {
+                gf_msg_debug(this->name, 0,
+                             "subvol %s is down. Unable to "
+                             " save mds subvol on inode for "
+                             " path %s gfid is %s ",
+                             conf->subvolumes[i]->name, local->loc.path,
+                             gfid_local);
+                goto out;
+            }
+        }
 
-                DHT_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno,
-				  local->inode, &local->stbuf, local->xattr,
-                                  &local->postparent);
-	}
+        /* Calculate hashed subvol based on inode and parent node
+         */
+        hashed_subvol = dht_inode_get_hashed_subvol(local->inode, this,
+                                                    &local->loc);
+        if (!hashed_subvol) {
+            gf_msg(this->name, GF_LOG_DEBUG, 0,
+                   DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+                   "Failed to get hashed subvol for path %s"
+                   "gfid is %s ",
+                   local->loc.path, gfid_local);
+            if (errst)
+                (*errst) = 1;
+            ret = -1;
+            goto out;
+        }
+        xattrs = dict_new();
+        if (!xattrs) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+                   "dict_new failed");
+            ret = -1;
+            goto out;
+        }
+        /* Add internal MDS xattr on disk for hashed subvol
+         */
+        ret = dht_dict_set_array(xattrs, conf->mds_xattr_key, zero, 1);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_WARNING, ENOMEM, DHT_MSG_DICT_SET_FAILED,
+                   "Failed to set dictionary"
+                   "  value:key = %s for "
+                   "path %s",
+                   conf->mds_xattr_key, local->loc.path);
+            ret = -1;
+            goto out;
+        }
+        /* Create a new frame to wind a call only while
+           this function call by revalidate_cbk code path
+           To wind a call parallel need to create a new frame
+        */
+        if (mark_during_fresh_lookup) {
+            xattr_frame = create_frame(this, this->ctx->pool);
+            if (!xattr_frame) {
+                ret = -1;
+                goto out;
+            }
+            copy_local = dht_local_init(xattr_frame, &(local->loc), NULL, 0);
+            if (!copy_local) {
+                ret = -1;
+                DHT_STACK_DESTROY(xattr_frame);
+                goto out;
+            }
+            copy_local->stbuf = local->stbuf;
+            copy_local->mds_heal_fresh_lookup = mark_during_fresh_lookup;
+            if (!copy_local->inode)
+                copy_local->inode = inode_ref(local->inode);
+            gf_uuid_copy(copy_local->loc.gfid, local->gfid);
+            FRAME_SU_DO(xattr_frame, dht_local_t);
+            STACK_WIND_COOKIE(xattr_frame, dht_common_mark_mdsxattr_cbk,
+                              hashed_subvol, hashed_subvol,
+                              hashed_subvol->fops->setxattr, &local->loc,
+                              xattrs, 0, NULL);
+        } else {
+            STACK_WIND_COOKIE(frame, dht_common_mark_mdsxattr_cbk,
+                              (void *)hashed_subvol, hashed_subvol,
+                              hashed_subvol->fops->setxattr, &local->loc,
+                              xattrs, 0, NULL);
+        }
+    } else {
+        gf_msg_debug(this->name, 0,
+                     "internal xattr %s is present on subvol"
+                     "on path %s gfid is %s ",
+                     conf->mds_xattr_key, local->loc.path, gfid_local);
+        if (!mark_during_fresh_lookup)
+            dht_selfheal_dir_setattr(frame, &local->loc, &local->stbuf,
+                                     0xffffffff, layout);
+    }
 
-        return 0;
+out:
+    if (xattrs)
+        dict_unref(xattrs);
+    return ret;
 }
 
+/* Get the value of key from dict in the bytewise and save in array after
+   convert from network byte order to host byte order
+*/
+static int32_t
+dht_dict_get_array(dict_t *dict, char *key, int32_t value[], int32_t size,
+                   int *errst)
+{
+    void *ptr = NULL;
+    int32_t len = -1;
+    int32_t vindex = -1;
+    int32_t err = -1;
+    int ret = 0;
+
+    if (dict == NULL) {
+        (*errst) = -1;
+        return -EINVAL;
+    }
+    err = dict_get_ptr_and_len(dict, key, &ptr, &len);
+    if (err != 0) {
+        (*errst) = -1;
+        return err;
+    }
+
+    if (len != (size * sizeof(int32_t))) {
+        (*errst) = -1;
+        return -EINVAL;
+    }
+
+    for (vindex = 0; vindex < size; vindex++) {
+        value[vindex] = ntoh32(*((int32_t *)ptr + vindex));
+        if (value[vindex] < 0)
+            ret = -1;
+    }
+
+    return ret;
+}
 
-int
-dht_lookup_linkfile_create_cbk (call_frame_t *frame, void *cookie,
-				xlator_t *this,
-				int32_t op_ret, int32_t op_errno,
-                                inode_t *inode, struct iatt *stbuf,
-                                struct iatt *preparent, struct iatt *postparent)
-{
-	dht_local_t  *local = NULL;
-	xlator_t     *cached_subvol = NULL;
-	dht_conf_t   *conf = NULL;
-        int           ret = -1;
-
-	local = frame->local;
-	cached_subvol = local->cached_subvol;
-	conf = this->private;
-
-        ret = dht_layout_preset (this, local->cached_subvol, inode);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "failed to set layout for subvolume %s",
-                        cached_subvol ? cached_subvol->name : "<nil>");
-                local->op_ret = -1;
-                local->op_errno = EINVAL;
-                goto unwind;
-        }
+static int
+dht_discover_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, inode_t *inode, struct iatt *stbuf,
+                 dict_t *xattr, struct iatt *postparent)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+    dht_layout_t *layout = NULL;
+    int ret = -1;
+    int is_dir = 0;
+    int32_t check_mds = 0;
+    int is_linkfile = 0;
+    int attempt_unwind = 0;
+    dht_conf_t *conf = 0;
+    char gfid_local[GF_UUID_BUF_SIZE] = {0};
+    char gfid_node[GF_UUID_BUF_SIZE] = {0};
+    int32_t mds_xattr_val[1] = {0};
+    int errst = 0;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", this->private, out);
+    GF_VALIDATE_OR_GOTO("dht", cookie, out);
+
+    local = frame->local;
+    prev = cookie;
+    conf = this->private;
+
+    layout = local->layout;
+
+    /* Check if the gfid is different for file from other node */
+    if (!op_ret && gf_uuid_compare(local->gfid, stbuf->ia_gfid)) {
+        gf_uuid_unparse(stbuf->ia_gfid, gfid_node);
+        gf_uuid_unparse(local->gfid, gfid_local);
+
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_GFID_MISMATCH,
+               "%s: gfid different on %s, gfid local = %s"
+               "gfid other = %s",
+               local->loc.path, prev->name, gfid_local, gfid_node);
+    }
+
+    LOCK(&frame->lock);
+    {
+        /* TODO: assert equal mode on stbuf->st_mode and
+           local->stbuf->st_mode
+
+           else mkdir/chmod/chown and fix
+        */
+
+        ret = dht_layout_merge(this, layout, prev, op_ret, op_errno, xattr);
+        if (ret)
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_MERGE_FAILED,
+                   "%s: failed to merge layouts for subvol %s", local->loc.path,
+                   prev->name);
 
-	local->op_ret = 0;
-	if ((local->stbuf.ia_nlink == 1)
-	    && (conf && conf->unhashed_sticky_bit)) {
-		local->stbuf.ia_prot.sticky = 1;
-	}
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+            gf_msg_debug(this->name, op_errno,
+                         "lookup of %s on %s returned error", local->loc.path,
+                         prev->name);
 
-        if (local->loc.parent)
-                local->postparent.ia_ino = local->loc.parent->ino;
+            goto unlock;
+        }
 
-unwind:
-        WIPE (&local->postparent);
+        is_linkfile = check_is_linkfile(inode, stbuf, xattr,
+                                        conf->link_xattr_name);
+        is_dir = check_is_dir(inode, stbuf, xattr);
 
-	DHT_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno,
-			  local->inode, &local->stbuf, local->xattr,
-                          &local->postparent);
-	return 0;
-}
+        if (is_dir) {
+            local->dir_count++;
+        } else {
+            local->file_count++;
+
+            if (!is_linkfile && !local->cached_subvol) {
+                /* real file */
+                /* Ok, we somehow managed to find a file on
+                 * more than one subvol. ignore this or we
+                 * will end up overwriting information while a
+                 * a thread is potentially unwinding from
+                 * dht_discover_complete
+                 */
+                local->cached_subvol = prev;
+                attempt_unwind = 1;
+            } else {
+                goto unlock;
+            }
+        }
 
+        local->op_ret = 0;
 
-int
-dht_lookup_everywhere_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			   int32_t op_ret, int32_t op_errno,
-                           inode_t *inode, struct iatt *buf, dict_t *xattr,
-                           struct iatt *postparent)
-{
-	dht_conf_t   *conf          = NULL;
-        dht_local_t  *local         = NULL;
-        int           this_call_cnt = 0;
-        call_frame_t *prev          = NULL;
-	int           is_linkfile   = 0;
-	int           is_dir        = 0;
-	xlator_t     *subvol        = NULL;
-	loc_t        *loc           = NULL;
-	xlator_t     *link_subvol   = NULL;
-	xlator_t     *hashed_subvol = NULL;
-	xlator_t     *cached_subvol = NULL;
-        int           ret = -1;
-
-	conf   = this->private;
-
-	local  = frame->local;
-	loc    = &local->loc;
-
-	prev   = cookie;
-	subvol = prev->this;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			if (op_errno != ENOENT)
-				local->op_errno = op_errno;
-			goto unlock;
-		}
-                if (uuid_is_null (local->gfid))
-                        memcpy (local->gfid, buf->ia_gfid, 16);
-
-		is_linkfile = check_is_linkfile (inode, buf, xattr);
-		is_dir = check_is_dir (inode, buf, xattr);
-
-		if (is_linkfile) {
-			link_subvol = dht_linkfile_subvol (this, inode, buf,
-							   xattr);
-			gf_log (this->name, GF_LOG_DEBUG,
-				"found on %s linkfile %s (-> %s)",
-				subvol->name, loc->path,
-				link_subvol ? link_subvol->name : "''");
-			goto unlock;
-		}
-
-                if (is_dir) {
-                        local->dir_count++;
-
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "found on %s directory %s",
-                                subvol->name, loc->path);
-                } else {
-                        local->file_count++;
-
-                        if (!local->cached_subvol) {
-                                /* found one file */
-                                dht_iatt_merge (this, &local->stbuf, buf,
-                                                subvol);
-                                local->xattr = dict_ref (xattr);
-                                local->cached_subvol = subvol;
-                                gf_log (this->name, GF_LOG_DEBUG,
-                                        "found on %s file %s",
-                                        subvol->name, loc->path);
-                                
-                                dht_iatt_merge (this, &local->postparent,
-                                                postparent, subvol);
-                        } else {
-                                gf_log (this->name, GF_LOG_DEBUG,
-                                        "multiple subvolumes (%s and %s) have "
-                                        "file %s", local->cached_subvol->name,
-                                        subvol->name, local->loc.path);
-                        }
-                }
-	}
-unlock:
-	UNLOCK (&frame->lock);
-
-	if (is_linkfile) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"deleting stale linkfile %s on %s",
-			loc->path, subvol->name);
-		dht_linkfile_unlink (frame, this, subvol, loc);
-	}
-
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt)) {
-		hashed_subvol = local->hashed_subvol;
-		cached_subvol = local->cached_subvol;
-
-                if (local->file_count && local->dir_count) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "path %s exists as a file on one subvolume " 
-                                "and directory on another. "
-                                "Please fix it manually",
-                                loc->path);
-                        DHT_STACK_UNWIND (lookup, frame, -1, EIO, NULL, NULL, NULL,
-                                          NULL);
-                        return 0;
-                }
+        if (local->xattr == NULL) {
+            local->xattr = dict_ref(xattr);
+        } else {
+            /* Don't aggregate for files. See BZ#1484709 */
+            if (is_dir)
+                dht_aggregate_xattr(local->xattr, xattr);
+        }
 
-                if (local->dir_count) {
-                        dht_lookup_directory (frame, this, &local->loc);
-                        return 0;
-                }
+        if (local->inode == NULL)
+            local->inode = inode_ref(inode);
 
-		if (!cached_subvol) {
-			DHT_STACK_UNWIND (lookup, frame, -1, ENOENT, NULL, NULL, NULL,
-                                          NULL);
-			return 0;
-		}
-
-                if (!hashed_subvol) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "cannot create linkfile file for %s on %s: "
-                                "hashed subvolume cannot be found.",
-                                loc->path, cached_subvol->name);
-                        
-                        local->op_ret = 0;
-                        local->op_errno = 0;
-
-                        ret = dht_layout_preset (frame->this, cached_subvol,
-                                                 local->inode);
-                        if (ret < 0) {
-                                gf_log (this->name, GF_LOG_DEBUG,
-                                        "failed to set layout for subvol %s",
-                                        cached_subvol ? cached_subvol->name :
-                                        "<nil>");
-                                local->op_ret = -1;
-                                local->op_errno = EINVAL;
-                        }
+        dht_iatt_merge(this, &local->stbuf, stbuf);
+        dht_iatt_merge(this, &local->postparent, postparent);
 
-                        if (local->loc.parent)
-                                local->postparent.ia_ino =
-                                        local->loc.parent->ino;
+        if (!dict_get(xattr, conf->mds_xattr_key)) {
+            goto unlock;
+        } else {
+            gf_msg_debug(this->name, 0,
+                         "internal xattr %s is present on subvol"
+                         "on path %s gfid is %s ",
+                         conf->mds_xattr_key, local->loc.path, gfid_local);
+        }
+        check_mds = dht_dict_get_array(xattr, conf->mds_xattr_key,
+                                       mds_xattr_val, 1, &errst);
+        /* save mds subvol on inode ctx */
+        ret = dht_inode_ctx_mdsvol_set(local->inode, this, prev);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SET_INODE_CTX_FAILED,
+                   "Failed to set hashed subvol for %s vol is %s",
+                   local->loc.path, prev->name);
+        }
+
+        if ((check_mds < 0) && !errst) {
+            local->mds_xattr = dict_ref(xattr);
+            gf_msg_debug(this->name, 0,
+                         "Value of %s is not zero on mds subvol"
+                         "so xattr needs to be healed on non mds"
+                         " path is %s and vol name is %s "
+                         " gfid is %s",
+                         conf->mds_xattr_key, local->loc.path, prev->name,
+                         gfid_local);
+            local->need_xattr_heal = 1;
+            local->mds_subvol = prev;
+        }
+    }
+unlock:
+    UNLOCK(&frame->lock);
+out:
+    /* Make sure, the thread executing dht_discover_complete is the one
+     * which calls STACK_DESTROY (frame). In the case of "attempt_unwind",
+     * this makes sure that the thread don't call dht_frame_return, till
+     * call to dht_discover_complete is done.
+     */
+    if (attempt_unwind) {
+        dht_discover_complete(this, frame);
+    }
 
-                        WIPE (&local->postparent);
+    this_call_cnt = dht_frame_return(frame);
 
-                        DHT_STACK_UNWIND (lookup, frame, local->op_ret,
-                                          local->op_errno, local->inode,
-                                          &local->stbuf, local->xattr,
-                                          &local->postparent);
-                        return 0;
-                }
+    if (is_last_call(this_call_cnt) && !attempt_unwind) {
+        dht_discover_complete(this, frame);
+    }
 
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "linking file %s existing on %s to %s (hash)",
-                        loc->path, cached_subvol->name,
-                        hashed_subvol->name);
-                        
-                dht_linkfile_create (frame, 
-                                     dht_lookup_linkfile_create_cbk,
-                                     cached_subvol, hashed_subvol, loc);
-	}
+    if (is_last_call(this_call_cnt))
+        DHT_STACK_DESTROY(frame);
 
-	return 0;
+    return 0;
 }
 
+static int
+dht_set_file_xattr_req(xlator_t *this, loc_t *loc, dict_t *xattr_req)
+{
+    int ret = -EINVAL;
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+    if (!conf) {
+        goto err;
+    }
+
+    if (!xattr_req) {
+        goto err;
+    }
+
+    /* Used to check whether this is a linkto file.
+     */
+    ret = dict_set_uint32(xattr_req, conf->link_xattr_name, 256);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, DHT_MSG_DICT_SET_FAILED,
+               "Failed to set dictionary value:key = %s for "
+               "path %s",
+               conf->link_xattr_name, loc->path);
+        goto err;
+    }
+
+    /* This is used to make sure we don't unlink linkto files
+     * which are the target of an ongoing file migration.
+     */
+    ret = dict_set_uint32(xattr_req, GLUSTERFS_OPEN_FD_COUNT, 4);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, DHT_MSG_DICT_SET_FAILED,
+               "Failed to set dictionary value:key = %s for "
+               "path %s",
+               GLUSTERFS_OPEN_FD_COUNT, loc->path);
+        goto err;
+    }
+
+    ret = 0;
+err:
+    return ret;
+}
 
-int
-dht_lookup_everywhere (call_frame_t *frame, xlator_t *this, loc_t *loc)
+/* This is a gfid based nameless lookup. Without a name, the hashed subvol
+ * cannot be calculated so a lookup is sent to all subvols.
+ */
+static int
+dht_do_discover(call_frame_t *frame, xlator_t *this, loc_t *loc)
 {
-	dht_conf_t     *conf = NULL;
-	dht_local_t    *local = NULL;
-	int             i = 0;
-	int             call_cnt = 0;
+    int ret;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int call_cnt = 0;
+    int op_errno = EINVAL;
+    int i = 0;
+    call_frame_t *discover_frame = NULL;
+
+    conf = this->private;
+    local = frame->local;
+
+    /* As we do not know if this is a file or directory, request
+     * both file and directory xattrs
+     */
+    ret = dht_set_file_xattr_req(this, loc, local->xattr_req);
+    if (ret) {
+        goto err;
+    }
+
+    ret = dht_set_dir_xattr_req(this, loc, local->xattr_req);
+    if (ret) {
+        goto err;
+    }
+
+    if (loc_is_root(loc)) {
+        /* Request the DHT commit hash xattr (trusted.glusterfs.dht.commithash)
+         * set on the brick root.
+         */
+        ret = dict_set_uint32(local->xattr_req, conf->commithash_xattr_name,
+                              sizeof(uint32_t));
+    }
 
-	conf = this->private;
-	local = frame->local;
-        if (!conf)
-                goto out;
+    call_cnt = conf->subvolume_cnt;
+    local->call_cnt = call_cnt;
 
-	call_cnt = conf->subvolume_cnt;
-	local->call_cnt = call_cnt;
+    local->layout = dht_layout_new(this, conf->subvolume_cnt);
 
-	if (!local->inode)
-		local->inode = inode_ref (loc->inode);
+    if (!local->layout) {
+        op_errno = ENOMEM;
+        goto err;
+    }
 
-	for (i = 0; i < call_cnt; i++) {
-		STACK_WIND (frame, dht_lookup_everywhere_cbk,
-			    conf->subvolumes[i],
-			    conf->subvolumes[i]->fops->lookup,
-			    loc, local->xattr_req);
-	}
+    gf_uuid_copy(local->gfid, loc->gfid);
 
-        return 0;
+    discover_frame = copy_frame(frame);
+    if (!discover_frame) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    discover_frame->local = local;
+    frame->local = NULL;
+    local->main_frame = frame;
+
+    for (i = 0; i < call_cnt; i++) {
+        STACK_WIND_COOKIE(discover_frame, dht_discover_cbk, conf->subvolumes[i],
+                          conf->subvolumes[i],
+                          conf->subvolumes[i]->fops->lookup, &local->loc,
+                          local->xattr_req);
+    }
+
+    return 0;
+
+err:
+    DHT_STACK_UNWIND(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+
+    return 0;
+}
+
+/* Code to call syntask to heal custom xattr from hashed subvol
+   to non hashed subvol
+*/
+int
+dht_dir_xattr_heal(xlator_t *this, dht_local_t *local, int *op_errno)
+{
+    dht_local_t *copy_local = NULL;
+    call_frame_t *copy = NULL;
+    int ret = -1;
+    char gfid_local[GF_UUID_BUF_SIZE] = {0};
+
+    if (gf_uuid_is_null(local->gfid)) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DIR_XATTR_HEAL_FAILED,
+               "No gfid exists for path %s "
+               "so healing xattr is not possible",
+               local->loc.path);
+        *op_errno = EIO;
+        goto out;
+    }
+
+    gf_uuid_unparse(local->gfid, gfid_local);
+    copy = create_frame(this, this->ctx->pool);
+    if (copy) {
+        copy_local = dht_local_init(copy, &(local->loc), NULL, 0);
+        if (!copy_local) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM,
+                   DHT_MSG_DIR_XATTR_HEAL_FAILED,
+                   "Memory allocation failed "
+                   "for path %s gfid %s ",
+                   local->loc.path, gfid_local);
+            *op_errno = ENOMEM;
+            DHT_STACK_DESTROY(copy);
+        } else {
+            copy_local->stbuf = local->stbuf;
+            gf_uuid_copy(copy_local->loc.gfid, local->gfid);
+            copy_local->mds_subvol = local->mds_subvol;
+            FRAME_SU_DO(copy, dht_local_t);
+            ret = synctask_new(this->ctx->env, dht_dir_heal_xattrs,
+                               dht_dir_heal_xattrs_done, copy, copy);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, ENOMEM,
+                       DHT_MSG_DIR_XATTR_HEAL_FAILED,
+                       "Synctask creation failed to heal xattr "
+                       "for path %s gfid %s ",
+                       local->loc.path, gfid_local);
+                *op_errno = ENOMEM;
+                DHT_STACK_DESTROY(copy);
+            }
+        }
+    }
 out:
-        DHT_STACK_UNWIND (lookup, frame, -1, EINVAL, NULL, NULL, NULL, NULL);
-        return 0;
+    return ret;
 }
 
+static int
+dht_needs_selfheal(call_frame_t *frame, xlator_t *this)
+{
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+    int needs_selfheal = 0;
+    int ret = 0;
+
+    local = frame->local;
+    layout = local->layout;
+
+    if (local->need_attrheal || local->need_xattr_heal ||
+        local->need_selfheal) {
+        needs_selfheal = 1;
+    }
+
+    ret = dht_layout_normalize(this, &local->loc, layout);
+
+    if (ret != 0) {
+        gf_msg_debug(this->name, 0, "fixing assignment on %s", local->loc.path);
+        needs_selfheal = 1;
+    }
+    return needs_selfheal;
+}
+
+static int
+is_permission_different(ia_prot_t *prot1, ia_prot_t *prot2)
+{
+    if ((prot1->owner.read != prot2->owner.read) ||
+        (prot1->owner.write != prot2->owner.write) ||
+        (prot1->owner.exec != prot2->owner.exec) ||
+        (prot1->group.read != prot2->group.read) ||
+        (prot1->group.write != prot2->group.write) ||
+        (prot1->group.exec != prot2->group.exec) ||
+        (prot1->other.read != prot2->other.read) ||
+        (prot1->other.write != prot2->other.write) ||
+        (prot1->other.exec != prot2->other.exec) ||
+        (prot1->suid != prot2->suid) || (prot1->sgid != prot2->sgid) ||
+        (prot1->sticky != prot2->sticky)) {
+        return 1;
+    } else {
+        return 0;
+    }
+}
 
 int
-dht_lookup_linkfile_cbk (call_frame_t *frame, void *cookie,
-                         xlator_t *this, int op_ret, int op_errno,
-                         inode_t *inode, struct iatt *stbuf, dict_t *xattr,
-                         struct iatt *postparent)
-{
-        call_frame_t *prev          = NULL;
-	dht_local_t  *local         = NULL;
-	xlator_t     *subvol        = NULL;
-	loc_t        *loc           = NULL;
-	dht_conf_t   *conf          = NULL;
-        int           ret           = 0;
-
-        prev   = cookie;
-	subvol = prev->this;
-	conf   = this->private;
-	local  = frame->local;
-	loc    = &local->loc;
+dht_lookup_dir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int op_ret, int op_errno, inode_t *inode, struct iatt *stbuf,
+                   dict_t *xattr, struct iatt *postparent)
+{
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+    dht_layout_t *layout = NULL;
+    int ret = -1;
+    int is_dir = 0;
+    int32_t check_mds = 0;
+    int errst = 0;
+    char gfid_local[GF_UUID_BUF_SIZE] = {0};
+    char gfid_node[GF_UUID_BUF_SIZE] = {0};
+    int32_t mds_xattr_val[1] = {0};
+
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", this->private, out);
+    GF_VALIDATE_OR_GOTO("dht", cookie, out);
+
+    local = frame->local;
+    prev = cookie;
+    conf = this->private;
+
+    layout = local->layout;
+    gf_msg_debug(this->name, op_errno,
+                 "%s: lookup on %s returned with op_ret = %d, op_errno = %d",
+                 local->loc.path, prev->name, op_ret, op_errno);
+
+    /* The first successful lookup*/
+    if (!op_ret && gf_uuid_is_null(local->gfid)) {
+        memcpy(local->gfid, stbuf->ia_gfid, 16);
+    }
+    if (!gf_uuid_is_null(local->gfid)) {
+        gf_uuid_unparse(local->gfid, gfid_local);
+    }
+
+    /* Check if the gfid is different for file from other node */
+    if (!op_ret && gf_uuid_compare(local->gfid, stbuf->ia_gfid)) {
+        gf_uuid_unparse(stbuf->ia_gfid, gfid_node);
+
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_GFID_MISMATCH,
+               "%s: gfid different on %s."
+               " gfid local = %s, gfid subvol = %s",
+               local->loc.path, prev->name, gfid_local, gfid_node);
+    }
+
+    LOCK(&frame->lock);
+    {
+        /* TODO: assert equal mode on stbuf->st_mode and
+           local->stbuf->st_mode
+           else mkdir/chmod/chown and fix
+        */
+        ret = dht_layout_merge(this, layout, prev, op_ret, op_errno, xattr);
 
         if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"lookup of %s on %s (following linkfile) failed (%s)",
-			local->loc.path, subvol->name, strerror (op_errno));
-                goto err;
-	}
+            local->op_errno = op_errno;
 
-        if (check_is_dir (inode, stbuf, xattr)) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "lookup of %s on %s (following linkfile) reached dir",
-                        local->loc.path, subvol->name);
-                goto err;
+            /* The GFID is missing on this subvol. Force a heal. */
+            if (op_errno == ENODATA) {
+                local->need_lookup_everywhere = 1;
+            }
+            goto unlock;
         }
 
-        if (check_is_linkfile (inode, stbuf, xattr)) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "lookup of %s on %s (following linkfile) reached link",
-                        local->loc.path, subvol->name);
-                goto err;
+        is_dir = check_is_dir(inode, stbuf, xattr);
+        if (!is_dir) {
+            gf_msg_debug(this->name, 0,
+                         "%s: lookup on %s returned non dir 0%o"
+                         "calling lookup_everywhere",
+                         local->loc.path, prev->name, stbuf->ia_type);
+
+            local->need_lookup_everywhere = 1;
+            goto unlock;
         }
 
-	if ((stbuf->ia_nlink == 1)
-	    && (conf && conf->unhashed_sticky_bit)) {
-		stbuf->ia_prot.sticky = 1;
-	}
-        dht_itransform (this, prev->this, stbuf->ia_ino, &stbuf->ia_ino);
-        if (local->loc.parent)
-                postparent->ia_ino = local->loc.parent->ino;
-        
-	ret = dht_layout_preset (this, prev->this, inode);
-	if (ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "failed to set layout for subvolume %s",
-                        prev->this->name);
-		op_ret   = -1;
-		op_errno = EINVAL;
-		goto out;
-	}
+        local->op_ret = 0;
+        if (local->xattr == NULL) {
+            local->xattr = dict_ref(xattr);
+        } else {
+            dht_aggregate_xattr(local->xattr, xattr);
+        }
 
-out:
-        WIPE (postparent);
+        if (__is_root_gfid(stbuf->ia_gfid)) {
+            ret = dht_dir_has_layout(xattr, conf->xattr_name);
+            if (ret >= 0) {
+                if (is_greater_time(local->prebuf.ia_ctime,
+                                    local->prebuf.ia_ctime_nsec,
+                                    stbuf->ia_ctime, stbuf->ia_ctime_nsec)) {
+                    /* Choose source */
+                    local->prebuf.ia_gid = stbuf->ia_gid;
+                    local->prebuf.ia_uid = stbuf->ia_uid;
+
+                    local->prebuf.ia_ctime = stbuf->ia_ctime;
+                    local->prebuf.ia_ctime_nsec = stbuf->ia_ctime_nsec;
+                    local->prebuf.ia_prot = stbuf->ia_prot;
+                }
+            }
+        }
 
-        DHT_STACK_UNWIND (lookup, frame, op_ret, op_errno, inode, stbuf, xattr,
-                          postparent);
+        if (local->stbuf.ia_type != IA_INVAL) {
+            /* This is not the first subvol to respond
+             * Compare values to see if attrs need to be healed
+             */
+            if ((local->stbuf.ia_gid != stbuf->ia_gid) ||
+                (local->stbuf.ia_uid != stbuf->ia_uid) ||
+                (is_permission_different(&local->stbuf.ia_prot,
+                                         &stbuf->ia_prot))) {
+                local->need_attrheal = 1;
+            }
+        }
 
-        return 0;
+        if (local->inode == NULL)
+            local->inode = inode_ref(inode);
 
-err:
-        dht_lookup_everywhere (frame, this, loc);
+        dht_iatt_merge(this, &local->stbuf, stbuf);
+        dht_iatt_merge(this, &local->postparent, postparent);
 
-        return 0;
-}
+        if (!dict_get(xattr, conf->mds_xattr_key)) {
+            gf_msg_debug(this->name, 0,
+                         "%s: mds xattr %s is not present "
+                         "on %s(gfid = %s)",
+                         local->loc.path, conf->mds_xattr_key, prev->name,
+                         gfid_local);
+            goto unlock;
+        }
 
+        /* Save the mds subvol info and stbuf. This is the value that will
+         * be used for healing
+         */
+        local->mds_subvol = prev;
+        local->mds_stbuf = *stbuf;
 
-int
-dht_lookup_directory (call_frame_t *frame, xlator_t *this, loc_t *loc)
-{
-        int           call_cnt = 0;
-        int           i = 0;
-        dht_conf_t   *conf = NULL;
-        dht_local_t  *local = NULL;
+        /* Save mds subvol on inode ctx */
 
-        conf = this->private;
-        local = frame->local;
+        ret = dht_inode_ctx_mdsvol_set(local->inode, this, prev);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SET_INODE_CTX_FAILED,
+                   "%s: Failed to set mds (%s)", local->loc.path, prev->name);
+        }
+        check_mds = dht_dict_get_array(xattr, conf->mds_xattr_key,
+                                       mds_xattr_val, 1, &errst);
+        if ((check_mds < 0) && !errst) {
+            /* Check if xattrs need to be healed on the directories */
+            local->mds_xattr = dict_ref(xattr);
+            gf_msg_debug(this->name, 0,
+                         "%s: %s is not zero on %s. Xattrs need to be healed."
+                         "(gfid = %s)",
+                         local->loc.path, conf->mds_xattr_key, prev->name,
+                         gfid_local);
+            local->need_xattr_heal = 1;
+        }
+    }
 
-        if (!conf)
-                goto unwind;
+unlock:
+    UNLOCK(&frame->lock);
 
-        call_cnt        = conf->subvolume_cnt;
-        local->call_cnt = call_cnt;
+    this_call_cnt = dht_frame_return(frame);
 
-        local->layout = dht_layout_new (this, conf->subvolume_cnt);
-        if (!local->layout) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                goto unwind;
+    if (is_last_call(this_call_cnt)) {
+        /* If the mds subvol is not set correctly*/
+        if (!__is_root_gfid(local->gfid) &&
+            (!dict_get(local->xattr, conf->mds_xattr_key))) {
+            local->need_selfheal = 1;
         }
 
-        for (i = 0; i < call_cnt; i++) {
-                STACK_WIND (frame, dht_lookup_dir_cbk,
-                            conf->subvolumes[i],
-                            conf->subvolumes[i]->fops->lookup,
-                            &local->loc, local->xattr_req);
+        /* No need to call xattr heal code if volume count is 1
+         */
+        if (conf->subvolume_cnt == 1) {
+            local->need_xattr_heal = 0;
         }
-        return 0;
-unwind:
-        DHT_STACK_UNWIND (lookup, frame, -1, ENOMEM, NULL, NULL, NULL, NULL);
-        return 0;
 
-}
+        if (local->need_selfheal || local->need_lookup_everywhere) {
+            /* Set the gfid-req so posix will set the GFID*/
+            if (!gf_uuid_is_null(local->gfid)) {
+                /* Ok, this should _never_ happen */
+                ret = dict_set_static_bin(local->xattr_req, "gfid-req",
+                                          local->gfid, 16);
+            } else {
+                if (!gf_uuid_is_null(local->gfid_req))
+                    ret = dict_set_static_bin(local->xattr_req, "gfid-req",
+                                              local->gfid_req, 16);
+            }
+        }
 
+        if (local->need_lookup_everywhere) {
+            local->need_lookup_everywhere = 0;
+            dht_lookup_everywhere(frame, this, &local->loc);
+            return 0;
+        }
 
-int
-dht_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                int op_ret, int op_errno,
-                inode_t *inode, struct iatt *stbuf, dict_t *xattr,
-                struct iatt *postparent)
-{
-        char          is_linkfile   = 0;
-        char          is_dir        = 0;
-        xlator_t     *subvol        = NULL;
-        dht_conf_t   *conf          = NULL;
-        dht_local_t  *local         = NULL;
-        loc_t        *loc           = NULL;
-        call_frame_t *prev          = NULL;
-        int           ret           = 0;
-        uint64_t      tmp_layout    = 0;
-        dht_layout_t *parent_layout = NULL;
-
-        conf  = this->private;
-        if (!conf)
-                goto out;
+        if (local->op_ret == 0) {
+            if (dht_needs_selfheal(frame, this)) {
+                goto selfheal;
+            }
 
-        prev  = cookie;
-        local = frame->local;
-        loc   = &local->loc;
+            dht_layout_set(this, local->inode, layout);
+            if (local->inode) {
+                dht_inode_ctx_time_update(local->inode, this, &local->stbuf, 1);
+            }
 
-        /* This is required for handling stale linkfile deletion,
-         * or any more call which happens from this 'loc'.
-         */
-        if (uuid_is_null (local->gfid) && !op_ret)
-                memcpy (local->gfid, stbuf->ia_gfid, 16);
-
-	if (ENTRY_MISSING (op_ret, op_errno)) {
-                if (conf->search_unhashed == GF_DHT_LOOKUP_UNHASHED_ON) {
-			local->op_errno = ENOENT;
-			dht_lookup_everywhere (frame, this, loc);
-			return 0;
-		}
-                if ((conf->search_unhashed == GF_DHT_LOOKUP_UNHASHED_AUTO) &&
-                    (loc->parent)) {
-                        ret = inode_ctx_get (loc->parent, this, &tmp_layout);
-                        parent_layout = (dht_layout_t *)(long)tmp_layout;
-                        if (parent_layout->search_unhashed) {
-                                local->op_errno = ENOENT;
-                                dht_lookup_everywhere (frame, this, loc);
-                                return 0;
-                        }
-                }
-	}
-
- 	if (op_ret == 0) {
- 		is_dir      = check_is_dir (inode, stbuf, xattr);
- 		if (is_dir) {
- 			local->inode = inode_ref (inode);
- 			local->xattr = dict_ref (xattr);
- 		}
- 	}
-
- 	if (is_dir || (op_ret == -1 && op_errno == ENOTCONN)) {
-                dht_lookup_directory (frame, this, &local->loc);
-                return 0;
- 	}
- 
-        if (op_ret == -1)
-                goto out;
+            if (local->loc.parent) {
+                dht_inode_ctx_time_update(local->loc.parent, this,
+                                          &local->postparent, 1);
+            }
+        }
 
-        is_linkfile = check_is_linkfile (inode, stbuf, xattr);
-        is_dir      = check_is_dir (inode, stbuf, xattr);
-
-        if (!is_dir && !is_linkfile) {
-                /* non-directory and not a linkfile */
-
-		dht_itransform (this, prev->this, stbuf->ia_ino,
-				&stbuf->ia_ino);
-                if (loc->parent)
-                        postparent->ia_ino = loc->parent->ino;
-
-		ret = dht_layout_preset (this, prev->this, inode);
-		if (ret < 0) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"could not set pre-set layout for subvolume %s",
-				prev->this->name);
-			op_ret   = -1;
-			op_errno = EINVAL;
-			goto out;
-		}
-                goto out; 
-	}
+        DHT_STRIP_PHASE1_FLAGS(&local->stbuf);
+        dht_set_fixed_dir_stat(&local->postparent);
+        /* Delete mds xattr at the time of STACK UNWIND */
+        if (local->xattr)
+            GF_REMOVE_INTERNAL_XATTR(conf->mds_xattr_key, local->xattr);
 
-        if (is_linkfile) {
-                subvol = dht_linkfile_subvol (this, inode, stbuf, xattr);
-
-                if (!subvol) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "linkfile not having link subvolume. path=%s",
-                                loc->path);
-			dht_lookup_everywhere (frame, this, loc);
-			return 0;
-                }
+        DHT_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno,
+                         local->inode, &local->stbuf, local->xattr,
+                         &local->postparent);
+    }
 
-		STACK_WIND (frame, dht_lookup_linkfile_cbk,
-			    subvol, subvol->fops->lookup,
-			    &local->loc, local->xattr_req);
-        }
+    return 0;
 
-        return 0;
+selfheal:
+    FRAME_SU_DO(frame, dht_local_t);
+    ret = dht_selfheal_directory(frame, dht_lookup_selfheal_cbk, &local->loc,
+                                 layout);
+out:
+    return ret;
+}
 
+static int
+dht_lookup_directory(call_frame_t *frame, xlator_t *this, loc_t *loc)
+{
+    int call_cnt = 0;
+    int i = 0;
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    int ret = 0;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", this, unwind);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, unwind);
+    GF_VALIDATE_OR_GOTO("dht", this->private, unwind);
+    GF_VALIDATE_OR_GOTO("dht", loc, unwind);
+
+    conf = this->private;
+    local = frame->local;
+
+    call_cnt = conf->subvolume_cnt;
+    local->call_cnt = call_cnt;
+
+    local->layout = dht_layout_new(this, conf->subvolume_cnt);
+    if (!local->layout) {
+        goto unwind;
+    }
+
+    if (local->xattr != NULL) {
+        dict_unref(local->xattr);
+        local->xattr = NULL;
+    }
+
+    if (!gf_uuid_is_null(local->gfid)) {
+        /* use this gfid in order to heal any missing ones */
+        ret = dict_set_gfuuid(local->xattr_req, "gfid-req", local->gfid, true);
+        if (ret)
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+                   "%s: Failed to set dictionary value:"
+                   " key = gfid-req",
+                   local->loc.path);
+    }
+
+    for (i = 0; i < call_cnt; i++) {
+        STACK_WIND_COOKIE(
+            frame, dht_lookup_dir_cbk, conf->subvolumes[i], conf->subvolumes[i],
+            conf->subvolumes[i]->fops->lookup, &local->loc, local->xattr_req);
+    }
+    return 0;
+unwind:
+    DHT_STACK_UNWIND(lookup, frame, -1, ENOMEM, NULL, NULL, NULL, NULL);
 out:
-        /* 
-         * FIXME: postparent->ia_size and postparent->st_blocks do not have 
-         * correct values. since, postparent corresponds to a directory these 
-         * two members should have values equal to sum of corresponding values
-         * from each of the subvolume. See dht_iatt_merge for reference.
+    return 0;
+}
+
+int
+dht_revalidate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int op_ret, int op_errno, inode_t *inode, struct iatt *stbuf,
+                   dict_t *xattr, struct iatt *postparent)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+    dht_layout_t *layout = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+    int is_dir = 0;
+    int is_linkfile = 0;
+    int follow_link = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    uint32_t vol_commit_hash = 0;
+    xlator_t *subvol = NULL;
+    int32_t check_mds = 0;
+    int errst = 0, i = 0;
+    int32_t mds_xattr_val[1] = {0};
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO("dht", this, err);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, err);
+    GF_VALIDATE_OR_GOTO("dht", cookie, err);
+    GF_VALIDATE_OR_GOTO("dht", this->private, err);
+
+    local = frame->local;
+    prev = cookie;
+    conf = this->private;
+
+    if (!conf->vch_forced) {
+        /* Update the commithash value if available
          */
+        ret = dict_get_uint32(xattr, conf->commithash_xattr_name,
+                              &vol_commit_hash);
+        if (ret == 0) {
+            conf->vol_commit_hash = vol_commit_hash;
+        }
+    }
 
-        WIPE (postparent);
+    gf_uuid_unparse(local->loc.gfid, gfid);
 
-        DHT_STACK_UNWIND (lookup, frame, op_ret, op_errno, inode, stbuf, xattr,
-                          postparent);
-        return 0;
-}
+    gf_msg_debug(this->name, op_errno,
+                 "%s: revalidate lookup on %s returned op_ret %d",
+                 local->loc.path, prev->name, op_ret);
 
+    LOCK(&frame->lock);
+    {
+        if (gf_uuid_is_null(local->gfid)) {
+            memcpy(local->gfid, local->loc.gfid, 16);
+        }
 
-int
-dht_lookup (call_frame_t *frame, xlator_t *this,
-            loc_t *loc, dict_t *xattr_req)
-{
-        xlator_t     *subvol = NULL;
-        xlator_t     *hashed_subvol = NULL;
-        xlator_t     *cached_subvol = NULL;
-        dht_local_t  *local  = NULL;
-	dht_conf_t   *conf = NULL;
-        int           ret    = -1;
-        int           op_errno = -1;
-	dht_layout_t *layout = NULL;
-	int           i = 0;
-	int           call_cnt = 0;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	conf = this->private;
-        if (!conf)
-                goto err;
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+
+            if ((op_errno != ENOTCONN) && (op_errno != ENOENT) &&
+                (op_errno != ESTALE)) {
+                gf_msg(this->name, GF_LOG_INFO, op_errno,
+                       DHT_MSG_REVALIDATE_CBK_INFO,
+                       "Revalidate: subvolume %s for %s "
+                       "(gfid = %s) returned -1",
+                       prev->name, local->loc.path, gfid);
+            }
+            if (op_errno == ESTALE) {
+                /* propagate the ESTALE to parent.
+                 * setting local->return_estale would send
+                 * ESTALE to parent. */
+                local->return_estale = 1;
+            }
+
+            /* if it is ENOENT, we may have to do a
+             * 'lookup_everywhere()' to make sure
+             * the file is not migrated */
+            if (op_errno == ENOENT) {
+                if (IA_ISREG(local->loc.inode->ia_type)) {
+                    gf_msg_debug(this->name, 0,
+                                 "found ENOENT for %s. "
+                                 "Setting "
+                                 "need_lookup_everywhere"
+                                 " flag to 1",
+                                 local->loc.path);
+
+                    local->need_lookup_everywhere = 1;
+                } else if (IA_ISDIR(local->loc.inode->ia_type)) {
+                    layout = local->layout;
+                    for (i = 0; i < layout->cnt; i++) {
+                        if (layout->list[i].xlator == prev) {
+                            layout->list[i].err = op_errno;
+                            break;
+                        }
+                    }
 
-        local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-        if (!dht_filter_loc_subvol_key (this, loc, &local->loc,
-                                        &hashed_subvol)) {
-                ret = loc_dup (loc, &local->loc);
-                if (ret == -1) {
-                        op_errno = errno;
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "copying location failed for path=%s",
-                                loc->path);
-                        goto err;
+                    local->need_selfheal = 1;
                 }
+            }
+
+            /* The GFID is missing on this subvol. Lookup everywhere to force a
+             * gfid heal
+             */
+            if ((op_errno == ENODATA) &&
+                (IA_ISDIR(local->loc.inode->ia_type))) {
+                local->need_lookup_everywhere = 1;
+            }
+
+            goto unlock;
         }
 
-	if (xattr_req) {
-		local->xattr_req = dict_ref (xattr_req);
-	} else {
-		local->xattr_req = dict_new ();
-	}
+        if ((!IA_ISINVAL(local->inode->ia_type)) &&
+            stbuf->ia_type != local->inode->ia_type) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_FILE_TYPE_MISMATCH,
+                   "mismatching filetypes 0%o v/s 0%o for %s,"
+                   " gfid = %s",
+                   (stbuf->ia_type), (local->inode->ia_type), local->loc.path,
+                   gfid);
 
-        if (!hashed_subvol)
-                hashed_subvol = dht_subvol_get_hashed (this, loc);
-	cached_subvol = dht_subvol_get_cached (this, loc->inode);
+            local->op_ret = -1;
+            local->op_errno = EINVAL;
 
-	local->cached_subvol = cached_subvol;
-	local->hashed_subvol = hashed_subvol;
+            goto unlock;
+        }
 
-        if (is_revalidate (loc)) {
-		local->layout = layout = dht_layout_get (this, loc->inode);
+        layout = local->layout;
 
-                if (!layout) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "revalidate without cache. path=%s",
-                                loc->path);
-                        op_errno = EINVAL;
-                        goto err;
+        is_dir = check_is_dir(inode, stbuf, xattr);
+        is_linkfile = check_is_linkfile(inode, stbuf, xattr,
+                                        conf->link_xattr_name);
+        if (is_linkfile) {
+            follow_link = 1;
+            goto unlock;
+        }
+        if (is_dir) {
+            ret = dht_dir_has_layout(xattr, conf->xattr_name);
+            if (ret >= 0) {
+                if (is_greater_time(local->prebuf.ia_ctime,
+                                    local->prebuf.ia_ctime_nsec,
+                                    stbuf->ia_ctime, stbuf->ia_ctime_nsec)) {
+                    /* Choose source */
+                    local->prebuf.ia_gid = stbuf->ia_gid;
+                    local->prebuf.ia_uid = stbuf->ia_uid;
+
+                    local->prebuf.ia_ctime = stbuf->ia_ctime;
+                    local->prebuf.ia_ctime_nsec = stbuf->ia_ctime_nsec;
+
+                    if (__is_root_gfid(stbuf->ia_gfid))
+                        local->prebuf.ia_prot = stbuf->ia_prot;
+                }
+            }
+
+            if (local->stbuf.ia_type != IA_INVAL) {
+                if ((local->stbuf.ia_gid != stbuf->ia_gid) ||
+                    (local->stbuf.ia_uid != stbuf->ia_uid) ||
+                    is_permission_different(&local->stbuf.ia_prot,
+                                            &stbuf->ia_prot)) {
+                    local->need_attrheal = 1;
                 }
+            }
+
+            if (!dict_get(xattr, conf->mds_xattr_key)) {
+                gf_msg_debug(this->name, 0,
+                             "%s: internal xattr %s is not present"
+                             " on subvol %s(gfid is %s)",
+                             local->loc.path, conf->mds_xattr_key, prev->name,
+                             gfid);
+            } else {
+                check_mds = dht_dict_get_array(xattr, conf->mds_xattr_key,
+                                               mds_xattr_val, 1, &errst);
+                local->mds_subvol = prev;
+                local->mds_stbuf.ia_gid = stbuf->ia_gid;
+                local->mds_stbuf.ia_uid = stbuf->ia_uid;
+                local->mds_stbuf.ia_prot = stbuf->ia_prot;
+
+                /* save mds subvol on inode ctx */
+                ret = dht_inode_ctx_mdsvol_set(local->inode, this, prev);
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0,
+                           DHT_MSG_SET_INODE_CTX_FAILED,
+                           "Failed to set MDS subvol for %s vol is %s",
+                           local->loc.path, prev->name);
+                }
+                if ((check_mds < 0) && !errst) {
+                    /* Check if xattrs need to be healed on the directory
+                     */
+                    local->mds_xattr = dict_ref(xattr);
+                    gf_msg_debug(this->name, 0,
+                                 "Value of %s is not zero on "
+                                 "hashed subvol so xattr needs to"
+                                 " be healed on non hashed"
+                                 " path is %s and vol name is %s "
+                                 " gfid is %s",
+                                 conf->mds_xattr_key, local->loc.path,
+                                 prev->name, gfid);
+                    local->need_xattr_heal = 1;
+                }
+            }
+            ret = dht_layout_dir_mismatch(this, layout, prev, &local->loc,
+                                          xattr);
+            if (ret != 0) {
+                /* In memory layout does not match on-disk layout.
+                 */
+                gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_LAYOUT_MISMATCH,
+                       "Mismatching layouts for %s, gfid = %s", local->loc.path,
+                       gfid);
+
+                local->layout_mismatch = 1;
+
+                goto unlock;
+            }
+        }
 
-		if (layout->gen && (layout->gen < conf->gen)) {
-			gf_log (this->name, GF_LOG_TRACE,
-				"incomplete layout failure for path=%s",
-				loc->path);
-
-                        dht_layout_unref (this, local->layout);
-                        local->layout = NULL;
-			goto do_fresh_lookup;
-		}
-
-		local->inode    = inode_ref (loc->inode);
-		local->ia_ino   = loc->inode->ino;
-		
-		local->call_cnt = layout->cnt;
-		call_cnt = local->call_cnt;
-		
-		/* NOTE: we don't require 'trusted.glusterfs.dht.linkto' attribute,
-		 *       revalidates directly go to the cached-subvolume.
-		 */
-		ret = dict_set_uint32 (local->xattr_req, 
-				       "trusted.glusterfs.dht", 4 * 4);
-
-		for (i = 0; i < layout->cnt; i++) {
-			subvol = layout->list[i].xlator;
-			
-			STACK_WIND (frame, dht_revalidate_cbk,
-				    subvol, subvol->fops->lookup,
-				    &local->loc, local->xattr_req);
-
-			if (!--call_cnt)
-				break;
-		}
+        gf_uuid_copy(local->stbuf.ia_gfid, stbuf->ia_gfid);
+        dht_iatt_merge(this, &local->stbuf, stbuf);
+        dht_iatt_merge(this, &local->postparent, postparent);
+
+        local->op_ret = 0;
+
+        if (!local->xattr) {
+            local->xattr = dict_ref(xattr);
+        } else if (is_dir) {
+            dht_aggregate_xattr(local->xattr, xattr);
+        }
+    }
+unlock:
+    UNLOCK(&frame->lock);
+
+    if (follow_link) {
+        /* Found a linkto file. Follow it to see if the target file exists
+         */
+        gf_uuid_copy(local->gfid, stbuf->ia_gfid);
+
+        subvol = dht_linkfile_subvol(this, inode, stbuf, xattr);
+        if (!subvol) {
+            op_errno = ESTALE;
+            local->op_ret = -1;
         } else {
-        do_fresh_lookup:
-		/* TODO: remove the hard-coding */
-		ret = dict_set_uint32 (local->xattr_req, 
-				       "trusted.glusterfs.dht", 4 * 4);
-
-		ret = dict_set_uint32 (local->xattr_req, 
-				       "trusted.glusterfs.dht.linkto", 256);
-
-                if (!hashed_subvol) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"no subvolume in layout for path=%s, "
-				"checking on all the subvols to see if "
-				"it is a directory", loc->path);
- 			call_cnt        = conf->subvolume_cnt;
- 			local->call_cnt = call_cnt;
- 			
- 			local->layout = dht_layout_new (this,
-                                                        conf->subvolume_cnt);
- 			if (!local->layout) {
- 				op_errno = ENOMEM;
- 				gf_log (this->name, GF_LOG_ERROR,
- 					"Out of memory");
- 				goto err;
- 			}
-
-			for (i = 0; i < call_cnt; i++) {
- 				STACK_WIND (frame, dht_lookup_dir_cbk,
- 					    conf->subvolumes[i],
- 					    conf->subvolumes[i]->fops->lookup,
- 					    &local->loc, local->xattr_req);
- 			}
- 			return 0;
+            STACK_WIND_COOKIE(frame, dht_lookup_linkfile_cbk, subvol, subvol,
+                              subvol->fops->lookup, &local->loc,
+                              local->xattr_req);
+            return 0;
+        }
+    }
+
+    this_call_cnt = dht_frame_return(frame);
+
+    if (is_last_call(this_call_cnt)) {
+        if (!IA_ISDIR(local->stbuf.ia_type) &&
+            (local->hashed_subvol != local->cached_subvol) &&
+            (local->stbuf.ia_nlink == 1) &&
+            (conf && conf->unhashed_sticky_bit)) {
+            local->stbuf.ia_prot.sticky = 1;
+        }
+        /* No need to call heal code if volume count is 1
+         */
+        if (conf->subvolume_cnt == 1)
+            local->need_xattr_heal = 0;
+
+        if (IA_ISDIR(local->stbuf.ia_type)) {
+            /* No mds xattr found. Trigger a heal to set it */
+            if (!__is_root_gfid(local->loc.inode->gfid) &&
+                (!dict_get(local->xattr, conf->mds_xattr_key)))
+                local->need_selfheal = 1;
+
+            if (dht_needs_selfheal(frame, this)) {
+                if (!__is_root_gfid(local->loc.inode->gfid)) {
+                    if (local->mds_subvol) {
+                        local->stbuf.ia_gid = local->mds_stbuf.ia_gid;
+                        local->stbuf.ia_uid = local->mds_stbuf.ia_uid;
+                        local->stbuf.ia_prot = local->mds_stbuf.ia_prot;
+                    }
+                } else {
+                    local->stbuf.ia_gid = local->prebuf.ia_gid;
+                    local->stbuf.ia_uid = local->prebuf.ia_uid;
+                    local->stbuf.ia_prot = local->prebuf.ia_prot;
                 }
 
-                STACK_WIND (frame, dht_lookup_cbk,
-                            hashed_subvol, hashed_subvol->fops->lookup,
-                            loc, local->xattr_req);
+                layout = local->layout;
+                dht_selfheal_directory(frame, dht_lookup_selfheal_cbk,
+                                       &local->loc, layout);
+                return 0;
+            }
         }
 
-        return 0;
+        if (local->layout_mismatch) {
+            /* Found layout mismatch in the directory, need to
+               fix this in the inode context */
+            dht_layout_unref(this, local->layout);
+            local->layout = NULL;
+            dht_lookup_directory(frame, this, &local->loc);
+            return 0;
+        }
+
+        if (local->need_lookup_everywhere) {
+            /* As the current layout gave ENOENT error, we would
+               need a new layout */
+            dht_layout_unref(this, local->layout);
+            local->layout = NULL;
+
+            /* We know that current cached subvol is no longer
+               valid, get the new one */
+            local->cached_subvol = NULL;
+            if (local->xattr_req) {
+                if (!gf_uuid_is_null(local->gfid)) {
+                    ret = dict_set_static_bin(local->xattr_req, "gfid-req",
+                                              local->gfid, 16);
+                }
+            }
+
+            dht_lookup_everywhere(frame, this, &local->loc);
+            return 0;
+        }
+        if (local->return_estale) {
+            local->op_ret = -1;
+            local->op_errno = ESTALE;
+        }
+
+        if (local->loc.parent) {
+            dht_inode_ctx_time_update(local->loc.parent, this,
+                                      &local->postparent, 1);
+        }
+
+        DHT_STRIP_PHASE1_FLAGS(&local->stbuf);
+        dht_set_fixed_dir_stat(&local->postparent);
+
+        /* local->stbuf is updated only from subvols which have a layout
+         * The reason is to avoid choosing attr heal source from newly
+         * added bricks. In case e.g we have only one subvol and for
+         * some reason layout is not present on it, then local->stbuf
+         * will be EINVAL. This is an indication that the subvols
+         * active in the cluster do not have layouts on disk.
+         * Unwind with ESTALE to trigger a fresh lookup */
+        if (is_dir && local->stbuf.ia_type == IA_INVAL) {
+            local->op_ret = -1;
+            local->op_errno = ESTALE;
+        }
+        /* Delete mds xattr at the time of STACK UNWIND */
+        if (local->xattr)
+            GF_REMOVE_INTERNAL_XATTR(conf->mds_xattr_key, local->xattr);
+
+        DHT_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno,
+                         local->inode, &local->stbuf, local->xattr,
+                         &local->postparent);
+    }
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
-        return 0;
+    return ret;
 }
 
+static int
+dht_lookup_linkfile_create_cbk(call_frame_t *frame, void *cooie, xlator_t *this,
+                               int32_t op_ret, int32_t op_errno, inode_t *inode,
+                               struct iatt *stbuf, struct iatt *preparent,
+                               struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *cached_subvol = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", this->private, out);
+
+    local = frame->local;
+    cached_subvol = local->cached_subvol;
+    conf = this->private;
+
+    gf_uuid_unparse(local->loc.gfid, gfid);
+
+    if (local->locked)
+        dht_unlock_namespace(frame, &local->lock[0]);
+
+    ret = dht_layout_preset(this, local->cached_subvol, local->loc.inode);
+    if (ret < 0) {
+        gf_msg_debug(this->name, EINVAL,
+                     "Failed to set layout for subvolume %s, "
+                     "(gfid = %s)",
+                     cached_subvol ? cached_subvol->name : "<nil>", gfid);
+        local->op_ret = -1;
+        local->op_errno = EINVAL;
+        goto unwind;
+    }
+
+    local->op_ret = 0;
+    if ((local->stbuf.ia_nlink == 1) && (conf && conf->unhashed_sticky_bit)) {
+        local->stbuf.ia_prot.sticky = 1;
+    }
+
+    if (local->loc.parent) {
+        dht_inode_ctx_time_update(local->loc.parent, this, postparent, 1);
+    }
 
-int
-dht_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-	      int op_ret, int op_errno, struct iatt *prebuf,
-              struct iatt *postbuf)
-{
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = 0;
-	call_frame_t *prev = NULL;
-
-
-	local = frame->local;
-	prev = cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-                        local->op_ret = -1;
-			gf_log (this->name, GF_LOG_DEBUG,
-				"subvolume %s returned -1 (%s)",
-				prev->this->name, strerror (op_errno));
-			goto unlock;
-		}
-
-                dht_iatt_merge (this, &local->prebuf, prebuf, prev->this);
-		dht_iatt_merge (this, &local->stbuf, postbuf, prev->this);
-
-		if (local->inode) {
-			local->stbuf.ia_ino = local->inode->ino;
-                        local->prebuf.ia_ino = local->inode->ino;
-                }
+unwind:
+    gf_msg_debug(this->name, 0,
+                 "creation of linkto on hashed subvol:%s, "
+                 "returned with op_ret %d and op_errno %d: %s",
+                 local->hashed_subvol->name, op_ret, op_errno,
+                 uuid_utoa(local->loc.gfid));
 
-		local->op_ret = 0;
-	}
-unlock:
-	UNLOCK (&frame->lock);
+    if (local->linked == _gf_true)
+        dht_linkfile_attr_heal(frame, this);
 
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt))
-		DHT_STACK_UNWIND (truncate, frame, local->op_ret, local->op_errno,
-				  &local->prebuf, &local->stbuf);
+    dht_set_fixed_dir_stat(&local->postparent);
 
-        return 0;
+    DHT_STRIP_PHASE1_FLAGS(&local->stbuf);
+    DHT_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno,
+                     local->inode, &local->stbuf, local->xattr,
+                     &local->postparent);
+out:
+    return ret;
 }
 
+static int
+dht_lookup_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, struct iatt *preparent,
+                      struct iatt *postparent, dict_t *xdata)
+{
+    int this_call_cnt = 0;
+    dht_local_t *local = NULL;
+    const char *path = NULL;
 
+    local = (dht_local_t *)frame->local;
+    path = local->loc.path;
+    FRAME_SU_UNDO(frame, dht_local_t);
 
-int
-dht_attr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-	      int op_ret, int op_errno, struct iatt *stbuf)
-{
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = 0;
-	call_frame_t *prev = NULL;
-
-
-	local = frame->local;
-	prev = cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			gf_log (this->name, GF_LOG_DEBUG,
-				"subvolume %s returned -1 (%s)",
-				prev->this->name, strerror (op_errno));
-			goto unlock;
-		}
-
-		dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
-		
-		if (local->inode)
-			local->stbuf.ia_ino = local->inode->ino;
-		local->op_ret = 0;
-	}
-unlock:
-	UNLOCK (&frame->lock);
+    gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_UNLINK_LOOKUP_INFO,
+           "lookup_unlink returned with "
+           "op_ret -> %d and op-errno -> %d for %s",
+           op_ret, op_errno, ((path == NULL) ? "null" : path));
 
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt))
-		DHT_STACK_UNWIND (stat, frame, local->op_ret, local->op_errno,
-				  &local->stbuf);
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        dht_lookup_everywhere_done(frame, this);
+    }
 
-        return 0;
+    return 0;
 }
 
-
-int
-dht_stat (call_frame_t *frame, xlator_t *this,
-	  loc_t *loc)
+static int
+dht_lookup_unlink_of_false_linkto_cbk(call_frame_t *frame, void *cookie,
+                                      xlator_t *this, int op_ret, int op_errno,
+                                      struct iatt *preparent,
+                                      struct iatt *postparent, dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
-	dht_layout_t *layout = NULL;
-	int           i = 0;
+    int this_call_cnt = 0;
+    dht_local_t *local = NULL;
+    const char *path = NULL;
 
+    local = (dht_local_t *)frame->local;
+    path = local->loc.path;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
+    FRAME_SU_UNDO(frame, dht_local_t);
 
+    gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_UNLINK_LOOKUP_INFO,
+           "lookup_unlink returned with "
+           "op_ret -> %d and op-errno -> %d for %s",
+           op_ret, op_errno, ((path == NULL) ? "null" : path));
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        if ((op_ret == 0) || ((op_errno != EBUSY) && (op_errno != ENOTCONN))) {
+            dht_lookup_everywhere_done(frame, this);
+        } else {
+            /*When dht_lookup_everywhere is performed, one cached
+             *and one hashed file was found and hashed file does
+             *not point to the above mentioned cached node. So it
+             *was considered as stale and an unlink was performed.
+             *But unlink fails. So may be rebalance is in progress.
+             *now ideally we have two data-files. One obtained during
+             *lookup_everywhere and one where unlink-failed. So
+             *at this point in time we cannot decide which one to
+             *choose because there are chances of first cached
+             *file is truncated after rebalance and if it is chosen
+             *as cached node, application will fail. So return EIO.*/
+
+            if (op_errno == EBUSY) {
+                gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                       DHT_MSG_UNLINK_FAILED,
+                       "Could not unlink the linkto file as "
+                       "either fd is open and/or linkto xattr "
+                       "is set for %s",
+                       ((path == NULL) ? "null" : path));
+            }
+            DHT_STACK_UNWIND(lookup, frame, -1, EIO, NULL, NULL, NULL, NULL);
+        }
+    }
+
+    return 0;
+}
 
-	local->layout = layout = dht_layout_get (this, loc->inode);
-	if (!layout) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no layout for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+static int
+dht_lookup_unlink_stale_linkto_cbk(call_frame_t *frame, void *cookie,
+                                   xlator_t *this, int op_ret, int op_errno,
+                                   struct iatt *preparent,
+                                   struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    const char *path = NULL;
 
-	local->inode = inode_ref (loc->inode);
-	local->call_cnt = layout->cnt;
+    /* NOTE:
+     * If stale file unlink fails either there is an open-fd or is not an
+     * dht-linkto-file then posix_unlink returns EBUSY, which is overwritten
+     *  to ENOENT
+     */
 
-	for (i = 0; i < layout->cnt; i++) {
-		subvol = layout->list[i].xlator;
+    local = frame->local;
 
-		STACK_WIND (frame, dht_attr_cbk,
-			    subvol, subvol->fops->stat,
-			    loc);
-	}
+    if (local) {
+        FRAME_SU_UNDO(frame, dht_local_t);
+        if (local->loc.path)
+            path = local->loc.path;
+    }
 
-	return 0;
+    gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_UNLINK_LOOKUP_INFO,
+           "Returned with op_ret %d and "
+           "op_errno %d for %s",
+           op_ret, op_errno, ((path == NULL) ? "null" : path));
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (stat, frame, -1, op_errno, NULL);
+    DHT_STACK_UNWIND(lookup, frame, -1, ENOENT, NULL, NULL, NULL, NULL);
 
-	return 0;
+    return 0;
 }
 
+static int
+dht_fill_dict_to_avoid_unlink_of_migrating_file(dict_t *dict)
+{
+    int ret = 0;
 
-int
-dht_fstat (call_frame_t *frame, xlator_t *this,
-	   fd_t *fd)
-{
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
-	dht_layout_t *layout = NULL;
-	int           i = 0;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	local->layout = layout = dht_layout_get (this, fd->inode);
-	if (!layout) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no layout for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	local->inode    = inode_ref (fd->inode);
-	local->call_cnt = layout->cnt;;
-
-	for (i = 0; i < layout->cnt; i++) {
-		subvol = layout->list[i].xlator;
-		STACK_WIND (frame, dht_attr_cbk,
-			    subvol, subvol->fops->fstat,
-			    fd);
-	}
-
-	return 0;
+    ret = dict_set_int32_sizen(dict, DHT_SKIP_NON_LINKTO_UNLINK, 1);
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (fstat, frame, -1, op_errno, NULL);
+    if (ret)
+        return -1;
 
-	return 0;
+    ret = dict_set_int32_sizen(dict, DHT_SKIP_OPEN_FD_UNLINK, 1);
+
+    if (ret)
+        return -1;
+
+    return 0;
 }
 
+static int32_t
+dht_linkfile_create_lookup_cbk(call_frame_t *frame, void *cookie,
+                               xlator_t *this, int32_t op_ret, int32_t op_errno,
+                               inode_t *inode, struct iatt *buf, dict_t *xdata,
+                               struct iatt *postparent)
+{
+    dht_local_t *local = NULL;
+    int call_cnt = 0, ret = 0;
+    xlator_t *subvol = NULL;
+    uuid_t gfid = {
+        0,
+    };
+    char gfid_str[GF_UUID_BUF_SIZE] = {0};
+
+    subvol = cookie;
+    local = frame->local;
+
+    if (subvol == local->hashed_subvol) {
+        if ((op_ret == 0) || (op_errno != ENOENT))
+            local->dont_create_linkto = _gf_true;
+    } else {
+        if (gf_uuid_is_null(local->gfid))
+            gf_uuid_copy(gfid, local->loc.gfid);
+        else
+            gf_uuid_copy(gfid, local->gfid);
+
+        if ((op_ret == 0) && gf_uuid_compare(gfid, buf->ia_gfid)) {
+            gf_uuid_unparse(gfid, gfid_str);
+            gf_msg_debug(this->name, 0,
+                         "gfid (%s) different on cached subvol "
+                         "(%s) and looked up inode (%s), not "
+                         "creating linkto",
+                         uuid_utoa(buf->ia_gfid), subvol->name, gfid_str);
+            local->dont_create_linkto = _gf_true;
+        } else if (op_ret == -1) {
+            local->dont_create_linkto = _gf_true;
+        }
+    }
+
+    call_cnt = dht_frame_return(frame);
+    if (is_last_call(call_cnt)) {
+        if (local->dont_create_linkto)
+            goto no_linkto;
+        else {
+            gf_msg_debug(this->name, 0,
+                         "Creating linkto file on %s(hash) to "
+                         "%s on %s (gfid = %s)",
+                         local->hashed_subvol->name, local->loc.path,
+                         local->cached_subvol->name, gfid_str);
+
+            ret = dht_linkfile_create(frame, dht_lookup_linkfile_create_cbk,
+                                      this, local->cached_subvol,
+                                      local->hashed_subvol, &local->loc);
+
+            if (ret < 0)
+                goto no_linkto;
+        }
+    }
+
+    return 0;
 
-int
-dht_truncate (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, off_t offset)
+no_linkto:
+    gf_msg_debug(this->name, 0,
+                 "skipped linkto creation (path:%s) (gfid:%s) "
+                 "(hashed-subvol:%s) (cached-subvol:%s)",
+                 local->loc.path, gfid_str, local->hashed_subvol->name,
+                 local->cached_subvol->name);
+
+    dht_lookup_linkfile_create_cbk(frame, NULL, this, 0, 0, local->loc.inode,
+                                   &local->stbuf, &local->preparent,
+                                   &local->postparent, local->xattr);
+    return 0;
+}
+
+static int32_t
+dht_call_lookup_linkfile_create(call_frame_t *frame, void *cookie,
+                                xlator_t *this, int32_t op_ret,
+                                int32_t op_errno, dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
+    dht_local_t *local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    int i = 0;
+    xlator_t *subvol = NULL;
 
+    local = frame->local;
+    if (gf_uuid_is_null(local->gfid))
+        gf_uuid_unparse(local->loc.gfid, gfid);
+    else
+        gf_uuid_unparse(local->gfid, gfid);
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
+    if (op_ret < 0) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "protecting namespace failed, skipping linkto "
+               "creation (path:%s)(gfid:%s)(hashed-subvol:%s)"
+               "(cached-subvol:%s)",
+               local->loc.path, gfid, local->hashed_subvol->name,
+               local->cached_subvol->name);
+        goto err;
+    }
 
-	subvol = dht_subvol_get_cached (this, loc->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+    local->locked = _gf_true;
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
+    local->call_cnt = 2;
 
-	local->inode = inode_ref (loc->inode);
-	local->call_cnt = 1;
+    for (i = 0; i < 2; i++) {
+        subvol = (subvol == NULL) ? local->hashed_subvol : local->cached_subvol;
 
-	STACK_WIND (frame, dht_truncate_cbk,
-		    subvol, subvol->fops->truncate,
-		    loc, offset);
+        STACK_WIND_COOKIE(frame, dht_linkfile_create_lookup_cbk, subvol, subvol,
+                          subvol->fops->lookup, &local->loc, NULL);
+    }
 
-	return 0;
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (truncate, frame, -1, op_errno, NULL, NULL);
-
-	return 0;
+    dht_lookup_linkfile_create_cbk(frame, NULL, this, 0, 0, local->loc.inode,
+                                   &local->stbuf, &local->preparent,
+                                   &local->postparent, local->xattr);
+    return 0;
 }
 
+/* Rebalance is performed from cached_node to hashed_node. Initial cached_node
+ * contains a non-linkto file. After migration it is converted to linkto and
+ * then unlinked. And at hashed_subvolume, first a linkto file is present,
+ * then after migration it is converted to a non-linkto file.
+ *
+ * Lets assume a file is present on cached subvolume and a new brick is added
+ * and new brick is the new_hashed subvolume. So fresh lookup on newly added
+ * hashed subvolume will fail and dht_lookup_everywhere gets called.  If just
+ * before sending the dht_lookup_everywhere request rebalance is in progress,
+ *
+ * from cached subvolume it may see: Nonlinkto or linkto or No file
+ * from hashed subvolume it may see: No file or linkto file or non-linkto file
+ *
+ * So this boils down to 9 cases:
+ *   at cached_subvol            at hashed_subvol
+ *   ----------------           -----------------
+ *
+ *a)   No file                     No file
+ *    [request reached after    [Request reached before
+ *       migration]                Migration]
+ *
+ *b)   No file                     Linkto File
+ *
+ *c)   No file                     Non-Linkto File
+ *
+ *d)   Linkto                      No-File
+ *
+ *e)   Linkto                      Linkto
+ *
+ *f)   Linkto                      Non-Linkto
+ *
+ *g)   NonLinkto                   No-File
+ *
+ *h)   NonLinkto                   Linkto
+ *
+ *i)   NonLinkto                   NonLinkto
+ *
+ * dht_lookup_everywhere_done takes decision based on any of the above case
+ */
 
-int
-dht_ftruncate (call_frame_t *frame, xlator_t *this,
-	       fd_t *fd, off_t offset)
+static int
+dht_lookup_everywhere_done(call_frame_t *frame, xlator_t *this)
 {
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
+    int ret = 0;
+    dht_local_t *local = NULL;
+    xlator_t *hashed_subvol = NULL;
+    xlator_t *cached_subvol = NULL;
+    dht_layout_t *layout = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    gf_boolean_t found_non_linkto_on_hashed = _gf_false;
+
+    local = frame->local;
+    hashed_subvol = local->hashed_subvol;
+    cached_subvol = local->cached_subvol;
+
+    gf_uuid_unparse(local->loc.gfid, gfid);
+
+    if (local->file_count && local->dir_count) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_FILE_TYPE_MISMATCH,
+               "path %s (gfid = %s)exists as a file on one "
+               "subvolume and directory on another. "
+               "Please fix it manually",
+               local->loc.path, gfid);
+        DHT_STACK_UNWIND(lookup, frame, -1, EIO, NULL, NULL, NULL, NULL);
+        return 0;
+    }
+    if (local->op_ret && local->gfid_missing) {
+        if (gf_uuid_is_null(local->gfid_req)) {
+            DHT_STACK_UNWIND(lookup, frame, -1, ENODATA, NULL, NULL, NULL,
+                             NULL);
+            return 0;
+        }
+        /* A hack */
+        dht_lookup_directory(frame, this, &local->loc);
+        return 0;
+    }
 
+    if (local->dir_count) {
+        dht_lookup_directory(frame, this, &local->loc);
+        return 0;
+    }
+
+    gf_msg_debug(this->name, 0,
+                 "STATUS: hashed_subvol %s "
+                 "cached_subvol %s",
+                 (hashed_subvol == NULL) ? "null" : hashed_subvol->name,
+                 (cached_subvol == NULL) ? "null" : cached_subvol->name);
+
+    if (!cached_subvol) {
+        if (local->skip_unlink.handle_valid_link && hashed_subvol) {
+            /*Purpose of "DHT_SKIP_NON_LINKTO_UNLINK":
+             * If this lookup is performed by rebalance and this
+             * rebalance process detected hashed file and by
+             * the time it sends the lookup request to cached node,
+             * file got migrated and now at initial hashed_node,
+             * final migrated file is present. With current logic,
+             * because this process fails to find the cached_node,
+             * it will unlink the file at initial hashed_node.
+             *
+             * So we avoid this by setting key, and checking at the
+             * posix_unlink that unlink the file only if file is a
+             * linkto file and not a migrated_file.
+             */
+
+            ret = dht_fill_dict_to_avoid_unlink_of_migrating_file(
+                local->xattr_req);
+
+            if (ret) {
+                /* If for some reason, setting key in the dict
+                 * fails, return with ENOENT, as with respect to
+                 * this process, it detected only a stale link
+                 * file.
+                 *
+                 * Next lookup will delete it.
+                 *
+                 * Performing deletion of stale link file when
+                 * setting key in dict fails, may cause the data
+                 * loss because of the above mentioned race.
+                 */
+
+                DHT_STACK_UNWIND(lookup, frame, -1, ENOENT, NULL, NULL, NULL,
+                                 NULL);
+            } else {
+                local->skip_unlink.handle_valid_link = _gf_false;
+
+                gf_msg_debug(this->name, 0,
+                             "No Cached was found and "
+                             "unlink on hashed was skipped"
+                             " so performing now: %s",
+                             local->loc.path);
+                FRAME_SU_DO(frame, dht_local_t);
+                STACK_WIND(frame, dht_lookup_unlink_stale_linkto_cbk,
+                           hashed_subvol, hashed_subvol->fops->unlink,
+                           &local->loc, 0, local->xattr_req);
+            }
+
+        } else {
+            gf_msg_debug(this->name, 0,
+                         "There was no cached file and  "
+                         "unlink on hashed is not skipped %s",
+                         local->loc.path);
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+            DHT_STACK_UNWIND(lookup, frame, -1, ENOENT, NULL, NULL, NULL, NULL);
+        }
+        return 0;
+    }
+
+    /* At the time of dht_lookup, no file was found on hashed and that is
+     * why dht_lookup_everywhere is called, but by the time
+     * dht_lookup_everywhere
+     * reached to server, file might have already migrated. In that case we
+     * will find a migrated file at the hashed_node. In this case store the
+     * layout in context and return successfully.
+     */
+
+    if (hashed_subvol || local->need_lookup_everywhere) {
+        if (local->need_lookup_everywhere) {
+            found_non_linkto_on_hashed = _gf_true;
+
+        } else if ((local->file_count == 1) &&
+                   (hashed_subvol == cached_subvol)) {
+            gf_msg_debug(this->name, 0,
+                         "found cached file on hashed subvolume "
+                         "so store in context and return for %s",
+                         local->loc.path);
+
+            found_non_linkto_on_hashed = _gf_true;
+        }
 
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
+        if (found_non_linkto_on_hashed)
+            goto preset_layout;
+    }
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
+    if (hashed_subvol) {
+        if (local->skip_unlink.handle_valid_link == _gf_true) {
+            if (cached_subvol == local->skip_unlink.hash_links_to) {
+                if (gf_uuid_compare(local->skip_unlink.cached_gfid,
+                                    local->skip_unlink.hashed_gfid)) {
+                    /*GFID different, return error*/
+                    DHT_STACK_UNWIND(lookup, frame, -1, ESTALE, NULL, NULL,
+                                     NULL, NULL);
 
-	local->inode = inode_ref (fd->inode);
-	local->call_cnt = 1;
+                    return 0;
+                }
 
-	STACK_WIND (frame, dht_truncate_cbk,
-		    subvol, subvol->fops->ftruncate,
-		    fd, offset);
+                ret = dht_layout_preset(this, cached_subvol, local->loc.inode);
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_INFO, 0,
+                           DHT_MSG_LAYOUT_PRESET_FAILED,
+                           "Could not set pre-set layout "
+                           "for subvolume %s",
+                           cached_subvol->name);
+                }
 
-	return 0;
+                local->op_ret = (ret == 0) ? ret : -1;
+                local->op_errno = (ret == 0) ? ret : EINVAL;
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (ftruncate, frame, -1, op_errno, NULL, NULL);
+                /* Presence of local->cached_subvol validates
+                 * that lookup from cached node is successful
+                 */
 
-	return 0;
-}
+                if (!local->op_ret && local->loc.parent) {
+                    dht_inode_ctx_time_update(local->loc.parent, this,
+                                              &local->postparent, 1);
+                }
 
+                gf_msg_debug(this->name, 0,
+                             "Skipped unlinking linkto file "
+                             "on the hashed subvolume. "
+                             "Returning success as it is a "
+                             "valid linkto file. Path:%s",
+                             local->loc.path);
+
+                goto unwind_hashed_and_cached;
+            } else {
+                local->skip_unlink.handle_valid_link = _gf_false;
+
+                gf_msg_debug(this->name, 0,
+                             "Linkto file found on hashed "
+                             "subvol "
+                             "and data file found on cached "
+                             "subvolume. But linkto points to "
+                             "different cached subvolume (%s) "
+                             "path %s",
+                             (local->skip_unlink.hash_links_to
+                                  ? local->skip_unlink.hash_links_to->name
+                                  : " <nil>"),
+                             local->loc.path);
+
+                if (local->skip_unlink.opend_fd_count == 0) {
+                    ret = dht_fill_dict_to_avoid_unlink_of_migrating_file(
+                        local->xattr_req);
+
+                    if (ret) {
+                        DHT_STACK_UNWIND(lookup, frame, -1, EIO, NULL, NULL,
+                                         NULL, NULL);
+                    } else {
+                        local->call_cnt = 1;
+                        FRAME_SU_DO(frame, dht_local_t);
+                        STACK_WIND(frame, dht_lookup_unlink_of_false_linkto_cbk,
+                                   hashed_subvol, hashed_subvol->fops->unlink,
+                                   &local->loc, 0, local->xattr_req);
+                    }
+
+                    return 0;
+                }
+            }
+        }
+    }
 
-int
-dht_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                int op_ret, int op_errno, struct iatt *preparent,
-                struct iatt *postparent)
-{
-	dht_local_t  *local = NULL;
-	call_frame_t *prev  = NULL;
-
-	local = frame->local;
-	prev  = cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-                        local->op_ret   = -1;
-			local->op_errno = op_errno;
-			gf_log (this->name, GF_LOG_DEBUG,
-				"subvolume %s returned -1 (%s)",
-				prev->this->name, strerror (op_errno));
-			goto unlock;
-		}
-
-                preparent->ia_ino = local->loc.parent->ino;
-                postparent->ia_ino = local->loc.parent->ino;
-		local->op_ret = 0;
-
-                local->postparent = *postparent;
-                local->preparent = *preparent;
-
-                WIPE (&local->postparent);
-                WIPE (&local->preparent);
-	}
-unlock:
-	UNLOCK (&frame->lock);
+preset_layout:
+
+    if (found_non_linkto_on_hashed) {
+        if (local->need_lookup_everywhere) {
+            if (gf_uuid_compare(local->gfid, local->inode->gfid)) {
+                /* GFID different, return error */
+                DHT_STACK_UNWIND(lookup, frame, -1, ENOENT, NULL, NULL, NULL,
+                                 NULL);
+                return 0;
+            }
+        }
+
+        local->op_ret = 0;
+        local->op_errno = 0;
+        layout = dht_layout_for_subvol(this, cached_subvol);
+        if (!layout) {
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_INFO,
+                   "%s: no pre-set layout for subvolume %s,"
+                   " gfid = %s",
+                   local->loc.path,
+                   (cached_subvol ? cached_subvol->name : "<nil>"), gfid);
+        }
+
+        ret = dht_layout_set(this, local->inode, layout);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_INFO,
+                   "%s: failed to set layout for subvol %s, "
+                   "gfid = %s",
+                   local->loc.path,
+                   (cached_subvol ? cached_subvol->name : "<nil>"), gfid);
+        }
+
+        if (local->loc.parent) {
+            dht_inode_ctx_time_update(local->loc.parent, this,
+                                      &local->postparent, 1);
+        }
+
+        DHT_STRIP_PHASE1_FLAGS(&local->stbuf);
+        dht_set_fixed_dir_stat(&local->postparent);
+        DHT_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno,
+                         local->inode, &local->stbuf, local->xattr,
+                         &local->postparent);
+        return 0;
+    }
+
+    if (!hashed_subvol) {
+        gf_msg_debug(this->name, 0,
+                     "Cannot create linkfile for %s on %s: "
+                     "hashed subvolume cannot be found, gfid = %s.",
+                     local->loc.path, cached_subvol->name, gfid);
+
+        local->op_ret = 0;
+        local->op_errno = 0;
+
+        ret = dht_layout_preset(frame->this, cached_subvol, local->inode);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_LAYOUT_PRESET_FAILED,
+                   "Failed to set layout for subvol %s"
+                   ", gfid = %s",
+                   cached_subvol ? cached_subvol->name : "<nil>", gfid);
+            local->op_ret = -1;
+            local->op_errno = EINVAL;
+        }
 
-        DHT_STACK_UNWIND (unlink, frame, local->op_ret, local->op_errno,
-                          &local->preparent, &local->postparent);
+        if (local->loc.parent) {
+            dht_inode_ctx_time_update(local->loc.parent, this,
+                                      &local->postparent, 1);
+        }
 
+        DHT_STRIP_PHASE1_FLAGS(&local->stbuf);
+        dht_set_fixed_dir_stat(&local->postparent);
+        DHT_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno,
+                         local->inode, &local->stbuf, local->xattr,
+                         &local->postparent);
         return 0;
+    }
+
+    if (frame->root->op != GF_FOP_RENAME) {
+        local->current = &local->lock[0];
+        ret = dht_protect_namespace(frame, &local->loc, hashed_subvol,
+                                    &local->current->ns,
+                                    dht_call_lookup_linkfile_create);
+    } else {
+        gf_msg_debug(this->name, 0,
+                     "Creating linkto file on %s(hash) to %s on %s "
+                     "(gfid = %s)",
+                     hashed_subvol->name, local->loc.path, cached_subvol->name,
+                     gfid);
+
+        ret = dht_linkfile_create(frame, dht_lookup_linkfile_create_cbk, this,
+                                  cached_subvol, hashed_subvol, &local->loc);
+    }
+
+    return ret;
+
+unwind_hashed_and_cached:
+    DHT_STRIP_PHASE1_FLAGS(&local->stbuf);
+    dht_set_fixed_dir_stat(&local->postparent);
+    DHT_STACK_UNWIND(lookup, frame, local->op_ret, local->op_errno,
+                     local->inode, &local->stbuf, local->xattr,
+                     &local->postparent);
+    return 0;
 }
 
+static int
+dht_lookup_everywhere_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, inode_t *inode,
+                          struct iatt *buf, dict_t *xattr,
+                          struct iatt *postparent)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+    int is_linkfile = 0;
+    int is_dir = 0;
+    loc_t *loc = NULL;
+    xlator_t *link_subvol = NULL;
+    int ret = -1;
+    int32_t fd_count = 0;
+    dht_conf_t *conf = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    dict_t *dict_req = {0};
+
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", cookie, out);
+    GF_VALIDATE_OR_GOTO("dht", this->private, out);
+
+    local = frame->local;
+    loc = &local->loc;
+    conf = this->private;
+
+    prev = cookie;
+
+    gf_msg_debug(this->name, 0,
+                 "returned with op_ret %d and op_errno %d (%s) "
+                 "from subvol %s",
+                 op_ret, op_errno, loc->path, prev->name);
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            if (op_errno != ENOENT)
+                local->op_errno = op_errno;
+            if (op_errno == ENODATA)
+                local->gfid_missing = _gf_true;
+            goto unlock;
+        }
+
+        if (gf_uuid_is_null(local->gfid))
+            gf_uuid_copy(local->gfid, buf->ia_gfid);
 
-int
-dht_unlink_linkfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int op_ret, int op_errno, struct iatt *preparent,
-                         struct iatt *postparent)
-{
-	dht_local_t  *local = NULL;
-	call_frame_t *prev = NULL;
-
-        xlator_t *cached_subvol = NULL;
-
-	local = frame->local;
-	prev  = cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			gf_log (this->name, GF_LOG_DEBUG,
-				"subvolume %s returned -1 (%s)",
-				prev->this->name, strerror (op_errno));
-			goto unlock;
-		}
-
-		local->op_ret = 0;
-	}
+        gf_uuid_unparse(local->gfid, gfid);
+
+        if (gf_uuid_compare(local->gfid, buf->ia_gfid)) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_GFID_MISMATCH,
+                   "%s: gfid differs on subvolume %s,"
+                   " gfid local = %s, gfid node = %s",
+                   loc->path, prev->name, gfid, uuid_utoa(buf->ia_gfid));
+        }
+
+        is_linkfile = check_is_linkfile(inode, buf, xattr,
+                                        conf->link_xattr_name);
+
+        if (is_linkfile) {
+            link_subvol = dht_linkfile_subvol(this, inode, buf, xattr);
+            gf_msg_debug(this->name, 0, "found on %s linkfile %s (-> %s)",
+                         prev->name, loc->path,
+                         link_subvol ? link_subvol->name : "''");
+            goto unlock;
+        }
+
+        is_dir = check_is_dir(inode, buf, xattr);
+
+        /* non linkfile GFID takes precedence but don't overwrite
+         gfid if we have already found a cached file*/
+        if (!local->cached_subvol)
+            gf_uuid_copy(local->gfid, buf->ia_gfid);
+
+        if (is_dir) {
+            local->dir_count++;
+
+            gf_msg_debug(this->name, 0, "found on %s directory %s", prev->name,
+                         loc->path);
+        } else {
+            local->file_count++;
+
+            gf_msg_debug(this->name, 0, "found cached file on %s for %s",
+                         prev->name, loc->path);
+
+            if (!local->cached_subvol) {
+                /* found one file */
+                dht_iatt_merge(this, &local->stbuf, buf);
+
+                local->xattr = dict_ref(xattr);
+                local->cached_subvol = prev;
+
+                gf_msg_debug(this->name, 0,
+                             "storing cached on %s file"
+                             " %s",
+                             prev->name, loc->path);
+
+                dht_iatt_merge(this, &local->postparent, postparent);
+
+                gf_uuid_copy(local->skip_unlink.cached_gfid, buf->ia_gfid);
+            } else {
+                /* This is where we need 'rename' both entries logic */
+                gf_msg(this->name, GF_LOG_WARNING, 0,
+                       DHT_MSG_FILE_ON_MULT_SUBVOL,
+                       "multiple subvolumes (%s and %s) have "
+                       "file %s (preferably rename the file "
+                       "in the backend,and do a fresh lookup)",
+                       local->cached_subvol->name, prev->name, local->loc.path);
+            }
+        }
+    }
 unlock:
-	UNLOCK (&frame->lock);
+    UNLOCK(&frame->lock);
+
+    if (is_linkfile) {
+        ret = dict_get_int32(xattr, GLUSTERFS_OPEN_FD_COUNT, &fd_count);
+
+        /*  Any linkto file found on the non-hashed subvolume should
+         *  be unlinked (performed in the "else if" block below)
+         *
+         *  But if a linkto file is found on hashed subvolume, it may be
+         *  pointing to valid cached node. So unlinking of linkto
+         *  file on hashed subvolume is skipped and inside
+         *  dht_lookup_everywhere_done, checks are performed. If this
+         *  linkto file is found as stale linkto file, it is deleted
+         *  otherwise unlink is skipped.
+         */
 
-        if (op_ret == -1)
-                goto err;
+        if (local->hashed_subvol && local->hashed_subvol == prev) {
+            local->skip_unlink.handle_valid_link = _gf_true;
+            local->skip_unlink.opend_fd_count = fd_count;
+            local->skip_unlink.hash_links_to = link_subvol;
+            gf_uuid_copy(local->skip_unlink.hashed_gfid, buf->ia_gfid);
+
+            gf_msg_debug(this->name, 0,
+                         "Found"
+                         " one linkto file on hashed subvol %s "
+                         "for %s: Skipping unlinking till "
+                         "everywhere_done",
+                         prev->name, loc->path);
+
+        } else if (!ret && (fd_count == 0)) {
+            dict_req = dict_new();
+
+            ret = dht_fill_dict_to_avoid_unlink_of_migrating_file(dict_req);
+
+            if (ret) {
+                /* Skip unlinking for dict_failure
+                 *File is found as a linkto file on non-hashed,
+                 *subvolume. In the current implementation,
+                 *finding a linkto-file on non-hashed does not
+                 *always implies that it is stale. So deletion
+                 *of file should be done only when both fd is
+                 *closed and linkto-xattr is set. In case of
+                 *dict_set failure, avoid skipping of file.
+                 *NOTE: dht_frame_return should get called for
+                 *      this block.
+                 */
+
+                dict_unref(dict_req);
+
+            } else {
+                gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_INFO,
+                       "attempting deletion of stale linkfile "
+                       "%s on %s (hashed subvol is %s)",
+                       loc->path, prev->name,
+                       (local->hashed_subvol ? local->hashed_subvol->name
+                                             : "<null>"));
+                /* *
+                 * These stale files may be created using root
+                 * user. Hence deletion will work only with
+                 * root.
+                 */
+                FRAME_SU_DO(frame, dht_local_t);
+                STACK_WIND(frame, dht_lookup_unlink_cbk, prev,
+                           prev->fops->unlink, loc, 0, dict_req);
+
+                dict_unref(dict_req);
 
-        cached_subvol = dht_subvol_get_cached (this, local->loc.inode);
-        if (!cached_subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no cached subvolume for path=%s",
-                        local->loc.path);
-                local->op_errno = EINVAL;
-                goto err;
+                return 0;
+            }
         }
+    }
 
-        STACK_WIND (frame, dht_unlink_cbk,
-                    cached_subvol, cached_subvol->fops->unlink,
-                    &local->loc);
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        dht_lookup_everywhere_done(frame, this);
+    }
 
-        return 0;
+out:
+    return ret;
+}
+
+int
+dht_lookup_everywhere(call_frame_t *frame, xlator_t *this, loc_t *loc)
+{
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    int i = 0;
+    int call_cnt = 0;
 
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", this->private, out);
+    GF_VALIDATE_OR_GOTO("dht", loc, out);
+
+    conf = this->private;
+    local = frame->local;
+
+    call_cnt = conf->subvolume_cnt;
+    local->call_cnt = call_cnt;
+
+    if (!local->inode)
+        local->inode = inode_ref(loc->inode);
+
+    gf_msg_debug(this->name, 0, "winding lookup call to %d subvols", call_cnt);
+
+    for (i = 0; i < call_cnt; i++) {
+        STACK_WIND_COOKIE(frame, dht_lookup_everywhere_cbk, conf->subvolumes[i],
+                          conf->subvolumes[i],
+                          conf->subvolumes[i]->fops->lookup, loc,
+                          local->xattr_req);
+    }
+
+    return 0;
+out:
+    DHT_STACK_UNWIND(lookup, frame, -1, EINVAL, NULL, NULL, NULL, NULL);
 err:
-        DHT_STACK_UNWIND (unlink, frame, -1, local->op_errno,
-                          NULL, NULL);
-        return 0;
+    return -1;
 }
 
-
 int
-dht_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
-               int op_errno, struct iatt *prebuf, struct iatt *postbuf)
+dht_lookup_linkfile_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, inode_t *inode,
+                        struct iatt *stbuf, dict_t *xattr,
+                        struct iatt *postparent)
 {
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = 0;
-	call_frame_t *prev = NULL;
+    xlator_t *prev = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *subvol = NULL;
+    loc_t *loc = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", this, unwind);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, unwind);
+    GF_VALIDATE_OR_GOTO("dht", this->private, unwind);
+    GF_VALIDATE_OR_GOTO("dht", cookie, unwind);
+
+    prev = cookie;
+    subvol = prev;
+    conf = this->private;
+    local = frame->local;
+    loc = &local->loc;
+
+    gf_uuid_unparse(loc->gfid, gfid);
+
+    if (op_ret == -1) {
+        gf_msg(this->name, GF_LOG_INFO, op_errno, DHT_MSG_LINK_FILE_LOOKUP_INFO,
+               "Lookup of %s on %s (following linkfile) failed "
+               ",gfid = %s",
+               local->loc.path, subvol->name, gfid);
+
+        /* If cached subvol returned ENOTCONN, do not do
+        lookup_everywhere. We need to make sure linkfile does not get
+        removed, which can take away the namespace, and subvol is
+        anyways down. */
+
+        local->cached_subvol = NULL;
+        if (op_errno != ENOTCONN)
+            goto err;
+        else
+            goto unwind;
+    }
+
+    if (check_is_dir(inode, stbuf, xattr)) {
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_LINK_FILE_LOOKUP_INFO,
+               "Lookup of %s on %s (following linkfile) reached dir,"
+               " gfid = %s",
+               local->loc.path, subvol->name, gfid);
+        goto err;
+    }
+
+    if (check_is_linkfile(inode, stbuf, xattr, conf->link_xattr_name)) {
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_LINK_FILE_LOOKUP_INFO,
+               "lookup of %s on %s (following linkfile) reached link,"
+               "gfid = %s",
+               local->loc.path, subvol->name, gfid);
+        goto err;
+    }
+
+    if (gf_uuid_compare(local->gfid, stbuf->ia_gfid)) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_GFID_MISMATCH,
+               "%s: gfid different on data file on %s,"
+               " gfid local = %s, gfid node = %s ",
+               local->loc.path, subvol->name, gfid, uuid_utoa(stbuf->ia_gfid));
+        goto err;
+    }
+
+    if ((stbuf->ia_nlink == 1) && (conf && conf->unhashed_sticky_bit)) {
+        stbuf->ia_prot.sticky = 1;
+    }
+
+    ret = dht_layout_preset(this, prev, inode);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_LAYOUT_PRESET_FAILED,
+               "Failed to set layout for subvolume %s,"
+               "gfid = %s",
+               prev->name, gfid);
+        op_ret = -1;
+        op_errno = EINVAL;
+    }
+
+    if (local->loc.parent) {
+        dht_inode_ctx_time_update(local->loc.parent, this, postparent, 1);
+    }
+
+unwind:
+    DHT_STRIP_PHASE1_FLAGS(stbuf);
+    dht_set_fixed_dir_stat(postparent);
+    DHT_STACK_UNWIND(lookup, frame, op_ret, op_errno, inode, stbuf, xattr,
+                     postparent);
+
+    return 0;
+
+err:
+    dht_lookup_everywhere(frame, this, loc);
+out:
+    return 0;
+}
 
+/* Code to get hashed subvol based on inode and loc
+   First it check if loc->parent and loc->path exist then it get
+   hashed subvol based on loc.
+*/
 
-	local = frame->local;
-	prev = cookie;
+static gf_boolean_t
+dht_should_lookup_everywhere(xlator_t *this, dht_conf_t *conf, loc_t *loc)
+{
+    dht_layout_t *parent_layout = NULL;
+    int ret = 0;
+    gf_boolean_t lookup_everywhere = _gf_true;
+
+    /* lookup-optimize supersedes lookup-unhashed settings.
+     * If it is set, do not process search_unhashed
+     * If lookup-optimize if enabled, lookup everywhere if:
+     *  - this is the rebalance daemon.
+     *  - loc->parent is unavailable.
+     *  - parent_layout is unavailable
+     *  - parent_layout->commit_hash != conf->vol_commit_hash
+     */
+
+    if (conf->lookup_optimize) {
+        if (!conf->defrag && loc->parent) {
+            ret = dht_inode_ctx_layout_get(loc->parent, this, &parent_layout);
+            if (!ret && parent_layout &&
+                (parent_layout->commit_hash == conf->vol_commit_hash)) {
+                lookup_everywhere = _gf_false;
+            }
+        }
+        goto out;
+    } else {
+        if (conf->search_unhashed == GF_DHT_LOOKUP_UNHASHED_AUTO) {
+            if (loc->parent) {
+                ret = dht_inode_ctx_layout_get(loc->parent, this,
+                                               &parent_layout);
+                if (ret || !parent_layout ||
+                    (!parent_layout->search_unhashed)) {
+                    lookup_everywhere = _gf_false;
+                }
+            } else {
+                lookup_everywhere = _gf_false;
+            }
 
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			gf_log (this->name, GF_LOG_DEBUG,
-				"subvolume %s returned -1 (%s)",
-				prev->this->name, strerror (op_errno));
-			goto unlock;
-		}
+            goto out;
+        }
+    }
+out:
+    return lookup_everywhere;
+}
 
-		local->op_ret = 0;
-	}
-unlock:
-	UNLOCK (&frame->lock);
+int
+dht_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, inode_t *inode, struct iatt *stbuf, dict_t *xattr,
+               struct iatt *postparent)
+{
+    char is_linkfile = 0;
+    char is_dir = 0;
+    xlator_t *subvol = NULL;
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    loc_t *loc = NULL;
+    xlator_t *prev = NULL;
+    int ret = 0;
+    uint32_t vol_commit_hash = 0;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", cookie, out);
+    GF_VALIDATE_OR_GOTO("dht", this->private, out);
+
+    conf = this->private;
+
+    prev = cookie;
+    local = frame->local;
+    loc = &local->loc;
+
+    gf_msg_debug(this->name, op_errno,
+                 "%s: fresh_lookup on %s returned with op_ret %d", loc->path,
+                 prev->name, op_ret);
+
+    if (op_ret == -1) {
+        if (ENTRY_MISSING(op_ret, op_errno)) {
+            if (1 == conf->subvolume_cnt) {
+                /* No need to lookup again */
+                goto out;
+            }
+
+            gf_msg_debug(this->name, 0, "Entry %s missing on subvol %s",
+                         loc->path, prev->name);
+
+            if (dht_should_lookup_everywhere(this, conf, loc)) {
+                local->op_errno = ENOENT;
+                dht_lookup_everywhere(frame, this, loc);
+                return 0;
+            }
 
-        if (local && (op_ret == 0)) {
-                prebuf->ia_ino = local->ia_ino;
-                postbuf->ia_ino = local->ia_ino;
+        } else {
+            /* posix returns ENODATA if the gfid is not set but the client and
+             * server protocol layers do not send the stbuf. We need to
+             * heal this so check if this is a directory on the other subvols.
+             */
+            if ((op_errno == ENOTCONN) || (op_errno == ENODATA)) {
+                dht_lookup_directory(frame, this, &local->loc);
+                return 0;
+            }
+        }
+        gf_msg_debug(this->name, op_errno, "%s: Lookup on subvolume %s failed",
+                     loc->path, prev->name);
+        goto out;
+    }
+
+    /* Lookup succeeded - op_ret = 0 */
+
+    /* This is required for handling stale linkfile deletion,
+     * or any more call which happens from this 'loc'.
+     */
+    if (gf_uuid_is_null(local->gfid)) {
+        /*This is set from the first successful response*/
+        memcpy(local->gfid, stbuf->ia_gfid, 16);
+    }
+
+    if (!conf->vch_forced) {
+        /* Update the commit hash in conf if it is found */
+        ret = dict_get_uint32(xattr, conf->commithash_xattr_name,
+                              &vol_commit_hash);
+        if (ret == 0) {
+            conf->vol_commit_hash = vol_commit_hash;
         }
+    }
+
+    is_dir = check_is_dir(inode, stbuf, xattr);
+    if (is_dir) {
+        /* A directory is present on all subvols, send the lookup to
+         * all subvols now */
+        local->inode = inode_ref(inode);
+        local->xattr = dict_ref(xattr);
+        dht_lookup_directory(frame, this, &local->loc);
+        return 0;
+    }
+
+    is_linkfile = check_is_linkfile(inode, stbuf, xattr, conf->link_xattr_name);
 
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt))
-		DHT_STACK_UNWIND (fsync, frame, local->op_ret, local->op_errno,
-                                  prebuf, postbuf);
+    if (!is_linkfile) {
+        /* non-directory and not a linkto file. This is a data file
+         * Update the layout to point to the cached subvol
+         */
 
+        ret = dht_layout_preset(this, prev, inode);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_LAYOUT_PRESET_FAILED,
+                   "%s: could not set pre-set layout for subvolume %s",
+                   loc->path, prev->name);
+            op_ret = -1;
+            op_errno = EINVAL;
+            goto out;
+        }
+        goto out;
+    }
+
+    /* This is a linkto file. Get the value of the target subvol from the
+     * linkto xattr and lookup there to see if the file exists
+     */
+    subvol = dht_linkfile_subvol(this, inode, stbuf, xattr);
+    if (!subvol) {
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_INFO,
+               "%s: No link subvol for linkto", loc->path);
+        dht_lookup_everywhere(frame, this, loc);
         return 0;
+    }
+
+    gf_msg_debug(this->name, 0, "%s: Calling lookup on linkto target %s",
+                 loc->path, subvol->name);
+
+    STACK_WIND_COOKIE(frame, dht_lookup_linkfile_cbk, subvol, subvol,
+                      subvol->fops->lookup, &local->loc, local->xattr_req);
+
+    return 0;
+
+out:
+    /*
+     * FIXME: postparent->ia_size and postparent->st_blocks do not have
+     * correct values. since, postparent corresponds to a directory these
+     * two members should have values equal to sum of corresponding values
+     * from each of the subvolume. See dht_iatt_merge for reference.
+     */
+
+    if (!op_ret && local && local->loc.parent) {
+        dht_inode_ctx_time_update(local->loc.parent, this, postparent, 1);
+    }
+
+    DHT_STRIP_PHASE1_FLAGS(stbuf);
+    dht_set_fixed_dir_stat(postparent);
+    DHT_STACK_UNWIND(lookup, frame, op_ret, op_errno, inode, stbuf, xattr,
+                     postparent);
+err:
+    return 0;
 }
 
+/* For directories, check if acl xattrs have been requested (by the acl
+ * xlator), if not, request for them. These xattrs are needed for dht dir
+ * self-heal to perform proper self-healing of dirs
+ */
+static void
+dht_check_and_set_acl_xattr_req(xlator_t *this, dict_t *xattr_req)
+{
+    int ret = 0;
 
+    GF_ASSERT(xattr_req);
 
-int
-dht_err_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-	     int op_ret, int op_errno)
+    if (!dict_get(xattr_req, POSIX_ACL_ACCESS_XATTR)) {
+        ret = dict_set_int8(xattr_req, POSIX_ACL_ACCESS_XATTR, 0);
+        if (ret)
+            gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_DICT_SET_FAILED,
+                   "Failed to set dictionary value:key = %s",
+                   POSIX_ACL_ACCESS_XATTR);
+    }
+
+    if (!dict_get(xattr_req, POSIX_ACL_DEFAULT_XATTR)) {
+        ret = dict_set_int8(xattr_req, POSIX_ACL_DEFAULT_XATTR, 0);
+        if (ret)
+            gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_DICT_SET_FAILED,
+                   "Failed to set dictionary value:key = %s",
+                   POSIX_ACL_DEFAULT_XATTR);
+    }
+
+    return;
+}
+
+/* for directories, we need the following info:
+ * the layout : trusted.glusterfs.dht
+ * the mds information : trusted.glusterfs.dht.mds
+ * the acl info: See above
+ */
+static int
+dht_set_dir_xattr_req(xlator_t *this, loc_t *loc, dict_t *xattr_req)
 {
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = 0;
-	call_frame_t *prev = NULL;
+    int ret = -EINVAL;
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+    if (!conf) {
+        goto err;
+    }
+
+    if (!xattr_req) {
+        goto err;
+    }
+
+    /* Xattr to get the layout for a directory
+     */
+    ret = dict_set_uint32(xattr_req, conf->xattr_name, 4 * 4);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, DHT_MSG_DICT_SET_FAILED,
+               "Failed to set dictionary value:key = %s for "
+               "path %s",
+               conf->xattr_name, loc->path);
+        goto err;
+    }
+
+    /*Non-fatal failure */
+    ret = dict_set_uint32(xattr_req, conf->mds_xattr_key, 4);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, DHT_MSG_DICT_SET_FAILED,
+               "Failed to set dictionary value:key = %s for "
+               "path %s",
+               conf->mds_xattr_key, loc->path);
+    }
+
+    dht_check_and_set_acl_xattr_req(this, xattr_req);
+    ret = 0;
+err:
+    return ret;
+}
+
+/* If the hashed subvol is present, send the lookup to only that subvol first.
+ * If no hashed subvol, send a lookup to all subvols and proceed based on the
+ * responses.
+ */
+static int
+dht_do_fresh_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc)
+{
+    int ret = -1;
+    dht_conf_t *conf = NULL;
+    xlator_t *hashed_subvol = NULL;
+    dht_local_t *local = NULL;
+    int op_errno = -1;
+    int call_cnt = 0;
+    int i = 0;
+
+    conf = this->private;
+    if (!conf) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    local = frame->local;
+    if (!local) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    /* Since we don't know whether this is a file or a directory,
+     * request all xattrs*/
+    ret = dht_set_file_xattr_req(this, loc, local->xattr_req);
+    if (ret) {
+        op_errno = -ret;
+        goto err;
+    }
+
+    ret = dht_set_dir_xattr_req(this, loc, local->xattr_req);
+    if (ret) {
+        op_errno = -ret;
+        goto err;
+    }
+
+    /* Fuse sets a random value in gfid-req. If the gfid is missing
+     * on one or more subvols, posix will set the gfid to this value,
+     * causing GFID mismatches for directories. Remove the value fuse
+     * has sent before sending the lookup.
+     */
+    ret = dict_get_gfuuid(local->xattr_req, "gfid-req", &local->gfid_req);
+    if (ret) {
+        gf_msg_debug(this->name, 0, "%s: No gfid-req available", loc->path);
+    } else {
+        dict_del(local->xattr_req, "gfid-req");
+    }
+    /* This should have been set in dht_lookup */
+    hashed_subvol = local->hashed_subvol;
+
+    if (!hashed_subvol) {
+        gf_msg_debug(this->name, 0,
+                     "%s: no subvolume in layout for path, "
+                     "checking on all the subvols to see if "
+                     "it is a directory",
+                     loc->path);
+
+        call_cnt = conf->subvolume_cnt;
+        local->call_cnt = call_cnt;
+
+        /* Allocate a layout. This will be populated and saved in
+         * the dht inode_ctx on successful lookup
+         */
+        local->layout = dht_layout_new(this, conf->subvolume_cnt);
+        if (!local->layout) {
+            op_errno = ENOMEM;
+            goto err;
+        }
 
+        gf_msg_debug(this->name, 0,
+                     "%s: Found null hashed subvol. Calling lookup"
+                     " on all nodes.",
+                     loc->path);
 
-	local = frame->local;
-	prev = cookie;
+        for (i = 0; i < call_cnt; i++) {
+            STACK_WIND_COOKIE(frame, dht_lookup_dir_cbk, conf->subvolumes[i],
+                              conf->subvolumes[i],
+                              conf->subvolumes[i]->fops->lookup, &local->loc,
+                              local->xattr_req);
+        }
+        return 0;
+    }
 
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			gf_log (this->name, GF_LOG_DEBUG,
-				"subvolume %s returned -1 (%s)",
-				prev->this->name, strerror (op_errno));
-			goto unlock;
-		}
+    /* if the hashed_subvol is non-null, send the lookup there first so
+     * as to see whether we have a file or a directory */
+    gf_msg_debug(this->name, 0, "%s: Calling fresh lookup on %s", loc->path,
+                 hashed_subvol->name);
 
-		local->op_ret = 0;
-	}
-unlock:
-	UNLOCK (&frame->lock);
+    STACK_WIND_COOKIE(frame, dht_lookup_cbk, hashed_subvol, hashed_subvol,
+                      hashed_subvol->fops->lookup, loc, local->xattr_req);
+    return 0;
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+    return 0;
+}
 
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt)) {
-		DHT_STACK_UNWIND (setxattr, frame, local->op_ret, local->op_errno);
+static int
+dht_do_revalidate(call_frame_t *frame, xlator_t *this, loc_t *loc)
+{
+    xlator_t *subvol = NULL;
+    xlator_t *mds_subvol = NULL;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+    int op_errno = -1;
+    dht_layout_t *layout = NULL;
+    int i = 0;
+    int call_cnt = 0;
+    int gen = 0;
+
+    conf = this->private;
+    if (!conf) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    local = frame->local;
+    if (!local) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!layout) {
+        gf_msg_debug(this->name, 0,
+                     "path = %s. No layout found in the inode ctx.", loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    /* Generation number has changed. This layout may be stale. */
+    if (layout->gen && (layout->gen < conf->gen)) {
+        gen = layout->gen;
+        dht_layout_unref(this, local->layout);
+        local->layout = NULL;
+        local->cached_subvol = NULL;
+
+        gf_msg_debug(this->name, 0,
+                     "path = %s. In memory layout may be stale."
+                     "(layout->gen (%d) is less than "
+                     "conf->gen (%d)). Calling fresh lookup.",
+                     loc->path, gen, conf->gen);
+
+        dht_do_fresh_lookup(frame, this, loc);
+        return 0;
+    }
+
+    local->inode = inode_ref(loc->inode);
+
+    /* Since we don't know whether this has changed,
+     * request all xattrs*/
+    ret = dht_set_file_xattr_req(this, loc, local->xattr_req);
+    if (ret) {
+        op_errno = -ret;
+        goto err;
+    }
+
+    ret = dht_set_dir_xattr_req(this, loc, local->xattr_req);
+    if (ret) {
+        op_errno = -ret;
+        goto err;
+    }
+
+    if (IA_ISDIR(local->inode->ia_type)) {
+        ret = dht_inode_ctx_mdsvol_get(local->inode, this, &mds_subvol);
+        if (ret || !mds_subvol) {
+            gf_msg_debug(this->name, 0, "path = %s. No mds subvol in inode ctx",
+                         local->loc.path);
         }
+        local->mds_subvol = mds_subvol;
+        local->call_cnt = conf->subvolume_cnt;
+
+        /* local->call_cnt will change as responses are processed. Always use a
+         * local copy to loop through the STACK_WIND calls
+         */
 
+        call_cnt = local->call_cnt;
+
+        for (i = 0; i < call_cnt; i++) {
+            STACK_WIND_COOKIE(frame, dht_revalidate_cbk, conf->subvolumes[i],
+                              conf->subvolumes[i],
+                              conf->subvolumes[i]->fops->lookup, loc,
+                              local->xattr_req);
+        }
         return 0;
+    }
+
+    /* If not a dir, this should be 1 */
+    local->call_cnt = layout->cnt;
+    call_cnt = local->call_cnt;
+
+    for (i = 0; i < call_cnt; i++) {
+        subvol = layout->list[i].xlator;
+
+        gf_msg_debug(this->name, 0,
+                     "path = %s. Calling "
+                     "revalidate lookup on %s",
+                     loc->path, subvol->name);
+
+        STACK_WIND_COOKIE(frame, dht_revalidate_cbk, subvol, subvol,
+                          subvol->fops->lookup, &local->loc, local->xattr_req);
+    }
+    return 0;
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+    return 0;
 }
 
+/* Depending on the input, decide if this is a:
+ * fresh-lookup: loc->name is provided but no dht inode ctx
+ * revalidation: loc->name is provided, dht inode ctx is present
+ * discover: gfid based nameless lookup.
+ */
 
 int
-dht_access (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc, int32_t mask)
+dht_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
+{
+    xlator_t *hashed_subvol = NULL;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+    int op_errno = -1;
+    loc_t new_loc = {
+        0,
+    };
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+
+    conf = this->private;
+    if (!conf)
+        goto err;
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_LOOKUP);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    ret = dht_filter_loc_subvol_key(this, loc, &new_loc, &hashed_subvol);
+    if (ret) {
+        loc_wipe(&local->loc);
+        ret = loc_dup(&new_loc, &local->loc);
+
+        /* we no longer need 'new_loc' entries */
+        loc_wipe(&new_loc);
+
+        /* check if loc_dup() is successful */
+        if (ret == -1) {
+            op_errno = errno;
+            gf_msg_debug(this->name, errno,
+                         "copying location failed for path=%s", loc->path);
+            goto err;
+        }
+    }
+
+    if (xattr_req) {
+        local->xattr_req = dict_ref(xattr_req);
+    } else {
+        local->xattr_req = dict_new();
+    }
+
+    /* Nameless lookup */
+
+    /* This is usually sent by NFS. Lookups are done based on the gfid and
+     * no name information is available. Without the name, dht cannot calculate
+     * the hash and has to send a lookup to all subvols.
+     */
+    if (gf_uuid_is_null(loc->pargfid) && !gf_uuid_is_null(loc->gfid) &&
+        !__is_root_gfid(loc->inode->gfid)) {
+        local->cached_subvol = NULL;
+        dht_do_discover(frame, this, loc);
+        return 0;
+    }
+
+    if (loc_is_root(loc)) {
+        /* Request the DHT commit hash xattr (trusted.glusterfs.dht.commithash)
+         * set on the brick root.
+         */
+        ret = dict_set_uint32(local->xattr_req, conf->commithash_xattr_name,
+                              sizeof(uint32_t));
+    }
+
+    if (!hashed_subvol)
+        hashed_subvol = dht_subvol_get_hashed(this, loc);
+    local->hashed_subvol = hashed_subvol;
+
+    if (is_revalidate(loc)) {
+        /* The entry has been looked up before and has a dht inode_ctx
+         */
+        dht_do_revalidate(frame, this, loc);
+        return 0;
+    } else {
+        /* Entry has not been looked up before
+         */
+        dht_do_fresh_lookup(frame, this, loc);
+        return 0;
+    }
+
+    return 0;
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+    return 0;
+}
+
+static int
+dht_unlink_linkfile_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, struct iatt *preparent,
+                        struct iatt *postparent, dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+
+    local = frame->local;
+    prev = cookie;
+
+    LOCK(&frame->lock);
+    {
+        if ((op_ret == -1) &&
+            !((op_errno == ENOENT) || (op_errno == ENOTCONN))) {
+            local->op_errno = op_errno;
+            UNLOCK(&frame->lock);
+            gf_msg_debug(this->name, op_errno,
+                         "Unlink link: subvolume %s returned -1", prev->name);
+            goto post_unlock;
+        }
 
+        local->op_ret = 0;
+    }
+    UNLOCK(&frame->lock);
+post_unlock:
+    dht_set_fixed_dir_stat(&local->preparent);
+    dht_set_fixed_dir_stat(&local->postparent);
+    DHT_STACK_UNWIND(unlink, frame, local->op_ret, local->op_errno,
+                     &local->preparent, &local->postparent, xdata);
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
+    return 0;
+}
 
-	subvol = dht_subvol_get_cached (this, loc->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+static int
+dht_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, struct iatt *preparent, struct iatt *postparent,
+               dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *hashed_subvol = NULL;
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
+    local = frame->local;
+    prev = cookie;
 
-	local->call_cnt = 1;
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            if (op_errno != ENOENT) {
+                local->op_ret = -1;
+                local->op_errno = op_errno;
+            } else {
+                local->op_ret = 0;
+            }
+            UNLOCK(&frame->lock);
+            gf_msg_debug(this->name, op_errno,
+                         "Unlink: subvolume %s returned -1", prev->name);
+            goto post_unlock;
+        }
 
-	STACK_WIND (frame, dht_err_cbk,
-		    subvol, subvol->fops->access,
-		    loc, mask);
+        local->op_ret = 0;
 
-	return 0;
+        local->postparent = *postparent;
+        local->preparent = *preparent;
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (access, frame, -1, op_errno);
+        if (local->loc.parent) {
+            dht_inode_ctx_time_update(local->loc.parent, this,
+                                      &local->preparent, 0);
+            dht_inode_ctx_time_update(local->loc.parent, this,
+                                      &local->postparent, 1);
+        }
+    }
+    UNLOCK(&frame->lock);
+post_unlock:
+    if (!local->op_ret) {
+        hashed_subvol = dht_subvol_get_hashed(this, &local->loc);
+        if (hashed_subvol && hashed_subvol != local->cached_subvol) {
+            /*
+             * If hashed and cached are different, then we need
+             * to unlink linkfile from hashed subvol if data
+             * file is deleted successfully
+             */
+            STACK_WIND_COOKIE(frame, dht_unlink_linkfile_cbk, hashed_subvol,
+                              hashed_subvol, hashed_subvol->fops->unlink,
+                              &local->loc, local->flags, xdata);
+            return 0;
+        }
+    }
 
-	return 0;
+    dht_set_fixed_dir_stat(&local->preparent);
+    dht_set_fixed_dir_stat(&local->postparent);
+    DHT_STACK_UNWIND(unlink, frame, local->op_ret, local->op_errno,
+                     &local->preparent, &local->postparent, xdata);
+
+    return 0;
 }
 
+static int
+dht_common_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    DHT_STACK_UNWIND(setxattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
-int
-dht_readlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		  int op_ret, int op_errno, const char *path, struct iatt *sbuf)
+static int
+dht_fix_layout_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        dht_local_t *local = NULL;
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
 
+    if (op_ret == 0) {
+        /* update the layout in the inode ctx */
         local = frame->local;
-        if (op_ret == -1)
-                goto err;
+        layout = local->selfheal.layout;
 
-        if (local) {
-                sbuf->ia_ino = local->ia_ino;
-        } else {
-                op_ret = -1;
-                op_errno = EINVAL;
+        dht_layout_set(this, local->loc.inode, layout);
+    }
+
+    DHT_STACK_UNWIND(setxattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+static int
+dht_err_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+            int op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+
+    local = frame->local;
+    prev = cookie;
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+            UNLOCK(&frame->lock);
+            gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                         prev->name);
+            goto post_unlock;
         }
 
-err:
-        DHT_STACK_UNWIND (readlink, frame, op_ret, op_errno, path, sbuf);
+        local->op_ret = 0;
+    }
+    UNLOCK(&frame->lock);
+post_unlock:
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        if ((local->fop == GF_FOP_SETXATTR) ||
+            (local->fop == GF_FOP_FSETXATTR)) {
+            DHT_STACK_UNWIND(setxattr, frame, local->op_ret, local->op_errno,
+                             NULL);
+            /* 'local' itself may not be valid after this */
+            goto out;
+        }
+        if ((local->fop == GF_FOP_REMOVEXATTR) ||
+            (local->fop == GF_FOP_FREMOVEXATTR)) {
+            DHT_STACK_UNWIND(removexattr, frame, local->op_ret, local->op_errno,
+                             NULL);
+        }
+    }
 
-        return 0;
+out:
+    return 0;
+}
+
+/* Set the value[] of key into dict after convert from
+   host byte order to network byte order
+*/
+int32_t
+dht_dict_set_array(dict_t *dict, char *key, int32_t value[], int32_t size)
+{
+    int ret = -1;
+    int32_t *ptr = NULL;
+    int32_t vindex;
+
+    if (value == NULL) {
+        return -EINVAL;
+    }
+
+    ptr = GF_MALLOC(sizeof(int32_t) * size, gf_common_mt_char);
+    if (ptr == NULL) {
+        return -ENOMEM;
+    }
+    for (vindex = 0; vindex < size; vindex++) {
+        ptr[vindex] = hton32(value[vindex]);
+    }
+    ret = dict_set_bin(dict, key, ptr, sizeof(int32_t) * size);
+    if (ret)
+        GF_FREE(ptr);
+    return ret;
 }
 
+static int
+dht_common_mds_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno, dict_t *dict,
+                           dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    call_frame_t *prev = cookie;
 
-int
-dht_readlink (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, size_t size)
-{
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = dht_subvol_get_cached (this, loc->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-        local->ia_ino = loc->inode->ino;
- 
-	STACK_WIND (frame, dht_readlink_cbk,
-		    subvol, subvol->fops->readlink,
-		    loc, size);
-
-	return 0;
+    local = frame->local;
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (readlink, frame, -1, op_errno, NULL, NULL);
+    if (op_ret)
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->this->name);
+
+    if (local->fop == GF_FOP_SETXATTR) {
+        DHT_STACK_UNWIND(setxattr, frame, 0, op_errno, local->xdata);
+        /* 'local' itself may not be valid after this */
+        goto out;
+    }
+
+    if (local->fop == GF_FOP_FSETXATTR) {
+        DHT_STACK_UNWIND(fsetxattr, frame, 0, op_errno, local->xdata);
+        /* 'local' itself may not be valid after this */
+        goto out;
+    }
 
-	return 0;
+    if (local->fop == GF_FOP_REMOVEXATTR) {
+        DHT_STACK_UNWIND(removexattr, frame, 0, op_errno, NULL);
+        /* 'local' itself may not be valid after this */
+        goto out;
+    }
+
+    if (local->fop == GF_FOP_FREMOVEXATTR) {
+        DHT_STACK_UNWIND(fremovexattr, frame, 0, op_errno, NULL);
+    }
+
+out:
+    return 0;
 }
 
+/* Code to wind a xattrop call to add 1 on current mds internal xattr
+   value
+*/
+static int
+dht_setxattr_non_mds_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int op_ret, int op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    int ret = 0;
+    dict_t *xattrop = NULL;
+    int32_t addone[1] = {1};
+    call_frame_t *prev = NULL;
+    dht_conf_t *conf = NULL;
+
+    local = frame->local;
+    prev = cookie;
+    conf = this->private;
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret && !local->op_ret) {
+            local->op_ret = op_ret;
+            local->op_errno = op_errno;
+            UNLOCK(&frame->lock);
+            gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                         prev->this->name);
+            goto post_unlock;
+        }
+    }
+    UNLOCK(&frame->lock);
+post_unlock:
+    this_call_cnt = dht_frame_return(frame);
+
+    if (is_last_call(this_call_cnt)) {
+        if (!local->op_ret) {
+            xattrop = dict_new();
+            if (!xattrop) {
+                gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_NO_MEMORY, 0,
+                       "dictionary creation failed");
+                ret = -1;
+                goto out;
+            }
+            ret = dht_dict_set_array(xattrop, conf->mds_xattr_key, addone, 1);
+            if (ret != 0) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                       "dictionary set array failed ");
+                ret = -1;
+                goto out;
+            }
+            if ((local->fop == GF_FOP_SETXATTR) ||
+                (local->fop == GF_FOP_REMOVEXATTR)) {
+                STACK_WIND(frame, dht_common_mds_xattrop_cbk, local->mds_subvol,
+                           local->mds_subvol->fops->xattrop, &local->loc,
+                           GF_XATTROP_ADD_ARRAY, xattrop, NULL);
+            } else {
+                STACK_WIND(frame, dht_common_mds_xattrop_cbk, local->mds_subvol,
+                           local->mds_subvol->fops->fxattrop, local->fd,
+                           GF_XATTROP_ADD_ARRAY, xattrop, NULL);
+            }
+        } else {
+            if (local->fop == GF_FOP_SETXATTR) {
+                DHT_STACK_UNWIND(setxattr, frame, 0, 0, local->xdata);
+                /* 'local' itself may not be valid after this */
+                goto just_return;
+            }
+
+            if (local->fop == GF_FOP_FSETXATTR) {
+                DHT_STACK_UNWIND(fsetxattr, frame, 0, 0, local->xdata);
+                /* 'local' itself may not be valid after this */
+                goto just_return;
+            }
+
+            if (local->fop == GF_FOP_REMOVEXATTR) {
+                DHT_STACK_UNWIND(removexattr, frame, 0, 0, NULL);
+                /* 'local' itself may not be valid after this */
+                goto just_return;
+            }
+
+            if (local->fop == GF_FOP_FREMOVEXATTR) {
+                DHT_STACK_UNWIND(fremovexattr, frame, 0, 0, NULL);
+                /* 'local' itself may not be valid after this */
+                goto just_return;
+            }
+        }
+    }
+out:
+    if (ret) {
+        if (local->fop == GF_FOP_SETXATTR) {
+            DHT_STACK_UNWIND(setxattr, frame, 0, 0, local->xdata);
+            /* 'local' itself may not be valid after this */
+            goto just_return;
+        }
+
+        if (local->fop == GF_FOP_FSETXATTR) {
+            DHT_STACK_UNWIND(fsetxattr, frame, 0, 0, local->xdata);
+            /* 'local' itself may not be valid after this */
+            goto just_return;
+        }
 
-int
-dht_fix_layout_cbk (call_frame_t *frame, void *cookie,
-                    xlator_t *this, int32_t op_ret, int32_t op_errno)
+        if (local->fop == GF_FOP_REMOVEXATTR) {
+            DHT_STACK_UNWIND(removexattr, frame, 0, 0, NULL);
+            /* 'local' itself may not be valid after this */
+            goto just_return;
+        }
+
+        if (local->fop == GF_FOP_FREMOVEXATTR) {
+            DHT_STACK_UNWIND(fremovexattr, frame, 0, 0, NULL);
+        }
+    }
+just_return:
+    if (xattrop)
+        dict_unref(xattrop);
+    return 0;
+}
+
+static int
+dht_setxattr_mds_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, dict_t *xdata)
 {
-        DHT_STACK_UNWIND (getxattr, frame, -1, ENODATA, NULL);
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    call_frame_t *prev = NULL;
+    xlator_t *mds_subvol = NULL;
+    int i = 0;
+
+    local = frame->local;
+    prev = cookie;
+    conf = this->private;
+    mds_subvol = local->mds_subvol;
+
+    if (op_ret == -1) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->this->name);
+        goto out;
+    }
+
+    local->op_ret = 0;
+    local->call_cnt = conf->subvolume_cnt - 1;
+    local->xdata = dict_ref(xdata);
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (mds_subvol && (mds_subvol == conf->subvolumes[i]))
+            continue;
+        if (local->fop == GF_FOP_SETXATTR) {
+            STACK_WIND(frame, dht_setxattr_non_mds_cbk, conf->subvolumes[i],
+                       conf->subvolumes[i]->fops->setxattr, &local->loc,
+                       local->xattr, local->flags, local->xattr_req);
+        }
 
-        return 0;
+        if (local->fop == GF_FOP_FSETXATTR) {
+            STACK_WIND(frame, dht_setxattr_non_mds_cbk, conf->subvolumes[i],
+                       conf->subvolumes[i]->fops->fsetxattr, local->fd,
+                       local->xattr, local->flags, local->xattr_req);
+        }
+
+        if (local->fop == GF_FOP_REMOVEXATTR) {
+            STACK_WIND(frame, dht_setxattr_non_mds_cbk, conf->subvolumes[i],
+                       conf->subvolumes[i]->fops->removexattr, &local->loc,
+                       local->key, local->xattr_req);
+        }
+
+        if (local->fop == GF_FOP_FREMOVEXATTR) {
+            STACK_WIND(frame, dht_setxattr_non_mds_cbk, conf->subvolumes[i],
+                       conf->subvolumes[i]->fops->fremovexattr, local->fd,
+                       local->key, local->xattr_req);
+        }
+    }
+
+    return 0;
+out:
+    if (local->fop == GF_FOP_SETXATTR) {
+        DHT_STACK_UNWIND(setxattr, frame, local->op_ret, local->op_errno,
+                         xdata);
+        /* 'local' itself may not be valid after this */
+        goto just_return;
+    }
+
+    if (local->fop == GF_FOP_FSETXATTR) {
+        DHT_STACK_UNWIND(fsetxattr, frame, local->op_ret, local->op_errno,
+                         xdata);
+        /* 'local' itself may not be valid after this */
+        goto just_return;
+    }
+
+    if (local->fop == GF_FOP_REMOVEXATTR) {
+        DHT_STACK_UNWIND(removexattr, frame, local->op_ret, local->op_errno,
+                         NULL);
+        /* 'local' itself may not be valid after this */
+        goto just_return;
+    }
+
+    if (local->fop == GF_FOP_FREMOVEXATTR) {
+        DHT_STACK_UNWIND(fremovexattr, frame, local->op_ret, local->op_errno,
+                         NULL);
+    }
+
+just_return:
+    return 0;
+}
+
+static int
+dht_xattrop_mds_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int op_ret, int op_errno, dict_t *dict, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    call_frame_t *prev = NULL;
+
+    local = frame->local;
+    prev = cookie;
+
+    if (op_ret == -1) {
+        local->op_errno = op_errno;
+        local->op_ret = op_ret;
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->this->name);
+        goto out;
+    }
+
+    if (local->fop == GF_FOP_SETXATTR) {
+        STACK_WIND(frame, dht_setxattr_mds_cbk, local->mds_subvol,
+                   local->mds_subvol->fops->setxattr, &local->loc, local->xattr,
+                   local->flags, local->xattr_req);
+    }
+
+    if (local->fop == GF_FOP_FSETXATTR) {
+        STACK_WIND(frame, dht_setxattr_mds_cbk, local->mds_subvol,
+                   local->mds_subvol->fops->fsetxattr, local->fd, local->xattr,
+                   local->flags, local->xattr_req);
+    }
+
+    if (local->fop == GF_FOP_REMOVEXATTR) {
+        STACK_WIND(frame, dht_setxattr_mds_cbk, local->mds_subvol,
+                   local->mds_subvol->fops->removexattr, &local->loc,
+                   local->key, local->xattr_req);
+    }
+
+    if (local->fop == GF_FOP_FREMOVEXATTR) {
+        STACK_WIND(frame, dht_setxattr_mds_cbk, local->mds_subvol,
+                   local->mds_subvol->fops->fremovexattr, local->fd, local->key,
+                   local->xattr_req);
+    }
+
+    return 0;
+out:
+    if (local->fop == GF_FOP_SETXATTR) {
+        DHT_STACK_UNWIND(setxattr, frame, local->op_ret, local->op_errno,
+                         xdata);
+        /* 'local' itself may not be valid after this */
+        goto just_return;
+    }
+
+    if (local->fop == GF_FOP_FSETXATTR) {
+        DHT_STACK_UNWIND(fsetxattr, frame, local->op_ret, local->op_errno,
+                         xdata);
+        /* 'local' itself may not be valid after this */
+        goto just_return;
+    }
+
+    if (local->fop == GF_FOP_REMOVEXATTR) {
+        DHT_STACK_UNWIND(removexattr, frame, local->op_ret, local->op_errno,
+                         NULL);
+        /* 'local' itself may not be valid after this */
+        goto just_return;
+    }
+
+    if (local->fop == GF_FOP_FREMOVEXATTR) {
+        DHT_STACK_UNWIND(fremovexattr, frame, local->op_ret, local->op_errno,
+                         NULL);
+    }
+
+just_return:
+    return 0;
 }
 
 static void
-fill_layout_info (dht_layout_t *layout, char *buf)
+fill_layout_info(dht_layout_t *layout, char *buf)
 {
-        int i = 0;
-        char tmp_buf[128] = {0,};
+    int i = 0;
+    char tmp_buf[128] = {
+        0,
+    };
+
+    for (i = 0; i < layout->cnt; i++) {
+        snprintf(tmp_buf, sizeof(tmp_buf), "(%s %u %u)",
+                 layout->list[i].xlator->name, layout->list[i].start,
+                 layout->list[i].stop);
+        if (i)
+            strcat(buf, " ");
+        strcat(buf, tmp_buf);
+    }
+}
 
-        for (i = 0; i < layout->cnt; i++) {
-                snprintf (tmp_buf, 128, "(%s %u %u)",
-                          layout->list[i].xlator->name,
-                          layout->list[i].start,
-                          layout->list[i].stop);
-                if (i)
-                        strcat (buf, " ");
-                strcat (buf, tmp_buf);
+static void
+dht_fill_pathinfo_xattr(xlator_t *this, dht_local_t *local, char *xattr_buf,
+                        int32_t alloc_len, int flag, char *layout_buf)
+{
+    if (flag) {
+        if (local->xattr_val) {
+            snprintf(xattr_buf, alloc_len,
+                     "((<" DHT_PATHINFO_HEADER "%s> %s) (%s-layout %s))",
+                     this->name, local->xattr_val, this->name, layout_buf);
+        } else {
+            snprintf(xattr_buf, alloc_len, "(%s-layout %s)", this->name,
+                     layout_buf);
         }
+    } else if (local->xattr_val) {
+        snprintf(xattr_buf, alloc_len, "(<" DHT_PATHINFO_HEADER "%s> %s)",
+                 this->name, local->xattr_val);
+    } else {
+        xattr_buf[0] = '\0';
+    }
 }
 
-int
-dht_pathinfo_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int op_ret, int op_errno, dict_t *xattr)
-{
-	dht_local_t  *local = NULL;
-        int ret = 0;
-        int flag = 0;
-        int this_call_cnt = 0;
-        char *value_got = NULL;
-        char  layout_buf[8192] = {0,};
-        char  xattr_buf[8192 + 1024] = {0,};
-        dict_t *dict = NULL;
+static int
+dht_vgetxattr_alloc_and_fill(dht_local_t *local, dict_t *xattr, xlator_t *this,
+                             int op_errno)
+{
+    int ret = -1;
+    char *value = NULL;
+
+    ret = dict_get_str(xattr, local->xsel, &value);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, DHT_MSG_GET_XATTR_FAILED,
+               "Subvolume %s returned -1", this->name);
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+        goto out;
+    }
+
+    local->alloc_len += strlen(value);
+
+    if (!local->xattr_val) {
+        local->alloc_len += (SLEN(DHT_PATHINFO_HEADER) + 10);
+        local->xattr_val = GF_MALLOC(local->alloc_len, gf_common_mt_char);
+        if (!local->xattr_val) {
+            ret = -1;
+            goto out;
+        }
+        local->xattr_val[0] = '\0';
+    }
+
+    int plen = strlen(local->xattr_val);
+    if (plen) {
+        /* extra byte(s) for \0 to be safe */
+        local->alloc_len += (plen + 2);
+        local->xattr_val = GF_REALLOC(local->xattr_val, local->alloc_len);
+        if (!local->xattr_val) {
+            ret = -1;
+            goto out;
+        }
+    }
 
-        local = frame->local;
+    (void)strcat(local->xattr_val, value);
+    (void)strcat(local->xattr_val, " ");
+    local->op_ret = 0;
 
-        if (op_ret != -1) {
-                ret = dict_get_str (xattr, GF_XATTR_PATHINFO_KEY, &value_got);
-                if (!ret) {
-                        if (!local->pathinfo)
-                                local->pathinfo = GF_CALLOC (8192, sizeof (char),
-                                                             gf_common_mt_char);
-                        if (local->pathinfo)
-                                strcat (local->pathinfo, value_got);
-                }
-        }
+    ret = 0;
 
-        this_call_cnt = dht_frame_return (frame);
-        if (is_last_call (this_call_cnt)) {
-                if (local->layout->cnt > 1) {
-                        /* Set it for directory */
-                        fill_layout_info (local->layout, layout_buf);
-                        flag = 1;
-                }
+out:
+    return ret;
+}
 
-                dict = dict_new ();
+static int
+dht_vgetxattr_fill_and_set(dht_local_t *local, dict_t **dict, xlator_t *this,
+                           gf_boolean_t flag)
+{
+    int ret = -1;
+    char *xattr_buf = NULL;
+    char layout_buf[8192] = {
+        0,
+    };
+
+    if (flag)
+        fill_layout_info(local->layout, layout_buf);
+
+    *dict = dict_new();
+    if (!*dict)
+        goto out;
+
+    local->xattr_val[strlen(local->xattr_val) - 1] = '\0';
+
+    /* we would need max this many bytes to create xattr string
+     * extra 40 bytes is just an estimated amount of additional
+     * space required as we include translator name and some
+     * spaces, brackets etc. when forming the pathinfo string.
+     *
+     * For node-uuid we just don't have all the pretty formatting,
+     * but since this is a generic routine for pathinfo & node-uuid
+     * we don't have conditional space allocation and try to be
+     * generic
+     */
+    local->alloc_len += (2 * strlen(this->name)) + strlen(layout_buf) + 40;
+    xattr_buf = GF_MALLOC(local->alloc_len, gf_common_mt_char);
+    if (!xattr_buf)
+        goto out;
+
+    if (XATTR_IS_PATHINFO(local->xsel)) {
+        (void)dht_fill_pathinfo_xattr(this, local, xattr_buf, local->alloc_len,
+                                      flag, layout_buf);
+    } else if ((XATTR_IS_NODE_UUID(local->xsel)) ||
+               (XATTR_IS_NODE_UUID_LIST(local->xsel))) {
+        (void)snprintf(xattr_buf, local->alloc_len, "%s", local->xattr_val);
+    } else {
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_GET_XATTR_FAILED,
+               "Unknown local->xsel (%s)", local->xsel);
+        GF_FREE(xattr_buf);
+        goto out;
+    }
+
+    ret = dict_set_dynstr(*dict, local->xsel, xattr_buf);
+    if (ret)
+        GF_FREE(xattr_buf);
+    GF_FREE(local->xattr_val);
 
-                if (flag && local->pathinfo)
-                        snprintf (xattr_buf, 9216, "((%s %s) (%s-layout %s))",
-                                  this->name, local->pathinfo, this->name,
-                                  layout_buf);
-                else if (local->pathinfo)
-                        snprintf (xattr_buf, 9216, "(%s %s)",
-                                  this->name, local->pathinfo);
-                else if (flag)
-                        snprintf (xattr_buf, 9216, "(%s-layout %s)",
-                                  this->name, layout_buf);
+out:
+    return ret;
+}
 
-                ret = dict_set_str (dict, GF_XATTR_PATHINFO_KEY,
-                                    xattr_buf);
+static int
+dht_find_local_subvol_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int op_ret, int op_errno, dict_t *xattr,
+                          dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    xlator_t *prev = NULL;
+    int this_call_cnt = 0;
+    int ret = 0;
+    char *uuid_str = NULL;
+    char *uuid_list = NULL;
+    char *next_uuid_str = NULL;
+    char *saveptr = NULL;
+    uuid_t node_uuid = {
+        0,
+    };
+    char *uuid_list_copy = NULL;
+    int count = 0;
+    int i = 0;
+    int index = 0;
+    int found = 0;
+    nodeuuid_info_t *tmp_ptr = NULL;
+
+    VALIDATE_OR_GOTO(frame, out);
+    VALIDATE_OR_GOTO(frame->local, out);
+
+    local = frame->local;
+    prev = cookie;
+    conf = this->private;
+
+    VALIDATE_OR_GOTO(conf->defrag, out);
+
+    gf_msg_debug(this->name, 0, "subvol %s returned", prev->name);
+
+    LOCK(&frame->lock);
+    {
+        this_call_cnt = --local->call_cnt;
+        if (op_ret < 0) {
+            local->op_ret = -1;
+            local->op_errno = op_errno;
+            UNLOCK(&frame->lock);
+            if (op_errno == ENODATA)
+                gf_msg_debug(this->name, 0, "failed to get node-uuid");
+            else
+                gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                       DHT_MSG_GET_XATTR_FAILED, "failed to get node-uuid");
+            goto post_unlock;
+        }
 
-                if (local->pathinfo)
-                        GF_FREE (local->pathinfo);
-                GF_FREE (local->key);
+        ret = dict_get_str(xattr, local->xsel, &uuid_list);
 
-                DHT_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_GET_FAILED,
+                   "Failed to get %s", local->xsel);
+            local->op_ret = -1;
+            local->op_errno = EINVAL;
+            goto unlock;
+        }
 
-                if (dict)
-                        dict_unref (dict);
+        /* As DHT will not know details of its child xlators
+         * we need to parse this twice to get the count first
+         * and allocate memory later.
+         */
+        count = 0;
+        index = conf->local_subvols_cnt;
 
-                return 0;
+        uuid_list_copy = gf_strdup(uuid_list);
+        if (!uuid_list_copy)
+            goto unlock;
+
+        for (uuid_str = strtok_r(uuid_list, " ", &saveptr); uuid_str;
+             uuid_str = next_uuid_str) {
+            next_uuid_str = strtok_r(NULL, " ", &saveptr);
+            if (gf_uuid_parse(uuid_str, node_uuid)) {
+                local->op_ret = -1;
+                local->op_errno = EINVAL;
+                UNLOCK(&frame->lock);
+                gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_UUID_PARSE_ERROR,
+                       "Failed to parse uuid for %s", prev->name);
+                goto post_unlock;
+            }
+
+            count++;
+            if (gf_uuid_compare(node_uuid, conf->defrag->node_uuid)) {
+                gf_msg_debug(this->name, 0,
+                             "subvol %s does not"
+                             "belong to this node",
+                             prev->name);
+            } else {
+                /* handle multiple bricks of the same replica
+                 * on the same node */
+                if (found)
+                    continue;
+                conf->local_subvols[(conf->local_subvols_cnt)++] = prev;
+                found = 1;
+                gf_msg_debug(this->name, 0,
+                             "subvol %s belongs to"
+                             " this node",
+                             prev->name);
+            }
         }
 
-        if (local->pathinfo)
-                strcat (local->pathinfo, " Link: ");
-        if (local->hashed_subvol) {
-                /* This will happen if there pending */
-                STACK_WIND (frame, dht_pathinfo_getxattr_cbk, local->hashed_subvol,
-                            local->hashed_subvol->fops->getxattr,
-                            &local->loc, local->key);
+        if (!found) {
+            local->op_ret = 0;
+            goto unlock;
+        }
 
-                return 0;
+        conf->local_nodeuuids[index].count = count;
+        conf->local_nodeuuids[index].elements = GF_CALLOC(
+            count, sizeof(nodeuuid_info_t), 1);
+
+        /* The node-uuids are guaranteed to be returned in the same
+         * order as the bricks
+         * A null node-uuid is returned for a brick that is down.
+         */
+
+        saveptr = NULL;
+        i = 0;
+
+        for (uuid_str = strtok_r(uuid_list_copy, " ", &saveptr); uuid_str;
+             uuid_str = next_uuid_str) {
+            next_uuid_str = strtok_r(NULL, " ", &saveptr);
+            tmp_ptr = &(conf->local_nodeuuids[index].elements[i]);
+            gf_uuid_parse(uuid_str, tmp_ptr->uuid);
+
+            if (!gf_uuid_compare(tmp_ptr->uuid, conf->defrag->node_uuid)) {
+                tmp_ptr->info = REBAL_NODEUUID_MINE;
+            }
+            i++;
+            tmp_ptr = NULL;
         }
+    }
 
-        gf_log ("this->name", GF_LOG_ERROR, "Unable to find hashed_subvol for path"
-                " %s", local->pathinfo);
-      
-        DHT_STACK_UNWIND (getxattr, frame, -1, op_errno, dict);
-        return 0;
+    local->op_ret = 0;
+unlock:
+    UNLOCK(&frame->lock);
+post_unlock:
+    if (!is_last_call(this_call_cnt))
+        goto out;
+
+    if (local->op_ret == -1) {
+        goto unwind;
+    }
+
+    DHT_STACK_UNWIND(getxattr, frame, 0, 0, xattr, xdata);
+    goto out;
+
+unwind:
+
+    GF_FREE(conf->local_nodeuuids[index].elements);
+    conf->local_nodeuuids[index].elements = NULL;
+
+    DHT_STACK_UNWIND(getxattr, frame, -1, local->op_errno, NULL, xdata);
+out:
+    GF_FREE(uuid_list_copy);
+    return 0;
 }
 
-int
-dht_linkinfo_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int op_ret, int op_errno, dict_t *xattr)
-{
-        int   ret   = 0;
-        char *value = NULL;
-
-        if (op_ret != -1) {
-                ret = dict_get_str (xattr, GF_XATTR_PATHINFO_KEY, &value);
-                if (!ret) {
-                        ret = dict_set_str (xattr, GF_XATTR_LINKINFO_KEY, value);
-                        if (!ret)
-                                gf_log (this->name, GF_LOG_TRACE,
-                                        "failed to set linkinfo");
-                }
+static int
+dht_vgetxattr_dir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
+{
+    int ret = 0;
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    dict_t *dict = NULL;
+
+    VALIDATE_OR_GOTO(frame, out);
+    VALIDATE_OR_GOTO(frame->local, out);
+
+    local = frame->local;
+
+    LOCK(&frame->lock);
+    {
+        this_call_cnt = --local->call_cnt;
+        if (op_ret < 0) {
+            if (op_errno != ENOTCONN) {
+                local->op_ret = -1;
+                local->op_errno = op_errno;
+                UNLOCK(&frame->lock);
+                gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                       DHT_MSG_GET_XATTR_FAILED, "getxattr err for dir");
+                goto post_unlock;
+            }
+
+            goto unlock;
+        }
+
+        ret = dht_vgetxattr_alloc_and_fill(local, xattr, this, op_errno);
+        if (ret) {
+            UNLOCK(&frame->lock);
+            gf_msg(this->name, GF_LOG_ERROR, op_errno, DHT_MSG_DICT_SET_FAILED,
+                   "alloc or fill failure");
+            goto post_unlock;
         }
+    }
+unlock:
+    UNLOCK(&frame->lock);
+post_unlock:
+    if (!is_last_call(this_call_cnt))
+        goto out;
 
-        DHT_STACK_UNWIND (getxattr, frame, op_ret, op_errno, xattr);
+    /* -- last call: do patch ups -- */
 
-        return 0;
+    if (local->op_ret == -1) {
+        goto unwind;
+    }
+
+    ret = dht_vgetxattr_fill_and_set(local, &dict, this, _gf_true);
+    if (ret)
+        goto unwind;
+
+    DHT_STACK_UNWIND(getxattr, frame, 0, 0, dict, xdata);
+    goto cleanup;
+
+unwind:
+    DHT_STACK_UNWIND(getxattr, frame, -1, local->op_errno, NULL, NULL);
+cleanup:
+    if (dict)
+        dict_unref(dict);
+out:
+    return 0;
 }
 
-int
-dht_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		  int op_ret, int op_errno, dict_t *xattr)
+static int
+dht_vgetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                  int op_errno, dict_t *xattr, dict_t *xdata)
 {
-        if (op_ret != -1) {
-                if (dict_get (xattr, "trusted.glusterfs.dht")) {
-                        dict_del (xattr, "trusted.glusterfs.dht");
-                }
+    dht_local_t *local = NULL;
+    int ret = 0;
+    dict_t *dict = NULL;
+    xlator_t *prev = NULL;
+    gf_boolean_t flag = _gf_true;
+
+    local = frame->local;
+    prev = cookie;
+
+    if (op_ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, DHT_MSG_GET_XATTR_FAILED,
+               "vgetxattr: Subvolume %s returned -1", prev->name);
+        goto unwind;
+    }
+
+    ret = dht_vgetxattr_alloc_and_fill(local, xattr, this, op_errno);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_NO_MEMORY,
+               "Allocation or fill failure");
+        goto unwind;
+    }
+
+    flag = (local->layout->cnt > 1) ? _gf_true : _gf_false;
+
+    ret = dht_vgetxattr_fill_and_set(local, &dict, this, flag);
+    if (ret)
+        goto unwind;
+
+    DHT_STACK_UNWIND(getxattr, frame, 0, 0, dict, xdata);
+    goto cleanup;
+
+unwind:
+    DHT_STACK_UNWIND(getxattr, frame, -1, local->op_errno, NULL, NULL);
+cleanup:
+    if (dict)
+        dict_unref(dict);
+
+    return 0;
+}
+
+static int
+dht_linkinfo_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int op_ret, int op_errno, dict_t *xattr,
+                          dict_t *xdata)
+{
+    int ret = 0;
+    char *value = NULL;
+
+    if (op_ret != -1) {
+        ret = dict_get_str(xattr, GF_XATTR_PATHINFO_KEY, &value);
+        if (!ret) {
+            ret = dict_set_str(xattr, GF_XATTR_LINKINFO_KEY, value);
+            if (!ret)
+                gf_msg_trace(this->name, 0, "failed to set linkinfo");
         }
+    }
 
-        DHT_STACK_UNWIND (getxattr, frame, op_ret, op_errno, xattr);
+    DHT_STACK_UNWIND(getxattr, frame, op_ret, op_errno, xattr, xdata);
 
-        return 0;
+    return 0;
 }
 
+static int
+dht_mds_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(frame->local, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+    local = frame->local;
+
+    if (!xattr || (op_ret == -1)) {
+        local->op_ret = op_ret;
+        goto out;
+    }
+    dict_del(xattr, conf->xattr_name);
+    local->op_ret = 0;
+
+    if (!local->xattr) {
+        local->xattr = dict_copy_with_ref(xattr, NULL);
+    }
+
+out:
+    DHT_STACK_UNWIND(getxattr, frame, local->op_ret, op_errno, local->xattr,
+                     xdata);
+    return 0;
+err:
+    DHT_STACK_UNWIND(getxattr, frame, -1, EINVAL, NULL, NULL);
+    return 0;
+}
 
 int
-dht_getxattr (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, const char *key)
-{
-	xlator_t     *subvol        = NULL;
-	xlator_t     *hashed_subvol = NULL;
-	xlator_t     *cached_subvol = NULL;
-	dht_conf_t   *conf          = NULL;
-	dht_local_t  *local         = NULL;
-        dht_layout_t *layout        = NULL;
-        int           op_errno      = -1;
-        int           ret           = 0;
-        int           flag          = 0;
-        int           i             = 0;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (this->private, err);
-
-        conf   = this->private;
-        layout = dht_layout_get (this, loc->inode);
-        if (key && (strcmp (key, GF_XATTR_PATHINFO_KEY) == 0)) {
-                hashed_subvol = dht_subvol_get_hashed (this, loc);
-                cached_subvol = dht_subvol_get_cached (this, loc->inode);
-
-                local = dht_local_init (frame);
-                if (!local) {
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Out of memory");
-                        goto err;
-                }
+dht_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, dict_t *xattr, dict_t *xdata)
+{
+    int this_call_cnt = 0;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = 0;
 
-                ret = loc_dup (loc, &local->loc);
-                if (ret == -1) {
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Out of memory");
-                        goto err;
-                }
-                local->key = gf_strdup (key);
-                if (!local->key) {
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Out of memory");
-                        goto err;
-                }
-                local->layout = layout;
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(frame->local, err);
+    VALIDATE_OR_GOTO(this->private, err);
 
-                local->call_cnt = 1;
-                if (hashed_subvol != cached_subvol) {
-                        local->call_cnt = 2;
-                        local->hashed_subvol = hashed_subvol;
-                }
+    conf = this->private;
+    local = frame->local;
+
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto err;
+        return 0;
+    }
 
-                STACK_WIND (frame, dht_pathinfo_getxattr_cbk, cached_subvol,
-                            cached_subvol->fops->getxattr, loc, key);
+    LOCK(&frame->lock);
+    {
+        if (!xattr || (op_ret == -1)) {
+            local->op_ret = op_ret;
+            goto unlock;
+        }
 
-                return 0;
+        dict_del(xattr, conf->xattr_name);
+        dict_del(xattr, conf->mds_xattr_key);
+
+        dict_del(xattr, conf->commithash_xattr_name);
+
+        if (frame->root->pid >= 0) {
+            GF_REMOVE_INTERNAL_XATTR("trusted.glusterfs.quota*", xattr);
+            GF_REMOVE_INTERNAL_XATTR("trusted.pgfid*", xattr);
+        }
+
+        local->op_ret = 0;
+
+        if (!local->xattr) {
+            local->xattr = dict_copy_with_ref(xattr, NULL);
+        } else {
+            dht_aggregate_xattr(local->xattr, xattr);
+        }
+
+        if (!local->xdata) {
+            local->xdata = dict_ref(xdata);
+        } else if ((local->inode && IA_ISDIR(local->inode->ia_type)) ||
+                   (local->fd && IA_ISDIR(local->fd->inode->ia_type))) {
+            dht_aggregate_xattr(local->xdata, xdata);
+        }
+    }
+unlock:
+    UNLOCK(&frame->lock);
+
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        /* If we have a valid xattr received from any one of the
+         * subvolume, let's return it */
+        if (local->xattr) {
+            local->op_ret = 0;
+        }
+
+        DHT_STACK_UNWIND(getxattr, frame, local->op_ret, op_errno, local->xattr,
+                         local->xdata);
+    }
+    return 0;
+err:
+    DHT_STACK_UNWIND(getxattr, frame, -1, EINVAL, NULL, NULL);
+    return 0;
+}
+
+static int32_t
+dht_getxattr_unwind(call_frame_t *frame, int op_ret, int op_errno, dict_t *dict,
+                    dict_t *xdata)
+{
+    DHT_STACK_UNWIND(getxattr, frame, op_ret, op_errno, dict, xdata);
+    return 0;
+}
+
+static int
+dht_getxattr_get_real_filename_cbk(call_frame_t *frame, void *cookie,
+                                   xlator_t *this, int op_ret, int op_errno,
+                                   dict_t *xattr, dict_t *xdata)
+{
+    int this_call_cnt = 0;
+    dht_local_t *local = NULL;
+
+    local = frame->local;
+
+    LOCK(&frame->lock);
+    {
+        if (local->op_errno == EOPNOTSUPP) {
+            /* Nothing to do here, we have already found
+             * a subvol which does not have the get_real_filename
+             * optimization. If condition is for simple logic.
+             */
+            goto unlock;
         }
-        if (key && (strcmp (key, GF_XATTR_LINKINFO_KEY) == 0)) {
-                hashed_subvol = dht_subvol_get_hashed (this, loc);
-                cached_subvol = dht_subvol_get_cached (this, loc->inode);
-                if (hashed_subvol == cached_subvol) {
-                        op_errno = ENODATA;
-                        goto err;
+
+        if (op_ret == -1) {
+            if (op_errno == EOPNOTSUPP) {
+                /* This subvol does not have the optimization.
+                 * Better let the user know we don't support it.
+                 * Remove previous results if any.
+                 */
+
+                if (local->xattr) {
+                    dict_unref(local->xattr);
+                    local->xattr = NULL;
                 }
-                if (hashed_subvol) {
-                        STACK_WIND (frame, dht_linkinfo_getxattr_cbk, hashed_subvol,
-                                    hashed_subvol->fops->getxattr, loc,
-                                    GF_XATTR_PATHINFO_KEY);
-                        return 0;
+
+                if (local->xattr_req) {
+                    dict_unref(local->xattr_req);
+                    local->xattr_req = NULL;
                 }
-                op_errno = ENODATA;
-                goto err;
+
+                local->op_ret = op_ret;
+                local->op_errno = op_errno;
+                UNLOCK(&frame->lock);
+                gf_msg(this->name, GF_LOG_WARNING, op_errno,
+                       DHT_MSG_UPGRADE_BRICKS,
+                       "At least "
+                       "one of the bricks does not support "
+                       "this operation. Please upgrade all "
+                       "bricks.");
+                goto post_unlock;
+            }
+
+            if (op_errno == ENOATTR) {
+                /* Do nothing, our defaults are set to this.
+                 */
+                goto unlock;
+            }
+
+            /* This is a place holder for every other error
+             * case. I am not sure of how to interpret
+             * ENOTCONN etc. As of now, choosing to ignore
+             * down subvol and return a good result(if any)
+             * from other subvol.
+             */
+            UNLOCK(&frame->lock);
+            gf_msg(this->name, GF_LOG_WARNING, op_errno,
+                   DHT_MSG_GET_XATTR_FAILED, "Failed to get real filename.");
+            goto post_unlock;
         }
-        if (key && (strcmp (key, GF_XATTR_FIX_LAYOUT_KEY) == 0)) {
-                for (i = 0; i < layout->cnt; i++) {
-                        if (layout->list[i].start == layout->list[i].stop) {
-                                flag = 1;
-                                break;
-                        }
-                }
-                if ((layout->cnt < conf->subvolume_cnt) || flag) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "expanding layout of %s from %d to %d",
-                                loc->path, layout->cnt, conf->subvolume_cnt);
-                        local = dht_local_init (frame);
-                        if (!local) {
-                                op_errno = ENOMEM;
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "Out of memory");
-                                goto err;
-                        }
 
-                        ret = loc_dup (loc, &local->loc);
-                        if (ret == -1) {
-                                op_errno = ENOMEM;
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "Out of memory");
-                                goto err;
-                        }
-                        local->layout = layout;
-                        //layout = dht_layout_new (this, conf->subvolume_cnt);
+        /* This subvol has the required file.
+         * There could be other subvols which have returned
+         * success already, choosing to return the latest good
+         * result.
+         */
+        if (local->xattr)
+            dict_unref(local->xattr);
+        local->xattr = dict_ref(xattr);
 
-                        dht_selfheal_new_directory (frame, dht_fix_layout_cbk,
-                                                    layout);
-                        return 0;
-                }
-                op_errno = ENODATA;
-                goto err;
+        if (local->xattr_req) {
+            dict_unref(local->xattr_req);
+            local->xattr_req = NULL;
         }
-	subvol = dht_subvol_get_cached (this, loc->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+        if (xdata)
+            local->xattr_req = dict_ref(xdata);
+
+        local->op_ret = op_ret;
+        local->op_errno = 0;
+        UNLOCK(&frame->lock);
+        gf_msg_debug(this->name, 0, "Found a matching file.");
+        goto post_unlock;
+    }
+unlock:
+    UNLOCK(&frame->lock);
+post_unlock:
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        DHT_STACK_UNWIND(getxattr, frame, local->op_ret, local->op_errno,
+                         local->xattr, local->xattr_req);
+    }
+
+    return 0;
+}
 
-	STACK_WIND (frame, dht_getxattr_cbk,
-		    subvol, subvol->fops->getxattr,
-		    loc, key);
+static int
+dht_getxattr_get_real_filename(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                               const char *key, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int i = 0;
+    dht_layout_t *layout = NULL;
+    int cnt = 0;
+    xlator_t *subvol = NULL;
 
-	return 0;
+    local = frame->local;
+    layout = local->layout;
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (getxattr, frame, -1, op_errno, NULL);
+    cnt = local->call_cnt = layout->cnt;
+
+    local->op_ret = -1;
+    local->op_errno = ENOATTR;
 
-	return 0;
+    for (i = 0; i < cnt; i++) {
+        subvol = layout->list[i].xlator;
+        STACK_WIND(frame, dht_getxattr_get_real_filename_cbk, subvol,
+                   subvol->fops->getxattr, loc, key, xdata);
+    }
+
+    return 0;
 }
 
+static int
+dht_marker_populate_args(call_frame_t *frame, int type, int *gauge,
+                         xlator_t **subvols)
+{
+    dht_local_t *local = NULL;
+    int i = 0;
+    dht_layout_t *layout = NULL;
 
-int
-dht_setxattr (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, dict_t *xattr, int flags)
-{
-	xlator_t     *subvol   = NULL;
-	dht_local_t  *local    = NULL;
-        dht_conf_t   *conf     = NULL;
-        dht_layout_t *layout   = NULL;
-        int           i        = 0;
-        int           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-        conf   = this->private;
-	subvol = dht_subvol_get_cached (this, loc->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	local->layout = layout = dht_layout_get (this, loc->inode);
-	if (!layout) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no layout for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	local->call_cnt = layout->cnt;
-
-	for (i = 0; i < layout->cnt; i++) {
-		STACK_WIND (frame, dht_err_cbk,
-			    layout->list[i].xlator,
-			    layout->list[i].xlator->fops->setxattr,
-			    loc, xattr, flags);
-	}
-
-	return 0;
+    local = frame->local;
+    layout = local->layout;
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (setxattr, frame, -1, op_errno);
+    for (i = 0; i < layout->cnt; i++)
+        subvols[i] = layout->list[i].xlator;
 
-	return 0;
+    return layout->cnt;
 }
 
-
-int
-dht_removexattr (call_frame_t *frame, xlator_t *this,
-		 loc_t *loc, const char *key)
+static int
+dht_is_debug_xattr_key(const char **array, char *key)
 {
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
+    int i = 0;
 
+    for (i = 0; array[i]; i++) {
+        if (fnmatch(array[i], key, FNM_NOESCAPE) == 0)
+            return i;
+    }
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
+    return -1;
+}
 
-	subvol = dht_subvol_get_cached (this, loc->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
+/* Note we already have frame->local initialised here*/
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
+static int
+dht_handle_debug_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                          const char *key)
+{
+    dht_local_t *local = NULL;
+    int ret = -1;
+    int op_errno = ENODATA;
+    char *value = NULL;
+    loc_t file_loc = {0};
+    const char *name = NULL;
+
+    local = frame->local;
+
+    if (dht_is_debug_xattr_key(dht_dbg_vxattrs, (char *)key) == -1) {
+        goto out;
+    }
+
+    local->xattr = dict_new();
+    if (!local->xattr) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    if (strncmp(key, DHT_DBG_HASHED_SUBVOL_KEY,
+                SLEN(DHT_DBG_HASHED_SUBVOL_KEY)) == 0) {
+        name = key + strlen(DHT_DBG_HASHED_SUBVOL_KEY);
+        if (strlen(name) == 0) {
+            op_errno = EINVAL;
+            goto out;
+        }
 
-	local->call_cnt = 1;
+        ret = dht_build_child_loc(this, &file_loc, loc, (char *)name);
+        if (ret) {
+            op_errno = ENOMEM;
+            goto out;
+        }
 
-	STACK_WIND (frame, dht_err_cbk,
-		    subvol, subvol->fops->removexattr,
-		    loc, key);
+        local->hashed_subvol = dht_subvol_get_hashed(this, &file_loc);
+        if (local->hashed_subvol == NULL) {
+            op_errno = ENODATA;
+            goto out;
+        }
 
-	return 0;
+        value = gf_strdup(local->hashed_subvol->name);
+        if (!value) {
+            op_errno = ENOMEM;
+            goto out;
+        }
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (removexattr, frame, -1, op_errno);
+        ret = dict_set_dynstr(local->xattr, (char *)key, value);
+        if (ret < 0) {
+            op_errno = -ret;
+            ret = -1;
+            goto out;
+        }
+        ret = 0;
+        goto out;
+    }
 
-	return 0;
+out:
+    loc_wipe(&file_loc);
+    DHT_STACK_UNWIND(getxattr, frame, ret, op_errno, local->xattr, NULL);
+    return 0;
 }
 
+/* Virtual Xattr which returns 1 if all subvols are up,
+   else returns 0. Geo-rep then uses this virtual xattr
+   after a fresh mount and starts the I/O.
+*/
+
+enum dht_vxattr_subvol {
+    DHT_VXATTR_SUBVOLS_UP = 1,
+    DHT_VXATTR_SUBVOLS_DOWN = 0,
+};
 
 int
-dht_fd_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-	      int op_ret, int op_errno, fd_t *fd)
+dht_vgetxattr_subvol_status(call_frame_t *frame, xlator_t *this,
+                            const char *key)
 {
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = 0;
-	call_frame_t *prev = NULL;
+    dht_local_t *local = NULL;
+    int ret = -1;
+    int op_errno = ENODATA;
+    int value = DHT_VXATTR_SUBVOLS_UP;
+    int i = 0;
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+    local = frame->local;
+
+    if (!key) {
+        op_errno = EINVAL;
+        goto out;
+    }
+    local->xattr = dict_new();
+    if (!local->xattr) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (!conf->subvolume_status[i]) {
+            value = DHT_VXATTR_SUBVOLS_DOWN;
+            gf_msg_debug(this->name, 0, "subvol %s is down ",
+                         conf->subvolumes[i]->name);
+            break;
+        }
+    }
+    ret = dict_set_int8(local->xattr, (char *)key, value);
+    if (ret < 0) {
+        op_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+    ret = 0;
 
+out:
+    DHT_STACK_UNWIND(getxattr, frame, ret, op_errno, local->xattr, NULL);
+    return 0;
+}
 
-	local = frame->local;
-	prev = cookie;
+int
+dht_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, const char *key,
+             dict_t *xdata)
+#define DHT_IS_DIR(layout) (layout->cnt > 1)
+{
+    xlator_t *subvol = NULL;
+    xlator_t *hashed_subvol = NULL;
+    xlator_t *mds_subvol = NULL;
+    xlator_t *cached_subvol = NULL;
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+    int op_errno = -1;
+    int i = 0;
+    int cnt = 0;
+    char *node_uuid_key = NULL;
+    int ret = -1;
+
+    GF_CHECK_XATTR_KEY_AND_GOTO(key, IO_THREADS_QUEUE_SIZE_KEY, op_errno, err);
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_GETXATTR);
+    if (!local) {
+        op_errno = ENOMEM;
+
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!layout) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LAYOUT_NULL,
+               "Layout is NULL");
+        op_errno = ENOENT;
+        goto err;
+    }
+
+    /* skip over code which is irrelevant without a valid key */
+    if (!key)
+        goto no_key;
+
+    local->key = gf_strdup(key);
+    if (!local->key) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    if (strncmp(key, conf->mds_xattr_key, strlen(key)) == 0) {
+        op_errno = ENOTSUP;
+        goto err;
+    }
+
+    if (strncmp(key, DHT_SUBVOL_STATUS_KEY, SLEN(DHT_SUBVOL_STATUS_KEY)) == 0) {
+        dht_vgetxattr_subvol_status(frame, this, key);
+        return 0;
+    }
 
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			gf_log (this->name, GF_LOG_DEBUG,
-				"subvolume %s returned -1 (%s)",
-				prev->this->name, strerror (op_errno));
-			goto unlock;
-		}
+    /* skip over code which is irrelevant if !DHT_IS_DIR(layout) */
+    if (!DHT_IS_DIR(layout))
+        goto no_dht_is_dir;
 
-		local->op_ret = 0;
-	}
-unlock:
-	UNLOCK (&frame->lock);
+    if ((strncmp(key, GF_XATTR_GET_REAL_FILENAME_KEY,
+                 SLEN(GF_XATTR_GET_REAL_FILENAME_KEY)) == 0) &&
+        DHT_IS_DIR(layout)) {
+        dht_getxattr_get_real_filename(frame, this, loc, key, xdata);
+        return 0;
+    }
+
+    if (!strcmp(key, GF_REBAL_FIND_LOCAL_SUBVOL)) {
+        ret = gf_asprintf(&node_uuid_key, "%s", GF_XATTR_LIST_NODE_UUIDS_KEY);
+        if (ret == -1 || !node_uuid_key) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_NO_MEMORY,
+                   "Failed to copy node uuid key");
+            op_errno = ENOMEM;
+            goto err;
+        }
+        (void)snprintf(local->xsel, sizeof(local->xsel), "%s", node_uuid_key);
+        cnt = local->call_cnt = conf->subvolume_cnt;
+        for (i = 0; i < cnt; i++) {
+            STACK_WIND_COOKIE(frame, dht_find_local_subvol_cbk,
+                              conf->subvolumes[i], conf->subvolumes[i],
+                              conf->subvolumes[i]->fops->getxattr, loc,
+                              node_uuid_key, xdata);
+        }
+        if (node_uuid_key)
+            GF_FREE(node_uuid_key);
+        return 0;
+    }
+
+    if (!strcmp(key, GF_REBAL_OLD_FIND_LOCAL_SUBVOL)) {
+        ret = gf_asprintf(&node_uuid_key, "%s", GF_XATTR_NODE_UUID_KEY);
+        if (ret == -1 || !node_uuid_key) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_NO_MEMORY,
+                   "Failed to copy node uuid key");
+            op_errno = ENOMEM;
+            goto err;
+        }
+        (void)snprintf(local->xsel, sizeof(local->xsel), "%s", node_uuid_key);
+        cnt = local->call_cnt = conf->subvolume_cnt;
+        for (i = 0; i < cnt; i++) {
+            STACK_WIND_COOKIE(frame, dht_find_local_subvol_cbk,
+                              conf->subvolumes[i], conf->subvolumes[i],
+                              conf->subvolumes[i]->fops->getxattr, loc,
+                              node_uuid_key, xdata);
+        }
+        if (node_uuid_key)
+            GF_FREE(node_uuid_key);
+        return 0;
+    }
+
+    /* for file use cached subvolume (obviously!): see if {}
+     * below
+     * for directory:
+     *  wind to all subvolumes and exclude subvolumes which
+     *  return ENOTCONN (in callback)
+     *
+     * NOTE: Don't trust inode here, as that may not be valid
+     *       (until inode_link() happens)
+     */
+
+    if (XATTR_IS_PATHINFO(key) || (strcmp(key, GF_XATTR_NODE_UUID_KEY) == 0) ||
+        (strcmp(key, GF_XATTR_LIST_NODE_UUIDS_KEY) == 0)) {
+        (void)snprintf(local->xsel, sizeof(local->xsel), "%s", key);
+        cnt = local->call_cnt = layout->cnt;
+        for (i = 0; i < cnt; i++) {
+            subvol = layout->list[i].xlator;
+            STACK_WIND(frame, dht_vgetxattr_dir_cbk, subvol,
+                       subvol->fops->getxattr, loc, key, xdata);
+        }
+        return 0;
+    }
+
+no_dht_is_dir:
+    /* node-uuid or pathinfo for files */
+    if (XATTR_IS_PATHINFO(key) || (strcmp(key, GF_XATTR_NODE_UUID_KEY) == 0)) {
+        cached_subvol = local->cached_subvol;
+        (void)snprintf(local->xsel, sizeof(local->xsel), "%s", key);
+        local->call_cnt = 1;
+        STACK_WIND_COOKIE(frame, dht_vgetxattr_cbk, cached_subvol,
+                          cached_subvol, cached_subvol->fops->getxattr, loc,
+                          key, xdata);
+
+        return 0;
+    }
+
+    if (strcmp(key, GF_XATTR_LINKINFO_KEY) == 0) {
+        hashed_subvol = dht_subvol_get_hashed(this, loc);
+        if (!hashed_subvol) {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+                   "Failed to get hashed subvol for %s", loc->path);
+            op_errno = EINVAL;
+            goto err;
+        }
+
+        cached_subvol = dht_subvol_get_cached(this, loc->inode);
+        if (!cached_subvol) {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   DHT_MSG_CACHED_SUBVOL_GET_FAILED,
+                   "Failed to get cached subvol for %s", loc->path);
+            op_errno = EINVAL;
+            goto err;
+        }
 
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt))
-		DHT_STACK_UNWIND (open, frame, local->op_ret, local->op_errno,
-				  local->fd);
+        if (hashed_subvol == cached_subvol) {
+            op_errno = ENODATA;
+            goto err;
+        }
 
+        STACK_WIND(frame, dht_linkinfo_getxattr_cbk, hashed_subvol,
+                   hashed_subvol->fops->getxattr, loc, GF_XATTR_PATHINFO_KEY,
+                   xdata);
+        return 0;
+    }
+
+    if (dht_is_debug_xattr_key(dht_dbg_vxattrs, (char *)key) >= 0) {
+        dht_handle_debug_getxattr(frame, this, loc, key);
+        return 0;
+    }
+
+no_key:
+    if (cluster_handle_marker_getxattr(frame, loc, key, conf->vol_uuid,
+                                       dht_getxattr_unwind,
+                                       dht_marker_populate_args) == 0)
         return 0;
-}
 
+    if (DHT_IS_DIR(layout)) {
+        local->call_cnt = conf->subvolume_cnt;
+        cnt = conf->subvolume_cnt;
+        ret = dht_inode_ctx_mdsvol_get(loc->inode, this, &mds_subvol);
+        if (!mds_subvol) {
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+                   "Cannot determine MDS, fetching xattr %s randomly"
+                   " from a subvol for path %s ",
+                   key, loc->path);
+        } else {
+            /* TODO need to handle it, As of now we are
+               choosing availability instead of chossing
+               consistencty, in case of mds_subvol is
+               down winding a getxattr call on other subvol
+               and return xattr
+            */
+            local->mds_subvol = mds_subvol;
+            for (i = 0; i < cnt; i++) {
+                if (conf->subvolumes[i] == mds_subvol) {
+                    if (!conf->subvolume_status[i]) {
+                        gf_msg(this->name, GF_LOG_INFO, 0,
+                               DHT_MSG_HASHED_SUBVOL_DOWN,
+                               "MDS %s is down for path"
+                               " path %s so fetching xattr "
+                               "%s randomly from a subvol ",
+                               local->mds_subvol->name, loc->path, key);
+                        ret = 1;
+                    }
+                }
+            }
+        }
+
+        if (!ret && key && local->mds_subvol && dht_match_xattr(key)) {
+            STACK_WIND(frame, dht_mds_getxattr_cbk, local->mds_subvol,
+                       local->mds_subvol->fops->getxattr, loc, key, xdata);
+
+            return 0;
+        }
+    } else {
+        cnt = local->call_cnt = 1;
+    }
+
+    for (i = 0; i < cnt; i++) {
+        subvol = layout->list[i].xlator;
+        STACK_WIND(frame, dht_getxattr_cbk, subvol, subvol->fops->getxattr, loc,
+                   key, xdata);
+    }
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(getxattr, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
+#undef DHT_IS_DIR
 
 int
-dht_open (call_frame_t *frame, xlator_t *this,
-	  loc_t *loc, int flags, fd_t *fd, int wbflags)
-{
-	xlator_t     *subvol = NULL;
-	int           ret = -1;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	local->fd = fd_ref (fd);
-	ret = loc_dup (loc, &local->loc);
-	if (ret == -1) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	local->call_cnt = 1;
-
-	STACK_WIND (frame, dht_fd_cbk,
-		    subvol, subvol->fops->open,
-		    loc, flags, fd, wbflags);
-
-	return 0;
+dht_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *key,
+              dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+    int op_errno = -1;
+    int i = 0;
+    int cnt = 0;
+    xlator_t *mds_subvol = NULL;
+    int ret = -1;
+    dht_conf_t *conf = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+    VALIDATE_OR_GOTO(fd->inode, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_FGETXATTR);
+    if (!local) {
+        op_errno = ENOMEM;
+
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!layout) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LAYOUT_NULL,
+               "Layout is NULL");
+        op_errno = ENOENT;
+        goto err;
+    }
+
+    if (key) {
+        local->key = gf_strdup(key);
+        if (!local->key) {
+            op_errno = ENOMEM;
+            goto err;
+        }
+    }
+
+    gf_uuid_unparse(fd->inode->gfid, gfid);
+
+    if ((fd->inode->ia_type == IA_IFDIR) && key &&
+        (strncmp(key, GF_XATTR_LOCKINFO_KEY, SLEN(GF_XATTR_LOCKINFO_KEY)) !=
+         0)) {
+        local->call_cnt = conf->subvolume_cnt;
+        cnt = conf->subvolume_cnt;
+        ret = dht_inode_ctx_mdsvol_get(fd->inode, this, &mds_subvol);
+
+        if (!mds_subvol) {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+                   "cannot determine MDS, fetching xattr %s "
+                   " randomly from a subvol for gfid %s ",
+                   key, gfid);
+        } else {
+            /* TODO need to handle it, As of now we are
+               choosing availability instead of chossing
+               consistencty, in case of hashed_subvol is
+               down winding a getxattr call on other subvol
+               and return xattr
+            */
+            local->mds_subvol = mds_subvol;
+            for (i = 0; i < cnt; i++) {
+                if (conf->subvolumes[i] == mds_subvol) {
+                    if (!conf->subvolume_status[i]) {
+                        gf_msg(this->name, GF_LOG_WARNING, 0,
+                               DHT_MSG_HASHED_SUBVOL_DOWN,
+                               "MDS subvolume %s is down"
+                               " for gfid %s so fetching xattr "
+                               " %s randomly from a subvol ",
+                               local->mds_subvol->name, gfid, key);
+                        ret = 1;
+                    }
+                }
+            }
+        }
+
+        if (!ret && key && local->mds_subvol && dht_match_xattr(key)) {
+            STACK_WIND(frame, dht_mds_getxattr_cbk, local->mds_subvol,
+                       local->mds_subvol->fops->fgetxattr, fd, key, NULL);
+
+            return 0;
+        }
+
+    } else {
+        cnt = local->call_cnt = 1;
+    }
+
+    for (i = 0; i < cnt; i++) {
+        subvol = layout->list[i].xlator;
+        STACK_WIND(frame, dht_getxattr_cbk, subvol, subvol->fops->fgetxattr, fd,
+                   key, NULL);
+    }
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (open, frame, -1, op_errno, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(fgetxattr, frame, -1, op_errno, NULL, NULL);
 
-	return 0;
+    return 0;
 }
 
+static int
+dht_setxattr2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
+{
+    dht_local_t *local = NULL;
+    int op_errno = EINVAL;
+
+    if (!frame || !frame->local)
+        goto err;
+
+    local = frame->local;
+    op_errno = local->op_errno;
+
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        DHT_STACK_UNWIND(setxattr, frame, local->op_ret, local->op_errno,
+                         local->rebalance.xdata);
+        return 0;
+    }
+
+    if (subvol == NULL)
+        goto err;
+
+    local->call_cnt = 2; /* This is the second attempt */
+
+    if (local->fop == GF_FOP_SETXATTR) {
+        STACK_WIND_COOKIE(frame, dht_file_setxattr_cbk, subvol, subvol,
+                          subvol->fops->setxattr, &local->loc,
+                          local->rebalance.xattr, local->rebalance.flags,
+                          local->xattr_req);
+    } else {
+        STACK_WIND_COOKIE(frame, dht_file_setxattr_cbk, subvol, subvol,
+                          subvol->fops->fsetxattr, local->fd,
+                          local->rebalance.xattr, local->rebalance.flags,
+                          local->xattr_req);
+    }
+
+    return 0;
+
+err:
+    DHT_STACK_UNWIND(setxattr, frame, (local ? local->op_ret : -1), op_errno,
+                     NULL);
+    return 0;
+}
 
 int
-dht_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-	       int op_ret, int op_errno,
-	       struct iovec *vector, int count, struct iatt *stbuf,
-               struct iobref *iobref)
+dht_file_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, dict_t *xdata)
 {
-        dht_local_t     *local = frame->local;
+    int ret = -1;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    struct iatt *stbuf = NULL;
+    inode_t *inode = NULL;
+    xlator_t *subvol1 = NULL, *subvol2 = NULL;
 
-        if (!local) {
-                op_ret = -1;
-                op_errno = EINVAL;
-                goto out;
+    local = frame->local;
+    prev = cookie;
+
+    local->op_errno = op_errno;
+
+    if ((local->fop == GF_FOP_FSETXATTR) &&
+        dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
+
+    if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1.",
+                     prev->name);
+        goto out;
+    }
+
+    if (local->call_cnt != 1)
+        goto out;
+
+    ret = dict_get_bin(xdata, DHT_IATT_IN_XDATA_KEY, (void **)&stbuf);
+
+    if ((!op_ret) && !stbuf) {
+        goto out;
+    }
+
+    local->op_ret = op_ret;
+    local->rebalance.target_op_fn = dht_setxattr2;
+    if (xdata)
+        local->rebalance.xdata = dict_ref(xdata);
+
+    /* Phase 2 of migration */
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(stbuf)) {
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+    /* Phase 1 of migration */
+    if (IS_DHT_MIGRATION_PHASE1(stbuf)) {
+        inode = (local->fd) ? local->fd->inode : local->loc.inode;
+
+        ret = dht_inode_ctx_get_mig_info(this, inode, &subvol1, &subvol2);
+        if (!dht_mig_info_is_invalid(local->cached_subvol, subvol1, subvol2)) {
+            dht_setxattr2(this, subvol2, frame, 0);
+            return 0;
         }
 
-        if (op_ret != -1)
-                stbuf->ia_ino = local->ia_ino;
+        ret = dht_rebalance_in_progress_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
 out:
-        DHT_STACK_UNWIND (readv, frame, op_ret, op_errno, vector, count, stbuf,
-                          iobref);
 
-        return 0;
+    if (local->fop == GF_FOP_SETXATTR) {
+        DHT_STACK_UNWIND(setxattr, frame, op_ret, op_errno, xdata);
+    } else {
+        DHT_STACK_UNWIND(fsetxattr, frame, op_ret, op_errno, xdata);
+    }
+
+    return 0;
 }
 
+/* Function is call by dict_foreach_fnmatch if key is match with
+   user.* and set boolean flag to true
+*/
+static int
+dht_is_user_xattr(dict_t *this, char *key, data_t *value, void *data)
+{
+    gf_boolean_t *user_xattr_found = data;
+    *user_xattr_found = _gf_true;
+    return 0;
+}
 
-int
-dht_readv (call_frame_t *frame, xlator_t *this,
-	   fd_t *fd, size_t size, off_t off)
-{
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-        local = dht_local_init (frame);
-        if (!local) {
-                gf_log (this->name, GF_LOG_ERROR, "Out of memory");
-                op_errno = ENOMEM;
+/* Common code to wind a (f)(set|remove)xattr call to set xattr on directory
+ */
+static int
+dht_dir_common_set_remove_xattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                                fd_t *fd, dict_t *xattr, int flags,
+                                dict_t *xdata, int *op_errno)
+
+{
+    dict_t *xattrop = NULL;
+    int32_t subone[1] = {-1};
+    gf_boolean_t uxattr_key_found = _gf_false;
+    xlator_t *mds_subvol = NULL;
+    xlator_t *travvol = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+    int i = 0;
+    int call_cnt = 0;
+    dht_local_t *local = NULL;
+    char gfid_local[GF_UUID_BUF_SIZE] = {0};
+    char **xattrs_to_heal;
+
+    conf = this->private;
+    local = frame->local;
+    call_cnt = conf->subvolume_cnt;
+    local->flags = flags;
+    xattrs_to_heal = get_xattrs_to_heal();
+
+    if (!gf_uuid_is_null(local->gfid)) {
+        gf_uuid_unparse(local->gfid, gfid_local);
+    }
+
+    if ((local->fop == GF_FOP_SETXATTR) || (local->fop == GF_FOP_FSETXATTR)) {
+        /* Check if any user xattr present in xattr
+         */
+        dict_foreach_fnmatch(xattr, "user*", dht_is_user_xattr,
+                             &uxattr_key_found);
+
+        /* Check if any custom key xattr present in dict xattr
+           and start index from 1 because user xattr already
+           checked in previous line
+        */
+        for (i = 1; xattrs_to_heal[i]; i++)
+            if (dict_get(xattr, xattrs_to_heal[i]))
+                uxattr_key_found = _gf_true;
+    }
+
+    if ((local->fop == GF_FOP_REMOVEXATTR) ||
+        (local->fop == GF_FOP_FREMOVEXATTR)) {
+        /* Check if any custom key xattr present in local->key
+         */
+        for (i = 0; xattrs_to_heal[i]; i++)
+            if (strstr(local->key, xattrs_to_heal[i]))
+                uxattr_key_found = _gf_true;
+    }
+
+    /* If there is no custom key xattr present or gfid is root
+       or call_cnt is 1 then wind a (f)setxattr call on all subvols
+    */
+    if (!uxattr_key_found || __is_root_gfid(local->gfid) || call_cnt == 1) {
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            travvol = conf->subvolumes[i];
+            if ((local->fop == GF_FOP_SETXATTR) ||
+                (local->fop == GF_FOP_FSETXATTR)) {
+                if (fd) {
+                    STACK_WIND_COOKIE(frame, dht_err_cbk, travvol, travvol,
+                                      travvol->fops->fsetxattr, fd, xattr,
+                                      flags, xdata);
+                } else {
+                    STACK_WIND_COOKIE(frame, dht_err_cbk, travvol, travvol,
+                                      travvol->fops->setxattr, loc, xattr,
+                                      flags, xdata);
+                }
+            }
+
+            if ((local->fop == GF_FOP_REMOVEXATTR) ||
+                (local->fop == GF_FOP_FREMOVEXATTR)) {
+                if (fd) {
+                    STACK_WIND_COOKIE(frame, dht_err_cbk, travvol, travvol,
+                                      travvol->fops->fremovexattr, fd,
+                                      local->key, local->xattr_req);
+                } else {
+                    STACK_WIND_COOKIE(frame, dht_err_cbk, travvol, travvol,
+                                      travvol->fops->removexattr, loc,
+                                      local->key, local->xattr_req);
+                }
+            }
+        }
+
+        return 0;
+    }
+
+    /* Calculate hash subvol based on inode and parent inode
+     */
+    if (fd) {
+        ret = dht_inode_ctx_mdsvol_get(fd->inode, this, &mds_subvol);
+    } else {
+        ret = dht_inode_ctx_mdsvol_get(loc->inode, this, &mds_subvol);
+    }
+    if (ret || !mds_subvol) {
+        if (fd) {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+                   "Failed to get mds subvol for fd %p"
+                   "gfid is %s ",
+                   fd, gfid_local);
+        } else {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+                   "%s: Failed to get mds subvol. (gfid is %s)", loc->path,
+                   gfid_local);
+        }
+        (*op_errno) = ENOENT;
+        goto err;
+    }
+
+    local->mds_subvol = mds_subvol;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (conf->subvolumes[i] == mds_subvol) {
+            if (!conf->subvolume_status[i]) {
+                gf_msg(this->name, GF_LOG_WARNING, 0,
+                       DHT_MSG_HASHED_SUBVOL_DOWN,
+                       "MDS subvol is down for path "
+                       " %s gfid is %s Unable to set xattr ",
+                       local->loc.path, gfid_local);
+                (*op_errno) = ENOTCONN;
                 goto err;
+            }
         }
+    }
+
+    if (uxattr_key_found) {
+        xattrop = dict_new();
+        if (!xattrop) {
+            gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_NO_MEMORY, 0,
+                   "dictionary creation failed for path %s "
+                   "for gfid is %s ",
+                   local->loc.path, gfid_local);
+            (*op_errno) = ENOMEM;
+            goto err;
+        }
+        local->xattr = dict_ref(xattr);
+        /* Subtract current MDS xattr value to -1 , value of MDS
+           xattr represents no. of times xattr modification failed
+           on non MDS subvols.
+        */
+        ret = dht_dict_set_array(xattrop, conf->mds_xattr_key, subone, 1);
+        if (ret != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                   "dictionary set array failed for path %s "
+                   "for gfid is %s ",
+                   local->loc.path, gfid_local);
+            if (xattrop)
+                dict_unref(xattrop);
+            (*op_errno) = ret;
+            goto err;
+        }
+        /* Wind a xattrop call to use ref counting approach
+           update mds xattr to -1 before update xattr on
+           hashed subvol and update mds xattr to +1 after update
+           xattr on all non hashed subvol
+        */
+        if (fd) {
+            STACK_WIND(frame, dht_xattrop_mds_cbk, local->mds_subvol,
+                       local->mds_subvol->fops->fxattrop, fd,
+                       GF_XATTROP_ADD_ARRAY, xattrop, NULL);
+        } else {
+            STACK_WIND(frame, dht_xattrop_mds_cbk, local->mds_subvol,
+                       local->mds_subvol->fops->xattrop, loc,
+                       GF_XATTROP_ADD_ARRAY, xattrop, NULL);
+        }
+        if (xattrop)
+            dict_unref(xattrop);
+    }
+
+    return 0;
+err:
+    return -1;
+}
 
-        local->ia_ino = fd->inode->ino;
-	STACK_WIND (frame, dht_readv_cbk,
-		    subvol, subvol->fops->readv,
-		    fd, size, off);
+int
+dht_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xattr,
+              int flags, dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    dht_local_t *local = NULL;
+    int op_errno = EINVAL;
+    dht_conf_t *conf = NULL;
+    dht_layout_t *layout = NULL;
+    int ret = -1;
+    int call_cnt = 0;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+    VALIDATE_OR_GOTO(fd->inode, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+
+    if (!conf->defrag)
+        GF_IF_INTERNAL_XATTR_GOTO(conf->wild_xattr_name, xattr, op_errno, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_FSETXATTR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!layout) {
+        gf_msg_debug(this->name, 0, "no layout for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    local->xattr_req = xdata ? dict_ref(xdata) : dict_new();
+    local->call_cnt = call_cnt = layout->cnt;
+
+    if (IA_ISDIR(fd->inode->ia_type)) {
+        local->hashed_subvol = NULL;
+        ret = dht_dir_common_set_remove_xattr(frame, this, NULL, fd, xattr,
+                                              flags, xdata, &op_errno);
+        if (ret)
+            goto err;
+    } else {
+        local->call_cnt = 1;
+        local->rebalance.xattr = dict_ref(xattr);
+        local->rebalance.flags = flags;
+
+        ret = dict_set_int8(local->xattr_req, DHT_IATT_IN_XDATA_KEY, 1);
+        if (ret) {
+            gf_msg_debug(this->name, 0,
+                         "Failed to set dictionary key %s for fd=%p",
+                         DHT_IATT_IN_XDATA_KEY, fd);
+        }
 
-	return 0;
+        STACK_WIND_COOKIE(frame, dht_file_setxattr_cbk, subvol, subvol,
+                          subvol->fops->fsetxattr, fd, xattr, flags,
+                          local->xattr_req);
+    }
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (readv, frame, -1, op_errno, NULL, 0, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(fsetxattr, frame, -1, op_errno, NULL);
+
+    return 0;
+}
+
+static int
+dht_checking_pathinfo_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int op_ret, int op_errno, dict_t *xattr,
+                          dict_t *xdata)
+{
+    int i = -1;
+    int ret = -1;
+    char *value = NULL;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    xlator_t *prev = NULL;
+    int this_call_cnt = 0;
+
+    local = frame->local;
+    prev = cookie;
+    conf = this->private;
+
+    if (op_ret == -1)
+        goto out;
+
+    ret = dict_get_str(xattr, GF_XATTR_PATHINFO_KEY, &value);
+    if (ret)
+        goto out;
+
+    if (!strcmp(value, local->key)) {
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (conf->subvolumes[i] == prev)
+                conf->decommissioned_bricks[i] = prev;
+        }
+    }
+
+out:
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        DHT_STACK_UNWIND(setxattr, frame, local->op_ret, ENOTSUP, NULL);
+    }
+    return 0;
+}
 
-	return 0;
+static int
+dht_nuke_dir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                 struct iatt *postparent, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(setxattr, frame, op_ret, op_errno, NULL);
+    return 0;
 }
 
+static int
+dht_nuke_dir(call_frame_t *frame, xlator_t *this, loc_t *loc, data_t *tmp)
+{
+    if (!IA_ISDIR(loc->inode->ia_type)) {
+        DHT_STACK_UNWIND(setxattr, frame, -1, ENOTSUP, NULL);
+        return 0;
+    }
+
+    /* Setxattr didn't need the parent, but rmdir does. */
+    loc->parent = inode_parent(loc->inode, NULL, NULL);
+    if (!loc->parent) {
+        DHT_STACK_UNWIND(setxattr, frame, -1, ENOENT, NULL);
+        return 0;
+    }
+    gf_uuid_copy(loc->pargfid, loc->parent->gfid);
+
+    if (!loc->name && loc->path) {
+        loc->name = strrchr(loc->path, '/');
+        if (loc->name) {
+            ++(loc->name);
+        }
+    }
+
+    /*
+     * We do this instead of calling dht_rmdir_do directly for two reasons.
+     * The first is that we want to reuse all of the initialization that
+     * dht_rmdir does, so if it ever changes we'll just follow along.  The
+     * second (i.e. why we don't use STACK_WIND_TAIL) is so that we don't
+     * obscure the fact that we came in via this path instead of a genuine
+     * rmdir.  That makes debugging just a tiny bit easier.
+     */
+    STACK_WIND(frame, dht_nuke_dir_cbk, this, this->fops->rmdir, loc, 1, NULL);
+
+    return 0;
+}
 
 int
-dht_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		int op_ret, int op_errno, struct iatt *prebuf,
-                struct iatt *postbuf)
+dht_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr,
+             int flags, dict_t *xdata)
 {
-        dht_local_t *local = NULL;
+    xlator_t *subvol = NULL;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    dht_methods_t *methods = NULL;
+    dht_layout_t *layout = NULL;
+    int i = 0;
+    int op_errno = EINVAL;
+    int ret = -1;
+    data_t *tmp = NULL;
+    uint32_t dir_spread = 0;
+    char value[4096] = {
+        0,
+    };
+    gf_dht_migrate_data_type_t forced_rebalance = GF_DHT_MIGRATE_DATA;
+    int call_cnt = 0;
+    uint32_t new_hash = 0;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+
+    conf = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, conf, err);
+
+    methods = &(conf->methods);
+
+    /* Rebalance daemon is allowed to set internal keys */
+    if (!conf->defrag)
+        GF_IF_INTERNAL_XATTR_GOTO(conf->wild_xattr_name, xattr, op_errno, err);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_SETXATTR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for path=%s",
+                     loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!layout) {
+        gf_msg_debug(this->name, 0, "no layout for path=%s", loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    local->call_cnt = call_cnt = layout->cnt;
+    tmp = dict_get(xattr, conf->mds_xattr_key);
+    if (tmp) {
+        op_errno = ENOTSUP;
+        goto err;
+    }
+
+    tmp = dict_get(xattr, GF_XATTR_FILE_MIGRATE_KEY);
+    if (tmp) {
+        if (IA_ISDIR(loc->inode->ia_type)) {
+            op_errno = ENOTSUP;
+            goto err;
+        }
 
-        if (op_ret == -1) {
-                goto out;
+        /* TODO: need to interpret the 'value' for more meaning
+           (ie, 'target' subvolume given there, etc) */
+        memcpy(value, tmp->data, tmp->len);
+        if (strcmp(value, "force") == 0)
+            forced_rebalance = GF_DHT_MIGRATE_DATA_EVEN_IF_LINK_EXISTS;
+
+        if (conf->decommission_in_progress)
+            forced_rebalance = GF_DHT_MIGRATE_HARDLINK;
+
+        if (!loc->path) {
+            op_errno = EINVAL;
+            goto err;
         }
 
-        local = frame->local;
-        if (!local) {
-                op_ret = -1;
-                op_errno = EINVAL;
-                goto out;
-        } 
-        
-        prebuf->ia_ino = local->ia_ino;
-        postbuf->ia_ino = local->ia_ino;
+        if (!local->loc.name)
+            local->loc.name = strrchr(local->loc.path, '/') + 1;
 
-out:
-        DHT_STACK_UNWIND (writev, frame, op_ret, op_errno, prebuf, postbuf);
+        if (!local->loc.parent)
+            local->loc.parent = inode_parent(local->loc.inode, NULL, NULL);
+
+        if ((!local->loc.name) || (!local->loc.parent)) {
+            op_errno = EINVAL;
+            goto err;
+        }
+
+        if (gf_uuid_is_null(local->loc.pargfid))
+            gf_uuid_copy(local->loc.pargfid, local->loc.parent->gfid);
+
+        methods->migration_get_dst_subvol(this, local);
 
+        if (!local->rebalance.target_node) {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+                   "Failed to get hashed subvol for %s", loc->path);
+            op_errno = EINVAL;
+            goto err;
+        }
+
+        local->rebalance.from_subvol = local->cached_subvol;
+
+        if (local->rebalance.target_node == local->rebalance.from_subvol) {
+            op_errno = EEXIST;
+            goto err;
+        }
+        if (local->rebalance.target_node) {
+            local->flags = forced_rebalance;
+
+            frame->root->pid = GF_CLIENT_PID_DEFRAG;
+
+            ret = dht_start_rebalance_task(this, frame);
+            if (!ret)
+                return 0;
+
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_REBALANCE_START_FAILED,
+                   "%s: failed to create a new rebalance synctask", loc->path);
+        }
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    tmp = dict_get(xattr, "decommission-brick");
+    if (tmp) {
+        /* This operation should happen only on '/' */
+        if (!__is_root_gfid(loc->inode->gfid)) {
+            op_errno = ENOTSUP;
+            goto err;
+        }
+
+        memcpy(value, tmp->data, min(tmp->len, 4095));
+        local->key = gf_strdup(value);
+        local->call_cnt = conf->subvolume_cnt;
+
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            /* Get the pathinfo, and then compare */
+            STACK_WIND_COOKIE(frame, dht_checking_pathinfo_cbk,
+                              conf->subvolumes[i], conf->subvolumes[i],
+                              conf->subvolumes[i]->fops->getxattr, loc,
+                              GF_XATTR_PATHINFO_KEY, NULL);
+        }
         return 0;
-}
+    }
+
+    tmp = dict_get(xattr, GF_XATTR_FIX_LAYOUT_KEY);
+    if (tmp) {
+        ret = dict_get_uint32(xattr, "new-commit-hash", &new_hash);
+        if (ret == 0) {
+            gf_msg_debug(this->name, 0,
+                         "updating commit hash for %s from %u to %u",
+                         uuid_utoa(loc->gfid), layout->commit_hash, new_hash);
+            layout->commit_hash = new_hash;
+
+            ret = dht_update_commit_hash_for_layout(frame);
+            if (ret) {
+                op_errno = ENOTCONN;
+                goto err;
+            }
+            return ret;
+        }
 
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_FIX_LAYOUT_INFO,
+               "fixing the layout of %s", loc->path);
 
-int
-dht_writev (call_frame_t *frame, xlator_t *this,
-	    fd_t *fd, struct iovec *vector, int count, off_t off,
-            struct iobref *iobref)
-{
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-        dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-        local = dht_local_init (frame);
-        if (!local) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                op_errno = ENOMEM;
+        ret = dht_fix_directory_layout(frame, dht_fix_layout_setxattr_cbk,
+                                       layout);
+        if (ret) {
+            op_errno = ENOTCONN;
+            goto err;
+        }
+        return ret;
+    }
+
+    tmp = dict_get(xattr, "distribute.directory-spread-count");
+    if (tmp) {
+        /* Setxattr value is packed as 'binary', not string */
+        memcpy(value, tmp->data, min(tmp->len, 4095));
+        ret = gf_string2uint32(value, &dir_spread);
+        if (!ret && ((dir_spread <= conf->subvolume_cnt) && (dir_spread > 0))) {
+            layout->spread_cnt = dir_spread;
+
+            ret = dht_fix_directory_layout(frame, dht_common_setxattr_cbk,
+                                           layout);
+            if (ret) {
+                op_errno = ENOTCONN;
                 goto err;
+            }
+            return ret;
         }
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_OPERATION_NOT_SUP,
+               "wrong 'directory-spread-count' value (%s)", value);
+        op_errno = ENOTSUP;
+        goto err;
+    }
+
+    tmp = dict_get(xattr, "glusterfs.dht.nuke");
+    if (tmp) {
+        return dht_nuke_dir(frame, this, loc, tmp);
+    }
+    local->xattr_req = xdata ? dict_ref(xdata) : dict_new();
+
+    if (IA_ISDIR(loc->inode->ia_type)) {
+        local->hashed_subvol = NULL;
+        ret = dht_dir_common_set_remove_xattr(frame, this, loc, NULL, xattr,
+                                              flags, xdata, &op_errno);
+        if (ret)
+            goto err;
+    } else {
+        local->rebalance.xattr = dict_ref(xattr);
+        local->rebalance.flags = flags;
+        local->call_cnt = 1;
 
-        local->ia_ino = fd->inode->ino;
+        ret = dict_set_int8(local->xattr_req, DHT_IATT_IN_XDATA_KEY, 1);
 
-	STACK_WIND (frame, dht_writev_cbk,
-		    subvol, subvol->fops->writev,
-		    fd, vector, count, off, iobref);
+        STACK_WIND_COOKIE(frame, dht_file_setxattr_cbk, subvol, subvol,
+                          subvol->fops->setxattr, loc, xattr, flags,
+                          local->xattr_req);
+    }
 
-	return 0;
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (writev, frame, -1, op_errno, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(setxattr, frame, -1, op_errno, NULL);
 
-	return 0;
+    return 0;
 }
 
+static int
+dht_removexattr2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
+{
+    dht_local_t *local = NULL;
+    int op_errno = EINVAL;
+
+    if (!frame || !frame->local)
+        goto err;
+
+    local = frame->local;
+    op_errno = local->op_errno;
+
+    local->call_cnt = 2; /* This is the second attempt */
+
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        DHT_STACK_UNWIND(removexattr, frame, local->op_ret, local->op_errno,
+                         local->rebalance.xdata);
+        return 0;
+    }
+
+    if (subvol == NULL)
+        goto err;
+
+    if (local->fop == GF_FOP_REMOVEXATTR) {
+        STACK_WIND_COOKIE(frame, dht_file_removexattr_cbk, subvol, subvol,
+                          subvol->fops->removexattr, &local->loc, local->key,
+                          local->xattr_req);
+    } else {
+        STACK_WIND_COOKIE(frame, dht_file_removexattr_cbk, subvol, subvol,
+                          subvol->fops->fremovexattr, local->fd, local->key,
+                          local->xattr_req);
+    }
+
+    return 0;
+
+err:
+    DHT_STACK_UNWIND(removexattr, frame, -1, op_errno, NULL);
+    return 0;
+}
 
 int
-dht_flush (call_frame_t *frame, xlator_t *this, fd_t *fd)
+dht_file_removexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int op_ret, int op_errno, dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
+    int ret = -1;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    struct iatt *stbuf = NULL;
+    inode_t *inode = NULL;
+    xlator_t *subvol1 = NULL, *subvol2 = NULL;
 
+    local = frame->local;
+    prev = cookie;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+    local->op_errno = op_errno;
 
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
+    if ((local->fop == GF_FOP_FREMOVEXATTR) &&
+        dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
+    if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->name);
+        goto out;
+    }
 
-	local->fd = fd_ref (fd);
-	local->call_cnt = 1;
+    if (local->call_cnt != 1)
+        goto out;
 
-	STACK_WIND (frame, dht_err_cbk,
-		    subvol, subvol->fops->flush, fd);
+    ret = dict_get_bin(xdata, DHT_IATT_IN_XDATA_KEY, (void **)&stbuf);
 
-	return 0;
+    if ((!op_ret) && !stbuf) {
+        goto out;
+    }
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (flush, frame, -1, op_errno);
+    local->op_ret = 0;
 
-	return 0;
-}
+    local->rebalance.target_op_fn = dht_removexattr2;
+    if (xdata)
+        local->rebalance.xdata = dict_ref(xdata);
+
+    /* Phase 2 of migration */
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(stbuf)) {
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+    /* Phase 1 of migration */
+    if (IS_DHT_MIGRATION_PHASE1(stbuf)) {
+        inode = (local->fd) ? local->fd->inode : local->loc.inode;
+
+        ret = dht_inode_ctx_get_mig_info(this, inode, &subvol1, &subvol2);
+        if (!dht_mig_info_is_invalid(local->cached_subvol, subvol1, subvol2)) {
+            dht_removexattr2(this, subvol2, frame, 0);
+            return 0;
+        }
+
+        ret = dht_rebalance_in_progress_check(this, frame);
+        if (!ret)
+            return 0;
+    }
 
+out:
+    if (local->fop == GF_FOP_REMOVEXATTR) {
+        DHT_STACK_UNWIND(removexattr, frame, op_ret, op_errno, xdata);
+    } else {
+        DHT_STACK_UNWIND(fremovexattr, frame, op_ret, op_errno, xdata);
+    }
+    return 0;
+}
 
 int
-dht_fsync (call_frame_t *frame, xlator_t *this,
-	   fd_t *fd, int datasync)
+dht_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                const char *key, dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+    int call_cnt = 0;
+    dht_conf_t *conf = NULL;
+    int ret = 0;
+
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+
+    GF_IF_NATIVE_XATTR_GOTO(conf->wild_xattr_name, key, op_errno, err);
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_REMOVEXATTR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for path=%s",
+                     loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!local->layout) {
+        gf_msg_debug(this->name, 0, "no layout for path=%s", loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+    local->xattr_req = (xdata) ? dict_ref(xdata) : dict_new();
+
+    local->call_cnt = call_cnt = layout->cnt;
+    local->key = gf_strdup(key);
+
+    if (key && (strncmp(key, conf->mds_xattr_key, strlen(key)) == 0)) {
+        op_errno = ENOTSUP;
+        goto err;
+    }
+
+    if (IA_ISDIR(loc->inode->ia_type)) {
+        local->hashed_subvol = NULL;
+        ret = dht_dir_common_set_remove_xattr(frame, this, loc, NULL, NULL, 0,
+                                              local->xattr_req, &op_errno);
+        if (ret)
+            goto err;
+
+    } else {
+        local->call_cnt = 1;
+        ret = dict_set_int8(local->xattr_req, DHT_IATT_IN_XDATA_KEY, 1);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_DICT_SET_FAILED,
+                   "Failed to "
+                   "set dictionary key %s for %s",
+                   DHT_IATT_IN_XDATA_KEY, loc->path);
+        }
 
+        STACK_WIND_COOKIE(frame, dht_file_removexattr_cbk, subvol, subvol,
+                          subvol->fops->removexattr, loc, key,
+                          local->xattr_req);
+    }
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+    return 0;
 
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(removexattr, frame, -1, op_errno, NULL);
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-	local->call_cnt = 1;
+    return 0;
+}
 
-        local->ia_ino = fd->inode->ino;
+int
+dht_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *key,
+                 dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+    int call_cnt = 0;
+    dht_conf_t *conf = 0;
+    int ret = 0;
+
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+
+    GF_IF_NATIVE_XATTR_GOTO(conf->wild_xattr_name, key, op_errno, err);
+
+    VALIDATE_OR_GOTO(frame, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_FREMOVEXATTR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for inode=%s",
+                     uuid_utoa(fd->inode->gfid));
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!local->layout) {
+        gf_msg_debug(this->name, 0, "no layout for inode=%s",
+                     uuid_utoa(fd->inode->gfid));
+        op_errno = EINVAL;
+        goto err;
+    }
+    local->xattr_req = xdata ? dict_ref(xdata) : dict_new();
+
+    local->call_cnt = call_cnt = layout->cnt;
+    local->key = gf_strdup(key);
+
+    if (IA_ISDIR(fd->inode->ia_type)) {
+        local->hashed_subvol = NULL;
+        ret = dht_dir_common_set_remove_xattr(frame, this, NULL, fd, NULL, 0,
+                                              local->xattr_req, &op_errno);
+        if (ret)
+            goto err;
+
+    } else {
+        local->call_cnt = 1;
+        ret = dict_set_int8(local->xattr_req, DHT_IATT_IN_XDATA_KEY, 1);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_DICT_SET_FAILED,
+                   "Failed to "
+                   "set dictionary key %s for fd=%p",
+                   DHT_IATT_IN_XDATA_KEY, fd);
+        }
 
-	STACK_WIND (frame, dht_fsync_cbk,
-		    subvol, subvol->fops->fsync,
-		    fd, datasync);
+        STACK_WIND_COOKIE(frame, dht_file_removexattr_cbk, subvol, subvol,
+                          subvol->fops->fremovexattr, fd, key,
+                          local->xattr_req);
+    }
 
-	return 0;
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (fsync, frame, -1, op_errno, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(fremovexattr, frame, -1, op_errno, NULL);
 
-	return 0;
+    return 0;
 }
 
-
 int
-dht_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-	    int op_ret, int op_errno, struct gf_flock *flock)
+dht_fd_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+           int op_errno, fd_t *fd, dict_t *xdata)
 {
-        DHT_STACK_UNWIND (lk, frame, op_ret, op_errno, flock);
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
 
-        return 0;
+    local = frame->local;
+    prev = cookie;
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+            UNLOCK(&frame->lock);
+            gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                         prev->name);
+            goto post_unlock;
+        }
+
+        local->op_ret = 0;
+    }
+    UNLOCK(&frame->lock);
+post_unlock:
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt))
+        DHT_STACK_UNWIND(open, frame, local->op_ret, local->op_errno, local->fd,
+                         NULL);
+
+    return 0;
+}
+
+/*
+ * dht_normalize_stats -
+ */
+static void
+dht_normalize_stats(struct statvfs *buf, unsigned long bsize,
+                    unsigned long frsize)
+{
+    double factor = 0;
+
+    if (buf->f_bsize != bsize) {
+        buf->f_bsize = bsize;
+    }
+
+    if (buf->f_frsize != frsize) {
+        factor = ((double)buf->f_frsize) / frsize;
+        buf->f_frsize = frsize;
+        buf->f_blocks = (fsblkcnt_t)(factor * buf->f_blocks);
+        buf->f_bfree = (fsblkcnt_t)(factor * buf->f_bfree);
+        buf->f_bavail = (fsblkcnt_t)(factor * buf->f_bavail);
+    }
 }
 
+static int
+dht_statfs_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, struct statvfs *statvfs, dict_t *xdata)
+{
+    gf_boolean_t event = _gf_false;
+    qdstatfs_action_t action = qdstatfs_action_OFF;
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    int bsize = 0;
+    int frsize = 0;
+    GF_UNUSED int ret = 0;
+    unsigned long new_usage = 0;
+    unsigned long cur_usage = 0;
+
+    local = frame->local;
+    GF_ASSERT(local);
+
+    if (xdata)
+        ret = dict_get_int8(xdata, "quota-deem-statfs", (int8_t *)&event);
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+            goto unlock;
+        }
+        if (!statvfs) {
+            op_errno = EINVAL;
+            local->op_ret = -1;
+            goto unlock;
+        }
+        local->op_ret = 0;
+
+        if (local->quota_deem_statfs) {
+            if (event == _gf_true) {
+                action = qdstatfs_action_COMPARE;
+            } else {
+                action = qdstatfs_action_NEGLECT;
+            }
+        } else {
+            if (event == _gf_true) {
+                action = qdstatfs_action_REPLACE;
+                local->quota_deem_statfs = _gf_true;
+            }
+        }
+
+        if (local->quota_deem_statfs) {
+            switch (action) {
+                case qdstatfs_action_NEGLECT:
+                    goto unlock;
+
+                case qdstatfs_action_REPLACE:
+                    local->statvfs = *statvfs;
+                    goto unlock;
+
+                case qdstatfs_action_COMPARE:
+                    new_usage = statvfs->f_blocks - statvfs->f_bfree;
+                    cur_usage = local->statvfs.f_blocks -
+                                local->statvfs.f_bfree;
+
+                    /* Take the max of the usage from subvols */
+                    if (new_usage >= cur_usage)
+                        local->statvfs = *statvfs;
+                    goto unlock;
+
+                default:
+                    break;
+            }
+        }
+
+        if (local->statvfs.f_bsize != 0) {
+            bsize = max(local->statvfs.f_bsize, statvfs->f_bsize);
+            frsize = max(local->statvfs.f_frsize, statvfs->f_frsize);
+            dht_normalize_stats(&local->statvfs, bsize, frsize);
+            dht_normalize_stats(statvfs, bsize, frsize);
+        } else {
+            local->statvfs.f_bsize = statvfs->f_bsize;
+            local->statvfs.f_frsize = statvfs->f_frsize;
+        }
+
+        local->statvfs.f_blocks += statvfs->f_blocks;
+        local->statvfs.f_bfree += statvfs->f_bfree;
+        local->statvfs.f_bavail += statvfs->f_bavail;
+        local->statvfs.f_files += statvfs->f_files;
+        local->statvfs.f_ffree += statvfs->f_ffree;
+        local->statvfs.f_favail += statvfs->f_favail;
+        local->statvfs.f_fsid = statvfs->f_fsid;
+        local->statvfs.f_flag = statvfs->f_flag;
+        local->statvfs.f_namemax = statvfs->f_namemax;
+    }
+unlock:
+    UNLOCK(&frame->lock);
+
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt))
+        DHT_STACK_UNWIND(statfs, frame, local->op_ret, local->op_errno,
+                         &local->statvfs, xdata);
+
+    return 0;
+}
 
 int
-dht_lk (call_frame_t *frame, xlator_t *this,
-	fd_t *fd, int cmd, struct gf_flock *flock)
+dht_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int op_errno = -1;
+    int i = -1;
+    inode_t *inode = NULL;
+    inode_table_t *itable = NULL;
+    static uuid_t root_gfid = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1};
+    loc_t newloc = {
+        0,
+    };
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+
+    local = dht_local_init(frame, NULL, NULL, GF_FOP_STATFS);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    if (loc->inode && !IA_ISDIR(loc->inode->ia_type)) {
+        itable = loc->inode->table;
+        if (!itable) {
+            op_errno = EINVAL;
+            goto err;
+        }
 
+        loc = &local->loc2;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+        inode = inode_find(itable, root_gfid);
+        if (!inode) {
+            op_errno = EINVAL;
+            goto err;
+        }
 
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
+        dht_build_root_loc(inode, &newloc);
+        loc = &newloc;
+    }
 
-	STACK_WIND (frame, dht_lk_cbk,
-		    subvol, subvol->fops->lk,
-		    fd, cmd, flock);
+    local->call_cnt = conf->subvolume_cnt;
 
-	return 0;
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        STACK_WIND(frame, dht_statfs_cbk, conf->subvolumes[i],
+                   conf->subvolumes[i]->fops->statfs, loc, xdata);
+    }
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (lk, frame, -1, op_errno, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(statfs, frame, -1, op_errno, NULL, NULL);
 
-	return 0;
+    return 0;
 }
 
-
 int
-dht_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		int op_ret, int op_errno, struct statvfs *statvfs)
-{
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = 0;
-
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			goto unlock;
-		}
-		local->op_ret = 0;
-
-		/* TODO: normalize sizes */
-		local->statvfs.f_bsize    = statvfs->f_bsize;
-		local->statvfs.f_frsize   = statvfs->f_frsize;
-
-		local->statvfs.f_blocks  += statvfs->f_blocks;
-		local->statvfs.f_bfree   += statvfs->f_bfree;
-		local->statvfs.f_bavail  += statvfs->f_bavail;
-		local->statvfs.f_files   += statvfs->f_files;
-		local->statvfs.f_ffree   += statvfs->f_ffree;
-		local->statvfs.f_favail  += statvfs->f_favail;
-		local->statvfs.f_fsid     = statvfs->f_fsid;
-		local->statvfs.f_flag     = statvfs->f_flag;
-		local->statvfs.f_namemax  = statvfs->f_namemax;
-
-	}
-unlock:
-	UNLOCK (&frame->lock);
+dht_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+            dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int op_errno = -1;
+    int i = -1;
+    int ret = 0;
+    gf_boolean_t new_xdata = _gf_false;
+    xlator_t **subvolumes = NULL;
+    int call_count = 0;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+
+    local = dht_local_init(frame, loc, fd, GF_FOP_OPENDIR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+    local->first_up_subvol = dht_first_up_subvol(this);
+
+    if (!xdata) {
+        xdata = dict_new();
+        if (!xdata) {
+            op_errno = ENOMEM;
+            goto err;
+        }
+        new_xdata = _gf_true;
+    }
+
+    ret = dict_set_uint32(xdata, conf->link_xattr_name, 256);
+    if (ret)
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+               "Failed to set dictionary value : key = %s",
+               conf->link_xattr_name);
+
+    /* dht_readdirp will wind to all subvols so open has to be sent to
+     * all subvols whether or not conf->local_subvols is set */
+
+    call_count = local->call_cnt = conf->subvolume_cnt;
+    subvolumes = conf->subvolumes;
+
+    /* In case of parallel-readdir, the readdir-ahead will be loaded
+     * below dht, in this case, if we want to enable or disable SKIP_DIRs
+     * it has to be done in opendir, so that prefetching logic in
+     * readdir-ahead, honors it */
+    for (i = 0; i < call_count; i++) {
+        if (conf->readdir_optimize == _gf_true) {
+            if (subvolumes[i] != local->first_up_subvol) {
+                ret = dict_set_int32(xdata, GF_READDIR_SKIP_DIRS, 1);
+                if (ret)
+                    gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                           "Failed to set dictionary"
+                           " value :key = %s, ret:%d",
+                           GF_READDIR_SKIP_DIRS, ret);
+            }
+        }
 
+        STACK_WIND_COOKIE(frame, dht_fd_cbk, subvolumes[i], subvolumes[i],
+                          subvolumes[i]->fops->opendir, loc, fd, xdata);
+        dict_del(xdata, GF_READDIR_SKIP_DIRS);
+    }
 
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt))
-		DHT_STACK_UNWIND (statfs, frame, local->op_ret, local->op_errno,
-				  &local->statvfs);
+    if (new_xdata)
+        dict_unref(xdata);
 
-        return 0;
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(opendir, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
 }
 
+/* dht_readdirp_cbk creates a new dentry and dentry->inode is not assigned.
+   This functions assigns an inode if all of the following conditions are
+   true:
 
-int
-dht_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc)
+   * DHT has only one child. In this case the entire layout is present on
+   this single child and hence we can set complete layout in inode.
+   * backend has complete layout and there are no anomalies in it and from
+   this information layout can be constructed and set in inode.
+*/
+
+static void
+dht_populate_inode_for_dentry(xlator_t *this, xlator_t *subvol,
+                              gf_dirent_t *entry, gf_dirent_t *orig_entry)
 {
-	dht_local_t  *local  = NULL;
-	dht_conf_t   *conf = NULL;
-        int           op_errno = -1;
-	int           i = -1;
+    dht_layout_t *layout = NULL;
+    int ret = 0;
+    loc_t loc = {
+        0,
+    };
+
+    if (gf_uuid_is_null(orig_entry->d_stat.ia_gfid)) {
+        /* this skips the '..' entry for the root of the volume */
+        return;
+    }
+
+    gf_uuid_copy(loc.gfid, orig_entry->d_stat.ia_gfid);
+    loc.inode = inode_ref(orig_entry->inode);
+
+    if (is_revalidate(&loc)) {
+        goto out;
+    }
+
+    layout = dht_layout_new(this, 1);
+    if (!layout)
+        goto out;
+
+    ret = dht_layout_merge(this, layout, subvol, 0, 0, orig_entry->dict);
+    if (!ret) {
+        ret = dht_layout_normalize(this, &loc, layout);
+        if (ret == 0) {
+            dht_layout_set(this, orig_entry->inode, layout);
+            entry->inode = inode_ref(orig_entry->inode);
+            layout = NULL;
+        }
+    }
 
+    if (layout)
+        dht_layout_unref(this, layout);
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (this->private, err);
+out:
+    loc_wipe(&loc);
+    return;
+}
 
-	conf = this->private;
+/* Posix returns op_errno = ENOENT to indicate that there are no more
+ * entries
+ */
+static int
+dht_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, gf_dirent_t *orig_entries, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    gf_dirent_t entries;
+    gf_dirent_t *orig_entry = NULL;
+    gf_dirent_t *entry = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *next_subvol = NULL;
+    off_t next_offset = 0;
+    int count = 0;
+    dht_layout_t *layout = NULL;
+    dht_conf_t *conf = NULL;
+    dht_methods_t *methods = NULL;
+    xlator_t *subvol = 0;
+    xlator_t *hashed_subvol = 0;
+    int ret = 0;
+    int readdir_optimize = 0;
+    inode_table_t *itable = NULL;
+    inode_t *inode = NULL;
+    gf_boolean_t skip_hashed_check = _gf_false;
+
+    INIT_LIST_HEAD(&entries.list);
+
+    prev = cookie;
+    local = frame->local;
+    GF_VALIDATE_OR_GOTO(this->name, local->fd, unwind);
+
+    itable = local->fd->inode->table;
+
+    conf = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, conf, unwind);
+
+    methods = &(conf->methods);
+
+    if (op_ret <= 0) {
+        goto done;
+    }
+
+    /* Why aren't we skipping DHT entirely in case of a single subvol?
+     * Because if this was a larger volume earlier and all but one subvol
+     * was removed, there might be stale linkto files on the subvol.
+     */
+    if (conf->subvolume_cnt == 1) {
+        /* return all directory and file entries except
+         * linkto files for a single child DHT
+         */
+        skip_hashed_check = _gf_true;
+    }
 
-	local = dht_local_init (frame);
-	local->call_cnt = conf->subvolume_cnt;
+    if (!local->layout)
+        local->layout = dht_layout_get(this, local->fd->inode);
 
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		STACK_WIND (frame, dht_statfs_cbk,
-			    conf->subvolumes[i],
-			    conf->subvolumes[i]->fops->statfs, loc);
-	}
+    layout = local->layout;
 
-	return 0;
+    /* This will skip the entries on the subvol without a layout,
+     * hence preventing the crash but rmdir might fail with
+     * "directory not empty" errors*/
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (statfs, frame, -1, op_errno, NULL);
+    if (layout == NULL)
+        goto done;
 
-	return 0;
-}
+    if (conf->readdir_optimize == _gf_true)
+        readdir_optimize = 1;
 
+    gf_msg_debug(this->name, 0, "Processing entries from %s", prev->name);
 
-int
-dht_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd)
-{
-	dht_local_t  *local  = NULL;
-	dht_conf_t   *conf = NULL;
-	int           ret = -1;
-        int           op_errno = -1;
-	int           i = -1;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (this->private, err);
-
-	conf = this->private;
-
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	local->fd = fd_ref (fd);
-	ret = loc_dup (loc, &local->loc);
-	if (ret == -1) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	local->call_cnt = conf->subvolume_cnt;
-
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		STACK_WIND (frame, dht_fd_cbk,
-			    conf->subvolumes[i],
-			    conf->subvolumes[i]->fops->opendir,
-			    loc, fd);
-	}
-
-	return 0;
+    list_for_each_entry(orig_entry, (&orig_entries->list), list)
+    {
+        next_offset = orig_entry->d_off;
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (opendir, frame, -1, op_errno, NULL);
+        gf_msg_debug(this->name, 0, "%s: entry = %s, type = %d", prev->name,
+                     orig_entry->d_name, orig_entry->d_type);
 
-	return 0;
-}
+        if (IA_ISINVAL(orig_entry->d_stat.ia_type)) {
+            /*stat failed somewhere- display this entry but the data may
+             * be inaccurate.
+             */
+            gf_msg_debug(this->name, EINVAL, "Invalid stat for %s (gfid %s)",
+                         orig_entry->d_name,
+                         uuid_utoa(orig_entry->d_stat.ia_gfid));
+        }
 
+        if (check_is_linkfile(NULL, (&orig_entry->d_stat), orig_entry->dict,
+                              conf->link_xattr_name)) {
+            gf_msg_debug(this->name, 0, "%s: %s is a linkto file", prev->name,
+                         orig_entry->d_name);
+            continue;
+        }
 
-int
-dht_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
-                  int op_errno, gf_dirent_t *orig_entries)
-{
-	dht_local_t  *local = NULL;
-	gf_dirent_t   entries;
-	gf_dirent_t  *orig_entry = NULL;
-	gf_dirent_t  *entry = NULL;
-	call_frame_t *prev = NULL;
-	xlator_t     *next_subvol = NULL;
-        off_t         next_offset = 0;
-	int           count = 0;
-        dht_layout_t *layout = 0;
-        dht_conf_t   *conf   = NULL;
-        xlator_t     *subvol = 0;
-
-	INIT_LIST_HEAD (&entries.list);
-	prev = cookie;
-	local = frame->local;
-	conf  = this->private;
-
-	if (op_ret < 0)
-		goto done;
-
-        if (!local->layout)
-                local->layout = dht_layout_get (this, local->fd->inode);
+        if (skip_hashed_check) {
+            goto list;
+        }
 
-        layout = local->layout;
+        if (check_is_dir(NULL, (&orig_entry->d_stat), NULL)) {
+            /*Directory entries filtering :
+             * a) If rebalance is running, pick from first_up_subvol
+             * b) (rebalance not running)hashed subvolume is NULL or
+             * down then filter in first_up_subvolume. Other wise the
+             * corresponding hashed subvolume will take care of the
+             * directory entry.
+             */
+            if (readdir_optimize) {
+                if (prev == local->first_up_subvol)
+                    goto list;
+                else
+                    continue;
+            }
+
+            hashed_subvol = methods->layout_search(this, layout,
+                                                   orig_entry->d_name);
+
+            if (prev == hashed_subvol)
+                goto list;
+            if ((hashed_subvol && dht_subvol_status(conf, hashed_subvol)) ||
+                (prev != local->first_up_subvol))
+                continue;
+
+            goto list;
+        }
 
-	list_for_each_entry (orig_entry, (&orig_entries->list), list) {
-                next_offset = orig_entry->d_off;
+    list:
+        entry = gf_dirent_for_name(orig_entry->d_name);
+        if (!entry) {
+            goto unwind;
+        }
 
-                if (check_is_linkfile (NULL, (&orig_entry->d_stat), NULL)
-                    || (check_is_dir (NULL, (&orig_entry->d_stat), NULL)
-                        && (prev->this != dht_first_up_subvol (this)))) {
-                        continue;
-                }
+        /* Do this if conf->search_unhashed is set to "auto" */
+        if (conf->search_unhashed == GF_DHT_LOOKUP_UNHASHED_AUTO) {
+            subvol = methods->layout_search(this, layout, orig_entry->d_name);
+            if (!subvol || (subvol != prev)) {
+                /* TODO: Count the number of entries which need
+                   linkfile to prove its existence in fs */
+                layout->search_unhashed++;
+            }
+        }
 
-                entry = gf_dirent_for_name (orig_entry->d_name);
-                if (!entry) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Out of memory");
-                        goto unwind;
+        entry->d_off = orig_entry->d_off;
+        entry->d_stat = orig_entry->d_stat;
+        entry->d_ino = orig_entry->d_ino;
+        entry->d_type = orig_entry->d_type;
+        entry->d_len = orig_entry->d_len;
+
+        if (orig_entry->dict)
+            entry->dict = dict_ref(orig_entry->dict);
+
+        /* making sure we set the inode ctx right with layout,
+           currently possible only for non-directories, so for
+           directories don't set entry inodes */
+        if (IA_ISDIR(entry->d_stat.ia_type)) {
+            entry->d_stat.ia_blocks = DHT_DIR_STAT_BLOCKS;
+            entry->d_stat.ia_size = DHT_DIR_STAT_SIZE;
+            if (orig_entry->inode) {
+                dht_inode_ctx_time_update(orig_entry->inode, this,
+                                          &entry->d_stat, 1);
+
+                if (conf->subvolume_cnt == 1) {
+                    dht_populate_inode_for_dentry(this, prev, entry,
+                                                  orig_entry);
                 }
-
-                /* Do this if conf->search_unhashed is set to "auto" */
-                if (conf->search_unhashed == GF_DHT_LOOKUP_UNHASHED_AUTO) {
-                        subvol = dht_layout_search (this, layout,
-                                                    orig_entry->d_name);
-                        if (!subvol || (subvol != prev->this)) {
-                                /* TODO: Count the number of entries which need
-                                   linkfile to prove its existance in fs */
-                                layout->search_unhashed++;
-                        }
+            }
+        } else {
+            if (orig_entry->dict &&
+                dict_get(orig_entry->dict, conf->link_xattr_name)) {
+                /* Strip out the S and T flags set by rebalance*/
+                DHT_STRIP_PHASE1_FLAGS(&entry->d_stat);
+            }
+
+            if (orig_entry->inode) {
+                ret = dht_layout_preset(this, prev, orig_entry->inode);
+                if (ret)
+                    gf_msg(this->name, GF_LOG_WARNING, 0,
+                           DHT_MSG_LAYOUT_SET_FAILED,
+                           "failed to link the layout "
+                           "in inode for %s",
+                           orig_entry->d_name);
+
+                entry->inode = inode_ref(orig_entry->inode);
+            } else if (itable) {
+                /*
+                 * orig_entry->inode might be null if any upper
+                 * layer xlators below client set to null, to
+                 * force a lookup on the inode even if the inode
+                 * is present in the inode table. In that case
+                 * we just update the ctx to make sure we didn't
+                 * missed anything.
+                 */
+                inode = inode_find(itable, orig_entry->d_stat.ia_gfid);
+                if (inode) {
+                    ret = dht_layout_preset(this, prev, inode);
+                    if (ret)
+                        gf_msg(this->name, GF_LOG_WARNING, 0,
+                               DHT_MSG_LAYOUT_SET_FAILED,
+                               "failed to link the layout"
+                               " in inode for %s",
+                               orig_entry->d_name);
+                    inode_unref(inode);
+                    inode = NULL;
                 }
-                entry->d_stat = orig_entry->d_stat;
-
-                dht_itransform (this, prev->this, orig_entry->d_ino,
-                                &entry->d_ino);
-                dht_itransform (this, prev->this, orig_entry->d_off,
-                                &entry->d_off);
-
-                entry->d_stat.ia_ino = entry->d_ino;
-                entry->d_type = orig_entry->d_type;
-                entry->d_len  = orig_entry->d_len;
-
-                list_add_tail (&entry->list, &entries.list);
-                count++;
-	}
-	op_ret = count;
-        /* We need to ensure that only the last subvolume's end-of-directory
-         * notification is respected so that directory reading does not stop
-         * before all subvolumes have been read. That could happen because the
-         * posix for each subvolume sends a ENOENT on end-of-directory but in
-         * distribute we're not concerned only with a posix's view of the
-         * directory but the aggregated namespace' view of the directory.
-         */
-        if (prev->this != dht_last_up_subvol (this))
-                op_errno = 0;
+            }
+        }
 
-done:
-	if (count == 0) {
-                /* non-zero next_offset means that
-                   EOF is not yet hit on the current subvol
-                */
-                if (next_offset == 0) {
-                        next_subvol = dht_subvol_next (this, prev->this);
-                } else {
-                        next_subvol = prev->this;
-                }
+        gf_msg_debug(this->name, 0, "%s: Adding entry = %s", prev->name,
+                     entry->d_name);
 
-		if (!next_subvol) {
-			goto unwind;
-		}
+        list_add_tail(&entry->list, &entries.list);
+        count++;
+    }
 
-		STACK_WIND (frame, dht_readdirp_cbk,
-			    next_subvol, next_subvol->fops->readdirp,
-			    local->fd, local->size, next_offset);
-		return 0;
-	}
+done:
 
-unwind:
-	if (op_ret < 0)
-		op_ret = 0;
+    /* We need to ensure that only the last subvolume's end-of-directory
+     * notification is respected so that directory reading does not stop
+     * before all subvolumes have been read. That could happen because the
+     * posix for each subvolume sends a ENOENT on end-of-directory but in
+     * distribute we're not concerned only with a posix's view of the
+     * directory but the aggregated namespace' view of the directory.
+     * Possible values:
+     * op_ret == 0 and op_errno != 0
+     *   if op_errno != ENOENT : Error.Unwind.
+     *   if op_errno == ENOENT : There are no more entries on this subvol.
+     *                           Move to the next one.
+     * op_ret > 0 and count == 0 :
+     *    The subvol returned entries to dht but all were stripped out.
+     *    For example, if they were linkto files or dirs where
+     *    hashed_subvol != prev. Try to get some entries by winding
+     *    to the next subvol. This can be dangerous if parallel readdir
+     *    is enabled as it grows the stack.
+     *
+     * op_ret > 0 and count > 0:
+     *   We found some entries. Unwind even if the buffer is not full.
+     *
+     */
+
+    op_ret = count;
+    if (count == 0) {
+        /* non-zero next_offset means that
+         * EOF is not yet hit on the current subvol
+         */
+        if ((next_offset == 0) || (op_errno == ENOENT)) {
+            next_offset = 0;
+            next_subvol = dht_subvol_next(this, prev);
+        } else {
+            next_subvol = prev;
+        }
 
-	DHT_STACK_UNWIND (readdirp, frame, op_ret, op_errno, &entries);
+        if (!next_subvol) {
+            goto unwind;
+        }
 
-	gf_dirent_free (&entries);
+        if (conf->readdir_optimize == _gf_true) {
+            if (next_subvol != local->first_up_subvol) {
+                ret = dict_set_int32(local->xattr, GF_READDIR_SKIP_DIRS, 1);
+                if (ret)
+                    gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                           "Failed to set dictionary value"
+                           ":key = %s",
+                           GF_READDIR_SKIP_DIRS);
+            } else {
+                dict_del(local->xattr, GF_READDIR_SKIP_DIRS);
+            }
+        }
 
+        STACK_WIND_COOKIE(frame, dht_readdirp_cbk, next_subvol, next_subvol,
+                          next_subvol->fops->readdirp, local->fd, local->size,
+                          next_offset, local->xattr);
         return 0;
+    }
+
+unwind:
+    /* We need to ensure that only the last subvolume's end-of-directory
+     * notification is respected so that directory reading does not stop
+     * before all subvolumes have been read. That could happen because the
+     * posix for each subvolume sends a ENOENT on end-of-directory but in
+     * distribute we're not concerned only with a posix's view of the
+     * directory but the aggregated namespace' view of the directory.
+     */
+    if (op_ret < 0)
+        op_ret = 0;
+
+    if (prev != dht_last_up_subvol(this))
+        op_errno = 0;
+
+    DHT_STACK_UNWIND(readdirp, frame, op_ret, op_errno, &entries, NULL);
+
+    gf_dirent_free(&entries);
+    return 0;
 }
 
+static int
+dht_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                int op_errno, gf_dirent_t *orig_entries, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    gf_dirent_t entries;
+    gf_dirent_t *orig_entry = NULL;
+    gf_dirent_t *entry = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *next_subvol = NULL;
+    off_t next_offset = 0;
+    int count = 0;
+    dht_layout_t *layout = 0;
+    xlator_t *subvol = 0;
+    dht_conf_t *conf = NULL;
+    dht_methods_t *methods = NULL;
+    gf_boolean_t skip_hashed_check = _gf_false;
 
+    INIT_LIST_HEAD(&entries.list);
 
-int
-dht_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		 int op_ret, int op_errno, gf_dirent_t *orig_entries)
-{
-	dht_local_t  *local = NULL;
-	gf_dirent_t   entries;
-	gf_dirent_t  *orig_entry = NULL;
-	gf_dirent_t  *entry = NULL;
-	call_frame_t *prev = NULL;
-	xlator_t     *next_subvol = NULL;
-        off_t         next_offset = 0;
-	int           count = 0;
-        dht_layout_t *layout = 0;
-        dht_conf_t   *conf   = NULL;
-        xlator_t     *subvol = 0;
-
-	INIT_LIST_HEAD (&entries.list);
-	prev = cookie;
-	local = frame->local;
-	conf  = this->private;
-
-	if (op_ret < 0)
-		goto done;
-
-        if (!local->layout)
-                local->layout = dht_layout_get (this, local->fd->inode);
+    prev = cookie;
+    local = frame->local;
 
-        layout = local->layout;
+    conf = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, conf, done);
 
-	list_for_each_entry (orig_entry, (&orig_entries->list), list) {
-                next_offset = orig_entry->d_off;
+    methods = &(conf->methods);
 
-                subvol = dht_layout_search (this, layout, orig_entry->d_name);
+    if (op_ret <= 0)
+        goto done;
 
-                if (!subvol || (subvol == prev->this)) {
-                        entry = gf_dirent_for_name (orig_entry->d_name);
-                        if (!entry) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "memory allocation failed :(");
-                                goto unwind;
-                        }
+    if (!local->layout)
+        local->layout = dht_layout_get(this, local->fd->inode);
 
-                        dht_itransform (this, prev->this, orig_entry->d_ino,
-                                        &entry->d_ino);
-                        dht_itransform (this, prev->this, orig_entry->d_off,
-                                        &entry->d_off);
+    layout = local->layout;
 
-                        entry->d_type = orig_entry->d_type;
-                        entry->d_len  = orig_entry->d_len;
+    gf_msg_debug(this->name, 0, "Processing entries from %s", prev->name);
 
-                        list_add_tail (&entry->list, &entries.list);
-                        count++;
-                }
-        }
-	op_ret = count;
-        /* We need to ensure that only the last subvolume's end-of-directory
-         * notification is respected so that directory reading does not stop
-         * before all subvolumes have been read. That could happen because the
-         * posix for each subvolume sends a ENOENT on end-of-directory but in
-         * distribute we're not concerned only with a posix's view of the
-         * directory but the aggregated namespace' view of the directory.
-         */
-        if (prev->this != dht_last_up_subvol (this))
-                op_errno = 0;
+    if (conf->subvolume_cnt == 1) {
+        /*return everything*/
+        skip_hashed_check = _gf_true;
+        count = op_ret;
+        goto done;
+    }
+
+    list_for_each_entry(orig_entry, (&orig_entries->list), list)
+    {
+        next_offset = orig_entry->d_off;
+
+        gf_msg_debug(this->name, 0, "%s: entry = %s, type = %d", prev->name,
+                     orig_entry->d_name, orig_entry->d_type);
+
+        subvol = methods->layout_search(this, layout, orig_entry->d_name);
+
+        if (!subvol || (subvol == prev)) {
+            entry = gf_dirent_for_name(orig_entry->d_name);
+            if (!entry) {
+                gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+                       "Memory allocation failed ");
+                goto unwind;
+            }
 
+            entry->d_off = orig_entry->d_off;
+            entry->d_ino = orig_entry->d_ino;
+            entry->d_type = orig_entry->d_type;
+            entry->d_len = orig_entry->d_len;
+
+            gf_msg_debug(this->name, 0, "%s: Adding = entry %s", prev->name,
+                         entry->d_name);
+
+            list_add_tail(&entry->list, &entries.list);
+            count++;
+        }
+    }
 done:
-	if (count == 0) {
-                /* non-zero next_offset means that
-                   EOF is not yet hit on the current subvol
-                */
-                if (next_offset == 0) {
-                        next_subvol = dht_subvol_next (this, prev->this);
-                } else {
-                        next_subvol = prev->this;
-                }
+    op_ret = count;
+    /* We need to ensure that only the last subvolume's end-of-directory
+     * notification is respected so that directory reading does not stop
+     * before all subvolumes have been read. That could happen because the
+     * posix for each subvolume sends a ENOENT on end-of-directory but in
+     * distribute we're not concerned only with a posix's view of the
+     * directory but the aggregated namespace' view of the directory.
+     */
+    if (count == 0) {
+        if ((next_offset == 0) || (op_errno == ENOENT)) {
+            next_offset = 0;
+            next_subvol = dht_subvol_next(this, prev);
+        } else {
+            next_subvol = prev;
+        }
 
-		if (!next_subvol) {
-			goto unwind;
-		}
+        if (!next_subvol) {
+            goto unwind;
+        }
 
-		STACK_WIND (frame, dht_readdir_cbk,
-			    next_subvol, next_subvol->fops->readdir,
-			    local->fd, local->size, next_offset);
-		return 0;
-	}
+        STACK_WIND_COOKIE(frame, dht_readdir_cbk, next_subvol, next_subvol,
+                          next_subvol->fops->readdir, local->fd, local->size,
+                          next_offset, NULL);
+        return 0;
+    }
 
 unwind:
-	if (op_ret < 0)
-		op_ret = 0;
+    /* We need to ensure that only the last subvolume's end-of-directory
+     * notification is respected so that directory reading does not stop
+     * before all subvolumes have been read. That could happen because the
+     * posix for each subvolume sends a ENOENT on end-of-directory but in
+     * distribute we're not concerned only with a posix's view of the
+     * directory but the aggregated namespace' view of the directory.
+     */
+
+    if (prev != dht_last_up_subvol(this))
+        op_errno = 0;
+
+    if (!skip_hashed_check) {
+        DHT_STACK_UNWIND(readdir, frame, op_ret, op_errno, &entries, NULL);
+        gf_dirent_free(&entries);
+
+    } else {
+        DHT_STACK_UNWIND(readdir, frame, op_ret, op_errno, orig_entries, NULL);
+    }
+    return 0;
+}
+
+static int
+dht_do_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+               off_t yoff, int whichop, dict_t *dict)
+{
+    dht_local_t *local = NULL;
+    int op_errno = -1;
+    xlator_t *xvol = NULL;
+    int ret = 0;
+    dht_conf_t *conf = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+
+    local = dht_local_init(frame, NULL, NULL, whichop);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->fd = fd_ref(fd);
+    local->size = size;
+    local->xattr_req = (dict) ? dict_ref(dict) : NULL;
+    local->first_up_subvol = dht_first_up_subvol(this);
+    local->op_ret = -1;
+
+    dht_deitransform(this, yoff, &xvol);
+
+    /* TODO: do proper readdir */
+    if (whichop == GF_FOP_READDIRP) {
+        if (dict)
+            local->xattr = dict_ref(dict);
+        else
+            local->xattr = dict_new();
+
+        if (local->xattr) {
+            ret = dict_set_uint32(local->xattr, conf->link_xattr_name, 256);
+            if (ret)
+                gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+                       "Failed to set dictionary value"
+                       " : key = %s",
+                       conf->link_xattr_name);
+
+            if (conf->readdir_optimize == _gf_true) {
+                if (xvol != local->first_up_subvol) {
+                    ret = dict_set_int32(local->xattr, GF_READDIR_SKIP_DIRS, 1);
+                    if (ret)
+                        gf_msg(this->name, GF_LOG_ERROR, 0,
+                               DHT_MSG_DICT_SET_FAILED,
+                               "Failed to set "
+                               "dictionary value: "
+                               "key = %s",
+                               GF_READDIR_SKIP_DIRS);
+                } else {
+                    dict_del(local->xattr, GF_READDIR_SKIP_DIRS);
+                }
+            }
+
+            if (conf->subvolume_cnt == 1) {
+                ret = dict_set_uint32(local->xattr, conf->xattr_name, 4 * 4);
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_WARNING, ENOMEM,
+                           DHT_MSG_DICT_SET_FAILED,
+                           "Failed to set dictionary "
+                           "value:key = %s ",
+                           conf->xattr_name);
+                }
+            }
+        }
 
-	DHT_STACK_UNWIND (readdir, frame, op_ret, op_errno, &entries);
+        STACK_WIND_COOKIE(frame, dht_readdirp_cbk, xvol, xvol,
+                          xvol->fops->readdirp, fd, size, yoff, local->xattr);
+    } else {
+        STACK_WIND_COOKIE(frame, dht_readdir_cbk, xvol, xvol,
+                          xvol->fops->readdir, fd, size, yoff, local->xattr);
+    }
 
-	gf_dirent_free (&entries);
+    return 0;
 
-        return 0;
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(readdir, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
 }
 
+int
+dht_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t yoff, dict_t *xdata)
+{
+    int op = GF_FOP_READDIR;
+    dht_conf_t *conf = NULL;
+    int i = 0;
+
+    conf = this->private;
+    if (!conf)
+        goto out;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (!conf->subvolume_status[i]) {
+            op = GF_FOP_READDIRP;
+            break;
+        }
+    }
+
+    if (conf->use_readdirp)
+        op = GF_FOP_READDIRP;
+
+out:
+    dht_do_readdir(frame, this, fd, size, yoff, op, 0);
+    return 0;
+}
 
 int
-dht_do_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-                off_t yoff, int whichop)
+dht_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+             off_t yoff, dict_t *dict)
 {
-	dht_local_t  *local  = NULL;
-	dht_conf_t   *conf = NULL;
-        int           op_errno = -1;
-	xlator_t     *xvol = NULL;
-	off_t         xoff = 0;
+    dht_do_readdir(frame, this, fd, size, yoff, GF_FOP_READDIRP, dict);
+    return 0;
+}
 
+static int
+dht_fsyncdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+
+    local = frame->local;
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1)
+            local->op_errno = op_errno;
+        else if (op_ret == 0)
+            local->op_ret = 0;
+    }
+    UNLOCK(&frame->lock);
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt))
+        DHT_STACK_UNWIND(fsyncdir, frame, local->op_ret, local->op_errno,
+                         xdata);
+
+    return 0;
+}
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+int
+dht_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int datasync,
+             dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int op_errno = -1;
+    int i = -1;
 
-	conf = this->private;
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+    VALIDATE_OR_GOTO(this->private, err);
 
-	local = dht_local_init (frame);
-	if (!local) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		op_errno = ENOMEM;
-		goto err;
-	}
+    conf = this->private;
 
-	local->fd = fd_ref (fd);
-	local->size = size;
+    local = dht_local_init(frame, NULL, NULL, GF_FOP_FSYNCDIR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
 
-	dht_deitransform (this, yoff, &xvol, (uint64_t *)&xoff);
+    local->fd = fd_ref(fd);
+    local->call_cnt = conf->subvolume_cnt;
 
-	/* TODO: do proper readdir */
-        if (whichop == GF_FOP_READDIR)
-                STACK_WIND (frame, dht_readdir_cbk, xvol, xvol->fops->readdir,
-                            fd, size, xoff);
-        else
-                STACK_WIND (frame, dht_readdirp_cbk, xvol, xvol->fops->readdirp,
-                            fd, size, xoff);
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        STACK_WIND(frame, dht_fsyncdir_cbk, conf->subvolumes[i],
+                   conf->subvolumes[i]->fops->fsyncdir, fd, datasync, xdata);
+    }
 
-	return 0;
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (readdir, frame, -1, op_errno, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(fsyncdir, frame, -1, op_errno, NULL);
 
-	return 0;
+    return 0;
 }
 
-
 int
-dht_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-             off_t yoff)
+dht_newfile_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                int op_errno, inode_t *inode, struct iatt *stbuf,
+                struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
 {
-        int          op = GF_FOP_READDIR;
-        dht_conf_t  *conf = NULL;
-        int          i = 0;
+    xlator_t *prev = NULL;
+    int ret = -1;
+    dht_local_t *local = NULL;
+
+    if (op_ret == -1)
+        goto out;
+
+    local = frame->local;
+    if (!local) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    prev = cookie;
+
+    if (local->loc.parent) {
+        dht_inode_ctx_time_update(local->loc.parent, this, preparent, 0);
+        dht_inode_ctx_time_update(local->loc.parent, this, postparent, 1);
+    }
+
+    ret = dht_layout_preset(this, prev, inode);
+    if (ret < 0) {
+        gf_msg_debug(this->name, EINVAL,
+                     "could not set pre-set layout for subvolume %s",
+                     prev ? prev->name : NULL);
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    }
+    if (local->linked == _gf_true)
+        dht_linkfile_attr_heal(frame, this);
+out:
+    /*
+     * FIXME: ia_size and st_blocks of preparent and postparent do not have
+     * correct values. since, preparent and postparent buffers correspond
+     * to a directory these two members should have values equal to sum of
+     * corresponding values from each of the subvolume.
+     * See dht_iatt_merge for reference.
+     */
+    DHT_STRIP_PHASE1_FLAGS(stbuf);
+    dht_set_fixed_dir_stat(postparent);
+    dht_set_fixed_dir_stat(preparent);
+
+    if (local && local->lock[0].layout.parent_layout.locks) {
+        /* store op_errno for failure case*/
+        local->op_errno = op_errno;
+        local->refresh_layout_unlock(frame, this, op_ret, 1);
 
-        conf = this->private;
-        if (!conf)
-                goto out;
+        if (op_ret == 0) {
+            DHT_STACK_UNWIND(mknod, frame, op_ret, op_errno, inode, stbuf,
+                             preparent, postparent, xdata);
+        }
+    } else {
+        DHT_STACK_UNWIND(mknod, frame, op_ret, op_errno, inode, stbuf,
+                         preparent, postparent, xdata);
+    }
 
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                if (!conf->subvolume_status[i]) {
-                        op = GF_FOP_READDIRP;
-                        break;
-                }
+    return 0;
+}
+
+static int
+dht_mknod_linkfile_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                              int32_t op_ret, int32_t op_errno, inode_t *inode,
+                              struct iatt *stbuf, struct iatt *preparent,
+                              struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *cached_subvol = NULL;
+    dht_conf_t *conf = NULL;
+
+    local = frame->local;
+
+    if (!local || !local->cached_subvol) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (op_ret == -1) {
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    conf = this->private;
+    if (!conf) {
+        local->op_errno = EINVAL;
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    cached_subvol = local->cached_subvol;
+
+    if (local->params) {
+        dict_del(local->params, conf->link_xattr_name);
+        dict_del(local->params, GLUSTERFS_INTERNAL_FOP_KEY);
+    }
+
+    STACK_WIND_COOKIE(frame, dht_newfile_cbk, (void *)cached_subvol,
+                      cached_subvol, cached_subvol->fops->mknod, &local->loc,
+                      local->mode, local->rdev, local->umask, local->params);
+
+    return 0;
+err:
+    if (local && local->lock[0].layout.parent_layout.locks) {
+        local->refresh_layout_unlock(frame, this, -1, 1);
+    } else {
+        DHT_STACK_UNWIND(mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                         NULL);
+    }
+    return 0;
+}
+
+static int
+dht_mknod_wind_to_avail_subvol(call_frame_t *frame, xlator_t *this,
+                               xlator_t *subvol, loc_t *loc, dev_t rdev,
+                               mode_t mode, mode_t umask, dict_t *params)
+{
+    dht_local_t *local = NULL;
+    xlator_t *avail_subvol = NULL;
+
+    local = frame->local;
+
+    if (!dht_is_subvol_filled(this, subvol)) {
+        gf_msg_debug(this->name, 0, "creating %s on %s", loc->path,
+                     subvol->name);
+
+        STACK_WIND_COOKIE(frame, dht_newfile_cbk, (void *)subvol, subvol,
+                          subvol->fops->mknod, loc, mode, rdev, umask, params);
+    } else {
+        avail_subvol = dht_free_disk_available_subvol(this, subvol, local);
+
+        if (avail_subvol != subvol) {
+            local->params = dict_ref(params);
+            local->rdev = rdev;
+            local->mode = mode;
+            local->umask = umask;
+            local->cached_subvol = avail_subvol;
+            local->hashed_subvol = subvol;
+
+            gf_msg_debug(this->name, 0, "creating %s on %s (link at %s)",
+                         loc->path, avail_subvol->name, subvol->name);
+
+            dht_linkfile_create(frame, dht_mknod_linkfile_create_cbk, this,
+                                avail_subvol, subvol, loc);
+
+            goto out;
         }
 
-        if (conf->use_readdirp)
-                op = GF_FOP_READDIRP;
+        gf_msg_debug(this->name, 0, "creating %s on %s", loc->path,
+                     subvol->name);
 
+        STACK_WIND_COOKIE(frame, dht_newfile_cbk, (void *)subvol, subvol,
+                          subvol->fops->mknod, loc, mode, rdev, umask, params);
+    }
 out:
-        dht_do_readdir (frame, this, fd, size, yoff, op);
-        return 0;
+    return 0;
 }
 
-int
-dht_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-              off_t yoff)
+static int32_t
+dht_mknod_do(call_frame_t *frame)
 {
-        dht_do_readdir (frame, this, fd, size, yoff, GF_FOP_READDIRP);
-        return 0;
-}
+    dht_local_t *local = NULL;
+    dht_layout_t *refreshed = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *this = NULL;
+    dht_conf_t *conf = NULL;
+    dht_methods_t *methods = NULL;
 
+    local = frame->local;
 
+    this = THIS;
 
-int
-dht_fsyncdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		  int op_ret, int op_errno)
-{
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = 0;
+    conf = this->private;
+
+    GF_VALIDATE_OR_GOTO(this->name, conf, err);
+
+    methods = &(conf->methods);
+
+    /* We don't need parent_loc anymore */
+    loc_wipe(&local->loc);
+
+    loc_copy(&local->loc, &local->loc2);
+
+    loc_wipe(&local->loc2);
+
+    refreshed = local->selfheal.refreshed_layout;
 
+    subvol = methods->layout_search(this, refreshed, local->loc.name);
 
-	local = frame->local;
+    if (!subvol) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+               "no subvolume in "
+               "layout for path=%s",
+               local->loc.path);
+        local->op_errno = ENOENT;
+        goto err;
+    }
 
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1)
-			local->op_errno = op_errno;
+    dht_mknod_wind_to_avail_subvol(frame, this, subvol, &local->loc,
+                                   local->rdev, local->mode, local->umask,
+                                   local->params);
+    return 0;
+err:
+    local->refresh_layout_unlock(frame, this, -1, 1);
+
+    return 0;
+}
 
-		if (op_ret == 0)
-			local->op_ret = 0;
-	}
-	UNLOCK (&frame->lock);
+static int32_t
+dht_mknod_unlock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    DHT_STACK_DESTROY(frame);
+    return 0;
+}
+
+static int32_t
+dht_mknod_finish(call_frame_t *frame, xlator_t *this, int op_ret,
+                 int invoke_cbk)
+{
+    dht_local_t *local = NULL, *lock_local = NULL;
+    call_frame_t *lock_frame = NULL;
+    int lock_count = 0;
+
+    local = frame->local;
+    lock_count = dht_lock_count(local->lock[0].layout.parent_layout.locks,
+                                local->lock[0].layout.parent_layout.lk_count);
+    if (lock_count == 0)
+        goto done;
+
+    lock_frame = copy_frame(frame);
+    if (lock_frame == NULL) {
+        goto done;
+    }
+
+    lock_local = dht_local_init(lock_frame, &local->loc, NULL,
+                                lock_frame->root->op);
+    if (lock_local == NULL) {
+        goto done;
+    }
+
+    lock_local->lock[0]
+        .layout.parent_layout.locks = local->lock[0].layout.parent_layout.locks;
+    lock_local->lock[0].layout.parent_layout.lk_count =
+        local->lock[0].layout.parent_layout.lk_count;
+
+    local->lock[0].layout.parent_layout.locks = NULL;
+    local->lock[0].layout.parent_layout.lk_count = 0;
+
+    dht_unlock_inodelk(lock_frame,
+                       lock_local->lock[0].layout.parent_layout.locks,
+                       lock_local->lock[0].layout.parent_layout.lk_count,
+                       dht_mknod_unlock_cbk);
+    lock_frame = NULL;
 
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt))
-		DHT_STACK_UNWIND (fsyncdir, frame, local->op_ret, local->op_errno);
+done:
+    if (lock_frame != NULL) {
+        DHT_STACK_DESTROY(lock_frame);
+    }
 
+    if (op_ret == 0)
         return 0;
+
+    DHT_STACK_UNWIND(mknod, frame, op_ret, local->op_errno, NULL, NULL, NULL,
+                     NULL, NULL);
+    return 0;
 }
 
+static int32_t
+dht_mknod_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
 
-int
-dht_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd, int datasync)
+    local = frame->local;
+
+    if (!local) {
+        goto err;
+    }
+
+    if (op_ret < 0) {
+        gf_msg("DHT", GF_LOG_ERROR, 0, DHT_MSG_INODE_LK_ERROR,
+               "mknod lock failed for file: %s", local->loc2.name);
+
+        local->op_errno = op_errno;
+
+        goto err;
+    }
+
+    local->refresh_layout_unlock = dht_mknod_finish;
+
+    local->refresh_layout_done = dht_mknod_do;
+
+    dht_refresh_layout(frame);
+
+    return 0;
+err:
+    if (local)
+        dht_mknod_finish(frame, this, -1, 0);
+    else
+        DHT_STACK_UNWIND(mknod, frame, -1, EINVAL, NULL, NULL, NULL, NULL,
+                         NULL);
+    return 0;
+}
+
+static int32_t
+dht_mknod_lock(call_frame_t *frame, xlator_t *subvol)
 {
-	dht_local_t  *local  = NULL;
-	dht_conf_t   *conf = NULL;
-        int           op_errno = -1;
-	int           i = -1;
+    dht_local_t *local = NULL;
+    int count = 1, ret = -1;
+    dht_lock_t **lk_array = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO(frame->this->name, frame->local, err);
 
+    local = frame->local;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (this->private, err);
+    lk_array = GF_CALLOC(count, sizeof(*lk_array), gf_common_mt_pointer);
 
-	conf = this->private;
+    if (lk_array == NULL)
+        goto err;
 
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
+    lk_array[0] = dht_lock_new(frame->this, subvol, &local->loc, F_RDLCK,
+                               DHT_LAYOUT_HEAL_DOMAIN, NULL,
+                               IGNORE_ENOENT_ESTALE);
 
-	local->fd = fd_ref (fd);
-	local->call_cnt = conf->subvolume_cnt;
+    if (lk_array[0] == NULL)
+        goto err;
 
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		STACK_WIND (frame, dht_fsyncdir_cbk,
-			    conf->subvolumes[i],
-			    conf->subvolumes[i]->fops->fsyncdir,
-			    fd, datasync);
-	}
+    local->lock[0].layout.parent_layout.locks = lk_array;
+    local->lock[0].layout.parent_layout.lk_count = count;
 
-	return 0;
+    ret = dht_blocking_inodelk(frame, lk_array, count, dht_mknod_lock_cbk);
 
+    if (ret < 0) {
+        local->lock[0].layout.parent_layout.locks = NULL;
+        local->lock[0].layout.parent_layout.lk_count = 0;
+        goto err;
+    }
+
+    return 0;
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (fsyncdir, frame, -1, op_errno);
+    if (lk_array != NULL) {
+        dht_lock_array_free(lk_array, count);
+        GF_FREE(lk_array);
+    }
 
-	return 0;
+    return -1;
 }
 
-
-int
-dht_newfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		 int op_ret, int op_errno,
-                 inode_t *inode, struct iatt *stbuf, struct iatt *preparent,
-                 struct iatt *postparent)
+static int
+dht_refresh_parent_layout_resume(call_frame_t *frame, xlator_t *this, int ret,
+                                 int invoke_cbk)
 {
-	call_frame_t *prev = NULL;
-	int           ret = -1;
-        dht_local_t  *local = NULL;
+    dht_local_t *local = NULL, *parent_local = NULL;
+    call_stub_t *stub = NULL;
+    call_frame_t *parent_frame = NULL;
 
+    local = frame->local;
 
-	if (op_ret == -1)
-		goto out;
+    stub = local->stub;
+    local->stub = NULL;
 
-        local = frame->local;
-        if (!local) {
-                op_ret = -1;
-                op_errno = EINVAL;
-                goto out;
-        }
+    parent_frame = stub->frame;
+    parent_local = parent_frame->local;
 
-	prev = cookie;
+    if (ret < 0) {
+        parent_local->op_ret = -1;
+        parent_local->op_errno = local->op_errno ? local->op_errno : EIO;
+    } else {
+        parent_local->op_ret = 0;
+    }
 
-	dht_itransform (this, prev->this, stbuf->ia_ino, &stbuf->ia_ino);
-        if (local->loc.parent) {
-                preparent->ia_ino = local->loc.parent->ino;
-                postparent->ia_ino = local->loc.parent->ino;
-
-                WIPE (preparent);
-                WIPE (postparent);
-        }
-
-        ret = dht_layout_preset (this, prev->this, inode);
-	if (ret < 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"could not set pre-set layout for subvolume %s",
-			prev->this->name);
-		op_ret   = -1;
-		op_errno = EINVAL;
-		goto out;
-	}
-out:
-        /* 
-         * FIXME: ia_size and st_blocks of preparent and postparent do not have 
-         * correct values. since, preparent and postparent buffers correspond
-         * to a directory these two members should have values equal to sum of
-         * corresponding values from each of the subvolume.
-         * See dht_iatt_merge for reference.
-         */ 
+    call_resume(stub);
 
-	DHT_STACK_UNWIND (mknod, frame, op_ret, op_errno, inode, stbuf, preparent,
-                          postparent);
-	return 0;
+    DHT_STACK_DESTROY(frame);
+
+    return 0;
 }
 
-int
-dht_mknod_linkfile_create_cbk (call_frame_t *frame, void *cookie,
-                               xlator_t *this,
-                               int32_t op_ret, int32_t op_errno,
-                               inode_t *inode, struct iatt *stbuf,
-                               struct iatt *preparent, struct iatt *postparent)
+static int
+dht_refresh_parent_layout_done(call_frame_t *frame)
 {
-	dht_local_t  *local = NULL;
-	xlator_t     *cached_subvol = NULL;
+    dht_local_t *local = NULL;
+    int ret = 0;
 
-        if (op_ret == -1)
-                goto err;
+    local = frame->local;
 
-	local = frame->local;
-	cached_subvol = local->cached_subvol;
+    if (local->op_ret < 0) {
+        ret = -1;
+        goto resume;
+    }
 
-        STACK_WIND (frame, dht_newfile_cbk,
-                    cached_subvol, cached_subvol->fops->mknod,
-                    &local->loc, local->mode, local->rdev,
-                    local->params);
+    dht_layout_set(frame->this, local->loc.inode,
+                   local->selfheal.refreshed_layout);
 
-        return 0;
- err:
- 	DHT_STACK_UNWIND (mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL);	
- 	return 0;
+resume:
+    dht_refresh_parent_layout_resume(frame, frame->this, ret, 1);
+    return 0;
 }
 
-int
-dht_mknod (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, mode_t mode, dev_t rdev, dict_t *params)
-{
-	xlator_t    *subvol = NULL;
-	int          op_errno = -1;
-        int          ret = -1;
-        xlator_t    *avail_subvol = NULL;
-	dht_conf_t  *conf = NULL;
-	dht_local_t *local = NULL;
-
-	VALIDATE_OR_GOTO (frame, err);
-	VALIDATE_OR_GOTO (this, err);
-	VALIDATE_OR_GOTO (loc, err);
-
-	conf = this->private;
-
-        dht_get_du_info (frame, this, loc);
-
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	subvol = dht_subvol_get_hashed (this, loc);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no subvolume in layout for path=%s",
-			loc->path);
-		op_errno = ENOENT;
-		goto err;
-	}
-
-        ret = loc_dup (loc, &local->loc);
-        if (ret == -1) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                goto err;
-        }
+static int
+dht_handle_parent_layout_change(xlator_t *this, call_stub_t *stub)
+{
+    call_frame_t *refresh_frame = NULL, *frame = NULL;
+    dht_local_t *refresh_local = NULL, *local = NULL;
 
-        if (!dht_is_subvol_filled (this, subvol)) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "creating %s on %s", loc->path, subvol->name);
-                
-                STACK_WIND (frame, dht_newfile_cbk,
-                            subvol, subvol->fops->mknod,
-                            loc, mode, rdev, params);
-        } else {
-                avail_subvol = dht_free_disk_available_subvol (this, subvol);
-                if (avail_subvol != subvol) {
-                        /* Choose the minimum filled volume, and create the 
-                           files there */
-
-                        local->params = dict_ref (params);
-                        local->cached_subvol = avail_subvol;
-                        local->mode = mode; 
-                        local->rdev = rdev;
-                        
-                        dht_linkfile_create (frame, 
-                                             dht_mknod_linkfile_create_cbk,
-                                             avail_subvol, subvol, loc);
-                } else {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "creating %s on %s", loc->path, subvol->name);
-                        
-                        STACK_WIND (frame, dht_newfile_cbk,
-                                    subvol, subvol->fops->mknod,
-                                    loc, mode, rdev, params);
-                }
-        }
+    frame = stub->frame;
+    local = frame->local;
 
-	return 0;
+    refresh_frame = copy_frame(frame);
+    if (!refresh_frame) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+               "mem allocation failed for refresh_frame");
+        return -1;
+    }
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (mknod, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL);
+    refresh_local = dht_local_init(refresh_frame, NULL, NULL, stub->fop);
+    if (!refresh_local) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+               "mem allocation failed for refresh_local");
+        return -1;
+    }
+
+    refresh_local->loc.inode = inode_ref(local->loc.parent);
+    gf_uuid_copy(refresh_local->loc.gfid, local->loc.parent->gfid);
+
+    refresh_local->stub = stub;
+
+    refresh_local->refresh_layout_unlock = dht_refresh_parent_layout_resume;
+    refresh_local->refresh_layout_done = dht_refresh_parent_layout_done;
 
-	return 0;
+    dht_refresh_layout(refresh_frame);
+    return 0;
 }
 
+static int32_t
+dht_call_mkdir_stub(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    call_stub_t *stub = NULL;
 
-int
-dht_symlink (call_frame_t *frame, xlator_t *this,
-	     const char *linkname, loc_t *loc, dict_t *params)
-{
-	xlator_t    *subvol = NULL;
-	int          op_errno = -1;
-        dht_local_t *local = NULL;
-        int          ret = -1;
-
-
-	VALIDATE_OR_GOTO (frame, err);
-	VALIDATE_OR_GOTO (this, err);
-	VALIDATE_OR_GOTO (loc, err);
-
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	subvol = dht_subvol_get_hashed (this, loc);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no subvolume in layout for path=%s",
-			loc->path);
-		op_errno = ENOENT;
-		goto err;
-	}
-
-        ret = loc_copy (&local->loc, loc);
-        if (ret == -1) {
-                gf_log (this->name, GF_LOG_TRACE, "Failed to copy loc");
-                op_errno = ENOMEM;
-                goto err;
-        }
+    local = frame->local;
+    stub = local->stub;
+    local->stub = NULL;
 
-	gf_log (this->name, GF_LOG_TRACE,
-		"creating %s on %s", loc->path, subvol->name);
+    if (op_ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+    } else {
+        local->op_ret = 0;
+    }
 
-	STACK_WIND (frame, dht_newfile_cbk,
-		    subvol, subvol->fops->symlink,
-		    linkname, loc, params);
+    call_resume(stub);
 
-	return 0;
+    return 0;
+}
 
+static int32_t
+dht_guard_parent_layout_and_namespace(xlator_t *subvol, call_stub_t *stub)
+{
+    dht_local_t *local = NULL;
+    int ret = -1;
+    loc_t *loc = NULL;
+    xlator_t *hashed_subvol = NULL, *this = NULL;
+    ;
+    call_frame_t *frame = NULL;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    int32_t *parent_disk_layout = NULL;
+    dht_layout_t *parent_layout = NULL;
+    dht_conf_t *conf = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht", stub, err);
+
+    frame = stub->frame;
+    this = frame->this;
+
+    conf = this->private;
+
+    local = frame->local;
+
+    local->stub = stub;
+
+    /* TODO: recheck whether we should lock on src or dst if we do similar
+     * stale layout checks for rename.
+     */
+    loc = &stub->args.loc;
+
+    gf_uuid_unparse(loc->parent->gfid, pgfid);
+
+    if (local->params == NULL) {
+        local->params = dict_new();
+        if (local->params == NULL) {
+            local->op_errno = ENOMEM;
+            gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+                   DHT_MSG_PARENT_LAYOUT_CHANGED,
+                   "%s (%s/%s) (path: %s): "
+                   "dict allocation failed",
+                   gf_fop_list[stub->fop], pgfid, loc->name, loc->path);
+            goto err;
+        }
+    }
+
+    hashed_subvol = dht_subvol_get_hashed(this, loc);
+    if (hashed_subvol == NULL) {
+        local->op_errno = EINVAL;
+
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "%s (%s/%s) (path: %s): "
+               "hashed subvolume not found",
+               gf_fop_list[stub->fop], pgfid, loc->name, loc->path);
+        goto err;
+    }
+
+    parent_layout = dht_layout_get(this, loc->parent);
+
+    ret = dht_disk_layout_extract_for_subvol(this, parent_layout, hashed_subvol,
+                                             &parent_disk_layout);
+    if (ret == -1) {
+        local->op_errno = EINVAL;
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "%s (%s/%s) (path: %s): "
+               "extracting in-memory layout of parent failed. ",
+               gf_fop_list[stub->fop], pgfid, loc->name, loc->path);
+        goto err;
+    }
+
+    memcpy((void *)local->parent_disk_layout, (void *)parent_disk_layout,
+           sizeof(local->parent_disk_layout));
+
+    dht_layout_unref(this, parent_layout);
+    parent_layout = NULL;
+
+    ret = dict_set_str(local->params, GF_PREOP_PARENT_KEY, conf->xattr_name);
+    if (ret < 0) {
+        local->op_errno = -ret;
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "%s (%s/%s) (path: %s): "
+               "setting %s key in params dictionary failed. ",
+               gf_fop_list[stub->fop], pgfid, loc->name, loc->path,
+               GF_PREOP_PARENT_KEY);
+        goto err;
+    }
+
+    ret = dict_set_bin(local->params, conf->xattr_name, parent_disk_layout,
+                       4 * 4);
+    if (ret < 0) {
+        local->op_errno = -ret;
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "%s (%s/%s) (path: %s): "
+               "setting parent-layout in params dictionary failed. ",
+               gf_fop_list[stub->fop], pgfid, loc->name, loc->path);
+        goto err;
+    }
+
+    parent_disk_layout = NULL;
+    local->hashed_subvol = hashed_subvol;
+
+    local->current = &local->lock[0];
+    ret = dht_protect_namespace(frame, loc, hashed_subvol, &local->current->ns,
+                                dht_call_mkdir_stub);
+    if (ret < 0)
+        goto err;
+
+    return 0;
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (link, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL);
 
-	return 0;
-}
+    if (parent_disk_layout != NULL)
+        GF_FREE(parent_disk_layout);
+
+    if (parent_layout != NULL)
+        dht_layout_unref(this, parent_layout);
 
+    return -1;
+}
 
 int
-dht_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc)
-{
-	xlator_t    *cached_subvol = NULL;
-	xlator_t    *hashed_subvol = NULL;
-        int          ret = -1;
-	int          op_errno = -1;
-	dht_local_t *local = NULL;
-
-
-	VALIDATE_OR_GOTO (frame, err);
-	VALIDATE_OR_GOTO (this, err);
-	VALIDATE_OR_GOTO (loc, err);
-
-        if (dht_filter_loc_subvol_key (this, loc, &local->loc,
-                                       &cached_subvol)) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "unlinking %s on %s (given path %s)",
-                        local->loc.path, cached_subvol->name, loc->path);
-                STACK_WIND (frame, dht_unlink_cbk,
-                            cached_subvol, cached_subvol->fops->unlink,
-                            &local->loc);
+dht_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+          dev_t rdev, mode_t umask, dict_t *params)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    int i = 0;
+    int ret = 0;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+
+    conf = this->private;
+
+    dht_get_du_info(frame, this, loc);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_MKNOD);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = dht_subvol_get_hashed(this, loc);
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no subvolume in layout for path=%s",
+                     loc->path);
+        op_errno = EIO;
+        goto err;
+    }
+
+    /* Post remove-brick, the client layout may not be in sync with
+     * disk layout because of lack of lookup. Hence,a mknod call
+     * may fall on the decommissioned brick.  Hence, if the
+     * hashed_subvol is part of decommissioned bricks  list, do a
+     * lookup on parent dir. If a fix-layout is already done by the
+     * remove-brick process, the parent directory layout will be in
+     * sync with that of the disk. If fix-layout is still ending
+     * on the parent directory, we can let the file get created on
+     * the decommissioned brick which will be eventually migrated to
+     * non-decommissioned brick based on the new layout.
+     */
+
+    if (conf->decommission_subvols_cnt) {
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (conf->decommissioned_bricks[i] &&
+                conf->decommissioned_bricks[i] == subvol) {
+                gf_msg_debug(this->name, 0,
+                             "hashed subvol:%s is "
+                             "part of decommission brick list for "
+                             "file: %s",
+                             subvol->name, loc->path);
+
+                /* dht_refresh_layout needs directory info in
+                 * local->loc. Hence, storing the parent_loc in
+                 * local->loc and storing the create context in
+                 * local->loc2. We will restore this information
+                 * in dht_creation do */
+
+                ret = loc_copy(&local->loc2, &local->loc);
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+                           "loc_copy failed %s", loc->path);
+
+                    goto err;
+                }
+
+                local->params = dict_ref(params);
+                local->rdev = rdev;
+                local->mode = mode;
+                local->umask = umask;
+
+                loc_wipe(&local->loc);
+
+                ret = dht_build_parent_loc(this, &local->loc, loc, &op_errno);
+
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_LOC_FAILED,
+                           "parent loc build failed");
+                    goto err;
+                }
+
+                ret = dht_mknod_lock(frame, subvol);
+
+                if (ret < 0) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INODE_LK_ERROR,
+                           "locking parent failed");
+                    goto err;
+                }
+
                 goto done;
+            }
         }
+    }
+
+    dht_mknod_wind_to_avail_subvol(frame, this, subvol, loc, rdev, mode, umask,
+                                   params);
 
-	cached_subvol = dht_subvol_get_cached (this, loc->inode);
-	if (!cached_subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	hashed_subvol = dht_subvol_get_hashed (this, loc);
-	if (!hashed_subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no subvolume in layout for path=%s",
-			loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	ret = loc_copy (&local->loc, loc);
-	if (ret == -1) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	if (hashed_subvol != cached_subvol) {
-		STACK_WIND (frame, dht_unlink_linkfile_cbk,
-			    hashed_subvol, hashed_subvol->fops->unlink, loc);
-        } else {
-                STACK_WIND (frame, dht_unlink_cbk,
-                            cached_subvol, cached_subvol->fops->unlink, loc);
-        }
 done:
-	return 0;
+    return 0;
+
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (unlink, frame, -1, op_errno, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
 
-	return 0;
+    return 0;
 }
 
+int
+dht_symlink(call_frame_t *frame, xlator_t *this, const char *linkname,
+            loc_t *loc, mode_t umask, dict_t *params)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_SYMLINK);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = dht_subvol_get_hashed(this, loc);
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no subvolume in layout for path=%s",
+                     loc->path);
+        op_errno = EIO;
+        goto err;
+    }
+
+    gf_msg_trace(this->name, 0, "creating %s on %s", loc->path, subvol->name);
+
+    STACK_WIND_COOKIE(frame, dht_newfile_cbk, (void *)subvol, subvol,
+                      subvol->fops->symlink, linkname, loc, umask, params);
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(link, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+
+    return 0;
+}
 
 int
-dht_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-	      int op_ret, int op_errno,
-              inode_t *inode, struct iatt *stbuf, struct iatt *preparent,
-              struct iatt *postparent)
+dht_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+           dict_t *xdata)
 {
-        call_frame_t *prev = NULL;
-	dht_layout_t *layout = NULL;
-	dht_local_t  *local = NULL;
+    xlator_t *cached_subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_UNLINK);
+    if (!local) {
+        op_errno = ENOMEM;
+
+        goto err;
+    }
+
+    cached_subvol = local->cached_subvol;
+    if (!cached_subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for path=%s",
+                     loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    local->flags = xflag;
+    STACK_WIND_COOKIE(frame, dht_unlink_cbk, cached_subvol, cached_subvol,
+                      cached_subvol->fops->unlink, loc, xflag, xdata);
+
+    return 0;
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(unlink, frame, -1, op_errno, NULL, NULL, NULL);
 
-        prev = cookie;
-	local = frame->local;
+    return 0;
+}
 
-        if (op_ret == -1)
-                goto out;
+static int
+dht_remove_stale_linkto_cbk(int ret, call_frame_t *sync_frame, void *data)
+{
+    DHT_STACK_DESTROY(sync_frame);
+    return 0;
+}
 
-	layout = dht_layout_for_subvol (this, prev->this);
-	if (!layout) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no pre-set layout for subvolume %s",
-			prev->this->name);
-		op_ret   = -1;
-		op_errno = EINVAL;
-		goto out;
-	}
+static int
+dht_remove_stale_linkto(void *data)
+{
+    call_frame_t *frame = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *this = NULL;
+    dict_t *xdata_in = NULL;
+    int ret = 0;
+
+    GF_VALIDATE_OR_GOTO("dht", data, out);
+
+    frame = data;
+    local = frame->local;
+    this = frame->this;
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", local, out);
+    GF_VALIDATE_OR_GOTO("dht", local->link_subvol, out);
+
+    xdata_in = dict_new();
+    if (!xdata_in)
+        goto out;
+
+    ret = dht_fill_dict_to_avoid_unlink_of_migrating_file(xdata_in);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, 0,
+               "Failed to set keys for stale linkto"
+               "deletion on path %s",
+               local->loc.path);
+        goto out;
+    }
+
+    ret = syncop_unlink(local->link_subvol, &local->loc, xdata_in, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, 0,
+               "Removal of linkto failed"
+               " on path %s at subvol %s",
+               local->loc.path, local->link_subvol->name);
+    }
+out:
+    if (xdata_in)
+        dict_unref(xdata_in);
+    return ret;
+}
 
-	stbuf->ia_ino = local->loc.inode->ino;
+static int
+dht_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+             int op_errno, inode_t *inode, struct iatt *stbuf,
+             struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int ret = -1;
+    gf_boolean_t stbuf_merged = _gf_false;
+    xlator_t *subvol = NULL;
+    call_frame_t *cleanup_frame = NULL;
+    dht_local_t *cleanup_local = NULL;
+
+    local = frame->local;
+
+    if (op_ret == -1) {
+        /* Remove the linkto if exists */
+        if (local->linked) {
+            cleanup_frame = create_frame(this, this->ctx->pool);
+            if (cleanup_frame) {
+                cleanup_local = dht_local_init(cleanup_frame, &local->loc2,
+                                               NULL, 0);
+                if (!cleanup_local || !local->link_subvol) {
+                    DHT_STACK_DESTROY(cleanup_frame);
+                    goto out;
+                }
+                cleanup_local->link_subvol = local->link_subvol;
+                FRAME_SU_DO(cleanup_frame, dht_local_t);
+                ret = synctask_new(this->ctx->env, dht_remove_stale_linkto,
+                                   dht_remove_stale_linkto_cbk, cleanup_frame,
+                                   cleanup_frame);
+            }
+        }
+        /* No continuation on DHT inode missing errors, as we should
+         * then have a good stbuf that states P2 happened. We would
+         * get inode missing if, the file completed migrated between
+         * the lookup and the link call */
+        goto out;
+    }
+
+    /* Update parent on success, even if P1/2 checks are positive.
+     * The second call on success will further update the parent */
+    if (local->loc.parent) {
+        dht_inode_ctx_time_update(local->loc.parent, this, preparent, 0);
+        dht_inode_ctx_time_update(local->loc.parent, this, postparent, 1);
+    }
+
+    /* Update linkto attrs, if this is the first call and non-P2,
+     * if we detect P2 then we need to trust the attrs from the
+     * second call, not the first */
+    if (local->linked == _gf_true &&
+        ((local->call_cnt == 1 && !IS_DHT_MIGRATION_PHASE2(stbuf)) ||
+         (local->call_cnt != 1 && IS_DHT_MIGRATION_PHASE2(&local->stbuf)))) {
+        dht_iatt_merge(this, &local->stbuf, stbuf);
+        stbuf_merged = _gf_true;
+        dht_linkfile_attr_heal(frame, this);
+    }
+
+    /* No further P1/2 checks if we are in the second iteration of
+     * the call */
+    if (local->call_cnt != 1) {
+        goto out;
+    } else {
+        /* Preserve the return values, in case the migration decides
+         * to recreate the link on the same subvol that the current
+         * hased for the link was created on. */
+        dht_iatt_merge(this, &local->preparent, preparent);
+        dht_iatt_merge(this, &local->postparent, postparent);
+        if (!stbuf_merged) {
+            dht_iatt_merge(this, &local->stbuf, stbuf);
+            stbuf_merged = _gf_true;
+        }
 
-        preparent->ia_ino = local->loc2.parent->ino;
-        postparent->ia_ino = local->loc2.parent->ino;
+        local->inode = inode_ref(inode);
+    }
 
-        WIPE (preparent);
-        WIPE (postparent);
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
+    local->rebalance.target_op_fn = dht_link2;
+    dht_set_local_rebalance(this, local, stbuf, preparent, postparent, xdata);
 
+    /* Check if the rebalance phase2 is true */
+    if (IS_DHT_MIGRATION_PHASE2(stbuf)) {
+        ret = dht_inode_ctx_get_mig_info(this, local->loc.inode, NULL, &subvol);
+        if (!subvol) {
+            /* Phase 2 of migration */
+            ret = dht_rebalance_complete_check(this, frame);
+            if (!ret)
+                return 0;
+        } else {
+            dht_link2(this, subvol, frame, 0);
+            return 0;
+        }
+    }
+
+    /* Check if the rebalance phase1 is true */
+    if (IS_DHT_MIGRATION_PHASE1(stbuf)) {
+        ret = dht_inode_ctx_get_mig_info(this, local->loc.inode, NULL, &subvol);
+        if (subvol) {
+            dht_link2(this, subvol, frame, 0);
+            return 0;
+        }
+        ret = dht_rebalance_in_progress_check(this, frame);
+        if (!ret)
+            return 0;
+    }
 out:
-        DHT_STACK_UNWIND (link, frame, op_ret, op_errno, inode, stbuf, preparent,
-                          postparent);
+    DHT_STRIP_PHASE1_FLAGS(stbuf);
 
-	return 0;
-}
+    dht_set_fixed_dir_stat(preparent);
+    dht_set_fixed_dir_stat(postparent);
+    DHT_STACK_UNWIND(link, frame, op_ret, op_errno, inode, stbuf, preparent,
+                     postparent, NULL);
 
+    return 0;
+}
 
-int
-dht_link_linkfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int op_ret, int op_errno,
-                       inode_t *inode, struct iatt *stbuf,
-                       struct iatt *preparent, struct iatt *postparent)
+static int
+dht_link2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
 {
-	dht_local_t  *local = NULL;
-	xlator_t     *srcvol = NULL;
+    dht_local_t *local = NULL;
+    int op_errno = EINVAL;
+
+    local = frame->local;
+    if (!local)
+        goto err;
 
+    op_errno = local->op_errno;
 
-	if (op_ret == -1)
-		goto err;
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        dht_set_fixed_dir_stat(&local->preparent);
+        dht_set_fixed_dir_stat(&local->postparent);
+
+        DHT_STACK_UNWIND(link, frame, local->op_ret, op_errno, local->inode,
+                         &local->stbuf, &local->preparent, &local->postparent,
+                         NULL);
+        return 0;
+    }
+
+    if (subvol == NULL) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    /* Second call to create link file could result in EEXIST as the
+     * first call created the linkto in the currently
+     * migrating subvol, which could be the new hashed subvol */
+    if (local->link_subvol == subvol) {
+        DHT_STRIP_PHASE1_FLAGS(&local->stbuf);
+        dht_set_fixed_dir_stat(&local->preparent);
+        dht_set_fixed_dir_stat(&local->postparent);
+        DHT_STACK_UNWIND(link, frame, 0, 0, local->inode, &local->stbuf,
+                         &local->preparent, &local->postparent, NULL);
 
-	local = frame->local;
-	srcvol = local->linkfile.srcvol;
+        return 0;
+    }
 
-	STACK_WIND (frame, dht_link_cbk,
-		    srcvol, srcvol->fops->link,
-		    &local->loc, &local->loc2);
+    local->call_cnt = 2;
 
-	return 0;
+    STACK_WIND(frame, dht_link_cbk, subvol, subvol->fops->link, &local->loc,
+               &local->loc2, local->xattr_req);
 
+    return 0;
 err:
-	DHT_STACK_UNWIND (link, frame, op_ret, op_errno, inode, stbuf, preparent,
-                          postparent);
+    DHT_STACK_UNWIND(link, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
 
-	return 0;
+    return 0;
 }
 
+static int
+dht_link_linkfile_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, inode_t *inode,
+                      struct iatt *stbuf, struct iatt *preparent,
+                      struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *srcvol = NULL;
+
+    if (op_ret == -1)
+        goto err;
 
-int
-dht_link (call_frame_t *frame, xlator_t *this,
-	  loc_t *oldloc, loc_t *newloc)
-{
-	xlator_t    *cached_subvol = NULL;
-	xlator_t    *hashed_subvol = NULL;
-	int          op_errno = -1;
-	int          ret = -1;
-	dht_local_t *local = NULL;
-
-
-	VALIDATE_OR_GOTO (frame, err);
-	VALIDATE_OR_GOTO (this, err);
-	VALIDATE_OR_GOTO (oldloc, err);
-	VALIDATE_OR_GOTO (newloc, err);
-
-	cached_subvol = dht_subvol_get_cached (this, oldloc->inode);
-	if (!cached_subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for path=%s", oldloc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	hashed_subvol = dht_subvol_get_hashed (this, newloc);
-	if (!hashed_subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no subvolume in layout for path=%s",
-			newloc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	ret = loc_copy (&local->loc, oldloc);
-	if (ret == -1) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	ret = loc_copy (&local->loc2, newloc);
-	if (ret == -1) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	if (hashed_subvol != cached_subvol) {
-                memcpy (local->gfid, oldloc->inode->gfid, 16);
-		dht_linkfile_create (frame, dht_link_linkfile_cbk,
-				     cached_subvol, hashed_subvol, newloc);
-	} else {
-		STACK_WIND (frame, dht_link_cbk,
-			    cached_subvol, cached_subvol->fops->link,
-			    oldloc, newloc);
-	}
-
-	return 0;
+    local = frame->local;
+    srcvol = local->linkfile.srcvol;
+
+    STACK_WIND(frame, dht_link_cbk, srcvol, srcvol->fops->link, &local->loc,
+               &local->loc2, local->xattr_req);
+
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (link, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+    DHT_STRIP_PHASE1_FLAGS(stbuf);
+    dht_set_fixed_dir_stat(preparent);
+    dht_set_fixed_dir_stat(postparent);
+    DHT_STACK_UNWIND(link, frame, op_ret, op_errno, inode, stbuf, preparent,
+                     postparent, xdata);
 
-	return 0;
+    return 0;
 }
 
+int
+dht_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+         dict_t *xdata)
+{
+    xlator_t *cached_subvol = NULL;
+    xlator_t *hashed_subvol = NULL;
+    int op_errno = -1;
+    int ret = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(oldloc, err);
+    VALIDATE_OR_GOTO(newloc, err);
+
+    local = dht_local_init(frame, oldloc, NULL, GF_FOP_LINK);
+    if (!local) {
+        op_errno = ENOMEM;
+
+        goto err;
+    }
+    local->call_cnt = 1;
+
+    cached_subvol = local->cached_subvol;
+    if (!cached_subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for path=%s",
+                     oldloc->path);
+        op_errno = ENOENT;
+        goto err;
+    }
+
+    hashed_subvol = dht_subvol_get_hashed(this, newloc);
+    if (!hashed_subvol) {
+        gf_msg_debug(this->name, 0, "no subvolume in layout for path=%s",
+                     newloc->path);
+        op_errno = EIO;
+        goto err;
+    }
+
+    ret = loc_copy(&local->loc2, newloc);
+    if (ret == -1) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    if (hashed_subvol != cached_subvol) {
+        gf_uuid_copy(local->gfid, oldloc->inode->gfid);
+        dht_linkfile_create(frame, dht_link_linkfile_cbk, this, cached_subvol,
+                            hashed_subvol, newloc);
+    } else {
+        STACK_WIND(frame, dht_link_cbk, cached_subvol,
+                   cached_subvol->fops->link, oldloc, newloc, xdata);
+    }
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(link, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+
+    return 0;
+}
 
 int
-dht_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		 int op_ret, int op_errno,
-		 fd_t *fd, inode_t *inode, struct iatt *stbuf,
-                 struct iatt *preparent, struct iatt *postparent)
+dht_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, fd_t *fd, inode_t *inode, struct iatt *stbuf,
+               struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
 {
-	call_frame_t *prev = NULL;
-	int           ret = -1;
-        dht_local_t  *local = NULL;
+    xlator_t *prev = NULL;
+    int ret = -1;
+    dht_local_t *local = NULL;
+    gf_boolean_t parent_layout_changed = _gf_false;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    xlator_t *subvol = NULL;
+
+    local = frame->local;
+
+    local = frame->local;
+    if (!local) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    if (op_ret == -1) {
+        local->op_errno = op_errno;
+        parent_layout_changed = (xdata &&
+                                 dict_get(xdata, GF_PREOP_CHECK_FAILED))
+                                    ? _gf_true
+                                    : _gf_false;
+
+        if (parent_layout_changed) {
+            if (local && local->lock[0].layout.parent_layout.locks) {
+                /* Returning failure as the layout could not be fixed even under
+                 * the lock */
+                goto out;
+            }
 
-	if (op_ret == -1)
-		goto out;
+            gf_uuid_unparse(local->loc.parent->gfid, pgfid);
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_PARENT_LAYOUT_CHANGED,
+                   "create (%s/%s) (path: %s): parent layout "
+                   "changed. Attempting a layout refresh and then a "
+                   "retry",
+                   pgfid, local->loc.name, local->loc.path);
+
+            /*
+              dht_refresh_layout needs directory info in local->loc.Hence,
+              storing the parent_loc in local->loc and storing the create
+              context in local->loc2. We will restore this information in
+              dht_creation_do.
+             */
+
+            loc_wipe(&local->loc2);
+
+            ret = loc_copy(&local->loc2, &local->loc);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+                       "loc_copy failed %s", local->loc.path);
 
-        local = frame->local;
-        if (!local) {
-                op_ret = -1;
-                op_errno = EINVAL;
                 goto out;
-        }
+            }
 
-	prev = cookie;
+            loc_wipe(&local->loc);
 
-	dht_itransform (this, prev->this, stbuf->ia_ino, &stbuf->ia_ino);
-        if (local->loc.parent) {
-                preparent->ia_ino = local->loc.parent->ino;
-                postparent->ia_ino = local->loc.parent->ino;
+            ret = dht_build_parent_loc(this, &local->loc, &local->loc2,
+                                       &op_errno);
+
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_LOC_FAILED,
+                       "parent loc build failed");
+                goto out;
+            }
+
+            subvol = dht_subvol_get_hashed(this, &local->loc2);
+
+            ret = dht_create_lock(frame, subvol);
+            if (ret < 0) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INODE_LK_ERROR,
+                       "locking parent failed");
+                goto out;
+            }
 
-                WIPE (preparent);
-                WIPE (postparent);
+            return 0;
         }
 
-        ret = dht_layout_preset (this, prev->this, inode);
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"could not set preset layout for subvol %s",
-                        prev->this->name);
-		op_ret   = -1;
-		op_errno = EINVAL;
-		goto out;
-	}
+        goto out;
+    }
+
+    prev = cookie;
+
+    if (local->loc.parent) {
+        dht_inode_ctx_time_update(local->loc.parent, this, preparent, 0);
+
+        dht_inode_ctx_time_update(local->loc.parent, this, postparent, 1);
+    }
+
+    ret = dht_fd_ctx_set(this, fd, prev);
+    if (ret != 0) {
+        gf_msg_debug(this->name, 0,
+                     "Possible fd leak. "
+                     "Could not set fd ctx for subvol %s",
+                     prev->name);
+    }
+
+    ret = dht_layout_preset(this, prev, inode);
+    if (ret != 0) {
+        gf_msg_debug(this->name, 0, "could not set preset layout for subvol %s",
+                     prev->name);
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    local->op_errno = op_errno;
 
+    if (local->linked == _gf_true) {
+        local->stbuf = *stbuf;
+        dht_linkfile_attr_heal(frame, this);
+    }
 out:
-	DHT_STACK_UNWIND (create, frame, op_ret, op_errno, fd, inode, stbuf, preparent,
-                          postparent);
-	return 0;
+
+    DHT_STRIP_PHASE1_FLAGS(stbuf);
+    dht_set_fixed_dir_stat(preparent);
+    dht_set_fixed_dir_stat(postparent);
+
+    if (local && local->lock[0].layout.parent_layout.locks) {
+        /* store op_errno for failure case*/
+        local->op_errno = op_errno;
+        local->refresh_layout_unlock(frame, this, op_ret, 1);
+
+        if (op_ret == 0) {
+            DHT_STACK_UNWIND(create, frame, op_ret, op_errno, fd, inode, stbuf,
+                             preparent, postparent, xdata);
+        }
+    } else {
+        DHT_STACK_UNWIND(create, frame, op_ret, op_errno, fd, inode, stbuf,
+                         preparent, postparent, xdata);
+    }
+    return 0;
 }
 
+static int
+dht_create_linkfile_create_cbk(call_frame_t *frame, void *cookie,
+                               xlator_t *this, int32_t op_ret, int32_t op_errno,
+                               inode_t *inode, struct iatt *stbuf,
+                               struct iatt *preparent, struct iatt *postparent,
+                               dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *cached_subvol = NULL;
+    dht_conf_t *conf = NULL;
+
+    local = frame->local;
+    if (!local) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (op_ret == -1) {
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    conf = this->private;
+    if (!conf) {
+        local->op_errno = EINVAL;
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    cached_subvol = local->cached_subvol;
+
+    if (local->params) {
+        dict_del(local->params, conf->link_xattr_name);
+        dict_del(local->params, GLUSTERFS_INTERNAL_FOP_KEY);
+    }
+
+    STACK_WIND_COOKIE(frame, dht_create_cbk, cached_subvol, cached_subvol,
+                      cached_subvol->fops->create, &local->loc, local->flags,
+                      local->mode, local->umask, local->fd, local->params);
+
+    return 0;
+err:
+    if (local && local->lock[0].layout.parent_layout.locks) {
+        local->refresh_layout_unlock(frame, this, -1, 1);
+    } else {
+        DHT_STACK_UNWIND(create, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                         NULL, NULL);
+    }
+    return 0;
+}
 
-int
-dht_create_linkfile_create_cbk (call_frame_t *frame, void *cookie,
-				xlator_t *this,
-				int32_t op_ret, int32_t op_errno,
-                                inode_t *inode, struct iatt *stbuf,
-                                struct iatt *preparent, struct iatt *postparent)
+static int
+dht_create_wind_to_avail_subvol(call_frame_t *frame, xlator_t *this,
+                                xlator_t *subvol, loc_t *loc, int32_t flags,
+                                mode_t mode, mode_t umask, fd_t *fd,
+                                dict_t *params)
 {
-	dht_local_t  *local = NULL;
-	xlator_t     *cached_subvol = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *avail_subvol = NULL;
 
-        if (op_ret == -1)
-                goto err;
+    local = frame->local;
 
-	local = frame->local;
-	cached_subvol = local->cached_subvol;
+    if (!dht_is_subvol_filled(this, subvol)) {
+        gf_msg_debug(this->name, 0, "creating %s on %s", loc->path,
+                     subvol->name);
 
-        STACK_WIND (frame, dht_create_cbk,
-                    cached_subvol, cached_subvol->fops->create,
-                    &local->loc, local->flags, local->mode,
-                    local->fd, local->params);
+        dht_set_parent_layout_in_dict(loc, this, local);
 
-        return 0;
- err:
- 	DHT_STACK_UNWIND (create, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
- 	return 0;
+        STACK_WIND_COOKIE(frame, dht_create_cbk, subvol, subvol,
+                          subvol->fops->create, loc, flags, mode, umask, fd,
+                          params);
+
+    } else {
+        avail_subvol = dht_free_disk_available_subvol(this, subvol, local);
+
+        if (avail_subvol != subvol) {
+            local->cached_subvol = avail_subvol;
+            local->hashed_subvol = subvol;
+
+            gf_msg_debug(this->name, 0, "creating %s on %s (link at %s)",
+                         loc->path, avail_subvol->name, subvol->name);
+
+            dht_linkfile_create(frame, dht_create_linkfile_create_cbk, this,
+                                avail_subvol, subvol, loc);
+
+            goto out;
+        }
+
+        gf_msg_debug(this->name, 0, "creating %s on %s", loc->path,
+                     subvol->name);
+
+        dht_set_parent_layout_in_dict(loc, this, local);
+
+        STACK_WIND_COOKIE(frame, dht_create_cbk, subvol, subvol,
+                          subvol->fops->create, loc, flags, mode, umask, fd,
+                          params);
+    }
+out:
+    return 0;
 }
 
 int
-dht_create (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc, int32_t flags, mode_t mode,
-            fd_t *fd, dict_t *params)
-{
-	int          op_errno = -1;
-        int          ret = -1;
-	xlator_t    *subvol = NULL;
-	dht_conf_t  *conf = NULL;
-        dht_local_t *local = NULL;
-        xlator_t    *avail_subvol = NULL;
-
-	VALIDATE_OR_GOTO (frame, err);
-	VALIDATE_OR_GOTO (this, err);
-	VALIDATE_OR_GOTO (loc, err);
-
-	conf = this->private;
-
-        dht_get_du_info (frame, this, loc);
-
-	local = dht_local_init (frame);
-	if (!local) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		op_errno = ENOMEM;
-		goto err;
-	}
-
-        if (dht_filter_loc_subvol_key (this, loc, &local->loc,
-                                       &subvol)) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "creating %s on %s (got create on %s)",
-                        local->loc.path, subvol->name, loc->path);
-                STACK_WIND (frame, dht_create_cbk,
-                            subvol, subvol->fops->create,
-                            &local->loc, flags, mode, fd, params);
-                goto done;
+dht_build_parent_loc(xlator_t *this, loc_t *parent, loc_t *child,
+                     int32_t *op_errno)
+{
+    inode_table_t *table = NULL;
+    int ret = -1;
+
+    if (!parent || !child) {
+        if (op_errno)
+            *op_errno = EINVAL;
+        goto out;
+    }
+
+    if (child->parent) {
+        parent->inode = inode_ref(child->parent);
+        if (!parent->inode) {
+            if (op_errno)
+                *op_errno = EINVAL;
+            goto out;
         }
 
-        ret = loc_dup (loc, &local->loc);
-        if (ret == -1) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                goto err;
-        }
-        subvol = dht_subvol_get_hashed (this, loc);
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no subvolume in layout for path=%s",
-                        loc->path);
-                op_errno = ENOENT;
-                goto err;
+        gf_uuid_copy(parent->gfid, child->pargfid);
+
+        ret = 0;
+
+        goto out;
+    } else {
+        if (gf_uuid_is_null(child->pargfid)) {
+            if (op_errno)
+                *op_errno = EINVAL;
+            goto out;
         }
 
-        if (!dht_is_subvol_filled (this, subvol)) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "creating %s on %s", loc->path, subvol->name);
-                STACK_WIND (frame, dht_create_cbk,
-                            subvol, subvol->fops->create,
-                            loc, flags, mode, fd, params);
-                goto done;
+        table = this->itable;
+
+        if (!table) {
+            if (op_errno) {
+                *op_errno = EINVAL;
+                goto out;
+            }
         }
-        /* Choose the minimum filled volume, and create the 
-           files there */
-        /* TODO */
-        avail_subvol = dht_free_disk_available_subvol (this, subvol);
-        if (avail_subvol != subvol) {
-                local->fd = fd_ref (fd);
-                local->params = dict_ref (params);
-                local->flags = flags;
-                local->mode = mode;
 
-                local->cached_subvol = avail_subvol;
-                local->hashed_subvol = subvol;
-                gf_log (this->name, GF_LOG_TRACE,
-                        "creating %s on %s (link at %s)", loc->path,
-                        avail_subvol->name, subvol->name);
-                dht_linkfile_create (frame,
-                                     dht_create_linkfile_create_cbk,
-                                     avail_subvol, subvol, loc);
-                goto done;
+        parent->inode = inode_find(table, child->pargfid);
+
+        if (!parent->inode) {
+            if (op_errno) {
+                *op_errno = EINVAL;
+                goto out;
+            }
         }
-        gf_log (this->name, GF_LOG_TRACE,
-                "creating %s on %s", loc->path, subvol->name);
-        STACK_WIND (frame, dht_create_cbk,
-                    subvol, subvol->fops->create,
-                    loc, flags, mode, fd, params);
-done:
-	return 0;
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (create, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+        gf_uuid_copy(parent->gfid, child->pargfid);
 
-	return 0;
-}
+        ret = 0;
+    }
 
+out:
+    return ret;
+}
 
-int
-dht_mkdir_selfheal_cbk (call_frame_t *frame, void *cookie,
-			xlator_t *this,
-			int32_t op_ret, int32_t op_errno)
+static int32_t
+dht_create_do(call_frame_t *frame)
 {
-	dht_local_t   *local = NULL;
-	dht_layout_t  *layout = NULL;
+    dht_local_t *local = NULL;
+    dht_layout_t *refreshed = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *this = NULL;
+    dht_conf_t *conf = NULL;
+    dht_methods_t *methods = NULL;
 
+    local = frame->local;
 
-	local = frame->local;
-	layout = local->selfheal.layout;
+    this = THIS;
 
-	if (op_ret == 0) {
-                dht_layout_set (this, local->inode, layout);
-		local->stbuf.ia_ino = local->ia_ino;
-                if (local->loc.parent) {
-                        local->preparent.ia_ino = local->loc.parent->ino;
-                        local->postparent.ia_ino = local->loc.parent->ino;
+    conf = this->private;
 
-                        WIPE (&local->preparent);
-                        WIPE (&local->postparent);
-                }
-	}
+    GF_VALIDATE_OR_GOTO(this->name, conf, err);
 
-	DHT_STACK_UNWIND (mkdir, frame, op_ret, op_errno,
-			  local->inode, &local->stbuf, &local->preparent,
-                          &local->postparent);
+    methods = &(conf->methods);
 
-	return 0;
-}
+    /* We don't need parent_loc anymore */
+    loc_wipe(&local->loc);
 
-int
-dht_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-               int op_ret, int op_errno, inode_t *inode, struct iatt *stbuf,
-               struct iatt *preparent, struct iatt *postparent)
-{
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = 0;
-	int           ret = -1;
-        int           subvol_filled = 0;
-	call_frame_t *prev = NULL;
-	dht_layout_t *layout = NULL;
-	dht_conf_t   *conf = NULL;
-
-	conf = this->private;
-	local = frame->local;
-	prev  = cookie;
-	layout = local->layout;
-
-        subvol_filled = dht_is_subvol_filled (this, prev->this);
-
-	LOCK (&frame->lock);
-	{
-                if (subvol_filled && (op_ret != -1)) {
-                        ret = dht_layout_merge (this, layout, prev->this,
-                                                -1, ENOSPC, NULL);
-                } else {
-                        ret = dht_layout_merge (this, layout, prev->this,
-                                                op_ret, op_errno, NULL);
-                }
+    loc_copy(&local->loc, &local->loc2);
 
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			goto unlock;
-		}
-		dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
-                dht_iatt_merge (this, &local->preparent, preparent, prev->this);
-                dht_iatt_merge (this, &local->postparent, postparent,
-                                prev->this);
-
-                if (prev->this == dht_first_up_subvol (this)) {
-                        local->ia_ino = local->stbuf.ia_ino;
-                }
+    loc_wipe(&local->loc2);
 
-	}
-unlock:
-	UNLOCK (&frame->lock);
+    refreshed = local->selfheal.refreshed_layout;
 
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt)) {
-		dht_selfheal_new_directory (frame, dht_mkdir_selfheal_cbk,
-					    layout);
-	}
+    subvol = methods->layout_search(this, refreshed, local->loc.name);
 
-        return 0;
-}
+    if (!subvol) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+               "no subvolume in "
+               "layout for path=%s",
+               local->loc.path);
+        local->op_errno = ENOENT;
+        goto err;
+    }
 
-int
-dht_mkdir_hashed_cbk (call_frame_t *frame, void *cookie, 
-		      xlator_t *this, int op_ret, int op_errno,
-                      inode_t *inode, struct iatt *stbuf,
-                      struct iatt *preparent, struct iatt *postparent)
-{
-	dht_local_t  *local = NULL;
-	int           ret = -1;
-	call_frame_t *prev = NULL;
-	dht_layout_t *layout = NULL;
-	dht_conf_t   *conf = NULL;
-	int           i = 0;
-	xlator_t     *hashed_subvol = NULL;
-
-        VALIDATE_OR_GOTO (this->private, err);
-
-	local = frame->local;
-	prev  = cookie;
-	layout = local->layout;
-	conf = this->private;
-	hashed_subvol = local->hashed_subvol;
-
-        if (uuid_is_null (local->loc.inode->gfid) && !op_ret)
-                memcpy (local->loc.inode->gfid, stbuf->ia_gfid, 16);
-
-        if (dht_is_subvol_filled (this, hashed_subvol))
-                ret = dht_layout_merge (this, layout, prev->this,
-                                        -1, ENOSPC, NULL);
-        else
-                ret = dht_layout_merge (this, layout, prev->this,
-                                        op_ret, op_errno, NULL);
-        
-	if (op_ret == -1) {
-		local->op_errno = op_errno;
-		goto err;
-	}
-	local->op_ret = 0;
-
-	dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
-        dht_iatt_merge (this, &local->preparent, preparent, prev->this);
-        dht_iatt_merge (this, &local->postparent, postparent, prev->this);
-
-	local->ia_ino = local->stbuf.ia_ino;
-
-	local->call_cnt = conf->subvolume_cnt - 1;
-	
-	if (local->call_cnt == 0) {
-		dht_selfheal_directory (frame, dht_mkdir_selfheal_cbk,
-					&local->loc, layout);
-	}
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		if (conf->subvolumes[i] == hashed_subvol)
-			continue;
-		STACK_WIND (frame, dht_mkdir_cbk,
-			    conf->subvolumes[i],
-			    conf->subvolumes[i]->fops->mkdir,
-			    &local->loc, local->mode, local->params);
-	}
-	return 0;
+    dht_create_wind_to_avail_subvol(frame, this, subvol, &local->loc,
+                                    local->flags, local->mode, local->umask,
+                                    local->fd, local->params);
+    return 0;
 err:
-	DHT_STACK_UNWIND (mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL);
-        return 0;
+    local->refresh_layout_unlock(frame, this, -1, 1);
+
+    return 0;
+}
+
+static int32_t
+dht_create_unlock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    DHT_STACK_DESTROY(frame);
+    return 0;
 }
 
+static int32_t
+dht_create_finish(call_frame_t *frame, xlator_t *this, int op_ret,
+                  int invoke_cbk)
+{
+    dht_local_t *local = NULL, *lock_local = NULL;
+    call_frame_t *lock_frame = NULL;
+    int lock_count = 0;
+
+    local = frame->local;
+    lock_count = dht_lock_count(local->lock[0].layout.parent_layout.locks,
+                                local->lock[0].layout.parent_layout.lk_count);
+    if (lock_count == 0)
+        goto done;
+
+    lock_frame = copy_frame(frame);
+    if (lock_frame == NULL) {
+        goto done;
+    }
+
+    lock_local = dht_local_init(lock_frame, &local->loc, NULL,
+                                lock_frame->root->op);
+    if (lock_local == NULL) {
+        goto done;
+    }
+
+    lock_local->lock[0]
+        .layout.parent_layout.locks = local->lock[0].layout.parent_layout.locks;
+    lock_local->lock[0].layout.parent_layout.lk_count =
+        local->lock[0].layout.parent_layout.lk_count;
+
+    local->lock[0].layout.parent_layout.locks = NULL;
+    local->lock[0].layout.parent_layout.lk_count = 0;
+
+    dht_unlock_inodelk(lock_frame,
+                       lock_local->lock[0].layout.parent_layout.locks,
+                       lock_local->lock[0].layout.parent_layout.lk_count,
+                       dht_create_unlock_cbk);
+    lock_frame = NULL;
 
-int
-dht_mkdir (call_frame_t *frame, xlator_t *this,
-	   loc_t *loc, mode_t mode, dict_t *params)
-{
-	dht_local_t  *local  = NULL;
-	dht_conf_t   *conf = NULL;
-        int           op_errno = -1;
-	int           ret = -1;
-	xlator_t     *hashed_subvol = NULL;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (this->private, err);
-
-	conf = this->private;
-
-        dht_get_du_info (frame, this, loc);
-
-	local = dht_local_init (frame);
-	if (!local) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		op_errno = ENOMEM;
-		goto err;
-	}
-
-	hashed_subvol = dht_subvol_get_hashed (this, loc);
-
-	if (hashed_subvol == NULL) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"hashed subvol not found for %s",
-                        loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	local->hashed_subvol = hashed_subvol;
-	local->inode = inode_ref (loc->inode);
-	ret = loc_copy (&local->loc, loc);
-	local->mode = mode;
-
-	if (ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		op_errno = ENOMEM;
-		goto err;
-	}
-
-        local->params = dict_ref (params);
-
-	local->layout = dht_layout_new (this, conf->subvolume_cnt);
-	if (!local->layout) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		op_errno = ENOMEM;
-		goto err;
-	}
-
-	STACK_WIND (frame, dht_mkdir_hashed_cbk,
-		    hashed_subvol,
-		    hashed_subvol->fops->mkdir,
-		    loc, mode, params);
-
-	return 0;
+done:
+    if (lock_frame != NULL) {
+        DHT_STACK_DESTROY(lock_frame);
+    }
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+    if (op_ret == 0)
+        return 0;
 
-	return 0;
+    DHT_STACK_UNWIND(create, frame, op_ret, local->op_errno, NULL, NULL, NULL,
+                     NULL, NULL, NULL);
+    return 0;
 }
 
-
-int
-dht_rmdir_selfheal_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int op_ret, int op_errno)
+static int32_t
+dht_create_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	dht_local_t  *local = NULL;
+    dht_local_t *local = NULL;
 
-	local = frame->local;
+    local = frame->local;
 
-        if (local->loc.parent) {
-                local->preparent.ia_ino = local->loc.parent->ino;
-                local->postparent.ia_ino = local->loc.parent->ino;
-        }
+    if (!local) {
+        goto err;
+    }
 
-	DHT_STACK_UNWIND (rmdir, frame, local->op_ret, local->op_errno,
-                          &local->preparent, &local->postparent);
+    if (op_ret < 0) {
+        gf_msg("DHT", GF_LOG_ERROR, 0, DHT_MSG_INODE_LK_ERROR,
+               "Create lock failed for file: %s", local->loc2.name);
 
-	return 0;
-}
+        local->op_errno = op_errno;
 
+        goto err;
+    }
 
-int
-dht_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-	       int op_ret, int op_errno, struct iatt *preparent,
-               struct iatt *postparent)
+    local->refresh_layout_unlock = dht_create_finish;
+
+    local->refresh_layout_done = dht_create_do;
+
+    dht_refresh_layout(frame);
+
+    return 0;
+err:
+    if (local)
+        dht_create_finish(frame, this, -1, 0);
+    else
+        DHT_STACK_UNWIND(create, frame, -1, EINVAL, NULL, NULL, NULL, NULL,
+                         NULL, NULL);
+    return 0;
+}
+
+int32_t
+dht_create_lock(call_frame_t *frame, xlator_t *subvol)
 {
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = 0;
-	call_frame_t *prev = NULL;
-
-	local = frame->local;
-	prev  = cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			local->op_ret   = -1;
-
-			if (op_errno != ENOENT)
-				local->need_selfheal = 1;
-
-			gf_log (this->name, GF_LOG_DEBUG,
-				"rmdir on %s for %s failed (%s)",
-				prev->this->name, local->loc.path,
-				strerror (op_errno));
-			goto unlock;
-		}
-
-                dht_iatt_merge (this, &local->preparent, preparent, prev->this);
-                dht_iatt_merge (this, &local->postparent, postparent,
-                                prev->this);
-	}
-unlock:
-	UNLOCK (&frame->lock);
+    dht_local_t *local = NULL;
+    int count = 1, ret = -1;
+    dht_lock_t **lk_array = NULL;
 
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO(frame->this->name, frame->local, err);
 
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt)) {
-		if (local->need_selfheal) {
-                        local->layout =
-                                dht_layout_get (this, local->loc.inode);
+    local = frame->local;
 
-			/* TODO: neater interface needed below */
-			local->stbuf.ia_type = local->loc.inode->ia_type;
+    lk_array = GF_CALLOC(count, sizeof(*lk_array), gf_common_mt_pointer);
 
-			dht_selfheal_restore (frame, dht_rmdir_selfheal_cbk,
-					      &local->loc, local->layout);
-		} else {
-                        if (local->loc.parent) {
-                                local->preparent.ia_ino =
-                                        local->loc.parent->ino;
-                                local->postparent.ia_ino =
-                                        local->loc.parent->ino;
+    if (lk_array == NULL)
+        goto err;
 
-                                WIPE (&local->preparent);
-                                WIPE (&local->postparent);
-                        }
+    lk_array[0] = dht_lock_new(frame->this, subvol, &local->loc, F_RDLCK,
+                               DHT_LAYOUT_HEAL_DOMAIN, NULL,
+                               IGNORE_ENOENT_ESTALE);
 
-			DHT_STACK_UNWIND (rmdir, frame, local->op_ret,
-					  local->op_errno, &local->preparent,
-                                          &local->postparent);
-		}
-	}
+    if (lk_array[0] == NULL)
+        goto err;
 
-        return 0;
+    local->lock[0].layout.parent_layout.locks = lk_array;
+    local->lock[0].layout.parent_layout.lk_count = count;
+
+    ret = dht_blocking_inodelk(frame, lk_array, count, dht_create_lock_cbk);
+
+    if (ret < 0) {
+        local->lock[0].layout.parent_layout.locks = NULL;
+        local->lock[0].layout.parent_layout.lk_count = 0;
+        goto err;
+    }
+
+    return 0;
+err:
+    if (lk_array != NULL) {
+        dht_lock_array_free(lk_array, count);
+        GF_FREE(lk_array);
+    }
+
+    return -1;
 }
 
+int
+dht_set_parent_layout_in_dict(loc_t *loc, xlator_t *this, dht_local_t *local)
+{
+    dht_conf_t *conf = this->private;
+    dht_layout_t *parent_layout = NULL;
+    int *parent_disk_layout = NULL;
+    xlator_t *hashed_subvol = NULL;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    int ret = 0;
+
+    gf_uuid_unparse(loc->parent->gfid, pgfid);
+
+    parent_layout = dht_layout_get(this, loc->parent);
+    hashed_subvol = dht_subvol_get_hashed(this, loc);
+
+    ret = dht_disk_layout_extract_for_subvol(this, parent_layout, hashed_subvol,
+                                             &parent_disk_layout);
+    if (ret == -1) {
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "%s (%s/%s) (path: %s): "
+               "extracting in-memory layout of parent failed. ",
+               gf_fop_list[local->fop], pgfid, loc->name, loc->path);
+        goto err;
+    }
+
+    ret = dict_set_str_sizen(local->params, GF_PREOP_PARENT_KEY,
+                             conf->xattr_name);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "%s (%s/%s) (path: %s): "
+               "setting %s key in params dictionary failed. ",
+               gf_fop_list[local->fop], pgfid, loc->name, loc->path,
+               GF_PREOP_PARENT_KEY);
+        goto err;
+    }
+
+    ret = dict_set_bin(local->params, conf->xattr_name, parent_disk_layout,
+                       4 * 4);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "%s (%s/%s) (path: %s): "
+               "setting parent-layout in params dictionary failed. ",
+               gf_fop_list[local->fop], pgfid, loc->name, loc->path);
+        goto err;
+    }
+
+err:
+    dht_layout_unref(this, parent_layout);
+    return ret;
+}
 
 int
-dht_rmdir_do (call_frame_t *frame, xlator_t *this)
+dht_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+           mode_t mode, mode_t umask, fd_t *fd, dict_t *params)
 {
-	dht_local_t  *local = NULL;
-	dht_conf_t   *conf = NULL;
-	int           i = 0;
+    int op_errno = -1;
+    xlator_t *subvol = NULL;
+    xlator_t *hashed_subvol = NULL;
+    dht_local_t *local = NULL;
+    int i = 0;
+    dht_conf_t *conf = NULL;
+    int ret = 0;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+
+    conf = this->private;
+
+    dht_get_du_info(frame, this, loc);
+
+    local = dht_local_init(frame, loc, fd, GF_FOP_CREATE);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->params = dict_ref(params);
+    local->flags = flags;
+    local->mode = mode;
+    local->umask = umask;
+
+    if (dht_filter_loc_subvol_key(this, loc, &local->loc, &subvol)) {
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_INFO,
+               "creating %s on %s (got create on %s)", local->loc.path,
+               subvol->name, loc->path);
+
+        /* Since lookup-optimize is enabled by default, we need
+         * to create the linkto file if required.
+         * Note this does not check for decommisioned bricks
+         * and min-free-disk limits as this is a debugging tool
+         * and not expected to be used in production.
+         */
+        hashed_subvol = dht_subvol_get_hashed(this, &local->loc);
+
+        if (hashed_subvol && (hashed_subvol != subvol)) {
+            /* Create the linkto file and then the data file */
+            local->cached_subvol = subvol;
+            local->hashed_subvol = hashed_subvol;
+
+            dht_linkfile_create(frame, dht_create_linkfile_create_cbk, this,
+                                subvol, hashed_subvol, &local->loc);
+            goto done;
+        }
+        /* We either don't have a hashed subvol or the hashed subvol is
+         * the same as the one specified. No need to create the linkto
+         * file as we expect a lookup everywhere if there are problems
+         * with the parent layout
+         */
+
+        dht_set_parent_layout_in_dict(loc, this, local);
+
+        STACK_WIND_COOKIE(frame, dht_create_cbk, subvol, subvol,
+                          subvol->fops->create, &local->loc, flags, mode, umask,
+                          fd, params);
+        goto done;
+    }
+
+    subvol = dht_subvol_get_hashed(this, loc);
+    if (!subvol) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+               "no subvolume in layout for path=%s", loc->path);
+
+        op_errno = EIO;
+        goto err;
+    }
+
+    /* Post remove-brick, the client layout may not be in sync with
+     * disk layout because of lack of lookup. Hence,a create call
+     * may fall on the decommissioned brick.  Hence, if the
+     * hashed_subvol is part of decommissioned bricks  list, do a
+     * lookup on parent dir. If a fix-layout is already done by the
+     * remove-brick process, the parent directory layout will be in
+     * sync with that of the disk. If fix-layout is still ending
+     * on the parent directory, we can let the file get created on
+     * the decommissioned brick which will be eventually migrated to
+     * non-decommissioned brick based on the new layout.
+     */
+
+    if (conf->decommission_subvols_cnt) {
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (conf->decommissioned_bricks[i] &&
+                conf->decommissioned_bricks[i] == subvol) {
+                gf_msg_debug(this->name, 0,
+                             "hashed subvol:%s is "
+                             "part of decommission brick list for "
+                             "file: %s",
+                             subvol->name, loc->path);
+
+                /* dht_refresh_layout needs directory info in
+                 * local->loc. Hence, storing the parent_loc in
+                 * local->loc and storing the create context in
+                 * local->loc2. We will restore this information
+                 * in dht_creation do */
+
+                ret = loc_copy(&local->loc2, &local->loc);
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+                           "loc_copy failed %s", loc->path);
+
+                    goto err;
+                }
 
-        VALIDATE_OR_GOTO (this->private, err);
+                loc_wipe(&local->loc);
 
-	conf = this->private;
-	local = frame->local;
+                ret = dht_build_parent_loc(this, &local->loc, loc, &op_errno);
 
-	if (local->op_ret == -1)
-		goto err;
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_LOC_FAILED,
+                           "parent loc build failed");
+                    goto err;
+                }
 
-	local->call_cnt = conf->subvolume_cnt;
+                ret = dht_create_lock(frame, subvol);
 
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		STACK_WIND (frame, dht_rmdir_cbk,
-			    conf->subvolumes[i],
-			    conf->subvolumes[i]->fops->rmdir,
-			    &local->loc, local->flags);
-	}
+                if (ret < 0) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INODE_LK_ERROR,
+                           "locking parent failed");
+                    goto err;
+                }
 
-	return 0;
+                goto done;
+            }
+        }
+    }
+
+    dht_create_wind_to_avail_subvol(frame, this, subvol, loc, flags, mode,
+                                    umask, fd, params);
+done:
+    return 0;
 
 err:
-	DHT_STACK_UNWIND (rmdir, frame, local->op_ret, local->op_errno,
-                          &local->preparent, &local->postparent);
-	return 0;
-}
 
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(create, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL,
+                     NULL);
 
-int
-dht_rmdir_linkfile_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                               int op_ret, int op_errno, struct iatt *preparent,
-                               struct iatt *postparent)
+    return 0;
+}
+
+static int
+dht_mkdir_selfheal_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        dht_local_t    *local = NULL;
-        call_frame_t   *prev = NULL;
-        xlator_t       *src = NULL;
-        call_frame_t   *main_frame = NULL;
-        dht_local_t    *main_local = NULL;
-        int             this_call_cnt = 0;
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
 
-        local  = frame->local;
-        prev   = cookie;
-        src    = prev->this;
+    local = frame->local;
+    layout = local->selfheal.layout;
 
-        main_frame = local->main_frame;
-        main_local = main_frame->local;
+    FRAME_SU_UNDO(frame, dht_local_t);
+    dht_set_fixed_dir_stat(&local->preparent);
+    dht_set_fixed_dir_stat(&local->postparent);
 
-        if (op_ret == 0) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "unlinked linkfile %s on %s",
-                        local->loc.path, src->name);
+    if (op_ret == 0) {
+        dht_layout_set(this, local->inode, layout);
+
+        dht_inode_ctx_time_update(local->inode, this, &local->stbuf, 1);
+        if (local->loc.parent) {
+            dht_inode_ctx_time_update(local->loc.parent, this,
+                                      &local->preparent, 0);
+
+            dht_inode_ctx_time_update(local->loc.parent, this,
+                                      &local->postparent, 1);
+        }
+    }
+
+    DHT_STACK_UNWIND(mkdir, frame, op_ret, op_errno, local->inode,
+                     &local->stbuf, &local->preparent, &local->postparent,
+                     NULL);
+
+    return 0;
+}
+
+static int
+dht_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, inode_t *inode, struct iatt *stbuf,
+              struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    int ret = -1;
+    gf_boolean_t subvol_filled = _gf_false;
+    gf_boolean_t dir_exists = _gf_false;
+    xlator_t *prev = NULL;
+    dht_layout_t *layout = NULL;
+
+    local = frame->local;
+    prev = cookie;
+    layout = local->layout;
+
+    subvol_filled = dht_is_subvol_filled(this, prev);
+
+    LOCK(&frame->lock);
+    {
+        if (subvol_filled && (op_ret != -1)) {
+            ret = dht_layout_merge(this, layout, prev, -1, ENOSPC, NULL);
         } else {
-                main_local->op_ret   = -1;
-                main_local->op_errno = op_errno;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "unlink of %s on %s failed (%s)",
-                        local->loc.path, src->name, strerror (op_errno));
+            if (op_ret == -1 && op_errno == EEXIST) {
+                /* Very likely just a race between mkdir and
+                   self-heal (from lookup of a concurrent mkdir
+                   attempt).
+                   Ignore error for now. layout setting will
+                   anyways fail if this was a different (old)
+                   pre-existing different directory.
+                */
+                op_ret = 0;
+                dir_exists = _gf_true;
+            }
+            ret = dht_layout_merge(this, layout, prev, op_ret, op_errno, NULL);
+        }
+        if (ret)
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_MERGE_FAILED,
+                   "%s: failed to merge layouts for subvol %s", local->loc.path,
+                   prev->name);
+
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+            goto unlock;
         }
 
-        this_call_cnt = dht_frame_return (main_frame);
-        if (is_last_call (this_call_cnt))
-                dht_rmdir_do (main_frame, this);
+        if (dir_exists)
+            goto unlock;
 
-        DHT_STACK_DESTROY (frame);
-        return 0;
+        dht_iatt_merge(this, &local->stbuf, stbuf);
+        dht_iatt_merge(this, &local->preparent, preparent);
+        dht_iatt_merge(this, &local->postparent, postparent);
+    }
+unlock:
+    UNLOCK(&frame->lock);
+
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        /*Unlock entrylk and inodelk once mkdir is done on all subvols*/
+        dht_unlock_namespace(frame, &local->lock[0]);
+        FRAME_SU_DO(frame, dht_local_t);
+        dht_selfheal_new_directory(frame, dht_mkdir_selfheal_cbk, layout);
+    }
+
+    return 0;
 }
 
+static int
+dht_mkdir_hashed_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, inode_t *inode,
+                     struct iatt *stbuf, struct iatt *preparent,
+                     struct iatt *postparent, dict_t *xdata);
 
-int
-dht_rmdir_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int op_ret, int op_errno, inode_t *inode,
-                      struct iatt *stbuf, dict_t *xattr, struct iatt *parent)
+static int
+dht_mkdir_helper(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+                 mode_t umask, dict_t *params)
 {
-        dht_local_t    *local = NULL;
-        call_frame_t   *prev = NULL;
-        xlator_t       *src = NULL;
-        call_frame_t   *main_frame = NULL;
-        dht_local_t    *main_local = NULL;
-        int             this_call_cnt = 0;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int op_errno = -1, ret = -1;
+    xlator_t *hashed_subvol = NULL;
+    int32_t *parent_disk_layout = NULL;
+    dht_layout_t *parent_layout = NULL;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+    VALIDATE_OR_GOTO(loc->path, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    gf_uuid_unparse(loc->parent->gfid, pgfid);
+
+    conf = this->private;
+    local = frame->local;
+
+    if (local->op_ret == -1) {
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "mkdir (%s/%s) (path: %s): refreshing parent layout "
+               "failed.",
+               pgfid, loc->name, loc->path);
+
+        op_errno = local->op_errno;
+        goto err;
+    }
+
+    local->op_ret = -1;
+
+    hashed_subvol = dht_subvol_get_hashed(this, loc);
+    if (hashed_subvol == NULL) {
+        gf_msg_debug(this->name, 0,
+                     "mkdir (%s/%s) (path: %s): hashed subvol not "
+                     "found",
+                     pgfid, loc->name, loc->path);
+        op_errno = ENOENT;
+        goto err;
+    }
+
+    local->hashed_subvol = hashed_subvol;
+
+    parent_layout = dht_layout_get(this, loc->parent);
+
+    ret = dht_disk_layout_extract_for_subvol(this, parent_layout, hashed_subvol,
+                                             &parent_disk_layout);
+    if (ret == -1) {
+        gf_msg(this->name, GF_LOG_WARNING, EIO, DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "mkdir (%s/%s) (path: %s): "
+               "extracting in-memory layout of parent failed. ",
+               pgfid, loc->name, loc->path);
+        goto err;
+    }
+
+    if (memcmp(local->parent_disk_layout, parent_disk_layout,
+               sizeof(local->parent_disk_layout)) == 0) {
+        gf_msg(this->name, GF_LOG_WARNING, EIO, DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "mkdir (%s/%s) (path: %s): loop detected. "
+               "parent layout didn't change even though "
+               "previous attempt of mkdir failed because of "
+               "in-memory layout not matching with that on disk.",
+               pgfid, loc->name, loc->path);
+        op_errno = EIO;
+        goto err;
+    }
+
+    memcpy((void *)local->parent_disk_layout, (void *)parent_disk_layout,
+           sizeof(local->parent_disk_layout));
+
+    dht_layout_unref(this, parent_layout);
+    parent_layout = NULL;
+
+    ret = dict_set_str(params, GF_PREOP_PARENT_KEY, conf->xattr_name);
+    if (ret < 0) {
+        local->op_errno = -ret;
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "mkdir (%s/%s) (path: %s): "
+               "setting %s key in params dictionary failed. ",
+               pgfid, loc->name, loc->path, GF_PREOP_PARENT_KEY);
+        goto err;
+    }
+
+    ret = dict_set_bin(params, conf->xattr_name, parent_disk_layout, 4 * 4);
+    if (ret < 0) {
+        local->op_errno = -ret;
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "setting parent-layout in params dictionary failed. "
+               "mkdir (%s/%s) (path: %s)",
+               pgfid, loc->name, loc->path);
+        goto err;
+    }
+
+    parent_disk_layout = NULL;
+
+    STACK_WIND_COOKIE(frame, dht_mkdir_hashed_cbk, hashed_subvol, hashed_subvol,
+                      hashed_subvol->fops->mkdir, loc, mode, umask, params);
+
+    return 0;
 
-        local = frame->local;
-        prev  = cookie;
-        src   = prev->this;
+err:
+    dht_unlock_namespace(frame, &local->lock[0]);
 
-        main_frame = local->main_frame;
-        main_local = main_frame->local;
+    op_errno = local ? local->op_errno : op_errno;
+    DHT_STACK_UNWIND(mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
 
-        if (op_ret != 0)
-                goto err;
+    if (parent_disk_layout != NULL)
+        GF_FREE(parent_disk_layout);
+
+    if (parent_layout != NULL)
+        dht_layout_unref(this, parent_layout);
 
-        if (check_is_linkfile (inode, stbuf, xattr) == 0) {
-                main_local->op_ret  = -1;
-                main_local->op_errno = ENOTEMPTY;
+    return 0;
+}
 
-                gf_log (this->name, GF_LOG_WARNING,
-                        "%s on %s found to be not a linkfile (type=0%o)",
-                        local->loc.path, src->name, stbuf->ia_type);
+static int
+dht_mkdir_hashed_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, inode_t *inode,
+                     struct iatt *stbuf, struct iatt *preparent,
+                     struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int ret = -1;
+    xlator_t *prev = NULL;
+    dht_layout_t *layout = NULL;
+    dht_conf_t *conf = NULL;
+    int i = 0;
+    xlator_t *hashed_subvol = NULL;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    gf_boolean_t parent_layout_changed = _gf_false;
+    call_stub_t *stub = NULL;
+
+    local = frame->local;
+    prev = cookie;
+    layout = local->layout;
+    conf = this->private;
+    hashed_subvol = local->hashed_subvol;
+
+    gf_uuid_unparse(local->loc.parent->gfid, pgfid);
+
+    if (gf_uuid_is_null(local->loc.gfid) && !op_ret)
+        gf_uuid_copy(local->loc.gfid, stbuf->ia_gfid);
+
+    if (op_ret == -1) {
+        local->op_errno = op_errno;
+
+        parent_layout_changed = (xdata &&
+                                 dict_get(xdata, GF_PREOP_CHECK_FAILED))
+                                    ? 1
+                                    : 0;
+        if (parent_layout_changed) {
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_PARENT_LAYOUT_CHANGED,
+                   "mkdir (%s/%s) (path: %s): parent layout "
+                   "changed. Attempting a refresh and then a "
+                   "retry",
+                   pgfid, local->loc.name, local->loc.path);
+
+            stub = fop_mkdir_stub(frame, dht_mkdir_helper, &local->loc,
+                                  local->mode, local->umask, local->params);
+            if (stub == NULL) {
                 goto err;
+            }
+
+            ret = dht_handle_parent_layout_change(this, stub);
+            if (ret) {
+                goto err;
+            }
+
+            stub = NULL;
+
+            return 0;
         }
 
-        STACK_WIND (frame, dht_rmdir_linkfile_unlink_cbk,
-                    src, src->fops->unlink, &local->loc);
+        goto err;
+    }
+
+    dict_del(local->params, GF_PREOP_PARENT_KEY);
+    dict_del(local->params, conf->xattr_name);
+
+    if (dht_is_subvol_filled(this, hashed_subvol))
+        ret = dht_layout_merge(this, layout, prev, -1, ENOSPC, NULL);
+    else
+        ret = dht_layout_merge(this, layout, prev, op_ret, op_errno, NULL);
+
+    /* TODO: we may have to return from the function
+       if layout merge fails. For now, lets just log an error */
+    if (ret)
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_MERGE_FAILED,
+               "%s: failed to merge layouts for subvol %s", local->loc.path,
+               prev->name);
+
+    local->op_ret = 0;
+
+    dht_iatt_merge(this, &local->stbuf, stbuf);
+    dht_iatt_merge(this, &local->preparent, preparent);
+    dht_iatt_merge(this, &local->postparent, postparent);
+
+    local->call_cnt = conf->subvolume_cnt - 1;
+    /* Delete internal mds xattr from params dict to avoid store
+      internal mds xattr on other subvols
+    */
+    dict_del(local->params, conf->mds_xattr_key);
+
+    if (gf_uuid_is_null(local->loc.gfid))
+        gf_uuid_copy(local->loc.gfid, stbuf->ia_gfid);
+
+    /* Set hashed subvol as a mds subvol on inode ctx */
+    /*if (!local->inode)
+            local->inode  = inode_ref (inode);
+    */
+    ret = dht_inode_ctx_mdsvol_set(local->inode, this, hashed_subvol);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SET_INODE_CTX_FAILED,
+               "Failed to set hashed subvol for %s on inode vol is %s",
+               local->loc.path, hashed_subvol->name);
+    }
+
+    if (local->call_cnt == 0) {
+        /*Unlock namespace lock once mkdir is done on all subvols*/
+        dht_unlock_namespace(frame, &local->lock[0]);
+        FRAME_SU_DO(frame, dht_local_t);
+        dht_selfheal_directory(frame, dht_mkdir_selfheal_cbk, &local->loc,
+                               layout);
         return 0;
+    }
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (conf->subvolumes[i] == hashed_subvol)
+            continue;
+        STACK_WIND_COOKIE(frame, dht_mkdir_cbk, conf->subvolumes[i],
+                          conf->subvolumes[i], conf->subvolumes[i]->fops->mkdir,
+                          &local->loc, local->mode, local->umask,
+                          local->params);
+    }
+
+    return 0;
 err:
+    if (local->op_ret != 0) {
+        dht_unlock_namespace(frame, &local->lock[0]);
+    }
 
-        this_call_cnt = dht_frame_return (main_frame);
-        if (is_last_call (this_call_cnt))
-                dht_rmdir_do (main_frame, this);
+    DHT_STACK_UNWIND(mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
 
-        DHT_STACK_DESTROY (frame);
-        return 0;
+    return 0;
 }
 
+static int
+dht_mkdir_guard_parent_layout_cbk(call_frame_t *frame, xlator_t *this,
+                                  loc_t *loc, mode_t mode, mode_t umask,
+                                  dict_t *params)
+{
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = 0;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    int ret = -1;
+    int32_t zero[1] = {0};
+
+    local = frame->local;
+    conf = this->private;
+
+    gf_uuid_unparse(loc->parent->gfid, pgfid);
+
+    if (local->op_ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, local->op_errno,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "mkdir (%s/%s) (path: %s): "
+               "Acquiring lock on parent to guard against "
+               "layout-change failed.",
+               pgfid, loc->name, loc->path);
+        goto err;
+    }
+
+    local->op_ret = -1;
+    /* Add internal MDS xattr on disk for hashed subvol
+     */
+    ret = dht_dict_set_array(params, conf->mds_xattr_key, zero, 1);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, DHT_MSG_DICT_SET_FAILED,
+               "Failed to set dictionary value:key = %s for "
+               "path %s",
+               conf->mds_xattr_key, loc->path);
+    }
+
+    STACK_WIND_COOKIE(frame, dht_mkdir_hashed_cbk, local->hashed_subvol,
+                      local->hashed_subvol, local->hashed_subvol->fops->mkdir,
+                      loc, mode, umask, params);
+
+    return 0;
+err:
+    DHT_STACK_UNWIND(mkdir, frame, -1, local->op_errno, NULL, NULL, NULL, NULL,
+                     NULL);
+
+    return 0;
+}
 
 int
-dht_rmdir_is_subvol_empty (call_frame_t *frame, xlator_t *this,
-                           gf_dirent_t *entries, xlator_t *src)
+dht_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+          mode_t umask, dict_t *params)
 {
-        int                 ret = 0;
-        int                 build_ret = 0;
-        gf_dirent_t        *trav = NULL;
-        call_frame_t       *lookup_frame = NULL;
-        dht_local_t        *lookup_local = NULL;
-        dht_local_t        *local = NULL;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int op_errno = EINVAL, ret = -1;
+    xlator_t *hashed_subvol = NULL;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    call_stub_t *stub = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+    VALIDATE_OR_GOTO(loc->path, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    gf_uuid_unparse(loc->parent->gfid, pgfid);
+
+    conf = this->private;
+
+    if (!params || !dict_get(params, "gfid-req")) {
+        op_errno = EPERM;
+        gf_msg_callingfn(this->name, GF_LOG_WARNING, op_errno,
+                         DHT_MSG_GFID_NULL,
+                         "mkdir: %s is received "
+                         "without gfid-req %p",
+                         loc->path, params);
+        goto err;
+    }
+
+    dht_get_du_info(frame, this, loc);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_MKDIR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    hashed_subvol = dht_subvol_get_hashed(this, loc);
+    if (hashed_subvol == NULL) {
+        gf_msg_debug(this->name, 0, "hashed subvol not found for %s",
+                     loc->path);
+        local->op_errno = EIO;
+        goto err;
+    }
+
+    local->hashed_subvol = hashed_subvol;
+    local->mode = mode;
+    local->umask = umask;
+    if (params)
+        local->params = dict_ref(params);
+
+    local->inode = inode_ref(loc->inode);
+
+    local->layout = dht_layout_new(this, conf->subvolume_cnt);
+    if (!local->layout) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    /* set the newly created directory hash to the commit hash
+     * if the configuration option is set. If configuration option
+     * is not set, the older clients may still be connecting to the
+     * volume and hence we need to preserve the 1 in disk[0] part of the
+     * layout xattr */
+    if (conf->lookup_optimize)
+        local->layout->commit_hash = conf->vol_commit_hash;
+    else
+        local->layout->commit_hash = DHT_LAYOUT_HASH_INVALID;
+
+    stub = fop_mkdir_stub(frame, dht_mkdir_guard_parent_layout_cbk, loc, mode,
+                          umask, params);
+    if (stub == NULL) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM,
+               DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "mkdir (%s/%s) (path: %s): "
+               "creating stub failed.",
+               pgfid, loc->name, loc->path);
+        local->op_errno = ENOMEM;
+        goto err;
+    }
+
+    ret = dht_guard_parent_layout_and_namespace(this, stub);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_PARENT_LAYOUT_CHANGED,
+               "mkdir (%s/%s) (path: %s) cannot wind lock request to "
+               "guard parent layout",
+               pgfid, loc->name, loc->path);
+        goto err;
+    }
+
+    return 0;
 
-        local = frame->local;
+err:
+    op_errno = local ? local->op_errno : op_errno;
+    DHT_STACK_UNWIND(mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+
+    return 0;
+}
+
+static int
+dht_rmdir_selfheal_cbk(call_frame_t *heal_frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    dht_local_t *heal_local = NULL;
+    call_frame_t *main_frame = NULL;
+
+    heal_local = heal_frame->local;
+    main_frame = heal_local->main_frame;
+    local = main_frame->local;
 
-        list_for_each_entry (trav, &entries->list, list) {
-                if (strcmp (trav->d_name, ".") == 0)
-                        continue;
-                if (strcmp (trav->d_name, "..") == 0)
-                        continue;
-                if (check_is_linkfile (NULL, (&trav->d_stat), NULL) == 1) {
-                        ret++;
-                        continue;
+    DHT_STACK_DESTROY(heal_frame);
+    dht_set_fixed_dir_stat(&local->preparent);
+    dht_set_fixed_dir_stat(&local->postparent);
+
+    DHT_STACK_UNWIND(rmdir, main_frame, local->op_ret, local->op_errno,
+                     &local->preparent, &local->postparent, NULL);
+
+    return 0;
+}
+
+static int
+dht_rmdir_hashed_subvol_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int op_ret, int op_errno, struct iatt *preparent,
+                            struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    dht_local_t *heal_local = NULL;
+    call_frame_t *heal_frame = NULL;
+    dht_conf_t *conf = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    prev = cookie;
+    conf = this->private;
+
+    gf_uuid_unparse(local->loc.gfid, gfid);
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+            local->op_ret = -1;
+            if (conf->subvolume_cnt != 1) {
+                if (op_errno != ENOENT && op_errno != EACCES &&
+                    op_errno != ESTALE) {
+                    local->need_selfheal = 1;
                 }
+            }
 
-                /* this entry is either a directory which is neither "." nor "..",
-                   or a non directory which is not a linkfile. the directory is to
-                   be treated as non-empty
-                */
-                return 0;
+            gf_msg_debug(this->name, op_errno,
+                         "rmdir on %s for %s failed "
+                         "(gfid = %s)",
+                         prev->name, local->loc.path, gfid);
+            goto unlock;
         }
 
-        list_for_each_entry (trav, &entries->list, list) {
-                if (strcmp (trav->d_name, ".") == 0)
-                        continue;
-                if (strcmp (trav->d_name, "..") == 0)
-                        continue;
+        dht_iatt_merge(this, &local->preparent, preparent);
+        dht_iatt_merge(this, &local->postparent, postparent);
+    }
+unlock:
+    UNLOCK(&frame->lock);
 
-                lookup_frame = NULL;
-                lookup_local = NULL;
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        if (local->need_selfheal) {
+            dht_rmdir_unlock(frame, this);
+            local->layout = dht_layout_get(this, local->loc.inode);
 
-                lookup_frame = copy_frame (frame);
-                if (!lookup_frame) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Out of Memory");
-                        /* out of memory, let the rmdir fail
-                           (as non-empty, unfortunately) */
-                        goto err;
-                }
+            /* TODO: neater interface needed below */
+            local->stbuf.ia_type = local->loc.inode->ia_type;
 
-                lookup_local = GF_CALLOC (sizeof (*local), 1,
-                                          gf_dht_mt_dht_local_t);
-                if (!lookup_local) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Out of Memory");
-                        goto err;
-                }
+            gf_uuid_copy(local->gfid, local->loc.inode->gfid);
 
-                lookup_frame->local = lookup_local;
-                lookup_local->main_frame = frame;
+            /* Use a different frame or else the rmdir op_ret is
+             * overwritten by that of the selfheal */
 
-                build_ret = dht_build_child_loc (this, &lookup_local->loc,
-                                                 &local->loc, trav->d_name);
-                if (build_ret != 0)
-                        goto err;
+            heal_frame = copy_frame(frame);
 
-                gf_log (this->name, GF_LOG_TRACE,
-                        "looking up %s on %s",
-                        lookup_local->loc.path, src->name);
+            if (heal_frame == NULL) {
+                goto err;
+            }
 
-                LOCK (&frame->lock);
-                {
-                        local->call_cnt++;
-                }
-                UNLOCK (&frame->lock);
+            heal_local = dht_local_init(heal_frame, &local->loc, NULL, 0);
+            if (!heal_local) {
+                DHT_STACK_DESTROY(heal_frame);
+                goto err;
+            }
+
+            heal_local->inode = inode_ref(local->loc.inode);
+            heal_local->main_frame = frame;
+            gf_uuid_copy(heal_local->gfid, local->loc.inode->gfid);
+
+            dht_selfheal_restore(heal_frame, dht_rmdir_selfheal_cbk,
+                                 &heal_local->loc, heal_local->layout);
+            return 0;
+        } else {
+            if (local->loc.parent) {
+                dht_inode_ctx_time_update(local->loc.parent, this,
+                                          &local->preparent, 0);
 
-                STACK_WIND (lookup_frame, dht_rmdir_lookup_cbk,
-                            src, src->fops->lookup,
-                            &lookup_local->loc, NULL);
-                ret++;
+                dht_inode_ctx_time_update(local->loc.parent, this,
+                                          &local->postparent, 1);
+            }
+
+            dht_set_fixed_dir_stat(&local->preparent);
+            dht_set_fixed_dir_stat(&local->postparent);
+
+            dht_rmdir_unlock(frame, this);
+            DHT_STACK_UNWIND(rmdir, frame, local->op_ret, local->op_errno,
+                             &local->preparent, &local->postparent, NULL);
         }
+    }
+
+    return 0;
 
-        return ret;
 err:
-        DHT_STACK_DESTROY (lookup_frame);
-        return 0;
+    DHT_STACK_UNWIND(rmdir, frame, local->op_ret, local->op_errno, NULL, NULL,
+                     NULL);
+    return 0;
 }
 
+static int
+dht_rmdir_unlock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    DHT_STACK_DESTROY(frame);
+    return 0;
+}
 
-int
-dht_rmdir_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int op_ret, int op_errno, gf_dirent_t *entries)
-{
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = -1;
-	call_frame_t *prev = NULL;
-        xlator_t     *src = NULL;
-        int           ret = 0;
-
-	local = frame->local;
-	prev  = cookie;
-        src   = prev->this;
-
-	if (op_ret > 2) {
-                ret = dht_rmdir_is_subvol_empty (frame, this, entries, src);
-
-                switch (ret) {
-                case 0: /* non linkfiles exist */
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "readdir on %s for %s returned %d entries",
-                                prev->this->name, local->loc.path, op_ret);
-                        local->op_ret = -1;
-                        local->op_errno = ENOTEMPTY;
-                        break;
-                default:
-                        /* @ret number of linkfiles are getting unlinked */
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "readdir on %s for %s found %d linkfiles",
-                                prev->this->name, local->loc.path, ret);
-                        break;
-                }
-	}
+static int
+dht_rmdir_unlock(call_frame_t *frame, xlator_t *this)
+{
+    dht_local_t *local = NULL, *lock_local = NULL;
+    call_frame_t *lock_frame = NULL;
+    int lock_count = 0;
 
-	this_call_cnt = dht_frame_return (frame);
+    local = frame->local;
 
-	if (is_last_call (this_call_cnt)) {
-		dht_rmdir_do (frame, this);
-	}
+    /* Unlock entrylk */
+    dht_unlock_entrylk_wrapper(frame, &local->lock[0].ns.directory_ns);
 
-	return 0;
-}
+    /* Unlock inodelk */
+    lock_count = dht_lock_count(local->lock[0].ns.parent_layout.locks,
+                                local->lock[0].ns.parent_layout.lk_count);
 
+    if (lock_count == 0)
+        goto done;
 
-int
-dht_rmdir_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int op_ret, int op_errno, fd_t *fd)
+    lock_frame = copy_frame(frame);
+    if (lock_frame == NULL)
+        goto done;
+
+    lock_local = dht_local_init(lock_frame, &local->loc, NULL,
+                                lock_frame->root->op);
+    if (lock_local == NULL)
+        goto done;
+
+    lock_local->lock[0].ns.parent_layout.locks = local->lock[0]
+                                                     .ns.parent_layout.locks;
+    lock_local->lock[0]
+        .ns.parent_layout.lk_count = local->lock[0].ns.parent_layout.lk_count;
+
+    local->lock[0].ns.parent_layout.locks = NULL;
+    local->lock[0].ns.parent_layout.lk_count = 0;
+    dht_unlock_inodelk(lock_frame, lock_local->lock[0].ns.parent_layout.locks,
+                       lock_local->lock[0].ns.parent_layout.lk_count,
+                       dht_rmdir_unlock_cbk);
+    lock_frame = NULL;
+
+done:
+    if (lock_frame != NULL) {
+        DHT_STACK_DESTROY(lock_frame);
+    }
+
+    return 0;
+}
+
+static int
+dht_rmdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, struct iatt *preparent, struct iatt *postparent,
+              dict_t *xdata)
 {
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = -1;
-	call_frame_t *prev = NULL;
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+    int done = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    dht_local_t *heal_local = NULL;
+    call_frame_t *heal_frame = NULL;
+    int ret = -1;
+
+    local = frame->local;
+    prev = cookie;
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            if ((op_errno != ENOENT) && (op_errno != ESTALE)) {
+                local->op_errno = op_errno;
+                local->op_ret = -1;
 
+                if (op_errno != EACCES)
+                    local->need_selfheal = 1;
+            }
 
-	local = frame->local;
-	prev  = cookie;
+            gf_uuid_unparse(local->loc.gfid, gfid);
 
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"opendir on %s for %s failed (%s)",
-			prev->this->name, local->loc.path,
-			strerror (op_errno));
-		goto err;
-	}
+            gf_msg_debug(this->name, op_errno,
+                         "rmdir on %s for %s failed."
+                         "(gfid = %s)",
+                         prev->name, local->loc.path, gfid);
+            goto unlock;
+        }
 
-	STACK_WIND (frame, dht_rmdir_readdirp_cbk,
-		    prev->this, prev->this->fops->readdirp,
-		    local->fd, 4096, 0);
+        /* Track if rmdir succeeded on at least one subvol*/
+        local->fop_succeeded = 1;
+        dht_iatt_merge(this, &local->preparent, preparent);
+        dht_iatt_merge(this, &local->postparent, postparent);
+    }
+unlock:
+    UNLOCK(&frame->lock);
 
-	return 0;
+    this_call_cnt = dht_frame_return(frame);
 
-err:
-	this_call_cnt = dht_frame_return (frame);
+    /* if local->hashed_subvol, we are yet to wind to hashed_subvol. */
+    if (local->hashed_subvol && (this_call_cnt == 1)) {
+        done = 1;
+    } else if (!local->hashed_subvol && !this_call_cnt) {
+        done = 1;
+    }
+
+    if (done) {
+        if (local->need_selfheal && local->fop_succeeded) {
+            dht_rmdir_unlock(frame, this);
+            local->layout = dht_layout_get(this, local->loc.inode);
+
+            /* TODO: neater interface needed below */
+            local->stbuf.ia_type = local->loc.inode->ia_type;
 
-	if (is_last_call (this_call_cnt)) {
-		dht_rmdir_do (frame, this);
-	}
+            gf_uuid_copy(local->gfid, local->loc.inode->gfid);
+            heal_frame = copy_frame(frame);
+            if (heal_frame == NULL) {
+                goto err;
+            }
+
+            heal_local = dht_local_init(heal_frame, &local->loc, NULL, 0);
+            if (!heal_local) {
+                DHT_STACK_DESTROY(heal_frame);
+                goto err;
+            }
+
+            heal_local->inode = inode_ref(local->loc.inode);
+            heal_local->main_frame = frame;
+            gf_uuid_copy(heal_local->gfid, local->loc.inode->gfid);
+            ret = dht_selfheal_restore(heal_frame, dht_rmdir_selfheal_cbk,
+                                       &heal_local->loc, heal_local->layout);
+            if (ret) {
+                DHT_STACK_DESTROY(heal_frame);
+                goto err;
+            }
+
+        } else if (this_call_cnt) {
+            /* If non-hashed subvol's have responded, proceed */
+            if (local->op_ret == 0) {
+                /* Delete the dir from the hashed subvol if:
+                 * The fop succeeded on at least one subvol
+                 *  and did not fail on any
+                 *  or
+                 *  The fop failed with ENOENT/ESTALE on
+                 *  all subvols */
+
+                STACK_WIND_COOKIE(frame, dht_rmdir_hashed_subvol_cbk,
+                                  local->hashed_subvol, local->hashed_subvol,
+                                  local->hashed_subvol->fops->rmdir,
+                                  &local->loc, local->flags, NULL);
+            } else {
+                /* hashed-subvol was non-NULL and rmdir failed on
+                 * all non hashed-subvols. Unwind rmdir with
+                 * local->op_ret and local->op_errno. */
+                dht_rmdir_unlock(frame, this);
+                DHT_STACK_UNWIND(rmdir, frame, local->op_ret, local->op_errno,
+                                 &local->preparent, &local->postparent, NULL);
+
+                return 0;
+            }
+        } else if (!this_call_cnt) {
+            /* All subvol's have responded, proceed */
+
+            if (local->loc.parent) {
+                dht_inode_ctx_time_update(local->loc.parent, this,
+                                          &local->preparent, 0);
 
-	return 0;
+                dht_inode_ctx_time_update(local->loc.parent, this,
+                                          &local->postparent, 1);
+            }
+
+            dht_set_fixed_dir_stat(&local->preparent);
+            dht_set_fixed_dir_stat(&local->postparent);
+
+            dht_rmdir_unlock(frame, this);
+            DHT_STACK_UNWIND(rmdir, frame, local->op_ret, local->op_errno,
+                             &local->preparent, &local->postparent, NULL);
+        }
+    }
+
+    return 0;
+
+err:
+    DHT_STACK_UNWIND(rmdir, frame, -1, local->op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
+static int
+dht_rmdir_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int i = 0;
+    xlator_t *hashed_subvol;
+
+    conf = this->private;
+    local = frame->local;
 
-int
-dht_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags)
-{
-	dht_local_t  *local  = NULL;
-	dht_conf_t   *conf = NULL;
-        int           op_errno = -1;
-	int           i = -1;
-	int           ret = -1;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (this->private, err);
-
-	conf = this->private;
-
-	local = dht_local_init (frame);
-	if (!local) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		op_errno = ENOMEM;
-		goto err;
-	}
-
-	local->call_cnt = conf->subvolume_cnt;
-	local->op_ret   = 0;
-
-	ret = loc_copy (&local->loc, loc);
-	if (ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		op_errno = ENOMEM;
-		goto err;
-	}
-
-        local->flags = flags;
-
-	local->fd = fd_create (local->loc.inode, frame->root->pid);
-	if (!local->fd) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		op_errno = ENOMEM;
-		goto err;
-	}
-
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		STACK_WIND (frame, dht_rmdir_opendir_cbk,
-			    conf->subvolumes[i],
-			    conf->subvolumes[i]->fops->opendir,
-			    loc, local->fd);
-	}
-
-	return 0;
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_INODE_LK_ERROR,
+               "acquiring entrylk after inodelk failed rmdir for %s)",
+               local->loc.path);
+
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    hashed_subvol = local->hashed_subvol;
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (hashed_subvol && (hashed_subvol == conf->subvolumes[i]))
+            continue;
+
+        STACK_WIND_COOKIE(frame, dht_rmdir_cbk, conf->subvolumes[i],
+                          conf->subvolumes[i], conf->subvolumes[i]->fops->rmdir,
+                          &local->loc, local->flags, NULL);
+    }
+
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (rmdir, frame, -1, op_errno,
-                          NULL, NULL);
+    DHT_STACK_UNWIND(rmdir, frame, local->op_ret, local->op_errno,
+                     &local->preparent, &local->postparent, NULL);
 
-	return 0;
+    return 0;
 }
 
+static int
+dht_rmdir_do(call_frame_t *frame, xlator_t *this)
+{
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+    xlator_t *hashed_subvol = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    VALIDATE_OR_GOTO(frame->local, err);
+    local = frame->local;
+    VALIDATE_OR_GOTO(this->private, out);
+    conf = this->private;
+
+    if (local->op_ret == -1)
+        goto out;
+
+    local->call_cnt = conf->subvolume_cnt;
+
+    /* first remove from non-hashed_subvol */
+    hashed_subvol = dht_subvol_get_hashed(this, &local->loc);
+
+    if (!hashed_subvol) {
+        gf_uuid_unparse(local->loc.gfid, gfid);
+
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+               "Failed to get hashed subvol for %s (gfid = %s)",
+               local->loc.path, gfid);
+    } else {
+        local->hashed_subvol = hashed_subvol;
+    }
+
+    /* When DHT has only 1 child */
+    if (conf->subvolume_cnt == 1) {
+        STACK_WIND_COOKIE(frame, dht_rmdir_hashed_subvol_cbk,
+                          conf->subvolumes[0], conf->subvolumes[0],
+                          conf->subvolumes[0]->fops->rmdir, &local->loc,
+                          local->flags, NULL);
+        return 0;
+    }
 
-int
-dht_xattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		 int32_t op_ret, int32_t op_errno, dict_t *dict)
+    local->current = &local->lock[0];
+    ret = dht_protect_namespace(frame, &local->loc, local->hashed_subvol,
+                                &local->current->ns, dht_rmdir_lock_cbk);
+    if (ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = errno ? errno : EINVAL;
+        goto out;
+    }
+
+    return 0;
+
+out:
+    dht_set_fixed_dir_stat(&local->preparent);
+    dht_set_fixed_dir_stat(&local->postparent);
+
+    DHT_STACK_UNWIND(rmdir, frame, local->op_ret, local->op_errno,
+                     &local->preparent, &local->postparent, NULL);
+    return 0;
+err:
+    DHT_STACK_UNWIND(rmdir, frame, -1, EINVAL, NULL, NULL, NULL);
+    return 0;
+}
+
+static void
+dht_rmdir_readdirp_done(call_frame_t *readdirp_frame, xlator_t *this)
 {
-	DHT_STACK_UNWIND (xattrop, frame, op_ret, op_errno, dict);
-	return 0;
+    call_frame_t *main_frame = NULL;
+    dht_local_t *main_local = NULL;
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+
+    local = readdirp_frame->local;
+    main_frame = local->main_frame;
+    main_local = main_frame->local;
+
+    /* At least one readdirp failed.
+     * This is a bit hit or miss - if readdirp failed on more than
+     * one subvol, we don't know which error is returned.
+     */
+    if (local->op_ret == -1) {
+        main_local->op_ret = local->op_ret;
+        main_local->op_errno = local->op_errno;
+    }
+
+    this_call_cnt = dht_frame_return(main_frame);
+
+    if (is_last_call(this_call_cnt))
+        dht_rmdir_do(main_frame, this);
+
+    DHT_STACK_DESTROY(readdirp_frame);
 }
 
+/* Keep sending readdirp on the subvol until it returns no more entries
+ * It is possible that not all entries will fit in a single readdirp in
+ * which case the rmdir will keep failing with ENOTEMPTY
+ */
 
-int
-dht_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc,
-	     gf_xattrop_flags_t flags, dict_t *dict)
-{
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = dht_subvol_get_cached (this, loc->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	local->inode = inode_ref (loc->inode);
-	local->call_cnt = 1;
-
-	STACK_WIND (frame,
-		    dht_xattrop_cbk,
-		    subvol, subvol->fops->xattrop,
-		    loc, flags, dict);
-
-	return 0;
+static int
+dht_rmdir_readdirp_do(call_frame_t *readdirp_frame, xlator_t *this)
+{
+    dht_local_t *local = NULL;
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (xattrop, frame, -1, op_errno, NULL);
+    local = readdirp_frame->local;
+
+    if (local->op_ret == -1) {
+        /* there is no point doing another readdirp on this
+         * subvol . */
+        dht_rmdir_readdirp_done(readdirp_frame, this);
+        return 0;
+    }
 
-	return 0;
+    STACK_WIND_COOKIE(readdirp_frame, dht_rmdir_readdirp_cbk,
+                      local->hashed_subvol, local->hashed_subvol,
+                      local->hashed_subvol->fops->readdirp, local->fd, 4096, 0,
+                      local->xattr);
+
+    return 0;
 }
 
+static int
+dht_rmdir_linkfile_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                              int op_ret, int op_errno, struct iatt *preparent,
+                              struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *src = NULL;
+    call_frame_t *readdirp_frame = NULL;
+    dht_local_t *readdirp_local = NULL;
+    int this_call_cnt = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    prev = cookie;
+    src = prev;
+
+    readdirp_frame = local->main_frame;
+    readdirp_local = readdirp_frame->local;
+
+    gf_uuid_unparse(local->loc.gfid, gfid);
+
+    if (op_ret == 0) {
+        gf_msg_trace(this->name, 0, "Unlinked linkfile %s on %s, gfid = %s",
+                     local->loc.path, src->name, gfid);
+    } else {
+        if (op_errno != ENOENT) {
+            readdirp_local->op_ret = -1;
+            readdirp_local->op_errno = op_errno;
+        }
+        gf_msg_debug(this->name, op_errno,
+                     "Unlink of %s on %s failed. (gfid = %s)", local->loc.path,
+                     src->name, gfid);
+    }
+
+    this_call_cnt = dht_frame_return(readdirp_frame);
 
-int
-dht_fxattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		  int32_t op_ret, int32_t op_errno, dict_t *dict)
+    if (is_last_call(this_call_cnt))
+        dht_rmdir_readdirp_do(readdirp_frame, this);
+
+    DHT_STACK_DESTROY(frame);
+    return 0;
+}
+
+static int
+dht_rmdir_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, inode_t *inode,
+                     struct iatt *stbuf, dict_t *xattr, struct iatt *parent)
 {
-	DHT_STACK_UNWIND (fxattrop, frame, op_ret, op_errno, dict);
-	return 0;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *src = NULL;
+    call_frame_t *readdirp_frame = NULL;
+    dht_local_t *readdirp_local = NULL;
+    int this_call_cnt = 0;
+    dht_conf_t *conf = this->private;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    prev = cookie;
+    src = prev;
+
+    gf_msg_debug(this->name, 0, "dht_rmdir_lookup_cbk %s", local->loc.path);
+
+    readdirp_frame = local->main_frame;
+    readdirp_local = readdirp_frame->local;
+
+    if (op_ret != 0) {
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_FILE_LOOKUP_FAILED,
+               "lookup failed for %s on %s", local->loc.path, src->name);
+        goto err;
+    }
+
+    if (!check_is_linkfile(inode, stbuf, xattr, conf->link_xattr_name)) {
+        readdirp_local->op_ret = -1;
+        readdirp_local->op_errno = ENOTEMPTY;
+
+        gf_uuid_unparse(local->loc.gfid, gfid);
+
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_NOT_LINK_FILE_ERROR,
+               "%s on %s is not a linkfile (type=0%o, gfid = %s)",
+               local->loc.path, src->name, stbuf->ia_type, gfid);
+        goto err;
+    }
+
+    STACK_WIND_COOKIE(frame, dht_rmdir_linkfile_unlink_cbk, src, src,
+                      src->fops->unlink, &local->loc, 0, NULL);
+    return 0;
+err:
+
+    this_call_cnt = dht_frame_return(readdirp_frame);
+    if (is_last_call(this_call_cnt)) {
+        dht_rmdir_readdirp_do(readdirp_frame, this);
+    }
+
+    DHT_STACK_DESTROY(frame);
+    return 0;
 }
 
+static int
+dht_rmdir_cached_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int op_ret, int op_errno, inode_t *inode,
+                            struct iatt *stbuf, dict_t *xattr,
+                            struct iatt *parent)
+{
+    dht_local_t *local = NULL;
+    xlator_t *src = NULL;
+    call_frame_t *readdirp_frame = NULL;
+    dht_local_t *readdirp_local = NULL;
+    int this_call_cnt = 0;
+    dht_conf_t *conf = this->private;
+    dict_t *xattrs = NULL;
+    int ret = 0;
+
+    local = frame->local;
+    src = local->hashed_subvol;
+
+    /* main_frame here is the readdirp_frame */
+
+    readdirp_frame = local->main_frame;
+    readdirp_local = readdirp_frame->local;
+
+    gf_msg_debug(this->name, 0, "returning for %s ", local->loc.path);
+
+    if (op_ret == 0) {
+        readdirp_local->op_ret = -1;
+        readdirp_local->op_errno = ENOTEMPTY;
+
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_SUBVOL_ERROR,
+               "%s found on cached subvol %s", local->loc.path, src->name);
+        goto err;
+    } else if (op_errno != ENOENT) {
+        readdirp_local->op_ret = -1;
+        readdirp_local->op_errno = op_errno;
+
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_SUBVOL_ERROR,
+               "%s not found on cached subvol %s", local->loc.path, src->name);
+        goto err;
+    }
+
+    xattrs = dict_new();
+    if (!xattrs) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+               "dict_new failed");
+        goto err;
+    }
+
+    ret = dict_set_uint32(xattrs, conf->link_xattr_name, 256);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+               "Failed to set dictionary value: key = %s",
+               conf->link_xattr_name);
+        if (xattrs)
+            dict_unref(xattrs);
+        goto err;
+    }
+    STACK_WIND_COOKIE(frame, dht_rmdir_lookup_cbk, src, src, src->fops->lookup,
+                      &local->loc, xattrs);
+    if (xattrs)
+        dict_unref(xattrs);
+
+    return 0;
+err:
 
-int
-dht_fxattrop (call_frame_t *frame, xlator_t *this,
-	      fd_t *fd, gf_xattrop_flags_t flags, dict_t *dict)
+    this_call_cnt = dht_frame_return(readdirp_frame);
+
+    /* Once all the lookups/unlinks etc have returned, proceed to wind
+     * readdirp on the subvol again until no entries are returned.
+     * This is required if there are more entries than can be returned
+     * in a single readdirp call.
+     */
+
+    if (is_last_call(this_call_cnt))
+        dht_rmdir_readdirp_do(readdirp_frame, this);
+
+    DHT_STACK_DESTROY(frame);
+    return 0;
+}
+
+static int
+dht_rmdir_is_subvol_empty(call_frame_t *frame, xlator_t *this,
+                          gf_dirent_t *entries, xlator_t *src)
 {
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
+    int ret = 0;
+    int build_ret = 0;
+    gf_dirent_t *trav = NULL;
+    call_frame_t *lookup_frame = NULL;
+    dht_local_t *lookup_local = NULL;
+    dht_local_t *local = NULL;
+    dict_t *xattrs = NULL;
+    dht_conf_t *conf = this->private;
+    xlator_t *subvol = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    int count = 0;
+    gf_boolean_t unwind = _gf_false;
+
+    local = frame->local;
+
+    list_for_each_entry(trav, &entries->list, list)
+    {
+        if (strcmp(trav->d_name, ".") == 0)
+            continue;
+        if (strcmp(trav->d_name, "..") == 0)
+            continue;
+        if (check_is_linkfile(NULL, (&trav->d_stat), trav->dict,
+                              conf->link_xattr_name)) {
+            count++;
+            continue;
+        }
+
+        /* this entry is either a directory which is neither "." nor "..",
+           or a non directory which is not a linkfile. the directory is to
+           be treated as non-empty
+        */
+        return 0;
+    }
+
+    xattrs = dict_new();
+    if (!xattrs) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+               "dict_new failed");
+        return -1;
+    }
+
+    ret = dict_set_uint32(xattrs, conf->link_xattr_name, 256);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+               "Failed to set dictionary value: key = %s",
+               conf->link_xattr_name);
+
+        if (xattrs)
+            dict_unref(xattrs);
+        return -1;
+    }
+
+    local->call_cnt = count;
+    ret = 0;
+
+    list_for_each_entry(trav, &entries->list, list)
+    {
+        if (strcmp(trav->d_name, ".") == 0)
+            continue;
+        if (strcmp(trav->d_name, "..") == 0)
+            continue;
+
+        lookup_frame = copy_frame(frame);
+
+        if (!lookup_frame) {
+            /* out of memory, let the rmdir fail
+               (as non-empty, unfortunately) */
+            goto err;
+        }
+
+        lookup_local = dht_local_init(lookup_frame, NULL, NULL, GF_FOP_LOOKUP);
+        if (!lookup_local) {
+            goto err;
+        }
+
+        lookup_frame->local = lookup_local;
+        lookup_local->main_frame = frame;
+        lookup_local->hashed_subvol = src;
+
+        build_ret = dht_build_child_loc(this, &lookup_local->loc, &local->loc,
+                                        trav->d_name);
+        if (build_ret != 0)
+            goto err;
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+        gf_uuid_copy(lookup_local->loc.gfid, trav->d_stat.ia_gfid);
 
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
+        gf_uuid_unparse(lookup_local->loc.gfid, gfid);
 
-	STACK_WIND (frame,
-		    dht_fxattrop_cbk,
-		    subvol, subvol->fops->fxattrop,
-		    fd, flags, dict);
+        gf_msg_trace(this->name, 0, "looking up %s on subvolume %s, gfid = %s",
+                     lookup_local->loc.path, src->name, gfid);
 
-	return 0;
+        subvol = dht_linkfile_subvol(this, NULL, &trav->d_stat, trav->dict);
+        if (!subvol || (subvol == src)) {
+            /* we need to delete the linkto file if it does not have a
+             * valid subvol or it points to itself.
+             */
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_INVALID_LINKFILE,
+                   "Linkfile does not have link subvolume. "
+                   "path = %s, gfid = %s",
+                   lookup_local->loc.path, gfid);
 
+            gf_msg_debug(this->name, 0, "looking up %s on subvol %s, gfid = %s",
+                         lookup_local->loc.path, src->name, gfid);
+
+            STACK_WIND_COOKIE(lookup_frame, dht_rmdir_lookup_cbk, src, src,
+                              src->fops->lookup, &lookup_local->loc, xattrs);
+        } else {
+            gf_msg_debug(this->name, 0,
+                         "Looking up linkfile target %s on "
+                         " subvol %s, gfid = %s",
+                         lookup_local->loc.path, subvol->name, gfid);
+
+            STACK_WIND(lookup_frame, dht_rmdir_cached_lookup_cbk, subvol,
+                       subvol->fops->lookup, &lookup_local->loc, xattrs);
+        }
+        ret++;
+
+        lookup_frame = NULL;
+        lookup_local = NULL;
+    }
+
+    if (xattrs)
+        dict_unref(xattrs);
+
+    return ret;
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (fxattrop, frame, -1, op_errno, NULL);
+    if (xattrs)
+        dict_unref(xattrs);
 
-	return 0;
-}
+    if (lookup_frame)
+        DHT_STACK_DESTROY(lookup_frame);
 
+    /* Handle the case where the wound calls have unwound before the
+     * loop processing is done
+     */
 
-int
-dht_inodelk_cbk (call_frame_t *frame, void *cookie,
-		 xlator_t *this, int32_t op_ret, int32_t op_errno)
+    LOCK(&frame->lock);
+    {
+        local->op_ret = -1;
+        local->op_errno = ENOTEMPTY;
+
+        local->call_cnt -= (count - ret);
+        if (!local->call_cnt)
+            unwind = _gf_true;
+    }
+    UNLOCK(&frame->lock);
+
+    if (!unwind) {
+        return ret;
+    }
+    return 0;
+}
+
+/*
+ * No more entries on this subvol. Proceed to the actual rmdir operation.
+ */
 
+static int
+dht_rmdir_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, gf_dirent_t *entries,
+                       dict_t *xdata)
 {
-	DHT_STACK_UNWIND (inodelk, frame, op_ret, op_errno);
-	return 0;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *src = NULL;
+    int ret = 0;
+    char *path = NULL;
+
+    local = frame->local;
+    prev = cookie;
+    src = prev;
+
+    if (op_ret > 2) {
+        /* dht_rmdir_is_subvol_empty() may free the frame,
+         * copy path for logging.
+         */
+        path = gf_strdup(local->loc.path);
+
+        ret = dht_rmdir_is_subvol_empty(frame, this, entries, src);
+
+        switch (ret) {
+            case 0: /* non linkfiles exist */
+                gf_msg_trace(this->name, 0,
+                             "readdir on %s for %s returned %d "
+                             "entries",
+                             prev->name, local->loc.path, op_ret);
+                local->op_ret = -1;
+                local->op_errno = ENOTEMPTY;
+                break;
+            default:
+                /* @ret number of linkfiles are getting unlinked */
+                gf_msg_trace(this->name, 0,
+                             "readdir on %s for %s found %d "
+                             "linkfiles",
+                             prev->name, path, ret);
+                break;
+        }
+    }
+
+    /* readdirp failed or no linkto files were found on this subvol */
+    if (!ret)
+        dht_rmdir_readdirp_done(frame, this);
+
+    GF_FREE(path);
+    return 0;
 }
 
+static int
+dht_rmdir_opendir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, fd_t *fd, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = -1;
+    xlator_t *prev = NULL;
+    int ret = 0;
+    dht_conf_t *conf = this->private;
+    dict_t *dict = NULL;
+    int i = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    dht_local_t *readdirp_local = NULL;
+    call_frame_t *readdirp_frame = NULL;
+    int cnt = 0;
+
+    local = frame->local;
+    prev = cookie;
+
+    this_call_cnt = dht_frame_return(frame);
+    if (op_ret == -1) {
+        gf_uuid_unparse(local->loc.gfid, gfid);
+
+        gf_msg_debug(this->name, op_errno,
+                     "opendir on %s for %s failed, "
+                     "gfid = %s,",
+                     prev->name, local->loc.path, gfid);
+        if ((op_errno != ENOENT) && (op_errno != ESTALE)) {
+            local->op_ret = -1;
+            local->op_errno = op_errno;
+        }
+        goto err;
+    }
 
-int32_t
-dht_inodelk (call_frame_t *frame, xlator_t *this,
-	     const char *volume, loc_t *loc, int32_t cmd, struct gf_flock *lock)
-{
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = dht_subvol_get_cached (this, loc->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	local->inode = inode_ref (loc->inode);
-	local->call_cnt = 1;
-
-	STACK_WIND (frame,
-		    dht_inodelk_cbk,
-		    subvol, subvol->fops->inodelk,
-		    volume, loc, cmd, lock);
-
-	return 0;
+    if (!is_last_call(this_call_cnt))
+        return 0;
+
+    if (local->op_ret == -1)
+        goto err;
+
+    fd_bind(fd);
+
+    dict = dict_new();
+    if (!dict) {
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        goto err;
+    }
+
+    ret = dict_set_uint32(dict, conf->link_xattr_name, 256);
+    if (ret)
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+               "%s: Failed to set dictionary value:key = %s", local->loc.path,
+               conf->link_xattr_name);
+
+    cnt = local->call_cnt = conf->subvolume_cnt;
+
+    /* Create a separate frame per subvol as we might need
+     * to resend readdirp multiple times to get all the
+     * entries.
+     */
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        readdirp_frame = copy_frame(frame);
+
+        if (!readdirp_frame) {
+            cnt--;
+            /* Reduce the local->call_cnt as well */
+            (void)dht_frame_return(frame);
+            continue;
+        }
+
+        readdirp_local = dht_local_init(readdirp_frame, &local->loc, local->fd,
+                                        0);
+
+        if (!readdirp_local) {
+            DHT_STACK_DESTROY(readdirp_frame);
+            cnt--;
+            /* Reduce the local->call_cnt as well */
+            dht_frame_return(frame);
+            continue;
+        }
+        readdirp_local->main_frame = frame;
+        readdirp_local->op_ret = 0;
+        readdirp_local->xattr = dict_ref(dict);
+        /* overload this field to save the subvol info */
+        readdirp_local->hashed_subvol = conf->subvolumes[i];
+
+        STACK_WIND_COOKIE(readdirp_frame, dht_rmdir_readdirp_cbk,
+                          conf->subvolumes[i], conf->subvolumes[i],
+                          conf->subvolumes[i]->fops->readdirp,
+                          readdirp_local->fd, 4096, 0, readdirp_local->xattr);
+    }
+
+    if (dict)
+        dict_unref(dict);
+
+    /* Could not wind readdirp to any subvol */
+
+    if (!cnt)
+        goto err;
+
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (inodelk, frame, -1, op_errno);
+    if (is_last_call(this_call_cnt)) {
+        dht_rmdir_do(frame, this);
+    }
 
-	return 0;
+    return 0;
 }
 
-
 int
-dht_finodelk_cbk (call_frame_t *frame, void *cookie,
-		  xlator_t *this, int32_t op_ret, int32_t op_errno)
-
+dht_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+          dict_t *xdata)
 {
-	DHT_STACK_UNWIND (finodelk, frame, op_ret, op_errno);
-	return 0;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int op_errno = -1;
+    int i = -1;
+    int ret = -1;
+    dict_t *xattr_req = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+    VALIDATE_OR_GOTO(loc->path, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_RMDIR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->call_cnt = conf->subvolume_cnt;
+    local->op_ret = 0;
+    local->fop_succeeded = 0;
+
+    local->flags = flags;
+
+    local->fd = fd_create(local->loc.inode, frame->root->pid);
+    if (!local->fd) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    if (flags) {
+        return dht_rmdir_do(frame, this);
+    }
+    if (xdata) {
+        xattr_req = dict_ref(xdata);
+    } else {
+        xattr_req = dict_new();
+    }
+    if (xattr_req) {
+        ret = dict_set_uint32(xattr_req, conf->link_xattr_name, 256);
+        /* If parallel-readdir is enabled, this is required
+         * to handle stale linkto files in the directory
+         * being deleted. If this fails, log an error but
+         * do not prevent the operation.
+         */
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, 0, "%s: failed to set key %s",
+                   loc->path, conf->link_xattr_name);
+        }
+    } else {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "%s: failed to set key %s",
+               loc->path, conf->link_xattr_name);
+    }
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        STACK_WIND_COOKIE(frame, dht_rmdir_opendir_cbk, conf->subvolumes[i],
+                          conf->subvolumes[i],
+                          conf->subvolumes[i]->fops->opendir, loc, local->fd,
+                          xattr_req);
+    }
+
+    if (xattr_req) {
+        dict_unref(xattr_req);
+    }
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(rmdir, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
 }
 
+static int
+dht_entrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata)
+
+{
+    DHT_STACK_UNWIND(entrylk, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
+/* TODO
+ * Sending entrylk to cached subvol can result in stale lock
+ * as described in the bug 1311002.
+ */
 int
-dht_finodelk (call_frame_t *frame, xlator_t *this,
-	      const char *volume, fd_t *fd, int32_t cmd, struct gf_flock *lock)
+dht_entrylk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+            const char *basename, entrylk_cmd cmd, entrylk_type type,
+            dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
 
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
+    local = dht_local_init(frame, loc, NULL, GF_FOP_ENTRYLK);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
 
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_uuid_unparse(loc->gfid, gfid);
 
-	STACK_WIND (frame,
-		    dht_finodelk_cbk,
-		    subvol, subvol->fops->finodelk,
-		    volume, fd, cmd, lock);
+        gf_msg_debug(this->name, 0,
+                     "no cached subvolume for path=%s, "
+                     "gfid = %s",
+                     loc->path, gfid);
+        op_errno = EINVAL;
+        goto err;
+    }
 
-	return 0;
+    local->call_cnt = 1;
+
+    STACK_WIND(frame, dht_entrylk_cbk, subvol, subvol->fops->entrylk, volume,
+               loc, basename, cmd, type, xdata);
+
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (finodelk, frame, -1, op_errno);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(entrylk, frame, -1, op_errno, NULL);
 
-	return 0;
+    return 0;
 }
 
-
-int
-dht_entrylk_cbk (call_frame_t *frame, void *cookie,
-		 xlator_t *this, int32_t op_ret, int32_t op_errno)
+static int
+dht_fentrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
 
 {
-	DHT_STACK_UNWIND (entrylk, frame, op_ret, op_errno);
-	return 0;
+    DHT_STACK_UNWIND(fentrylk, frame, op_ret, op_errno, NULL);
+    return 0;
 }
 
-
 int
-dht_entrylk (call_frame_t *frame, xlator_t *this,
-	     const char *volume, loc_t *loc, const char *basename,
-	     entrylk_cmd cmd, entrylk_type type)
-{
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
-	dht_local_t  *local = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = dht_subvol_get_cached (this, loc->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	local->inode = inode_ref (loc->inode);
-	local->call_cnt = 1;
-
-	STACK_WIND (frame, dht_entrylk_cbk,
-		    subvol, subvol->fops->entrylk,
-		    volume, loc, basename, cmd, type);
-
-	return 0;
+dht_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+             const char *basename, entrylk_cmd cmd, entrylk_type type,
+             dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+    VALIDATE_OR_GOTO(fd->inode, err);
+
+    gf_uuid_unparse(fd->inode->gfid, gfid);
+
+    subvol = dht_subvol_get_cached(this, fd->inode);
+    if (!subvol) {
+        gf_msg_debug(this->name, 0,
+                     "No cached subvolume for fd=%p,"
+                     " gfid = %s",
+                     fd, gfid);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    STACK_WIND(frame, dht_fentrylk_cbk, subvol, subvol->fops->fentrylk, volume,
+               fd, basename, cmd, type, xdata);
+
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (entrylk, frame, -1, op_errno);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(fentrylk, frame, -1, op_errno, NULL);
 
-	return 0;
+    return 0;
 }
 
+static int32_t
+dht_ipc_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
 
-int
-dht_fentrylk_cbk (call_frame_t *frame, void *cookie,
-		  xlator_t *this, int32_t op_ret, int32_t op_errno)
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
 
-{
-	DHT_STACK_UNWIND (fentrylk, frame, op_ret, op_errno);
-	return 0;
-}
+    local = frame->local;
 
+    LOCK(&frame->lock);
+    {
+        if (op_ret < 0 && op_errno != ENOTCONN) {
+            local->op_errno = op_errno;
+            goto unlock;
+        }
+        local->op_ret = 0;
+    }
+unlock:
+    UNLOCK(&frame->lock);
 
-int
-dht_fentrylk (call_frame_t *frame, xlator_t *this,
-	      const char *volume, fd_t *fd, const char *basename,
-	      entrylk_cmd cmd, entrylk_type type)
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        DHT_STACK_UNWIND(ipc, frame, local->op_ret, local->op_errno, NULL);
+    }
+
+out:
+    return 0;
+}
+
+int32_t
+dht_ipc(call_frame_t *frame, xlator_t *this, int32_t op, dict_t *xdata)
 {
-	xlator_t     *subvol = NULL;
-        int           op_errno = -1;
+    dht_local_t *local = NULL;
+    int op_errno = EINVAL;
+    dht_conf_t *conf = NULL;
+    int call_cnt = 0;
+    int i = 0;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+
+    if (op != GF_IPC_TARGET_UPCALL)
+        goto wind_default;
+
+    VALIDATE_OR_GOTO(this->private, err);
+    conf = this->private;
+
+    local = dht_local_init(frame, NULL, NULL, GF_FOP_IPC);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
 
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
+    call_cnt = conf->subvolume_cnt;
+    local->call_cnt = call_cnt;
 
-	subvol = dht_subvol_get_cached (this, fd->inode);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
+    if (xdata) {
+        if (dict_set_int8(xdata, conf->xattr_name, 0) < 0)
+            goto err;
+    }
 
-	STACK_WIND (frame, dht_fentrylk_cbk,
-		    subvol, subvol->fops->fentrylk,
-		    volume, fd, basename, cmd, type);
+    for (i = 0; i < call_cnt; i++) {
+        STACK_WIND(frame, dht_ipc_cbk, conf->subvolumes[i],
+                   conf->subvolumes[i]->fops->ipc, op, xdata);
+    }
 
-	return 0;
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (fentrylk, frame, -1, op_errno);
+    DHT_STACK_UNWIND(ipc, frame, -1, op_errno, NULL);
 
-	return 0;
+    return 0;
+
+wind_default:
+    STACK_WIND(frame, default_ipc_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ipc, op, xdata);
+    return 0;
 }
 
+int
+dht_forget(xlator_t *this, inode_t *inode)
+{
+    uint64_t ctx_int = 0;
+    dht_inode_ctx_t *ctx = NULL;
+    dht_layout_t *layout = NULL;
+
+    inode_ctx_del(inode, this, &ctx_int);
+
+    if (!ctx_int)
+        return 0;
+
+    ctx = (dht_inode_ctx_t *)(long)ctx_int;
+
+    layout = ctx->layout;
+    ctx->layout = NULL;
+    dht_layout_unref(this, layout);
+    GF_FREE(ctx);
+
+    return 0;
+}
 
 int
-dht_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int op_ret, int op_errno, struct iatt *statpre,
-                 struct iatt *statpost)
-{
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = 0;
-	call_frame_t *prev = NULL;
-
-
-	local = frame->local;
-	prev = cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			gf_log (this->name, GF_LOG_DEBUG,
-				"subvolume %s returned -1 (%s)",
-				prev->this->name, strerror (op_errno));
-			goto unlock;
-		}
-
-		dht_iatt_merge (this, &local->prebuf, statpre, prev->this);
-                dht_iatt_merge (this, &local->stbuf, statpost, prev->this);
-		
-		if (local->inode) {
-			local->prebuf.ia_ino = local->inode->ino;
-                        local->stbuf.ia_ino = local->inode->ino;
+dht_notify(xlator_t *this, int event, void *data, ...)
+{
+    xlator_t *subvol = NULL;
+    int cnt = -1;
+    int i = -1;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+    int propagate = 0;
+
+    int had_heard_from_all = 0;
+    int have_heard_from_all = 0;
+    gf_defrag_info_t *defrag = NULL;
+    dict_t *dict = NULL;
+    gf_defrag_type cmd = 0;
+    dict_t *output = NULL;
+    va_list ap;
+    struct gf_upcall *up_data = NULL;
+    struct gf_upcall_cache_invalidation *up_ci = NULL;
+
+    conf = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, conf, out);
+
+    /* had all subvolumes reported status once till now? */
+    had_heard_from_all = 1;
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (!conf->last_event[i]) {
+            had_heard_from_all = 0;
+        }
+    }
+
+    switch (event) {
+        case GF_EVENT_CHILD_UP:
+            subvol = data;
+
+            conf->gen++;
+
+            for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (subvol == conf->subvolumes[i]) {
+                    cnt = i;
+                    break;
+                }
+            }
+
+            if (cnt == -1) {
+                gf_msg_debug(this->name, 0,
+                             "got GF_EVENT_CHILD_UP bad "
+                             "subvolume %s",
+                             subvol->name);
+                break;
+            }
+
+            LOCK(&conf->subvolume_lock);
+            {
+                conf->subvolume_status[cnt] = 1;
+                conf->last_event[cnt] = event;
+                conf->subvol_up_time[cnt] = gf_time();
+            }
+            UNLOCK(&conf->subvolume_lock);
+
+            /* one of the node came back up, do a stat update */
+            dht_get_du_info_for_subvol(this, cnt);
+
+            break;
+
+        case GF_EVENT_SOME_DESCENDENT_UP:
+            subvol = data;
+            conf->gen++;
+            propagate = 1;
+
+            break;
+
+        case GF_EVENT_SOME_DESCENDENT_DOWN:
+            subvol = data;
+            propagate = 1;
+
+            break;
+
+        case GF_EVENT_CHILD_DOWN:
+            subvol = data;
+
+            if (conf->assert_no_child_down) {
+                gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_CHILD_DOWN,
+                       "Received CHILD_DOWN. Exiting");
+                if (conf->defrag) {
+                    gf_defrag_stop(conf, GF_DEFRAG_STATUS_FAILED, NULL);
+                } else {
+                    kill(getpid(), SIGTERM);
                 }
+            }
 
-		local->op_ret = 0;
-	}
-unlock:
-	UNLOCK (&frame->lock);
+            for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (subvol == conf->subvolumes[i]) {
+                    cnt = i;
+                    break;
+                }
+            }
+
+            if (cnt == -1) {
+                gf_msg_debug(this->name, 0,
+                             "got GF_EVENT_CHILD_DOWN bad "
+                             "subvolume %s",
+                             subvol->name);
+                break;
+            }
+
+            LOCK(&conf->subvolume_lock);
+            {
+                conf->subvolume_status[cnt] = 0;
+                conf->last_event[cnt] = event;
+                conf->subvol_up_time[cnt] = 0;
+            }
+            UNLOCK(&conf->subvolume_lock);
+
+            for (i = 0; i < conf->subvolume_cnt; i++)
+                if (conf->last_event[i] != event)
+                    event = GF_EVENT_SOME_DESCENDENT_DOWN;
+            break;
+
+        case GF_EVENT_CHILD_CONNECTING:
+            subvol = data;
+
+            for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (subvol == conf->subvolumes[i]) {
+                    cnt = i;
+                    break;
+                }
+            }
+
+            if (cnt == -1) {
+                gf_msg_debug(this->name, 0,
+                             "got GF_EVENT_CHILD_CONNECTING"
+                             " bad subvolume %s",
+                             subvol->name);
+                break;
+            }
+
+            LOCK(&conf->subvolume_lock);
+            {
+                conf->last_event[cnt] = event;
+            }
+            UNLOCK(&conf->subvolume_lock);
+
+            break;
+        case GF_EVENT_VOLUME_DEFRAG: {
+            if (!conf->defrag) {
+                return ret;
+            }
+            defrag = conf->defrag;
 
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt))
-		DHT_STACK_UNWIND (setattr, frame, local->op_ret, local->op_errno,
-				  &local->prebuf, &local->stbuf);
+            dict = data;
+            va_start(ap, data);
+            output = va_arg(ap, dict_t *);
 
-        return 0;
-}
+            ret = dict_get_int32(dict, "rebalance-command", (int32_t *)&cmd);
+            if (ret) {
+                va_end(ap);
+                return ret;
+            }
+            LOCK(&defrag->lock);
+            {
+                if (defrag->is_exiting)
+                    goto unlock;
+                if ((cmd == GF_DEFRAG_CMD_STATUS) ||
+                    (cmd == GF_DEFRAG_CMD_DETACH_STATUS))
+                    gf_defrag_status_get(conf, output);
+                else if (cmd == GF_DEFRAG_CMD_DETACH_START)
+                    defrag->cmd = GF_DEFRAG_CMD_DETACH_START;
+                else if (cmd == GF_DEFRAG_CMD_STOP ||
+                         cmd == GF_DEFRAG_CMD_DETACH_STOP)
+                    gf_defrag_stop(conf, GF_DEFRAG_STATUS_STOPPED, output);
+            }
+        unlock:
+            UNLOCK(&defrag->lock);
+            va_end(ap);
+            return ret;
+            break;
+        }
+        case GF_EVENT_UPCALL:
+            up_data = (struct gf_upcall *)data;
+            if (up_data->event_type != GF_UPCALL_CACHE_INVALIDATION)
+                break;
+            up_ci = (struct gf_upcall_cache_invalidation *)up_data->data;
+
+            /* Since md-cache will be aggressively filtering lookups,
+             * the stale layout issue will be more pronounced. Hence
+             * when a layout xattr is changed by the rebalance process
+             * notify all the md-cache clients to invalidate the existing
+             * stat cache and send the lookup next time*/
+            if (up_ci->dict && dict_get(up_ci->dict, conf->xattr_name))
+                up_ci->flags |= UP_EXPLICIT_LOOKUP;
+
+            /* TODO: Instead of invalidating iatt, update the new
+             * hashed/cached subvolume in dht inode_ctx */
+            if (IS_DHT_LINKFILE_MODE(&up_ci->stat))
+                up_ci->flags |= UP_EXPLICIT_LOOKUP;
+
+            propagate = 1;
+            break;
+        default:
+            propagate = 1;
+            break;
+    }
+
+    /* have all subvolumes reported status once by now? */
+    have_heard_from_all = 1;
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (!conf->last_event[i])
+            have_heard_from_all = 0;
+    }
+
+    /* if all subvols have reported status, no need to hide anything
+       or wait for anything else. Just propagate blindly */
+    if (have_heard_from_all) {
+        propagate = 1;
+    }
+
+    if (!had_heard_from_all && have_heard_from_all) {
+        static int run_defrag = 0;
+        /* This is the first event which completes aggregation
+           of events from all subvolumes. If at least one subvol
+           had come up, propagate CHILD_UP, but only this time
+        */
+        event = GF_EVENT_CHILD_DOWN;
+
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (conf->last_event[i] == GF_EVENT_CHILD_UP) {
+                event = GF_EVENT_CHILD_UP;
+                break;
+            }
+
+            if (conf->last_event[i] == GF_EVENT_CHILD_CONNECTING) {
+                event = GF_EVENT_CHILD_CONNECTING;
+                /* continue to check other events for CHILD_UP */
+            }
+        }
 
+        /* Rebalance is started with assert_no_child_down. So we do
+         * not need to handle CHILD_DOWN event here.
+         *
+         * If there is a graph switch, we should not restart the
+         * rebalance daemon. Use 'run_defrag' to indicate if the
+         * thread has already started.
+         */
+        if (conf->defrag && !run_defrag) {
+            run_defrag = 1;
+            ret = gf_thread_create(&conf->defrag->th, NULL, gf_defrag_start,
+                                   this, "dhtdg");
+            if (ret) {
+                GF_FREE(conf->defrag);
+                conf->defrag = NULL;
+                kill(getpid(), SIGTERM);
+            }
+        }
+    }
+
+    ret = 0;
+    if (propagate)
+        ret = default_notify(this, event, data);
+out:
+    return ret;
+}
 
 int
-dht_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-             struct iatt *stbuf, int32_t valid)
-{
-	dht_layout_t *layout = NULL;
-	dht_local_t  *local  = NULL;
-        int           op_errno = -1;
-	int           i = -1;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_DEBUG,
-			"memory allocation failed :(");
-		goto err;
-	}
-
-	local->layout = layout = dht_layout_get (this, loc->inode);
-	if (!layout) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no layout for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	if (!layout_is_sane (layout)) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"layout is not sane for path=%s", loc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	local->inode = inode_ref (loc->inode);
-	local->call_cnt = layout->cnt;
-
-	for (i = 0; i < layout->cnt; i++) {
-		STACK_WIND (frame, dht_setattr_cbk,
-			    layout->list[i].xlator,
-			    layout->list[i].xlator->fops->setattr,
-			    loc, stbuf, valid);
-	}
-
-	return 0;
+dht_inode_ctx_layout_get(inode_t *inode, xlator_t *this, dht_layout_t **layout)
+{
+    dht_inode_ctx_t *ctx = NULL;
+    int ret = -1;
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (setattr, frame, -1, op_errno, NULL, NULL);
+    ret = dht_inode_ctx_get(inode, this, &ctx);
 
-	return 0;
+    if (!ret && ctx) {
+        if (ctx->layout) {
+            if (layout)
+                *layout = ctx->layout;
+            ret = 0;
+        } else {
+            ret = -1;
+        }
+    }
+
+    return ret;
 }
 
+void
+dht_log_new_layout_for_dir_selfheal(xlator_t *this, loc_t *loc,
+                                    dht_layout_t *layout)
+{
+    char string[2048] = {0};
+    char *output_string = NULL;
+    int len = 0;
+    int off = 0;
+    int i = 0;
+    gf_loglevel_t log_level = gf_log_get_loglevel();
+    int ret = 0;
 
-int
-dht_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
-              int32_t valid)
-{
-	dht_layout_t *layout = NULL;
-	dht_local_t  *local  = NULL;
-        int           op_errno = -1;
-	int           i = -1;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	local->layout = layout = dht_layout_get (this, fd->inode);
-	if (!layout) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no layout for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	if (!layout_is_sane (layout)) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"layout is not sane for fd=%p", fd);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	local->inode = inode_ref (fd->inode);
-	local->call_cnt = layout->cnt;
-
-	for (i = 0; i < layout->cnt; i++) {
-		STACK_WIND (frame, dht_setattr_cbk,
-			    layout->list[i].xlator,
-			    layout->list[i].xlator->fops->fsetattr,
-			    fd, stbuf, valid);
-	}
-
-	return 0;
+    if (log_level < GF_LOG_INFO)
+        return;
 
-err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (fsetattr, frame, -1, op_errno, NULL, NULL);
+    if (!layout)
+        return;
 
-	return 0;
-}
+    if (!layout->cnt)
+        return;
 
+    if (!loc)
+        return;
 
-int
-dht_forget (xlator_t *this, inode_t *inode)
+    if (!loc->path)
+        return;
+
+    ret = snprintf(string, sizeof(string), "Setting layout of %s with ",
+                   loc->path);
+
+    if (ret < 0)
+        return;
+
+    len += ret;
+
+    /* Calculation  of total length of the string required to calloc
+     * output_string. Log includes subvolume-name, start-range, end-range
+     * and err value.
+     *
+     * This log will help to debug cases where:
+     * a) Different processes set different layout of a directory.
+     * b) Error captured in lookup, which will be filled in layout->err
+     * (like ENOENT, ESTALE etc)
+     */
+
+    for (i = 0; i < layout->cnt; i++) {
+        ret = snprintf(string, sizeof(string),
+                       "[Subvol_name: %s, Err: %d , Start: "
+                       "0x%x, Stop: 0x%x, Hash: 0x%x], ",
+                       layout->list[i].xlator->name, layout->list[i].err,
+                       layout->list[i].start, layout->list[i].stop,
+                       layout->list[i].commit_hash);
+
+        if (ret < 0)
+            return;
+
+        len += ret;
+    }
+
+    len++;
+
+    output_string = GF_MALLOC(len + 1, gf_common_mt_char);
+
+    if (!output_string)
+        return;
+
+    ret = snprintf(output_string, len + 1, "Setting layout of %s with ",
+                   loc->path);
+
+    if (ret < 0)
+        goto err;
+
+    off += ret;
+
+    for (i = 0; i < layout->cnt; i++) {
+        ret = snprintf(output_string + off, len - off,
+                       "[Subvol_name: %s, Err: %d , Start: "
+                       "0x%x, Stop: 0x%x, Hash: 0x%x], ",
+                       layout->list[i].xlator->name, layout->list[i].err,
+                       layout->list[i].start, layout->list[i].stop,
+                       layout->list[i].commit_hash);
+
+        if (ret < 0)
+            goto err;
+
+        off += ret;
+    }
+
+    gf_msg(this->name, GF_LOG_DEBUG, 0, DHT_MSG_LOG_FIXED_LAYOUT, "%s",
+           output_string);
+
+err:
+    GF_FREE(output_string);
+}
+
+int32_t
+dht_migration_get_dst_subvol(xlator_t *this, dht_local_t *local)
 {
-	uint64_t      tmp_layout = 0;
-	dht_layout_t *layout = NULL;
+    int ret = -1;
 
-	inode_ctx_get (inode, this, &tmp_layout);
+    if (!local)
+        goto out;
 
-	if (!tmp_layout)
-		return 0;
+    local->rebalance.target_node = dht_subvol_get_hashed(this, &local->loc);
 
-	layout = (dht_layout_t *)(long)tmp_layout;
-        dht_layout_unref (this, layout);
+    if (local->rebalance.target_node)
+        ret = 0;
 
-	return 0;
+out:
+    return ret;
 }
 
+/*
+This function should not be called more then once during a FOP
+handling path. It is valid only for for ops on files
+*/
+int32_t
+dht_set_local_rebalance(xlator_t *this, dht_local_t *local, struct iatt *stbuf,
+                        struct iatt *prebuf, struct iatt *postbuf,
+                        dict_t *xdata)
+{
+    if (!local)
+        return -1;
+
+    if (local->rebalance.set) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_REBAL_STRUCT_SET,
+               "local->rebalance already set");
+    }
 
+    if (stbuf)
+        memcpy(&local->rebalance.stbuf, stbuf, sizeof(struct iatt));
 
-int
-dht_init_subvolumes (xlator_t *this, dht_conf_t *conf)
+    if (prebuf)
+        memcpy(&local->rebalance.prebuf, prebuf, sizeof(struct iatt));
+
+    if (postbuf)
+        memcpy(&local->rebalance.postbuf, postbuf, sizeof(struct iatt));
+
+    if (xdata)
+        local->rebalance.xdata = dict_ref(xdata);
+
+    local->rebalance.set = 1;
+
+    return 0;
+}
+
+int32_t
+dht_release(xlator_t *this, fd_t *fd)
 {
-        xlator_list_t *subvols = NULL;
-        int            cnt = 0;
+    return dht_fd_ctx_destroy(this, fd);
+}
 
-        if (!conf)
-                return -1;
+static int
+dht_pt_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, inode_t *inode, struct iatt *stbuf,
+                 struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
 
-        for (subvols = this->children; subvols; subvols = subvols->next)
-                cnt++;
+    local = frame->local;
 
-        conf->subvolumes = GF_CALLOC (cnt, sizeof (xlator_t *),
-                                      gf_dht_mt_xlator_t);
-        if (!conf->subvolumes) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                return -1;
-        }
-        conf->subvolume_cnt = cnt;
+    if (!op_ret) {
+        dht_layout_set(this, inode, local->layout);
+    }
+
+    DHT_STACK_UNWIND(mkdir, frame, op_ret, op_errno, inode, stbuf, preparent,
+                     postparent, NULL);
 
-        cnt = 0;
-        for (subvols = this->children; subvols; subvols = subvols->next)
-                conf->subvolumes[cnt++] = subvols->xlator;
+    return 0;
+}
 
-	conf->subvolume_status = GF_CALLOC (cnt, sizeof (char),
-                                            gf_dht_mt_char);
-	if (!conf->subvolume_status) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		return -1;
-	}
+int32_t
+dht_pt_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+             mode_t umask, dict_t *xdata)
+{
+    dht_layout_t *layout = NULL;
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    bool free_xdata = false;
+    int ret = 0;
+    int op_errno = 0;
+    int32_t *disk_layout_p = NULL;
+
+    conf = this->private;
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_MKDIR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    layout = dht_layout_new(this, conf->subvolume_cnt);
+    if (!layout)
+        goto wind;
+
+    local->layout = layout;
+
+    if (!xdata) {
+        xdata = dict_new();
+        if (!xdata)
+            goto wind;
+        free_xdata = true;
+    }
+
+    /*Set the xlator or the following will crash*/
+    layout->list[0].xlator = conf->subvolumes[0];
+
+    dht_selfheal_layout_new_directory(frame, loc, layout);
+
+    dht_disk_layout_extract(this, layout, 0, &disk_layout_p);
+
+    ret = dict_set_bin(xdata, conf->xattr_name, disk_layout_p, 4 * 4);
+    if (ret) {
+        gf_msg("dht", GF_LOG_DEBUG, EINVAL, DHT_MSG_DICT_SET_FAILED,
+               "dht layout dict set failed");
+    }
+wind:
+    STACK_WIND(frame, dht_pt_mkdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mkdir, loc, mode, umask, xdata);
+    if (free_xdata)
+        dict_unref(xdata);
+    return 0;
 
-        return 0;
+err:
+    op_errno = local ? local->op_errno : op_errno;
+    DHT_STACK_UNWIND(mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+
+    return 0;
 }
 
+static int
+dht_pt_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
+{
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+    dict_del(xattr, conf->xattr_name);
+    dict_del(xattr, conf->mds_xattr_key);
+    dict_del(xattr, conf->commithash_xattr_name);
+
+    if (frame->root->pid >= 0) {
+        GF_REMOVE_INTERNAL_XATTR("trusted.glusterfs.quota*", xattr);
+        GF_REMOVE_INTERNAL_XATTR("trusted.pgfid*", xattr);
+    }
+
+    DHT_STACK_UNWIND(getxattr, frame, op_ret, op_errno, xattr, xdata);
+    return 0;
+}
 
 int
-dht_notify (xlator_t *this, int event, void *data, ...)
+dht_pt_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                const char *key, dict_t *xdata)
 {
-	xlator_t   *subvol = NULL;
-	int         cnt    = -1;
-	int         i      = -1;
-	dht_conf_t *conf   = NULL;
-	int         ret    = -1;
+    STACK_WIND(frame, dht_pt_getxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->getxattr, loc, key, xdata);
+    return 0;
+}
 
-        /* Notify is on by default for all events except CHILD_UP. Do not
-         * notify parent for CHILD_UP till all distribute children have come up.
-         */
-        int         notify = 1;
+static int
+dht_pt_fgetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
+{
+    dht_conf_t *conf = NULL;
 
-	conf = this->private;
-        if (!conf)
-                return ret;
+    conf = this->private;
+    dict_del(xattr, conf->xattr_name);
 
-	switch (event) {
-	case GF_EVENT_CHILD_UP:
-		subvol = data;
-
-		conf->gen++;
-
-		for (i = 0; i < conf->subvolume_cnt; i++) {
-			if (subvol == conf->subvolumes[i]) {
-				cnt = i;
-				break;
-			}
-		}
-
-		if (cnt == -1) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"got GF_EVENT_CHILD_UP bad subvolume %s",
-				subvol->name);
-			break;
-		}
-
-		LOCK (&conf->subvolume_lock);
-		{
-                        if (!conf->subvolume_status[cnt]) {
-                                conf->subvolume_status[cnt] = 1;
-                                ++conf->upsubvols;
-                        }
+    if (frame->root->pid >= 0) {
+        GF_REMOVE_INTERNAL_XATTR("trusted.glusterfs.quota*", xattr);
+        GF_REMOVE_INTERNAL_XATTR("trusted.pgfid*", xattr);
+    }
 
-                        if (conf->upsubvols != conf->subvolume_cnt)
-                                notify = 0;
-		}
-		UNLOCK (&conf->subvolume_lock);
-
-                /* one of the node came back up, do a stat update */
-                dht_get_du_info_for_subvol (this, cnt);
-
-		break;
-
-	case GF_EVENT_CHILD_DOWN:
-		subvol = data;
-
-		for (i = 0; i < conf->subvolume_cnt; i++) {
-			if (subvol == conf->subvolumes[i]) {
-				cnt = i;
-				break;
-			}
-		}
-
-		if (cnt == -1) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"got GF_EVENT_CHILD_DOWN bad subvolume %s",
-				subvol->name);
-			break;
-		}
-
-		LOCK (&conf->subvolume_lock);
-		{
-                        if (conf->subvolume_status[cnt]) {
-                                conf->subvolume_status[cnt] = 0;
-                                --conf->upsubvols;
-                        }
-		}
-		UNLOCK (&conf->subvolume_lock);
+    DHT_STACK_UNWIND(fgetxattr, frame, op_ret, op_errno, xattr, xdata);
+    return 0;
+}
+
+int
+dht_pt_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *key,
+                 dict_t *xdata)
+{
+    STACK_WIND(frame, dht_pt_fgetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fgetxattr, fd, key, xdata);
+    return 0;
+}
 
-		break;
-	}
+/* The job of this function is to check if all the xlators have updated
+ * error in the layout. */
+int
+dht_dir_layout_error_check(xlator_t *this, inode_t *inode)
+{
+    dht_layout_t *layout = NULL;
+    int i = 0;
 
-        if (notify)
-                ret = default_notify (this, event, data);
+    layout = dht_layout_get(this, inode);
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].err == 0) {
+            return 0;
+        }
+    }
 
-	return ret;
+    /* Returning the first xlator error as all xlators have errors */
+    return layout->list[0].err;
 }
-
diff --git a/xlators/cluster/dht/src/dht-common.h b/xlators/cluster/dht/src/dht-common.h
index 5c486e4a52c..fe0dc3db34a 100644
--- a/xlators/cluster/dht/src/dht-common.h
+++ b/xlators/cluster/dht/src/dht-common.h
@@ -1,305 +1,1384 @@
 /*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+#include <regex.h>
 
 #include "dht-mem-types.h"
+#include "dht-messages.h"
+#include <glusterfs/call-stub.h>
+#include "libxlator.h"
+#include <glusterfs/syncop.h>
+#include <glusterfs/refcount.h>
+#include <glusterfs/timer.h>
+#include "protocol-common.h"
+#include <glusterfs/glusterfs-acl.h>
 
 #ifndef _DHT_H
 #define _DHT_H
 
-#define GF_XATTR_FIX_LAYOUT_KEY   "trusted.distribute.fix.layout"
-#define GF_DHT_LOOKUP_UNHASHED_ON   1
+#define GF_XATTR_FIX_LAYOUT_KEY "distribute.fix.layout"
+#define GF_XATTR_FILE_MIGRATE_KEY "trusted.distribute.migrate-data"
+#define DHT_MDS_STR "mds"
+#define GF_DHT_LOOKUP_UNHASHED_OFF 0
+#define GF_DHT_LOOKUP_UNHASHED_ON 1
 #define GF_DHT_LOOKUP_UNHASHED_AUTO 2
+#define DHT_PATHINFO_HEADER "DISTRIBUTE:"
+#define DHT_FILE_MIGRATE_DOMAIN "dht.file.migrate"
+/* Layout synchronization */
+#define DHT_LAYOUT_HEAL_DOMAIN "dht.layout.heal"
+/* Namespace synchronization */
+#define DHT_ENTRY_SYNC_DOMAIN "dht.entry.sync"
+#define DHT_LAYOUT_HASH_INVALID 1
+#define MAX_REBAL_THREADS sysconf(_SC_NPROCESSORS_ONLN)
+
+#define DHT_DIR_STAT_BLOCKS 8
+#define DHT_DIR_STAT_SIZE 4096
+
+/* Virtual xattr for subvols status */
+
+#define DHT_SUBVOL_STATUS_KEY "dht.subvol.status"
+
+/* Virtual xattrs for debugging */
+
+#define DHT_DBG_HASHED_SUBVOL_PATTERN "dht.file.hashed-subvol.*"
+#define DHT_DBG_HASHED_SUBVOL_KEY "dht.file.hashed-subvol."
 
-#include <fnmatch.h>
+/* Rebalance nodeuuid flags */
+#define REBAL_NODEUUID_MINE 0x01
 
-typedef int (*dht_selfheal_dir_cbk_t) (call_frame_t *frame, void *cookie,
-				       xlator_t *this,
-				       int32_t op_ret, int32_t op_errno);
+typedef int (*dht_selfheal_dir_cbk_t)(call_frame_t *frame, void *cookie,
+                                      xlator_t *this, int32_t op_ret,
+                                      int32_t op_errno, dict_t *xdata);
+typedef int (*dht_defrag_cbk_fn_t)(xlator_t *this, xlator_t *dst_node,
+                                   call_frame_t *frame, int ret);
 
+typedef int (*dht_refresh_layout_unlock)(call_frame_t *frame, xlator_t *this,
+                                         int op_ret, int invoke_cbk);
+
+typedef int (*dht_refresh_layout_done_handle)(call_frame_t *frame);
 
 struct dht_layout {
-        int               cnt;
-	int               preset;
-        int               gen;
-	int               type;
-        int               ref;   /* use with dht_conf_t->layout_lock */
-        int               search_unhashed;
-        struct {
-		int       err;   /* 0 = normal
-				   -1 = dir exists and no xattr
-				   >0 = dir lookup failed with errno
-				 */
-                uint32_t  start;
-                uint32_t  stop;
-                xlator_t *xlator;
-        } list[0];
+    int spread_cnt; /* layout spread count per directory,
+                       is controlled by 'setxattr()' with
+                       special key */
+    int cnt;
+    int preset;
+    /*
+     * The last *configuration* state for which this directory was known
+     * to be in balance.  The corresponding vol_commit_hash changes
+     * whenever bricks are added or removed.  This value changes when a
+     * (full) rebalance is complete.  If they match, it's safe to assume
+     * that every file is where it should be and there's no need to do
+     * lookups for files elsewhere.  If they don't, then we have to do a
+     * global lookup to be sure.
+     */
+    uint32_t commit_hash;
+    /*
+     * The *runtime* state of the volume, changes when connections to
+     * bricks are made or lost.
+     */
+    int gen;
+    int type;
+    gf_atomic_t ref; /* use with dht_conf_t->layout_lock */
+    uint32_t search_unhashed;
+    struct {
+        int err; /* 0 = normal
+                    -1 = dir exists and no xattr
+                    >0 = dir lookup failed with errno
+                 */
+        uint32_t start;
+        uint32_t stop;
+        uint32_t commit_hash;
+        xlator_t *xlator;
+    } list[];
 };
 typedef struct dht_layout dht_layout_t;
 
+struct dht_stat_time {
+    uint32_t atime;
+    uint32_t atime_nsec;
+    uint32_t ctime;
+    uint32_t ctime_nsec;
+    uint32_t mtime;
+    uint32_t mtime_nsec;
+};
+
+typedef struct dht_stat_time dht_stat_time_t;
+
+struct dht_inode_ctx {
+    dht_layout_t *layout;
+    dht_stat_time_t time;
+    xlator_t *lock_subvol;
+    xlator_t *mds_subvol; /* This is only used for directories */
+};
+
+typedef struct dht_inode_ctx dht_inode_ctx_t;
 
 typedef enum {
-	DHT_HASH_TYPE_DM,
+    DHT_HASH_TYPE_DM,
+    DHT_HASH_TYPE_DM_USER,
 } dht_hashfn_type_t;
 
+typedef enum {
+    DHT_INODELK,
+    DHT_ENTRYLK,
+} dht_lock_type_t;
+
+/* rebalance related */
+struct dht_rebalance_ {
+    xlator_t *from_subvol;
+    xlator_t *target_node;
+    off_t offset;
+    size_t size;
+    int32_t flags;
+    int count;
+    struct iobref *iobref;
+    struct iovec *vector;
+    struct iatt stbuf;
+    struct iatt prebuf;
+    struct iatt postbuf;
+    dht_defrag_cbk_fn_t target_op_fn;
+    dict_t *xdata;
+    dict_t *xattr;
+    dict_t *dict;
+    struct gf_flock flock;
+    int32_t set;
+    int lock_cmd;
+};
+
+/**
+ * Enum to store decided action based on the qdstatfs (quota-deem-statfs)
+ * events
+ **/
+typedef enum {
+    qdstatfs_action_OFF = 0,
+    qdstatfs_action_REPLACE,
+    qdstatfs_action_NEGLECT,
+    qdstatfs_action_COMPARE,
+} qdstatfs_action_t;
+
+typedef enum {
+    REACTION_INVALID,
+    FAIL_ON_ANY_ERROR,
+    IGNORE_ENOENT_ESTALE,
+    IGNORE_ENOENT_ESTALE_EIO,
+} dht_reaction_type_t;
+
+struct dht_skip_linkto_unlink {
+    xlator_t *hash_links_to;
+    uuid_t cached_gfid;
+    uuid_t hashed_gfid;
+    int opend_fd_count;
+    gf_boolean_t handle_valid_link;
+};
+
+typedef struct {
+    xlator_t *xl;
+    loc_t loc;      /* contains/points to inode to lock on. */
+    char *domain;   /* Only locks within a single domain
+                     * contend with each other
+                     */
+    char *basename; /* Required for entrylk */
+    gf_boolean_t locked;
+    dht_reaction_type_t do_on_failure;
+    short type; /* read/write lock.                     */
+    gf_lkowner_t lk_owner;
+} dht_lock_t;
+
+/* The lock structure represents inodelk. */
+typedef struct {
+    fop_inodelk_cbk_t inodelk_cbk;
+    dht_lock_t **locks;
+    int lk_count;
+    dht_reaction_type_t reaction;
+
+    /* whether locking failed on _any_ of the "locks" above */
+    int op_ret;
+    int op_errno;
+} dht_ilock_wrap_t;
+
+/* The lock structure represents entrylk. */
+typedef struct {
+    fop_entrylk_cbk_t entrylk_cbk;
+    dht_lock_t **locks;
+    int lk_count;
+    dht_reaction_type_t reaction;
+
+    /* whether locking failed on _any_ of the "locks" above */
+    int op_ret;
+    int op_errno;
+} dht_elock_wrap_t;
+
+/* The first member of dht_dir_transaction_t should be of type dht_ilock_wrap_t.
+ * Otherwise it can result in subtle memory corruption issues as in most of the
+ * places we use lock[0].layout.my_layout or lock[0].layout.parent_layout and
+ * lock[0].ns.parent_layout (like in dht_local_wipe).
+ */
+typedef union {
+    union {
+        dht_ilock_wrap_t my_layout;
+        dht_ilock_wrap_t parent_layout;
+    } layout;
+    struct dht_namespace {
+        dht_ilock_wrap_t parent_layout;
+        dht_elock_wrap_t directory_ns;
+        fop_entrylk_cbk_t ns_cbk;
+    } ns;
+} dht_dir_transaction_t;
+
+typedef int (*dht_selfheal_layout_t)(call_frame_t *frame, loc_t *loc,
+                                     dht_layout_t *layout);
+
+typedef gf_boolean_t (*dht_need_heal_t)(call_frame_t *frame,
+                                        dht_layout_t **inmem,
+                                        dht_layout_t **ondisk);
 
 struct dht_local {
-	int                      call_cnt;
-	loc_t                    loc;
-	loc_t                    loc2;
-	int                      op_ret;
-	int                      op_errno;
-	int                      layout_mismatch;
-        /* Use stbuf as the postbuf, when we require both
-         * pre and post attrs */
-	struct iatt              stbuf;
-        struct iatt              prebuf;
-        struct iatt              preoldparent;
-        struct iatt              postoldparent;
-        struct iatt              preparent;
-        struct iatt              postparent;
-	struct statvfs           statvfs;
-	fd_t                    *fd;
-	inode_t                 *inode;
-        dict_t                  *params;
-	dict_t                  *xattr;
-	dict_t                  *xattr_req;
-	dht_layout_t            *layout;
-	size_t                   size;
-	ino_t                    ia_ino;
-	xlator_t                *src_hashed, *src_cached;
-	xlator_t                *dst_hashed, *dst_cached;
-	xlator_t                *cached_subvol;
-	xlator_t                *hashed_subvol;
-	char                     need_selfheal;
-        int                      file_count;
-        int                      dir_count;
-        call_frame_t            *main_frame;
-	struct {
-		fop_mknod_cbk_t  linkfile_cbk;
-		struct iatt      stbuf;
-		loc_t            loc;
-		inode_t         *inode;
-		dict_t          *xattr;
-		xlator_t        *srcvol;
-	} linkfile;
-	struct {
-		uint32_t         hole_cnt;
-		uint32_t         overlaps_cnt;
-		uint32_t         missing;
-		uint32_t         down;
-		uint32_t         misc;
-		dht_selfheal_dir_cbk_t   dir_cbk;
-		dht_layout_t    *layout;
-	} selfheal;
-        uint32_t                 uid;
-        uint32_t                 gid;
-
-	/* needed by nufa */
-	int32_t flags;
-	mode_t  mode;
-	dev_t   rdev;
-
-        /* need for file-info */
-        char   *pathinfo;
-        char   *key;
-
-        char   *newpath;
-
-        /* gfid related */
-        uuid_t  gfid;
+    loc_t loc;
+    loc_t loc2;
+    int call_cnt;
+    int op_ret;
+    int op_errno;
+    int layout_mismatch;
+    /* Use stbuf as the postbuf, when we require both
+     * pre and post attrs */
+    struct iatt stbuf;
+    struct iatt mds_stbuf;
+    struct iatt prebuf;
+    struct iatt preoldparent;
+    struct iatt postoldparent;
+    struct iatt preparent;
+    struct iatt postparent;
+    struct statvfs statvfs;
+    fd_t *fd;
+    inode_t *inode;
+    dict_t *params;
+    dict_t *xattr;
+    dict_t *mds_xattr;
+    dict_t *xdata; /* dict used to save xdata response by xattr fop */
+    dict_t *xattr_req;
+    dht_layout_t *layout;
+    size_t size;
+    ino_t ia_ino;
+    xlator_t *src_hashed, *src_cached;
+    xlator_t *dst_hashed, *dst_cached;
+    xlator_t *cached_subvol;
+    xlator_t *hashed_subvol;
+    xlator_t *mds_subvol; /* This is use for dir only */
+    int file_count;
+    int dir_count;
+    call_frame_t *main_frame;
+    int fop_succeeded;
+    struct {
+        fop_mknod_cbk_t linkfile_cbk;
+        struct iatt stbuf;
+        loc_t loc;
+        inode_t *inode;
+        dict_t *xattr;
+        xlator_t *srcvol;
+    } linkfile;
+    struct {
+        uint32_t hole_cnt;
+        uint32_t overlaps_cnt;
+        uint32_t down;
+        uint32_t misc;
+        dht_selfheal_dir_cbk_t dir_cbk;
+        dht_selfheal_layout_t healer;
+        dht_need_heal_t should_heal;
+        dht_layout_t *layout, *refreshed_layout;
+        uint32_t missing_cnt;
+        gf_boolean_t force_mkdir;
+    } selfheal;
+
+    dht_refresh_layout_unlock refresh_layout_unlock;
+    dht_refresh_layout_done_handle refresh_layout_done;
+
+    uint32_t uid;
+    uint32_t gid;
+    pid_t pid;
+
+    glusterfs_fop_t fop;
+
+    /* need for file-info */
+    char *xattr_val;
+    char *key;
+
+    /* needed by nufa */
+    int32_t flags;
+    mode_t mode;
+    dev_t rdev;
+    mode_t umask;
+
+    /* which xattr request? */
+    char xsel[256];
+    int32_t alloc_len;
+
+    /* gfid related */
+    uuid_t gfid;
+    uuid_t gfid_req;
+
+    xlator_t *link_subvol;
+
+    struct dht_rebalance_ rebalance;
+    xlator_t *first_up_subvol;
+
+    struct dht_skip_linkto_unlink skip_unlink;
+
+    dht_dir_transaction_t lock[2], *current;
+
+    /* inodelks during filerename for backward compatibility */
+    dht_lock_t **rename_inodelk_backward_compatible;
+
+    call_stub_t *stub;
+    int32_t parent_disk_layout[4];
+
+    /* rename rollback */
+    int *ret_cache;
+
+    loc_t loc2_copy;
+
+    int rename_inodelk_bc_count;
+    /* This is use only for directory operation */
+    int32_t valid;
+    int32_t mds_heal_fresh_lookup;
+    short lock_type;
+    char need_selfheal;
+    char need_xattr_heal;
+    char need_attrheal;
+    /* flag used to make sure we need to return estale in
+       {lookup,revalidate}_cbk */
+    char return_estale;
+    char need_lookup_everywhere;
+    /* fd open check */
+    gf_boolean_t fd_checked;
+    gf_boolean_t linked;
+    gf_boolean_t added_link;
+    gf_boolean_t is_linkfile;
+    gf_boolean_t quota_deem_statfs;
+    gf_boolean_t heal_layout;
+    gf_boolean_t locked;
+    gf_boolean_t dont_create_linkto;
+    gf_boolean_t gfid_missing;
 };
 typedef struct dht_local dht_local_t;
 
 /* du - disk-usage */
 struct dht_du {
-        double   avail_percent;
-        uint64_t avail_space;
-        uint32_t log;
+    double avail_percent;
+    double avail_inodes;
+    uint64_t avail_space;
+    uint32_t log;
+    uint32_t chunks;
+    uint32_t total_blocks;
+    uint32_t avail_blocks;
+    uint32_t frsize; /*fragment size*/
 };
 typedef struct dht_du dht_du_t;
 
+enum gf_defrag_type {
+    GF_DEFRAG_CMD_NONE = 0,
+    GF_DEFRAG_CMD_START = 1,
+    GF_DEFRAG_CMD_STOP = 1 + 1,
+    GF_DEFRAG_CMD_STATUS = 1 + 2,
+    GF_DEFRAG_CMD_START_LAYOUT_FIX = 1 + 3,
+    GF_DEFRAG_CMD_START_FORCE = 1 + 4,
+    GF_DEFRAG_CMD_DETACH_STATUS = 1 + 11,
+    GF_DEFRAG_CMD_DETACH_START = 1 + 13,
+    GF_DEFRAG_CMD_DETACH_COMMIT = 1 + 14,
+    GF_DEFRAG_CMD_DETACH_COMMIT_FORCE = 1 + 15,
+    GF_DEFRAG_CMD_DETACH_STOP = 1 + 16,
+    /* new labels are used so it will help
+     * while removing old labels by easily differentiating.
+     * A few labels are added so that the count remains same
+     * between this enum and the ones on the xdr file.
+     * different values for the same enum cause errors and
+     * confusion.
+     */
+};
+typedef enum gf_defrag_type gf_defrag_type;
+
+enum gf_defrag_status_t {
+    GF_DEFRAG_STATUS_NOT_STARTED,
+    GF_DEFRAG_STATUS_STARTED,
+    GF_DEFRAG_STATUS_STOPPED,
+    GF_DEFRAG_STATUS_COMPLETE,
+    GF_DEFRAG_STATUS_FAILED,
+    GF_DEFRAG_STATUS_LAYOUT_FIX_STARTED,
+    GF_DEFRAG_STATUS_LAYOUT_FIX_STOPPED,
+    GF_DEFRAG_STATUS_LAYOUT_FIX_COMPLETE,
+    GF_DEFRAG_STATUS_LAYOUT_FIX_FAILED,
+};
+typedef enum gf_defrag_status_t gf_defrag_status_t;
+
+typedef struct gf_defrag_pattern_list gf_defrag_pattern_list_t;
+
+struct gf_defrag_pattern_list {
+    char path_pattern[256];
+    uint64_t size;
+    gf_defrag_pattern_list_t *next;
+};
+
+struct dht_container {
+    union {
+        struct list_head list;
+        struct {
+            struct _gf_dirent_t *next;
+            struct _gf_dirent_t *prev;
+        };
+    };
+    gf_dirent_t *df_entry;
+    xlator_t *this;
+    loc_t *parent_loc;
+    dict_t *migrate_data;
+    int local_subvol_index;
+};
+
+typedef struct nodeuuid_info {
+    char info;   /* Set to 1 is this is my node's uuid*/
+    uuid_t uuid; /* Store the nodeuuid as well for debugging*/
+} nodeuuid_info_t;
+
+typedef struct subvol_nodeuuids_info {
+    nodeuuid_info_t *elements;
+    int count;
+} subvol_nodeuuids_info_t;
+
+struct gf_defrag_info_ {
+    uint64_t total_files;
+    uint64_t total_data;
+    uint64_t num_files_lookedup;
+    uint64_t total_failures;
+    uint64_t skipped;
+    uint64_t num_dirs_processed;
+    uint64_t size_processed;
+    gf_lock_t lock;
+    pthread_t th;
+    struct rpc_clnt *rpc;
+    uint32_t connected;
+    uint32_t is_exiting;
+    pid_t pid;
+    int cmd;
+    inode_t *root_inode;
+    uuid_t node_uuid;
+    time_t start_time;
+    uint32_t new_commit_hash;
+    gf_defrag_status_t defrag_status;
+    gf_defrag_pattern_list_t *defrag_pattern;
+
+    pthread_cond_t parallel_migration_cond;
+    pthread_mutex_t dfq_mutex;
+    pthread_cond_t rebalance_crawler_alarm;
+    int32_t q_entry_count;
+    int32_t global_error;
+    struct dht_container *queue;
+    int32_t crawl_done;
+    int32_t abort;
+    int32_t wakeup_crawler;
+
+    /*Throttle params*/
+    /*stands for reconfigured thread count*/
+    int32_t recon_thread_count;
+    pthread_cond_t df_wakeup_thread;
+
+    /* backpointer to make it easier to write functions for rebalance */
+    xlator_t *this;
+
+    pthread_cond_t fc_wakeup_cond;
+    pthread_mutex_t fc_mutex;
+
+    /*stands for current running thread count*/
+    int32_t current_thread_count;
+
+    gf_boolean_t stats;
+    /* lock migration flag */
+    gf_boolean_t lock_migration_enabled;
+};
+
+typedef struct gf_defrag_info_ gf_defrag_info_t;
+
+struct dht_methods_s {
+    int32_t (*migration_get_dst_subvol)(xlator_t *this, dht_local_t *local);
+    int32_t (*migration_other)(xlator_t *this, gf_defrag_info_t *defrag);
+    xlator_t *(*layout_search)(xlator_t *this, dht_layout_t *layout,
+                               const char *name);
+};
+
+typedef struct dht_methods_s dht_methods_t;
+
 struct dht_conf {
-	gf_lock_t      subvolume_lock;
-        int            subvolume_cnt;
-        xlator_t     **subvolumes;
-	char          *subvolume_status;
-	dht_layout_t **file_layouts;
-	dht_layout_t **dir_layouts;
-	dht_layout_t  *default_dir_layout;
-	gf_boolean_t   search_unhashed;
-	int            gen;
-        dht_du_t      *du_stats;
-        uint64_t       min_free_disk;
-        char           disk_unit;
-        int32_t        refresh_interval;
-        gf_boolean_t   unhashed_sticky_bit;
-	struct timeval last_stat_fetch;
-        gf_lock_t      layout_lock;
-        void          *private;     /* Can be used by wrapper xlators over
-                                       dht */
-        gf_boolean_t   use_readdirp;
-        int             upsubvols;      /* Count the number of online children*/
+    xlator_t **subvolumes;
+    char *subvolume_status;
+    int *last_event;
+    dht_layout_t **file_layouts;
+    dht_layout_t **dir_layouts;
+    unsigned int search_unhashed;
+    int gen;
+    dht_du_t *du_stats;
+    double min_free_disk;
+    double min_free_inodes;
+    int subvolume_cnt;
+    int32_t refresh_interval;
+    gf_lock_t subvolume_lock;
+    time_t last_stat_fetch;
+    gf_lock_t layout_lock;
+    dict_t *leaf_to_subvol;
+    void *private; /* Can be used by wrapper xlators over
+                      dht */
+    time_t *subvol_up_time;
+
+    /* to keep track of nodes which are decommissioned */
+    xlator_t **decommissioned_bricks;
+    int decommission_in_progress;
+    int decommission_subvols_cnt;
+
+    /* defrag related */
+    gf_defrag_info_t *defrag;
+
+    /* Support regex-based name reinterpretation. */
+    regex_t rsync_regex;
+    regex_t extra_regex;
+
+    /* Support variable xattr names. */
+    char *xattr_name;
+    char *mds_xattr_key;
+    char *link_xattr_name;
+    char *commithash_xattr_name;
+    char *wild_xattr_name;
+
+    dht_methods_t methods;
+
+    struct mem_pool *lock_pool;
+
+    /*local subvol storage for rebalance*/
+    xlator_t **local_subvols;
+    subvol_nodeuuids_info_t *local_nodeuuids;
+    int32_t local_subvols_cnt;
+
+    int dthrottle;
+
+    /* Hard link handle requirement for migration triggered from client*/
+    synclock_t link_lock;
+
+    /* lock migration */
+    gf_lock_t lock;
+
+    /* This is the count used as the distribute layout for a directory */
+    /* Will be a global flag to control the layout spread count */
+    uint32_t dir_spread_cnt;
+
+    /*
+     * "Commit hash" for this volume topology.  Changed whenever bricks
+     * are added or removed.
+     */
+    uint32_t vol_commit_hash;
+
+    char vol_uuid[UUID_SIZE + 1];
+
+    char disk_unit;
+
+    gf_boolean_t lock_migration_enabled;
+
+    gf_boolean_t vch_forced;
+
+    gf_boolean_t use_fallocate;
+
+    gf_boolean_t force_migration;
+
+    gf_boolean_t lookup_optimize;
+
+    gf_boolean_t unhashed_sticky_bit;
+
+    gf_boolean_t assert_no_child_down;
+
+    gf_boolean_t use_readdirp;
+
+    /* Request to filter directory entries in readdir request */
+    gf_boolean_t readdir_optimize;
+
+    gf_boolean_t rsync_regex_valid;
+
+    gf_boolean_t extra_regex_valid;
+
+    /* Support size-weighted rebalancing (heterogeneous bricks). */
+    gf_boolean_t do_weighting;
+
+    gf_boolean_t randomize_by_gfid;
 };
 typedef struct dht_conf dht_conf_t;
 
+struct dht_dfoffset_ctx {
+    xlator_t *this;
+    off_t offset;
+    int32_t readdir_done;
+};
+typedef struct dht_dfoffset_ctx dht_dfoffset_ctx_t;
 
 struct dht_disk_layout {
-	uint32_t           cnt;
-	uint32_t           type;
-	struct {
-		uint32_t   start;
-		uint32_t   stop;
-	} list[1];
+    uint32_t cnt;
+    uint32_t type;
+    struct {
+        uint32_t start;
+        uint32_t stop;
+    } list[1];
 };
 typedef struct dht_disk_layout dht_disk_layout_t;
 
-#define WIPE(statp) do { typeof(*statp) z = {0,}; if (statp) *statp = z; } while (0)
+typedef enum {
+    GF_DHT_MIGRATE_DATA,
+    GF_DHT_MIGRATE_DATA_EVEN_IF_LINK_EXISTS,
+    GF_DHT_MIGRATE_HARDLINK,
+    GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS
+} gf_dht_migrate_data_type_t;
+
+typedef enum {
+    GF_DHT_EQUAL_DISTRIBUTION,
+    GF_DHT_WEIGHTED_DISTRIBUTION
+} dht_distribution_type_t;
 
-#define ENTRY_MISSING(op_ret, op_errno) (op_ret == -1 && op_errno == ENOENT)
+struct dir_dfmeta {
+    gf_dirent_t *equeue;
+    dht_dfoffset_ctx_t *offset_var;
+    struct list_head **head;
+    struct list_head **iterator;
+    int *fetch_entries;
+    /* fds corresponding to local subvols only */
+    fd_t **lfd;
+};
 
-#define is_fs_root(loc) (strcmp (loc->path, "/") == 0)
+typedef struct dht_migrate_info {
+    xlator_t *src_subvol;
+    xlator_t *dst_subvol;
+    GF_REF_DECL;
+} dht_migrate_info_t;
 
-#define is_revalidate(loc) (inode_ctx_get (loc->inode, this, NULL) == 0)
+typedef struct dht_fd_ctx {
+    uint64_t opened_on_dst;
+    GF_REF_DECL;
+} dht_fd_ctx_t;
+
+#define ENTRY_MISSING(op_ret, op_errno) (op_ret == -1 && op_errno == ENOENT)
+
+#define is_revalidate(loc)                                                     \
+    (dht_inode_ctx_layout_get((loc)->inode, this, NULL) == 0)
 
 #define is_last_call(cnt) (cnt == 0)
 
-#define DHT_LINKFILE_MODE (S_ISVTX)
-#define check_is_linkfile(i,s,x) (                                      \
-                ((st_mode_from_ia (s->ia_prot, s->ia_type) & ~S_IFMT)   \
-                 == DHT_LINKFILE_MODE) &&                               \
-                (s->ia_size == 0))
+#define DHT_MIGRATION_IN_PROGRESS 1
+#define DHT_MIGRATION_COMPLETED 2
+
+#define check_is_linkfile(i, s, x, n)                                          \
+    (IS_DHT_LINKFILE_MODE(s) && dict_get(x, n))
+
+#define IS_DHT_MIGRATION_PHASE2(buf)                                           \
+    (IA_ISREG((buf)->ia_type) &&                                               \
+     ((st_mode_from_ia((buf)->ia_prot, (buf)->ia_type) & ~S_IFMT) ==           \
+      DHT_LINKFILE_MODE))
+
+#define IS_DHT_MIGRATION_PHASE1(buf)                                           \
+    (IA_ISREG((buf)->ia_type) && ((buf)->ia_prot.sticky == 1) &&               \
+     ((buf)->ia_prot.sgid == 1))
 
-#define check_is_dir(i,s,x) (IA_ISDIR(s->ia_type))
+#define DHT_STRIP_PHASE1_FLAGS(buf)                                            \
+    do {                                                                       \
+        if ((buf) && IS_DHT_MIGRATION_PHASE1(buf)) {                           \
+            (buf)->ia_prot.sticky = 0;                                         \
+            (buf)->ia_prot.sgid = 0;                                           \
+        }                                                                      \
+    } while (0)
+
+#define dht_inode_missing(op_errno) (op_errno == ENOENT || op_errno == ESTALE)
+
+#define check_is_dir(i, s, x) (IA_ISDIR(s->ia_type))
 
 #define layout_is_sane(layout) ((layout) && (layout->cnt > 0))
 
-#define DHT_STACK_UNWIND(fop, frame, params ...) do {           \
-		dht_local_t *__local = NULL;                    \
-                xlator_t *__xl = NULL;                          \
-                if (frame) {                                    \
-                        __xl = frame->this;                     \
-                        __local = frame->local;                 \
-                        frame->local = NULL;                    \
-                }                                               \
-		STACK_UNWIND_STRICT (fop, frame, params);       \
-		dht_local_wipe (__xl, __local);                 \
-	} while (0)
-
-#define DHT_STACK_DESTROY(frame) do {		       \
-		dht_local_t *__local = NULL;           \
-                xlator_t *__xl = NULL;                 \
-                __xl = frame->this;                    \
-		__local = frame->local;                \
-		frame->local = NULL;		       \
-		STACK_DESTROY (frame->root);	       \
-		dht_local_wipe (__xl, __local);	       \
-	} while (0)
-
-dht_layout_t *dht_layout_new (xlator_t *this, int cnt);
-dht_layout_t *dht_layout_get (xlator_t *this, inode_t *inode);
-dht_layout_t *dht_layout_for_subvol (xlator_t *this, xlator_t *subvol);
-xlator_t *dht_layout_search (xlator_t *this, dht_layout_t *layout,
-			     const char *name);
-int dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout);
-int dht_layout_anomalies (xlator_t *this, loc_t *loc, dht_layout_t *layout,
-			  uint32_t *holes_p, uint32_t *overlaps_p,
-			  uint32_t *missing_p, uint32_t *down_p,
-			  uint32_t *misc_p);
-int dht_layout_dir_mismatch (xlator_t *this, dht_layout_t *layout,
-			     xlator_t *subvol, loc_t *loc, dict_t *xattr);
-
-xlator_t *dht_linkfile_subvol (xlator_t *this, inode_t *inode,
-			       struct iatt *buf, dict_t *xattr);
-int dht_linkfile_unlink (call_frame_t *frame, xlator_t *this,
-			 xlator_t *subvol, loc_t *loc);
-
-int dht_layouts_init (xlator_t *this, dht_conf_t *conf);
-int dht_layout_merge (xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
-		      int op_ret, int op_errno, dict_t *xattr);
-
-int dht_disk_layout_extract (xlator_t *this, dht_layout_t *layout,
-			     int pos, int32_t **disk_layout_p);
-int dht_disk_layout_merge (xlator_t *this, dht_layout_t *layout,
-			   int pos, void *disk_layout_raw);
-
-
-int dht_frame_return (call_frame_t *frame);
-
-int dht_itransform (xlator_t *this, xlator_t *subvol, uint64_t x, uint64_t *y);
-int dht_deitransform (xlator_t *this, uint64_t y, xlator_t **subvol,
-		      uint64_t *x);
-
-void dht_local_wipe (xlator_t *this, dht_local_t *local);
-dht_local_t *dht_local_init (call_frame_t *frame);
-int dht_iatt_merge (xlator_t *this, struct iatt *to, struct iatt *from,
-		    xlator_t *subvol);
-
-xlator_t *dht_subvol_get_hashed (xlator_t *this, loc_t *loc);
-xlator_t *dht_subvol_get_cached (xlator_t *this, inode_t *inode);
-xlator_t *dht_subvol_next (xlator_t *this, xlator_t *prev);
-int dht_subvol_cnt (xlator_t *this, xlator_t *subvol);
-
-int dht_hash_compute (int type, const char *name, uint32_t *hash_p);
-
-int dht_linkfile_create (call_frame_t *frame, fop_mknod_cbk_t linkfile_cbk,
-			 xlator_t *tovol, xlator_t *fromvol, loc_t *loc);
-int dht_lookup_directory (call_frame_t *frame, xlator_t *this, loc_t *loc);
-int dht_lookup_everywhere (call_frame_t *frame, xlator_t *this, loc_t *loc);
-int
-dht_selfheal_directory (call_frame_t *frame, dht_selfheal_dir_cbk_t cbk,
-			loc_t *loc, dht_layout_t *layout);
-int
-dht_selfheal_new_directory (call_frame_t *frame, dht_selfheal_dir_cbk_t cbk,
-			    dht_layout_t *layout);
-int
-dht_selfheal_restore (call_frame_t *frame, dht_selfheal_dir_cbk_t cbk,
-		      loc_t *loc, dht_layout_t *layout);
-int
-dht_layout_sort_volname (dht_layout_t *layout);
-
-int dht_rename (call_frame_t *frame, xlator_t *this,
-		loc_t *oldloc, loc_t *newloc);
-
-int dht_get_du_info (call_frame_t *frame, xlator_t *this, loc_t *loc);
-
-int dht_is_subvol_filled (xlator_t *this, xlator_t *subvol);
-xlator_t *dht_free_disk_available_subvol (xlator_t *this, xlator_t *subvol);
-int dht_get_du_info_for_subvol (xlator_t *this, int subvol_idx);
-
-int dht_layout_preset (xlator_t *this, xlator_t *subvol, inode_t *inode);
-int dht_layout_set (xlator_t *this, inode_t *inode, dht_layout_t *layout);
-void dht_layout_unref (xlator_t *this, dht_layout_t *layout);
-dht_layout_t *dht_layout_ref (xlator_t *this, dht_layout_t *layout);
-xlator_t *dht_first_up_subvol (xlator_t *this);
-xlator_t *dht_last_up_subvol (xlator_t *this);
-int dht_frame_su_do (call_frame_t *frame);
-int dht_frame_su_undo (call_frame_t *frame);
-
-int dht_build_child_loc (xlator_t *this, loc_t *child, loc_t *parent, char *name);
-
-int dht_filter_loc_subvol_key (xlator_t *this, loc_t *loc, loc_t *new_loc,
-                               xlator_t **subvol);
+#define we_are_not_migrating(x) ((x) == 1)
+
+#define DHT_STACK_UNWIND(fop, frame, params...)                                \
+    do {                                                                       \
+        dht_local_t *__local = NULL;                                           \
+        xlator_t *__xl = NULL;                                                 \
+        if (frame) {                                                           \
+            __xl = frame->this;                                                \
+            __local = frame->local;                                            \
+            frame->local = NULL;                                               \
+        }                                                                      \
+        STACK_UNWIND_STRICT(fop, frame, params);                               \
+        dht_local_wipe(__xl, __local);                                         \
+    } while (0)
+
+#define DHT_STACK_DESTROY(frame)                                               \
+    do {                                                                       \
+        dht_local_t *__local = NULL;                                           \
+        xlator_t *__xl = NULL;                                                 \
+        __xl = frame->this;                                                    \
+        __local = frame->local;                                                \
+        frame->local = NULL;                                                   \
+        STACK_DESTROY(frame->root);                                            \
+        dht_local_wipe(__xl, __local);                                         \
+    } while (0)
+
+#define DHT_UPDATE_TIME(ctx_sec, ctx_nsec, new_sec, new_nsec, post)            \
+    do {                                                                       \
+        if (ctx_sec == new_sec)                                                \
+            new_nsec = max(new_nsec, ctx_nsec);                                \
+        else if (ctx_sec > new_sec) {                                          \
+            new_sec = ctx_sec;                                                 \
+            new_nsec = ctx_nsec;                                               \
+        }                                                                      \
+        if (post) {                                                            \
+            ctx_sec = new_sec;                                                 \
+            ctx_nsec = new_nsec;                                               \
+        }                                                                      \
+    } while (0)
+
+#define is_greater_time(a, an, b, bn)                                          \
+    (((a) < (b)) || (((a) == (b)) && ((an) < (bn))))
+
+#define DHT_MARK_FOP_INTERNAL(xattr)                                           \
+    do {                                                                       \
+        int tmp = -1;                                                          \
+        if (!xattr) {                                                          \
+            xattr = dict_new();                                                \
+            if (!xattr)                                                        \
+                break;                                                         \
+        }                                                                      \
+        tmp = dict_set_str(xattr, GLUSTERFS_INTERNAL_FOP_KEY, "yes");          \
+        if (tmp) {                                                             \
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,       \
+                   "Failed to set dictionary value: key = %s,"                 \
+                   " path = %s",                                               \
+                   GLUSTERFS_INTERNAL_FOP_KEY, local->loc.path);               \
+        }                                                                      \
+    } while (0)
+
+dht_layout_t *
+dht_layout_new(xlator_t *this, int cnt);
+dht_layout_t *
+dht_layout_get(xlator_t *this, inode_t *inode);
+dht_layout_t *
+dht_layout_for_subvol(xlator_t *this, xlator_t *subvol);
+xlator_t *
+dht_layout_search(xlator_t *this, dht_layout_t *layout, const char *name);
+int32_t
+dht_migration_get_dst_subvol(xlator_t *this, dht_local_t *local);
+int32_t
+dht_migration_needed(xlator_t *this);
+int
+dht_layout_normalize(xlator_t *this, loc_t *loc, dht_layout_t *layout);
+void
+dht_layout_anomalies(xlator_t *this, loc_t *loc, dht_layout_t *layout,
+                     uint32_t *holes_p, uint32_t *overlaps_p,
+                     uint32_t *missing_p, uint32_t *down_p, uint32_t *misc_p,
+                     uint32_t *no_space_p);
+int
+dht_layout_dir_mismatch(xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
+                        loc_t *loc, dict_t *xattr);
+xlator_t *
+dht_linkfile_subvol(xlator_t *this, inode_t *inode, struct iatt *buf,
+                    dict_t *xattr);
+int
+dht_linkfile_unlink(call_frame_t *frame, xlator_t *this, xlator_t *subvol,
+                    loc_t *loc);
+
+int
+dht_layouts_init(xlator_t *this, dht_conf_t *conf);
+int
+dht_layout_merge(xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
+                 int op_ret, int op_errno, dict_t *xattr);
+
+int
+dht_disk_layout_extract(xlator_t *this, dht_layout_t *layout, int pos,
+                        int32_t **disk_layout_p);
+int
+dht_disk_layout_extract_for_subvol(xlator_t *this, dht_layout_t *layout,
+                                   xlator_t *subvol, int32_t **disk_layout_p);
+
+int
+dht_frame_return(call_frame_t *frame);
+
+int
+dht_deitransform(xlator_t *this, uint64_t y, xlator_t **subvol);
 
+void
+dht_local_wipe(xlator_t *this, dht_local_t *local);
+dht_local_t *
+dht_local_init(call_frame_t *frame, loc_t *loc, fd_t *fd, glusterfs_fop_t fop);
+int
+dht_iatt_merge(xlator_t *this, struct iatt *to, struct iatt *from);
+
+xlator_t *
+dht_subvol_get_hashed(xlator_t *this, loc_t *loc);
+xlator_t *
+dht_subvol_get_cached(xlator_t *this, inode_t *inode);
+xlator_t *
+dht_subvol_next(xlator_t *this, xlator_t *prev);
+xlator_t *
+dht_subvol_next_available(xlator_t *this, xlator_t *prev);
+int
+dht_subvol_cnt(xlator_t *this, xlator_t *subvol);
+
+int
+dht_hash_compute(xlator_t *this, int type, const char *name, uint32_t *hash_p);
+
+int
+dht_linkfile_create(call_frame_t *frame, fop_mknod_cbk_t linkfile_cbk,
+                    xlator_t *this, xlator_t *tovol, xlator_t *fromvol,
+                    loc_t *loc);
+int
+dht_lookup_everywhere(call_frame_t *frame, xlator_t *this, loc_t *loc);
+int
+dht_selfheal_directory(call_frame_t *frame, dht_selfheal_dir_cbk_t cbk,
+                       loc_t *loc, dht_layout_t *layout);
+int
+dht_selfheal_new_directory(call_frame_t *frame, dht_selfheal_dir_cbk_t cbk,
+                           dht_layout_t *layout);
+int
+dht_selfheal_restore(call_frame_t *frame, dht_selfheal_dir_cbk_t cbk,
+                     loc_t *loc, dht_layout_t *layout);
+void
+dht_layout_sort_volname(dht_layout_t *layout);
+
+int
+dht_get_du_info(call_frame_t *frame, xlator_t *this, loc_t *loc);
+
+gf_boolean_t
+dht_is_subvol_filled(xlator_t *this, xlator_t *subvol);
+xlator_t *
+dht_free_disk_available_subvol(xlator_t *this, xlator_t *subvol,
+                               dht_local_t *layout);
+int
+dht_get_du_info_for_subvol(xlator_t *this, int subvol_idx);
 
+int
+dht_layout_preset(xlator_t *this, xlator_t *subvol, inode_t *inode);
+int
+dht_layout_set(xlator_t *this, inode_t *inode, dht_layout_t *layout);
+;
+void
+dht_layout_unref(xlator_t *this, dht_layout_t *layout);
+dht_layout_t *
+dht_layout_ref(xlator_t *this, dht_layout_t *layout);
+int
+dht_layout_index_for_subvol(dht_layout_t *layout, xlator_t *subvol);
+xlator_t *
+dht_first_up_subvol(xlator_t *this);
+xlator_t *
+dht_last_up_subvol(xlator_t *this);
+
+int
+dht_build_child_loc(xlator_t *this, loc_t *child, loc_t *parent, char *name);
+
+int
+dht_filter_loc_subvol_key(xlator_t *this, loc_t *loc, loc_t *new_loc,
+                          xlator_t **subvol);
+
+int
+dht_rename_cleanup(call_frame_t *frame);
+int
+dht_rename_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, inode_t *inode,
+                    struct iatt *stbuf, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata);
+
+int
+dht_update_commit_hash_for_layout(call_frame_t *frame);
+int
+dht_fix_directory_layout(call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk,
+                         dht_layout_t *layout);
+
+int
+dht_init_subvolumes(xlator_t *this, dht_conf_t *conf);
+
+/* migration/rebalance */
+int
+dht_start_rebalance_task(xlator_t *this, call_frame_t *frame);
+
+int
+dht_rebalance_in_progress_check(xlator_t *this, call_frame_t *frame);
+int
+dht_rebalance_complete_check(xlator_t *this, call_frame_t *frame);
+
+int
+dht_init_local_subvolumes(xlator_t *this, dht_conf_t *conf);
+
+/* FOPS */
+int32_t
+dht_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req);
+
+int32_t
+dht_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata);
+
+int32_t
+dht_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata);
+
+int32_t
+dht_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+             dict_t *xdata);
+
+int32_t
+dht_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+              dict_t *xdata);
+
+int32_t
+dht_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+           dict_t *xdata);
+
+int32_t
+dht_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+             dict_t *xdata);
+
+int32_t
+dht_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+          dev_t rdev, mode_t umask, dict_t *xdata);
+
+int32_t
+dht_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+          mode_t umask, dict_t *xdata);
+
+int32_t
+dht_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+           dict_t *xdata);
+
+int32_t
+dht_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+          dict_t *xdata);
+
+int32_t
+dht_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+            loc_t *loc, mode_t umask, dict_t *xdata);
+
+int32_t
+dht_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+           dict_t *xdata);
+
+int32_t
+dht_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+         dict_t *xdata);
+
+int32_t
+dht_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+           mode_t mode, mode_t umask, fd_t *fd, dict_t *params);
+
+int32_t
+dht_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+         fd_t *fd, dict_t *xdata);
+
+int32_t
+dht_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+          off_t offset, uint32_t flags, dict_t *xdata);
+
+int32_t
+dht_writev(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
+           int32_t count, off_t offset, uint32_t flags, struct iobref *iobref,
+           dict_t *xdata);
+
+int32_t
+dht_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata);
+
+int32_t
+dht_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+          dict_t *xdata);
+
+int32_t
+dht_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+            dict_t *xdata);
+
+int32_t
+dht_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+             dict_t *xdata);
+
+int32_t
+dht_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata);
+
+int32_t
+dht_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+             int32_t flags, dict_t *xdata);
+
+int32_t
+dht_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, const char *name,
+             dict_t *xdata);
+
+int32_t
+dht_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+              int32_t flags, dict_t *xdata);
+
+int32_t
+dht_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+              dict_t *xdata);
+
+int32_t
+dht_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                const char *name, dict_t *xdata);
+int32_t
+dht_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                 const char *name, dict_t *xdata);
+
+int32_t
+dht_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+       struct gf_flock *flock, dict_t *xdata);
+
+int32_t
+dht_lease(call_frame_t *frame, xlator_t *this, loc_t *loc,
+          struct gf_lease *lease, dict_t *xdata);
+
+int32_t
+dht_inodelk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+            int32_t cmd, struct gf_flock *flock, dict_t *xdata);
+
+int32_t
+dht_finodelk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+             int32_t cmd, struct gf_flock *flock, dict_t *xdata);
+
+int32_t
+dht_entrylk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+            const char *basename, entrylk_cmd cmd, entrylk_type type,
+            dict_t *xdata);
+
+int32_t
+dht_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+             const char *basename, entrylk_cmd cmd, entrylk_type type,
+             dict_t *xdata);
+
+int32_t
+dht_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t off, dict_t *xdata);
+
+int32_t
+dht_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+             off_t off, dict_t *dict);
+
+int32_t
+dht_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+            gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata);
+
+int32_t
+dht_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+             gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata);
+
+int32_t
+dht_forget(xlator_t *this, inode_t *inode);
+int32_t
+dht_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *stbuf,
+            int32_t valid, dict_t *xdata);
+int32_t
+dht_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
+             int32_t valid, dict_t *xdata);
+int32_t
+dht_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+              off_t offset, size_t len, dict_t *xdata);
+int32_t
+dht_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+            size_t len, dict_t *xdata);
+int32_t
+dht_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             off_t len, dict_t *xdata);
+int32_t
+dht_ipc(call_frame_t *frame, xlator_t *this, int32_t op, dict_t *xdata);
+
+int
+dht_set_subvol_range(xlator_t *this);
+int32_t
+dht_init(xlator_t *this);
+void
+dht_fini(xlator_t *this);
+int
+dht_reconfigure(xlator_t *this, dict_t *options);
+int32_t
+dht_notify(xlator_t *this, int32_t event, void *data, ...);
+
+/* definitions for nufa/switch */
+int
+dht_revalidate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int op_ret, int op_errno, inode_t *inode, struct iatt *stbuf,
+                   dict_t *xattr, struct iatt *postparent);
+int
+dht_lookup_dir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int op_ret, int op_errno, inode_t *inode, struct iatt *stbuf,
+                   dict_t *xattr, struct iatt *postparent);
+int
+dht_lookup_linkfile_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, inode_t *inode,
+                        struct iatt *stbuf, dict_t *xattr,
+                        struct iatt *postparent);
+int
+dht_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, inode_t *inode, struct iatt *stbuf, dict_t *xattr,
+               struct iatt *postparent);
+int
+dht_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, fd_t *fd, inode_t *inode, struct iatt *stbuf,
+               struct iatt *preparent, struct iatt *postparent, dict_t *xdata);
+int
+dht_newfile_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                int op_errno, inode_t *inode, struct iatt *stbuf,
+                struct iatt *preparent, struct iatt *postparent, dict_t *xdata);
+
+int
+dht_finodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata);
+
+int
+dht_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, dict_t *xattr, dict_t *xdata);
+
+int
+dht_common_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *dict,
+                       dict_t *xdata);
+int
+gf_defrag_status_get(dht_conf_t *conf, dict_t *dict);
+
+int
+gf_defrag_stop(dht_conf_t *conf, gf_defrag_status_t status, dict_t *output);
+
+void *
+gf_defrag_start(void *this);
+
+int32_t
+gf_defrag_handle_hardlink(xlator_t *this, loc_t *loc, int *fop_errno);
+int
+dht_migrate_file(xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
+                 int flag, int *fop_errno);
+int
+dht_inode_ctx_layout_get(inode_t *inode, xlator_t *this,
+                         dht_layout_t **layout_int);
+int
+dht_inode_ctx_layout_set(inode_t *inode, xlator_t *this,
+                         dht_layout_t *layout_int);
+int
+dht_inode_ctx_time_update(inode_t *inode, xlator_t *this, struct iatt *stat,
+                          int32_t update_ctx);
+void
+dht_inode_ctx_time_set(inode_t *inode, xlator_t *this, struct iatt *stat);
+
+int
+dht_inode_ctx_get(inode_t *inode, xlator_t *this, dht_inode_ctx_t **ctx);
+int
+dht_inode_ctx_set(inode_t *inode, xlator_t *this, dht_inode_ctx_t *ctx);
+int
+dht_dir_attr_heal(void *data);
+int
+dht_dir_attr_heal_done(int ret, call_frame_t *sync_frame, void *data);
+xlator_t *
+dht_subvol_with_free_space_inodes(xlator_t *this, xlator_t *subvol,
+                                  xlator_t *ignore, dht_layout_t *layout,
+                                  uint64_t filesize);
+xlator_t *
+dht_subvol_maxspace_nonzeroinode(xlator_t *this, xlator_t *subvol,
+                                 dht_layout_t *layout);
+int
+dht_dir_has_layout(dict_t *xattr, char *name);
+int
+dht_linkfile_attr_heal(call_frame_t *frame, xlator_t *this);
+
+int32_t
+dht_priv_dump(xlator_t *this);
+int32_t
+dht_inodectx_dump(xlator_t *this, inode_t *inode);
+
+gf_boolean_t
+dht_is_subvol_in_layout(dht_layout_t *layout, xlator_t *xlator);
+
+int
+dht_inode_ctx_get_mig_info(xlator_t *this, inode_t *inode,
+                           xlator_t **src_subvol, xlator_t **dst_subvol);
+gf_boolean_t
+dht_mig_info_is_invalid(xlator_t *current, xlator_t *src_subvol,
+                        xlator_t *dst_subvol);
+
+int
+dht_subvol_status(dht_conf_t *conf, xlator_t *subvol);
+
+void
+dht_log_new_layout_for_dir_selfheal(xlator_t *this, loc_t *loc,
+                                    dht_layout_t *layout);
+
+int
+dht_layout_sort(dht_layout_t *layout);
+
+int
+dht_heal_full_path(void *data);
+
+int
+dht_heal_full_path_done(int op_ret, call_frame_t *frame, void *data);
+
+int
+dht_layout_missing_dirs(dht_layout_t *layout);
+
+int
+dht_refresh_layout(call_frame_t *frame);
+
+int
+dht_build_parent_loc(xlator_t *this, loc_t *parent, loc_t *child,
+                     int32_t *op_errno);
+
+int32_t
+dht_set_local_rebalance(xlator_t *this, dht_local_t *local, struct iatt *stbuf,
+                        struct iatt *prebuf, struct iatt *postbuf,
+                        dict_t *xdata);
+void
+dht_build_root_loc(inode_t *inode, loc_t *loc);
+
+gf_boolean_t
+dht_fd_open_on_dst(xlator_t *this, fd_t *fd, xlator_t *dst);
+
+int32_t
+dht_fd_ctx_destroy(xlator_t *this, fd_t *fd);
+
+int32_t
+dht_release(xlator_t *this, fd_t *fd);
+
+int32_t
+dht_set_fixed_dir_stat(struct iatt *stat);
+
+xlator_t *
+dht_get_lock_subvolume(xlator_t *this, struct gf_flock *lock,
+                       dht_local_t *local);
+
+int
+dht_lk_inode_unref(call_frame_t *frame, int32_t op_ret);
+
+int
+dht_fd_ctx_set(xlator_t *this, fd_t *fd, xlator_t *subvol);
+
+int
+dht_check_and_open_fd_on_subvol(xlator_t *this, call_frame_t *frame);
+
+/* FD fop callbacks */
+
+int
+dht_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+               dict_t *xdata);
+
+int
+dht_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, dict_t *xdata);
+
+int
+dht_file_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata);
+
+int
+dht_zerofill_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+                 dict_t *xdata);
+
+int
+dht_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+                dict_t *xdata);
+
+int
+dht_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                  int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+                  dict_t *xdata);
+
+int
+dht_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+                 dict_t *xdata);
+
+int
+dht_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+              dict_t *xdata);
+
+int
+dht_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, struct iovec *vector, int count, struct iatt *stbuf,
+              struct iobref *iobref, dict_t *xdata);
+
+int
+dht_file_attr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                  int op_errno, struct iatt *stbuf, dict_t *xdata);
+
+int
+dht_file_removexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int op_ret, int op_errno, dict_t *xdata);
+
+int
+dht_file_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, dict_t *xdata);
+
+/* All custom xattr heal functions */
+int
+dht_dir_heal_xattrs(void *data);
+
+int
+dht_dir_heal_xattrs_done(int ret, call_frame_t *sync_frame, void *data);
+
+int32_t
+dht_dict_set_array(dict_t *dict, char *key, int32_t value[], int32_t size);
+
+int
+dht_set_user_xattr(dict_t *dict, char *k, data_t *v, void *data);
+
+void
+dht_dir_set_heal_xattr(xlator_t *this, dht_local_t *local, dict_t *dst,
+                       dict_t *src, int *uret, int *uflag);
+
+int
+dht_dir_xattr_heal(xlator_t *this, dht_local_t *local, int *op_errno);
+
+int
+dht_common_mark_mdsxattr(call_frame_t *frame, int *errst, int flag);
+
+int
+dht_inode_ctx_mdsvol_get(inode_t *inode, xlator_t *this, xlator_t **mdsvol);
+
+int
+dht_selfheal_dir_setattr(call_frame_t *frame, loc_t *loc, struct iatt *stbuf,
+                         int32_t valid, dht_layout_t *layout);
+
+/* Abstract out the DHT-IATT-IN-DICT */
+
+void
+dht_selfheal_layout_new_directory(call_frame_t *frame, loc_t *loc,
+                                  dht_layout_t *new_layout);
+
+int
+dht_pt_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *key,
+                 dict_t *xdata);
+
+int
+dht_pt_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                const char *key, dict_t *xdata);
+
+int32_t
+dht_pt_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+             mode_t umask, dict_t *xdata);
+
+int
+dht_pt_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+              dict_t *xdata);
+
+int32_t
+dht_check_remote_fd_failed_error(dht_local_t *local, int op_ret, int op_errno);
+
+int
+dht_common_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *dict,
+                       dict_t *xdata);
+
+int32_t
+dht_create_lock(call_frame_t *frame, xlator_t *subvol);
+
+int
+dht_set_parent_layout_in_dict(loc_t *loc, xlator_t *this, dht_local_t *local);
+
+int
+dht_dir_layout_error_check(xlator_t *this, inode_t *inode);
+
+int
+dht_inode_ctx_mdsvol_set(inode_t *inode, xlator_t *this, xlator_t *mds_subvol);
 #endif /* _DHT_H */
diff --git a/xlators/cluster/dht/src/dht-diskusage.c b/xlators/cluster/dht/src/dht-diskusage.c
index fbc1085755e..c0588828fdb 100644
--- a/xlators/cluster/dht/src/dht-diskusage.c
+++ b/xlators/cluster/dht/src/dht-diskusage.c
@@ -1,263 +1,487 @@
 /*
-   Copyright (c) 2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
 
 /* TODO: add NS locking */
 
-#include "glusterfs.h"
-#include "xlator.h"
 #include "dht-common.h"
-#include "defaults.h"
 
 #include <sys/time.h>
+#include <glusterfs/events.h>
 
-
-int 
-dht_du_info_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int op_ret, int op_errno, struct statvfs *statvfs)
+int
+dht_du_info_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                int op_errno, struct statvfs *statvfs, dict_t *xdata)
 {
-	dht_conf_t    *conf         = NULL;
-        call_frame_t  *prev          = NULL;
-	int            this_call_cnt = 0;
-        int            i = 0;
-        double         percent = 0;
-        uint64_t       bytes = 0;
-
-        conf = this->private;
-        prev = cookie;
-
-        if (op_ret == -1) 
-                goto out;
-
-        if (statvfs && statvfs->f_blocks) {
-                percent = (statvfs->f_bfree * 100) / statvfs->f_blocks;
-                bytes = (statvfs->f_bfree * statvfs->f_bsize);
-        }
-        
-        LOCK (&conf->subvolume_lock);
-        {
-                for (i = 0; i < conf->subvolume_cnt; i++)
-                        if (prev->this == conf->subvolumes[i]) {
-                                conf->du_stats[i].avail_percent = percent;
-                                conf->du_stats[i].avail_space   = bytes;
-                                gf_log (this->name, GF_LOG_DEBUG,
-                                        "on subvolume '%s': avail_percent is: "
-                                        "%.2f and avail_space is: %"PRIu64"",
-                                        prev->this->name, 
-                                        conf->du_stats[i].avail_percent,
-                                        conf->du_stats[i].avail_space);
-                        }
-        }
-        UNLOCK (&conf->subvolume_lock);
-
- out:
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt))
-		DHT_STACK_DESTROY (frame);
-
-        return 0;
+    dht_conf_t *conf = NULL;
+    xlator_t *prev = NULL;
+    int this_call_cnt = 0;
+    int i = 0;
+    double percent = 0;
+    double percent_inodes = 0;
+    uint64_t bytes = 0;
+    uint32_t bpc; /* blocks per chunk */
+    uint32_t chunks = 0;
+
+    conf = this->private;
+    prev = cookie;
+
+    if (op_ret == -1 || !statvfs) {
+        gf_msg(this->name, GF_LOG_WARNING, op_errno,
+               DHT_MSG_GET_DISK_INFO_ERROR, "failed to get disk info from %s",
+               prev->name);
+        goto out;
+    }
+
+    if (statvfs->f_blocks) {
+        percent = (statvfs->f_bavail * 100) / statvfs->f_blocks;
+        bytes = (statvfs->f_bavail * statvfs->f_frsize);
+        /*
+         * A 32-bit count of 1MB chunks allows a maximum brick size of
+         * ~4PB.  It's possible that we could see a single local FS
+         * bigger than that some day, but this code is likely to be
+         * irrelevant by then.  Meanwhile, it's more important to keep
+         * the chunk size small so the layout-calculation code that
+         * uses this value can be tested on normal machines.
+         */
+        bpc = (1 << 20) / statvfs->f_bsize;
+        chunks = (statvfs->f_blocks + bpc - 1) / bpc;
+    }
+
+    if (statvfs->f_files) {
+        percent_inodes = (statvfs->f_ffree * 100) / statvfs->f_files;
+    } else {
+        /*
+         * Set percent inodes to 100 for dynamically allocated inode
+         * filesystems. The rationale is that distribute need not
+         * worry about total inodes; rather, let the 'create()' be
+         * scheduled on the hashed subvol regardless of the total
+         * inodes.
+         */
+        percent_inodes = 100;
+    }
+
+    LOCK(&conf->subvolume_lock);
+    {
+        for (i = 0; i < conf->subvolume_cnt; i++)
+            if (prev == conf->subvolumes[i]) {
+                conf->du_stats[i].avail_percent = percent;
+                conf->du_stats[i].avail_space = bytes;
+                conf->du_stats[i].avail_inodes = percent_inodes;
+                conf->du_stats[i].chunks = chunks;
+                conf->du_stats[i].total_blocks = statvfs->f_blocks;
+                conf->du_stats[i].avail_blocks = statvfs->f_bavail;
+                conf->du_stats[i].frsize = statvfs->f_frsize;
+
+                gf_msg_debug(this->name, 0,
+                             "subvolume '%s': avail_percent "
+                             "is: %.2f and avail_space "
+                             "is: %" PRIu64
+                             " and avail_inodes"
+                             " is: %.2f",
+                             prev->name, conf->du_stats[i].avail_percent,
+                             conf->du_stats[i].avail_space,
+                             conf->du_stats[i].avail_inodes);
+                break; /* no point in looping further */
+            }
+    }
+    UNLOCK(&conf->subvolume_lock);
+
+out:
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt))
+        DHT_STACK_DESTROY(frame);
+
+    return 0;
 }
 
 int
-dht_get_du_info_for_subvol (xlator_t *this, int subvol_idx)
+dht_get_du_info_for_subvol(xlator_t *this, int subvol_idx)
 {
-	dht_conf_t    *conf         = NULL;
-	call_frame_t  *statfs_frame = NULL;
-	dht_local_t   *statfs_local = NULL;
-        call_pool_t   *pool         = NULL;
+    dht_conf_t *conf = NULL;
+    call_frame_t *statfs_frame = NULL;
+    dht_local_t *statfs_local = NULL;
+    call_pool_t *pool = NULL;
+    loc_t tmp_loc = {
+        0,
+    };
+
+    conf = this->private;
+    pool = this->ctx->pool;
+
+    statfs_frame = create_frame(this, pool);
+    if (!statfs_frame) {
+        goto err;
+    }
+
+    /* local->fop value is not used in this case */
+    statfs_local = dht_local_init(statfs_frame, NULL, NULL, GF_FOP_MAXVALUE);
+    if (!statfs_local) {
+        goto err;
+    }
+
+    /* make it root gfid, should be enough to get the proper info back */
+    tmp_loc.gfid[15] = 1;
+
+    statfs_local->call_cnt = 1;
+    STACK_WIND_COOKIE(
+        statfs_frame, dht_du_info_cbk, conf->subvolumes[subvol_idx],
+        conf->subvolumes[subvol_idx],
+        conf->subvolumes[subvol_idx]->fops->statfs, &tmp_loc, NULL);
+
+    return 0;
+err:
+    if (statfs_frame)
+        DHT_STACK_DESTROY(statfs_frame);
 
-	conf = this->private;
-        pool = this->ctx->pool;
+    return -1;
+}
 
-        statfs_frame = create_frame (this, pool);
+int
+dht_get_du_info(call_frame_t *frame, xlator_t *this, loc_t *loc)
+{
+    int i = 0;
+    int ret = -1;
+    dht_conf_t *conf = NULL;
+    call_frame_t *statfs_frame = NULL;
+    dht_local_t *statfs_local = NULL;
+    loc_t tmp_loc = {
+        0,
+    };
+    time_t now;
+
+    conf = this->private;
+    now = gf_time();
+    /* make it root gfid, should be enough to get the proper
+       info back */
+    tmp_loc.gfid[15] = 1;
+
+    if (now > (conf->refresh_interval + conf->last_stat_fetch)) {
+        statfs_frame = copy_frame(frame);
         if (!statfs_frame) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                goto err;
+            goto err;
         }
-        
-        statfs_local = dht_local_init (statfs_frame);
+
+        /* In this case, 'local->fop' is not used */
+        statfs_local = dht_local_init(statfs_frame, loc, NULL, GF_FOP_MAXVALUE);
         if (!statfs_local) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                goto err;
+            goto err;
+        }
+
+        statfs_local->params = dict_new();
+        if (!statfs_local->params)
+            goto err;
+
+        ret = dict_set_int8(statfs_local->params,
+                            GF_INTERNAL_IGNORE_DEEM_STATFS, 1);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                   "Failed to set " GF_INTERNAL_IGNORE_DEEM_STATFS " in dict");
+            goto err;
         }
-        
-        loc_t tmp_loc = { .inode = NULL,
-                          .path = "/",
-        };
-        
-        statfs_local->call_cnt = 1;
-        STACK_WIND (statfs_frame, dht_du_info_cbk,
-                    conf->subvolumes[subvol_idx],
-                    conf->subvolumes[subvol_idx]->fops->statfs,
-                    &tmp_loc);
-
-        return 0;
- err:
-	if (statfs_frame)
-		DHT_STACK_DESTROY (statfs_frame);
-        
-        return -1;
+
+        statfs_local->call_cnt = conf->subvolume_cnt;
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            STACK_WIND_COOKIE(statfs_frame, dht_du_info_cbk,
+                              conf->subvolumes[i], conf->subvolumes[i],
+                              conf->subvolumes[i]->fops->statfs, &tmp_loc,
+                              statfs_local->params);
+        }
+
+        conf->last_stat_fetch = now;
+    }
+    return 0;
+err:
+    if (statfs_frame)
+        DHT_STACK_DESTROY(statfs_frame);
+
+    return -1;
 }
 
-int
-dht_get_du_info (call_frame_t *frame, xlator_t *this, loc_t *loc)
+gf_boolean_t
+dht_is_subvol_filled(xlator_t *this, xlator_t *subvol)
 {
-        int            i = 0;
-	dht_conf_t    *conf         = NULL;
-	call_frame_t  *statfs_frame = NULL;
-	dht_local_t   *statfs_local = NULL;
-        struct timeval tv = {0,};
-
-	conf  = this->private;
-
-	gettimeofday (&tv, NULL);
-	if (tv.tv_sec > (conf->refresh_interval 
-			 + conf->last_stat_fetch.tv_sec)) {
-
-                statfs_frame = copy_frame (frame);
-                if (!statfs_frame) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Out of memory");
-                        goto err;
+    int i = 0;
+    char vol_name[256];
+    dht_conf_t *conf = NULL;
+    gf_boolean_t subvol_filled_inodes = _gf_false;
+    gf_boolean_t subvol_filled_space = _gf_false;
+    gf_boolean_t is_subvol_filled = _gf_false;
+    double usage = 0;
+
+    conf = this->private;
+
+    /* Check for values above specified percent or free disk */
+    LOCK(&conf->subvolume_lock);
+    {
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (subvol == conf->subvolumes[i]) {
+                if (conf->disk_unit == 'p') {
+                    if (conf->du_stats[i].avail_percent < conf->min_free_disk) {
+                        subvol_filled_space = _gf_true;
+                        break;
+                    }
+
+                } else {
+                    if (conf->du_stats[i].avail_space < conf->min_free_disk) {
+                        subvol_filled_space = _gf_true;
+                        break;
+                    }
                 }
-
-                statfs_local = dht_local_init (statfs_frame);
-                if (!statfs_local) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Out of memory");
-                        goto err;
+                if (conf->du_stats[i].avail_inodes < conf->min_free_inodes) {
+                    subvol_filled_inodes = _gf_true;
+                    break;
                 }
+            }
+        }
+    }
+    UNLOCK(&conf->subvolume_lock);
 
-                loc_copy (&statfs_local->loc, loc);
-                loc_t tmp_loc = { .inode = NULL,
-                              .path = "/",
-                };
-                
-                statfs_local->call_cnt = conf->subvolume_cnt;
-                for (i = 0; i < conf->subvolume_cnt; i++) {
-                        STACK_WIND (statfs_frame, dht_du_info_cbk,
-                                    conf->subvolumes[i],
-                                    conf->subvolumes[i]->fops->statfs,
-                                    &tmp_loc);
-                }
+    if (subvol_filled_space && conf->subvolume_status[i]) {
+        if (!(conf->du_stats[i].log++ % (GF_UNIVERSAL_ANSWER * 10))) {
+            usage = 100 - conf->du_stats[i].avail_percent;
+
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_SUBVOL_INSUFF_SPACE,
+                   "disk space on subvolume '%s' is getting "
+                   "full (%.2f %%), consider adding more bricks",
+                   subvol->name, usage);
 
-                conf->last_stat_fetch.tv_sec = tv.tv_sec;
+            (void)snprintf(vol_name, sizeof(vol_name), "%s", this->name);
+            vol_name[(strlen(this->name) - 4)] = '\0';
+
+            gf_event(EVENT_DHT_DISK_USAGE, "volume=%s;subvol=%s;usage=%.2f %%",
+                     vol_name, subvol->name, usage);
         }
-        return 0;
-err:
-	if (statfs_frame)
-		DHT_STACK_DESTROY (statfs_frame);
+    }
+
+    if (subvol_filled_inodes && conf->subvolume_status[i]) {
+        if (!(conf->du_stats[i].log++ % (GF_UNIVERSAL_ANSWER * 10))) {
+            usage = 100 - conf->du_stats[i].avail_inodes;
+            gf_msg(this->name, GF_LOG_CRITICAL, 0, DHT_MSG_SUBVOL_INSUFF_INODES,
+                   "inodes on subvolume '%s' are at "
+                   "(%.2f %%), consider adding more bricks",
+                   subvol->name, usage);
+
+            (void)snprintf(vol_name, sizeof(vol_name), "%s", this->name);
+            vol_name[(strlen(this->name) - 4)] = '\0';
+
+            gf_event(EVENT_DHT_INODES_USAGE,
+                     "volume=%s;subvol=%s;usage=%.2f %%", vol_name,
+                     subvol->name, usage);
+        }
+    }
 
-        return -1;
+    is_subvol_filled = (subvol_filled_space || subvol_filled_inodes);
+
+    return is_subvol_filled;
 }
 
+/*Get the best subvolume to create the file in*/
+xlator_t *
+dht_free_disk_available_subvol(xlator_t *this, xlator_t *subvol,
+                               dht_local_t *local)
+{
+    xlator_t *avail_subvol = NULL;
+    dht_conf_t *conf = NULL;
+    dht_layout_t *layout = NULL;
+    loc_t *loc = NULL;
+
+    conf = this->private;
+    if (!local)
+        goto out;
+    loc = &local->loc;
+    if (!local->layout) {
+        layout = dht_layout_get(this, loc->parent);
+
+        if (!layout) {
+            gf_msg_debug(this->name, 0,
+                         "Missing layout. path=%s,"
+                         " parent gfid = %s",
+                         loc->path, uuid_utoa(loc->parent->gfid));
+            goto out;
+        }
+    } else {
+        layout = dht_layout_ref(this, local->layout);
+    }
+
+    LOCK(&conf->subvolume_lock);
+    {
+        avail_subvol = dht_subvol_with_free_space_inodes(this, subvol, NULL,
+                                                         layout, 0);
+        if (!avail_subvol) {
+            avail_subvol = dht_subvol_maxspace_nonzeroinode(this, subvol,
+                                                            layout);
+        }
+    }
+    UNLOCK(&conf->subvolume_lock);
+out:
+    if (!avail_subvol) {
+        gf_msg_debug(this->name, 0,
+                     "No subvolume has enough free space \
+                              and/or inodes to create");
+        avail_subvol = subvol;
+    }
+
+    if (layout)
+        dht_layout_unref(this, layout);
+    return avail_subvol;
+}
 
-int
-dht_is_subvol_filled (xlator_t *this, xlator_t *subvol)
+static inline int32_t
+dht_subvol_has_err(dht_conf_t *conf, xlator_t *this, xlator_t *ignore,
+                   dht_layout_t *layout)
 {
-        int         i = 0;
-        int         subvol_filled = 0;
-	dht_conf_t *conf = NULL;
-
-        conf = this->private;
-
-        /* Check for values above specified percent or free disk */
-        LOCK (&conf->subvolume_lock);
-        {
-                for (i = 0; i < conf->subvolume_cnt; i++) {
-                        if (subvol == conf->subvolumes[i]) {
-                                if (conf->disk_unit == 'p') {
-                                        if (conf->du_stats[i].avail_percent <
-                                            conf->min_free_disk) {
-                                                subvol_filled = 1;
-                                                break;
-                                        }
-                                } else {
-                                        if (conf->du_stats[i].avail_space <
-                                            conf->min_free_disk) {
-                                                subvol_filled = 1;
-                                                break;
-                                        }
-                                }
-                        }
-                }
+    int ret = -1;
+    int i = 0;
+
+    if (!this || !layout)
+        goto out;
+
+    /* this check is meant for rebalance process. The source of the file
+     * should be ignored for space check */
+    if (this == ignore) {
+        goto out;
+    }
+
+    /* check if subvol has layout errors, before selecting it */
+    for (i = 0; i < layout->cnt; i++) {
+        if (!strcmp(layout->list[i].xlator->name, this->name) &&
+            (layout->list[i].err != 0)) {
+            ret = -1;
+            goto out;
         }
-        UNLOCK (&conf->subvolume_lock);
-
-        if (subvol_filled && conf->subvolume_status[i]) {
-                if (!(conf->du_stats[i].log++ % (GF_UNIVERSAL_ANSWER * 10))) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "disk space on subvolume '%s' is getting "
-                                "full (%.2f %%), consider adding more nodes", 
-                                subvol->name, 
-                                (100 - conf->du_stats[i].avail_percent));
-                }
+    }
+
+    /* discard decommissioned subvol */
+    if (conf->decommission_subvols_cnt) {
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (conf->decommissioned_bricks[i] &&
+                conf->decommissioned_bricks[i] == this) {
+                ret = -1;
+                goto out;
+            }
         }
+    }
 
-        return subvol_filled;
+    ret = 0;
+out:
+    return ret;
 }
 
+/*Get subvolume which has both space and inodes more than the min criteria*/
 xlator_t *
-dht_free_disk_available_subvol (xlator_t *this, xlator_t *subvol) 
+dht_subvol_with_free_space_inodes(xlator_t *this, xlator_t *subvol,
+                                  xlator_t *ignore, dht_layout_t *layout,
+                                  uint64_t filesize)
 {
-        int         i = 0;
-        double      max= 0;
-        xlator_t   *avail_subvol = NULL;
-	dht_conf_t *conf = NULL;
-
-        conf = this->private;
-
-        LOCK (&conf->subvolume_lock);
-        {
-                for (i = 0; i < conf->subvolume_cnt; i++) {
-                        if (conf->disk_unit == 'p') {
-                                if (conf->du_stats[i].avail_percent > max) {
-                                        max = conf->du_stats[i].avail_percent;
-                                        avail_subvol = conf->subvolumes[i];
-                                }
-                        } else {
-                                if (conf->du_stats[i].avail_space > max) {
-                                        max = conf->du_stats[i].avail_space;
-                                        avail_subvol = conf->subvolumes[i];
-                                }
-                        }
-                }
+    int i = 0;
+    double max = 0;
+    double max_inodes = 0;
+    int ignore_subvol = 0;
+    uint64_t total_blocks = 0;
+    uint64_t avail_blocks = 0;
+    uint64_t frsize = 0;
+    double post_availspace = 0;
+    double post_percent = 0;
+
+    xlator_t *avail_subvol = NULL;
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        /* check if subvol has layout errors and also it is not a
+         * decommissioned brick, before selecting it */
+        ignore_subvol = dht_subvol_has_err(conf, conf->subvolumes[i], ignore,
+                                           layout);
+        if (ignore_subvol)
+            continue;
+
+        if ((conf->disk_unit == 'p') &&
+            (conf->du_stats[i].avail_percent > conf->min_free_disk) &&
+            (conf->du_stats[i].avail_inodes > conf->min_free_inodes)) {
+            if ((conf->du_stats[i].avail_inodes > max_inodes) ||
+                (conf->du_stats[i].avail_percent > max)) {
+                max = conf->du_stats[i].avail_percent;
+                max_inodes = conf->du_stats[i].avail_inodes;
+                avail_subvol = conf->subvolumes[i];
+                total_blocks = conf->du_stats[i].total_blocks;
+                avail_blocks = conf->du_stats[i].avail_blocks;
+                frsize = conf->du_stats[i].frsize;
+            }
         }
-        UNLOCK (&conf->subvolume_lock);
 
-        if (!avail_subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no subvolume has enough free space to create");
+        if ((conf->disk_unit != 'p') &&
+            (conf->du_stats[i].avail_space > conf->min_free_disk) &&
+            (conf->du_stats[i].avail_inodes > conf->min_free_inodes)) {
+            if ((conf->du_stats[i].avail_inodes > max_inodes) ||
+                (conf->du_stats[i].avail_space > max)) {
+                max = conf->du_stats[i].avail_space;
+                max_inodes = conf->du_stats[i].avail_inodes;
+                avail_subvol = conf->subvolumes[i];
+            }
         }
+    }
+
+    if (avail_subvol) {
+        if (conf->disk_unit == 'p') {
+            post_availspace = (avail_blocks * frsize) - filesize;
+            post_percent = (post_availspace * 100) / (total_blocks * frsize);
+            if (post_percent < conf->min_free_disk)
+                avail_subvol = NULL;
+        }
+        if (conf->disk_unit != 'p') {
+            if ((max - filesize) < conf->min_free_disk)
+                avail_subvol = NULL;
+        }
+    }
 
-        if (max < conf->min_free_disk)
-                avail_subvol = subvol;
+    return avail_subvol;
+}
 
-        if (!avail_subvol)
-                avail_subvol = subvol;                
+/* Get subvol which has at least one inode and maximum space */
+xlator_t *
+dht_subvol_maxspace_nonzeroinode(xlator_t *this, xlator_t *subvol,
+                                 dht_layout_t *layout)
+{
+    int i = 0;
+    double max = 0;
+    int ignore_subvol = 0;
+
+    xlator_t *avail_subvol = NULL;
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        /* check if subvol has layout errors and also it is not a
+         * decommissioned brick, before selecting it*/
+
+        ignore_subvol = dht_subvol_has_err(conf, conf->subvolumes[i], NULL,
+                                           layout);
+        if (ignore_subvol)
+            continue;
+
+        if (conf->disk_unit == 'p') {
+            if ((conf->du_stats[i].avail_percent > max) &&
+                (conf->du_stats[i].avail_inodes > 0)) {
+                max = conf->du_stats[i].avail_percent;
+                avail_subvol = conf->subvolumes[i];
+            }
+        } else {
+            if ((conf->du_stats[i].avail_space > max) &&
+                (conf->du_stats[i].avail_inodes > 0)) {
+                max = conf->du_stats[i].avail_space;
+                avail_subvol = conf->subvolumes[i];
+            }
+        }
+    }
 
-        return avail_subvol;
+    return avail_subvol;
 }
diff --git a/xlators/cluster/dht/src/dht-hashfn.c b/xlators/cluster/dht/src/dht-hashfn.c
index 09394a81428..acda67c312a 100644
--- a/xlators/cluster/dht/src/dht-hashfn.c
+++ b/xlators/cluster/dht/src/dht-hashfn.c
@@ -1,81 +1,110 @@
 /*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
 
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
 
-#include "glusterfs.h"
-#include "xlator.h"
 #include "dht-common.h"
-#include "hashfn.h"
+#include <glusterfs/hashfn.h>
 
-
-int
-dht_hash_compute_internal (int type, const char *name, uint32_t *hash_p)
+static int
+dht_hash_compute_internal(int type, const char *name, const int len,
+                          uint32_t *hash_p)
 {
-	int      ret = 0;
-	uint32_t hash = 0;
-
-	switch (type) {
-	case DHT_HASH_TYPE_DM:
-		hash = gf_dm_hashfn (name, strlen (name));
-		break;
-	default:
-		ret = -1;
-		break;
-	}
-
-	if (ret == 0) {
-		*hash_p = hash;
-	}
-
-	return ret;
+    int ret = 0;
+    uint32_t hash = 0;
+
+    switch (type) {
+        case DHT_HASH_TYPE_DM:
+        case DHT_HASH_TYPE_DM_USER:
+            hash = gf_dm_hashfn(name, len);
+            break;
+        default:
+            ret = -1;
+            break;
+    }
+
+    if (ret == 0) {
+        *hash_p = hash;
+    }
+
+    return ret;
 }
 
-
-#define MAKE_RSYNC_FRIENDLY_NAME(rsync_frndly_name, name) do {          \
-                rsync_frndly_name = (char *) name;			\
-                if (name[0] == '.') {                                   \
-                        char *dot   = 0;                                \
-                        int namelen = 0;                                \
-                                                                        \
-                        dot = strrchr (name, '.');                      \
-                        if (dot && dot > (name + 1) && *(dot + 1)) {    \
-                                namelen = (dot - name);                 \
-                                rsync_frndly_name = alloca (namelen);   \
-                                strncpy (rsync_frndly_name, name + 1,   \
-                                         namelen);                      \
-                                rsync_frndly_name[namelen - 1] = 0;     \
-                        }                                               \
-                }                                                       \
-        } while (0);
-
+/* The function returns:
+ * 0  : in case no munge took place
+ * >0 : the length (inc. terminating NULL!) of the newly modified string,
+ *      if it was munged.
+ */
+static int
+dht_munge_name(const char *original, char *modified, size_t len, regex_t *re)
+{
+    regmatch_t matches[2] = {
+        {0},
+    };
+    size_t new_len = 0;
+    int ret = 0;
+
+    ret = regexec(re, original, 2, matches, 0);
+
+    if (ret != REG_NOMATCH) {
+        if (matches[1].rm_so != -1) {
+            new_len = matches[1].rm_eo - matches[1].rm_so;
+            /* Equal would fail due to the NUL at the end. */
+            if (new_len < len) {
+                memcpy(modified, original + matches[1].rm_so, new_len);
+                modified[new_len] = '\0';
+                return new_len + 1; /* +1 for the terminating NULL */
+            }
+        }
+    }
+
+    /* This is guaranteed safe because of how the dest was allocated. */
+    strcpy(modified, original);
+    return 0;
+}
 
 int
-dht_hash_compute (int type, const char *name, uint32_t *hash_p)
+dht_hash_compute(xlator_t *this, int type, const char *name, uint32_t *hash_p)
 {
-	char     *rsync_friendly_name = NULL;
-
-	MAKE_RSYNC_FRIENDLY_NAME (rsync_friendly_name, name);
-
-	return dht_hash_compute_internal (type, rsync_friendly_name, hash_p);
+    char *rsync_friendly_name = NULL;
+    dht_conf_t *priv = NULL;
+    size_t len = 0;
+    int munged = 0;
+
+    priv = this->private;
+
+    if (name == NULL)
+        return -1;
+
+    len = strlen(name) + 1;
+    rsync_friendly_name = alloca(len);
+
+    LOCK(&priv->lock);
+    {
+        if (priv->extra_regex_valid) {
+            munged = dht_munge_name(name, rsync_friendly_name, len,
+                                    &priv->extra_regex);
+        }
+
+        if (!munged && priv->rsync_regex_valid) {
+            gf_msg_trace(this->name, 0, "trying regex for %s", name);
+            munged = dht_munge_name(name, rsync_friendly_name, len,
+                                    &priv->rsync_regex);
+        }
+    }
+    UNLOCK(&priv->lock);
+    if (munged) {
+        gf_msg_debug(this->name, 0, "munged down to %s", rsync_friendly_name);
+        len = munged;
+    } else {
+        rsync_friendly_name = (char *)name;
+    }
+
+    return dht_hash_compute_internal(type, rsync_friendly_name, len - 1,
+                                     hash_p);
 }
diff --git a/xlators/cluster/dht/src/dht-helper.c b/xlators/cluster/dht/src/dht-helper.c
index 9e21713f7a6..3f2fe43d5f3 100644
--- a/xlators/cluster/dht/src/dht-helper.c
+++ b/xlators/cluster/dht/src/dht-helper.c
@@ -1,525 +1,2304 @@
 /*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+#include "dht-common.h"
+#include "dht-lock.h"
+#include "glusterfs/compat-errno.h"  // for ENODATA on BSD
 
+static void
+dht_free_fd_ctx(dht_fd_ctx_t *fd_ctx)
+{
+    GF_FREE(fd_ctx);
+}
 
-#include "glusterfs.h"
-#include "xlator.h"
-#include "dht-common.h"
+int32_t
+dht_fd_ctx_destroy(xlator_t *this, fd_t *fd)
+{
+    dht_fd_ctx_t *fd_ctx = NULL;
+    uint64_t value = 0;
+    int32_t ret = -1;
+
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    ret = fd_ctx_del(fd, this, &value);
+    if (ret) {
+        goto out;
+    }
+
+    fd_ctx = (dht_fd_ctx_t *)(uintptr_t)value;
+    if (fd_ctx) {
+        GF_REF_PUT(fd_ctx);
+    }
+out:
+    return ret;
+}
+
+static int
+__dht_fd_ctx_set(xlator_t *this, fd_t *fd, xlator_t *dst)
+{
+    dht_fd_ctx_t *fd_ctx = NULL;
+    uint64_t value = 0;
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
 
+    fd_ctx = GF_CALLOC(1, sizeof(*fd_ctx), gf_dht_mt_fd_ctx_t);
+
+    if (!fd_ctx) {
+        goto out;
+    }
+
+    fd_ctx->opened_on_dst = (uint64_t)(uintptr_t)dst;
+    GF_REF_INIT(fd_ctx, dht_free_fd_ctx);
+
+    value = (uint64_t)(uintptr_t)fd_ctx;
+
+    ret = __fd_ctx_set(fd, this, value);
+    if (ret < 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_FD_CTX_SET_FAILED,
+                "fd=0x%p", fd, NULL);
+        GF_REF_PUT(fd_ctx);
+    }
+out:
+    return ret;
+}
 
 int
-dht_frame_return (call_frame_t *frame)
+dht_fd_ctx_set(xlator_t *this, fd_t *fd, xlator_t *dst)
 {
-	dht_local_t *local = NULL;
-	int          this_call_cnt = -1;
+    dht_fd_ctx_t *fd_ctx = NULL;
+    uint64_t value = 0;
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    LOCK(&fd->lock);
+    {
+        ret = __fd_ctx_get(fd, this, &value);
+        if (ret && value) {
+            fd_ctx = (dht_fd_ctx_t *)(uintptr_t)value;
+            if (fd_ctx->opened_on_dst == (uint64_t)(uintptr_t)dst) {
+                /* This could happen due to racing
+                 * check_progress tasks*/
+                goto unlock;
+            } else {
+                /* This would be a big problem*/
+                /* Overwrite and hope for the best*/
+                fd_ctx->opened_on_dst = (uint64_t)(uintptr_t)dst;
+                UNLOCK(&fd->lock);
+                gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_INVALID_VALUE,
+                        NULL);
 
-	if (!frame)
-		return -1;
+                goto out;
+            }
+        }
+        ret = __dht_fd_ctx_set(this, fd, dst);
+    }
+unlock:
+    UNLOCK(&fd->lock);
+out:
+    return ret;
+}
 
-	local = frame->local;
+static dht_fd_ctx_t *
+dht_fd_ctx_get(xlator_t *this, fd_t *fd)
+{
+    dht_fd_ctx_t *fd_ctx = NULL;
+    int ret = -1;
+    uint64_t tmp_val = 0;
+
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    LOCK(&fd->lock);
+    {
+        ret = __fd_ctx_get(fd, this, &tmp_val);
+        if ((ret < 0) || (tmp_val == 0)) {
+            goto unlock;
+        }
 
-	LOCK (&frame->lock);
-	{
-		this_call_cnt = --local->call_cnt;
-	}
-	UNLOCK (&frame->lock);
+        fd_ctx = (dht_fd_ctx_t *)(uintptr_t)tmp_val;
+        GF_REF_GET(fd_ctx);
+    }
+unlock:
+    UNLOCK(&fd->lock);
 
-	return this_call_cnt;
+out:
+    return fd_ctx;
 }
 
+gf_boolean_t
+dht_fd_open_on_dst(xlator_t *this, fd_t *fd, xlator_t *dst)
+{
+    dht_fd_ctx_t *fd_ctx = NULL;
+    gf_boolean_t opened = _gf_false;
 
-int
-dht_itransform (xlator_t *this, xlator_t *subvol, uint64_t x, uint64_t *y_p)
+    fd_ctx = dht_fd_ctx_get(this, fd);
+
+    if (fd_ctx) {
+        if (fd_ctx->opened_on_dst == (uint64_t)(uintptr_t)dst) {
+            opened = _gf_true;
+        }
+        GF_REF_PUT(fd_ctx);
+    }
+
+    return opened;
+}
+
+void
+dht_free_mig_info(void *data)
 {
-	dht_conf_t *conf = NULL;
-	int         cnt = 0;
-	int         max = 0;
-	uint64_t    y = 0;
+    dht_migrate_info_t *miginfo = NULL;
 
+    miginfo = data;
+    GF_FREE(miginfo);
 
-	if (x == ((uint64_t) -1)) {
-		y = (uint64_t) -1;
-		goto out;
-	}
+    return;
+}
 
-	conf = this->private;
-        if (!conf)
-                goto out;
+static int
+dht_inode_ctx_set_mig_info(xlator_t *this, inode_t *inode, xlator_t *src_subvol,
+                           xlator_t *dst_subvol)
+{
+    dht_migrate_info_t *miginfo = NULL;
+    uint64_t value = 0;
+    int ret = -1;
+
+    miginfo = GF_CALLOC(1, sizeof(*miginfo), gf_dht_mt_miginfo_t);
+    if (miginfo == NULL)
+        goto out;
 
-	max = conf->subvolume_cnt;
-	cnt = dht_subvol_cnt (this, subvol);
+    miginfo->src_subvol = src_subvol;
+    miginfo->dst_subvol = dst_subvol;
+    GF_REF_INIT(miginfo, dht_free_mig_info);
 
-	y = ((x * max) + cnt);
+    value = (uint64_t)(uintptr_t)miginfo;
+
+    ret = inode_ctx_set1(inode, this, &value);
+    if (ret < 0) {
+        GF_REF_PUT(miginfo);
+    }
 
 out:
-	if (y_p)
-		*y_p = y;
+    return ret;
+}
 
-	return 0;
+int
+dht_inode_ctx_get_mig_info(xlator_t *this, inode_t *inode,
+                           xlator_t **src_subvol, xlator_t **dst_subvol)
+{
+    int ret = -1;
+    uint64_t tmp_miginfo = 0;
+    dht_migrate_info_t *miginfo = NULL;
+
+    LOCK(&inode->lock);
+    {
+        ret = __inode_ctx_get1(inode, this, &tmp_miginfo);
+        if ((ret < 0) || (tmp_miginfo == 0)) {
+            UNLOCK(&inode->lock);
+            goto out;
+        }
+
+        miginfo = (dht_migrate_info_t *)(uintptr_t)tmp_miginfo;
+        GF_REF_GET(miginfo);
+    }
+    UNLOCK(&inode->lock);
+
+    if (src_subvol)
+        *src_subvol = miginfo->src_subvol;
+
+    if (dst_subvol)
+        *dst_subvol = miginfo->dst_subvol;
+
+    GF_REF_PUT(miginfo);
+
+out:
+    return ret;
 }
 
+gf_boolean_t
+dht_mig_info_is_invalid(xlator_t *current, xlator_t *src_subvol,
+                        xlator_t *dst_subvol)
+{
+    /* Not set
+     */
+    if (!src_subvol || !dst_subvol)
+        return _gf_true;
+
+    /* Invalid scenarios:
+     * The src_subvol does not match the subvol on which the current op was sent
+     * so the cached subvol has changed between the last mig_info_set and now.
+     * src_subvol == dst_subvol. The file was migrated without any FOP detecting
+     * a P2 so the old dst is now the current subvol.
+     *
+     * There is still one scenario where the info could be outdated - if
+     * file has undergone multiple migrations and ends up on the same src_subvol
+     * on which the mig_info was first set.
+     */
+    if ((current == dst_subvol) || (current != src_subvol))
+        return _gf_true;
+
+    return _gf_false;
+}
+
+/* Used to check if fd fops have the fd opened on the cached subvol
+ * This is required when:
+ * 1. an fd is opened on FILE1 on subvol1
+ * 2. the file is migrated to subvol2
+ * 3. a lookup updates the cached subvol in the inode_ctx to subvol2
+ * 4. a write comes on the fd
+ * The write is sent to subvol2 on an fd which has been opened only on fd1
+ * Since the migration phase checks don't kick in, the fop fails with EBADF
+ *
+ */
+
 int
-dht_filter_loc_subvol_key (xlator_t *this, loc_t *loc, loc_t *new_loc,
-                           xlator_t **subvol)
+dht_check_and_open_fd_on_subvol_complete(int ret, call_frame_t *frame,
+                                         void *data)
 {
-        char          *new_name  = NULL;
-        char          *new_path  = NULL;
-        xlator_list_t *trav      = NULL;
-        char           key[1024] = {0,};
-        int            ret       = 0; /* not found */
+    glusterfs_fop_t fop = 0;
+    dht_local_t *local = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *this = NULL;
+    fd_t *fd = NULL;
+    int op_errno = -1;
+
+    local = frame->local;
+    this = frame->this;
+    fop = local->fop;
+    subvol = local->cached_subvol;
+    fd = local->fd;
+
+    if (ret) {
+        op_errno = local->op_errno;
+        goto handle_err;
+    }
+
+    switch (fop) {
+        case GF_FOP_WRITE:
+            STACK_WIND_COOKIE(frame, dht_writev_cbk, subvol, subvol,
+                              subvol->fops->writev, fd, local->rebalance.vector,
+                              local->rebalance.count, local->rebalance.offset,
+                              local->rebalance.flags, local->rebalance.iobref,
+                              local->xattr_req);
+            break;
+
+        case GF_FOP_FLUSH:
+            STACK_WIND(frame, dht_flush_cbk, subvol, subvol->fops->flush, fd,
+                       local->xattr_req);
+            break;
+
+        case GF_FOP_FSETATTR:
+            STACK_WIND_COOKIE(frame, dht_file_setattr_cbk, subvol, subvol,
+                              subvol->fops->fsetattr, fd,
+                              &local->rebalance.stbuf, local->rebalance.flags,
+                              local->xattr_req);
+            break;
+
+        case GF_FOP_ZEROFILL:
+            STACK_WIND_COOKIE(frame, dht_zerofill_cbk, subvol, subvol,
+                              subvol->fops->zerofill, fd,
+                              local->rebalance.offset, local->rebalance.size,
+                              local->xattr_req);
+
+            break;
+
+        case GF_FOP_DISCARD:
+            STACK_WIND_COOKIE(frame, dht_discard_cbk, subvol, subvol,
+                              subvol->fops->discard, local->fd,
+                              local->rebalance.offset, local->rebalance.size,
+                              local->xattr_req);
+            break;
+
+        case GF_FOP_FALLOCATE:
+            STACK_WIND_COOKIE(frame, dht_fallocate_cbk, subvol, subvol,
+                              subvol->fops->fallocate, fd,
+                              local->rebalance.flags, local->rebalance.offset,
+                              local->rebalance.size, local->xattr_req);
+            break;
+
+        case GF_FOP_FTRUNCATE:
+            STACK_WIND_COOKIE(frame, dht_truncate_cbk, subvol, subvol,
+                              subvol->fops->ftruncate, fd,
+                              local->rebalance.offset, local->xattr_req);
+            break;
+
+        case GF_FOP_FSYNC:
+            STACK_WIND_COOKIE(frame, dht_fsync_cbk, subvol, subvol,
+                              subvol->fops->fsync, local->fd,
+                              local->rebalance.flags, local->xattr_req);
+            break;
+
+        case GF_FOP_READ:
+            STACK_WIND(frame, dht_readv_cbk, subvol, subvol->fops->readv,
+                       local->fd, local->rebalance.size,
+                       local->rebalance.offset, local->rebalance.flags,
+                       local->xattr_req);
+            break;
+
+        case GF_FOP_FSTAT:
+            STACK_WIND_COOKIE(frame, dht_file_attr_cbk, subvol, subvol,
+                              subvol->fops->fstat, fd, local->xattr_req);
+            break;
+
+        case GF_FOP_FSETXATTR:
+            STACK_WIND_COOKIE(frame, dht_file_setxattr_cbk, subvol, subvol,
+                              subvol->fops->fsetxattr, local->fd,
+                              local->rebalance.xattr, local->rebalance.flags,
+                              local->xattr_req);
+            break;
+
+        case GF_FOP_FREMOVEXATTR:
+            STACK_WIND_COOKIE(frame, dht_file_removexattr_cbk, subvol, subvol,
+                              subvol->fops->fremovexattr, local->fd, local->key,
+                              local->xattr_req);
+
+            break;
+
+        case GF_FOP_FXATTROP:
+            STACK_WIND(frame, dht_common_xattrop_cbk, subvol,
+                       subvol->fops->fxattrop, local->fd,
+                       local->rebalance.flags, local->rebalance.xattr,
+                       local->xattr_req);
+            break;
+
+        case GF_FOP_FGETXATTR:
+            STACK_WIND(frame, dht_getxattr_cbk, subvol, subvol->fops->fgetxattr,
+                       local->fd, local->key, NULL);
+            break;
+
+        case GF_FOP_FINODELK:
+            STACK_WIND(frame, dht_finodelk_cbk, subvol, subvol->fops->finodelk,
+                       local->key, local->fd, local->rebalance.lock_cmd,
+                       &local->rebalance.flock, local->xattr_req);
+            break;
+        default:
+            gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_UNKNOWN_FOP, "fd=%p",
+                    fd, "gfid=%s", uuid_utoa(fd->inode->gfid), "name=%s",
+                    subvol->name, NULL);
+            break;
+    }
+
+    goto out;
+
+    /* Could not open the fd on the dst. Unwind */
+
+handle_err:
+
+    switch (fop) {
+        case GF_FOP_WRITE:
+            DHT_STACK_UNWIND(writev, frame, -1, op_errno, NULL, NULL, NULL);
+            break;
+
+        case GF_FOP_FLUSH:
+            DHT_STACK_UNWIND(flush, frame, -1, op_errno, NULL);
+            break;
+
+        case GF_FOP_FSETATTR:
+            DHT_STACK_UNWIND(fsetattr, frame, -1, op_errno, NULL, NULL, NULL);
+            break;
+
+        case GF_FOP_ZEROFILL:
+            DHT_STACK_UNWIND(zerofill, frame, -1, op_errno, NULL, NULL, NULL);
+            break;
+
+        case GF_FOP_DISCARD:
+            DHT_STACK_UNWIND(discard, frame, -1, op_errno, NULL, NULL, NULL);
+            break;
+
+        case GF_FOP_FALLOCATE:
+            DHT_STACK_UNWIND(fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+            break;
+
+        case GF_FOP_FTRUNCATE:
+            DHT_STACK_UNWIND(ftruncate, frame, -1, op_errno, NULL, NULL, NULL);
+            break;
+
+        case GF_FOP_FSYNC:
+            DHT_STACK_UNWIND(fsync, frame, -1, op_errno, NULL, NULL, NULL);
+            break;
+
+        case GF_FOP_READ:
+            DHT_STACK_UNWIND(readv, frame, -1, op_errno, NULL, 0, NULL, NULL,
+                             NULL);
+            break;
+
+        case GF_FOP_FSTAT:
+            DHT_STACK_UNWIND(fstat, frame, -1, op_errno, NULL, NULL);
+            break;
+
+        case GF_FOP_FSETXATTR:
+            DHT_STACK_UNWIND(fsetxattr, frame, -1, op_errno, NULL);
+            break;
+
+        case GF_FOP_FREMOVEXATTR:
+            DHT_STACK_UNWIND(fremovexattr, frame, -1, op_errno, NULL);
+            break;
+
+        case GF_FOP_FXATTROP:
+            DHT_STACK_UNWIND(fxattrop, frame, -1, op_errno, NULL, NULL);
+            break;
+
+        case GF_FOP_FGETXATTR:
+            DHT_STACK_UNWIND(fgetxattr, frame, -1, op_errno, NULL, NULL);
+            break;
+
+        case GF_FOP_FINODELK:
+            DHT_STACK_UNWIND(finodelk, frame, -1, op_errno, NULL);
+            break;
+
+        default:
+            gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_UNKNOWN_FOP, "fd=%p",
+                    fd, "gfid=%s", uuid_utoa(fd->inode->gfid), "name=%s",
+                    subvol->name, NULL);
+            break;
+    }
 
-        /* Why do other tasks if first required 'char' itself is not there */
-        if (loc->name && !strchr (loc->name, '@'))
-                goto out;
+out:
 
-        trav = this->children;
-        while (trav) {
-                snprintf (key, 1024, "*@%s:%s", this->name, trav->xlator->name);
-                if (fnmatch (key, loc->name, FNM_NOESCAPE) == 0) {
-                        new_name = GF_CALLOC(strlen (loc->name),
-                                             sizeof (char),
-                                             gf_common_mt_char);
-                        if (!new_name)
-                                goto out;
-                        if (fnmatch (key, loc->path, FNM_NOESCAPE) == 0) {
-                                new_path = GF_CALLOC(strlen (loc->path),
-                                                     sizeof (char),
-                                                     gf_common_mt_char);
-                                if (!new_path)
-                                        goto out;
-                                strncpy (new_path, loc->path, (strlen (loc->path) -
-                                                               strlen (key) + 1));
-                        }
-                        strncpy (new_name, loc->name, (strlen (loc->name) -
-                                                       strlen (key) + 1));
-
-                        if (new_loc) {
-                                new_loc->path   = ((new_path) ? new_path:
-                                                   gf_strdup (loc->path));
-                                new_loc->name   = new_name;
-                                new_loc->ino    = loc->ino;
-                                new_loc->inode  = inode_ref (loc->inode);
-                                new_loc->parent = inode_ref (loc->parent);
-                        }
-                        *subvol         = trav->xlator;
-                        ret = 1;  /* success */
-                        goto out;
-                }
-                trav = trav->next;
+    return 0;
+}
+
+/* Check once again if the fd has been opened on the cached subvol.
+ * If not, open and update the fd_ctx.
+ */
+
+int
+dht_check_and_open_fd_on_subvol_task(void *data)
+{
+    loc_t loc = {
+        0,
+    };
+    int ret = -1;
+    call_frame_t *frame = NULL;
+    dht_local_t *local = NULL;
+    fd_t *fd = NULL;
+    xlator_t *this = NULL;
+    xlator_t *subvol = NULL;
+
+    frame = data;
+    local = frame->local;
+    this = THIS;
+    fd = local->fd;
+    subvol = local->cached_subvol;
+
+    local->fd_checked = _gf_true;
+
+    if (fd_is_anonymous(fd) || dht_fd_open_on_dst(this, fd, subvol)) {
+        ret = 0;
+        goto out;
+    }
+
+    gf_msg_debug(this->name, 0, "Opening fd (%p, flags=0%o) on file %s @ %s",
+                 fd, fd->flags, uuid_utoa(fd->inode->gfid), subvol->name);
+
+    loc.inode = inode_ref(fd->inode);
+    gf_uuid_copy(loc.gfid, fd->inode->gfid);
+
+    /* Open this on the dst subvol */
+
+    SYNCTASK_SETID(0, 0);
+
+    ret = syncop_open(subvol, &loc, (fd->flags & ~(O_CREAT | O_EXCL | O_TRUNC)),
+                      fd, NULL, NULL);
+
+    if (ret < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_OPEN_FD_ON_DST_FAILED,
+                "fd=%p", fd, "flags=0%o", fd->flags, "gfid=%s",
+                uuid_utoa(fd->inode->gfid), "name=%s", subvol->name, NULL);
+        /* This can happen if the cached subvol was updated in the
+         * inode_ctx and the fd was opened on the new cached suvol
+         * after this fop was wound on the old cached subvol.
+         * As we do not close the fd on the old subvol (a leak)
+         * don't treat ENOENT as an error and allow the phase1/phase2
+         * checks to handle it.
+         */
+
+        if ((-ret != ENOENT) && (-ret != ESTALE)) {
+            local->op_errno = -ret;
+            ret = -1;
+        } else {
+            ret = 0;
         }
+
+        local->op_errno = -ret;
+        ret = -1;
+
+    } else {
+        dht_fd_ctx_set(this, fd, subvol);
+    }
+
+    SYNCTASK_SETID(frame->root->uid, frame->root->gid);
 out:
-        if (!ret) {
-                /* !success */
-                if (new_path)
-                        GF_FREE (new_path);
-                if (new_name)
-                        GF_FREE (new_name);
-        }
-        return ret;
+    loc_wipe(&loc);
+
+    return ret;
+}
+
+int
+dht_check_and_open_fd_on_subvol(xlator_t *this, call_frame_t *frame)
+{
+    int ret = -1;
+    dht_local_t *local = NULL;
+
+    /*
+            if (dht_fd_open_on_dst (this, fd, subvol))
+                    goto out;
+    */
+    local = frame->local;
+
+    ret = synctask_new(this->ctx->env, dht_check_and_open_fd_on_subvol_task,
+                       dht_check_and_open_fd_on_subvol_complete, frame, frame);
+
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SYNCTASK_CREATE_FAILED,
+                "to-check-and-open fd=%p", local->fd, NULL);
+    }
+
+    return ret;
 }
 
 int
-dht_deitransform (xlator_t *this, uint64_t y, xlator_t **subvol_p,
-		  uint64_t *x_p)
+dht_frame_return(call_frame_t *frame)
 {
-	dht_conf_t *conf = NULL;
-	int         cnt = 0;
-	int         max = 0;
-	uint64_t    x = 0;
-	xlator_t   *subvol = 0;
+    dht_local_t *local = NULL;
+    int this_call_cnt = -1;
+
+    if (!frame)
+        return -1;
+
+    local = frame->local;
+
+    LOCK(&frame->lock);
+    {
+        this_call_cnt = --local->call_cnt;
+    }
+    UNLOCK(&frame->lock);
+
+    return this_call_cnt;
+}
+
+/*
+ * Use this function to specify which subvol you want the file created
+ * on - this need not be the hashed subvol.
+ * Format: <filename>@<this->name>:<subvol-name>
+ * Eg: file-1@vol1-dht:vol1-client-0
+ *     where vol1 is a pure distribute volume
+ *     will create file-1 on vol1-client-0
+ */
 
-        if (!this->private)
+int
+dht_filter_loc_subvol_key(xlator_t *this, loc_t *loc, loc_t *new_loc,
+                          xlator_t **subvol)
+{
+    char *new_name = NULL;
+    char *new_path = NULL;
+    xlator_list_t *trav = NULL;
+    char key[1024] = {
+        0,
+    };
+    int ret = 0; /* not found */
+    int keylen = 0;
+    int name_len = 0;
+    int path_len = 0;
+
+    /* Why do other tasks if first required 'char' itself is not there */
+    if (!new_loc || !loc || !loc->name || !strchr(loc->name, '@')) {
+        /* Skip the GF_FREE checks here */
+        return ret;
+    }
+
+    trav = this->children;
+    while (trav) {
+        keylen = snprintf(key, sizeof(key), "*@%s:%s", this->name,
+                          trav->xlator->name);
+        /* Ignore '*' */
+        keylen = keylen - 1;
+        if (fnmatch(key, loc->name, FNM_NOESCAPE) == 0) {
+            name_len = strlen(loc->name) - keylen;
+            new_name = GF_MALLOC(name_len + 1, gf_common_mt_char);
+            if (!new_name)
                 goto out;
+            if (fnmatch(key, loc->path, FNM_NOESCAPE) == 0) {
+                path_len = strlen(loc->path) - keylen;
+                new_path = GF_MALLOC(path_len + 1, gf_common_mt_char);
+                if (!new_path)
+                    goto out;
+                snprintf(new_path, path_len + 1, "%s", loc->path);
+            }
+            snprintf(new_name, name_len + 1, "%s", loc->name);
+
+            if (new_loc) {
+                new_loc->path = ((new_path) ? new_path : gf_strdup(loc->path));
+                new_loc->name = new_name;
+                new_loc->inode = inode_ref(loc->inode);
+                new_loc->parent = inode_ref(loc->parent);
+            }
+            *subvol = trav->xlator;
+            ret = 1; /* success */
+            goto out;
+        }
+        trav = trav->next;
+    }
+out:
+    if (!ret) {
+        /* !success */
+        GF_FREE(new_path);
+        GF_FREE(new_name);
+    }
+    return ret;
+}
 
-	conf = this->private;
-	max = conf->subvolume_cnt;
+static xlator_t *
+dht_get_subvol_from_id(xlator_t *this, int client_id)
+{
+    xlator_t *xl = NULL;
+    dht_conf_t *conf = NULL;
+    char *sid = NULL;
+    int32_t ret = -1;
 
-	cnt = y % max;
-	x   = y / max;
+    conf = this->private;
 
-	subvol = conf->subvolumes[cnt];
+    ret = gf_asprintf(&sid, "%d", client_id);
+    if (ret == -1) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_ASPRINTF_FAILED, NULL);
+        goto out;
+    }
 
-	if (subvol_p)
-		*subvol_p = subvol;
+    if (dict_get_ptr(conf->leaf_to_subvol, sid, (void **)&xl))
+        xl = NULL;
 
-	if (x_p)
-		*x_p = x;
+    GF_FREE(sid);
 
 out:
-	return 0;
+    return xl;
 }
 
+int
+dht_deitransform(xlator_t *this, uint64_t y, xlator_t **subvol_p)
+{
+    int client_id = 0;
+    xlator_t *subvol = 0;
+    dht_conf_t *conf = NULL;
+
+    if (!this->private)
+        return -1;
+
+    conf = this->private;
+
+    client_id = gf_deitransform(this, y);
+
+    subvol = dht_get_subvol_from_id(this, client_id);
+
+    if (!subvol)
+        subvol = conf->subvolumes[0];
+
+    if (subvol_p)
+        *subvol_p = subvol;
+
+    return 0;
+}
 
 void
-dht_local_wipe (xlator_t *this, dht_local_t *local)
+dht_local_wipe(xlator_t *this, dht_local_t *local)
 {
-	if (!local)
-		return;
+    int i = 0;
 
-	loc_wipe (&local->loc);
-	loc_wipe (&local->loc2);
+    if (!local)
+        return;
 
-	if (local->xattr)
-		dict_unref (local->xattr);
+    loc_wipe(&local->loc);
+    loc_wipe(&local->loc2);
+    loc_wipe(&local->loc2_copy);
 
-	if (local->inode)
-		inode_unref (local->inode);
+    if (local->xattr)
+        dict_unref(local->xattr);
 
-	if (local->layout) {
-		dht_layout_unref (this, local->layout);
-                local->layout = NULL;
-        }
+    if (local->inode)
+        inode_unref(local->inode);
 
-	loc_wipe (&local->linkfile.loc);
+    if (local->layout) {
+        dht_layout_unref(this, local->layout);
+        local->layout = NULL;
+    }
 
-	if (local->linkfile.xattr)
-		dict_unref (local->linkfile.xattr);
+    loc_wipe(&local->linkfile.loc);
 
-	if (local->linkfile.inode)
-		inode_unref (local->linkfile.inode);
+    if (local->linkfile.xattr)
+        dict_unref(local->linkfile.xattr);
 
-	if (local->fd) {
-		fd_unref (local->fd);
-		local->fd = NULL;
-	}
+    if (local->linkfile.inode)
+        inode_unref(local->linkfile.inode);
 
-        if (local->params) {
-                dict_unref (local->params);
-                local->params = NULL;
-        }
+    if (local->fd) {
+        fd_unref(local->fd);
+        local->fd = NULL;
+    }
 
-	if (local->xattr_req)
-		dict_unref (local->xattr_req);
+    if (local->params) {
+        dict_unref(local->params);
+        local->params = NULL;
+    }
 
-        if (local->selfheal.layout) {
-                dht_layout_unref (this, local->selfheal.layout);
-                local->selfheal.layout = NULL;
-        }
+    if (local->xattr_req)
+        dict_unref(local->xattr_req);
+    if (local->mds_xattr)
+        dict_unref(local->mds_xattr);
+    if (local->xdata)
+        dict_unref(local->xdata);
 
-        if (local->newpath) {
-                GF_FREE (local->newpath);
-        }
+    if (local->selfheal.layout) {
+        dht_layout_unref(this, local->selfheal.layout);
+        local->selfheal.layout = NULL;
+    }
 
-	GF_FREE (local);
-}
+    if (local->selfheal.refreshed_layout) {
+        dht_layout_unref(this, local->selfheal.refreshed_layout);
+        local->selfheal.refreshed_layout = NULL;
+    }
 
+    for (i = 0; i < 2; i++) {
+        dht_lock_array_free(local->lock[i].ns.parent_layout.locks,
+                            local->lock[i].ns.parent_layout.lk_count);
 
-dht_local_t *
-dht_local_init (call_frame_t *frame)
-{
-	dht_local_t *local = NULL;
+        GF_FREE(local->lock[i].ns.parent_layout.locks);
 
-	/* TODO: use mem-pool */
-	local = GF_CALLOC (1, sizeof (*local),
-                           gf_dht_mt_dht_local_t);
+        dht_lock_array_free(local->lock[i].ns.directory_ns.locks,
+                            local->lock[i].ns.directory_ns.lk_count);
+        GF_FREE(local->lock[i].ns.directory_ns.locks);
+    }
 
-	if (!local)
-		return NULL;
+    GF_FREE(local->key);
 
-	local->op_ret = -1;
-	local->op_errno = EUCLEAN;
+    if (local->rebalance.xdata)
+        dict_unref(local->rebalance.xdata);
 
-	frame->local = local;
+    if (local->rebalance.xattr)
+        dict_unref(local->rebalance.xattr);
 
-	return local;
-}
+    if (local->rebalance.dict)
+        dict_unref(local->rebalance.dict);
 
+    GF_FREE(local->rebalance.vector);
 
-char *
-basestr (const char *str)
-{
-        char *basestr = NULL;
+    if (local->rebalance.iobref)
+        iobref_unref(local->rebalance.iobref);
 
-        basestr = strrchr (str, '/');
-        if (basestr)
-                basestr ++;
+    if (local->stub) {
+        call_stub_destroy(local->stub);
+        local->stub = NULL;
+    }
 
-        return basestr;
-}
+    if (local->ret_cache)
+        GF_FREE(local->ret_cache);
 
+    mem_put(local);
+}
 
-xlator_t *
-dht_first_up_subvol (xlator_t *this)
+dht_local_t *
+dht_local_init(call_frame_t *frame, loc_t *loc, fd_t *fd, glusterfs_fop_t fop)
 {
-	dht_conf_t *conf = NULL;
-	xlator_t   *child = NULL;
-	int         i = 0;
+    dht_local_t *local = NULL;
+    inode_t *inode = NULL;
+    int ret = 0;
 
-	conf = this->private;
-        if (!conf)
-                goto out;
+    local = mem_get0(THIS->local_pool);
+    if (!local)
+        goto out;
+
+    if (loc) {
+        ret = loc_copy(&local->loc, loc);
+        if (ret)
+            goto out;
+
+        inode = loc->inode;
+    }
+
+    if (fd) {
+        local->fd = fd_ref(fd);
+        if (!inode)
+            inode = fd->inode;
+    }
+
+    local->op_ret = -1;
+    local->op_errno = EUCLEAN;
+    local->fop = fop;
 
-	LOCK (&conf->subvolume_lock);
-	{
-		for (i = 0; i < conf->subvolume_cnt; i++) {
-			if (conf->subvolume_status[i]) {
-				child = conf->subvolumes[i];
-				break;
-			}
-		}
-	}
-	UNLOCK (&conf->subvolume_lock);
+    if (inode) {
+        local->layout = dht_layout_get(frame->this, inode);
+        local->cached_subvol = dht_subvol_get_cached(frame->this, inode);
+    }
+
+    frame->local = local;
 
 out:
-	return child;
+    if (ret) {
+        if (local)
+            mem_put(local);
+        local = NULL;
+    }
+    return local;
 }
 
 xlator_t *
-dht_last_up_subvol (xlator_t *this)
+dht_first_up_subvol(xlator_t *this)
 {
-        dht_conf_t *conf = NULL;
-        xlator_t   *child = NULL;
-        int         i = 0;
+    dht_conf_t *conf = NULL;
+    xlator_t *child = NULL;
+    int i = 0;
+    time_t time = 0;
+
+    conf = this->private;
+    if (!conf)
+        goto out;
+
+    LOCK(&conf->subvolume_lock);
+    {
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (conf->subvol_up_time[i]) {
+                if (!time) {
+                    time = conf->subvol_up_time[i];
+                    child = conf->subvolumes[i];
+                } else if (time > conf->subvol_up_time[i]) {
+                    time = conf->subvol_up_time[i];
+                    child = conf->subvolumes[i];
+                }
+            }
+        }
+    }
+    UNLOCK(&conf->subvolume_lock);
 
-        conf = this->private;
-        if (!conf)
-                goto out;
+out:
+    return child;
+}
 
-        LOCK (&conf->subvolume_lock);
-        {
-                for (i = conf->subvolume_cnt-1; i >= 0; i--) {
-                        if (conf->subvolume_status[i]) {
-                                child = conf->subvolumes[i];
-                                break;
-                        }
-                }
+xlator_t *
+dht_last_up_subvol(xlator_t *this)
+{
+    dht_conf_t *conf = NULL;
+    xlator_t *child = NULL;
+    int i = 0;
+
+    conf = this->private;
+    if (!conf)
+        goto out;
+
+    LOCK(&conf->subvolume_lock);
+    {
+        for (i = conf->subvolume_cnt - 1; i >= 0; i--) {
+            if (conf->subvolume_status[i]) {
+                child = conf->subvolumes[i];
+                break;
+            }
         }
-        UNLOCK (&conf->subvolume_lock);
+    }
+    UNLOCK(&conf->subvolume_lock);
 
 out:
-        return child;
+    return child;
 }
 
 xlator_t *
-dht_subvol_get_hashed (xlator_t *this, loc_t *loc)
+dht_subvol_get_hashed(xlator_t *this, loc_t *loc)
 {
-        dht_layout_t *layout = NULL;
-        xlator_t     *subvol = NULL;
+    dht_layout_t *layout = NULL;
+    xlator_t *subvol = NULL;
+    dht_conf_t *conf = NULL;
+    dht_methods_t *methods = NULL;
 
-        if (is_fs_root (loc)) {
-                subvol = dht_first_up_subvol (this);
-                goto out;
-        }
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
 
-        layout = dht_layout_get (this, loc->parent);
+    conf = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, conf, out);
 
-        if (!layout) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "layout missing path=%s parent=%"PRId64,
-                        loc->path, loc->parent->ino);
-                goto out;
-        }
+    methods = &(conf->methods);
 
-        subvol = dht_layout_search (this, layout, loc->name);
+    if (__is_root_gfid(loc->gfid)) {
+        subvol = dht_first_up_subvol(this);
+        goto out;
+    }
 
-        if (!subvol) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "could not find subvolume for path=%s",
-                        loc->path);
-                goto out;
-        }
+    GF_VALIDATE_OR_GOTO(this->name, loc->parent, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->name, out);
+
+    layout = dht_layout_get(this, loc->parent);
+
+    if (!layout) {
+        gf_msg_debug(this->name, 0, "Missing layout. path=%s, parent gfid =%s",
+                     loc->path, uuid_utoa(loc->parent->gfid));
+        goto out;
+    }
+
+    subvol = methods->layout_search(this, layout, loc->name);
+
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "No hashed subvolume for path=%s",
+                     loc->path);
+        goto out;
+    }
 
 out:
-        if (layout) {
-                dht_layout_unref (this, layout);
-        }
+    if (layout) {
+        dht_layout_unref(this, layout);
+    }
 
-        return subvol;
+    return subvol;
 }
 
-
 xlator_t *
-dht_subvol_get_cached (xlator_t *this, inode_t *inode)
+dht_subvol_get_cached(xlator_t *this, inode_t *inode)
 {
-        dht_layout_t *layout = NULL;
-        xlator_t     *subvol = NULL;
+    dht_layout_t *layout = NULL;
+    xlator_t *subvol = NULL;
 
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
 
-        layout = dht_layout_get (this, inode);
+    layout = dht_layout_get(this, inode);
 
-        if (!layout) {
-                goto out;
+    if (!layout) {
+        goto out;
+    }
+
+    subvol = layout->list[0].xlator;
+
+out:
+    if (layout) {
+        dht_layout_unref(this, layout);
+    }
+
+    return subvol;
+}
+
+xlator_t *
+dht_subvol_next(xlator_t *this, xlator_t *prev)
+{
+    dht_conf_t *conf = NULL;
+    int i = 0;
+    xlator_t *next = NULL;
+
+    conf = this->private;
+    if (!conf)
+        goto out;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (conf->subvolumes[i] == prev) {
+            if ((i + 1) < conf->subvolume_cnt)
+                next = conf->subvolumes[i + 1];
+            break;
         }
+    }
 
-	subvol = layout->list[0].xlator;
+out:
+    return next;
+}
+
+/* This func wraps around, if prev is actually the last subvol.
+ */
+xlator_t *
+dht_subvol_next_available(xlator_t *this, xlator_t *prev)
+{
+    dht_conf_t *conf = NULL;
+    int i = 0;
+    xlator_t *next = NULL;
+
+    conf = this->private;
+    if (!conf)
+        goto out;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (conf->subvolumes[i] == prev) {
+            /* if prev is last in conf->subvolumes, then wrap
+             * around.
+             */
+            if ((i + 1) < conf->subvolume_cnt) {
+                next = conf->subvolumes[i + 1];
+            } else {
+                next = conf->subvolumes[0];
+            }
+            break;
+        }
+    }
 
 out:
-        if (layout) {
-                dht_layout_unref (this, layout);
+    return next;
+}
+int
+dht_subvol_cnt(xlator_t *this, xlator_t *subvol)
+{
+    int i = 0;
+    int ret = -1;
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+    if (!conf)
+        goto out;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (subvol == conf->subvolumes[i]) {
+            ret = i;
+            break;
         }
+    }
 
-        return subvol;
+out:
+    return ret;
 }
 
+#define set_if_greater(a, b)                                                   \
+    do {                                                                       \
+        if ((a) < (b))                                                         \
+            (a) = (b);                                                         \
+    } while (0)
 
-xlator_t *
-dht_subvol_next (xlator_t *this, xlator_t *prev)
+#define set_if_greater_time(a, an, b, bn)                                      \
+    do {                                                                       \
+        if (((a) < (b)) || (((a) == (b)) && ((an) < (bn)))) {                  \
+            (a) = (b);                                                         \
+            (an) = (bn);                                                       \
+        }                                                                      \
+    } while (0)
+
+int
+dht_iatt_merge(xlator_t *this, struct iatt *to, struct iatt *from)
 {
-	dht_conf_t *conf = NULL;
-	int         i = 0;
-	xlator_t   *next = NULL;
+    if (!from || !to)
+        return 0;
 
-	conf = this->private;
-        if (!conf)
-                goto out;
+    to->ia_dev = from->ia_dev;
+
+    gf_uuid_copy(to->ia_gfid, from->ia_gfid);
+
+    to->ia_ino = from->ia_ino;
+    to->ia_prot = from->ia_prot;
+    to->ia_type = from->ia_type;
+    to->ia_nlink = from->ia_nlink;
+    to->ia_rdev = from->ia_rdev;
+    to->ia_size += from->ia_size;
+    to->ia_blksize = from->ia_blksize;
+    to->ia_blocks += from->ia_blocks;
+
+    if (IA_ISDIR(from->ia_type)) {
+        to->ia_blocks = DHT_DIR_STAT_BLOCKS;
+        to->ia_size = DHT_DIR_STAT_SIZE;
+    }
+    set_if_greater(to->ia_uid, from->ia_uid);
+    set_if_greater(to->ia_gid, from->ia_gid);
+
+    set_if_greater_time(to->ia_atime, to->ia_atime_nsec, from->ia_atime,
+                        from->ia_atime_nsec);
+    set_if_greater_time(to->ia_mtime, to->ia_mtime_nsec, from->ia_mtime,
+                        from->ia_mtime_nsec);
+    set_if_greater_time(to->ia_ctime, to->ia_ctime_nsec, from->ia_ctime,
+                        from->ia_ctime_nsec);
+
+    return 0;
+}
+
+int
+dht_build_child_loc(xlator_t *this, loc_t *child, loc_t *parent, char *name)
+{
+    if (!child) {
+        goto err;
+    }
 
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		if (conf->subvolumes[i] == prev) {
-			if ((i + 1) < conf->subvolume_cnt)
-				next = conf->subvolumes[i + 1];
-			break;
-		}
-	}
+    if (strcmp(parent->path, "/") == 0)
+        gf_asprintf((char **)&child->path, "/%s", name);
+    else
+        gf_asprintf((char **)&child->path, "%s/%s", parent->path, name);
 
-out:
-	return next;
+    if (!child->path) {
+        goto err;
+    }
+
+    child->name = strrchr(child->path, '/');
+    if (child->name)
+        child->name++;
+
+    child->parent = inode_ref(parent->inode);
+    child->inode = inode_new(parent->inode->table);
+
+    if (!child->inode) {
+        goto err;
+    }
+
+    return 0;
+err:
+    if (child) {
+        loc_wipe(child);
+    }
+    return -1;
 }
 
+int
+dht_init_local_subvolumes(xlator_t *this, dht_conf_t *conf)
+{
+    xlator_list_t *subvols = NULL;
+    int cnt = 0;
+
+    if (!conf)
+        return -1;
+
+    for (subvols = this->children; subvols; subvols = subvols->next)
+        cnt++;
+
+    conf->local_subvols = GF_CALLOC(cnt, sizeof(xlator_t *),
+                                    gf_dht_mt_xlator_t);
+
+    /* FIX FIX : do this dynamically*/
+    conf->local_nodeuuids = GF_CALLOC(cnt, sizeof(subvol_nodeuuids_info_t),
+                                      gf_dht_nodeuuids_t);
+
+    if (!conf->local_subvols || !conf->local_nodeuuids) {
+        return -1;
+    }
+
+    conf->local_subvols_cnt = 0;
+
+    return 0;
+}
 
 int
-dht_subvol_cnt (xlator_t *this, xlator_t *subvol)
+dht_init_subvolumes(xlator_t *this, dht_conf_t *conf)
 {
-	int i = 0;
-	int ret = -1;
-	dht_conf_t *conf = NULL;
+    xlator_list_t *subvols = NULL;
+    int cnt = 0;
 
-	conf = this->private;
-        if (!conf)
-                goto out;
+    if (!conf)
+        return -1;
+
+    for (subvols = this->children; subvols; subvols = subvols->next)
+        cnt++;
+
+    conf->subvolumes = GF_CALLOC(cnt, sizeof(xlator_t *), gf_dht_mt_xlator_t);
+    if (!conf->subvolumes) {
+        return -1;
+    }
+    conf->subvolume_cnt = cnt;
+    /* Doesn't make sense to do any dht layer tasks
+       if the subvol count is 1. Set it as pass_through */
+    if (cnt == 1)
+        this->pass_through = _gf_true;
+
+    conf->local_subvols_cnt = 0;
+
+    dht_set_subvol_range(this);
+
+    cnt = 0;
+    for (subvols = this->children; subvols; subvols = subvols->next)
+        conf->subvolumes[cnt++] = subvols->xlator;
+
+    conf->subvolume_status = GF_CALLOC(cnt, sizeof(char), gf_dht_mt_char);
+    if (!conf->subvolume_status) {
+        return -1;
+    }
+
+    conf->last_event = GF_CALLOC(cnt, sizeof(int), gf_dht_mt_char);
+    if (!conf->last_event) {
+        return -1;
+    }
+
+    conf->subvol_up_time = GF_CALLOC(cnt, sizeof(time_t),
+                                     gf_dht_mt_subvol_time);
+    if (!conf->subvol_up_time) {
+        return -1;
+    }
 
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		if (subvol == conf->subvolumes[i]) {
-			ret = i;
-			break;
-		}
-	}
+    conf->du_stats = GF_CALLOC(conf->subvolume_cnt, sizeof(dht_du_t),
+                               gf_dht_mt_dht_du_t);
+    if (!conf->du_stats) {
+        return -1;
+    }
+
+    conf->decommissioned_bricks = GF_CALLOC(cnt, sizeof(xlator_t *),
+                                            gf_dht_mt_xlator_t);
+    if (!conf->decommissioned_bricks) {
+        return -1;
+    }
+
+    return 0;
+}
+
+/*
+ op_ret values :
+  0 : Success.
+ -1 : Failure.
+  1 : File is being migrated but not by this DHT layer.
+*/
+
+static int
+dht_migration_complete_check_done(int op_ret, call_frame_t *frame, void *data)
+{
+    dht_local_t *local = NULL;
+    xlator_t *subvol = NULL;
+
+    local = frame->local;
+
+    if (op_ret != 0)
+        goto out;
+
+    if (local->cached_subvol == NULL) {
+        local->op_errno = EINVAL;
+        goto out;
+    }
+
+    subvol = local->cached_subvol;
 
 out:
-	return ret;
+    local->rebalance.target_op_fn(THIS, subvol, frame, op_ret);
+
+    return 0;
 }
 
+int
+dht_migration_complete_check_task(void *data)
+{
+    int ret = -1;
+    xlator_t *src_node = NULL;
+    xlator_t *dst_node = NULL, *linkto_target = NULL;
+    dht_local_t *local = NULL;
+    dict_t *dict = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+    xlator_t *this = NULL;
+    call_frame_t *frame = NULL;
+    loc_t tmp_loc = {
+        0,
+    };
+    char *path = NULL;
+    dht_conf_t *conf = NULL;
+    inode_t *inode = NULL;
+    fd_t *iter_fd = NULL;
+    fd_t *tmp = NULL;
+    uint64_t tmp_miginfo = 0;
+    dht_migrate_info_t *miginfo = NULL;
+    gf_boolean_t skip_open = _gf_false;
+    int open_failed = 0;
+
+    this = THIS;
+    frame = data;
+    local = frame->local;
+    conf = this->private;
+
+    src_node = local->cached_subvol;
+
+    if (!local->loc.inode && !local->fd) {
+        local->op_errno = EINVAL;
+        goto out;
+    }
+
+    inode = (!local->fd) ? local->loc.inode : local->fd->inode;
+
+    /* getxattr on cached_subvol for 'linkto' value. Do path based getxattr
+     * as root:root. If a fd is already open, access check won't be done*/
+
+    if (!local->loc.inode) {
+        ret = syncop_fgetxattr(src_node, local->fd, &dict,
+                               conf->link_xattr_name, NULL, NULL);
+    } else {
+        SYNCTASK_SETID(0, 0);
+        ret = syncop_getxattr(src_node, &local->loc, &dict,
+                              conf->link_xattr_name, NULL, NULL);
+        SYNCTASK_SETID(frame->root->uid, frame->root->gid);
+    }
+
+    /*
+     * Each DHT xlator layer has its own name for the linkto xattr.
+     * If the file mode bits indicate the the file is being migrated but
+     * this layer's linkto xattr is not set, it means that another
+     * DHT layer is migrating the file. In this case, return 1 so
+     * the mode bits can be passed on to the higher layer for appropriate
+     * action.
+     */
+    if (-ret == ENODATA) {
+        /* This DHT translator is not migrating this file */
+
+        ret = inode_ctx_reset1(inode, this, &tmp_miginfo);
+        if (tmp_miginfo) {
+            /* This can be a problem if the file was
+             * migrated by two different layers. Raise
+             * a warning here.
+             */
+            gf_smsg(
+                this->name, GF_LOG_WARNING, 0, DHT_MSG_HAS_MIGINFO, "tmp=%s",
+                tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid), NULL);
+
+            miginfo = (void *)(uintptr_t)tmp_miginfo;
+            GF_REF_PUT(miginfo);
+        }
+        ret = 1;
+        goto out;
+    }
+
+    if (!ret)
+        linkto_target = dht_linkfile_subvol(this, NULL, NULL, dict);
+
+    if (local->loc.inode) {
+        loc_copy(&tmp_loc, &local->loc);
+    } else {
+        tmp_loc.inode = inode_ref(inode);
+        gf_uuid_copy(tmp_loc.gfid, inode->gfid);
+    }
+
+    ret = syncop_lookup(this, &tmp_loc, &stbuf, 0, 0, 0);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_FILE_LOOKUP_FAILED,
+                "tmp=%s", tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid),
+                "name=%s", this->name, NULL);
+        local->op_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    dst_node = dht_subvol_get_cached(this, tmp_loc.inode);
+    if (linkto_target && dst_node != linkto_target) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_INVALID_LINKFILE,
+                "linkto_target_name=%s", linkto_target->name, "dst_name=%s",
+                dst_node->name, NULL);
+    }
+
+    if (gf_uuid_compare(stbuf.ia_gfid, tmp_loc.inode->gfid)) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_GFID_MISMATCH, "tmp=%s",
+                tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid),
+                "dst_name=%s", dst_node->name, NULL);
+        ret = -1;
+        local->op_errno = EIO;
+        goto out;
+    }
+
+    /* update local. A layout is set in inode-ctx in lookup already */
+
+    dht_layout_unref(this, local->layout);
+
+    local->layout = dht_layout_get(frame->this, inode);
+    local->cached_subvol = dst_node;
+
+    ret = 0;
+
+    /* once we detect the migration complete, the inode-ctx2 is no more
+       required.. delete the ctx and also, it means, open() already
+       done on all the fd of inode */
+    ret = inode_ctx_reset1(inode, this, &tmp_miginfo);
+    if (tmp_miginfo) {
+        miginfo = (void *)(uintptr_t)tmp_miginfo;
+        GF_REF_PUT(miginfo);
+        goto out;
+    }
+
+    /* perform 'open()' on all the fd's present on the inode */
+    if (tmp_loc.path == NULL) {
+        inode_path(inode, NULL, &path);
+        if (path)
+            tmp_loc.path = path;
+    }
+
+    LOCK(&inode->lock);
+
+    if (list_empty(&inode->fd_list))
+        goto unlock;
+
+    /* perform open as root:root. There is window between linkfile
+     * creation(root:root) and setattr with the correct uid/gid
+     */
+    SYNCTASK_SETID(0, 0);
+
+    /* It's possible that we are the last user of iter_fd after each
+     * iteration. In this case the fd_unref() of iter_fd at the end of
+     * the loop will cause the destruction of the fd. So we need to
+     * iterate the list safely because iter_fd cannot be trusted.
+     */
+    iter_fd = list_entry((&inode->fd_list)->next, typeof(*iter_fd), inode_list);
+    while (&iter_fd->inode_list != (&inode->fd_list)) {
+        if (fd_is_anonymous(iter_fd) ||
+            (dht_fd_open_on_dst(this, iter_fd, dst_node))) {
+            if (!tmp) {
+                iter_fd = list_entry(iter_fd->inode_list.next, typeof(*iter_fd),
+                                     inode_list);
+                continue;
+            }
+            skip_open = _gf_true;
+        }
+        /* We need to release the inode->lock before calling
+         * syncop_open() to avoid possible deadlocks. However this
+         * can cause the iter_fd to be released by other threads.
+         * To avoid this, we take a reference before releasing the
+         * lock.
+         */
+        fd_ref(iter_fd);
+
+        UNLOCK(&inode->lock);
+
+        if (tmp) {
+            fd_unref(tmp);
+            tmp = NULL;
+        }
+        if (skip_open)
+            goto next;
+
+        /* flags for open are stripped down to allow following the
+         * new location of the file, otherwise we can get EEXIST or
+         * truncate the file again as rebalance is moving the data */
+        ret = syncop_open(dst_node, &tmp_loc,
+                          (iter_fd->flags & ~(O_CREAT | O_EXCL | O_TRUNC)),
+                          iter_fd, NULL, NULL);
+        if (ret < 0) {
+            gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                    DHT_MSG_OPEN_FD_ON_DST_FAILED, "id=%p", iter_fd,
+                    "flags=0%o", iter_fd->flags, "path=%s", path, "name=%s",
+                    dst_node->name, NULL);
+
+            open_failed = 1;
+            local->op_errno = -ret;
+            ret = -1;
+        } else {
+            dht_fd_ctx_set(this, iter_fd, dst_node);
+        }
+
+    next:
+        LOCK(&inode->lock);
+        skip_open = _gf_false;
+        tmp = iter_fd;
+        iter_fd = list_entry(tmp->inode_list.next, typeof(*tmp), inode_list);
+    }
+
+    SYNCTASK_SETID(frame->root->uid, frame->root->gid);
+
+    if (open_failed) {
+        ret = -1;
+        goto unlock;
+    }
+    ret = 0;
+
+unlock:
+    UNLOCK(&inode->lock);
+    if (tmp) {
+        fd_unref(tmp);
+        tmp = NULL;
+    }
 
-#define set_if_greater(a, b) do {		\
-		if ((a) < (b))			\
-			(a) = (b);		\
-	} while (0)
+out:
+    if (dict) {
+        dict_unref(dict);
+    }
+
+    loc_wipe(&tmp_loc);
+
+    return ret;
+}
 
 int
-dht_iatt_merge (xlator_t *this, struct iatt *to,
-		struct iatt *from, xlator_t *subvol)
+dht_rebalance_complete_check(xlator_t *this, call_frame_t *frame)
 {
-        if (!from || !to)
-                return 0;
+    int ret = -1;
+
+    ret = synctask_new(this->ctx->env, dht_migration_complete_check_task,
+                       dht_migration_complete_check_done, frame, frame);
+    return ret;
+}
 
-	to->ia_dev      = from->ia_dev;
+/* During 'in-progress' state, both nodes should have the file */
+/*
+ op_ret values :
+  0 : Success
+ -1 : Failure.
+  1 : File is being migrated but not by this DHT layer.
+*/
+static int
+dht_inprogress_check_done(int op_ret, call_frame_t *frame, void *data)
+{
+    dht_local_t *local = NULL;
+    xlator_t *dst_subvol = NULL, *src_subvol = NULL;
+    inode_t *inode = NULL;
 
-        uuid_copy (to->ia_gfid, from->ia_gfid);
+    local = frame->local;
 
-	dht_itransform (this, subvol, from->ia_ino, &to->ia_ino);
+    if (op_ret != 0)
+        goto out;
 
-	to->ia_prot     = from->ia_prot;
-	to->ia_type     = from->ia_type;
-	to->ia_nlink    = from->ia_nlink;
-	to->ia_rdev     = from->ia_rdev;
-	to->ia_size    += from->ia_size;
-	to->ia_blksize  = from->ia_blksize;
-	to->ia_blocks  += from->ia_blocks;
+    inode = local->loc.inode ? local->loc.inode : local->fd->inode;
 
-	set_if_greater (to->ia_uid, from->ia_uid);
-	set_if_greater (to->ia_gid, from->ia_gid);
+    dht_inode_ctx_get_mig_info(THIS, inode, &src_subvol, &dst_subvol);
+    if (dht_mig_info_is_invalid(local->cached_subvol, src_subvol, dst_subvol)) {
+        dst_subvol = dht_subvol_get_cached(THIS, inode);
+        if (!dst_subvol) {
+            local->op_errno = EINVAL;
+            goto out;
+        }
+    }
 
-	set_if_greater (to->ia_atime, from->ia_atime);
-	set_if_greater (to->ia_mtime, from->ia_mtime);
-	set_if_greater (to->ia_ctime, from->ia_ctime);
+out:
+    local->rebalance.target_op_fn(THIS, dst_subvol, frame, op_ret);
 
-	return 0;
+    return 0;
+}
+
+static int
+dht_rebalance_inprogress_task(void *data)
+{
+    int ret = -1;
+    xlator_t *src_node = NULL;
+    xlator_t *dst_node = NULL;
+    dht_local_t *local = NULL;
+    dict_t *dict = NULL;
+    call_frame_t *frame = NULL;
+    xlator_t *this = NULL;
+    char *path = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+    loc_t tmp_loc = {
+        0,
+    };
+    dht_conf_t *conf = NULL;
+    inode_t *inode = NULL;
+    fd_t *iter_fd = NULL;
+    fd_t *tmp = NULL;
+    int open_failed = 0;
+    uint64_t tmp_miginfo = 0;
+    dht_migrate_info_t *miginfo = NULL;
+    gf_boolean_t skip_open = _gf_false;
+
+    this = THIS;
+    frame = data;
+    local = frame->local;
+    conf = this->private;
+
+    src_node = local->cached_subvol;
+
+    if (!local->loc.inode && !local->fd)
+        goto out;
+
+    inode = (!local->fd) ? local->loc.inode : local->fd->inode;
+
+    /* getxattr on cached_subvol for 'linkto' value. Do path based getxattr
+     * as root:root. If a fd is already open, access check won't be done*/
+    if (local->loc.inode) {
+        SYNCTASK_SETID(0, 0);
+        ret = syncop_getxattr(src_node, &local->loc, &dict,
+                              conf->link_xattr_name, NULL, NULL);
+        SYNCTASK_SETID(frame->root->uid, frame->root->gid);
+    } else {
+        ret = syncop_fgetxattr(src_node, local->fd, &dict,
+                               conf->link_xattr_name, NULL, NULL);
+    }
+
+    /*
+     * Each DHT xlator layer has its own name for the linkto xattr.
+     * If the file mode bits indicate the the file is being migrated but
+     * this layer's linkto xattr is not present, it means that another
+     * DHT layer is migrating the file. In this case, return 1 so
+     * the mode bits can be passed on to the higher layer for appropriate
+     * action.
+     */
+
+    if (-ret == ENODATA) {
+        /* This DHT layer is not migrating this file */
+        ret = inode_ctx_reset1(inode, this, &tmp_miginfo);
+        if (tmp_miginfo) {
+            /* This can be a problem if the file was
+             * migrated by two different layers. Raise
+             * a warning here.
+             */
+            gf_smsg(
+                this->name, GF_LOG_WARNING, 0, DHT_MSG_HAS_MIGINFO, "tmp=%s",
+                tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid), NULL);
+            miginfo = (void *)(uintptr_t)tmp_miginfo;
+            GF_REF_PUT(miginfo);
+        }
+        ret = 1;
+        goto out;
+    }
+
+    if (ret < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_GET_XATTR_FAILED,
+                "path=%s", local->loc.path, NULL);
+        ret = -1;
+        goto out;
+    }
+
+    dst_node = dht_linkfile_subvol(this, NULL, NULL, dict);
+    if (!dst_node) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_GET_XATTR_FAILED,
+                "path=%s", local->loc.path, NULL);
+        ret = -1;
+        goto out;
+    }
+
+    local->rebalance.target_node = dst_node;
+
+    if (local->loc.inode) {
+        loc_copy(&tmp_loc, &local->loc);
+    } else {
+        tmp_loc.inode = inode_ref(inode);
+        gf_uuid_copy(tmp_loc.gfid, inode->gfid);
+    }
+
+    /* lookup on dst */
+    ret = syncop_lookup(dst_node, &tmp_loc, &stbuf, NULL, NULL, NULL);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_FILE_LOOKUP_FAILED,
+                "tmp=%s", tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid),
+                "name=%s", dst_node->name, NULL);
+        ret = -1;
+        goto out;
+    }
+
+    if (gf_uuid_compare(stbuf.ia_gfid, tmp_loc.inode->gfid)) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_GFID_MISMATCH, "tmp=%s",
+                tmp_loc.path ? tmp_loc.path : uuid_utoa(tmp_loc.gfid),
+                "name=%s", dst_node->name, NULL);
+        ret = -1;
+        goto out;
+    }
+    ret = 0;
+
+    if (tmp_loc.path == NULL) {
+        inode_path(inode, NULL, &path);
+        if (path)
+            tmp_loc.path = path;
+    }
+
+    LOCK(&inode->lock);
+
+    if (list_empty(&inode->fd_list))
+        goto unlock;
+
+    /* perform open as root:root. There is window between linkfile
+     * creation(root:root) and setattr with the correct uid/gid
+     */
+    SYNCTASK_SETID(0, 0);
+
+    /* It's possible that we are the last user of iter_fd after each
+     * iteration. In this case the fd_unref() of iter_fd at the end of
+     * the loop will cause the destruction of the fd. So we need to
+     * iterate the list safely because iter_fd cannot be trusted.
+     */
+    iter_fd = list_entry((&inode->fd_list)->next, typeof(*iter_fd), inode_list);
+    while (&iter_fd->inode_list != (&inode->fd_list)) {
+        /* We need to release the inode->lock before calling
+         * syncop_open() to avoid possible deadlocks. However this
+         * can cause the iter_fd to be released by other threads.
+         * To avoid this, we take a reference before releasing the
+         * lock.
+         */
+
+        if (fd_is_anonymous(iter_fd) ||
+            (dht_fd_open_on_dst(this, iter_fd, dst_node))) {
+            if (!tmp) {
+                iter_fd = list_entry(iter_fd->inode_list.next, typeof(*iter_fd),
+                                     inode_list);
+                continue;
+            }
+            skip_open = _gf_true;
+        }
+
+        /* Yes, this is ugly but there isn't a cleaner way to do this
+         * the fd_ref is an atomic increment so not too bad. We want to
+         * reduce the number of inode locks and unlocks.
+         */
+
+        fd_ref(iter_fd);
+        UNLOCK(&inode->lock);
+
+        if (tmp) {
+            fd_unref(tmp);
+            tmp = NULL;
+        }
+        if (skip_open)
+            goto next;
+
+        /* flags for open are stripped down to allow following the
+         * new location of the file, otherwise we can get EEXIST or
+         * truncate the file again as rebalance is moving the data */
+        ret = syncop_open(dst_node, &tmp_loc,
+                          (iter_fd->flags & ~(O_CREAT | O_EXCL | O_TRUNC)),
+                          iter_fd, NULL, NULL);
+        if (ret < 0) {
+            gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                    DHT_MSG_OPEN_FD_ON_DST_FAILED, "fd=%p", iter_fd,
+                    "flags=0%o", iter_fd->flags, "path=%s", path, "name=%s",
+                    dst_node->name, NULL);
+            ret = -1;
+            open_failed = 1;
+        } else {
+            /* Potential fd leak if this fails here as it will be
+               reopened at the next Phase1/2 check */
+            dht_fd_ctx_set(this, iter_fd, dst_node);
+        }
+
+    next:
+        LOCK(&inode->lock);
+        skip_open = _gf_false;
+        tmp = iter_fd;
+        iter_fd = list_entry(tmp->inode_list.next, typeof(*tmp), inode_list);
+    }
+
+    SYNCTASK_SETID(frame->root->uid, frame->root->gid);
+
+unlock:
+    UNLOCK(&inode->lock);
+
+    if (tmp) {
+        fd_unref(tmp);
+        tmp = NULL;
+    }
+    if (open_failed) {
+        ret = -1;
+        goto out;
+    }
+
+    ret = dht_inode_ctx_set_mig_info(this, inode, src_node, dst_node);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SET_INODE_CTX_FAILED,
+                "path=%s", local->loc.path, "name=%s", dst_node->name, NULL);
+        goto out;
+    }
+
+    ret = 0;
+out:
+    if (dict) {
+        dict_unref(dict);
+    }
+
+    loc_wipe(&tmp_loc);
+    return ret;
+}
+
+int
+dht_rebalance_in_progress_check(xlator_t *this, call_frame_t *frame)
+{
+    int ret = -1;
+
+    ret = synctask_new(this->ctx->env, dht_rebalance_inprogress_task,
+                       dht_inprogress_check_done, frame, frame);
+    return ret;
 }
 
 int
-dht_frame_su_do (call_frame_t *frame)
+dht_inode_ctx_layout_set(inode_t *inode, xlator_t *this,
+                         dht_layout_t *layout_int)
+{
+    dht_inode_ctx_t *ctx = NULL;
+    int ret = -1;
+
+    ret = dht_inode_ctx_get(inode, this, &ctx);
+    if (!ret && ctx) {
+        ctx->layout = layout_int;
+    } else {
+        ctx = GF_CALLOC(1, sizeof(*ctx), gf_dht_mt_inode_ctx_t);
+        if (!ctx)
+            return ret;
+        ctx->layout = layout_int;
+    }
+
+    ret = dht_inode_ctx_set(inode, this, ctx);
+
+    return ret;
+}
+
+void
+dht_inode_ctx_time_set(inode_t *inode, xlator_t *this, struct iatt *stat)
 {
-        dht_local_t     *local = NULL;
+    dht_inode_ctx_t *ctx = NULL;
+    dht_stat_time_t *time = 0;
+    int ret = -1;
 
-        local = frame->local;
+    ret = dht_inode_ctx_get(inode, this, &ctx);
 
-        local->uid = frame->root->uid;
-        local->gid = frame->root->gid;
+    if (ret)
+        return;
 
-        frame->root->uid = 0;
-        frame->root->gid = 0;
+    time = &ctx->time;
 
-        return 0;
+    time->mtime = stat->ia_mtime;
+    time->mtime_nsec = stat->ia_mtime_nsec;
+
+    time->ctime = stat->ia_ctime;
+    time->ctime_nsec = stat->ia_ctime_nsec;
+
+    time->atime = stat->ia_atime;
+    time->atime_nsec = stat->ia_atime_nsec;
+
+    return;
 }
 
+int
+dht_inode_ctx_time_update(inode_t *inode, xlator_t *this, struct iatt *stat,
+                          int32_t post)
+{
+    dht_inode_ctx_t *ctx = NULL;
+    dht_stat_time_t *time = 0;
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO(this->name, stat, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+
+    ret = dht_inode_ctx_get(inode, this, &ctx);
+
+    if (ret) {
+        ctx = GF_CALLOC(1, sizeof(*ctx), gf_dht_mt_inode_ctx_t);
+        if (!ctx)
+            return -1;
+    }
+
+    time = &ctx->time;
+
+    LOCK(&inode->lock);
+    {
+        DHT_UPDATE_TIME(time->mtime, time->mtime_nsec, stat->ia_mtime,
+                        stat->ia_mtime_nsec, post);
+        DHT_UPDATE_TIME(time->ctime, time->ctime_nsec, stat->ia_ctime,
+                        stat->ia_ctime_nsec, post);
+        DHT_UPDATE_TIME(time->atime, time->atime_nsec, stat->ia_atime,
+                        stat->ia_atime_nsec, post);
+    }
+    UNLOCK(&inode->lock);
+
+    ret = dht_inode_ctx_set(inode, this, ctx);
+out:
+    return 0;
+}
 
 int
-dht_frame_su_undo (call_frame_t *frame)
+dht_inode_ctx_get(inode_t *inode, xlator_t *this, dht_inode_ctx_t **ctx)
 {
-        dht_local_t     *local = NULL;
+    int ret = -1;
+    uint64_t ctx_int = 0;
 
-        local = frame->local;
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
 
-        frame->root->uid = local->uid;
-        frame->root->gid = local->gid;
+    ret = inode_ctx_get(inode, this, &ctx_int);
 
-        return 0;
+    if (ret)
+        return ret;
+
+    if (ctx)
+        *ctx = (dht_inode_ctx_t *)(uintptr_t)ctx_int;
+out:
+    return ret;
 }
 
+int
+dht_inode_ctx_set(inode_t *inode, xlator_t *this, dht_inode_ctx_t *ctx)
+{
+    int ret = -1;
+    uint64_t ctx_int = 0;
+
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+    GF_VALIDATE_OR_GOTO(this->name, ctx, out);
+
+    ctx_int = (long)ctx;
+    ret = inode_ctx_set(inode, this, &ctx_int);
+out:
+    return ret;
+}
 
 int
-dht_build_child_loc (xlator_t *this, loc_t *child, loc_t *parent, char *name)
+dht_subvol_status(dht_conf_t *conf, xlator_t *subvol)
+{
+    int i;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (conf->subvolumes[i] == subvol) {
+            return conf->subvolume_status[i];
+        }
+    }
+    return 0;
+}
+
+inode_t *
+dht_heal_path(xlator_t *this, char *path, inode_table_t *itable)
 {
-        if (!child) {
-                goto err;
+    int ret = -1;
+    struct iatt iatt = {
+        0,
+    };
+    inode_t *linked_inode = NULL;
+    loc_t loc = {
+        0,
+    };
+    char *bname = NULL;
+    char *save_ptr = NULL;
+    static uuid_t gfid = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1};
+    char *tmp_path = NULL;
+
+    tmp_path = gf_strdup(path);
+    if (!tmp_path) {
+        goto out;
+    }
+
+    gf_uuid_copy(loc.pargfid, gfid);
+    loc.parent = inode_ref(itable->root);
+
+    bname = strtok_r(tmp_path, "/", &save_ptr);
+
+    /* sending a lookup on parent directory,
+     * Eg:  if  path is like /a/b/c/d/e/f/g/
+     * then we will send a lookup on a first and then b,c,d,etc
+     */
+
+    while (bname) {
+        linked_inode = NULL;
+        loc.inode = inode_grep(itable, loc.parent, bname);
+        if (loc.inode == NULL) {
+            loc.inode = inode_new(itable);
+            if (loc.inode == NULL) {
+                ret = -ENOMEM;
+                goto out;
+            }
+        } else {
+            /*
+             * Inode is already populated in the inode table.
+             * Which means we already looked up the inode and
+             * linked with a dentry. So that we will skip
+             * lookup on this entry, and proceed to next.
+             */
+            linked_inode = loc.inode;
+            bname = strtok_r(NULL, "/", &save_ptr);
+            if (!bname) {
+                goto out;
+            }
+            inode_unref(loc.parent);
+            loc.parent = loc.inode;
+            gf_uuid_copy(loc.pargfid, loc.inode->gfid);
+            loc.inode = NULL;
+            continue;
         }
 
-        if (strcmp (parent->path, "/") == 0)
-                gf_asprintf ((char **)&child->path, "/%s", name);
-        else
-                gf_asprintf ((char **)&child->path, "%s/%s", parent->path, name);
+        loc.name = bname;
+        ret = loc_path(&loc, bname);
 
-        if (!child->path) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                goto err;
+        ret = syncop_lookup(this, &loc, &iatt, NULL, NULL, NULL);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_INFO, -ret, DHT_MSG_DIR_SELFHEAL_FAILED,
+                    "path=%s", path, "subvolume=%s", this->name, "bname=%s",
+                    bname, NULL);
+            goto out;
         }
 
-        child->name = strrchr (child->path, '/');
-        if (child->name)
-                child->name++;
+        linked_inode = inode_link(loc.inode, loc.parent, bname, &iatt);
+        if (!linked_inode)
+            goto out;
+
+        loc_wipe(&loc);
+        gf_uuid_copy(loc.pargfid, linked_inode->gfid);
+        loc.inode = NULL;
+
+        bname = strtok_r(NULL, "/", &save_ptr);
+        if (bname)
+            loc.parent = linked_inode;
+    }
+out:
+    inode_ref(linked_inode);
+    loc_wipe(&loc);
+    GF_FREE(tmp_path);
 
-        child->parent = inode_ref (parent->inode);
-        child->inode = inode_new (parent->inode->table);
+    return linked_inode;
+}
 
-        if (!child->inode) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                goto err;
+int
+dht_heal_full_path(void *data)
+{
+    call_frame_t *heal_frame = data;
+    dht_local_t *local = NULL;
+    loc_t loc = {
+        0,
+    };
+    dict_t *dict = NULL;
+    char *path = NULL;
+    int ret = -1;
+    xlator_t *source = NULL;
+    xlator_t *this = NULL;
+    inode_table_t *itable = NULL;
+    inode_t *inode = NULL;
+    inode_t *tmp_inode = NULL;
+
+    GF_VALIDATE_OR_GOTO("DHT", heal_frame, out);
+
+    local = heal_frame->local;
+    this = heal_frame->this;
+    source = heal_frame->cookie;
+    heal_frame->cookie = NULL;
+    gf_uuid_copy(loc.gfid, local->gfid);
+
+    if (local->loc.inode)
+        loc.inode = inode_ref(local->loc.inode);
+    else
+        goto out;
+
+    itable = loc.inode->table;
+    ret = syncop_getxattr(source, &loc, &dict, GET_ANCESTRY_PATH_KEY, NULL,
+                          NULL);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_INFO, -ret, DHT_MSG_DIR_HEAL_ABORT,
+                "subvol=%s", source->name, NULL);
+        goto out;
+    }
+
+    ret = dict_get_str(dict, GET_ANCESTRY_PATH_KEY, &path);
+    if (path) {
+        inode = dht_heal_path(this, path, itable);
+        if (inode && inode != local->inode) {
+            /*
+             * if inode returned by heal function is different
+             * from what we passed, which means a racing thread
+             * already linked a different inode for dentry.
+             * So we will update our local->inode, so that we can
+             * retrurn proper inode.
+             */
+            tmp_inode = local->inode;
+            local->inode = inode;
+            inode_unref(tmp_inode);
+            tmp_inode = NULL;
+        } else {
+            inode_unref(inode);
         }
+    }
 
-        return 0;
-err:
-        loc_wipe (child);
+out:
+    loc_wipe(&loc);
+    if (dict)
+        dict_unref(dict);
+    return 0;
+}
+
+int
+dht_heal_full_path_done(int op_ret, call_frame_t *heal_frame, void *data)
+{
+    call_frame_t *main_frame = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *this = NULL;
+    int ret = -1;
+    int op_errno = 0;
+
+    local = heal_frame->local;
+    main_frame = local->main_frame;
+    local->main_frame = NULL;
+    this = heal_frame->this;
+
+    dht_set_fixed_dir_stat(&local->postparent);
+    if (local->need_xattr_heal) {
+        local->need_xattr_heal = 0;
+        ret = dht_dir_xattr_heal(this, local, &op_errno);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                    DHT_MSG_DIR_XATTR_HEAL_FAILED, "path=%s", local->loc.path,
+                    NULL);
+        }
+    }
+
+    DHT_STACK_UNWIND(lookup, main_frame, 0, 0, local->inode, &local->stbuf,
+                     local->xattr, &local->postparent);
+
+    DHT_STACK_DESTROY(heal_frame);
+    return 0;
+}
+
+/* This function must be called inside an inode lock */
+int
+__dht_lock_subvol_set(inode_t *inode, xlator_t *this, xlator_t *lock_subvol)
+{
+    dht_inode_ctx_t *ctx = NULL;
+    int ret = -1;
+    uint64_t value = 0;
+
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+
+    ret = __inode_ctx_get0(inode, this, &value);
+    if (ret || !value) {
         return -1;
+    }
+
+    ctx = (dht_inode_ctx_t *)(uintptr_t)value;
+    ctx->lock_subvol = lock_subvol;
+out:
+    return ret;
+}
+
+xlator_t *
+dht_get_lock_subvolume(xlator_t *this, struct gf_flock *lock,
+                       dht_local_t *local)
+{
+    xlator_t *subvol = NULL;
+    inode_t *inode = NULL;
+    int32_t ret = -1;
+    uint64_t value = 0;
+    xlator_t *cached_subvol = NULL;
+    dht_inode_ctx_t *ctx = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    GF_VALIDATE_OR_GOTO(this->name, lock, out);
+    GF_VALIDATE_OR_GOTO(this->name, local, out);
+
+    cached_subvol = local->cached_subvol;
+
+    if (local->loc.inode || local->fd) {
+        inode = local->loc.inode ? local->loc.inode : local->fd->inode;
+    }
+
+    if (!inode)
+        goto out;
+
+    if (!(IA_ISDIR(inode->ia_type) || IA_ISINVAL(inode->ia_type))) {
+        /*
+         * We may get non-linked inode for directories as part
+         * of the selfheal code path. So checking  for IA_INVAL
+         * type also. This will only happen for directory.
+         */
+        subvol = local->cached_subvol;
+        goto out;
+    }
+
+    if (lock->l_type != F_UNLCK) {
+        /*
+         * inode purging might happen on NFS between a lk
+         * and unlk. Due to this lk and unlk might be sent
+         * to different subvols.
+         * So during a lock request, taking a ref on inode
+         * to prevent inode purging. inode unref will happen
+         * in unlock cbk code path.
+         */
+        inode_ref(inode);
+    }
+
+    LOCK(&inode->lock);
+    ret = __inode_ctx_get0(inode, this, &value);
+    if (!ret && value) {
+        ctx = (dht_inode_ctx_t *)(uintptr_t)value;
+        subvol = ctx->lock_subvol;
+    }
+    if (!subvol && lock->l_type != F_UNLCK && cached_subvol) {
+        ret = __dht_lock_subvol_set(inode, this, cached_subvol);
+        if (ret) {
+            gf_uuid_unparse(inode->gfid, gfid);
+            UNLOCK(&inode->lock);
+            gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_SET_INODE_CTX_FAILED,
+                    "lock_subvol gfid=%s", gfid, NULL);
+            goto post_unlock;
+        }
+        subvol = cached_subvol;
+    }
+    UNLOCK(&inode->lock);
+post_unlock:
+    if (!subvol && inode && lock->l_type != F_UNLCK) {
+        inode_unref(inode);
+    }
+out:
+    return subvol;
+}
+
+int
+dht_lk_inode_unref(call_frame_t *frame, int32_t op_ret)
+{
+    int ret = -1;
+    dht_local_t *local = NULL;
+    inode_t *inode = NULL;
+    xlator_t *this = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    this = frame->this;
+
+    if (local->loc.inode || local->fd) {
+        inode = local->loc.inode ? local->loc.inode : local->fd->inode;
+    }
+    if (!inode) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LOCK_INODE_UNREF_FAILED,
+                NULL);
+        goto out;
+    }
+
+    if (!(IA_ISDIR(inode->ia_type) || IA_ISINVAL(inode->ia_type))) {
+        ret = 0;
+        goto out;
+    }
+
+    switch (local->lock_type) {
+        case F_RDLCK:
+        case F_WRLCK:
+            if (op_ret) {
+                gf_uuid_unparse(inode->gfid, gfid);
+                gf_msg_debug(this->name, 0, "lock request failed for gfid %s",
+                             gfid);
+                inode_unref(inode);
+                goto out;
+            }
+            break;
+
+        case F_UNLCK:
+            if (!op_ret) {
+                inode_unref(inode);
+            } else {
+                gf_uuid_unparse(inode->gfid, gfid);
+                gf_smsg(this->name, GF_LOG_WARNING, 0,
+                        DHT_MSG_LOCK_INODE_UNREF_FAILED, "gfid=%s", gfid, NULL);
+                goto out;
+            }
+        default:
+            break;
+    }
+    ret = 0;
+out:
+    return ret;
+}
+
+/* Code to update custom extended attributes from src dict to dst dict
+ */
+void
+dht_dir_set_heal_xattr(xlator_t *this, dht_local_t *local, dict_t *dst,
+                       dict_t *src, int *uret, int *uflag)
+{
+    int ret = -1;
+    data_t *keyval = NULL;
+    int luret = -1;
+    int luflag = -1;
+    int i = 0;
+    char **xattrs_to_heal;
+
+    if (!src || !dst) {
+        gf_smsg(this->name, GF_LOG_WARNING, EINVAL, DHT_MSG_DST_NULL_SET_FAILED,
+                "path=%s", local->loc.path, NULL);
+        return;
+    }
+    /* Check if any user xattr present in src dict and set
+       it to dst dict
+    */
+    luret = dict_foreach_fnmatch(src, "user.*", dht_set_user_xattr, dst);
+    /* Check if any other custom xattr present in src dict
+       and set it to dst dict, here index start from 1 because
+       user xattr already checked in previous statement
+    */
+
+    xattrs_to_heal = get_xattrs_to_heal();
+
+    for (i = 1; xattrs_to_heal[i]; i++) {
+        keyval = dict_get(src, xattrs_to_heal[i]);
+        if (keyval) {
+            luflag = 1;
+            ret = dict_set(dst, xattrs_to_heal[i], keyval);
+            if (ret)
+                gf_smsg(this->name, GF_LOG_WARNING, ENOMEM,
+                        DHT_MSG_DICT_SET_FAILED, "key=%s", xattrs_to_heal[i],
+                        "path=%s", local->loc.path, NULL);
+            keyval = NULL;
+        }
+    }
+    if (uret)
+        (*uret) = luret;
+    if (uflag)
+        (*uflag) = luflag;
 }
diff --git a/xlators/cluster/dht/src/dht-inode-read.c b/xlators/cluster/dht/src/dht-inode-read.c
new file mode 100644
index 00000000000..dbb8070b0da
--- /dev/null
+++ b/xlators/cluster/dht/src/dht-inode-read.c
@@ -0,0 +1,1658 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "dht-common.h"
+
+static int
+dht_access2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret);
+static int
+dht_readv2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret);
+static int
+dht_attr2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret);
+static int
+dht_open2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret);
+static int
+dht_flush2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret);
+static int
+dht_lk2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret);
+static int
+dht_fsync2(xlator_t *this, xlator_t *dst_node, call_frame_t *frame, int ret);
+static int
+dht_common_xattrop2(xlator_t *this, xlator_t *subvol, call_frame_t *frame,
+                    int ret);
+
+static int
+dht_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+             int op_errno, fd_t *fd, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    int ret = 0;
+
+    local = frame->local;
+    prev = cookie;
+
+    local->op_errno = op_errno;
+    if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->name);
+        goto out;
+    }
+
+    /* Update ctx if the fd has been opened on the target*/
+    if (!op_ret && (local->call_cnt == 1)) {
+        dht_fd_ctx_set(this, fd, prev);
+        goto out;
+    }
+
+    if (!op_ret || (local->call_cnt != 1))
+        goto out;
+
+    /* rebalance would have happened */
+    local->rebalance.target_op_fn = dht_open2;
+    ret = dht_rebalance_complete_check(this, frame);
+    if (!ret)
+        return 0;
+
+out:
+    DHT_STACK_UNWIND(open, frame, op_ret, op_errno, local->fd, xdata);
+
+    return 0;
+}
+
+static int
+dht_open2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
+{
+    dht_local_t *local = NULL;
+    int op_errno = EINVAL;
+
+    if (!frame || !frame->local)
+        goto out;
+
+    local = frame->local;
+    op_errno = local->op_errno;
+
+    if (we_are_not_migrating(ret)) {
+        /* This DHT layer is not migrating the file */
+        DHT_STACK_UNWIND(open, frame, -1, local->op_errno, NULL,
+                         local->rebalance.xdata);
+        return 0;
+    }
+
+    if (subvol == NULL)
+        goto out;
+
+    local->call_cnt = 2;
+
+    STACK_WIND_COOKIE(frame, dht_open_cbk, subvol, subvol, subvol->fops->open,
+                      &local->loc, local->rebalance.flags, local->fd,
+                      local->xattr_req);
+    return 0;
+
+out:
+    DHT_STACK_UNWIND(open, frame, -1, op_errno, NULL, NULL);
+    return 0;
+}
+
+int
+dht_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags, fd_t *fd,
+         dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    local = dht_local_init(frame, loc, fd, GF_FOP_OPEN);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    local->rebalance.flags = flags;
+    local->call_cnt = 1;
+
+    STACK_WIND_COOKIE(frame, dht_open_cbk, subvol, subvol, subvol->fops->open,
+                      loc, flags, fd, xdata);
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(open, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
+
+int
+dht_file_attr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                  int op_errno, struct iatt *stbuf, dict_t *xdata)
+{
+    xlator_t *subvol1 = 0;
+    xlator_t *subvol2 = 0;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    int ret = -1;
+    inode_t *inode = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", cookie, out);
+
+    local = frame->local;
+    prev = cookie;
+
+    if ((local->fop == GF_FOP_FSTAT) &&
+        dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
+
+    if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+        local->op_errno = op_errno;
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->name);
+        goto out;
+    }
+
+    if (local->call_cnt != 1)
+        goto out;
+
+    local->op_errno = op_errno;
+    local->op_ret = op_ret;
+
+    /* Check if the rebalance phase2 is true */
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(stbuf)) {
+        local->rebalance.target_op_fn = dht_attr2;
+        dht_set_local_rebalance(this, local, NULL, NULL, stbuf, xdata);
+        inode = (local->fd) ? local->fd->inode : local->loc.inode;
+
+        dht_inode_ctx_get_mig_info(this, inode, &subvol1, &subvol2);
+        if (dht_mig_info_is_invalid(local->cached_subvol, subvol1, subvol2)) {
+            /* Phase 2 of migration */
+            ret = dht_rebalance_complete_check(this, frame);
+            if (!ret)
+                return 0;
+        } else {
+            /* it is a non-fd op or it is an fd based Fop and
+               opened on the dst.*/
+            if (local->fd && !dht_fd_open_on_dst(this, local->fd, subvol2)) {
+                ret = dht_rebalance_complete_check(this, frame);
+                if (!ret)
+                    return 0;
+            } else {
+                dht_attr2(this, subvol2, frame, 0);
+                return 0;
+            }
+        }
+    }
+
+out:
+    DHT_STRIP_PHASE1_FLAGS(stbuf);
+    DHT_STACK_UNWIND(stat, frame, op_ret, op_errno, stbuf, xdata);
+err:
+    return 0;
+}
+
+static int
+dht_attr2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
+{
+    dht_local_t *local = NULL;
+    int op_errno = EINVAL;
+
+    local = frame->local;
+    if (!local)
+        goto out;
+
+    op_errno = local->op_errno;
+
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        DHT_STACK_UNWIND(stat, frame, local->op_ret, op_errno,
+                         &local->rebalance.postbuf, local->rebalance.xdata);
+        return 0;
+    }
+
+    if (subvol == NULL)
+        goto out;
+
+    local->call_cnt = 2;
+
+    if (local->fop == GF_FOP_FSTAT) {
+        STACK_WIND_COOKIE(frame, dht_file_attr_cbk, subvol, subvol,
+                          subvol->fops->fstat, local->fd, local->xattr_req);
+    } else {
+        STACK_WIND_COOKIE(frame, dht_file_attr_cbk, subvol, subvol,
+                          subvol->fops->stat, &local->loc, local->xattr_req);
+    }
+
+    return 0;
+
+out:
+    DHT_STACK_UNWIND(stat, frame, -1, op_errno, NULL, NULL);
+    return 0;
+}
+
+static int
+dht_attr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+             int op_errno, struct iatt *stbuf, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+    local = frame->local;
+    prev = cookie;
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+            UNLOCK(&frame->lock);
+            gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                         prev->name);
+
+            goto post_unlock;
+        }
+
+        dht_iatt_merge(this, &local->stbuf, stbuf);
+
+        local->op_ret = 0;
+    }
+    UNLOCK(&frame->lock);
+post_unlock:
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        DHT_STACK_UNWIND(stat, frame, local->op_ret, local->op_errno,
+                         &local->stbuf, xdata);
+    }
+
+    return 0;
+}
+
+int
+dht_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+    int i = 0;
+    int call_cnt = 0;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+    VALIDATE_OR_GOTO(loc->path, err);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_STAT);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!layout) {
+        gf_msg_debug(this->name, 0, "no layout for path=%s", loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    if (IA_ISREG(loc->inode->ia_type)) {
+        local->call_cnt = 1;
+
+        subvol = local->cached_subvol;
+
+        STACK_WIND_COOKIE(frame, dht_file_attr_cbk, subvol, subvol,
+                          subvol->fops->stat, loc, xdata);
+
+        return 0;
+    }
+
+    local->call_cnt = call_cnt = layout->cnt;
+
+    for (i = 0; i < call_cnt; i++) {
+        subvol = layout->list[i].xlator;
+
+        STACK_WIND_COOKIE(frame, dht_attr_cbk, subvol, subvol,
+                          subvol->fops->stat, loc, xdata);
+    }
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(stat, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
+
+int
+dht_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+    int i = 0;
+    int call_cnt = 0;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_FSTAT);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!layout) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "no layout for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    if (IA_ISREG(fd->inode->ia_type)) {
+        local->call_cnt = 1;
+
+        subvol = local->cached_subvol;
+
+        STACK_WIND_COOKIE(frame, dht_file_attr_cbk, subvol, subvol,
+                          subvol->fops->fstat, fd, xdata);
+        return 0;
+    }
+
+    local->call_cnt = call_cnt = layout->cnt;
+
+    for (i = 0; i < call_cnt; i++) {
+        subvol = layout->list[i].xlator;
+        STACK_WIND_COOKIE(frame, dht_attr_cbk, subvol, subvol,
+                          subvol->fops->fstat, fd, xdata);
+    }
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(fstat, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
+
+int
+dht_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, struct iovec *vector, int count, struct iatt *stbuf,
+              struct iobref *iobref, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int ret = 0;
+    xlator_t *src_subvol = 0;
+    xlator_t *dst_subvol = 0;
+
+    local = frame->local;
+    if (!local) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    /* This is already second try, no need for re-check */
+    if (local->call_cnt != 1)
+        goto out;
+
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
+
+    if ((op_ret == -1) && !dht_inode_missing(op_errno))
+        goto out;
+
+    local->op_errno = op_errno;
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(stbuf)) {
+        local->op_ret = op_ret;
+        local->rebalance.target_op_fn = dht_readv2;
+        dht_set_local_rebalance(this, local, NULL, NULL, stbuf, xdata);
+        /* File would be migrated to other node */
+        ret = dht_inode_ctx_get_mig_info(this, local->fd->inode, &src_subvol,
+                                         &dst_subvol);
+
+        if (dht_mig_info_is_invalid(local->cached_subvol, src_subvol,
+                                    dst_subvol) ||
+            !dht_fd_open_on_dst(this, local->fd, dst_subvol)) {
+            ret = dht_rebalance_complete_check(this, frame);
+            if (!ret)
+                return 0;
+        } else {
+            /* value is already set in fd_ctx, that means no need
+               to check for whether its complete or not. */
+            dht_readv2(this, dst_subvol, frame, 0);
+            return 0;
+        }
+    }
+
+out:
+    DHT_STRIP_PHASE1_FLAGS(stbuf);
+
+    DHT_STACK_UNWIND(readv, frame, op_ret, op_errno, vector, count, stbuf,
+                     iobref, xdata);
+
+    return 0;
+}
+
+static int
+dht_readv2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
+{
+    dht_local_t *local = NULL;
+    int op_errno = EINVAL;
+
+    local = frame->local;
+    if (!local)
+        goto out;
+
+    op_errno = local->op_errno;
+
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        DHT_STACK_UNWIND(readv, frame, local->op_ret, op_errno, NULL, 0,
+                         &local->rebalance.postbuf, NULL,
+                         local->rebalance.xdata);
+        return 0;
+    }
+
+    if (subvol == NULL)
+        goto out;
+
+    local->call_cnt = 2;
+
+    STACK_WIND(frame, dht_readv_cbk, subvol, subvol->fops->readv, local->fd,
+               local->rebalance.size, local->rebalance.offset,
+               local->rebalance.flags, local->xattr_req);
+
+    return 0;
+
+out:
+    DHT_STACK_UNWIND(readv, frame, -1, op_errno, NULL, 0, NULL, NULL, NULL);
+    return 0;
+}
+
+int
+dht_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size, off_t off,
+          uint32_t flags, dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_READ);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    local->rebalance.offset = off;
+    local->rebalance.size = size;
+    local->rebalance.flags = flags;
+    local->call_cnt = 1;
+
+    STACK_WIND(frame, dht_readv_cbk, subvol, subvol->fops->readv, local->fd,
+               local->rebalance.size, local->rebalance.offset,
+               local->rebalance.flags, local->xattr_req);
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(readv, frame, -1, op_errno, NULL, 0, NULL, NULL, NULL);
+
+    return 0;
+}
+
+static int
+dht_access_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, dict_t *xdata)
+{
+    int ret = -1;
+    dht_local_t *local = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *prev = NULL;
+
+    local = frame->local;
+    prev = cookie;
+
+    if (!prev)
+        goto out;
+    if (local->call_cnt != 1)
+        goto out;
+    if ((op_ret == -1) &&
+        ((op_errno == ENOTCONN) || dht_inode_missing(op_errno)) &&
+        IA_ISDIR(local->loc.inode->ia_type)) {
+        subvol = dht_subvol_next_available(this, prev);
+        if (!subvol)
+            goto out;
+
+        /* check if we are done with visiting every node */
+        if (subvol == local->cached_subvol) {
+            goto out;
+        }
+
+        STACK_WIND_COOKIE(frame, dht_access_cbk, subvol, subvol,
+                          subvol->fops->access, &local->loc,
+                          local->rebalance.flags, NULL);
+        return 0;
+    }
+    if ((op_ret == -1) && dht_inode_missing(op_errno) &&
+        !(IA_ISDIR(local->loc.inode->ia_type))) {
+        /* File would be migrated to other node */
+        local->op_errno = op_errno;
+        local->rebalance.target_op_fn = dht_access2;
+        ret = dht_rebalance_complete_check(frame->this, frame);
+        if (!ret)
+            return 0;
+    }
+
+out:
+    DHT_STACK_UNWIND(access, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+static int
+dht_access2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
+{
+    dht_local_t *local = NULL;
+    int op_errno = EINVAL;
+
+    local = frame->local;
+    if (!local)
+        goto out;
+
+    op_errno = local->op_errno;
+
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+
+        DHT_STACK_UNWIND(access, frame, -1, op_errno, NULL);
+        return 0;
+    }
+
+    if (subvol == NULL)
+        goto out;
+
+    local->call_cnt = 2;
+
+    STACK_WIND_COOKIE(frame, dht_access_cbk, subvol, subvol,
+                      subvol->fops->access, &local->loc, local->rebalance.flags,
+                      local->xattr_req);
+
+    return 0;
+
+out:
+    DHT_STACK_UNWIND(access, frame, -1, op_errno, NULL);
+    return 0;
+}
+
+int
+dht_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+           dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+    VALIDATE_OR_GOTO(loc->path, err);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_ACCESS);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->rebalance.flags = mask;
+    local->call_cnt = 1;
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for path=%s",
+                     loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    STACK_WIND_COOKIE(frame, dht_access_cbk, subvol, subvol,
+                      subvol->fops->access, loc, mask, xdata);
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(access, frame, -1, op_errno, NULL);
+
+    return 0;
+}
+
+int
+dht_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *subvol = 0;
+    int ret = 0;
+
+    local = frame->local;
+
+    local->op_errno = op_errno;
+
+    if (local->call_cnt != 1)
+        goto out;
+
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
+
+    local->rebalance.target_op_fn = dht_flush2;
+
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
+
+    /* If context is set, then send flush() it to the destination */
+    dht_inode_ctx_get_mig_info(this, local->fd->inode, NULL, &subvol);
+    if (subvol && dht_fd_open_on_dst(this, local->fd, subvol)) {
+        dht_flush2(this, subvol, frame, 0);
+        return 0;
+    }
+
+    if (op_errno == EREMOTE) {
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret) {
+            return 0;
+        }
+    }
+
+out:
+    DHT_STACK_UNWIND(flush, frame, op_ret, op_errno, xdata);
+
+    return 0;
+}
+
+static int
+dht_flush2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
+{
+    dht_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+
+    if ((frame == NULL) || (frame->local == NULL))
+        goto out;
+
+    local = frame->local;
+
+    op_errno = local->op_errno;
+
+    if (subvol == NULL)
+        goto out;
+
+    local->call_cnt = 2; /* This is the second attempt */
+
+    STACK_WIND(frame, dht_flush_cbk, subvol, subvol->fops->flush, local->fd,
+               local->xattr_req);
+
+    return 0;
+
+out:
+    DHT_STACK_UNWIND(flush, frame, -1, op_errno, NULL);
+    return 0;
+}
+
+int
+dht_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_FLUSH);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    local->call_cnt = 1;
+
+    STACK_WIND(frame, dht_flush_cbk, subvol, subvol->fops->flush, fd,
+               local->xattr_req);
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(flush, frame, -1, op_errno, NULL);
+
+    return 0;
+}
+
+int
+dht_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+              dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    int ret = -1;
+    inode_t *inode = NULL;
+    xlator_t *src_subvol = 0;
+    xlator_t *dst_subvol = 0;
+
+    local = frame->local;
+    prev = cookie;
+
+    local->op_errno = op_errno;
+
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
+
+    if (op_ret == -1 && !dht_inode_missing(op_errno)) {
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->name);
+        goto out;
+    }
+
+    if (local->call_cnt != 1) {
+        if (local->stbuf.ia_blocks) {
+            dht_iatt_merge(this, postbuf, &local->stbuf);
+            dht_iatt_merge(this, prebuf, &local->prebuf);
+        }
+        goto out;
+    }
+
+    local->op_ret = op_ret;
+    inode = local->fd->inode;
+
+    local->rebalance.target_op_fn = dht_fsync2;
+    dht_set_local_rebalance(this, local, NULL, prebuf, postbuf, xdata);
+
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(postbuf)) {
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+    /* Check if the rebalance phase1 is true */
+    if (IS_DHT_MIGRATION_PHASE1(postbuf)) {
+        dht_iatt_merge(this, &local->stbuf, postbuf);
+        dht_iatt_merge(this, &local->prebuf, prebuf);
+
+        dht_inode_ctx_get_mig_info(this, inode, &src_subvol, &dst_subvol);
+
+        if (dht_mig_info_is_invalid(local->cached_subvol, src_subvol,
+                                    dst_subvol) ||
+            !dht_fd_open_on_dst(this, local->fd, dst_subvol)) {
+            ret = dht_rebalance_in_progress_check(this, frame);
+            if (!ret)
+                return 0;
+        } else {
+            dht_fsync2(this, dst_subvol, frame, 0);
+            return 0;
+        }
+    }
+
+out:
+    DHT_STRIP_PHASE1_FLAGS(postbuf);
+    DHT_STRIP_PHASE1_FLAGS(prebuf);
+
+    DHT_STACK_UNWIND(fsync, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+
+    return 0;
+}
+
+static int
+dht_fsync2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
+{
+    dht_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+
+    if ((frame == NULL) || (frame->local == NULL))
+        goto out;
+
+    local = frame->local;
+    op_errno = local->op_errno;
+
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        DHT_STACK_UNWIND(fsync, frame, local->op_ret, op_errno,
+                         &local->rebalance.prebuf, &local->rebalance.postbuf,
+                         local->rebalance.xdata);
+        return 0;
+    }
+
+    if (subvol == NULL)
+        goto out;
+
+    local->call_cnt = 2; /* This is the second attempt */
+
+    STACK_WIND_COOKIE(frame, dht_fsync_cbk, subvol, subvol, subvol->fops->fsync,
+                      local->fd, local->rebalance.flags, local->xattr_req);
+
+    return 0;
+
+out:
+    DHT_STACK_UNWIND(fsync, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
+}
+
+int
+dht_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int datasync,
+          dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_FSYNC);
+    if (!local) {
+        op_errno = ENOMEM;
+
+        goto err;
+    }
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    local->call_cnt = 1;
+    local->rebalance.flags = datasync;
+
+    subvol = local->cached_subvol;
+
+    STACK_WIND_COOKIE(frame, dht_fsync_cbk, subvol, subvol, subvol->fops->fsync,
+                      local->fd, local->rebalance.flags, local->xattr_req);
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(fsync, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+/* TODO: for 'lk()' call, we need some other special error, may be ESTALE to
+   indicate that lock migration happened on the fd, so we can consider it as
+   phase 2 of migration */
+static int
+dht_lk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+           int op_errno, struct gf_flock *flock, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int ret = -1;
+    xlator_t *subvol = NULL;
+
+    local = frame->local;
+
+    if (!local) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    if (local->call_cnt != 1)
+        goto out;
+
+    local->rebalance.target_op_fn = dht_lk2;
+
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
+
+    if (xdata)
+        local->rebalance.xdata = dict_ref(xdata);
+
+    if (op_errno == EREMOTE) {
+        dht_inode_ctx_get_mig_info(this, local->fd->inode, NULL, &subvol);
+        if (subvol && dht_fd_open_on_dst(this, local->fd, subvol)) {
+            dht_lk2(this, subvol, frame, 0);
+            return 0;
+        } else {
+            ret = dht_rebalance_complete_check(this, frame);
+            if (!ret) {
+                return 0;
+            }
+        }
+    }
+
+out:
+    dht_lk_inode_unref(frame, op_ret);
+    DHT_STACK_UNWIND(lk, frame, op_ret, op_errno, flock, xdata);
+
+    return 0;
+}
+
+static int
+dht_lk2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
+{
+    dht_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+
+    if ((frame == NULL) || (frame->local == NULL))
+        goto out;
+
+    local = frame->local;
+
+    op_errno = local->op_errno;
+
+    if (subvol == NULL)
+        goto out;
+
+    local->call_cnt = 2; /* This is the second attempt */
+
+    STACK_WIND(frame, dht_lk_cbk, subvol, subvol->fops->lk, local->fd,
+               local->rebalance.lock_cmd, &local->rebalance.flock,
+               local->xattr_req);
+
+    return 0;
+
+out:
+    DHT_STACK_UNWIND(lk, frame, -1, op_errno, NULL, NULL);
+    return 0;
+}
+
+int
+dht_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int cmd,
+       struct gf_flock *flock, dict_t *xdata)
+{
+    xlator_t *lock_subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_LK);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->lock_type = flock->l_type;
+    lock_subvol = dht_get_lock_subvolume(this, flock, local);
+    if (!lock_subvol) {
+        gf_msg_debug(this->name, 0, "no lock subvolume for path=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    /*
+            local->cached_subvol = lock_subvol;
+            ret = dht_check_and_open_fd_on_subvol (this, frame);
+            if (ret)
+                    goto err;
+    */
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    local->rebalance.flock = *flock;
+    local->rebalance.lock_cmd = cmd;
+
+    local->call_cnt = 1;
+
+    STACK_WIND(frame, dht_lk_cbk, lock_subvol, lock_subvol->fops->lk, fd, cmd,
+               flock, xdata);
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(lk, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
+
+static int
+dht_lease_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, struct gf_lease *lease, dict_t *xdata)
+{
+    DHT_STACK_UNWIND(lease, frame, op_ret, op_errno, lease, xdata);
+
+    return 0;
+}
+
+int
+dht_lease(call_frame_t *frame, xlator_t *this, loc_t *loc,
+          struct gf_lease *lease, dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+
+    subvol = dht_subvol_get_cached(this, loc->inode);
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for path=%s",
+                     loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    /* TODO: for rebalance, we need to preserve the fop arguments */
+    STACK_WIND(frame, dht_lease_cbk, subvol, subvol->fops->lease, loc, lease,
+               xdata);
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(lease, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
+
+/* Symlinks are currently not migrated, so no need for any check here */
+static int
+dht_readlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, const char *path, struct iatt *stbuf,
+                 dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+
+    local = frame->local;
+    if (op_ret == -1)
+        goto err;
+
+    if (!local) {
+        op_ret = -1;
+        op_errno = EINVAL;
+    }
+
+err:
+    DHT_STRIP_PHASE1_FLAGS(stbuf);
+    DHT_STACK_UNWIND(readlink, frame, op_ret, op_errno, path, stbuf, xdata);
+
+    return 0;
+}
+
+int
+dht_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+             dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+    VALIDATE_OR_GOTO(loc->path, err);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_READLINK);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for path=%s",
+                     loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    STACK_WIND(frame, dht_readlink_cbk, subvol, subvol->fops->readlink, loc,
+               size, xdata);
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(readlink, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+/* Get both DHT_IATT_IN_XDATA_KEY and DHT_MODE_IN_XDATA_KEY
+ * Use DHT_MODE_IN_XDATA_KEY if available, else fall back to
+ * DHT_IATT_IN_XDATA_KEY
+ * This will return a dummy iatt with only the mode and type set
+ */
+static int
+dht_read_iatt_from_xdata(dict_t *xdata, struct iatt *stbuf)
+{
+    int ret = -1;
+    int32_t mode = 0;
+
+    ret = dict_get_int32(xdata, DHT_MODE_IN_XDATA_KEY, &mode);
+
+    if (ret) {
+        ret = dict_get_bin(xdata, DHT_IATT_IN_XDATA_KEY, (void **)&stbuf);
+    } else {
+        stbuf->ia_prot = ia_prot_from_st_mode(mode);
+        stbuf->ia_type = ia_type_from_st_mode(mode);
+    }
+
+    return ret;
+}
+
+int
+dht_common_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *dict,
+                       dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    call_frame_t *call_frame = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *src_subvol = NULL;
+    xlator_t *dst_subvol = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+    int ret = -1;
+    inode_t *inode = NULL;
+
+    local = frame->local;
+    call_frame = cookie;
+    prev = call_frame->this;
+
+    local->op_errno = op_errno;
+
+    if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1.",
+                     prev->name);
+        goto out;
+    }
+
+    if (local->call_cnt != 1)
+        goto out;
+
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
+
+    ret = dht_read_iatt_from_xdata(xdata, &stbuf);
+
+    if ((!op_ret) && (ret)) {
+        /* This is a potential problem and can cause corruption
+         * with sharding.
+         * Oh well. We tried.
+         */
+        goto out;
+    }
+
+    local->op_ret = op_ret;
+    local->rebalance.target_op_fn = dht_common_xattrop2;
+    if (xdata)
+        local->rebalance.xdata = dict_ref(xdata);
+
+    if (dict)
+        local->rebalance.dict = dict_ref(dict);
+
+    /* Phase 2 of migration */
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(&stbuf)) {
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+    /* Check if the rebalance phase1 is true */
+    if (IS_DHT_MIGRATION_PHASE1(&stbuf)) {
+        inode = local->loc.inode ? local->loc.inode : local->fd->inode;
+        dht_inode_ctx_get_mig_info(this, inode, &src_subvol, &dst_subvol);
+
+        if (dht_mig_info_is_invalid(local->cached_subvol, src_subvol,
+                                    dst_subvol) ||
+            !dht_fd_open_on_dst(this, local->fd, dst_subvol)) {
+            ret = dht_rebalance_in_progress_check(this, frame);
+            if (!ret)
+                return 0;
+        } else {
+            dht_common_xattrop2(this, dst_subvol, frame, 0);
+            return 0;
+        }
+    }
+
+out:
+    if (local->fop == GF_FOP_XATTROP) {
+        DHT_STACK_UNWIND(xattrop, frame, op_ret, op_errno, dict, xdata);
+    } else {
+        DHT_STACK_UNWIND(fxattrop, frame, op_ret, op_errno, dict, xdata);
+    }
+
+    return 0;
+}
+
+static int
+dht_common_xattrop2(xlator_t *this, xlator_t *subvol, call_frame_t *frame,
+                    int ret)
+{
+    dht_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+
+    if ((frame == NULL) || (frame->local == NULL))
+        goto out;
+
+    local = frame->local;
+    op_errno = local->op_errno;
+
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        if (local->fop == GF_FOP_XATTROP) {
+            DHT_STACK_UNWIND(xattrop, frame, local->op_ret, op_errno,
+                             local->rebalance.dict, local->rebalance.xdata);
+        } else {
+            DHT_STACK_UNWIND(fxattrop, frame, local->op_ret, op_errno,
+                             local->rebalance.dict, local->rebalance.xdata);
+        }
+
+        return 0;
+    }
+
+    if (subvol == NULL)
+        goto out;
+
+    local->call_cnt = 2; /* This is the second attempt */
+
+    if (local->fop == GF_FOP_XATTROP) {
+        STACK_WIND(frame, dht_common_xattrop_cbk, subvol, subvol->fops->xattrop,
+                   &local->loc, local->rebalance.flags, local->rebalance.xattr,
+                   local->xattr_req);
+    } else {
+        STACK_WIND(frame, dht_common_xattrop_cbk, subvol,
+                   subvol->fops->fxattrop, local->fd, local->rebalance.flags,
+                   local->rebalance.xattr, local->xattr_req);
+    }
+
+    return 0;
+
+out:
+
+    /* If local is unavailable we could be unwinding the wrong
+     * function here */
+
+    if (local && (local->fop == GF_FOP_XATTROP)) {
+        DHT_STACK_UNWIND(xattrop, frame, -1, op_errno, NULL, NULL);
+    } else {
+        DHT_STACK_UNWIND(fxattrop, frame, -1, op_errno, NULL, NULL);
+    }
+    return 0;
+}
+
+static int
+dht_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+    DHT_STACK_UNWIND(xattrop, frame, op_ret, op_errno, dict, xdata);
+    return 0;
+}
+
+/* Set both DHT_IATT_IN_XDATA_KEY and DHT_MODE_IN_XDATA_KEY
+ * Use DHT_MODE_IN_XDATA_KEY if available. Else fall back to
+ * DHT_IATT_IN_XDATA_KEY
+ */
+static int
+dht_request_iatt_in_xdata(dict_t *xattr_req)
+{
+    int ret = -1;
+
+    ret = dict_set_int8(xattr_req, DHT_MODE_IN_XDATA_KEY, 1);
+    ret = dict_set_int8(xattr_req, DHT_IATT_IN_XDATA_KEY, 1);
+
+    /* At least one call succeeded */
+    return ret;
+}
+
+int
+dht_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+            gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+    int ret = -1;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_XATTROP);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for gfid=%s",
+                     uuid_utoa(loc->inode->gfid));
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    /* Todo : Handle dirs as well. At the moment the only xlator above dht
+     * that uses xattrop is sharding and that is only for files */
+
+    if (IA_ISDIR(loc->inode->ia_type)) {
+        STACK_WIND(frame, dht_xattrop_cbk, subvol, subvol->fops->xattrop, loc,
+                   flags, dict, xdata);
+
+    } else {
+        local->xattr_req = xdata ? dict_ref(xdata) : dict_new();
+        local->call_cnt = 1;
+
+        local->rebalance.xattr = dict_ref(dict);
+        local->rebalance.flags = flags;
+
+        ret = dht_request_iatt_in_xdata(local->xattr_req);
+
+        if (ret) {
+            gf_msg_debug(this->name, 0,
+                         "Failed to set dictionary key %s file=%s",
+                         DHT_IATT_IN_XDATA_KEY, loc->path);
+        }
+
+        STACK_WIND(frame, dht_common_xattrop_cbk, subvol, subvol->fops->xattrop,
+                   loc, local->rebalance.flags, local->rebalance.xattr,
+                   local->xattr_req);
+    }
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(xattrop, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
+
+static int
+dht_fxattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+    DHT_STACK_UNWIND(fxattrop, frame, op_ret, op_errno, dict, xdata);
+    return 0;
+}
+
+int
+dht_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+             gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+    int ret = -1;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    subvol = dht_subvol_get_cached(this, fd->inode);
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_FXATTROP);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    /* Todo : Handle dirs as well. At the moment the only xlator above dht
+     * that uses xattrop is sharding and that is only for files */
+
+    if (IA_ISDIR(fd->inode->ia_type)) {
+        STACK_WIND(frame, dht_fxattrop_cbk, subvol, subvol->fops->fxattrop, fd,
+                   flags, dict, xdata);
+
+    } else {
+        local->xattr_req = xdata ? dict_ref(xdata) : dict_new();
+        local->call_cnt = 1;
+
+        local->rebalance.xattr = dict_ref(dict);
+        local->rebalance.flags = flags;
+
+        ret = dht_request_iatt_in_xdata(local->xattr_req);
+
+        if (ret) {
+            gf_msg_debug(this->name, 0, "Failed to set dictionary key %s fd=%p",
+                         DHT_IATT_IN_XDATA_KEY, fd);
+        }
+
+        STACK_WIND(frame, dht_common_xattrop_cbk, subvol,
+                   subvol->fops->fxattrop, fd, local->rebalance.flags,
+                   local->rebalance.xattr, local->xattr_req);
+    }
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(fxattrop, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
+
+/* Currently no translators on top of 'distribute' will be using
+ * below fops, hence not implementing 'migration' related checks
+ */
+
+static int
+dht_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata)
+
+{
+    dht_lk_inode_unref(frame, op_ret);
+    DHT_STACK_UNWIND(inodelk, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+int32_t
+dht_inodelk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+            int32_t cmd, struct gf_flock *lock, dict_t *xdata)
+{
+    xlator_t *lock_subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_INODELK);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->lock_type = lock->l_type;
+    lock_subvol = dht_get_lock_subvolume(this, lock, local);
+    if (!lock_subvol) {
+        gf_msg_debug(this->name, 0, "no lock subvolume for path=%s", loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    local->call_cnt = 1;
+
+    STACK_WIND(frame, dht_inodelk_cbk, lock_subvol, lock_subvol->fops->inodelk,
+               volume, loc, cmd, lock, xdata);
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(inodelk, frame, -1, op_errno, NULL);
+
+    return 0;
+}
+
+int
+dht_finodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
+
+{
+    dht_local_t *local = NULL;
+    int ret = 0;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+
+    local = frame->local;
+
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
+
+out:
+    dht_lk_inode_unref(frame, op_ret);
+    DHT_STACK_UNWIND(finodelk, frame, op_ret, op_errno, xdata);
+
+    return 0;
+}
+
+int
+dht_finodelk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+             int32_t cmd, struct gf_flock *lock, dict_t *xdata)
+{
+    xlator_t *lock_subvol = NULL;
+    dht_local_t *local = NULL;
+    int op_errno = -1;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_INODELK);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->call_cnt = 1;
+    local->lock_type = lock->l_type;
+
+    lock_subvol = dht_get_lock_subvolume(this, lock, local);
+    if (!lock_subvol) {
+        gf_msg_debug(this->name, 0, "no lock subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    /*
+            local->cached_subvol = lock_subvol;
+            ret = dht_check_and_open_fd_on_subvol (this, frame);
+            if (ret)
+                    goto err;
+    */
+    local->rebalance.flock = *lock;
+    local->rebalance.lock_cmd = cmd;
+    local->key = gf_strdup(volume);
+
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    STACK_WIND(frame, dht_finodelk_cbk, lock_subvol,
+               lock_subvol->fops->finodelk, volume, fd, cmd, lock, xdata);
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(finodelk, frame, -1, op_errno, NULL);
+
+    return 0;
+}
diff --git a/xlators/cluster/dht/src/dht-inode-write.c b/xlators/cluster/dht/src/dht-inode-write.c
new file mode 100644
index 00000000000..2f23ce90fbd
--- /dev/null
+++ b/xlators/cluster/dht/src/dht-inode-write.c
@@ -0,0 +1,1404 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "dht-common.h"
+
+static int
+dht_writev2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret);
+static int
+dht_truncate2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret);
+static int
+dht_setattr2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret);
+static int
+dht_fallocate2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret);
+static int
+dht_discard2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret);
+static int
+dht_zerofill2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret);
+
+int
+dht_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+               dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    int ret = -1;
+    xlator_t *subvol1 = NULL;
+    xlator_t *subvol2 = NULL;
+
+    local = frame->local;
+    prev = cookie;
+
+    if (!local) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    /* writev fails with EBADF if dht has not yet opened the fd
+     * on the cached subvol. This could happen if the file was migrated
+     * and a lookup updated the cached subvol in the inode ctx.
+     * We only check once as this could be a valid bad fd error.
+     */
+
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
+
+    if (op_ret == -1 && !dht_inode_missing(op_errno)) {
+        local->op_errno = op_errno;
+        local->op_ret = -1;
+        gf_msg_debug(this->name, 0, "subvolume %s returned -1 (%s)", prev->name,
+                     strerror(op_errno));
+        goto out;
+    }
+
+    if (local->call_cnt != 1) {
+        /* preserve the modes of source */
+        if (local->stbuf.ia_blocks) {
+            dht_iatt_merge(this, postbuf, &local->stbuf);
+            dht_iatt_merge(this, prebuf, &local->prebuf);
+        }
+        goto out;
+    }
+
+    local->rebalance.target_op_fn = dht_writev2;
+
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
+
+    /* We might need to pass the stbuf information to the higher DHT
+     * layer for appropriate handling.
+     */
+
+    dht_set_local_rebalance(this, local, NULL, prebuf, postbuf, xdata);
+
+    /* Phase 2 of migration */
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(postbuf)) {
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+    /* Check if the rebalance phase1 is true */
+    if (IS_DHT_MIGRATION_PHASE1(postbuf)) {
+        if (!local->xattr_req) {
+            local->xattr_req = dict_new();
+            if (!local->xattr_req) {
+                gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_NO_MEMORY, ENOMEM,
+                       "insufficient memory");
+                local->op_errno = ENOMEM;
+                local->op_ret = -1;
+                goto out;
+            }
+        }
+
+        ret = dict_set_uint32(local->xattr_req, GF_PROTECT_FROM_EXTERNAL_WRITES,
+                              1);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_DICT_SET_FAILED, 0,
+                   "Failed to set key %s in dictionary",
+                   GF_PROTECT_FROM_EXTERNAL_WRITES);
+            local->op_errno = ENOMEM;
+            local->op_ret = -1;
+            goto out;
+        }
+
+        dht_iatt_merge(this, &local->stbuf, postbuf);
+        dht_iatt_merge(this, &local->prebuf, prebuf);
+
+        ret = dht_inode_ctx_get_mig_info(this, local->fd->inode, &subvol1,
+                                         &subvol2);
+        if (!dht_mig_info_is_invalid(local->cached_subvol, subvol1, subvol2)) {
+            if (dht_fd_open_on_dst(this, local->fd, subvol2)) {
+                dht_writev2(this, subvol2, frame, 0);
+                return 0;
+            }
+        }
+        ret = dht_rebalance_in_progress_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+out:
+    DHT_STRIP_PHASE1_FLAGS(postbuf);
+    DHT_STRIP_PHASE1_FLAGS(prebuf);
+
+    DHT_STACK_UNWIND(writev, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+
+    return 0;
+}
+
+static int
+dht_writev2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
+{
+    dht_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+
+    if ((frame == NULL) || (frame->local == NULL))
+        goto out;
+
+    local = frame->local;
+    op_errno = local->op_errno;
+
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        DHT_STACK_UNWIND(writev, frame, local->op_ret, local->op_errno,
+                         &local->rebalance.prebuf, &local->rebalance.postbuf,
+                         local->rebalance.xdata);
+        return 0;
+    }
+
+    if (subvol == NULL)
+        goto out;
+
+    local->call_cnt = 2; /* This is the second attempt */
+
+    STACK_WIND_COOKIE(frame, dht_writev_cbk, subvol, subvol,
+                      subvol->fops->writev, local->fd, local->rebalance.vector,
+                      local->rebalance.count, local->rebalance.offset,
+                      local->rebalance.flags, local->rebalance.iobref,
+                      local->xattr_req);
+
+    return 0;
+
+out:
+    DHT_STACK_UNWIND(writev, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int
+dht_writev(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
+           int count, off_t off, uint32_t flags, struct iobref *iobref,
+           dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_WRITE);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    local->rebalance.vector = iov_dup(vector, count);
+    local->rebalance.offset = off;
+    local->rebalance.count = count;
+    local->rebalance.flags = flags;
+    local->rebalance.iobref = iobref_ref(iobref);
+    local->call_cnt = 1;
+
+    STACK_WIND_COOKIE(frame, dht_writev_cbk, subvol, subvol,
+                      subvol->fops->writev, fd, local->rebalance.vector,
+                      local->rebalance.count, local->rebalance.offset,
+                      local->rebalance.flags, local->rebalance.iobref,
+                      local->xattr_req);
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(writev, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int
+dht_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+                 dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    int ret = -1;
+    xlator_t *src_subvol = NULL;
+    xlator_t *dst_subvol = NULL;
+    inode_t *inode = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", cookie, out);
+
+    local = frame->local;
+    prev = cookie;
+
+    /* Needs to be checked only for ftruncate.
+     * ftruncate fails with EBADF/EINVAL if dht has not yet opened the fd
+     * on the cached subvol. This could happen if the file was migrated
+     * and a lookup updated the cached subvol in the inode ctx.
+     * We only check once as this could actually be a valid error.
+     */
+
+    if ((local->fop == GF_FOP_FTRUNCATE) &&
+        dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
+
+    if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+        local->op_errno = op_errno;
+        local->op_ret = -1;
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->name);
+
+        goto out;
+    }
+
+    if (local->call_cnt != 1) {
+        if (local->stbuf.ia_blocks) {
+            dht_iatt_merge(this, postbuf, &local->stbuf);
+            dht_iatt_merge(this, prebuf, &local->prebuf);
+        }
+        goto out;
+    }
+
+    local->rebalance.target_op_fn = dht_truncate2;
+
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
+
+    /* We might need to pass the stbuf information to the higher DHT
+     * layer for appropriate handling.
+     */
+
+    dht_set_local_rebalance(this, local, NULL, prebuf, postbuf, xdata);
+
+    /* Phase 2 of migration */
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(postbuf)) {
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+    /* Check if the rebalance phase1 is true */
+    if (IS_DHT_MIGRATION_PHASE1(postbuf)) {
+        dht_iatt_merge(this, &local->stbuf, postbuf);
+        dht_iatt_merge(this, &local->prebuf, prebuf);
+
+        inode = (local->fd) ? local->fd->inode : local->loc.inode;
+
+        dht_inode_ctx_get_mig_info(this, inode, &src_subvol, &dst_subvol);
+        if (!dht_mig_info_is_invalid(local->cached_subvol, src_subvol,
+                                     dst_subvol)) {
+            if ((!local->fd) ||
+                ((local->fd) &&
+                 dht_fd_open_on_dst(this, local->fd, dst_subvol))) {
+                dht_truncate2(this, dst_subvol, frame, 0);
+                return 0;
+            }
+        }
+        ret = dht_rebalance_in_progress_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+out:
+    DHT_STRIP_PHASE1_FLAGS(postbuf);
+    DHT_STRIP_PHASE1_FLAGS(prebuf);
+
+    DHT_STACK_UNWIND(truncate, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+err:
+    return 0;
+}
+
+static int
+dht_truncate2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
+{
+    dht_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+
+    if (!frame || !frame->local)
+        goto out;
+
+    local = frame->local;
+    op_errno = local->op_errno;
+
+    /* This dht xlator is not migrating the file  */
+    if (we_are_not_migrating(ret)) {
+        DHT_STACK_UNWIND(truncate, frame, local->op_ret, local->op_errno,
+                         &local->rebalance.prebuf, &local->rebalance.postbuf,
+                         local->rebalance.xdata);
+        return 0;
+    }
+
+    if (subvol == NULL)
+        goto out;
+
+    local->call_cnt = 2; /* This is the second attempt */
+
+    if (local->fop == GF_FOP_TRUNCATE) {
+        STACK_WIND_COOKIE(frame, dht_truncate_cbk, subvol, subvol,
+                          subvol->fops->truncate, &local->loc,
+                          local->rebalance.offset, local->xattr_req);
+    } else {
+        STACK_WIND_COOKIE(frame, dht_truncate_cbk, subvol, subvol,
+                          subvol->fops->ftruncate, local->fd,
+                          local->rebalance.offset, local->xattr_req);
+    }
+
+    return 0;
+
+out:
+    DHT_STACK_UNWIND(truncate, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
+}
+
+int
+dht_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+             dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_TRUNCATE);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->rebalance.offset = offset;
+    local->call_cnt = 1;
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for gfid=%s",
+                     uuid_utoa(loc->inode->gfid));
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    STACK_WIND_COOKIE(frame, dht_truncate_cbk, subvol, subvol,
+                      subvol->fops->truncate, loc, offset, xdata);
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(truncate, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int
+dht_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+              dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_FTRUNCATE);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->rebalance.offset = offset;
+    local->call_cnt = 1;
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    STACK_WIND_COOKIE(frame, dht_truncate_cbk, subvol, subvol,
+                      subvol->fops->ftruncate, fd, local->rebalance.offset,
+                      local->xattr_req);
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(ftruncate, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int
+dht_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                  int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+                  dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    int ret = -1;
+    xlator_t *src_subvol = NULL;
+    xlator_t *dst_subvol = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", cookie, out);
+
+    local = frame->local;
+    prev = cookie;
+
+    /* fallocate fails with EBADF if dht has not yet opened the fd
+     * on the cached subvol. This could happen if the file was migrated
+     * and a lookup updated the cached subvol in the inode ctx.
+     * We only check once as this could actually be a valid error.
+     */
+
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
+
+    if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+        local->op_errno = op_errno;
+        local->op_ret = -1;
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->name);
+
+        goto out;
+    }
+
+    if (local->call_cnt != 1) {
+        if (local->stbuf.ia_blocks) {
+            dht_iatt_merge(this, postbuf, &local->stbuf);
+            dht_iatt_merge(this, prebuf, &local->prebuf);
+        }
+        goto out;
+    }
+
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
+    local->rebalance.target_op_fn = dht_fallocate2;
+
+    dht_set_local_rebalance(this, local, NULL, prebuf, postbuf, xdata);
+
+    /* Phase 2 of migration */
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(postbuf)) {
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+    /* Check if the rebalance phase1 is true */
+    if (IS_DHT_MIGRATION_PHASE1(postbuf)) {
+        dht_iatt_merge(this, &local->stbuf, postbuf);
+        dht_iatt_merge(this, &local->prebuf, prebuf);
+
+        dht_inode_ctx_get_mig_info(this, local->fd->inode, &src_subvol,
+                                   &dst_subvol);
+        if (!dht_mig_info_is_invalid(local->cached_subvol, src_subvol,
+                                     dst_subvol)) {
+            if (dht_fd_open_on_dst(this, local->fd, dst_subvol)) {
+                dht_fallocate2(this, dst_subvol, frame, 0);
+                return 0;
+            }
+        }
+        ret = dht_rebalance_in_progress_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+out:
+    DHT_STRIP_PHASE1_FLAGS(postbuf);
+    DHT_STRIP_PHASE1_FLAGS(prebuf);
+
+    DHT_STACK_UNWIND(fallocate, frame, op_ret, op_errno, prebuf, postbuf,
+                     xdata);
+err:
+    return 0;
+}
+
+static int
+dht_fallocate2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
+{
+    dht_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+
+    if (!frame || !frame->local)
+        goto out;
+
+    local = frame->local;
+    op_errno = local->op_errno;
+
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        DHT_STACK_UNWIND(fallocate, frame, local->op_ret, local->op_errno,
+                         &local->rebalance.prebuf, &local->rebalance.postbuf,
+                         local->rebalance.xdata);
+        return 0;
+    }
+
+    if (subvol == NULL)
+        goto out;
+
+    local->call_cnt = 2; /* This is the second attempt */
+
+    STACK_WIND_COOKIE(frame, dht_fallocate_cbk, subvol, subvol,
+                      subvol->fops->fallocate, local->fd,
+                      local->rebalance.flags, local->rebalance.offset,
+                      local->rebalance.size, local->xattr_req);
+
+    return 0;
+
+out:
+    DHT_STACK_UNWIND(fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
+}
+
+int
+dht_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+              off_t offset, size_t len, dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_FALLOCATE);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->rebalance.flags = mode;
+    local->rebalance.offset = offset;
+    local->rebalance.size = len;
+
+    local->call_cnt = 1;
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    STACK_WIND_COOKIE(frame, dht_fallocate_cbk, subvol, subvol,
+                      subvol->fops->fallocate, fd, local->rebalance.flags,
+                      local->rebalance.offset, local->rebalance.size,
+                      local->xattr_req);
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int
+dht_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+                dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    int ret = -1;
+    xlator_t *src_subvol = NULL;
+    xlator_t *dst_subvol = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", cookie, out);
+
+    local = frame->local;
+    prev = cookie;
+
+    /* discard fails with EBADF if dht has not yet opened the fd
+     * on the cached subvol. This could happen if the file was migrated
+     * and a lookup updated the cached subvol in the inode ctx.
+     * We only check once as this could actually be a valid error.
+     */
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
+
+    if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+        local->op_errno = op_errno;
+        local->op_ret = -1;
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->name);
+
+        goto out;
+    }
+
+    if (local->call_cnt != 1) {
+        if (local->stbuf.ia_blocks) {
+            dht_iatt_merge(this, postbuf, &local->stbuf);
+            dht_iatt_merge(this, prebuf, &local->prebuf);
+        }
+        goto out;
+    }
+
+    local->rebalance.target_op_fn = dht_discard2;
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
+
+    dht_set_local_rebalance(this, local, NULL, prebuf, postbuf, xdata);
+
+    /* Phase 2 of migration */
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(postbuf)) {
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+    /* Check if the rebalance phase1 is true */
+    if (IS_DHT_MIGRATION_PHASE1(postbuf)) {
+        dht_iatt_merge(this, &local->stbuf, postbuf);
+        dht_iatt_merge(this, &local->prebuf, prebuf);
+
+        dht_inode_ctx_get_mig_info(this, local->fd->inode, &src_subvol,
+                                   &dst_subvol);
+        if (!dht_mig_info_is_invalid(local->cached_subvol, src_subvol,
+                                     dst_subvol)) {
+            if (dht_fd_open_on_dst(this, local->fd, dst_subvol)) {
+                dht_discard2(this, dst_subvol, frame, 0);
+                return 0;
+            }
+        }
+        ret = dht_rebalance_in_progress_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+out:
+    DHT_STRIP_PHASE1_FLAGS(postbuf);
+    DHT_STRIP_PHASE1_FLAGS(prebuf);
+
+    DHT_STACK_UNWIND(discard, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+err:
+    return 0;
+}
+
+static int
+dht_discard2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
+{
+    dht_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+
+    if (!frame || !frame->local)
+        goto out;
+
+    local = frame->local;
+    op_errno = local->op_errno;
+
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        DHT_STACK_UNWIND(discard, frame, local->op_ret, local->op_errno,
+                         &local->rebalance.prebuf, &local->rebalance.postbuf,
+                         local->rebalance.xdata);
+        return 0;
+    }
+
+    if (subvol == NULL)
+        goto out;
+
+    local->call_cnt = 2; /* This is the second attempt */
+
+    STACK_WIND_COOKIE(frame, dht_discard_cbk, subvol, subvol,
+                      subvol->fops->discard, local->fd, local->rebalance.offset,
+                      local->rebalance.size, local->xattr_req);
+
+    return 0;
+
+out:
+    DHT_STACK_UNWIND(discard, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
+}
+
+int
+dht_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+            size_t len, dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_DISCARD);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->rebalance.offset = offset;
+    local->rebalance.size = len;
+
+    local->call_cnt = 1;
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    STACK_WIND_COOKIE(frame, dht_discard_cbk, subvol, subvol,
+                      subvol->fops->discard, fd, local->rebalance.offset,
+                      local->rebalance.size, local->xattr_req);
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(discard, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int
+dht_zerofill_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+                 dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    int ret = -1;
+    xlator_t *subvol1 = NULL, *subvol2 = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+    GF_VALIDATE_OR_GOTO("dht", cookie, out);
+
+    local = frame->local;
+    prev = cookie;
+
+    /* zerofill fails with EBADF if dht has not yet opened the fd
+     * on the cached subvol. This could happen if the file was migrated
+     * and a lookup updated the cached subvol in the inode ctx.
+     * We only check once as this could actually be a valid error.
+     */
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
+
+    if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+        local->op_errno = op_errno;
+        local->op_ret = -1;
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->name);
+        goto out;
+    }
+
+    if (local->call_cnt != 1) {
+        if (local->stbuf.ia_blocks) {
+            dht_iatt_merge(this, postbuf, &local->stbuf);
+            dht_iatt_merge(this, prebuf, &local->prebuf);
+        }
+        goto out;
+    }
+
+    local->rebalance.target_op_fn = dht_zerofill2;
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
+
+    dht_set_local_rebalance(this, local, NULL, prebuf, postbuf, xdata);
+
+    /* Phase 2 of migration */
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(postbuf)) {
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+    /* Check if the rebalance phase1 is true */
+    if (IS_DHT_MIGRATION_PHASE1(postbuf)) {
+        dht_iatt_merge(this, &local->stbuf, postbuf);
+        dht_iatt_merge(this, &local->prebuf, prebuf);
+
+        ret = dht_inode_ctx_get_mig_info(this, local->fd->inode, &subvol1,
+                                         &subvol2);
+        if (!dht_mig_info_is_invalid(local->cached_subvol, subvol1, subvol2)) {
+            if (dht_fd_open_on_dst(this, local->fd, subvol2)) {
+                dht_zerofill2(this, subvol2, frame, 0);
+                return 0;
+            }
+        }
+
+        ret = dht_rebalance_in_progress_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+out:
+    DHT_STRIP_PHASE1_FLAGS(postbuf);
+    DHT_STRIP_PHASE1_FLAGS(prebuf);
+
+    DHT_STACK_UNWIND(zerofill, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+err:
+    return 0;
+}
+
+static int
+dht_zerofill2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
+{
+    dht_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+
+    if (!frame || !frame->local)
+        goto out;
+
+    local = frame->local;
+
+    op_errno = local->op_errno;
+
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        DHT_STACK_UNWIND(zerofill, frame, local->op_ret, local->op_errno,
+                         &local->rebalance.prebuf, &local->rebalance.postbuf,
+                         local->rebalance.xdata);
+
+        return 0;
+    }
+
+    if (subvol == NULL)
+        goto out;
+
+    local->call_cnt = 2; /* This is the second attempt */
+
+    STACK_WIND_COOKIE(frame, dht_zerofill_cbk, subvol, subvol,
+                      subvol->fops->zerofill, local->fd,
+                      local->rebalance.offset, local->rebalance.size,
+                      local->xattr_req);
+
+    return 0;
+
+out:
+
+    DHT_STACK_UNWIND(zerofill, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
+}
+
+int
+dht_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             off_t len, dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    int op_errno = -1;
+    dht_local_t *local = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_ZEROFILL);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->rebalance.offset = offset;
+    local->rebalance.size = len;
+
+    local->call_cnt = 1;
+    subvol = local->cached_subvol;
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no cached subvolume for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    STACK_WIND_COOKIE(frame, dht_zerofill_cbk, subvol, subvol,
+                      subvol->fops->zerofill, fd, local->rebalance.offset,
+                      local->rebalance.size, local->xattr_req);
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(zerofill, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+/* handle cases of migration here for 'setattr()' calls */
+int
+dht_file_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    int ret = -1;
+
+    local = frame->local;
+    prev = cookie;
+
+    local->op_errno = op_errno;
+
+    if ((local->fop == GF_FOP_FSETATTR) &&
+        dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
+
+    if ((op_ret == -1) && !dht_inode_missing(op_errno)) {
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->name);
+        goto out;
+    }
+
+    if (local->call_cnt != 1)
+        goto out;
+
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
+
+    local->rebalance.target_op_fn = dht_setattr2;
+
+    /* Phase 2 of migration */
+    if ((op_ret == -1) || IS_DHT_MIGRATION_PHASE2(postbuf)) {
+        dht_set_local_rebalance(this, local, NULL, prebuf, postbuf, xdata);
+
+        ret = dht_rebalance_complete_check(this, frame);
+        if (!ret)
+            return 0;
+    }
+
+    /* At the end of the migration process, whatever 'attr' we
+       have on source file will be migrated to destination file
+       in one shot, hence we don't need to check for in progress
+       state here (ie, PHASE1) */
+out:
+    DHT_STRIP_PHASE1_FLAGS(postbuf);
+    DHT_STRIP_PHASE1_FLAGS(prebuf);
+
+    DHT_STACK_UNWIND(setattr, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+
+    return 0;
+}
+
+static int
+dht_setattr2(xlator_t *this, xlator_t *subvol, call_frame_t *frame, int ret)
+{
+    dht_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+
+    if (!frame || !frame->local)
+        goto out;
+
+    local = frame->local;
+    op_errno = local->op_errno;
+
+    if (we_are_not_migrating(ret)) {
+        /* This dht xlator is not migrating the file. Unwind and
+         * pass on the original mode bits so the higher DHT layer
+         * can handle this.
+         */
+        DHT_STACK_UNWIND(setattr, frame, local->op_ret, local->op_errno,
+                         &local->rebalance.prebuf, &local->rebalance.postbuf,
+                         local->rebalance.xdata);
+        return 0;
+    }
+
+    if (subvol == NULL)
+        goto out;
+
+    local->call_cnt = 2; /* This is the second attempt */
+
+    if (local->fop == GF_FOP_SETATTR) {
+        STACK_WIND_COOKIE(frame, dht_file_setattr_cbk, subvol, subvol,
+                          subvol->fops->setattr, &local->loc,
+                          &local->rebalance.stbuf, local->rebalance.flags,
+                          local->xattr_req);
+    } else {
+        STACK_WIND_COOKIE(frame, dht_file_setattr_cbk, subvol, subvol,
+                          subvol->fops->fsetattr, local->fd,
+                          &local->rebalance.stbuf, local->rebalance.flags,
+                          local->xattr_req);
+    }
+
+    return 0;
+
+out:
+    DHT_STACK_UNWIND(setattr, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
+}
+
+/* Keep the existing code same for all the cases other than regular file */
+int
+dht_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                int op_errno, struct iatt *statpre, struct iatt *statpost,
+                dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+
+    local = frame->local;
+    prev = cookie;
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret == -1) {
+            local->op_errno = op_errno;
+            UNLOCK(&frame->lock);
+            gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                         prev->name);
+            goto post_unlock;
+        }
+
+        dht_iatt_merge(this, &local->prebuf, statpre);
+        dht_iatt_merge(this, &local->stbuf, statpost);
+
+        local->op_ret = 0;
+        local->op_errno = 0;
+    }
+    UNLOCK(&frame->lock);
+post_unlock:
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        if (local->op_ret == 0)
+            dht_inode_ctx_time_set(local->loc.inode, this, &local->stbuf);
+        DHT_STACK_UNWIND(setattr, frame, local->op_ret, local->op_errno,
+                         &local->prebuf, &local->stbuf, xdata);
+    }
+
+    return 0;
+}
+
+/* Keep the existing code same for all the cases other than regular file */
+int
+dht_non_mds_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, struct iatt *statpre,
+                        struct iatt *statpost, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+
+    local = frame->local;
+    prev = cookie;
+
+    if (op_ret == -1) {
+        gf_msg(this->name, op_errno, 0, 0, "subvolume %s returned -1",
+               prev->name);
+        goto post_unlock;
+    }
+
+    LOCK(&frame->lock);
+    {
+        dht_iatt_merge(this, &local->prebuf, statpre);
+        dht_iatt_merge(this, &local->stbuf, statpost);
+
+        local->op_ret = 0;
+        local->op_errno = 0;
+    }
+    UNLOCK(&frame->lock);
+post_unlock:
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        dht_inode_ctx_time_set(local->loc.inode, this, &local->stbuf);
+        DHT_STACK_UNWIND(setattr, frame, 0, 0, &local->prebuf, &local->stbuf,
+                         xdata);
+    }
+
+    return 0;
+}
+
+int
+dht_mds_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int op_ret, int op_errno, struct iatt *statpre,
+                    struct iatt *statpost, dict_t *xdata)
+
+{
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *mds_subvol = NULL;
+    struct iatt loc_stbuf = {
+        0,
+    };
+    int i = 0;
+
+    local = frame->local;
+    prev = cookie;
+    conf = this->private;
+    mds_subvol = local->mds_subvol;
+
+    if (op_ret == -1) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        gf_msg_debug(this->name, op_errno, "subvolume %s returned -1",
+                     prev->name);
+        goto out;
+    }
+
+    local->op_ret = 0;
+    loc_stbuf = local->stbuf;
+    dht_iatt_merge(this, &local->prebuf, statpre);
+    dht_iatt_merge(this, &local->stbuf, statpost);
+
+    local->call_cnt = conf->subvolume_cnt - 1;
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (mds_subvol == conf->subvolumes[i])
+            continue;
+        STACK_WIND_COOKIE(frame, dht_non_mds_setattr_cbk, conf->subvolumes[i],
+                          conf->subvolumes[i],
+                          conf->subvolumes[i]->fops->setattr, &local->loc,
+                          &loc_stbuf, local->valid, local->xattr_req);
+    }
+
+    return 0;
+out:
+    DHT_STACK_UNWIND(setattr, frame, local->op_ret, local->op_errno,
+                     &local->prebuf, &local->stbuf, xdata);
+
+    return 0;
+}
+
+int
+dht_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *stbuf,
+            int32_t valid, dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    xlator_t *mds_subvol = NULL;
+    dht_layout_t *layout = NULL;
+    dht_local_t *local = NULL;
+    int op_errno = -1;
+    int i = -1;
+    int ret = -1;
+    int call_cnt = 0;
+    dht_conf_t *conf = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+    VALIDATE_OR_GOTO(loc->path, err);
+
+    conf = this->private;
+    local = dht_local_init(frame, loc, NULL, GF_FOP_SETATTR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!layout) {
+        gf_msg_debug(this->name, 0, "no layout for path=%s", loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (!layout_is_sane(layout)) {
+        gf_msg_debug(this->name, 0, "layout is not sane for path=%s",
+                     loc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    if (IA_ISREG(loc->inode->ia_type)) {
+        /* in the regular file _cbk(), we need to check for
+           migration possibilities */
+        local->rebalance.stbuf = *stbuf;
+        local->rebalance.flags = valid;
+        local->call_cnt = 1;
+        subvol = local->cached_subvol;
+
+        STACK_WIND_COOKIE(frame, dht_file_setattr_cbk, subvol, subvol,
+                          subvol->fops->setattr, loc, stbuf, valid, xdata);
+
+        return 0;
+    }
+
+    local->call_cnt = call_cnt = layout->cnt;
+
+    if (IA_ISDIR(loc->inode->ia_type) && !__is_root_gfid(loc->inode->gfid) &&
+        call_cnt != 1) {
+        ret = dht_inode_ctx_mdsvol_get(loc->inode, this, &mds_subvol);
+        if (ret || !mds_subvol) {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+                   "Failed to get mds subvol for path %s", local->loc.path);
+            op_errno = EINVAL;
+            goto err;
+        }
+
+        local->mds_subvol = mds_subvol;
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (conf->subvolumes[i] == mds_subvol) {
+                if (!conf->subvolume_status[i]) {
+                    gf_msg(this->name, GF_LOG_WARNING, layout->list[i].err,
+                           DHT_MSG_HASHED_SUBVOL_DOWN,
+                           "MDS subvol is down for path "
+                           " %s Unable to set attr ",
+                           local->loc.path);
+                    op_errno = ENOTCONN;
+                    goto err;
+                }
+            }
+        }
+        local->valid = valid;
+        local->stbuf = *stbuf;
+
+        STACK_WIND_COOKIE(frame, dht_mds_setattr_cbk, local->mds_subvol,
+                          local->mds_subvol, local->mds_subvol->fops->setattr,
+                          loc, stbuf, valid, xdata);
+        return 0;
+    } else {
+        for (i = 0; i < call_cnt; i++) {
+            STACK_WIND_COOKIE(frame, dht_setattr_cbk, layout->list[i].xlator,
+                              layout->list[i].xlator,
+                              layout->list[i].xlator->fops->setattr, loc, stbuf,
+                              valid, xdata);
+        }
+    }
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(setattr, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int
+dht_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
+             int32_t valid, dict_t *xdata)
+{
+    xlator_t *subvol = NULL;
+    dht_layout_t *layout = NULL;
+    dht_local_t *local = NULL;
+    int op_errno = -1;
+    int i = -1;
+    int call_cnt = 0;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    local = dht_local_init(frame, NULL, fd, GF_FOP_FSETATTR);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    layout = local->layout;
+    if (!layout) {
+        gf_msg_debug(this->name, 0, "no layout for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (!layout_is_sane(layout)) {
+        gf_msg_debug(this->name, 0, "layout is not sane for fd=%p", fd);
+        op_errno = EINVAL;
+        goto err;
+    }
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    if (IA_ISREG(fd->inode->ia_type)) {
+        /* in the regular file _cbk(), we need to check for
+           migration possibilities */
+        local->rebalance.stbuf = *stbuf;
+        local->rebalance.flags = valid;
+        local->call_cnt = 1;
+        subvol = local->cached_subvol;
+
+        STACK_WIND_COOKIE(frame, dht_file_setattr_cbk, subvol, subvol,
+                          subvol->fops->fsetattr, fd, &local->rebalance.stbuf,
+                          local->rebalance.flags, local->xattr_req);
+        return 0;
+    }
+
+    local->call_cnt = call_cnt = layout->cnt;
+
+    for (i = 0; i < call_cnt; i++) {
+        STACK_WIND_COOKIE(frame, dht_setattr_cbk, layout->list[i].xlator,
+                          layout->list[i].xlator,
+                          layout->list[i].xlator->fops->fsetattr, fd, stbuf,
+                          valid, xdata);
+    }
+
+    return 0;
+
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(fsetattr, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
diff --git a/xlators/cluster/dht/src/dht-layout.c b/xlators/cluster/dht/src/dht-layout.c
index a11b78f8d06..fda904c92c9 100644
--- a/xlators/cluster/dht/src/dht-layout.c
+++ b/xlators/cluster/dht/src/dht-layout.c
@@ -1,728 +1,808 @@
 /*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
 
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
 
-#include "glusterfs.h"
-#include "xlator.h"
 #include "dht-common.h"
-#include "byte-order.h"
+#include <glusterfs/byte-order.h>
+#include "unittest/unittest.h"
 
-#define layout_base_size (sizeof (dht_layout_t))
+#define layout_base_size (sizeof(dht_layout_t))
 
-#define layout_entry_size (sizeof ((dht_layout_t *)NULL)->list[0])
+#define layout_entry_size (sizeof((dht_layout_t *)NULL)->list[0])
 
 #define layout_size(cnt) (layout_base_size + (cnt * layout_entry_size))
 
-
 dht_layout_t *
-dht_layout_new (xlator_t *this, int cnt)
+dht_layout_new(xlator_t *this, int cnt)
 {
-	dht_layout_t *layout = NULL;
-        dht_conf_t   *conf = NULL;
+    dht_layout_t *layout = NULL;
+    dht_conf_t *conf = NULL;
 
+    REQUIRE(NULL != this);
+    REQUIRE(cnt >= 0);
 
-        conf = this->private;
+    conf = this->private;
 
-	layout = GF_CALLOC (1, layout_size (cnt),
-                            gf_dht_mt_dht_layout_t);
-	if (!layout) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto out;
-	}
+    layout = GF_CALLOC(1, layout_size(cnt), gf_dht_mt_dht_layout_t);
+    if (!layout) {
+        goto out;
+    }
 
-        layout->type = DHT_HASH_TYPE_DM;
-	layout->cnt = cnt;
-        if (conf)
-                layout->gen = conf->gen;
+    layout->type = DHT_HASH_TYPE_DM;
+    layout->cnt = cnt;
 
-        layout->ref = 1;
+    if (conf) {
+        layout->spread_cnt = conf->dir_spread_cnt;
+        layout->gen = conf->gen;
+    }
+
+    GF_ATOMIC_INIT(layout->ref, 1);
+
+    ENSURE(NULL != layout);
+    ENSURE(layout->type == DHT_HASH_TYPE_DM);
+    ENSURE(layout->cnt == cnt);
+    ENSURE(GF_ATOMIC_GET(layout->ref) == 1);
 out:
-	return layout;
+    return layout;
 }
 
-
 dht_layout_t *
-dht_layout_get (xlator_t *this, inode_t *inode)
+dht_layout_get(xlator_t *this, inode_t *inode)
 {
-        dht_conf_t   *conf = NULL;
-        uint64_t      layout_int = 0;
-        dht_layout_t *layout = NULL;
-        int           ret    = -1;
-
-        conf = this->private;
-        if (!conf)
-                goto out;
-
-        LOCK (&conf->layout_lock);
-        {
-                ret = inode_ctx_get (inode, this, &layout_int);
-                if (ret == 0) {
-                        layout = (dht_layout_t *) (unsigned long) layout_int;
-                        layout->ref++;
-                }
-        }
-        UNLOCK (&conf->layout_lock);
-
-out:
-        return layout;
+    dht_layout_t *layout = NULL;
+    int ret = 0;
+
+    ret = dht_inode_ctx_layout_get(inode, this, &layout);
+    if ((!ret) && layout) {
+        GF_ATOMIC_INC(layout->ref);
+    }
+    return layout;
 }
 
-
 int
-dht_layout_set (xlator_t *this, inode_t *inode, dht_layout_t *layout)
+dht_layout_set(xlator_t *this, inode_t *inode, dht_layout_t *layout)
 {
-        dht_conf_t   *conf = NULL;
-        int           oldret = -1;
-        int           ret = 0;
-        dht_layout_t *old_layout;
-        uint64_t      old_layout_int;
-
-        conf = this->private;
-        if (!conf)
-                goto out;
-
-        LOCK (&conf->layout_lock);
-        {
-                oldret = inode_ctx_get (inode, this, &old_layout_int);
-
-                layout->ref++;
-                ret = inode_ctx_put (inode, this, (uint64_t) (unsigned long)
-                                     layout);
-        }
-        UNLOCK (&conf->layout_lock);
-
-        if (oldret == 0) {
-                old_layout = (dht_layout_t *) (unsigned long) old_layout_int;
-                dht_layout_unref (this, old_layout);
-        }
+    dht_conf_t *conf = NULL;
+    int oldret = -1;
+    int ret = -1;
+    dht_layout_t *old_layout;
+
+    conf = this->private;
+    if (!conf || !layout)
+        goto out;
+
+    LOCK(&conf->layout_lock);
+    {
+        oldret = dht_inode_ctx_layout_get(inode, this, &old_layout);
+        if (layout)
+            GF_ATOMIC_INC(layout->ref);
+        ret = dht_inode_ctx_layout_set(inode, this, layout);
+    }
+    UNLOCK(&conf->layout_lock);
+
+    if (!oldret) {
+        dht_layout_unref(this, old_layout);
+    }
+    if (ret)
+        GF_ATOMIC_DEC(layout->ref);
 
 out:
-        return ret;
+    return ret;
 }
 
-
 void
-dht_layout_unref (xlator_t *this, dht_layout_t *layout)
+dht_layout_unref(xlator_t *this, dht_layout_t *layout)
 {
-        dht_conf_t  *conf = NULL;
-        int          ref = 0;
+    int ref = 0;
 
-        if (layout->preset || !this->private)
-                return;
+    if (!layout || layout->preset || !this->private)
+        return;
 
-        conf = this->private;
-
-        LOCK (&conf->layout_lock);
-        {
-                ref = --layout->ref;
-        }
-        UNLOCK (&conf->layout_lock);
+    ref = GF_ATOMIC_DEC(layout->ref);
 
-        if (!ref)
-                GF_FREE (layout);
+    if (!ref)
+        GF_FREE(layout);
 }
 
-
 dht_layout_t *
-dht_layout_ref (xlator_t *this, dht_layout_t *layout)
+dht_layout_ref(xlator_t *this, dht_layout_t *layout)
 {
-        dht_conf_t  *conf = NULL;
-
-        if (layout->preset || !this->private)
-                return layout;
+    if (layout->preset || !this->private)
+        return layout;
 
-        conf = this->private;
-        LOCK (&conf->layout_lock);
-        {
-                layout->ref++;
-        }
-        UNLOCK (&conf->layout_lock);
+    GF_ATOMIC_INC(layout->ref);
 
-        return layout;
+    return layout;
 }
 
-
 xlator_t *
-dht_layout_search (xlator_t *this, dht_layout_t *layout, const char *name)
+dht_layout_search(xlator_t *this, dht_layout_t *layout, const char *name)
 {
-	uint32_t   hash = 0;
-        xlator_t  *subvol = NULL;
-	int        i = 0;
-	int        ret = 0;
-
-
-	ret = dht_hash_compute (layout->type, name, &hash);
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"hash computation failed for type=%d name=%s",
-			layout->type, name);
-		goto out;
-	}
-
-	for (i = 0; i < layout->cnt; i++) {
-		if (layout->list[i].start <= hash
-		    && layout->list[i].stop >= hash) {
-			subvol = layout->list[i].xlator;
-			break;
-		}
-	}
-
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no subvolume for hash (value) = %u", hash);
-	}
+    uint32_t hash = 0;
+    xlator_t *subvol = NULL;
+    int i = 0;
+    int ret = 0;
+
+    ret = dht_hash_compute(this, layout->type, name, &hash);
+    if (ret != 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_COMPUTE_HASH_FAILED,
+                "type=%d", layout->type, "name=%s", name, NULL);
+        goto out;
+    }
+
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].start <= hash && layout->list[i].stop >= hash) {
+            subvol = layout->list[i].xlator;
+            break;
+        }
+    }
+
+    if (!subvol) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+                "hash-value=0x%x", hash, NULL);
+    }
 
 out:
-	return subvol;
+    return subvol;
 }
 
-
 dht_layout_t *
-dht_layout_for_subvol (xlator_t *this, xlator_t *subvol)
+dht_layout_for_subvol(xlator_t *this, xlator_t *subvol)
 {
-	dht_conf_t   *conf = NULL;
-	dht_layout_t *layout = NULL;
-	int           i = 0;
-
-	conf = this->private;
-        if (!conf)
-                goto out;
-
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		if (conf->subvolumes[i] == subvol) {
-			layout = conf->file_layouts[i];
-			break;
-		}
-	}
+    dht_conf_t *conf = NULL;
+    dht_layout_t *layout = NULL;
+    int i = 0;
+
+    conf = this->private;
+    if (!conf)
+        goto out;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (conf->subvolumes[i] == subvol) {
+            layout = conf->file_layouts[i];
+            break;
+        }
+    }
 
 out:
-	return layout;
+    return layout;
 }
 
-
 int
-dht_layouts_init (xlator_t *this, dht_conf_t *conf)
+dht_layouts_init(xlator_t *this, dht_conf_t *conf)
 {
-	dht_layout_t *layout = NULL;
-	int           i = 0;
-	int           ret = -1;
+    dht_layout_t *layout = NULL;
+    int i = 0;
+    int ret = -1;
 
-        if (!conf)
-                goto out;
+    if (!conf)
+        goto out;
 
-	conf->file_layouts = GF_CALLOC (conf->subvolume_cnt,
-				        sizeof (dht_layout_t *),
-                                        gf_dht_mt_dht_layout_t);
-	if (!conf->file_layouts) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto out;
-	}
+    conf->file_layouts = GF_CALLOC(conf->subvolume_cnt, sizeof(dht_layout_t *),
+                                   gf_dht_mt_dht_layout_t);
+    if (!conf->file_layouts) {
+        goto out;
+    }
 
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		layout = dht_layout_new (this, 1);
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        layout = dht_layout_new(this, 1);
 
-		if (!layout) {
-			goto out;
-		}
+        if (!layout) {
+            goto out;
+        }
 
-		layout->preset = 1;
+        layout->preset = 1;
 
-		layout->list[0].xlator = conf->subvolumes[i];
+        layout->list[0].xlator = conf->subvolumes[i];
 
-		conf->file_layouts[i] = layout;
-	}
+        conf->file_layouts[i] = layout;
+    }
 
-	ret = 0;
+    ret = 0;
 out:
-	return ret;
+    return ret;
 }
 
-
 int
-dht_disk_layout_extract (xlator_t *this, dht_layout_t *layout,
-			 int pos, int32_t **disk_layout_p)
+dht_disk_layout_extract(xlator_t *this, dht_layout_t *layout, int pos,
+                        int32_t **disk_layout_p)
 {
-	int      ret = -1;
-	int32_t *disk_layout = NULL;
-
-	disk_layout = GF_CALLOC (5, sizeof (int),
-                                 gf_dht_mt_int32_t);
-	if (!disk_layout) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto out;
-	}
-
-	disk_layout[0] = hton32 (1);
-	disk_layout[1] = hton32 (layout->type);
-	disk_layout[2] = hton32 (layout->list[pos].start);
-	disk_layout[3] = hton32 (layout->list[pos].stop);
-
-	if (disk_layout_p)
-		*disk_layout_p = disk_layout;
-	ret = 0;
+    int ret = -1;
+    int32_t *disk_layout = NULL;
+
+    disk_layout = GF_CALLOC(5, sizeof(int), gf_dht_mt_int32_t);
+    if (!disk_layout) {
+        goto out;
+    }
+
+    disk_layout[0] = hton32(layout->list[pos].commit_hash);
+    disk_layout[1] = hton32(layout->type);
+    disk_layout[2] = hton32(layout->list[pos].start);
+    disk_layout[3] = hton32(layout->list[pos].stop);
+
+    if (disk_layout_p)
+        *disk_layout_p = disk_layout;
+    else
+        GF_FREE(disk_layout);
+
+    ret = 0;
 
 out:
-	return ret;
+    return ret;
 }
 
-
 int
-dht_disk_layout_merge (xlator_t *this, dht_layout_t *layout,
-		       int pos, void *disk_layout_raw)
+dht_disk_layout_extract_for_subvol(xlator_t *this, dht_layout_t *layout,
+                                   xlator_t *subvol, int32_t **disk_layout_p)
 {
-	int      cnt = 0;
-	int      type = 0;
-	int      start_off = 0;
-	int      stop_off = 0;
-        int      disk_layout[4];
+    int i = 0;
 
-	/* TODO: assert disk_layout_ptr is of required length */
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].xlator == subvol)
+            break;
+    }
 
-        memcpy (disk_layout, disk_layout_raw, sizeof (disk_layout));
+    if (i == layout->cnt)
+        return -1;
 
-	cnt  = ntoh32 (disk_layout[0]);
-	if (cnt != 1) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"disk layout has invalid count %d", cnt);
-		return -1;
-	}
+    return dht_disk_layout_extract(this, layout, i, disk_layout_p);
+}
 
-	/* TODO: assert type is compatible */
-	type      = ntoh32 (disk_layout[1]);
-	start_off = ntoh32 (disk_layout[2]);
-	stop_off  = ntoh32 (disk_layout[3]);
+static int
+dht_disk_layout_merge(xlator_t *this, dht_layout_t *layout, int pos,
+                      void *disk_layout_raw, int disk_layout_len)
+{
+    int type = 0;
+    int start_off = 0;
+    int stop_off = 0;
+    int commit_hash = 0;
+    int disk_layout[4];
+
+    if (!disk_layout_raw) {
+        gf_smsg(this->name, GF_LOG_CRITICAL, 0, DHT_MSG_LAYOUT_MERGE_FAILED,
+                NULL);
+        return -1;
+    }
+
+    GF_ASSERT(disk_layout_len == sizeof(disk_layout));
+
+    memcpy(disk_layout, disk_layout_raw, disk_layout_len);
+
+    type = ntoh32(disk_layout[1]);
+    switch (type) {
+        case DHT_HASH_TYPE_DM_USER:
+            gf_msg_debug(this->name, 0, "found user-set layout");
+            layout->type = type;
+            /* Fall through. */
+        case DHT_HASH_TYPE_DM:
+            break;
+        default:
+            gf_smsg(this->name, GF_LOG_CRITICAL, 0, DHT_MSG_INVALID_DISK_LAYOUT,
+                    "layout=%d", disk_layout[1], NULL);
+            return -1;
+    }
+
+    commit_hash = ntoh32(disk_layout[0]);
+    start_off = ntoh32(disk_layout[2]);
+    stop_off = ntoh32(disk_layout[3]);
+
+    layout->list[pos].commit_hash = commit_hash;
+    layout->list[pos].start = start_off;
+    layout->list[pos].stop = stop_off;
+
+    gf_msg_trace(this->name, 0,
+                 "merged to layout: 0x%x - 0x%x (hash 0x%x, type %d) from %s",
+                 start_off, stop_off, commit_hash, type,
+                 layout->list[pos].xlator->name);
+
+    return 0;
+}
 
-	layout->list[pos].start = start_off;
-	layout->list[pos].stop  = stop_off;
+int
+dht_layout_merge(xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
+                 int op_ret, int op_errno, dict_t *xattr)
+{
+    int i = 0;
+    int ret = -1;
+    int err = -1;
+    void *disk_layout_raw = NULL;
+    int disk_layout_len = 0;
+    dht_conf_t *conf = this->private;
+
+    if (op_ret != 0) {
+        err = op_errno;
+    }
+
+    if (!layout)
+        goto out;
+
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].xlator == NULL) {
+            layout->list[i].err = err;
+            layout->list[i].xlator = subvol;
+            break;
+        }
+    }
 
-	gf_log (this->name, GF_LOG_TRACE,
-		"merged to layout: %u - %u (type %d) from %s",
-		start_off, stop_off, type,
-		layout->list[pos].xlator->name);
+    if (op_ret != 0) {
+        ret = 0;
+        goto out;
+    }
+
+    if (xattr) {
+        /* during lookup and not mkdir */
+        ret = dict_get_ptr_and_len(xattr, conf->xattr_name, &disk_layout_raw,
+                                   &disk_layout_len);
+    }
+
+    if (ret != 0) {
+        layout->list[i].err = 0;
+        gf_msg_trace(this->name, 0, "Missing disk layout on %s. err = %d",
+                     subvol->name, err);
+        ret = 0;
+        goto out;
+    }
 
-	return 0;
-}
+    ret = dht_disk_layout_merge(this, layout, i, disk_layout_raw,
+                                disk_layout_len);
+    if (ret != 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_MERGE_FAILED,
+                "subvolume=%s", subvol->name, NULL);
+        goto out;
+    }
 
+    if (layout->commit_hash == 0) {
+        layout->commit_hash = layout->list[i].commit_hash;
+    } else if (layout->commit_hash != layout->list[i].commit_hash) {
+        layout->commit_hash = DHT_LAYOUT_HASH_INVALID;
+    }
 
-int
-dht_layout_merge (xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
-		  int op_ret, int op_errno, dict_t *xattr)
-{
-	int      i     = 0;
-	int      ret   = -1;
-	int      err   = -1;
-	void    *disk_layout_raw = NULL;
-
-
-	if (op_ret != 0) {
-		err = op_errno;
-	}
-
-	for (i = 0; i < layout->cnt; i++) {
-		if (layout->list[i].xlator == NULL) {
-			layout->list[i].err    = err;
-			layout->list[i].xlator = subvol;
-			break;
-		}
-	}
-
-	if (op_ret != 0) {
-		ret = 0;
-		goto out;
-	}
-
-	if (xattr) {
-		/* during lookup and not mkdir */
-		ret = dict_get_ptr (xattr, "trusted.glusterfs.dht",
-				    &disk_layout_raw);
-	}
-
-	if (ret != 0) {
-		layout->list[i].err = -1;
-		gf_log (this->name, GF_LOG_TRACE,
-			"missing disk layout on %s. err = %d",
-			subvol->name, err);
-		ret = 0;
-		goto out;
-	}
-
-	ret = dht_disk_layout_merge (this, layout, i, disk_layout_raw);
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"layout merge from subvolume %s failed",
-			subvol->name);
-		goto out;
-	}
-	layout->list[i].err = 0;
+    layout->list[i].err = 0;
 
 out:
-	return ret;
+    return ret;
 }
 
+void
+dht_layout_entry_swap(dht_layout_t *layout, int i, int j)
+{
+    uint32_t start_swap = 0;
+    uint32_t stop_swap = 0;
+    uint32_t commit_hash_swap = 0;
+    xlator_t *xlator_swap = 0;
+    int err_swap = 0;
+
+    start_swap = layout->list[i].start;
+    stop_swap = layout->list[i].stop;
+    xlator_swap = layout->list[i].xlator;
+    err_swap = layout->list[i].err;
+    commit_hash_swap = layout->list[i].commit_hash;
+
+    layout->list[i].start = layout->list[j].start;
+    layout->list[i].stop = layout->list[j].stop;
+    layout->list[i].xlator = layout->list[j].xlator;
+    layout->list[i].err = layout->list[j].err;
+    layout->list[i].commit_hash = layout->list[j].commit_hash;
+
+    layout->list[j].start = start_swap;
+    layout->list[j].stop = stop_swap;
+    layout->list[j].xlator = xlator_swap;
+    layout->list[j].err = err_swap;
+    layout->list[j].commit_hash = commit_hash_swap;
+}
 
 void
-dht_layout_entry_swap (dht_layout_t *layout, int i, int j)
+dht_layout_range_swap(dht_layout_t *layout, int i, int j)
 {
-	uint32_t  start_swap = 0;
-	uint32_t  stop_swap = 0;
-	xlator_t *xlator_swap = 0;
-	int       err_swap = 0;
-
-
-	start_swap  = layout->list[i].start;
-	stop_swap   = layout->list[i].stop;
-	xlator_swap = layout->list[i].xlator;
-	err_swap    = layout->list[i].err;
-
-	layout->list[i].start  = layout->list[j].start;
-	layout->list[i].stop   = layout->list[j].stop;
-	layout->list[i].xlator = layout->list[j].xlator;
-	layout->list[i].err    = layout->list[j].err;
-
-	layout->list[j].start  = start_swap;
-	layout->list[j].stop   = stop_swap;
-	layout->list[j].xlator = xlator_swap;
-	layout->list[j].err    = err_swap;
+    uint32_t start_swap = 0;
+    uint32_t stop_swap = 0;
+
+    start_swap = layout->list[i].start;
+    stop_swap = layout->list[i].stop;
+
+    layout->list[i].start = layout->list[j].start;
+    layout->list[i].stop = layout->list[j].stop;
+
+    layout->list[j].start = start_swap;
+    layout->list[j].stop = stop_swap;
+}
+static int64_t
+dht_layout_entry_cmp_volname(dht_layout_t *layout, int i, int j)
+{
+    return (strcmp(layout->list[i].xlator->name, layout->list[j].xlator->name));
 }
 
-int64_t
-dht_layout_entry_cmp_volname (dht_layout_t *layout, int i, int j)
+gf_boolean_t
+dht_is_subvol_in_layout(dht_layout_t *layout, xlator_t *xlator)
 {
-	return (strcmp (layout->list[i].xlator->name, 
-			layout->list[j].xlator->name));
+    int i = 0;
+
+    for (i = 0; i < layout->cnt; i++) {
+        /* Check if xlator is already part of layout, and layout is
+         * non-zero. */
+        if (!strcmp(layout->list[i].xlator->name, xlator->name)) {
+            if (layout->list[i].start != layout->list[i].stop)
+                return _gf_true;
+            break;
+        }
+    }
+    return _gf_false;
 }
 
-int64_t
-dht_layout_entry_cmp (dht_layout_t *layout, int i, int j)
+static int64_t
+dht_layout_entry_cmp(dht_layout_t *layout, int i, int j)
 {
-	int64_t diff = 0;
+    int64_t diff = 0;
 
-	if (layout->list[i].err || layout->list[j].err)
-		diff = layout->list[i].err - layout->list[j].err;
-	else
-		diff = (int64_t) layout->list[i].start
-			- (int64_t) layout->list[j].start;
+    /* swap zero'ed out layouts to front, if needed */
+    if (!layout->list[j].start && !layout->list[j].stop) {
+        diff = (int64_t)layout->list[i].stop - (int64_t)layout->list[j].stop;
+        goto out;
+    }
+    diff = (int64_t)layout->list[i].start - (int64_t)layout->list[j].start;
 
-	return diff;
+out:
+    return diff;
 }
 
-
 int
-dht_layout_sort (dht_layout_t *layout)
+dht_layout_sort(dht_layout_t *layout)
 {
-	int       i = 0;
-	int       j = 0;
-	int64_t   ret = 0;
+    int i = 0;
+    int j = 0;
+    int64_t ret = 0;
 
-	/* TODO: O(n^2) -- bad bad */
+    /* TODO: O(n^2) -- bad bad */
 
-	for (i = 0; i < layout->cnt - 1; i++) {
-		for (j = i + 1; j < layout->cnt; j++) {
-			ret = dht_layout_entry_cmp (layout, i, j);
-			if (ret > 0)
-				dht_layout_entry_swap (layout, i, j);
-		}
-	}
+    for (i = 0; i < layout->cnt - 1; i++) {
+        for (j = i + 1; j < layout->cnt; j++) {
+            ret = dht_layout_entry_cmp(layout, i, j);
+            if (ret > 0)
+                dht_layout_entry_swap(layout, i, j);
+        }
+    }
 
-	return 0;
+    return 0;
 }
 
-int
-dht_layout_sort_volname (dht_layout_t *layout)
+void
+dht_layout_sort_volname(dht_layout_t *layout)
+{
+    int i = 0;
+    int j = 0;
+    int64_t ret = 0;
+
+    /* TODO: O(n^2) -- bad bad */
+
+    for (i = 0; i < layout->cnt - 1; i++) {
+        for (j = i + 1; j < layout->cnt; j++) {
+            ret = dht_layout_entry_cmp_volname(layout, i, j);
+            if (ret > 0)
+                dht_layout_entry_swap(layout, i, j);
+        }
+    }
+}
+
+void
+dht_layout_anomalies(xlator_t *this, loc_t *loc, dht_layout_t *layout,
+                     uint32_t *holes_p, uint32_t *overlaps_p,
+                     uint32_t *missing_p, uint32_t *down_p, uint32_t *misc_p,
+                     uint32_t *no_space_p)
 {
-	int       i = 0;
-	int       j = 0;
-	int64_t   ret = 0;
+    uint32_t overlaps = 0;
+    uint32_t missing = 0;
+    uint32_t down = 0;
+    uint32_t misc = 0;
+    uint32_t hole_cnt = 0;
+    uint32_t overlap_cnt = 0;
+    int i = 0;
+    uint32_t prev_stop = 0;
+    uint32_t last_stop = 0;
+    char is_virgin = 1;
+    uint32_t no_space = 0;
+
+    /* This function scans through the layout spread of a directory to
+       check if there are any anomalies. Prior to calling this function
+       the layout entries should be sorted in the ascending order.
+
+       If the layout entry has err != 0
+            then increment the corresponding anomaly.
+       else
+            if (start of the current layout entry > stop + 1 of previous
+               non erroneous layout entry)
+                    then it indicates a hole in the layout
+            if (start of the current layout entry < stop + 1 of previous
+                non erroneous layout entry)
+                     then it indicates an overlap in the layout
+    */
+    last_stop = layout->list[0].start - 1;
+    prev_stop = last_stop;
+
+    for (i = 0; i < layout->cnt; i++) {
+        switch (layout->list[i].err) {
+            case -1:
+            case ENOENT:
+            case ESTALE:
+                missing++;
+                continue;
+            case ENOTCONN:
+                down++;
+                continue;
+            case ENOSPC:
+                no_space++;
+                continue;
+            case 0:
+                /* if err == 0 and start == stop, then it is a non misc++;
+                 * participating subvolume(spread-cnt). Then, do not
+                 * check for anomalies. If start != stop, then treat it
+                 * as misc err */
+                if (layout->list[i].start == layout->list[i].stop) {
+                    continue;
+                }
+                break;
+            default:
+                misc++;
+                continue;
+        }
+
+        is_virgin = 0;
+
+        if ((prev_stop + 1) < layout->list[i].start) {
+            hole_cnt++;
+        }
+
+        if ((prev_stop + 1) > layout->list[i].start) {
+            overlap_cnt++;
+            overlaps += ((prev_stop + 1) - layout->list[i].start);
+        }
+        prev_stop = layout->list[i].stop;
+    }
 
-	/* TODO: O(n^2) -- bad bad */
+    if ((last_stop - prev_stop) || is_virgin)
+        hole_cnt++;
 
-	for (i = 0; i < layout->cnt - 1; i++) {
-		for (j = i + 1; j < layout->cnt; j++) {
-			ret = dht_layout_entry_cmp_volname (layout, i, j);
-			if (ret > 0)
-				dht_layout_entry_swap (layout, i, j);
-		}
-	}
+    if (holes_p)
+        *holes_p = hole_cnt;
 
-	return 0;
+    if (overlaps_p)
+        *overlaps_p = overlap_cnt;
+
+    if (missing_p)
+        *missing_p = missing;
+
+    if (down_p)
+        *down_p = down;
+
+    if (misc_p)
+        *misc_p = misc;
+
+    if (no_space_p)
+        *no_space_p = no_space;
 }
 
+int
+dht_layout_missing_dirs(dht_layout_t *layout)
+{
+    int i = 0, missing = 0;
+
+    if (layout == NULL)
+        goto out;
+
+    for (i = 0; i < layout->cnt; i++) {
+        if ((layout->list[i].err == ENOENT) ||
+            ((layout->list[i].err == -1) && (layout->list[i].start == 0) &&
+             (layout->list[i].stop == 0))) {
+            missing++;
+        }
+    }
+
+out:
+    return missing;
+}
 
 int
-dht_layout_anomalies (xlator_t *this, loc_t *loc, dht_layout_t *layout,
-		      uint32_t *holes_p, uint32_t *overlaps_p,
-		      uint32_t *missing_p, uint32_t *down_p, uint32_t *misc_p)
+dht_layout_normalize(xlator_t *this, loc_t *loc, dht_layout_t *layout)
 {
-	uint32_t    overlaps = 0;
-	uint32_t    missing  = 0;
-	uint32_t    down     = 0;
-	uint32_t    misc     = 0;
-	uint32_t    hole_cnt = 0;
-	uint32_t    overlap_cnt = 0;
-	int         i = 0;
-	int         ret = 0;
-	uint32_t    prev_stop = 0;
-	uint32_t    last_stop = 0;
-	char        is_virgin = 1;
-
-	/* TODO: explain WTF is happening */
-
-	last_stop = layout->list[0].start - 1;
-	prev_stop = last_stop;
-
-	for (i = 0; i < layout->cnt; i++) {
-		if (layout->list[i].err) {
-			switch (layout->list[i].err) {
-			case -1:
-			case ENOENT:
-				missing++;
-				break;
-			case ENOTCONN:
-				down++;
-				break;
-			case ENOSPC:
-				down++;
-				break;
-			default:
-				misc++;
-			}
-			continue;
-		}
-
-		is_virgin = 0;
-
-		if ((prev_stop + 1) < layout->list[i].start) {
-			hole_cnt++;
-		}
-
-		if ((prev_stop + 1) > layout->list[i].start) {
-			overlap_cnt++;
-			overlaps += ((prev_stop + 1) - layout->list[i].start);
-		}
-		prev_stop = layout->list[i].stop;
-	}
-
-	if ((last_stop - prev_stop) || is_virgin)
-	    hole_cnt++;
-
-	if (holes_p)
-		*holes_p = hole_cnt;
-
-	if (overlaps_p)
-		*overlaps_p = overlap_cnt;
-
-	if (missing_p)
-		*missing_p = missing;
-
-	if (down_p)
-		*down_p = down;
-
-	if (misc_p)
-		*misc_p = misc;
-
-	return ret;
+    int ret = 0;
+    uint32_t holes = 0;
+    uint32_t overlaps = 0;
+    uint32_t missing = 0;
+    uint32_t down = 0;
+    uint32_t misc = 0, missing_dirs = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    ret = dht_layout_sort(layout);
+    if (ret == -1) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_SORT_FAILED,
+                NULL);
+        goto out;
+    }
+
+    gf_uuid_unparse(loc->gfid, gfid);
+
+    dht_layout_anomalies(this, loc, layout, &holes, &overlaps, &missing, &down,
+                         &misc, NULL);
+
+    if (holes || overlaps) {
+        if (missing == layout->cnt) {
+            gf_msg_debug(this->name, 0,
+                         "Directory %s looked up first time"
+                         " gfid = %s",
+                         loc->path, gfid);
+        } else {
+            gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_ANOMALIES_INFO,
+                    "path=%s", loc->path, "gfid=%s", gfid, "holes=%d", holes,
+                    "overlaps=%d", overlaps, NULL);
+        }
+        ret = -1;
+    }
+
+    if (ret >= 0) {
+        missing_dirs = dht_layout_missing_dirs(layout);
+        /* TODO During DHT selfheal rewrite (almost) find a better place
+         * to detect this - probably in dht_layout_anomalies()
+         */
+        if (missing_dirs > 0)
+            ret += missing_dirs;
+    }
+
+out:
+    return ret;
 }
 
+int
+dht_dir_has_layout(dict_t *xattr, char *name)
+{
+    void *disk_layout_raw = NULL;
+
+    return dict_get_ptr(xattr, name, &disk_layout_raw);
+}
 
 int
-dht_layout_normalize (xlator_t *this, loc_t *loc, dht_layout_t *layout)
+dht_layout_dir_mismatch(xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
+                        loc_t *loc, dict_t *xattr)
 {
-	int          ret   = 0;
-	int          i = 0;
-	uint32_t     holes = 0;
-	uint32_t     overlaps = 0;
-	uint32_t     missing = 0;
-	uint32_t     down = 0;
-	uint32_t     misc = 0;
-
-
-	ret = dht_layout_sort (layout);
-	if (ret == -1) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"sort failed?! how the ....");
-		goto out;
-	}
-
-	ret = dht_layout_anomalies (this, loc, layout,
-				    &holes, &overlaps,
-				    &missing, &down, &misc);
-	if (ret == -1) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"error while finding anomalies in %s -- not good news",
-			loc->path);
-		goto out;
-	}
-
-	if (holes || overlaps) {
-		if (missing == layout->cnt) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"directory %s looked up first time",
-				loc->path);
-		} else {
-			gf_log (this->name, GF_LOG_INFO,
-				"found anomalies in %s. holes=%d overlaps=%d",
-				loc->path, holes, overlaps);
-		}
-		ret = 1;
-	}
-
-	for (i = 0; i < layout->cnt; i++) {
-	/* TODO During DHT selfheal rewrite (almost) find a better place to 
-	 * detect this - probably in dht_layout_anomalies() 
-	 */
-		if (layout->list[i].err > 0) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"path=%s err=%s on subvol=%s",
-                                loc->path, strerror (layout->list[i].err),
-                                (layout->list[i].xlator ?
-                                 layout->list[i].xlator->name : "<>"));
-                        if (layout->list[i].err == ENOENT)
-                                ret = 1;
-		}
-	}
+    int idx = 0;
+    int pos = -1;
+    int ret = 0;
+    int err = 0;
+    int dict_ret = 0;
+    int32_t disk_layout[4];
+    void *disk_layout_raw = NULL;
+    uint32_t start_off = -1;
+    uint32_t stop_off = -1;
+    uint32_t commit_hash = -1;
+    dht_conf_t *conf = this->private;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    if (loc && loc->inode)
+        gf_uuid_unparse(loc->inode->gfid, gfid);
+
+    for (idx = 0; idx < layout->cnt; idx++) {
+        if (layout->list[idx].xlator == subvol) {
+            pos = idx;
+            break;
+        }
+    }
 
+    if (pos == -1) {
+        if (loc) {
+            gf_msg_debug(this->name, 0, "%s - no layout info for subvolume %s",
+                         loc ? loc->path : "path not found", subvol->name);
+        }
+        ret = 1;
+        goto out;
+    }
+
+    err = layout->list[pos].err;
+
+    if (!xattr) {
+        if (err == 0) {
+            if (loc) {
+                gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_XATTR_DICT_NULL,
+                        "path=%s", loc->path, NULL);
+            } else {
+                gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_XATTR_DICT_NULL,
+                        "path not found", NULL);
+            }
+            ret = -1;
+        }
+        goto out;
+    }
+
+    dict_ret = dict_get_ptr(xattr, conf->xattr_name, &disk_layout_raw);
+
+    if (dict_ret < 0) {
+        if (err == 0 && layout->list[pos].stop) {
+            if (loc) {
+                gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_DISK_LAYOUT_MISSING,
+                        "path=%s", loc->path, "gfid=%s", gfid, NULL);
+            } else {
+                gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_DISK_LAYOUT_MISSING,
+                        "path not found"
+                        "gfid=%s",
+                        gfid, NULL);
+            }
+            ret = -1;
+        }
+        goto out;
+    }
+
+    memcpy(disk_layout, disk_layout_raw, sizeof(disk_layout));
+
+    start_off = ntoh32(disk_layout[2]);
+    stop_off = ntoh32(disk_layout[3]);
+    commit_hash = ntoh32(disk_layout[0]);
+
+    if ((layout->list[pos].start != start_off) ||
+        (layout->list[pos].stop != stop_off) ||
+        (layout->list[pos].commit_hash != commit_hash)) {
+        gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_LAYOUT_INFO, "subvol=%s",
+                layout->list[pos].xlator->name, "inode-layout:start=0x%x",
+                layout->list[pos].start, "inode-layout:stop=0x%x",
+                layout->list[pos].stop, "layout-commit-hash=0x%x; ",
+                layout->list[pos].commit_hash, "disk-layout:start-off=0x%x",
+                start_off, "disk-layout:top-off=0x%x", stop_off,
+                "commit-hash=0x%x", commit_hash, NULL);
+        ret = 1;
+    } else {
+        ret = 0;
+    }
 out:
-	return ret;
+    return ret;
 }
 
-
 int
-dht_layout_dir_mismatch (xlator_t *this, dht_layout_t *layout, xlator_t *subvol,
-			 loc_t *loc, dict_t *xattr)
+dht_layout_preset(xlator_t *this, xlator_t *subvol, inode_t *inode)
 {
-	int       idx = 0;
-	int       pos = -1;
-	int       ret = 0;
-        int       err = 0;
-        int       dict_ret = 0;
-	int32_t   disk_layout[4];
-        void     *disk_layout_raw = NULL;
-	int32_t   count = -1;
-	uint32_t  start_off = -1;
-	uint32_t  stop_off = -1;
-
-
-	for (idx = 0; idx < layout->cnt; idx++) {
-		if (layout->list[idx].xlator == subvol) {
-			pos = idx;
-			break;
-		}
-	}
-	
-	if (pos == -1) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"%s - no layout info for subvolume %s",
-			loc->path, subvol->name);
-		ret = 1;
-		goto out;
-	}
-
-        err = layout->list[pos].err;
-
-	if (!xattr) {
-                if (err == 0) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s - xattr dictionary is NULL",
-                                loc->path);
-                        ret = -1;
-                }
-		goto out;
-	}
+    dht_layout_t *layout = NULL;
+    int ret = -1;
+    dht_conf_t *conf = NULL;
 
-	dict_ret = dict_get_ptr (xattr, "trusted.glusterfs.dht",
-                                 &disk_layout_raw);
+    conf = this->private;
+    if (!conf)
+        goto out;
 
-	if (dict_ret < 0) {
-                if (err == 0) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s - disk layout missing", loc->path);
-                        ret = -1;
-                }
-		goto out;
-	}
-
-        memcpy (disk_layout, disk_layout_raw, sizeof (disk_layout));
-
-	count  = ntoh32 (disk_layout[0]);
-	if (count != 1) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"%s - disk layout has invalid count %d",
-			loc->path, count);
-		ret = -1;
-		goto out;
-	}
-
-	start_off = ntoh32 (disk_layout[2]);
-	stop_off  = ntoh32 (disk_layout[3]);
-
-	if ((layout->list[pos].start != start_off)
-	    || (layout->list[pos].stop != stop_off)) {
-		gf_log (this->name, GF_LOG_INFO,
-			"subvol: %s; inode layout - %"PRIu32" - %"PRIu32"; "
-			"disk layout - %"PRIu32" - %"PRIu32,
-			layout->list[pos].xlator->name,
-			layout->list[pos].start, layout->list[pos].stop,
-			start_off, stop_off);
-		ret = 1;
-	} else {
-		ret = 0;
-	}
+    layout = dht_layout_for_subvol(this, subvol);
+    if (!layout) {
+        gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_NO_LAYOUT_INFO,
+                "subvolume=%s", subvol ? subvol->name : "<nil>", NULL);
+        ret = -1;
+        goto out;
+    }
+
+    gf_msg_debug(this->name, 0, "file = %s, subvol = %s",
+                 uuid_utoa(inode->gfid), subvol ? subvol->name : "<nil>");
+
+    LOCK(&conf->layout_lock);
+    {
+        dht_inode_ctx_layout_set(inode, this, layout);
+    }
+
+    UNLOCK(&conf->layout_lock);
+
+    ret = 0;
 out:
-	return ret;
+    return ret;
 }
 
-
 int
-dht_layout_preset (xlator_t *this, xlator_t *subvol, inode_t *inode)
+dht_layout_index_for_subvol(dht_layout_t *layout, xlator_t *subvol)
 {
-        dht_layout_t *layout = NULL;
-        int           ret = -1;
-        dht_conf_t   *conf = NULL;
-
-        conf = this->private;
-        if (!conf)
-                goto out;
-
-	layout = dht_layout_for_subvol (this, subvol);
-	if (!layout) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no pre-set layout for subvolume %s",
-			subvol ? subvol->name : "<nil>");
-		ret = -1;
-		goto out;
-	}
-
-        LOCK (&conf->layout_lock);
-        {
-                inode_ctx_put (inode, this, (uint64_t)(long)layout);
+    int i = 0, ret = -1;
+
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].xlator == subvol) {
+            ret = i;
+            break;
         }
-        UNLOCK (&conf->layout_lock);
+    }
 
-        ret = 0;
-out:
-        return ret;
+    return ret;
 }
diff --git a/xlators/cluster/dht/src/dht-linkfile.c b/xlators/cluster/dht/src/dht-linkfile.c
index c72579e47d9..89ec6cca56e 100644
--- a/xlators/cluster/dht/src/dht-linkfile.c
+++ b/xlators/cluster/dht/src/dht-linkfile.c
@@ -1,247 +1,328 @@
 /*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+#include <glusterfs/compat.h>
+#include "dht-common.h"
 
+static int
+dht_linkfile_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, inode_t *inode,
+                        struct iatt *stbuf, dict_t *xattr,
+                        struct iatt *postparent)
+{
+    char is_linkfile = 0;
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    prev = cookie;
+    conf = this->private;
+
+    if (op_ret)
+        goto out;
+
+    gf_uuid_unparse(local->loc.gfid, gfid);
+
+    is_linkfile = check_is_linkfile(inode, stbuf, xattr, conf->link_xattr_name);
+    if (!is_linkfile)
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_NOT_LINK_FILE_ERROR,
+                "name=%s", prev->name, "path=%s", local->loc.path, "gfid=%s",
+                gfid, NULL);
+out:
+    local->linkfile.linkfile_cbk(frame, cookie, this, op_ret, op_errno, inode,
+                                 stbuf, postparent, postparent, xattr);
+    return 0;
+}
 
-#include "glusterfs.h"
-#include "xlator.h"
-#include "compat.h"
-#include "dht-common.h"
+static int
+dht_linkfile_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, inode_t *inode,
+                        struct iatt *stbuf, struct iatt *preparent,
+                        struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *subvol = NULL;
+    dict_t *xattrs = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+
+    local = frame->local;
+
+    if (!op_ret)
+        local->linked = _gf_true;
+
+    FRAME_SU_UNDO(frame, dht_local_t);
+
+    if (op_ret && (op_errno == EEXIST)) {
+        conf = this->private;
+        subvol = cookie;
+        if (!subvol)
+            goto out;
+        xattrs = dict_new();
+        if (!xattrs)
+            goto out;
+        ret = dict_set_uint32(xattrs, conf->link_xattr_name, 256);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                    "mame=%s", conf->link_xattr_name, NULL);
+            goto out;
+        }
+
+        STACK_WIND_COOKIE(frame, dht_linkfile_lookup_cbk, subvol, subvol,
+                          subvol->fops->lookup, &local->linkfile.loc, xattrs);
+        if (xattrs)
+            dict_unref(xattrs);
+        return 0;
+    }
+out:
+    local->linkfile.linkfile_cbk(frame, cookie, this, op_ret, op_errno, inode,
+                                 stbuf, preparent, postparent, xdata);
+    if (xattrs)
+        dict_unref(xattrs);
+    return 0;
+}
 
+int
+dht_linkfile_create(call_frame_t *frame, fop_mknod_cbk_t linkfile_cbk,
+                    xlator_t *this, xlator_t *tovol, xlator_t *fromvol,
+                    loc_t *loc)
+{
+    dht_local_t *local = NULL;
+    dict_t *dict = NULL;
+    int need_unref = 0;
+    int ret = 0;
+    dht_conf_t *conf = this->private;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    local->linkfile.linkfile_cbk = linkfile_cbk;
+    local->linkfile.srcvol = tovol;
+    loc_copy(&local->linkfile.loc, loc);
+
+    local->linked = _gf_false;
+
+    dict = local->params;
+    if (!dict) {
+        dict = dict_new();
+        if (!dict)
+            goto out;
+        need_unref = 1;
+    }
+
+    if (!gf_uuid_is_null(local->gfid)) {
+        gf_uuid_unparse(local->gfid, gfid);
+
+        ret = dict_set_gfuuid(dict, "gfid-req", local->gfid, true);
+        if (ret)
+            gf_smsg("dht-linkfile", GF_LOG_INFO, 0, DHT_MSG_DICT_SET_FAILED,
+                    "path=%s", loc->path, "gfid=%s", gfid, NULL);
+    } else {
+        gf_uuid_unparse(loc->gfid, gfid);
+    }
+
+    ret = dict_set_str(dict, GLUSTERFS_INTERNAL_FOP_KEY, "yes");
+    if (ret)
+        gf_smsg("dht-linkfile", GF_LOG_INFO, 0, DHT_MSG_DICT_SET_FAILED,
+                "path=%s", loc->path, "key=%s", GLUSTERFS_INTERNAL_FOP_KEY,
+                "gfid=%s", gfid, NULL);
+
+    ret = dict_set_str(dict, conf->link_xattr_name, tovol->name);
+
+    if (ret < 0) {
+        gf_smsg(frame->this->name, GF_LOG_INFO, 0, DHT_MSG_CREATE_LINK_FAILED,
+                "path=%s", loc->path, "gfid=%s", gfid, NULL);
+        goto out;
+    }
+
+    local->link_subvol = fromvol;
+    /* Always create as root:root. dht_linkfile_attr_heal fixes the
+     * ownsership */
+    FRAME_SU_DO(frame, dht_local_t);
+    STACK_WIND_COOKIE(frame, dht_linkfile_create_cbk, fromvol, fromvol,
+                      fromvol->fops->mknod, loc, S_IFREG | DHT_LINKFILE_MODE, 0,
+                      0, dict);
+
+    if (need_unref && dict)
+        dict_unref(dict);
+
+    return 0;
+out:
+    local->linkfile.linkfile_cbk(frame, frame->this, frame->this, -1, ENOMEM,
+                                 loc->inode, NULL, NULL, NULL, NULL);
 
+    if (need_unref && dict)
+        dict_unref(dict);
+
+    return 0;
+}
 
 int
-dht_linkfile_xattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int op_ret, int op_errno)
+dht_linkfile_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno,
+                        struct iatt *preparent, struct iatt *postparent,
+                        dict_t *xdata)
 {
-	dht_local_t *local = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *subvol = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
 
+    local = frame->local;
+    subvol = cookie;
 
-	local = frame->local;
-	local->linkfile.linkfile_cbk (frame, cookie, this, op_ret, op_errno,
-				      local->linkfile.inode,
-				      &local->linkfile.stbuf, NULL, NULL);
+    if (op_ret == -1) {
+        gf_uuid_unparse(local->loc.gfid, gfid);
+        gf_smsg(this->name, GF_LOG_INFO, op_errno, DHT_MSG_UNLINK_FAILED,
+                "path=%s", local->loc.path, "gfid=%s", gfid, "subvolume=%s",
+                subvol->name, NULL);
+    }
 
-	return 0;
-}
+    DHT_STACK_DESTROY(frame);
 
+    return 0;
+}
 
 int
-dht_linkfile_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int op_ret, int op_errno, inode_t *inode,
-                         struct iatt *stbuf, struct iatt *preparent,
-                         struct iatt *postparent)
+dht_linkfile_unlink(call_frame_t *frame, xlator_t *this, xlator_t *subvol,
+                    loc_t *loc)
 {
-	dht_local_t  *local = NULL;
-	call_frame_t *prev = NULL;
-	dict_t       *xattr = NULL;
-	data_t       *str_data = NULL;
-	int           ret = -1;
-
-	local = frame->local;
-	prev  = cookie;
-
-	if (op_ret == -1)
-		goto err;
-
-	xattr = get_new_dict ();
-	if (!xattr) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		op_errno = ENOMEM;
-		goto err;
-	}
-
-	local->linkfile.xattr = dict_ref (xattr);
-	local->linkfile.inode = inode_ref (inode);
-
-	str_data = str_to_data (local->linkfile.srcvol->name);
-	if (!str_data) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		op_errno = ENOMEM;
-		goto err;
-	}
-
-	ret = dict_set (xattr, "trusted.glusterfs.dht.linkto", str_data);
-	if (ret < 0) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"failed to initialize linkfile data");
-	}
-	str_data = NULL;
-
-	local->linkfile.stbuf = *stbuf;
-
-	STACK_WIND (frame, dht_linkfile_xattr_cbk,
-		    prev->this, prev->this->fops->setxattr,
-		    &local->linkfile.loc, local->linkfile.xattr, 0);
-
-	return 0;
+    call_frame_t *unlink_frame = NULL;
+    dht_local_t *unlink_local = NULL;
+
+    unlink_frame = copy_frame(frame);
+    if (!unlink_frame) {
+        goto err;
+    }
+
+    /* Using non-fop value here, as anyways, 'local->fop' is not used in
+       this particular case */
+    unlink_local = dht_local_init(unlink_frame, loc, NULL, GF_FOP_MAXVALUE);
+    if (!unlink_local) {
+        goto err;
+    }
 
+    STACK_WIND_COOKIE(unlink_frame, dht_linkfile_unlink_cbk, subvol, subvol,
+                      subvol->fops->unlink, &unlink_local->loc, 0, NULL);
+
+    return 0;
 err:
-	if (str_data) {
-		data_destroy (str_data);
-		str_data = NULL;
-	}
-
-        local->linkfile.linkfile_cbk (frame, cookie, this, op_ret, op_errno,
-                                      inode, stbuf, preparent, postparent);
-	return 0;
-}
+    if (unlink_frame)
+        DHT_STACK_DESTROY(unlink_frame);
 
+    return -1;
+}
 
-int
-dht_linkfile_create (call_frame_t *frame, fop_mknod_cbk_t linkfile_cbk,
-		     xlator_t *tovol, xlator_t *fromvol, loc_t *loc)
+xlator_t *
+dht_linkfile_subvol(xlator_t *this, inode_t *inode, struct iatt *stbuf,
+                    dict_t *xattr)
 {
-	dht_local_t *local = NULL;
-        dict_t      *dict = NULL;
-        int          ret = 0;
-
-	local = frame->local;
-	local->linkfile.linkfile_cbk = linkfile_cbk;
-	local->linkfile.srcvol = tovol;
-	loc_copy (&local->linkfile.loc, loc);
-
-        if (!uuid_is_null (local->gfid)) {
-                dict = dict_new ();
-                if (!dict)
-                        goto out;
-                ret = dict_set_static_bin (dict, "gfid-req", local->gfid, 16);
-                if (ret)
-                        gf_log ("dht-linkfile", GF_LOG_DEBUG, "gfid set failed");
-        } else if (local->params) {
-                dict = dict_ref (local->params);
-        }
-        if (!dict)
-                gf_log ("", GF_LOG_DEBUG,
-                        "dict is NULL, need to make sure gfid's are same");
+    dht_conf_t *conf = NULL;
+    xlator_t *subvol = NULL;
+    void *volname = NULL;
+    int i = 0, ret = 0;
 
-	STACK_WIND (frame, dht_linkfile_create_cbk,
-		    fromvol, fromvol->fops->mknod, loc,
-		    S_IFREG | DHT_LINKFILE_MODE, 0, dict);
+    conf = this->private;
 
-        if (dict)
-                dict_unref (dict);
+    if (!xattr)
+        goto out;
+
+    ret = dict_get_ptr(xattr, conf->link_xattr_name, &volname);
+
+    if ((-1 == ret) || !volname)
+        goto out;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (strcmp(conf->subvolumes[i]->name, (char *)volname) == 0) {
+            subvol = conf->subvolumes[i];
+            break;
+        }
+    }
 
-        return 0;
 out:
-	local->linkfile.linkfile_cbk (frame, NULL, frame->this, -1, ENOMEM,
-				      loc->inode, NULL, NULL, NULL);
-	return 0;
+    return subvol;
 }
 
-
-int
-dht_linkfile_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			 int32_t op_ret, int32_t op_errno,
-                         struct iatt *preparent, struct iatt *postparent)
+static int
+dht_linkfile_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int op_ret, int op_errno, struct iatt *statpre,
+                         struct iatt *statpost, dict_t *xdata)
 {
-	dht_local_t   *local = NULL;
-	call_frame_t  *prev = NULL;
-	xlator_t      *subvol = NULL;
+    dht_local_t *local = NULL;
+    loc_t *loc = NULL;
 
-	local = frame->local;
-	prev = cookie;
-	subvol = prev->this;
+    local = frame->local;
+    loc = &local->loc;
 
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"unlinking linkfile %s on %s failed (%s)",
-			local->loc.path, subvol->name, strerror (op_errno));
-	}
+    if (op_ret)
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno, DHT_MSG_SETATTR_FAILED,
+                "path=%s", (loc->path ? loc->path : "NULL"), "gfid=%s",
+                uuid_utoa(local->gfid), NULL);
 
-	DHT_STACK_DESTROY (frame);
+    DHT_STACK_DESTROY(frame);
 
-	return 0;
+    return 0;
 }
 
-
 int
-dht_linkfile_unlink (call_frame_t *frame, xlator_t *this,
-		     xlator_t *subvol, loc_t *loc)
+dht_linkfile_attr_heal(call_frame_t *frame, xlator_t *this)
 {
-	call_frame_t *unlink_frame = NULL;
-	dht_local_t  *unlink_local = NULL;
-
-	unlink_frame = copy_frame (frame);
-	if (!unlink_frame) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	unlink_local = dht_local_init (unlink_frame);
-	if (!unlink_local) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	loc_copy (&unlink_local->loc, loc);
-
-	STACK_WIND (unlink_frame, dht_linkfile_unlink_cbk,
-		    subvol, subvol->fops->unlink,
-		    &unlink_local->loc);
-
-	return 0;
-err:
-	if (unlink_frame)
-		DHT_STACK_DESTROY (unlink_frame);
+    int ret = -1;
+    call_frame_t *copy = NULL;
+    dht_local_t *local = NULL;
+    dht_local_t *copy_local = NULL;
+    xlator_t *subvol = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+    dict_t *xattr = NULL;
+
+    local = frame->local;
+
+    GF_VALIDATE_OR_GOTO("dht", local, out);
+    GF_VALIDATE_OR_GOTO("dht", local->link_subvol, out);
+
+    if (local->stbuf.ia_type == IA_INVAL)
+        return 0;
 
-	return -1;
-}
+    DHT_MARK_FOP_INTERNAL(xattr);
 
+    gf_uuid_copy(local->loc.gfid, local->stbuf.ia_gfid);
 
-xlator_t *
-dht_linkfile_subvol (xlator_t *this, inode_t *inode, struct iatt *stbuf,
-		     dict_t *xattr)
-{
-	dht_conf_t *conf = NULL;
-	xlator_t   *subvol = NULL;
-	void       *volname = NULL;
-	int         i = 0, ret = 0;
+    copy = copy_frame(frame);
 
+    if (!copy)
+        goto out;
 
-	conf = this->private;
+    copy_local = dht_local_init(copy, &local->loc, NULL, 0);
 
-	if (!xattr)
-		goto out;
+    if (!copy_local)
+        goto out;
 
-	ret = dict_get_ptr (xattr, "trusted.glusterfs.dht.linkto", &volname);
+    stbuf = local->stbuf;
+    subvol = local->link_subvol;
 
-	if ((-1 == ret) || !volname)
-		goto out;
+    copy->local = copy_local;
 
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		if (strcmp (conf->subvolumes[i]->name, (char *)volname) == 0) {
-			subvol = conf->subvolumes[i];
-			break;
-		}
-	}
+    FRAME_SU_DO(copy, dht_local_t);
 
+    STACK_WIND(copy, dht_linkfile_setattr_cbk, subvol, subvol->fops->setattr,
+               &copy_local->loc, &stbuf, (GF_SET_ATTR_UID | GF_SET_ATTR_GID),
+               xattr);
+    ret = 0;
 out:
-	return subvol;
-}
+    if ((ret < 0) && (copy))
+        DHT_STACK_DESTROY(copy);
 
+    if (xattr)
+        dict_unref(xattr);
 
+    return ret;
+}
diff --git a/xlators/cluster/dht/src/dht-lock.c b/xlators/cluster/dht/src/dht-lock.c
new file mode 100644
index 00000000000..638821ccee5
--- /dev/null
+++ b/xlators/cluster/dht/src/dht-lock.c
@@ -0,0 +1,1392 @@
+/*
+  Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "dht-lock.h"
+
+static char *
+dht_lock_asprintf(dht_lock_t *lock)
+{
+    char *lk_buf = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {
+        0,
+    };
+
+    if (lock == NULL)
+        goto out;
+
+    uuid_utoa_r(lock->loc.gfid, gfid);
+
+    gf_asprintf(&lk_buf, "%s:%s", lock->xl->name, gfid);
+
+out:
+    return lk_buf;
+}
+
+static void
+dht_log_lk_array(char *name, gf_loglevel_t log_level, dht_lock_t **lk_array,
+                 int count)
+{
+    int i = 0;
+    char *lk_buf = NULL;
+
+    if ((lk_array == NULL) || (count == 0))
+        goto out;
+
+    for (i = 0; i < count; i++) {
+        lk_buf = dht_lock_asprintf(lk_array[i]);
+        if (!lk_buf)
+            goto out;
+
+        gf_smsg(name, log_level, 0, DHT_MSG_LK_ARRAY_INFO, "index=%d", i,
+                "lk_buf=%s", lk_buf, NULL);
+        GF_FREE(lk_buf);
+    }
+
+out:
+    return;
+}
+
+static void
+dht_lock_stack_destroy(call_frame_t *lock_frame, dht_lock_type_t lk)
+{
+    dht_local_t *local = NULL;
+
+    local = lock_frame->local;
+
+    if (lk == DHT_INODELK) {
+        local->lock[0].layout.my_layout.locks = NULL;
+        local->lock[0].layout.my_layout.lk_count = 0;
+    } else {
+        local->lock[0].ns.directory_ns.locks = NULL;
+        local->lock[0].ns.directory_ns.lk_count = 0;
+    }
+
+    DHT_STACK_DESTROY(lock_frame);
+    return;
+}
+
+static void
+dht_lock_free(dht_lock_t *lock)
+{
+    if (lock == NULL)
+        goto out;
+
+    loc_wipe(&lock->loc);
+    GF_FREE(lock->domain);
+    GF_FREE(lock->basename);
+    mem_put(lock);
+
+out:
+    return;
+}
+
+static void
+dht_set_lkowner(dht_lock_t **lk_array, int count, gf_lkowner_t *lkowner)
+{
+    int i = 0;
+
+    if (!lk_array || !lkowner)
+        goto out;
+
+    for (i = 0; i < count; i++) {
+        lk_array[i]->lk_owner = *lkowner;
+    }
+
+out:
+    return;
+}
+
+static int
+dht_lock_request_cmp(const void *val1, const void *val2)
+{
+    dht_lock_t *lock1 = NULL;
+    dht_lock_t *lock2 = NULL;
+    int ret = -1;
+
+    lock1 = *(dht_lock_t **)val1;
+    lock2 = *(dht_lock_t **)val2;
+
+    GF_VALIDATE_OR_GOTO("dht-locks", lock1, out);
+    GF_VALIDATE_OR_GOTO("dht-locks", lock2, out);
+
+    ret = strcmp(lock1->xl->name, lock2->xl->name);
+
+    if (ret == 0) {
+        ret = gf_uuid_compare(lock1->loc.gfid, lock2->loc.gfid);
+    }
+
+out:
+    return ret;
+}
+
+static int
+dht_lock_order_requests(dht_lock_t **locks, int count)
+{
+    int ret = -1;
+
+    if (!locks || !count)
+        goto out;
+
+    qsort(locks, count, sizeof(*locks), dht_lock_request_cmp);
+    ret = 0;
+
+out:
+    return ret;
+}
+
+void
+dht_lock_array_free(dht_lock_t **lk_array, int count)
+{
+    int i = 0;
+    dht_lock_t *lock = NULL;
+
+    if (lk_array == NULL)
+        goto out;
+
+    for (i = 0; i < count; i++) {
+        lock = lk_array[i];
+        lk_array[i] = NULL;
+        dht_lock_free(lock);
+    }
+
+out:
+    return;
+}
+
+int32_t
+dht_lock_count(dht_lock_t **lk_array, int lk_count)
+{
+    int i = 0, locked = 0;
+
+    if ((lk_array == NULL) || (lk_count == 0))
+        goto out;
+
+    for (i = 0; i < lk_count; i++) {
+        if (lk_array[i]->locked)
+            locked++;
+    }
+out:
+    return locked;
+}
+
+static call_frame_t *
+dht_lock_frame(call_frame_t *parent_frame)
+{
+    call_frame_t *lock_frame = NULL;
+
+    lock_frame = copy_frame(parent_frame);
+    if (lock_frame == NULL)
+        goto out;
+
+    set_lk_owner_from_ptr(&lock_frame->root->lk_owner, parent_frame->root);
+
+out:
+    return lock_frame;
+}
+
+dht_lock_t *
+dht_lock_new(xlator_t *this, xlator_t *xl, loc_t *loc, short type,
+             const char *domain, const char *basename,
+             dht_reaction_type_t do_on_failure)
+{
+    dht_conf_t *conf = NULL;
+    dht_lock_t *lock = NULL;
+
+    conf = this->private;
+
+    lock = mem_get0(conf->lock_pool);
+    if (lock == NULL)
+        goto out;
+
+    lock->xl = xl;
+    lock->type = type;
+    lock->do_on_failure = do_on_failure;
+
+    lock->domain = gf_strdup(domain);
+    if (lock->domain == NULL) {
+        dht_lock_free(lock);
+        lock = NULL;
+        goto out;
+    }
+
+    if (basename) {
+        lock->basename = gf_strdup(basename);
+        if (lock->basename == NULL) {
+            dht_lock_free(lock);
+            lock = NULL;
+            goto out;
+        }
+    }
+
+    /* Fill only inode and gfid.
+       posix and protocol/server give preference to pargfid/basename over
+       gfid/inode for resolution if all the three parameters of loc_t are
+       present. I want to avoid the following hypothetical situation:
+
+       1. rebalance did a lookup on a dentry and got a gfid.
+       2. rebalance acquires lock on loc_t which was filled with gfid and
+          path (pargfid/bname) from step 1.
+       3. somebody deleted and recreated the same file
+       4. rename on the same path acquires lock on loc_t which now points
+          to a different inode (and hence gets the lock).
+       5. rebalance continues to migrate file (note that not all fops done
+          by rebalance during migration are inode/gfid based Eg., unlink)
+       6. rename continues.
+    */
+    lock->loc.inode = inode_ref(loc->inode);
+    loc_gfid(loc, lock->loc.gfid);
+
+out:
+    return lock;
+}
+
+static int
+dht_local_entrylk_init(call_frame_t *frame, dht_lock_t **lk_array, int lk_count,
+                       fop_entrylk_cbk_t entrylk_cbk)
+{
+    int ret = -1;
+    dht_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local == NULL) {
+        local = dht_local_init(frame, NULL, NULL, 0);
+    }
+
+    if (local == NULL) {
+        goto out;
+    }
+
+    local->lock[0].ns.directory_ns.entrylk_cbk = entrylk_cbk;
+    local->lock[0].ns.directory_ns.locks = lk_array;
+    local->lock[0].ns.directory_ns.lk_count = lk_count;
+
+    ret = dht_lock_order_requests(local->lock[0].ns.directory_ns.locks,
+                                  local->lock[0].ns.directory_ns.lk_count);
+    if (ret < 0)
+        goto out;
+
+    ret = 0;
+out:
+    return ret;
+}
+
+static void
+dht_entrylk_done(call_frame_t *lock_frame)
+{
+    fop_entrylk_cbk_t entrylk_cbk = NULL;
+    call_frame_t *main_frame = NULL;
+    dht_local_t *local = NULL;
+
+    local = lock_frame->local;
+    main_frame = local->main_frame;
+
+    local->lock[0].ns.directory_ns.locks = NULL;
+    local->lock[0].ns.directory_ns.lk_count = 0;
+
+    entrylk_cbk = local->lock[0].ns.directory_ns.entrylk_cbk;
+    local->lock[0].ns.directory_ns.entrylk_cbk = NULL;
+
+    entrylk_cbk(main_frame, NULL, main_frame->this,
+                local->lock[0].ns.directory_ns.op_ret,
+                local->lock[0].ns.directory_ns.op_errno, NULL);
+
+    dht_lock_stack_destroy(lock_frame, DHT_ENTRYLK);
+    return;
+}
+
+static int32_t
+dht_unlock_entrylk_done(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    gf_uuid_unparse(local->lock[0].ns.directory_ns.locks[0]->loc.inode->gfid,
+                    gfid);
+
+    if (op_ret < 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, op_errno,
+                DHT_MSG_UNLOCK_GFID_FAILED, "gfid=%s", gfid,
+                "DHT_LAYOUT_HEAL_DOMAIN", NULL);
+    }
+
+    DHT_STACK_DESTROY(frame);
+    return 0;
+}
+
+static int32_t
+dht_unlock_entrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int lk_index = 0, call_cnt = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    lk_index = (long)cookie;
+
+    local = frame->local;
+
+    uuid_utoa_r(local->lock[0].ns.directory_ns.locks[lk_index]->loc.gfid, gfid);
+
+    if (op_ret < 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_UNLOCKING_FAILED,
+                "name=%s",
+                local->lock[0].ns.directory_ns.locks[lk_index]->xl->name,
+                "gfid=%s", gfid, NULL);
+    } else {
+        local->lock[0].ns.directory_ns.locks[lk_index]->locked = 0;
+    }
+
+    call_cnt = dht_frame_return(frame);
+    if (is_last_call(call_cnt)) {
+        dht_entrylk_done(frame);
+    }
+
+    return 0;
+}
+
+static int32_t
+dht_unlock_entrylk(call_frame_t *frame, dht_lock_t **lk_array, int lk_count,
+                   fop_entrylk_cbk_t entrylk_cbk)
+{
+    dht_local_t *local = NULL;
+    int ret = -1, i = 0;
+    call_frame_t *lock_frame = NULL;
+    int call_cnt = 0;
+
+    GF_VALIDATE_OR_GOTO("dht-locks", frame, done);
+    GF_VALIDATE_OR_GOTO(frame->this->name, lk_array, done);
+    GF_VALIDATE_OR_GOTO(frame->this->name, entrylk_cbk, done);
+
+    call_cnt = dht_lock_count(lk_array, lk_count);
+    if (call_cnt == 0) {
+        ret = 0;
+        goto done;
+    }
+
+    lock_frame = dht_lock_frame(frame);
+    if (lock_frame == NULL) {
+        gf_smsg(frame->this->name, GF_LOG_WARNING, 0,
+                DHT_MSG_ALLOC_FRAME_FAILED_NOT_UNLOCKING_FOLLOWING_ENTRYLKS,
+                NULL);
+
+        dht_log_lk_array(frame->this->name, GF_LOG_WARNING, lk_array, lk_count);
+        goto done;
+    }
+
+    ret = dht_local_entrylk_init(lock_frame, lk_array, lk_count, entrylk_cbk);
+    if (ret < 0) {
+        gf_smsg(frame->this->name, GF_LOG_WARNING, 0,
+                DHT_MSG_LOCAL_LOCKS_STORE_FAILED_UNLOCKING_FOLLOWING_ENTRYLK,
+                NULL);
+
+        dht_log_lk_array(frame->this->name, GF_LOG_WARNING, lk_array, lk_count);
+
+        goto done;
+    }
+
+    local = lock_frame->local;
+    local->main_frame = frame;
+    local->call_cnt = call_cnt;
+
+    for (i = 0; i < local->lock[0].ns.directory_ns.lk_count; i++) {
+        if (!local->lock[0].ns.directory_ns.locks[i]->locked)
+            continue;
+
+        lock_frame->root
+            ->lk_owner = local->lock[0].ns.directory_ns.locks[i]->lk_owner;
+        STACK_WIND_COOKIE(
+            lock_frame, dht_unlock_entrylk_cbk, (void *)(long)i,
+            local->lock[0].ns.directory_ns.locks[i]->xl,
+            local->lock[0].ns.directory_ns.locks[i]->xl->fops->entrylk,
+            local->lock[0].ns.directory_ns.locks[i]->domain,
+            &local->lock[0].ns.directory_ns.locks[i]->loc,
+            local->lock[0].ns.directory_ns.locks[i]->basename, ENTRYLK_UNLOCK,
+            ENTRYLK_WRLCK, NULL);
+        if (!--call_cnt)
+            break;
+    }
+
+    return 0;
+
+done:
+    if (lock_frame)
+        dht_lock_stack_destroy(lock_frame, DHT_ENTRYLK);
+
+    /* no locks acquired, invoke entrylk_cbk */
+    if (ret == 0)
+        entrylk_cbk(frame, NULL, frame->this, 0, 0, NULL);
+
+    return ret;
+}
+
+int32_t
+dht_unlock_entrylk_wrapper(call_frame_t *frame, dht_elock_wrap_t *entrylk)
+{
+    dht_local_t *local = NULL, *lock_local = NULL;
+    call_frame_t *lock_frame = NULL;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    int ret = 0;
+
+    local = frame->local;
+
+    if (!entrylk || !entrylk->locks)
+        goto out;
+
+    gf_uuid_unparse(local->loc.parent->gfid, pgfid);
+
+    lock_frame = copy_frame(frame);
+    if (lock_frame == NULL) {
+        gf_smsg(frame->this->name, GF_LOG_WARNING, ENOMEM,
+                DHT_MSG_COPY_FRAME_FAILED, "pgfid=%s", pgfid, "name=%s",
+                local->loc.name, "path=%s", local->loc.path, NULL);
+        goto done;
+    }
+
+    lock_local = dht_local_init(lock_frame, NULL, NULL, 0);
+    if (lock_local == NULL) {
+        gf_smsg(frame->this->name, GF_LOG_WARNING, ENOMEM,
+                DHT_MSG_CREATE_FAILED, "local", "pgfid=%s", pgfid, "name=%s",
+                local->loc.name, "path=%s", local->loc.path, NULL);
+        goto done;
+    }
+
+    lock_frame->local = lock_local;
+
+    lock_local->lock[0].ns.directory_ns.locks = entrylk->locks;
+    lock_local->lock[0].ns.directory_ns.lk_count = entrylk->lk_count;
+    entrylk->locks = NULL;
+    entrylk->lk_count = 0;
+
+    ret = dht_unlock_entrylk(
+        lock_frame, lock_local->lock[0].ns.directory_ns.locks,
+        lock_local->lock[0].ns.directory_ns.lk_count, dht_unlock_entrylk_done);
+    if (ret)
+        goto done;
+
+    lock_frame = NULL;
+
+done:
+    if (lock_frame != NULL) {
+        DHT_STACK_DESTROY(lock_frame);
+    }
+
+out:
+    return 0;
+}
+
+static int
+dht_entrylk_cleanup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_entrylk_done(frame);
+    return 0;
+}
+
+static void
+dht_entrylk_cleanup(call_frame_t *lock_frame)
+{
+    dht_lock_t **lk_array = NULL;
+    int lk_count = 0, lk_acquired = 0;
+    dht_local_t *local = NULL;
+
+    local = lock_frame->local;
+
+    lk_array = local->lock[0].ns.directory_ns.locks;
+    lk_count = local->lock[0].ns.directory_ns.lk_count;
+
+    lk_acquired = dht_lock_count(lk_array, lk_count);
+    if (lk_acquired != 0) {
+        dht_unlock_entrylk(lock_frame, lk_array, lk_count,
+                           dht_entrylk_cleanup_cbk);
+    } else {
+        dht_entrylk_done(lock_frame);
+    }
+
+    return;
+}
+
+static int32_t
+dht_blocking_entrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    int lk_index = 0;
+    int i = 0;
+    dht_local_t *local = NULL;
+
+    lk_index = (long)cookie;
+
+    local = frame->local;
+    if (op_ret == 0) {
+        local->lock[0].ns.directory_ns.locks[lk_index]->locked = _gf_true;
+    } else {
+        switch (op_errno) {
+            case ESTALE:
+            case ENOENT:
+                if (local->lock[0]
+                        .ns.directory_ns.locks[lk_index]
+                        ->do_on_failure != IGNORE_ENOENT_ESTALE) {
+                    local->lock[0].ns.directory_ns.op_ret = -1;
+                    local->lock[0].ns.directory_ns.op_errno = op_errno;
+                    goto cleanup;
+                }
+                break;
+            default:
+                local->lock[0].ns.directory_ns.op_ret = -1;
+                local->lock[0].ns.directory_ns.op_errno = op_errno;
+                goto cleanup;
+        }
+    }
+
+    if (lk_index == (local->lock[0].ns.directory_ns.lk_count - 1)) {
+        for (i = 0; (i < local->lock[0].ns.directory_ns.lk_count) &&
+                    (!local->lock[0].ns.directory_ns.locks[i]->locked);
+             i++)
+            ;
+
+        if (i == local->lock[0].ns.directory_ns.lk_count) {
+            local->lock[0].ns.directory_ns.op_ret = -1;
+            local->lock[0].ns.directory_ns.op_errno = op_errno;
+        }
+
+        dht_entrylk_done(frame);
+    } else {
+        dht_blocking_entrylk_rec(frame, ++lk_index);
+    }
+
+    return 0;
+
+cleanup:
+    dht_entrylk_cleanup(frame);
+
+    return 0;
+}
+
+void
+dht_blocking_entrylk_rec(call_frame_t *frame, int i)
+{
+    dht_local_t *local = NULL;
+
+    local = frame->local;
+
+    STACK_WIND_COOKIE(
+        frame, dht_blocking_entrylk_cbk, (void *)(long)i,
+        local->lock[0].ns.directory_ns.locks[i]->xl,
+        local->lock[0].ns.directory_ns.locks[i]->xl->fops->entrylk,
+        local->lock[0].ns.directory_ns.locks[i]->domain,
+        &local->lock[0].ns.directory_ns.locks[i]->loc,
+        local->lock[0].ns.directory_ns.locks[i]->basename, ENTRYLK_LOCK,
+        ENTRYLK_WRLCK, NULL);
+
+    return;
+}
+
+int
+dht_blocking_entrylk(call_frame_t *frame, dht_lock_t **lk_array, int lk_count,
+                     fop_entrylk_cbk_t entrylk_cbk)
+{
+    int ret = -1;
+    call_frame_t *lock_frame = NULL;
+    dht_local_t *local = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht-locks", frame, out);
+    GF_VALIDATE_OR_GOTO(frame->this->name, lk_array, out);
+    GF_VALIDATE_OR_GOTO(frame->this->name, entrylk_cbk, out);
+
+    lock_frame = dht_lock_frame(frame);
+    if (lock_frame == NULL)
+        goto out;
+
+    ret = dht_local_entrylk_init(lock_frame, lk_array, lk_count, entrylk_cbk);
+    if (ret < 0) {
+        goto out;
+    }
+
+    dht_set_lkowner(lk_array, lk_count, &lock_frame->root->lk_owner);
+
+    local = lock_frame->local;
+    local->main_frame = frame;
+
+    dht_blocking_entrylk_rec(lock_frame, 0);
+
+    return 0;
+out:
+    if (lock_frame)
+        dht_lock_stack_destroy(lock_frame, DHT_ENTRYLK);
+
+    return -1;
+}
+
+static int
+dht_local_inodelk_init(call_frame_t *frame, dht_lock_t **lk_array, int lk_count,
+                       fop_inodelk_cbk_t inodelk_cbk)
+{
+    int ret = -1;
+    dht_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local == NULL) {
+        local = dht_local_init(frame, NULL, NULL, 0);
+    }
+
+    if (local == NULL) {
+        goto out;
+    }
+
+    local->lock[0].layout.my_layout.inodelk_cbk = inodelk_cbk;
+    local->lock[0].layout.my_layout.locks = lk_array;
+    local->lock[0].layout.my_layout.lk_count = lk_count;
+
+    ret = dht_lock_order_requests(local->lock[0].layout.my_layout.locks,
+                                  local->lock[0].layout.my_layout.lk_count);
+    if (ret < 0)
+        goto out;
+
+    ret = 0;
+out:
+    return ret;
+}
+
+static void
+dht_inodelk_done(call_frame_t *lock_frame)
+{
+    fop_inodelk_cbk_t inodelk_cbk = NULL;
+    call_frame_t *main_frame = NULL;
+    dht_local_t *local = NULL;
+
+    local = lock_frame->local;
+    main_frame = local->main_frame;
+
+    local->lock[0].layout.my_layout.locks = NULL;
+    local->lock[0].layout.my_layout.lk_count = 0;
+
+    inodelk_cbk = local->lock[0].layout.my_layout.inodelk_cbk;
+    local->lock[0].layout.my_layout.inodelk_cbk = NULL;
+
+    inodelk_cbk(main_frame, NULL, main_frame->this,
+                local->lock[0].layout.my_layout.op_ret,
+                local->lock[0].layout.my_layout.op_errno, NULL);
+
+    dht_lock_stack_destroy(lock_frame, DHT_INODELK);
+    return;
+}
+
+static int32_t
+dht_unlock_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int lk_index = 0, call_cnt = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    lk_index = (long)cookie;
+
+    local = frame->local;
+    if (op_ret < 0) {
+        uuid_utoa_r(local->lock[0].layout.my_layout.locks[lk_index]->loc.gfid,
+                    gfid);
+
+        gf_smsg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_UNLOCKING_FAILED,
+                "name=%s",
+                local->lock[0].layout.my_layout.locks[lk_index]->xl->name,
+                "gfid=%s", gfid, NULL);
+    } else {
+        local->lock[0].layout.my_layout.locks[lk_index]->locked = 0;
+    }
+
+    call_cnt = dht_frame_return(frame);
+    if (is_last_call(call_cnt)) {
+        dht_inodelk_done(frame);
+    }
+
+    return 0;
+}
+
+static int32_t
+dht_unlock_inodelk_done(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    gf_uuid_unparse(local->lock[0].layout.my_layout.locks[0]->loc.inode->gfid,
+                    gfid);
+
+    if (op_ret < 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, op_errno,
+                DHT_MSG_UNLOCK_GFID_FAILED, "DHT_LAYOUT_HEAL_DOMAIN gfid=%s",
+                gfid, NULL);
+    }
+
+    DHT_STACK_DESTROY(frame);
+    return 0;
+}
+
+int32_t
+dht_unlock_inodelk(call_frame_t *frame, dht_lock_t **lk_array, int lk_count,
+                   fop_inodelk_cbk_t inodelk_cbk)
+{
+    dht_local_t *local = NULL;
+    struct gf_flock flock = {
+        0,
+    };
+    int ret = -1, i = 0;
+    call_frame_t *lock_frame = NULL;
+    int call_cnt = 0;
+
+    GF_VALIDATE_OR_GOTO("dht-locks", frame, done);
+    GF_VALIDATE_OR_GOTO(frame->this->name, lk_array, done);
+    GF_VALIDATE_OR_GOTO(frame->this->name, inodelk_cbk, done);
+
+    call_cnt = dht_lock_count(lk_array, lk_count);
+    if (call_cnt == 0) {
+        ret = 0;
+        goto done;
+    }
+
+    lock_frame = dht_lock_frame(frame);
+    if (lock_frame == NULL) {
+        gf_smsg(frame->this->name, GF_LOG_WARNING, 0,
+                DHT_MSG_ALLOC_FRAME_FAILED_NOT_UNLOCKING_FOLLOWING_ENTRYLKS,
+                NULL);
+
+        dht_log_lk_array(frame->this->name, GF_LOG_WARNING, lk_array, lk_count);
+        goto done;
+    }
+
+    ret = dht_local_inodelk_init(lock_frame, lk_array, lk_count, inodelk_cbk);
+    if (ret < 0) {
+        gf_smsg(frame->this->name, GF_LOG_WARNING, 0,
+                DHT_MSG_LOCAL_LOCKS_STORE_FAILED_UNLOCKING_FOLLOWING_ENTRYLK,
+                NULL);
+
+        dht_log_lk_array(frame->this->name, GF_LOG_WARNING, lk_array, lk_count);
+
+        goto done;
+    }
+
+    local = lock_frame->local;
+    local->main_frame = frame;
+    local->call_cnt = call_cnt;
+
+    flock.l_type = F_UNLCK;
+
+    for (i = 0; i < local->lock[0].layout.my_layout.lk_count; i++) {
+        if (!local->lock[0].layout.my_layout.locks[i]->locked)
+            continue;
+
+        lock_frame->root
+            ->lk_owner = local->lock[0].layout.my_layout.locks[i]->lk_owner;
+        STACK_WIND_COOKIE(
+            lock_frame, dht_unlock_inodelk_cbk, (void *)(long)i,
+            local->lock[0].layout.my_layout.locks[i]->xl,
+            local->lock[0].layout.my_layout.locks[i]->xl->fops->inodelk,
+            local->lock[0].layout.my_layout.locks[i]->domain,
+            &local->lock[0].layout.my_layout.locks[i]->loc, F_SETLK, &flock,
+            NULL);
+        if (!--call_cnt)
+            break;
+    }
+
+    return 0;
+
+done:
+    if (lock_frame)
+        dht_lock_stack_destroy(lock_frame, DHT_INODELK);
+
+    /* no locks acquired, invoke inodelk_cbk */
+    if (ret == 0)
+        inodelk_cbk(frame, NULL, frame->this, 0, 0, NULL);
+
+    return ret;
+}
+
+int32_t
+dht_unlock_inodelk_wrapper(call_frame_t *frame, dht_ilock_wrap_t *inodelk)
+{
+    dht_local_t *local = NULL, *lock_local = NULL;
+    call_frame_t *lock_frame = NULL;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    int ret = 0;
+
+    local = frame->local;
+
+    if (!inodelk || !inodelk->locks)
+        goto out;
+
+    gf_uuid_unparse(local->loc.parent->gfid, pgfid);
+
+    lock_frame = copy_frame(frame);
+    if (lock_frame == NULL) {
+        gf_smsg(frame->this->name, GF_LOG_WARNING, ENOMEM,
+                DHT_MSG_COPY_FRAME_FAILED, "pgfid=%s", pgfid, "name=%s",
+                local->loc.name, "path=%s", local->loc.path, NULL);
+        goto done;
+    }
+
+    lock_local = dht_local_init(lock_frame, NULL, NULL, 0);
+    if (lock_local == NULL) {
+        gf_smsg(frame->this->name, GF_LOG_WARNING, ENOMEM,
+                DHT_MSG_CREATE_FAILED, "local", "gfid=%s", pgfid, "name=%s",
+                local->loc.name, "path=%s", local->loc.path, NULL);
+        goto done;
+    }
+
+    lock_frame->local = lock_local;
+
+    lock_local->lock[0].layout.my_layout.locks = inodelk->locks;
+    lock_local->lock[0].layout.my_layout.lk_count = inodelk->lk_count;
+    inodelk->locks = NULL;
+    inodelk->lk_count = 0;
+
+    ret = dht_unlock_inodelk(
+        lock_frame, lock_local->lock[0].layout.my_layout.locks,
+        lock_local->lock[0].layout.my_layout.lk_count, dht_unlock_inodelk_done);
+
+    if (ret)
+        goto done;
+
+    lock_frame = NULL;
+
+done:
+    if (lock_frame != NULL) {
+        DHT_STACK_DESTROY(lock_frame);
+    }
+out:
+    return 0;
+}
+
+static int
+dht_inodelk_cleanup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_inodelk_done(frame);
+    return 0;
+}
+
+static void
+dht_inodelk_cleanup(call_frame_t *lock_frame)
+{
+    dht_lock_t **lk_array = NULL;
+    int lk_count = 0, lk_acquired = 0;
+    dht_local_t *local = NULL;
+
+    local = lock_frame->local;
+
+    lk_array = local->lock[0].layout.my_layout.locks;
+    lk_count = local->lock[0].layout.my_layout.lk_count;
+
+    lk_acquired = dht_lock_count(lk_array, lk_count);
+    if (lk_acquired != 0) {
+        dht_unlock_inodelk(lock_frame, lk_array, lk_count,
+                           dht_inodelk_cleanup_cbk);
+    } else {
+        dht_inodelk_done(lock_frame);
+    }
+
+    return;
+}
+
+static int32_t
+dht_nonblocking_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int lk_index = 0, call_cnt = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    lk_index = (long)cookie;
+
+    if (op_ret == -1) {
+        local->lock[0].layout.my_layout.op_ret = -1;
+        local->lock[0].layout.my_layout.op_errno = op_errno;
+
+        if (local && local->lock[0].layout.my_layout.locks[lk_index]) {
+            uuid_utoa_r(local->lock[0]
+                            .layout.my_layout.locks[lk_index]
+                            ->loc.inode->gfid,
+                        gfid);
+
+            gf_msg_debug(
+                this->name, op_errno,
+                "inodelk failed on gfid: %s "
+                "subvolume: %s",
+                gfid,
+                local->lock[0].layout.my_layout.locks[lk_index]->xl->name);
+        }
+
+        goto out;
+    }
+
+    local->lock[0].layout.my_layout.locks[lk_index]->locked = _gf_true;
+
+out:
+    call_cnt = dht_frame_return(frame);
+    if (is_last_call(call_cnt)) {
+        if (local->lock[0].layout.my_layout.op_ret < 0) {
+            dht_inodelk_cleanup(frame);
+            return 0;
+        }
+
+        dht_inodelk_done(frame);
+    }
+
+    return 0;
+}
+
+int
+dht_nonblocking_inodelk(call_frame_t *frame, dht_lock_t **lk_array,
+                        int lk_count, fop_inodelk_cbk_t inodelk_cbk)
+{
+    struct gf_flock flock = {
+        0,
+    };
+    int i = 0, ret = 0;
+    dht_local_t *local = NULL;
+    call_frame_t *lock_frame = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht-locks", frame, out);
+    GF_VALIDATE_OR_GOTO(frame->this->name, lk_array, out);
+    GF_VALIDATE_OR_GOTO(frame->this->name, inodelk_cbk, out);
+
+    lock_frame = dht_lock_frame(frame);
+    if (lock_frame == NULL)
+        goto out;
+
+    ret = dht_local_inodelk_init(lock_frame, lk_array, lk_count, inodelk_cbk);
+    if (ret < 0) {
+        goto out;
+    }
+
+    dht_set_lkowner(lk_array, lk_count, &lock_frame->root->lk_owner);
+
+    local = lock_frame->local;
+    local->main_frame = frame;
+
+    local->call_cnt = lk_count;
+
+    for (i = 0; i < lk_count; i++) {
+        flock.l_type = local->lock[0].layout.my_layout.locks[i]->type;
+
+        STACK_WIND_COOKIE(
+            lock_frame, dht_nonblocking_inodelk_cbk, (void *)(long)i,
+            local->lock[0].layout.my_layout.locks[i]->xl,
+            local->lock[0].layout.my_layout.locks[i]->xl->fops->inodelk,
+            local->lock[0].layout.my_layout.locks[i]->domain,
+            &local->lock[0].layout.my_layout.locks[i]->loc, F_SETLK, &flock,
+            NULL);
+    }
+
+    return 0;
+
+out:
+    if (lock_frame)
+        dht_lock_stack_destroy(lock_frame, DHT_INODELK);
+
+    return -1;
+}
+
+static int32_t
+dht_blocking_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    int lk_index = 0;
+    int i = 0;
+    dht_local_t *local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {
+        0,
+    };
+    dht_reaction_type_t reaction = 0;
+
+    lk_index = (long)cookie;
+
+    local = frame->local;
+    if (op_ret == 0) {
+        local->lock[0].layout.my_layout.locks[lk_index]->locked = _gf_true;
+    } else {
+        switch (op_errno) {
+            case ESTALE:
+            case ENOENT:
+                reaction = local->lock[0]
+                               .layout.my_layout.locks[lk_index]
+                               ->do_on_failure;
+                if ((reaction != IGNORE_ENOENT_ESTALE) &&
+                    (reaction != IGNORE_ENOENT_ESTALE_EIO)) {
+                    gf_uuid_unparse(local->lock[0]
+                                        .layout.my_layout.locks[lk_index]
+                                        ->loc.gfid,
+                                    gfid);
+                    local->lock[0].layout.my_layout.op_ret = -1;
+                    local->lock[0].layout.my_layout.op_errno = op_errno;
+                    gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                            DHT_MSG_INODELK_FAILED, "subvol=%s",
+                            local->lock[0]
+                                .layout.my_layout.locks[lk_index]
+                                ->xl->name,
+                            "gfid=%s", gfid, NULL);
+                    goto cleanup;
+                }
+                break;
+            case EIO:
+                reaction = local->lock[0]
+                               .layout.my_layout.locks[lk_index]
+                               ->do_on_failure;
+                if (reaction != IGNORE_ENOENT_ESTALE_EIO) {
+                    gf_uuid_unparse(local->lock[0]
+                                        .layout.my_layout.locks[lk_index]
+                                        ->loc.gfid,
+                                    gfid);
+                    local->lock[0].layout.my_layout.op_ret = -1;
+                    local->lock[0].layout.my_layout.op_errno = op_errno;
+                    gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                            DHT_MSG_INODELK_FAILED, "subvol=%s",
+                            local->lock[0]
+                                .layout.my_layout.locks[lk_index]
+                                ->xl->name,
+                            "gfid=%s", gfid, NULL);
+                    goto cleanup;
+                }
+                break;
+
+            default:
+                gf_uuid_unparse(
+                    local->lock[0].layout.my_layout.locks[lk_index]->loc.gfid,
+                    gfid);
+                local->lock[0].layout.my_layout.op_ret = -1;
+                local->lock[0].layout.my_layout.op_errno = op_errno;
+                gf_smsg(
+                    this->name, GF_LOG_ERROR, op_errno, DHT_MSG_INODELK_FAILED,
+                    "subvol=%s",
+                    local->lock[0].layout.my_layout.locks[lk_index]->xl->name,
+                    "gfid=%s", gfid, NULL);
+                goto cleanup;
+        }
+    }
+
+    if (lk_index == (local->lock[0].layout.my_layout.lk_count - 1)) {
+        for (i = 0; (i < local->lock[0].layout.my_layout.lk_count) &&
+                    (!local->lock[0].layout.my_layout.locks[i]->locked);
+             i++)
+            ;
+
+        if (i == local->lock[0].layout.my_layout.lk_count) {
+            local->lock[0].layout.my_layout.op_ret = -1;
+            local->lock[0].layout.my_layout.op_errno = op_errno;
+        }
+
+        dht_inodelk_done(frame);
+    } else {
+        dht_blocking_inodelk_rec(frame, ++lk_index);
+    }
+
+    return 0;
+
+cleanup:
+    dht_inodelk_cleanup(frame);
+
+    return 0;
+}
+
+void
+dht_blocking_inodelk_rec(call_frame_t *frame, int i)
+{
+    dht_local_t *local = NULL;
+    struct gf_flock flock = {
+        0,
+    };
+
+    local = frame->local;
+
+    flock.l_type = local->lock[0].layout.my_layout.locks[i]->type;
+
+    STACK_WIND_COOKIE(
+        frame, dht_blocking_inodelk_cbk, (void *)(long)i,
+        local->lock[0].layout.my_layout.locks[i]->xl,
+        local->lock[0].layout.my_layout.locks[i]->xl->fops->inodelk,
+        local->lock[0].layout.my_layout.locks[i]->domain,
+        &local->lock[0].layout.my_layout.locks[i]->loc, F_SETLKW, &flock, NULL);
+
+    return;
+}
+
+int
+dht_blocking_inodelk(call_frame_t *frame, dht_lock_t **lk_array, int lk_count,
+                     fop_inodelk_cbk_t inodelk_cbk)
+{
+    int ret = -1;
+    call_frame_t *lock_frame = NULL;
+    dht_local_t *local = NULL;
+    dht_local_t *tmp_local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("dht-locks", frame, out);
+    GF_VALIDATE_OR_GOTO(frame->this->name, lk_array, out);
+    GF_VALIDATE_OR_GOTO(frame->this->name, inodelk_cbk, out);
+
+    tmp_local = frame->local;
+
+    lock_frame = dht_lock_frame(frame);
+    if (lock_frame == NULL) {
+        gf_uuid_unparse(tmp_local->loc.gfid, gfid);
+        gf_smsg("dht", GF_LOG_ERROR, ENOMEM, DHT_MSG_LOCK_FRAME_FAILED,
+                "gfid=%s", gfid, "path=%s", tmp_local->loc.path, NULL);
+        goto out;
+    }
+
+    ret = dht_local_inodelk_init(lock_frame, lk_array, lk_count, inodelk_cbk);
+    if (ret < 0) {
+        gf_uuid_unparse(tmp_local->loc.gfid, gfid);
+        gf_smsg("dht", GF_LOG_ERROR, ENOMEM, DHT_MSG_LOCAL_LOCK_INIT_FAILED,
+                "gfid=%s", gfid, "path=%s", tmp_local->loc.path, NULL);
+        goto out;
+    }
+
+    dht_set_lkowner(lk_array, lk_count, &lock_frame->root->lk_owner);
+
+    local = lock_frame->local;
+    local->main_frame = frame;
+
+    dht_blocking_inodelk_rec(lock_frame, 0);
+
+    return 0;
+out:
+    if (lock_frame)
+        dht_lock_stack_destroy(lock_frame, DHT_INODELK);
+
+    return -1;
+}
+
+void
+dht_unlock_namespace(call_frame_t *frame, dht_dir_transaction_t *lock)
+{
+    GF_VALIDATE_OR_GOTO("dht-locks", frame, out);
+    GF_VALIDATE_OR_GOTO(frame->this->name, lock, out);
+
+    dht_unlock_entrylk_wrapper(frame, &lock->ns.directory_ns);
+    dht_unlock_inodelk_wrapper(frame, &lock->ns.parent_layout);
+
+out:
+    return;
+}
+
+static int32_t
+dht_protect_namespace_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+
+    local = frame->local;
+    if (op_ret != 0)
+        dht_unlock_inodelk_wrapper(frame, &local->current->ns.parent_layout);
+
+    local->current->ns.ns_cbk(frame, cookie, this, op_ret, op_errno, xdata);
+    return 0;
+}
+
+int32_t
+dht_blocking_entrylk_after_inodelk(call_frame_t *frame, void *cookie,
+                                   xlator_t *this, int32_t op_ret,
+                                   int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int ret = -1;
+    loc_t *loc = NULL;
+    dht_lock_t **lk_array = NULL;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    int count = 0;
+    dht_elock_wrap_t *entrylk = NULL;
+
+    local = frame->local;
+    entrylk = &local->current->ns.directory_ns;
+
+    if (op_ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    loc = &entrylk->locks[0]->loc;
+    gf_uuid_unparse(loc->gfid, pgfid);
+
+    local->op_ret = 0;
+    lk_array = entrylk->locks;
+    count = entrylk->lk_count;
+
+    ret = dht_blocking_entrylk(frame, lk_array, count,
+                               dht_protect_namespace_cbk);
+
+    if (ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = EIO;
+        gf_smsg(this->name, GF_LOG_WARNING, local->op_errno,
+                DHT_MSG_ENTRYLK_FAILED_AFT_INODELK, "fop=%s",
+                gf_fop_list[local->fop], "pgfid=%s", pgfid, "basename=%s",
+                entrylk->locks[0]->basename, NULL);
+        goto err;
+    }
+
+    return 0;
+
+err:
+    if (lk_array != NULL) {
+        dht_lock_array_free(lk_array, count);
+        GF_FREE(lk_array);
+        entrylk->locks = NULL;
+        entrylk->lk_count = 0;
+    }
+
+    /* Unlock inodelk. No harm calling unlock twice */
+    dht_unlock_inodelk_wrapper(frame, &local->current->ns.parent_layout);
+    /* Call ns_cbk. It will take care of unwinding */
+    local->current->ns.ns_cbk(frame, NULL, this, local->op_ret, local->op_errno,
+                              NULL);
+    return 0;
+}
+
+/* Given the loc and the subvol, this routine takes the inodelk on
+ * the parent inode and entrylk on (parent, loc->name). This routine
+ * is specific as it supports only one subvol on which it takes inodelk
+ * and then entrylk serially.
+ */
+int
+dht_protect_namespace(call_frame_t *frame, loc_t *loc, xlator_t *subvol,
+                      struct dht_namespace *ns, fop_entrylk_cbk_t ns_cbk)
+{
+    dht_ilock_wrap_t *inodelk = NULL;
+    dht_elock_wrap_t *entrylk = NULL;
+    dht_lock_t **lk_array = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *this = NULL;
+    loc_t parent = {
+        0,
+    };
+    int ret = -1;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    int32_t op_errno = 0;
+    int count = 1;
+
+    GF_VALIDATE_OR_GOTO("dht-locks", frame, out);
+    GF_VALIDATE_OR_GOTO(frame->this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(frame->this->name, loc->parent, out);
+    GF_VALIDATE_OR_GOTO(frame->this->name, subvol, out);
+
+    local = frame->local;
+    this = frame->this;
+
+    inodelk = &ns->parent_layout;
+    entrylk = &ns->directory_ns;
+
+    /* Initialize entrylk_cbk and parent loc */
+    ns->ns_cbk = ns_cbk;
+
+    ret = dht_build_parent_loc(this, &parent, loc, &op_errno);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno, DHT_MSG_LOC_FAILED,
+                "gfid=%s", loc->gfid, "name=%s", loc->name, "path=%s",
+                loc->path, NULL);
+        goto out;
+    }
+    gf_uuid_unparse(parent.gfid, pgfid);
+
+    /* Alloc inodelk */
+    inodelk->locks = GF_CALLOC(count, sizeof(*lk_array), gf_common_mt_pointer);
+    if (inodelk->locks == NULL) {
+        local->op_errno = ENOMEM;
+        gf_smsg(this->name, GF_LOG_WARNING, local->op_errno,
+                DHT_MSG_CALLOC_FAILED, "fop=%s", gf_fop_list[local->fop],
+                "pgfid=%s", pgfid, "name=%s", loc->name, "path=%s", loc->path,
+                NULL);
+        goto out;
+    }
+
+    inodelk->locks[0] = dht_lock_new(this, subvol, &parent, F_RDLCK,
+                                     DHT_LAYOUT_HEAL_DOMAIN, NULL,
+                                     FAIL_ON_ANY_ERROR);
+    if (inodelk->locks[0] == NULL) {
+        local->op_errno = ENOMEM;
+        gf_smsg(this->name, GF_LOG_WARNING, local->op_errno,
+                DHT_MSG_LOCK_ALLOC_FAILED, "inodelk-fop=%s",
+                gf_fop_list[local->fop], "pgfid=%s", pgfid, "name=%s",
+                loc->name, "path=%s", loc->path, NULL);
+        goto err;
+    }
+    inodelk->lk_count = count;
+
+    /* Allock entrylk */
+    entrylk->locks = GF_CALLOC(count, sizeof(*lk_array), gf_common_mt_pointer);
+    if (entrylk->locks == NULL) {
+        local->op_errno = ENOMEM;
+        gf_smsg(this->name, GF_LOG_WARNING, local->op_errno,
+                DHT_MSG_CALLOC_FAILED, "entrylk-fop=%s",
+                gf_fop_list[local->fop], "pgfid=%s", pgfid, "name=%s",
+                loc->name, "path=%s", loc->path, NULL);
+
+        goto err;
+    }
+
+    entrylk->locks[0] = dht_lock_new(this, subvol, &parent, F_WRLCK,
+                                     DHT_ENTRY_SYNC_DOMAIN, loc->name,
+                                     FAIL_ON_ANY_ERROR);
+    if (entrylk->locks[0] == NULL) {
+        local->op_errno = ENOMEM;
+        gf_smsg(this->name, GF_LOG_WARNING, local->op_errno,
+                DHT_MSG_LOCK_ALLOC_FAILED, "entrylk-fop=%s",
+                gf_fop_list[local->fop], "pgfid=%s", pgfid, "name=%s",
+                loc->name, "path=%s", loc->path, NULL);
+
+        goto err;
+    }
+    entrylk->lk_count = count;
+
+    /* Take read inodelk on parent. If it is successful, take write entrylk
+     * on name in cbk.
+     */
+    lk_array = inodelk->locks;
+    ret = dht_blocking_inodelk(frame, lk_array, count,
+                               dht_blocking_entrylk_after_inodelk);
+    if (ret < 0) {
+        local->op_errno = EIO;
+        gf_smsg(this->name, GF_LOG_WARNING, local->op_errno,
+                DHT_MSG_BLOCK_INODELK_FAILED, "fop=%s", gf_fop_list[local->fop],
+                "pgfid=%s", pgfid, "name=%s", loc->name, "path=%s", loc->path,
+                NULL);
+
+        goto err;
+    }
+
+    loc_wipe(&parent);
+
+    return 0;
+err:
+    if (entrylk->locks != NULL) {
+        dht_lock_array_free(entrylk->locks, count);
+        GF_FREE(entrylk->locks);
+        entrylk->locks = NULL;
+        entrylk->lk_count = 0;
+    }
+
+    if (inodelk->locks != NULL) {
+        dht_lock_array_free(inodelk->locks, count);
+        GF_FREE(inodelk->locks);
+        inodelk->locks = NULL;
+        inodelk->lk_count = 0;
+    }
+
+    loc_wipe(&parent);
+out:
+    return -1;
+}
diff --git a/xlators/cluster/dht/src/dht-lock.h b/xlators/cluster/dht/src/dht-lock.h
new file mode 100644
index 00000000000..6485c03fb6e
--- /dev/null
+++ b/xlators/cluster/dht/src/dht-lock.h
@@ -0,0 +1,91 @@
+/*
+  Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _DHT_LOCK_H
+#define _DHT_LOCK_H
+
+#include "dht-common.h"
+
+void
+dht_lock_array_free(dht_lock_t **lk_array, int count);
+
+int32_t
+dht_lock_count(dht_lock_t **lk_array, int lk_count);
+
+dht_lock_t *
+dht_lock_new(xlator_t *this, xlator_t *xl, loc_t *loc, short type,
+             const char *domain, const char *basename,
+             dht_reaction_type_t do_on_failure);
+
+int32_t
+dht_unlock_entrylk_wrapper(call_frame_t *, dht_elock_wrap_t *);
+
+void
+dht_blocking_entrylk_rec(call_frame_t *frame, int i);
+
+int
+dht_blocking_entrylk(call_frame_t *frame, dht_lock_t **lk_array, int lk_count,
+                     fop_inodelk_cbk_t entrylk_cbk);
+
+int32_t
+dht_unlock_inodelk(call_frame_t *frame, dht_lock_t **lk_array, int lk_count,
+                   fop_inodelk_cbk_t inodelk_cbk);
+
+int32_t
+dht_unlock_inodelk_wrapper(call_frame_t *, dht_ilock_wrap_t *);
+
+/* Acquire non-blocking inodelk on a list of xlators.
+ *
+ * @lk_array: array of lock requests lock on.
+ *
+ * @lk_count: number of locks in @lk_array
+ *
+ * @inodelk_cbk: will be called after inodelk replies are received
+ *
+ * @retval: -1 if stack_winding inodelk fails. 0 otherwise.
+ *          inodelk_cbk is called with appropriate error on errors.
+ *          On failure to acquire lock on all members of list, successful
+ *          locks are unlocked before invoking cbk.
+ */
+
+int
+dht_nonblocking_inodelk(call_frame_t *frame, dht_lock_t **lk_array,
+                        int lk_count, fop_inodelk_cbk_t inodelk_cbk);
+
+void
+dht_blocking_inodelk_rec(call_frame_t *frame, int i);
+
+/* same as dht_nonblocking_inodelk, but issues sequential blocking locks on
+ * @lk_array directly. locks are issued on some order which remains same
+ * for a list of xlators (irrespective of order of xlators within list).
+ */
+
+int
+dht_blocking_inodelk(call_frame_t *frame, dht_lock_t **lk_array, int lk_count,
+                     fop_inodelk_cbk_t inodelk_cbk);
+
+int32_t
+dht_blocking_entrylk_after_inodelk(call_frame_t *frame, void *cookie,
+                                   xlator_t *this, int32_t op_ret,
+                                   int32_t op_errno, dict_t *xdata);
+
+int32_t
+dht_blocking_entrylk_after_inodelk_rename(call_frame_t *frame, void *cookie,
+                                          xlator_t *this, int32_t op_ret,
+                                          int32_t op_errno, dict_t *xdata);
+
+void
+dht_unlock_namespace(call_frame_t *, dht_dir_transaction_t *);
+
+int
+dht_protect_namespace(call_frame_t *frame, loc_t *loc, xlator_t *subvol,
+                      struct dht_namespace *ns, fop_entrylk_cbk_t ns_cbk);
+
+#endif /* _DHT_LOCK_H */
diff --git a/xlators/cluster/dht/src/dht-mem-types.h b/xlators/cluster/dht/src/dht-mem-types.h
index a5d28e8fbcc..e3c4471334a 100644
--- a/xlators/cluster/dht/src/dht-mem-types.h
+++ b/xlators/cluster/dht/src/dht-mem-types.h
@@ -1,42 +1,38 @@
-
 /*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
 
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
-
 #ifndef __DHT_MEM_TYPES_H__
 #define __DHT_MEM_TYPES_H__
 
-#include "mem-types.h"
+#include <glusterfs/mem-types.h>
 
 enum gf_dht_mem_types_ {
-        gf_dht_mt_dht_du_t = gf_common_mt_end + 1,
-        gf_dht_mt_dht_conf_t,
-        gf_dht_mt_char,
-        gf_dht_mt_int32_t,
-        gf_dht_mt_dht_local_t,
-        gf_dht_mt_xlator_t,
-        gf_dht_mt_dht_layout_t,
-        gf_switch_mt_dht_conf_t,
-        gf_switch_mt_dht_du_t,
-        gf_switch_mt_switch_sched_array,
-        gf_switch_mt_switch_struct,
-        gf_dht_mt_end
+    gf_dht_mt_dht_du_t = gf_common_mt_end + 1,
+    gf_dht_mt_dht_conf_t,
+    gf_dht_mt_char,
+    gf_dht_mt_int32_t,
+    gf_dht_mt_xlator_t,
+    gf_dht_mt_dht_layout_t,
+    gf_switch_mt_switch_sched_array,
+    gf_switch_mt_switch_struct,
+    gf_dht_mt_subvol_time,
+    gf_dht_mt_loc_t,
+    gf_defrag_info_mt,
+    gf_dht_mt_inode_ctx_t,
+    gf_dht_mt_dirent_t,
+    gf_dht_mt_container_t,
+    gf_dht_mt_octx_t,
+    gf_dht_mt_miginfo_t,
+    gf_dht_mt_fd_ctx_t,
+    gf_dht_ret_cache_t,
+    gf_dht_nodeuuids_t,
+    gf_dht_mt_end
 };
 #endif
-
diff --git a/xlators/cluster/dht/src/dht-messages.h b/xlators/cluster/dht/src/dht-messages.h
new file mode 100644
index 00000000000..601f8dad78b
--- /dev/null
+++ b/xlators/cluster/dht/src/dht-messages.h
@@ -0,0 +1,386 @@
+/*Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _DHT_MESSAGES_H_
+#define _DHT_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(
+    DHT, DHT_MSG_CACHED_SUBVOL_GET_FAILED, DHT_MSG_CREATE_LINK_FAILED,
+    DHT_MSG_DICT_SET_FAILED, DHT_MSG_DIR_ATTR_HEAL_FAILED,
+    DHT_MSG_DIR_SELFHEAL_FAILED, DHT_MSG_DIR_SELFHEAL_XATTR_FAILED,
+    DHT_MSG_FILE_ON_MULT_SUBVOL, DHT_MSG_FILE_TYPE_MISMATCH,
+    DHT_MSG_GFID_MISMATCH, DHT_MSG_GFID_NULL, DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+    DHT_MSG_INIT_FAILED, DHT_MSG_INVALID_CONFIGURATION,
+    DHT_MSG_INVALID_DISK_LAYOUT, DHT_MSG_INVALID_OPTION,
+    DHT_MSG_LAYOUT_FIX_FAILED, DHT_MSG_LAYOUT_MERGE_FAILED,
+    DHT_MSG_LAYOUT_MISMATCH, DHT_MSG_LAYOUT_NULL, DHT_MSG_MIGRATE_DATA_COMPLETE,
+    DHT_MSG_MIGRATE_DATA_FAILED, DHT_MSG_MIGRATE_FILE_COMPLETE,
+    DHT_MSG_MIGRATE_FILE_FAILED, DHT_MSG_NO_MEMORY, DHT_MSG_OPENDIR_FAILED,
+    DHT_MSG_REBALANCE_FAILED, DHT_MSG_REBALANCE_START_FAILED,
+    DHT_MSG_REBALANCE_STATUS, DHT_MSG_REBALANCE_STOPPED, DHT_MSG_RENAME_FAILED,
+    DHT_MSG_SETATTR_FAILED, DHT_MSG_SUBVOL_INSUFF_INODES,
+    DHT_MSG_SUBVOL_INSUFF_SPACE, DHT_MSG_UNLINK_FAILED,
+    DHT_MSG_LAYOUT_SET_FAILED, DHT_MSG_LOG_FIXED_LAYOUT,
+    DHT_MSG_GET_XATTR_FAILED, DHT_MSG_FILE_LOOKUP_FAILED,
+    DHT_MSG_OPEN_FD_FAILED, DHT_MSG_SET_INODE_CTX_FAILED,
+    DHT_MSG_UNLOCKING_FAILED, DHT_MSG_DISK_LAYOUT_NULL, DHT_MSG_SUBVOL_INFO,
+    DHT_MSG_CHUNK_SIZE_INFO, DHT_MSG_LAYOUT_FORM_FAILED, DHT_MSG_SUBVOL_ERROR,
+    DHT_MSG_LAYOUT_SORT_FAILED, DHT_MSG_REGEX_INFO, DHT_MSG_FOPEN_FAILED,
+    DHT_MSG_SET_HOSTNAME_FAILED, DHT_MSG_BRICK_ERROR, DHT_MSG_SYNCOP_FAILED,
+    DHT_MSG_MIGRATE_INFO, DHT_MSG_SOCKET_ERROR, DHT_MSG_CREATE_FD_FAILED,
+    DHT_MSG_READDIR_ERROR, DHT_MSG_CHILD_LOC_BUILD_FAILED,
+    DHT_MSG_SET_SWITCH_PATTERN_ERROR, DHT_MSG_COMPUTE_HASH_FAILED,
+    DHT_MSG_FIND_LAYOUT_ANOMALIES_ERROR, DHT_MSG_ANOMALIES_INFO,
+    DHT_MSG_LAYOUT_INFO, DHT_MSG_INODE_LK_ERROR, DHT_MSG_RENAME_INFO,
+    DHT_MSG_DATA_NULL, DHT_MSG_AGGREGATE_QUOTA_XATTR_FAILED,
+    DHT_MSG_UNLINK_LOOKUP_INFO, DHT_MSG_LINK_FILE_LOOKUP_INFO,
+    DHT_MSG_OPERATION_NOT_SUP, DHT_MSG_NOT_LINK_FILE_ERROR, DHT_MSG_CHILD_DOWN,
+    DHT_MSG_UUID_PARSE_ERROR, DHT_MSG_GET_DISK_INFO_ERROR,
+    DHT_MSG_INVALID_VALUE, DHT_MSG_SWITCH_PATTERN_INFO,
+    DHT_MSG_SUBVOL_OP_FAILED, DHT_MSG_LAYOUT_PRESET_FAILED,
+    DHT_MSG_INVALID_LINKFILE, DHT_MSG_FIX_LAYOUT_INFO,
+    DHT_MSG_GET_HOSTNAME_FAILED, DHT_MSG_WRITE_FAILED,
+    DHT_MSG_MIGRATE_HARDLINK_FILE_FAILED, DHT_MSG_FSYNC_FAILED,
+    DHT_MSG_SUBVOL_DECOMMISSION_INFO, DHT_MSG_BRICK_QUERY_FAILED,
+    DHT_MSG_SUBVOL_NO_LAYOUT_INFO, DHT_MSG_OPEN_FD_ON_DST_FAILED,
+    DHT_MSG_SUBVOL_NOT_FOUND, DHT_MSG_FILE_LOOKUP_ON_DST_FAILED,
+    DHT_MSG_DISK_LAYOUT_MISSING, DHT_MSG_DICT_GET_FAILED,
+    DHT_MSG_REVALIDATE_CBK_INFO, DHT_MSG_UPGRADE_BRICKS, DHT_MSG_LK_ARRAY_INFO,
+    DHT_MSG_RENAME_NOT_LOCAL, DHT_MSG_RECONFIGURE_INFO,
+    DHT_MSG_INIT_LOCAL_SUBVOL_FAILED, DHT_MSG_SYS_CALL_GET_TIME_FAILED,
+    DHT_MSG_NO_DISK_USAGE_STATUS, DHT_MSG_SUBVOL_DOWN_ERROR,
+    DHT_MSG_REBAL_THROTTLE_INFO, DHT_MSG_COMMIT_HASH_INFO,
+    DHT_MSG_REBAL_STRUCT_SET, DHT_MSG_HAS_MIGINFO, DHT_MSG_SETTLE_HASH_FAILED,
+    DHT_MSG_DEFRAG_PROCESS_DIR_FAILED, DHT_MSG_FD_CTX_SET_FAILED,
+    DHT_MSG_STALE_LOOKUP, DHT_MSG_PARENT_LAYOUT_CHANGED,
+    DHT_MSG_LOCK_MIGRATION_FAILED, DHT_MSG_LOCK_INODE_UNREF_FAILED,
+    DHT_MSG_ASPRINTF_FAILED, DHT_MSG_DIR_LOOKUP_FAILED, DHT_MSG_INODELK_FAILED,
+    DHT_MSG_LOCK_FRAME_FAILED, DHT_MSG_LOCAL_LOCK_INIT_FAILED,
+    DHT_MSG_ENTRYLK_ERROR, DHT_MSG_INODELK_ERROR, DHT_MSG_LOC_FAILED,
+    DHT_MSG_UNKNOWN_FOP, DHT_MSG_MIGRATE_FILE_SKIPPED,
+    DHT_MSG_DIR_XATTR_HEAL_FAILED, DHT_MSG_HASHED_SUBVOL_DOWN,
+    DHT_MSG_NON_HASHED_SUBVOL_DOWN, DHT_MSG_SYNCTASK_CREATE_FAILED,
+    DHT_MSG_DIR_HEAL_ABORT, DHT_MSG_MIGRATE_SKIP, DHT_MSG_FD_CREATE_FAILED,
+    DHT_MSG_DICT_NEW_FAILED, DHT_MSG_FAILED_TO_OPEN, DHT_MSG_CREATE_FAILED,
+    DHT_MSG_FILE_NOT_EXIST, DHT_MSG_CHOWN_FAILED, DHT_MSG_FALLOCATE_FAILED,
+    DHT_MSG_FTRUNCATE_FAILED, DHT_MSG_STATFS_FAILED, DHT_MSG_WRITE_CROSS,
+    DHT_MSG_NEW_TARGET_FOUND, DHT_MSG_INSUFF_MEMORY, DHT_MSG_SET_XATTR_FAILED,
+    DHT_MSG_SET_MODE_FAILED, DHT_MSG_FILE_EXISTS_IN_DEST,
+    DHT_MSG_SYMLINK_FAILED, DHT_MSG_LINKFILE_DEL_FAILED, DHT_MSG_MKNOD_FAILED,
+    DHT_MSG_MIGRATE_CLEANUP_FAILED, DHT_MSG_LOCK_MIGRATE,
+    DHT_MSG_PARENT_BUILD_FAILED, DHT_MSG_HASHED_SUBVOL_NOT_FOUND,
+    DHT_MSG_ACQUIRE_ENTRYLK_FAILED, DHT_MSG_CREATE_DST_FAILED,
+    DHT_MSG_MIGRATION_EXIT, DHT_MSG_CHANGED_DST, DHT_MSG_TRACE_FAILED,
+    DHT_MSG_WRITE_LOCK_FAILED, DHT_MSG_GETACTIVELK_FAILED, DHT_MSG_STAT_FAILED,
+    DHT_MSG_UNLINK_PERFORM_FAILED, DHT_MSG_CLANUP_SOURCE_FILE_FAILED,
+    DHT_MSG_UNLOCK_FILE_FAILED, DHT_MSG_REMOVE_XATTR_FAILED,
+    DHT_MSG_DATA_MIGRATE_ABORT, DHT_MSG_DEFRAG_NULL, DHT_MSG_PARENT_NULL,
+    DHT_MSG_GFID_NOT_PRESENT, DHT_MSG_CHILD_LOC_FAILED,
+    DHT_MSG_SET_LOOKUP_FAILED, DHT_MSG_DIR_REMOVED, DHT_MSG_FIX_NOT_COMP,
+    DHT_MSG_SUBVOL_DETER_FAILED, DHT_MSG_LOCAL_SUBVOL, DHT_MSG_NODE_UUID,
+    DHT_MSG_SIZE_FILE, DHT_MSG_GET_DATA_SIZE_FAILED,
+    DHT_MSG_PTHREAD_JOIN_FAILED, DHT_MSG_COUNTER_THREAD_CREATE_FAILED,
+    DHT_MSG_MIGRATION_INIT_QUEUE_FAILED, DHT_MSG_PAUSED_TIMEOUT, DHT_MSG_WOKE,
+    DHT_MSG_ABORT_REBALANCE, DHT_MSG_CREATE_TASK_REBAL_FAILED,
+    DHT_MSG_REBAL_ESTIMATE_NOT_AVAIL, DHT_MSG_ADD_CHOICES_ERROR,
+    DHT_MSG_GET_CHOICES_ERROR, DHT_MSG_PREPARE_STATUS_ERROR,
+    DHT_MSG_SET_CHOICE_FAILED, DHT_MSG_SET_HASHED_SUBVOL_FAILED,
+    DHT_MSG_XATTR_HEAL_NOT_POSS, DHT_MSG_LINKTO_FILE_FAILED,
+    DHT_MSG_STALE_LINKFILE_DELETE, DHT_MSG_NO_SUBVOL_FOR_LINKTO,
+    DHT_MSG_SUBVOL_RETURNED, DHT_MSG_UNKNOWN_LOCAL_XSEL, DHT_MSG_GET_XATTR_ERR,
+    DHT_MSG_ALLOC_OR_FILL_FAILED, DHT_MSG_GET_REAL_NAME_FAILED,
+    DHT_MSG_COPY_UUID_FAILED, DHT_MSG_MDS_DETER_FAILED,
+    DHT_MSG_CREATE_REBAL_FAILED, DHT_MSG_LINK_LAYOUT_FAILED,
+    DHT_MSG_NO_SUBVOL_IN_LAYOUT, DHT_MSG_MEM_ALLOC_FAILED,
+    DHT_MSG_SET_IN_PARAMS_DICT_FAILED, DHT_MSG_LOC_COPY_FAILED,
+    DHT_MSG_PARENT_LOC_FAILED, DHT_MSG_CREATE_LOCK_FAILED,
+    DHT_MSG_PREV_ATTEMPT_FAILED, DHT_MSG_REFRESH_ATTEMPT,
+    DHT_MSG_ACQUIRE_LOCK_FAILED, DHT_MSG_CREATE_STUB_FAILED,
+    DHT_MSG_WIND_LOCK_REQ_FAILED, DHT_MSG_REFRESH_FAILED,
+    DHT_MSG_CACHED_SUBVOL_ERROR, DHT_MSG_NO_LINK_SUBVOL, DHT_MSG_SET_KEY_FAILED,
+    DHT_MSG_REMOVE_LINKTO_FAILED, DHT_MSG_LAYOUT_DICT_SET_FAILED,
+    DHT_MSG_XATTR_DICT_NULL, DHT_MSG_DUMMY_ALLOC_FAILED, DHT_MSG_DICT_IS_NULL,
+    DHT_MSG_LINK_INODE_FAILED, DHT_MSG_SELFHEAL_FAILED, DHT_MSG_NO_MDS_SUBVOL,
+    DHT_MSG_LIST_XATTRS_FAILED, DHT_MSG_RESET_INTER_XATTR_FAILED,
+    DHT_MSG_MDS_DOWN_UNABLE_TO_SET, DHT_MSG_WIND_UNLOCK_FAILED,
+    DHT_MSG_COMMIT_HASH_FAILED, DHT_MSG_UNLOCK_GFID_FAILED,
+    DHT_MSG_UNLOCK_FOLLOW_ENTRYLK, DHT_MSG_COPY_FRAME_FAILED,
+    DHT_MSG_UNLOCK_FOLLOW_LOCKS, DHT_MSG_ENTRYLK_FAILED_AFT_INODELK,
+    DHT_MSG_CALLOC_FAILED, DHT_MSG_LOCK_ALLOC_FAILED,
+    DHT_MSG_BLOCK_INODELK_FAILED,
+    DHT_MSG_LOCAL_LOCKS_STORE_FAILED_UNLOCKING_FOLLOWING_ENTRYLK,
+    DHT_MSG_ALLOC_FRAME_FAILED_NOT_UNLOCKING_FOLLOWING_ENTRYLKS,
+    DHT_MSG_DST_NULL_SET_FAILED);
+
+#define DHT_MSG_FD_CTX_SET_FAILED_STR "Failed to set fd ctx"
+#define DHT_MSG_INVALID_VALUE_STR "Different dst found in the fd ctx"
+#define DHT_MSG_UNKNOWN_FOP_STR "Unknown FOP on file"
+#define DHT_MSG_OPEN_FD_ON_DST_FAILED_STR "Failed to open the fd on file"
+#define DHT_MSG_SYNCTASK_CREATE_FAILED_STR "Failed to create synctask"
+#define DHT_MSG_ASPRINTF_FAILED_STR                                            \
+    "asprintf failed while fetching subvol from the id"
+#define DHT_MSG_HAS_MIGINFO_STR "Found miginfo in the inode ctx"
+#define DHT_MSG_FILE_LOOKUP_FAILED_STR "failed to lookup the file"
+#define DHT_MSG_INVALID_LINKFILE_STR                                           \
+    "linkto target is different from cached-subvol. treating as destination "  \
+    "subvol"
+#define DHT_MSG_GFID_MISMATCH_STR "gfid different on the target file"
+#define DHT_MSG_GET_XATTR_FAILED_STR "failed to get 'linkto' xattr"
+#define DHT_MSG_SET_INODE_CTX_FAILED_STR "failed to set inode-ctx target file"
+#define DHT_MSG_DIR_SELFHEAL_FAILED_STR "Healing of path failed"
+#define DHT_MSG_DIR_HEAL_ABORT_STR                                             \
+    "Failed to get path from subvol. Aborting directory healing"
+#define DHT_MSG_DIR_XATTR_HEAL_FAILED_STR "xattr heal failed for directory"
+#define DHT_MSG_LOCK_INODE_UNREF_FAILED_STR                                    \
+    "Found a NULL inode. Failed to unref the inode"
+#define DHT_MSG_DICT_SET_FAILED_STR "Failed to set dictionary value"
+#define DHT_MSG_NOT_LINK_FILE_ERROR_STR "got non-linkfile"
+#define DHT_MSG_CREATE_LINK_FAILED_STR "failed to initialize linkfile data"
+#define DHT_MSG_UNLINK_FAILED_STR "Unlinking linkfile on subvolume failed"
+#define DHT_MSG_MIGRATE_FILE_FAILED_STR "Migrate file failed"
+#define DHT_MSG_NO_MEMORY_STR "could not allocate memory for dict"
+#define DHT_MSG_SUBVOL_ERROR_STR "Failed to get linkto subvol"
+#define DHT_MSG_MIGRATE_HARDLINK_FILE_FAILED_STR "link failed on subvol"
+#define DHT_MSG_MIGRATE_FILE_SKIPPED_STR "Migration skipped"
+#define DHT_MSG_FD_CREATE_FAILED_STR "fd create failed"
+#define DHT_MSG_DICT_NEW_FAILED_STR "dict_new failed"
+#define DHT_MSG_FAILED_TO_OPEN_STR "failed to open"
+#define DHT_MSG_CREATE_FAILED_STR "failed to create"
+#define DHT_MSG_FILE_NOT_EXIST_STR "file does not exist"
+#define DHT_MSG_CHOWN_FAILED_STR "chown failed"
+#define DHT_MSG_FALLOCATE_FAILED_STR "fallocate failed"
+#define DHT_MSG_FTRUNCATE_FAILED_STR "ftruncate failed"
+#define DHT_MSG_STATFS_FAILED_STR "failed to get statfs"
+#define DHT_MSG_WRITE_CROSS_STR                                                \
+    "write will cross min-fre-disk for file on subvol. looking for new subvol"
+#define DHT_MSG_SUBVOL_INSUFF_SPACE_STR                                        \
+    "Could not find any subvol with space accommodating the file. Cosider "    \
+    "adding bricks"
+#define DHT_MSG_NEW_TARGET_FOUND_STR "New target found for file"
+#define DHT_MSG_INSUFF_MEMORY_STR "insufficient memory"
+#define DHT_MSG_SET_XATTR_FAILED_STR "failed to set xattr"
+#define DHT_MSG_SET_MODE_FAILED_STR "failed to set mode"
+#define DHT_MSG_FILE_EXISTS_IN_DEST_STR "file exists in destination"
+#define DHT_MSG_LINKFILE_DEL_FAILED_STR "failed to delete the linkfile"
+#define DHT_MSG_SYMLINK_FAILED_STR "symlink failed"
+#define DHT_MSG_MKNOD_FAILED_STR "mknod failed"
+#define DHT_MSG_SETATTR_FAILED_STR "failed to perform setattr"
+#define DHT_MSG_MIGRATE_CLEANUP_FAILED_STR                                     \
+    "Migrate file cleanup failed: failed to fstat file"
+#define DHT_MSG_LOCK_MIGRATE_STR "locks will be migrated for file"
+#define DHT_MSG_PARENT_BUILD_FAILED_STR                                        \
+    "failed to build parent loc, which is needed to acquire entrylk to "       \
+    "synchronize with renames on this path. Skipping migration"
+#define DHT_MSG_HASHED_SUBVOL_NOT_FOUND_STR                                    \
+    "cannot find hashed subvol which is needed to synchronize with renames "   \
+    "on this path. Skipping migration"
+#define DHT_MSG_ACQUIRE_ENTRYLK_FAILED_STR "failed to acquire entrylk on subvol"
+#define DHT_MSG_CREATE_DST_FAILED_STR "create dst failed for file"
+#define DHT_MSG_MIGRATION_EXIT_STR "Exiting migration"
+#define DHT_MSG_CHANGED_DST_STR "destination changed fo file"
+#define DHT_MSG_TRACE_FAILED_STR "Trace failed"
+#define DHT_MSG_WRITE_LOCK_FAILED_STR "write lock failed"
+#define DHT_MSG_GETACTIVELK_FAILED_STR "getactivelk failed for file"
+#define DHT_MSG_STAT_FAILED_STR "failed to do a stat"
+#define DHT_MSG_UNLINK_PERFORM_FAILED_STR "failed to perform unlink"
+#define DHT_MSG_MIGRATE_FILE_COMPLETE_STR "completed migration"
+#define DHT_MSG_CLANUP_SOURCE_FILE_FAILED_STR "failed to cleanup source file"
+#define DHT_MSG_UNLOCK_FILE_FAILED_STR "failed to unlock file"
+#define DHT_MSG_REMOVE_XATTR_FAILED_STR "remove xattr failed"
+#define DHT_MSG_SOCKET_ERROR_STR "Failed to unlink listener socket"
+#define DHT_MSG_HASHED_SUBVOL_GET_FAILED_STR "Failed to get hashed subvolume"
+#define DHT_MSG_CACHED_SUBVOL_GET_FAILED_STR "Failed to get cached subvolume"
+#define DHT_MSG_MIGRATE_DATA_FAILED_STR "migrate-data failed"
+#define DHT_MSG_DEFRAG_NULL_STR "defrag is NULL"
+#define DHT_MSG_DATA_MIGRATE_ABORT_STR                                         \
+    "Readdirp failed. Aborting data migration for dict"
+#define DHT_MSG_LAYOUT_FIX_FAILED_STR "fix layout failed"
+#define DHT_MSG_PARENT_NULL_STR "parent is NULL"
+#define DHT_MSG_GFID_NOT_PRESENT_STR "gfid not present"
+#define DHT_MSG_CHILD_LOC_FAILED_STR "Child loc build failed"
+#define DHT_MSG_SET_LOOKUP_FAILED_STR "Failed to set lookup"
+#define DHT_MSG_DIR_LOOKUP_FAILED_STR "lookup failed"
+#define DHT_MSG_DIR_REMOVED_STR "Dir renamed or removed. Skipping"
+#define DHT_MSG_READDIR_ERROR_STR "readdir failed, Aborting fix-layout"
+#define DHT_MSG_SETTLE_HASH_FAILED_STR "Settle hash failed"
+#define DHT_MSG_DEFRAG_PROCESS_DIR_FAILED_STR "gf_defrag_process_dir failed"
+#define DHT_MSG_FIX_NOT_COMP_STR                                               \
+    "Unable to retrieve fixlayout xattr. Assume background fix layout not "    \
+    "complete"
+#define DHT_MSG_SUBVOL_DETER_FAILED_STR                                        \
+    "local subvolume determination failed with error"
+#define DHT_MSG_LOCAL_SUBVOL_STR "local subvol"
+#define DHT_MSG_NODE_UUID_STR "node uuid"
+#define DHT_MSG_SIZE_FILE_STR "Total size files"
+#define DHT_MSG_GET_DATA_SIZE_FAILED_STR                                       \
+    "Failed to get the total data size. Unable to estimate time to complete "  \
+    "rebalance"
+#define DHT_MSG_PTHREAD_JOIN_FAILED_STR                                        \
+    "file_counter_thread: pthread_join failed"
+#define DHT_MSG_COUNTER_THREAD_CREATE_FAILED_STR                               \
+    "Failed to create the file counter thread"
+#define DHT_MSG_MIGRATION_INIT_QUEUE_FAILED_STR                                \
+    "Failed to initialise migration queue"
+#define DHT_MSG_REBALANCE_STOPPED_STR "Received stop command on rebalance"
+#define DHT_MSG_PAUSED_TIMEOUT_STR "Request pause timer timeout"
+#define DHT_MSG_WOKE_STR "woken"
+#define DHT_MSG_ABORT_REBALANCE_STR "Aborting rebalance"
+#define DHT_MSG_REBALANCE_START_FAILED_STR                                     \
+    "Failed to start rebalance: look up on / failed"
+#define DHT_MSG_CREATE_TASK_REBAL_FAILED_STR                                   \
+    "Could not create task for rebalance"
+#define DHT_MSG_REBAL_ESTIMATE_NOT_AVAIL_STR                                   \
+    "Rebalance estimates will not be available"
+#define DHT_MSG_REBALANCE_STATUS_STR "Rebalance status"
+#define DHT_MSG_DATA_NULL_STR "data value is NULL"
+#define DHT_MSG_ADD_CHOICES_ERROR_STR "Error to add choices in buffer"
+#define DHT_MSG_GET_CHOICES_ERROR_STR "Error to get choices"
+#define DHT_MSG_PREPARE_STATUS_ERROR_STR "Error to prepare status"
+#define DHT_MSG_SET_CHOICE_FAILED_STR "Failed to set full choice"
+#define DHT_MSG_AGGREGATE_QUOTA_XATTR_FAILED_STR                               \
+    "Failed to aggregate quota xattr"
+#define DHT_MSG_FILE_TYPE_MISMATCH_STR                                         \
+    "path exists as a file on one subvolume and directory on another. Please " \
+    "fix it manually"
+#define DHT_MSG_LAYOUT_SET_FAILED_STR "failed to set layout for subvolume"
+#define DHT_MSG_LAYOUT_MERGE_FAILED_STR "failed to merge layouts for subvolume"
+#define DHT_MSG_SET_HASHED_SUBVOL_FAILED_STR "Failed to set hashed subvolume"
+#define DHT_MSG_XATTR_HEAL_NOT_POSS_STR                                        \
+    "No gfid exists for path. so healing xattr is not possible"
+#define DHT_MSG_REVALIDATE_CBK_INFO_STR "Revalidate: subvolume returned -1"
+#define DHT_MSG_LAYOUT_MISMATCH_STR "Mismatching layouts"
+#define DHT_MSG_UNLINK_LOOKUP_INFO_STR "lookup_unlink retuened"
+#define DHT_MSG_LINKTO_FILE_FAILED_STR                                         \
+    "Could not unlink the linkto file as either fd is open and/or linkto "     \
+    "xattr is set"
+#define DHT_MSG_LAYOUT_PRESET_FAILED_STR                                       \
+    "Could not set pre-set layout for subvolume"
+#define DHT_MSG_FILE_ON_MULT_SUBVOL_STR                                        \
+    "multiple subvolumes have file (preferably rename the file in the "        \
+    "backend, and do a fresh lookup"
+#define DHT_MSG_STALE_LINKFILE_DELETE_STR                                      \
+    "attempting deletion of stale linkfile"
+#define DHT_MSG_LINK_FILE_LOOKUP_INFO_STR "Lookup on following linkfile"
+#define DHT_MSG_NO_SUBVOL_FOR_LINKTO_STR "No link subvolume for linkto"
+#define DHT_MSG_SUBVOL_RETURNED_STR "Subvolume returned -1"
+#define DHT_MSG_UNKNOWN_LOCAL_XSEL_STR "Unknown local->xsel"
+#define DHT_MSG_DICT_GET_FAILED_STR "Failed to get"
+#define DHT_MSG_UUID_PARSE_ERROR_STR "Failed to parse uuid"
+#define DHT_MSG_GET_XATTR_ERR_STR "getxattr err for dir"
+#define DHT_MSG_ALLOC_OR_FILL_FAILED_STR "alloc or fill failed"
+#define DHT_MSG_UPGRADE_BRICKS_STR                                             \
+    "At least one of the bricks does not support this operation. Please "      \
+    "upgrade all bricks"
+#define DHT_MSG_GET_REAL_NAME_FAILED_STR "Failed to get real filename"
+#define DHT_MSG_LAYOUT_NULL_STR "Layout is NULL"
+#define DHT_MSG_COPY_UUID_FAILED_STR "Failed to copy node uuid key"
+#define DHT_MSG_MDS_DETER_FAILED_STR                                           \
+    "Cannot determine MDS, fetching xattr randomly from a subvol"
+#define DHT_MSG_HASHED_SUBVOL_DOWN_STR                                         \
+    "MDS is down for path, so fetching xattr randomly from subvol"
+#define DHT_MSG_CREATE_REBAL_FAILED_STR                                        \
+    "failed to create a new rebalance synctask"
+#define DHT_MSG_FIX_LAYOUT_INFO_STR "fixing the layout"
+#define DHT_MSG_OPERATION_NOT_SUP_STR "wrong directory-spread-count value"
+#define DHT_MSG_LINK_LAYOUT_FAILED_STR "failed to link the layout in inode"
+#define DHT_MSG_NO_SUBVOL_IN_LAYOUT_STR "no subvolume in layout for path"
+#define DHT_MSG_INODE_LK_ERROR_STR "mknod lock failed for file"
+#define DHT_MSG_MEM_ALLOC_FAILED_STR "mem allocation failed"
+#define DHT_MSG_PARENT_LAYOUT_CHANGED_STR                                      \
+    "extracting in-memory layout of parent failed"
+#define DHT_MSG_SET_IN_PARAMS_DICT_FAILED_STR                                  \
+    "setting in params dictionary failed"
+#define DHT_MSG_LOC_COPY_FAILED_STR "loc_copy failed"
+#define DHT_MSG_LOC_FAILED_STR "parent loc build failed"
+#define DHT_MSG_PARENT_LOC_FAILED_STR "locking parent failed"
+#define DHT_MSG_CREATE_LOCK_FAILED_STR "Create lock failed"
+#define DHT_MSG_PREV_ATTEMPT_FAILED_STR                                        \
+    "mkdir loop detected. parent layout didn't change even though previous "   \
+    "attempt of mkdir failed because of in-memory layout not matching with "   \
+    "that on disk."
+#define DHT_MSG_REFRESH_ATTEMPT_STR                                            \
+    "mkdir parent layout changed. Attempting a refresh and then a retry"
+#define DHT_MSG_ACQUIRE_LOCK_FAILED_STR                                        \
+    "Acquiring lock on parent to guard against layout-change failed"
+#define DHT_MSG_CREATE_STUB_FAILED_STR "creating stub failed"
+#define DHT_MSG_WIND_LOCK_REQ_FAILED_STR                                       \
+    "cannot wind lock request to guard parent layout"
+#define DHT_MSG_REFRESH_FAILED_STR "refreshing parent layout failed."
+#define DHT_MSG_CACHED_SUBVOL_ERROR_STR "On cached subvol"
+#define DHT_MSG_NO_LINK_SUBVOL_STR "Linkfile does not have link subvolume"
+#define DHT_MSG_SET_KEY_FAILED_STR "failed to set key"
+#define DHT_MSG_CHILD_DOWN_STR "Received CHILD_DOWN. Exiting"
+#define DHT_MSG_LOG_FIXED_LAYOUT_STR "log layout fixed"
+#define DHT_MSG_REBAL_STRUCT_SET_STR "local->rebalance already set"
+#define DHT_MSG_REMOVE_LINKTO_FAILED_STR "Removal of linkto failed at subvol"
+#define DHT_MSG_LAYOUT_DICT_SET_FAILED_STR "dht layout dict set failed"
+#define DHT_MSG_SUBVOL_INFO_STR "creating subvolume"
+#define DHT_MSG_COMPUTE_HASH_FAILED_STR "hash computation failed"
+#define DHT_MSG_INVALID_DISK_LAYOUT_STR                                        \
+    "Invalid disk layout: Catastrophic error layout with unknown type found"
+#define DHT_MSG_LAYOUT_SORT_FAILED_STR "layout sort failed"
+#define DHT_MSG_ANOMALIES_INFO_STR "Found anomalies"
+#define DHT_MSG_XATTR_DICT_NULL_STR "xattr dictionary is NULL"
+#define DHT_MSG_DISK_LAYOUT_MISSING_STR "Disk layout missing"
+#define DHT_MSG_LAYOUT_INFO_STR "layout info"
+#define DHT_MSG_SUBVOL_NO_LAYOUT_INFO_STR "no pre-set layout for subvol"
+#define DHT_MSG_SELFHEAL_XATTR_FAILED_STR "layout setxattr failed"
+#define DHT_MSG_DIR_SELFHEAL_XATTR_FAILED_STR "Directory self heal xattr failed"
+#define DHT_MSG_DUMMY_ALLOC_FAILED_STR "failed to allocate dummy layout"
+#define DHT_MSG_DICT_IS_NULL_STR                                               \
+    "dict is NULL, need to make sure gfids are same"
+#define DHT_MSG_ENTRYLK_ERROR_STR "acquiring entrylk after inodelk failed"
+#define DHT_MSG_NO_DISK_USAGE_STATUS_STR "no du stats"
+#define DHT_MSG_LINK_INODE_FAILED_STR "linking inode failed"
+#define DHT_MSG_SELFHEAL_FAILED_STR "Directory selfheal failed"
+#define DHT_MSG_NO_MDS_SUBVOL_STR "No mds subvol"
+#define DHT_MSG_LIST_XATTRS_FAILED_STR "failed to list xattrs"
+#define DHT_MSG_RESET_INTER_XATTR_FAILED_STR "Failed to reset internal xattr"
+#define DHT_MSG_MDS_DOWN_UNABLE_TO_SET_STR                                     \
+    "mds subvol is down, unable to set xattr"
+#define DHT_MSG_DIR_ATTR_HEAL_FAILED_STR                                       \
+    "Directory attr heal failed. Failed to set uid/gid"
+#define DHT_MSG_WIND_UNLOCK_FAILED_STR                                         \
+    "Winding unlock failed: stale locks left on brick"
+#define DHT_MSG_COMMIT_HASH_FAILED_STR "Directory commit hash updaten failed"
+#define DHT_MSG_LK_ARRAY_INFO_STR "lk info"
+#define DHT_MSG_UNLOCK_GFID_FAILED_STR                                         \
+    "unlock failed on gfid: stale lock might be left"
+#define DHT_MSG_UNLOCKING_FAILED_STR "unlocking failed"
+#define DHT_MSG_UNLOCK_FOLLOW_ENTRYLK_STR "not unlocking following entrylks"
+#define DHT_MSG_COPY_FRAME_FAILED_STR "copy frame failed"
+#define DHT_MSG_UNLOCK_FOLLOW_LOCKS_STR "not unlocking following locks"
+#define DHT_MSG_INODELK_FAILED_STR "inodelk failed on subvol"
+#define DHT_MSG_LOCK_FRAME_FAILED_STR "memory allocation failed for lock_frame"
+#define DHT_MSG_LOCAL_LOCK_INIT_FAILED_STR "dht_local_lock_init failed"
+#define DHT_MSG_ENTRYLK_FAILED_AFT_INODELK_STR                                 \
+    "dht_blocking_entrylk failed after taking inodelk"
+#define DHT_MSG_BLOCK_INODELK_FAILED_STR "dht_blocking_inodelk failed"
+#define DHT_MSG_CALLOC_FAILED_STR "calloc failed"
+#define DHT_MSG_LOCK_ALLOC_FAILED_STR "lock allocation failed"
+#define DHT_MSG_ALLOC_FRAME_FAILED_NOT_UNLOCKING_FOLLOWING_ENTRYLKS_STR        \
+    "cannot allocate a frame, not unlocking following entrylks"
+#define DHT_MSG_LOCAL_LOCKS_STORE_FAILED_UNLOCKING_FOLLOWING_ENTRYLK_STR       \
+    "storing locks in local failed, not unlocking following entrylks"
+#define DHT_MSG_DST_NULL_SET_FAILED_STR                                        \
+    "src or dst is NULL, Failed to set dictionary value"
+
+#endif /* _DHT_MESSAGES_H_ */
diff --git a/xlators/cluster/dht/src/dht-rebalance.c b/xlators/cluster/dht/src/dht-rebalance.c
new file mode 100644
index 00000000000..8ba8082bd86
--- /dev/null
+++ b/xlators/cluster/dht/src/dht-rebalance.c
@@ -0,0 +1,4702 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "dht-common.h"
+#include <glusterfs/syscall.h>
+#include <fnmatch.h>
+#include <signal.h>
+#include <glusterfs/events.h>
+#include "glusterfs/compat-errno.h"  // for ENODATA on BSD
+
+#define GF_DISK_SECTOR_SIZE 512
+#define DHT_REBALANCE_PID 4242        /* Change it if required */
+#define DHT_REBALANCE_BLKSIZE 1048576 /* 1 MB */
+#define MAX_MIGRATE_QUEUE_COUNT 500
+#define MIN_MIGRATE_QUEUE_COUNT 200
+#define MAX_REBAL_TYPE_SIZE 16
+#define FILE_CNT_INTERVAL 600       /* 10 mins */
+#define ESTIMATE_START_INTERVAL 600 /* 10 mins */
+#define HARDLINK_MIG_INPROGRESS -2
+#define SKIP_MIGRATION_FD_POSITIVE -3
+#ifndef MAX
+#define MAX(a, b) (((a) > (b)) ? (a) : (b))
+#endif
+
+#define GF_CRAWL_INDEX_MOVE(idx, sv_cnt)                                       \
+    {                                                                          \
+        idx++;                                                                 \
+        idx %= sv_cnt;                                                         \
+    }
+
+uint64_t g_totalfiles = 0;
+uint64_t g_totalsize = 0;
+
+void
+gf_defrag_free_dir_dfmeta(struct dir_dfmeta *meta, int local_subvols_cnt)
+{
+    int i = 0;
+
+    if (meta) {
+        for (i = 0; i < local_subvols_cnt; i++) {
+            if (meta->equeue)
+                gf_dirent_free(&meta->equeue[i]);
+            if (meta->lfd && meta->lfd[i])
+                fd_unref(meta->lfd[i]);
+        }
+
+        GF_FREE(meta->equeue);
+        GF_FREE(meta->head);
+        GF_FREE(meta->iterator);
+        GF_FREE(meta->offset_var);
+        GF_FREE(meta->fetch_entries);
+        GF_FREE(meta->lfd);
+        GF_FREE(meta);
+    }
+}
+
+void
+gf_defrag_free_container(struct dht_container *container)
+{
+    if (container) {
+        gf_dirent_entry_free(container->df_entry);
+
+        if (container->parent_loc) {
+            loc_wipe(container->parent_loc);
+        }
+
+        GF_FREE(container->parent_loc);
+
+        GF_FREE(container);
+    }
+}
+
+void
+dht_set_global_defrag_error(gf_defrag_info_t *defrag, int ret)
+{
+    LOCK(&defrag->lock);
+    {
+        defrag->global_error = ret;
+    }
+    UNLOCK(&defrag->lock);
+    return;
+}
+
+static int
+dht_send_rebalance_event(xlator_t *this, int cmd, gf_defrag_status_t status)
+{
+    int ret = -1;
+    char *volname = NULL;
+    char *tmpstr = NULL;
+    char *ptr = NULL;
+    char *suffix = "-dht";
+    int len = 0;
+
+    eventtypes_t event = EVENT_LAST;
+
+    switch (status) {
+        case GF_DEFRAG_STATUS_COMPLETE:
+            event = EVENT_VOLUME_REBALANCE_COMPLETE;
+            break;
+        case GF_DEFRAG_STATUS_FAILED:
+            event = EVENT_VOLUME_REBALANCE_FAILED;
+            break;
+        case GF_DEFRAG_STATUS_STOPPED:
+            event = EVENT_VOLUME_REBALANCE_STOP;
+            break;
+        default:
+            break;
+    }
+
+    /* DHT volume */
+    len = strlen(this->name) - strlen(suffix);
+    tmpstr = gf_strdup(this->name);
+    if (tmpstr) {
+        ptr = tmpstr + len;
+        if (!strcmp(ptr, suffix)) {
+            tmpstr[len] = '\0';
+            volname = tmpstr;
+        }
+    }
+
+    if (!volname) {
+        /* Better than nothing */
+        volname = this->name;
+    }
+
+    if (event != EVENT_LAST) {
+        gf_event(event, "volume=%s", volname);
+    }
+
+    GF_FREE(tmpstr);
+    return ret;
+}
+
+static void
+dht_strip_out_acls(dict_t *dict)
+{
+    if (dict) {
+        dict_del(dict, "trusted.SGI_ACL_FILE");
+        dict_del(dict, POSIX_ACL_ACCESS_XATTR);
+    }
+}
+
+/*
+   return values:
+   -1 : failure
+   -2 : success
+
+Hard link migration is carried out in three stages.
+
+(Say there are n hardlinks)
+Stage 1: Setting the new hashed subvol information on the 1st hardlink
+         encountered (linkto setxattr)
+
+Stage 2: Creating hardlinks on new hashed subvol for the 2nd to (n-1)th
+         hardlink
+
+Stage 3: Physical migration of the data file for nth hardlink
+
+Why to deem "-2" as success and not "0":
+
+   dht_migrate_file expects return value "0" from _is_file_migratable if
+the file has to be migrated.
+
+   _is_file_migratable returns zero only when it is called with the
+flag "GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS".
+
+   gf_defrag_handle_hardlink calls dht_migrate_file for physical migration
+of the data file with the flag "GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS"
+
+Hence, gf_defrag_handle_hardlink returning "0" for success will force
+"dht_migrate_file" to migrate each of the hardlink which is not intended.
+
+For each of the three stage mentioned above "-2" will be returned and will
+be converted to "0" in dht_migrate_file.
+
+*/
+
+int32_t
+gf_defrag_handle_hardlink(xlator_t *this, loc_t *loc, int *fop_errno)
+{
+    int32_t ret = -1;
+    xlator_t *cached_subvol = NULL;
+    xlator_t *hashed_subvol = NULL;
+    xlator_t *linkto_subvol = NULL;
+    data_t *data = NULL;
+    struct iatt iatt = {
+        0,
+    };
+    int32_t op_errno = 0;
+    dht_conf_t *conf = NULL;
+    gf_loglevel_t loglevel = 0;
+    dict_t *link_xattr = NULL;
+    dict_t *dict = NULL;
+    dict_t *xattr_rsp = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+
+    *fop_errno = EINVAL;
+
+    GF_VALIDATE_OR_GOTO("defrag", loc, out);
+    GF_VALIDATE_OR_GOTO("defrag", loc->name, out);
+    GF_VALIDATE_OR_GOTO("defrag", this, out);
+    GF_VALIDATE_OR_GOTO("defrag", this->private, out);
+
+    conf = this->private;
+
+    if (gf_uuid_is_null(loc->pargfid)) {
+        gf_msg("", GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed :"
+               "loc->pargfid is NULL for %s",
+               loc->path);
+        *fop_errno = EINVAL;
+        ret = -1;
+        goto out;
+    }
+
+    if (gf_uuid_is_null(loc->gfid)) {
+        gf_msg("", GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed :"
+               "loc->gfid is NULL for %s",
+               loc->path);
+        *fop_errno = EINVAL;
+        ret = -1;
+        goto out;
+    }
+
+    link_xattr = dict_new();
+    if (!link_xattr) {
+        ret = -1;
+        *fop_errno = ENOMEM;
+        goto out;
+    }
+
+    /*
+      Parallel migration can lead to migration of the hard link multiple
+      times which can lead to data loss. Hence, adding a fresh lookup to
+      decide whether migration is required or not.
+
+      Elaborating the scenario for let say 10 hardlinks [link{1..10}]:
+          Let say the first hard link "link1"  does the setxattr of the
+      new hashed subvolume info on the cached file. As there are multiple
+      threads working, we might have already all the links created on the
+      new hashed by the time we reach hardlink let say link5. Now the
+      number of links on hashed is equal to that of cached. Hence, file
+      migration will happen for link6.
+
+             Cached                                 Hashed
+      --------T link6                        rwxrwxrwx   link6
+
+      Now post above state all the link file on the cached will be zero
+      byte linkto files. Hence, if we still do migration for the following
+      files link{7..10}, we will end up migrating 0 data leading to data
+      loss.
+            Hence, a lookup can make sure whether we need to migrate the
+      file or not.
+    */
+
+    dict = dict_new();
+    if (!dict) {
+        ret = -1;
+        *fop_errno = ENOMEM;
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+               "could not allocate memory for dict");
+        goto out;
+    }
+
+    ret = dict_set_int32(dict, conf->link_xattr_name, 256);
+    if (ret) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed:"
+               "%s: failed to set 'linkto' key in dict",
+               loc->path);
+        goto out;
+    }
+
+    ret = syncop_lookup(this, loc, &stbuf, NULL, dict, &xattr_rsp);
+    if (ret) {
+        /*Ignore ENOENT and ESTALE as file might have been
+          migrated already*/
+        if (-ret == ENOENT || -ret == ESTALE) {
+            ret = -2;
+            goto out;
+        }
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed:%s lookup failed with ret = %d", loc->path,
+               ret);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    cached_subvol = dht_subvol_get_cached(this, loc->inode);
+    if (!cached_subvol) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed :"
+               "Failed to get cached subvol"
+               " for %s on %s",
+               loc->name, this->name);
+        *fop_errno = EINVAL;
+        ret = -1;
+        goto out;
+    }
+
+    hashed_subvol = dht_subvol_get_hashed(this, loc);
+    if (!hashed_subvol) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed :"
+               "Failed to get hashed subvol"
+               " for %s on %s",
+               loc->name, this->name);
+        *fop_errno = EINVAL;
+        ret = -1;
+        goto out;
+    }
+
+    /* Hardlink migration happens only with remove-brick. So this condition will
+     * be true only when the migration has happened. In case hardlinks are
+     * migrated for rebalance case, remove this check. Having this check here
+     * avoid redundant calls below*/
+    if (hashed_subvol == cached_subvol) {
+        ret = -2;
+        goto out;
+    }
+
+    gf_log(this->name, GF_LOG_INFO,
+           "Attempting to migrate hardlink %s "
+           "with gfid %s from %s -> %s",
+           loc->name, uuid_utoa(loc->gfid), cached_subvol->name,
+           hashed_subvol->name);
+
+    data = dict_get(xattr_rsp, conf->link_xattr_name);
+    /* set linkto on cached -> hashed if not present, else link it */
+    if (!data) {
+        ret = dict_set_str(link_xattr, conf->link_xattr_name,
+                           hashed_subvol->name);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Migrate file failed :"
+                   "Failed to set dictionary value:"
+                   " key = %s for %s",
+                   conf->link_xattr_name, loc->name);
+            *fop_errno = ENOMEM;
+            ret = -1;
+            goto out;
+        }
+
+        ret = syncop_setxattr(cached_subvol, loc, link_xattr, 0, NULL, NULL);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Migrate file failed :"
+                   "Linkto setxattr failed %s -> %s",
+                   cached_subvol->name, loc->name);
+            *fop_errno = -ret;
+            ret = -1;
+            goto out;
+        }
+
+        gf_msg_debug(this->name, 0,
+                     "hardlink target subvol created on %s "
+                     ",cached %s, file %s",
+                     hashed_subvol->name, cached_subvol->name, loc->path);
+
+        ret = -2;
+        goto out;
+    } else {
+        linkto_subvol = dht_linkfile_subvol(this, NULL, NULL, xattr_rsp);
+        if (!linkto_subvol) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SUBVOL_ERROR,
+                   "Failed to get "
+                   "linkto subvol for %s",
+                   loc->name);
+        } else {
+            hashed_subvol = linkto_subvol;
+        }
+
+        ret = syncop_link(hashed_subvol, loc, loc, &iatt, NULL, NULL);
+        if (ret) {
+            op_errno = -ret;
+            ret = -1;
+
+            loglevel = (op_errno == EEXIST) ? GF_LOG_DEBUG : GF_LOG_ERROR;
+            gf_msg(this->name, loglevel, op_errno,
+                   DHT_MSG_MIGRATE_HARDLINK_FILE_FAILED,
+                   "link of %s -> %s"
+                   " failed on  subvol %s",
+                   loc->name, uuid_utoa(loc->gfid), hashed_subvol->name);
+            if (op_errno != EEXIST) {
+                *fop_errno = op_errno;
+                goto out;
+            }
+        } else {
+            gf_msg_debug(this->name, 0,
+                         "syncop_link successful for"
+                         " hardlink %s on subvol %s, cached %s",
+                         loc->path, hashed_subvol->name, cached_subvol->name);
+        }
+    }
+
+    ret = syncop_lookup(hashed_subvol, loc, &iatt, NULL, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed :Failed lookup %s on %s ", loc->name,
+               hashed_subvol->name);
+
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    /* There is a race where on the target subvol for the hardlink
+     * (note: hash subvol for the hardlink might differ from this), some
+     * other client(non-rebalance) would have created a linkto file for that
+     * hardlink as part of lookup. So let say there are 10 hardlinks, on the
+     * 5th hardlink it self the hardlinks might have migrated. Now for
+     * (6..10th) hardlinks the cached and target would be same as the file
+     * has already migrated. Hence this check is needed  */
+    if (cached_subvol == hashed_subvol) {
+        gf_msg_debug(this->name, 0,
+                     "source %s and destination %s "
+                     "for hardlink %s are same",
+                     cached_subvol->name, hashed_subvol->name, loc->path);
+        ret = -2;
+        goto out;
+    }
+
+    if (iatt.ia_nlink == stbuf.ia_nlink) {
+        ret = dht_migrate_file(this, loc, cached_subvol, hashed_subvol,
+                               GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS, fop_errno);
+        if (ret) {
+            goto out;
+        }
+    }
+    ret = -2;
+out:
+    if (link_xattr)
+        dict_unref(link_xattr);
+
+    if (xattr_rsp)
+        dict_unref(xattr_rsp);
+
+    if (dict)
+        dict_unref(dict);
+
+    return ret;
+}
+
+static int
+__check_file_has_hardlink(xlator_t *this, loc_t *loc, struct iatt *stbuf,
+                          dict_t *xattrs, int flags, gf_defrag_info_t *defrag,
+                          dht_conf_t *conf, int *fop_errno)
+{
+    int ret = 0;
+
+    if (flags == GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS) {
+        ret = 0;
+        return ret;
+    }
+    if (stbuf->ia_nlink > 1) {
+        /* support for decomission */
+        if (flags == GF_DHT_MIGRATE_HARDLINK) {
+            synclock_lock(&conf->link_lock);
+            ret = gf_defrag_handle_hardlink(this, loc, fop_errno);
+            synclock_unlock(&conf->link_lock);
+            /*
+            Returning zero will force the file to be remigrated.
+            Checkout gf_defrag_handle_hardlink for more information.
+            */
+            if (ret && ret != -2) {
+                gf_msg(this->name, GF_LOG_WARNING, 0,
+                       DHT_MSG_MIGRATE_FILE_FAILED,
+                       "Migrate file failed:"
+                       "%s: failed to migrate file with link",
+                       loc->path);
+            }
+        } else {
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Migration skipped for:"
+                   "%s: file has hardlinks",
+                   loc->path);
+            *fop_errno = ENOTSUP;
+            ret = 1;
+        }
+    }
+
+    return ret;
+}
+
+/*
+     return values
+     0 : File will be migrated
+    -2 : File will not be migrated
+         (This is the return value from gf_defrag_handle_hardlink. Checkout
+         gf_defrag_handle_hardlink for description of "returning -2")
+    -1 : failure
+*/
+static int
+__is_file_migratable(xlator_t *this, loc_t *loc, struct iatt *stbuf,
+                     dict_t *xattrs, int flags, gf_defrag_info_t *defrag,
+                     dht_conf_t *conf, int *fop_errno)
+{
+    int ret = -1;
+    int lock_count = 0;
+
+    if (IA_ISDIR(stbuf->ia_type)) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed:"
+               "%s: migrate-file called on directory",
+               loc->path);
+        *fop_errno = EISDIR;
+        ret = -1;
+        goto out;
+    }
+
+    if (!conf->lock_migration_enabled) {
+        ret = dict_get_int32(xattrs, GLUSTERFS_POSIXLK_COUNT, &lock_count);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Migrate file failed:"
+                   "%s: Unable to get lock count for file",
+                   loc->path);
+            *fop_errno = EINVAL;
+            ret = -1;
+            goto out;
+        }
+
+        if (lock_count) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Migrate file failed: %s: File has locks."
+                   " Skipping file migration",
+                   loc->path);
+            *fop_errno = ENOTSUP;
+            ret = 1;
+            goto out;
+        }
+    }
+
+    /* Check if file has hardlink*/
+    ret = __check_file_has_hardlink(this, loc, stbuf, xattrs, flags, defrag,
+                                    conf, fop_errno);
+out:
+    return ret;
+}
+
+static int
+__dht_rebalance_create_dst_file(xlator_t *this, xlator_t *to, xlator_t *from,
+                                loc_t *loc, struct iatt *stbuf, fd_t **dst_fd,
+                                int *fop_errno, int file_has_holes)
+{
+    int ret = -1;
+    int ret2 = -1;
+    fd_t *fd = NULL;
+    struct iatt new_stbuf = {
+        0,
+    };
+    struct iatt check_stbuf = {
+        0,
+    };
+    dht_conf_t *conf = NULL;
+    dict_t *dict = NULL;
+    dict_t *xdata = NULL;
+
+    conf = this->private;
+
+    dict = dict_new();
+    if (!dict) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+               "dictionary allocation failed for"
+               "path:%s",
+               loc->path);
+        goto out;
+    }
+    ret = dict_set_gfuuid(dict, "gfid-req", stbuf->ia_gfid, true);
+    if (ret) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+               "%s: failed to set dictionary value: key = gfid-req", loc->path);
+        goto out;
+    }
+
+    ret = dict_set_str(dict, conf->link_xattr_name, from->name);
+    if (ret) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+               "%s: failed to set dictionary value: key = %s ", loc->path,
+               conf->link_xattr_name);
+        goto out;
+    }
+
+    fd = fd_create(loc->inode, DHT_REBALANCE_PID);
+    if (!fd) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: fd create failed (destination)", loc->path);
+        goto out;
+    }
+
+    xdata = dict_new();
+    if (!xdata) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: dict_new failed)", loc->path);
+        goto out;
+    }
+
+    ret = dict_set_int32_sizen(xdata, GF_CLEAN_WRITE_PROTECTION, 1);
+    if (ret) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+               "%s: failed to set dictionary value: key = %s ", loc->path,
+               GF_CLEAN_WRITE_PROTECTION);
+        goto out;
+    }
+
+    ret = syncop_lookup(to, loc, &new_stbuf, NULL, xdata, NULL);
+    if (!ret) {
+        /* File exits in the destination, check if gfid matches */
+        if (gf_uuid_compare(stbuf->ia_gfid, new_stbuf.ia_gfid) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_GFID_MISMATCH,
+                   "file %s exists in %s with different gfid", loc->path,
+                   to->name);
+            *fop_errno = EINVAL;
+            ret = -1;
+            goto out;
+        }
+    }
+    if ((ret < 0) && (-ret != ENOENT)) {
+        /* File exists in destination, but not accessible */
+        gf_msg(THIS->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: failed to lookup file", loc->path);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    /* Create the destination with LINKFILE mode, and linkto xattr,
+       if the linkfile already exists, just open the file */
+    if (!ret) {
+        /*
+         * File already present, just open the file.
+         */
+        ret = syncop_open(to, loc, O_RDWR, fd, NULL, NULL);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "failed to open %s on %s", loc->path, to->name);
+            *fop_errno = -ret;
+            ret = -1;
+            goto out;
+        }
+    } else {
+        ret = syncop_create(to, loc, O_RDWR, DHT_LINKFILE_MODE, fd, &new_stbuf,
+                            dict, NULL);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "failed to create %s on %s", loc->path, to->name);
+            *fop_errno = -ret;
+            ret = -1;
+            goto out;
+        }
+    }
+
+    fd_bind(fd);
+
+    /*Reason of doing lookup after create again:
+     *In the create, there is some time-gap between opening fd at the
+     *server (posix_layer) and binding it in server (incrementing fd count),
+     *so if in that time-gap, if other process sends unlink considering it
+     *as a linkto file, because inode->fd count will be 0, so file will be
+     *unlinked at the backend. And because further operations are performed
+     *on fd, so though migration will be done but will end with no file
+     *at  the backend.
+     */
+
+    ret = syncop_lookup(to, loc, &check_stbuf, NULL, NULL, NULL);
+    if (!ret) {
+        if (gf_uuid_compare(stbuf->ia_gfid, check_stbuf.ia_gfid) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_GFID_MISMATCH,
+                   "file %s exists in %s with different gfid,"
+                   "found in lookup after create",
+                   loc->path, to->name);
+            *fop_errno = EINVAL;
+            ret = -1;
+            goto out;
+        }
+    }
+
+    if (-ret == ENOENT) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: file does not exist"
+               "on %s",
+               loc->path, to->name);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    ret = syncop_fsetattr(to, fd, stbuf, (GF_SET_ATTR_UID | GF_SET_ATTR_GID),
+                          NULL, NULL, NULL, NULL);
+    if (ret < 0) {
+        *fop_errno = -ret;
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "chown failed for %s on %s", loc->path, to->name);
+    }
+
+    /* No need to bother about 0 byte size files */
+    if (stbuf->ia_size > 0) {
+        if (conf->use_fallocate && !file_has_holes) {
+            ret = syncop_fallocate(to, fd, 0, 0, stbuf->ia_size, NULL, NULL);
+            if (ret < 0) {
+                if (ret == -EOPNOTSUPP || ret == -EINVAL || ret == -ENOSYS) {
+                    conf->use_fallocate = _gf_false;
+                } else {
+                    gf_msg(this->name, GF_LOG_ERROR, -ret,
+                           DHT_MSG_MIGRATE_FILE_FAILED,
+                           "fallocate failed for %s on %s", loc->path,
+                           to->name);
+
+                    *fop_errno = -ret;
+
+                    /* fallocate does not release the space
+                     * in some cases
+                     */
+                    ret2 = syncop_ftruncate(to, fd, 0, NULL, NULL, NULL, NULL);
+                    if (ret2 < 0) {
+                        gf_msg(this->name, GF_LOG_WARNING, -ret2,
+                               DHT_MSG_MIGRATE_FILE_FAILED,
+                               "ftruncate failed for "
+                               "%s on %s",
+                               loc->path, to->name);
+                    }
+                    goto out;
+                }
+            }
+        } else {
+            ret = syncop_ftruncate(to, fd, stbuf->ia_size, NULL, NULL, NULL,
+                                   NULL);
+            if (ret < 0) {
+                *fop_errno = -ret;
+                gf_msg(this->name, GF_LOG_WARNING, -ret,
+                       DHT_MSG_MIGRATE_FILE_FAILED,
+                       "ftruncate failed for %s on %s", loc->path, to->name);
+            }
+        }
+    }
+
+    /* success */
+    ret = 0;
+
+    if (dst_fd)
+        *dst_fd = fd;
+
+out:
+    if (ret) {
+        if (fd) {
+            fd_unref(fd);
+        }
+    }
+    if (dict)
+        dict_unref(dict);
+
+    if (xdata)
+        dict_unref(xdata);
+
+    return ret;
+}
+
+static int
+__dht_check_free_space(xlator_t *this, xlator_t *to, xlator_t *from, loc_t *loc,
+                       struct iatt *stbuf, int flag, dht_conf_t *conf,
+                       gf_boolean_t *target_changed, xlator_t **new_subvol,
+                       int *fop_errno)
+{
+    struct statvfs src_statfs = {
+        0,
+    };
+    struct statvfs dst_statfs = {
+        0,
+    };
+    int ret = -1;
+    dict_t *xdata = NULL;
+    dht_layout_t *layout = NULL;
+    uint64_t src_statfs_blocks = 1;
+    uint64_t dst_statfs_blocks = 1;
+    double dst_post_availspacepercent = 0;
+    double src_post_availspacepercent = 0;
+    uint64_t file_blocks = 0;
+    uint64_t src_total_blocks = 0;
+    uint64_t dst_total_blocks = 0;
+
+    xdata = dict_new();
+    if (!xdata) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+               "failed to allocate dictionary");
+        goto out;
+    }
+
+    ret = dict_set_int8(xdata, GF_INTERNAL_IGNORE_DEEM_STATFS, 1);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Failed to set " GF_INTERNAL_IGNORE_DEEM_STATFS " in dict");
+        ret = -1;
+        *fop_errno = ENOMEM;
+        goto out;
+    }
+
+    ret = syncop_statfs(from, loc, &src_statfs, xdata, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "failed to get statfs of %s on %s", loc->path, from->name);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    ret = syncop_statfs(to, loc, &dst_statfs, xdata, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "failed to get statfs of %s on %s", loc->path, to->name);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    gf_msg_debug(this->name, 0,
+                 "min_free_disk - %f , block available - %" PRId64
+                 ", block size - %lu",
+                 conf->min_free_disk, dst_statfs.f_bavail, dst_statfs.f_bsize);
+
+    dst_statfs_blocks = dst_statfs.f_bavail *
+                        (dst_statfs.f_frsize / GF_DISK_SECTOR_SIZE);
+
+    src_statfs_blocks = src_statfs.f_bavail *
+                        (src_statfs.f_frsize / GF_DISK_SECTOR_SIZE);
+
+    dst_total_blocks = dst_statfs.f_blocks *
+                       (dst_statfs.f_frsize / GF_DISK_SECTOR_SIZE);
+
+    src_total_blocks = src_statfs.f_blocks *
+                       (src_statfs.f_frsize / GF_DISK_SECTOR_SIZE);
+
+    /* if force option is given, do not check for space @ dst.
+     * Check only if space is avail for the file */
+    if (flag != GF_DHT_MIGRATE_DATA)
+        goto check_avail_space;
+
+    /* Check:
+       During rebalance `migrate-data` - Destination subvol experiences
+       a `reduction` in 'blocks' of free space, at the same time source
+       subvol gains certain 'blocks' of free space. A valid check is
+       necessary here to avoid erroneous move to destination where
+       the space could be scantily available.
+       With heterogeneous brick support, an actual space comparison could
+       prevent any files being migrated to newly added bricks if they are
+       smaller then the free space available on the existing bricks.
+     */
+    if (!conf->use_fallocate) {
+        file_blocks = stbuf->ia_size + GF_DISK_SECTOR_SIZE - 1;
+        file_blocks /= GF_DISK_SECTOR_SIZE;
+
+        if (file_blocks >= dst_statfs_blocks) {
+            dst_statfs_blocks = 0;
+        } else {
+            dst_statfs_blocks -= file_blocks;
+        }
+    }
+
+    src_post_availspacepercent = ((src_statfs_blocks + file_blocks) * 100) /
+                                 src_total_blocks;
+
+    dst_post_availspacepercent = (dst_statfs_blocks * 100) / dst_total_blocks;
+
+    if (dst_post_availspacepercent < src_post_availspacepercent) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "data movement of file "
+               "{blocks:%" PRIu64
+               " name:(%s)} would result in "
+               "dst node (%s:%" PRIu64
+               ") having lower disk "
+               "space than the source node (%s:%" PRIu64
+               ")"
+               ".Skipping file.",
+               stbuf->ia_blocks, loc->path, to->name, dst_statfs_blocks,
+               from->name, src_statfs_blocks);
+
+        /* this is not a 'failure', but we don't want to
+           consider this as 'success' too :-/ */
+        *fop_errno = ENOSPC;
+        ret = 1;
+        goto out;
+    }
+
+check_avail_space:
+    if (conf->disk_unit == 'p' && dst_statfs.f_blocks) {
+        dst_post_availspacepercent = (dst_statfs_blocks * 100) /
+                                     dst_total_blocks;
+
+        gf_msg_debug(this->name, 0,
+                     "file : %s, post_availspacepercent"
+                     " : %lf f_bavail : %" PRIu64 " min-free-disk: %lf",
+                     loc->path, dst_post_availspacepercent, dst_statfs.f_bavail,
+                     conf->min_free_disk);
+
+        if (dst_post_availspacepercent < conf->min_free_disk) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, 0,
+                   "Write will cross min-free-disk for "
+                   "file - %s on subvol - %s. Looking "
+                   "for new subvol",
+                   loc->path, to->name);
+
+            goto find_new_subvol;
+        } else {
+            ret = 0;
+            goto out;
+        }
+    }
+
+    if (conf->disk_unit != 'p') {
+        if ((dst_statfs_blocks * GF_DISK_SECTOR_SIZE) < conf->min_free_disk) {
+            gf_msg_debug(this->name, 0,
+                         "file : %s,  destination frsize: %lu "
+                         "f_bavail : %" PRIu64 " min-free-disk: %lf",
+                         loc->path, dst_statfs.f_frsize, dst_statfs.f_bavail,
+                         conf->min_free_disk);
+
+            gf_msg(this->name, GF_LOG_WARNING, 0, 0,
+                   "write will"
+                   " cross min-free-disk for file - %s on subvol -"
+                   " %s. looking for new subvol",
+                   loc->path, to->name);
+
+            goto find_new_subvol;
+
+        } else {
+            ret = 0;
+            goto out;
+        }
+    }
+
+find_new_subvol:
+    layout = dht_layout_get(this, loc->parent);
+    if (!layout) {
+        gf_log(this->name, GF_LOG_ERROR, "Layout is NULL");
+        *fop_errno = EINVAL;
+        ret = -1;
+        goto out;
+    }
+
+    *new_subvol = dht_subvol_with_free_space_inodes(this, to, from, layout,
+                                                    stbuf->ia_size);
+    if ((!(*new_subvol)) || (*new_subvol == from)) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_SUBVOL_INSUFF_SPACE,
+               "Could not find any subvol"
+               " with space accommodating the file - %s. Consider "
+               "adding bricks",
+               loc->path);
+
+        *target_changed = _gf_false;
+        *fop_errno = ENOSPC;
+        ret = -1;
+    } else {
+        gf_msg(this->name, GF_LOG_INFO, 0, 0,
+               "new target found - %s"
+               " for file - %s",
+               (*new_subvol)->name, loc->path);
+        *target_changed = _gf_true;
+        ret = 0;
+    }
+
+out:
+    if (xdata)
+        dict_unref(xdata);
+    return ret;
+}
+
+static int
+__dht_rebalance_migrate_data(xlator_t *this, gf_defrag_info_t *defrag,
+                             xlator_t *from, xlator_t *to, fd_t *src, fd_t *dst,
+                             uint64_t ia_size, int hole_exists, int *fop_errno)
+{
+    int ret = 0;
+    int count = 0;
+    off_t offset = 0;
+    off_t data_offset = 0;
+    off_t hole_offset = 0;
+    struct iovec *vector = NULL;
+    struct iobref *iobref = NULL;
+    uint64_t total = 0;
+    size_t read_size = 0;
+    size_t data_block_size = 0;
+    dict_t *xdata = NULL;
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    /* if file size is '0', no need to enter this loop */
+    while (total < ia_size) {
+        /* This is a regular file - read it sequentially */
+        if (!hole_exists) {
+            read_size = (((ia_size - total) > DHT_REBALANCE_BLKSIZE)
+                             ? DHT_REBALANCE_BLKSIZE
+                             : (ia_size - total));
+        } else {
+            /* This is a sparse file - read only the data segments in the file
+             */
+
+            /* If the previous data block is fully copied, find the next data
+             * segment
+             * starting at the offset of the last read and written byte,  */
+            if (data_block_size <= 0) {
+                ret = syncop_seek(from, src, offset, GF_SEEK_DATA, NULL,
+                                  &data_offset);
+                if (ret) {
+                    if (ret == -ENXIO)
+                        ret = 0; /* No more data segments */
+                    else
+                        *fop_errno = -ret; /* Error occurred */
+
+                    break;
+                }
+
+                /* If the position of the current data segment is greater than
+                 * the position of the next hole, find the next hole in order to
+                 * calculate the length of the new data segment */
+                if (data_offset > hole_offset) {
+                    /* Starting at the offset of the last data segment, find the
+                     * next hole */
+                    ret = syncop_seek(from, src, data_offset, GF_SEEK_HOLE,
+                                      NULL, &hole_offset);
+                    if (ret) {
+                        /* If an error occurred here it's a real error because
+                         * if the seek for a data segment was successful then
+                         * necessarily another hole must exist (EOF is a hole)
+                         */
+                        *fop_errno = -ret;
+                        break;
+                    }
+
+                    /* Calculate the total size of the current data block */
+                    data_block_size = hole_offset - data_offset;
+                }
+            } else {
+                /* There is still data in the current segment, move the
+                 * data_offset to the position of the last written byte */
+                data_offset = offset;
+            }
+
+            /* Calculate how much data needs to be read and written. If the data
+             * segment's length is bigger than DHT_REBALANCE_BLKSIZE, read and
+             * write DHT_REBALANCE_BLKSIZE data length and the rest in the
+             * next iteration(s) */
+            read_size = ((data_block_size > DHT_REBALANCE_BLKSIZE)
+                             ? DHT_REBALANCE_BLKSIZE
+                             : data_block_size);
+
+            /* Calculate the remaining size of the data block - maybe there's no
+             * need to seek for data in the next iteration */
+            data_block_size -= read_size;
+
+            /* Set offset to the offset of the data segment so read and write
+             * will have the correct position */
+            offset = data_offset;
+        }
+
+        ret = syncop_readv(from, src, read_size, offset, 0, &vector, &count,
+                           &iobref, NULL, NULL, NULL);
+
+        if (!ret || (ret < 0)) {
+            if (!ret) {
+                /* File was probably truncated*/
+                ret = -1;
+                *fop_errno = ENOSPC;
+            } else {
+                *fop_errno = -ret;
+            }
+            break;
+        }
+
+        if (!conf->force_migration) {
+            if (!xdata) {
+                xdata = dict_new();
+                if (!xdata) {
+                    gf_msg("dht", GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                           "insufficient memory");
+                    ret = -1;
+                    *fop_errno = ENOMEM;
+                    break;
+                }
+
+                /* Fail this write and abort rebalance if we
+                 * detect a write from client since migration of
+                 * this file started. This is done to avoid
+                 * potential data corruption due to out of order
+                 * writes from rebalance and client to the same
+                 * region (as compared between src and dst
+                 * files). See
+                 * https://github.com/gluster/glusterfs/issues/308
+                 * for more details.
+                 */
+                ret = dict_set_int32_sizen(xdata, GF_AVOID_OVERWRITE, 1);
+                if (ret) {
+                    gf_msg("dht", GF_LOG_ERROR, 0, ENOMEM,
+                           "failed to set dict");
+                    ret = -1;
+                    *fop_errno = ENOMEM;
+                    break;
+                }
+            }
+        }
+
+        ret = syncop_writev(to, dst, vector, count, offset, iobref, 0, NULL,
+                            NULL, xdata, NULL);
+        if (ret < 0) {
+            *fop_errno = -ret;
+            break;
+        }
+
+        offset += ret;
+        total += ret;
+
+        GF_FREE(vector);
+        if (iobref)
+            iobref_unref(iobref);
+        iobref = NULL;
+        vector = NULL;
+    }
+    if (iobref)
+        iobref_unref(iobref);
+    GF_FREE(vector);
+
+    if (ret >= 0)
+        ret = 0;
+    else
+        ret = -1;
+
+    if (xdata) {
+        dict_unref(xdata);
+    }
+
+    return ret;
+}
+
+static int
+__dht_rebalance_open_src_file(xlator_t *this, xlator_t *from, xlator_t *to,
+                              loc_t *loc, struct iatt *stbuf, fd_t **src_fd,
+                              gf_boolean_t *clean_src, int *fop_errno)
+{
+    int ret = 0;
+    fd_t *fd = NULL;
+    dict_t *dict = NULL;
+    struct iatt iatt = {
+        0,
+    };
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    *clean_src = _gf_false;
+
+    fd = fd_create(loc->inode, DHT_REBALANCE_PID);
+    if (!fd) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: fd create failed (source)", loc->path);
+        *fop_errno = ENOMEM;
+        ret = -1;
+        goto out;
+    }
+
+    ret = syncop_open(from, loc, O_RDWR, fd, NULL, NULL);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "failed to open file %s on %s", loc->path, from->name);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    fd_bind(fd);
+
+    if (src_fd)
+        *src_fd = fd;
+
+    ret = -1;
+    dict = dict_new();
+    if (!dict) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: Could not allocate memory for dict", loc->path);
+        *fop_errno = ENOMEM;
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_set_str(dict, conf->link_xattr_name, to->name);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to set xattr in dict for %s (linkto:%s)", loc->path,
+               to->name);
+        *fop_errno = ENOMEM;
+        ret = -1;
+        goto out;
+    }
+
+    /* Once the migration starts, the source should have 'linkto' key set
+       to show which is the target, so other clients can work around it */
+    ret = syncop_setxattr(from, loc, dict, 0, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "failed to set xattr on %s in %s", loc->path, from->name);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    /* Reset source mode/xattr if migration fails*/
+    *clean_src = _gf_true;
+
+    /* mode should be (+S+T) to indicate migration is in progress */
+    iatt.ia_prot = stbuf->ia_prot;
+    iatt.ia_type = stbuf->ia_type;
+    iatt.ia_prot.sticky = 1;
+    iatt.ia_prot.sgid = 1;
+
+    ret = syncop_setattr(from, loc, &iatt, GF_SET_ATTR_MODE, NULL, NULL, NULL,
+                         NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "failed to set mode on %s in %s", loc->path, from->name);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    /* success */
+    ret = 0;
+out:
+    if (dict)
+        dict_unref(dict);
+
+    return ret;
+}
+
+int
+migrate_special_files(xlator_t *this, xlator_t *from, xlator_t *to, loc_t *loc,
+                      struct iatt *buf, int *fop_errno)
+{
+    int ret = -1;
+    dict_t *rsp_dict = NULL;
+    dict_t *dict = NULL;
+    char *link = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+    dht_conf_t *conf = this->private;
+
+    dict = dict_new();
+    if (!dict) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        goto out;
+    }
+    ret = dict_set_int32(dict, conf->link_xattr_name, 256);
+    if (ret) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_log(this->name, GF_LOG_ERROR,
+               "%s: failed to set 'linkto' key in dict", loc->path);
+        goto out;
+    }
+
+    /* check in the destination if the file is link file */
+    ret = syncop_lookup(to, loc, &stbuf, NULL, dict, &rsp_dict);
+    if ((ret < 0) && (-ret != ENOENT)) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: lookup failed", loc->path);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    /* we no more require this key */
+    dict_del(dict, conf->link_xattr_name);
+
+    /* file exists in target node, only if it is 'linkfile' its valid,
+       otherwise, error out */
+    if (!ret) {
+        if (!check_is_linkfile(loc->inode, &stbuf, rsp_dict,
+                               conf->link_xattr_name)) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "%s: file exists in destination", loc->path);
+            *fop_errno = EINVAL;
+            ret = -1;
+            goto out;
+        }
+
+        /* as file is linkfile, delete it */
+        ret = syncop_unlink(to, loc, NULL, NULL);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_WARNING, -ret,
+                   DHT_MSG_MIGRATE_FILE_FAILED,
+                   "%s: failed to delete the linkfile", loc->path);
+            *fop_errno = -ret;
+            ret = -1;
+            goto out;
+        }
+    }
+
+    /* Set the gfid of the source file in dict */
+    ret = dict_set_gfuuid(dict, "gfid-req", buf->ia_gfid, true);
+    if (ret) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_log(this->name, GF_LOG_ERROR,
+               "%s: failed to set gfid in dict for create", loc->path);
+        goto out;
+    }
+
+    /* Create the file in target */
+    if (IA_ISLNK(buf->ia_type)) {
+        /* Handle symlinks separately */
+        ret = syncop_readlink(from, loc, &link, buf->ia_size, NULL, NULL);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_WARNING, -ret,
+                   DHT_MSG_MIGRATE_FILE_FAILED,
+                   "%s: readlink on symlink failed", loc->path);
+            *fop_errno = -ret;
+            ret = -1;
+            goto out;
+        }
+
+        ret = syncop_symlink(to, loc, link, 0, dict, NULL);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_WARNING, -ret,
+                   DHT_MSG_MIGRATE_FILE_FAILED, "%s: creating symlink failed",
+                   loc->path);
+            *fop_errno = -ret;
+            ret = -1;
+            goto out;
+        }
+
+        goto done;
+    }
+
+    ret = syncop_mknod(to, loc, st_mode_from_ia(buf->ia_prot, buf->ia_type),
+                       makedev(ia_major(buf->ia_rdev), ia_minor(buf->ia_rdev)),
+                       0, dict, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: mknod failed", loc->path);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+done:
+    ret = syncop_setattr(to, loc, buf,
+                         (GF_SET_ATTR_MTIME | GF_SET_ATTR_UID |
+                          GF_SET_ATTR_GID | GF_SET_ATTR_MODE),
+                         NULL, NULL, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: failed to perform setattr on %s", loc->path, to->name);
+        *fop_errno = -ret;
+    }
+
+    ret = syncop_unlink(from, loc, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: unlink failed", loc->path);
+        *fop_errno = -ret;
+        ret = -1;
+    }
+
+out:
+    GF_FREE(link);
+    if (dict)
+        dict_unref(dict);
+
+    if (rsp_dict)
+        dict_unref(rsp_dict);
+
+    return ret;
+}
+
+static int
+__dht_migration_cleanup_src_file(xlator_t *this, loc_t *loc, fd_t *fd,
+                                 xlator_t *from, ia_prot_t *src_ia_prot)
+{
+    int ret = -1;
+    dht_conf_t *conf = NULL;
+    struct iatt new_stbuf = {
+        0,
+    };
+
+    if (!this || !fd || !from || !src_ia_prot) {
+        goto out;
+    }
+
+    conf = this->private;
+
+    /*Revert source mode and xattr changes*/
+    ret = syncop_fstat(from, fd, &new_stbuf, NULL, NULL);
+    if (ret < 0) {
+        /* Failed to get the stat info */
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file cleanup failed: failed to fstat "
+               "file %s on %s ",
+               loc->path, from->name);
+        ret = -1;
+        goto out;
+    }
+
+    /* Remove the sticky bit and sgid bit set, reset it to 0*/
+    if (!src_ia_prot->sticky)
+        new_stbuf.ia_prot.sticky = 0;
+
+    if (!src_ia_prot->sgid)
+        new_stbuf.ia_prot.sgid = 0;
+
+    ret = syncop_fsetattr(from, fd, &new_stbuf,
+                          (GF_SET_ATTR_GID | GF_SET_ATTR_MODE), NULL, NULL,
+                          NULL, NULL);
+
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file cleanup failed:"
+               "%s: failed to perform fsetattr on %s ",
+               loc->path, from->name);
+        ret = -1;
+        goto out;
+    }
+
+    ret = syncop_fremovexattr(from, fd, conf->link_xattr_name, 0, NULL);
+    if (ret) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "%s: failed to remove linkto xattr on %s (%s)", loc->path,
+               from->name, strerror(-ret));
+        ret = -1;
+        goto out;
+    }
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+/*
+  return values:
+
+   -1 : failure
+    0 : successfully migrated data
+    1 : not a failure, but we can't migrate data as of now
+*/
+int
+dht_migrate_file(xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
+                 int flag, int *fop_errno)
+{
+    int ret = -1;
+    struct iatt new_stbuf = {
+        0,
+    };
+    struct iatt stbuf = {
+        0,
+    };
+    struct iatt empty_iatt = {
+        0,
+    };
+    ia_prot_t src_ia_prot = {
+        0,
+    };
+    fd_t *src_fd = NULL;
+    fd_t *dst_fd = NULL;
+    dict_t *dict = NULL;
+    dict_t *xattr = NULL;
+    dict_t *xattr_rsp = NULL;
+    int file_has_holes = 0;
+    dht_conf_t *conf = this->private;
+    int rcvd_enoent_from_src = 0;
+    struct gf_flock flock = {
+        0,
+    };
+    struct gf_flock plock = {
+        0,
+    };
+    loc_t tmp_loc = {
+        0,
+    };
+    loc_t parent_loc = {
+        0,
+    };
+    gf_boolean_t inodelk_locked = _gf_false;
+    gf_boolean_t entrylk_locked = _gf_false;
+    gf_boolean_t p_locked = _gf_false;
+    int lk_ret = -1;
+    gf_defrag_info_t *defrag = NULL;
+    gf_boolean_t clean_src = _gf_false;
+    gf_boolean_t clean_dst = _gf_false;
+    int log_level = GF_LOG_INFO;
+    gf_boolean_t delete_src_linkto = _gf_true;
+    lock_migration_info_t locklist;
+    dict_t *meta_dict = NULL;
+    gf_boolean_t meta_locked = _gf_false;
+    gf_boolean_t target_changed = _gf_false;
+    xlator_t *new_target = NULL;
+    xlator_t *old_target = NULL;
+    xlator_t *hashed_subvol = NULL;
+    fd_t *linkto_fd = NULL;
+    dict_t *xdata = NULL;
+
+    if (from == to) {
+        gf_msg_debug(this->name, 0,
+                     "destination and source are same. file %s"
+                     " might have migrated already",
+                     loc->path);
+        ret = 0;
+        goto out;
+    }
+
+    gf_log(this->name, log_level, "%s: attempting to move from %s to %s",
+           loc->path, from->name, to->name);
+
+    dict = dict_new();
+    if (!dict) {
+        ret = -1;
+        *fop_errno = ENOMEM;
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+               "Could not allocate memory for dict");
+        goto out;
+    }
+    ret = dict_set_int32(dict, conf->link_xattr_name, 256);
+    if (ret) {
+        *fop_errno = ENOMEM;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed:"
+               "%s: failed to set 'linkto' key in dict",
+               loc->path);
+        goto out;
+    }
+
+    /* Do not migrate file in case lock migration is not enabled on the
+     * volume*/
+    if (!conf->lock_migration_enabled) {
+        ret = dict_set_int32(dict, GLUSTERFS_POSIXLK_COUNT, sizeof(int32_t));
+        if (ret) {
+            *fop_errno = ENOMEM;
+            ret = -1;
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Migrate file failed: %s: failed to "
+                   "set " GLUSTERFS_POSIXLK_COUNT " key in dict",
+                   loc->path);
+            goto out;
+        }
+    } else {
+        gf_msg(this->name, GF_LOG_INFO, 0, 0,
+               "locks will be migrated"
+               " for file: %s",
+               loc->path);
+    }
+
+    /* The file is locked to prevent a rename during a migration. Renames
+     * and migrations on the file at the same time can lead to data loss.
+     */
+
+    ret = dht_build_parent_loc(this, &parent_loc, loc, fop_errno);
+    if (ret < 0) {
+        ret = -1;
+        gf_msg(this->name, GF_LOG_WARNING, *fop_errno,
+               DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: failed to build parent loc, which is needed to "
+               "acquire entrylk to synchronize with renames on this "
+               "path. Skipping migration",
+               loc->path);
+        goto out;
+    }
+
+    hashed_subvol = dht_subvol_get_hashed(this, loc);
+    if (hashed_subvol == NULL) {
+        ret = -1;
+        gf_msg(this->name, GF_LOG_WARNING, EINVAL, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: cannot find hashed subvol which is needed to "
+               "synchronize with renames on this path. "
+               "Skipping migration",
+               loc->path);
+        goto out;
+    }
+
+    flock.l_type = F_WRLCK;
+
+    tmp_loc.inode = inode_ref(loc->inode);
+    gf_uuid_copy(tmp_loc.gfid, loc->gfid);
+    tmp_loc.path = gf_strdup(loc->path);
+
+    /* this inodelk happens with flock.owner being zero. But to synchronize
+     * hardlink migration we need to have different lkowner for each migration
+     * Filed a bug here: https://bugzilla.redhat.com/show_bug.cgi?id=1468202 to
+     * track the fix for this. Currently synclock takes care of synchronizing
+     * hardlink migration. Once this bug is fixed we can avoid taking synclock
+     */
+    ret = syncop_inodelk(from, DHT_FILE_MIGRATE_DOMAIN, &tmp_loc, F_SETLKW,
+                         &flock, NULL, NULL);
+    if (ret < 0) {
+        *fop_errno = -ret;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_WARNING, *fop_errno,
+               DHT_MSG_MIGRATE_FILE_FAILED,
+               "migrate file failed: "
+               "%s: failed to lock file on %s",
+               loc->path, from->name);
+        goto out;
+    }
+
+    inodelk_locked = _gf_true;
+
+    /* dht_rename has changed to use entrylk on hashed subvol for
+     * synchronization. So, rebalance too has to acquire an entrylk on
+     * hashed subvol.
+     */
+    ret = syncop_entrylk(hashed_subvol, DHT_ENTRY_SYNC_DOMAIN, &parent_loc,
+                         loc->name, ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL, NULL);
+    if (ret < 0) {
+        *fop_errno = -ret;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_WARNING, *fop_errno,
+               DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: failed to acquire entrylk on subvol %s", loc->path,
+               hashed_subvol->name);
+        goto out;
+    }
+
+    entrylk_locked = _gf_true;
+
+    /* Phase 1 - Data migration is in progress from now on */
+    ret = syncop_lookup(from, loc, &stbuf, NULL, dict, &xattr_rsp);
+    if (ret) {
+        *fop_errno = -ret;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, *fop_errno,
+               DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed:"
+               "%s: lookup failed on %s",
+               loc->path, from->name);
+        goto out;
+    }
+
+    /* preserve source mode, so set the same to the destination */
+    src_ia_prot = stbuf.ia_prot;
+
+    /* Check if file can be migrated */
+    ret = __is_file_migratable(this, loc, &stbuf, xattr_rsp, flag, defrag, conf,
+                               fop_errno);
+    if (ret) {
+        if (ret == HARDLINK_MIG_INPROGRESS)
+            ret = 0;
+        goto out;
+    }
+
+    /* Take care of the special files */
+    if (!IA_ISREG(stbuf.ia_type)) {
+        /* Special files */
+        ret = migrate_special_files(this, from, to, loc, &stbuf, fop_errno);
+        goto out;
+    }
+
+    /* Try to preserve 'holes' while migrating data */
+    if (stbuf.ia_size > (stbuf.ia_blocks * GF_DISK_SECTOR_SIZE))
+        file_has_holes = 1;
+
+    /* create the destination, with required modes/xattr */
+    ret = __dht_rebalance_create_dst_file(this, to, from, loc, &stbuf, &dst_fd,
+                                          fop_errno, file_has_holes);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "Create dst failed"
+               " on - %s for file - %s",
+               to->name, loc->path);
+        goto out;
+    }
+
+    clean_dst = _gf_true;
+
+    ret = __dht_check_free_space(this, to, from, loc, &stbuf, flag, conf,
+                                 &target_changed, &new_target, fop_errno);
+    if (target_changed) {
+        /* Can't handle for hardlinks. Marking this as failure */
+        if (flag == GF_DHT_MIGRATE_HARDLINK_IN_PROGRESS || stbuf.ia_nlink > 1) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SUBVOL_INSUFF_SPACE,
+                   "Exiting migration for"
+                   " file - %s. flag - %d, stbuf.ia_nlink - %d",
+                   loc->path, flag, stbuf.ia_nlink);
+            ret = -1;
+            goto out;
+        }
+
+        ret = syncop_ftruncate(to, dst_fd, 0, NULL, NULL, NULL, NULL);
+        if (ret) {
+            gf_log(this->name, GF_LOG_WARNING,
+                   "%s: failed to perform truncate on %s (%s)", loc->path,
+                   to->name, strerror(-ret));
+        }
+
+        syncop_close(dst_fd);
+        dst_fd = NULL;
+
+        old_target = to;
+        to = new_target;
+
+        clean_dst = _gf_false;
+
+        /* if the file migration is successful to this new target, then
+         * update the xattr on the old destination to point the new
+         * destination. We need to do update this only post migration
+         * as in case of failure the linkto needs to point to the source
+         * subvol */
+        ret = __dht_rebalance_create_dst_file(
+            this, to, from, loc, &stbuf, &dst_fd, fop_errno, file_has_holes);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "Create dst failed"
+                   " on - %s for file - %s",
+                   to->name, loc->path);
+            goto out;
+        } else {
+            gf_msg(this->name, GF_LOG_INFO, 0, 0,
+                   "destination for file "
+                   "- %s is changed to - %s",
+                   loc->path, to->name);
+            clean_dst = _gf_true;
+        }
+    }
+
+    if (ret) {
+        goto out;
+    }
+
+    /* Open the source, and also update mode/xattr */
+    ret = __dht_rebalance_open_src_file(this, from, to, loc, &stbuf, &src_fd,
+                                        &clean_src, fop_errno);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed: failed to open %s on %s", loc->path,
+               from->name);
+        goto out;
+    }
+
+    /* TODO: move all xattr related operations to fd based operations */
+    ret = syncop_listxattr(from, loc, &xattr, NULL, NULL);
+    if (ret < 0) {
+        *fop_errno = -ret;
+        gf_msg(this->name, GF_LOG_WARNING, *fop_errno,
+               DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed:"
+               "%s: failed to get xattr from %s",
+               loc->path, from->name);
+        ret = -1;
+        goto out;
+    }
+
+    /* Copying posix acls to the linkto file messes up the permissions*/
+    dht_strip_out_acls(xattr);
+
+    /* Remove the linkto xattr as we don't want to overwrite the value
+     * set on the dst.
+     */
+    dict_del(xattr, conf->link_xattr_name);
+
+    /* We need to error out if this fails as having the wrong shard xattrs
+     * set on the dst could cause data corruption
+     */
+    ret = syncop_fsetxattr(to, dst_fd, xattr, 0, NULL, NULL);
+    if (ret < 0) {
+        *fop_errno = -ret;
+        gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: failed to set xattr on %s", loc->path, to->name);
+        ret = -1;
+        goto out;
+    }
+
+    if (xattr_rsp) {
+        /* we no more require this key */
+        dict_del(dict, conf->link_xattr_name);
+        dict_unref(xattr_rsp);
+    }
+
+    ret = syncop_fstat(from, src_fd, &stbuf, dict, &xattr_rsp);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed:failed to lookup %s on %s ", loc->path,
+               from->name);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    /* Check again if file has hardlink */
+    ret = __check_file_has_hardlink(this, loc, &stbuf, xattr_rsp, flag, defrag,
+                                    conf, fop_errno);
+    if (ret) {
+        if (ret == HARDLINK_MIG_INPROGRESS)
+            ret = 0;
+        goto out;
+    }
+
+    ret = __dht_rebalance_migrate_data(this, defrag, from, to, src_fd, dst_fd,
+                                       stbuf.ia_size, file_has_holes,
+                                       fop_errno);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed: %s: failed to migrate data", loc->path);
+
+        ret = -1;
+        goto out;
+    }
+
+    /* TODO: Sync the locks */
+
+    xdata = dict_new();
+    if (!xdata || dict_set_int8(xdata, "last-fsync", 1)) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "%s: failed to set last-fsync flag on "
+               "%s (%s)",
+               loc->path, to->name, strerror(ENOMEM));
+    }
+
+    ret = syncop_fsync(to, dst_fd, 0, NULL, NULL, xdata, NULL);
+    if (ret) {
+        gf_log(this->name, GF_LOG_WARNING, "%s: failed to fsync on %s (%s)",
+               loc->path, to->name, strerror(-ret));
+        *fop_errno = -ret;
+    }
+
+    /* Phase 2 - Data-Migration Complete, Housekeeping updates pending */
+
+    ret = syncop_fstat(from, src_fd, &new_stbuf, NULL, NULL);
+    if (ret < 0) {
+        /* Failed to get the stat info */
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed: failed to fstat file %s on %s ", loc->path,
+               from->name);
+        *fop_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    /* Lock the entire source file to prevent clients from taking a
+       lock on it as dht_lk does not handle file migration.
+
+       This still leaves a small window where conflicting locks can
+       be granted to different clients. If client1 requests a blocking
+       lock on the src file, it will be granted after the migrating
+       process releases its lock. If client2 requests a lock on the dst
+       data file, it will also be granted, but all FOPs will be redirected
+       to the dst data file.
+    */
+
+    /* Take meta lock  */
+
+    if (conf->lock_migration_enabled) {
+        meta_dict = dict_new();
+        if (!meta_dict) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "dict_new failed");
+
+            *fop_errno = ENOMEM;
+            ret = -1;
+            goto out;
+        }
+
+        ret = dict_set_str(meta_dict, GLUSTERFS_INTERNAL_FOP_KEY, "yes");
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                   "Failed to set dictionary value: key = %s,"
+                   " path = %s",
+                   GLUSTERFS_INTERNAL_FOP_KEY, loc->path);
+            *fop_errno = ENOMEM;
+            ret = -1;
+            goto out;
+        }
+
+        ret = dict_set_int32(meta_dict, GF_META_LOCK_KEY, 1);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Trace dict_set failed");
+            *fop_errno = ENOMEM;
+            ret = -1;
+            goto out;
+        }
+
+        ret = syncop_setxattr(from, loc, meta_dict, 0, NULL, NULL);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Trace syncop_setxattr metalock failed");
+
+            *fop_errno = -ret;
+            ret = -1;
+            goto out;
+        } else {
+            meta_locked = _gf_true;
+        }
+    }
+
+    if (!conf->lock_migration_enabled) {
+        plock.l_type = F_WRLCK;
+        plock.l_start = 0;
+        plock.l_len = 0;
+        plock.l_whence = SEEK_SET;
+
+        ret = syncop_lk(from, src_fd, F_SETLK, &plock, NULL, NULL);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Migrate file failed:"
+                   "%s: Failed to lock on %s",
+                   loc->path, from->name);
+            *fop_errno = -ret;
+            ret = -1;
+            goto out;
+        }
+
+        p_locked = _gf_true;
+
+    } else {
+        INIT_LIST_HEAD(&locklist.list);
+
+        ret = syncop_getactivelk(from, loc, &locklist, NULL, NULL);
+        if (ret == 0) {
+            gf_log(this->name, GF_LOG_INFO, "No active locks on:%s", loc->path);
+
+        } else if (ret > 0) {
+            ret = syncop_setactivelk(to, loc, &locklist, NULL, NULL);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, -ret,
+                       DHT_MSG_LOCK_MIGRATION_FAILED, "write lock failed on:%s",
+                       loc->path);
+
+                *fop_errno = -ret;
+                ret = -1;
+                goto metaunlock;
+            }
+        } else {
+            gf_msg(this->name, GF_LOG_ERROR, -ret,
+                   DHT_MSG_LOCK_MIGRATION_FAILED,
+                   "getactivelk failed for file: %s", loc->path);
+            *fop_errno = -ret;
+        }
+    }
+
+    /* source would have both sticky bit and sgid bit set, reset it to 0,
+       and set the source permission on destination, if it was not set
+       prior to setting rebalance-modes in source  */
+    if (!src_ia_prot.sticky)
+        new_stbuf.ia_prot.sticky = 0;
+
+    if (!src_ia_prot.sgid)
+        new_stbuf.ia_prot.sgid = 0;
+
+    /* TODO: if the source actually had sticky bit, or sgid bit set,
+       we are not handling it */
+
+    ret = syncop_fsetattr(
+        to, dst_fd, &new_stbuf,
+        (GF_SET_ATTR_UID | GF_SET_ATTR_GID | GF_SET_ATTR_MODE), NULL, NULL,
+        NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed:"
+               "%s: failed to perform setattr on %s ",
+               loc->path, to->name);
+        *fop_errno = -ret;
+        ret = -1;
+        goto metaunlock;
+    }
+
+    /* Because 'futimes' is not portable */
+    ret = syncop_setattr(to, loc, &new_stbuf,
+                         (GF_SET_ATTR_MTIME | GF_SET_ATTR_ATIME), NULL, NULL,
+                         NULL, NULL);
+    if (ret) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "%s: failed to perform setattr on %s ", loc->path, to->name);
+        *fop_errno = -ret;
+    }
+
+    if (target_changed) {
+        dict_del(dict, GLUSTERFS_POSIXLK_COUNT);
+        ret = dict_set_str(dict, conf->link_xattr_name, to->name);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "failed to set xattr in dict for %s (linkto:%s)", loc->path,
+                   to->name);
+            *fop_errno = ENOMEM;
+            ret = -1;
+            goto out;
+        }
+
+        ret = syncop_setxattr(old_target, loc, dict, 0, NULL, NULL);
+        if (ret && -ret != ESTALE && -ret != ENOENT) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "failed to set xattr on %s in %s", loc->path,
+                   old_target->name);
+            *fop_errno = -ret;
+            ret = -1;
+            goto out;
+        } else if (-ret == ESTALE || -ret == ENOENT) {
+            /* The failure ESTALE indicates that the linkto
+             * file on the hashed subvol might have been deleted.
+             * In this case will create a linkto file with new target
+             * as linkto xattr value*/
+            linkto_fd = fd_create(loc->inode, DHT_REBALANCE_PID);
+            if (!linkto_fd) {
+                gf_msg(this->name, GF_LOG_ERROR, errno,
+                       DHT_MSG_MIGRATE_FILE_FAILED, "%s: fd create failed",
+                       loc->path);
+                *fop_errno = ENOMEM;
+                ret = -1;
+                goto out;
+            }
+            ret = syncop_create(old_target, loc, O_RDWR, DHT_LINKFILE_MODE,
+                                linkto_fd, NULL, dict, NULL);
+            if (ret != 0 && -ret != EEXIST && -ret != ESTALE) {
+                *fop_errno = -ret;
+                ret = -1;
+                gf_msg(this->name, GF_LOG_ERROR, -ret,
+                       DHT_MSG_MIGRATE_FILE_FAILED,
+                       "failed to create linkto file on %s in %s", loc->path,
+                       old_target->name);
+                goto out;
+            } else if (ret == 0) {
+                ret = syncop_fsetattr(old_target, linkto_fd, &stbuf,
+                                      (GF_SET_ATTR_UID | GF_SET_ATTR_GID), NULL,
+                                      NULL, NULL, NULL);
+                if (ret < 0) {
+                    *fop_errno = -ret;
+                    gf_msg(this->name, GF_LOG_ERROR, -ret,
+                           DHT_MSG_MIGRATE_FILE_FAILED,
+                           "chown failed for %s on %s", loc->path,
+                           old_target->name);
+                }
+            }
+        }
+    }
+
+    clean_dst = _gf_false;
+
+    /* Posix acls are not set on DHT linkto files as part of the initial
+     * initial xattrs set on the dst file, so these need
+     * to be set on the dst file after the linkto attrs are removed.
+     * TODO: Optimize this.
+     */
+    if (xattr) {
+        dict_unref(xattr);
+        xattr = NULL;
+    }
+
+    /* Set only the Posix ACLs this time */
+    ret = syncop_getxattr(from, loc, &xattr, POSIX_ACL_ACCESS_XATTR, NULL,
+                          NULL);
+    if (ret < 0) {
+        if ((-ret != ENODATA) && (-ret != ENOATTR)) {
+            gf_msg(this->name, GF_LOG_WARNING, -ret,
+                   DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Migrate file failed:"
+                   "%s: failed to get xattr from %s",
+                   loc->path, from->name);
+            *fop_errno = -ret;
+        }
+    } else {
+        ret = syncop_setxattr(to, loc, xattr, 0, NULL, NULL);
+        if (ret < 0) {
+            /* Potential problem here where Posix ACLs will
+             * not be set on the target file */
+
+            gf_msg(this->name, GF_LOG_WARNING, -ret,
+                   DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Migrate file failed:"
+                   "%s: failed to set xattr on %s",
+                   loc->path, to->name);
+            *fop_errno = -ret;
+        }
+    }
+
+    /* The src file is being unlinked after this so we don't need
+       to clean it up */
+    clean_src = _gf_false;
+
+    /* Make the source as a linkfile first before deleting it */
+    empty_iatt.ia_prot.sticky = 1;
+    ret = syncop_fsetattr(from, src_fd, &empty_iatt, GF_SET_ATTR_MODE, NULL,
+                          NULL, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed:"
+               "%s: failed to perform setattr on %s ",
+               loc->path, from->name);
+        *fop_errno = -ret;
+        ret = -1;
+        goto metaunlock;
+    }
+
+    /* Free up the data blocks on the source node, as the whole
+        file is migrated */
+    ret = syncop_ftruncate(from, src_fd, 0, NULL, NULL, NULL, NULL);
+    if (ret) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "%s: failed to perform truncate on %s (%s)", loc->path,
+               from->name, strerror(-ret));
+        *fop_errno = -ret;
+    }
+
+    /* remove the 'linkto' xattr from the destination */
+    ret = syncop_fremovexattr(to, dst_fd, conf->link_xattr_name, 0, NULL);
+    if (ret) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "%s: failed to perform removexattr on %s (%s)", loc->path,
+               to->name, strerror(-ret));
+        *fop_errno = -ret;
+    }
+
+    /* Do a stat and check the gfid before unlink */
+
+    /*
+     * Cached file changes its state from non-linkto to linkto file after
+     * migrating data. If lookup from any other mount-point is performed,
+     * converted-linkto-cached file will be treated as a stale and will be
+     * unlinked. But by this time, file is already migrated. So further
+     * failure because of ENOENT should  not be treated as error
+     */
+
+    ret = syncop_stat(from, loc, &empty_iatt, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "%s: failed to do a stat on %s", loc->path, from->name);
+
+        if (-ret != ENOENT) {
+            *fop_errno = -ret;
+            ret = -1;
+            goto metaunlock;
+        }
+
+        rcvd_enoent_from_src = 1;
+    }
+
+    if ((gf_uuid_compare(empty_iatt.ia_gfid, loc->gfid) == 0) &&
+        (!rcvd_enoent_from_src) && delete_src_linkto) {
+        /* take out the source from namespace */
+        ret = syncop_unlink(from, loc, NULL, NULL);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_WARNING, -ret,
+                   DHT_MSG_MIGRATE_FILE_FAILED,
+                   "%s: failed to perform unlink on %s", loc->path, from->name);
+            *fop_errno = -ret;
+            ret = -1;
+            goto metaunlock;
+        }
+    }
+
+    ret = syncop_lookup(this, loc, NULL, NULL, NULL, NULL);
+    if (ret) {
+        gf_msg_debug(this->name, -ret,
+                     "%s: failed to lookup the file on subvolumes", loc->path);
+        *fop_errno = -ret;
+    }
+
+    gf_msg(this->name, log_level, 0, DHT_MSG_MIGRATE_FILE_COMPLETE,
+           "completed migration of %s from subvolume %s to %s", loc->path,
+           from->name, to->name);
+
+    ret = 0;
+
+metaunlock:
+
+    if (conf->lock_migration_enabled && meta_locked) {
+        dict_del(meta_dict, GF_META_LOCK_KEY);
+
+        ret = dict_set_int32(meta_dict, GF_META_UNLOCK_KEY, 1);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Trace dict_set failed");
+
+            *fop_errno = ENOMEM;
+            ret = -1;
+            goto out;
+        }
+
+        if (clean_dst == _gf_false)
+            ret = dict_set_int32(meta_dict, "status", 1);
+        else
+            ret = dict_set_int32(meta_dict, "status", 0);
+
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Trace dict_set failed");
+
+            *fop_errno = ENOMEM;
+            ret = -1;
+            goto out;
+        }
+
+        ret = syncop_setxattr(from, loc, meta_dict, 0, NULL, NULL);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Trace syncop_setxattr meta unlock failed");
+
+            *fop_errno = -ret;
+            ret = -1;
+            goto out;
+        }
+    }
+
+out:
+    if (clean_src) {
+        /* Revert source mode and xattr changes*/
+        lk_ret = __dht_migration_cleanup_src_file(this, loc, src_fd, from,
+                                                  &src_ia_prot);
+        if (lk_ret) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+                   "%s: failed to cleanup source file on %s", loc->path,
+                   from->name);
+        }
+    }
+
+    /* reset the destination back to 0 */
+    if (clean_dst) {
+        lk_ret = syncop_ftruncate(to, dst_fd, 0, NULL, NULL, NULL, NULL);
+        if (lk_ret) {
+            gf_msg(this->name, GF_LOG_ERROR, -lk_ret,
+                   DHT_MSG_MIGRATE_FILE_FAILED,
+                   "Migrate file failed: "
+                   "%s: failed to reset target size back to 0",
+                   loc->path);
+        }
+    }
+
+    if (inodelk_locked) {
+        flock.l_type = F_UNLCK;
+
+        lk_ret = syncop_inodelk(from, DHT_FILE_MIGRATE_DOMAIN, &tmp_loc,
+                                F_SETLK, &flock, NULL, NULL);
+        if (lk_ret < 0) {
+            gf_msg(this->name, GF_LOG_WARNING, -lk_ret,
+                   DHT_MSG_MIGRATE_FILE_FAILED,
+                   "%s: failed to unlock file on %s", loc->path, from->name);
+        }
+    }
+
+    if (entrylk_locked) {
+        lk_ret = syncop_entrylk(hashed_subvol, DHT_ENTRY_SYNC_DOMAIN,
+                                &parent_loc, loc->name, ENTRYLK_UNLOCK,
+                                ENTRYLK_UNLOCK, NULL, NULL);
+        if (lk_ret < 0) {
+            gf_msg(this->name, GF_LOG_WARNING, -lk_ret,
+                   DHT_MSG_MIGRATE_FILE_FAILED,
+                   "%s: failed to unlock entrylk on %s", loc->path,
+                   hashed_subvol->name);
+        }
+    }
+
+    if (p_locked) {
+        plock.l_type = F_UNLCK;
+        lk_ret = syncop_lk(from, src_fd, F_SETLK, &plock, NULL, NULL);
+
+        if (lk_ret < 0) {
+            gf_msg(this->name, GF_LOG_WARNING, -lk_ret,
+                   DHT_MSG_MIGRATE_FILE_FAILED,
+                   "%s: failed to unlock file on %s", loc->path, from->name);
+        }
+    }
+
+    lk_ret = syncop_removexattr(to, loc, GF_PROTECT_FROM_EXTERNAL_WRITES, NULL,
+                                NULL);
+    if (lk_ret && (lk_ret != -ENODATA) && (lk_ret != -ENOATTR)) {
+        gf_msg(this->name, GF_LOG_WARNING, -lk_ret, 0,
+               "%s: removexattr failed key %s", loc->path,
+               GF_PROTECT_FROM_EXTERNAL_WRITES);
+    }
+
+    if (dict)
+        dict_unref(dict);
+
+    if (xattr)
+        dict_unref(xattr);
+    if (xattr_rsp)
+        dict_unref(xattr_rsp);
+
+    if (dst_fd)
+        syncop_close(dst_fd);
+
+    if (src_fd)
+        syncop_close(src_fd);
+    if (linkto_fd)
+        syncop_close(linkto_fd);
+
+    if (xdata)
+        dict_unref(xdata);
+
+    loc_wipe(&tmp_loc);
+    loc_wipe(&parent_loc);
+
+    return ret;
+}
+
+static int
+rebalance_task(void *data)
+{
+    int ret = -1;
+    dht_local_t *local = NULL;
+    call_frame_t *frame = NULL;
+    int fop_errno = 0;
+
+    frame = data;
+
+    local = frame->local;
+
+    /* This function is 'synchrounous', hence if it returns,
+       we are done with the task */
+    ret = dht_migrate_file(THIS, &local->loc, local->rebalance.from_subvol,
+                           local->rebalance.target_node, local->flags,
+                           &fop_errno);
+
+    return ret;
+}
+
+static int
+rebalance_task_completion(int op_ret, call_frame_t *sync_frame, void *data)
+{
+    int32_t op_errno = EINVAL;
+
+    if (op_ret == -1) {
+        /* Failure of migration process, mostly due to write process.
+           as we can't preserve the exact errno, lets say there was
+           no space to migrate-data
+        */
+        op_errno = ENOSPC;
+    } else if (op_ret == 1) {
+        /* migration didn't happen, but is not a failure, let the user
+           understand that he doesn't have permission to migrate the
+           file.
+        */
+        op_ret = -1;
+        op_errno = EPERM;
+    } else if (op_ret != 0) {
+        op_errno = -op_ret;
+        op_ret = -1;
+    }
+
+    DHT_STACK_UNWIND(setxattr, sync_frame, op_ret, op_errno, NULL);
+    return 0;
+}
+
+int
+dht_start_rebalance_task(xlator_t *this, call_frame_t *frame)
+{
+    int ret = -1;
+
+    ret = synctask_new(this->ctx->env, rebalance_task,
+                       rebalance_task_completion, frame, frame);
+    return ret;
+}
+
+int
+gf_listener_stop(xlator_t *this)
+{
+    glusterfs_ctx_t *ctx = NULL;
+    cmd_args_t *cmd_args = NULL;
+    int ret = 0;
+
+    ctx = this->ctx;
+    GF_ASSERT(ctx);
+    cmd_args = &ctx->cmd_args;
+    if (cmd_args->sock_file) {
+        ret = sys_unlink(cmd_args->sock_file);
+        if (ret && (ENOENT == errno)) {
+            ret = 0;
+        }
+    }
+
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, DHT_MSG_SOCKET_ERROR,
+               "Failed to unlink listener "
+               "socket %s",
+               cmd_args->sock_file);
+    }
+    return ret;
+}
+
+void
+dht_build_root_inode(xlator_t *this, inode_t **inode)
+{
+    inode_table_t *itable = NULL;
+    static uuid_t root_gfid = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1};
+
+    itable = inode_table_new(0, this);
+    if (!itable)
+        return;
+
+    *inode = inode_find(itable, root_gfid);
+}
+
+void
+dht_build_root_loc(inode_t *inode, loc_t *loc)
+{
+    loc->path = "/";
+    loc->inode = inode;
+    loc->inode->ia_type = IA_IFDIR;
+    memset(loc->gfid, 0, 16);
+    loc->gfid[15] = 1;
+}
+
+/* return values: 1 -> error, bug ignore and continue
+                  0 -> proceed
+                 -1 -> error, handle it */
+int32_t
+gf_defrag_handle_migrate_error(int32_t op_errno, gf_defrag_info_t *defrag)
+{
+    int ret = 0;
+    /* if errno is not ENOTCONN, we can still continue
+       with rebalance process */
+    if (op_errno != ENOTCONN) {
+        ret = 1;
+        goto out;
+    }
+
+    if (op_errno == ENOTCONN) {
+        /* Most probably mount point went missing (mostly due
+           to a brick down), say rebalance failure to user,
+           let him restart it if everything is fine */
+        defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
+        ret = -1;
+        goto out;
+    }
+
+out:
+    return ret;
+}
+
+static gf_boolean_t
+gf_defrag_pattern_match(gf_defrag_info_t *defrag, char *name, uint64_t size)
+{
+    gf_defrag_pattern_list_t *trav = NULL;
+    gf_boolean_t match = _gf_false;
+    gf_boolean_t ret = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("dht", defrag, out);
+
+    trav = defrag->defrag_pattern;
+    while (trav) {
+        if (!fnmatch(trav->path_pattern, name, FNM_NOESCAPE)) {
+            match = _gf_true;
+            break;
+        }
+        trav = trav->next;
+    }
+
+    if ((match == _gf_true) && (size >= trav->size))
+        ret = _gf_true;
+
+out:
+    return ret;
+}
+
+int
+dht_dfreaddirp_done(dht_dfoffset_ctx_t *offset_var, int cnt)
+{
+    int i;
+    int result = 1;
+
+    for (i = 0; i < cnt; i++) {
+        if (offset_var[i].readdir_done == 0) {
+            result = 0;
+            break;
+        }
+    }
+    return result;
+}
+
+int static gf_defrag_ctx_subvols_init(dht_dfoffset_ctx_t *offset_var,
+                                      xlator_t *this)
+{
+    int i;
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf)
+        return -1;
+
+    for (i = 0; i < conf->local_subvols_cnt; i++) {
+        offset_var[i].this = conf->local_subvols[i];
+        offset_var[i].offset = (off_t)0;
+        offset_var[i].readdir_done = 0;
+    }
+
+    return 0;
+}
+
+static int
+dht_get_first_non_null_index(subvol_nodeuuids_info_t *entry)
+{
+    int i = 0;
+    int index = 0;
+
+    for (i = 0; i < entry->count; i++) {
+        if (!gf_uuid_is_null(entry->elements[i].uuid)) {
+            index = i;
+            goto out;
+        }
+    }
+
+    if (i == entry->count) {
+        index = -1;
+    }
+out:
+    return index;
+}
+
+/* Return value
+ * 0 : this node does not migrate the file
+ * 1 : this node migrates the file
+ *
+ * Use the hash value of the gfid to determine which node will migrate files.
+ * Using the gfid instead of the name also ensures that the same node handles
+ * all hardlinks.
+ */
+
+gf_boolean_t
+gf_defrag_should_i_migrate(xlator_t *this, int local_subvol_index, uuid_t gfid)
+{
+    gf_boolean_t ret = _gf_false;
+    int i = local_subvol_index;
+    char *str = NULL;
+    uint32_t hashval = 0;
+    int32_t index = 0;
+    dht_conf_t *conf = NULL;
+    char buf[UUID_CANONICAL_FORM_LEN + 1] = {
+        0,
+    };
+    subvol_nodeuuids_info_t *entry = NULL;
+
+    conf = this->private;
+
+    /* Pure distribute. A subvol in this case
+        will be handled by only one node */
+
+    entry = &(conf->local_nodeuuids[i]);
+    if (entry->count == 1) {
+        return 1;
+    }
+
+    str = uuid_utoa_r(gfid, buf);
+    if (dht_hash_compute(this, 0, str, &hashval) == 0) {
+        index = (hashval % entry->count);
+        if (entry->elements[index].info == REBAL_NODEUUID_MINE) {
+            /* Index matches this node's nodeuuid.*/
+            ret = _gf_true;
+            goto out;
+        }
+
+        /* Brick down - some other node has to migrate these files*/
+        if (gf_uuid_is_null(entry->elements[index].uuid)) {
+            /* Fall back to the first non-null index */
+            index = dht_get_first_non_null_index(entry);
+
+            if (index == -1) {
+                /* None of the bricks in the subvol are up.
+                 * CHILD_DOWN will kill the process soon */
+
+                return _gf_false;
+            }
+
+            if (entry->elements[index].info == REBAL_NODEUUID_MINE) {
+                /* Index matches this node's nodeuuid.*/
+                ret = _gf_true;
+                goto out;
+            }
+        }
+    }
+out:
+    return ret;
+}
+
+int
+gf_defrag_migrate_single_file(void *opaque)
+{
+    xlator_t *this = NULL;
+    dht_conf_t *conf = NULL;
+    gf_defrag_info_t *defrag = NULL;
+    int ret = 0;
+    gf_dirent_t *entry = NULL;
+    struct timeval start = {
+        0,
+    };
+    loc_t entry_loc = {
+        0,
+    };
+    loc_t *loc = NULL;
+    struct iatt iatt = {
+        0,
+    };
+    dict_t *migrate_data = NULL;
+    struct timeval end = {
+        0,
+    };
+    double elapsed = {
+        0,
+    };
+    struct dht_container *rebal_entry = NULL;
+    inode_t *inode = NULL;
+    xlator_t *hashed_subvol = NULL;
+    xlator_t *cached_subvol = NULL;
+    call_frame_t *statfs_frame = NULL;
+    xlator_t *old_THIS = NULL;
+    data_t *tmp = NULL;
+    int fop_errno = 0;
+    gf_dht_migrate_data_type_t rebal_type = GF_DHT_MIGRATE_DATA;
+    char value[MAX_REBAL_TYPE_SIZE] = {
+        0,
+    };
+    struct iatt *iatt_ptr = NULL;
+    gf_boolean_t update_skippedcount = _gf_true;
+    int i = 0;
+    gf_boolean_t should_i_migrate = 0;
+
+    rebal_entry = (struct dht_container *)opaque;
+    if (!rebal_entry) {
+        gf_log("DHT", GF_LOG_ERROR, "rebal_entry is NULL");
+        ret = -1;
+        goto out;
+    }
+
+    this = rebal_entry->this;
+
+    conf = this->private;
+
+    defrag = conf->defrag;
+
+    loc = rebal_entry->parent_loc;
+
+    migrate_data = rebal_entry->migrate_data;
+
+    entry = rebal_entry->df_entry;
+    iatt_ptr = &entry->d_stat;
+
+    if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) {
+        ret = -1;
+        goto out;
+    }
+
+    if (defrag->stats == _gf_true) {
+        gettimeofday(&start, NULL);
+    }
+
+    if (defrag->defrag_pattern &&
+        (gf_defrag_pattern_match(defrag, entry->d_name,
+                                 entry->d_stat.ia_size) == _gf_false)) {
+        gf_log(this->name, GF_LOG_ERROR, "pattern_match failed");
+        goto out;
+    }
+
+    memset(&entry_loc, 0, sizeof(entry_loc));
+
+    ret = dht_build_child_loc(this, &entry_loc, loc, entry->d_name);
+    if (ret) {
+        LOCK(&defrag->lock);
+        {
+            defrag->total_failures += 1;
+        }
+        UNLOCK(&defrag->lock);
+
+        ret = 0;
+
+        gf_log(this->name, GF_LOG_ERROR, "Child loc build failed");
+
+        goto out;
+    }
+
+    should_i_migrate = gf_defrag_should_i_migrate(
+        this, rebal_entry->local_subvol_index, entry->d_stat.ia_gfid);
+
+    gf_uuid_copy(entry_loc.gfid, entry->d_stat.ia_gfid);
+
+    gf_uuid_copy(entry_loc.pargfid, loc->gfid);
+
+    ret = syncop_lookup(this, &entry_loc, &iatt, NULL, NULL, NULL);
+
+    if (!should_i_migrate) {
+        /* this node isn't supposed to migrate the file. suppressing any
+         * potential error from lookup as this file is under migration by
+         * another node */
+        if (ret) {
+            gf_msg_debug(this->name, -ret,
+                         "Ignoring lookup failure: node isn't migrating %s",
+                         entry_loc.path);
+            ret = 0;
+        }
+        gf_msg_debug(this->name, 0, "Don't migrate %s ", entry_loc.path);
+        goto out;
+    }
+
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
+               "Migrate file failed: %s lookup failed", entry_loc.path);
+
+        /* Increase failure count only for remove-brick op, so that
+         * user is warned to check the removed-brick for any files left
+         * unmigrated
+         */
+        if (conf->decommission_subvols_cnt) {
+            LOCK(&defrag->lock);
+            {
+                defrag->total_failures += 1;
+            }
+            UNLOCK(&defrag->lock);
+        }
+
+        ret = 0;
+        goto out;
+    }
+
+    iatt_ptr = &iatt;
+
+    hashed_subvol = dht_subvol_get_hashed(this, &entry_loc);
+    if (!hashed_subvol) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_HASHED_SUBVOL_GET_FAILED,
+               "Failed to get hashed subvol for %s", entry_loc.path);
+        ret = 0;
+        goto out;
+    }
+
+    cached_subvol = dht_subvol_get_cached(this, entry_loc.inode);
+    if (!cached_subvol) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_CACHED_SUBVOL_GET_FAILED,
+               "Failed to get cached subvol for %s", entry_loc.path);
+
+        ret = 0;
+        goto out;
+    }
+
+    if (hashed_subvol == cached_subvol) {
+        ret = 0;
+        goto out;
+    }
+
+    inode = inode_link(entry_loc.inode, entry_loc.parent, entry->d_name, &iatt);
+    inode_unref(entry_loc.inode);
+    /* use the inode returned by inode_link */
+    entry_loc.inode = inode;
+
+    old_THIS = THIS;
+    THIS = this;
+    statfs_frame = create_frame(this, this->ctx->pool);
+    if (!statfs_frame) {
+        gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_NO_MEMORY, ENOMEM,
+               "Insufficient memory. Frame creation failed");
+        ret = -1;
+        goto out;
+    }
+
+    /* async statfs information for honoring min-free-disk */
+    dht_get_du_info(statfs_frame, this, loc);
+    THIS = old_THIS;
+
+    tmp = dict_get(migrate_data, GF_XATTR_FILE_MIGRATE_KEY);
+    if (tmp) {
+        memcpy(value, tmp->data, tmp->len);
+        if (strcmp(value, "force") == 0)
+            rebal_type = GF_DHT_MIGRATE_DATA_EVEN_IF_LINK_EXISTS;
+
+        if (conf->decommission_in_progress)
+            rebal_type = GF_DHT_MIGRATE_HARDLINK;
+    }
+
+    ret = dht_migrate_file(this, &entry_loc, cached_subvol, hashed_subvol,
+                           rebal_type, &fop_errno);
+    if (ret == 1) {
+        if (fop_errno == ENOSPC) {
+            gf_msg_debug(this->name, 0,
+                         "migrate-data skipped for"
+                         " %s due to space constraints",
+                         entry_loc.path);
+
+            /* For remove-brick case if the source is not one of the
+             * removed-brick, do not mark the error as failure */
+            if (conf->decommission_subvols_cnt) {
+                for (i = 0; i < conf->subvolume_cnt; i++) {
+                    if (conf->decommissioned_bricks[i] == cached_subvol) {
+                        LOCK(&defrag->lock);
+                        {
+                            defrag->total_failures += 1;
+                            update_skippedcount = _gf_false;
+                        }
+                        UNLOCK(&defrag->lock);
+
+                        break;
+                    }
+                }
+            }
+
+            if (update_skippedcount) {
+                LOCK(&defrag->lock);
+                {
+                    defrag->skipped += 1;
+                }
+                UNLOCK(&defrag->lock);
+
+                gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_MIGRATE_FILE_SKIPPED,
+                       "File migration skipped for %s.", entry_loc.path);
+            }
+
+        } else if (fop_errno == ENOTSUP) {
+            gf_msg_debug(this->name, 0,
+                         "migrate-data skipped for"
+                         " hardlink %s ",
+                         entry_loc.path);
+            LOCK(&defrag->lock);
+            {
+                defrag->skipped += 1;
+            }
+            UNLOCK(&defrag->lock);
+
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_MIGRATE_FILE_SKIPPED,
+                   "File migration skipped for %s.", entry_loc.path);
+        }
+
+        ret = 0;
+        goto out;
+    } else if (ret < 0) {
+        if (fop_errno != EEXIST) {
+            gf_msg(this->name, GF_LOG_ERROR, fop_errno,
+                   DHT_MSG_MIGRATE_FILE_FAILED, "migrate-data failed for %s",
+                   entry_loc.path);
+
+            LOCK(&defrag->lock);
+            {
+                defrag->total_failures += 1;
+            }
+            UNLOCK(&defrag->lock);
+        }
+
+        ret = gf_defrag_handle_migrate_error(fop_errno, defrag);
+
+        if (!ret) {
+            gf_msg(this->name, GF_LOG_ERROR, fop_errno,
+                   DHT_MSG_MIGRATE_FILE_FAILED,
+                   "migrate-data on %s failed:", entry_loc.path);
+        } else if (ret == 1) {
+            ret = 0;
+        }
+
+        goto out;
+    }
+
+    LOCK(&defrag->lock);
+    {
+        defrag->total_files += 1;
+        defrag->total_data += iatt.ia_size;
+    }
+    UNLOCK(&defrag->lock);
+
+    if (defrag->stats == _gf_true) {
+        gettimeofday(&end, NULL);
+        elapsed = gf_tvdiff(&start, &end);
+        gf_log(this->name, GF_LOG_INFO,
+               "Migration of "
+               "file:%s size:%" PRIu64
+               " bytes took %.2f"
+               "secs and ret: %d",
+               entry_loc.name, iatt.ia_size, elapsed / 1e6, ret);
+    }
+
+out:
+    if (statfs_frame) {
+        STACK_DESTROY(statfs_frame->root);
+    }
+
+    if (iatt_ptr) {
+        LOCK(&defrag->lock);
+        {
+            defrag->size_processed += iatt_ptr->ia_size;
+        }
+        UNLOCK(&defrag->lock);
+    }
+    loc_wipe(&entry_loc);
+
+    return ret;
+}
+
+void *
+gf_defrag_task(void *opaque)
+{
+    struct list_head *q_head = NULL;
+    struct dht_container *iterator = NULL;
+    gf_defrag_info_t *defrag = NULL;
+    int ret = 0;
+    pid_t pid = GF_CLIENT_PID_DEFRAG;
+
+    defrag = (gf_defrag_info_t *)opaque;
+    if (!defrag) {
+        gf_msg("dht", GF_LOG_ERROR, 0, 0, "defrag is NULL");
+        goto out;
+    }
+
+    syncopctx_setfspid(&pid);
+
+    q_head = &(defrag->queue[0].list);
+
+    /* The following while loop will dequeue one entry from the defrag->queue
+       under lock. We will update the defrag->global_error only when there
+       is an error which is critical to stop the rebalance process. The stop
+       message will be intimated to other migrator threads by setting the
+       defrag->defrag_status to GF_DEFRAG_STATUS_FAILED.
+
+       In defrag->queue, a low watermark (MIN_MIGRATE_QUEUE_COUNT) is
+       maintained so that crawler does not starve the file migration
+       workers and a high watermark (MAX_MIGRATE_QUEUE_COUNT) so that
+       crawler does not go far ahead in filling up the queue.
+     */
+
+    while (_gf_true) {
+        if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) {
+            pthread_cond_broadcast(&defrag->rebalance_crawler_alarm);
+            pthread_cond_broadcast(&defrag->parallel_migration_cond);
+            goto out;
+        }
+
+        pthread_mutex_lock(&defrag->dfq_mutex);
+        {
+            /*Throttle down:
+              If the reconfigured count is less than current thread
+              count, then the current thread will sleep */
+
+            /*TODO: Need to refactor the following block to work
+             *under defrag->lock. For now access
+             * defrag->current_thread_count and rthcount under
+             * dfq_mutex lock */
+            while (!defrag->crawl_done && (defrag->recon_thread_count <
+                                           defrag->current_thread_count)) {
+                defrag->current_thread_count--;
+                gf_msg_debug("DHT", 0,
+                             "Thread sleeping. "
+                             "current thread count: %d",
+                             defrag->current_thread_count);
+
+                pthread_cond_wait(&defrag->df_wakeup_thread,
+                                  &defrag->dfq_mutex);
+
+                defrag->current_thread_count++;
+                gf_msg_debug("DHT", 0,
+                             "Thread wokeup. "
+                             "current thread count: %d",
+                             defrag->current_thread_count);
+            }
+
+            if (defrag->q_entry_count) {
+                iterator = list_entry(q_head->next, typeof(*iterator), list);
+
+                gf_msg_debug("DHT", 0,
+                             "picking entry "
+                             "%s",
+                             iterator->df_entry->d_name);
+
+                list_del_init(&(iterator->list));
+
+                defrag->q_entry_count--;
+
+                if ((defrag->q_entry_count < MIN_MIGRATE_QUEUE_COUNT) &&
+                    defrag->wakeup_crawler) {
+                    pthread_cond_broadcast(&defrag->rebalance_crawler_alarm);
+                }
+                pthread_mutex_unlock(&defrag->dfq_mutex);
+                ret = gf_defrag_migrate_single_file((void *)iterator);
+
+                /*Critical errors: ENOTCONN and ENOSPACE*/
+                if (ret) {
+                    dht_set_global_defrag_error(defrag, ret);
+
+                    defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
+
+                    pthread_cond_broadcast(&defrag->rebalance_crawler_alarm);
+
+                    pthread_cond_broadcast(&defrag->parallel_migration_cond);
+
+                    goto out;
+                }
+
+                gf_defrag_free_container(iterator);
+
+                continue;
+            } else {
+                /* defrag->crawl_done flag is set means crawling
+                 file system is done and hence a list_empty when
+                 the above flag is set indicates there are no more
+                 entries to be added to the queue and rebalance is
+                 finished */
+
+                if (!defrag->crawl_done) {
+                    defrag->current_thread_count--;
+                    gf_msg_debug("DHT", 0,
+                                 "Thread "
+                                 "sleeping while  waiting "
+                                 "for migration entries. "
+                                 "current thread  count:%d",
+                                 defrag->current_thread_count);
+
+                    pthread_cond_wait(&defrag->parallel_migration_cond,
+                                      &defrag->dfq_mutex);
+                }
+
+                if (defrag->crawl_done && !defrag->q_entry_count) {
+                    defrag->current_thread_count++;
+                    gf_msg_debug("DHT", 0, "Exiting thread");
+
+                    pthread_cond_broadcast(&defrag->parallel_migration_cond);
+                    goto unlock;
+                } else {
+                    defrag->current_thread_count++;
+                    gf_msg_debug("DHT", 0,
+                                 "Thread woke up"
+                                 " as found migrating entries. "
+                                 "current thread count:%d",
+                                 defrag->current_thread_count);
+
+                    pthread_mutex_unlock(&defrag->dfq_mutex);
+                    continue;
+                }
+            }
+        }
+    unlock:
+        pthread_mutex_unlock(&defrag->dfq_mutex);
+        break;
+    }
+out:
+    return NULL;
+}
+
+int static gf_defrag_get_entry(xlator_t *this, int i,
+                               struct dht_container **container, loc_t *loc,
+                               dht_conf_t *conf, gf_defrag_info_t *defrag,
+                               fd_t *fd, dict_t *migrate_data,
+                               struct dir_dfmeta *dir_dfmeta, dict_t *xattr_req,
+                               int *perrno)
+{
+    int ret = 0;
+    char is_linkfile = 0;
+    gf_dirent_t *df_entry = NULL;
+    struct dht_container *tmp_container = NULL;
+
+    if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) {
+        ret = -1;
+        goto out;
+    }
+
+    if (dir_dfmeta->offset_var[i].readdir_done == 1) {
+        ret = 0;
+        goto out;
+    }
+
+    if (dir_dfmeta->fetch_entries[i] == 1) {
+        if (!fd) {
+            dir_dfmeta->fetch_entries[i] = 0;
+            dir_dfmeta->offset_var[i].readdir_done = 1;
+            ret = 0;
+            goto out;
+        }
+
+        ret = syncop_readdirp(conf->local_subvols[i], fd, 131072,
+                              dir_dfmeta->offset_var[i].offset,
+                              &(dir_dfmeta->equeue[i]), xattr_req, NULL);
+        if (ret == 0) {
+            dir_dfmeta->offset_var[i].readdir_done = 1;
+            ret = 0;
+            goto out;
+        }
+
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_WARNING, -ret,
+                   DHT_MSG_MIGRATE_DATA_FAILED,
+                   "Readdirp failed. Aborting data migration for "
+                   "directory: %s",
+                   loc->path);
+            *perrno = -ret;
+            ret = -1;
+            goto out;
+        }
+
+        if (list_empty(&(dir_dfmeta->equeue[i].list))) {
+            dir_dfmeta->offset_var[i].readdir_done = 1;
+            ret = 0;
+            goto out;
+        }
+
+        dir_dfmeta->fetch_entries[i] = 0;
+    }
+
+    while (1) {
+        if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) {
+            ret = -1;
+            goto out;
+        }
+
+        df_entry = list_entry(dir_dfmeta->iterator[i]->next, typeof(*df_entry),
+                              list);
+
+        if (&df_entry->list == dir_dfmeta->head[i]) {
+            gf_dirent_free(&(dir_dfmeta->equeue[i]));
+            INIT_LIST_HEAD(&(dir_dfmeta->equeue[i].list));
+            dir_dfmeta->fetch_entries[i] = 1;
+            dir_dfmeta->iterator[i] = dir_dfmeta->head[i];
+            ret = 0;
+            goto out;
+        }
+
+        dir_dfmeta->iterator[i] = dir_dfmeta->iterator[i]->next;
+
+        dir_dfmeta->offset_var[i].offset = df_entry->d_off;
+        if (!strcmp(df_entry->d_name, ".") || !strcmp(df_entry->d_name, ".."))
+            continue;
+
+        if (IA_ISDIR(df_entry->d_stat.ia_type)) {
+            defrag->size_processed += df_entry->d_stat.ia_size;
+            continue;
+        }
+
+        defrag->num_files_lookedup++;
+
+        if (defrag->defrag_pattern &&
+            (gf_defrag_pattern_match(defrag, df_entry->d_name,
+                                     df_entry->d_stat.ia_size) == _gf_false)) {
+            defrag->size_processed += df_entry->d_stat.ia_size;
+            continue;
+        }
+
+        is_linkfile = check_is_linkfile(NULL, &df_entry->d_stat, df_entry->dict,
+                                        conf->link_xattr_name);
+
+        if (is_linkfile) {
+            /* No need to add linkto file to the queue for
+               migration. Only the actual data file need to
+               be checked for migration criteria.
+            */
+
+            gf_msg_debug(this->name, 0,
+                         "Skipping linkfile"
+                         " %s on subvol: %s",
+                         df_entry->d_name, conf->local_subvols[i]->name);
+            continue;
+        }
+
+        /*Build Container Structure */
+
+        tmp_container = GF_CALLOC(1, sizeof(struct dht_container),
+                                  gf_dht_mt_container_t);
+        if (!tmp_container) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "Failed to allocate "
+                   "memory for container");
+            ret = -1;
+            goto out;
+        }
+        tmp_container->df_entry = gf_dirent_for_name(df_entry->d_name);
+        if (!tmp_container->df_entry) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "Failed to allocate "
+                   "memory for df_entry");
+            ret = -1;
+            goto out;
+        }
+
+        tmp_container->local_subvol_index = i;
+
+        tmp_container->df_entry->d_stat = df_entry->d_stat;
+
+        tmp_container->df_entry->d_ino = df_entry->d_ino;
+
+        tmp_container->df_entry->d_type = df_entry->d_type;
+
+        tmp_container->df_entry->d_len = df_entry->d_len;
+
+        tmp_container->parent_loc = GF_CALLOC(1, sizeof(*loc), gf_dht_mt_loc_t);
+        if (!tmp_container->parent_loc) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "Failed to allocate "
+                   "memory for loc");
+            ret = -1;
+            goto out;
+        }
+
+        ret = loc_copy(tmp_container->parent_loc, loc);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR, "loc_copy failed");
+            ret = -1;
+            goto out;
+        }
+
+        tmp_container->migrate_data = migrate_data;
+
+        tmp_container->this = this;
+
+        if (df_entry->dict)
+            tmp_container->df_entry->dict = dict_ref(df_entry->dict);
+
+        /*Build Container Structure >> END*/
+
+        ret = 0;
+        goto out;
+    }
+
+out:
+    if (ret == 0) {
+        *container = tmp_container;
+    } else {
+        if (tmp_container) {
+            gf_defrag_free_container(tmp_container);
+        }
+    }
+
+    return ret;
+}
+
+int
+gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
+                      dict_t *migrate_data, int *perrno)
+{
+    int ret = -1;
+    dht_conf_t *conf = NULL;
+    gf_dirent_t entries;
+    dict_t *xattr_req = NULL;
+    struct timeval dir_start = {
+        0,
+    };
+    struct timeval end = {
+        0,
+    };
+    double elapsed = {
+        0,
+    };
+    int local_subvols_cnt = 0;
+    int i = 0;
+    int j = 0;
+    struct dht_container *container = NULL;
+    int ldfq_count = 0;
+    int dfc_index = 0;
+    int throttle_up = 0;
+    struct dir_dfmeta *dir_dfmeta = NULL;
+    xlator_t *old_THIS = NULL;
+
+    gf_log(this->name, GF_LOG_INFO, "migrate data called on %s", loc->path);
+    gettimeofday(&dir_start, NULL);
+
+    conf = this->private;
+    local_subvols_cnt = conf->local_subvols_cnt;
+
+    if (!local_subvols_cnt) {
+        ret = 0;
+        goto out;
+    }
+
+    old_THIS = THIS;
+    THIS = this;
+
+    dir_dfmeta = GF_CALLOC(1, sizeof(*dir_dfmeta), gf_common_mt_pointer);
+    if (!dir_dfmeta) {
+        gf_log(this->name, GF_LOG_ERROR, "dir_dfmeta is NULL");
+        ret = -1;
+        goto out;
+    }
+
+    dir_dfmeta->lfd = GF_CALLOC(local_subvols_cnt, sizeof(fd_t *),
+                                gf_common_mt_pointer);
+    if (!dir_dfmeta->lfd) {
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_INSUFF_MEMORY,
+                "for dir_dfmeta", NULL);
+        ret = -1;
+        *perrno = ENOMEM;
+        goto out;
+    }
+
+    for (i = 0; i < local_subvols_cnt; i++) {
+        dir_dfmeta->lfd[i] = fd_create(loc->inode, defrag->pid);
+        if (!dir_dfmeta->lfd[i]) {
+            gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_FD_CREATE_FAILED,
+                    NULL);
+            *perrno = ENOMEM;
+            ret = -1;
+            goto out;
+        }
+
+        ret = syncop_opendir(conf->local_subvols[i], loc, dir_dfmeta->lfd[i],
+                             NULL, NULL);
+        if (ret) {
+            fd_unref(dir_dfmeta->lfd[i]);
+            dir_dfmeta->lfd[i] = NULL;
+            gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_FAILED_TO_OPEN,
+                    "dir: %s", loc->path, "subvol: %s",
+                    conf->local_subvols[i]->name, NULL);
+
+            if (conf->decommission_in_progress) {
+                *perrno = -ret;
+                ret = -1;
+                goto out;
+            }
+        } else {
+            fd_bind(dir_dfmeta->lfd[i]);
+        }
+    }
+
+    dir_dfmeta->head = GF_CALLOC(local_subvols_cnt, sizeof(*(dir_dfmeta->head)),
+                                 gf_common_mt_pointer);
+    if (!dir_dfmeta->head) {
+        gf_log(this->name, GF_LOG_ERROR, "dir_dfmeta->head is NULL");
+        ret = -1;
+        goto out;
+    }
+
+    dir_dfmeta->iterator = GF_CALLOC(local_subvols_cnt,
+                                     sizeof(*(dir_dfmeta->iterator)),
+                                     gf_common_mt_pointer);
+    if (!dir_dfmeta->iterator) {
+        gf_log(this->name, GF_LOG_ERROR, "dir_dfmeta->iterator is NULL");
+        ret = -1;
+        goto out;
+    }
+
+    dir_dfmeta->equeue = GF_CALLOC(local_subvols_cnt, sizeof(entries),
+                                   gf_dht_mt_dirent_t);
+    if (!dir_dfmeta->equeue) {
+        gf_log(this->name, GF_LOG_ERROR, "dir_dfmeta->equeue is NULL");
+        ret = -1;
+        goto out;
+    }
+
+    dir_dfmeta->offset_var = GF_CALLOC(
+        local_subvols_cnt, sizeof(dht_dfoffset_ctx_t), gf_dht_mt_octx_t);
+    if (!dir_dfmeta->offset_var) {
+        gf_log(this->name, GF_LOG_ERROR, "dir_dfmeta->offset_var is NULL");
+        ret = -1;
+        goto out;
+    }
+
+    ret = gf_defrag_ctx_subvols_init(dir_dfmeta->offset_var, this);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "dht_dfoffset_ctx_t"
+               "initialization failed");
+        ret = -1;
+        goto out;
+    }
+
+    dir_dfmeta->fetch_entries = GF_CALLOC(local_subvols_cnt, sizeof(int),
+                                          gf_common_mt_int);
+    if (!dir_dfmeta->fetch_entries) {
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_INSUFF_MEMORY,
+                "for dir_dfmeta->fetch_entries", NULL);
+        ret = -1;
+        goto out;
+    }
+
+    for (i = 0; i < local_subvols_cnt; i++) {
+        INIT_LIST_HEAD(&(dir_dfmeta->equeue[i].list));
+        dir_dfmeta->head[i] = &(dir_dfmeta->equeue[i].list);
+        dir_dfmeta->iterator[i] = dir_dfmeta->head[i];
+        dir_dfmeta->fetch_entries[i] = 1;
+    }
+
+    xattr_req = dict_new();
+    if (!xattr_req) {
+        gf_log(this->name, GF_LOG_ERROR, "dict_new failed");
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_set_uint32(xattr_req, conf->link_xattr_name, 256);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to set dict for "
+               "key: %s",
+               conf->link_xattr_name);
+        ret = -1;
+        goto out;
+    }
+
+    /*
+     Job: Read entries from each local subvol and store the entries
+          in equeue array of linked list. Now pick one entry from the
+          equeue array in a round robin basis and add them to defrag Queue.
+    */
+
+    while (!dht_dfreaddirp_done(dir_dfmeta->offset_var, local_subvols_cnt)) {
+        pthread_mutex_lock(&defrag->dfq_mutex);
+        {
+            /*Throttle up: If reconfigured count is higher than
+              current thread count, wake up the sleeping threads
+              TODO: Need to refactor this. Instead of making the
+              thread sleep and wake, we should terminate and spawn
+              threads on-demand*/
+
+            if (defrag->recon_thread_count > defrag->current_thread_count) {
+                throttle_up = (defrag->recon_thread_count -
+                               defrag->current_thread_count);
+                for (j = 0; j < throttle_up; j++) {
+                    pthread_cond_signal(&defrag->df_wakeup_thread);
+                }
+            }
+
+            while (defrag->q_entry_count > MAX_MIGRATE_QUEUE_COUNT) {
+                defrag->wakeup_crawler = 1;
+                pthread_cond_wait(&defrag->rebalance_crawler_alarm,
+                                  &defrag->dfq_mutex);
+            }
+
+            ldfq_count = defrag->q_entry_count;
+
+            if (defrag->wakeup_crawler) {
+                defrag->wakeup_crawler = 0;
+            }
+        }
+        pthread_mutex_unlock(&defrag->dfq_mutex);
+
+        while (
+            ldfq_count <= MAX_MIGRATE_QUEUE_COUNT &&
+            !dht_dfreaddirp_done(dir_dfmeta->offset_var, local_subvols_cnt)) {
+            ret = gf_defrag_get_entry(this, dfc_index, &container, loc, conf,
+                                      defrag, dir_dfmeta->lfd[dfc_index],
+                                      migrate_data, dir_dfmeta, xattr_req,
+                                      perrno);
+
+            if (defrag->defrag_status == GF_DEFRAG_STATUS_STOPPED) {
+                goto out;
+            }
+
+            if (ret) {
+                gf_log(this->name, GF_LOG_WARNING,
+                       "Found "
+                       "error from gf_defrag_get_entry");
+
+                ret = -1;
+                goto out;
+            }
+
+            /* Check if we got an entry, else we need to move the
+               index to the next subvol */
+            if (!container) {
+                GF_CRAWL_INDEX_MOVE(dfc_index, local_subvols_cnt);
+                continue;
+            }
+
+            /* Q this entry in the dfq */
+            pthread_mutex_lock(&defrag->dfq_mutex);
+            {
+                list_add_tail(&container->list, &(defrag->queue[0].list));
+                defrag->q_entry_count++;
+                ldfq_count = defrag->q_entry_count;
+
+                gf_msg_debug(this->name, 0,
+                             "added "
+                             "file:%s parent:%s to the queue ",
+                             container->df_entry->d_name,
+                             container->parent_loc->path);
+
+                pthread_cond_signal(&defrag->parallel_migration_cond);
+            }
+            pthread_mutex_unlock(&defrag->dfq_mutex);
+
+            GF_CRAWL_INDEX_MOVE(dfc_index, local_subvols_cnt);
+        }
+    }
+
+    gettimeofday(&end, NULL);
+    elapsed = gf_tvdiff(&dir_start, &end);
+    gf_log(this->name, GF_LOG_INFO,
+           "Migration operation on dir %s took "
+           "%.2f secs",
+           loc->path, elapsed / 1e6);
+    ret = 0;
+out:
+    THIS = old_THIS;
+    gf_defrag_free_dir_dfmeta(dir_dfmeta, local_subvols_cnt);
+
+    if (xattr_req)
+        dict_unref(xattr_req);
+
+    /* It does not matter if it errored out - this number is
+     * used to calculate rebalance estimated time to complete.
+     * No locking required as dirs are processed by a single thread.
+     */
+    defrag->num_dirs_processed++;
+    return ret;
+}
+
+int
+gf_defrag_settle_hash(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
+                      dict_t *fix_layout)
+{
+    int ret;
+    dht_conf_t *conf = NULL;
+    /*
+     * Now we're ready to update the directory commit hash for the volume
+     * root, so that hash miscompares and broadcast lookups can stop.
+     * However, we want to skip that if fix-layout is all we did.  In
+     * that case, we want the miscompares etc. to continue until a real
+     * rebalance is complete.
+     */
+    if (defrag->cmd == GF_DEFRAG_CMD_START_LAYOUT_FIX ||
+        defrag->cmd == GF_DEFRAG_CMD_DETACH_START) {
+        return 0;
+    }
+
+    conf = this->private;
+    if (!conf) {
+        /*Uh oh
+         */
+        return -1;
+    }
+
+    if (conf->local_subvols_cnt == 0 || !conf->lookup_optimize) {
+        /* Commit hash updates are only done on local subvolumes and
+         * only when lookup optimization is needed (for older client
+         * support)
+         */
+        return 0;
+    }
+
+    ret = dict_set_uint32(fix_layout, "new-commit-hash",
+                          defrag->new_commit_hash);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR, "Failed to set new-commit-hash");
+        return -1;
+    }
+
+    ret = syncop_setxattr(this, loc, fix_layout, 0, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_LAYOUT_FIX_FAILED,
+               "fix layout on %s failed", loc->path);
+
+        if (-ret == ENOENT || -ret == ESTALE) {
+            /* Dir most likely is deleted */
+            return 0;
+        }
+
+        return -1;
+    }
+
+    /* TBD: find more efficient solution than adding/deleting every time */
+    dict_del(fix_layout, "new-commit-hash");
+
+    return 0;
+}
+
+int
+gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
+                     dict_t *fix_layout, dict_t *migrate_data)
+{
+    int ret = -1;
+    loc_t entry_loc = {
+        0,
+    };
+    fd_t *fd = NULL;
+    gf_dirent_t entries;
+    gf_dirent_t *tmp = NULL;
+    gf_dirent_t *entry = NULL;
+    gf_boolean_t free_entries = _gf_false;
+    off_t offset = 0;
+    struct iatt iatt = {
+        0,
+    };
+    inode_t *linked_inode = NULL, *inode = NULL;
+    dht_conf_t *conf = NULL;
+    int perrno = 0;
+
+    conf = this->private;
+    if (!conf) {
+        ret = -1;
+        goto out;
+    }
+
+    ret = syncop_lookup(this, loc, &iatt, NULL, NULL, NULL);
+    if (ret) {
+        if (strcmp(loc->path, "/") == 0) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_DIR_LOOKUP_FAILED,
+                   "lookup failed for:%s", loc->path);
+
+            defrag->total_failures++;
+            ret = -1;
+            goto out;
+        }
+
+        if (-ret == ENOENT || -ret == ESTALE) {
+            gf_msg(this->name, GF_LOG_INFO, -ret, DHT_MSG_DIR_LOOKUP_FAILED,
+                   "Dir:%s renamed or removed. Skipping", loc->path);
+            if (conf->decommission_subvols_cnt) {
+                defrag->total_failures++;
+            }
+            ret = 0;
+            goto out;
+        } else {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_DIR_LOOKUP_FAILED,
+                   "lookup failed for:%s", loc->path);
+
+            defrag->total_failures++;
+            goto out;
+        }
+    }
+
+    fd = fd_create(loc->inode, defrag->pid);
+    if (!fd) {
+        gf_log(this->name, GF_LOG_ERROR, "Failed to create fd");
+        ret = -1;
+        goto out;
+    }
+
+    ret = syncop_opendir(this, loc, fd, NULL, NULL);
+    if (ret) {
+        if (-ret == ENOENT || -ret == ESTALE) {
+            if (conf->decommission_subvols_cnt) {
+                defrag->total_failures++;
+            }
+            ret = 0;
+            goto out;
+        }
+
+        gf_log(this->name, GF_LOG_ERROR,
+               "Failed to open dir %s, "
+               "err:%d",
+               loc->path, -ret);
+
+        ret = -1;
+        goto out;
+    }
+
+    fd_bind(fd);
+    INIT_LIST_HEAD(&entries.list);
+
+    while ((ret = syncop_readdirp(this, fd, 131072, offset, &entries, NULL,
+                                  NULL)) != 0) {
+        if (ret < 0) {
+            if (-ret == ENOENT || -ret == ESTALE) {
+                if (conf->decommission_subvols_cnt) {
+                    defrag->total_failures++;
+                }
+                ret = 0;
+                goto out;
+            }
+
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_READDIR_ERROR,
+                   "readdirp failed for "
+                   "path %s. Aborting fix-layout",
+                   loc->path);
+
+            ret = -1;
+            goto out;
+        }
+
+        if (list_empty(&entries.list))
+            break;
+
+        free_entries = _gf_true;
+
+        list_for_each_entry_safe(entry, tmp, &entries.list, list)
+        {
+            if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED) {
+                ret = 1;
+                goto out;
+            }
+
+            offset = entry->d_off;
+
+            if (!strcmp(entry->d_name, ".") || !strcmp(entry->d_name, ".."))
+                continue;
+            if (!IA_ISDIR(entry->d_stat.ia_type)) {
+                continue;
+            }
+            loc_wipe(&entry_loc);
+
+            ret = dht_build_child_loc(this, &entry_loc, loc, entry->d_name);
+            if (ret) {
+                gf_log(this->name, GF_LOG_ERROR,
+                       "Child loc"
+                       " build failed for entry: %s",
+                       entry->d_name);
+
+                if (conf->decommission_in_progress) {
+                    defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
+
+                    goto out;
+                } else {
+                    continue;
+                }
+            }
+
+            if (gf_uuid_is_null(entry->d_stat.ia_gfid)) {
+                gf_log(this->name, GF_LOG_ERROR,
+                       "%s/%s"
+                       " gfid not present",
+                       loc->path, entry->d_name);
+                continue;
+            }
+
+            gf_uuid_copy(entry_loc.gfid, entry->d_stat.ia_gfid);
+
+            /*In case the gfid stored in the inode by inode_link
+             * and the gfid obtained in the lookup differs, then
+             * client3_3_lookup_cbk will return ESTALE and proper
+             * error will be captured
+             */
+
+            linked_inode = inode_link(entry_loc.inode, loc->inode,
+                                      entry->d_name, &entry->d_stat);
+
+            inode = entry_loc.inode;
+            entry_loc.inode = linked_inode;
+            inode_unref(inode);
+
+            if (gf_uuid_is_null(loc->gfid)) {
+                gf_log(this->name, GF_LOG_ERROR,
+                       "%s/%s"
+                       " gfid not present",
+                       loc->path, entry->d_name);
+                continue;
+            }
+
+            gf_uuid_copy(entry_loc.pargfid, loc->gfid);
+
+            ret = syncop_lookup(this, &entry_loc, &iatt, NULL, NULL, NULL);
+            if (ret) {
+                if (-ret == ENOENT || -ret == ESTALE) {
+                    gf_msg(this->name, GF_LOG_INFO, -ret,
+                           DHT_MSG_DIR_LOOKUP_FAILED,
+                           "Dir:%s renamed or removed. "
+                           "Skipping",
+                           loc->path);
+                    ret = 0;
+                    if (conf->decommission_subvols_cnt) {
+                        defrag->total_failures++;
+                    }
+                    continue;
+                } else {
+                    gf_msg(this->name, GF_LOG_ERROR, -ret,
+                           DHT_MSG_DIR_LOOKUP_FAILED, "lookup failed for:%s",
+                           entry_loc.path);
+
+                    defrag->total_failures++;
+
+                    if (conf->decommission_in_progress) {
+                        defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
+                        ret = -1;
+                        goto out;
+                    } else {
+                        continue;
+                    }
+                }
+            }
+
+            /* A return value of 2 means, either process_dir or
+             * lookup of a dir failed. Hence, don't commit hash
+             * for the current directory*/
+
+            ret = gf_defrag_fix_layout(this, defrag, &entry_loc, fix_layout,
+                                       migrate_data);
+
+            if (defrag->defrag_status == GF_DEFRAG_STATUS_STOPPED ||
+                defrag->defrag_status == GF_DEFRAG_STATUS_FAILED) {
+                goto out;
+            }
+
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LAYOUT_FIX_FAILED,
+                       "Fix layout failed for %s", entry_loc.path);
+
+                defrag->total_failures++;
+
+                if (conf->decommission_in_progress) {
+                    defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
+
+                    goto out;
+                } else {
+                    /* Let's not commit-hash if
+                     * gf_defrag_fix_layout failed*/
+                    continue;
+                }
+            }
+        }
+
+        gf_dirent_free(&entries);
+        free_entries = _gf_false;
+        INIT_LIST_HEAD(&entries.list);
+    }
+
+    /* A directory layout is fixed only after its subdirs are healed to
+     * any newly added bricks. If the layout is fixed before subdirs are
+     * healed, the newly added brick will get a non-null layout.
+     * Any subdirs which hash to that layout will no longer show up
+     * in a directory listing until they are healed.
+     */
+
+    ret = syncop_setxattr(this, loc, fix_layout, 0, NULL, NULL);
+
+    /* In case of a race where the directory is deleted just before
+     * layout setxattr, the errors are updated in the layout structure.
+     * We can use this information to make a decision whether the directory
+     * is deleted entirely.
+     */
+    if (ret == 0) {
+        ret = dht_dir_layout_error_check(this, loc->inode);
+        ret = -ret;
+    }
+
+    if (ret) {
+        if (-ret == ENOENT || -ret == ESTALE) {
+            gf_msg(this->name, GF_LOG_INFO, -ret, DHT_MSG_LAYOUT_FIX_FAILED,
+                   "Setxattr failed. Dir %s "
+                   "renamed or removed",
+                   loc->path);
+            if (conf->decommission_subvols_cnt) {
+                defrag->total_failures++;
+            }
+            ret = 0;
+            goto out;
+        } else {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_LAYOUT_FIX_FAILED,
+                   "Setxattr failed for %s", loc->path);
+
+            defrag->total_failures++;
+
+            if (conf->decommission_in_progress) {
+                defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
+                ret = -1;
+                goto out;
+            }
+        }
+    }
+
+    if (defrag->cmd != GF_DEFRAG_CMD_START_LAYOUT_FIX) {
+        ret = gf_defrag_process_dir(this, defrag, loc, migrate_data, &perrno);
+
+        if (ret) {
+            if (perrno == ENOENT || perrno == ESTALE) {
+                ret = 0;
+                goto out;
+            } else {
+                defrag->total_failures++;
+
+                gf_msg(this->name, GF_LOG_ERROR, 0,
+                       DHT_MSG_DEFRAG_PROCESS_DIR_FAILED,
+                       "gf_defrag_process_dir failed for "
+                       "directory: %s",
+                       loc->path);
+
+                if (conf->decommission_in_progress) {
+                    goto out;
+                }
+            }
+        }
+    }
+
+    gf_msg_trace(this->name, 0, "fix layout called on %s", loc->path);
+
+    if (gf_defrag_settle_hash(this, defrag, loc, fix_layout) != 0) {
+        defrag->total_failures++;
+
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SETTLE_HASH_FAILED,
+               "Settle hash failed for %s", loc->path);
+
+        ret = -1;
+
+        if (conf->decommission_in_progress) {
+            defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
+            goto out;
+        }
+    }
+
+    ret = 0;
+out:
+    if (free_entries)
+        gf_dirent_free(&entries);
+
+    loc_wipe(&entry_loc);
+
+    if (fd)
+        fd_unref(fd);
+
+    return ret;
+}
+
+int
+dht_init_local_subvols_and_nodeuuids(xlator_t *this, dht_conf_t *conf,
+                                     loc_t *loc)
+{
+    dict_t *dict = NULL;
+    uuid_t *uuid_ptr = NULL;
+    int ret = -1;
+    int i = 0;
+    int j = 0;
+
+    /* Find local subvolumes */
+    ret = syncop_getxattr(this, loc, &dict, GF_REBAL_FIND_LOCAL_SUBVOL, NULL,
+                          NULL);
+    if (ret && (ret != -ENODATA)) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, 0,
+               "local "
+               "subvolume determination failed with error: %d",
+               -ret);
+        ret = -1;
+        goto out;
+    }
+
+    if (!ret)
+        goto out;
+
+    ret = syncop_getxattr(this, loc, &dict, GF_REBAL_OLD_FIND_LOCAL_SUBVOL,
+                          NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, 0,
+               "local "
+               "subvolume determination failed with error: %d",
+               -ret);
+        ret = -1;
+        goto out;
+    }
+    ret = 0;
+
+out:
+    if (ret) {
+        return ret;
+    }
+
+    for (i = 0; i < conf->local_subvols_cnt; i++) {
+        gf_msg(this->name, GF_LOG_INFO, 0, 0,
+               "local subvol: "
+               "%s",
+               conf->local_subvols[i]->name);
+
+        for (j = 0; j < conf->local_nodeuuids[i].count; j++) {
+            uuid_ptr = &(conf->local_nodeuuids[i].elements[j].uuid);
+            gf_msg(this->name, GF_LOG_INFO, 0, 0, "node uuid : %s",
+                   uuid_utoa(*uuid_ptr));
+        }
+    }
+
+    return ret;
+}
+
+/* Functions for the rebalance estimates feature */
+
+uint64_t
+gf_defrag_subvol_file_size(xlator_t *this, loc_t *root_loc)
+{
+    int ret = -1;
+    struct statvfs buf = {
+        0,
+    };
+
+    ret = syncop_statfs(this, root_loc, &buf, NULL, NULL);
+    if (ret) {
+        /* Aargh! */
+        return 0;
+    }
+    return ((buf.f_blocks - buf.f_bfree) * buf.f_frsize);
+}
+
+uint64_t
+gf_defrag_total_file_size(xlator_t *this, loc_t *root_loc)
+{
+    dht_conf_t *conf = NULL;
+    int i = 0;
+    uint64_t size_files = 0;
+    uint64_t total_size = 0;
+
+    conf = this->private;
+    if (!conf) {
+        return 0;
+    }
+
+    for (i = 0; i < conf->local_subvols_cnt; i++) {
+        size_files = gf_defrag_subvol_file_size(conf->local_subvols[i],
+                                                root_loc);
+        total_size += size_files;
+        gf_msg(this->name, GF_LOG_INFO, 0, 0,
+               "local subvol: %s,"
+               "cnt = %" PRIu64,
+               conf->local_subvols[i]->name, size_files);
+    }
+
+    gf_msg(this->name, GF_LOG_INFO, 0, 0, "Total size files = %" PRIu64,
+           total_size);
+
+    return total_size;
+}
+
+static void *
+dht_file_counter_thread(void *args)
+{
+    gf_defrag_info_t *defrag = NULL;
+    loc_t root_loc = {
+        0,
+    };
+    struct timespec time_to_wait = {
+        0,
+    };
+    uint64_t tmp_size = 0;
+
+    if (!args)
+        return NULL;
+
+    defrag = (gf_defrag_info_t *)args;
+    dht_build_root_loc(defrag->root_inode, &root_loc);
+
+    while (defrag->defrag_status == GF_DEFRAG_STATUS_STARTED) {
+        timespec_now(&time_to_wait);
+        time_to_wait.tv_sec += 600;
+
+        pthread_mutex_lock(&defrag->fc_mutex);
+        pthread_cond_timedwait(&defrag->fc_wakeup_cond, &defrag->fc_mutex,
+                               &time_to_wait);
+
+        pthread_mutex_unlock(&defrag->fc_mutex);
+
+        if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED)
+            break;
+
+        tmp_size = gf_defrag_total_file_size(defrag->this, &root_loc);
+
+        gf_log("dht", GF_LOG_INFO, "tmp data size =%" PRIu64, tmp_size);
+
+        if (!tmp_size) {
+            gf_msg("dht", GF_LOG_ERROR, 0, 0,
+                   "Failed to get "
+                   "the total data size. Unable to estimate "
+                   "time to complete rebalance.");
+        } else {
+            g_totalsize = tmp_size;
+            gf_msg_debug("dht", 0, "total data size =%" PRIu64, g_totalsize);
+        }
+    }
+
+    return NULL;
+}
+
+int
+gf_defrag_estimates_cleanup(xlator_t *this, gf_defrag_info_t *defrag,
+                            pthread_t filecnt_thread)
+{
+    int ret = -1;
+
+    /* Wake up the filecounter thread.
+     * By now the defrag status will no longer be
+     * GF_DEFRAG_STATUS_STARTED so the thread will exit the loop.
+     */
+    pthread_mutex_lock(&defrag->fc_mutex);
+    {
+        pthread_cond_broadcast(&defrag->fc_wakeup_cond);
+    }
+    pthread_mutex_unlock(&defrag->fc_mutex);
+
+    ret = pthread_join(filecnt_thread, NULL);
+    if (ret) {
+        gf_msg("dht", GF_LOG_ERROR, ret, 0,
+               "file_counter_thread: pthread_join failed.");
+        ret = -1;
+    }
+    return ret;
+}
+
+int
+gf_defrag_estimates_init(xlator_t *this, loc_t *loc, pthread_t *filecnt_thread)
+{
+    int ret = -1;
+    dht_conf_t *conf = NULL;
+    gf_defrag_info_t *defrag = NULL;
+
+    conf = this->private;
+    defrag = conf->defrag;
+
+    g_totalsize = gf_defrag_total_file_size(this, loc);
+    if (!g_totalsize) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "Failed to get "
+               "the total data size. Unable to estimate "
+               "time to complete rebalance.");
+        goto out;
+    }
+
+    ret = gf_thread_create(filecnt_thread, NULL, dht_file_counter_thread,
+                           (void *)defrag, "dhtfcnt");
+
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, ret, 0,
+               "Failed to "
+               "create the file counter thread ");
+        ret = -1;
+        goto out;
+    }
+    ret = 0;
+out:
+    return ret;
+}
+
+/* Init and cleanup functions for parallel file migration*/
+int
+gf_defrag_parallel_migration_init(xlator_t *this, gf_defrag_info_t *defrag,
+                                  pthread_t **tid_array, int *thread_index)
+{
+    int ret = -1;
+    int thread_spawn_count = 0;
+    int index = 0;
+    pthread_t *tid = NULL;
+
+    if (!defrag)
+        goto out;
+
+    /* Initialize global entry queue */
+    defrag->queue = GF_CALLOC(1, sizeof(struct dht_container),
+                              gf_dht_mt_container_t);
+
+    if (!defrag->queue) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, 0,
+               "Failed to initialise migration queue");
+        ret = -1;
+        goto out;
+    }
+
+    INIT_LIST_HEAD(&(defrag->queue[0].list));
+
+    thread_spawn_count = MAX(MAX_REBAL_THREADS, 4);
+
+    gf_msg_debug(this->name, 0, "thread_spawn_count: %d", thread_spawn_count);
+
+    tid = GF_CALLOC(thread_spawn_count, sizeof(pthread_t),
+                    gf_common_mt_pthread_t);
+    if (!tid) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, 0,
+               "Failed to create migration threads");
+        ret = -1;
+        goto out;
+    }
+    defrag->current_thread_count = thread_spawn_count;
+
+    /*Spawn Threads Here*/
+    while (index < thread_spawn_count) {
+        ret = gf_thread_create(&(tid[index]), NULL, gf_defrag_task,
+                               (void *)defrag, "dhtmig%d", (index + 1) & 0x3ff);
+        if (ret != 0) {
+            gf_msg("DHT", GF_LOG_ERROR, ret, 0, "Thread[%d] creation failed. ",
+                   index);
+            ret = -1;
+            goto out;
+        } else {
+            gf_log("DHT", GF_LOG_INFO,
+                   "Thread[%d] "
+                   "creation successful",
+                   index);
+        }
+        index++;
+    }
+
+    ret = 0;
+out:
+    *thread_index = index;
+    *tid_array = tid;
+
+    return ret;
+}
+
+int
+gf_defrag_parallel_migration_cleanup(gf_defrag_info_t *defrag,
+                                     pthread_t *tid_array, int thread_index)
+{
+    int ret = -1;
+    int i = 0;
+
+    if (!defrag)
+        goto out;
+
+    /* Wake up all migration threads */
+    pthread_mutex_lock(&defrag->dfq_mutex);
+    {
+        defrag->crawl_done = 1;
+
+        pthread_cond_broadcast(&defrag->parallel_migration_cond);
+        pthread_cond_broadcast(&defrag->df_wakeup_thread);
+    }
+    pthread_mutex_unlock(&defrag->dfq_mutex);
+
+    /*Wait for all the threads to complete their task*/
+    for (i = 0; i < thread_index; i++) {
+        pthread_join(tid_array[i], NULL);
+    }
+
+    GF_FREE(tid_array);
+
+    /* Cleanup the migration queue */
+    if (defrag->queue) {
+        gf_dirent_free(defrag->queue[0].df_entry);
+        INIT_LIST_HEAD(&(defrag->queue[0].list));
+    }
+
+    GF_FREE(defrag->queue);
+
+    ret = 0;
+out:
+    return ret;
+}
+
+int
+gf_defrag_start_crawl(void *data)
+{
+    xlator_t *this = NULL;
+    dht_conf_t *conf = NULL;
+    gf_defrag_info_t *defrag = NULL;
+    dict_t *fix_layout = NULL;
+    dict_t *migrate_data = NULL;
+    dict_t *status = NULL;
+    glusterfs_ctx_t *ctx = NULL;
+    call_frame_t *statfs_frame = NULL;
+    xlator_t *old_THIS = NULL;
+    int ret = -1;
+    loc_t loc = {
+        0,
+    };
+    struct iatt iatt = {
+        0,
+    };
+    struct iatt parent = {
+        0,
+    };
+    int thread_index = 0;
+    pthread_t *tid = NULL;
+    pthread_t filecnt_thread;
+    gf_boolean_t fc_thread_started = _gf_false;
+
+    this = data;
+    if (!this)
+        goto exit;
+
+    ctx = this->ctx;
+    if (!ctx)
+        goto exit;
+
+    conf = this->private;
+    if (!conf)
+        goto exit;
+
+    defrag = conf->defrag;
+    if (!defrag)
+        goto exit;
+
+    defrag->start_time = gf_time();
+
+    dht_build_root_inode(this, &defrag->root_inode);
+    if (!defrag->root_inode)
+        goto out;
+
+    dht_build_root_loc(defrag->root_inode, &loc);
+
+    /* fix-layout on '/' first */
+
+    ret = syncop_lookup(this, &loc, &iatt, &parent, NULL, NULL);
+
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_REBALANCE_START_FAILED,
+               "Failed to start rebalance: look up on / failed");
+        ret = -1;
+        goto out;
+    }
+
+    old_THIS = THIS;
+    THIS = this;
+
+    statfs_frame = create_frame(this, this->ctx->pool);
+    if (!statfs_frame) {
+        gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_NO_MEMORY, ENOMEM,
+               "Insufficient memory. Frame creation failed");
+        ret = -1;
+        goto out;
+    }
+
+    /* async statfs update for honoring min-free-disk */
+    dht_get_du_info(statfs_frame, this, &loc);
+    THIS = old_THIS;
+
+    fix_layout = dict_new();
+    if (!fix_layout) {
+        ret = -1;
+        goto out;
+    }
+
+    /*
+     * Unfortunately, we can't do special xattrs (like fix.layout) and
+     * real ones in the same call currently, and changing it seems
+     * riskier than just doing two calls.
+     */
+
+    gf_log(this->name, GF_LOG_INFO, "%s using commit hash %u", __func__,
+           conf->vol_commit_hash);
+
+    ret = dict_set_uint32(fix_layout, conf->commithash_xattr_name,
+                          conf->vol_commit_hash);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR, "Failed to set %s",
+               conf->commithash_xattr_name);
+        defrag->total_failures++;
+        ret = -1;
+        goto out;
+    }
+
+    ret = syncop_setxattr(this, &loc, fix_layout, 0, NULL, NULL);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Failed to set commit hash on %s. "
+               "Rebalance cannot proceed.",
+               loc.path);
+        defrag->total_failures++;
+        ret = -1;
+        goto out;
+    }
+
+    /* We now return to our regularly scheduled program. */
+
+    ret = dict_set_str(fix_layout, GF_XATTR_FIX_LAYOUT_KEY, "yes");
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_REBALANCE_START_FAILED,
+               "Failed to start rebalance:"
+               "Failed to set dictionary value: key = %s",
+               GF_XATTR_FIX_LAYOUT_KEY);
+        defrag->total_failures++;
+        ret = -1;
+        goto out;
+    }
+
+    defrag->new_commit_hash = conf->vol_commit_hash;
+
+    ret = syncop_setxattr(this, &loc, fix_layout, 0, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_REBALANCE_FAILED,
+               "fix layout on %s failed", loc.path);
+        defrag->total_failures++;
+        ret = -1;
+        goto out;
+    }
+
+    if (defrag->cmd != GF_DEFRAG_CMD_START_LAYOUT_FIX) {
+        /* We need to migrate files */
+
+        migrate_data = dict_new();
+        if (!migrate_data) {
+            defrag->total_failures++;
+            ret = -1;
+            goto out;
+        }
+        ret = dict_set_str(
+            migrate_data, GF_XATTR_FILE_MIGRATE_KEY,
+            (defrag->cmd == GF_DEFRAG_CMD_START_FORCE) ? "force" : "non-force");
+        if (ret) {
+            defrag->total_failures++;
+            ret = -1;
+            goto out;
+        }
+
+        ret = dht_init_local_subvols_and_nodeuuids(this, conf, &loc);
+        if (ret) {
+            ret = -1;
+            goto out;
+        }
+
+        /* Initialise the structures required for parallel migration */
+        ret = gf_defrag_parallel_migration_init(this, defrag, &tid,
+                                                &thread_index);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, 0, "Aborting rebalance.");
+            goto out;
+        }
+
+        ret = gf_defrag_estimates_init(this, &loc, &filecnt_thread);
+        if (ret) {
+            /* Not a fatal error. Allow the rebalance to proceed*/
+            ret = 0;
+        } else {
+            fc_thread_started = _gf_true;
+        }
+    }
+
+    ret = gf_defrag_fix_layout(this, defrag, &loc, fix_layout, migrate_data);
+    if (ret) {
+        defrag->total_failures++;
+        ret = -1;
+        goto out;
+    }
+
+    if (gf_defrag_settle_hash(this, defrag, &loc, fix_layout) != 0) {
+        defrag->total_failures++;
+        ret = -1;
+        goto out;
+    }
+
+    gf_log("DHT", GF_LOG_INFO, "crawling file-system completed");
+out:
+
+    /* We are here means crawling the entire file system is done
+       or something failed. Set defrag->crawl_done flag to intimate
+       the migrator threads to exhaust the defrag->queue and terminate*/
+
+    if (ret) {
+        defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
+    }
+
+    gf_defrag_parallel_migration_cleanup(defrag, tid, thread_index);
+
+    if ((defrag->defrag_status != GF_DEFRAG_STATUS_STOPPED) &&
+        (defrag->defrag_status != GF_DEFRAG_STATUS_FAILED)) {
+        defrag->defrag_status = GF_DEFRAG_STATUS_COMPLETE;
+    }
+
+    if (fc_thread_started) {
+        gf_defrag_estimates_cleanup(this, defrag, filecnt_thread);
+    }
+
+    dht_send_rebalance_event(this, defrag->cmd, defrag->defrag_status);
+
+    status = dict_new();
+    LOCK(&defrag->lock);
+    {
+        gf_defrag_status_get(conf, status);
+        if (ctx && ctx->notify)
+            ctx->notify(GF_EN_DEFRAG_STATUS, status);
+        if (status)
+            dict_unref(status);
+        defrag->is_exiting = 1;
+    }
+    UNLOCK(&defrag->lock);
+
+    GF_FREE(defrag);
+    conf->defrag = NULL;
+
+    if (migrate_data)
+        dict_unref(migrate_data);
+
+    if (statfs_frame) {
+        STACK_DESTROY(statfs_frame->root);
+    }
+exit:
+    return ret;
+}
+
+static int
+gf_defrag_done(int ret, call_frame_t *sync_frame, void *data)
+{
+    gf_listener_stop(sync_frame->this);
+
+    STACK_DESTROY(sync_frame->root);
+    kill(getpid(), SIGTERM);
+    return 0;
+}
+
+void *
+gf_defrag_start(void *data)
+{
+    int ret = -1;
+    call_frame_t *frame = NULL;
+    dht_conf_t *conf = NULL;
+    gf_defrag_info_t *defrag = NULL;
+    xlator_t *this = NULL;
+    xlator_t *old_THIS = NULL;
+
+    this = data;
+    conf = this->private;
+    if (!conf)
+        goto out;
+
+    defrag = conf->defrag;
+    if (!defrag)
+        goto out;
+
+    frame = create_frame(this, this->ctx->pool);
+    if (!frame)
+        goto out;
+
+    frame->root->pid = GF_CLIENT_PID_DEFRAG;
+
+    defrag->pid = frame->root->pid;
+
+    defrag->defrag_status = GF_DEFRAG_STATUS_STARTED;
+
+    old_THIS = THIS;
+    THIS = this;
+    ret = synctask_new(this->ctx->env, gf_defrag_start_crawl, gf_defrag_done,
+                       frame, this);
+
+    if (ret)
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_REBALANCE_START_FAILED,
+               "Could not create task for rebalance");
+    THIS = old_THIS;
+out:
+    return NULL;
+}
+
+uint64_t
+gf_defrag_get_estimates_based_on_size(dht_conf_t *conf)
+{
+    gf_defrag_info_t *defrag = NULL;
+    double rate_processed = 0;
+    uint64_t total_processed = 0;
+    uint64_t tmp_count = 0;
+    uint64_t time_to_complete = 0;
+    double elapsed = 0;
+
+    defrag = conf->defrag;
+
+    if (!g_totalsize)
+        goto out;
+
+    elapsed = gf_time() - defrag->start_time;
+
+    /* Don't calculate the estimates for the first 10 minutes.
+     * It is unlikely to be accurate and estimates are not required
+     * if the process finishes in less than 10 mins.
+     */
+
+    if (elapsed < ESTIMATE_START_INTERVAL) {
+        gf_msg(THIS->name, GF_LOG_INFO, 0, 0,
+               "Rebalance estimates will not be available for the "
+               "first %d seconds.",
+               ESTIMATE_START_INTERVAL);
+
+        goto out;
+    }
+
+    total_processed = defrag->size_processed;
+
+    /* rate at which files processed */
+    rate_processed = (total_processed) / elapsed;
+
+    tmp_count = g_totalsize;
+
+    if (rate_processed) {
+        time_to_complete = (tmp_count) / rate_processed;
+
+    } else {
+        gf_msg(THIS->name, GF_LOG_ERROR, 0, 0,
+               "Unable to calculate estimated time for rebalance");
+    }
+
+    gf_log(THIS->name, GF_LOG_INFO,
+           "TIME: (size) total_processed=%" PRIu64 " tmp_cnt = %" PRIu64
+           ","
+           "rate_processed=%f, elapsed = %f",
+           total_processed, tmp_count, rate_processed, elapsed);
+
+out:
+    return time_to_complete;
+}
+
+int
+gf_defrag_status_get(dht_conf_t *conf, dict_t *dict)
+{
+    int ret = 0;
+    uint64_t files = 0;
+    uint64_t size = 0;
+    uint64_t lookup = 0;
+    uint64_t failures = 0;
+    uint64_t skipped = 0;
+    char *status = "";
+    double elapsed = 0;
+    uint64_t time_to_complete = 0;
+    uint64_t time_left = 0;
+    gf_defrag_info_t *defrag = conf->defrag;
+
+    if (!defrag)
+        goto out;
+
+    ret = 0;
+    if (defrag->defrag_status == GF_DEFRAG_STATUS_NOT_STARTED)
+        goto out;
+
+    files = defrag->total_files;
+    size = defrag->total_data;
+    lookup = defrag->num_files_lookedup;
+    failures = defrag->total_failures;
+    skipped = defrag->skipped;
+
+    elapsed = gf_time() - defrag->start_time;
+
+    /* The rebalance is still in progress */
+
+    if (defrag->defrag_status == GF_DEFRAG_STATUS_STARTED) {
+        time_to_complete = gf_defrag_get_estimates_based_on_size(conf);
+
+        if (time_to_complete && (time_to_complete > elapsed))
+            time_left = time_to_complete - elapsed;
+
+        gf_log(THIS->name, GF_LOG_INFO,
+               "TIME: Estimated total time to complete (size)= %" PRIu64
+               " seconds, seconds left = %" PRIu64 "",
+               time_to_complete, time_left);
+    }
+
+    if (!dict)
+        goto log;
+
+    ret = dict_set_uint64(dict, "files", files);
+    if (ret)
+        gf_log(THIS->name, GF_LOG_WARNING, "failed to set file count");
+
+    ret = dict_set_uint64(dict, "size", size);
+    if (ret)
+        gf_log(THIS->name, GF_LOG_WARNING, "failed to set size of xfer");
+
+    ret = dict_set_uint64(dict, "lookups", lookup);
+    if (ret)
+        gf_log(THIS->name, GF_LOG_WARNING, "failed to set lookedup file count");
+
+    ret = dict_set_int32(dict, "status", defrag->defrag_status);
+    if (ret)
+        gf_log(THIS->name, GF_LOG_WARNING, "failed to set status");
+
+    ret = dict_set_double(dict, "run-time", elapsed);
+    if (ret)
+        gf_log(THIS->name, GF_LOG_WARNING, "failed to set run-time");
+
+    ret = dict_set_uint64(dict, "failures", failures);
+    if (ret)
+        gf_log(THIS->name, GF_LOG_WARNING, "failed to set failure count");
+
+    ret = dict_set_uint64(dict, "skipped", skipped);
+    if (ret)
+        gf_log(THIS->name, GF_LOG_WARNING, "failed to set skipped file count");
+
+    ret = dict_set_uint64(dict, "time-left", time_left);
+    if (ret)
+        gf_log(THIS->name, GF_LOG_WARNING, "failed to set time-left");
+
+log:
+    switch (defrag->defrag_status) {
+        case GF_DEFRAG_STATUS_NOT_STARTED:
+            status = "not started";
+            break;
+        case GF_DEFRAG_STATUS_STARTED:
+            status = "in progress";
+            break;
+        case GF_DEFRAG_STATUS_STOPPED:
+            status = "stopped";
+            break;
+        case GF_DEFRAG_STATUS_COMPLETE:
+            status = "completed";
+            break;
+        case GF_DEFRAG_STATUS_FAILED:
+            status = "failed";
+            break;
+        default:
+            break;
+    }
+
+    gf_msg(THIS->name, GF_LOG_INFO, 0, DHT_MSG_REBALANCE_STATUS,
+           "Rebalance is %s. Time taken is %.2f secs", status, elapsed);
+    gf_msg(THIS->name, GF_LOG_INFO, 0, DHT_MSG_REBALANCE_STATUS,
+           "Files migrated: %" PRIu64 ", size: %" PRIu64 ", lookups: %" PRIu64
+           ", failures: %" PRIu64
+           ", skipped: "
+           "%" PRIu64,
+           files, size, lookup, failures, skipped);
+out:
+    return 0;
+}
+
+int
+gf_defrag_stop(dht_conf_t *conf, gf_defrag_status_t status, dict_t *output)
+{
+    /* TODO: set a variable 'stop_defrag' here, it should be checked
+       in defrag loop */
+    int ret = -1;
+    gf_defrag_info_t *defrag = conf->defrag;
+
+    GF_ASSERT(defrag);
+
+    if (defrag->defrag_status == GF_DEFRAG_STATUS_NOT_STARTED) {
+        goto out;
+    }
+
+    gf_msg("", GF_LOG_INFO, 0, DHT_MSG_REBALANCE_STOPPED,
+           "Received stop command on rebalance");
+    defrag->defrag_status = status;
+
+    if (output)
+        gf_defrag_status_get(conf, output);
+    ret = 0;
+out:
+    gf_msg_debug("", 0, "Returning %d", ret);
+    return ret;
+}
diff --git a/xlators/cluster/dht/src/dht-rename.c b/xlators/cluster/dht/src/dht-rename.c
index dcdd5c7ec1f..d9dbf50492f 100644
--- a/xlators/cluster/dht/src/dht-rename.c
+++ b/xlators/cluster/dht/src/dht-rename.c
@@ -1,713 +1,1997 @@
 /*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
 /* TODO: link(oldpath, newpath) fails if newpath already exists. DHT should
  *       delete the newpath if it gets EEXISTS from link() call.
  */
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "xlator.h"
 #include "dht-common.h"
-#include "defaults.h"
+#include "dht-lock.h"
+#include <glusterfs/defaults.h>
 
+int
+dht_rename_unlock(call_frame_t *frame, xlator_t *this);
+int32_t
+dht_rename_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *xdata);
 
 int
-dht_rename_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		    int32_t op_ret, int32_t op_errno, struct iatt *stbuf,
-                    struct iatt *preoldparent, struct iatt *postoldparent,
-                    struct iatt *prenewparent, struct iatt *postnewparent)
+dht_rename_unlock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = 0;
-	call_frame_t *prev = NULL;
+    dht_local_t *local = NULL;
 
+    local = frame->local;
 
-	local = frame->local;
-	prev = cookie;
+    dht_set_fixed_dir_stat(&local->preoldparent);
+    dht_set_fixed_dir_stat(&local->postoldparent);
+    dht_set_fixed_dir_stat(&local->preparent);
+    dht_set_fixed_dir_stat(&local->postparent);
 
-	if (op_ret == -1) {
-		/* TODO: undo the damage */
+    if (IA_ISREG(local->stbuf.ia_type))
+        DHT_STRIP_PHASE1_FLAGS(&local->stbuf);
 
-		gf_log (this->name, GF_LOG_DEBUG,
-			"rename %s -> %s on %s failed (%s)",
-			local->loc.path, local->loc2.path,
-			prev->this->name, strerror (op_errno));
+    DHT_STACK_UNWIND(rename, frame, local->op_ret, local->op_errno,
+                     &local->stbuf, &local->preoldparent, &local->postoldparent,
+                     &local->preparent, &local->postparent, local->xattr);
+    return 0;
+}
 
-		local->op_ret   = op_ret;
-		local->op_errno = op_errno;
-	} else {
-		/* TODO: construct proper stbuf for dir */
-                /* 
-                 * FIXME: is this the correct way to build stbuf and
-                 * parent bufs?
-                */
-                dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
-                dht_iatt_merge (this, &local->preoldparent, preoldparent,
-                                prev->this);
-                dht_iatt_merge (this, &local->postoldparent, postoldparent,
-                                prev->this);
-                dht_iatt_merge (this, &local->preparent, prenewparent,
-                                prev->this);
-                dht_iatt_merge (this, &local->postparent, postnewparent,
-                                prev->this);
-	}
+static void
+dht_rename_dir_unlock_src(call_frame_t *frame, xlator_t *this)
+{
+    dht_local_t *local = NULL;
 
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt)) {
-                local->stbuf.ia_ino = local->loc.inode->ino;
+    local = frame->local;
+    dht_unlock_namespace(frame, &local->lock[0]);
+    return;
+}
 
-                local->preoldparent.ia_ino = local->loc.parent->ino;
-                local->postoldparent.ia_ino = local->loc.parent->ino;
+static void
+dht_rename_dir_unlock_dst(call_frame_t *frame, xlator_t *this)
+{
+    dht_local_t *local = NULL;
+    int op_ret = -1;
+    char src_gfid[GF_UUID_BUF_SIZE] = {0};
+    char dst_gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+
+    /* Unlock entrylk */
+    dht_unlock_entrylk_wrapper(frame, &local->lock[1].ns.directory_ns);
+
+    /* Unlock inodelk */
+    op_ret = dht_unlock_inodelk(frame, local->lock[1].ns.parent_layout.locks,
+                                local->lock[1].ns.parent_layout.lk_count,
+                                dht_rename_unlock_cbk);
+    if (op_ret < 0) {
+        uuid_utoa_r(local->loc.inode->gfid, src_gfid);
+
+        if (local->loc2.inode)
+            uuid_utoa_r(local->loc2.inode->gfid, dst_gfid);
+
+        if (IA_ISREG(local->stbuf.ia_type))
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_UNLOCKING_FAILED,
+                   "winding unlock inodelk failed "
+                   "rename (%s:%s:%s %s:%s:%s), "
+                   "stale locks left on bricks",
+                   local->loc.path, src_gfid, local->src_cached->name,
+                   local->loc2.path, dst_gfid,
+                   local->dst_cached ? local->dst_cached->name : NULL);
+        else
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_UNLOCKING_FAILED,
+                   "winding unlock inodelk failed "
+                   "rename (%s:%s %s:%s), "
+                   "stale locks left on bricks",
+                   local->loc.path, src_gfid, local->loc2.path, dst_gfid);
+
+        dht_rename_unlock_cbk(frame, NULL, this, 0, 0, NULL);
+    }
+
+    return;
+}
 
-                local->preparent.ia_ino = local->loc2.parent->ino;
-                local->postparent.ia_ino = local->loc2.parent->ino;
+static int
+dht_rename_dir_unlock(call_frame_t *frame, xlator_t *this)
+{
+    dht_rename_dir_unlock_src(frame, this);
+    dht_rename_dir_unlock_dst(frame, this);
+    return 0;
+}
+int
+dht_rename_dir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *stbuf,
+                   struct iatt *preoldparent, struct iatt *postoldparent,
+                   struct iatt *prenewparent, struct iatt *postnewparent,
+                   dict_t *xdata)
+{
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+    int i = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    int subvol_cnt = -1;
+
+    conf = this->private;
+    local = frame->local;
+    prev = cookie;
+    subvol_cnt = dht_subvol_cnt(this, prev);
+    local->ret_cache[subvol_cnt] = op_ret;
+
+    if (op_ret == -1) {
+        gf_uuid_unparse(local->loc.inode->gfid, gfid);
+
+        gf_msg(this->name, GF_LOG_INFO, op_errno, DHT_MSG_RENAME_FAILED,
+               "Rename %s -> %s on %s failed, (gfid = %s)", local->loc.path,
+               local->loc2.path, prev->name, gfid);
+
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        goto unwind;
+    }
+    /* TODO: construct proper stbuf for dir */
+    /*
+     * FIXME: is this the correct way to build stbuf and
+     * parent bufs?
+     */
+    dht_iatt_merge(this, &local->stbuf, stbuf);
+    dht_iatt_merge(this, &local->preoldparent, preoldparent);
+    dht_iatt_merge(this, &local->postoldparent, postoldparent);
+    dht_iatt_merge(this, &local->preparent, prenewparent);
+    dht_iatt_merge(this, &local->postparent, postnewparent);
 
-                WIPE (&local->preoldparent);
-                WIPE (&local->postoldparent);
-                WIPE (&local->preparent);
-                WIPE (&local->postparent);
+unwind:
+    this_call_cnt = dht_frame_return(frame);
+    if (is_last_call(this_call_cnt)) {
+        /* We get here with local->call_cnt == 0. Which means
+         * we are the only one executing this code, there is
+         * no contention. Therefore it's safe to manipulate or
+         * deref local->call_cnt directly (without locking).
+         */
+        if (local->ret_cache[conf->subvolume_cnt] == 0) {
+            /* count errant subvols in last field of ret_cache */
+            for (i = 0; i < conf->subvolume_cnt; i++) {
+                if (local->ret_cache[i] != 0)
+                    ++local->ret_cache[conf->subvolume_cnt];
+            }
+            if (local->ret_cache[conf->subvolume_cnt]) {
+                /* undoing the damage:
+                 * for all subvolumes, where rename
+                 * succeeded, we perform the reverse operation
+                 */
+                for (i = 0; i < conf->subvolume_cnt; i++) {
+                    if (local->ret_cache[i] == 0)
+                        ++local->call_cnt;
+                }
+                for (i = 0; i < conf->subvolume_cnt; i++) {
+                    if (local->ret_cache[i])
+                        continue;
 
-		DHT_STACK_UNWIND (rename, frame, local->op_ret, local->op_errno,
-				  &local->stbuf, &local->preoldparent,
-                                  &local->postoldparent,
-                                  &local->preparent, &local->postparent);
-	}
+                    STACK_WIND(frame, dht_rename_dir_cbk, conf->subvolumes[i],
+                               conf->subvolumes[i]->fops->rename, &local->loc2,
+                               &local->loc, NULL);
+                }
 
-	return 0;
-}
+                return 0;
+            }
+        }
 
+        WIPE(&local->preoldparent);
+        WIPE(&local->postoldparent);
+        WIPE(&local->preparent);
+        WIPE(&local->postparent);
 
+        dht_rename_dir_unlock(frame, this);
+    }
+
+    return 0;
+}
 
 int
-dht_rename_dir_do (call_frame_t *frame, xlator_t *this)
+dht_rename_hashed_dir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, struct iatt *stbuf,
+                          struct iatt *preoldparent, struct iatt *postoldparent,
+                          struct iatt *prenewparent, struct iatt *postnewparent,
+                          dict_t *xdata)
 {
-	dht_local_t  *local = NULL;
-	dht_conf_t   *conf = NULL;
-	int           i = 0;
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    int call_cnt = 0;
+    xlator_t *prev = NULL;
+    int i = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    conf = this->private;
+    local = frame->local;
+    prev = cookie;
+
+    if (op_ret == -1) {
+        gf_uuid_unparse(local->loc.inode->gfid, gfid);
+
+        gf_msg(this->name, GF_LOG_INFO, op_errno, DHT_MSG_RENAME_FAILED,
+               "rename %s -> %s on %s failed, (gfid = %s) ", local->loc.path,
+               local->loc2.path, prev->name, gfid);
+
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        goto unwind;
+    }
+    /* TODO: construct proper stbuf for dir */
+    /*
+     * FIXME: is this the correct way to build stbuf and
+     * parent bufs?
+     */
+    dht_iatt_merge(this, &local->stbuf, stbuf);
+    dht_iatt_merge(this, &local->preoldparent, preoldparent);
+    dht_iatt_merge(this, &local->postoldparent, postoldparent);
+    dht_iatt_merge(this, &local->preparent, prenewparent);
+    dht_iatt_merge(this, &local->postparent, postnewparent);
+
+    call_cnt = local->call_cnt = conf->subvolume_cnt - 1;
+
+    if (!local->call_cnt)
+        goto unwind;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (conf->subvolumes[i] == local->dst_hashed)
+            continue;
+        STACK_WIND_COOKIE(
+            frame, dht_rename_dir_cbk, conf->subvolumes[i], conf->subvolumes[i],
+            conf->subvolumes[i]->fops->rename, &local->loc, &local->loc2, NULL);
+        if (!--call_cnt)
+            break;
+    }
+
+    return 0;
+unwind:
+    WIPE(&local->preoldparent);
+    WIPE(&local->postoldparent);
+    WIPE(&local->preparent);
+    WIPE(&local->postparent);
 
-	conf = this->private;
-	local = frame->local;
+    dht_rename_dir_unlock(frame, this);
+    return 0;
+}
 
-	if (local->op_ret == -1)
-		goto err;
+int
+dht_rename_dir_do(call_frame_t *frame, xlator_t *this)
+{
+    dht_local_t *local = NULL;
+
+    local = frame->local;
 
-	local->call_cnt = conf->subvolume_cnt;
-	local->op_ret = 0;
+    if (local->op_ret == -1)
+        goto err;
 
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		STACK_WIND (frame, dht_rename_dir_cbk,
-			    conf->subvolumes[i],
-			    conf->subvolumes[i]->fops->rename,
-			    &local->loc, &local->loc2);
-	}
+    local->op_ret = 0;
 
-	return 0;
+    STACK_WIND_COOKIE(frame, dht_rename_hashed_dir_cbk, local->dst_hashed,
+                      local->dst_hashed, local->dst_hashed->fops->rename,
+                      &local->loc, &local->loc2, NULL);
+    return 0;
 
 err:
-	DHT_STACK_UNWIND (rename, frame, local->op_ret, local->op_errno, NULL, NULL,
-                          NULL, NULL, NULL);
-	return 0;
+    dht_rename_dir_unlock(frame, this);
+    return 0;
 }
 
-
 int
-dht_rename_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int op_ret, int op_errno, gf_dirent_t *entries)
+dht_rename_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, gf_dirent_t *entries,
+                       dict_t *xdata)
 {
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = -1;
-	call_frame_t *prev = NULL;
+    dht_local_t *local = NULL;
+    int this_call_cnt = -1;
+    xlator_t *prev = NULL;
 
-	local = frame->local;
-	prev  = cookie;
+    local = frame->local;
+    prev = cookie;
 
-	if (op_ret > 2) {
-		gf_log (this->name, GF_LOG_TRACE,
-			"readdir on %s for %s returned %d entries",
-			prev->this->name, local->loc.path, op_ret);
-		local->op_ret = -1;
-		local->op_errno = ENOTEMPTY;
-	}
+    if (op_ret > 2) {
+        gf_msg_trace(this->name, 0, "readdir on %s for %s returned %d entries",
+                     prev->name, local->loc.path, op_ret);
+        local->op_ret = -1;
+        local->op_errno = ENOTEMPTY;
+    }
 
-	this_call_cnt = dht_frame_return (frame);
+    this_call_cnt = dht_frame_return(frame);
 
-	if (is_last_call (this_call_cnt)) {
-		dht_rename_dir_do (frame, this);
-	}
+    if (is_last_call(this_call_cnt)) {
+        dht_rename_dir_do(frame, this);
+    }
 
-	return 0;
+    return 0;
 }
 
-
 int
-dht_rename_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int op_ret, int op_errno, fd_t *fd)
+dht_rename_opendir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, fd_t *fd, dict_t *xdata)
 {
-	dht_local_t  *local = NULL;
-	int           this_call_cnt = -1;
-	call_frame_t *prev = NULL;
-
+    dht_local_t *local = NULL;
+    int this_call_cnt = -1;
+    xlator_t *prev = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
 
-	local = frame->local;
-	prev  = cookie;
+    local = frame->local;
+    prev = cookie;
 
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"opendir on %s for %s failed (%s)",
-			prev->this->name, local->loc.path,
-			strerror (op_errno));
-		goto err;
-	}
+    if (op_ret == -1) {
+        gf_uuid_unparse(local->loc.inode->gfid, gfid);
+        gf_msg(this->name, GF_LOG_INFO, op_errno, DHT_MSG_OPENDIR_FAILED,
+               "opendir on %s for %s failed,(gfid = %s) ", prev->name,
+               local->loc.path, gfid);
+        goto err;
+    }
 
-	STACK_WIND (frame, dht_rename_readdir_cbk,
-		    prev->this, prev->this->fops->readdir,
-		    local->fd, 4096, 0);
+    fd_bind(fd);
+    STACK_WIND_COOKIE(frame, dht_rename_readdir_cbk, prev, prev,
+                      prev->fops->readdir, local->fd, 4096, 0, NULL);
 
-	return 0;
+    return 0;
 
 err:
-	this_call_cnt = dht_frame_return (frame);
+    this_call_cnt = dht_frame_return(frame);
 
-	if (is_last_call (this_call_cnt)) {
-		dht_rename_dir_do (frame, this);
-	}
+    if (is_last_call(this_call_cnt)) {
+        dht_rename_dir_do(frame, this);
+    }
 
-	return 0;
+    return 0;
 }
 
-
 int
-dht_rename_dir (call_frame_t *frame, xlator_t *this)
+dht_rename_dir_lock2_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	dht_conf_t  *conf = NULL;
-	dht_local_t *local = NULL;
-	int          i = 0;
-	int          op_errno = -1;
+    dht_local_t *local = NULL;
+    char src_gfid[GF_UUID_BUF_SIZE] = {0};
+    char dst_gfid[GF_UUID_BUF_SIZE] = {0};
+    dht_conf_t *conf = NULL;
+    int i = 0;
+
+    local = frame->local;
+    conf = this->private;
+
+    if (op_ret < 0) {
+        uuid_utoa_r(local->loc.inode->gfid, src_gfid);
+
+        if (local->loc2.inode)
+            uuid_utoa_r(local->loc2.inode->gfid, dst_gfid);
+
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_INODE_LK_ERROR,
+               "acquiring entrylk after inodelk failed"
+               "rename (%s:%s:%s %s:%s:%s)",
+               local->loc.path, src_gfid, local->src_cached->name,
+               local->loc2.path, dst_gfid,
+               local->dst_cached ? local->dst_cached->name : NULL);
+
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    local->fd = fd_create(local->loc.inode, frame->root->pid);
+    if (!local->fd) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->op_ret = 0;
+
+    if (!local->dst_cached) {
+        dht_rename_dir_do(frame, this);
+        return 0;
+    }
 
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        STACK_WIND_COOKIE(frame, dht_rename_opendir_cbk, conf->subvolumes[i],
+                          conf->subvolumes[i],
+                          conf->subvolumes[i]->fops->opendir, &local->loc2,
+                          local->fd, NULL);
+    }
 
-	conf = frame->this->private;
-	local = frame->local;
+    return 0;
 
-	local->call_cnt = conf->subvolume_cnt;
+err:
+    /* No harm in calling an extra unlock */
+    dht_rename_dir_unlock(frame, this);
+    return 0;
+}
 
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                if (!conf->subvolume_status[i]) {
-                        op_errno = ENOTCONN;
-                        goto err;
-                }
+int
+dht_rename_dir_lock1_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    char src_gfid[GF_UUID_BUF_SIZE] = {0};
+    char dst_gfid[GF_UUID_BUF_SIZE] = {0};
+    int ret = 0;
+    loc_t *loc = NULL;
+    xlator_t *subvol = NULL;
+
+    local = frame->local;
+
+    if (op_ret < 0) {
+        uuid_utoa_r(local->loc.inode->gfid, src_gfid);
+
+        if (local->loc2.inode)
+            uuid_utoa_r(local->loc2.inode->gfid, dst_gfid);
+
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_INODE_LK_ERROR,
+               "acquiring entrylk after inodelk failed"
+               "rename (%s:%s:%s %s:%s:%s)",
+               local->loc.path, src_gfid, local->src_cached->name,
+               local->loc2.path, dst_gfid,
+               local->dst_cached ? local->dst_cached->name : NULL);
+
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    if (local->current == &local->lock[0]) {
+        loc = &local->loc2;
+        subvol = local->dst_hashed;
+        local->current = &local->lock[1];
+    } else {
+        loc = &local->loc;
+        subvol = local->src_hashed;
+        local->current = &local->lock[0];
+    }
+    ret = dht_protect_namespace(frame, loc, subvol, &local->current->ns,
+                                dht_rename_dir_lock2_cbk);
+    if (ret < 0) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    return 0;
+err:
+    /* No harm in calling an extra unlock */
+    dht_rename_dir_unlock(frame, this);
+    return 0;
+}
+
+/*
+ * If the hashed subvolumes of both source and dst are the different,
+ * lock in dictionary order of hashed subvol->name. This is important
+ * in case the parent directory is the same for both src and dst to
+ * prevent inodelk deadlocks when racing with a fix-layout op on the parent.
+ *
+ * If the hashed subvols are the same, use the gfid/name to determine
+ * the order of taking locks to prevent entrylk deadlocks when the parent
+ * dirs are the same.
+ *
+ */
+static int
+dht_order_rename_lock(call_frame_t *frame, loc_t **loc, xlator_t **subvol)
+{
+    int ret = 0;
+    int op_ret = 0;
+    dht_local_t *local = NULL;
+    char *src = NULL;
+    char *dst = NULL;
+
+    local = frame->local;
+
+    if (local->src_hashed->name == local->dst_hashed->name) {
+        ret = 0;
+    } else {
+        ret = strcmp(local->src_hashed->name, local->dst_hashed->name);
+    }
+
+    if (ret == 0) {
+        /* hashed subvols are the same for src and dst */
+        /* Entrylks need to be ordered*/
+
+        src = alloca(GF_UUID_BNAME_BUF_SIZE + strlen(local->loc.name) + 1);
+        if (!src) {
+            gf_msg(frame->this->name, GF_LOG_ERROR, ENOMEM, 0,
+                   "Insufficient memory for src");
+            op_ret = -1;
+            goto out;
         }
 
-	local->fd = fd_create (local->loc.inode, frame->root->pid);
-	if (!local->fd) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		op_errno = ENOMEM;
-		goto err;
-	}
+        if (!gf_uuid_is_null(local->loc.pargfid))
+            uuid_utoa_r(local->loc.pargfid, src);
+        else if (local->loc.parent)
+            uuid_utoa_r(local->loc.parent->gfid, src);
+        else
+            src[0] = '\0';
+
+        strcat(src, local->loc.name);
+
+        dst = alloca(GF_UUID_BNAME_BUF_SIZE + strlen(local->loc2.name) + 1);
+        if (!dst) {
+            gf_msg(frame->this->name, GF_LOG_ERROR, ENOMEM, 0,
+                   "Insufficient memory for dst");
+            op_ret = -1;
+            goto out;
+        }
 
-	local->op_ret = 0;
+        if (!gf_uuid_is_null(local->loc2.pargfid))
+            uuid_utoa_r(local->loc2.pargfid, dst);
+        else if (local->loc2.parent)
+            uuid_utoa_r(local->loc2.parent->gfid, dst);
+        else
+            dst[0] = '\0';
 
-	if (!local->dst_cached) {
-		dht_rename_dir_do (frame, this);
-		return 0;
-	}
+        strcat(dst, local->loc2.name);
+        ret = strcmp(src, dst);
+    }
 
-	for (i = 0; i < conf->subvolume_cnt; i++) {
-		STACK_WIND (frame, dht_rename_opendir_cbk,
-			    conf->subvolumes[i],
-			    conf->subvolumes[i]->fops->opendir,
-			    &local->loc2, local->fd);
-	}
+    if (ret <= 0) {
+        /*inodelk in dictionary order of hashed subvol names*/
+        /*entrylk in dictionary order of gfid/basename */
+        local->current = &local->lock[0];
+        *loc = &local->loc;
+        *subvol = local->src_hashed;
 
-	return 0;
+    } else {
+        local->current = &local->lock[1];
+        *loc = &local->loc2;
+        *subvol = local->dst_hashed;
+    }
+
+    op_ret = 0;
+
+out:
+    return op_ret;
+}
+
+int
+dht_rename_dir(call_frame_t *frame, xlator_t *this)
+{
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    loc_t *loc = NULL;
+    xlator_t *subvol = NULL;
+    int i = 0;
+    int ret = 0;
+    int op_errno = -1;
+
+    conf = frame->this->private;
+    local = frame->local;
+
+    local->ret_cache = GF_CALLOC(conf->subvolume_cnt + 1, sizeof(int),
+                                 gf_dht_ret_cache_t);
+
+    if (local->ret_cache == NULL) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->call_cnt = conf->subvolume_cnt;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (!conf->subvolume_status[i]) {
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_RENAME_FAILED,
+                   "Rename dir failed: subvolume down (%s)",
+                   conf->subvolumes[i]->name);
+            op_errno = ENOTCONN;
+            goto err;
+        }
+    }
+
+    /* Locks on src and dst needs to ordered which otherwise might cause
+     * deadlocks when rename (src, dst) and rename (dst, src) is done from
+     * two different clients
+     */
+    ret = dht_order_rename_lock(frame, &loc, &subvol);
+    if (ret) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    /* Rename must take locks on src to avoid lookup selfheal from
+     * recreating src on those subvols where the rename was successful.
+     * The locks can't be issued parallel as two different clients might
+     * attempt same rename command and be in dead lock.
+     */
+    ret = dht_protect_namespace(frame, loc, subvol, &local->current->ns,
+                                dht_rename_dir_lock1_cbk);
+    if (ret < 0) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (rename, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
-	return 0;
+    DHT_STACK_UNWIND(rename, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL,
+                     NULL);
+    return 0;
+}
+
+static int
+dht_rename_track_for_changelog(xlator_t *this, dict_t *xattr, loc_t *oldloc,
+                               loc_t *newloc)
+{
+    int ret = -1;
+    dht_changelog_rename_info_t *info = NULL;
+    char *name = NULL;
+    int len1 = 0;
+    int len2 = 0;
+    int size = 0;
+
+    if (!xattr || !oldloc || !newloc || !this)
+        return ret;
+
+    len1 = strlen(oldloc->name) + 1;
+    len2 = strlen(newloc->name) + 1;
+    size = sizeof(dht_changelog_rename_info_t) + len1 + len2;
+
+    info = GF_CALLOC(size, sizeof(char), gf_common_mt_char);
+    if (!info) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+               "Failed to calloc memory");
+        return ret;
+    }
+
+    gf_uuid_copy(info->old_pargfid, oldloc->pargfid);
+    gf_uuid_copy(info->new_pargfid, newloc->pargfid);
+
+    info->oldname_len = len1;
+    info->newname_len = len2;
+    strncpy(info->buffer, oldloc->name, len1);
+    name = info->buffer + len1;
+    strncpy(name, newloc->name, len2);
+
+    ret = dict_set_bin(xattr, DHT_CHANGELOG_RENAME_OP_KEY, info, size);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+               "Failed to set dictionary value: key = %s,"
+               " path = %s",
+               DHT_CHANGELOG_RENAME_OP_KEY, oldloc->name);
+        GF_FREE(info);
+    }
+
+    return ret;
 }
 
+#define DHT_MARKER_DONT_ACCOUNT(xattr)                                         \
+    do {                                                                       \
+        int tmp = -1;                                                          \
+        if (!xattr) {                                                          \
+            xattr = dict_new();                                                \
+            if (!xattr)                                                        \
+                break;                                                         \
+        }                                                                      \
+        tmp = dict_set_str(xattr, GLUSTERFS_MARKER_DONT_ACCOUNT_KEY, "yes");   \
+        if (tmp) {                                                             \
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,       \
+                   "Failed to set dictionary value: key = %s,"                 \
+                   " path = %s",                                               \
+                   GLUSTERFS_MARKER_DONT_ACCOUNT_KEY, local->loc.path);        \
+        }                                                                      \
+    } while (0)
+
+#define DHT_CHANGELOG_TRACK_AS_RENAME(xattr, oldloc, newloc)                   \
+    do {                                                                       \
+        int tmp = -1;                                                          \
+        if (!xattr) {                                                          \
+            xattr = dict_new();                                                \
+            if (!xattr) {                                                      \
+                gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,   \
+                       "Failed to create dictionary to "                       \
+                       "track rename");                                        \
+                break;                                                         \
+            }                                                                  \
+        }                                                                      \
+                                                                               \
+        tmp = dht_rename_track_for_changelog(this, xattr, oldloc, newloc);     \
+                                                                               \
+        if (tmp) {                                                             \
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,       \
+                   "Failed to set dictionary value: key = %s,"                 \
+                   " path = %s",                                               \
+                   DHT_CHANGELOG_RENAME_OP_KEY, (oldloc)->path);               \
+        }                                                                      \
+    } while (0)
 
 int
-dht_rename_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno, struct iatt *preparent,
-                       struct iatt *postparent)
+dht_rename_unlock(call_frame_t *frame, xlator_t *this)
 {
-	dht_local_t  *local = NULL;
-	call_frame_t *prev = NULL;
-	int           this_call_cnt = 0;
+    dht_local_t *local = NULL;
+    int op_ret = -1;
+    char src_gfid[GF_UUID_BUF_SIZE] = {0};
+    char dst_gfid[GF_UUID_BUF_SIZE] = {0};
+    dht_ilock_wrap_t inodelk_wrapper = {
+        0,
+    };
+
+    local = frame->local;
+    inodelk_wrapper.locks = local->rename_inodelk_backward_compatible;
+    inodelk_wrapper.lk_count = local->rename_inodelk_bc_count;
+
+    op_ret = dht_unlock_inodelk_wrapper(frame, &inodelk_wrapper);
+    if (op_ret < 0) {
+        uuid_utoa_r(local->loc.inode->gfid, src_gfid);
+
+        if (local->loc2.inode)
+            uuid_utoa_r(local->loc2.inode->gfid, dst_gfid);
+
+        if (IA_ISREG(local->stbuf.ia_type))
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_UNLOCKING_FAILED,
+                   "winding unlock inodelk failed "
+                   "rename (%s:%s:%s %s:%s:%s), "
+                   "stale locks left on bricks",
+                   local->loc.path, src_gfid, local->src_cached->name,
+                   local->loc2.path, dst_gfid,
+                   local->dst_cached ? local->dst_cached->name : NULL);
+        else
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_UNLOCKING_FAILED,
+                   "winding unlock inodelk failed "
+                   "rename (%s:%s %s:%s), "
+                   "stale locks left on bricks",
+                   local->loc.path, src_gfid, local->loc2.path, dst_gfid);
+    }
+
+    dht_unlock_namespace(frame, &local->lock[0]);
+    dht_unlock_namespace(frame, &local->lock[1]);
+
+    dht_rename_unlock_cbk(frame, NULL, this, local->op_ret, local->op_errno,
+                          NULL);
+    return 0;
+}
 
-	local = frame->local;
-	prev  = cookie;
+int
+dht_rename_done(call_frame_t *frame, xlator_t *this)
+{
+    dht_local_t *local = NULL;
 
-        if (!local) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "!local, should not happen");
-                goto out;
-        }
+    local = frame->local;
 
-	this_call_cnt = dht_frame_return (frame);
+    if (local->linked == _gf_true) {
+        local->linked = _gf_false;
+        dht_linkfile_attr_heal(frame, this);
+    }
 
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"unlink on %s failed (%s)",
-			prev->this->name, strerror (op_errno));
-	}
+    dht_rename_unlock(frame, this);
+    return 0;
+}
 
-        WIPE (&local->preoldparent);
-        WIPE (&local->postoldparent);
-        WIPE (&local->preparent);
-        WIPE (&local->postparent);
+int
+dht_rename_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                      struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    int this_call_cnt = 0;
 
-	if (is_last_call (this_call_cnt)) {
-		DHT_STACK_UNWIND (rename, frame, local->op_ret, local->op_errno,
-				  &local->stbuf, &local->preoldparent,
-                                  &local->postoldparent, &local->preparent,
-                                  &local->postparent);
-        }
+    local = frame->local;
+    prev = cookie;
+
+    FRAME_SU_UNDO(frame, dht_local_t);
+    if (!local) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INVALID_VALUE,
+               "!local, should not happen");
+        goto out;
+    }
+
+    this_call_cnt = dht_frame_return(frame);
+
+    if (op_ret == -1) {
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_UNLINK_FAILED,
+               "%s: Rename: unlink on %s failed ", local->loc.path, prev->name);
+    }
+
+    WIPE(&local->preoldparent);
+    WIPE(&local->postoldparent);
+    WIPE(&local->preparent);
+    WIPE(&local->postparent);
+
+    if (is_last_call(this_call_cnt)) {
+        dht_rename_done(frame, this);
+    }
 
 out:
-	return 0;
+    return 0;
 }
 
-
 int
-dht_rename_cleanup (call_frame_t *frame)
+dht_rename_cleanup(call_frame_t *frame)
 {
-	dht_local_t *local = NULL;
-	xlator_t    *this = NULL;
-	xlator_t    *src_hashed = NULL;
-	xlator_t    *src_cached = NULL;
-	xlator_t    *dst_hashed = NULL;
-	xlator_t    *dst_cached = NULL;
-	int          call_cnt = 0;
+    dht_local_t *local = NULL;
+    xlator_t *this = NULL;
+    xlator_t *src_hashed = NULL;
+    xlator_t *src_cached = NULL;
+    xlator_t *dst_hashed = NULL;
+    xlator_t *dst_cached = NULL;
+    int call_cnt = 0;
+    dict_t *xattr = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
 
+    local = frame->local;
+    this = frame->this;
 
-	local = frame->local;
-	this  = frame->this;
+    src_hashed = local->src_hashed;
+    src_cached = local->src_cached;
+    dst_hashed = local->dst_hashed;
+    dst_cached = local->dst_cached;
 
-	src_hashed = local->src_hashed;
-	src_cached = local->src_cached;
-	dst_hashed = local->dst_hashed;
-	dst_cached = local->dst_cached;
+    if (src_cached == dst_cached)
+        goto nolinks;
 
-	if (src_cached == dst_cached)
-		goto nolinks;
+    if (local->linked && (dst_hashed != src_hashed) &&
+        (dst_hashed != src_cached)) {
+        call_cnt++;
+    }
 
-	if (dst_hashed != src_hashed && dst_hashed != src_cached)
-		call_cnt++;
+    if (local->added_link && (src_cached != dst_hashed)) {
+        call_cnt++;
+    }
 
-	if (src_cached != dst_hashed)
-		call_cnt++;
+    local->call_cnt = call_cnt;
 
-	local->call_cnt = call_cnt;
+    if (!call_cnt)
+        goto nolinks;
 
-        if (!call_cnt)
-                goto nolinks;
+    DHT_MARK_FOP_INTERNAL(xattr);
 
-	if (dst_hashed != src_hashed && dst_hashed != src_cached) {
-		gf_log (this->name, GF_LOG_TRACE,
-			"unlinking linkfile %s @ %s => %s",
-			local->loc.path, dst_hashed->name, src_cached->name);
-                STACK_WIND (frame, dht_rename_unlink_cbk,
-                            dst_hashed, dst_hashed->fops->unlink,
-                            &local->loc);
-	}
+    gf_uuid_unparse(local->loc.inode->gfid, gfid);
 
-	if (src_cached != dst_hashed) {
-		gf_log (this->name, GF_LOG_TRACE,
-			"unlinking link %s => %s (%s)", local->loc.path,
-			local->loc2.path, src_cached->name);
-		STACK_WIND (frame, dht_rename_unlink_cbk,
-			    src_cached, src_cached->fops->unlink,
-			    &local->loc2);
-	}
+    if (local->linked && (dst_hashed != src_hashed) &&
+        (dst_hashed != src_cached)) {
+        dict_t *xattr_new = NULL;
 
-        return 0;
+        gf_msg_trace(this->name, 0,
+                     "unlinking linkfile %s @ %s => %s, (gfid = %s)",
+                     local->loc.path, dst_hashed->name, src_cached->name, gfid);
+
+        xattr_new = dict_copy_with_ref(xattr, NULL);
+
+        DHT_MARKER_DONT_ACCOUNT(xattr_new);
+
+        FRAME_SU_DO(frame, dht_local_t);
+        STACK_WIND_COOKIE(frame, dht_rename_unlink_cbk, dst_hashed, dst_hashed,
+                          dst_hashed->fops->unlink, &local->loc, 0, xattr_new);
+
+        dict_unref(xattr_new);
+        xattr_new = NULL;
+    }
+
+    if (local->added_link && (src_cached != dst_hashed)) {
+        dict_t *xattr_new = NULL;
+
+        gf_msg_trace(this->name, 0, "unlinking link %s => %s (%s), (gfid = %s)",
+                     local->loc.path, local->loc2.path, src_cached->name, gfid);
+
+        xattr_new = dict_copy_with_ref(xattr, NULL);
+
+        if (gf_uuid_compare(local->loc.pargfid, local->loc2.pargfid) == 0) {
+            DHT_MARKER_DONT_ACCOUNT(xattr_new);
+        }
+        /* *
+         * The link to file is created using root permission.
+         * Hence deletion should happen using root. Otherwise
+         * it will fail.
+         */
+        FRAME_SU_DO(frame, dht_local_t);
+        STACK_WIND_COOKIE(frame, dht_rename_unlink_cbk, src_cached, src_cached,
+                          src_cached->fops->unlink, &local->loc2, 0, xattr_new);
+
+        dict_unref(xattr_new);
+        xattr_new = NULL;
+    }
+
+    if (xattr)
+        dict_unref(xattr);
+
+    return 0;
 
 nolinks:
-        WIPE (&local->preoldparent);
-        WIPE (&local->postoldparent);
-        WIPE (&local->preparent);
-        WIPE (&local->postparent);
+    WIPE(&local->preoldparent);
+    WIPE(&local->postoldparent);
+    WIPE(&local->preparent);
+    WIPE(&local->postparent);
 
-	DHT_STACK_UNWIND (rename, frame, local->op_ret, local->op_errno,
-			  &local->stbuf, &local->preoldparent,
-                          &local->postoldparent, &local->preparent,
-                          &local->postparent);
+    dht_rename_unlock(frame, this);
+    return 0;
+}
 
-        return 0;
+int
+dht_rename_unlink(call_frame_t *frame, xlator_t *this)
+{
+    dht_local_t *local = NULL;
+    xlator_t *src_hashed = NULL;
+    xlator_t *src_cached = NULL;
+    xlator_t *dst_hashed = NULL;
+    xlator_t *dst_cached = NULL;
+    xlator_t *rename_subvol = NULL;
+    dict_t *xattr = NULL;
+
+    local = frame->local;
+
+    src_hashed = local->src_hashed;
+    src_cached = local->src_cached;
+    dst_hashed = local->dst_hashed;
+    dst_cached = local->dst_cached;
+
+    local->call_cnt = 0;
+
+    /* NOTE: rename_subvol is the same subvolume from which dht_rename_cbk
+     * is called. since rename has already happened on rename_subvol,
+     * unlink shouldn't be sent for oldpath (either linkfile or cached-file)
+     * on rename_subvol. */
+    if (src_cached == dst_cached)
+        rename_subvol = src_cached;
+    else
+        rename_subvol = dst_hashed;
+
+    /* TODO: delete files in background */
+
+    if (src_cached != dst_hashed && src_cached != dst_cached)
+        local->call_cnt++;
+
+    if (src_hashed != rename_subvol && src_hashed != src_cached)
+        local->call_cnt++;
+
+    if (dst_cached && dst_cached != dst_hashed && dst_cached != src_cached)
+        local->call_cnt++;
+
+    if (local->call_cnt == 0)
+        goto unwind;
+
+    DHT_MARK_FOP_INTERNAL(xattr);
+
+    if (src_cached != dst_hashed && src_cached != dst_cached) {
+        dict_t *xattr_new = NULL;
+
+        xattr_new = dict_copy_with_ref(xattr, NULL);
+
+        gf_msg_trace(this->name, 0, "deleting old src datafile %s @ %s",
+                     local->loc.path, src_cached->name);
+
+        if (gf_uuid_compare(local->loc.pargfid, local->loc2.pargfid) == 0) {
+            DHT_MARKER_DONT_ACCOUNT(xattr_new);
+        }
+
+        DHT_CHANGELOG_TRACK_AS_RENAME(xattr_new, &local->loc, &local->loc2);
+        STACK_WIND_COOKIE(frame, dht_rename_unlink_cbk, src_cached, src_cached,
+                          src_cached->fops->unlink, &local->loc, 0, xattr_new);
+
+        dict_unref(xattr_new);
+        xattr_new = NULL;
+    }
+
+    if (src_hashed != rename_subvol && src_hashed != src_cached) {
+        dict_t *xattr_new = NULL;
+
+        xattr_new = dict_copy_with_ref(xattr, NULL);
+
+        gf_msg_trace(this->name, 0, "deleting old src linkfile %s @ %s",
+                     local->loc.path, src_hashed->name);
+
+        DHT_MARKER_DONT_ACCOUNT(xattr_new);
+
+        STACK_WIND_COOKIE(frame, dht_rename_unlink_cbk, src_hashed, src_hashed,
+                          src_hashed->fops->unlink, &local->loc, 0, xattr_new);
+
+        dict_unref(xattr_new);
+        xattr_new = NULL;
+    }
+
+    if (dst_cached && (dst_cached != dst_hashed) &&
+        (dst_cached != src_cached)) {
+        gf_msg_trace(this->name, 0, "deleting old dst datafile %s @ %s",
+                     local->loc2.path, dst_cached->name);
+
+        STACK_WIND_COOKIE(frame, dht_rename_unlink_cbk, dst_cached, dst_cached,
+                          dst_cached->fops->unlink, &local->loc2, 0, xattr);
+    }
+    if (xattr)
+        dict_unref(xattr);
+    return 0;
+
+unwind:
+    WIPE(&local->preoldparent);
+    WIPE(&local->postoldparent);
+    WIPE(&local->preparent);
+    WIPE(&local->postparent);
+
+    dht_rename_done(frame, this);
+
+    return 0;
 }
 
+int
+dht_rename_links_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int32_t op_ret, int32_t op_errno, inode_t *inode,
+                            struct iatt *stbuf, struct iatt *preparent,
+                            struct iatt *postparent, dict_t *xdata)
+{
+    xlator_t *prev = NULL;
+    dht_local_t *local = NULL;
+    call_frame_t *main_frame = NULL;
+
+    prev = cookie;
+    local = frame->local;
+    main_frame = local->main_frame;
+
+    /* TODO: Handle this case in lookup-optimize */
+    if (op_ret == -1) {
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_CREATE_LINK_FAILED,
+               "link/file %s on %s failed", local->loc.path, prev->name);
+    }
+
+    if (local->linked == _gf_true) {
+        local->linked = _gf_false;
+        dht_linkfile_attr_heal(frame, this);
+    }
+
+    dht_rename_unlink(main_frame, this);
+    DHT_STACK_DESTROY(frame);
+    return 0;
+}
 
 int
-dht_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		int32_t op_ret, int32_t op_errno, struct iatt *stbuf,
-                struct iatt *preoldparent, struct iatt *postoldparent,
-                struct iatt *prenewparent, struct iatt *postnewparent)
-{
-	dht_local_t  *local = NULL;
-	call_frame_t *prev = NULL;
-	xlator_t     *src_hashed = NULL;
-	xlator_t     *src_cached = NULL;
-	xlator_t     *dst_hashed = NULL;
-	xlator_t     *dst_cached = NULL;
-	xlator_t     *rename_subvol = NULL;
-
-	local = frame->local;
-	prev = cookie;
-
-	src_hashed = local->src_hashed;
-	src_cached = local->src_cached;
-	dst_hashed = local->dst_hashed;
-	dst_cached = local->dst_cached;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"rename on %s failed (%s)", prev->this->name,
-			strerror (op_errno));
-		local->op_ret   = op_ret;
-		local->op_errno = op_errno;
-		goto cleanup;
-	}
-
-        dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
-        dht_iatt_merge (this, &local->preoldparent, preoldparent, prev->this);
-        dht_iatt_merge (this, &local->postoldparent, postoldparent, prev->this);
-        dht_iatt_merge (this, &local->preparent, prenewparent, prev->this);
-        dht_iatt_merge (this, &local->postparent, postnewparent, prev->this);
-
-        local->stbuf.ia_ino = local->loc.inode->ino;
-
-        local->preoldparent.ia_ino = local->loc.parent->ino;
-        local->postoldparent.ia_ino = local->loc.parent->ino;
-
-        local->preparent.ia_ino = local->loc2.parent->ino;
-        local->postparent.ia_ino = local->loc2.parent->ino;
-	
-	/* NOTE: rename_subvol is the same subvolume from which dht_rename_cbk
-	 *       is called. since rename has already happened on rename_subvol,
-	 *       unlink should not be sent for oldpath (either linkfile or cached-file)
-	 *       on rename_subvol. */
-	if (src_cached == dst_cached)
-		rename_subvol = src_cached;
-	else
-		rename_subvol = dst_hashed;
-
-	/* TODO: delete files in background */
-
-	if (src_cached != dst_hashed && src_cached != dst_cached)
-		local->call_cnt++;
-
-	if (src_hashed != rename_subvol && src_hashed != src_cached)
-		local->call_cnt++;
-
-	if (dst_cached && dst_cached != dst_hashed && dst_cached != src_cached)
-		local->call_cnt++;
-
-	if (local->call_cnt == 0)
-		goto unwind;
-
-	if (src_cached != dst_hashed && src_cached != dst_cached) {
-		gf_log (this->name, GF_LOG_TRACE,
-			"deleting old src datafile %s @ %s",
-			local->loc.path, src_cached->name);
-
-		STACK_WIND (frame, dht_rename_unlink_cbk,
-			    src_cached, src_cached->fops->unlink,
-			    &local->loc);
-	}
-
-	if (src_hashed != rename_subvol && src_hashed != src_cached) {
-		gf_log (this->name, GF_LOG_TRACE,
-			"deleting old src linkfile %s @ %s",
-			local->loc.path, src_hashed->name);
-
-		STACK_WIND (frame, dht_rename_unlink_cbk,
-			    src_hashed, src_hashed->fops->unlink,
-			    &local->loc);
-	}
-
-	if (dst_cached
-	    && (dst_cached != dst_hashed)
-	    && (dst_cached != src_cached)) {
-		gf_log (this->name, GF_LOG_TRACE,
-			"deleting old dst datafile %s @ %s",
-			local->loc2.path, dst_cached->name);
-
-		STACK_WIND (frame, dht_rename_unlink_cbk,
-			    dst_cached, dst_cached->fops->unlink,
-			    &local->loc2);
-	}
-	return 0;
+dht_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, struct iatt *stbuf,
+               struct iatt *preoldparent, struct iatt *postoldparent,
+               struct iatt *prenewparent, struct iatt *postnewparent,
+               dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *src_cached = NULL;
+    xlator_t *dst_hashed = NULL;
+    xlator_t *dst_cached = NULL;
+    call_frame_t *link_frame = NULL;
+    dht_local_t *link_local = NULL;
+
+    local = frame->local;
+    prev = cookie;
+
+    src_cached = local->src_cached;
+    dst_hashed = local->dst_hashed;
+    dst_cached = local->dst_cached;
+
+    if (local->linked == _gf_true)
+        FRAME_SU_UNDO(frame, dht_local_t);
+
+    /* It is a critical failure iff we fail to rename the cached file
+     * if the rename of the linkto failed, it is not a critical failure,
+     * and we do not want to lose the created hard link for the new
+     * name as that could have been read by other clients.
+     *
+     * NOTE: If another client is attempting the same oldname -> newname
+     * rename, and finds both file names as existing, and are hard links
+     * to each other, then FUSE would send in an unlink for oldname. In
+     * this time duration if we treat the linkto as a critical error and
+     * unlink the newname we created, we would have effectively lost the
+     * file to rename operations.
+     *
+     * Repercussions of treating this as a non-critical error is that
+     * we could leave behind a stale linkto file and/or not create the new
+     * linkto file, the second case would be rectified by a subsequent
+     * lookup, the first case by a rebalance, like for all stale linkto
+     * files */
+
+    if (op_ret == -1) {
+        /* Critical failure: unable to rename the cached file */
+        if (prev == src_cached) {
+            gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_RENAME_FAILED,
+                   "%s: Rename on %s failed, (gfid = %s) ", local->loc.path,
+                   prev->name,
+                   local->loc.inode ? uuid_utoa(local->loc.inode->gfid) : "");
+            local->op_ret = op_ret;
+            local->op_errno = op_errno;
+            goto cleanup;
+        } else {
+            /* Non-critical failure, unable to rename the linkto
+             * file
+             */
+            gf_msg(this->name, GF_LOG_INFO, op_errno, DHT_MSG_RENAME_FAILED,
+                   "%s: Rename (linkto file) on %s failed, "
+                   "(gfid = %s) ",
+                   local->loc.path, prev->name,
+                   local->loc.inode ? uuid_utoa(local->loc.inode->gfid) : "");
+        }
+    }
+    if (xdata) {
+        if (!local->xattr)
+            local->xattr = dict_ref(xdata);
+        else
+            local->xattr = dict_copy_with_ref(xdata, local->xattr);
+    }
+
+    /* Merge attrs only from src_cached. In case there of src_cached !=
+     * dst_hashed, this ignores linkfile attrs. */
+    if (prev == src_cached) {
+        dht_iatt_merge(this, &local->stbuf, stbuf);
+        dht_iatt_merge(this, &local->preoldparent, preoldparent);
+        dht_iatt_merge(this, &local->postoldparent, postoldparent);
+        dht_iatt_merge(this, &local->preparent, prenewparent);
+        dht_iatt_merge(this, &local->postparent, postnewparent);
+    }
+
+    /* Create the linkto file for the dst file */
+    if ((src_cached == dst_cached) && (dst_hashed != dst_cached)) {
+        link_frame = copy_frame(frame);
+        if (!link_frame) {
+            goto unlink;
+        }
 
-unwind:
-        WIPE (&local->preoldparent);
-        WIPE (&local->postoldparent);
-        WIPE (&local->preparent);
-        WIPE (&local->postparent);
+        /* fop value sent as maxvalue because it is not used
+         * anywhere in this case */
+        link_local = dht_local_init(link_frame, &local->loc2, NULL,
+                                    GF_FOP_MAXVALUE);
+        if (!link_local) {
+            goto unlink;
+        }
 
-	DHT_STACK_UNWIND (rename, frame, local->op_ret, local->op_errno,
-			  &local->stbuf, &local->preoldparent,
-                          &local->postoldparent, &local->preparent,
-                          &local->postparent);
+        if (link_local->loc.inode)
+            inode_unref(link_local->loc.inode);
+        link_local->loc.inode = inode_ref(local->loc.inode);
+        link_local->main_frame = frame;
+        link_local->stbuf = local->stbuf;
+        gf_uuid_copy(link_local->gfid, local->loc.inode->gfid);
 
-	return 0;
+        dht_linkfile_create(link_frame, dht_rename_links_create_cbk, this,
+                            src_cached, dst_hashed, &link_local->loc);
+        return 0;
+    }
+
+unlink:
+
+    if (link_frame) {
+        DHT_STACK_DESTROY(link_frame);
+    }
+    dht_rename_unlink(frame, this);
+    return 0;
 
 cleanup:
-        dht_rename_cleanup (frame);
+    dht_rename_cleanup(frame);
 
-        return 0;
+    return 0;
 }
 
+int
+dht_do_rename(call_frame_t *frame)
+{
+    dht_local_t *local = NULL;
+    xlator_t *dst_hashed = NULL;
+    xlator_t *src_cached = NULL;
+    xlator_t *dst_cached = NULL;
+    xlator_t *this = NULL;
+    xlator_t *rename_subvol = NULL;
+
+    local = frame->local;
+    this = frame->this;
+
+    dst_hashed = local->dst_hashed;
+    dst_cached = local->dst_cached;
+    src_cached = local->src_cached;
+
+    if (src_cached == dst_cached)
+        rename_subvol = src_cached;
+    else
+        rename_subvol = dst_hashed;
+
+    if ((src_cached != dst_hashed) && (rename_subvol == dst_hashed)) {
+        DHT_MARKER_DONT_ACCOUNT(local->xattr_req);
+    }
+
+    if (rename_subvol == src_cached) {
+        DHT_CHANGELOG_TRACK_AS_RENAME(local->xattr_req, &local->loc,
+                                      &local->loc2);
+    }
+
+    gf_msg_trace(this->name, 0, "renaming %s => %s (%s)", local->loc.path,
+                 local->loc2.path, rename_subvol->name);
+
+    if (local->linked == _gf_true)
+        FRAME_SU_DO(frame, dht_local_t);
+    STACK_WIND_COOKIE(frame, dht_rename_cbk, rename_subvol, rename_subvol,
+                      rename_subvol->fops->rename, &local->loc, &local->loc2,
+                      local->xattr_req);
+    return 0;
+}
 
 int
-dht_do_rename (call_frame_t *frame)
+dht_rename_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, inode_t *inode,
+                    struct iatt *stbuf, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata)
 {
-	dht_local_t *local = NULL;
-	xlator_t    *dst_hashed = NULL;
-	xlator_t    *src_cached = NULL;
-	xlator_t    *dst_cached = NULL;
-	xlator_t    *this = NULL;
-	xlator_t    *rename_subvol = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
 
+    local = frame->local;
+    prev = cookie;
 
-	local = frame->local;
-	this  = frame->this;
+    if (op_ret == -1) {
+        gf_msg_debug(this->name, 0, "link/file on %s failed (%s)", prev->name,
+                     strerror(op_errno));
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+        local->added_link = _gf_false;
+    } else
+        dht_iatt_merge(this, &local->stbuf, stbuf);
 
-	dst_hashed = local->dst_hashed;
-	dst_cached = local->dst_cached;
-	src_cached = local->src_cached;
+    if (local->op_ret == -1)
+        goto cleanup;
 
-	if (src_cached == dst_cached)
-		rename_subvol = src_cached;
-	else
-		rename_subvol = dst_hashed;
+    dht_do_rename(frame);
 
-	gf_log (this->name, GF_LOG_TRACE,
-		"renaming %s => %s (%s)",
-		local->loc.path, local->loc2.path, rename_subvol->name);
+    return 0;
 
-	STACK_WIND (frame, dht_rename_cbk,
-		    rename_subvol, rename_subvol->fops->rename,
-		    &local->loc, &local->loc2);
+cleanup:
+    dht_rename_cleanup(frame);
 
-	return 0;
+    return 0;
 }
 
-
 int
-dht_rename_links_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno,
-                      inode_t *inode, struct iatt *stbuf,
-                      struct iatt *preparent, struct iatt *postparent)
-{
-	dht_local_t  *local = NULL;
-	call_frame_t *prev = NULL;
-	int           this_call_cnt  = 0;
-
-
-	local = frame->local;
-	prev = cookie;
-	
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"link/file on %s failed (%s)",
-			prev->this->name, strerror (op_errno));
-		local->op_ret   = -1;
-		local->op_errno = op_errno;
-	}
-
-	this_call_cnt = dht_frame_return (frame);
-	if (is_last_call (this_call_cnt)) {
-		if (local->op_ret == -1)
-			goto cleanup;
-		
-		dht_do_rename (frame);
-	}
-
-	return 0;
+dht_rename_linkto_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, inode_t *inode,
+                      struct iatt *stbuf, struct iatt *preparent,
+                      struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *src_cached = NULL;
+    dict_t *xattr = NULL;
+
+    local = frame->local;
+    DHT_MARK_FOP_INTERNAL(xattr);
+    prev = cookie;
+    src_cached = local->src_cached;
+
+    if (op_ret == -1) {
+        gf_msg_debug(this->name, 0, "link/file on %s failed (%s)", prev->name,
+                     strerror(op_errno));
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+    }
+
+    /* If linkto creation failed move to failure cleanup code,
+     * instead of continuing with creating the link file */
+    if (local->op_ret != 0) {
+        goto cleanup;
+    }
+
+    gf_msg_trace(this->name, 0, "link %s => %s (%s)", local->loc.path,
+                 local->loc2.path, src_cached->name);
+    if (gf_uuid_compare(local->loc.pargfid, local->loc2.pargfid) == 0) {
+        DHT_MARKER_DONT_ACCOUNT(xattr);
+    }
+
+    local->added_link = _gf_true;
+
+    STACK_WIND_COOKIE(frame, dht_rename_link_cbk, src_cached, src_cached,
+                      src_cached->fops->link, &local->loc, &local->loc2, xattr);
+
+    if (xattr)
+        dict_unref(xattr);
+
+    return 0;
 
 cleanup:
-        dht_rename_cleanup (frame);
+    dht_rename_cleanup(frame);
 
-	return 0;
+    if (xattr)
+        dict_unref(xattr);
+
+    return 0;
 }
 
+int
+dht_rename_unlink_links_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int32_t op_ret, int32_t op_errno,
+                            struct iatt *preparent, struct iatt *postparent,
+                            dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *prev = NULL;
+
+    local = frame->local;
+    prev = cookie;
+
+    if ((op_ret == -1) && (op_errno != ENOENT)) {
+        gf_msg_debug(this->name, 0, "unlink of %s on %s failed (%s)",
+                     local->loc2.path, prev->name, strerror(op_errno));
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+    }
+
+    if (local->op_ret == -1)
+        goto cleanup;
+
+    dht_do_rename(frame);
+
+    return 0;
+
+cleanup:
+    dht_rename_cleanup(frame);
+
+    return 0;
+}
 
 int
-dht_rename_create_links (call_frame_t *frame)
-{
-	dht_local_t *local = NULL;
-	xlator_t    *this = NULL;
-	xlator_t    *src_hashed = NULL;
-	xlator_t    *src_cached = NULL;
-	xlator_t    *dst_hashed = NULL;
-	xlator_t    *dst_cached = NULL;
-	int          call_cnt = 0;
-
-
-	local = frame->local;
-	this  = frame->this;
-
-	src_hashed = local->src_hashed;
-	src_cached = local->src_cached;
-	dst_hashed = local->dst_hashed;
-	dst_cached = local->dst_cached;
-
-	if (src_cached == dst_cached)
-		goto nolinks;
-
-	if (dst_hashed != src_hashed && dst_hashed != src_cached)
-		call_cnt++;
-
-	if (src_cached != dst_hashed)
-		call_cnt++;
-
-	local->call_cnt = call_cnt;
-
-	if (dst_hashed != src_hashed && dst_hashed != src_cached) {
-		gf_log (this->name, GF_LOG_TRACE,
-			"linkfile %s @ %s => %s",
-			local->loc.path, dst_hashed->name, src_cached->name);
-                memcpy (local->gfid, local->loc.inode->gfid, 16);
-		dht_linkfile_create (frame, dht_rename_links_cbk,
-				     src_cached, dst_hashed, &local->loc);
-	}
-
-	if (src_cached != dst_hashed) {
-		gf_log (this->name, GF_LOG_TRACE,
-			"link %s => %s (%s)", local->loc.path,
-			local->loc2.path, src_cached->name);
-		STACK_WIND (frame, dht_rename_links_cbk,
-			    src_cached, src_cached->fops->link,
-			    &local->loc, &local->loc2);
-	}
+dht_rename_create_links(call_frame_t *frame)
+{
+    dht_local_t *local = NULL;
+    xlator_t *this = NULL;
+    xlator_t *src_hashed = NULL;
+    xlator_t *src_cached = NULL;
+    xlator_t *dst_hashed = NULL;
+    xlator_t *dst_cached = NULL;
+    int call_cnt = 0;
+    dict_t *xattr = NULL;
+
+    local = frame->local;
+    this = frame->this;
+
+    src_hashed = local->src_hashed;
+    src_cached = local->src_cached;
+    dst_hashed = local->dst_hashed;
+    dst_cached = local->dst_cached;
+
+    DHT_MARK_FOP_INTERNAL(xattr);
+
+    if (src_cached == dst_cached) {
+        dict_t *xattr_new = NULL;
+
+        if (dst_hashed == dst_cached)
+            goto nolinks;
+
+        xattr_new = dict_copy_with_ref(xattr, NULL);
+
+        gf_msg_trace(this->name, 0, "unlinking dst linkfile %s @ %s",
+                     local->loc2.path, dst_hashed->name);
+
+        DHT_MARKER_DONT_ACCOUNT(xattr_new);
+
+        STACK_WIND_COOKIE(frame, dht_rename_unlink_links_cbk, dst_hashed,
+                          dst_hashed, dst_hashed->fops->unlink, &local->loc2, 0,
+                          xattr_new);
+
+        dict_unref(xattr_new);
+        if (xattr)
+            dict_unref(xattr);
+
+        return 0;
+    }
+
+    if (src_cached != dst_hashed) {
+        /* needed to create the link file */
+        call_cnt++;
+        if (dst_hashed != src_hashed)
+            /* needed to create the linkto file */
+            call_cnt++;
+    }
+
+    /* We should not have any failures post the link creation, as this
+     * introduces the newname into the namespace. Clients could have cached
+     * the existence of the newname and may start taking actions based on
+     * the same. Hence create the linkto first, and then attempt the link.
+     *
+     * NOTE: If another client is attempting the same oldname -> newname
+     * rename, and finds both file names as existing, and are hard links
+     * to each other, then FUSE would send in an unlink for oldname. In
+     * this time duration if we treat the linkto as a critical error and
+     * unlink the newname we created, we would have effectively lost the
+     * file to rename operations. */
+    if (dst_hashed != src_hashed && src_cached != dst_hashed) {
+        gf_msg_trace(this->name, 0, "linkfile %s @ %s => %s", local->loc.path,
+                     dst_hashed->name, src_cached->name);
+
+        memcpy(local->gfid, local->loc.inode->gfid, 16);
+        dht_linkfile_create(frame, dht_rename_linkto_cbk, this, src_cached,
+                            dst_hashed, &local->loc);
+    } else if (src_cached != dst_hashed) {
+        dict_t *xattr_new = NULL;
+
+        xattr_new = dict_copy_with_ref(xattr, NULL);
+
+        gf_msg_trace(this->name, 0, "link %s => %s (%s)", local->loc.path,
+                     local->loc2.path, src_cached->name);
+        if (gf_uuid_compare(local->loc.pargfid, local->loc2.pargfid) == 0) {
+            DHT_MARKER_DONT_ACCOUNT(xattr_new);
+        }
+
+        local->added_link = _gf_true;
+
+        STACK_WIND_COOKIE(frame, dht_rename_link_cbk, src_cached, src_cached,
+                          src_cached->fops->link, &local->loc, &local->loc2,
+                          xattr_new);
+
+        dict_unref(xattr_new);
+    }
 
 nolinks:
-	if (!call_cnt) {
-		/* skip to next step */
-		dht_do_rename (frame);
-	}
+    if (!call_cnt) {
+        /* skip to next step */
+        dht_do_rename(frame);
+    }
+    if (xattr)
+        dict_unref(xattr);
+
+    return 0;
+}
 
-	return 0;
+int
+dht_rename_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, inode_t *inode,
+                      struct iatt *stbuf, dict_t *xattr,
+                      struct iatt *postparent)
+{
+    dht_local_t *local = NULL;
+    int call_cnt = 0;
+    dht_conf_t *conf = NULL;
+    char gfid_local[GF_UUID_BUF_SIZE] = {0};
+    char gfid_server[GF_UUID_BUF_SIZE] = {0};
+    int child_index = -1;
+    gf_boolean_t is_src = _gf_false;
+    loc_t *loc = NULL;
+
+    child_index = (long)cookie;
+
+    local = frame->local;
+    conf = this->private;
+
+    is_src = (child_index == 0);
+    if (is_src)
+        loc = &local->loc;
+    else
+        loc = &local->loc2;
+
+    if (op_ret >= 0) {
+        if (is_src)
+            local->src_cached = dht_subvol_get_cached(this, local->loc.inode);
+        else {
+            if (loc->inode)
+                gf_uuid_unparse(loc->inode->gfid, gfid_local);
+
+            gf_msg_debug(this->name, 0,
+                         "dst_cached before lookup: %s, "
+                         "(path:%s)(gfid:%s),",
+                         local->loc2.path,
+                         local->dst_cached ? local->dst_cached->name : NULL,
+                         local->dst_cached ? gfid_local : NULL);
+
+            local->dst_cached = dht_subvol_get_cached(this,
+                                                      local->loc2_copy.inode);
+
+            gf_uuid_unparse(stbuf->ia_gfid, gfid_local);
+
+            gf_msg_debug(this->name, GF_LOG_WARNING,
+                         "dst_cached after lookup: %s, "
+                         "(path:%s)(gfid:%s)",
+                         local->loc2.path,
+                         local->dst_cached ? local->dst_cached->name : NULL,
+                         local->dst_cached ? gfid_local : NULL);
+
+            if ((local->loc2.inode == NULL) ||
+                gf_uuid_compare(stbuf->ia_gfid, local->loc2.inode->gfid)) {
+                if (local->loc2.inode != NULL) {
+                    inode_unlink(local->loc2.inode, local->loc2.parent,
+                                 local->loc2.name);
+                    inode_unref(local->loc2.inode);
+                }
+
+                local->loc2.inode = inode_link(local->loc2_copy.inode,
+                                               local->loc2_copy.parent,
+                                               local->loc2_copy.name, stbuf);
+                gf_uuid_copy(local->loc2.gfid, stbuf->ia_gfid);
+            }
+        }
+    }
+
+    if (op_ret < 0) {
+        if (is_src) {
+            /* The meaning of is_linkfile is overloaded here. For locking
+             * to work properly both rebalance and rename should acquire
+             * lock on datafile. The reason for sending this lookup is to
+             * find out whether we've acquired a lock on data file.
+             * Between the lookup before rename and this rename, the
+             * file could be migrated by a rebalance process and now this
+             * file this might be a linkto file. We verify that by sending
+             * this lookup. However, if this lookup fails we cannot really
+             * say whether we've acquired lock on a datafile or linkto file.
+             * So, we act conservatively and _assume_
+             * that this is a linkfile and fail the rename operation.
+             */
+            local->is_linkfile = _gf_true;
+            local->op_errno = op_errno;
+        } else {
+            if (local->dst_cached)
+                gf_msg_debug(this->name, op_errno,
+                             "file %s (gfid:%s) was present "
+                             "(hashed-subvol=%s, "
+                             "cached-subvol=%s) before rename,"
+                             " but lookup failed",
+                             local->loc2.path,
+                             uuid_utoa(local->loc2.inode->gfid),
+                             local->dst_hashed->name, local->dst_cached->name);
+            if (dht_inode_missing(op_errno))
+                local->dst_cached = NULL;
+        }
+    } else if (is_src && xattr &&
+               check_is_linkfile(inode, stbuf, xattr, conf->link_xattr_name)) {
+        local->is_linkfile = _gf_true;
+        /* Found linkto file instead of data file, passdown ENOENT
+         * based on the above comment */
+        local->op_errno = ENOENT;
+    }
+
+    if (!local->is_linkfile && (op_ret >= 0) &&
+        gf_uuid_compare(loc->gfid, stbuf->ia_gfid)) {
+        gf_uuid_unparse(loc->gfid, gfid_local);
+        gf_uuid_unparse(stbuf->ia_gfid, gfid_server);
+
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_GFID_MISMATCH,
+               "path:%s, received a different gfid, local_gfid= %s"
+               " server_gfid: %s",
+               local->loc.path, gfid_local, gfid_server);
+
+        /* Will passdown ENOENT anyway since the file we sent on
+         * rename is replaced with a different file */
+        local->op_errno = ENOENT;
+        /* Since local->is_linkfile is used here to detect failure,
+         * marking this to true */
+        local->is_linkfile = _gf_true;
+    }
+
+    call_cnt = dht_frame_return(frame);
+    if (is_last_call(call_cnt)) {
+        if (local->is_linkfile) {
+            local->op_ret = -1;
+            goto fail;
+        }
+
+        dht_rename_create_links(frame);
+    }
+
+    return 0;
+fail:
+    dht_rename_unlock(frame, this);
+    return 0;
 }
 
+int
+dht_rename_file_lock1_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    char src_gfid[GF_UUID_BUF_SIZE] = {0};
+    char dst_gfid[GF_UUID_BUF_SIZE] = {0};
+    int ret = 0;
+    loc_t *loc = NULL;
+    xlator_t *subvol = NULL;
+
+    local = frame->local;
+
+    if (op_ret < 0) {
+        uuid_utoa_r(local->loc.inode->gfid, src_gfid);
+
+        if (local->loc2.inode)
+            uuid_utoa_r(local->loc2.inode->gfid, dst_gfid);
+
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_INODE_LK_ERROR,
+               "protecting namespace of %s failed"
+               "rename (%s:%s:%s %s:%s:%s)",
+               local->current == &local->lock[0] ? local->loc.path
+                                                 : local->loc2.path,
+               local->loc.path, src_gfid, local->src_hashed->name,
+               local->loc2.path, dst_gfid,
+               local->dst_hashed ? local->dst_hashed->name : NULL);
+
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    if (local->current == &local->lock[0]) {
+        loc = &local->loc2;
+        subvol = local->dst_hashed;
+        local->current = &local->lock[1];
+    } else {
+        loc = &local->loc;
+        subvol = local->src_hashed;
+        local->current = &local->lock[0];
+    }
+
+    ret = dht_protect_namespace(frame, loc, subvol, &local->current->ns,
+                                dht_rename_lock_cbk);
+    if (ret < 0) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    return 0;
+err:
+    /* No harm in calling an extra unlock */
+    dht_rename_unlock(frame, this);
+    return 0;
+}
+
+int32_t
+dht_rename_file_protect_namespace(call_frame_t *frame, void *cookie,
+                                  xlator_t *this, int32_t op_ret,
+                                  int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    char src_gfid[GF_UUID_BUF_SIZE] = {0};
+    char dst_gfid[GF_UUID_BUF_SIZE] = {0};
+    int ret = 0;
+    loc_t *loc = NULL;
+    xlator_t *subvol = NULL;
+
+    local = frame->local;
+
+    if (op_ret < 0) {
+        uuid_utoa_r(local->loc.inode->gfid, src_gfid);
+
+        if (local->loc2.inode)
+            uuid_utoa_r(local->loc2.inode->gfid, dst_gfid);
+
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_INODE_LK_ERROR,
+               "acquiring inodelk failed "
+               "rename (%s:%s:%s %s:%s:%s)",
+               local->loc.path, src_gfid, local->src_cached->name,
+               local->loc2.path, dst_gfid,
+               local->dst_cached ? local->dst_cached->name : NULL);
+
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+
+        goto err;
+    }
+
+    /* Locks on src and dst needs to ordered which otherwise might cause
+     * deadlocks when rename (src, dst) and rename (dst, src) is done from
+     * two different clients
+     */
+    ret = dht_order_rename_lock(frame, &loc, &subvol);
+    if (ret) {
+        local->op_errno = ENOMEM;
+        goto err;
+    }
+
+    ret = dht_protect_namespace(frame, loc, subvol, &local->current->ns,
+                                dht_rename_file_lock1_cbk);
+    if (ret < 0) {
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    return 0;
+
+err:
+    /* Its fine to call unlock even when no locks are acquired, as we check
+     * for lock->locked before winding a unlock call.
+     */
+    dht_rename_unlock(frame, this);
+
+    return 0;
+}
+
+int32_t
+dht_rename_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    char src_gfid[GF_UUID_BUF_SIZE] = {0};
+    char dst_gfid[GF_UUID_BUF_SIZE] = {0};
+    dict_t *xattr_req = NULL;
+    dht_conf_t *conf = NULL;
+    int i = 0;
+    xlator_t *subvol = NULL;
+    dht_lock_t *lock = NULL;
+
+    local = frame->local;
+    conf = this->private;
+
+    if (op_ret < 0) {
+        uuid_utoa_r(local->loc.inode->gfid, src_gfid);
+
+        if (local->loc2.inode)
+            uuid_utoa_r(local->loc2.inode->gfid, dst_gfid);
+
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_INODE_LK_ERROR,
+               "protecting namespace of %s failed. "
+               "rename (%s:%s:%s %s:%s:%s)",
+               local->current == &local->lock[0] ? local->loc.path
+                                                 : local->loc2.path,
+               local->loc.path, src_gfid, local->src_hashed->name,
+               local->loc2.path, dst_gfid,
+               local->dst_hashed ? local->dst_hashed->name : NULL);
+
+        local->op_ret = -1;
+        local->op_errno = op_errno;
+
+        goto done;
+    }
+
+    xattr_req = dict_new();
+    if (xattr_req == NULL) {
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        goto done;
+    }
+
+    op_ret = dict_set_uint32(xattr_req, conf->link_xattr_name, 256);
+    if (op_ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = -op_ret;
+        goto done;
+    }
+
+    /* dst_cached might've changed. This normally happens for two reasons:
+     * 1. rebalance migrated dst
+     * 2. Another parallel rename was done overwriting dst
+     *
+     * Doing a lookup on local->loc2 when dst exists, but is associated
+     * with a different gfid will result in an ESTALE error. So, do a fresh
+     * lookup with a new inode on dst-path and handle change of dst-cached
+     * in the cbk. Also, to identify dst-cached changes we do a lookup on
+     * "this" rather than the subvol.
+     */
+    loc_copy(&local->loc2_copy, &local->loc2);
+    inode_unref(local->loc2_copy.inode);
+    local->loc2_copy.inode = inode_new(local->loc.inode->table);
+
+    /* Why not use local->lock.locks[?].loc for lookup post lock phase
+     * ---------------------------------------------------------------
+     * "layout.parent_layout.locks[?].loc" does not have the name and pargfid
+     * populated.
+     * Reason: If we had populated the name and pargfid, server might
+     * resolve to a successful lookup even if there is a file with same name
+     * with a different gfid(unlink & create) as server does name based
+     * resolution on first priority. And this can result in operating on a
+     * different inode entirely.
+     *
+     * Now consider a scenario where source file was renamed by some other
+     * client to a new name just before this lock was granted. So if a
+     * lookup would be done on local->lock[0].layout.parent_layout.locks[?].loc,
+     * server will send success even if the entry was renamed (since server will
+     * do a gfid based resolution). So once a lock is granted, make sure the
+     * file exists with the name that the client requested with.
+     * */
+
+    local->call_cnt = 2;
+    for (i = 0; i < 2; i++) {
+        if (i == 0) {
+            lock = local->rename_inodelk_backward_compatible[0];
+            if (gf_uuid_compare(local->loc.gfid, lock->loc.gfid) == 0)
+                subvol = lock->xl;
+            else {
+                lock = local->rename_inodelk_backward_compatible[1];
+                subvol = lock->xl;
+            }
+        } else {
+            subvol = this;
+        }
+
+        STACK_WIND_COOKIE(frame, dht_rename_lookup_cbk, (void *)(long)i, subvol,
+                          subvol->fops->lookup,
+                          (i == 0) ? &local->loc : &local->loc2_copy,
+                          xattr_req);
+    }
+
+    dict_unref(xattr_req);
+    return 0;
+
+done:
+    /* Its fine to call unlock even when no locks are acquired, as we check
+     * for lock->locked before winding a unlock call.
+     */
+    dht_rename_unlock(frame, this);
+
+    if (xattr_req)
+        dict_unref(xattr_req);
+
+    return 0;
+}
+
+int
+dht_rename_lock(call_frame_t *frame)
+{
+    dht_local_t *local = NULL;
+    int count = 1, ret = -1;
+    dht_lock_t **lk_array = NULL;
+
+    local = frame->local;
+
+    if (local->dst_cached)
+        count++;
+
+    lk_array = GF_CALLOC(count, sizeof(*lk_array), gf_common_mt_pointer);
+    if (lk_array == NULL)
+        goto err;
+
+    lk_array[0] = dht_lock_new(frame->this, local->src_cached, &local->loc,
+                               F_WRLCK, DHT_FILE_MIGRATE_DOMAIN, NULL,
+                               FAIL_ON_ANY_ERROR);
+    if (lk_array[0] == NULL)
+        goto err;
+
+    if (local->dst_cached) {
+        /* dst might be removed by the time inodelk reaches bricks,
+         * which can result in ESTALE errors. POSIX imposes no
+         * restriction for dst to be present for renames to be
+         * successful. So, we'll ignore ESTALE errors. As far as
+         * synchronization on dst goes, we'll achieve the same by
+         * holding entrylk on parent directory of dst in the namespace
+         * of basename(dst). Also, there might not be quorum in cluster
+         * xlators like EC/disperse on errno, in which case they return
+         * EIO. For eg., in a disperse (4 + 2), 3 might return success
+         * and three might return ESTALE. Disperse, having no Quorum
+         * unwinds inodelk with EIO. So, ignore EIO too.
+         */
+        lk_array[1] = dht_lock_new(frame->this, local->dst_cached, &local->loc2,
+                                   F_WRLCK, DHT_FILE_MIGRATE_DOMAIN, NULL,
+                                   IGNORE_ENOENT_ESTALE_EIO);
+        if (lk_array[1] == NULL)
+            goto err;
+    }
+
+    local->rename_inodelk_backward_compatible = lk_array;
+    local->rename_inodelk_bc_count = count;
+
+    /* retaining inodelks for the sake of backward compatibility. Please
+     * make sure to remove this inodelk once all of 3.10, 3.12 and 3.13
+     * reach EOL. Better way of getting synchronization would be to acquire
+     * entrylks on src and dst parent directories in the namespace of
+     * basenames of src and dst
+     */
+    ret = dht_blocking_inodelk(frame, lk_array, count,
+                               dht_rename_file_protect_namespace);
+    if (ret < 0) {
+        local->rename_inodelk_backward_compatible = NULL;
+        local->rename_inodelk_bc_count = 0;
+        goto err;
+    }
+
+    return 0;
+err:
+    if (lk_array != NULL) {
+        int tmp_count = 0, i = 0;
+
+        for (i = 0; (i < count) && (lk_array[i]); i++, tmp_count++)
+            ;
+
+        dht_lock_array_free(lk_array, tmp_count);
+        GF_FREE(lk_array);
+    }
+
+    return -1;
+}
 
 int
-dht_rename (call_frame_t *frame, xlator_t *this,
-	    loc_t *oldloc, loc_t *newloc)
-{
-	xlator_t    *src_cached = NULL;
-	xlator_t    *src_hashed = NULL;
-	xlator_t    *dst_cached = NULL;
-	xlator_t    *dst_hashed = NULL;
-	int          op_errno = -1;
-	int          ret = -1;
-	dht_local_t *local = NULL;
-
-
-	VALIDATE_OR_GOTO (frame, err);
-	VALIDATE_OR_GOTO (this, err);
-	VALIDATE_OR_GOTO (oldloc, err);
-	VALIDATE_OR_GOTO (newloc, err);
-
-	src_hashed = dht_subvol_get_hashed (this, oldloc);
-	if (!src_hashed) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no subvolume in layout for path=%s",
-			oldloc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	src_cached = dht_subvol_get_cached (this, oldloc->inode);
-	if (!src_cached) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no cached subvolume for path=%s", oldloc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	dst_hashed = dht_subvol_get_hashed (this, newloc);
-	if (!dst_hashed) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no subvolume in layout for path=%s",
-			newloc->path);
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	if (newloc->inode)
-		dst_cached = dht_subvol_get_cached (this, newloc->inode);
-
-	local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	ret = loc_copy (&local->loc, oldloc);
-	if (ret == -1) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	ret = loc_copy (&local->loc2, newloc);
-	if (ret == -1) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	local->src_hashed = src_hashed;
-	local->src_cached = src_cached;
-	local->dst_hashed = dst_hashed;
-	local->dst_cached = dst_cached;
-
-	gf_log (this->name, GF_LOG_TRACE,
-		"renaming %s (hash=%s/cache=%s) => %s (hash=%s/cache=%s)",
-		oldloc->path, src_hashed->name, src_cached->name,
-		newloc->path, dst_hashed->name,
-		dst_cached ? dst_cached->name : "<nul>");
-
-	if (IA_ISDIR (oldloc->inode->ia_type)) {
-		dht_rename_dir (frame, this);
-	} else {
-		local->op_ret = 0;
-		dht_rename_create_links (frame);
-	}
-
-	return 0;
+dht_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+           dict_t *xdata)
+{
+    xlator_t *src_cached = NULL;
+    xlator_t *src_hashed = NULL;
+    xlator_t *dst_cached = NULL;
+    xlator_t *dst_hashed = NULL;
+    int op_errno = -1;
+    int ret = -1;
+    dht_local_t *local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    char newgfid[GF_UUID_BUF_SIZE] = {0};
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(oldloc, err);
+    VALIDATE_OR_GOTO(newloc, err);
+
+    gf_uuid_unparse(oldloc->inode->gfid, gfid);
+
+    src_hashed = dht_subvol_get_hashed(this, oldloc);
+    if (!src_hashed) {
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_RENAME_FAILED,
+               "No hashed subvolume in layout for path=%s,"
+               "(gfid = %s)",
+               oldloc->path, gfid);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    src_cached = dht_subvol_get_cached(this, oldloc->inode);
+    if (!src_cached) {
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_RENAME_FAILED,
+               "No cached subvolume for path = %s,"
+               "(gfid = %s)",
+               oldloc->path, gfid);
+
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    dst_hashed = dht_subvol_get_hashed(this, newloc);
+    if (!dst_hashed) {
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_RENAME_FAILED,
+               "No hashed subvolume in layout for path=%s", newloc->path);
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    if (newloc->inode)
+        dst_cached = dht_subvol_get_cached(this, newloc->inode);
+
+    local = dht_local_init(frame, oldloc, NULL, GF_FOP_RENAME);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+    /* cached_subvol will be set from dht_local_init, reset it to NULL,
+       as the logic of handling rename is different  */
+    local->cached_subvol = NULL;
+
+    ret = loc_copy(&local->loc2, newloc);
+    if (ret == -1) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->src_hashed = src_hashed;
+    local->src_cached = src_cached;
+    local->dst_hashed = dst_hashed;
+    local->dst_cached = dst_cached;
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    if (newloc->inode)
+        gf_uuid_unparse(newloc->inode->gfid, newgfid);
+
+    gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_RENAME_INFO,
+           "renaming %s (%s) (hash=%s/cache=%s) => %s (%s) "
+           "(hash=%s/cache=%s) ",
+           oldloc->path, gfid, src_hashed->name, src_cached->name, newloc->path,
+           newloc->inode ? newgfid : NULL, dst_hashed->name,
+           dst_cached ? dst_cached->name : "<nul>");
+
+    if (IA_ISDIR(oldloc->inode->ia_type)) {
+        dht_rename_dir(frame, this);
+    } else {
+        local->op_ret = 0;
+        ret = dht_rename_lock(frame);
+        if (ret < 0) {
+            op_errno = ENOMEM;
+            goto err;
+        }
+    }
+
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (rename, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(rename, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL,
+                     NULL);
 
-	return 0;
+    return 0;
+}
+
+int
+dht_pt_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+              dict_t *xdata)
+{
+    gf_boolean_t free_xdata = _gf_false;
+
+    /* Just a pass through */
+    if (!IA_ISDIR(oldloc->inode->ia_type)) {
+        if (!xdata) {
+            free_xdata = _gf_true;
+        }
+        DHT_CHANGELOG_TRACK_AS_RENAME(xdata, oldloc, newloc);
+    }
+    default_rename(frame, this, oldloc, newloc, xdata);
+    if (free_xdata && xdata) {
+        dict_unref(xdata);
+        xdata = NULL;
+    }
+    return 0;
 }
diff --git a/xlators/cluster/dht/src/dht-selfheal.c b/xlators/cluster/dht/src/dht-selfheal.c
index c12847f581d..3e24065227c 100644
--- a/xlators/cluster/dht/src/dht-selfheal.c
+++ b/xlators/cluster/dht/src/dht-selfheal.c
@@ -1,558 +1,2600 @@
 /*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
 
+#include "dht-lock.h"
+
+#define DHT_SET_LAYOUT_RANGE(layout, i, srt, chunk, path)                      \
+    do {                                                                       \
+        layout->list[i].start = srt;                                           \
+        layout->list[i].stop = srt + chunk - 1;                                \
+        layout->list[i].commit_hash = layout->commit_hash;                     \
+                                                                               \
+        gf_msg_trace(this->name, 0,                                            \
+                     "gave fix: 0x%x - 0x%x, with commit-hash 0x%x"            \
+                     " on %s for %s",                                          \
+                     layout->list[i].start, layout->list[i].stop,              \
+                     layout->list[i].commit_hash,                              \
+                     layout->list[i].xlator->name, path);                      \
+    } while (0)
+
+#define DHT_RESET_LAYOUT_RANGE(layout)                                         \
+    do {                                                                       \
+        int cnt = 0;                                                           \
+        for (cnt = 0; cnt < layout->cnt; cnt++) {                              \
+            layout->list[cnt].start = 0;                                       \
+            layout->list[cnt].stop = 0;                                        \
+        }                                                                      \
+    } while (0)
+
+static int
+dht_selfheal_layout_lock(call_frame_t *frame, dht_layout_t *layout,
+                         gf_boolean_t newdir, dht_selfheal_layout_t healer,
+                         dht_need_heal_t should_heal);
+
+static uint32_t
+dht_overlap_calc(dht_layout_t *old, int o, dht_layout_t *new, int n)
+{
+    if (o >= old->cnt || n >= new->cnt)
+        return 0;
 
-#include "glusterfs.h"
-#include "xlator.h"
-#include "dht-common.h"
+    if (old->list[o].err > 0 || new->list[n].err > 0)
+        return 0;
 
+    if (old->list[o].start == old->list[o].stop) {
+        return 0;
+    }
 
-int
-dht_selfheal_dir_finish (call_frame_t *frame, xlator_t *this, int ret)
-{
-	dht_local_t  *local = NULL;
+    if (new->list[n].start == new->list[n].stop) {
+        return 0;
+    }
 
+    if ((old->list[o].start > new->list[n].stop) ||
+        (old->list[o].stop < new->list[n].start))
+        return 0;
 
-	local = frame->local;
-	local->selfheal.dir_cbk (frame, NULL, frame->this, ret,
-				 local->op_errno);
+    return min(old->list[o].stop, new->list[n].stop) -
+           max(old->list[o].start, new->list[n].start) + 1;
+}
 
-	return 0;
+int
+dht_selfheal_unlock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    DHT_STACK_DESTROY(frame);
+    return 0;
 }
 
+int
+dht_selfheal_dir_finish(call_frame_t *frame, xlator_t *this, int ret,
+                        int invoke_cbk)
+{
+    dht_local_t *local = NULL, *lock_local = NULL;
+    call_frame_t *lock_frame = NULL;
+    int lock_count = 0;
+
+    local = frame->local;
+
+    /* Unlock entrylk */
+    dht_unlock_entrylk_wrapper(frame, &local->lock[0].ns.directory_ns);
+
+    /* Unlock inodelk */
+    lock_count = dht_lock_count(local->lock[0].ns.parent_layout.locks,
+                                local->lock[0].ns.parent_layout.lk_count);
+    if (lock_count == 0)
+        goto done;
+
+    lock_frame = copy_frame(frame);
+    if (lock_frame == NULL) {
+        goto done;
+    }
+
+    lock_local = dht_local_init(lock_frame, &local->loc, NULL,
+                                lock_frame->root->op);
+    if (lock_local == NULL) {
+        goto done;
+    }
+
+    lock_local->lock[0].ns.parent_layout.locks = local->lock[0]
+                                                     .ns.parent_layout.locks;
+    lock_local->lock[0]
+        .ns.parent_layout.lk_count = local->lock[0].ns.parent_layout.lk_count;
+
+    local->lock[0].ns.parent_layout.locks = NULL;
+    local->lock[0].ns.parent_layout.lk_count = 0;
+
+    dht_unlock_inodelk(lock_frame, lock_local->lock[0].ns.parent_layout.locks,
+                       lock_local->lock[0].ns.parent_layout.lk_count,
+                       dht_selfheal_unlock_cbk);
+    lock_frame = NULL;
+
+done:
+    if (invoke_cbk)
+        local->selfheal.dir_cbk(frame, NULL, frame->this, ret, local->op_errno,
+                                NULL);
+    if (lock_frame != NULL) {
+        DHT_STACK_DESTROY(lock_frame);
+    }
+
+    return 0;
+}
 
 int
-dht_selfheal_dir_xattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			    int op_ret, int op_errno)
+dht_refresh_layout_done(call_frame_t *frame)
 {
-	dht_local_t  *local = NULL;
-	call_frame_t *prev = NULL;
-	xlator_t     *subvol = NULL;
-	int           i = 0;
-	dht_layout_t *layout = NULL;
-	int           err = 0;
-	int           this_call_cnt = 0;
+    int ret = -1;
+    dht_layout_t *refreshed = NULL, *heal = NULL;
+    dht_local_t *local = NULL;
+    dht_need_heal_t should_heal = NULL;
+    dht_selfheal_layout_t healer = NULL;
 
-	local = frame->local;
-	layout = local->selfheal.layout;
-	prev = cookie;
-	subvol = prev->this;
+    local = frame->local;
 
-	if (op_ret == 0)
-		err = 0;
-	else
-		err = op_errno;
+    refreshed = local->selfheal.refreshed_layout;
+    heal = local->selfheal.layout;
 
-	for (i = 0; i < layout->cnt; i++) {
-		if (layout->list[i].xlator == subvol) {
-			layout->list[i].err = err;
-			break;
-		}
-	}
+    healer = local->selfheal.healer;
+    should_heal = local->selfheal.should_heal;
 
-	this_call_cnt = dht_frame_return (frame);
+    ret = dht_layout_sort(refreshed);
+    if (ret == -1) {
+        gf_smsg(frame->this->name, GF_LOG_WARNING, 0,
+                DHT_MSG_LAYOUT_SORT_FAILED, NULL);
+        goto err;
+    }
 
-	if (is_last_call (this_call_cnt)) {
-		dht_selfheal_dir_finish (frame, this, 0);
-	}
+    if (should_heal(frame, &heal, &refreshed)) {
+        healer(frame, &local->loc, heal);
+    } else {
+        local->selfheal.layout = NULL;
+        local->selfheal.refreshed_layout = NULL;
+        local->selfheal.layout = refreshed;
 
-	return 0;
+        dht_layout_unref(frame->this, heal);
+
+        dht_selfheal_dir_finish(frame, frame->this, 0, 1);
+    }
+
+    return 0;
+
+err:
+    dht_selfheal_dir_finish(frame, frame->this, -1, 1);
+    return 0;
 }
 
+int
+dht_refresh_layout_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, inode_t *inode,
+                       struct iatt *stbuf, dict_t *xattr,
+                       struct iatt *postparent)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+    xlator_t *prev = NULL;
+    dht_layout_t *layout = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO("dht", this, err);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, err);
+    GF_VALIDATE_OR_GOTO("dht", this->private, err);
+
+    local = frame->local;
+    prev = cookie;
+
+    layout = local->selfheal.refreshed_layout;
+
+    LOCK(&frame->lock);
+    {
+        op_ret = dht_layout_merge(this, layout, prev, op_ret, op_errno, xattr);
+
+        dht_iatt_merge(this, &local->stbuf, stbuf);
+
+        if (op_ret == -1) {
+            gf_uuid_unparse(local->loc.gfid, gfid);
+            local->op_errno = op_errno;
+            gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                    DHT_MSG_FILE_LOOKUP_FAILED, "path=%s", local->loc.path,
+                    "name=%s", prev->name, "gfid=%s", gfid, NULL);
+
+            goto unlock;
+        }
+
+        local->op_ret = 0;
+    }
+unlock:
+    UNLOCK(&frame->lock);
+
+    this_call_cnt = dht_frame_return(frame);
+
+    if (is_last_call(this_call_cnt)) {
+        if (local->op_ret == 0) {
+            local->refresh_layout_done(frame);
+        } else {
+            goto err;
+        }
+    }
+
+    return 0;
+
+err:
+    if (local) {
+        local->refresh_layout_unlock(frame, this, -1, 1);
+    }
+    return 0;
+}
 
 int
-dht_selfheal_dir_xattr_persubvol (call_frame_t *frame, loc_t *loc,
-				  dht_layout_t *layout, int i)
+dht_refresh_layout(call_frame_t *frame)
 {
-	xlator_t          *subvol = NULL;
-	dict_t            *xattr = NULL;
-	int                ret = 0;
-	xlator_t          *this = NULL;
-	int32_t           *disk_layout = NULL;
+    int call_cnt = 0;
+    int i = 0, ret = -1;
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *this = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+
+    this = frame->this;
+    conf = this->private;
+    local = frame->local;
+
+    call_cnt = conf->subvolume_cnt;
+    local->call_cnt = call_cnt;
+    local->op_ret = -1;
+
+    if (local->selfheal.refreshed_layout) {
+        dht_layout_unref(this, local->selfheal.refreshed_layout);
+        local->selfheal.refreshed_layout = NULL;
+    }
+
+    local->selfheal.refreshed_layout = dht_layout_new(this,
+                                                      conf->subvolume_cnt);
+    if (!local->selfheal.refreshed_layout) {
+        gf_uuid_unparse(local->loc.gfid, gfid);
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MEM_ALLOC_FAILED,
+                "path=%s", local->loc.path, "gfid=%s", gfid, NULL);
+        goto out;
+    }
+
+    if (local->xattr != NULL) {
+        dict_del(local->xattr, conf->xattr_name);
+    }
+
+    if (local->xattr_req == NULL) {
+        gf_uuid_unparse(local->loc.gfid, gfid);
+        local->xattr_req = dict_new();
+        if (local->xattr_req == NULL) {
+            gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+                    "path=%s", local->loc.path, "gfid=%s", gfid, NULL);
+            goto out;
+        }
+    }
 
+    if (dict_get(local->xattr_req, conf->xattr_name) == 0) {
+        ret = dict_set_uint32(local->xattr_req, conf->xattr_name, 4 * 4);
+        if (ret)
+            gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+                    "path=%s", local->loc.path, "key=%s", conf->xattr_name,
+                    NULL);
+    }
 
-	subvol = layout->list[i].xlator;
-	this = frame->this;
+    for (i = 0; i < call_cnt; i++) {
+        STACK_WIND_COOKIE(frame, dht_refresh_layout_cbk, conf->subvolumes[i],
+                          conf->subvolumes[i],
+                          conf->subvolumes[i]->fops->lookup, &local->loc,
+                          local->xattr_req);
+    }
 
-	xattr = get_new_dict ();
-	if (!xattr) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
+    return 0;
 
-	ret = dht_disk_layout_extract (this, layout, i, &disk_layout);
-	if (ret == -1) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"failed to extract disk layout");
-		goto err;
-	}
+out:
+    if (local) {
+        local->refresh_layout_unlock(frame, this, -1, 1);
+    }
+    return 0;
+}
 
-	ret = dict_set_bin (xattr, "trusted.glusterfs.dht",
-			    disk_layout, 4 * 4);
-	if (ret == -1) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"failed to set xattr dictionary");
-		goto err;
-	}
-	disk_layout = NULL;
+int32_t
+dht_selfheal_layout_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                             int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
 
-	gf_log (this->name, GF_LOG_TRACE,
-		"setting hash range %u - %u (type %d) on subvolume %s for %s",
-		layout->list[i].start, layout->list[i].stop,
-		layout->type, subvol->name, loc->path);
+    local = frame->local;
 
-	dict_ref (xattr);
+    if (!local) {
+        goto err;
+    }
 
-	STACK_WIND (frame, dht_selfheal_dir_xattr_cbk,
-		    subvol, subvol->fops->setxattr,
-		    loc, xattr, 0);
+    if (op_ret < 0) {
+        local->op_errno = op_errno;
+        goto err;
+    }
 
-	dict_unref (xattr);
+    local->refresh_layout_unlock = dht_selfheal_dir_finish;
+    local->refresh_layout_done = dht_refresh_layout_done;
 
-	return 0;
+    dht_refresh_layout(frame);
+    return 0;
 
 err:
-	if (xattr)
-		dict_destroy (xattr);
+    dht_selfheal_dir_finish(frame, this, -1, 1);
+    return 0;
+}
 
-	if (disk_layout)
-		GF_FREE (disk_layout);
+gf_boolean_t
+dht_should_heal_layout(call_frame_t *frame, dht_layout_t **heal,
+                       dht_layout_t **ondisk)
+{
+    gf_boolean_t fixit = _gf_true;
+    dht_local_t *local = NULL;
+    int heal_missing_dirs = 0;
+
+    local = frame->local;
+
+    if ((heal == NULL) || (*heal == NULL) || (ondisk == NULL) ||
+        (*ondisk == NULL))
+        goto out;
+
+    dht_layout_anomalies(
+        frame->this, &local->loc, *ondisk, &local->selfheal.hole_cnt,
+        &local->selfheal.overlaps_cnt, &local->selfheal.missing_cnt,
+        &local->selfheal.down, &local->selfheal.misc, NULL);
+
+    /* Directories might've been created as part of this self-heal. We've to
+     * sync non-layout xattrs and set range 0-0 on new directories
+     */
+    heal_missing_dirs = local->selfheal.force_mkdir
+                            ? local->selfheal.force_mkdir
+                            : dht_layout_missing_dirs(*heal);
+
+    if ((local->selfheal.hole_cnt == 0) &&
+        (local->selfheal.overlaps_cnt == 0) && heal_missing_dirs) {
+        dht_layout_t *tmp = NULL;
+
+        /* Just added a brick and need to set 0-0 range on this brick.
+         * But ondisk layout is well-formed. So, swap layouts "heal" and
+         * "ondisk". Now "ondisk" layout will be used for healing
+         * xattrs. If there are any non-participating subvols in
+         * "ondisk" layout, dht_selfheal_dir_xattr_persubvol will set
+         * 0-0 and non-layout xattrs. This way we won't end up in
+         * "corrupting" already set and well-formed "ondisk" layout.
+         */
+        tmp = *heal;
+        *heal = *ondisk;
+        *ondisk = tmp;
+
+        /* Current selfheal code, heals non-layout xattrs only after
+         * an add-brick. In fact non-layout xattrs are considered as
+         * secondary citizens which are healed only if layout xattrs
+         * need to be healed. This is wrong, since for eg., quota can be
+         * set when layout is well-formed, but a node is down. Also,
+         * just for healing non-layout xattrs, we don't need locking.
+         * This issue is _NOT FIXED_ by this patch.
+         */
+    }
+
+    fixit = (local->selfheal.hole_cnt || local->selfheal.overlaps_cnt ||
+             heal_missing_dirs);
 
-	dht_selfheal_dir_xattr_cbk (frame, subvol, frame->this,
-				    -1, ENOMEM);
-	return 0;
+out:
+    return fixit;
 }
 
+int
+dht_layout_span(dht_layout_t *layout)
+{
+    int i = 0, count = 0;
+
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].err)
+            continue;
+
+        if (layout->list[i].start != layout->list[i].stop)
+            count++;
+    }
+
+    return count;
+}
 
 int
-dht_selfheal_dir_xattr (call_frame_t *frame, loc_t *loc, dht_layout_t *layout)
+dht_decommissioned_bricks_in_layout(xlator_t *this, dht_layout_t *layout)
 {
-	dht_local_t *local = NULL;
-	int          missing_xattr = 0;
-	int          i = 0;
-	xlator_t    *this = NULL;
+    dht_conf_t *conf = NULL;
+    int count = 0, i = 0, j = 0;
 
-	local = frame->local;
-	this = frame->this;
+    if ((this == NULL) || (layout == NULL))
+        goto out;
 
-	for (i = 0; i < layout->cnt; i++) {
-		if (layout->list[i].err != -1 || !layout->list[i].stop) {
-			/* err != -1 would mean xattr present on the directory
-			 * or the directory is itself non existant.
-			 * !layout->list[i].stop would mean layout absent
-			 */
-			continue;
-		}
-		missing_xattr++;
-	}
+    conf = this->private;
 
-	gf_log (this->name, GF_LOG_TRACE,
-		"%d subvolumes missing xattr for %s",
-		missing_xattr, loc->path);
+    for (i = 0; i < layout->cnt; i++) {
+        for (j = 0; j < conf->subvolume_cnt; j++) {
+            if (conf->decommissioned_bricks[j] &&
+                conf->decommissioned_bricks[j] == layout->list[i].xlator) {
+                count++;
+            }
+        }
+    }
 
-	if (missing_xattr == 0) {
-		dht_selfheal_dir_finish (frame, this, 0);
-		return 0;
-	}
+out:
+    return count;
+}
+
+dht_distribution_type_t
+dht_distribution_type(xlator_t *this, dht_layout_t *layout)
+{
+    dht_distribution_type_t type = GF_DHT_EQUAL_DISTRIBUTION;
+    int i = 0;
+    uint32_t start_range = 0, range = 0, diff = 0;
+
+    if ((this == NULL) || (layout == NULL) || (layout->cnt < 1)) {
+        goto out;
+    }
+
+    for (i = 0; i < layout->cnt; i++) {
+        if (start_range == 0) {
+            start_range = layout->list[i].stop - layout->list[i].start;
+            continue;
+        }
+
+        range = layout->list[i].stop - layout->list[i].start;
+        diff = (range >= start_range) ? range - start_range
+                                      : start_range - range;
+
+        if ((range != 0) && (diff > layout->cnt)) {
+            type = GF_DHT_WEIGHTED_DISTRIBUTION;
+            break;
+        }
+    }
+
+out:
+    return type;
+}
+
+gf_boolean_t
+dht_should_fix_layout(call_frame_t *frame, dht_layout_t **inmem,
+                      dht_layout_t **ondisk)
+{
+    gf_boolean_t fixit = _gf_true;
+
+    dht_local_t *local = NULL;
+    int layout_span = 0;
+    int decommissioned_bricks = 0;
+    dht_conf_t *conf = NULL;
+    dht_distribution_type_t inmem_dist_type = 0;
+    dht_distribution_type_t ondisk_dist_type = 0;
+
+    conf = frame->this->private;
+
+    local = frame->local;
+
+    if ((inmem == NULL) || (*inmem == NULL) || (ondisk == NULL) ||
+        (*ondisk == NULL))
+        goto out;
+
+    dht_layout_anomalies(frame->this, &local->loc, *ondisk,
+                         &local->selfheal.hole_cnt,
+                         &local->selfheal.overlaps_cnt, NULL,
+                         &local->selfheal.down, &local->selfheal.misc, NULL);
+
+    if (local->selfheal.down || local->selfheal.misc) {
+        fixit = _gf_false;
+        goto out;
+    }
+
+    if (local->selfheal.hole_cnt || local->selfheal.overlaps_cnt)
+        goto out;
+
+    /* If commit hashes are being updated, let it through */
+    if ((*inmem)->commit_hash != (*ondisk)->commit_hash)
+        goto out;
 
-	local->call_cnt = missing_xattr;
+    layout_span = dht_layout_span(*ondisk);
 
-	for (i = 0; i < layout->cnt; i++) {
-		if (layout->list[i].err != -1 || !layout->list[i].stop)
-			continue;
+    decommissioned_bricks = dht_decommissioned_bricks_in_layout(frame->this,
+                                                                *ondisk);
+    inmem_dist_type = dht_distribution_type(frame->this, *inmem);
+    ondisk_dist_type = dht_distribution_type(frame->this, *ondisk);
 
-		dht_selfheal_dir_xattr_persubvol (frame, loc, layout, i);
+    if ((decommissioned_bricks == 0) &&
+        (layout_span ==
+         (conf->subvolume_cnt - conf->decommission_subvols_cnt)) &&
+        (inmem_dist_type == ondisk_dist_type))
+        fixit = _gf_false;
 
-		if (--missing_xattr == 0)
-			break;
-	}
-	return 0;
+out:
+
+    return fixit;
 }
 
+static int
+dht_selfheal_layout_lock(call_frame_t *frame, dht_layout_t *layout,
+                         gf_boolean_t newdir, dht_selfheal_layout_t healer,
+                         dht_need_heal_t should_heal)
+{
+    dht_local_t *local = NULL;
+    int count = 1, ret = -1, i = 0;
+    dht_lock_t **lk_array = NULL;
+    dht_conf_t *conf = NULL;
+    dht_layout_t *tmp = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO(frame->this->name, frame->local, err);
+
+    local = frame->local;
+
+    conf = frame->this->private;
+
+    local->selfheal.healer = healer;
+    local->selfheal.should_heal = should_heal;
+
+    tmp = local->selfheal.layout;
+    local->selfheal.layout = dht_layout_ref(frame->this, layout);
+    dht_layout_unref(frame->this, tmp);
+
+    if (!newdir) {
+        count = conf->subvolume_cnt;
+
+        lk_array = GF_CALLOC(count, sizeof(*lk_array), gf_common_mt_char);
+        if (lk_array == NULL) {
+            gf_uuid_unparse(local->stbuf.ia_gfid, gfid);
+            gf_smsg("dht", GF_LOG_ERROR, ENOMEM, DHT_MSG_MEM_ALLOC_FAILED,
+                    "lk_array-gfid=%s", gfid, "path=%s", local->loc.path, NULL);
+            goto err;
+        }
+
+        for (i = 0; i < count; i++) {
+            lk_array[i] = dht_lock_new(
+                frame->this, conf->subvolumes[i], &local->loc, F_WRLCK,
+                DHT_LAYOUT_HEAL_DOMAIN, NULL, FAIL_ON_ANY_ERROR);
+            if (lk_array[i] == NULL) {
+                gf_uuid_unparse(local->stbuf.ia_gfid, gfid);
+                gf_smsg(THIS->name, GF_LOG_ERROR, ENOMEM,
+                        DHT_MSG_MEM_ALLOC_FAILED, "lk_array-gfid=%s", gfid,
+                        "path=%s", local->loc.path, NULL);
+                goto err;
+            }
+        }
+    } else {
+        count = 1;
+        lk_array = GF_CALLOC(count, sizeof(*lk_array), gf_common_mt_char);
+        if (lk_array == NULL) {
+            gf_uuid_unparse(local->stbuf.ia_gfid, gfid);
+            gf_smsg(THIS->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MEM_ALLOC_FAILED,
+                    "lk_array-gfid=%s", gfid, "path=%s", local->loc.path, NULL);
+            goto err;
+        }
+
+        lk_array[0] = dht_lock_new(frame->this, local->hashed_subvol,
+                                   &local->loc, F_WRLCK, DHT_LAYOUT_HEAL_DOMAIN,
+                                   NULL, FAIL_ON_ANY_ERROR);
+        if (lk_array[0] == NULL) {
+            gf_uuid_unparse(local->stbuf.ia_gfid, gfid);
+            gf_smsg(THIS->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MEM_ALLOC_FAILED,
+                    "lk_array-gfid=%s", gfid, "path=%s", local->loc.path, NULL);
+            goto err;
+        }
+    }
+
+    local->lock[0].layout.my_layout.locks = lk_array;
+    local->lock[0].layout.my_layout.lk_count = count;
+
+    ret = dht_blocking_inodelk(frame, lk_array, count,
+                               dht_selfheal_layout_lock_cbk);
+    if (ret < 0) {
+        local->lock[0].layout.my_layout.locks = NULL;
+        local->lock[0].layout.my_layout.lk_count = 0;
+        goto err;
+    }
+
+    return 0;
+err:
+    if (lk_array != NULL) {
+        dht_lock_array_free(lk_array, count);
+        GF_FREE(lk_array);
+    }
 
+    return -1;
+}
+
+static int
+dht_selfheal_dir_xattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int op_ret, int op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    xlator_t *subvol = NULL;
+    struct iatt *stbuf = NULL;
+    int i = 0;
+    int ret = 0;
+    dht_layout_t *layout = NULL;
+    int err = 0;
+    int this_call_cnt = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    layout = local->selfheal.layout;
+    subvol = cookie;
+
+    if (op_ret == 0) {
+        err = 0;
+    } else {
+        gf_uuid_unparse(local->loc.gfid, gfid);
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                DHT_MSG_DIR_SELFHEAL_XATTR_FAILED, "name=%s", subvol->name,
+                "path=%s", local->loc.path, "gfid=%s", gfid, NULL);
+        err = op_errno;
+    }
+
+    ret = dict_get_bin(xdata, DHT_IATT_IN_XDATA_KEY, (void **)&stbuf);
+    if (ret < 0) {
+        gf_uuid_unparse(local->loc.gfid, gfid);
+        gf_msg_debug(this->name, 0,
+                     "key = %s not present in dict"
+                     ", path:%s gfid:%s",
+                     DHT_IATT_IN_XDATA_KEY, local->loc.path, gfid);
+    }
+
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].xlator == subvol) {
+            layout->list[i].err = err;
+            break;
+        }
+    }
+
+    LOCK(&frame->lock);
+    {
+        dht_iatt_merge(this, &local->stbuf, stbuf);
+    }
+    UNLOCK(&frame->lock);
+
+    this_call_cnt = dht_frame_return(frame);
+
+    if (is_last_call(this_call_cnt)) {
+        dht_selfheal_dir_finish(frame, this, 0, 1);
+    }
+
+    return 0;
+}
+
+/* Code is required to set user xattr to local->xattr
+ */
 int
-dht_selfheal_dir_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			    int op_ret, int op_errno,
-                            inode_t *inode, struct iatt *stbuf,
-                            struct iatt *preparent, struct iatt *postparent)
-{
-	dht_local_t   *local = NULL;
-	dht_layout_t  *layout = NULL;
-	call_frame_t  *prev = NULL;
-	xlator_t      *subvol = NULL;
-	int            i = 0;
-	int            this_call_cnt = 0;
-
-
-	local  = frame->local;
-	layout = local->selfheal.layout;
-	prev   = cookie;
-	subvol = prev->this;
-
-	if ((op_ret == 0) || ((op_ret == -1) && (op_errno == EEXIST))) {
-		for (i = 0; i < layout->cnt; i++) {
-			if (layout->list[i].xlator == subvol) {
-				layout->list[i].err = -1;
-				break;
-			}
-		}
-	}
-
-        if (op_ret)
-                goto out;
+dht_set_user_xattr(dict_t *dict, char *k, data_t *v, void *data)
+{
+    dict_t *set_xattr = data;
+    int ret = -1;
+
+    ret = dict_set(set_xattr, k, v);
+    return ret;
+}
+
+static int
+dht_selfheal_dir_xattr_persubvol(call_frame_t *frame, loc_t *loc,
+                                 dht_layout_t *layout, int i,
+                                 xlator_t *req_subvol)
+{
+    xlator_t *subvol = NULL;
+    dict_t *xattr = NULL;
+    dict_t *xdata = NULL;
+    int ret = 0;
+    xlator_t *this = NULL;
+    int32_t *disk_layout = NULL;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    data_t *data = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    if (req_subvol)
+        subvol = req_subvol;
+    else
+        subvol = layout->list[i].xlator;
+    this = frame->this;
+
+    GF_VALIDATE_OR_GOTO("", this, err);
+    GF_VALIDATE_OR_GOTO(this->name, layout, err);
+    GF_VALIDATE_OR_GOTO(this->name, local, err);
+    GF_VALIDATE_OR_GOTO(this->name, subvol, err);
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+
+    xattr = dict_new();
+    if (!xattr) {
+        goto err;
+    }
+
+    xdata = dict_new();
+    if (!xdata)
+        goto err;
+
+    ret = dict_set_str(xdata, GLUSTERFS_INTERNAL_FOP_KEY, "yes");
+    if (ret < 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+                "path=%s", loc->path, "key=%s", GLUSTERFS_INTERNAL_FOP_KEY,
+                "gfid=%s", gfid, NULL);
+        goto err;
+    }
+
+    ret = dict_set_int8(xdata, DHT_IATT_IN_XDATA_KEY, 1);
+    if (ret < 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+                "path=%s", loc->path, "key=%s", DHT_IATT_IN_XDATA_KEY,
+                "gfid=%s", gfid, NULL);
+        goto err;
+    }
+
+    gf_uuid_unparse(loc->inode->gfid, gfid);
+
+    ret = dht_disk_layout_extract(this, layout, i, &disk_layout);
+    if (ret == -1) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0,
+                DHT_MSG_DIR_SELFHEAL_XATTR_FAILED,
+                "extract-disk-layout-failed, path=%s", loc->path, "subvol=%s",
+                subvol->name, "gfid=%s", gfid, NULL);
+        goto err;
+    }
+
+    ret = dict_set_bin(xattr, conf->xattr_name, disk_layout, 4 * 4);
+    if (ret == -1) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0,
+                DHT_MSG_DIR_SELFHEAL_XATTR_FAILED, "path=%s", loc->path,
+                "subvol=%s", subvol->name,
+                "set-xattr-dictionary-failed"
+                "gfid=%s",
+                gfid, NULL);
+        goto err;
+    }
+    disk_layout = NULL;
+
+    gf_msg_trace(this->name, 0,
+                 "setting hash range 0x%x - 0x%x (type %d) on subvolume %s"
+                 " for %s",
+                 layout->list[i].start, layout->list[i].stop, layout->type,
+                 subvol->name, loc->path);
+
+    if (local->xattr) {
+        data = dict_get(local->xattr, QUOTA_LIMIT_KEY);
+        if (data) {
+            ret = dict_add(xattr, QUOTA_LIMIT_KEY, data);
+            if (ret) {
+                gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                        "path=%s", loc->path, "key=%s", QUOTA_LIMIT_KEY, NULL);
+            }
+        }
+        data = dict_get(local->xattr, QUOTA_LIMIT_OBJECTS_KEY);
+        if (data) {
+            ret = dict_add(xattr, QUOTA_LIMIT_OBJECTS_KEY, data);
+            if (ret) {
+                gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                        "path=%s", loc->path, "key=%s", QUOTA_LIMIT_OBJECTS_KEY,
+                        NULL);
+            }
+        }
+    }
+
+    if (!gf_uuid_is_null(local->gfid))
+        gf_uuid_copy(loc->gfid, local->gfid);
+
+    STACK_WIND_COOKIE(frame, dht_selfheal_dir_xattr_cbk, (void *)subvol, subvol,
+                      subvol->fops->setxattr, loc, xattr, 0, xdata);
 
+    dict_unref(xattr);
+    dict_unref(xdata);
 
-        dht_iatt_merge (this, &local->stbuf, stbuf, prev->this);
-        if (prev->this == local->hashed_subvol)
-                local->ia_ino = local->stbuf.ia_ino;
+    return 0;
 
-        dht_iatt_merge (this, &local->preparent, preparent, prev->this);
-        dht_iatt_merge (this, &local->postparent, postparent, prev->this);
+err:
+    if (xattr)
+        dict_unref(xattr);
+    if (xdata)
+        dict_unref(xdata);
+
+    GF_FREE(disk_layout);
+
+    dht_selfheal_dir_xattr_cbk(frame, (void *)subvol, frame->this, -1, ENOMEM,
+                               NULL);
+    return 0;
+}
+
+static int
+dht_fix_dir_xattr(call_frame_t *frame, loc_t *loc, dht_layout_t *layout)
+{
+    dht_local_t *local = NULL;
+    int i = 0;
+    int count = 0;
+    xlator_t *this = NULL;
+    dht_conf_t *conf = NULL;
+    dht_layout_t *dummy = NULL;
+
+    local = frame->local;
+    this = frame->this;
+    conf = this->private;
+
+    gf_msg_debug(this->name, 0, "%s: Writing the new range for all subvolumes",
+                 loc->path);
+
+    local->call_cnt = count = conf->subvolume_cnt;
+
+    if (gf_log_get_loglevel() >= GF_LOG_DEBUG)
+        dht_log_new_layout_for_dir_selfheal(this, loc, layout);
+
+    for (i = 0; i < layout->cnt; i++) {
+        dht_selfheal_dir_xattr_persubvol(frame, loc, layout, i, NULL);
+
+        if (--count == 0)
+            goto out;
+    }
+    /* if we are here, subvolcount > layout_count. subvols-per-directory
+     * option might be set here. We need to clear out layout from the
+     * non-participating subvolumes, else it will result in overlaps */
+    dummy = dht_layout_new(this, 1);
+    if (!dummy)
+        goto out;
+    dummy->commit_hash = layout->commit_hash;
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (_gf_false == dht_is_subvol_in_layout(layout, conf->subvolumes[i])) {
+            dht_selfheal_dir_xattr_persubvol(frame, loc, dummy, 0,
+                                             conf->subvolumes[i]);
+            if (--count == 0)
+                break;
+        }
+    }
 
+    dht_layout_unref(this, dummy);
 out:
-	this_call_cnt = dht_frame_return (frame);
+    return 0;
+}
 
-	if (is_last_call (this_call_cnt)) {
-		dht_selfheal_dir_xattr (frame, &local->loc, layout);
-	}
+static int
+dht_selfheal_dir_xattr(call_frame_t *frame, loc_t *loc, dht_layout_t *layout)
+{
+    dht_local_t *local = NULL;
+    int missing_xattr = 0;
+    int i = 0;
+    xlator_t *this = NULL;
+    dht_conf_t *conf = NULL;
+    dht_layout_t *dummy = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {
+        0,
+    };
+
+    local = frame->local;
+    this = frame->this;
+    conf = this->private;
+
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].err != -1 || !layout->list[i].stop) {
+            /* err != -1 would mean xattr present on the directory
+             * or the directory is non existent.
+             * !layout->list[i].stop would mean layout absent
+             */
+
+            continue;
+        }
+        missing_xattr++;
+    }
+    /* Also account for subvolumes with no-layout. Used for zero'ing out
+     * the layouts and for setting quota key's if present */
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (_gf_false == dht_is_subvol_in_layout(layout, conf->subvolumes[i])) {
+            missing_xattr++;
+        }
+    }
+    gf_msg_trace(this->name, 0, "%d subvolumes missing xattr for %s",
+                 missing_xattr, loc->path);
+
+    if (missing_xattr == 0) {
+        dht_selfheal_dir_finish(frame, this, 0, 1);
+        return 0;
+    }
+
+    local->call_cnt = missing_xattr;
+
+    if (gf_log_get_loglevel() >= GF_LOG_DEBUG)
+        dht_log_new_layout_for_dir_selfheal(this, loc, layout);
+
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].err != -1 || !layout->list[i].stop)
+            continue;
+
+        dht_selfheal_dir_xattr_persubvol(frame, loc, layout, i, NULL);
+
+        if (--missing_xattr == 0)
+            break;
+    }
+    dummy = dht_layout_new(this, 1);
+    if (!dummy) {
+        gf_uuid_unparse(loc->gfid, gfid);
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_DUMMY_ALLOC_FAILED,
+                "path=%s", loc->path, "gfid=%s", gfid, NULL);
+        goto out;
+    }
+    for (i = 0; i < conf->subvolume_cnt && missing_xattr; i++) {
+        if (_gf_false == dht_is_subvol_in_layout(layout, conf->subvolumes[i])) {
+            dht_selfheal_dir_xattr_persubvol(frame, loc, dummy, 0,
+                                             conf->subvolumes[i]);
+            missing_xattr--;
+        }
+    }
 
-	return 0;
+    dht_layout_unref(this, dummy);
+out:
+    return 0;
 }
 
+int
+dht_selfheal_dir_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                             int op_ret, int op_errno, struct iatt *statpre,
+                             struct iatt *statpost, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+    int this_call_cnt = 0, ret = -1;
+
+    local = frame->local;
+    layout = local->selfheal.layout;
+
+    this_call_cnt = dht_frame_return(frame);
+
+    if (is_last_call(this_call_cnt)) {
+        if (!local->heal_layout) {
+            gf_msg_trace(this->name, 0, "Skip heal layout for %s gfid = %s ",
+                         local->loc.path, uuid_utoa(local->gfid));
+
+            dht_selfheal_dir_finish(frame, this, 0, 1);
+            return 0;
+        }
+        ret = dht_selfheal_layout_lock(frame, layout, _gf_false,
+                                       dht_selfheal_dir_xattr,
+                                       dht_should_heal_layout);
+
+        if (ret < 0) {
+            dht_selfheal_dir_finish(frame, this, -1, 1);
+        }
+    }
+
+    return 0;
+}
 
 int
-dht_selfheal_dir_mkdir (call_frame_t *frame, loc_t *loc,
-			dht_layout_t *layout, int force)
-{
-	int           missing_dirs = 0;
-	int           i     = 0;
-	int           ret   = -1;
-	dht_local_t  *local = NULL;
-	xlator_t     *this = NULL;
-        dict_t       *dict = NULL;
-
-	local = frame->local;
-	this = frame->this;
-
-	for (i = 0; i < layout->cnt; i++) {
-		if (layout->list[i].err == ENOENT || force)
-			missing_dirs++;
-	}
-
-	if (missing_dirs == 0) {
-		dht_selfheal_dir_xattr (frame, loc, layout);
-		return 0;
-	}
-
-	local->call_cnt = missing_dirs;
-        if (!uuid_is_null (local->gfid)) {
-                dict = dict_new ();
-                if (!dict)
-                        return -1;
-
-                ret = dict_set_static_bin (dict, "gfid-req", local->gfid, 16);
-                if (ret)
-                        gf_log (this->name, GF_LOG_INFO,
-                                "failed to set gfid in dict");
-        } else if (local->params) {
-                /* Send the dictionary from higher layers directly */
-                dict = dict_ref (local->params);
+dht_selfheal_dir_setattr(call_frame_t *frame, loc_t *loc, struct iatt *stbuf,
+                         int32_t valid, dht_layout_t *layout)
+{
+    int missing_attr = 0;
+    int i = 0, ret = -1;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    xlator_t *this = NULL;
+    int cnt = 0;
+
+    local = frame->local;
+    this = frame->this;
+    conf = this->private;
+
+    /* We need to heal the attrs if:
+     * 1. Any directories were missing - the newly created dirs will need
+     *    to have the correct attrs set
+     * 2. An existing dir does not have the correct permissions -they may
+     *    have been changed when a brick was down.
+     */
+
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].err == -1)
+            missing_attr++;
+    }
+
+    if ((missing_attr == 0) && (local->need_attrheal == 0)) {
+        if (!local->heal_layout) {
+            gf_msg_trace(this->name, 0, "Skip heal layout for %s gfid = %s ",
+                         loc->path, uuid_utoa(loc->gfid));
+            dht_selfheal_dir_finish(frame, this, 0, 1);
+            return 0;
+        }
+        ret = dht_selfheal_layout_lock(frame, layout, _gf_false,
+                                       dht_selfheal_dir_xattr,
+                                       dht_should_heal_layout);
+
+        if (ret < 0) {
+            dht_selfheal_dir_finish(frame, this, -1, 1);
+        }
+
+        return 0;
+    }
+
+    cnt = local->call_cnt = conf->subvolume_cnt;
+
+    for (i = 0; i < cnt; i++) {
+        STACK_WIND(frame, dht_selfheal_dir_setattr_cbk, layout->list[i].xlator,
+                   layout->list[i].xlator->fops->setattr, loc, stbuf, valid,
+                   NULL);
+    }
+
+    return 0;
+}
+
+static int
+dht_selfheal_dir_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int op_ret, int op_errno, inode_t *inode,
+                           struct iatt *stbuf, struct iatt *preparent,
+                           struct iatt *postparent, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    dht_layout_t *layout = NULL;
+    xlator_t *prev = NULL;
+    xlator_t *subvol = NULL;
+    int i = 0, ret = -1;
+    int this_call_cnt = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+    layout = local->selfheal.layout;
+    prev = cookie;
+    subvol = prev;
+
+    if ((op_ret == 0) || ((op_ret == -1) && (op_errno == EEXIST))) {
+        for (i = 0; i < layout->cnt; i++) {
+            if (layout->list[i].xlator == subvol) {
+                layout->list[i].err = -1;
+                break;
+            }
         }
+    }
+
+    if (op_ret) {
+        gf_uuid_unparse(local->loc.gfid, gfid);
+        gf_smsg(this->name,
+                ((op_errno == EEXIST) ? GF_LOG_DEBUG : GF_LOG_WARNING),
+                op_errno, DHT_MSG_DIR_SELFHEAL_FAILED, "path=%s",
+                local->loc.path, "gfid=%s", gfid, NULL);
+        goto out;
+    }
+    dht_iatt_merge(this, &local->preparent, preparent);
+    dht_iatt_merge(this, &local->postparent, postparent);
+    ret = 0;
+
+out:
+    this_call_cnt = dht_frame_return(frame);
+
+    if (is_last_call(this_call_cnt)) {
+        dht_selfheal_dir_finish(frame, this, ret, 0);
+        dht_selfheal_dir_setattr(frame, &local->loc, &local->stbuf, 0xffffff,
+                                 layout);
+    }
+
+    return 0;
+}
 
+static int
+dht_selfheal_dir_mkdir_lookup_done(call_frame_t *frame, xlator_t *this)
+{
+    dht_local_t *local = NULL;
+    int i = 0;
+    dict_t *dict = NULL;
+    dht_layout_t *layout = NULL;
+    loc_t *loc = NULL;
+    int cnt = 0;
+    int ret = -1;
+
+    VALIDATE_OR_GOTO(this->private, err);
+
+    local = frame->local;
+    layout = local->layout;
+    loc = &local->loc;
+
+    if (!gf_uuid_is_null(local->gfid)) {
+        dict = dict_new();
+        if (!dict)
+            return -1;
+
+        ret = dict_set_gfuuid(dict, "gfid-req", local->gfid, true);
+        if (ret)
+            gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+                    "path=%s", loc->path, "key=gfid-req", NULL);
+    } else if (local->params) {
+        /* Send the dictionary from higher layers directly */
+
+        dict = dict_ref(local->params);
+    }
+    /* Code to update all extended attributed from local->xattr
+       to dict
+    */
+    dht_dir_set_heal_xattr(this, local, dict, local->xattr, NULL, NULL);
+
+    if (!dict) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_IS_NULL, NULL);
+        dict = dict_new();
         if (!dict)
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "dict is NULL, need to make sure gfid's are same");
+            return -1;
+    }
+    ret = dict_set_flag(dict, GF_INTERNAL_CTX_KEY, GF_DHT_HEAL_DIR);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, "key=%s",
+                GF_INTERNAL_CTX_KEY, "path=%s", loc->path, NULL);
+        /* We can still continue. As heal can still happen
+         * unless quota limits have reached for the dir.
+         */
+    }
+
+    cnt = layout->cnt;
+    for (i = 0; i < cnt; i++) {
+        if (layout->list[i].err == ESTALE || layout->list[i].err == ENOENT ||
+            local->selfheal.force_mkdir) {
+            gf_msg_debug(this->name, 0, "Creating directory %s on subvol %s",
+                         loc->path, layout->list[i].xlator->name);
+
+            STACK_WIND_COOKIE(
+                frame, dht_selfheal_dir_mkdir_cbk, layout->list[i].xlator,
+                layout->list[i].xlator, layout->list[i].xlator->fops->mkdir,
+                loc,
+                st_mode_from_ia(local->stbuf.ia_prot, local->stbuf.ia_type), 0,
+                dict);
+        }
+    }
+
+    if (dict)
+        dict_unref(dict);
+
+    return 0;
 
-	for (i = 0; i < layout->cnt; i++) {
-		if (layout->list[i].err == ENOENT || force) {
-			gf_log (this->name, GF_LOG_TRACE,
-				"creating directory %s on subvol %s",
-				loc->path, layout->list[i].xlator->name);
+err:
+    dht_selfheal_dir_finish(frame, this, -1, 1);
+    return 0;
+}
+
+static int
+dht_selfheal_dir_mkdir_lookup_cbk(call_frame_t *frame, void *cookie,
+                                  xlator_t *this, int op_ret, int op_errno,
+                                  inode_t *inode, struct iatt *stbuf,
+                                  dict_t *xattr, struct iatt *postparent)
+{
+    dht_local_t *local = NULL;
+    int i = 0;
+    int this_call_cnt = 0;
+    int missing_dirs = 0;
+    dht_layout_t *layout = NULL;
+    xlator_t *prev = 0;
+    loc_t *loc = NULL;
+    char gfid_local[GF_UUID_BUF_SIZE] = {0};
+    int index = -1;
+
+    VALIDATE_OR_GOTO(this->private, err);
+
+    local = frame->local;
+    layout = local->layout;
+    loc = &local->loc;
+    prev = cookie;
+
+    if (!gf_uuid_is_null(local->gfid))
+        gf_uuid_unparse(local->gfid, gfid_local);
+
+    LOCK(&frame->lock);
+    {
+        index = dht_layout_index_for_subvol(layout, prev);
+        if ((op_ret < 0) && (op_errno == ENOENT || op_errno == ESTALE)) {
+            local->selfheal.hole_cnt = !local->selfheal.hole_cnt
+                                           ? 1
+                                           : local->selfheal.hole_cnt + 1;
+            /* the status might have changed. Update the layout with the
+             * new status
+             */
+            if (index >= 0) {
+                layout->list[index].err = op_errno;
+            }
+        }
 
-			STACK_WIND (frame, dht_selfheal_dir_mkdir_cbk,
-				    layout->list[i].xlator,
-				    layout->list[i].xlator->fops->mkdir,
-				    loc,
-                                    st_mode_from_ia (local->stbuf.ia_prot,
-                                                     local->stbuf.ia_type),
-                                    dict);
-		}
-	}
+        if (!op_ret) {
+            dht_iatt_merge(this, &local->stbuf, stbuf);
+            if (prev == local->mds_subvol) {
+                dict_unref(local->xattr);
+                local->xattr = dict_ref(xattr);
+            }
+            /* the status might have changed. Update the layout with the
+             * new status
+             */
+            if (index >= 0) {
+                layout->list[index].err = -1;
+            }
+        }
+    }
+    UNLOCK(&frame->lock);
+
+    this_call_cnt = dht_frame_return(frame);
+
+    if (is_last_call(this_call_cnt)) {
+        if (local->selfheal.hole_cnt == layout->cnt) {
+            gf_msg_debug(this->name, op_errno,
+                         "Lookup failed, an rmdir could have "
+                         "deleted this entry %s",
+                         loc->name);
+            local->op_errno = op_errno;
+            goto err;
+        } else {
+            for (i = 0; i < layout->cnt; i++) {
+                if (layout->list[i].err == ENOENT ||
+                    layout->list[i].err == ESTALE ||
+                    local->selfheal.force_mkdir)
+                    missing_dirs++;
+            }
+
+            if (missing_dirs == 0) {
+                dht_selfheal_dir_finish(frame, this, 0, 0);
+                dht_selfheal_dir_setattr(frame, loc, &local->stbuf, 0xffffffff,
+                                         layout);
+                return 0;
+            }
+
+            local->call_cnt = missing_dirs;
+            dht_selfheal_dir_mkdir_lookup_done(frame, this);
+        }
+    }
 
-        if (dict)
-                dict_unref (dict);
+    return 0;
 
-	return 0;
+err:
+    dht_selfheal_dir_finish(frame, this, -1, 1);
+    return 0;
 }
 
+static int
+dht_selfheal_dir_mkdir_lock_cbk(call_frame_t *frame, void *cookie,
+                                xlator_t *this, int32_t op_ret,
+                                int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int i = 0;
+    int ret = -1;
+    xlator_t *mds_subvol = NULL;
 
-int
-dht_selfheal_layout_alloc_start (xlator_t *this, loc_t *loc,
-                                 dht_layout_t *layout)
+    VALIDATE_OR_GOTO(this->private, err);
+
+    conf = this->private;
+    local = frame->local;
+    mds_subvol = local->mds_subvol;
+
+    local->call_cnt = conf->subvolume_cnt;
+
+    if (op_ret < 0) {
+        if (op_errno == EINVAL) {
+            local->call_cnt = 1;
+            dht_selfheal_dir_mkdir_lookup_done(frame, this);
+            return 0;
+        }
+
+        gf_smsg(this->name, GF_LOG_WARNING, op_errno, DHT_MSG_ENTRYLK_ERROR,
+                "path=%s", local->loc.path, NULL);
+
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    /* After getting locks, perform lookup again to ensure that the
+       directory was not deleted by a racing rmdir
+    */
+    if (!local->xattr_req)
+        local->xattr_req = dict_new();
+
+    ret = dict_set_int32(local->xattr_req, "list-xattr", 1);
+    if (ret)
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED, "path=%s",
+                local->loc.path, NULL);
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (mds_subvol && conf->subvolumes[i] == mds_subvol) {
+            STACK_WIND_COOKIE(frame, dht_selfheal_dir_mkdir_lookup_cbk,
+                              conf->subvolumes[i], conf->subvolumes[i],
+                              conf->subvolumes[i]->fops->lookup, &local->loc,
+                              local->xattr_req);
+        } else {
+            STACK_WIND_COOKIE(frame, dht_selfheal_dir_mkdir_lookup_cbk,
+                              conf->subvolumes[i], conf->subvolumes[i],
+                              conf->subvolumes[i]->fops->lookup, &local->loc,
+                              NULL);
+        }
+    }
+
+    return 0;
+
+err:
+    dht_selfheal_dir_finish(frame, this, -1, 1);
+    return 0;
+}
+
+static int
+dht_selfheal_dir_mkdir(call_frame_t *frame, loc_t *loc, dht_layout_t *layout,
+                       int force)
+{
+    int missing_dirs = 0;
+    int i = 0;
+    int op_errno = 0;
+    int ret = -1;
+    dht_local_t *local = NULL;
+    xlator_t *this = NULL;
+    dht_conf_t *conf = NULL;
+
+    local = frame->local;
+    this = frame->this;
+    conf = this->private;
+
+    local->selfheal.force_mkdir = force;
+    local->selfheal.hole_cnt = 0;
+
+    for (i = 0; i < layout->cnt; i++) {
+        if (layout->list[i].err == ENOENT || force)
+            missing_dirs++;
+    }
+
+    if (missing_dirs == 0) {
+        /* We don't need to create any directories. Proceed to heal the
+         * attrs and xattrs
+         */
+        if (!__is_root_gfid(local->stbuf.ia_gfid)) {
+            if (local->need_xattr_heal) {
+                local->need_xattr_heal = 0;
+                ret = dht_dir_xattr_heal(this, local, &op_errno);
+                if (ret) {
+                    gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                            DHT_MSG_DIR_XATTR_HEAL_FAILED, "path=%s",
+                            local->loc.path, "gfid=%s", local->gfid, NULL);
+                }
+            } else {
+                if (!gf_uuid_is_null(local->gfid))
+                    gf_uuid_copy(loc->gfid, local->gfid);
+
+                ret = dht_common_mark_mdsxattr(frame, NULL, 0);
+                if (!ret)
+                    return 0;
+
+                gf_smsg(this->name, GF_LOG_INFO, 0, DHT_MSG_SET_XATTR_FAILED,
+                        "path=%s", local->loc.path, "gfid=%s", local->gfid,
+                        NULL);
+            }
+        }
+        dht_selfheal_dir_setattr(frame, loc, &local->stbuf, 0xffffffff, layout);
+        return 0;
+    }
+
+    /* MDS xattr is populated only while DHT is having more than one
+     subvol.In case of graph switch while adding more dht subvols need to
+     consider hash subvol as a MDS to avoid MDS check failure at the time
+     of running fop on directory
+    */
+    if (!dict_get(local->xattr, conf->mds_xattr_key) &&
+        (conf->subvolume_cnt > 1)) {
+        if (local->hashed_subvol == NULL) {
+            local->hashed_subvol = dht_subvol_get_hashed(this, loc);
+            if (local->hashed_subvol == NULL) {
+                local->op_errno = EINVAL;
+                gf_smsg(this->name, GF_LOG_WARNING, local->op_errno,
+                        DHT_MSG_HASHED_SUBVOL_GET_FAILED, "gfid=%s",
+                        loc->pargfid, "name=%s", loc->name, "path=%s",
+                        loc->path, NULL);
+                goto err;
+            }
+        }
+        ret = dht_inode_ctx_mdsvol_set(local->inode, this,
+                                       local->hashed_subvol);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SET_INODE_CTX_FAILED,
+                   "Failed to set hashed subvol for %s on inode vol is %s",
+                   local->loc.path,
+                   local->hashed_subvol ? local->hashed_subvol->name : "NULL");
+            goto err;
+        }
+    }
+
+    if (local->hashed_subvol == NULL) {
+        local->hashed_subvol = dht_subvol_get_hashed(this, loc);
+        if (local->hashed_subvol == NULL) {
+            local->op_errno = EINVAL;
+            gf_smsg(this->name, GF_LOG_WARNING, local->op_errno,
+                    DHT_MSG_HASHED_SUBVOL_GET_FAILED, "gfid=%s", loc->pargfid,
+                    "name=%s", loc->name, "path=%s", loc->path, NULL);
+            goto err;
+        }
+    }
+
+    local->current = &local->lock[0];
+    ret = dht_protect_namespace(frame, loc, local->hashed_subvol,
+                                &local->current->ns,
+                                dht_selfheal_dir_mkdir_lock_cbk);
+
+    if (ret < 0)
+        goto err;
+
+    return 0;
+err:
+    return -1;
+}
+
+static int
+dht_selfheal_layout_alloc_start(xlator_t *this, loc_t *loc,
+                                dht_layout_t *layout)
 {
-        int           start = 0;
-        uint32_t      hashval = 0;
-        int           ret = 0;
+    int start = 0;
+    uint32_t hashval = 0;
+    int ret = 0;
+    const char *str = NULL;
+    dht_conf_t *conf = NULL;
+    char buf[UUID_CANONICAL_FORM_LEN + 1] = {
+        0,
+    };
+
+    conf = this->private;
+
+    if (conf->randomize_by_gfid) {
+        str = uuid_utoa_r(loc->gfid, buf);
+    } else {
+        str = loc->path;
+    }
+
+    ret = dht_hash_compute(this, layout->type, str, &hashval);
+    if (ret == 0) {
+        start = (hashval % layout->cnt);
+    }
+
+    return start;
+}
 
-        ret = dht_hash_compute (layout->type, loc->path, &hashval);
-        if (ret == 0) {
-                start = (hashval % layout->cnt);
+static int
+dht_get_layout_count(xlator_t *this, dht_layout_t *layout, int new_layout)
+{
+    int i = 0;
+    int j = 0;
+    int err = 0;
+    int count = 0;
+    dht_conf_t *conf = NULL;
+
+    /* Gets in use only for replace-brick, remove-brick */
+    conf = this->private;
+    for (i = 0; i < layout->cnt; i++) {
+        for (j = 0; j < conf->subvolume_cnt; j++) {
+            if (conf->decommissioned_bricks[j] &&
+                conf->decommissioned_bricks[j] == layout->list[i].xlator) {
+                layout->list[i].err = EINVAL;
+                break;
+            }
         }
+    }
+
+    for (i = 0; i < layout->cnt; i++) {
+        err = layout->list[i].err;
+        if (err == -1 || err == 0 || err == ENOENT) {
+            /* Take this with a pinch of salt. The behaviour seems
+             * to be slightly different when this function is
+             * invoked from mkdir codepath. For eg., err == 0 in
+             * mkdir codepath means directory created but xattr
+             * is not set yet.
+             */
+
+            /* Setting list[i].err = -1 is an indication for
+               dht_selfheal_layout_new_directory() to assign
+               a range. We set it to -1 based on any one of
+               the three criteria:
+
+               - err == -1 already, which means directory
+                 existed but layout was not set on it.
+
+               - err == 0, which means directory exists and
+                 has an old layout piece which will be
+                 overwritten now.
+
+               - err == ENOENT, which means directory does
+                 not exist (possibly racing with mkdir or
+                 finishing half done mkdir). The missing
+                 directory will be attempted to be recreated.
+            */
+            count++;
+            if (!err)
+                layout->list[i].err = -1;
+        }
+    }
+
+    /* no subvolume has enough space, but can't stop directory creation */
+    if (!count || !new_layout) {
+        for (i = 0; i < layout->cnt; i++) {
+            err = layout->list[i].err;
+            if (err == ENOSPC) {
+                layout->list[i].err = -1;
+                count++;
+            }
+        }
+    }
+
+    /* if layout->spread_cnt is set, check if it is <= available
+     * subvolumes (down brick and decommissioned bricks are considered
+     * un-available). Else return count (available up bricks) */
+    count = ((layout->spread_cnt && (layout->spread_cnt <= count))
+                 ? layout->spread_cnt
+                 : ((count) ? count : 1));
 
-        return start;
+    return count;
 }
 
+void
+dht_selfheal_layout_new_directory(call_frame_t *frame, loc_t *loc,
+                                  dht_layout_t *new_layout);
 
 void
-dht_selfheal_layout_new_directory (call_frame_t *frame, loc_t *loc,
-				   dht_layout_t *layout)
-{
-	xlator_t    *this = NULL;
-	uint32_t     chunk = 0;
-	int          i = 0;
-	uint32_t     start = 0;
-	int          cnt = 0;
-	int          err = 0;
-        int          start_subvol = 0;
-
-	this = frame->this;
-
-	for (i = 0; i < layout->cnt; i++) {
-		err = layout->list[i].err;
-		if (err == -1 || err == 0) {
-			layout->list[i].err = -1;
-			cnt++;
-		}
-	}
-
-        /* no subvolume has enough space, but can't stop directory creation */
-        if (!cnt) {
-                for (i = 0; i < layout->cnt; i++) {
-                        err = layout->list[i].err;
-                        if (err == ENOSPC) {
-                                layout->list[i].err = -1;
-                                cnt++;
-                        }
+dht_layout_range_swap(dht_layout_t *layout, int i, int j);
+
+/*
+ * It's a bit icky using local variables in a macro, but it makes the rest
+ * of the code a lot clearer.
+ */
+#define OV_ENTRY(x, y) table[x * new->cnt + y]
+
+static void
+dht_selfheal_layout_maximize_overlap(call_frame_t *frame, loc_t *loc,
+                                     dht_layout_t *new, dht_layout_t *old)
+{
+    int i = 0;
+    int j = 0;
+    uint32_t curr_overlap = 0;
+    uint32_t max_overlap = 0;
+    int max_overlap_idx = -1;
+    uint32_t overlap = 0;
+    uint32_t *table = NULL;
+
+    dht_layout_sort_volname(old);
+    /* Now both old_layout->list[] and new_layout->list[]
+       are match the same xlators/subvolumes. i.e,
+       old_layout->[i] and new_layout->[i] are referring
+       to the same subvolumes
+    */
+
+    /* Build a table of overlaps between new[i] and old[j]. */
+    table = alloca(sizeof(overlap) * old->cnt * new->cnt);
+    if (!table) {
+        return;
+    }
+    memset(table, 0, sizeof(overlap) * old->cnt * new->cnt);
+    for (i = 0; i < new->cnt; ++i) {
+        for (j = 0; j < old->cnt; ++j) {
+            OV_ENTRY(i, j) = dht_overlap_calc(old, j, new, i);
+        }
+    }
+
+    for (i = 0; i < new->cnt; i++) {
+        if (new->list[i].err > 0) {
+            /* Subvol might be marked for decommission
+               with EINVAL, or some other serious error
+               marked with positive errno.
+            */
+            continue;
+        }
+
+        max_overlap = 0;
+        max_overlap_idx = i;
+        for (j = (i + 1); j < new->cnt; ++j) {
+            if (new->list[j].err > 0) {
+                /* Subvol might be marked for decommission
+                with EINVAL, or some other serious error
+                marked with positive errno.
+                */
+                continue;
+            }
+            /* Calculate the overlap now. */
+            curr_overlap = OV_ENTRY(i, i) + OV_ENTRY(j, j);
+            /* Calculate the overlap after the proposed swap. */
+            overlap = OV_ENTRY(i, j) + OV_ENTRY(j, i);
+            /* Are we better than status quo? */
+            if (overlap > curr_overlap) {
+                overlap -= curr_overlap;
+                /* Are we better than the previous choice? */
+                if (overlap > max_overlap) {
+                    max_overlap = overlap;
+                    max_overlap_idx = j;
                 }
+            }
         }
 
-	chunk = ((unsigned long) 0xffffffff) / ((cnt) ? cnt : 1);
-
-	start_subvol = dht_selfheal_layout_alloc_start (this, loc, layout);
-
-	for (i = start_subvol; i < layout->cnt; i++) {
-		err = layout->list[i].err;
-		if (err == -1) {
-			layout->list[i].start = start;
-			layout->list[i].stop  = start + chunk - 1;
-			
-			start = start + chunk;
-
-			gf_log (this->name, GF_LOG_TRACE,
-				"gave fix: %u - %u on %s for %s",
-				layout->list[i].start, layout->list[i].stop,
-				layout->list[i].xlator->name, loc->path);
-			if (--cnt == 0) {
-				layout->list[i].stop = 0xffffffff;
-				break;
-			}
-		}
-	}
-
-	for (i = 0; i < start_subvol; i++) {
-		err = layout->list[i].err;
-		if (err == -1) {
-			layout->list[i].start = start;
-			layout->list[i].stop  = start + chunk - 1;
-			
-			start = start + chunk;
-
-			gf_log (this->name, GF_LOG_TRACE,
-				"gave fix: %u - %u on %s for %s",
-				layout->list[i].start, layout->list[i].stop,
-				layout->list[i].xlator->name, loc->path);
-			if (--cnt == 0) {
-				layout->list[i].stop = 0xffffffff;
-				break;
-			}
-		}
-	}
+        if (max_overlap_idx != i) {
+            dht_layout_range_swap(new, i, max_overlap_idx);
+            /* Need to swap the table values too. */
+            for (j = 0; j < old->cnt; ++j) {
+                overlap = OV_ENTRY(i, j);
+                OV_ENTRY(i, j) = OV_ENTRY(max_overlap_idx, j);
+                OV_ENTRY(max_overlap_idx, j) = overlap;
+            }
+        }
+    }
 }
 
+static dht_layout_t *
+dht_fix_layout_of_directory(call_frame_t *frame, loc_t *loc,
+                            dht_layout_t *layout)
+{
+    int i = 0;
+    xlator_t *this = NULL;
+    dht_layout_t *new_layout = NULL;
+    dht_conf_t *priv = NULL;
+    dht_local_t *local = NULL;
+    uint32_t subvol_down = 0;
+    gf_boolean_t maximize_overlap = _gf_true;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    this = frame->this;
+    priv = this->private;
+    local = frame->local;
+
+    if (layout->type == DHT_HASH_TYPE_DM_USER) {
+        gf_msg_debug(THIS->name, 0, "leaving %s alone", loc->path);
+        goto done;
+    }
+
+    new_layout = dht_layout_new(this, priv->subvolume_cnt);
+    if (!new_layout) {
+        gf_uuid_unparse(loc->gfid, gfid);
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MEM_ALLOC_FAILED,
+                "new_layout, path=%s", loc->path, "gfid=%s", gfid, NULL);
+        goto done;
+    }
+
+    /* If a subvolume is down, do not re-write the layout. */
+    dht_layout_anomalies(this, loc, layout, NULL, NULL, NULL, &subvol_down,
+                         NULL, NULL);
+
+    if (subvol_down) {
+        gf_uuid_unparse(loc->gfid, gfid);
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LAYOUT_FIX_FAILED,
+                "subvol-down=%u", subvol_down, "Skipping-fix-layout", "path=%s",
+                loc->path, "gfid=%s", gfid, NULL);
+        GF_FREE(new_layout);
+        return NULL;
+    }
+
+    for (i = 0; i < new_layout->cnt; i++) {
+        if (layout->list[i].err != ENOSPC)
+            new_layout->list[i].err = layout->list[i].err;
+        else
+            new_layout->list[i].err = -1;
+
+        new_layout->list[i].xlator = layout->list[i].xlator;
+    }
+
+    new_layout->commit_hash = layout->commit_hash;
+
+    if (priv->du_stats) {
+        for (i = 0; i < priv->subvolume_cnt; ++i) {
+            gf_smsg(this->name, GF_LOG_DEBUG, 0, DHT_MSG_SUBVOL_INFO,
+                    "index=%d", i, "name=%s", priv->subvolumes[i]->name,
+                    "chunks=%u", priv->du_stats[i].chunks, "path=%s", loc->path,
+                    NULL);
+
+            /* Maximize overlap if the bricks are all the same
+             *  size.
+             * This is probably not going to be very common on
+             * live setups but will benefit our regression tests
+             */
+            if (i && (priv->du_stats[i].chunks != priv->du_stats[0].chunks)) {
+                maximize_overlap = _gf_false;
+            }
+        }
+    } else {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_NO_DISK_USAGE_STATUS,
+                NULL);
+    }
+
+    /* First give it a layout as though it is a new directory. This
+       ensures rotation to kick in */
+    dht_layout_sort_volname(new_layout);
+    dht_selfheal_layout_new_directory(frame, loc, new_layout);
+
+    /* Maximize overlap if weighted-rebalance is disabled */
+    if (!priv->do_weighting)
+        maximize_overlap = _gf_true;
+
+    /* Now selectively re-assign ranges only when it helps */
+    if (maximize_overlap) {
+        dht_selfheal_layout_maximize_overlap(frame, loc, new_layout, layout);
+    }
+done:
+    if (new_layout) {
+        /* Make sure the extra 'ref' for existing layout is removed */
+        dht_layout_unref(this, local->layout);
+
+        local->layout = new_layout;
+    }
+
+    return local->layout;
+}
 
-int
-dht_selfheal_dir_getafix (call_frame_t *frame, loc_t *loc,
-			  dht_layout_t *layout)
-{
-	dht_conf_t  *conf = NULL;
-	xlator_t    *this = NULL;
-	dht_local_t *local = NULL;
-	int          missing = -1;
-	int          down = -1;
-	int          holes = -1;
-	int          ret = -1;
-	int          i = -1;
-	int          overlaps = -1;
-
-	this = frame->this;
-	conf = this->private;
-	local = frame->local;
-
-	missing = local->selfheal.missing;
-	down = local->selfheal.down;
-	holes = local->selfheal.hole_cnt;
-	overlaps = local->selfheal.overlaps_cnt;
-
-	if ((missing + down) == conf->subvolume_cnt) {
-		dht_selfheal_layout_new_directory (frame, loc, layout);
-		ret = 0;
-	}
-
-	if (holes <= down) {
-		/* the down subvol might fill up the holes */
-		ret = 0;
-	}
-
-	if (holes || overlaps) {
-		dht_selfheal_layout_new_directory (frame, loc, layout);
-		ret = 0;
-	}
-
-	for (i = 0; i < layout->cnt; i++) {
-		/* directory not present */
-		if (layout->list[i].err == ENOENT) {
-			ret = 0;
-			break;
-		}
-	}
-
-	/* TODO: give a fix to these non-virgins */
-
-	return ret;
+/*
+ * Having to call this 2x for each entry in the layout is pretty horrible, but
+ * that's what all of this layout-sorting nonsense gets us.
+ */
+static uint32_t
+dht_get_chunks_from_xl(xlator_t *parent, xlator_t *child)
+{
+    dht_conf_t *priv = parent->private;
+    xlator_list_t *trav;
+    uint32_t index = 0;
+
+    if (!priv->du_stats) {
+        return 0;
+    }
+
+    for (trav = parent->children; trav; trav = trav->next) {
+        if (trav->xlator == child) {
+            return priv->du_stats[index].chunks;
+        }
+        ++index;
+    }
+
+    return 0;
+}
+
+void
+dht_selfheal_layout_new_directory(call_frame_t *frame, loc_t *loc,
+                                  dht_layout_t *layout)
+{
+    xlator_t *this = NULL;
+    double chunk = 0;
+    int i = 0;
+    uint32_t start = 0;
+    int bricks_to_use = 0;
+    int err = 0;
+    int start_subvol = 0;
+    uint32_t curr_size;
+    uint32_t range_size;
+    uint64_t total_size = 0;
+    int real_i;
+    dht_conf_t *priv;
+    gf_boolean_t weight_by_size;
+    int bricks_used = 0;
+
+    this = frame->this;
+    priv = this->private;
+    weight_by_size = priv->do_weighting;
+
+    bricks_to_use = dht_get_layout_count(this, layout, 1);
+    GF_ASSERT(bricks_to_use > 0);
+
+    bricks_used = 0;
+    for (i = 0; i < layout->cnt; ++i) {
+        err = layout->list[i].err;
+        if ((err != -1) && (err != ENOENT)) {
+            continue;
+        }
+        curr_size = dht_get_chunks_from_xl(this, layout->list[i].xlator);
+        if (!curr_size) {
+            weight_by_size = _gf_false;
+            break;
+        }
+        total_size += curr_size;
+        if (++bricks_used >= bricks_to_use) {
+            break;
+        }
+    }
+
+    if (weight_by_size && total_size) {
+        /* We know total_size is not zero. */
+        chunk = ((double)0xffffffff) / ((double)total_size);
+        gf_msg_debug(this->name, 0,
+                     "chunk size = 0xffffffff / %" PRIu64 " = %f", total_size,
+                     chunk);
+    } else {
+        weight_by_size = _gf_false;
+        chunk = ((unsigned long)0xffffffff) / bricks_to_use;
+    }
+
+    start_subvol = dht_selfheal_layout_alloc_start(this, loc, layout);
+
+    /* clear out the range, as we are re-computing here */
+    DHT_RESET_LAYOUT_RANGE(layout);
+
+    /*
+     * OK, what's this "real_i" stuff about?  This used to be two loops -
+     * from start_subvol to layout->cnt-1, then from 0 to start_subvol-1.
+     * That way is practically an open invitation to bugs when only one
+     * of the loops is updated.  Using real_i and modulo operators to make
+     * it one loop avoids this problem.  Remember, folks: it's everyone's
+     * responsibility to help stamp out copy/paste abuse.
+     */
+    bricks_used = 0;
+    for (real_i = 0; real_i < layout->cnt; real_i++) {
+        i = (real_i + start_subvol) % layout->cnt;
+        err = layout->list[i].err;
+        if ((err != -1) && (err != ENOENT)) {
+            continue;
+        }
+        if (weight_by_size) {
+            curr_size = dht_get_chunks_from_xl(this, layout->list[i].xlator);
+            if (!curr_size) {
+                continue;
+            }
+        } else {
+            curr_size = 1;
+        }
+        range_size = chunk * curr_size;
+        gf_msg_debug(this->name, 0, "assigning range size 0x%x to %s",
+                     range_size, layout->list[i].xlator->name);
+        DHT_SET_LAYOUT_RANGE(layout, i, start, range_size, loc->path);
+        if (++bricks_used >= bricks_to_use) {
+            layout->list[i].stop = 0xffffffff;
+            goto done;
+        }
+        start += range_size;
+    }
+
+done:
+    return;
+}
+
+static int
+dht_selfheal_dir_getafix(call_frame_t *frame, loc_t *loc, dht_layout_t *layout)
+{
+    dht_local_t *local = NULL;
+    uint32_t holes = 0;
+    int ret = -1;
+    int i = -1;
+    uint32_t overlaps = 0;
+
+    local = frame->local;
+
+    holes = local->selfheal.hole_cnt;
+    overlaps = local->selfheal.overlaps_cnt;
+
+    if (holes || overlaps) {
+        /* If the layout has anomalies which would change the hash
+         * ranges, then we need to reset the commit_hash for this
+         * directory, as the layout would change and things may not
+         * be in place as expected */
+        layout->commit_hash = DHT_LAYOUT_HASH_INVALID;
+        dht_selfheal_layout_new_directory(frame, loc, layout);
+        ret = 0;
+    }
+
+    for (i = 0; i < layout->cnt; i++) {
+        /* directory not present */
+        if (layout->list[i].err == ENOENT) {
+            ret = 0;
+            break;
+        }
+    }
+
+    /* TODO: give a fix to these non-virgins */
+
+    return ret;
 }
 
 int
-dht_selfheal_new_directory (call_frame_t *frame, 
-			    dht_selfheal_dir_cbk_t dir_cbk,
-			    dht_layout_t *layout)
+dht_selfheal_new_directory(call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk,
+                           dht_layout_t *layout)
 {
-	dht_local_t *local = NULL;
+    dht_local_t *local = NULL;
+    int ret = 0;
+    inode_t *linked_inode = NULL, *inode = NULL;
+    loc_t *loc = NULL;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    int32_t op_errno = EIO;
+
+    local = frame->local;
+
+    loc = &local->loc;
 
-	local = frame->local;
+    gf_uuid_unparse(local->stbuf.ia_gfid, gfid);
+    gf_uuid_unparse(loc->parent->gfid, pgfid);
 
-	local->selfheal.dir_cbk = dir_cbk;
-	local->selfheal.layout = dht_layout_ref (frame->this, layout);
+    linked_inode = inode_link(loc->inode, loc->parent, loc->name,
+                              &local->stbuf);
+    if (!linked_inode) {
+        gf_smsg(frame->this->name, GF_LOG_WARNING, 0, DHT_MSG_LINK_INODE_FAILED,
+                "pgfid=%s", pgfid, "name=%s", loc->name, "gfid=%s", gfid, NULL);
+        ret = -1;
+        goto out;
+    }
 
-	dht_layout_sort_volname (layout);
-	dht_selfheal_layout_new_directory (frame, &local->loc, layout);
-	dht_selfheal_dir_xattr (frame, &local->loc, layout);
-	return 0;
+    inode = loc->inode;
+    loc->inode = linked_inode;
+    inode_unref(inode);
+
+    local->selfheal.dir_cbk = dir_cbk;
+    local->selfheal.layout = dht_layout_ref(frame->this, layout);
+
+    dht_layout_sort_volname(layout);
+    dht_selfheal_layout_new_directory(frame, &local->loc, layout);
+
+    op_errno = ENOMEM;
+    ret = dht_selfheal_layout_lock(frame, layout, _gf_true,
+                                   dht_selfheal_dir_xattr,
+                                   dht_should_heal_layout);
+
+out:
+    if (ret < 0) {
+        dir_cbk(frame, NULL, frame->this, -1, op_errno, NULL);
+    }
+
+    return 0;
 }
 
+int
+dht_fix_directory_layout(call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk,
+                         dht_layout_t *layout)
+{
+    dht_local_t *local = NULL;
+    dht_layout_t *tmp_layout = NULL;
+    int ret = 0;
+
+    local = frame->local;
+
+    local->selfheal.dir_cbk = dir_cbk;
+    local->selfheal.layout = dht_layout_ref(frame->this, layout);
+
+    /* No layout sorting required here */
+    tmp_layout = dht_fix_layout_of_directory(frame, &local->loc, layout);
+    if (!tmp_layout) {
+        return -1;
+    }
+
+    ret = dht_selfheal_layout_lock(frame, tmp_layout, _gf_false,
+                                   dht_fix_dir_xattr, dht_should_fix_layout);
+
+    return ret;
+}
 
 int
-dht_selfheal_directory (call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk,
-			loc_t *loc, dht_layout_t *layout)
-{
-	dht_local_t *local    = NULL;
-	uint32_t     holes    = 0;
-	uint32_t     down     = 0;
-	uint32_t     misc     = 0;
-	int          ret      = 0;
-	xlator_t    *this     = NULL;
-
-	local = frame->local;
-	this = frame->this;
-
-	dht_layout_anomalies (this, loc, layout,
-                              &local->selfheal.hole_cnt,
-                              &local->selfheal.overlaps_cnt,
-                              &local->selfheal.missing,
-                              &local->selfheal.down,
-                              &local->selfheal.misc);
-
-	holes    = local->selfheal.hole_cnt;
-	down     = local->selfheal.down;
-	misc     = local->selfheal.misc;
-
-	local->selfheal.dir_cbk = dir_cbk;
-	local->selfheal.layout = dht_layout_ref (this, layout);
-
-	if (down) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"%d subvolumes down -- not fixing", down);
-                ret = -1;
-                local->op_errno = ESTALE;
-		goto sorry_no_fix;
-	}
-
-	if (misc) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"%d subvolumes have unrecoverable errors", misc);
-                ret = -1;
-                local->op_errno = ESTALE;
-		goto sorry_no_fix;
-	}
-
-	dht_layout_sort_volname (layout);
-	ret = dht_selfheal_dir_getafix (frame, loc, layout);
-
-	if (ret == -1) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"not able to form layout for the directory");
-                ret = -1;
-                local->op_errno = ESTALE;
-		goto sorry_no_fix;
-	}
-
-	dht_selfheal_dir_mkdir (frame, loc, layout, 0);
-
-	return 0;
+dht_selfheal_directory(call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk,
+                       loc_t *loc, dht_layout_t *layout)
+{
+    dht_local_t *local = NULL;
+    xlator_t *this = NULL;
+    uint32_t down = 0;
+    uint32_t misc = 0;
+    int ret = 0;
+    char pgfid[GF_UUID_BUF_SIZE] = {0};
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    inode_t *linked_inode = NULL, *inode = NULL;
+
+    local = frame->local;
+    this = frame->this;
+
+    local->selfheal.dir_cbk = dir_cbk;
+    local->selfheal.layout = dht_layout_ref(this, layout);
+
+    if (local->need_attrheal) {
+        if (__is_root_gfid(local->stbuf.ia_gfid)) {
+            local->stbuf.ia_gid = local->prebuf.ia_gid;
+            local->stbuf.ia_uid = local->prebuf.ia_uid;
+
+            local->stbuf.ia_ctime = local->prebuf.ia_ctime;
+            local->stbuf.ia_ctime_nsec = local->prebuf.ia_ctime_nsec;
+            local->stbuf.ia_prot = local->prebuf.ia_prot;
+
+        } else if (!IA_ISINVAL(local->mds_stbuf.ia_type)) {
+            local->stbuf = local->mds_stbuf;
+        }
+    }
+
+    if (!__is_root_gfid(local->stbuf.ia_gfid)) {
+        gf_uuid_unparse(local->stbuf.ia_gfid, gfid);
+        gf_uuid_unparse(loc->parent->gfid, pgfid);
+
+        linked_inode = inode_link(loc->inode, loc->parent, loc->name,
+                                  &local->stbuf);
+        if (!linked_inode) {
+            gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_LINK_INODE_FAILED,
+                    "pgfid=%s", pgfid, "name=%s", loc->name, "gfid=%s", gfid,
+                    NULL);
+            ret = 0;
+            goto sorry_no_fix;
+        }
+
+        inode = loc->inode;
+        loc->inode = linked_inode;
+        inode_unref(inode);
+    }
+
+    if (local->need_xattr_heal && (local->mds_xattr)) {
+        dht_dir_set_heal_xattr(this, local, local->xattr, local->mds_xattr,
+                               NULL, NULL);
+        dict_unref(local->mds_xattr);
+        local->mds_xattr = NULL;
+    }
+
+    dht_layout_anomalies(this, loc, layout, &local->selfheal.hole_cnt,
+                         &local->selfheal.overlaps_cnt,
+                         &local->selfheal.missing_cnt, &local->selfheal.down,
+                         &local->selfheal.misc, NULL);
+
+    down = local->selfheal.down;
+    misc = local->selfheal.misc;
+
+    if (down) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_SELFHEAL_FAILED,
+                "path=%s", loc->path, "subvol-down=%d", down, "Not-fixing",
+                "gfid=%s", gfid, NULL);
+        ret = 0;
+        goto sorry_no_fix;
+    }
+
+    if (misc) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_SELFHEAL_FAILED,
+                "path=%s", loc->path, "misc=%d", misc, "unrecoverable-errors",
+                "gfid=%s", gfid, NULL);
+
+        ret = 0;
+        goto sorry_no_fix;
+    }
+
+    dht_layout_sort_volname(layout);
+    local->heal_layout = _gf_true;
+
+    /* Ignore return value as it can be inferred from result of
+     * dht_layout_anomalies
+     */
+    dht_selfheal_dir_getafix(frame, loc, layout);
+
+    if (!(local->selfheal.hole_cnt || local->selfheal.overlaps_cnt ||
+          local->selfheal.missing_cnt)) {
+        local->heal_layout = _gf_false;
+    }
+
+    ret = dht_selfheal_dir_mkdir(frame, loc, layout, 0);
+    if (ret < 0) {
+        ret = 0;
+        goto sorry_no_fix;
+    }
+
+    return 0;
 
 sorry_no_fix:
-	/* TODO: need to put appropriate local->op_errno */
-	dht_selfheal_dir_finish (frame, this, ret);
+    /* TODO: need to put appropriate local->op_errno */
+    dht_selfheal_dir_finish(frame, this, ret, 1);
+
+    return 0;
+}
 
-	return 0;
+int
+dht_selfheal_restore(call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk,
+                     loc_t *loc, dht_layout_t *layout)
+{
+    int ret = 0;
+    dht_local_t *local = NULL;
+
+    local = frame->local;
+
+    local->selfheal.dir_cbk = dir_cbk;
+    local->selfheal.layout = dht_layout_ref(frame->this, layout);
+
+    ret = dht_selfheal_dir_mkdir(frame, loc, layout, 1);
+
+    return ret;
+}
+
+int
+dht_dir_heal_xattrs(void *data)
+{
+    call_frame_t *frame = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *mds_subvol = NULL;
+    xlator_t *this = NULL;
+    dht_conf_t *conf = NULL;
+    dict_t *user_xattr = NULL;
+    dict_t *internal_xattr = NULL;
+    dict_t *mds_xattr = NULL;
+    dict_t *xdata = NULL;
+    int call_cnt = 0;
+    int ret = -1;
+    int uret = 0;
+    int uflag = 0;
+    int i = 0;
+    int xattr_hashed = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    int32_t allzero[1] = {0};
+
+    GF_VALIDATE_OR_GOTO("dht", data, out);
+
+    frame = data;
+    local = frame->local;
+    this = frame->this;
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, local, out);
+    mds_subvol = local->mds_subvol;
+    conf = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, conf, out);
+    gf_uuid_unparse(local->loc.gfid, gfid);
+
+    if (!mds_subvol) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_NO_MDS_SUBVOL, "path=%s",
+                local->loc.path, "gfid=%s", gfid, NULL);
+        goto out;
+    }
+
+    if ((local->loc.inode && gf_uuid_is_null(local->loc.inode->gfid)) ||
+        gf_uuid_is_null(local->loc.gfid)) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_GFID_NOT_PRESENT,
+                "skip-heal path=%s", local->loc.path, "gfid=%s", gfid, NULL);
+        goto out;
+    }
+
+    internal_xattr = dict_new();
+    if (!internal_xattr) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_CREATE_FAILED,
+                "dictionary", NULL);
+        goto out;
+    }
+    xdata = dict_new();
+    if (!xdata) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_CREATE_FAILED,
+                "dictionary", NULL);
+        goto out;
+    }
+
+    call_cnt = conf->subvolume_cnt;
+
+    user_xattr = dict_new();
+    if (!user_xattr) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_CREATE_FAILED,
+                "dictionary", NULL);
+        goto out;
+    }
+
+    ret = syncop_listxattr(local->mds_subvol, &local->loc, &mds_xattr, NULL,
+                           NULL);
+    if (ret < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_LIST_XATTRS_FAILED,
+                "path=%s", local->loc.path, "name=%s", local->mds_subvol->name,
+                NULL);
+    }
+
+    if (!mds_xattr)
+        goto out;
+
+    dht_dir_set_heal_xattr(this, local, user_xattr, mds_xattr, &uret, &uflag);
+
+    /* To set quota related xattr need to set GLUSTERFS_INTERNAL_FOP_KEY
+     * key value to 1
+     */
+    if (dict_get(user_xattr, QUOTA_LIMIT_KEY) ||
+        dict_get(user_xattr, QUOTA_LIMIT_OBJECTS_KEY)) {
+        ret = dict_set_int32(xdata, GLUSTERFS_INTERNAL_FOP_KEY, 1);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                    "key=%s", GLUSTERFS_INTERNAL_FOP_KEY, "path=%s",
+                    local->loc.path, NULL);
+            goto out;
+        }
+    }
+    if (uret <= 0 && !uflag)
+        goto out;
+
+    for (i = 0; i < call_cnt; i++) {
+        subvol = conf->subvolumes[i];
+        if (subvol == mds_subvol)
+            continue;
+        if (uret || uflag) {
+            /* Custom xattr heal is required - let posix handle it */
+            ret = dict_set_int8(xdata, "sync_backend_xattrs", _gf_true);
+            if (ret) {
+                gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+                        "path=%s", local->loc.path, "key=%s",
+                        "sync_backend_xattrs", NULL);
+                goto out;
+            }
+
+            ret = syncop_setxattr(subvol, &local->loc, user_xattr, 0, xdata,
+                                  NULL);
+            if (ret) {
+                xattr_hashed = 1;
+                gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                        DHT_MSG_DIR_XATTR_HEAL_FAILED,
+                        "set-user-xattr-failed path=%s", local->loc.path,
+                        "subvol=%s", subvol->name, "gfid=%s", gfid, NULL);
+            } else {
+                dict_del(xdata, "sync_backend_xattrs");
+            }
+        }
+    }
+    /* After heal all custom xattr reset internal MDS xattr to 0 */
+    if (!xattr_hashed) {
+        ret = dht_dict_set_array(internal_xattr, conf->mds_xattr_key, allzero,
+                                 1);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_WARNING, ENOMEM, DHT_MSG_DICT_SET_FAILED,
+                    "key=%s", conf->mds_xattr_key, "path=%s", local->loc.path,
+                    NULL);
+            goto out;
+        }
+        ret = syncop_setxattr(mds_subvol, &local->loc, internal_xattr, 0, NULL,
+                              NULL);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                    DHT_MSG_DIR_XATTR_HEAL_FAILED, "path=%s", local->loc.path,
+                    "subvol=%s", mds_subvol->name, "gfid=%s", gfid, NULL);
+        }
+    }
+
+out:
+    if (user_xattr)
+        dict_unref(user_xattr);
+    if (mds_xattr)
+        dict_unref(mds_xattr);
+    if (internal_xattr)
+        dict_unref(internal_xattr);
+    if (xdata)
+        dict_unref(xdata);
+    return 0;
+}
+
+int
+dht_dir_heal_xattrs_done(int ret, call_frame_t *sync_frame, void *data)
+{
+    DHT_STACK_DESTROY(sync_frame);
+    return 0;
 }
 
+int
+dht_dir_attr_heal(void *data)
+{
+    call_frame_t *frame = NULL;
+    dht_local_t *local = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *mds_subvol = NULL;
+    xlator_t *this = NULL;
+    dht_conf_t *conf = NULL;
+    int call_cnt = 0;
+    int ret = -1;
+    int i = 0;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    GF_VALIDATE_OR_GOTO("dht", data, out);
+
+    frame = data;
+    local = frame->local;
+    this = frame->this;
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", local, out);
+    conf = this->private;
+    GF_VALIDATE_OR_GOTO("dht", conf, out);
+
+    mds_subvol = local->mds_subvol;
+    call_cnt = conf->subvolume_cnt;
+
+    if (!__is_root_gfid(local->stbuf.ia_gfid) && (!mds_subvol)) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_NO_MDS_SUBVOL, "path=%s",
+                local->loc.path, "gfid=%s", gfid, NULL);
+        goto out;
+    }
+
+    if (!__is_root_gfid(local->stbuf.ia_gfid)) {
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (conf->subvolumes[i] == mds_subvol) {
+                if (!conf->subvolume_status[i]) {
+                    gf_smsg(this->name, GF_LOG_ERROR, 0,
+                            DHT_MSG_MDS_DOWN_UNABLE_TO_SET, "path=%s",
+                            local->loc.path, "gfid=%s", gfid, NULL);
+                    goto out;
+                }
+            }
+        }
+    }
+
+    for (i = 0; i < call_cnt; i++) {
+        subvol = conf->subvolumes[i];
+        if (!subvol || subvol == mds_subvol)
+            continue;
+        if (__is_root_gfid(local->stbuf.ia_gfid)) {
+            ret = syncop_setattr(
+                subvol, &local->loc, &local->stbuf,
+                (GF_SET_ATTR_UID | GF_SET_ATTR_GID | GF_SET_ATTR_MODE), NULL,
+                NULL, NULL, NULL);
+        } else {
+            ret = syncop_setattr(
+                subvol, &local->loc, &local->mds_stbuf,
+                (GF_SET_ATTR_UID | GF_SET_ATTR_GID | GF_SET_ATTR_MODE), NULL,
+                NULL, NULL, NULL);
+        }
+
+        if (ret) {
+            gf_uuid_unparse(local->loc.gfid, gfid);
+
+            gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                    DHT_MSG_DIR_ATTR_HEAL_FAILED, "path=%s", local->loc.path,
+                    "subvol=%s", subvol->name, "gfid=%s", gfid, NULL);
+        }
+    }
+out:
+    return 0;
+}
 
 int
-dht_selfheal_restore (call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk,
-		      loc_t *loc, dht_layout_t *layout)
+dht_dir_attr_heal_done(int ret, call_frame_t *sync_frame, void *data)
+{
+    DHT_STACK_DESTROY(sync_frame);
+    return 0;
+}
+
+/* EXIT: dht_update_commit_hash_for_layout */
+static int
+dht_update_commit_hash_for_layout_done(call_frame_t *frame, void *cookie,
+                                       xlator_t *this, int32_t op_ret,
+                                       int32_t op_errno, dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+
+    local = frame->local;
+
+    /* preserve oldest error */
+    if (op_ret && !local->op_ret) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+    }
+
+    DHT_STACK_UNWIND(setxattr, frame, local->op_ret, local->op_errno, NULL);
+
+    return 0;
+}
+
+static int
+dht_update_commit_hash_for_layout_unlock(call_frame_t *frame, xlator_t *this)
+{
+    dht_local_t *local = NULL;
+    int ret = 0;
+
+    local = frame->local;
+
+    ret = dht_unlock_inodelk(frame, local->lock[0].layout.my_layout.locks,
+                             local->lock[0].layout.my_layout.lk_count,
+                             dht_update_commit_hash_for_layout_done);
+    if (ret < 0) {
+        /* preserve oldest error, just ... */
+        if (!local->op_ret) {
+            local->op_errno = errno;
+            local->op_ret = -1;
+        }
+
+        gf_smsg(this->name, GF_LOG_WARNING, errno, DHT_MSG_WIND_UNLOCK_FAILED,
+                "path=%s", local->loc.path, NULL);
+
+        dht_update_commit_hash_for_layout_done(frame, NULL, this, 0, 0, NULL);
+    }
+
+    return 0;
+}
+
+static int
+dht_update_commit_hash_for_layout_cbk(call_frame_t *frame, void *cookie,
+                                      xlator_t *this, int op_ret, int op_errno,
+                                      dict_t *xdata)
+{
+    dht_local_t *local = NULL;
+    int this_call_cnt = 0;
+
+    local = frame->local;
+
+    LOCK(&frame->lock);
+    /* store first failure, just because */
+    if (op_ret && !local->op_ret) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+    }
+    UNLOCK(&frame->lock);
+
+    this_call_cnt = dht_frame_return(frame);
+
+    if (is_last_call(this_call_cnt)) {
+        dht_update_commit_hash_for_layout_unlock(frame, this);
+    }
+
+    return 0;
+}
+
+static int
+dht_update_commit_hash_for_layout_resume(call_frame_t *frame, void *cookie,
+                                         xlator_t *this, int32_t op_ret,
+                                         int32_t op_errno, dict_t *xdata)
 {
-	int          ret = 0;
-	dht_local_t *local    = NULL;
+    dht_local_t *local = NULL;
+    int count = 1, ret = -1, i = 0, j = 0;
+    dht_conf_t *conf = NULL;
+    dht_layout_t *layout = NULL;
+    int32_t *disk_layout = NULL;
+    dict_t **xattr = NULL;
+
+    local = frame->local;
+    conf = frame->this->private;
+    count = conf->local_subvols_cnt;
+    layout = local->layout;
+
+    if (op_ret < 0) {
+        goto err_done;
+    }
+
+    /* We precreate the xattr list as we cannot change call count post the
+     * first wind as we may never continue from there. So we finish prep
+     * work before winding the setxattrs */
+    xattr = GF_CALLOC(count, sizeof(*xattr), gf_common_mt_char);
+    if (!xattr) {
+        local->op_errno = errno;
+
+        gf_smsg(this->name, GF_LOG_WARNING, errno, DHT_MSG_COMMIT_HASH_FAILED,
+                "allocation-failed path=%s", local->loc.path, NULL);
+
+        goto err;
+    }
+
+    for (i = 0; i < count; i++) {
+        /* find the layout index for the subvolume */
+        ret = dht_layout_index_for_subvol(layout, conf->local_subvols[i]);
+        if (ret < 0) {
+            local->op_errno = ENOENT;
+
+            gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_COMMIT_HASH_FAILED,
+                    "path=%s", local->loc.path, "subvol=%s",
+                    conf->local_subvols[i]->name, "find-disk-layout-failed",
+                    NULL);
+
+            goto err;
+        }
+        j = ret;
+
+        /* update the commit hash for the layout */
+        layout->list[j].commit_hash = layout->commit_hash;
+
+        /* extract the current layout */
+        ret = dht_disk_layout_extract(this, layout, j, &disk_layout);
+        if (ret == -1) {
+            local->op_errno = errno;
+
+            gf_smsg(this->name, GF_LOG_WARNING, errno,
+                    DHT_MSG_COMMIT_HASH_FAILED, "path=%s", local->loc.path,
+                    "subvol=%s", conf->local_subvols[i]->name,
+                    "extract-disk-layout-failed", NULL);
 
+            goto err;
+        }
+
+        xattr[i] = dict_new();
+        if (!xattr[i]) {
+            local->op_errno = errno;
+
+            gf_smsg(this->name, GF_LOG_WARNING, errno,
+                    DHT_MSG_COMMIT_HASH_FAILED, "path=%s Allocation-failed",
+                    local->loc.path, NULL);
+
+            goto err;
+        }
+
+        ret = dict_set_bin(xattr[i], conf->xattr_name, disk_layout, 4 * 4);
+        if (ret != 0) {
+            local->op_errno = ENOMEM;
+
+            gf_smsg(this->name, GF_LOG_WARNING, 0,
+                    DHT_MSG_DIR_SELFHEAL_XATTR_FAILED, "path=%s",
+                    local->loc.path, "subvol=%s", conf->local_subvols[i]->name,
+                    "set-xattr-failed", NULL);
+
+            goto err;
+        }
+        disk_layout = NULL;
+
+        gf_msg_trace(this->name, 0,
+                     "setting commit hash %u on subvolume %s"
+                     " for %s",
+                     layout->list[j].commit_hash, conf->local_subvols[i]->name,
+                     local->loc.path);
+    }
+
+    /* wind the setting of the commit hash across the local subvols */
+    local->call_cnt = count;
+    local->op_ret = 0;
+    local->op_errno = 0;
+    for (i = 0; i < count; i++) {
+        STACK_WIND(frame, dht_update_commit_hash_for_layout_cbk,
+                   conf->local_subvols[i],
+                   conf->local_subvols[i]->fops->setxattr, &local->loc,
+                   xattr[i], 0, NULL);
+    }
+    for (i = 0; i < count; i++)
+        dict_unref(xattr[i]);
+    GF_FREE(xattr);
+
+    return 0;
+err:
+    if (xattr) {
+        for (i = 0; i < count; i++) {
+            if (xattr[i])
+                dict_unref(xattr[i]);
+        }
+
+        GF_FREE(xattr);
+    }
+
+    GF_FREE(disk_layout);
+
+    local->op_ret = -1;
+
+    dht_update_commit_hash_for_layout_unlock(frame, this);
 
-	local = frame->local;
+    return 0;
+err_done:
+    local->op_ret = -1;
 
-	local->selfheal.dir_cbk = dir_cbk;
-	local->selfheal.layout = dht_layout_ref (frame->this, layout);
+    dht_update_commit_hash_for_layout_done(frame, NULL, this, 0, 0, NULL);
 
-	ret = dht_selfheal_dir_mkdir (frame, loc, layout, 1);
+    return 0;
+}
+
+/* ENTER: dht_update_commit_hash_for_layout (see EXIT above)
+ * This function is invoked from rebalance only.
+ * As a result, the check here is simple enough to see if defrag is present
+ * in the conf, as other data would be populated appropriately if so.
+ * If ever this was to be used in other code paths, checks would need to
+ * change.
+ *
+ * Functional details:
+ *  - Lock the inodes on the subvols that we want the commit hash updated
+ *  - Update each layout with the inode layout, modified to take in the new
+ *    commit hash.
+ *  - Unlock and return.
+ */
+int
+dht_update_commit_hash_for_layout(call_frame_t *frame)
+{
+    dht_local_t *local = NULL;
+    int count = 1, ret = -1, i = 0;
+    dht_lock_t **lk_array = NULL;
+    dht_conf_t *conf = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, err);
+    GF_VALIDATE_OR_GOTO(frame->this->name, frame->local, err);
+
+    local = frame->local;
+    conf = frame->this->private;
+
+    if (!conf->defrag)
+        goto err;
+
+    count = conf->local_subvols_cnt;
+    lk_array = GF_CALLOC(count, sizeof(*lk_array), gf_common_mt_char);
+    if (lk_array == NULL)
+        goto err;
+
+    for (i = 0; i < count; i++) {
+        lk_array[i] = dht_lock_new(frame->this, conf->local_subvols[i],
+                                   &local->loc, F_WRLCK, DHT_LAYOUT_HEAL_DOMAIN,
+                                   NULL, FAIL_ON_ANY_ERROR);
+        if (lk_array[i] == NULL)
+            goto err;
+    }
+
+    local->lock[0].layout.my_layout.locks = lk_array;
+    local->lock[0].layout.my_layout.lk_count = count;
+
+    ret = dht_blocking_inodelk(frame, lk_array, count,
+                               dht_update_commit_hash_for_layout_resume);
+    if (ret < 0) {
+        local->lock[0].layout.my_layout.locks = NULL;
+        local->lock[0].layout.my_layout.lk_count = 0;
+        goto err;
+    }
+
+    return 0;
+err:
+    if (lk_array != NULL) {
+        dht_lock_array_free(lk_array, count);
+        GF_FREE(lk_array);
+    }
 
-	return ret;
+    return -1;
 }
diff --git a/xlators/cluster/dht/src/dht-shared.c b/xlators/cluster/dht/src/dht-shared.c
new file mode 100644
index 00000000000..bb72b0ffbb5
--- /dev/null
+++ b/xlators/cluster/dht/src/dht-shared.c
@@ -0,0 +1,1104 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+/* TODO: add NS locking */
+#include <glusterfs/statedump.h>
+#include "dht-common.h"
+#include "dht-messages.h"
+
+#ifndef MAX
+#define MAX(a, b) (((a) > (b)) ? (a) : (b))
+#endif
+
+/* TODO:
+   - use volumename in xattr instead of "dht"
+   - use NS locks
+   - handle all cases in self heal layout reconstruction
+   - complete linkfile selfheal
+*/
+
+static void
+dht_layout_dump(dht_layout_t *layout, const char *prefix)
+{
+    char key[GF_DUMP_MAX_BUF_LEN];
+    int i = 0;
+
+    if (!layout)
+        goto out;
+
+    gf_proc_dump_build_key(key, prefix, "cnt");
+    gf_proc_dump_write(key, "%d", layout->cnt);
+    gf_proc_dump_build_key(key, prefix, "preset");
+    gf_proc_dump_write(key, "%d", layout->preset);
+    gf_proc_dump_build_key(key, prefix, "gen");
+    gf_proc_dump_write(key, "%d", layout->gen);
+    if (layout->type != IA_INVAL) {
+        gf_proc_dump_build_key(key, prefix, "inode type");
+        gf_proc_dump_write(key, "%d", layout->type);
+    }
+
+    if (!IA_ISDIR(layout->type))
+        goto out;
+
+    for (i = 0; i < layout->cnt; i++) {
+        gf_proc_dump_build_key(key, prefix, "list[%d].err", i);
+        gf_proc_dump_write(key, "%d", layout->list[i].err);
+        gf_proc_dump_build_key(key, prefix, "list[%d].start", i);
+        gf_proc_dump_write(key, "0x%x", layout->list[i].start);
+        gf_proc_dump_build_key(key, prefix, "list[%d].stop", i);
+        gf_proc_dump_write(key, "0x%x", layout->list[i].stop);
+        if (layout->list[i].xlator) {
+            gf_proc_dump_build_key(key, prefix, "list[%d].xlator.type", i);
+            gf_proc_dump_write(key, "%s", layout->list[i].xlator->type);
+            gf_proc_dump_build_key(key, prefix, "list[%d].xlator.name", i);
+            gf_proc_dump_write(key, "%s", layout->list[i].xlator->name);
+        }
+    }
+
+out:
+    return;
+}
+
+int32_t
+dht_priv_dump(xlator_t *this)
+{
+    char key_prefix[GF_DUMP_MAX_BUF_LEN];
+    char key[GF_DUMP_MAX_BUF_LEN];
+    int i = 0;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+
+    if (!this)
+        goto out;
+
+    conf = this->private;
+    if (!conf)
+        goto out;
+
+    ret = TRY_LOCK(&conf->subvolume_lock);
+    if (ret != 0) {
+        return ret;
+    }
+
+    gf_proc_dump_add_section("xlator.cluster.dht.%s.priv", this->name);
+    gf_proc_dump_build_key(key_prefix, "xlator.cluster.dht", "%s.priv",
+                           this->name);
+    gf_proc_dump_write("subvol_cnt", "%d", conf->subvolume_cnt);
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        snprintf(key, sizeof(key), "subvolumes[%d]", i);
+        gf_proc_dump_write(key, "%s.%s", conf->subvolumes[i]->type,
+                           conf->subvolumes[i]->name);
+        if (conf->file_layouts && conf->file_layouts[i]) {
+            snprintf(key, sizeof(key), "file_layouts[%d]", i);
+            dht_layout_dump(conf->file_layouts[i], key);
+        }
+        if (conf->dir_layouts && conf->dir_layouts[i]) {
+            snprintf(key, sizeof(key), "dir_layouts[%d]", i);
+            dht_layout_dump(conf->dir_layouts[i], key);
+        }
+        if (conf->subvolume_status) {
+            snprintf(key, sizeof(key), "subvolume_status[%d]", i);
+            gf_proc_dump_write(key, "%d", (int)conf->subvolume_status[i]);
+        }
+    }
+
+    gf_proc_dump_write("search_unhashed", "%d", conf->search_unhashed);
+    gf_proc_dump_write("gen", "%d", conf->gen);
+    gf_proc_dump_write("min_free_disk", "%lf", conf->min_free_disk);
+    gf_proc_dump_write("min_free_inodes", "%lf", conf->min_free_inodes);
+    gf_proc_dump_write("disk_unit", "%c", conf->disk_unit);
+    gf_proc_dump_write("refresh_interval", "%d", conf->refresh_interval);
+    gf_proc_dump_write("unhashed_sticky_bit", "%d", conf->unhashed_sticky_bit);
+    gf_proc_dump_write("use-readdirp", "%d", conf->use_readdirp);
+
+    if (conf->du_stats && conf->subvolume_status) {
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (!conf->subvolume_status[i])
+                continue;
+
+            snprintf(key, sizeof(key), "subvolumes[%d]", i);
+            gf_proc_dump_write(key, "%s", conf->subvolumes[i]->name);
+
+            snprintf(key, sizeof(key), "du_stats[%d].avail_percent", i);
+            gf_proc_dump_write(key, "%lf", conf->du_stats[i].avail_percent);
+
+            snprintf(key, sizeof(key), "du_stats[%d].avail_space", i);
+            gf_proc_dump_write(key, "%" PRIu64, conf->du_stats[i].avail_space);
+
+            snprintf(key, sizeof(key), "du_stats[%d].avail_inodes", i);
+            gf_proc_dump_write(key, "%lf", conf->du_stats[i].avail_inodes);
+
+            snprintf(key, sizeof(key), "du_stats[%d].log", i);
+            gf_proc_dump_write(key, "%" PRIu32, conf->du_stats[i].log);
+        }
+    }
+
+    if (conf->last_stat_fetch)
+        gf_proc_dump_write("last_stat_fetch", "%s",
+                           ctime(&conf->last_stat_fetch));
+
+    UNLOCK(&conf->subvolume_lock);
+
+out:
+    return ret;
+}
+
+int32_t
+dht_inodectx_dump(xlator_t *this, inode_t *inode)
+{
+    int ret = -1;
+    dht_layout_t *layout = NULL;
+
+    if (!this)
+        goto out;
+    if (!inode)
+        goto out;
+
+    ret = dht_inode_ctx_layout_get(inode, this, &layout);
+
+    if ((ret != 0) || !layout)
+        return ret;
+
+    gf_proc_dump_add_section("xlator.cluster.dht.%s.inode", this->name);
+    dht_layout_dump(layout, "layout");
+
+out:
+    return ret;
+}
+
+void
+dht_fini(xlator_t *this)
+{
+    int i = 0;
+    dht_conf_t *conf = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+
+    conf = this->private;
+    this->private = NULL;
+    if (conf) {
+        if (conf->file_layouts) {
+            for (i = 0; i < conf->subvolume_cnt; i++) {
+                GF_FREE(conf->file_layouts[i]);
+            }
+            GF_FREE(conf->file_layouts);
+        }
+
+        dict_unref(conf->leaf_to_subvol);
+
+        /* allocated in dht_init_subvolumes() */
+        GF_FREE(conf->subvolumes);
+        GF_FREE(conf->subvolume_status);
+        GF_FREE(conf->last_event);
+        GF_FREE(conf->subvol_up_time);
+        GF_FREE(conf->du_stats);
+        GF_FREE(conf->decommissioned_bricks);
+
+        /* allocated in dht_init() */
+        GF_FREE(conf->mds_xattr_key);
+        GF_FREE(conf->link_xattr_name);
+        GF_FREE(conf->commithash_xattr_name);
+        GF_FREE(conf->wild_xattr_name);
+
+        /* allocated in dht_init_regex() */
+        if (conf->rsync_regex_valid)
+            regfree(&conf->rsync_regex);
+        if (conf->extra_regex_valid)
+            regfree(&conf->extra_regex);
+
+        synclock_destroy(&conf->link_lock);
+
+        if (conf->lock_pool)
+            mem_pool_destroy(conf->lock_pool);
+
+        GF_FREE(conf);
+    }
+out:
+    return;
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+
+    ret = xlator_mem_acct_init(this, gf_dht_mt_end + 1);
+
+    if (ret != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_NO_MEMORY,
+               "Memory accounting init failed");
+        return ret;
+    }
+out:
+    return ret;
+}
+
+static int
+dht_parse_decommissioned_bricks(xlator_t *this, dht_conf_t *conf,
+                                const char *bricks)
+{
+    int i = 0;
+    int ret = -1;
+    char *tmpstr = NULL;
+    char *dup_brick = NULL;
+    char *node = NULL;
+
+    if (!conf || !bricks)
+        goto out;
+
+    dup_brick = gf_strdup(bricks);
+    if (dup_brick == NULL) {
+        goto out;
+    }
+
+    node = strtok_r(dup_brick, ",", &tmpstr);
+    while (node) {
+        for (i = 0; i < conf->subvolume_cnt; i++) {
+            if (!strcmp(conf->subvolumes[i]->name, node)) {
+                conf->decommissioned_bricks[i] = conf->subvolumes[i];
+                conf->decommission_subvols_cnt++;
+                gf_msg(this->name, GF_LOG_INFO, 0,
+                       DHT_MSG_SUBVOL_DECOMMISSION_INFO,
+                       "decommissioning subvolume %s",
+                       conf->subvolumes[i]->name);
+                break;
+            }
+        }
+        if (i == conf->subvolume_cnt) {
+            /* Wrong node given. */
+            goto out;
+        }
+        node = strtok_r(NULL, ",", &tmpstr);
+    }
+
+    ret = 0;
+    conf->decommission_in_progress = 1;
+out:
+    GF_FREE(dup_brick);
+
+    return ret;
+}
+
+static void
+dht_decommissioned_remove(xlator_t *this, dht_conf_t *conf)
+{
+    int i = 0;
+
+    for (i = 0; i < conf->subvolume_cnt; i++) {
+        if (conf->decommissioned_bricks[i]) {
+            conf->decommissioned_bricks[i] = NULL;
+            conf->decommission_subvols_cnt--;
+        }
+    }
+}
+
+static void
+dht_init_regex(xlator_t *this, dict_t *odict, char *name, regex_t *re,
+               gf_boolean_t *re_valid, dht_conf_t *conf)
+{
+    char *temp_str = NULL;
+
+    if (dict_get_str(odict, name, &temp_str) != 0) {
+        if (strcmp(name, "rsync-hash-regex")) {
+            return;
+        }
+        temp_str = "^\\.(.+)\\.[^.]+$";
+    }
+
+    LOCK(&conf->lock);
+    {
+        if (*re_valid) {
+            regfree(re);
+            *re_valid = _gf_false;
+        }
+
+        if (!strcmp(temp_str, "none")) {
+            goto unlock;
+        }
+
+        if (regcomp(re, temp_str, REG_EXTENDED) == 0) {
+            gf_msg_debug(this->name, 0, "using regex %s = %s", name, temp_str);
+            *re_valid = _gf_true;
+        } else {
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_REGEX_INFO,
+                   "compiling regex %s failed", temp_str);
+        }
+    }
+unlock:
+    UNLOCK(&conf->lock);
+}
+
+int
+dht_set_subvol_range(xlator_t *this)
+{
+    int ret = -1;
+    dht_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf)
+        goto out;
+
+    conf->leaf_to_subvol = dict_new();
+    if (!conf->leaf_to_subvol)
+        goto out;
+
+    ret = glusterfs_reachable_leaves(this, conf->leaf_to_subvol);
+
+out:
+    return ret;
+}
+
+static int
+dht_configure_throttle(xlator_t *this, dht_conf_t *conf, char *temp_str)
+{
+    int rebal_thread_count = 0;
+    int ret = 0;
+
+    pthread_mutex_lock(&conf->defrag->dfq_mutex);
+    {
+        if (!strcasecmp(temp_str, "lazy")) {
+            conf->defrag->recon_thread_count = 1;
+        } else if (!strcasecmp(temp_str, "normal")) {
+            conf->defrag->recon_thread_count = 2;
+        } else if (!strcasecmp(temp_str, "aggressive")) {
+            conf->defrag->recon_thread_count = MAX(MAX_REBAL_THREADS - 4, 4);
+        } else if ((gf_string2int(temp_str, &rebal_thread_count) == 0)) {
+            if ((rebal_thread_count > 0) &&
+                (rebal_thread_count <= MAX_REBAL_THREADS)) {
+                conf->defrag->recon_thread_count = rebal_thread_count;
+                pthread_mutex_unlock(&conf->defrag->dfq_mutex);
+                gf_msg(this->name, GF_LOG_INFO, 0, 0,
+                       "rebal thread count configured to %d",
+                       rebal_thread_count);
+                goto out;
+            } else {
+                pthread_mutex_unlock(&conf->defrag->dfq_mutex);
+                gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INVALID_OPTION,
+                       "Invalid option: Reconfigure: "
+                       "rebal-throttle should be "
+                       "within range of 0 and maximum number of"
+                       " cores available");
+                ret = -1;
+                goto out;
+            }
+        } else {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INVALID_OPTION,
+                   "Invalid option: Reconfigure: "
+                   "rebal-throttle should be {lazy|normal|aggressive}"
+                   " or a number up to the number of cores available,"
+                   " not (%s), defaulting to (%d)",
+                   temp_str, conf->dthrottle);
+            ret = -1;
+        }
+    }
+    pthread_mutex_unlock(&conf->defrag->dfq_mutex);
+
+out:
+    return ret;
+}
+
+int
+dht_reconfigure(xlator_t *this, dict_t *options)
+{
+    dht_conf_t *conf = NULL;
+    char *temp_str = NULL;
+    gf_boolean_t search_unhashed;
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", options, out);
+
+    conf = this->private;
+    if (!conf)
+        return 0;
+
+    if (dict_get_str(options, "lookup-unhashed", &temp_str) == 0) {
+        /* If option is not "auto", other options _should_ be boolean*/
+        if (strcasecmp(temp_str, "auto")) {
+            if (!gf_string2boolean(temp_str, &search_unhashed)) {
+                gf_msg_debug(this->name, 0,
+                             "Reconfigure: "
+                             "lookup-unhashed reconfigured(%s)",
+                             temp_str);
+                conf->search_unhashed = search_unhashed;
+            } else {
+                gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INVALID_OPTION,
+                       "Invalid option: Reconfigure: "
+                       "lookup-unhashed should be boolean,"
+                       " not (%s), defaulting to (%d)",
+                       temp_str, conf->search_unhashed);
+                ret = -1;
+                goto out;
+            }
+        } else {
+            gf_msg_debug(this->name, 0,
+                         "Reconfigure:"
+                         " lookup-unhashed reconfigured auto ");
+            conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
+        }
+    }
+
+    GF_OPTION_RECONF("lookup-optimize", conf->lookup_optimize, options, bool,
+                     out);
+
+    GF_OPTION_RECONF("min-free-disk", conf->min_free_disk, options,
+                     percent_or_size, out);
+    /* option can be any one of percent or bytes */
+    conf->disk_unit = 0;
+    if (conf->min_free_disk < 100.0)
+        conf->disk_unit = 'p';
+
+    GF_OPTION_RECONF("min-free-inodes", conf->min_free_inodes, options, percent,
+                     out);
+
+    GF_OPTION_RECONF("directory-layout-spread", conf->dir_spread_cnt, options,
+                     uint32, out);
+
+    GF_OPTION_RECONF("readdir-optimize", conf->readdir_optimize, options, bool,
+                     out);
+    GF_OPTION_RECONF("randomize-hash-range-by-gfid", conf->randomize_by_gfid,
+                     options, bool, out);
+
+    GF_OPTION_RECONF("lock-migration", conf->lock_migration_enabled, options,
+                     bool, out);
+
+    GF_OPTION_RECONF("force-migration", conf->force_migration, options, bool,
+                     out);
+
+    if (conf->defrag) {
+        if (dict_get_str(options, "rebal-throttle", &temp_str) == 0) {
+            ret = dht_configure_throttle(this, conf, temp_str);
+            if (ret == -1)
+                goto out;
+        }
+    }
+
+    if (conf->defrag) {
+        conf->defrag->lock_migration_enabled = conf->lock_migration_enabled;
+    }
+
+    if (conf->defrag) {
+        GF_OPTION_RECONF("rebalance-stats", conf->defrag->stats, options, bool,
+                         out);
+    }
+
+    if (dict_get_str(options, "decommissioned-bricks", &temp_str) == 0) {
+        ret = dht_parse_decommissioned_bricks(this, conf, temp_str);
+        if (ret == -1)
+            goto out;
+    } else {
+        dht_decommissioned_remove(this, conf);
+    }
+
+    dht_init_regex(this, options, "rsync-hash-regex", &conf->rsync_regex,
+                   &conf->rsync_regex_valid, conf);
+    dht_init_regex(this, options, "extra-hash-regex", &conf->extra_regex,
+                   &conf->extra_regex_valid, conf);
+
+    GF_OPTION_RECONF("weighted-rebalance", conf->do_weighting, options, bool,
+                     out);
+
+    GF_OPTION_RECONF("use-readdirp", conf->use_readdirp, options, bool, out);
+    ret = 0;
+out:
+    return ret;
+}
+
+static int
+gf_defrag_pattern_list_fill(xlator_t *this, gf_defrag_info_t *defrag,
+                            char *data)
+{
+    int ret = -1;
+    char *tmp_str = NULL;
+    char *tmp_str1 = NULL;
+    char *dup_str = NULL;
+    char *num = NULL;
+    char *pattern_str = NULL;
+    char *pattern = NULL;
+    gf_defrag_pattern_list_t *temp_list = NULL;
+    gf_defrag_pattern_list_t *pattern_list = NULL;
+
+    if (!this || !defrag || !data)
+        goto out;
+
+    /* Get the pattern for pattern list. "pattern:<optional-size>"
+     * eg: *avi, *pdf:10MB, *:1TB
+     */
+    pattern_str = strtok_r(data, ",", &tmp_str);
+    while (pattern_str) {
+        dup_str = gf_strdup(pattern_str);
+        if (!dup_str)
+            goto out;
+        pattern_list = GF_CALLOC(1, sizeof(gf_defrag_pattern_list_t), 1);
+        if (!pattern_list) {
+            goto out;
+        }
+        pattern = strtok_r(dup_str, ":", &tmp_str1);
+        num = strtok_r(NULL, ":", &tmp_str1);
+        if (!pattern)
+            goto out;
+        if (!num) {
+            if (gf_string2bytesize_uint64(pattern, &pattern_list->size) == 0) {
+                pattern = "*";
+            }
+        } else if (gf_string2bytesize_uint64(num, &pattern_list->size) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INVALID_OPTION,
+                   "Invalid option. Defrag pattern:"
+                   " Invalid number format \"%s\"",
+                   num);
+            goto out;
+        }
+        memcpy(pattern_list->path_pattern, pattern, strlen(dup_str));
+
+        if (!defrag->defrag_pattern)
+            temp_list = NULL;
+        else
+            temp_list = defrag->defrag_pattern;
+
+        pattern_list->next = temp_list;
+
+        defrag->defrag_pattern = pattern_list;
+        pattern_list = NULL;
+
+        GF_FREE(dup_str);
+        dup_str = NULL;
+
+        pattern_str = strtok_r(NULL, ",", &tmp_str);
+    }
+
+    ret = 0;
+out:
+    if (ret)
+        GF_FREE(pattern_list);
+    GF_FREE(dup_str);
+
+    return ret;
+}
+
+static int
+dht_init_methods(xlator_t *this)
+{
+    int ret = -1;
+    dht_conf_t *conf = NULL;
+    dht_methods_t *methods = NULL;
+
+    GF_VALIDATE_OR_GOTO("dht", this, err);
+
+    conf = this->private;
+    methods = &(conf->methods);
+
+    methods->migration_get_dst_subvol = dht_migration_get_dst_subvol;
+    methods->migration_other = NULL;
+    methods->layout_search = dht_layout_search;
+
+    ret = 0;
+err:
+    return ret;
+}
+
+int
+dht_init(xlator_t *this)
+{
+    dht_conf_t *conf = NULL;
+    char *temp_str = NULL;
+    int ret = -1;
+    int i = 0;
+    gf_defrag_info_t *defrag = NULL;
+    int cmd = 0;
+    char *node_uuid = NULL;
+    uint32_t commit_hash = 0;
+
+    GF_VALIDATE_OR_GOTO("dht", this, err);
+
+    if (!this->children) {
+        gf_msg(this->name, GF_LOG_CRITICAL, 0, DHT_MSG_INVALID_CONFIGURATION,
+               "Distribute needs more than one subvolume");
+        return -1;
+    }
+
+    if (!this->parents) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_INVALID_CONFIGURATION,
+               "dangling volume. check volfile");
+    }
+
+    conf = GF_CALLOC(1, sizeof(*conf), gf_dht_mt_dht_conf_t);
+    if (!conf) {
+        goto err;
+    }
+
+    LOCK_INIT(&conf->subvolume_lock);
+    LOCK_INIT(&conf->layout_lock);
+    LOCK_INIT(&conf->lock);
+    synclock_init(&conf->link_lock, SYNC_LOCK_DEFAULT);
+
+    /* We get the commit-hash to set only for rebalance process */
+    if (dict_get_uint32(this->options, "commit-hash", &commit_hash) == 0) {
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_COMMIT_HASH_INFO,
+               "%s using commit hash %u", __func__, commit_hash);
+        conf->vol_commit_hash = commit_hash;
+        conf->vch_forced = _gf_true;
+    }
+
+    ret = dict_get_int32(this->options, "rebalance-cmd", &cmd);
+
+    if (cmd) {
+        defrag = GF_CALLOC(1, sizeof(gf_defrag_info_t), gf_defrag_info_mt);
+
+        GF_VALIDATE_OR_GOTO(this->name, defrag, err);
+
+        LOCK_INIT(&defrag->lock);
+
+        defrag->is_exiting = 0;
+
+        conf->defrag = defrag;
+        defrag->this = this;
+
+        ret = dict_get_str(this->options, "node-uuid", &node_uuid);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INVALID_CONFIGURATION,
+                   "Invalid volume configuration: "
+                   "node-uuid not specified");
+            goto err;
+        }
+
+        if (gf_uuid_parse(node_uuid, defrag->node_uuid)) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INVALID_OPTION,
+                   "Invalid option:"
+                   " Cannot parse glusterd node uuid");
+            goto err;
+        }
+
+        defrag->cmd = cmd;
+
+        defrag->stats = _gf_false;
+
+        defrag->queue = NULL;
+
+        defrag->crawl_done = 0;
+
+        defrag->global_error = 0;
+
+        defrag->q_entry_count = 0;
+
+        defrag->wakeup_crawler = 0;
+
+        pthread_mutex_init(&defrag->dfq_mutex, 0);
+        pthread_cond_init(&defrag->parallel_migration_cond, 0);
+        pthread_cond_init(&defrag->rebalance_crawler_alarm, 0);
+        pthread_cond_init(&defrag->df_wakeup_thread, 0);
+
+        pthread_mutex_init(&defrag->fc_mutex, 0);
+        pthread_cond_init(&defrag->fc_wakeup_cond, 0);
+
+        defrag->global_error = 0;
+    }
+
+    conf->use_fallocate = 1;
+
+    conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_ON;
+    if (dict_get_str(this->options, "lookup-unhashed", &temp_str) == 0) {
+        /* If option is not "auto", other options _should_ be boolean */
+        if (strcasecmp(temp_str, "auto")) {
+            gf_boolean_t search_unhashed_bool;
+            ret = gf_string2boolean(temp_str, &search_unhashed_bool);
+            if (ret == -1) {
+                goto err;
+            }
+            conf->search_unhashed = search_unhashed_bool
+                                        ? GF_DHT_LOOKUP_UNHASHED_ON
+                                        : GF_DHT_LOOKUP_UNHASHED_OFF;
+        } else {
+            conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
+        }
+    }
+
+    GF_OPTION_INIT("lookup-optimize", conf->lookup_optimize, bool, err);
+
+    GF_OPTION_INIT("unhashed-sticky-bit", conf->unhashed_sticky_bit, bool, err);
+
+    GF_OPTION_INIT("use-readdirp", conf->use_readdirp, bool, err);
+
+    GF_OPTION_INIT("min-free-disk", conf->min_free_disk, percent_or_size, err);
+
+    GF_OPTION_INIT("min-free-inodes", conf->min_free_inodes, percent, err);
+
+    conf->dir_spread_cnt = conf->subvolume_cnt;
+    GF_OPTION_INIT("directory-layout-spread", conf->dir_spread_cnt, uint32,
+                   err);
+
+    GF_OPTION_INIT("assert-no-child-down", conf->assert_no_child_down, bool,
+                   err);
+
+    GF_OPTION_INIT("readdir-optimize", conf->readdir_optimize, bool, err);
+
+    GF_OPTION_INIT("lock-migration", conf->lock_migration_enabled, bool, err);
+
+    GF_OPTION_INIT("force-migration", conf->force_migration, bool, err);
+
+    if (defrag) {
+        defrag->lock_migration_enabled = conf->lock_migration_enabled;
+
+        GF_OPTION_INIT("rebalance-stats", defrag->stats, bool, err);
+        if (dict_get_str(this->options, "rebalance-filter", &temp_str) == 0) {
+            if (gf_defrag_pattern_list_fill(this, defrag, temp_str) == -1) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INVALID_OPTION,
+                       "Invalid option:"
+                       " Cannot parse rebalance-filter (%s)",
+                       temp_str);
+
+                goto err;
+            }
+        }
+    }
+
+    /* option can be any one of percent or bytes */
+    conf->disk_unit = 0;
+    if (conf->min_free_disk < 100)
+        conf->disk_unit = 'p';
+
+    ret = dht_init_subvolumes(this, conf);
+    if (ret == -1) {
+        goto err;
+    }
+
+    if (cmd) {
+        ret = dht_init_local_subvolumes(this, conf);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   DHT_MSG_INIT_LOCAL_SUBVOL_FAILED,
+                   "dht_init_local_subvolumes failed");
+            goto err;
+        }
+    }
+
+    if (dict_get_str(this->options, "decommissioned-bricks", &temp_str) == 0) {
+        ret = dht_parse_decommissioned_bricks(this, conf, temp_str);
+        if (ret == -1)
+            goto err;
+    }
+
+    dht_init_regex(this, this->options, "rsync-hash-regex", &conf->rsync_regex,
+                   &conf->rsync_regex_valid, conf);
+    dht_init_regex(this, this->options, "extra-hash-regex", &conf->extra_regex,
+                   &conf->extra_regex_valid, conf);
+
+    ret = dht_layouts_init(this, conf);
+    if (ret == -1) {
+        goto err;
+    }
+
+    conf->gen = 1;
+
+    this->local_pool = mem_pool_new(dht_local_t, 512);
+    if (!this->local_pool) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_NO_MEMORY,
+               " DHT initialisation failed. "
+               "failed to create local_t's memory pool");
+        goto err;
+    }
+
+    GF_OPTION_INIT("randomize-hash-range-by-gfid", conf->randomize_by_gfid,
+                   bool, err);
+
+    if (defrag) {
+        GF_OPTION_INIT("rebal-throttle", temp_str, str, err);
+        if (temp_str) {
+            ret = dht_configure_throttle(this, conf, temp_str);
+            if (ret == -1)
+                goto err;
+        }
+    }
+
+    GF_OPTION_INIT("xattr-name", conf->xattr_name, str, err);
+    gf_asprintf(&conf->mds_xattr_key, "%s." DHT_MDS_STR, conf->xattr_name);
+    gf_asprintf(&conf->link_xattr_name, "%s." DHT_LINKFILE_STR,
+                conf->xattr_name);
+    gf_asprintf(&conf->commithash_xattr_name, "%s." DHT_COMMITHASH_STR,
+                conf->xattr_name);
+    gf_asprintf(&conf->wild_xattr_name, "%s*", conf->xattr_name);
+    if (!conf->link_xattr_name || !conf->wild_xattr_name) {
+        goto err;
+    }
+
+    GF_OPTION_INIT("weighted-rebalance", conf->do_weighting, bool, err);
+
+    conf->lock_pool = mem_pool_new(dht_lock_t, 512);
+    if (!conf->lock_pool) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_INIT_FAILED,
+               "failed to create lock mem_pool, failing "
+               "initialization");
+        goto err;
+    }
+
+    this->private = conf;
+
+    if (dht_set_subvol_range(this))
+        goto err;
+
+    if (dht_init_methods(this))
+        goto err;
+
+    return 0;
+
+err:
+    if (conf) {
+        if (conf->file_layouts) {
+            for (i = 0; i < conf->subvolume_cnt; i++) {
+                GF_FREE(conf->file_layouts[i]);
+            }
+            GF_FREE(conf->file_layouts);
+        }
+
+        GF_FREE(conf->subvolumes);
+
+        GF_FREE(conf->subvolume_status);
+
+        GF_FREE(conf->du_stats);
+
+        GF_FREE(conf->defrag);
+
+        GF_FREE(conf->xattr_name);
+        GF_FREE(conf->link_xattr_name);
+        GF_FREE(conf->wild_xattr_name);
+        GF_FREE(conf->mds_xattr_key);
+
+        if (conf->lock_pool)
+            mem_pool_destroy(conf->lock_pool);
+
+        GF_FREE(conf);
+    }
+
+    return -1;
+}
+
+struct volume_options dht_options[] = {
+    {
+        .key = {"lookup-unhashed"},
+        .value = {"auto", "yes", "no", "enable", "disable", "1", "0", "on",
+                  "off"},
+        .type = GF_OPTION_TYPE_STR,
+        .default_value = "on",
+        .description =
+            "This option if set to ON, does a lookup through "
+            "all the sub-volumes, in case a lookup didn't return any result "
+            "from the hash subvolume. If set to OFF, it does not do a lookup "
+            "on the remaining subvolumes.",
+        .op_version = {1},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE,
+        .level = OPT_STATUS_BASIC,
+    },
+    {.key = {"lookup-optimize"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .description =
+         "This option if set to ON enables the optimization "
+         "of -ve lookups, by not doing a lookup on non-hashed subvolumes for "
+         "files, in case the hashed subvolume does not return any result. "
+         "This option disregards the lookup-unhashed setting, when enabled.",
+     .op_version = {GD_OP_VERSION_3_7_2},
+     .level = OPT_STATUS_ADVANCED,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC},
+    {.key = {"min-free-disk"},
+     .type = GF_OPTION_TYPE_PERCENT_OR_SIZET,
+     .default_value = "10%",
+     .description =
+         "Percentage/Size of disk space, after which the "
+         "process starts balancing out the cluster, and logs will appear "
+         "in log files",
+     .op_version = {1},
+     .level = OPT_STATUS_BASIC,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC},
+    {.key = {"min-free-inodes"},
+     .type = GF_OPTION_TYPE_PERCENT,
+     .default_value = "5%",
+     .description = "after system has only N% of inodes, warnings "
+                    "starts to appear in log files",
+     .op_version = {1},
+     .level = OPT_STATUS_BASIC,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC},
+    {
+        .key = {"unhashed-sticky-bit"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "off",
+    },
+    {.key = {"use-readdirp"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .description = "This option if set to ON, forces the use of "
+                    "readdirp, and hence also displays the stats of the files.",
+     .level = OPT_STATUS_ADVANCED,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC},
+    {.key = {"assert-no-child-down"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = "This option if set to ON, in the event of "
+                    "CHILD_DOWN, will call exit."},
+    {
+        .key = {"directory-layout-spread"},
+        .type = GF_OPTION_TYPE_INT,
+        .min = 1,
+        .validate = GF_OPT_VALIDATE_MIN,
+        .description = "Specifies the directory layout spread. Takes number "
+                       "of subvolumes as default value.",
+
+        .op_version = {2},
+    },
+    {
+        .key = {"decommissioned-bricks"},
+        .type = GF_OPTION_TYPE_ANY,
+        .description =
+            "This option if set to ON, decommissions "
+            "the brick, so that no new data is allowed to be created "
+            "on that brick.",
+        .level = OPT_STATUS_ADVANCED,
+    },
+    {
+        .key = {"rebalance-cmd"},
+        .type = GF_OPTION_TYPE_INT,
+    },
+    {
+        .key = {"commit-hash"},
+        .type = GF_OPTION_TYPE_INT,
+    },
+    {
+        .key = {"node-uuid"},
+        .type = GF_OPTION_TYPE_STR,
+    },
+    {
+        .key = {"rebalance-stats"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "off",
+        .description =
+            "This option if set to ON displays and logs the "
+            " time taken for migration of each file, during the rebalance "
+            "process. If set to OFF, the rebalance logs will only display the "
+            "time spent in each directory.",
+        .op_version = {2},
+        .level = OPT_STATUS_BASIC,
+    },
+    {.key = {"readdir-optimize"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description =
+         "This option if set to ON enables the optimization "
+         "that allows DHT to requests non-first subvolumes to filter out "
+         "directory entries.",
+     .op_version = {1},
+     .level = OPT_STATUS_ADVANCED,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC},
+    {.key = {"rsync-hash-regex"},
+     .type = GF_OPTION_TYPE_STR,
+     /* Setting a default here doesn't work.  See dht_init_regex. */
+     .description =
+         "Regular expression for stripping temporary-file "
+         "suffix and prefix used by rsync, to prevent relocation when the "
+         "file is renamed.",
+     .op_version = {3},
+     .level = OPT_STATUS_BASIC,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC},
+    {.key = {"extra-hash-regex"},
+     .type = GF_OPTION_TYPE_STR,
+     /* Setting a default here doesn't work.  See dht_init_regex. */
+     .description =
+         "Regular expression for stripping temporary-file "
+         "suffix and prefix used by an application, to prevent relocation when "
+         "the file is renamed.",
+     .op_version = {3},
+     .level = OPT_STATUS_BASIC,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC},
+    {
+        .key = {"rebalance-filter"},
+        .type = GF_OPTION_TYPE_STR,
+    },
+
+    {
+        .key = {"xattr-name"},
+        .type = GF_OPTION_TYPE_STR,
+        .default_value = "trusted.glusterfs.dht",
+        .description =
+            "Base for extended attributes used by this "
+            "translator instance, to avoid conflicts with others above or "
+            "below it.",
+        .op_version = {3},
+    },
+
+    {.key = {"weighted-rebalance"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .description =
+         "When enabled, files will be allocated to bricks "
+         "with a probability proportional to their size.  Otherwise, all "
+         "bricks will have the same probability (legacy behavior).",
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .level = OPT_STATUS_BASIC,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC},
+
+    /* NUFA option */
+    {.key = {"local-volume-name"}, .type = GF_OPTION_TYPE_XLATOR},
+
+    /* switch option */
+    {.key = {"pattern.switch.case"}, .type = GF_OPTION_TYPE_ANY},
+
+    {
+        .key = {"randomize-hash-range-by-gfid"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "off",
+        .description =
+            "Use gfid of directory to determine the subvolume "
+            "from which hash ranges are allocated starting with 0. "
+            "Note that we still use a directory/file's name to determine the "
+            "subvolume to which it hashes",
+        .op_version = {GD_OP_VERSION_3_6_0},
+    },
+
+    {.key = {"rebal-throttle"},
+     .type = GF_OPTION_TYPE_STR,
+     .default_value = "normal",
+     .description = " Sets the maximum number of parallel file migrations "
+                    "allowed on a node during the rebalance operation. The"
+                    " default value is normal and allows a max of "
+                    "[($(processing units) - 4) / 2), 2]  files to be "
+                    "migrated at a time. Lazy will allow only one file to "
+                    "be migrated at a time and aggressive will allow "
+                    "max of [($(processing units) - 4) / 2), 4]",
+     .op_version = {GD_OP_VERSION_3_7_0},
+     .level = OPT_STATUS_BASIC,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC
+
+    },
+
+    {.key = {"lock-migration"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = " If enabled this feature will migrate the posix locks"
+                    " associated with a file during rebalance",
+     .op_version = {GD_OP_VERSION_3_8_0},
+     .level = OPT_STATUS_ADVANCED,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC},
+
+    {.key = {"force-migration"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = "If disabled, rebalance will not migrate files that "
+                    "are being written to by an application",
+     .op_version = {GD_OP_VERSION_4_0_0},
+     .level = OPT_STATUS_ADVANCED,
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC},
+
+    {.key = {NULL}},
+};
+
+#define NUM_DHT_OPTIONS (sizeof(dht_options) / sizeof(dht_options[0]))
+
+extern struct volume_options options[NUM_DHT_OPTIONS]
+    __attribute__((alias("dht_options")));
diff --git a/xlators/cluster/dht/src/dht.c b/xlators/cluster/dht/src/dht.c
index 2c8f2e9a224..53de8292704 100644
--- a/xlators/cluster/dht/src/dht.c
+++ b/xlators/cluster/dht/src/dht.c
@@ -1,552 +1,123 @@
 /*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
 
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-/* TODO: add NS locking */
-
-#include "statedump.h"
-#include "dht-common.c"
-
-/* TODO:
-   - use volumename in xattr instead of "dht"
-   - use NS locks
-   - handle all cases in self heal layout reconstruction
-   - complete linkfile selfheal
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
-
-void
-dht_layout_dump (dht_layout_t  *layout, const char *prefix)
-{
-
-        char    key[GF_DUMP_MAX_BUF_LEN];
-        int     i = 0;
-
-        if (!layout)
-                return;
-
-        gf_proc_dump_build_key(key, prefix, "cnt");
-        gf_proc_dump_write(key, "%d", layout->cnt);
-        gf_proc_dump_build_key(key, prefix, "preset");
-        gf_proc_dump_write(key, "%d", layout->preset);
-        gf_proc_dump_build_key(key, prefix, "gen");
-        gf_proc_dump_write(key, "%d", layout->gen);
-        gf_proc_dump_build_key(key, prefix, "type");
-        gf_proc_dump_write(key, "%d", layout->type);
-
-        for (i = 0; i < layout->cnt; i++) {
-                gf_proc_dump_build_key(key, prefix,"list[%d].err", i);
-                gf_proc_dump_write(key, "%d", layout->list[i].err);
-                gf_proc_dump_build_key(key, prefix,"list[%d].start", i);
-                gf_proc_dump_write(key, "%u", layout->list[i].start);
-                gf_proc_dump_build_key(key, prefix,"list[%d].stop", i);
-                gf_proc_dump_write(key, "%u", layout->list[i].stop);
-                if (layout->list[i].xlator) {
-                        gf_proc_dump_build_key(key, prefix,
-                                               "list[%d].xlator.type", i);
-                        gf_proc_dump_write(key, "%s",
-                                           layout->list[i].xlator->type);
-                        gf_proc_dump_build_key(key, prefix,
-                                               "list[%d].xlator.name", i);
-                        gf_proc_dump_write(key, "%s",
-                                           layout->list[i].xlator->name);
-                }
-        }
-}
-
-
-int32_t
-dht_priv_dump (xlator_t *this)
-{
-	char            key_prefix[GF_DUMP_MAX_BUF_LEN];
-	char            key[GF_DUMP_MAX_BUF_LEN];
-        int             i = 0;
-        dht_conf_t      *conf = NULL;
-        int             ret = 0;
-
-        if (!this)
-                return -1;
-
-        conf = this->private;
-
-        if (!conf)
-                return -1;
-
-        ret = TRY_LOCK(&conf->subvolume_lock);
-
-        if (ret != 0) {
-                gf_log("", GF_LOG_WARNING, "Unable to lock dht subvolume %s",
-                                this->name);
-                return ret;
-        }
-
-        gf_proc_dump_add_section("xlator.cluster.dht.%s.priv", this->name);
-        gf_proc_dump_build_key(key_prefix,"xlator.cluster.dht","%s.priv",
-                               this->name);
-        gf_proc_dump_build_key(key, key_prefix, "subvolume_cnt");
-        gf_proc_dump_write(key,"%d", conf->subvolume_cnt);
-        for (i = 0; i < conf->subvolume_cnt; i++) {
-                gf_proc_dump_build_key(key, key_prefix, "subvolumes[%d]", i);
-                gf_proc_dump_write(key, "%s.%s", conf->subvolumes[i]->type,
-                                   conf->subvolumes[i]->name);
-                if (conf->file_layouts && conf->file_layouts[i]){
-                        gf_proc_dump_build_key(key, key_prefix,
-                                               "file_layouts[%d]",i);
-                        dht_layout_dump(conf->file_layouts[i], key);
-                }
-                if (conf->dir_layouts && conf->dir_layouts[i]) {
-                        gf_proc_dump_build_key(key, key_prefix,
-                                              "dir_layouts[%d]",i);
-                        dht_layout_dump(conf->dir_layouts[i], key);
-                }
-                if (conf->subvolume_status) {
-                        gf_proc_dump_build_key(key, key_prefix,
-                                                "subvolume_status[%d]", i);
-                        gf_proc_dump_write(key, "%d",
-                                               (int)conf->subvolume_status[i]);
-                }
-
-        }
-
-        gf_proc_dump_build_key(key, key_prefix,"default_dir_layout");
-        dht_layout_dump(conf->default_dir_layout, key);
-
-        gf_proc_dump_build_key(key, key_prefix, "search_unhashed");
-        gf_proc_dump_write(key, "%d", conf->search_unhashed);
-        gf_proc_dump_build_key(key, key_prefix, "gen");
-        gf_proc_dump_write(key, "%d", conf->gen);
-        gf_proc_dump_build_key(key, key_prefix, "min_free_disk");
-        gf_proc_dump_write(key, "%lu", conf->min_free_disk);
-        gf_proc_dump_build_key(key, key_prefix, "disk_unit");
-        gf_proc_dump_write(key, "%c", conf->disk_unit);
-        gf_proc_dump_build_key(key, key_prefix, "refresh_interval");
-        gf_proc_dump_write(key, "%d", conf->refresh_interval);
-        gf_proc_dump_build_key(key, key_prefix, "unhashed_sticky_bit");
-        gf_proc_dump_write(key, "%d", conf->unhashed_sticky_bit);
-        if (conf ->du_stats) {
-                gf_proc_dump_build_key(key, key_prefix,
-                                "du_stats.avail_percent");
-                gf_proc_dump_write(key, "%lf", conf->du_stats->avail_percent);
-                gf_proc_dump_build_key(key, key_prefix,
-                                "du_stats.avail_space");
-                gf_proc_dump_write(key, "%lu", conf->du_stats->avail_space);
-                gf_proc_dump_build_key(key, key_prefix,
-                                "du_stats.log");
-                gf_proc_dump_write(key, "%lu", conf->du_stats->log);
-        }
-        gf_proc_dump_build_key(key, key_prefix, "last_stat_fetch");
-        gf_proc_dump_write(key, "%s", ctime(&conf->last_stat_fetch.tv_sec));
-
-        UNLOCK(&conf->subvolume_lock);
-
-        return 0;
-}
-
-int32_t
-dht_inodectx_dump (xlator_t *this, inode_t *inode)
-{
-        int             ret = -1;
-        char            key_prefix[GF_DUMP_MAX_BUF_LEN];
-        dht_layout_t    *layout = NULL;
-	uint64_t        tmp_layout = 0;
-
-        if (!inode)
-                return -1;
-
-	ret = inode_ctx_get (inode, this, &tmp_layout);
-
-        if (ret != 0)
-                return ret;
-
-        layout = (dht_layout_t *)(long)tmp_layout;
-
-        if (!layout)
-                return -1;
-
-        gf_proc_dump_build_key(key_prefix, "xlator.cluster.dht",
-                               "%s.inode.%ld", this->name, inode->ino);
-        dht_layout_dump(layout, key_prefix);
-
-        return 0;
-}
-
-int
-notify (xlator_t *this, int event, void *data, ...)
-{
-	int ret = -1;
-
-	ret = dht_notify (this, event, data);
-
-	return ret;
-}
-
-void
-fini (xlator_t *this)
-{
-        int         i = 0;
-        dht_conf_t *conf = NULL;
-
-	conf = this->private;
-        this->private = NULL;
-        if (conf) {
-                if (conf->file_layouts) {
-                        for (i = 0; i < conf->subvolume_cnt; i++) {
-                                GF_FREE (conf->file_layouts[i]);
-                        }
-                        GF_FREE (conf->file_layouts);
-                }
-
-                if (conf->default_dir_layout)
-                        GF_FREE (conf->default_dir_layout);
-
-                if (conf->subvolumes)
-                        GF_FREE (conf->subvolumes);
-
-		if (conf->subvolume_status)
-			GF_FREE (conf->subvolume_status);
-
-                GF_FREE (conf);
-        }
-
-	return;
-}
-
-int32_t
-mem_acct_init (xlator_t *this)
-{
-        int     ret = -1;
-
-        if (!this)
-                return ret;
-
-        ret = xlator_mem_acct_init (this, gf_dht_mt_end + 1);
-        
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
-                                "failed");
-                return ret;
-        }
-
-        return ret;
-}
-int
-validate_options (xlator_t *this, dict_t *options, char **op_errstr)
-{
-        char            *temp_str = NULL;
-        gf_boolean_t     search_unhashed;
-        int              ret = 0;
-        
-
-
-
-
-        if (dict_get_str (options, "lookup-unhashed", &temp_str) == 0) {
-                if (strcasecmp (temp_str, "auto")) {
-                        if (!gf_string2boolean (temp_str, &search_unhashed)) {
-                                gf_log(this->name, GF_LOG_DEBUG, "Validated"
-                                                " lookup-unahashed (%s)",
-                                                                temp_str);
-                        }
-                        else {
-                                gf_log(this->name, GF_LOG_ERROR, "Validation:"
-                                                " lookup-unahashed should be boolean,"
-                                                                " not (%s)",
-                                                                temp_str);
-                                *op_errstr = gf_strdup ("Error, lookup-"
-                                                "unhashed be boolean");
-                                ret = -1;
-                                goto out;
-                        }
-                
-                }
-        }
-
-        
-
-
-
-out:
-                return ret;
-}
-
-int
-reconfigure (xlator_t *this, dict_t *options)
-{
-	dht_conf_t	*conf = NULL;
-	char		*temp_str = NULL;
-	gf_boolean_t     search_unhashed;
-	uint32_t         temp_free_disk = 0;
-	int		 ret = 0;
-
-
-	conf = this->private;
-        if (!conf)
-                return 0;
-
-	if (dict_get_str (options, "lookup-unhashed", &temp_str) == 0) {
-                /* If option is not "auto", other options _should_ be boolean*/
-                if (strcasecmp (temp_str, "auto")) {
-                        if (!gf_string2boolean (temp_str, &search_unhashed)) {
-				gf_log(this->name, GF_LOG_DEBUG, "Reconfigure:"
-				       " lookup-unahashed reconfigured (%s)",
-				       temp_str);
-				conf->search_unhashed = search_unhashed;
-			}
-			else {
-				gf_log(this->name, GF_LOG_ERROR, "Reconfigure:"
-				       " lookup-unahashed should be boolean,"
-				        " not (%s), defaulting to (%d)",
-				       temp_str, conf->search_unhashed);
-				//return -1;
-				ret = -1;
-				goto out;
-			}
-		
-		}
-                else {
-			gf_log(this->name, GF_LOG_DEBUG, "Reconfigure:"
-			       " lookup-unahashed reconfigured auto ");
-                        conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
-		}
-	}
-
-	if (dict_get_str (options, "min-free-disk", &temp_str) == 0) {
-		if (gf_string2percent (temp_str, &temp_free_disk) == 0) {
-                        if (temp_free_disk > 100) {
-                                gf_string2bytesize (temp_str, 
-                                                    &conf->min_free_disk);
-                                conf->disk_unit = 'b';
-                        } else {
-                                conf->min_free_disk = (uint64_t)temp_free_disk;
-                        }
-                } else {
-                        gf_string2bytesize (temp_str, &conf->min_free_disk);
-                        conf->disk_unit = 'b';
-                }
-
-		gf_log(this->name, GF_LOG_DEBUG, "Reconfigure:"
-			       " min-free-disk reconfigured to ",
-			       temp_str);
-	}
-
-out:
-	return ret;
-}
-
-int
-init (xlator_t *this)
-{
-        dht_conf_t    *conf = NULL;
-	char          *temp_str = NULL;
-        int            ret = -1;
-        int            i = 0;
-        uint32_t       temp_free_disk = 0;
-
-
-	if (!this->children) {
-		gf_log (this->name, GF_LOG_CRITICAL,
-			"Distribute needs more than one subvolume");
-		return -1;
-	}
-  
-	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"dangling volume. check volfile");
-	}
-
-        conf = GF_CALLOC (1, sizeof (*conf), gf_dht_mt_dht_conf_t);
-        if (!conf) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                goto err;
-        }
-
-        conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_ON;
-	if (dict_get_str (this->options, "lookup-unhashed", &temp_str) == 0) {
-                /* If option is not "auto", other options _should_ be boolean */
-                if (strcasecmp (temp_str, "auto"))
-                        gf_string2boolean (temp_str, &conf->search_unhashed);
-                else
-                        conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
-	}
-
-	conf->unhashed_sticky_bit = 0;
-
-	if (dict_get_str (this->options, "unhashed-sticky-bit", 
-                          &temp_str) == 0) {
-	        gf_string2boolean (temp_str, &conf->unhashed_sticky_bit);
-	}
-
-	conf->use_readdirp = 1;
-
-	if (dict_get_str (this->options, "use-readdirp",
-                          &temp_str) == 0) {
-	        gf_string2boolean (temp_str, &conf->use_readdirp);
-	}
-
-        conf->disk_unit = 'p';
-        conf->min_free_disk = 10;
-
-	if (dict_get_str (this->options, "min-free-disk", &temp_str) == 0) {
-		if (gf_string2percent (temp_str, &temp_free_disk) == 0) {
-                        if (temp_free_disk > 100) {
-                                gf_string2bytesize (temp_str, 
-                                                    &conf->min_free_disk);
-                                conf->disk_unit = 'b';
-                        } else {
-                                conf->min_free_disk = (uint64_t)temp_free_disk;
-                        }
-                } else {
-                        gf_string2bytesize (temp_str, &conf->min_free_disk);
-                        conf->disk_unit = 'b';
-                }
-	}
-
-
-        ret = dht_init_subvolumes (this, conf);
-        if (ret == -1) {
-                goto err;
-        }
-
-        ret = dht_layouts_init (this, conf);
-        if (ret == -1) {
-                goto err;
-        }
-
-        conf->du_stats = GF_CALLOC (conf->subvolume_cnt, sizeof (dht_du_t),
-                                    gf_dht_mt_dht_du_t);
-        if (!conf->du_stats) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                goto err;
-        }
-
-	LOCK_INIT (&conf->subvolume_lock);
-	LOCK_INIT (&conf->layout_lock);
-
-	conf->gen = 1;
-
-        this->private = conf;
-
-        return 0;
-
-err:
-        if (conf) {
-                if (conf->file_layouts) {
-                        for (i = 0; i < conf->subvolume_cnt; i++) {
-                                GF_FREE (conf->file_layouts[i]);
-                        }
-                        GF_FREE (conf->file_layouts);
-                }
-
-                if (conf->default_dir_layout)
-                        GF_FREE (conf->default_dir_layout);
-
-                if (conf->subvolumes)
-                        GF_FREE (conf->subvolumes);
-
-		if (conf->subvolume_status)
-			GF_FREE (conf->subvolume_status);
-
-                if (conf->du_stats)
-                        GF_FREE (conf->du_stats);
-
-                GF_FREE (conf);
-        }
-
-        return -1;
-}
-
+#include "dht-common.h"
+
+struct xlator_fops dht_pt_fops = {
+    /* we need to keep mkdir to make sure we
+       have layout on new directory */
+    .mkdir = dht_pt_mkdir,
+    .getxattr = dht_pt_getxattr,
+    .fgetxattr = dht_pt_fgetxattr,
+
+    /* required to trace fop properly in changelog */
+    .rename = dht_pt_rename,
+
+    /* FIXME: commenting the '.lookup()' below made some of
+       the failing tests to pass. I would remove the below
+       line, but keeping it here as a reminder for people
+       to check for issues if they find concerns with DHT
+       pass-through logic  */
+    /*
+      .lookup = dht_lookup,
+      .readdir = dht_readdir,
+      .readdirp = dht_readdirp,
+    */
+    /* Keeping above as commented, mainly to support the
+       usecase of a gluster volume getting to 1x(anytype),
+       due to remove-brick (shrinking) exercise. In that case,
+       we would need above fops to be available, so we can
+       handle the case of dangling linkto files (if any) */
+};
 
 struct xlator_fops fops = {
-	.lookup      = dht_lookup,
-	.mknod       = dht_mknod,
-	.create      = dht_create,
-
-	.stat        = dht_stat,
-	.fstat       = dht_fstat,
-	.truncate    = dht_truncate,
-	.ftruncate   = dht_ftruncate,
-	.access      = dht_access,
-	.readlink    = dht_readlink,
-	.setxattr    = dht_setxattr,
-	.getxattr    = dht_getxattr,
-	.removexattr = dht_removexattr,
-	.open        = dht_open,
-	.readv       = dht_readv,
-	.writev      = dht_writev,
-	.flush       = dht_flush,
-	.fsync       = dht_fsync,
-	.statfs      = dht_statfs,
-	.lk          = dht_lk,
-	.opendir     = dht_opendir,
-	.readdir     = dht_readdir,
-	.readdirp    = dht_readdirp,
-	.fsyncdir    = dht_fsyncdir,
-	.symlink     = dht_symlink,
-	.unlink      = dht_unlink,
-	.link        = dht_link,
-	.mkdir       = dht_mkdir,
-	.rmdir       = dht_rmdir,
-	.rename      = dht_rename,
-	.inodelk     = dht_inodelk,
-	.finodelk    = dht_finodelk,
-	.entrylk     = dht_entrylk,
-	.fentrylk    = dht_fentrylk,
-	.xattrop     = dht_xattrop,
-	.fxattrop    = dht_fxattrop,
-        .setattr     = dht_setattr, 
-        .fsetattr    = dht_fsetattr,
+    .ipc = dht_ipc,
+    .lookup = dht_lookup,
+    .mknod = dht_mknod,
+    .create = dht_create,
+
+    .open = dht_open,
+    .statfs = dht_statfs,
+    .opendir = dht_opendir,
+    .readdir = dht_readdir,
+    .readdirp = dht_readdirp,
+    .fsyncdir = dht_fsyncdir,
+    .symlink = dht_symlink,
+    .unlink = dht_unlink,
+    .link = dht_link,
+    .mkdir = dht_mkdir,
+    .rmdir = dht_rmdir,
+    .rename = dht_rename,
+    .entrylk = dht_entrylk,
+    .fentrylk = dht_fentrylk,
+
+    /* Inode read operations */
+    .stat = dht_stat,
+    .fstat = dht_fstat,
+    .access = dht_access,
+    .readlink = dht_readlink,
+    .getxattr = dht_getxattr,
+    .fgetxattr = dht_fgetxattr,
+    .readv = dht_readv,
+    .flush = dht_flush,
+    .fsync = dht_fsync,
+    .inodelk = dht_inodelk,
+    .finodelk = dht_finodelk,
+    .lk = dht_lk,
+    .lease = dht_lease,
+
+    /* Inode write operations */
+    .fremovexattr = dht_fremovexattr,
+    .removexattr = dht_removexattr,
+    .setxattr = dht_setxattr,
+    .fsetxattr = dht_fsetxattr,
+    .truncate = dht_truncate,
+    .ftruncate = dht_ftruncate,
+    .writev = dht_writev,
+    .xattrop = dht_xattrop,
+    .fxattrop = dht_fxattrop,
+    .setattr = dht_setattr,
+    .fsetattr = dht_fsetattr,
+    .fallocate = dht_fallocate,
+    .discard = dht_discard,
+    .zerofill = dht_zerofill,
 };
 
 struct xlator_dumpops dumpops = {
-        .priv = dht_priv_dump,
-        .inodectx = dht_inodectx_dump,
+    .priv = dht_priv_dump,
+    .inodectx = dht_inodectx_dump,
 };
 
-
 struct xlator_cbks cbks = {
-//	.release    = dht_release,
-//      .releasedir = dht_releasedir,
-	.forget     = dht_forget
+    .release = dht_release,
+    //      .releasedir = dht_releasedir,
+    .forget = dht_forget,
 };
 
-
-struct volume_options options[] = {
-        { .key  = {"lookup-unhashed"},
-          .value = {"auto", "yes", "no", "enable", "disable", "1", "0",
-                    "on", "off"},
-	  .type = GF_OPTION_TYPE_STR
-	},
-        { .key  = {"min-free-disk"},
-          .type = GF_OPTION_TYPE_PERCENT_OR_SIZET,
-        },
-        { .key = {"unhashed-sticky-bit"},
-          .type = GF_OPTION_TYPE_BOOL
-        },
-        { .key = {"use-readdirp"},
-          .type = GF_OPTION_TYPE_BOOL
-        },
-	{ .key  = {NULL} },
+extern int32_t
+mem_acct_init(xlator_t *this);
+
+extern struct volume_options dht_options[];
+
+xlator_api_t xlator_api = {
+    .init = dht_init,
+    .fini = dht_fini,
+    .notify = dht_notify,
+    .reconfigure = dht_reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .dumpops = &dumpops,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = dht_options,
+    .identifier = "distribute",
+    .pass_through_fops = &dht_pt_fops,
+    .category = GF_MAINTAINED,
 };
diff --git a/xlators/cluster/dht/src/nufa.c b/xlators/cluster/dht/src/nufa.c
index c13fdaebe3b..3648a564840 100644
--- a/xlators/cluster/dht/src/nufa.c
+++ b/xlators/cluster/dht/src/nufa.c
@@ -1,762 +1,657 @@
 /*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
 
-#include "dht-common.c"
+#include "dht-common.h"
 
 /* TODO: all 'TODO's in dht.c holds good */
 
+extern struct volume_options dht_options[];
+
 int
-nufa_local_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int op_ret, int op_errno,
-                       inode_t *inode, struct iatt *stbuf, dict_t *xattr,
-                       struct iatt *postparent)
+nufa_local_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, inode_t *inode,
+                      struct iatt *stbuf, dict_t *xattr,
+                      struct iatt *postparent)
 {
-        xlator_t     *subvol      = NULL;
-        char          is_linkfile = 0;
-        char          is_dir      = 0;
-        dht_conf_t   *conf        = NULL;
-        dht_local_t  *local       = NULL;
-        loc_t        *loc         = NULL;
-        int           i           = 0;
-        call_frame_t *prev        = NULL;
-	int           call_cnt    = 0;
-        int           ret         = 0;
-
-
-        conf  = this->private;
-
-        prev  = cookie;
-        local = frame->local;
-        loc   = &local->loc;
-
-	if (ENTRY_MISSING (op_ret, op_errno)) {
-		if (conf->search_unhashed) {
-			local->op_errno = ENOENT;
-			dht_lookup_everywhere (frame, this, loc);
-			return 0;
-		}
-	}
-
-        if (op_ret == -1)
-                goto out;
-
-        is_linkfile = check_is_linkfile (inode, stbuf, xattr);
-        is_dir      = check_is_dir (inode, stbuf, xattr);
-
-        if (!is_dir && !is_linkfile) {
-                /* non-directory and not a linkfile */
-
-		dht_itransform (this, prev->this, stbuf->ia_ino,
-				&stbuf->ia_ino);
-
-		ret = dht_layout_preset (this, prev->this, inode);
-		if (ret < 0) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"could not set pre-set layout for subvol %s",
-				prev->this->name);
-			op_ret   = -1;
-			op_errno = EINVAL;
-			goto err;
-		}
-
-                goto out;
+    xlator_t *subvol = NULL;
+    char is_linkfile = 0;
+    char is_dir = 0;
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    loc_t *loc = NULL;
+    int i = 0;
+    xlator_t *prev = NULL;
+    int call_cnt = 0;
+    int ret = 0;
+
+    conf = this->private;
+
+    prev = cookie;
+    local = frame->local;
+    loc = &local->loc;
+
+    if (ENTRY_MISSING(op_ret, op_errno)) {
+        if (conf->search_unhashed) {
+            local->op_errno = ENOENT;
+            dht_lookup_everywhere(frame, this, loc);
+            return 0;
         }
-
-        if (is_dir) {
-                call_cnt        = conf->subvolume_cnt;
-		local->call_cnt = call_cnt;
-
-                local->inode = inode_ref (inode);
-                local->xattr = dict_ref (xattr);
-
-		local->op_ret = 0;
-		local->op_errno = 0;
-
-		local->layout = dht_layout_new (this, conf->subvolume_cnt);
-		if (!local->layout) {
-			op_ret   = -1;
-			op_errno = ENOMEM;
-			gf_log (this->name, GF_LOG_DEBUG,
-				"memory allocation failed :(");
-			goto err;
-		}
-
-                for (i = 0; i < call_cnt; i++) {
-                        STACK_WIND (frame, dht_lookup_dir_cbk,
-                                    conf->subvolumes[i],
-                                    conf->subvolumes[i]->fops->lookup,
-                                    &local->loc, local->xattr_req);
-                }
+    }
+
+    if (op_ret == -1)
+        goto out;
+
+    is_linkfile = check_is_linkfile(inode, stbuf, xattr, conf->link_xattr_name);
+    is_dir = check_is_dir(inode, stbuf, xattr);
+
+    if (!is_dir && !is_linkfile) {
+        /* non-directory and not a linkfile */
+        ret = dht_layout_preset(this, prev, inode);
+        if (ret < 0) {
+            gf_msg_debug(this->name, 0,
+                         "could not set pre-set layout for subvol"
+                         " %s",
+                         prev->name);
+            op_ret = -1;
+            op_errno = EINVAL;
+            goto err;
         }
 
-        if (is_linkfile) {
-                subvol = dht_linkfile_subvol (this, inode, stbuf, xattr);
+        goto out;
+    }
 
-                if (!subvol) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "linkfile not having link subvolume. path=%s",
-                                loc->path);
-			dht_lookup_everywhere (frame, this, loc);
-			return 0;
-                }
+    if (is_dir) {
+        call_cnt = conf->subvolume_cnt;
+        local->call_cnt = call_cnt;
 
-		STACK_WIND (frame, dht_lookup_linkfile_cbk,
-			    subvol, subvol->fops->lookup,
-			    &local->loc, local->xattr_req);
+        local->inode = inode_ref(inode);
+        local->xattr = dict_ref(xattr);
+
+        local->op_ret = 0;
+        local->op_errno = 0;
+
+        local->layout = dht_layout_new(this, conf->subvolume_cnt);
+        if (!local->layout) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto err;
         }
 
-        return 0;
+        for (i = 0; i < call_cnt; i++) {
+            STACK_WIND_COOKIE(frame, dht_lookup_dir_cbk, conf->subvolumes[i],
+                              conf->subvolumes[i],
+                              conf->subvolumes[i]->fops->lookup, &local->loc,
+                              local->xattr_req);
+        }
+    }
 
-out:
-	if (!local->hashed_subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no subvolume in layout for path=%s",
-			local->loc.path);
-                local->op_errno = ENOENT;
-                dht_lookup_everywhere (frame, this, loc);
-                return 0;
-	}
-
-	STACK_WIND (frame, dht_lookup_cbk,
-		    local->hashed_subvol, local->hashed_subvol->fops->lookup,
-		    &local->loc, local->xattr_req);
-
-	return 0;
-
- err:
-        DHT_STACK_UNWIND (lookup, frame, op_ret, op_errno,
-                          inode, stbuf, xattr, NULL);
-        return 0;
-}
+    if (is_linkfile) {
+        subvol = dht_linkfile_subvol(this, inode, stbuf, xattr);
 
-int
-nufa_lookup (call_frame_t *frame, xlator_t *this,
-	     loc_t *loc, dict_t *xattr_req)
-{
-        xlator_t     *hashed_subvol = NULL;
-        xlator_t     *cached_subvol = NULL;
-        xlator_t     *subvol = NULL;
-        dht_local_t  *local  = NULL;
-	dht_conf_t   *conf = NULL;
-        int           ret    = -1;
-        int           op_errno = -1;
-	dht_layout_t *layout = NULL;
-	int           i = 0;
-	int           call_cnt = 0;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	conf = this->private;
-
-        local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-        ret = loc_dup (loc, &local->loc);
-        if (ret == -1) {
-                op_errno = errno;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "copying location failed for path=%s",
-                        loc->path);
-                goto err;
+        if (!subvol) {
+            gf_msg_debug(this->name, 0,
+                         "linkfile has no link subvolume. path=%s", loc->path);
+            dht_lookup_everywhere(frame, this, loc);
+            return 0;
         }
 
-	if (xattr_req) {
-		local->xattr_req = dict_ref (xattr_req);
-	} else {
-		local->xattr_req = dict_new ();
-	}
-
-	hashed_subvol = dht_subvol_get_hashed (this, &local->loc);
-	cached_subvol = dht_subvol_get_cached (this, local->loc.inode);
-
-	local->cached_subvol = cached_subvol;
-	local->hashed_subvol = hashed_subvol;
-
-        if (is_revalidate (loc)) {
-		local->layout = layout = dht_layout_get (this, loc->inode);
-
-                if (!layout) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "revalidate without cache. path=%s",
-                                loc->path);
-                        op_errno = EINVAL;
-                        goto err;
-                }
-
-		if (layout->gen && (layout->gen < conf->gen)) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"incomplete layout failure for path=%s",
-				loc->path);
-                        dht_layout_unref (this, local->layout);
-			goto do_fresh_lookup;
-		}
-
-		local->inode    = inode_ref (loc->inode);
-		local->ia_ino   = loc->inode->ino;
-
-		local->call_cnt = layout->cnt;
-		call_cnt = local->call_cnt;
-
-		/* NOTE: we don't require 'trusted.glusterfs.dht.linkto' attribute,
-		 *       revalidates directly go to the cached-subvolume.
-		 */
-		ret = dict_set_uint32 (local->xattr_req,
-				       "trusted.glusterfs.dht", 4 * 4);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Failed to set dict value.");
-                        op_errno = -1;
-                        goto err;
-                }
-
-		for (i = 0; i < layout->cnt; i++) {
-			subvol = layout->list[i].xlator;
-
-			STACK_WIND (frame, dht_revalidate_cbk,
-				    subvol, subvol->fops->lookup,
-				    loc, local->xattr_req);
-
-			if (!--call_cnt)
-				break;
-		}
-	} else {
-        do_fresh_lookup:
-		ret = dict_set_uint32 (local->xattr_req,
-				       "trusted.glusterfs.dht", 4 * 4);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Failed to set dict value.");
-                        op_errno = -1;
-                        goto err;
-                }
-
-		ret = dict_set_uint32 (local->xattr_req,
-				       "trusted.glusterfs.dht.linkto", 256);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Failed to set dict value.");
-                        op_errno = -1;
-                        goto err;
-                }
-
-		/* Send it to only local volume */
-		STACK_WIND (frame, nufa_local_lookup_cbk,
-			    (xlator_t *)conf->private,
-			    ((xlator_t *)conf->private)->fops->lookup,
-			    loc, local->xattr_req);
-	}
+        STACK_WIND_COOKIE(frame, dht_lookup_linkfile_cbk, subvol, subvol,
+                          subvol->fops->lookup, &local->loc, local->xattr_req);
+    }
+
+    return 0;
 
+out:
+    if (!local->hashed_subvol) {
+        gf_msg_debug(this->name, 0, "no subvolume in layout for path=%s",
+                     local->loc.path);
+        local->op_errno = ENOENT;
+        dht_lookup_everywhere(frame, this, loc);
         return 0;
+    }
+
+    STACK_WIND_COOKIE(frame, dht_lookup_cbk, local->hashed_subvol,
+                      local->hashed_subvol, local->hashed_subvol->fops->lookup,
+                      &local->loc, local->xattr_req);
+
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
-	return 0;
+    DHT_STACK_UNWIND(lookup, frame, op_ret, op_errno, inode, stbuf, xattr,
+                     postparent);
+    return 0;
 }
 
 int
-nufa_create_linkfile_create_cbk (call_frame_t *frame, void *cookie,
-				 xlator_t *this, int op_ret, int op_errno,
-                                 inode_t *inode, struct iatt *stbuf,
-                                 struct iatt *preparent,
-                                 struct iatt *postparent)
+nufa_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
 {
-	dht_local_t  *local = NULL;
+    xlator_t *hashed_subvol = NULL;
+    xlator_t *subvol = NULL;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+    int op_errno = -1;
+    dht_layout_t *layout = NULL;
+    int i = 0;
+    int call_cnt = 0;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+    VALIDATE_OR_GOTO(loc->path, err);
+
+    conf = this->private;
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_LOOKUP);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    if (xattr_req) {
+        local->xattr_req = dict_ref(xattr_req);
+    } else {
+        local->xattr_req = dict_new();
+    }
+
+    hashed_subvol = dht_subvol_get_hashed(this, &local->loc);
+
+    local->hashed_subvol = hashed_subvol;
+
+    if (is_revalidate(loc)) {
+        layout = local->layout;
+        if (!layout) {
+            gf_msg_debug(this->name, 0,
+                         "revalidate lookup without cache. "
+                         "path=%s",
+                         loc->path);
+            op_errno = EINVAL;
+            goto err;
+        }
 
-	local = frame->local;
+        if (layout->gen && (layout->gen < conf->gen)) {
+            gf_msg_debug(this->name, 0, "incomplete layout failure for path=%s",
+                         loc->path);
+            dht_layout_unref(this, local->layout);
+            goto do_fresh_lookup;
+        }
 
-	if (op_ret == -1)
-		goto err;
+        local->inode = inode_ref(loc->inode);
 
-	STACK_WIND (frame, dht_create_cbk,
-		    local->cached_subvol, local->cached_subvol->fops->create,
-		    &local->loc, local->flags, local->mode, local->fd,
-                    local->params);
+        local->call_cnt = layout->cnt;
+        call_cnt = local->call_cnt;
 
-	return 0;
+        /* NOTE: we don't require 'trusted.glusterfs.dht.linkto' attribute,
+         *       revalidates directly go to the cached-subvolume.
+         */
+        ret = dict_set_uint32(local->xattr_req, conf->xattr_name, 4 * 4);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                   "Failed to set dict value.");
+            op_errno = -1;
+            goto err;
+        }
 
- err:
-	DHT_STACK_UNWIND (create, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL, NULL);
-	return 0;
-}
+        for (i = 0; i < layout->cnt; i++) {
+            subvol = layout->list[i].xlator;
 
-int
-nufa_create (call_frame_t *frame, xlator_t *this,
-	     loc_t *loc, int32_t flags, mode_t mode,
-             fd_t *fd, dict_t *params)
-{
-	dht_local_t *local = NULL;
-	dht_conf_t  *conf  = NULL;
-	xlator_t    *subvol = NULL;
-        xlator_t    *avail_subvol = NULL;
-	int          op_errno = -1;
-	int          ret = -1;
-
-	VALIDATE_OR_GOTO (frame, err);
-	VALIDATE_OR_GOTO (this, err);
-	VALIDATE_OR_GOTO (loc, err);
-
-	conf  = this->private;
-
-        dht_get_du_info (frame, this, loc);
-
-        local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	subvol = dht_subvol_get_hashed (this, loc);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no subvolume in layout for path=%s",
-			loc->path);
-		op_errno = ENOENT;
-		goto err;
-	}
-
-        avail_subvol = conf->private;
-        if (dht_is_subvol_filled (this, (xlator_t *)conf->private)) {
-                avail_subvol =
-                        dht_free_disk_available_subvol (this,
-                                                        (xlator_t *)conf->private);
-        }
+            STACK_WIND_COOKIE(frame, dht_revalidate_cbk, subvol, subvol,
+                              subvol->fops->lookup, loc, local->xattr_req);
 
-        if (subvol != avail_subvol) {
-                /* create a link file instead of actual file */
-                ret = loc_copy (&local->loc, loc);
-                if (ret == -1) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "Out of memory");
-                                op_errno = ENOMEM;
-                                goto err;
-                }
-
-                local->fd = fd_ref (fd);
-                local->params = dict_ref (params);
-                local->mode = mode;
-                local->flags = flags;
-
-                local->cached_subvol = avail_subvol;
-                dht_linkfile_create (frame,
-                                     nufa_create_linkfile_create_cbk,
-                                     avail_subvol, subvol, loc);
-                return 0;
+            if (!--call_cnt)
+                break;
+        }
+    } else {
+    do_fresh_lookup:
+        ret = dict_set_uint32(local->xattr_req, conf->xattr_name, 4 * 4);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                   "Failed to set dict value.");
+            op_errno = -1;
+            goto err;
         }
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "creating %s on %s", loc->path, subvol->name);
+        ret = dict_set_uint32(local->xattr_req, conf->link_xattr_name, 256);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+                   "Failed to set dict value.");
+            op_errno = -1;
+            goto err;
+        }
 
-        STACK_WIND (frame, dht_create_cbk,
-                    subvol, subvol->fops->create,
-                    loc, flags, mode, fd, params);
+        /* Send it to only local volume */
+        STACK_WIND_COOKIE(
+            frame, nufa_local_lookup_cbk, ((xlator_t *)conf->private),
+            ((xlator_t *)conf->private),
+            ((xlator_t *)conf->private)->fops->lookup, loc, local->xattr_req);
+    }
 
-        return 0;
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (create, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL, NULL);
-
-	return 0;
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+    return 0;
 }
 
 int
-nufa_mknod_linkfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int op_ret, int op_errno, inode_t *inode,
-                         struct iatt *stbuf, struct iatt *preparent,
-                         struct iatt *postparent)
+nufa_create_linkfile_create_cbk(call_frame_t *frame, void *cookie,
+                                xlator_t *this, int op_ret, int op_errno,
+                                inode_t *inode, struct iatt *stbuf,
+                                struct iatt *preparent, struct iatt *postparent,
+                                dict_t *xdata)
 {
-	dht_local_t  *local = NULL;
+    dht_local_t *local = NULL;
 
-	local = frame->local;
+    local = frame->local;
 
-	if (op_ret >= 0) {
-		STACK_WIND (frame, dht_newfile_cbk,
-			    local->cached_subvol,
-			    local->cached_subvol->fops->mknod,
-			    &local->loc, local->mode, local->rdev,
-                            local->params);
+    if (op_ret == -1)
+        goto err;
 
-		return 0;
-	}
+    STACK_WIND_COOKIE(frame, dht_create_cbk, local->cached_subvol,
+                      local->cached_subvol, local->cached_subvol->fops->create,
+                      &local->loc, local->flags, local->mode, local->umask,
+                      local->fd, local->params);
 
-        WIPE (postparent);
-        WIPE (preparent);
+    return 0;
 
-	DHT_STACK_UNWIND (link, frame, op_ret, op_errno,
-                          inode, stbuf, preparent, postparent);
-	return 0;
+err:
+    DHT_STACK_UNWIND(create, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL,
+                     NULL);
+    return 0;
 }
 
-
 int
-nufa_mknod (call_frame_t *frame, xlator_t *this,
-	    loc_t *loc, mode_t mode, dev_t rdev, dict_t *params)
+nufa_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+            mode_t mode, mode_t umask, fd_t *fd, dict_t *params)
 {
-	dht_local_t *local = NULL;
-	dht_conf_t  *conf  = NULL;
-	xlator_t    *subvol = NULL;
-        xlator_t    *avail_subvol = NULL;
-	int          op_errno = -1;
-	int          ret = -1;
-
-	VALIDATE_OR_GOTO (frame, err);
-	VALIDATE_OR_GOTO (this, err);
-	VALIDATE_OR_GOTO (loc, err);
-
-	conf  = this->private;
-
-        dht_get_du_info (frame, this, loc);
-
-        local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	subvol = dht_subvol_get_hashed (this, loc);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no subvolume in layout for path=%s",
-			loc->path);
-		op_errno = ENOENT;
-		goto err;
-	}
-
-        /* Consider the disksize in consideration */
-        avail_subvol = conf->private;
-        if (dht_is_subvol_filled (this, (xlator_t *)conf->private)) {
-                avail_subvol =
-                        dht_free_disk_available_subvol (this,
-                                                        (xlator_t *)conf->private);
-        }
-
-	if (avail_subvol != subvol) {
-		/* Create linkfile first */
-		ret = loc_copy (&local->loc, loc);
-		if (ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"Out of memory");
-			op_errno = ENOMEM;
-			goto err;
-		}
-
-                local->params = dict_ref (params);
-		local->mode = mode;
-		local->rdev = rdev;
-		local->cached_subvol = avail_subvol;
-
-		dht_linkfile_create (frame, nufa_mknod_linkfile_cbk,
-                                     avail_subvol, subvol, loc);
-		return 0;
-	}
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *avail_subvol = NULL;
+    int op_errno = -1;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+
+    conf = this->private;
+
+    dht_get_du_info(frame, this, loc);
+
+    local = dht_local_init(frame, loc, fd, GF_FOP_CREATE);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = dht_subvol_get_hashed(this, loc);
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no subvolume in layout for path=%s",
+                     loc->path);
+        op_errno = ENOENT;
+        goto err;
+    }
+
+    avail_subvol = conf->private;
+    if (dht_is_subvol_filled(this, (xlator_t *)conf->private)) {
+        avail_subvol = dht_free_disk_available_subvol(
+            this, (xlator_t *)conf->private, local);
+    }
+
+    if (subvol != avail_subvol) {
+        /* create a link file instead of actual file */
+        local->params = dict_ref(params);
+        local->mode = mode;
+        local->flags = flags;
+        local->umask = umask;
+        local->cached_subvol = avail_subvol;
+        dht_linkfile_create(frame, nufa_create_linkfile_create_cbk, this,
+                            avail_subvol, subvol, loc);
+        return 0;
+    }
 
-	gf_log (this->name, GF_LOG_TRACE,
-		"creating %s on %s", loc->path, subvol->name);
+    gf_msg_trace(this->name, 0, "creating %s on %s", loc->path, subvol->name);
 
-	STACK_WIND (frame, dht_newfile_cbk,
-		    subvol, subvol->fops->mknod,
-		    loc, mode, rdev, params);
+    STACK_WIND_COOKIE(frame, dht_create_cbk, subvol, subvol,
+                      subvol->fops->create, loc, flags, mode, umask, fd,
+                      params);
 
-	return 0;
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (mknod, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(create, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL,
+                     NULL);
 
-	return 0;
+    return 0;
 }
 
-
 int
-notify (xlator_t *this, int event, void *data, ...)
+nufa_mknod_linkfile_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, inode_t *inode,
+                        struct iatt *stbuf, struct iatt *preparent,
+                        struct iatt *postparent, dict_t *xdata)
 {
-	int ret = -1;
+    dht_local_t *local = NULL;
 
-	ret = dht_notify (this, event, data);
+    local = frame->local;
+    if (!local || !local->cached_subvol) {
+        op_errno = EINVAL;
+        op_ret = -1;
+        goto err;
+    }
 
-	return ret;
-}
+    if (op_ret >= 0) {
+        STACK_WIND_COOKIE(
+            frame, dht_newfile_cbk, (void *)local->cached_subvol,
+            local->cached_subvol, local->cached_subvol->fops->mknod,
+            &local->loc, local->mode, local->rdev, local->umask, local->params);
 
-void
-fini (xlator_t *this)
-{
-        int         i = 0;
-        dht_conf_t *conf = NULL;
+        return 0;
+    }
+err:
+    WIPE(postparent);
+    WIPE(preparent);
 
-	conf = this->private;
+    DHT_STACK_UNWIND(link, frame, op_ret, op_errno, inode, stbuf, preparent,
+                     postparent, xdata);
+    return 0;
+}
 
-        if (conf) {
-                if (conf->file_layouts) {
-                        for (i = 0; i < conf->subvolume_cnt; i++) {
-                                GF_FREE (conf->file_layouts[i]);
-                        }
-                        GF_FREE (conf->file_layouts);
-                }
+int
+nufa_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+           dev_t rdev, mode_t umask, dict_t *params)
+{
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *avail_subvol = NULL;
+    int op_errno = -1;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+
+    conf = this->private;
+
+    dht_get_du_info(frame, this, loc);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_MKNOD);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = dht_subvol_get_hashed(this, loc);
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no subvolume in layout for path=%s",
+                     loc->path);
+        op_errno = ENOENT;
+        goto err;
+    }
+
+    /* Consider the disksize in consideration */
+    avail_subvol = conf->private;
+    if (dht_is_subvol_filled(this, (xlator_t *)conf->private)) {
+        avail_subvol = dht_free_disk_available_subvol(
+            this, (xlator_t *)conf->private, local);
+    }
+
+    if (avail_subvol != subvol) {
+        /* Create linkfile first */
+
+        local->params = dict_ref(params);
+        local->mode = mode;
+        local->umask = umask;
+        local->rdev = rdev;
+        local->cached_subvol = avail_subvol;
+
+        dht_linkfile_create(frame, nufa_mknod_linkfile_cbk, this, avail_subvol,
+                            subvol, loc);
+        return 0;
+    }
 
-                if (conf->default_dir_layout)
-                        GF_FREE (conf->default_dir_layout);
+    gf_msg_trace(this->name, 0, "creating %s on %s", loc->path, subvol->name);
 
-                if (conf->subvolumes)
-                        GF_FREE (conf->subvolumes);
+    STACK_WIND_COOKIE(frame, dht_newfile_cbk, (void *)subvol, subvol,
+                      subvol->fops->mknod, loc, mode, rdev, umask, params);
 
-		if (conf->subvolume_status)
-			GF_FREE (conf->subvolume_status);
+    return 0;
 
-                GF_FREE (conf);
-        }
+err:
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
 
-	return;
+    return 0;
 }
 
-int
-init (xlator_t *this)
+gf_boolean_t
+same_first_part(char *str1, char term1, char *str2, char term2)
 {
-        dht_conf_t    *conf = NULL;
-	xlator_list_t *trav = NULL;
-	data_t        *data = NULL;
-	char          *local_volname = NULL;
-	char          *temp_str = NULL;
-        int            ret = -1;
-        int            i = 0;
-	char           my_hostname[256];
-	uint32_t       temp_free_disk = 0;
-
-	if (!this->children) {
-		gf_log (this->name, GF_LOG_CRITICAL,
-			"NUFA needs more than one subvolume");
-		return -1;
-	}
-
-	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"dangling volume. check volfile");
-	}
-
-        conf = GF_CALLOC (1, sizeof (*conf),
-                          gf_dht_mt_dht_conf_t);
-        if (!conf) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                goto err;
+    gf_boolean_t ended1;
+    gf_boolean_t ended2;
+
+    for (;;) {
+        ended1 = ((*str1 == '\0') || (*str1 == term1));
+        ended2 = ((*str2 == '\0') || (*str2 == term2));
+        if (ended1 && ended2) {
+            return _gf_true;
         }
-
-        conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_ON;
-	if (dict_get_str (this->options, "lookup-unhashed", &temp_str) == 0) {
-                /* If option is not "auto", other options _should_ be boolean */
-                if (strcasecmp (temp_str, "auto"))
-                        gf_string2boolean (temp_str, &conf->search_unhashed);
-                else
-                        conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
-	}
-
-        ret = dht_init_subvolumes (this, conf);
-        if (ret == -1) {
-                goto err;
+        if (ended1 || ended2 || (*str1 != *str2)) {
+            return _gf_false;
         }
+        ++str1;
+        ++str2;
+    }
+}
 
-        ret = dht_layouts_init (this, conf);
-        if (ret == -1) {
-                goto err;
-        }
-
-	LOCK_INIT (&conf->subvolume_lock);
-	LOCK_INIT (&conf->layout_lock);
-
-	conf->gen = 1;
-
-	local_volname = "localhost";
-	ret = gethostname (my_hostname, 256);
-	if (ret < 0) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"could not find hostname (%s)",
-			strerror (errno));
-	}
-
-	if (ret == 0)
-		local_volname = my_hostname;
-
-	data = dict_get (this->options, "local-volume-name");
-	if (data) {
-		local_volname = data->data;
-	}
-
-	trav = this->children;
-	while (trav) {
-		if (strcmp (trav->xlator->name, local_volname) == 0)
-			break;
-		trav = trav->next;
-	}
-
-	if (!trav) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Could not find subvolume named '%s'. "
-			"Please define volume with the name as the hostname "
-			"or override it with 'option local-volume-name'",
-			local_volname);
-		goto err;
-	}
-	/* The volume specified exists */
-	conf->private = trav->xlator;
-
-        conf->min_free_disk = 10;
-        conf->disk_unit = 'p';
-
-        if (dict_get_str (this->options, "min-free-disk",
-                          &temp_str) == 0) {
-                if (gf_string2percent (temp_str,
-                                       &temp_free_disk) == 0) {
-                        if (temp_free_disk > 100) {
-                                gf_string2bytesize (temp_str,
-                                                        &conf->min_free_disk);
-                                conf->disk_unit = 'b';
-                        } else {
-                                conf->min_free_disk = (uint64_t)temp_free_disk;
-                                conf->disk_unit = 'p';
-                        }
-                } else {
-                        gf_string2bytesize (temp_str,
-                                                &conf->min_free_disk);
-                        conf->disk_unit = 'b';
-                }
-        }
-
-        conf->du_stats = GF_CALLOC (conf->subvolume_cnt, sizeof (dht_du_t),
-                                    gf_dht_mt_dht_du_t);
-        if (!conf->du_stats) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                goto err;
-        }
-
-        this->private = conf;
-
-        return 0;
-
-err:
-        if (conf) {
-                if (conf->file_layouts) {
-                        for (i = 0; i < conf->subvolume_cnt; i++) {
-                                GF_FREE (conf->file_layouts[i]);
-                        }
-                        GF_FREE (conf->file_layouts);
-                }
-
-                if (conf->default_dir_layout)
-                        GF_FREE (conf->default_dir_layout);
+typedef struct nufa_args {
+    xlator_t *this;
+    char *volname;
+    gf_boolean_t addr_match;
+} nufa_args_t;
 
-                if (conf->subvolumes)
-                        GF_FREE (conf->subvolumes);
+static void
+nufa_find_local_brick(xlator_t *xl, void *data)
+{
+    nufa_args_t *args = data;
+    xlator_t *this = args->this;
+    char *local_volname = args->volname;
+    gf_boolean_t addr_match = args->addr_match;
+    char *brick_host = NULL;
+    dht_conf_t *conf = this->private;
+    int ret = -1;
+
+    /*This means a local subvol was already found. We pick the first brick
+     * that is local*/
+    if (conf->private)
+        return;
+
+    if (strcmp(xl->name, local_volname) == 0) {
+        conf->private = xl;
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_INFO,
+               "Using specified subvol %s", local_volname);
+        return;
+    }
+
+    if (!addr_match)
+        return;
+
+    ret = dict_get_str(xl->options, "remote-host", &brick_host);
+    if ((ret == 0) && (gf_is_same_address(local_volname, brick_host) ||
+                       gf_is_local_addr(brick_host))) {
+        conf->private = xl;
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_INFO,
+               "Using the first local "
+               "subvol %s",
+               xl->name);
+        return;
+    }
+}
 
-		if (conf->subvolume_status)
-			GF_FREE (conf->subvolume_status);
+static void
+nufa_to_dht(xlator_t *this)
+{
+    GF_ASSERT(this);
+    GF_ASSERT(this->fops);
 
-                if (conf->du_stats)
-                        GF_FREE (conf->du_stats);
+    this->fops->lookup = dht_lookup;
+    this->fops->create = dht_create;
+    this->fops->mknod = dht_mknod;
+}
 
-                GF_FREE (conf);
+int
+nufa_find_local_subvol(xlator_t *this, void (*fn)(xlator_t *each, void *data),
+                       void *data)
+{
+    int ret = -1;
+    dht_conf_t *conf = this->private;
+    xlator_list_t *trav = NULL;
+    xlator_t *parent = NULL;
+    xlator_t *candidate = NULL;
+
+    xlator_foreach_depth_first(this, fn, data);
+    if (!conf->private) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_BRICK_ERROR,
+               "Couldn't find a local "
+               "brick");
+        return -1;
+    }
+
+    candidate = conf->private;
+    trav = candidate->parents;
+    while (trav) {
+        parent = trav->xlator;
+        if (strcmp(parent->type, "cluster/nufa") == 0) {
+            gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_INFO,
+                   "Found local subvol, "
+                   "%s",
+                   candidate->name);
+            ret = 0;
+            conf->private = candidate;
+            break;
         }
 
-        return -1;
+        candidate = parent;
+        trav = parent->parents;
+    }
+
+    return ret;
 }
 
+int
+nufa_init(xlator_t *this)
+{
+    data_t *data = NULL;
+    char *local_volname = NULL;
+    int ret = -1;
+    char my_hostname[256];
+    gf_boolean_t addr_match = _gf_false;
+    nufa_args_t args = {
+        0,
+    };
+
+    ret = dht_init(this);
+    if (ret) {
+        return ret;
+    }
+
+    if ((data = dict_get(this->options, "local-volume-name"))) {
+        local_volname = data->data;
+
+    } else {
+        addr_match = _gf_true;
+        local_volname = "localhost";
+        ret = gethostname(my_hostname, 256);
+        if (ret == 0)
+            local_volname = my_hostname;
+
+        else
+            gf_msg(this->name, GF_LOG_WARNING, errno,
+                   DHT_MSG_GET_HOSTNAME_FAILED, "could not find hostname");
+    }
+
+    args.this = this;
+    args.volname = local_volname;
+    args.addr_match = addr_match;
+    ret = nufa_find_local_subvol(this, nufa_find_local_brick, &args);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_SUBVOL_INFO,
+               "Unable to find local subvolume, switching "
+               "to dht mode");
+        nufa_to_dht(this);
+    }
+    return 0;
+}
 
-struct xlator_fops fops = {
-	.lookup      = nufa_lookup,
-	.create      = nufa_create,
-	.mknod       = nufa_mknod,
-
-	.stat        = dht_stat,
-	.fstat       = dht_fstat,
-	.truncate    = dht_truncate,
-	.ftruncate   = dht_ftruncate,
-	.access      = dht_access,
-	.readlink    = dht_readlink,
-	.setxattr    = dht_setxattr,
-	.getxattr    = dht_getxattr,
-	.removexattr = dht_removexattr,
-	.open        = dht_open,
-	.readv       = dht_readv,
-	.writev      = dht_writev,
-	.flush       = dht_flush,
-	.fsync       = dht_fsync,
-	.statfs      = dht_statfs,
-	.lk          = dht_lk,
-	.opendir     = dht_opendir,
-	.readdir     = dht_readdir,
-	.readdirp    = dht_readdirp,
-	.fsyncdir    = dht_fsyncdir,
-	.symlink     = dht_symlink,
-	.unlink      = dht_unlink,
-	.link        = dht_link,
-	.mkdir       = dht_mkdir,
-	.rmdir       = dht_rmdir,
-	.rename      = dht_rename,
-	.inodelk     = dht_inodelk,
-	.finodelk    = dht_finodelk,
-	.entrylk     = dht_entrylk,
-	.fentrylk    = dht_fentrylk,
-	.xattrop     = dht_xattrop,
-	.fxattrop    = dht_fxattrop,
-        .setattr     = dht_setattr,
+dht_methods_t dht_methods = {
+    .migration_get_dst_subvol = dht_migration_get_dst_subvol,
+    .layout_search = dht_layout_search,
 };
 
-
-struct xlator_cbks cbks = {
-	.forget     = dht_forget
+struct xlator_fops fops = {
+    .lookup = nufa_lookup,
+    .create = nufa_create,
+    .mknod = nufa_mknod,
+
+    .stat = dht_stat,
+    .fstat = dht_fstat,
+    .truncate = dht_truncate,
+    .ftruncate = dht_ftruncate,
+    .access = dht_access,
+    .readlink = dht_readlink,
+    .setxattr = dht_setxattr,
+    .getxattr = dht_getxattr,
+    .removexattr = dht_removexattr,
+    .open = dht_open,
+    .readv = dht_readv,
+    .writev = dht_writev,
+    .flush = dht_flush,
+    .fsync = dht_fsync,
+    .statfs = dht_statfs,
+    .lk = dht_lk,
+    .opendir = dht_opendir,
+    .readdir = dht_readdir,
+    .readdirp = dht_readdirp,
+    .fsyncdir = dht_fsyncdir,
+    .symlink = dht_symlink,
+    .unlink = dht_unlink,
+    .link = dht_link,
+    .mkdir = dht_mkdir,
+    .rmdir = dht_rmdir,
+    .rename = dht_rename,
+    .inodelk = dht_inodelk,
+    .finodelk = dht_finodelk,
+    .entrylk = dht_entrylk,
+    .fentrylk = dht_fentrylk,
+    .xattrop = dht_xattrop,
+    .fxattrop = dht_fxattrop,
+    .setattr = dht_setattr,
 };
 
-
-struct volume_options options[] = {
-        { .key  = {"lookup-unhashed"},
-          .value = {"auto", "yes", "no", "enable", "disable", "1", "0",
-                    "on", "off"},
-	  .type = GF_OPTION_TYPE_STR
-	},
-	{ .key  = {"local-volume-name"},
-	  .type = GF_OPTION_TYPE_XLATOR
-	},
-        { .key  = {"min-free-disk"},
-          .type = GF_OPTION_TYPE_PERCENT_OR_SIZET,
-        },
-	{ .key  = {NULL} },
+struct xlator_cbks cbks = {.forget = dht_forget};
+extern int32_t
+mem_acct_init(xlator_t *this);
+
+xlator_api_t xlator_api = {
+    .init = nufa_init,
+    .fini = dht_fini,
+    .notify = dht_notify,
+    .reconfigure = dht_reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = dht_options,
+    .identifier = "nufa",
+    .category = GF_TECH_PREVIEW,
 };
diff --git a/xlators/cluster/dht/src/switch.c b/xlators/cluster/dht/src/switch.c
index dc56787eb4c..207d109a025 100644
--- a/xlators/cluster/dht/src/switch.c
+++ b/xlators/cluster/dht/src/switch.c
@@ -1,29 +1,14 @@
 /*
-   Copyright (c) 2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
 
-#include "dht-common.c"
+#include "dht-common.h"
 #include "dht-mem-types.h"
 
 #include <sys/time.h>
@@ -31,1036 +16,876 @@
 #include <fnmatch.h>
 #include <string.h>
 
+extern struct volume_options dht_options[];
+
 struct switch_sched_array {
-	xlator_t *xl;
-	int32_t   eligible;
-	int32_t   considered;
+    xlator_t *xl;
+    int32_t eligible;
+    int32_t considered;
 };
 
 /* Select one of this struct based on the path's pattern match */
 struct switch_struct {
-	struct switch_struct      *next;
-        struct switch_sched_array *array;
-	int32_t                    node_index; /* Index of the node in
-                                                  this pattern. */
-	int32_t                    num_child;  /* Total num of child nodes
-                                                  with this pattern. */
-	char                       path_pattern[256];
+    struct switch_struct *next;
+    struct switch_sched_array *array;
+    int32_t node_index; /* Index of the node in
+                           this pattern. */
+    int32_t num_child;  /* Total num of child nodes
+                           with this pattern. */
+    char path_pattern[256];
 };
 
 /* TODO: all 'TODO's in dht.c holds good */
 /* This function should return child node as '*:subvolumes' is inserterd */
 
 static int32_t
-gf_switch_valid_child (xlator_t *this, const char *child)
+gf_switch_valid_child(xlator_t *this, const char *child)
 {
-        xlator_list_t *children = NULL;
-        int32_t        ret = 0;
-
-        children = this->children;
-        while (children) {
-                if (!strcmp (child, children->xlator->name)) {
-                        ret = 1;
-                        break;
-                }
-                children = children->next;
+    xlator_list_t *children = NULL;
+    int32_t ret = 0;
+
+    children = this->children;
+    while (children) {
+        if (!strcmp(child, children->xlator->name)) {
+            ret = 1;
+            break;
         }
+        children = children->next;
+    }
 
-        return ret;
+    return ret;
 }
 
 static xlator_t *
-get_switch_matching_subvol (const char *path, dht_conf_t *conf,
-                            xlator_t *hashed_subvol)
+get_switch_matching_subvol(const char *path, dht_conf_t *conf,
+                           xlator_t *hashed_subvol)
 {
-        struct switch_struct *cond      = NULL;
-	struct switch_struct *trav      = NULL;
-	char                 *pathname  = NULL;
-	int                   idx     = 0;
-
-        cond = conf->private;
-        if (!cond)
-                return hashed_subvol;
-
-        trav = cond;
-        pathname = gf_strdup (path);
-	while (trav) {
-		if (fnmatch (trav->path_pattern,
-			     pathname, FNM_NOESCAPE) == 0) {
-                        for (idx = 0; idx < trav->num_child; idx++) {
-                                if (trav->array[idx].xl == hashed_subvol)
-                                        return hashed_subvol;
-                        }
-                        idx = trav->node_index++;
-                        trav->node_index %= trav->num_child;
-                        return trav->array[idx].xl;
-		}
-		trav = trav->next;
-	}
-	GF_FREE (pathname);
-	return hashed_subvol;
-}
+    struct switch_struct *cond = NULL;
+    struct switch_struct *trav = NULL;
+    char *pathname = NULL;
+    int idx = 0;
+    xlator_t *subvol = NULL;
+
+    cond = conf->private;
+    subvol = hashed_subvol;
+    if (!cond)
+        goto out;
+
+    pathname = gf_strdup(path);
+    if (!pathname)
+        goto out;
+
+    trav = cond;
+    while (trav) {
+        if (fnmatch(trav->path_pattern, pathname, FNM_NOESCAPE) == 0) {
+            for (idx = 0; idx < trav->num_child; idx++) {
+                if (trav->array[idx].xl == hashed_subvol)
+                    goto out;
+            }
+            idx = trav->node_index++;
+            trav->node_index %= trav->num_child;
+            subvol = trav->array[idx].xl;
+            goto out;
+        }
+        trav = trav->next;
+    }
+out:
+    GF_FREE(pathname);
 
+    return subvol;
+}
 
 int
-switch_local_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int op_ret, int op_errno,
-                         inode_t *inode, struct iatt *stbuf, dict_t *xattr,
-                         struct iatt *postparent)
+switch_local_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, inode_t *inode,
+                        struct iatt *stbuf, dict_t *xattr,
+                        struct iatt *postparent)
 {
-        xlator_t     *subvol      = NULL;
-        char          is_linkfile = 0;
-        char          is_dir      = 0;
-        dht_conf_t   *conf        = NULL;
-        dht_local_t  *local       = NULL;
-        loc_t        *loc         = NULL;
-        int           i           = 0;
-        call_frame_t *prev        = NULL;
-	int           call_cnt    = 0;
-        int           ret         = 0;
-
-        conf  = this->private;
-
-        prev  = cookie;
-        local = frame->local;
-        loc   = &local->loc;
-
-	if (ENTRY_MISSING (op_ret, op_errno)) {
-		if (conf->search_unhashed) {
-			local->op_errno = ENOENT;
-			dht_lookup_everywhere (frame, this, loc);
-			return 0;
-		}
-	}
-
-        if (op_ret == -1)
-                goto out;
-
-        is_linkfile = check_is_linkfile (inode, stbuf, xattr);
-        is_dir      = check_is_dir (inode, stbuf, xattr);
-
-        if (!is_dir && !is_linkfile) {
-                /* non-directory and not a linkfile */
-
-		dht_itransform (this, prev->this, stbuf->ia_ino,
-				&stbuf->ia_ino);
-
-		ret = dht_layout_preset (this, prev->this, inode);
-		if (ret < 0) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"could not set pre-set layout for subvol %s",
-				prev->this->name);
-			op_ret   = -1;
-			op_errno = EINVAL;
-			goto err;
-		}
-
-                goto out;
+    xlator_t *subvol = NULL;
+    char is_linkfile = 0;
+    char is_dir = 0;
+    dht_conf_t *conf = NULL;
+    dht_local_t *local = NULL;
+    loc_t *loc = NULL;
+    int i = 0;
+    xlator_t *prev = NULL;
+    int call_cnt = 0;
+    int ret = 0;
+
+    conf = this->private;
+
+    prev = cookie;
+    local = frame->local;
+    loc = &local->loc;
+
+    if (ENTRY_MISSING(op_ret, op_errno)) {
+        if (conf->search_unhashed) {
+            local->op_errno = ENOENT;
+            dht_lookup_everywhere(frame, this, loc);
+            return 0;
+        }
+    }
+
+    if (op_ret == -1)
+        goto out;
+
+    is_linkfile = check_is_linkfile(inode, stbuf, xattr, conf->link_xattr_name);
+    is_dir = check_is_dir(inode, stbuf, xattr);
+
+    if (!is_dir && !is_linkfile) {
+        /* non-directory and not a linkfile */
+
+        ret = dht_layout_preset(this, prev, inode);
+        if (ret < 0) {
+            gf_msg_debug(this->name, 0,
+                         "could not set pre-set layout "
+                         "for subvol %s",
+                         prev->name);
+            op_ret = -1;
+            op_errno = EINVAL;
+            goto err;
         }
 
-        if (is_dir) {
-                call_cnt        = conf->subvolume_cnt;
-		local->call_cnt = call_cnt;
-
-                local->inode = inode_ref (inode);
-                local->xattr = dict_ref (xattr);
-
-		local->op_ret = 0;
-		local->op_errno = 0;
-
-		local->layout = dht_layout_new (this, conf->subvolume_cnt);
-		if (!local->layout) {
-			op_ret   = -1;
-			op_errno = ENOMEM;
-			gf_log (this->name, GF_LOG_DEBUG,
-				"memory allocation failed :(");
-			goto err;
-		}
-
-                for (i = 0; i < call_cnt; i++) {
-                        STACK_WIND (frame, dht_lookup_dir_cbk,
-                                    conf->subvolumes[i],
-                                    conf->subvolumes[i]->fops->lookup,
-                                    &local->loc, local->xattr_req);
-                }
+        goto out;
+    }
+
+    if (is_dir) {
+        call_cnt = conf->subvolume_cnt;
+        local->call_cnt = call_cnt;
+
+        local->inode = inode_ref(inode);
+        local->xattr = dict_ref(xattr);
+
+        local->op_ret = 0;
+        local->op_errno = 0;
+
+        local->layout = dht_layout_new(this, conf->subvolume_cnt);
+        if (!local->layout) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            gf_msg_debug(this->name, 0, "memory allocation failed :(");
+            goto err;
         }
 
-        if (is_linkfile) {
-                subvol = dht_linkfile_subvol (this, inode, stbuf, xattr);
+        for (i = 0; i < call_cnt; i++) {
+            STACK_WIND_COOKIE(frame, dht_lookup_dir_cbk, conf->subvolumes[i],
+                              conf->subvolumes[i],
+                              conf->subvolumes[i]->fops->lookup, &local->loc,
+                              local->xattr_req);
+        }
+    }
 
-                if (!subvol) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "linkfile not having link subvolume. path=%s",
-                                loc->path);
-			dht_lookup_everywhere (frame, this, loc);
-			return 0;
-                }
+    if (is_linkfile) {
+        subvol = dht_linkfile_subvol(this, inode, stbuf, xattr);
 
-		STACK_WIND (frame, dht_lookup_linkfile_cbk,
-			    subvol, subvol->fops->lookup,
-			    &local->loc, local->xattr_req);
+        if (!subvol) {
+            gf_msg_debug(this->name, 0,
+                         "linkfile has no link subvolume.path=%s", loc->path);
+            dht_lookup_everywhere(frame, this, loc);
+            return 0;
         }
 
-        return 0;
+        STACK_WIND_COOKIE(frame, dht_lookup_linkfile_cbk, subvol, subvol,
+                          subvol->fops->lookup, &local->loc, local->xattr_req);
+    }
+
+    return 0;
 
 out:
-	if (!local->hashed_subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no subvolume in layout for path=%s",
-			local->loc.path);
-                local->op_errno = ENOENT;
-                dht_lookup_everywhere (frame, this, loc);
-                return 0;
-	}
-
-	STACK_WIND (frame, dht_lookup_cbk,
-		    local->hashed_subvol, local->hashed_subvol->fops->lookup,
-		    &local->loc, local->xattr_req);
-
-	return 0;
-
- err:
-        DHT_STACK_UNWIND (lookup, frame, op_ret, op_errno,
-                          inode, stbuf, xattr, NULL);
+    if (!local->hashed_subvol) {
+        gf_msg_debug(this->name, 0, "no subvolume in layout for path=%s",
+                     local->loc.path);
+        local->op_errno = ENOENT;
+        dht_lookup_everywhere(frame, this, loc);
         return 0;
+    }
+
+    STACK_WIND_COOKIE(frame, dht_lookup_cbk, local->hashed_subvol,
+                      local->hashed_subvol, local->hashed_subvol->fops->lookup,
+                      &local->loc, local->xattr_req);
+
+    return 0;
+
+err:
+    DHT_STACK_UNWIND(lookup, frame, op_ret, op_errno, inode, stbuf, xattr,
+                     NULL);
+    return 0;
 }
 
 int
-switch_lookup (call_frame_t *frame, xlator_t *this,
-	     loc_t *loc, dict_t *xattr_req)
+switch_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc,
+              dict_t *xattr_req)
 {
-        xlator_t     *hashed_subvol = NULL;
-        xlator_t     *cached_subvol = NULL;
-        xlator_t     *subvol = NULL;
-        dht_local_t  *local  = NULL;
-	dht_conf_t   *conf = NULL;
-        int           ret    = -1;
-        int           op_errno = -1;
-	dht_layout_t *layout = NULL;
-	int           i = 0;
-	int           call_cnt = 0;
-
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	conf = this->private;
-
-        local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-        ret = loc_dup (loc, &local->loc);
-        if (ret == -1) {
-                op_errno = errno;
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "copying location failed for path=%s",
-                        loc->path);
-                goto err;
+    xlator_t *hashed_subvol = NULL;
+    xlator_t *cached_subvol = NULL;
+    xlator_t *subvol = NULL;
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    int ret = -1;
+    int op_errno = -1;
+    dht_layout_t *layout = NULL;
+    int i = 0;
+    int call_cnt = 0;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+    VALIDATE_OR_GOTO(loc->path, err);
+
+    conf = this->private;
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_LOOKUP);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    if (xattr_req) {
+        local->xattr_req = dict_ref(xattr_req);
+    } else {
+        local->xattr_req = dict_new();
+    }
+
+    hashed_subvol = dht_subvol_get_hashed(this, &local->loc);
+    cached_subvol = local->cached_subvol;
+
+    local->hashed_subvol = hashed_subvol;
+
+    if (is_revalidate(loc)) {
+        layout = local->layout;
+        if (!layout) {
+            gf_msg_debug(this->name, 0,
+                         "revalidate lookup without cache. path=%s", loc->path);
+            op_errno = EINVAL;
+            goto err;
         }
 
-	if (xattr_req) {
-		local->xattr_req = dict_ref (xattr_req);
-	} else {
-		local->xattr_req = dict_new ();
-	}
+        if (layout->gen && (layout->gen < conf->gen)) {
+            gf_msg_debug(this->name, 0, "incomplete layout failure for path=%s",
+                         loc->path);
+            dht_layout_unref(this, local->layout);
+            goto do_fresh_lookup;
+        }
 
-	hashed_subvol = dht_subvol_get_hashed (this, &local->loc);
-	cached_subvol = dht_subvol_get_cached (this, local->loc.inode);
+        local->inode = inode_ref(loc->inode);
 
-	local->cached_subvol = cached_subvol;
-	local->hashed_subvol = hashed_subvol;
+        local->call_cnt = layout->cnt;
+        call_cnt = local->call_cnt;
 
-        if (is_revalidate (loc)) {
-		local->layout = layout = dht_layout_get (this, loc->inode);
+        /* NOTE: we don't require 'trusted.glusterfs.dht.linkto'
+         * attribute, revalidates directly go to the cached-subvolume.
+         */
+        ret = dict_set_uint32(local->xattr_req, conf->xattr_name, 4 * 4);
+        if (ret < 0)
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+                   "failed to set dict value for %s", conf->xattr_name);
 
-                if (!layout) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "revalidate without cache. path=%s",
-                                loc->path);
-                        op_errno = EINVAL;
-                        goto err;
-                }
+        for (i = 0; i < layout->cnt; i++) {
+            subvol = layout->list[i].xlator;
 
-		if (layout->gen && (layout->gen < conf->gen)) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"incomplete layout failure for path=%s",
-				loc->path);
-                        dht_layout_unref (this, local->layout);
-			goto do_fresh_lookup;
-		}
-
-		local->inode    = inode_ref (loc->inode);
-		local->ia_ino   = loc->inode->ino;
-
-		local->call_cnt = layout->cnt;
-		call_cnt = local->call_cnt;
-
-		/* NOTE: we don't require 'trusted.glusterfs.dht.linkto'
-                 * attribute, revalidates directly go to the cached-subvolume.
-		 */
-		ret = dict_set_uint32 (local->xattr_req,
-				       "trusted.glusterfs.dht", 4 * 4);
-                if (ret < 0)
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "failed to set dict value for "
-                                "trusted.glusterfs.dht");
-
-		for (i = 0; i < layout->cnt; i++) {
-			subvol = layout->list[i].xlator;
-
-			STACK_WIND (frame, dht_revalidate_cbk,
-				    subvol, subvol->fops->lookup,
-				    loc, local->xattr_req);
-
-			if (!--call_cnt)
-				break;
-		}
-	} else {
-        do_fresh_lookup:
-		ret = dict_set_uint32 (local->xattr_req,
-				       "trusted.glusterfs.dht", 4 * 4);
-                if (ret < 0)
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "failed to set dict value for "
-                                "trusted.glusterfs.dht");
-                
-		ret = dict_set_uint32 (local->xattr_req,
-				       "trusted.glusterfs.dht.linkto", 256);
-                if (ret < 0)
-                        gf_log (this->name, GF_LOG_WARNING, 
-                                "failed to set dict value for " 
-                                "trusted.glusterfs.dht.linkto");
-
-                if (!hashed_subvol) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"no subvolume in layout for path=%s, "
-				"checking on all the subvols to see if "
-				"it is a directory", loc->path);
- 			call_cnt        = conf->subvolume_cnt;
- 			local->call_cnt = call_cnt;
-
- 			local->layout = dht_layout_new (this,
-                                                        conf->subvolume_cnt);
- 			if (!local->layout) {
- 				op_errno = ENOMEM;
- 				gf_log (this->name, GF_LOG_ERROR,
- 					"Out of memory");
- 				goto err;
- 			}
-
-			for (i = 0; i < call_cnt; i++) {
- 				STACK_WIND (frame, dht_lookup_dir_cbk,
- 					    conf->subvolumes[i],
- 					    conf->subvolumes[i]->fops->lookup,
- 					    &local->loc, local->xattr_req);
- 			}
- 			return 0;
-                }
+            STACK_WIND_COOKIE(frame, dht_revalidate_cbk, subvol, subvol,
+                              subvol->fops->lookup, loc, local->xattr_req);
 
-		/*  */
-                cached_subvol = get_switch_matching_subvol (loc->path, conf,
-                                                            hashed_subvol);
-                if (cached_subvol == hashed_subvol) {
-                        STACK_WIND (frame, dht_lookup_cbk,
-                                    hashed_subvol,
-                                    hashed_subvol->fops->lookup,
-                                    loc, local->xattr_req);
-                } else {
-                        STACK_WIND (frame, switch_local_lookup_cbk,
-                                    cached_subvol,
-                                    cached_subvol->fops->lookup,
-                                    loc, local->xattr_req);
-                }
+            if (!--call_cnt)
+                break;
+        }
+    } else {
+    do_fresh_lookup:
+        ret = dict_set_uint32(local->xattr_req, conf->xattr_name, 4 * 4);
+        if (ret < 0)
+            gf_msg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+                   "failed to set dict value for %s", conf->xattr_name);
+
+        ret = dict_set_uint32(local->xattr_req, conf->link_xattr_name, 256);
+        if (ret < 0)
+            gf_msg(this->name, GF_LOG_WARNING, EINVAL, DHT_MSG_DICT_SET_FAILED,
+                   "failed to set dict value for %s", conf->link_xattr_name);
+
+        if (!hashed_subvol) {
+            gf_msg_debug(this->name, 0,
+                         "no subvolume in layout for path=%s, "
+                         "checking on all the subvols to see if "
+                         "it is a directory",
+                         loc->path);
+            call_cnt = conf->subvolume_cnt;
+            local->call_cnt = call_cnt;
+
+            local->layout = dht_layout_new(this, conf->subvolume_cnt);
+            if (!local->layout) {
+                op_errno = ENOMEM;
+                goto err;
+            }
+
+            for (i = 0; i < call_cnt; i++) {
+                STACK_WIND_COOKIE(frame, dht_lookup_dir_cbk,
+                                  conf->subvolumes[i], conf->subvolumes[i],
+                                  conf->subvolumes[i]->fops->lookup,
+                                  &local->loc, local->xattr_req);
+            }
+            return 0;
         }
 
-        return 0;
+        /*  */
+        cached_subvol = get_switch_matching_subvol(loc->path, conf,
+                                                   hashed_subvol);
+        if (cached_subvol == hashed_subvol) {
+            STACK_WIND_COOKIE(frame, dht_lookup_cbk, hashed_subvol,
+                              hashed_subvol, hashed_subvol->fops->lookup, loc,
+                              local->xattr_req);
+        } else {
+            STACK_WIND_COOKIE(frame, switch_local_lookup_cbk, cached_subvol,
+                              cached_subvol, cached_subvol->fops->lookup, loc,
+                              local->xattr_req);
+        }
+    }
+
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-        DHT_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
-	return 0;
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+    return 0;
 }
 
 int
-switch_create_linkfile_create_cbk (call_frame_t *frame, void *cookie,
-				 xlator_t *this, int op_ret, int op_errno,
-                                 inode_t *inode, struct iatt *stbuf,
-                                 struct iatt *preparent,
-                                 struct iatt *postparent)
+switch_create_linkfile_create_cbk(call_frame_t *frame, void *cookie,
+                                  xlator_t *this, int op_ret, int op_errno,
+                                  inode_t *inode, struct iatt *stbuf,
+                                  struct iatt *preparent,
+                                  struct iatt *postparent, dict_t *xdata)
 {
-	dht_local_t  *local = NULL;
+    dht_local_t *local = NULL;
 
-	local = frame->local;
+    local = frame->local;
 
-	if (op_ret == -1)
-		goto err;
+    if (op_ret == -1)
+        goto err;
 
-	STACK_WIND (frame, dht_create_cbk,
-		    local->cached_subvol, local->cached_subvol->fops->create,
-		    &local->loc, local->flags, local->mode, local->fd,
-                    local->params);
+    STACK_WIND_COOKIE(frame, dht_create_cbk, local->cached_subvol,
+                      local->cached_subvol, local->cached_subvol->fops->create,
+                      &local->loc, local->flags, local->mode, local->umask,
+                      local->fd, local->params);
 
-	return 0;
+    return 0;
 
- err:
-	DHT_STACK_UNWIND (create, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL, NULL);
-	return 0;
+err:
+    DHT_STACK_UNWIND(create, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL,
+                     NULL);
+    return 0;
 }
 
 int
-switch_create (call_frame_t *frame, xlator_t *this,
-               loc_t *loc, int32_t flags, mode_t mode,
-               fd_t *fd, dict_t *params)
+switch_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+              mode_t mode, mode_t umask, fd_t *fd, dict_t *params)
 {
-	dht_local_t *local = NULL;
-	dht_conf_t  *conf  = NULL;
-	xlator_t    *subvol = NULL;
-        xlator_t    *avail_subvol = NULL;
-	int          op_errno = -1;
-	int          ret = -1;
-
-	VALIDATE_OR_GOTO (frame, err);
-	VALIDATE_OR_GOTO (this, err);
-	VALIDATE_OR_GOTO (loc, err);
-
-	conf  = this->private;
-
-        dht_get_du_info (frame, this, loc);
-
-        local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	subvol = dht_subvol_get_hashed (this, loc);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no subvolume in layout for path=%s",
-			loc->path);
-		op_errno = ENOENT;
-		goto err;
-	}
-
-        avail_subvol = get_switch_matching_subvol (loc->path, conf, subvol);
-        if (dht_is_subvol_filled (this, avail_subvol)) {
-                avail_subvol =
-                        dht_free_disk_available_subvol (this, avail_subvol);
-        }
-
-        if (subvol != avail_subvol) {
-                /* create a link file instead of actual file */
-                ret = loc_copy (&local->loc, loc);
-                if (ret == -1) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "Out of memory");
-                                op_errno = ENOMEM;
-                                goto err;
-                }
-
-                local->fd = fd_ref (fd);
-                local->mode = mode;
-                local->flags = flags;
-
-                local->cached_subvol = avail_subvol;
-                dht_linkfile_create (frame,
-                                     switch_create_linkfile_create_cbk,
-                                     avail_subvol, subvol, loc);
-                return 0;
-        }
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *avail_subvol = NULL;
+    int op_errno = -1;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+
+    conf = this->private;
+
+    dht_get_du_info(frame, this, loc);
+
+    local = dht_local_init(frame, loc, fd, GF_FOP_CREATE);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = dht_subvol_get_hashed(this, loc);
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no subvolume in layout for path=%s",
+                     loc->path);
+        op_errno = ENOENT;
+        goto err;
+    }
+
+    avail_subvol = get_switch_matching_subvol(loc->path, conf, subvol);
+    if (dht_is_subvol_filled(this, avail_subvol)) {
+        avail_subvol = dht_free_disk_available_subvol(this, avail_subvol,
+                                                      local);
+    }
+
+    if (subvol != avail_subvol) {
+        /* create a link file instead of actual file */
+        local->mode = mode;
+        local->flags = flags;
+        local->umask = umask;
+        local->cached_subvol = avail_subvol;
+        dht_linkfile_create(frame, switch_create_linkfile_create_cbk, this,
+                            avail_subvol, subvol, loc);
+        return 0;
+    }
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "creating %s on %s", loc->path, subvol->name);
+    gf_msg_trace(this->name, 0, "creating %s on %s", loc->path, subvol->name);
 
-        STACK_WIND (frame, dht_create_cbk,
-                    subvol, subvol->fops->create,
-                    loc, flags, mode, fd, params);
+    STACK_WIND_COOKIE(frame, dht_create_cbk, subvol, subvol,
+                      subvol->fops->create, loc, flags, mode, umask, fd,
+                      params);
 
-        return 0;
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (create, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL, NULL);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(create, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL,
+                     NULL);
 
-	return 0;
+    return 0;
 }
 
 int
-switch_mknod_linkfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int op_ret, int op_errno, inode_t *inode,
-                         struct iatt *stbuf, struct iatt *preparent,
-                         struct iatt *postparent)
+switch_mknod_linkfile_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int op_ret, int op_errno, inode_t *inode,
+                          struct iatt *stbuf, struct iatt *preparent,
+                          struct iatt *postparent, dict_t *xdata)
 {
-	dht_local_t  *local = NULL;
-
-	local = frame->local;
+    dht_local_t *local = NULL;
 
-	if (op_ret >= 0) {
-		STACK_WIND (frame, dht_newfile_cbk,
-			    local->cached_subvol,
-			    local->cached_subvol->fops->mknod,
-			    &local->loc, local->mode, local->rdev,
-                            local->params);
+    local = frame->local;
+    if (!local || !local->cached_subvol) {
+        op_errno = EINVAL;
+        op_ret = -1;
+        goto err;
+    }
 
-		return 0;
-	}
+    if (op_ret >= 0) {
+        STACK_WIND_COOKIE(
+            frame, dht_newfile_cbk, (void *)local->cached_subvol,
+            local->cached_subvol, local->cached_subvol->fops->mknod,
+            &local->loc, local->mode, local->rdev, local->umask, local->params);
 
-	DHT_STACK_UNWIND (link, frame, op_ret, op_errno,
-                          inode, stbuf, preparent, postparent);
-	return 0;
+        return 0;
+    }
+err:
+    DHT_STACK_UNWIND(link, frame, op_ret, op_errno, inode, stbuf, preparent,
+                     postparent, xdata);
+    return 0;
 }
 
-
 int
-switch_mknod (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, mode_t mode, dev_t rdev, dict_t *params)
+switch_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+             dev_t rdev, mode_t umask, dict_t *params)
 {
-	dht_local_t *local = NULL;
-	dht_conf_t  *conf  = NULL;
-	xlator_t    *subvol = NULL;
-        xlator_t    *avail_subvol = NULL;
-	int          op_errno = -1;
-	int          ret = -1;
-
-	VALIDATE_OR_GOTO (frame, err);
-	VALIDATE_OR_GOTO (this, err);
-	VALIDATE_OR_GOTO (loc, err);
-
-	conf  = this->private;
-
-        dht_get_du_info (frame, this, loc);
-
-        local = dht_local_init (frame);
-	if (!local) {
-		op_errno = ENOMEM;
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		goto err;
-	}
-
-	subvol = dht_subvol_get_hashed (this, loc);
-	if (!subvol) {
-		gf_log (this->name, GF_LOG_DEBUG,
-			"no subvolume in layout for path=%s",
-			loc->path);
-		op_errno = ENOENT;
-		goto err;
-	}
-
-        /* Consider the disksize in consideration */
-        avail_subvol = get_switch_matching_subvol (loc->path, conf, subvol);
-        if (dht_is_subvol_filled (this, avail_subvol)) {
-                avail_subvol =
-                        dht_free_disk_available_subvol (this, avail_subvol);
-        }
-
-	if (avail_subvol != subvol) {
-		/* Create linkfile first */
-		ret = loc_copy (&local->loc, loc);
-		if (ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"Out of memory");
-			op_errno = ENOMEM;
-			goto err;
-		}
-
-                local->params = dict_ref (params);
-		local->mode = mode;
-		local->rdev = rdev;
-		local->cached_subvol = avail_subvol;
-
-		dht_linkfile_create (frame, switch_mknod_linkfile_cbk,
-                                     avail_subvol, subvol, loc);
-		return 0;
-	}
+    dht_local_t *local = NULL;
+    dht_conf_t *conf = NULL;
+    xlator_t *subvol = NULL;
+    xlator_t *avail_subvol = NULL;
+    int op_errno = -1;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+
+    conf = this->private;
+
+    dht_get_du_info(frame, this, loc);
+
+    local = dht_local_init(frame, loc, NULL, GF_FOP_MKNOD);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    subvol = dht_subvol_get_hashed(this, loc);
+    if (!subvol) {
+        gf_msg_debug(this->name, 0, "no subvolume in layout for path=%s",
+                     loc->path);
+        op_errno = ENOENT;
+        goto err;
+    }
+
+    /* Consider the disksize in consideration */
+    avail_subvol = get_switch_matching_subvol(loc->path, conf, subvol);
+    if (dht_is_subvol_filled(this, avail_subvol)) {
+        avail_subvol = dht_free_disk_available_subvol(this, avail_subvol,
+                                                      local);
+    }
+
+    if (avail_subvol != subvol) {
+        /* Create linkfile first */
+
+        local->params = dict_ref(params);
+        local->mode = mode;
+        local->umask = umask;
+        local->rdev = rdev;
+        local->cached_subvol = avail_subvol;
+
+        dht_linkfile_create(frame, switch_mknod_linkfile_cbk, this,
+                            avail_subvol, subvol, loc);
+        return 0;
+    }
 
-	gf_log (this->name, GF_LOG_TRACE,
-		"creating %s on %s", loc->path, subvol->name);
+    gf_msg_trace(this->name, 0, "creating %s on %s", loc->path, subvol->name);
 
-	STACK_WIND (frame, dht_newfile_cbk,
-		    subvol, subvol->fops->mknod,
-		    loc, mode, rdev, params);
+    STACK_WIND_COOKIE(frame, dht_newfile_cbk, (void *)subvol, subvol,
+                      subvol->fops->mknod, loc, mode, rdev, umask, params);
 
-	return 0;
+    return 0;
 
 err:
-	op_errno = (op_errno == -1) ? errno : op_errno;
-	DHT_STACK_UNWIND (mknod, frame, -1, op_errno,
-                          NULL, NULL, NULL, NULL);
-
-	return 0;
-}
-
-
-int
-notify (xlator_t *this, int event, void *data, ...)
-{
-	int ret = -1;
-
-	ret = dht_notify (this, event, data);
+    op_errno = (op_errno == -1) ? errno : op_errno;
+    DHT_STACK_UNWIND(mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
 
-	return ret;
+    return 0;
 }
 
 void
-fini (xlator_t *this)
+switch_fini(xlator_t *this)
 {
-        int                   i = 0;
-        dht_conf_t           *conf = NULL;
-        struct switch_struct *trav = NULL;
-        struct switch_struct *prev = NULL;
-
-	conf = this->private;
-
-        if (conf) {
-                trav = (struct switch_struct *)conf->private;
-                conf->private = NULL;
-                while (trav) {
-                        if (trav->array)
-                                GF_FREE (trav->array);
-                        prev = trav;
-                        trav = trav->next;
-                        GF_FREE (prev);
-                }
-
-                if (conf->file_layouts) {
-                        for (i = 0; i < conf->subvolume_cnt; i++) {
-                                GF_FREE (conf->file_layouts[i]);
-                        }
-                        GF_FREE (conf->file_layouts);
-                }
-
-                if (conf->default_dir_layout)
-                        GF_FREE (conf->default_dir_layout);
-
-                if (conf->subvolumes)
-                        GF_FREE (conf->subvolumes);
-
-		if (conf->subvolume_status)
-			GF_FREE (conf->subvolume_status);
-
-                GF_FREE (conf);
+    dht_conf_t *conf = NULL;
+    struct switch_struct *trav = NULL;
+    struct switch_struct *prev = NULL;
+
+    conf = this->private;
+
+    if (conf) {
+        trav = (struct switch_struct *)conf->private;
+        conf->private = NULL;
+        while (trav) {
+            GF_FREE(trav->array);
+            prev = trav;
+            trav = trav->next;
+            GF_FREE(prev);
         }
+    }
 
-	return;
+    dht_fini(this);
 }
 
 int
-set_switch_pattern (xlator_t *this, dht_conf_t *conf,
-                    const char *pattern_str)
+set_switch_pattern(xlator_t *this, dht_conf_t *conf, const char *pattern_str)
 {
-        int                         flag = 0;
-        int                         idx = 0;
-        int                         index = 0;
-        int                         child_count = 0;
-        char                       *tmp = NULL;
-        char                       *tmp1 = NULL;
-        char                       *child = NULL;
-        char                       *tmp_str = NULL;
-        char                       *tmp_str1 = NULL;
-        char                       *dup_str = NULL;
-        char                       *dup_childs = NULL;
-        char                       *switch_str = NULL;
-        char                       *pattern = NULL;
-        char                       *childs = NULL;
-        char                       *option_string = NULL;
-        struct switch_struct        *switch_buf = NULL;
-        struct switch_struct        *switch_opt = NULL;
-        struct switch_struct        *trav = NULL;
-	struct switch_sched_array  *switch_buf_array = NULL;
-	xlator_list_t              *trav_xl = NULL;
-
-        trav_xl = this->children;
-	while (trav_xl) {
-		index++;
-		trav_xl = trav_xl->next;
-	}
-	child_count = index;
-	switch_buf_array = GF_CALLOC ((index + 1),
-                                      sizeof (struct switch_sched_array),
-                                      gf_switch_mt_switch_sched_array);
-        if (!switch_buf_array)
-                goto err;
+    int flag = 0;
+    int idx = 0;
+    int index = 0;
+    int child_count = 0;
+    char *tmp = NULL;
+    char *tmp1 = NULL;
+    char *child = NULL;
+    char *tmp_str = NULL;
+    char *tmp_str1 = NULL;
+    char *dup_str = NULL;
+    char *dup_childs = NULL;
+    char *switch_str = NULL;
+    char *pattern = NULL;
+    char *childs = NULL;
+    char *option_string = NULL;
+    size_t pattern_length;
+    struct switch_struct *switch_buf = NULL;
+    struct switch_struct *switch_opt = NULL;
+    struct switch_struct *trav = NULL;
+    struct switch_sched_array *switch_buf_array = NULL;
+    xlator_list_t *trav_xl = NULL;
+
+    trav_xl = this->children;
+    while (trav_xl) {
+        index++;
+        trav_xl = trav_xl->next;
+    }
+    child_count = index;
+    switch_buf_array = GF_CALLOC((index + 1), sizeof(struct switch_sched_array),
+                                 gf_switch_mt_switch_sched_array);
+    if (!switch_buf_array)
+        goto err;
+
+    trav_xl = this->children;
+    index = 0;
+
+    while (trav_xl) {
+        switch_buf_array[index].xl = trav_xl->xlator;
+        switch_buf_array[index].eligible = 1;
+        trav_xl = trav_xl->next;
+        index++;
+    }
+
+    /*  *jpg:child1,child2;*mpg:child3;*:child4,child5,child6 */
+
+    /* Get the pattern for considering switch case.
+       "option block-size *avi:10MB" etc */
+    option_string = gf_strdup(pattern_str);
+    if (option_string == NULL) {
+        goto err;
+    }
+    switch_str = strtok_r(option_string, ";", &tmp_str);
+    while (switch_str) {
+        dup_str = gf_strdup(switch_str);
+        if (dup_str == NULL) {
+            goto err;
+        }
+        switch_opt = GF_CALLOC(1, sizeof(struct switch_struct),
+                               gf_switch_mt_switch_struct);
+        if (!switch_opt) {
+            GF_FREE(dup_str);
+            goto err;
+        }
 
-	trav_xl = this->children;
-	index = 0;
-
-	while (trav_xl) {
-		switch_buf_array[index].xl = trav_xl->xlator;
-		switch_buf_array[index].eligible = 1;
-		trav_xl = trav_xl->next;
-		index++;
-	}
-
-	/*  *jpg:child1,child2;*mpg:child3;*:child4,child5,child6 */
-
-        /* Get the pattern for considering switch case.
-           "option block-size *avi:10MB" etc */
-        option_string = gf_strdup (pattern_str);
-        switch_str = strtok_r (option_string, ";", &tmp_str);
-        while (switch_str) {
-                dup_str = gf_strdup (switch_str);
-                switch_opt = GF_CALLOC (1, sizeof (struct switch_struct),
-                                        gf_switch_mt_switch_struct);
-                if (!switch_opt)
-                        goto err;
-
-                pattern = strtok_r (dup_str, ":", &tmp_str1);
-                childs = strtok_r (NULL, ":", &tmp_str1);
-                if (strncmp (pattern, "*", 2) == 0) {
-                        gf_log ("switch", GF_LOG_NORMAL,
-                                "'*' pattern will be taken by default "
-                                "for all the unconfigured child nodes,"
-                                " hence neglecting current option");
-                        switch_str = strtok_r (NULL, ";", &tmp_str);
-                        GF_FREE (dup_str);
-                        continue;
-                }
-                memcpy (switch_opt->path_pattern, pattern, strlen (pattern));
-                if (childs) {
-                        dup_childs = gf_strdup (childs);
-                        child = strtok_r (dup_childs, ",", &tmp);
-                        while (child) {
-                                if (gf_switch_valid_child (this, child)) {
-                                        idx++;
-                                        child = strtok_r (NULL, ",", &tmp);
-                                } else {
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                "%s is not a subvolume of %s. "
-                                                "pattern can only be scheduled "
-                                                "only to a subvolume of %s",
-                                                child, this->name, this->name);
-                                        goto err;
-                                }
-                        }
-                        GF_FREE (dup_childs);
-                        child = strtok_r (childs, ",", &tmp1);
-                        switch_opt->num_child = idx;
-                        switch_opt->array = GF_CALLOC (1, (idx *
-                                                       sizeof (struct switch_sched_array)),
-                                                       gf_switch_mt_switch_sched_array);
-                        if (!switch_opt->array)
-                                goto err;
-                        idx = 0;
-                        while (child) {
-                                for (index = 0; index < child_count; index++) {
-                                        if (strcmp (switch_buf_array[index].xl->name,
-                                                    child) == 0) {
-                                                gf_log ("switch", GF_LOG_DEBUG,
-                                                        "'%s' pattern will be "
-                                                        "scheduled to \"%s\"",
-                                                        switch_opt->path_pattern, child);
-                                                /*
-                                                  if (switch_buf_array[index-1].considered) {
-                                                  gf_log ("switch", GF_LOG_DEBUG,
-                                                  "ambiguity found, exiting");
-                                                  return -1;
-                                                  }
-                                                */
-                                                switch_opt->array[idx].xl = switch_buf_array[index].xl;
-                                                switch_buf_array[index].considered = 1;
-                                                idx++;
-                                                break;
-                                        }
-                                }
-                                child = strtok_r (NULL, ",", &tmp1);
-                        }
+        pattern = strtok_r(dup_str, ":", &tmp_str1);
+        childs = strtok_r(NULL, ":", &tmp_str1);
+        if (strncmp(pattern, "*", 2) == 0) {
+            gf_msg("switch", GF_LOG_INFO, 0, DHT_MSG_SWITCH_PATTERN_INFO,
+                   "'*' pattern will be taken by default "
+                   "for all the unconfigured child nodes,"
+                   " hence neglecting current option");
+            switch_str = strtok_r(NULL, ";", &tmp_str);
+            GF_FREE(switch_opt);
+            switch_opt = NULL;
+            GF_FREE(dup_str);
+            continue;
+        }
+        GF_FREE(dup_str);
+
+        pattern_length = strlen(pattern);
+        if (pattern_length >= (sizeof(switch_opt->path_pattern))) {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   DHT_MSG_SET_SWITCH_PATTERN_ERROR, "Pattern (%s) too long",
+                   pattern);
+            goto err;
+        }
+        memcpy(switch_opt->path_pattern, pattern, pattern_length);
+        switch_opt->path_pattern[pattern_length] = '\0';
+
+        if (childs) {
+            dup_childs = gf_strdup(childs);
+            if (dup_childs == NULL) {
+                goto err;
+            }
+            child = strtok_r(dup_childs, ",", &tmp);
+            while (child) {
+                if (gf_switch_valid_child(this, child)) {
+                    idx++;
+                    child = strtok_r(NULL, ",", &tmp);
                 } else {
-                        /* error */
-                        gf_log ("switch", GF_LOG_ERROR,
-                                "Check \"scheduler.switch.case\" "
-                                "option in unify volume. Exiting");
-                        goto err;
+                    gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SUBVOL_ERROR,
+                           "%s is not a subvolume of %s. "
+                           "pattern can only be scheduled "
+                           "only to a subvolume of %s",
+                           child, this->name, this->name);
+                    GF_FREE(dup_childs);
+                    goto err;
                 }
-                GF_FREE (dup_str);
-                
-                /* Link it to the main structure */
-                if (switch_buf) {
-                        /* there are already few entries */
-                        trav = switch_buf;
-                        while (trav->next)
-                                trav = trav->next;
-                        trav->next = switch_opt;
-                } else {
-                        /* First entry */
-                        switch_buf = switch_opt;
+            }
+            GF_FREE(dup_childs);
+            child = strtok_r(childs, ",", &tmp1);
+            switch_opt->num_child = idx;
+            switch_opt->array = GF_CALLOC(
+                1, (idx * sizeof(struct switch_sched_array)),
+                gf_switch_mt_switch_sched_array);
+            if (!switch_opt->array)
+                goto err;
+            idx = 0;
+            while (child) {
+                for (index = 0; index < child_count; index++) {
+                    if (strcmp(switch_buf_array[index].xl->name, child) == 0) {
+                        gf_msg_debug("switch", 0,
+                                     "'%s' pattern will be "
+                                     "scheduled to \"%s\"",
+                                     switch_opt->path_pattern, child);
+                        /*
+                          if (switch_buf_array[index-1].considered) {
+                          gf_msg_debug ("switch", 0,
+                          "ambiguity found, exiting");
+                          return -1;
+                          }
+                        */
+                        switch_opt->array[idx].xl = switch_buf_array[index].xl;
+                        switch_buf_array[index].considered = 1;
+                        idx++;
+                        break;
+                    }
                 }
-                switch_str = strtok_r (NULL, ";", &tmp_str);
+                child = strtok_r(NULL, ",", &tmp1);
+            }
+        } else {
+            /* error */
+            gf_msg("switch", GF_LOG_ERROR, 0, DHT_MSG_SET_SWITCH_PATTERN_ERROR,
+                   "Check \"scheduler.switch.case\" "
+                   "option in unify volume. Exiting");
+            goto err;
         }
 
-	/* Now, all the pattern based considerations done, so for all the
-	 * remaining pattern, '*' to all the remaining child nodes
-	 */
-	{
-		for (index=0; index < child_count; index++) {
-			/* check for considered flag */
-			if (switch_buf_array[index].considered)
-				continue;
-			flag++;
-		}
-		if (!flag) {
-			gf_log ("switch", GF_LOG_ERROR,
-				"No nodes left for pattern '*'. Exiting");
-			goto err;
-		}
-		switch_opt = GF_CALLOC (1, sizeof (struct switch_struct),
-                                        gf_switch_mt_switch_struct);
-                if (!switch_opt)
-                        goto err;
-
-		/* Add the '*' pattern to the array */
-		memcpy (switch_opt->path_pattern, "*", 2);
-		switch_opt->num_child = flag;
-		switch_opt->array =
-			GF_CALLOC (1, 
-                                   flag * sizeof (struct switch_sched_array),
-                                   gf_switch_mt_switch_sched_array);
-                if (!switch_opt->array)
-                        goto err;
-		flag = 0;
-		for (index=0; index < child_count; index++) {
-			/* check for considered flag */
-			if (switch_buf_array[index].considered)
-				continue;
-			gf_log ("switch", GF_LOG_DEBUG,
-				"'%s' pattern will be scheduled to \"%s\"",
-				switch_opt->path_pattern,
-				switch_buf_array[index].xl->name);
-			switch_opt->array[flag].xl =
-				switch_buf_array[index].xl;
-			switch_buf_array[index].considered = 1;
-			flag++;
-                }
-		if (switch_buf) {
-			/* there are already few entries */
-			trav = switch_buf;
-			while (trav->next)
-				trav = trav->next;
-			trav->next = switch_opt;
-		} else {
-			/* First entry */
-			switch_buf = switch_opt;
-		}
-	}
-        /* */
-        conf->private = switch_buf;
-
-        return 0;
-err:
+        /* Link it to the main structure */
         if (switch_buf) {
-                if (switch_buf_array)
-                        GF_FREE (switch_buf_array);
-                trav = switch_buf;
-                while (trav) {
-                        if (trav->array)
-                                GF_FREE (trav->array);
-                        switch_opt = trav;
-                        trav = trav->next;
-                        GF_FREE (switch_opt);
-                }
+            /* there are already few entries */
+            trav = switch_buf;
+            while (trav->next)
+                trav = trav->next;
+            trav->next = switch_opt;
+        } else {
+            /* First entry */
+            switch_buf = switch_opt;
         }
-        return -1;
-}
-
-
-int
-init (xlator_t *this)
-{
-        dht_conf_t            *conf = NULL;
-	data_t                *data = NULL;
-	char                  *temp_str = NULL;
-        int                    ret = -1;
-        int                    i = 0;
-	uint32_t               temp_free_disk = 0;
-
-	if (!this->children) {
-		gf_log (this->name, GF_LOG_CRITICAL,
-			"SWITCH needs more than one subvolume");
-		return -1;
-	}
-
-	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"dangling volume. check volfile");
-	}
-
-        conf = GF_CALLOC (1, sizeof (*conf), gf_switch_mt_dht_conf_t);
-        if (!conf) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                goto err;
+        switch_opt = NULL;
+        switch_str = strtok_r(NULL, ";", &tmp_str);
+    }
+
+    /* Now, all the pattern based considerations done, so for all the
+     * remaining pattern, '*' to all the remaining child nodes
+     */
+    {
+        for (index = 0; index < child_count; index++) {
+            /* check for considered flag */
+            if (switch_buf_array[index].considered)
+                continue;
+            flag++;
         }
-
-        conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_ON;
-	if (dict_get_str (this->options, "lookup-unhashed", &temp_str) == 0) {
-                /* If option is not "auto", other options _should_ be boolean */
-                if (strcasecmp (temp_str, "auto"))
-                        gf_string2boolean (temp_str, &conf->search_unhashed);
-                else
-                        conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
-	}
-
-	conf->unhashed_sticky_bit = 0;
-	if (dict_get_str (this->options, "unhashed-sticky-bit",
-                          &temp_str) == 0) {
-	        gf_string2boolean (temp_str, &conf->unhashed_sticky_bit);
-	}
-
-        conf->min_free_disk = 10;
-        conf->disk_unit = 'p';
-
-        if (dict_get_str (this->options, "min-free-disk",
-                          &temp_str) == 0) {
-                if (gf_string2percent (temp_str,
-                                       &temp_free_disk) == 0) {
-                        if (temp_free_disk > 100) {
-                                gf_string2bytesize (temp_str,
-                                                        &conf->min_free_disk);
-                                conf->disk_unit = 'b';
-                        } else {
-                                conf->min_free_disk = (uint64_t)temp_free_disk;
-                                conf->disk_unit = 'p';
-                        }
-                } else {
-                        gf_string2bytesize (temp_str,
-                                                &conf->min_free_disk);
-                        conf->disk_unit = 'b';
-                }
+        if (!flag) {
+            gf_msg("switch", GF_LOG_ERROR, 0, DHT_MSG_SET_SWITCH_PATTERN_ERROR,
+                   "No nodes left for pattern '*'. Exiting");
+            goto err;
         }
-
-	data = dict_get (this->options, "pattern.switch.case");
-	if (data) {
-                /* TODO: */
-                ret = set_switch_pattern (this, conf, data->data);
-                if (ret) {
-                        goto err;
-                }
-	}
-
-        ret = dht_init_subvolumes (this, conf);
-        if (ret == -1) {
-                goto err;
+        switch_opt = GF_CALLOC(1, sizeof(struct switch_struct),
+                               gf_switch_mt_switch_struct);
+        if (!switch_opt)
+            goto err;
+
+        /* Add the '*' pattern to the array */
+        memcpy(switch_opt->path_pattern, "*", 2);
+        switch_opt->num_child = flag;
+        switch_opt->array = GF_CALLOC(1,
+                                      flag * sizeof(struct switch_sched_array),
+                                      gf_switch_mt_switch_sched_array);
+        if (!switch_opt->array)
+            goto err;
+        flag = 0;
+        for (index = 0; index < child_count; index++) {
+            /* check for considered flag */
+            if (switch_buf_array[index].considered)
+                continue;
+            gf_msg_debug("switch", 0,
+                         "'%s'"
+                         " pattern will be scheduled to \"%s\"",
+                         switch_opt->path_pattern,
+                         switch_buf_array[index].xl->name);
+
+            switch_opt->array[flag].xl = switch_buf_array[index].xl;
+            switch_buf_array[index].considered = 1;
+            flag++;
         }
-
-        ret = dht_layouts_init (this, conf);
-        if (ret == -1) {
-                goto err;
+        if (switch_buf) {
+            /* there are already few entries */
+            trav = switch_buf;
+            while (trav->next)
+                trav = trav->next;
+            trav->next = switch_opt;
+        } else {
+            /* First entry */
+            switch_buf = switch_opt;
         }
+        switch_opt = NULL;
+    }
+    /* */
+    conf->private = switch_buf;
 
-	LOCK_INIT (&conf->subvolume_lock);
-	LOCK_INIT (&conf->layout_lock);
+    GF_FREE(option_string);
+    return 0;
+err:
+    GF_FREE(switch_buf_array);
+    GF_FREE(switch_opt);
+    GF_FREE(option_string);
+
+    if (switch_buf) {
+        trav = switch_buf;
+        while (trav) {
+            GF_FREE(trav->array);
+            switch_opt = trav;
+            trav = trav->next;
+            GF_FREE(switch_opt);
+        }
+    }
+    return -1;
+}
 
-	conf->gen = 1;
+int32_t
+switch_init(xlator_t *this)
+{
+    dht_conf_t *conf = NULL;
+    data_t *data = NULL;
+    int ret = -1;
 
-        conf->du_stats = GF_CALLOC (conf->subvolume_cnt, sizeof (dht_du_t),
-                                    gf_switch_mt_dht_du_t);
-        if (!conf->du_stats) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                goto err;
+    ret = dht_init(this);
+    if (ret) {
+        return ret;
+    }
+    conf = this->private;
+
+    data = dict_get(this->options, "pattern.switch.case");
+    if (data) {
+        /* TODO: */
+        ret = set_switch_pattern(this, conf, data->data);
+        if (ret) {
+            goto err;
         }
+    }
 
-        this->private = conf;
-
-        return 0;
+    this->private = conf;
+    return 0;
 
 err:
-        if (conf) {
-                if (conf->file_layouts) {
-                        for (i = 0; i < conf->subvolume_cnt; i++) {
-                                GF_FREE (conf->file_layouts[i]);
-                        }
-                        GF_FREE (conf->file_layouts);
-                }
-
-                if (conf->default_dir_layout)
-                        GF_FREE (conf->default_dir_layout);
-
-                if (conf->subvolumes)
-                        GF_FREE (conf->subvolumes);
-
-		if (conf->subvolume_status)
-			GF_FREE (conf->subvolume_status);
-
-                if (conf->du_stats)
-                        GF_FREE (conf->du_stats);
-
-                GF_FREE (conf);
-        }
-
-        return -1;
+    dht_fini(this);
+    return -1;
 }
 
-
 struct xlator_fops fops = {
-	.lookup      = switch_lookup,
-	.create      = switch_create,
-	.mknod       = switch_mknod,
-
-	.stat        = dht_stat,
-	.fstat       = dht_fstat,
-	.truncate    = dht_truncate,
-	.ftruncate   = dht_ftruncate,
-	.access      = dht_access,
-	.readlink    = dht_readlink,
-	.setxattr    = dht_setxattr,
-	.getxattr    = dht_getxattr,
-	.removexattr = dht_removexattr,
-	.open        = dht_open,
-	.readv       = dht_readv,
-	.writev      = dht_writev,
-	.flush       = dht_flush,
-	.fsync       = dht_fsync,
-	.statfs      = dht_statfs,
-	.lk          = dht_lk,
-	.opendir     = dht_opendir,
-	.readdir     = dht_readdir,
-	.readdirp    = dht_readdirp,
-	.fsyncdir    = dht_fsyncdir,
-	.symlink     = dht_symlink,
-	.unlink      = dht_unlink,
-	.link        = dht_link,
-	.mkdir       = dht_mkdir,
-	.rmdir       = dht_rmdir,
-	.rename      = dht_rename,
-	.inodelk     = dht_inodelk,
-	.finodelk    = dht_finodelk,
-	.entrylk     = dht_entrylk,
-	.fentrylk    = dht_fentrylk,
-	.xattrop     = dht_xattrop,
-	.fxattrop    = dht_fxattrop,
-        .setattr     = dht_setattr,
+    .lookup = switch_lookup,
+    .create = switch_create,
+    .mknod = switch_mknod,
+
+    .stat = dht_stat,
+    .fstat = dht_fstat,
+    .truncate = dht_truncate,
+    .ftruncate = dht_ftruncate,
+    .access = dht_access,
+    .readlink = dht_readlink,
+    .setxattr = dht_setxattr,
+    .getxattr = dht_getxattr,
+    .removexattr = dht_removexattr,
+    .open = dht_open,
+    .readv = dht_readv,
+    .writev = dht_writev,
+    .flush = dht_flush,
+    .fsync = dht_fsync,
+    .statfs = dht_statfs,
+    .lk = dht_lk,
+    .opendir = dht_opendir,
+    .readdir = dht_readdir,
+    .readdirp = dht_readdirp,
+    .fsyncdir = dht_fsyncdir,
+    .symlink = dht_symlink,
+    .unlink = dht_unlink,
+    .link = dht_link,
+    .mkdir = dht_mkdir,
+    .rmdir = dht_rmdir,
+    .rename = dht_rename,
+    .inodelk = dht_inodelk,
+    .finodelk = dht_finodelk,
+    .entrylk = dht_entrylk,
+    .fentrylk = dht_fentrylk,
+    .xattrop = dht_xattrop,
+    .fxattrop = dht_fxattrop,
+    .setattr = dht_setattr,
 };
 
-
-struct xlator_cbks cbks = {
-	.forget     = dht_forget
-};
-
-
-struct volume_options options[] = {
-        { .key  = {"lookup-unhashed"},
-          .value = {"auto", "yes", "no", "enable", "disable", "1", "0",
-                    "on", "off"},
-	  .type = GF_OPTION_TYPE_STR
-	},
-	{ .key  = {"pattern.switch.case"},
-	  .type = GF_OPTION_TYPE_ANY
-	},
-        { .key  = {"min-free-disk"},
-          .type = GF_OPTION_TYPE_PERCENT_OR_SIZET,
-        },
-	{ .key  = {NULL} },
+struct xlator_cbks cbks = {.forget = dht_forget};
+extern int32_t
+mem_acct_init(xlator_t *this);
+
+xlator_api_t xlator_api = {
+    .init = switch_init,
+    .fini = switch_fini,
+    .notify = dht_notify,
+    .reconfigure = dht_reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = dht_options,
+    .identifier = "switch",
+    .category = GF_TECH_PREVIEW,
 };
diff --git a/xlators/cluster/dht/src/unittest/dht_layout_mock.c b/xlators/cluster/dht/src/unittest/dht_layout_mock.c
new file mode 100644
index 00000000000..771452963d1
--- /dev/null
+++ b/xlators/cluster/dht/src/unittest/dht_layout_mock.c
@@ -0,0 +1,73 @@
+/*
+  Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/xlator.h>
+#include "dht-common.h"
+#include <glusterfs/byte-order.h>
+
+int
+dht_hash_compute(xlator_t *this, int type, const char *name, uint32_t *hash_p)
+{
+    return 0;
+}
+
+int
+dht_inode_ctx_layout_get(inode_t *inode, xlator_t *this, dht_layout_t **layout)
+{
+    return 0;
+}
+
+int
+dht_inode_ctx_layout_set(inode_t *inode, xlator_t *this,
+                         dht_layout_t *layout_int)
+{
+    return 0;
+}
+
+int
+dict_get_ptr(dict_t *this, char *key, void **ptr)
+{
+    return 0;
+}
+
+int
+dict_get_ptr_and_len(dict_t *this, char *key, void **ptr, int *len)
+{
+    return 0;
+}
+
+int
+_gf_log(const char *domain, const char *file, const char *function,
+        int32_t line, gf_loglevel_t level, const char *fmt, ...)
+{
+    return 0;
+}
+
+int
+_gf_log_callingfn(const char *domain, const char *file, const char *function,
+                  int32_t line, gf_loglevel_t level, const char *fmt, ...)
+{
+    return 0;
+}
+
+void
+gf_uuid_unparse(const uuid_t uu, char *out)
+{
+    // could call a will-return function here
+    // to place the correct data in *out
+}
+
+int
+_gf_msg(const char *domain, const char *file, const char *function,
+        int32_t line, gf_loglevel_t level, int errnum, int trace,
+        uint64_t msgid, const char *fmt, ...)
+{
+    return 0;
+}
diff --git a/xlators/cluster/dht/src/unittest/dht_layout_unittest.c b/xlators/cluster/dht/src/unittest/dht_layout_unittest.c
new file mode 100644
index 00000000000..c94a1d0a2e1
--- /dev/null
+++ b/xlators/cluster/dht/src/unittest/dht_layout_unittest.c
@@ -0,0 +1,127 @@
+/*
+  Copyright (c) 2008-2014 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "dht-common.h"
+#include <glusterfs/logging.h>
+#include <glusterfs/xlator.h>
+
+#include <inttypes.h>
+#include <stdarg.h>
+#include <stddef.h>
+#include <setjmp.h>
+#include <cmocka_pbc.h>
+#include <cmocka.h>
+
+/*
+ * Helper functions
+ */
+
+static xlator_t *
+helper_xlator_init(uint32_t num_types)
+{
+    xlator_t *xl;
+    int i, ret;
+
+    REQUIRE(num_types > 0);
+
+    xl = test_calloc(1, sizeof(xlator_t));
+    assert_non_null(xl);
+    xl->mem_acct->num_types = num_types;
+    xl->mem_acct = test_calloc(sizeof(struct mem_acct) +
+                               sizeof(struct mem_acct_rec) + num_types);
+    assert_non_null(xl->mem_acct);
+
+    xl->ctx = test_calloc(1, sizeof(glusterfs_ctx_t));
+    assert_non_null(xl->ctx);
+
+    for (i = 0; i < num_types; i++) {
+        ret = LOCK_INIT(&(xl->mem_acct.rec[i].lock));
+        assert_false(ret);
+    }
+
+    ENSURE(num_types == xl->mem_acct.num_types);
+    ENSURE(NULL != xl);
+
+    return xl;
+}
+
+static int
+helper_xlator_destroy(xlator_t *xl)
+{
+    int i, ret;
+
+    for (i = 0; i < xl->mem_acct.num_types; i++) {
+        ret = LOCK_DESTROY(&(xl->mem_acct.rec[i].lock));
+        assert_int_equal(ret, 0);
+    }
+
+    free(xl->mem_acct.rec);
+    free(xl->ctx);
+    free(xl);
+    return 0;
+}
+
+/*
+ * Unit tests
+ */
+static void
+test_dht_layout_new(void **state)
+{
+    xlator_t *xl;
+    dht_layout_t *layout;
+    dht_conf_t *conf;
+    int cnt;
+
+    expect_assert_failure(dht_layout_new(NULL, 0));
+    expect_assert_failure(dht_layout_new((xlator_t *)0x12345, -1));
+    xl = helper_xlator_init(10);
+
+    // xl->private is NULL
+    assert_null(xl->private);
+    cnt = 100;
+    layout = dht_layout_new(xl, cnt);
+    assert_non_null(layout);
+    assert_int_equal(layout->type, DHT_HASH_TYPE_DM);
+    assert_int_equal(layout->cnt, cnt);
+    assert_int_equal(GF_ATOMIC_GET(layout->ref), 1);
+    assert_int_equal(layout->gen, 0);
+    assert_int_equal(layout->spread_cnt, 0);
+    free(layout);
+
+    // xl->private is not NULL
+    cnt = 110;
+    conf = (dht_conf_t *)test_calloc(1, sizeof(dht_conf_t));
+    assert_non_null(conf);
+    conf->dir_spread_cnt = 12345;
+    conf->gen = -123;
+    xl->private = conf;
+
+    layout = dht_layout_new(xl, cnt);
+    assert_non_null(layout);
+    assert_int_equal(layout->type, DHT_HASH_TYPE_DM);
+    assert_int_equal(layout->cnt, cnt);
+    assert_int_equal(GF_ATOMIC_GET(layout->ref), 1);
+    assert_int_equal(layout->gen, conf->gen);
+    assert_int_equal(layout->spread_cnt, conf->dir_spread_cnt);
+    free(layout);
+
+    free(conf);
+    helper_xlator_destroy(xl);
+}
+
+int
+main(void)
+{
+    const struct CMUnitTest xlator_dht_layout_tests[] = {
+        unit_test(test_dht_layout_new),
+    };
+
+    return cmocka_run_group_tests(xlator_dht_layout_tests, NULL, NULL);
+}
diff --git a/xlators/cluster/ha/Makefile.am b/xlators/cluster/ec/Makefile.am
index d471a3f9243..d471a3f9243 100644
--- a/xlators/cluster/ha/Makefile.am
+++ b/xlators/cluster/ec/Makefile.am
diff --git a/xlators/cluster/ec/src/Makefile.am b/xlators/cluster/ec/src/Makefile.am
new file mode 100644
index 00000000000..406a636bbc2
--- /dev/null
+++ b/xlators/cluster/ec/src/Makefile.am
@@ -0,0 +1,83 @@
+xlator_LTLIBRARIES = ec.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/cluster
+
+ec_sources := ec.c
+ec_sources += ec-data.c
+ec_sources += ec-helpers.c
+ec_sources += ec-common.c
+ec_sources += ec-generic.c
+ec_sources += ec-locks.c
+ec_sources += ec-dir-read.c
+ec_sources += ec-dir-write.c
+ec_sources += ec-inode-read.c
+ec_sources += ec-inode-write.c
+ec_sources += ec-combine.c
+ec_sources += ec-method.c
+ec_sources += ec-galois.c
+ec_sources += ec-code.c
+ec_sources += ec-code-c.c
+ec_sources += ec-gf8.c
+ec_sources += ec-heal.c
+ec_sources += ec-heald.c
+
+ec_headers := ec.h
+ec_headers += ec-mem-types.h
+ec_headers += ec-helpers.h
+ec_headers += ec-data.h
+ec_headers += ec-fops.h
+ec_headers += ec-common.h
+ec_headers += ec-combine.h
+ec_headers += ec-method.h
+ec_headers += ec-galois.h
+ec_headers += ec-code.h
+ec_headers += ec-code-c.h
+ec_headers += ec-gf8.h
+ec_headers += ec-heald.h
+ec_headers += ec-messages.h
+ec_headers += ec-types.h
+
+if ENABLE_EC_DYNAMIC_INTEL
+  ec_sources += ec-code-intel.c
+  ec_headers += ec-code-intel.h
+endif
+
+if ENABLE_EC_DYNAMIC_X64
+  ec_sources += ec-code-x64.c
+  ec_headers += ec-code-x64.h
+endif
+
+if ENABLE_EC_DYNAMIC_SSE
+  ec_sources += ec-code-sse.c
+  ec_headers += ec-code-sse.h
+endif
+
+if ENABLE_EC_DYNAMIC_AVX
+  ec_sources += ec-code-avx.c
+  ec_headers += ec-code-avx.h
+endif
+
+ec_ext_sources = $(top_builddir)/xlators/lib/src/libxlator.c
+
+ec_ext_headers = $(top_builddir)/xlators/lib/src/libxlator.h
+
+ec_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+ec_la_SOURCES = $(ec_sources) $(ec_headers) $(ec_ext_sources) $(ec_ext_headers)
+ec_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+AM_CPPFLAGS = $(GF_CPPFLAGS)
+AM_CPPFLAGS += -I$(top_srcdir)/libglusterfs/src
+AM_CPPFLAGS += -I$(top_srcdir)/xlators/lib/src
+AM_CPPFLAGS += -I$(top_srcdir)/rpc/rpc-lib/src
+AM_CPPFLAGS += -I$(top_srcdir)/rpc/xdr/src
+AM_CPPFLAGS += -I$(top_builddir)/rpc/xdr/src
+AM_CPPFLAGS += -DGLUSTERFS_LIBEXECDIR=\"$(GLUSTERFS_LIBEXECDIR)\"
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
+
+install-data-hook:
+	ln -sf ec.so $(DESTDIR)$(xlatordir)/disperse.so
+
+uninstall-local:
+	rm -f $(DESTDIR)$(xlatordir)/disperse.so
diff --git a/xlators/cluster/ec/src/ec-code-avx.c b/xlators/cluster/ec/src/ec-code-avx.c
new file mode 100644
index 00000000000..70afaa00f54
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code-avx.c
@@ -0,0 +1,109 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <errno.h>
+
+#include "ec-code-intel.h"
+
+static void
+ec_code_avx_prolog(ec_code_builder_t *builder)
+{
+    builder->loop = builder->address;
+}
+
+static void
+ec_code_avx_epilog(ec_code_builder_t *builder)
+{
+    ec_code_intel_op_add_i2r(builder, 32, REG_DX);
+    ec_code_intel_op_add_i2r(builder, 32, REG_DI);
+    ec_code_intel_op_test_i2r(builder, builder->width - 1, REG_DX);
+    ec_code_intel_op_jne(builder, builder->loop);
+
+    ec_code_intel_op_ret(builder, 0);
+}
+
+static void
+ec_code_avx_load(ec_code_builder_t *builder, uint32_t dst, uint32_t idx,
+                 uint32_t bit)
+{
+    if (builder->linear) {
+        ec_code_intel_op_mov_m2avx(
+            builder, REG_SI, REG_DX, 1,
+            idx * builder->width * builder->bits + bit * builder->width, dst);
+    } else {
+        if (builder->base != idx) {
+            ec_code_intel_op_mov_m2r(builder, REG_SI, REG_NULL, 0, idx * 8,
+                                     REG_AX);
+            builder->base = idx;
+        }
+        ec_code_intel_op_mov_m2avx(builder, REG_AX, REG_DX, 1,
+                                   bit * builder->width, dst);
+    }
+}
+
+static void
+ec_code_avx_store(ec_code_builder_t *builder, uint32_t src, uint32_t bit)
+{
+    ec_code_intel_op_mov_avx2m(builder, src, REG_DI, REG_NULL, 0,
+                               bit * builder->width);
+}
+
+static void
+ec_code_avx_copy(ec_code_builder_t *builder, uint32_t dst, uint32_t src)
+{
+    ec_code_intel_op_mov_avx2avx(builder, src, dst);
+}
+
+static void
+ec_code_avx_xor2(ec_code_builder_t *builder, uint32_t dst, uint32_t src)
+{
+    ec_code_intel_op_xor_avx2avx(builder, src, dst);
+}
+
+static void
+ec_code_avx_xor3(ec_code_builder_t *builder, uint32_t dst, uint32_t src1,
+                 uint32_t src2)
+{
+    ec_code_intel_op_mov_avx2avx(builder, src1, dst);
+    ec_code_intel_op_xor_avx2avx(builder, src2, dst);
+}
+
+static void
+ec_code_avx_xorm(ec_code_builder_t *builder, uint32_t dst, uint32_t idx,
+                 uint32_t bit)
+{
+    if (builder->linear) {
+        ec_code_intel_op_xor_m2avx(
+            builder, REG_SI, REG_DX, 1,
+            idx * builder->width * builder->bits + bit * builder->width, dst);
+    } else {
+        if (builder->base != idx) {
+            ec_code_intel_op_mov_m2r(builder, REG_SI, REG_NULL, 0, idx * 8,
+                                     REG_AX);
+            builder->base = idx;
+        }
+        ec_code_intel_op_xor_m2avx(builder, REG_AX, REG_DX, 1,
+                                   bit * builder->width, dst);
+    }
+}
+
+static char *ec_code_avx_needed_flags[] = {"avx2", NULL};
+
+ec_code_gen_t ec_code_gen_avx = {.name = "avx",
+                                 .flags = ec_code_avx_needed_flags,
+                                 .width = 32,
+                                 .prolog = ec_code_avx_prolog,
+                                 .epilog = ec_code_avx_epilog,
+                                 .load = ec_code_avx_load,
+                                 .store = ec_code_avx_store,
+                                 .copy = ec_code_avx_copy,
+                                 .xor2 = ec_code_avx_xor2,
+                                 .xor3 = ec_code_avx_xor3,
+                                 .xorm = ec_code_avx_xorm};
diff --git a/xlators/cluster/ec/src/ec-code-avx.h b/xlators/cluster/ec/src/ec-code-avx.h
new file mode 100644
index 00000000000..fdca4ad2c8f
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code-avx.h
@@ -0,0 +1,18 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_CODE_AVX_H__
+#define __EC_CODE_AVX_H__
+
+#include "ec-code.h"
+
+extern ec_code_gen_t ec_code_gen_avx;
+
+#endif /* __EC_CODE_AVX_H__ */
diff --git a/xlators/cluster/ec/src/ec-code-c.c b/xlators/cluster/ec/src/ec-code-c.c
new file mode 100644
index 00000000000..acdc665c2cf
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code-c.c
@@ -0,0 +1,11679 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <inttypes.h>
+#include <string.h>
+
+#include "ec-method.h"
+#include "ec-code-c.h"
+
+#define WIDTH (EC_METHOD_WORD_SIZE / sizeof(uint64_t))
+
+static void
+gf8_muladd_00(void *out, void *in)
+{
+    memcpy(out, in, EC_METHOD_WORD_SIZE * 8);
+}
+
+static void
+gf8_muladd_01(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        out_ptr[0] ^= in_ptr[0];
+        out_ptr[WIDTH] ^= in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] ^= in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] ^= in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] ^= in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] ^= in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] ^= in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] ^= in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_02(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in7;
+        out1 = in0;
+        out7 = in6;
+        out5 = in4;
+        out6 = in5;
+        out3 = in2 ^ in7;
+        out4 = in3 ^ in7;
+        out2 = in1 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_03(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in0 ^ in7;
+        tmp0 = in2 ^ in7;
+        out1 = in0 ^ in1;
+        out7 = in6 ^ in7;
+        out5 = in4 ^ in5;
+        out6 = in5 ^ in6;
+        out4 = in3 ^ in4 ^ in7;
+        out2 = tmp0 ^ in1;
+        out3 = tmp0 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_04(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in6;
+        out1 = in7;
+        out7 = in5;
+        out6 = in4;
+        tmp0 = in6 ^ in7;
+        out2 = in0 ^ in6;
+        out5 = in3 ^ in7;
+        out3 = tmp0 ^ in1;
+        out4 = tmp0 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_05(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in0 ^ in6;
+        out1 = in1 ^ in7;
+        out7 = in5 ^ in7;
+        out6 = in4 ^ in6;
+        out2 = out0 ^ in2;
+        out3 = out1 ^ in3 ^ in6;
+        out5 = out7 ^ in3;
+        out4 = out6 ^ in2 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_06(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in6 ^ in7;
+        tmp0 = in1 ^ in6;
+        out1 = in0 ^ in7;
+        out7 = in5 ^ in6;
+        out6 = in4 ^ in5;
+        out4 = in2 ^ in3 ^ in6;
+        out5 = in3 ^ in4 ^ in7;
+        out3 = tmp0 ^ in2;
+        out2 = tmp0 ^ out1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_07(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in6;
+        tmp1 = in5 ^ in6;
+        tmp2 = in0 ^ in7;
+        tmp3 = tmp0 ^ in3;
+        out6 = tmp1 ^ in4;
+        out7 = tmp1 ^ in7;
+        out0 = tmp2 ^ in6;
+        out1 = tmp2 ^ in1;
+        out3 = tmp3 ^ in1;
+        out4 = tmp3 ^ in4;
+        out5 = out4 ^ out7 ^ in2;
+        out2 = tmp0 ^ out1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_08(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in5;
+        out1 = in6;
+        out7 = in4;
+        out6 = in3 ^ in7;
+        out3 = in0 ^ in5 ^ in6;
+        out5 = in2 ^ in6 ^ in7;
+        out2 = in5 ^ in7;
+        out4 = out2 ^ in1 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_09(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in0 ^ in5;
+        tmp0 = in3 ^ in6;
+        out1 = in1 ^ in6;
+        out7 = in4 ^ in7;
+        out2 = in2 ^ in5 ^ in7;
+        out3 = tmp0 ^ out0;
+        out6 = tmp0 ^ in7;
+        out4 = out1 ^ out7 ^ in5;
+        out5 = out2 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_0A(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in5 ^ in7;
+        out1 = in0 ^ in6;
+        out7 = in4 ^ in6;
+        out2 = in1 ^ in5;
+        out6 = out0 ^ in3;
+        out3 = out0 ^ out1 ^ in2;
+        out5 = out7 ^ in2 ^ in7;
+        out4 = out2 ^ in3 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_0B(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in5;
+        tmp1 = in0 ^ in6;
+        tmp2 = in4 ^ in7;
+        out0 = in0 ^ in5 ^ in7;
+        out2 = tmp0 ^ in1;
+        out1 = tmp1 ^ in1;
+        out6 = tmp1 ^ out0 ^ in3;
+        out7 = tmp2 ^ in6;
+        out4 = tmp2 ^ out6 ^ in1;
+        out3 = out6 ^ in0 ^ in2;
+        out5 = tmp0 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_0C(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in5 ^ in6;
+        out1 = in6 ^ in7;
+        out7 = in4 ^ in5;
+        tmp0 = in1 ^ in5;
+        tmp1 = in0 ^ in7;
+        out5 = in2 ^ in3 ^ in6;
+        out6 = in3 ^ in4 ^ in7;
+        out2 = tmp1 ^ out0;
+        out4 = tmp0 ^ in2;
+        out3 = tmp0 ^ tmp1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_0D(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in4 ^ in5;
+        tmp1 = in5 ^ in6;
+        out1 = in1 ^ in6 ^ in7;
+        out7 = tmp0 ^ in7;
+        out4 = tmp0 ^ in1 ^ in2;
+        out0 = tmp1 ^ in0;
+        tmp2 = tmp1 ^ in3;
+        out6 = tmp2 ^ out7;
+        out2 = out0 ^ in2 ^ in7;
+        out3 = out0 ^ out1 ^ in3;
+        out5 = tmp2 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_0E(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in1;
+        tmp1 = in2 ^ in5;
+        tmp2 = in5 ^ in6;
+        out1 = in0 ^ in6 ^ in7;
+        out3 = tmp0 ^ tmp1;
+        out2 = tmp0 ^ tmp2;
+        tmp3 = tmp1 ^ in3;
+        out7 = tmp2 ^ in4;
+        out0 = tmp2 ^ in7;
+        out4 = tmp3 ^ in1 ^ in7;
+        out5 = tmp3 ^ out7;
+        out6 = out0 ^ out5 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_0F(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in6 ^ in7;
+        tmp1 = tmp0 ^ in1;
+        tmp2 = tmp0 ^ in5;
+        out1 = tmp1 ^ in0;
+        out7 = tmp2 ^ in4;
+        out0 = tmp2 ^ in0;
+        out6 = out7 ^ in3;
+        out5 = out6 ^ in2 ^ in7;
+        tmp3 = tmp1 ^ out0 ^ in2;
+        out4 = tmp1 ^ out5;
+        out2 = tmp3 ^ in6;
+        out3 = tmp3 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_10(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in4;
+        out1 = in5;
+        out7 = in3 ^ in7;
+        tmp0 = in6 ^ in7;
+        out2 = in4 ^ in6;
+        tmp1 = out2 ^ in5;
+        out6 = tmp0 ^ in2;
+        out3 = tmp0 ^ tmp1;
+        out5 = out2 ^ out3 ^ in1;
+        out4 = tmp1 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_11(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out7 = in3;
+        out0 = in0 ^ in4;
+        out1 = in1 ^ in5;
+        out6 = in2 ^ in7;
+        out4 = in0 ^ in5 ^ in6;
+        out5 = in1 ^ in6 ^ in7;
+        out2 = in2 ^ in4 ^ in6;
+        out3 = in3 ^ in4 ^ in5 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_12(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in4 ^ in7;
+        out1 = in0 ^ in5;
+        out3 = in2 ^ in4 ^ in5;
+        tmp0 = out0 ^ in6;
+        out2 = tmp0 ^ in1;
+        tmp1 = tmp0 ^ in3;
+        out6 = tmp0 ^ out3;
+        out5 = out2 ^ in5;
+        out7 = tmp1 ^ in4;
+        out4 = tmp1 ^ out1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_13(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out7 = in3 ^ in6;
+        tmp0 = in0 ^ in5;
+        tmp1 = in4 ^ in7;
+        out6 = in2 ^ in5 ^ in7;
+        out4 = tmp0 ^ out7 ^ in7;
+        out1 = tmp0 ^ in1;
+        out0 = tmp1 ^ in0;
+        out5 = tmp1 ^ in1 ^ in6;
+        out3 = tmp1 ^ out6 ^ in3;
+        out2 = out5 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_14(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in4 ^ in6;
+        out1 = in5 ^ in7;
+        out2 = in0 ^ in4;
+        tmp0 = out0 ^ in5;
+        out7 = out1 ^ in3;
+        tmp1 = out1 ^ in2;
+        out3 = tmp0 ^ in1;
+        out6 = tmp0 ^ tmp1;
+        out4 = tmp1 ^ out2;
+        out5 = out3 ^ in3 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_15(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out7 = in3 ^ in5;
+        tmp0 = in0 ^ in4;
+        out1 = in1 ^ in5 ^ in7;
+        out5 = in1 ^ in3 ^ in6;
+        out0 = tmp0 ^ in6;
+        out2 = tmp0 ^ in2;
+        out3 = out5 ^ in4 ^ in5;
+        out6 = out2 ^ in0 ^ in7;
+        out4 = tmp0 ^ out6 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_16(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in5;
+        tmp1 = in4 ^ in7;
+        tmp2 = in2 ^ in3 ^ in4;
+        out1 = tmp0 ^ in7;
+        out4 = tmp0 ^ tmp2;
+        out0 = tmp1 ^ in6;
+        tmp3 = tmp1 ^ in1;
+        out6 = out0 ^ in2 ^ in5;
+        out2 = tmp3 ^ in0;
+        out3 = out6 ^ in1;
+        out7 = tmp2 ^ out6;
+        out5 = tmp3 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_17(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in5;
+        tmp1 = in3 ^ in6;
+        tmp2 = tmp0 ^ in4;
+        out4 = tmp0 ^ in0 ^ in3;
+        out7 = tmp1 ^ in5;
+        tmp3 = tmp1 ^ in1;
+        out6 = tmp2 ^ in7;
+        out5 = tmp3 ^ in4;
+        out3 = tmp3 ^ out6;
+        out0 = out3 ^ out4 ^ in1;
+        out2 = out3 ^ out7 ^ in0;
+        out1 = tmp2 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_18(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in4 ^ in5;
+        out1 = in5 ^ in6;
+        tmp0 = in4 ^ in7;
+        out5 = in1 ^ in2 ^ in5;
+        out6 = in2 ^ in3 ^ in6;
+        out2 = tmp0 ^ out1;
+        out7 = tmp0 ^ in3;
+        tmp1 = tmp0 ^ in0;
+        out3 = tmp1 ^ in6;
+        out4 = tmp1 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_19(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out5 = in1 ^ in2;
+        out7 = in3 ^ in4;
+        tmp0 = in0 ^ in7;
+        out6 = in2 ^ in3;
+        out1 = in1 ^ in5 ^ in6;
+        out0 = in0 ^ in4 ^ in5;
+        out4 = tmp0 ^ in1;
+        tmp1 = tmp0 ^ in6;
+        out2 = tmp1 ^ out0 ^ in2;
+        out3 = tmp1 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_1A(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in4 ^ in5;
+        tmp1 = in5 ^ in6;
+        tmp2 = tmp0 ^ in1;
+        out0 = tmp0 ^ in7;
+        out1 = tmp1 ^ in0;
+        tmp3 = tmp1 ^ in3;
+        out5 = tmp2 ^ in2;
+        out2 = tmp2 ^ in6;
+        out7 = tmp3 ^ out0;
+        out6 = tmp3 ^ in2;
+        out4 = tmp3 ^ out2 ^ in0;
+        out3 = tmp0 ^ out1 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_1B(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in4;
+        tmp1 = in2 ^ in5;
+        tmp2 = in3 ^ in6;
+        out5 = tmp0 ^ in1;
+        tmp3 = tmp0 ^ in0;
+        out6 = tmp1 ^ in3;
+        out0 = tmp1 ^ tmp3 ^ in7;
+        out7 = tmp2 ^ in4;
+        tmp4 = out5 ^ in6;
+        out3 = tmp2 ^ tmp3;
+        out2 = tmp4 ^ in5;
+        out4 = tmp4 ^ out3;
+        out1 = tmp3 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_1C(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in3;
+        tmp1 = in4 ^ in6;
+        tmp2 = in5 ^ in7;
+        out6 = tmp0 ^ tmp1;
+        out0 = tmp1 ^ in5;
+        out1 = tmp2 ^ in6;
+        tmp3 = tmp2 ^ in1;
+        tmp4 = tmp2 ^ in4;
+        out2 = tmp4 ^ in0;
+        out7 = tmp4 ^ in3;
+        out5 = tmp0 ^ tmp3;
+        out3 = tmp3 ^ out2;
+        out4 = out3 ^ in2 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_1D(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in3;
+        tmp1 = in0 ^ in4;
+        tmp2 = in3 ^ in4;
+        tmp3 = in2 ^ in7;
+        out3 = tmp0 ^ tmp1;
+        out5 = tmp0 ^ tmp3;
+        tmp4 = tmp1 ^ in5;
+        out6 = tmp2 ^ in2;
+        out7 = tmp2 ^ in5;
+        out2 = tmp3 ^ tmp4;
+        out4 = out3 ^ out6 ^ in6;
+        out0 = tmp4 ^ in6;
+        out1 = out2 ^ out4 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_1E(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in4;
+        tmp1 = in2 ^ in7;
+        tmp2 = tmp0 ^ in1;
+        out3 = tmp1 ^ tmp2;
+        out2 = tmp2 ^ in5;
+        out4 = out3 ^ in3 ^ in6;
+        tmp3 = out4 ^ in7;
+        out6 = tmp3 ^ out2 ^ in4;
+        out7 = tmp1 ^ out6;
+        out0 = out7 ^ in3;
+        out1 = tmp0 ^ out0;
+        out5 = tmp3 ^ out1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_1F(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in4 ^ in6;
+        tmp1 = tmp0 ^ in5;
+        out7 = tmp1 ^ in3;
+        out0 = tmp1 ^ in0 ^ in7;
+        out6 = out7 ^ in2 ^ in6;
+        out1 = out0 ^ in1 ^ in4;
+        out4 = out0 ^ out6 ^ in1;
+        out3 = tmp0 ^ out4;
+        out2 = out4 ^ out7 ^ in7;
+        out5 = out3 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_20(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in4;
+        out0 = in3 ^ in7;
+        tmp0 = in3 ^ in4;
+        tmp1 = in6 ^ in7;
+        out2 = out0 ^ in5;
+        out4 = tmp0 ^ in5;
+        out3 = tmp0 ^ tmp1;
+        out7 = tmp1 ^ in2;
+        out6 = tmp1 ^ in1 ^ in5;
+        out5 = out2 ^ out3 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_21(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in1 ^ in4;
+        tmp0 = in4 ^ in6;
+        out4 = in3 ^ in5;
+        out7 = in2 ^ in6;
+        out0 = in0 ^ in3 ^ in7;
+        out6 = in1 ^ in5 ^ in7;
+        out3 = tmp0 ^ in7;
+        out5 = tmp0 ^ in0;
+        out2 = out4 ^ in2 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_22(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in3;
+        out1 = in0 ^ in4;
+        out7 = in2 ^ in7;
+        out4 = in4 ^ in5 ^ in7;
+        out5 = in0 ^ in5 ^ in6;
+        out6 = in1 ^ in6 ^ in7;
+        out3 = in2 ^ in3 ^ in4 ^ in6;
+        out2 = in1 ^ in3 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_23(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out7 = in2;
+        out0 = in0 ^ in3;
+        out4 = in5 ^ in7;
+        out5 = in0 ^ in6;
+        out6 = in1 ^ in7;
+        out3 = in2 ^ in4 ^ in6;
+        out1 = in0 ^ in1 ^ in4;
+        out2 = out4 ^ out6 ^ in2 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_24(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in4 ^ in7;
+        tmp0 = in3 ^ in4;
+        out0 = in3 ^ in6 ^ in7;
+        out3 = tmp0 ^ in1;
+        tmp1 = out0 ^ in5;
+        out6 = tmp1 ^ out3;
+        out2 = tmp1 ^ in0;
+        out7 = tmp1 ^ in2 ^ in3;
+        out5 = out2 ^ in4;
+        out4 = tmp0 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_25(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in1 ^ in4;
+        tmp0 = in2 ^ in5;
+        out1 = out3 ^ in7;
+        out7 = tmp0 ^ in6;
+        out6 = out1 ^ in5;
+        out4 = out7 ^ in3 ^ in7;
+        out2 = out4 ^ in0;
+        out0 = tmp0 ^ out2;
+        out5 = out0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_26(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in3 ^ in6;
+        tmp0 = in4 ^ in7;
+        out7 = in2 ^ in5 ^ in7;
+        tmp1 = out0 ^ in0 ^ in5;
+        out1 = tmp0 ^ in0;
+        tmp2 = tmp0 ^ in6;
+        out2 = tmp1 ^ in1;
+        out5 = tmp1 ^ in7;
+        out6 = tmp2 ^ in1;
+        out4 = tmp2 ^ out7;
+        out3 = out0 ^ out6 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_27(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out7 = in2 ^ in5;
+        out0 = in0 ^ in3 ^ in6;
+        out6 = in1 ^ in4 ^ in7;
+        out4 = out7 ^ in6;
+        out2 = out0 ^ out7 ^ in1;
+        out5 = out0 ^ in7;
+        out1 = out6 ^ in0;
+        out3 = out6 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_28(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in3;
+        out1 = in4 ^ in6;
+        out0 = in3 ^ in5 ^ in7;
+        tmp0 = out1 ^ in7;
+        tmp1 = out0 ^ in4;
+        out7 = tmp0 ^ in2;
+        tmp2 = tmp0 ^ in1;
+        out3 = tmp1 ^ in0;
+        out6 = tmp1 ^ tmp2;
+        out4 = tmp2 ^ in3;
+        out5 = out3 ^ in2 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_29(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in2 ^ in3;
+        tmp0 = in1 ^ in3;
+        tmp1 = in4 ^ in6;
+        tmp2 = in0 ^ in4 ^ in7;
+        out6 = tmp0 ^ in5;
+        out4 = tmp0 ^ in6 ^ in7;
+        out1 = tmp1 ^ in1;
+        out7 = tmp1 ^ in2;
+        out3 = tmp2 ^ in5;
+        out5 = tmp2 ^ in2;
+        out0 = out3 ^ in3 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_2A(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in3 ^ in5;
+        tmp0 = in1 ^ in3;
+        tmp1 = in0 ^ in4;
+        out7 = in2 ^ in4 ^ in7;
+        out3 = tmp1 ^ out0 ^ in2;
+        out2 = tmp0 ^ in7;
+        out6 = tmp0 ^ in6;
+        out1 = tmp1 ^ in6;
+        out5 = tmp1 ^ out7 ^ in5;
+        out4 = out1 ^ in0 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_2B(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in1 ^ in6;
+        out7 = in2 ^ in4;
+        tmp0 = in0 ^ in5;
+        tmp1 = in2 ^ in7;
+        out6 = in1 ^ in3;
+        out1 = out4 ^ in0 ^ in4;
+        out3 = tmp0 ^ out7;
+        out0 = tmp0 ^ in3;
+        out5 = tmp1 ^ in0;
+        out2 = tmp1 ^ out6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_2C(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in5;
+        tmp1 = in2 ^ in3 ^ in4;
+        tmp2 = tmp0 ^ in6;
+        out4 = tmp1 ^ in1;
+        out5 = tmp1 ^ in0 ^ in5;
+        tmp3 = tmp2 ^ in4;
+        out6 = tmp2 ^ out4;
+        out7 = tmp3 ^ in7;
+        out2 = tmp3 ^ out5;
+        out3 = out6 ^ in0;
+        out0 = tmp1 ^ out7;
+        out1 = tmp0 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_2D(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in3;
+        out4 = tmp0 ^ in1;
+        tmp1 = tmp0 ^ in0;
+        out2 = tmp1 ^ in6;
+        out5 = tmp1 ^ in4;
+        tmp2 = out2 ^ in2;
+        tmp3 = tmp2 ^ in5;
+        out0 = tmp3 ^ in7;
+        out7 = tmp3 ^ out5;
+        out6 = out4 ^ out7 ^ in6;
+        out3 = tmp2 ^ out6;
+        out1 = out0 ^ out6 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_2E(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in4 ^ in7;
+        out0 = in3 ^ in5 ^ in6;
+        tmp1 = tmp0 ^ in0;
+        tmp2 = tmp0 ^ in2;
+        out1 = tmp1 ^ in6;
+        out4 = tmp2 ^ in1;
+        out7 = tmp2 ^ in5;
+        out3 = out0 ^ out4 ^ in0;
+        out2 = out3 ^ out7 ^ in7;
+        out6 = tmp1 ^ out2;
+        out5 = tmp1 ^ out7 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_2F(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in3;
+        tmp1 = in2 ^ in5;
+        out4 = in1 ^ in2 ^ in7;
+        out6 = in1 ^ in3 ^ in4;
+        out5 = tmp0 ^ in2;
+        tmp2 = tmp0 ^ in6;
+        out7 = tmp1 ^ in4;
+        out0 = tmp2 ^ in5;
+        out2 = tmp2 ^ out4;
+        out1 = tmp2 ^ out6 ^ in7;
+        out3 = tmp1 ^ out1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_30(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in4 ^ in5;
+        tmp0 = in3 ^ in6;
+        tmp1 = in4 ^ in7;
+        out6 = in1 ^ in2 ^ in5;
+        out3 = tmp0 ^ in5;
+        out4 = tmp0 ^ in0;
+        out7 = tmp0 ^ in2;
+        out0 = tmp1 ^ in3;
+        out2 = tmp1 ^ out3;
+        out5 = tmp1 ^ in0 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_31(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in5 ^ in6;
+        tmp0 = in4 ^ in5;
+        tmp1 = in0 ^ in3 ^ in4;
+        tmp2 = out3 ^ in2;
+        out1 = tmp0 ^ in1;
+        out0 = tmp1 ^ in7;
+        out4 = tmp1 ^ in6;
+        out6 = tmp2 ^ in1;
+        out2 = tmp2 ^ out0 ^ in0;
+        out5 = out1 ^ in0 ^ in7;
+        out7 = tmp0 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_32(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in3 ^ in4;
+        out7 = in2 ^ in3;
+        tmp0 = in5 ^ in6;
+        tmp1 = in0 ^ in7;
+        out6 = in1 ^ in2;
+        out1 = in0 ^ in4 ^ in5;
+        out2 = tmp0 ^ out0 ^ in1;
+        out3 = tmp0 ^ out7 ^ in7;
+        out4 = tmp1 ^ in6;
+        out5 = tmp1 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_33(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in3;
+        tmp1 = in0 ^ in4;
+        tmp2 = in1 ^ in5;
+        out6 = in1 ^ in2 ^ in6;
+        out7 = tmp0 ^ in7;
+        out0 = tmp1 ^ in3;
+        out1 = tmp1 ^ tmp2;
+        tmp3 = tmp2 ^ in7;
+        tmp4 = tmp2 ^ in4 ^ in6;
+        out5 = tmp3 ^ in0;
+        out3 = tmp3 ^ out6;
+        out4 = tmp4 ^ out5;
+        out2 = tmp0 ^ tmp4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_34(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in4;
+        tmp1 = in4 ^ in5;
+        tmp2 = tmp0 ^ in1;
+        tmp3 = tmp0 ^ in6;
+        out1 = tmp1 ^ in7;
+        tmp4 = tmp1 ^ in2;
+        out5 = tmp2 ^ in0;
+        out3 = tmp2 ^ out1;
+        out0 = tmp3 ^ in7;
+        out7 = tmp3 ^ tmp4;
+        out6 = tmp4 ^ in1;
+        out2 = out3 ^ out5 ^ in3;
+        out4 = tmp4 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_35(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in6;
+        tmp1 = in5 ^ in7;
+        out7 = tmp0 ^ tmp1 ^ in3;
+        out3 = tmp1 ^ in1;
+        out1 = out3 ^ in4;
+        tmp2 = out1 ^ in7;
+        out5 = tmp2 ^ in0 ^ in3;
+        out6 = tmp0 ^ tmp2;
+        out0 = out3 ^ out5 ^ in6;
+        out4 = tmp0 ^ out0;
+        out2 = out4 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_36(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in0 ^ in2;
+        tmp0 = in1 ^ in3;
+        out0 = in3 ^ in4 ^ in6;
+        out6 = in1 ^ in2 ^ in4;
+        out5 = tmp0 ^ in0;
+        tmp1 = out5 ^ in5;
+        out2 = tmp1 ^ in4;
+        out3 = tmp1 ^ out4;
+        out1 = tmp0 ^ out2 ^ in7;
+        out7 = out3 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_37(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in2;
+        tmp1 = in2 ^ in4;
+        tmp2 = tmp0 ^ in6;
+        out3 = tmp0 ^ in5;
+        out4 = tmp1 ^ in0;
+        out6 = tmp2 ^ in4;
+        out1 = out3 ^ out4 ^ in7;
+        tmp3 = out4 ^ in1 ^ in3;
+        out7 = tmp3 ^ out1;
+        out2 = tmp3 ^ in5;
+        out5 = tmp1 ^ out2;
+        out0 = tmp2 ^ tmp3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_38(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in0 ^ in3;
+        tmp0 = in3 ^ in4;
+        tmp1 = in5 ^ in7;
+        tmp2 = out3 ^ in1;
+        out2 = tmp0 ^ in6;
+        out0 = tmp0 ^ tmp1;
+        out4 = tmp1 ^ tmp2;
+        out7 = out2 ^ in2;
+        out1 = out2 ^ in3 ^ in5;
+        out6 = out4 ^ in0 ^ in2;
+        out5 = tmp2 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_39(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in0;
+        tmp0 = in1 ^ in5;
+        tmp1 = tmp0 ^ in4;
+        out1 = tmp1 ^ in6;
+        out5 = out1 ^ in0 ^ in2;
+        tmp2 = tmp0 ^ out5;
+        out2 = tmp2 ^ in0 ^ in3;
+        out7 = out2 ^ in7;
+        out6 = tmp1 ^ out7;
+        out4 = tmp2 ^ out6;
+        out0 = out4 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_3A(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in1;
+        tmp1 = in0 ^ in2;
+        tmp2 = in3 ^ in4;
+        tmp3 = in1 ^ in6;
+        tmp4 = in3 ^ in7;
+        out4 = tmp0 ^ in5;
+        out5 = tmp1 ^ tmp3;
+        out3 = tmp1 ^ tmp4;
+        out0 = tmp2 ^ in5;
+        out7 = tmp2 ^ in2;
+        tmp5 = tmp3 ^ in4;
+        out2 = tmp4 ^ tmp5;
+        out1 = tmp5 ^ out4;
+        out6 = tmp0 ^ out3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_3B(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in6;
+        tmp1 = in2 ^ in7;
+        tmp2 = tmp0 ^ in3;
+        out3 = tmp1 ^ in0;
+        out6 = tmp1 ^ tmp2;
+        out2 = out6 ^ in4;
+        out7 = tmp0 ^ out2;
+        out0 = out3 ^ out7 ^ in5;
+        out5 = out0 ^ out2 ^ in7;
+        out1 = tmp2 ^ out0;
+        out4 = out1 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_3C(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in3;
+        tmp1 = in2 ^ in7;
+        tmp2 = in1 ^ in6 ^ in7;
+        out2 = tmp0 ^ in4;
+        out3 = tmp0 ^ tmp2;
+        out4 = tmp1 ^ out3 ^ in5;
+        out5 = tmp2 ^ out2 ^ in2;
+        out1 = out4 ^ out5 ^ in6;
+        out0 = out1 ^ in3;
+        out7 = tmp1 ^ out0;
+        out6 = tmp2 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_3D(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in2;
+        tmp1 = tmp0 ^ in3;
+        out2 = tmp1 ^ in4;
+        tmp2 = out2 ^ in5;
+        out4 = tmp2 ^ in1 ^ in6;
+        out5 = out4 ^ in7;
+        out6 = out5 ^ in0;
+        out7 = out6 ^ in1;
+        out0 = tmp0 ^ out7;
+        out1 = tmp1 ^ out5;
+        out3 = tmp2 ^ out6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_3E(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in5;
+        tmp1 = tmp0 ^ in4;
+        out0 = tmp1 ^ in6;
+        out7 = tmp1 ^ in2;
+        out6 = out7 ^ in1 ^ in5 ^ in7;
+        out2 = out6 ^ in0 ^ in2;
+        out4 = out0 ^ out6 ^ in0;
+        out5 = tmp0 ^ out4;
+        out3 = out5 ^ in7;
+        out1 = out3 ^ out6 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_3F(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in1;
+        out3 = tmp0 ^ in2 ^ in6;
+        tmp1 = out3 ^ in5 ^ in7;
+        out4 = tmp1 ^ in4;
+        out5 = tmp1 ^ in3;
+        out1 = out4 ^ in2;
+        out7 = out1 ^ out3 ^ in3;
+        out2 = tmp0 ^ out7 ^ in5;
+        tmp2 = out2 ^ in0;
+        out6 = tmp2 ^ in6;
+        out0 = tmp1 ^ tmp2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_40(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in3 ^ in7;
+        tmp0 = in3 ^ in4;
+        tmp1 = in6 ^ in7;
+        out4 = tmp0 ^ in2;
+        out5 = tmp0 ^ in5;
+        out0 = tmp1 ^ in2;
+        out7 = tmp1 ^ in1 ^ in5;
+        out2 = out0 ^ in4;
+        out3 = out2 ^ out5 ^ in7;
+        out6 = out3 ^ out4 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_41(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in2 ^ in3;
+        tmp0 = in5 ^ in6;
+        tmp1 = in6 ^ in7;
+        out5 = in3 ^ in4;
+        out1 = in1 ^ in3 ^ in7;
+        out6 = in0 ^ in4 ^ in5;
+        out3 = tmp0 ^ in2;
+        out7 = tmp0 ^ in1;
+        out2 = tmp1 ^ in4;
+        out0 = tmp1 ^ in0 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_42(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in2 ^ in6;
+        out5 = in3 ^ in5;
+        out1 = in0 ^ in3 ^ in7;
+        out7 = in1 ^ in5 ^ in7;
+        out4 = in2 ^ in4 ^ in7;
+        out6 = in0 ^ in4 ^ in6;
+        out2 = out0 ^ in1 ^ in4;
+        out3 = out5 ^ in6 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_43(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out5 = in3;
+        out7 = in1 ^ in5;
+        out4 = in2 ^ in7;
+        out6 = in0 ^ in4;
+        out0 = in0 ^ in2 ^ in6;
+        out3 = in5 ^ in6 ^ in7;
+        out2 = in1 ^ in4 ^ in6;
+        out1 = in0 ^ in1 ^ in3 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_44(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in3;
+        out0 = in2 ^ in7;
+        tmp0 = in4 ^ in7;
+        out7 = in1 ^ in6 ^ in7;
+        out6 = in0 ^ in5 ^ in6;
+        out4 = tmp0 ^ in3 ^ in6;
+        out3 = out0 ^ in1 ^ in3 ^ in5;
+        out2 = out0 ^ in0 ^ in4;
+        out5 = tmp0 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_45(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in1 ^ in3;
+        out7 = in1 ^ in6;
+        out5 = in4 ^ in7;
+        out6 = in0 ^ in5;
+        out0 = in0 ^ in2 ^ in7;
+        out4 = in3 ^ in6 ^ in7;
+        out2 = out5 ^ in0;
+        out3 = out0 ^ out6 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_46(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in2;
+        out1 = in0 ^ in3;
+        out7 = in1 ^ in7;
+        out4 = in4 ^ in6;
+        out5 = in5 ^ in7;
+        out6 = in0 ^ in6;
+        out3 = in1 ^ in3 ^ in5;
+        out2 = out4 ^ out6 ^ in1 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_47(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in6;
+        out7 = in1;
+        out5 = in7;
+        out6 = in0;
+        tmp0 = in0 ^ in1;
+        out3 = in1 ^ in5;
+        out0 = in0 ^ in2;
+        out1 = tmp0 ^ in3;
+        out2 = tmp0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_48(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in3;
+        out1 = in3 ^ in6 ^ in7;
+        out3 = tmp0 ^ in0;
+        out0 = tmp0 ^ out1 ^ in5;
+        tmp1 = out0 ^ in4;
+        out2 = tmp1 ^ in7;
+        out5 = tmp1 ^ in3;
+        out4 = out5 ^ in1;
+        out7 = tmp0 ^ out4;
+        out6 = tmp1 ^ out3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_49(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in0 ^ in2;
+        tmp0 = in2 ^ in5;
+        out2 = in4 ^ in5 ^ in6;
+        tmp1 = tmp0 ^ out2 ^ in3;
+        out7 = out2 ^ in1;
+        out5 = tmp1 ^ in7;
+        out4 = out5 ^ out7 ^ in6;
+        out1 = tmp0 ^ out4;
+        out6 = out1 ^ out7 ^ in0;
+        out0 = tmp1 ^ out6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_4A(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in6;
+        tmp1 = in3 ^ in7;
+        out0 = tmp0 ^ in5;
+        out3 = tmp1 ^ in0;
+        out5 = tmp1 ^ out0;
+        out4 = out0 ^ in1 ^ in4;
+        out1 = out3 ^ in6;
+        out2 = out4 ^ in7;
+        out6 = out1 ^ in4;
+        out7 = tmp0 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_4B(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in0 ^ in7;
+        tmp0 = in1 ^ in5;
+        tmp1 = in2 ^ in6;
+        tmp2 = out3 ^ in3;
+        out7 = tmp0 ^ in4;
+        out4 = tmp0 ^ tmp1;
+        tmp3 = tmp1 ^ in0;
+        out6 = tmp2 ^ in4;
+        out5 = tmp2 ^ tmp3;
+        out1 = tmp2 ^ in1 ^ in6;
+        out2 = out7 ^ in6 ^ in7;
+        out0 = tmp3 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_4C(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in3 ^ in6;
+        tmp0 = in2 ^ in5;
+        tmp1 = out1 ^ in5 ^ in7;
+        out0 = tmp0 ^ in7;
+        tmp2 = tmp0 ^ in4;
+        out6 = tmp1 ^ in0;
+        out2 = tmp2 ^ in0;
+        out5 = tmp2 ^ in6;
+        out3 = tmp0 ^ out6 ^ in1;
+        out7 = out0 ^ out5 ^ in1;
+        out4 = tmp1 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_4D(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in5;
+        tmp1 = in1 ^ in6;
+        out4 = in1 ^ in3 ^ in5;
+        tmp2 = tmp0 ^ in7;
+        out2 = tmp0 ^ in4;
+        out1 = tmp1 ^ in3;
+        out7 = tmp1 ^ in4;
+        out0 = tmp2 ^ in2;
+        out6 = tmp2 ^ in3;
+        out5 = out7 ^ in1 ^ in2;
+        out3 = tmp1 ^ out0 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_4E(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in2 ^ in5;
+        out7 = in1 ^ in4 ^ in7;
+        out1 = in0 ^ in3 ^ in6;
+        out5 = out0 ^ in6;
+        out4 = out7 ^ in5;
+        out3 = out1 ^ in1;
+        out6 = out1 ^ in7;
+        out2 = out4 ^ in0 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_4F(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out5 = in2 ^ in6;
+        out7 = in1 ^ in4;
+        out3 = in0 ^ in1 ^ in6;
+        out4 = in1 ^ in5 ^ in7;
+        out0 = in0 ^ in2 ^ in5;
+        out6 = in0 ^ in3 ^ in7;
+        out1 = out3 ^ in3;
+        out2 = out4 ^ in0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_50(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in2 ^ in7;
+        tmp0 = in3 ^ in5;
+        out0 = out2 ^ in4 ^ in6;
+        out1 = tmp0 ^ in7;
+        tmp1 = tmp0 ^ in6;
+        out3 = out0 ^ in3;
+        out7 = tmp1 ^ in1;
+        tmp2 = tmp1 ^ in0;
+        out5 = out3 ^ in1 ^ in2;
+        out4 = tmp2 ^ in2;
+        out6 = tmp2 ^ out3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_51(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in7;
+        out3 = in2 ^ in4 ^ in6 ^ in7;
+        out0 = out3 ^ in0;
+        out6 = out0 ^ in5;
+        out4 = out6 ^ in3 ^ in7;
+        out1 = out0 ^ out4 ^ in1;
+        out7 = out1 ^ in6;
+        out5 = out7 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_52(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in1 ^ in2;
+        tmp0 = in2 ^ in4;
+        tmp1 = in3 ^ in5;
+        tmp2 = in3 ^ in6;
+        tmp3 = in0 ^ in7;
+        out0 = tmp0 ^ in6;
+        out6 = tmp0 ^ tmp3;
+        out7 = tmp1 ^ in1;
+        out1 = tmp1 ^ tmp3;
+        out3 = tmp2 ^ in4;
+        out5 = tmp2 ^ in1 ^ in7;
+        out4 = tmp2 ^ out1 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_53(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in1;
+        out3 = in4 ^ in6;
+        out0 = out3 ^ in0 ^ in2;
+        out6 = out0 ^ in7;
+        out4 = out6 ^ in5;
+        out7 = out0 ^ out4 ^ in1 ^ in3;
+        out1 = out7 ^ in0;
+        out5 = out7 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_54(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in3 ^ in5;
+        tmp0 = in1 ^ in3;
+        tmp1 = in2 ^ in4;
+        tmp2 = in0 ^ in7;
+        out5 = in1 ^ in4 ^ in6;
+        out4 = tmp2 ^ out1;
+        out7 = tmp0 ^ in6;
+        out3 = tmp0 ^ tmp1;
+        out0 = tmp1 ^ in7;
+        tmp3 = tmp2 ^ in2;
+        out2 = tmp3 ^ in6;
+        out6 = tmp3 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_55(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in3;
+        tmp1 = in1 ^ in4;
+        tmp2 = in6 ^ in7;
+        out7 = tmp0 ^ tmp2;
+        out1 = tmp0 ^ in5;
+        out3 = tmp1 ^ in2;
+        out5 = tmp1 ^ in5 ^ in6;
+        out2 = tmp2 ^ in0;
+        out4 = out5 ^ out7 ^ in0;
+        out6 = out2 ^ in2 ^ in5;
+        out0 = out5 ^ out6 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_56(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in2 ^ in4;
+        tmp0 = in0 ^ in2;
+        out4 = in0 ^ in5;
+        out7 = in1 ^ in3;
+        out5 = in1 ^ in6;
+        out6 = tmp0 ^ in7;
+        out2 = tmp0 ^ out5;
+        out1 = out4 ^ in3;
+        out3 = out7 ^ in4 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_57(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in5;
+        tmp1 = in1 ^ in7;
+        out0 = in0 ^ in2 ^ in4;
+        out5 = in1 ^ in5 ^ in6;
+        out4 = tmp0 ^ in4;
+        out1 = tmp0 ^ in1 ^ in3;
+        out2 = tmp0 ^ out5;
+        out3 = tmp1 ^ in4;
+        out7 = tmp1 ^ in3;
+        out6 = tmp1 ^ out2 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_58(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in2 ^ in5;
+        tmp0 = in2 ^ in3 ^ in4;
+        out5 = tmp0 ^ in1;
+        out6 = tmp0 ^ in0 ^ in5;
+        out3 = out6 ^ in7;
+        tmp1 = out2 ^ out5;
+        out7 = tmp1 ^ in6;
+        out4 = tmp1 ^ out3 ^ in3;
+        out0 = out4 ^ out7 ^ in0;
+        out1 = tmp0 ^ out0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_59(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in5;
+        tmp0 = in0 ^ in5 ^ in7;
+        out3 = tmp0 ^ in2 ^ in4;
+        out0 = out3 ^ in6;
+        tmp1 = out0 ^ in7;
+        out6 = tmp1 ^ in3;
+        out5 = out6 ^ in0 ^ in1 ^ in6;
+        out4 = tmp0 ^ out5;
+        out1 = tmp1 ^ out4;
+        out7 = out1 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_5A(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in2;
+        tmp1 = in2 ^ in5;
+        out5 = tmp0 ^ in3;
+        out4 = tmp0 ^ in0;
+        tmp2 = tmp1 ^ in4;
+        out2 = tmp1 ^ in1 ^ in7;
+        out7 = tmp2 ^ out5;
+        out6 = out4 ^ out7 ^ in5;
+        out0 = tmp2 ^ in6;
+        out1 = out0 ^ out6 ^ in7;
+        out3 = tmp1 ^ out6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_5B(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in3;
+        tmp1 = in0 ^ in4;
+        tmp2 = in1 ^ in5;
+        out5 = tmp0 ^ tmp2;
+        tmp3 = tmp1 ^ in6;
+        out3 = tmp1 ^ in5;
+        out2 = tmp2 ^ in7;
+        tmp4 = out3 ^ in2;
+        out7 = out2 ^ in3 ^ in4;
+        out0 = tmp4 ^ in6;
+        out6 = tmp0 ^ tmp3;
+        out4 = tmp2 ^ tmp4;
+        out1 = tmp3 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_5C(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in6;
+        tmp1 = in0 ^ in2 ^ in5;
+        out1 = tmp0 ^ in5;
+        tmp2 = tmp0 ^ in1;
+        out2 = tmp1 ^ in6;
+        out6 = tmp1 ^ in3;
+        out4 = tmp2 ^ in0;
+        out7 = tmp2 ^ in4;
+        out3 = tmp1 ^ out7;
+        out0 = out3 ^ out4 ^ in7;
+        out5 = out0 ^ in1 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_5D(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in1;
+        tmp1 = in0 ^ in6;
+        out2 = tmp1 ^ in5;
+        tmp2 = out2 ^ in3;
+        out6 = tmp2 ^ in2;
+        out1 = tmp0 ^ tmp2;
+        tmp3 = out1 ^ in4 ^ in5;
+        out4 = tmp3 ^ in0;
+        out7 = tmp3 ^ in7;
+        tmp4 = out4 ^ out6;
+        out5 = tmp4 ^ in7;
+        out0 = tmp0 ^ out5;
+        out3 = tmp1 ^ tmp4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_5E(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in5;
+        tmp1 = in3 ^ in5;
+        tmp2 = in1 ^ in7;
+        out7 = in1 ^ in3 ^ in4;
+        out0 = tmp0 ^ in4;
+        tmp3 = tmp1 ^ in0;
+        out5 = tmp2 ^ in2;
+        out1 = tmp3 ^ in6;
+        out6 = tmp0 ^ tmp3;
+        tmp4 = tmp2 ^ out1;
+        out3 = tmp4 ^ in4;
+        out4 = tmp1 ^ tmp4;
+        out2 = tmp0 ^ out4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_5F(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in5;
+        tmp1 = in0 ^ in6;
+        tmp2 = tmp0 ^ in7;
+        tmp3 = tmp1 ^ in3;
+        out2 = tmp1 ^ tmp2;
+        out5 = tmp2 ^ in2;
+        out6 = tmp3 ^ in2;
+        out3 = out2 ^ in4;
+        out4 = out3 ^ in5;
+        out1 = tmp0 ^ tmp3;
+        out7 = tmp3 ^ out4;
+        out0 = out4 ^ out5 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_60(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in2 ^ in5;
+        tmp0 = in3 ^ in6;
+        out1 = in3 ^ in4 ^ in7;
+        out7 = out4 ^ in1;
+        tmp1 = out4 ^ in4;
+        out0 = tmp0 ^ in2;
+        out5 = tmp0 ^ in0;
+        out2 = tmp0 ^ tmp1;
+        out3 = tmp1 ^ in7;
+        out6 = out3 ^ out7 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_61(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in5;
+        out4 = tmp0 ^ in4;
+        tmp1 = out4 ^ in3;
+        out3 = tmp1 ^ in7;
+        out2 = tmp1 ^ in2 ^ in6;
+        out1 = tmp0 ^ out3 ^ in1;
+        out0 = out2 ^ out4 ^ in0;
+        out7 = tmp1 ^ out1;
+        out6 = out0 ^ out1 ^ in2;
+        out5 = tmp0 ^ out0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_62(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in4 ^ in5;
+        tmp0 = in0 ^ in3 ^ in4;
+        out1 = tmp0 ^ in7;
+        out5 = tmp0 ^ in6;
+        tmp1 = out1 ^ in0;
+        tmp2 = tmp1 ^ out3;
+        out4 = tmp2 ^ in2;
+        tmp3 = tmp2 ^ in1;
+        out0 = out4 ^ in5 ^ in6;
+        out7 = tmp3 ^ out0;
+        out6 = tmp0 ^ tmp3;
+        out2 = tmp1 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_63(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in4;
+        tmp1 = in1 ^ in7;
+        out3 = tmp0 ^ in5;
+        tmp2 = out3 ^ in6;
+        out4 = out3 ^ in2 ^ in7;
+        out5 = tmp2 ^ in0;
+        tmp3 = out5 ^ in3;
+        out0 = tmp3 ^ out4;
+        out2 = tmp1 ^ tmp2;
+        out6 = tmp1 ^ tmp3;
+        tmp4 = tmp0 ^ out2;
+        out1 = tmp4 ^ out5;
+        out7 = tmp4 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_64(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in2 ^ in3;
+        out1 = in3 ^ in4;
+        out7 = in1 ^ in2;
+        tmp0 = in4 ^ in5;
+        tmp1 = in0 ^ in7;
+        out4 = in5 ^ in6 ^ in7;
+        out2 = tmp0 ^ out0 ^ in0;
+        out3 = tmp0 ^ out7 ^ in6;
+        out5 = tmp1 ^ in6;
+        out6 = tmp1 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_65(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in3;
+        tmp1 = in4 ^ in5;
+        tmp2 = in6 ^ in7;
+        out7 = in1 ^ in2 ^ in7;
+        out1 = in1 ^ in3 ^ in4;
+        out0 = tmp0 ^ in2;
+        out2 = tmp0 ^ tmp1;
+        out4 = tmp1 ^ tmp2;
+        tmp3 = tmp2 ^ in0;
+        out3 = out4 ^ out7 ^ in3;
+        out5 = tmp3 ^ in5;
+        out6 = tmp3 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_66(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in2;
+        tmp1 = in2 ^ in3;
+        tmp2 = in0 ^ in4;
+        out7 = tmp0 ^ in6;
+        out0 = tmp1 ^ in7;
+        out1 = tmp2 ^ in3;
+        tmp3 = tmp2 ^ in6;
+        tmp4 = out1 ^ in5;
+        out5 = tmp3 ^ in7;
+        out4 = tmp3 ^ tmp4;
+        out2 = tmp0 ^ tmp4 ^ in7;
+        out6 = tmp1 ^ out2 ^ in4;
+        out3 = tmp3 ^ out6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_67(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in3;
+        tmp1 = tmp0 ^ in1;
+        tmp2 = tmp0 ^ in7;
+        out1 = tmp1 ^ in4;
+        out0 = tmp2 ^ in2;
+        tmp3 = out1 ^ in7;
+        out2 = tmp3 ^ in5;
+        out3 = out2 ^ in0 ^ in6;
+        out7 = tmp1 ^ out0 ^ in6;
+        out5 = tmp1 ^ out3;
+        out4 = tmp2 ^ out5;
+        out6 = tmp3 ^ out4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_68(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in4;
+        tmp1 = in2 ^ in3 ^ in5;
+        tmp2 = tmp0 ^ in1;
+        tmp3 = tmp0 ^ in6;
+        out0 = tmp1 ^ in6;
+        out6 = tmp2 ^ in0;
+        out7 = tmp1 ^ tmp2;
+        out1 = tmp3 ^ in7;
+        out2 = out1 ^ in2;
+        out4 = tmp2 ^ out2;
+        out3 = out4 ^ out6 ^ in3;
+        out5 = tmp3 ^ out3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_69(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in6 ^ in7;
+        out2 = tmp0 ^ in3 ^ in4;
+        out1 = out2 ^ in1;
+        out3 = out2 ^ in0 ^ in2;
+        out4 = out1 ^ in2 ^ in3;
+        out6 = out1 ^ in0 ^ in7;
+        out7 = out4 ^ in5 ^ in6;
+        out5 = out4 ^ out6 ^ in5;
+        out0 = tmp0 ^ out5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_6A(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in6;
+        out3 = in0 ^ in4 ^ in6;
+        tmp1 = tmp0 ^ in3;
+        out4 = tmp1 ^ in1;
+        tmp2 = tmp1 ^ in7;
+        out2 = out4 ^ in4;
+        out0 = tmp2 ^ in5;
+        out5 = tmp2 ^ out3;
+        out7 = out2 ^ in3 ^ in5;
+        out1 = tmp0 ^ out5;
+        out6 = tmp1 ^ out7 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_6B(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in4 ^ in6;
+        out2 = tmp0 ^ in1 ^ in3;
+        out4 = out2 ^ in2;
+        tmp1 = out2 ^ in0;
+        out7 = out4 ^ in3 ^ in5 ^ in7;
+        out1 = tmp1 ^ in7;
+        out3 = tmp1 ^ in1;
+        out6 = tmp1 ^ in5;
+        out0 = tmp1 ^ out7 ^ in6;
+        out5 = tmp0 ^ out0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_6C(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in1;
+        tmp0 = in2 ^ in3;
+        out5 = in0 ^ in2;
+        out1 = in3 ^ in4 ^ in6;
+        tmp1 = out5 ^ in1;
+        out0 = tmp0 ^ in5;
+        out6 = tmp0 ^ tmp1;
+        out3 = tmp1 ^ in4;
+        out7 = out3 ^ in0;
+        out2 = out6 ^ out7 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_6D(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in1 ^ in4;
+        tmp0 = in0 ^ in2;
+        tmp1 = out4 ^ in3;
+        out7 = out4 ^ in2 ^ in7;
+        out5 = tmp0 ^ in5;
+        out3 = tmp0 ^ tmp1;
+        out1 = tmp1 ^ in6;
+        out0 = out5 ^ in3;
+        out2 = out3 ^ out7 ^ in4;
+        out6 = out1 ^ in0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_6E(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in3;
+        tmp1 = in0 ^ in4;
+        out4 = tmp0 ^ in7;
+        out6 = tmp0 ^ in0 ^ in5;
+        out5 = tmp1 ^ in2;
+        tmp2 = tmp1 ^ in3;
+        out3 = tmp2 ^ out4;
+        out1 = tmp2 ^ in6;
+        out2 = tmp0 ^ out5;
+        out0 = out2 ^ out3 ^ in5;
+        out7 = out1 ^ out2 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_6F(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in7;
+        tmp1 = tmp0 ^ in4;
+        tmp2 = tmp0 ^ in0 ^ in2;
+        out4 = tmp1 ^ in1;
+        out0 = tmp2 ^ in5;
+        out3 = out4 ^ in0;
+        out2 = out3 ^ in7;
+        out1 = out2 ^ in6;
+        out6 = out1 ^ in4 ^ in5;
+        out7 = tmp2 ^ out1;
+        out5 = tmp1 ^ out0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_70(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in2;
+        tmp0 = in2 ^ in4;
+        out2 = in2 ^ in3 ^ in5;
+        tmp1 = tmp0 ^ in6;
+        tmp2 = out2 ^ in7;
+        out0 = tmp1 ^ in3;
+        out4 = tmp1 ^ in0;
+        out7 = tmp2 ^ in1;
+        out6 = out4 ^ in1;
+        out5 = out7 ^ in0 ^ in2;
+        out1 = tmp0 ^ tmp2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_71(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in3 ^ in5;
+        out3 = in2 ^ in3;
+        tmp0 = in0 ^ in2;
+        tmp1 = out2 ^ in1;
+        out4 = tmp0 ^ in6;
+        tmp2 = tmp0 ^ in1;
+        out7 = tmp1 ^ in2;
+        out1 = tmp1 ^ in4 ^ in7;
+        out0 = out4 ^ in3 ^ in4;
+        out6 = tmp2 ^ in4;
+        out5 = tmp2 ^ out3 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_72(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in7;
+        tmp0 = in0 ^ in4;
+        tmp1 = tmp0 ^ in3 ^ in7;
+        out1 = tmp1 ^ in5;
+        out5 = out1 ^ in1;
+        tmp2 = tmp0 ^ out5;
+        out2 = tmp2 ^ in2;
+        out7 = out2 ^ in6;
+        out6 = tmp1 ^ out7;
+        out4 = tmp2 ^ out6;
+        out0 = out4 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_73(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in3 ^ in7;
+        out2 = out3 ^ in1 ^ in5;
+        out1 = out2 ^ in0 ^ in4;
+        out5 = out1 ^ in5;
+        out6 = out1 ^ out3 ^ in2;
+        out0 = out2 ^ out6 ^ in6;
+        out7 = out0 ^ out1 ^ in3;
+        out4 = out0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_74(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in4;
+        tmp1 = in1 ^ in2 ^ in6;
+        out4 = in0 ^ in4 ^ in7;
+        out5 = in0 ^ in1 ^ in5;
+        out0 = tmp0 ^ in2;
+        out1 = tmp0 ^ in5;
+        out3 = tmp1 ^ in7;
+        out6 = tmp1 ^ in0;
+        out2 = tmp1 ^ out5 ^ in3;
+        out7 = out3 ^ in3 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_75(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in0 ^ in7;
+        tmp0 = in1 ^ in3;
+        out5 = in0 ^ in1;
+        out7 = tmp0 ^ in2;
+        tmp1 = tmp0 ^ in4;
+        out6 = out5 ^ in2;
+        tmp2 = out7 ^ in6;
+        out1 = tmp1 ^ in5;
+        out0 = tmp1 ^ out6;
+        out3 = tmp2 ^ in7;
+        out2 = tmp2 ^ out6 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_76(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in1 ^ in6;
+        tmp0 = in0 ^ in5;
+        tmp1 = in3 ^ in7;
+        tmp2 = tmp0 ^ in4;
+        tmp3 = tmp1 ^ in2;
+        out5 = tmp2 ^ in1;
+        out1 = tmp2 ^ in3;
+        out0 = tmp3 ^ in4;
+        out4 = out1 ^ in5;
+        out7 = tmp3 ^ out3;
+        out2 = tmp0 ^ out7;
+        out6 = tmp1 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_77(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in0 ^ in3;
+        tmp0 = in1 ^ in4;
+        tmp1 = in1 ^ in6;
+        tmp2 = out4 ^ in5;
+        out5 = tmp0 ^ in0;
+        out1 = tmp0 ^ tmp2;
+        out3 = tmp1 ^ in3;
+        out2 = tmp1 ^ tmp2 ^ in7;
+        out7 = out3 ^ in2;
+        tmp3 = out7 ^ in6;
+        out6 = tmp2 ^ tmp3;
+        out0 = tmp3 ^ out5 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_78(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in3;
+        tmp1 = in2 ^ in7;
+        tmp2 = in0 ^ in5 ^ in6;
+        out2 = tmp1 ^ in3;
+        out3 = tmp2 ^ in2;
+        out5 = out3 ^ in1 ^ in3;
+        out0 = tmp0 ^ out3 ^ in4;
+        out1 = tmp1 ^ out0;
+        out4 = out1 ^ out5 ^ in5;
+        out7 = tmp0 ^ out4;
+        out6 = tmp2 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_79(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in3 ^ in7;
+        tmp0 = in3 ^ in4;
+        tmp1 = in1 ^ in5;
+        tmp2 = tmp1 ^ in2;
+        out4 = tmp2 ^ in0 ^ in7;
+        tmp3 = out4 ^ in5;
+        out5 = tmp3 ^ out2 ^ in6;
+        out7 = tmp0 ^ tmp2;
+        out6 = tmp0 ^ tmp3;
+        out3 = tmp1 ^ out5;
+        out0 = out3 ^ in4;
+        out1 = tmp3 ^ out0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_7A(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in2;
+        out2 = tmp0 ^ in3;
+        tmp1 = out2 ^ in4;
+        out4 = tmp1 ^ in0 ^ in5;
+        out5 = out4 ^ in6;
+        out6 = out5 ^ in7;
+        out7 = out6 ^ in0;
+        out0 = out7 ^ in1;
+        out1 = tmp0 ^ out6;
+        out3 = tmp1 ^ out6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_7B(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in1 ^ in3;
+        tmp0 = in0 ^ in5;
+        out4 = tmp0 ^ out2 ^ in2;
+        tmp1 = out4 ^ in4;
+        out6 = tmp1 ^ in7;
+        out5 = tmp1 ^ in5 ^ in6;
+        out0 = out6 ^ in1 ^ in6;
+        tmp2 = out0 ^ in2;
+        out1 = tmp2 ^ in1;
+        out3 = tmp2 ^ in4;
+        out7 = tmp0 ^ out5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_7C(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in5;
+        tmp1 = tmp0 ^ in4;
+        out0 = tmp1 ^ in2;
+        out1 = tmp1 ^ in6;
+        out7 = out0 ^ in1 ^ in5 ^ in7;
+        out5 = out1 ^ out7 ^ in0;
+        out3 = out5 ^ in6;
+        out6 = tmp0 ^ out5;
+        out2 = out6 ^ in1;
+        out4 = out2 ^ out7 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_7D(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in2;
+        tmp1 = tmp0 ^ in3;
+        tmp2 = tmp0 ^ in6;
+        out7 = tmp1 ^ in4;
+        tmp3 = tmp2 ^ in0;
+        out5 = tmp3 ^ in7;
+        out4 = tmp3 ^ in2 ^ in5;
+        out2 = tmp1 ^ out5;
+        out6 = tmp2 ^ out2;
+        out0 = out4 ^ out7 ^ in6;
+        out1 = tmp3 ^ out0;
+        out3 = out6 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_7E(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in4;
+        tmp1 = in0 ^ in5;
+        out1 = tmp0 ^ tmp1 ^ in6;
+        out3 = tmp1 ^ in1;
+        out4 = out1 ^ in1 ^ in7;
+        tmp2 = out4 ^ in3;
+        out5 = tmp2 ^ in2;
+        out6 = tmp0 ^ out5;
+        out7 = tmp1 ^ out4 ^ in2;
+        out2 = out6 ^ in5 ^ in7;
+        out0 = tmp2 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_7F(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in7;
+        tmp1 = tmp0 ^ in3 ^ in5;
+        tmp2 = tmp1 ^ in0;
+        out0 = tmp2 ^ in4;
+        out6 = tmp2 ^ in1;
+        out3 = tmp0 ^ out6;
+        tmp3 = out3 ^ in6;
+        out1 = tmp3 ^ in4;
+        out2 = tmp3 ^ in5;
+        out4 = tmp3 ^ in7;
+        out5 = tmp1 ^ out1;
+        out7 = out0 ^ out4 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_80(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in3;
+        tmp1 = in4 ^ in5;
+        out1 = in2 ^ in6 ^ in7;
+        out5 = tmp0 ^ in4;
+        tmp2 = tmp0 ^ in1;
+        out6 = tmp1 ^ in3;
+        out7 = tmp1 ^ in0 ^ in6;
+        out4 = tmp2 ^ in7;
+        out3 = tmp2 ^ out6;
+        out2 = out3 ^ out5 ^ in6;
+        out0 = out2 ^ in3 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_81(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in4 ^ in6;
+        tmp1 = tmp0 ^ in3;
+        out6 = tmp1 ^ in5;
+        out5 = out6 ^ in2 ^ in6;
+        out3 = out5 ^ in1;
+        out2 = tmp0 ^ out3;
+        out1 = out3 ^ out6 ^ in7;
+        out4 = tmp1 ^ out1;
+        out7 = out2 ^ out4 ^ in0;
+        out0 = out7 ^ in1 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_82(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in1 ^ in2;
+        tmp0 = in6 ^ in7;
+        out5 = in2 ^ in3;
+        out6 = in3 ^ in4;
+        out7 = in0 ^ in4 ^ in5;
+        out0 = in1 ^ in5 ^ in6;
+        out1 = tmp0 ^ in0 ^ in2;
+        out2 = tmp0 ^ in3 ^ in5;
+        out3 = tmp0 ^ out0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_83(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in1;
+        tmp1 = in2 ^ in5;
+        tmp2 = in3 ^ in6;
+        out4 = in1 ^ in2 ^ in4;
+        out0 = tmp0 ^ in5 ^ in6;
+        out5 = tmp1 ^ in3;
+        tmp3 = tmp1 ^ in7;
+        out6 = tmp2 ^ in4;
+        out2 = tmp2 ^ tmp3;
+        tmp4 = tmp3 ^ out4;
+        out1 = tmp3 ^ out0;
+        out3 = tmp4 ^ in3;
+        out7 = tmp0 ^ tmp4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_84(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in2 ^ in6;
+        out6 = in3 ^ in5;
+        out0 = in1 ^ in5 ^ in7;
+        out7 = in0 ^ in4 ^ in6;
+        out4 = in1 ^ in3 ^ in6;
+        out5 = in2 ^ in4 ^ in7;
+        out2 = out6 ^ in0 ^ in1;
+        out3 = out5 ^ in5 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_85(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in6;
+        tmp1 = in3 ^ in6;
+        tmp2 = tmp0 ^ in4;
+        out1 = tmp0 ^ in2;
+        out6 = tmp1 ^ in5;
+        out4 = tmp2 ^ in3;
+        tmp3 = out1 ^ out6;
+        out2 = tmp3 ^ in0;
+        out3 = tmp2 ^ tmp3 ^ in7;
+        out7 = out2 ^ out3 ^ in1;
+        out5 = tmp1 ^ out3;
+        out0 = tmp2 ^ out7 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_86(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out6 = in3;
+        out7 = in0 ^ in4;
+        out0 = in1 ^ in5;
+        out5 = in2 ^ in7;
+        out3 = in4 ^ in5 ^ in6;
+        out1 = in0 ^ in2 ^ in6;
+        out4 = in1 ^ in6 ^ in7;
+        out2 = in0 ^ in3 ^ in5 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_87(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out6 = in3 ^ in6;
+        tmp0 = in0 ^ in1;
+        out7 = in0 ^ in4 ^ in7;
+        out5 = in2 ^ in5 ^ in7;
+        out3 = out6 ^ in4 ^ in5;
+        out0 = tmp0 ^ in5;
+        tmp1 = tmp0 ^ in6;
+        out2 = out5 ^ in0 ^ in3;
+        out1 = tmp1 ^ in2;
+        out4 = tmp1 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_88(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in2 ^ in7;
+        tmp0 = in5 ^ in6;
+        out0 = in1 ^ in6 ^ in7;
+        out6 = in4 ^ in5 ^ in7;
+        out3 = out0 ^ out1 ^ in0 ^ in4;
+        out7 = tmp0 ^ in0;
+        tmp1 = tmp0 ^ in3;
+        out2 = out0 ^ in3;
+        out4 = tmp1 ^ in2;
+        out5 = tmp1 ^ out6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_89(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in7;
+        tmp1 = in2 ^ in7;
+        tmp2 = tmp0 ^ in6;
+        out1 = tmp1 ^ in1;
+        out7 = tmp2 ^ in5;
+        out0 = tmp2 ^ in1;
+        out2 = out1 ^ in3 ^ in6;
+        out6 = out7 ^ in0 ^ in4;
+        out5 = out6 ^ in3;
+        out3 = tmp0 ^ out2 ^ in4;
+        out4 = tmp1 ^ out5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_8A(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in1 ^ in6;
+        out7 = in0 ^ in5;
+        out2 = in3 ^ in6;
+        out6 = in4 ^ in7;
+        out1 = in0 ^ in2 ^ in7;
+        out3 = out0 ^ out6 ^ in0;
+        out4 = out1 ^ out7 ^ in6;
+        out5 = out2 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_8B(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in1;
+        tmp1 = in3 ^ in6;
+        tmp2 = in5 ^ in7;
+        tmp3 = tmp0 ^ in7;
+        out0 = tmp0 ^ in6;
+        out2 = tmp1 ^ in2;
+        out5 = tmp1 ^ tmp2;
+        out7 = tmp2 ^ in0;
+        tmp4 = tmp3 ^ in4;
+        out1 = tmp3 ^ in2;
+        out6 = tmp4 ^ out0;
+        out4 = out6 ^ in2 ^ in5;
+        out3 = tmp1 ^ tmp4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_8C(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in2;
+        out0 = in1 ^ in7;
+        out7 = in0 ^ in6;
+        out5 = in4 ^ in6;
+        out6 = in5 ^ in7;
+        out2 = out0 ^ in0 ^ in3;
+        out3 = out5 ^ out7 ^ in2 ^ in7;
+        out4 = out6 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_8D(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in1 ^ in2;
+        tmp0 = in6 ^ in7;
+        out0 = in0 ^ in1 ^ in7;
+        out5 = in4 ^ in5 ^ in6;
+        out6 = tmp0 ^ in5;
+        out7 = tmp0 ^ in0;
+        out4 = tmp0 ^ out5 ^ in3;
+        out2 = out0 ^ in2 ^ in3;
+        out3 = out2 ^ in1 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_8E(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in1;
+        out4 = in5;
+        out7 = in0;
+        out5 = in6;
+        out6 = in7;
+        out3 = in0 ^ in4;
+        out1 = in0 ^ in2;
+        out2 = in0 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_8F(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in0 ^ in1;
+        tmp0 = in0 ^ in3;
+        out4 = in4 ^ in5;
+        out7 = in0 ^ in7;
+        out5 = in5 ^ in6;
+        out6 = in6 ^ in7;
+        out1 = out0 ^ in2;
+        out2 = tmp0 ^ in2;
+        out3 = tmp0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_90(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in2;
+        tmp1 = in2 ^ in6 ^ in7;
+        out3 = tmp0 ^ in7;
+        out1 = tmp1 ^ in5;
+        tmp2 = out1 ^ in4;
+        out6 = tmp2 ^ in3;
+        out5 = out6 ^ in1;
+        out4 = out5 ^ in0;
+        out0 = tmp0 ^ tmp2;
+        out7 = tmp0 ^ out4;
+        out2 = tmp1 ^ out5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_91(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in4;
+        tmp1 = tmp0 ^ in3 ^ in5;
+        out2 = tmp1 ^ in1;
+        out6 = tmp1 ^ in7;
+        tmp2 = out2 ^ in5 ^ in7;
+        out3 = tmp2 ^ in4;
+        out5 = tmp2 ^ in6;
+        out1 = tmp1 ^ out5 ^ in2;
+        tmp3 = out1 ^ in0;
+        out4 = tmp3 ^ in3;
+        out0 = tmp0 ^ tmp3;
+        out7 = tmp2 ^ tmp3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_92(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in1;
+        tmp0 = in4 ^ in5;
+        tmp1 = tmp0 ^ in1;
+        out2 = tmp0 ^ in3 ^ in7;
+        out0 = tmp1 ^ in6;
+        out7 = out2 ^ in0;
+        out4 = out0 ^ in0 ^ in2;
+        out5 = out4 ^ out7 ^ in5;
+        out6 = tmp1 ^ out5;
+        out1 = out6 ^ out7 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_93(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in1 ^ in3;
+        tmp0 = in2 ^ in7;
+        tmp1 = out3 ^ in6;
+        tmp2 = tmp0 ^ in4;
+        out5 = tmp0 ^ tmp1;
+        out6 = tmp2 ^ in3;
+        out2 = out6 ^ in5;
+        out0 = out2 ^ out5 ^ in0;
+        out7 = tmp1 ^ out0;
+        out1 = tmp2 ^ out0;
+        out4 = out1 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_94(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in2 ^ in6;
+        tmp0 = in1 ^ in4 ^ in5;
+        out1 = out3 ^ in5;
+        out5 = tmp0 ^ out3;
+        out0 = tmp0 ^ in7;
+        out4 = tmp0 ^ in0 ^ in3;
+        out6 = out1 ^ in3 ^ in7;
+        out2 = out4 ^ in6;
+        out7 = out0 ^ out2 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_95(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in3;
+        out3 = tmp0 ^ in6;
+        tmp1 = tmp0 ^ in7;
+        tmp2 = out3 ^ in0;
+        out6 = tmp1 ^ in5;
+        tmp3 = tmp2 ^ in4;
+        out7 = tmp3 ^ in2;
+        tmp4 = tmp3 ^ in5;
+        out2 = tmp4 ^ in1;
+        tmp5 = out2 ^ in6;
+        out0 = tmp1 ^ tmp5;
+        out1 = tmp5 ^ out7;
+        out4 = tmp2 ^ out1;
+        out5 = tmp4 ^ out4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_96(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in6 ^ in7;
+        tmp0 = in1 ^ in5;
+        tmp1 = in5 ^ in6;
+        out6 = out3 ^ in2 ^ in3;
+        out0 = tmp0 ^ in4;
+        tmp2 = tmp1 ^ in2;
+        out4 = out0 ^ in0 ^ in7;
+        out1 = tmp2 ^ in0;
+        out5 = tmp2 ^ in1;
+        out7 = tmp0 ^ out4 ^ in3;
+        out2 = tmp1 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_97(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in4;
+        tmp1 = in2 ^ in6;
+        out3 = in3 ^ in6 ^ in7;
+        out7 = tmp0 ^ in3;
+        tmp2 = tmp0 ^ in5;
+        out5 = tmp1 ^ in1;
+        out6 = tmp1 ^ out3;
+        out0 = tmp2 ^ in1;
+        out2 = tmp2 ^ out3 ^ in2;
+        tmp3 = out0 ^ in4;
+        out4 = tmp3 ^ in7;
+        out1 = tmp1 ^ tmp3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_98(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in5 ^ in7;
+        tmp1 = in1 ^ in4 ^ in7;
+        out1 = tmp0 ^ in2;
+        out0 = tmp1 ^ in6;
+        out2 = tmp1 ^ in3;
+        out6 = out0 ^ out1 ^ in1;
+        out5 = tmp0 ^ out2;
+        out3 = tmp1 ^ out6 ^ in0;
+        out7 = out0 ^ out5 ^ in0;
+        out4 = out6 ^ out7 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_99(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in3;
+        out5 = in1 ^ in3 ^ in4;
+        out6 = in2 ^ in4 ^ in5;
+        out4 = tmp0 ^ in2;
+        tmp1 = tmp0 ^ in6;
+        tmp2 = out5 ^ in7;
+        out7 = tmp1 ^ in5;
+        out0 = tmp1 ^ tmp2;
+        out2 = tmp2 ^ in2;
+        out3 = out0 ^ out6 ^ in3;
+        out1 = tmp1 ^ out3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_9A(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in3 ^ in4;
+        tmp0 = in0 ^ in5;
+        tmp1 = in1 ^ in6;
+        out5 = in1 ^ in3 ^ in5;
+        tmp2 = tmp0 ^ in7;
+        out3 = tmp0 ^ tmp1;
+        out0 = tmp1 ^ in4;
+        out7 = tmp2 ^ in3;
+        out1 = tmp2 ^ in2;
+        out6 = out0 ^ in1 ^ in2;
+        out4 = out1 ^ in4 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_9B(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out5 = in1 ^ in3;
+        tmp0 = in3 ^ in5;
+        out6 = in2 ^ in4;
+        out4 = in0 ^ in2 ^ in7;
+        out7 = tmp0 ^ in0;
+        out2 = out6 ^ in3;
+        out1 = out4 ^ in1 ^ in5;
+        out3 = out7 ^ in1 ^ in6;
+        out0 = tmp0 ^ out3 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_9C(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out1 = in2 ^ in5;
+        tmp0 = in0 ^ in3 ^ in6;
+        out3 = out1 ^ in0;
+        out6 = out1 ^ in6;
+        out7 = tmp0 ^ in7;
+        out4 = out7 ^ in4;
+        out2 = out4 ^ in1;
+        out0 = tmp0 ^ out2;
+        out5 = out0 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_9D(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out6 = in2 ^ in5;
+        tmp0 = in0 ^ in3;
+        out5 = in1 ^ in4 ^ in7;
+        out1 = out6 ^ in1;
+        out3 = tmp0 ^ out6;
+        out7 = tmp0 ^ in6;
+        out0 = out5 ^ in0;
+        out4 = out7 ^ in7;
+        out2 = out5 ^ out7 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_9E(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in1 ^ in4;
+        tmp0 = in0 ^ in5;
+        out6 = in2 ^ in6;
+        out7 = in0 ^ in3 ^ in7;
+        out4 = in0 ^ in4 ^ in6;
+        out5 = in1 ^ in5 ^ in7;
+        out1 = tmp0 ^ in2;
+        out3 = tmp0 ^ in7;
+        out2 = out4 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_9F(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out6 = in2;
+        out7 = in0 ^ in3;
+        tmp0 = in0 ^ in1;
+        out4 = in0 ^ in6;
+        out5 = in1 ^ in7;
+        out1 = tmp0 ^ in2 ^ in5;
+        out2 = out7 ^ in2 ^ in4 ^ in6;
+        out3 = out7 ^ in5 ^ in7;
+        out0 = tmp0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_A0(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in6;
+        out2 = tmp0 ^ in7;
+        tmp1 = tmp0 ^ in5;
+        out6 = out2 ^ in3 ^ in4;
+        out0 = tmp1 ^ in3;
+        tmp2 = out0 ^ in2;
+        out3 = tmp2 ^ in7;
+        tmp3 = tmp2 ^ in1;
+        out5 = tmp3 ^ in0;
+        out4 = tmp3 ^ out6;
+        out7 = out5 ^ out6 ^ in1;
+        out1 = tmp1 ^ out4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_A1(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in5;
+        tmp1 = tmp0 ^ in1;
+        tmp2 = tmp0 ^ in4;
+        out4 = tmp1 ^ in7;
+        out7 = tmp2 ^ in0;
+        out6 = tmp2 ^ out4 ^ in3;
+        out3 = out4 ^ in6;
+        out2 = out3 ^ in5;
+        out1 = out2 ^ in4;
+        out5 = out1 ^ out6 ^ in0;
+        out0 = tmp1 ^ out5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_A2(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in6;
+        tmp0 = in1 ^ in3 ^ in5;
+        out3 = tmp0 ^ in6;
+        out4 = tmp0 ^ in2 ^ in4;
+        out0 = out3 ^ in7;
+        out6 = out0 ^ in4;
+        out1 = out0 ^ out4 ^ in0;
+        out7 = out1 ^ in5;
+        out5 = out7 ^ in3 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_A3(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in2 ^ in6;
+        out3 = in1 ^ in5 ^ in6;
+        tmp0 = out2 ^ in0;
+        out4 = out2 ^ out3 ^ in3;
+        tmp1 = tmp0 ^ in4;
+        out0 = tmp0 ^ out4 ^ in7;
+        out5 = tmp1 ^ in3;
+        out7 = tmp1 ^ in5;
+        out1 = tmp1 ^ in1 ^ in7;
+        out6 = tmp1 ^ out0 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_A4(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in3;
+        tmp1 = in2 ^ in4;
+        tmp2 = in2 ^ in5;
+        tmp3 = in0 ^ in7;
+        out0 = tmp0 ^ in5;
+        out6 = tmp0 ^ in6 ^ in7;
+        out1 = tmp1 ^ in6;
+        out7 = tmp1 ^ tmp3;
+        out3 = tmp2 ^ in3;
+        tmp4 = tmp2 ^ out1;
+        out2 = tmp3 ^ in1;
+        out5 = tmp4 ^ out7;
+        out4 = tmp4 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_A5(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in2 ^ in5;
+        tmp0 = in1 ^ in6;
+        tmp1 = in0 ^ in1;
+        tmp2 = in2 ^ in4;
+        out6 = in1 ^ in3 ^ in7;
+        out4 = tmp0 ^ in5;
+        out1 = tmp0 ^ tmp2;
+        out0 = tmp1 ^ in3 ^ in5;
+        out2 = tmp1 ^ in2 ^ in7;
+        out7 = tmp2 ^ in0;
+        out5 = tmp0 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_A6(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in0;
+        out3 = in3 ^ in5 ^ in7;
+        out1 = in0 ^ in2 ^ in4 ^ in6;
+        out0 = out3 ^ in1;
+        out7 = out1 ^ in7;
+        out6 = out0 ^ in6;
+        out5 = out7 ^ in5;
+        out4 = out6 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_A7(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in0 ^ in2;
+        out3 = in5 ^ in7;
+        out7 = out2 ^ in4 ^ in6;
+        out6 = out3 ^ in1 ^ in3;
+        out1 = out7 ^ in1;
+        out5 = out7 ^ in7;
+        out0 = out6 ^ in0;
+        out4 = out6 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_A8(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in4;
+        tmp1 = in1 ^ in6;
+        tmp2 = in0 ^ in2 ^ in7;
+        out1 = tmp0 ^ in7;
+        out4 = tmp0 ^ in6;
+        out0 = tmp1 ^ in3;
+        out2 = tmp1 ^ in5;
+        out6 = tmp1 ^ in4;
+        out7 = tmp2 ^ in5;
+        out3 = tmp2 ^ out0 ^ in6;
+        out5 = out7 ^ in2 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_A9(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in2 ^ in6;
+        out6 = in1 ^ in4;
+        out7 = in0 ^ in2 ^ in5;
+        out5 = in0 ^ in3 ^ in7;
+        out2 = out4 ^ in1 ^ in5;
+        out1 = out6 ^ in2 ^ in7;
+        out0 = out2 ^ out7 ^ in3;
+        out3 = out1 ^ in0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_AA(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in2;
+        tmp1 = in1 ^ in3;
+        tmp2 = in6 ^ in7;
+        out1 = tmp0 ^ in4 ^ in7;
+        out3 = tmp1 ^ in0;
+        out0 = tmp1 ^ tmp2;
+        out2 = tmp2 ^ in5;
+        out7 = tmp0 ^ out2;
+        out6 = out1 ^ out7 ^ in1;
+        out5 = out0 ^ out6 ^ in0;
+        out4 = out5 ^ out7 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_AB(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in0 ^ in1;
+        tmp0 = in1 ^ in4;
+        tmp1 = in0 ^ in7;
+        out6 = tmp0 ^ in5;
+        out1 = tmp0 ^ tmp1 ^ in2;
+        out5 = tmp1 ^ in3 ^ in4;
+        out0 = tmp0 ^ out5 ^ in6;
+        out4 = out0 ^ out3 ^ in2;
+        out2 = out4 ^ in3 ^ in5;
+        out7 = tmp1 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_AC(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in1 ^ in3;
+        out1 = in2 ^ in4;
+        tmp0 = in0 ^ in2;
+        out4 = in4 ^ in7;
+        out5 = in0 ^ in5;
+        out6 = in1 ^ in6;
+        out7 = tmp0 ^ in7;
+        out3 = tmp0 ^ in3 ^ in6;
+        out2 = out5 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_AD(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in7;
+        out5 = in0;
+        out6 = in1;
+        out7 = in0 ^ in2;
+        out0 = in0 ^ in1 ^ in3;
+        out2 = out7 ^ in1 ^ in5;
+        out1 = in1 ^ in2 ^ in4;
+        out3 = out7 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_AE(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in3 ^ in4;
+        tmp0 = in0 ^ in4;
+        tmp1 = in0 ^ in7;
+        out0 = in1 ^ in3 ^ in7;
+        out1 = tmp0 ^ in2;
+        out5 = tmp0 ^ in5;
+        tmp2 = tmp1 ^ in6;
+        out2 = tmp1 ^ in5;
+        out3 = tmp2 ^ in3;
+        out7 = tmp2 ^ in2;
+        out6 = tmp2 ^ out2 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_AF(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in3;
+        tmp0 = in0 ^ in7;
+        out5 = in0 ^ in4;
+        out6 = in1 ^ in5;
+        out7 = in0 ^ in2 ^ in6;
+        out0 = tmp0 ^ in1 ^ in3;
+        out3 = tmp0 ^ in6;
+        out2 = tmp0 ^ in2 ^ in5;
+        out1 = out5 ^ in1 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_B0(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in4;
+        tmp1 = in3 ^ in6;
+        out2 = tmp0 ^ in7;
+        tmp2 = tmp0 ^ tmp1;
+        out0 = tmp2 ^ in5;
+        out3 = tmp2 ^ in2;
+        out6 = out3 ^ in6;
+        tmp3 = out6 ^ in0 ^ in1;
+        out7 = tmp3 ^ in5;
+        out5 = tmp3 ^ out2;
+        out1 = out0 ^ out5 ^ in0;
+        out4 = tmp1 ^ out5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_B1(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in4;
+        out2 = tmp0 ^ in2 ^ in7;
+        tmp1 = out2 ^ in6;
+        out1 = tmp1 ^ in5;
+        out3 = tmp1 ^ in7;
+        out4 = tmp1 ^ in0;
+        out6 = out3 ^ in3;
+        out0 = out6 ^ in0 ^ in2 ^ in5;
+        out5 = tmp1 ^ out0 ^ in1;
+        out7 = tmp0 ^ out5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_B2(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in4;
+        tmp0 = in4 ^ in7;
+        tmp1 = in1 ^ in3 ^ in6;
+        out3 = tmp0 ^ tmp1;
+        tmp2 = tmp1 ^ in0;
+        out0 = out3 ^ in5;
+        out4 = tmp2 ^ in2;
+        tmp3 = out4 ^ in6;
+        out5 = tmp0 ^ tmp3;
+        out1 = tmp3 ^ out0;
+        tmp4 = out1 ^ in7;
+        out7 = tmp4 ^ in3;
+        out6 = tmp2 ^ tmp4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_B3(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in2 ^ in4;
+        tmp0 = in0 ^ in5;
+        tmp1 = in1 ^ in6;
+        out3 = tmp1 ^ in4 ^ in7;
+        tmp2 = tmp0 ^ out3;
+        out0 = tmp2 ^ in3;
+        out1 = tmp2 ^ in2;
+        out5 = out0 ^ in2 ^ in6;
+        out7 = tmp1 ^ out5;
+        out4 = out7 ^ in1 ^ in5 ^ in7;
+        out6 = tmp0 ^ out4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_B4(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in0 ^ in1;
+        out5 = out4 ^ in2;
+        tmp0 = out4 ^ in4;
+        out6 = out5 ^ in0 ^ in3;
+        out7 = tmp0 ^ out6;
+        out2 = tmp0 ^ in6 ^ in7;
+        out3 = out7 ^ in0 ^ in7;
+        out0 = out5 ^ out7 ^ in5;
+        out1 = out0 ^ out6 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_B5(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in1;
+        tmp1 = in2 ^ in4;
+        out4 = tmp0 ^ in4;
+        out3 = tmp1 ^ in7;
+        tmp2 = out4 ^ in5;
+        out7 = out3 ^ in0 ^ in3;
+        out0 = tmp2 ^ in3;
+        out2 = tmp0 ^ out3 ^ in6;
+        out5 = tmp1 ^ tmp2;
+        out6 = out2 ^ out7 ^ in2;
+        out1 = tmp0 ^ out0 ^ out6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_B6(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in3 ^ in4;
+        tmp0 = in1 ^ in2;
+        tmp1 = in0 ^ in4;
+        tmp2 = in3 ^ in5;
+        tmp3 = out3 ^ in1 ^ in7;
+        out5 = tmp0 ^ tmp1;
+        out6 = tmp0 ^ tmp2;
+        out2 = tmp1 ^ in6;
+        out4 = tmp1 ^ tmp3;
+        out0 = tmp3 ^ in5;
+        out1 = out2 ^ in2 ^ in5;
+        out7 = tmp2 ^ out1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_B7(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in4;
+        tmp0 = in0 ^ in4;
+        out2 = tmp0 ^ in2 ^ in6;
+        tmp1 = out2 ^ in7;
+        out1 = out2 ^ in1 ^ in5;
+        out7 = tmp1 ^ in3;
+        out5 = out1 ^ in6;
+        out6 = tmp0 ^ out1 ^ in3;
+        out0 = tmp1 ^ out6;
+        out4 = out0 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_B8(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in4;
+        tmp1 = in2 ^ in5;
+        out2 = tmp0 ^ in5;
+        out4 = tmp1 ^ in0;
+        tmp2 = tmp1 ^ in7;
+        out6 = tmp2 ^ out2;
+        out7 = out4 ^ in3;
+        out1 = tmp2 ^ in4;
+        out3 = tmp0 ^ out7;
+        out0 = out3 ^ out4 ^ in6;
+        out5 = out0 ^ in0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_B9(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in2;
+        tmp1 = in4 ^ in5;
+        out4 = tmp0 ^ tmp1;
+        tmp2 = tmp0 ^ in3 ^ in7;
+        out3 = out4 ^ in1;
+        out7 = tmp2 ^ in5;
+        out2 = out3 ^ in0;
+        out1 = out2 ^ in7;
+        out6 = out1 ^ in5 ^ in6;
+        out0 = tmp2 ^ out6;
+        out5 = tmp1 ^ out0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_BA(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in5 ^ in7;
+        out2 = tmp0 ^ in4;
+        tmp1 = out2 ^ in2;
+        out1 = tmp1 ^ in0;
+        out6 = tmp1 ^ in1;
+        out4 = out1 ^ in3 ^ in4;
+        tmp2 = out4 ^ out6;
+        out7 = out4 ^ in6 ^ in7;
+        out5 = tmp2 ^ in6;
+        out3 = tmp0 ^ tmp2;
+        out0 = out6 ^ out7 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_BB(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in2 ^ in4 ^ in5 ^ in7;
+        tmp0 = out2 ^ in1;
+        out4 = out2 ^ in0 ^ in3;
+        out1 = tmp0 ^ in0;
+        out6 = tmp0 ^ in6;
+        out3 = out1 ^ in2;
+        tmp1 = out4 ^ out6 ^ in4;
+        out0 = tmp1 ^ in7;
+        out5 = tmp1 ^ in5;
+        out7 = tmp0 ^ tmp1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_BC(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in2;
+        tmp1 = in2 ^ in4;
+        out0 = in1 ^ in3 ^ in4;
+        out6 = in1 ^ in2 ^ in7;
+        out7 = tmp0 ^ in3;
+        out5 = tmp0 ^ out6 ^ in6;
+        out1 = tmp1 ^ in5;
+        tmp2 = out1 ^ out5 ^ in1;
+        out3 = tmp2 ^ in3;
+        out4 = tmp1 ^ tmp2;
+        out2 = tmp2 ^ out6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_BD(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in3;
+        tmp1 = in1 ^ in4;
+        out0 = tmp0 ^ tmp1;
+        out7 = tmp0 ^ in2 ^ in7;
+        out1 = tmp1 ^ in2 ^ in5;
+        tmp2 = out1 ^ in0;
+        out2 = tmp2 ^ in6;
+        out3 = out2 ^ in1 ^ in7;
+        out4 = out3 ^ in2;
+        out5 = tmp1 ^ out4;
+        out6 = tmp2 ^ out4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_BE(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in3 ^ in6;
+        out4 = tmp0 ^ in5;
+        out7 = tmp0 ^ in2;
+        out3 = out4 ^ in4;
+        out1 = out3 ^ out7 ^ in0;
+        out2 = out3 ^ in3 ^ in7;
+        out0 = out2 ^ out4 ^ in1;
+        out5 = tmp0 ^ out0;
+        out6 = out1 ^ out5 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_BF(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in4;
+        out3 = tmp0 ^ in5 ^ in6;
+        out4 = out3 ^ in3;
+        tmp1 = out3 ^ in7;
+        out2 = tmp1 ^ in2;
+        out5 = tmp1 ^ in1;
+        tmp2 = out2 ^ in5;
+        out7 = tmp2 ^ in3 ^ in4;
+        tmp3 = tmp0 ^ out5;
+        out0 = tmp3 ^ out4;
+        out1 = tmp2 ^ tmp3;
+        out6 = tmp3 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_C0(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out5 = in2 ^ in5;
+        tmp0 = in1 ^ in4;
+        tmp1 = in3 ^ in6;
+        out0 = out5 ^ in1;
+        out4 = tmp0 ^ in7;
+        out3 = tmp0 ^ tmp1;
+        out1 = tmp1 ^ in2;
+        out6 = tmp1 ^ in0;
+        out7 = out4 ^ in0;
+        out2 = out4 ^ out5 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_C1(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out5 = in2;
+        tmp0 = in0 ^ in1;
+        out4 = in1 ^ in7;
+        out6 = in0 ^ in3;
+        out3 = in1 ^ in4 ^ in6;
+        tmp1 = tmp0 ^ in2;
+        out7 = tmp0 ^ in4;
+        out0 = tmp1 ^ in5;
+        out1 = tmp1 ^ out6 ^ in6;
+        out2 = out6 ^ out7 ^ in5 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_C2(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in1 ^ in3 ^ in4;
+        tmp0 = in0 ^ in3 ^ in6;
+        out5 = in2 ^ in4 ^ in5;
+        tmp1 = out4 ^ in7;
+        out1 = tmp0 ^ in2;
+        out6 = tmp0 ^ in5;
+        out2 = out5 ^ in3;
+        out7 = tmp0 ^ tmp1;
+        out3 = tmp1 ^ in2 ^ in6;
+        out0 = tmp1 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_C3(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in1 ^ in3;
+        tmp0 = in0 ^ in2;
+        tmp1 = in3 ^ in5;
+        out5 = in2 ^ in4;
+        tmp2 = tmp0 ^ out4;
+        out2 = tmp1 ^ in4;
+        out6 = tmp1 ^ in0;
+        out0 = tmp1 ^ tmp2 ^ in7;
+        out1 = tmp2 ^ in6;
+        out7 = out1 ^ out5 ^ in3;
+        out3 = tmp0 ^ out7 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_C4(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in7;
+        out3 = tmp0 ^ in4;
+        tmp1 = tmp0 ^ in2;
+        out1 = tmp1 ^ in6;
+        out5 = tmp1 ^ in5;
+        out4 = out1 ^ out3 ^ in1;
+        out0 = out4 ^ in4 ^ in5;
+        out2 = out0 ^ out3 ^ in0;
+        out7 = out1 ^ out2 ^ in7;
+        out6 = tmp1 ^ out0 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_C5(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in4 ^ in7;
+        tmp0 = in3 ^ in7;
+        out4 = in1 ^ in2 ^ in6;
+        out6 = in0 ^ in3 ^ in4;
+        out5 = tmp0 ^ in2;
+        out1 = tmp0 ^ out4;
+        out0 = out4 ^ in0 ^ in5;
+        out2 = out0 ^ out5 ^ in4;
+        out7 = tmp0 ^ out2 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_C6(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in5 ^ in6;
+        tmp1 = in1 ^ in7;
+        tmp2 = tmp0 ^ in0;
+        tmp3 = tmp0 ^ tmp1;
+        tmp4 = tmp2 ^ in4;
+        out0 = tmp3 ^ in2;
+        out6 = tmp4 ^ in3;
+        out2 = out6 ^ in2;
+        out7 = tmp1 ^ tmp4;
+        out3 = tmp2 ^ out2;
+        tmp5 = out3 ^ in5;
+        out5 = tmp5 ^ in7;
+        out4 = tmp3 ^ tmp5;
+        out1 = tmp4 ^ out5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_C7(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in2 ^ in4;
+        tmp0 = in3 ^ in5;
+        tmp1 = out3 ^ in7;
+        out6 = tmp0 ^ in0 ^ in4;
+        out5 = tmp1 ^ in3;
+        out2 = out6 ^ in6;
+        out7 = out2 ^ in1 ^ in3;
+        out0 = tmp1 ^ out7;
+        out1 = tmp0 ^ out0;
+        out4 = out1 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_C8(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out0 = in1 ^ in2;
+        out1 = in2 ^ in3;
+        tmp0 = in5 ^ in6;
+        tmp1 = in0 ^ in7;
+        out2 = out1 ^ in1 ^ in4;
+        out4 = tmp0 ^ in4;
+        out5 = tmp0 ^ in7;
+        out6 = tmp1 ^ in6;
+        out7 = tmp1 ^ in1;
+        out3 = out2 ^ in0 ^ in2 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_C9(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in5 ^ in6;
+        out7 = in0 ^ in1;
+        tmp0 = in1 ^ in3;
+        out5 = in6 ^ in7;
+        out6 = in0 ^ in7;
+        out0 = out7 ^ in2;
+        out3 = out7 ^ in4 ^ in5;
+        out1 = tmp0 ^ in2;
+        out2 = tmp0 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_CA(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in7;
+        tmp1 = in2 ^ in7;
+        tmp2 = tmp0 ^ in6;
+        out0 = tmp1 ^ in1;
+        tmp3 = tmp1 ^ in3;
+        out6 = tmp2 ^ in5;
+        out7 = tmp2 ^ in1;
+        out2 = tmp3 ^ in4;
+        out5 = out6 ^ in0 ^ in4;
+        out4 = out5 ^ in3;
+        out1 = tmp0 ^ tmp3;
+        out3 = tmp3 ^ out5 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_CB(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in4 ^ in7;
+        tmp1 = in5 ^ in7;
+        out7 = in0 ^ in1 ^ in6;
+        out5 = tmp0 ^ in6;
+        out2 = tmp0 ^ in3;
+        out6 = tmp1 ^ in0;
+        out4 = tmp1 ^ in3 ^ in6;
+        tmp2 = out5 ^ out7 ^ in2;
+        out1 = tmp2 ^ out2;
+        out0 = tmp2 ^ in4;
+        out3 = tmp2 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_CC(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in5;
+        tmp1 = in1 ^ in6;
+        out1 = in2 ^ in3 ^ in7;
+        out5 = tmp0 ^ in6;
+        out0 = tmp1 ^ in2;
+        tmp2 = out5 ^ in0 ^ in7;
+        out3 = tmp2 ^ in4;
+        out6 = tmp0 ^ out3;
+        out7 = tmp1 ^ tmp2 ^ in3;
+        tmp3 = out1 ^ out6;
+        out4 = tmp2 ^ tmp3;
+        out2 = tmp3 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_CD(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out5 = in3 ^ in6;
+        tmp0 = in0 ^ in1;
+        tmp1 = in2 ^ in7;
+        out6 = in0 ^ in4 ^ in7;
+        out2 = tmp0 ^ out5 ^ in4;
+        out7 = tmp0 ^ in5;
+        out0 = tmp0 ^ in2 ^ in6;
+        out4 = tmp1 ^ in5;
+        out1 = tmp1 ^ in1 ^ in3;
+        out3 = out6 ^ in5 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_CE(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in5;
+        tmp1 = tmp0 ^ in3;
+        out4 = tmp1 ^ in4;
+        tmp2 = out4 ^ in6;
+        out3 = tmp2 ^ in0;
+        out5 = tmp2 ^ in2;
+        out2 = out3 ^ in5 ^ in7;
+        out6 = tmp1 ^ out2;
+        out7 = out2 ^ out4 ^ in1;
+        out1 = tmp2 ^ out6;
+        out0 = tmp0 ^ out7 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_CF(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in6;
+        tmp1 = in0 ^ in1 ^ in5;
+        out4 = in2 ^ in3 ^ in5;
+        out5 = tmp0 ^ in4;
+        out7 = tmp1 ^ in6;
+        out1 = tmp1 ^ out4 ^ in7;
+        tmp2 = out5 ^ in0;
+        out2 = tmp2 ^ in7;
+        out3 = tmp2 ^ out4;
+        out6 = tmp0 ^ out2 ^ in5;
+        out0 = tmp0 ^ out1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_D0(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in3;
+        tmp1 = in1 ^ in4;
+        tmp2 = in2 ^ in5;
+        out7 = tmp0 ^ tmp1;
+        out0 = tmp1 ^ tmp2;
+        tmp3 = tmp2 ^ in3;
+        out1 = tmp3 ^ in6;
+        tmp4 = out1 ^ in1;
+        out2 = tmp4 ^ in7;
+        out3 = out2 ^ in2;
+        out4 = tmp0 ^ out3;
+        out5 = tmp3 ^ out3;
+        out6 = tmp4 ^ out4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_D1(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in5 ^ in6;
+        tmp1 = tmp0 ^ in1;
+        out1 = tmp1 ^ in2;
+        out2 = tmp1 ^ in7;
+        out3 = out2 ^ in3;
+        out5 = out3 ^ in2;
+        tmp2 = out3 ^ in0;
+        out4 = tmp2 ^ in4;
+        out7 = tmp0 ^ out4;
+        out6 = tmp2 ^ out1 ^ in6;
+        out0 = out2 ^ out6 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_D2(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in5 ^ in6;
+        out2 = tmp0 ^ in2 ^ in3;
+        out1 = out2 ^ in0;
+        out3 = out2 ^ in1;
+        out4 = out1 ^ in1 ^ in2;
+        out6 = out1 ^ in6 ^ in7;
+        out7 = out4 ^ in4 ^ in5;
+        out5 = out4 ^ out6 ^ in4;
+        out0 = tmp0 ^ out5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_D3(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in3 ^ in5 ^ in6;
+        tmp0 = out2 ^ in2;
+        tmp1 = tmp0 ^ in1;
+        out1 = tmp1 ^ in0;
+        out3 = tmp1 ^ in3;
+        out4 = out1 ^ in2 ^ in4;
+        tmp2 = out4 ^ in5;
+        out7 = tmp2 ^ in7;
+        out0 = tmp0 ^ out7;
+        tmp3 = out0 ^ in0;
+        out5 = tmp3 ^ in6;
+        out6 = tmp2 ^ tmp3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_D4(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in3 ^ in5;
+        tmp0 = in1 ^ in5;
+        tmp1 = tmp0 ^ in2;
+        out4 = tmp1 ^ in0;
+        tmp2 = tmp1 ^ in6;
+        out2 = out4 ^ in3 ^ in7;
+        out0 = tmp2 ^ in4;
+        out5 = tmp2 ^ out3;
+        out1 = tmp0 ^ out5 ^ in7;
+        out6 = tmp0 ^ out2 ^ in4;
+        out7 = tmp1 ^ out6 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_D5(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in5;
+        tmp0 = in0 ^ in4;
+        tmp1 = tmp0 ^ in1 ^ in5;
+        out4 = tmp1 ^ in2;
+        out0 = out4 ^ in6;
+        tmp2 = tmp0 ^ out0;
+        out5 = tmp2 ^ in3;
+        out1 = out5 ^ in7;
+        out6 = tmp1 ^ out1;
+        out7 = tmp2 ^ out6;
+        out2 = out7 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_D6(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in2 ^ in4 ^ in6;
+        out5 = tmp0 ^ in3;
+        out0 = tmp0 ^ in5 ^ in7;
+        out3 = out0 ^ out5 ^ in2;
+        tmp1 = out3 ^ in0;
+        out1 = tmp1 ^ in6;
+        out2 = tmp1 ^ in7;
+        out4 = tmp1 ^ in1;
+        out6 = tmp1 ^ in4;
+        out7 = tmp0 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_D7(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in3;
+        out3 = in2 ^ in5 ^ in7;
+        out2 = tmp0 ^ in5;
+        tmp1 = tmp0 ^ out3 ^ in1;
+        out1 = tmp1 ^ in6;
+        out4 = tmp1 ^ in4;
+        tmp2 = out1 ^ in4;
+        out6 = tmp2 ^ in1;
+        out7 = tmp2 ^ in2;
+        out0 = tmp2 ^ in3;
+        out5 = tmp2 ^ in0 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_D8(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in0;
+        out5 = in1;
+        tmp0 = in1 ^ in2;
+        out6 = in0 ^ in2;
+        out0 = tmp0 ^ in4;
+        tmp1 = tmp0 ^ in3;
+        out7 = tmp1 ^ out6;
+        out2 = tmp1 ^ in6;
+        out3 = out7 ^ in7;
+        out1 = tmp1 ^ in1 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_D9(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in0 ^ in4;
+        out5 = in1 ^ in5;
+        out2 = in1 ^ in3 ^ in6;
+        out3 = in0 ^ in1 ^ in7;
+        out6 = in0 ^ in2 ^ in6;
+        out0 = out4 ^ in1 ^ in2;
+        out1 = out5 ^ in2 ^ in3;
+        out7 = out3 ^ in3;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_DA(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out5 = in1 ^ in4;
+        tmp0 = in2 ^ in7;
+        tmp1 = in0 ^ in2 ^ in3;
+        out0 = tmp0 ^ out5;
+        out4 = tmp0 ^ tmp1;
+        out2 = tmp0 ^ in3 ^ in6;
+        out1 = tmp1 ^ in5;
+        out3 = tmp1 ^ in1;
+        out6 = out1 ^ in3;
+        out7 = out3 ^ in2 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_DB(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in1;
+        tmp1 = in1 ^ in5;
+        tmp2 = in3 ^ in7;
+        out3 = tmp0 ^ in2;
+        out5 = tmp1 ^ in4;
+        out6 = tmp1 ^ out3 ^ in6;
+        out2 = tmp2 ^ in6;
+        tmp3 = tmp2 ^ in4;
+        tmp4 = out3 ^ in3;
+        out4 = tmp3 ^ in0;
+        out1 = tmp4 ^ in5;
+        out0 = tmp3 ^ tmp4;
+        out7 = tmp0 ^ out2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_DC(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in2;
+        tmp1 = in0 ^ in3;
+        out6 = tmp0 ^ in4;
+        tmp2 = tmp0 ^ in7;
+        out3 = tmp1 ^ in6;
+        tmp3 = tmp1 ^ in1;
+        out1 = tmp1 ^ tmp2 ^ in5;
+        out4 = tmp2 ^ in6;
+        out2 = tmp3 ^ in2;
+        out7 = tmp3 ^ in5;
+        out5 = tmp2 ^ out2;
+        out0 = out2 ^ out3 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_DD(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in0 ^ in6;
+        out2 = in0 ^ in1 ^ in3;
+        out6 = out3 ^ in2 ^ in4;
+        out7 = out2 ^ in5 ^ in7;
+        out0 = out6 ^ in1;
+        out4 = out6 ^ in7;
+        out5 = out7 ^ in0;
+        out1 = out5 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_DE(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in3 ^ in6;
+        tmp1 = in3 ^ in4 ^ in7;
+        out4 = tmp0 ^ in0;
+        out5 = tmp1 ^ in1;
+        out3 = out4 ^ in7;
+        out2 = out3 ^ in6;
+        out1 = out2 ^ in5;
+        out6 = tmp1 ^ out1;
+        out0 = tmp0 ^ out5;
+        out7 = out0 ^ out1 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_DF(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in0 ^ in3 ^ in7;
+        tmp0 = out2 ^ in1 ^ in5;
+        out1 = tmp0 ^ in2;
+        out7 = tmp0 ^ in6;
+        out5 = tmp0 ^ in0 ^ in4;
+        tmp1 = out1 ^ out5 ^ in6;
+        out4 = tmp1 ^ in3;
+        out6 = tmp1 ^ in5;
+        tmp2 = tmp1 ^ in7;
+        out0 = tmp2 ^ in1;
+        out3 = tmp2 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_E0(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in1 ^ in7;
+        tmp0 = in2 ^ in4;
+        out4 = out3 ^ in3 ^ in5;
+        out2 = tmp0 ^ in1;
+        tmp1 = tmp0 ^ in6;
+        out0 = out4 ^ in2;
+        out6 = out4 ^ in0;
+        out1 = tmp1 ^ in3;
+        out5 = tmp1 ^ in0;
+        out7 = out5 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_E1(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in1 ^ in4;
+        tmp0 = in1 ^ in7;
+        out3 = tmp0 ^ in3;
+        tmp1 = out3 ^ in5;
+        out4 = tmp1 ^ in4;
+        tmp2 = tmp1 ^ in0;
+        out0 = tmp2 ^ in2;
+        out6 = tmp2 ^ in6;
+        tmp3 = out0 ^ out4 ^ in6;
+        out5 = tmp3 ^ in5;
+        out7 = tmp0 ^ tmp3;
+        out1 = tmp2 ^ out5 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_E2(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in1 ^ in2;
+        out4 = in1 ^ in5;
+        out2 = in2 ^ in4 ^ in7;
+        out5 = in0 ^ in2 ^ in6;
+        out0 = out3 ^ in3 ^ in5;
+        out7 = out3 ^ in0 ^ in4;
+        out6 = out2 ^ out7 ^ in3;
+        out1 = out5 ^ in3 ^ in4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_E3(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in4 ^ in7;
+        tmp0 = in1 ^ in3;
+        out3 = tmp0 ^ in2;
+        tmp1 = out3 ^ in0;
+        out0 = tmp1 ^ in5;
+        tmp2 = tmp1 ^ in4;
+        out1 = tmp2 ^ in6;
+        tmp3 = tmp2 ^ in3;
+        out7 = tmp3 ^ in7;
+        out6 = out1 ^ out2 ^ in2;
+        tmp4 = tmp0 ^ out0;
+        out5 = tmp4 ^ in6;
+        out4 = tmp3 ^ tmp4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_E4(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in6;
+        tmp0 = in0 ^ in4;
+        tmp1 = tmp0 ^ in2 ^ in6;
+        out2 = tmp1 ^ in1;
+        out7 = out2 ^ in5;
+        tmp2 = tmp0 ^ out7;
+        out4 = tmp2 ^ in3;
+        out0 = out4 ^ in7;
+        out6 = tmp1 ^ out0;
+        out5 = tmp2 ^ out6;
+        out1 = out5 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_E5(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in3 ^ in6;
+        tmp0 = in0 ^ in1;
+        tmp1 = in5 ^ in7;
+        out2 = tmp0 ^ in4 ^ in6;
+        tmp2 = tmp1 ^ out2;
+        out6 = tmp2 ^ in3;
+        out7 = tmp2 ^ in2;
+        out0 = out6 ^ in2 ^ in4;
+        out5 = out6 ^ in1 ^ in2;
+        out1 = tmp0 ^ out5 ^ in5;
+        out4 = tmp1 ^ out1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_E6(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in2 ^ in6 ^ in7;
+        out2 = out3 ^ in0 ^ in4;
+        out4 = out3 ^ in1 ^ in5;
+        out1 = out2 ^ in3;
+        out7 = out2 ^ out4 ^ in2;
+        out0 = out4 ^ in3 ^ in7;
+        out5 = out1 ^ in4;
+        out6 = out0 ^ out2 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_E7(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in3;
+        out3 = tmp0 ^ in6 ^ in7;
+        tmp1 = out3 ^ in0;
+        out5 = tmp1 ^ in5;
+        tmp2 = tmp1 ^ in4;
+        tmp3 = out5 ^ in7;
+        out1 = tmp2 ^ in1;
+        out0 = tmp3 ^ in1;
+        out6 = out1 ^ in2;
+        out2 = tmp0 ^ tmp2;
+        tmp4 = tmp3 ^ out6;
+        out4 = tmp4 ^ in6;
+        out7 = tmp4 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_E8(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in3 ^ in6;
+        tmp0 = in4 ^ in7;
+        out1 = in2 ^ in3 ^ in4;
+        out5 = tmp0 ^ in0;
+        tmp1 = tmp0 ^ in1;
+        tmp2 = tmp1 ^ in5;
+        out0 = tmp1 ^ out1;
+        out2 = tmp2 ^ in2;
+        out6 = tmp2 ^ out5;
+        tmp3 = out6 ^ in6;
+        out3 = tmp3 ^ in7;
+        out7 = tmp3 ^ in2 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_E9(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in1;
+        tmp1 = in3 ^ in6;
+        tmp2 = tmp0 ^ in6;
+        out4 = tmp1 ^ in4;
+        out6 = tmp2 ^ in5;
+        out7 = tmp2 ^ in2 ^ in7;
+        out3 = out6 ^ in3 ^ in7;
+        out0 = tmp1 ^ out7;
+        out2 = out3 ^ out4 ^ in0;
+        out5 = tmp0 ^ out2;
+        out1 = out0 ^ out5 ^ in5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_EA(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in6 ^ in7;
+        out5 = in0 ^ in7;
+        out6 = in0 ^ in1;
+        out0 = in1 ^ in2 ^ in3;
+        out2 = in2 ^ in4 ^ in5;
+        out7 = out6 ^ in2;
+        out1 = out0 ^ out6 ^ in4;
+        out3 = out7 ^ in5 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_EB(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in4 ^ in5;
+        tmp0 = in0 ^ in1;
+        out4 = in4 ^ in6 ^ in7;
+        out5 = in0 ^ in5 ^ in7;
+        out6 = tmp0 ^ in6;
+        tmp1 = tmp0 ^ in2;
+        out0 = tmp1 ^ in3;
+        out7 = tmp1 ^ in7;
+        out1 = out0 ^ in4;
+        out3 = out0 ^ in5 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_EC(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out3 = in0 ^ in5;
+        out4 = in2 ^ in3 ^ in7;
+        out5 = in0 ^ in3 ^ in4;
+        out6 = out3 ^ in1 ^ in4;
+        out1 = out4 ^ in4;
+        out0 = out4 ^ in1 ^ in6;
+        out2 = out0 ^ out5 ^ in5;
+        out7 = out2 ^ in4 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_ED(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in2 ^ in4;
+        tmp1 = in3 ^ in5;
+        out4 = tmp0 ^ in3 ^ in7;
+        out3 = tmp1 ^ in0;
+        out1 = out4 ^ in1;
+        out5 = out3 ^ in4;
+        out7 = out1 ^ out5 ^ in6;
+        out2 = tmp0 ^ out7;
+        out0 = tmp1 ^ out7;
+        out6 = out2 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_EE(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in2;
+        tmp0 = in0 ^ in1;
+        out5 = in0 ^ in3;
+        tmp1 = tmp0 ^ in2;
+        out6 = tmp0 ^ in4;
+        tmp2 = tmp1 ^ out5;
+        out7 = tmp1 ^ in5;
+        out1 = tmp2 ^ out6 ^ in7;
+        out0 = tmp2 ^ in6;
+        tmp3 = out7 ^ in1;
+        out3 = tmp3 ^ in7;
+        out2 = tmp3 ^ in4 ^ in6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_EF(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out4 = in2 ^ in4;
+        tmp0 = in0 ^ in5;
+        tmp1 = in4 ^ in6;
+        out5 = tmp0 ^ in3;
+        out2 = tmp0 ^ tmp1;
+        out6 = tmp1 ^ in0 ^ in1;
+        out3 = out5 ^ in2 ^ in7;
+        out7 = out3 ^ in1 ^ in3;
+        out0 = out4 ^ out6 ^ in3;
+        out1 = tmp1 ^ out0 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_F0(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in2;
+        tmp1 = in4 ^ in5;
+        out2 = tmp0 ^ in6;
+        out3 = tmp1 ^ in1;
+        tmp2 = tmp1 ^ in7;
+        out1 = out2 ^ out3 ^ in3;
+        tmp3 = tmp0 ^ tmp2;
+        out0 = tmp3 ^ in3;
+        out5 = tmp3 ^ in0;
+        out4 = out1 ^ out5 ^ in4;
+        out7 = out4 ^ in2;
+        out6 = tmp2 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_F1(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in1 ^ in6;
+        tmp0 = in3 ^ in5;
+        out3 = tmp0 ^ in1 ^ in4;
+        tmp1 = out3 ^ in2;
+        out1 = tmp1 ^ in6;
+        tmp2 = tmp1 ^ in0;
+        tmp3 = out1 ^ in5;
+        out0 = tmp2 ^ in7;
+        out6 = tmp2 ^ in4;
+        out7 = tmp3 ^ in0;
+        out5 = tmp0 ^ out0;
+        out4 = tmp3 ^ out5 ^ in1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_F2(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in4 ^ in5;
+        out2 = in2 ^ in6 ^ in7;
+        tmp1 = tmp0 ^ in1;
+        tmp2 = tmp1 ^ in2;
+        out0 = tmp2 ^ in3;
+        out3 = tmp2 ^ in7;
+        out5 = out3 ^ in0 ^ in4;
+        tmp3 = tmp0 ^ out5;
+        out7 = tmp3 ^ in3;
+        out4 = tmp3 ^ out2;
+        out1 = out0 ^ out4 ^ in4;
+        out6 = tmp1 ^ out1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_F3(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in6 ^ in7;
+        tmp0 = in0 ^ in1;
+        out4 = tmp0 ^ in6;
+        tmp1 = tmp0 ^ in2;
+        out5 = tmp1 ^ in7;
+        out6 = tmp1 ^ in3;
+        out7 = out6 ^ in4;
+        out0 = out7 ^ in5;
+        out1 = out0 ^ in6;
+        out3 = out0 ^ in0 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_F4(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in0 ^ in1 ^ in2;
+        tmp0 = out2 ^ in3;
+        out4 = tmp0 ^ in4;
+        out5 = out4 ^ in5;
+        out6 = out5 ^ in6;
+        out7 = out6 ^ in7;
+        out0 = out7 ^ in0;
+        out1 = out0 ^ in1;
+        out3 = tmp0 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_F5(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in0 ^ in1;
+        tmp0 = out2 ^ in2;
+        out4 = tmp0 ^ in3;
+        out5 = out4 ^ in4;
+        out6 = out5 ^ in5;
+        out7 = out6 ^ in6;
+        out0 = out7 ^ in7;
+        out1 = out0 ^ in0;
+        out3 = tmp0 ^ out0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_F6(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in7;
+        out2 = tmp0 ^ in2;
+        out4 = out2 ^ in1 ^ in4;
+        out7 = out4 ^ in3 ^ in5;
+        out5 = out7 ^ in4 ^ in7;
+        out0 = tmp0 ^ out7 ^ in6;
+        tmp1 = out0 ^ in1;
+        out6 = out0 ^ in0 ^ in5;
+        out3 = tmp1 ^ in3;
+        out1 = tmp0 ^ tmp1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_F7(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in0 ^ in7;
+        tmp0 = out2 ^ in1;
+        out4 = tmp0 ^ in2;
+        out5 = out4 ^ in3 ^ in7;
+        out6 = out5 ^ in4;
+        out7 = out6 ^ in5;
+        out0 = out7 ^ in6;
+        out1 = out0 ^ in7;
+        out3 = tmp0 ^ out1;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_F8(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in4;
+        tmp1 = in3 ^ in5;
+        tmp2 = tmp0 ^ in6;
+        out4 = tmp0 ^ tmp1;
+        out1 = tmp1 ^ in2 ^ in4;
+        out3 = tmp2 ^ in1;
+        out5 = out3 ^ in5;
+        out7 = out1 ^ out5 ^ in7;
+        out6 = tmp1 ^ out7;
+        out0 = tmp2 ^ out7;
+        out2 = out6 ^ in0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_F9(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in3 ^ in5;
+        tmp1 = in0 ^ in6;
+        out4 = tmp0 ^ in0;
+        tmp2 = tmp1 ^ in4;
+        tmp3 = tmp1 ^ in2;
+        out5 = tmp2 ^ in1;
+        out3 = out5 ^ in3;
+        tmp4 = tmp3 ^ out3;
+        out1 = tmp4 ^ in5;
+        out0 = tmp4 ^ in0 ^ in7;
+        out6 = tmp0 ^ out0 ^ in4;
+        out7 = tmp2 ^ tmp4;
+        out2 = tmp3 ^ out6;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_FA(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in1;
+        tmp1 = tmp0 ^ in2;
+        tmp2 = tmp0 ^ in5;
+        tmp3 = tmp1 ^ in7;
+        out5 = tmp2 ^ in6;
+        out6 = tmp3 ^ in6;
+        out7 = tmp3 ^ in3;
+        out3 = out6 ^ in4;
+        out2 = tmp1 ^ out5;
+        out4 = out2 ^ out3 ^ in1;
+        out0 = out4 ^ out7 ^ in5;
+        out1 = tmp2 ^ out0;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_FB(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in5 ^ in6;
+        tmp0 = in0 ^ in1;
+        out4 = in0 ^ in5 ^ in7;
+        out5 = tmp0 ^ in6;
+        tmp1 = tmp0 ^ in2;
+        out6 = tmp1 ^ in7;
+        out7 = tmp1 ^ in3;
+        out0 = out7 ^ in4;
+        out1 = out0 ^ in5;
+        out3 = out0 ^ in6 ^ in7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_FC(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in1 ^ in2;
+        tmp1 = in0 ^ in7;
+        out2 = tmp0 ^ tmp1 ^ in5;
+        out3 = tmp1 ^ in4;
+        tmp2 = out2 ^ in6;
+        out6 = tmp2 ^ in4;
+        out7 = tmp2 ^ in3;
+        out4 = out6 ^ in1 ^ in3;
+        tmp3 = out4 ^ in0;
+        out1 = tmp3 ^ in6;
+        out0 = tmp3 ^ in1 ^ in5;
+        out5 = tmp0 ^ out4;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_FD(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in5;
+        tmp1 = in1 ^ in7;
+        out2 = tmp0 ^ tmp1;
+        out6 = out2 ^ in2 ^ in4;
+        tmp2 = out6 ^ in0;
+        out1 = tmp2 ^ in3;
+        out0 = tmp0 ^ out1 ^ in6;
+        out5 = out0 ^ in2;
+        tmp3 = out5 ^ in1;
+        out3 = tmp3 ^ in6;
+        out7 = tmp2 ^ tmp3;
+        out4 = tmp1 ^ out7;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_FE(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3, tmp4;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        tmp0 = in0 ^ in2;
+        out2 = tmp0 ^ in5;
+        out3 = tmp0 ^ in4;
+        tmp1 = out3 ^ in6;
+        out4 = tmp1 ^ in5;
+        tmp2 = tmp1 ^ in1;
+        out6 = tmp2 ^ in7;
+        tmp3 = tmp2 ^ in0;
+        out0 = tmp3 ^ in3;
+        tmp4 = out0 ^ out4 ^ in7;
+        out5 = tmp4 ^ in6;
+        out7 = tmp4 ^ in2;
+        out1 = tmp3 ^ out5;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void
+gf8_muladd_FF(void *out, void *in)
+{
+    unsigned int i;
+    uint64_t *in_ptr = (uint64_t *)in;
+    uint64_t *out_ptr = (uint64_t *)out;
+
+    for (i = 0; i < WIDTH; i++) {
+        uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+        uint64_t tmp0, tmp1, tmp2, tmp3;
+
+        uint64_t in0 = out_ptr[0];
+        uint64_t in1 = out_ptr[WIDTH];
+        uint64_t in2 = out_ptr[WIDTH * 2];
+        uint64_t in3 = out_ptr[WIDTH * 3];
+        uint64_t in4 = out_ptr[WIDTH * 4];
+        uint64_t in5 = out_ptr[WIDTH * 5];
+        uint64_t in6 = out_ptr[WIDTH * 6];
+        uint64_t in7 = out_ptr[WIDTH * 7];
+
+        out2 = in0 ^ in5;
+        tmp0 = in4 ^ in7;
+        tmp1 = out2 ^ in2;
+        out4 = tmp1 ^ in6;
+        out7 = tmp1 ^ in1 ^ in3;
+        out1 = tmp0 ^ out7;
+        tmp2 = out1 ^ in5;
+        out6 = tmp2 ^ in3;
+        tmp3 = tmp2 ^ in7;
+        out0 = tmp3 ^ in6;
+        out3 = tmp3 ^ in1;
+        out5 = tmp0 ^ out0 ^ in2;
+
+        out_ptr[0] = out0 ^ in_ptr[0];
+        out_ptr[WIDTH] = out1 ^ in_ptr[WIDTH];
+        out_ptr[WIDTH * 2] = out2 ^ in_ptr[WIDTH * 2];
+        out_ptr[WIDTH * 3] = out3 ^ in_ptr[WIDTH * 3];
+        out_ptr[WIDTH * 4] = out4 ^ in_ptr[WIDTH * 4];
+        out_ptr[WIDTH * 5] = out5 ^ in_ptr[WIDTH * 5];
+        out_ptr[WIDTH * 6] = out6 ^ in_ptr[WIDTH * 6];
+        out_ptr[WIDTH * 7] = out7 ^ in_ptr[WIDTH * 7];
+
+        in_ptr++;
+        out_ptr++;
+    }
+}
+
+static void (*gf8_muladd[])(void *out, void *in) = {
+    gf8_muladd_00, gf8_muladd_01, gf8_muladd_02, gf8_muladd_03, gf8_muladd_04,
+    gf8_muladd_05, gf8_muladd_06, gf8_muladd_07, gf8_muladd_08, gf8_muladd_09,
+    gf8_muladd_0A, gf8_muladd_0B, gf8_muladd_0C, gf8_muladd_0D, gf8_muladd_0E,
+    gf8_muladd_0F, gf8_muladd_10, gf8_muladd_11, gf8_muladd_12, gf8_muladd_13,
+    gf8_muladd_14, gf8_muladd_15, gf8_muladd_16, gf8_muladd_17, gf8_muladd_18,
+    gf8_muladd_19, gf8_muladd_1A, gf8_muladd_1B, gf8_muladd_1C, gf8_muladd_1D,
+    gf8_muladd_1E, gf8_muladd_1F, gf8_muladd_20, gf8_muladd_21, gf8_muladd_22,
+    gf8_muladd_23, gf8_muladd_24, gf8_muladd_25, gf8_muladd_26, gf8_muladd_27,
+    gf8_muladd_28, gf8_muladd_29, gf8_muladd_2A, gf8_muladd_2B, gf8_muladd_2C,
+    gf8_muladd_2D, gf8_muladd_2E, gf8_muladd_2F, gf8_muladd_30, gf8_muladd_31,
+    gf8_muladd_32, gf8_muladd_33, gf8_muladd_34, gf8_muladd_35, gf8_muladd_36,
+    gf8_muladd_37, gf8_muladd_38, gf8_muladd_39, gf8_muladd_3A, gf8_muladd_3B,
+    gf8_muladd_3C, gf8_muladd_3D, gf8_muladd_3E, gf8_muladd_3F, gf8_muladd_40,
+    gf8_muladd_41, gf8_muladd_42, gf8_muladd_43, gf8_muladd_44, gf8_muladd_45,
+    gf8_muladd_46, gf8_muladd_47, gf8_muladd_48, gf8_muladd_49, gf8_muladd_4A,
+    gf8_muladd_4B, gf8_muladd_4C, gf8_muladd_4D, gf8_muladd_4E, gf8_muladd_4F,
+    gf8_muladd_50, gf8_muladd_51, gf8_muladd_52, gf8_muladd_53, gf8_muladd_54,
+    gf8_muladd_55, gf8_muladd_56, gf8_muladd_57, gf8_muladd_58, gf8_muladd_59,
+    gf8_muladd_5A, gf8_muladd_5B, gf8_muladd_5C, gf8_muladd_5D, gf8_muladd_5E,
+    gf8_muladd_5F, gf8_muladd_60, gf8_muladd_61, gf8_muladd_62, gf8_muladd_63,
+    gf8_muladd_64, gf8_muladd_65, gf8_muladd_66, gf8_muladd_67, gf8_muladd_68,
+    gf8_muladd_69, gf8_muladd_6A, gf8_muladd_6B, gf8_muladd_6C, gf8_muladd_6D,
+    gf8_muladd_6E, gf8_muladd_6F, gf8_muladd_70, gf8_muladd_71, gf8_muladd_72,
+    gf8_muladd_73, gf8_muladd_74, gf8_muladd_75, gf8_muladd_76, gf8_muladd_77,
+    gf8_muladd_78, gf8_muladd_79, gf8_muladd_7A, gf8_muladd_7B, gf8_muladd_7C,
+    gf8_muladd_7D, gf8_muladd_7E, gf8_muladd_7F, gf8_muladd_80, gf8_muladd_81,
+    gf8_muladd_82, gf8_muladd_83, gf8_muladd_84, gf8_muladd_85, gf8_muladd_86,
+    gf8_muladd_87, gf8_muladd_88, gf8_muladd_89, gf8_muladd_8A, gf8_muladd_8B,
+    gf8_muladd_8C, gf8_muladd_8D, gf8_muladd_8E, gf8_muladd_8F, gf8_muladd_90,
+    gf8_muladd_91, gf8_muladd_92, gf8_muladd_93, gf8_muladd_94, gf8_muladd_95,
+    gf8_muladd_96, gf8_muladd_97, gf8_muladd_98, gf8_muladd_99, gf8_muladd_9A,
+    gf8_muladd_9B, gf8_muladd_9C, gf8_muladd_9D, gf8_muladd_9E, gf8_muladd_9F,
+    gf8_muladd_A0, gf8_muladd_A1, gf8_muladd_A2, gf8_muladd_A3, gf8_muladd_A4,
+    gf8_muladd_A5, gf8_muladd_A6, gf8_muladd_A7, gf8_muladd_A8, gf8_muladd_A9,
+    gf8_muladd_AA, gf8_muladd_AB, gf8_muladd_AC, gf8_muladd_AD, gf8_muladd_AE,
+    gf8_muladd_AF, gf8_muladd_B0, gf8_muladd_B1, gf8_muladd_B2, gf8_muladd_B3,
+    gf8_muladd_B4, gf8_muladd_B5, gf8_muladd_B6, gf8_muladd_B7, gf8_muladd_B8,
+    gf8_muladd_B9, gf8_muladd_BA, gf8_muladd_BB, gf8_muladd_BC, gf8_muladd_BD,
+    gf8_muladd_BE, gf8_muladd_BF, gf8_muladd_C0, gf8_muladd_C1, gf8_muladd_C2,
+    gf8_muladd_C3, gf8_muladd_C4, gf8_muladd_C5, gf8_muladd_C6, gf8_muladd_C7,
+    gf8_muladd_C8, gf8_muladd_C9, gf8_muladd_CA, gf8_muladd_CB, gf8_muladd_CC,
+    gf8_muladd_CD, gf8_muladd_CE, gf8_muladd_CF, gf8_muladd_D0, gf8_muladd_D1,
+    gf8_muladd_D2, gf8_muladd_D3, gf8_muladd_D4, gf8_muladd_D5, gf8_muladd_D6,
+    gf8_muladd_D7, gf8_muladd_D8, gf8_muladd_D9, gf8_muladd_DA, gf8_muladd_DB,
+    gf8_muladd_DC, gf8_muladd_DD, gf8_muladd_DE, gf8_muladd_DF, gf8_muladd_E0,
+    gf8_muladd_E1, gf8_muladd_E2, gf8_muladd_E3, gf8_muladd_E4, gf8_muladd_E5,
+    gf8_muladd_E6, gf8_muladd_E7, gf8_muladd_E8, gf8_muladd_E9, gf8_muladd_EA,
+    gf8_muladd_EB, gf8_muladd_EC, gf8_muladd_ED, gf8_muladd_EE, gf8_muladd_EF,
+    gf8_muladd_F0, gf8_muladd_F1, gf8_muladd_F2, gf8_muladd_F3, gf8_muladd_F4,
+    gf8_muladd_F5, gf8_muladd_F6, gf8_muladd_F7, gf8_muladd_F8, gf8_muladd_F9,
+    gf8_muladd_FA, gf8_muladd_FB, gf8_muladd_FC, gf8_muladd_FD, gf8_muladd_FE,
+    gf8_muladd_FF};
+
+static uint64_t zero[EC_METHOD_WORD_SIZE * 8] = {
+    0,
+};
+
+void
+ec_code_c_prepare(ec_gf_t *gf, uint32_t *values, uint32_t count)
+{
+    uint32_t i, last, tmp;
+
+    last = 1;
+    for (i = count; i > 0; i--) {
+        if (values[i - 1] != 0) {
+            tmp = values[i - 1];
+            values[i - 1] = ec_gf_div(gf, tmp, last);
+            last = tmp;
+        }
+    }
+}
+
+void
+ec_code_c_linear(void *dst, void *src, uint64_t offset, uint32_t *values,
+                 uint32_t count)
+{
+    src += offset;
+    gf8_muladd_00(dst, src);
+    while (--count > 0) {
+        src += EC_METHOD_CHUNK_SIZE;
+        gf8_muladd[*values](dst, src);
+        values++;
+    }
+}
+
+void
+ec_code_c_interleaved(void *dst, void **src, uint64_t offset, uint32_t *values,
+                      uint32_t count)
+{
+    uint32_t i, last, tmp;
+
+    i = 0;
+    while ((last = *values++) == 0) {
+        i++;
+    }
+    gf8_muladd_00(dst, src[i++] + offset);
+    while (i < count) {
+        tmp = *values++;
+        if (tmp != 0) {
+            gf8_muladd[last](dst, src[i] + offset);
+            last = tmp;
+        }
+        i++;
+    }
+    gf8_muladd[last](dst, zero);
+}
diff --git a/xlators/cluster/ec/src/ec-code-c.h b/xlators/cluster/ec/src/ec-code-c.h
new file mode 100644
index 00000000000..42b5a064eb8
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code-c.h
@@ -0,0 +1,27 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_CODE_C_H__
+#define __EC_CODE_C_H__
+
+#include "ec-types.h"
+
+void
+ec_code_c_prepare(ec_gf_t *gf, uint32_t *values, uint32_t count);
+
+void
+ec_code_c_linear(void *dst, void *src, uint64_t offset, uint32_t *values,
+                 uint32_t count);
+
+void
+ec_code_c_interleaved(void *dst, void **src, uint64_t offset, uint32_t *values,
+                      uint32_t count);
+
+#endif /* __EC_CODE_C_H__ */
diff --git a/xlators/cluster/ec/src/ec-code-intel.c b/xlators/cluster/ec/src/ec-code-intel.c
new file mode 100644
index 00000000000..f1c4e13e321
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code-intel.c
@@ -0,0 +1,594 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <inttypes.h>
+#include <string.h>
+#include <errno.h>
+
+#include "ec-code-intel.h"
+
+static void
+ec_code_intel_init(ec_code_intel_t *intel)
+{
+    memset(intel, 0, sizeof(ec_code_intel_t));
+}
+
+static void
+ec_code_intel_prefix(ec_code_intel_t *intel, uint8_t prefix)
+{
+    intel->prefix.data[intel->prefix.bytes++] = prefix;
+}
+
+static void
+ec_code_intel_rex(ec_code_intel_t *intel, gf_boolean_t w)
+{
+    gf_boolean_t present = _gf_false;
+
+    if (w) {
+        intel->rex.w = 1;
+        present = _gf_true;
+    }
+    if (intel->modrm.present) {
+        if (intel->modrm.reg > 7) {
+            intel->modrm.reg &= 7;
+            intel->rex.r = 1;
+            present = _gf_true;
+        }
+        if (intel->sib.present) {
+            if (intel->sib.index > 7) {
+                intel->sib.index &= 7;
+                intel->rex.x = 1;
+                present = _gf_true;
+            }
+            if (intel->sib.base > 7) {
+                intel->sib.base &= 7;
+                intel->rex.b = 1;
+                present = _gf_true;
+            }
+        } else if (intel->modrm.rm > 7) {
+            intel->modrm.rm &= 7;
+            intel->rex.b = 1;
+            present = _gf_true;
+        }
+    } else if (intel->reg > 7) {
+        intel->reg &= 7;
+        intel->rex.b = 1;
+        present = _gf_true;
+    }
+    intel->rex.present = present;
+}
+
+static void
+ec_code_intel_vex(ec_code_intel_t *intel, gf_boolean_t w, gf_boolean_t l,
+                  ec_code_vex_opcode_t opcode, ec_code_vex_prefix_t prefix,
+                  uint32_t reg)
+{
+    ec_code_intel_rex(intel, w);
+    if (((intel->rex.w == 1) || (intel->rex.x == 0) || (intel->rex.b == 0)) ||
+        ((opcode != VEX_OPCODE_NONE) && (opcode != VEX_OPCODE_0F))) {
+        intel->rex.present = _gf_false;
+
+        intel->vex.bytes = 3;
+        intel->vex.data[0] = 0xC4;
+        intel->vex.data[1] = ((intel->rex.r << 7) | (intel->rex.x << 6) |
+                              (intel->rex.b << 5) | opcode) ^
+                             0xE0;
+        intel->vex.data[2] = (intel->rex.w << 7) | ((~reg & 0x0F) << 3) |
+                             (l ? 0x04 : 0x00) | prefix;
+    } else {
+        intel->vex.bytes = 2;
+        intel->vex.data[0] = 0xC5;
+        intel->vex.data[1] = (intel->rex.r << 7) | ((~reg & 0x0F) << 3) |
+                             (l ? 0x04 : 0x00) | prefix;
+    }
+}
+
+static void
+ec_code_intel_modrm_reg(ec_code_intel_t *intel, uint32_t rm, uint32_t reg)
+{
+    intel->modrm.present = _gf_true;
+    intel->modrm.mod = 3;
+    intel->modrm.rm = rm;
+    intel->modrm.reg = reg;
+}
+
+static void
+ec_code_intel_modrm_mem(ec_code_intel_t *intel, uint32_t reg,
+                        ec_code_intel_reg_t base, ec_code_intel_reg_t index,
+                        uint32_t scale, int32_t offset)
+{
+    if (index == REG_SP) {
+        intel->invalid = _gf_true;
+        return;
+    }
+    if ((index != REG_NULL) && (scale != 1) && (scale != 2) && (scale != 4) &&
+        (scale != 8)) {
+        intel->invalid = _gf_true;
+        return;
+    }
+    scale >>= 1;
+    if (scale == 4) {
+        scale = 3;
+    }
+
+    intel->modrm.present = _gf_true;
+    intel->modrm.reg = reg;
+
+    intel->offset.value = offset;
+    if ((offset == 0) && (base != REG_BP)) {
+        intel->modrm.mod = 0;
+        intel->offset.bytes = 0;
+    } else if ((offset >= -128) && (offset <= 127)) {
+        intel->modrm.mod = 1;
+        intel->offset.bytes = 1;
+    } else {
+        intel->modrm.mod = 2;
+        intel->offset.bytes = 4;
+    }
+
+    intel->modrm.rm = base;
+    if ((index != REG_NULL) || (base == REG_SP)) {
+        intel->modrm.rm = 4;
+        intel->sib.present = _gf_true;
+        intel->sib.index = index;
+        if (index == REG_NULL) {
+            intel->sib.index = 4;
+        }
+        intel->sib.scale = scale;
+        intel->sib.base = base;
+        if (base == REG_NULL) {
+            intel->sib.base = 5;
+            intel->modrm.mod = 0;
+            intel->offset.bytes = 4;
+        }
+    } else if (base == REG_NULL) {
+        intel->modrm.mod = 0;
+        intel->modrm.rm = 5;
+        intel->offset.bytes = 4;
+    }
+}
+
+static void
+ec_code_intel_op_1(ec_code_intel_t *intel, uint8_t opcode, uint32_t reg)
+{
+    intel->reg = reg;
+    intel->opcode.bytes = 1;
+    intel->opcode.data[0] = opcode;
+}
+
+static void
+ec_code_intel_op_2(ec_code_intel_t *intel, uint8_t opcode1, uint8_t opcode2,
+                   uint32_t reg)
+{
+    intel->reg = reg;
+    intel->opcode.bytes = 2;
+    intel->opcode.data[0] = opcode1;
+    intel->opcode.data[1] = opcode2;
+}
+
+static void
+ec_code_intel_immediate_1(ec_code_intel_t *intel, uint32_t value)
+{
+    intel->immediate.bytes = 1;
+    intel->immediate.value = value;
+}
+
+static void
+ec_code_intel_immediate_2(ec_code_intel_t *intel, uint32_t value)
+{
+    intel->immediate.bytes = 2;
+    intel->immediate.value = value;
+}
+
+static void
+ec_code_intel_immediate_4(ec_code_intel_t *intel, uint32_t value)
+{
+    intel->immediate.bytes = 4;
+    intel->immediate.value = value;
+}
+
+static void
+ec_code_intel_emit(ec_code_builder_t *builder, ec_code_intel_t *intel)
+{
+    uint8_t insn[15];
+    uint32_t i, count;
+
+    if (intel->invalid) {
+        ec_code_error(builder, EINVAL);
+        return;
+    }
+
+    count = 0;
+    for (i = 0; i < intel->prefix.bytes; i++) {
+        insn[count++] = intel->prefix.data[i];
+    }
+    for (i = 0; i < intel->vex.bytes; i++) {
+        insn[count++] = intel->vex.data[i];
+    }
+    if (intel->rex.present) {
+        insn[count++] = 0x40 | (intel->rex.w << 3) | (intel->rex.r << 2) |
+                        (intel->rex.x << 1) | (intel->rex.b << 0);
+    }
+    for (i = 0; i < intel->opcode.bytes; i++) {
+        insn[count++] = intel->opcode.data[i];
+    }
+    if (intel->modrm.present) {
+        insn[count++] = (intel->modrm.mod << 6) | (intel->modrm.reg << 3) |
+                        (intel->modrm.rm << 0);
+        if (intel->sib.present) {
+            insn[count++] = (intel->sib.scale << 6) | (intel->sib.index << 3) |
+                            (intel->sib.base << 0);
+        }
+    }
+    for (i = 0; i < intel->offset.bytes; i++) {
+        insn[count++] = intel->offset.data[i];
+    }
+    for (i = 0; i < intel->immediate.bytes; i++) {
+        insn[count++] = intel->immediate.data[i];
+    }
+
+    ec_code_emit(builder, insn, count);
+}
+
+void
+ec_code_intel_op_push_r(ec_code_builder_t *builder, ec_code_intel_reg_t reg)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_op_1(&intel, 0x50 | (reg & 7), reg);
+    ec_code_intel_rex(&intel, _gf_false);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_pop_r(ec_code_builder_t *builder, ec_code_intel_reg_t reg)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_op_1(&intel, 0x58 | (reg & 7), reg);
+    ec_code_intel_rex(&intel, _gf_false);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_ret(ec_code_builder_t *builder, uint32_t size)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    if (size == 0) {
+        ec_code_intel_op_1(&intel, 0xC3, 0);
+    } else {
+        ec_code_intel_immediate_2(&intel, size);
+        ec_code_intel_op_1(&intel, 0xC2, 0);
+    }
+    ec_code_intel_rex(&intel, _gf_false);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_mov_r2r(ec_code_builder_t *builder, ec_code_intel_reg_t src,
+                         ec_code_intel_reg_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_modrm_reg(&intel, dst, src);
+    ec_code_intel_op_1(&intel, 0x89, 0);
+    ec_code_intel_rex(&intel, _gf_true);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_mov_r2m(ec_code_builder_t *builder, ec_code_intel_reg_t src,
+                         ec_code_intel_reg_t base, ec_code_intel_reg_t index,
+                         uint32_t scale, int32_t offset)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_modrm_mem(&intel, src, base, index, scale, offset);
+    ec_code_intel_op_1(&intel, 0x89, 0);
+    ec_code_intel_rex(&intel, _gf_true);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_mov_m2r(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                         ec_code_intel_reg_t index, uint32_t scale,
+                         int32_t offset, ec_code_intel_reg_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_modrm_mem(&intel, dst, base, index, scale, offset);
+    ec_code_intel_op_1(&intel, 0x8B, 0);
+    ec_code_intel_rex(&intel, _gf_true);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_xor_r2r(ec_code_builder_t *builder, ec_code_intel_reg_t src,
+                         ec_code_intel_reg_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_modrm_reg(&intel, dst, src);
+    ec_code_intel_op_1(&intel, 0x31, 0);
+    ec_code_intel_rex(&intel, _gf_true);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_xor_m2r(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                         ec_code_intel_reg_t index, uint32_t scale,
+                         int32_t offset, ec_code_intel_reg_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_modrm_mem(&intel, dst, base, index, scale, offset);
+    ec_code_intel_op_1(&intel, 0x33, 0);
+    ec_code_intel_rex(&intel, _gf_true);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_add_i2r(ec_code_builder_t *builder, int32_t value,
+                         ec_code_intel_reg_t reg)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    if ((value >= -128) && (value < 128)) {
+        ec_code_intel_modrm_reg(&intel, reg, 0);
+        ec_code_intel_op_1(&intel, 0x83, 0);
+        ec_code_intel_immediate_1(&intel, value);
+    } else {
+        if (reg == REG_AX) {
+            ec_code_intel_op_1(&intel, 0x05, reg);
+        } else {
+            ec_code_intel_modrm_reg(&intel, reg, 0);
+            ec_code_intel_op_1(&intel, 0x81, 0);
+        }
+        ec_code_intel_immediate_4(&intel, value);
+    }
+    ec_code_intel_rex(&intel, _gf_true);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_test_i2r(ec_code_builder_t *builder, uint32_t value,
+                          ec_code_intel_reg_t reg)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    if (reg == REG_AX) {
+        ec_code_intel_op_1(&intel, 0xA9, reg);
+    } else {
+        ec_code_intel_modrm_reg(&intel, reg, 0);
+        ec_code_intel_op_1(&intel, 0xF7, 0);
+    }
+    ec_code_intel_immediate_4(&intel, value);
+    ec_code_intel_rex(&intel, _gf_true);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_jne(ec_code_builder_t *builder, uint32_t address)
+{
+    ec_code_intel_t intel;
+    int32_t rel;
+
+    ec_code_intel_init(&intel);
+
+    rel = address - builder->address - 2;
+    if ((rel >= -128) && (rel < 128)) {
+        ec_code_intel_op_1(&intel, 0x75, 0);
+        ec_code_intel_immediate_1(&intel, rel);
+    } else {
+        rel -= 4;
+        ec_code_intel_op_2(&intel, 0x0F, 0x85, 0);
+        ec_code_intel_immediate_4(&intel, rel);
+    }
+    ec_code_intel_rex(&intel, _gf_false);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_mov_sse2sse(ec_code_builder_t *builder, uint32_t src,
+                             uint32_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_prefix(&intel, 0x66);
+    ec_code_intel_modrm_reg(&intel, src, dst);
+    ec_code_intel_op_2(&intel, 0x0F, 0x6F, 0);
+    ec_code_intel_rex(&intel, _gf_false);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_mov_sse2m(ec_code_builder_t *builder, uint32_t src,
+                           ec_code_intel_reg_t base, ec_code_intel_reg_t index,
+                           uint32_t scale, int32_t offset)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_prefix(&intel, 0x66);
+    ec_code_intel_modrm_mem(&intel, src, base, index, scale, offset);
+    ec_code_intel_op_2(&intel, 0x0F, 0x7F, 0);
+    ec_code_intel_rex(&intel, _gf_false);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_mov_m2sse(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                           ec_code_intel_reg_t index, uint32_t scale,
+                           int32_t offset, uint32_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_prefix(&intel, 0x66);
+    ec_code_intel_modrm_mem(&intel, dst, base, index, scale, offset);
+    ec_code_intel_op_2(&intel, 0x0F, 0x6F, 0);
+    ec_code_intel_rex(&intel, _gf_false);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_xor_sse2sse(ec_code_builder_t *builder, uint32_t src,
+                             uint32_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_prefix(&intel, 0x66);
+    ec_code_intel_modrm_reg(&intel, src, dst);
+    ec_code_intel_op_2(&intel, 0x0F, 0xEF, 0);
+    ec_code_intel_rex(&intel, _gf_false);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_xor_m2sse(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                           ec_code_intel_reg_t index, uint32_t scale,
+                           int32_t offset, uint32_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_prefix(&intel, 0x66);
+    ec_code_intel_modrm_mem(&intel, dst, base, index, scale, offset);
+    ec_code_intel_op_2(&intel, 0x0F, 0xEF, 0);
+    ec_code_intel_rex(&intel, _gf_false);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_mov_avx2avx(ec_code_builder_t *builder, uint32_t src,
+                             uint32_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_modrm_reg(&intel, src, dst);
+    ec_code_intel_op_1(&intel, 0x6F, 0);
+    ec_code_intel_vex(&intel, _gf_false, _gf_true, VEX_OPCODE_0F, VEX_PREFIX_66,
+                      VEX_REG_NONE);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_mov_avx2m(ec_code_builder_t *builder, uint32_t src,
+                           ec_code_intel_reg_t base, ec_code_intel_reg_t index,
+                           uint32_t scale, int32_t offset)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_modrm_mem(&intel, src, base, index, scale, offset);
+    ec_code_intel_op_1(&intel, 0x7F, 0);
+    ec_code_intel_vex(&intel, _gf_false, _gf_true, VEX_OPCODE_0F, VEX_PREFIX_66,
+                      VEX_REG_NONE);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_mov_m2avx(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                           ec_code_intel_reg_t index, uint32_t scale,
+                           int32_t offset, uint32_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_modrm_mem(&intel, dst, base, index, scale, offset);
+    ec_code_intel_op_1(&intel, 0x6F, 0);
+    ec_code_intel_vex(&intel, _gf_false, _gf_true, VEX_OPCODE_0F, VEX_PREFIX_66,
+                      VEX_REG_NONE);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_xor_avx2avx(ec_code_builder_t *builder, uint32_t src,
+                             uint32_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_modrm_reg(&intel, src, dst);
+    ec_code_intel_op_1(&intel, 0xEF, 0);
+    ec_code_intel_vex(&intel, _gf_false, _gf_true, VEX_OPCODE_0F, VEX_PREFIX_66,
+                      dst);
+
+    ec_code_intel_emit(builder, &intel);
+}
+
+void
+ec_code_intel_op_xor_m2avx(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                           ec_code_intel_reg_t index, uint32_t scale,
+                           int32_t offset, uint32_t dst)
+{
+    ec_code_intel_t intel;
+
+    ec_code_intel_init(&intel);
+
+    ec_code_intel_modrm_mem(&intel, dst, base, index, scale, offset);
+    ec_code_intel_op_1(&intel, 0xEF, 0);
+    ec_code_intel_vex(&intel, _gf_false, _gf_true, VEX_OPCODE_0F, VEX_PREFIX_66,
+                      dst);
+
+    ec_code_intel_emit(builder, &intel);
+}
diff --git a/xlators/cluster/ec/src/ec-code-intel.h b/xlators/cluster/ec/src/ec-code-intel.h
new file mode 100644
index 00000000000..3fa4a174765
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code-intel.h
@@ -0,0 +1,191 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_CODE_INTEL_H__
+#define __EC_CODE_INTEL_H__
+
+#include "ec-code.h"
+
+#define VEX_REG_NONE 0
+
+enum _ec_code_intel_reg;
+typedef enum _ec_code_intel_reg ec_code_intel_reg_t;
+
+enum _ec_code_vex_prefix;
+typedef enum _ec_code_vex_prefix ec_code_vex_prefix_t;
+
+enum _ec_code_vex_opcode;
+typedef enum _ec_code_vex_opcode ec_code_vex_opcode_t;
+
+struct _ec_code_intel_buffer;
+typedef struct _ec_code_intel_buffer ec_code_intel_buffer_t;
+
+struct _ec_code_intel_sib;
+typedef struct _ec_code_intel_sib ec_code_intel_sib_t;
+
+struct _ec_code_intel_modrm;
+typedef struct _ec_code_intel_modrm ec_code_intel_modrm_t;
+
+struct _ec_code_intel_rex;
+typedef struct _ec_code_intel_rex ec_code_intel_rex_t;
+
+struct _ec_code_intel;
+typedef struct _ec_code_intel ec_code_intel_t;
+
+enum _ec_code_intel_reg {
+    REG_NULL = -1,
+    REG_AX,
+    REG_CX,
+    REG_DX,
+    REG_BX,
+    REG_SP,
+    REG_BP,
+    REG_SI,
+    REG_DI,
+    REG_8,
+    REG_9,
+    REG_10,
+    REG_11,
+    REG_12,
+    REG_13,
+    REG_14,
+    REG_15
+};
+
+enum _ec_code_vex_prefix {
+    VEX_PREFIX_NONE = 0,
+    VEX_PREFIX_66,
+    VEX_PREFIX_F3,
+    VEX_PREFIX_F2
+};
+
+enum _ec_code_vex_opcode {
+    VEX_OPCODE_NONE = 0,
+    VEX_OPCODE_0F,
+    VEX_OPCODE_0F_38,
+    VEX_OPCODE_0F_3A
+};
+
+struct _ec_code_intel_buffer {
+    uint32_t bytes;
+    union {
+        uint8_t data[4];
+        uint32_t value;
+    };
+};
+
+struct _ec_code_intel_sib {
+    gf_boolean_t present;
+    uint32_t base;
+    uint32_t index;
+    uint32_t scale;
+};
+
+struct _ec_code_intel_modrm {
+    gf_boolean_t present;
+    uint32_t mod;
+    uint32_t rm;
+    uint32_t reg;
+};
+
+struct _ec_code_intel_rex {
+    gf_boolean_t present;
+    uint32_t w;
+    uint32_t r;
+    uint32_t x;
+    uint32_t b;
+};
+
+struct _ec_code_intel {
+    gf_boolean_t invalid;
+    ec_code_intel_buffer_t prefix;
+    ec_code_intel_buffer_t opcode;
+    ec_code_intel_buffer_t offset;
+    ec_code_intel_buffer_t immediate;
+    ec_code_intel_buffer_t vex;
+    ec_code_intel_rex_t rex;
+    ec_code_intel_modrm_t modrm;
+    ec_code_intel_sib_t sib;
+    uint32_t reg;
+};
+
+void
+ec_code_intel_op_push_r(ec_code_builder_t *builder, ec_code_intel_reg_t reg);
+void
+ec_code_intel_op_pop_r(ec_code_builder_t *builder, ec_code_intel_reg_t reg);
+void
+ec_code_intel_op_ret(ec_code_builder_t *builder, uint32_t size);
+
+void
+ec_code_intel_op_mov_r2r(ec_code_builder_t *builder, ec_code_intel_reg_t src,
+                         ec_code_intel_reg_t dst);
+void
+ec_code_intel_op_mov_r2m(ec_code_builder_t *builder, ec_code_intel_reg_t src,
+                         ec_code_intel_reg_t base, ec_code_intel_reg_t index,
+                         uint32_t scale, int32_t offset);
+void
+ec_code_intel_op_mov_m2r(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                         ec_code_intel_reg_t index, uint32_t scale,
+                         int32_t offset, ec_code_intel_reg_t dst);
+void
+ec_code_intel_op_xor_r2r(ec_code_builder_t *builder, ec_code_intel_reg_t src,
+                         ec_code_intel_reg_t dst);
+void
+ec_code_intel_op_xor_m2r(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                         ec_code_intel_reg_t index, uint32_t scale,
+                         int32_t offset, ec_code_intel_reg_t dst);
+void
+ec_code_intel_op_add_i2r(ec_code_builder_t *builder, int32_t value,
+                         ec_code_intel_reg_t reg);
+void
+ec_code_intel_op_test_i2r(ec_code_builder_t *builder, uint32_t value,
+                          ec_code_intel_reg_t reg);
+void
+ec_code_intel_op_jne(ec_code_builder_t *builder, uint32_t address);
+
+void
+ec_code_intel_op_mov_sse2sse(ec_code_builder_t *builder, uint32_t src,
+                             uint32_t dst);
+void
+ec_code_intel_op_mov_sse2m(ec_code_builder_t *builder, uint32_t src,
+                           ec_code_intel_reg_t base, ec_code_intel_reg_t index,
+                           uint32_t scale, int32_t offset);
+void
+ec_code_intel_op_mov_m2sse(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                           ec_code_intel_reg_t index, uint32_t scale,
+                           int32_t offset, uint32_t dst);
+void
+ec_code_intel_op_xor_sse2sse(ec_code_builder_t *builder, uint32_t src,
+                             uint32_t dst);
+void
+ec_code_intel_op_xor_m2sse(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                           ec_code_intel_reg_t index, uint32_t scale,
+                           int32_t offset, uint32_t dst);
+
+void
+ec_code_intel_op_mov_avx2avx(ec_code_builder_t *builder, uint32_t src,
+                             uint32_t dst);
+void
+ec_code_intel_op_mov_avx2m(ec_code_builder_t *builder, uint32_t src,
+                           ec_code_intel_reg_t base, ec_code_intel_reg_t index,
+                           uint32_t scale, int32_t offset);
+void
+ec_code_intel_op_mov_m2avx(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                           ec_code_intel_reg_t index, uint32_t scale,
+                           int32_t offset, uint32_t dst);
+void
+ec_code_intel_op_xor_avx2avx(ec_code_builder_t *builder, uint32_t src,
+                             uint32_t dst);
+void
+ec_code_intel_op_xor_m2avx(ec_code_builder_t *builder, ec_code_intel_reg_t base,
+                           ec_code_intel_reg_t index, uint32_t scale,
+                           int32_t offset, uint32_t dst);
+
+#endif /* __EC_CODE_INTEL_H__ */
diff --git a/xlators/cluster/ec/src/ec-code-sse.c b/xlators/cluster/ec/src/ec-code-sse.c
new file mode 100644
index 00000000000..e11e7ff8400
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code-sse.c
@@ -0,0 +1,101 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <errno.h>
+
+#include "ec-code-intel.h"
+
+static void
+ec_code_sse_prolog(ec_code_builder_t *builder)
+{
+    builder->loop = builder->address;
+}
+
+static void
+ec_code_sse_epilog(ec_code_builder_t *builder)
+{
+    ec_code_intel_op_add_i2r(builder, 16, REG_DX);
+    ec_code_intel_op_add_i2r(builder, 16, REG_DI);
+    ec_code_intel_op_test_i2r(builder, builder->width - 1, REG_DX);
+    ec_code_intel_op_jne(builder, builder->loop);
+
+    ec_code_intel_op_ret(builder, 0);
+}
+
+static void
+ec_code_sse_load(ec_code_builder_t *builder, uint32_t dst, uint32_t idx,
+                 uint32_t bit)
+{
+    if (builder->linear) {
+        ec_code_intel_op_mov_m2sse(
+            builder, REG_SI, REG_DX, 1,
+            idx * builder->width * builder->bits + bit * builder->width, dst);
+    } else {
+        if (builder->base != idx) {
+            ec_code_intel_op_mov_m2r(builder, REG_SI, REG_NULL, 0, idx * 8,
+                                     REG_AX);
+            builder->base = idx;
+        }
+        ec_code_intel_op_mov_m2sse(builder, REG_AX, REG_DX, 1,
+                                   bit * builder->width, dst);
+    }
+}
+
+static void
+ec_code_sse_store(ec_code_builder_t *builder, uint32_t src, uint32_t bit)
+{
+    ec_code_intel_op_mov_sse2m(builder, src, REG_DI, REG_NULL, 0,
+                               bit * builder->width);
+}
+
+static void
+ec_code_sse_copy(ec_code_builder_t *builder, uint32_t dst, uint32_t src)
+{
+    ec_code_intel_op_mov_sse2sse(builder, src, dst);
+}
+
+static void
+ec_code_sse_xor2(ec_code_builder_t *builder, uint32_t dst, uint32_t src)
+{
+    ec_code_intel_op_xor_sse2sse(builder, src, dst);
+}
+
+static void
+ec_code_sse_xorm(ec_code_builder_t *builder, uint32_t dst, uint32_t idx,
+                 uint32_t bit)
+{
+    if (builder->linear) {
+        ec_code_intel_op_xor_m2sse(
+            builder, REG_SI, REG_DX, 1,
+            idx * builder->width * builder->bits + bit * builder->width, dst);
+    } else {
+        if (builder->base != idx) {
+            ec_code_intel_op_mov_m2r(builder, REG_SI, REG_NULL, 0, idx * 8,
+                                     REG_AX);
+            builder->base = idx;
+        }
+        ec_code_intel_op_xor_m2sse(builder, REG_AX, REG_DX, 1,
+                                   bit * builder->width, dst);
+    }
+}
+
+static char *ec_code_sse_needed_flags[] = {"sse2", NULL};
+
+ec_code_gen_t ec_code_gen_sse = {.name = "sse",
+                                 .flags = ec_code_sse_needed_flags,
+                                 .width = 16,
+                                 .prolog = ec_code_sse_prolog,
+                                 .epilog = ec_code_sse_epilog,
+                                 .load = ec_code_sse_load,
+                                 .store = ec_code_sse_store,
+                                 .copy = ec_code_sse_copy,
+                                 .xor2 = ec_code_sse_xor2,
+                                 .xor3 = NULL,
+                                 .xorm = ec_code_sse_xorm};
diff --git a/xlators/cluster/ec/src/ec-code-sse.h b/xlators/cluster/ec/src/ec-code-sse.h
new file mode 100644
index 00000000000..f1acbcf894b
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code-sse.h
@@ -0,0 +1,18 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_CODE_SSE_H__
+#define __EC_CODE_SSE_H__
+
+#include "ec-code.h"
+
+extern ec_code_gen_t ec_code_gen_sse;
+
+#endif /* __EC_CODE_SSE_H__ */
diff --git a/xlators/cluster/ec/src/ec-code-x64.c b/xlators/cluster/ec/src/ec-code-x64.c
new file mode 100644
index 00000000000..26565b4493f
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code-x64.c
@@ -0,0 +1,144 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <errno.h>
+
+#include "ec-code-intel.h"
+
+static ec_code_intel_reg_t ec_code_x64_regmap[] = {
+    REG_AX, REG_CX, REG_BP, REG_8,  REG_9, REG_10,
+    REG_11, REG_12, REG_13, REG_14, REG_15};
+
+static void
+ec_code_x64_prolog(ec_code_builder_t *builder)
+{
+    uint32_t i;
+
+    ec_code_intel_op_push_r(builder, REG_BP);
+    if (!builder->linear) {
+        ec_code_intel_op_push_r(builder, REG_BX);
+    }
+    if (builder->regs > 11) {
+        ec_code_error(builder, EINVAL);
+        return;
+    }
+    for (i = 7; i < builder->regs; i++) {
+        ec_code_intel_op_push_r(builder, ec_code_x64_regmap[i]);
+    }
+
+    builder->loop = builder->address;
+}
+
+static void
+ec_code_x64_epilog(ec_code_builder_t *builder)
+{
+    uint32_t i;
+
+    ec_code_intel_op_add_i2r(builder, 8, REG_DX);
+    ec_code_intel_op_add_i2r(builder, 8, REG_DI);
+    ec_code_intel_op_test_i2r(builder, builder->width - 1, REG_DX);
+    ec_code_intel_op_jne(builder, builder->loop);
+
+    if (builder->regs > 11) {
+        ec_code_error(builder, EINVAL);
+        return;
+    }
+    for (i = builder->regs; i > 7; i--) {
+        ec_code_intel_op_pop_r(builder, ec_code_x64_regmap[i - 1]);
+    }
+    if (!builder->linear) {
+        ec_code_intel_op_pop_r(builder, REG_BX);
+    }
+    ec_code_intel_op_pop_r(builder, REG_BP);
+    ec_code_intel_op_ret(builder, 0);
+}
+
+static void
+ec_code_x64_load(ec_code_builder_t *builder, uint32_t dst, uint32_t idx,
+                 uint32_t bit)
+{
+    dst = ec_code_x64_regmap[dst];
+
+    if (builder->linear) {
+        ec_code_intel_op_mov_m2r(
+            builder, REG_SI, REG_DX, 1,
+            idx * builder->width * builder->bits + bit * builder->width, dst);
+    } else {
+        if (builder->base != idx) {
+            ec_code_intel_op_mov_m2r(builder, REG_SI, REG_NULL, 0, idx * 8,
+                                     REG_BX);
+            builder->base = idx;
+        }
+        ec_code_intel_op_mov_m2r(builder, REG_BX, REG_DX, 1,
+                                 bit * builder->width, dst);
+    }
+}
+
+static void
+ec_code_x64_store(ec_code_builder_t *builder, uint32_t src, uint32_t bit)
+{
+    src = ec_code_x64_regmap[src];
+
+    ec_code_intel_op_mov_r2m(builder, src, REG_DI, REG_NULL, 0,
+                             bit * builder->width);
+}
+
+static void
+ec_code_x64_copy(ec_code_builder_t *builder, uint32_t dst, uint32_t src)
+{
+    dst = ec_code_x64_regmap[dst];
+    src = ec_code_x64_regmap[src];
+
+    ec_code_intel_op_mov_r2r(builder, src, dst);
+}
+
+static void
+ec_code_x64_xor2(ec_code_builder_t *builder, uint32_t dst, uint32_t src)
+{
+    dst = ec_code_x64_regmap[dst];
+    src = ec_code_x64_regmap[src];
+
+    ec_code_intel_op_xor_r2r(builder, src, dst);
+}
+
+static void
+ec_code_x64_xorm(ec_code_builder_t *builder, uint32_t dst, uint32_t idx,
+                 uint32_t bit)
+{
+    dst = ec_code_x64_regmap[dst];
+
+    if (builder->linear) {
+        ec_code_intel_op_xor_m2r(
+            builder, REG_SI, REG_DX, 1,
+            idx * builder->width * builder->bits + bit * builder->width, dst);
+    } else {
+        if (builder->base != idx) {
+            ec_code_intel_op_mov_m2r(builder, REG_SI, REG_NULL, 0, idx * 8,
+                                     REG_BX);
+            builder->base = idx;
+        }
+        ec_code_intel_op_xor_m2r(builder, REG_BX, REG_DX, 1,
+                                 bit * builder->width, dst);
+    }
+}
+
+static char *ec_code_x64_needed_flags[] = {NULL};
+
+ec_code_gen_t ec_code_gen_x64 = {.name = "x64",
+                                 .flags = ec_code_x64_needed_flags,
+                                 .width = sizeof(uint64_t),
+                                 .prolog = ec_code_x64_prolog,
+                                 .epilog = ec_code_x64_epilog,
+                                 .load = ec_code_x64_load,
+                                 .store = ec_code_x64_store,
+                                 .copy = ec_code_x64_copy,
+                                 .xor2 = ec_code_x64_xor2,
+                                 .xor3 = NULL,
+                                 .xorm = ec_code_x64_xorm};
diff --git a/xlators/cluster/ec/src/ec-code-x64.h b/xlators/cluster/ec/src/ec-code-x64.h
new file mode 100644
index 00000000000..bd8174e4bf5
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code-x64.h
@@ -0,0 +1,18 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_CODE_X64_H__
+#define __EC_CODE_X64_H__
+
+#include "ec-code.h"
+
+extern ec_code_gen_t ec_code_gen_x64;
+
+#endif /* __EC_CODE_X64_H__ */
diff --git a/xlators/cluster/ec/src/ec-code.c b/xlators/cluster/ec/src/ec-code.c
new file mode 100644
index 00000000000..03162ae05a9
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code.c
@@ -0,0 +1,1060 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <string.h>
+#include <sys/mman.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <ctype.h>
+
+#include <glusterfs/syscall.h>
+
+#include "ec-mem-types.h"
+#include "ec-code.h"
+#include "ec-messages.h"
+#include "ec-code-c.h"
+#include "ec-helpers.h"
+
+#ifdef USE_EC_DYNAMIC_X64
+#include "ec-code-x64.h"
+#endif
+
+#ifdef USE_EC_DYNAMIC_SSE
+#include "ec-code-sse.h"
+#endif
+
+#ifdef USE_EC_DYNAMIC_AVX
+#include "ec-code-avx.h"
+#endif
+
+#define EC_CODE_SIZE (1024 * 64)
+#define EC_CODE_ALIGN 4096
+
+#define EC_CODE_CHUNK_MIN_SIZE 512
+
+#define EC_PROC_BUFFER_SIZE 4096
+
+#define PROC_CPUINFO "/proc/cpuinfo"
+
+struct _ec_code_proc;
+typedef struct _ec_code_proc ec_code_proc_t;
+
+struct _ec_code_proc {
+    int32_t fd;
+    gf_boolean_t eof;
+    gf_boolean_t error;
+    gf_boolean_t skip;
+    ssize_t size;
+    ssize_t pos;
+    char buffer[EC_PROC_BUFFER_SIZE];
+};
+
+static ec_code_gen_t *ec_code_gen_table[] = {
+#ifdef USE_EC_DYNAMIC_AVX
+    &ec_code_gen_avx,
+#endif
+#ifdef USE_EC_DYNAMIC_SSE
+    &ec_code_gen_sse,
+#endif
+#ifdef USE_EC_DYNAMIC_X64
+    &ec_code_gen_x64,
+#endif
+    NULL};
+
+static void
+ec_code_arg_set(ec_code_arg_t *arg, uint32_t value)
+{
+    arg->value = value;
+}
+
+static void
+ec_code_arg_assign(ec_code_builder_t *builder, ec_code_op_t *op,
+                   ec_code_arg_t *arg, uint32_t reg)
+{
+    arg->value = reg;
+
+    if (builder->regs <= reg) {
+        builder->regs = reg + 1;
+    }
+}
+
+static void
+ec_code_arg_use(ec_code_builder_t *builder, ec_code_op_t *op,
+                ec_code_arg_t *arg, uint32_t reg)
+{
+    arg->value = reg;
+}
+
+static void
+ec_code_arg_update(ec_code_builder_t *builder, ec_code_op_t *op,
+                   ec_code_arg_t *arg, uint32_t reg)
+{
+    arg->value = reg;
+}
+
+static ec_code_op_t *
+ec_code_op_next(ec_code_builder_t *builder)
+{
+    ec_code_op_t *op;
+
+    op = &builder->ops[builder->count++];
+    memset(op, 0, sizeof(ec_code_op_t));
+
+    return op;
+}
+
+static void
+ec_code_load(ec_code_builder_t *builder, uint32_t bit, uint32_t offset)
+{
+    ec_code_op_t *op;
+
+    op = ec_code_op_next(builder);
+
+    op->op = EC_GF_OP_LOAD;
+    ec_code_arg_assign(builder, op, &op->arg1, builder->map[bit]);
+    ec_code_arg_set(&op->arg2, offset);
+    ec_code_arg_set(&op->arg3, bit);
+}
+
+static void
+ec_code_store(ec_code_builder_t *builder, uint32_t reg, uint32_t bit)
+{
+    ec_code_op_t *op;
+
+    op = ec_code_op_next(builder);
+
+    op->op = EC_GF_OP_STORE;
+    ec_code_arg_use(builder, op, &op->arg1, builder->map[reg]);
+    ec_code_arg_set(&op->arg2, 0);
+    ec_code_arg_set(&op->arg3, bit);
+}
+
+static void
+ec_code_copy(ec_code_builder_t *builder, uint32_t dst, uint32_t src)
+{
+    ec_code_op_t *op;
+
+    op = ec_code_op_next(builder);
+
+    op->op = EC_GF_OP_COPY;
+    ec_code_arg_assign(builder, op, &op->arg1, builder->map[dst]);
+    ec_code_arg_use(builder, op, &op->arg2, builder->map[src]);
+    ec_code_arg_set(&op->arg3, 0);
+}
+
+static void
+ec_code_xor2(ec_code_builder_t *builder, uint32_t dst, uint32_t src)
+{
+    ec_code_op_t *op;
+
+    op = ec_code_op_next(builder);
+
+    op->op = EC_GF_OP_XOR2;
+    ec_code_arg_update(builder, op, &op->arg1, builder->map[dst]);
+    ec_code_arg_use(builder, op, &op->arg2, builder->map[src]);
+    ec_code_arg_set(&op->arg3, 0);
+}
+
+static void
+ec_code_xor3(ec_code_builder_t *builder, uint32_t dst, uint32_t src1,
+             uint32_t src2)
+{
+    ec_code_op_t *op;
+
+    if (builder->code->gen->xor3 == NULL) {
+        ec_code_copy(builder, dst, src1);
+        ec_code_xor2(builder, dst, src2);
+
+        return;
+    }
+
+    op = ec_code_op_next(builder);
+
+    op->op = EC_GF_OP_XOR3;
+    ec_code_arg_assign(builder, op, &op->arg1, builder->map[dst]);
+    ec_code_arg_use(builder, op, &op->arg2, builder->map[src1]);
+    ec_code_arg_use(builder, op, &op->arg3, builder->map[src2]);
+}
+
+static void
+ec_code_xorm(ec_code_builder_t *builder, uint32_t bit, uint32_t offset)
+{
+    ec_code_op_t *op;
+
+    op = ec_code_op_next(builder);
+
+    op->op = EC_GF_OP_XORM;
+    ec_code_arg_update(builder, op, &op->arg1, builder->map[bit]);
+    ec_code_arg_set(&op->arg2, offset);
+    ec_code_arg_set(&op->arg3, bit);
+}
+
+static void
+ec_code_dup(ec_code_builder_t *builder, ec_gf_op_t *op)
+{
+    switch (op->op) {
+        case EC_GF_OP_COPY:
+            ec_code_copy(builder, op->arg1, op->arg2);
+            break;
+        case EC_GF_OP_XOR2:
+            ec_code_xor2(builder, op->arg1, op->arg2);
+            break;
+        case EC_GF_OP_XOR3:
+            ec_code_xor3(builder, op->arg1, op->arg2, op->arg3);
+            break;
+        default:
+            break;
+    }
+}
+
+static void
+ec_code_gf_load(ec_code_builder_t *builder, uint32_t offset)
+{
+    uint32_t i;
+
+    for (i = 0; i < builder->code->gf->bits; i++) {
+        ec_code_load(builder, i, offset);
+    }
+}
+
+static void
+ec_code_gf_load_xor(ec_code_builder_t *builder, uint32_t offset)
+{
+    uint32_t i;
+
+    for (i = 0; i < builder->code->gf->bits; i++) {
+        ec_code_xorm(builder, i, offset);
+    }
+}
+
+static void
+ec_code_gf_store(ec_code_builder_t *builder)
+{
+    uint32_t i;
+
+    for (i = 0; i < builder->code->gf->bits; i++) {
+        ec_code_store(builder, i, i);
+    }
+}
+
+static void
+ec_code_gf_clear(ec_code_builder_t *builder)
+{
+    uint32_t i;
+
+    ec_code_xor2(builder, 0, 0);
+    for (i = 0; i < builder->code->gf->bits; i++) {
+        ec_code_store(builder, 0, i);
+    }
+}
+
+static void
+ec_code_gf_mul(ec_code_builder_t *builder, uint32_t value)
+{
+    ec_gf_mul_t *mul;
+    ec_gf_op_t *op;
+    uint32_t map[EC_GF_MAX_REGS];
+    int32_t i;
+
+    mul = builder->code->gf->table[value];
+    for (op = mul->ops; op->op != EC_GF_OP_END; op++) {
+        ec_code_dup(builder, op);
+    }
+
+    for (i = 0; i < mul->regs; i++) {
+        map[i] = builder->map[mul->map[i]];
+    }
+    memcpy(builder->map, map, sizeof(uint32_t) * mul->regs);
+}
+
+static ec_code_builder_t *
+ec_code_prepare(ec_code_t *code, uint32_t count, uint32_t width,
+                gf_boolean_t linear)
+{
+    ec_code_builder_t *builder;
+    uint32_t i;
+
+    count *= code->gf->bits + code->gf->max_ops;
+    count += code->gf->bits;
+    builder = GF_MALLOC(
+        sizeof(ec_code_builder_t) + sizeof(ec_code_op_t) * count,
+        ec_mt_ec_code_builder_t);
+    if (builder == NULL) {
+        return EC_ERR(ENOMEM);
+    }
+
+    builder->address = 0;
+    builder->code = code;
+    builder->size = 0;
+    builder->count = 0;
+    builder->regs = 0;
+    builder->error = 0;
+    builder->bits = code->gf->bits;
+    builder->width = width;
+    builder->data = NULL;
+    builder->linear = linear;
+    builder->base = -1;
+
+    for (i = 0; i < EC_GF_MAX_REGS; i++) {
+        builder->map[i] = i;
+    }
+
+    return builder;
+}
+
+static size_t
+ec_code_space_size(void)
+{
+    return (sizeof(ec_code_space_t) + 15) & ~15;
+}
+
+static size_t
+ec_code_chunk_size(void)
+{
+    return (sizeof(ec_code_chunk_t) + 15) & ~15;
+}
+
+static ec_code_chunk_t *
+ec_code_chunk_from_space(ec_code_space_t *space)
+{
+    return (ec_code_chunk_t *)((uintptr_t)space + ec_code_space_size());
+}
+
+static void *
+ec_code_to_executable(ec_code_space_t *space, void *addr)
+{
+    return (void *)((uintptr_t)addr - (uintptr_t)space +
+                    (uintptr_t)space->exec);
+}
+
+static void *
+ec_code_from_executable(ec_code_space_t *space, void *addr)
+{
+    return (void *)((uintptr_t)addr - (uintptr_t)space->exec +
+                    (uintptr_t)space);
+}
+
+static void *
+ec_code_func_from_chunk(ec_code_chunk_t *chunk, void **exec)
+{
+    void *addr;
+
+    addr = (void *)((uintptr_t)chunk + ec_code_chunk_size());
+
+    *exec = ec_code_to_executable(chunk->space, addr);
+
+    return addr;
+}
+
+static ec_code_chunk_t *
+ec_code_chunk_from_func(ec_code_func_linear_t func)
+{
+    ec_code_chunk_t *chunk;
+
+    chunk = (ec_code_chunk_t *)((uintptr_t)func - ec_code_chunk_size());
+
+    return ec_code_from_executable(chunk->space, chunk);
+}
+
+static ec_code_chunk_t *
+ec_code_chunk_split(ec_code_chunk_t *chunk, size_t size)
+{
+    ec_code_chunk_t *extra;
+    ssize_t avail;
+
+    avail = chunk->size - size - ec_code_chunk_size();
+    if (avail > 0) {
+        extra = (ec_code_chunk_t *)((uintptr_t)chunk + chunk->size - avail);
+        extra->space = chunk->space;
+        extra->size = avail;
+        list_add(&extra->list, &chunk->list);
+        chunk->size = size;
+    }
+    list_del_init(&chunk->list);
+
+    return chunk;
+}
+
+static gf_boolean_t
+ec_code_chunk_touch(ec_code_chunk_t *prev, ec_code_chunk_t *next)
+{
+    uintptr_t end;
+
+    end = (uintptr_t)prev + ec_code_chunk_size() + prev->size;
+    return (end == (uintptr_t)next);
+}
+
+static ec_code_space_t *
+ec_code_space_create(ec_code_t *code, size_t size)
+{
+    char path[] = GLUSTERFS_LIBEXECDIR "/ec-code-dynamic.XXXXXX";
+    ec_code_space_t *space;
+    void *exec;
+    int32_t fd, err;
+
+    /* We need to create memory areas to store the generated dynamic code.
+     * Obviously these areas need to be written to be able to create the
+     * code and they also need to be executable to execute it.
+     *
+     * However it's a bad practice to have a memory region that is both
+     * writable *and* executable. In fact, selinux forbids this and causes
+     * attempts to do so to fail (unless specifically configured).
+     *
+     * To solve the problem we'll use two distinct memory areas mapped to
+     * the same physical storage. One of the memory areas will have write
+     * permission, and the other will have execute permission. Both areas
+     * will have the same contents. The physical storage will be a regular
+     * file that will be mmapped to both areas.
+     */
+
+    /* We need to create a temporary file as the backend storage for the
+     * memory mapped areas. */
+    /* coverity[secure_temp] mkstemp uses 0600 as the mode and is safe */
+    fd = mkstemp(path);
+    if (fd < 0) {
+        err = errno;
+        gf_msg(THIS->name, GF_LOG_ERROR, err, EC_MSG_DYN_CREATE_FAILED,
+               "Unable to create a temporary file for the ec dynamic "
+               "code");
+        space = EC_ERR(err);
+        goto done;
+    }
+    /* Once created we don't need to keep it in the file system. It will
+     * still exist until we close the last file descriptor or unmap the
+     * memory areas bound to the file. */
+    sys_unlink(path);
+
+    size = (size + EC_CODE_ALIGN - 1) & ~(EC_CODE_ALIGN - 1);
+    if (sys_ftruncate(fd, size) < 0) {
+        err = errno;
+        gf_msg(THIS->name, GF_LOG_ERROR, err, EC_MSG_DYN_CREATE_FAILED,
+               "Unable to resize the file for the ec dynamic code");
+        space = EC_ERR(err);
+        goto done_close;
+    }
+
+    /* This creates an executable memory area to be able to run the
+     * generated fragments of code. */
+    exec = mmap(NULL, size, PROT_READ | PROT_EXEC, MAP_SHARED, fd, 0);
+    if (exec == MAP_FAILED) {
+        err = errno;
+        gf_msg(THIS->name, GF_LOG_ERROR, err, EC_MSG_DYN_CREATE_FAILED,
+               "Unable to map the executable area for the ec dynamic "
+               "code");
+        space = EC_ERR(err);
+        goto done_close;
+    }
+    /* It's not important to check the return value of mlock(). If it fails
+     * everything will continue to work normally. */
+    mlock(exec, size);
+
+    /* This maps a read/write memory area to be able to create the dynamici
+     * code. */
+    space = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
+    if (space == MAP_FAILED) {
+        err = errno;
+        gf_msg(THIS->name, GF_LOG_ERROR, err, EC_MSG_DYN_CREATE_FAILED,
+               "Unable to map the writable area for the ec dynamic "
+               "code");
+        space = EC_ERR(err);
+
+        munmap(exec, size);
+
+        goto done_close;
+    }
+
+    space->exec = exec;
+    space->size = size;
+    space->code = code;
+    list_add_tail(&space->list, &code->spaces);
+    INIT_LIST_HEAD(&space->chunks);
+
+done_close:
+    /* If everything has succeeded, we already have the memory areas
+     * mapped. We don't need the file descriptor anymore because the
+     * backend storage will be there until the mmap()'d regions are
+     * unmapped. */
+    sys_close(fd);
+done:
+    return space;
+}
+
+static void
+ec_code_space_destroy(ec_code_space_t *space)
+{
+    list_del_init(&space->list);
+
+    munmap(space->exec, space->size);
+    munmap(space, space->size);
+}
+
+static void
+ec_code_chunk_merge(ec_code_chunk_t *chunk)
+{
+    ec_code_chunk_t *item, *tmp;
+
+    list_for_each_entry_safe(item, tmp, &chunk->space->chunks, list)
+    {
+        if ((uintptr_t)item > (uintptr_t)chunk) {
+            list_add_tail(&chunk->list, &item->list);
+            if (ec_code_chunk_touch(chunk, item)) {
+                chunk->size += item->size + ec_code_chunk_size();
+                list_del_init(&item->list);
+            }
+
+            goto check;
+        }
+        if (ec_code_chunk_touch(item, chunk)) {
+            item->size += chunk->size + ec_code_chunk_size();
+            list_del_init(&item->list);
+            chunk = item;
+        }
+    }
+    list_add_tail(&chunk->list, &chunk->space->chunks);
+
+check:
+    if (chunk->size ==
+        chunk->space->size - ec_code_space_size() - ec_code_chunk_size()) {
+        ec_code_space_destroy(chunk->space);
+    }
+}
+
+static ec_code_chunk_t *
+ec_code_space_alloc(ec_code_t *code, size_t size)
+{
+    ec_code_space_t *space;
+    ec_code_chunk_t *chunk;
+    size_t map_size;
+
+    /* To minimize fragmentation, we only allocate chunks of sizes multiples
+     * of EC_CODE_CHUNK_MIN_SIZE. */
+    size = ((size + ec_code_chunk_size() + EC_CODE_CHUNK_MIN_SIZE - 1) &
+            ~(EC_CODE_CHUNK_MIN_SIZE - 1)) -
+           ec_code_chunk_size();
+    list_for_each_entry(space, &code->spaces, list)
+    {
+        list_for_each_entry(chunk, &space->chunks, list)
+        {
+            if (chunk->size >= size) {
+                goto out;
+            }
+        }
+    }
+
+    map_size = EC_CODE_SIZE - ec_code_space_size() - ec_code_chunk_size();
+    if (map_size < size) {
+        map_size = size;
+    }
+    space = ec_code_space_create(code, map_size);
+    if (EC_IS_ERR(space)) {
+        return (ec_code_chunk_t *)space;
+    }
+
+    chunk = ec_code_chunk_from_space(space);
+    chunk->size = map_size - ec_code_space_size() - ec_code_chunk_size();
+    list_add(&chunk->list, &space->chunks);
+
+out:
+    chunk->space = space;
+
+    return ec_code_chunk_split(chunk, size);
+}
+
+static ec_code_chunk_t *
+ec_code_alloc(ec_code_t *code, uint32_t size)
+{
+    ec_code_chunk_t *chunk;
+
+    LOCK(&code->lock);
+
+    chunk = ec_code_space_alloc(code, size);
+
+    UNLOCK(&code->lock);
+
+    return chunk;
+}
+
+static void
+ec_code_free(ec_code_chunk_t *chunk)
+{
+    gf_lock_t *lock;
+
+    lock = &chunk->space->code->lock;
+    LOCK(lock);
+
+    ec_code_chunk_merge(chunk);
+
+    UNLOCK(lock);
+}
+
+static int32_t
+ec_code_write(ec_code_builder_t *builder)
+{
+    ec_code_gen_t *gen;
+    ec_code_op_t *op;
+    uint32_t i;
+
+    builder->error = 0;
+    builder->size = 0;
+    builder->address = 0;
+    builder->base = -1;
+
+    gen = builder->code->gen;
+    gen->prolog(builder);
+    for (i = 0; i < builder->count; i++) {
+        op = &builder->ops[i];
+        switch (op->op) {
+            case EC_GF_OP_LOAD:
+                gen->load(builder, op->arg1.value, op->arg2.value,
+                          op->arg3.value);
+                break;
+            case EC_GF_OP_STORE:
+                gen->store(builder, op->arg1.value, op->arg3.value);
+                break;
+            case EC_GF_OP_COPY:
+                gen->copy(builder, op->arg1.value, op->arg2.value);
+                break;
+            case EC_GF_OP_XOR2:
+                gen->xor2(builder, op->arg1.value, op->arg2.value);
+                break;
+            case EC_GF_OP_XOR3:
+                gen->xor3(builder, op->arg1.value, op->arg2.value,
+                          op->arg3.value);
+                break;
+            case EC_GF_OP_XORM:
+                gen->xorm(builder, op->arg1.value, op->arg2.value,
+                          op->arg3.value);
+                break;
+            default:
+                break;
+        }
+    }
+    gen->epilog(builder);
+
+    return builder->error;
+}
+
+static void *
+ec_code_compile(ec_code_builder_t *builder)
+{
+    ec_code_chunk_t *chunk;
+    void *func;
+    int32_t err;
+
+    err = ec_code_write(builder);
+    if (err != 0) {
+        return EC_ERR(err);
+    }
+
+    chunk = ec_code_alloc(builder->code, builder->size);
+    if (EC_IS_ERR(chunk)) {
+        return chunk;
+    }
+    builder->data = ec_code_func_from_chunk(chunk, &func);
+
+    err = ec_code_write(builder);
+    if (err != 0) {
+        ec_code_free(chunk);
+
+        return EC_ERR(err);
+    }
+
+    GF_FREE(builder);
+
+    return func;
+}
+
+ec_code_t *
+ec_code_create(ec_gf_t *gf, ec_code_gen_t *gen)
+{
+    ec_code_t *code;
+
+    code = GF_MALLOC(sizeof(ec_code_t), ec_mt_ec_code_t);
+    if (code == NULL) {
+        return EC_ERR(ENOMEM);
+    }
+    memset(code, 0, sizeof(ec_code_t));
+    INIT_LIST_HEAD(&code->spaces);
+    LOCK_INIT(&code->lock);
+
+    code->gf = gf;
+    code->gen = gen;
+
+    return code;
+}
+
+void
+ec_code_destroy(ec_code_t *code)
+{
+    if (!list_empty(&code->spaces)) {
+    }
+
+    LOCK_DESTROY(&code->lock);
+
+    GF_FREE(code);
+}
+
+static uint32_t
+ec_code_value_next(uint32_t *values, uint32_t count, uint32_t *offset)
+{
+    uint32_t i, next;
+
+    next = 0;
+    for (i = *offset + 1; i < count; i++) {
+        next = values[i];
+        if (next != 0) {
+            break;
+        }
+    }
+    *offset = i;
+
+    return next;
+}
+
+static void *
+ec_code_build_dynamic(ec_code_t *code, uint32_t width, uint32_t *values,
+                      uint32_t count, gf_boolean_t linear)
+{
+    ec_code_builder_t *builder;
+    uint32_t offset, val, next;
+
+    builder = ec_code_prepare(code, count, width, linear);
+    if (EC_IS_ERR(builder)) {
+        return builder;
+    }
+
+    offset = -1;
+    next = ec_code_value_next(values, count, &offset);
+    if (next != 0) {
+        ec_code_gf_load(builder, offset);
+        do {
+            val = next;
+            next = ec_code_value_next(values, count, &offset);
+            if (next != 0) {
+                ec_code_gf_mul(builder, ec_gf_div(code->gf, val, next));
+                ec_code_gf_load_xor(builder, offset);
+            }
+        } while (next != 0);
+        ec_code_gf_mul(builder, val);
+        ec_code_gf_store(builder);
+    } else {
+        ec_code_gf_clear(builder);
+    }
+
+    return ec_code_compile(builder);
+}
+
+static void *
+ec_code_build(ec_code_t *code, uint32_t width, uint32_t *values, uint32_t count,
+              gf_boolean_t linear)
+{
+    void *func;
+
+    if (code->gen != NULL) {
+        func = ec_code_build_dynamic(code, width, values, count, linear);
+        if (!EC_IS_ERR(func)) {
+            return func;
+        }
+
+        gf_msg_debug(THIS->name, GF_LOG_DEBUG,
+                     "Unable to generate dynamic code. Falling back "
+                     "to precompiled code");
+
+        /* The dynamic code generation shouldn't fail in normal
+         * conditions, but if it fails at some point, it's very
+         * probable that it will fail again, so we completely disable
+         * dynamic code generation. */
+        code->gen = NULL;
+    }
+
+    ec_code_c_prepare(code->gf, values, count);
+
+    if (linear) {
+        return ec_code_c_linear;
+    }
+
+    return ec_code_c_interleaved;
+}
+
+ec_code_func_linear_t
+ec_code_build_linear(ec_code_t *code, uint32_t width, uint32_t *values,
+                     uint32_t count)
+{
+    return (ec_code_func_linear_t)ec_code_build(code, width, values, count,
+                                                _gf_true);
+}
+
+ec_code_func_interleaved_t
+ec_code_build_interleaved(ec_code_t *code, uint32_t width, uint32_t *values,
+                          uint32_t count)
+{
+    return (ec_code_func_interleaved_t)ec_code_build(code, width, values, count,
+                                                     _gf_false);
+}
+
+void
+ec_code_release(ec_code_t *code, ec_code_func_t *func)
+{
+    if ((func->linear != ec_code_c_linear) &&
+        (func->interleaved != ec_code_c_interleaved)) {
+        ec_code_free(ec_code_chunk_from_func(func->linear));
+    }
+}
+
+void
+ec_code_error(ec_code_builder_t *builder, int32_t error)
+{
+    if (builder->error == 0) {
+        gf_msg(THIS->name, GF_LOG_ERROR, error, EC_MSG_DYN_CODEGEN_FAILED,
+               "Failed to generate dynamic code");
+        builder->error = error;
+    }
+}
+
+void
+ec_code_emit(ec_code_builder_t *builder, uint8_t *bytes, uint32_t count)
+{
+    if (builder->error != 0) {
+        return;
+    }
+
+    if (builder->data != NULL) {
+        memcpy(builder->data + builder->size, bytes, count);
+    }
+
+    builder->size += count;
+    builder->address += count;
+}
+
+static char *
+ec_code_proc_trim_left(char *text, ssize_t *length)
+{
+    ssize_t len;
+
+    for (len = *length; (len > 0) && isspace(*text); len--) {
+        text++;
+    }
+    *length = len;
+
+    return text;
+}
+
+static char *
+ec_code_proc_trim_right(char *text, ssize_t *length, char sep)
+{
+    char *last;
+    ssize_t len;
+
+    len = *length;
+
+    last = text;
+    for (len = *length; (len > 0) && (*text != sep); len--) {
+        if (!isspace(*text)) {
+            last = text + 1;
+        }
+        text++;
+    }
+    *last = 0;
+    *length = len;
+
+    return text;
+}
+
+static char *
+ec_code_proc_line_parse(ec_code_proc_t *file, ssize_t *length)
+{
+    char *text, *end;
+    ssize_t len;
+
+    len = file->size - file->pos;
+    text = ec_code_proc_trim_left(file->buffer + file->pos, &len);
+    end = ec_code_proc_trim_right(text, &len, '\n');
+    if (len == 0) {
+        if (!file->eof) {
+            if (text == file->buffer) {
+                file->size = file->pos = 0;
+                file->skip = _gf_true;
+            } else {
+                file->size = file->pos = end - text;
+                memmove(file->buffer, text, file->pos + 1);
+            }
+            len = sys_read(file->fd, file->buffer + file->pos,
+                           sizeof(file->buffer) - file->pos - 1);
+            if (len > 0) {
+                file->size += len;
+            }
+            file->error = len < 0;
+            file->eof = len <= 0;
+
+            return NULL;
+        }
+        file->size = file->pos = 0;
+    } else {
+        file->pos = end - file->buffer + 1;
+    }
+
+    *length = end - text;
+
+    if (file->skip) {
+        file->skip = _gf_false;
+        text = NULL;
+    }
+
+    return text;
+}
+
+static char *
+ec_code_proc_line(ec_code_proc_t *file, ssize_t *length)
+{
+    char *text;
+
+    text = NULL;
+    while (!file->eof) {
+        text = ec_code_proc_line_parse(file, length);
+        if (text != NULL) {
+            break;
+        }
+    }
+
+    return text;
+}
+
+static char *
+ec_code_proc_split(char *text, ssize_t *length, char sep)
+{
+    text = ec_code_proc_trim_right(text, length, sep);
+    if (*length == 0) {
+        return NULL;
+    }
+    (*length)--;
+    text++;
+
+    return ec_code_proc_trim_left(text, length);
+}
+
+static uint32_t
+ec_code_cpu_check(uint32_t idx, char *list, uint32_t count)
+{
+    ec_code_gen_t *gen;
+    char **ptr;
+    char *table[count + 1];
+    uint32_t i;
+
+    for (i = 0; i < count; i++) {
+        table[i] = list;
+        list += strlen(list) + 1;
+    }
+
+    gen = ec_code_gen_table[idx];
+    while (gen != NULL) {
+        for (ptr = gen->flags; *ptr != NULL; ptr++) {
+            for (i = 0; i < count; i++) {
+                if (strcmp(*ptr, table[i]) == 0) {
+                    break;
+                }
+            }
+            if (i >= count) {
+                gen = ec_code_gen_table[++idx];
+                break;
+            }
+        }
+        if (*ptr == NULL) {
+            break;
+        }
+    }
+
+    return idx;
+}
+
+ec_code_gen_t *
+ec_code_detect(xlator_t *xl, const char *def)
+{
+    ec_code_proc_t file;
+    ec_code_gen_t *gen = NULL;
+    char *line, *data, *list;
+    ssize_t length;
+    uint32_t count, base, select;
+
+    if (strcmp(def, "none") == 0) {
+        gf_msg(xl->name, GF_LOG_INFO, 0, EC_MSG_EXTENSION_NONE,
+               "Not using any cpu extensions");
+
+        return NULL;
+    }
+
+    file.fd = sys_open(PROC_CPUINFO, O_RDONLY, 0);
+    if (file.fd < 0) {
+        goto out;
+    }
+    file.size = file.pos = 0;
+    file.eof = file.error = file.skip = _gf_false;
+
+    select = 0;
+    if (strcmp(def, "auto") != 0) {
+        while (ec_code_gen_table[select] != NULL) {
+            if (strcmp(ec_code_gen_table[select]->name, def) == 0) {
+                break;
+            }
+            select++;
+        }
+        if (ec_code_gen_table[select] == NULL) {
+            gf_msg(xl->name, GF_LOG_WARNING, EINVAL, EC_MSG_EXTENSION_UNKNOWN,
+                   "CPU extension '%s' is not known. Not using any cpu "
+                   "extensions",
+                   def);
+
+            return NULL;
+        }
+    } else {
+        def = NULL;
+    }
+
+    while ((line = ec_code_proc_line(&file, &length)) != NULL) {
+        data = ec_code_proc_split(line, &length, ':');
+        if ((data != NULL) && (strcmp(line, "flags") == 0)) {
+            list = data;
+            count = 0;
+            while ((data != NULL) && (*data != 0)) {
+                count++;
+                data = ec_code_proc_split(data, &length, ' ');
+            }
+            base = select;
+            select = ec_code_cpu_check(select, list, count);
+            if ((base != select) && (def != NULL)) {
+                gf_msg(xl->name, GF_LOG_WARNING, ENOTSUP,
+                       EC_MSG_EXTENSION_UNSUPPORTED,
+                       "CPU extension '%s' is not supported", def);
+                def = NULL;
+            }
+        }
+    }
+
+    if (file.error) {
+        gf_msg(xl->name, GF_LOG_WARNING, 0, EC_MSG_EXTENSION_FAILED,
+               "Unable to determine supported CPU extensions. Not using any "
+               "cpu extensions");
+
+        gen = NULL;
+    } else {
+        gen = ec_code_gen_table[select];
+        if (gen == NULL) {
+            gf_msg(xl->name, GF_LOG_INFO, 0, EC_MSG_EXTENSION_NONE,
+                   "Not using any cpu extensions");
+        } else {
+            gf_msg(xl->name, GF_LOG_INFO, 0, EC_MSG_EXTENSION,
+                   "Using '%s' CPU extensions", gen->name);
+        }
+    }
+
+    sys_close(file.fd);
+
+out:
+    return gen;
+}
diff --git a/xlators/cluster/ec/src/ec-code.h b/xlators/cluster/ec/src/ec-code.h
new file mode 100644
index 00000000000..75fb35d93e3
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-code.h
@@ -0,0 +1,44 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_CODE_H__
+#define __EC_CODE_H__
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/list.h>
+
+#include "ec-types.h"
+#include "ec-galois.h"
+
+ec_code_gen_t *
+ec_code_detect(xlator_t *xl, const char *def);
+
+ec_code_t *
+ec_code_create(ec_gf_t *gf, ec_code_gen_t *gen);
+
+void
+ec_code_destroy(ec_code_t *code);
+
+ec_code_func_linear_t
+ec_code_build_linear(ec_code_t *code, uint32_t width, uint32_t *values,
+                     uint32_t count);
+ec_code_func_interleaved_t
+ec_code_build_interleaved(ec_code_t *code, uint32_t width, uint32_t *values,
+                          uint32_t count);
+void
+ec_code_release(ec_code_t *code, ec_code_func_t *func);
+
+void
+ec_code_error(ec_code_builder_t *builder, int32_t error);
+
+void
+ec_code_emit(ec_code_builder_t *builder, uint8_t *bytes, uint32_t count);
+
+#endif /* __EC_CODE_H__ */
diff --git a/xlators/cluster/ec/src/ec-combine.c b/xlators/cluster/ec/src/ec-combine.c
new file mode 100644
index 00000000000..703a30e2485
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-combine.c
@@ -0,0 +1,995 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <fnmatch.h>
+
+#include "libxlator.h"
+#include <glusterfs/byte-order.h>
+
+#include "ec-types.h"
+#include "ec-helpers.h"
+#include "ec-common.h"
+#include "ec-combine.h"
+#include "ec-messages.h"
+#include <glusterfs/quota-common-utils.h>
+
+#define EC_QUOTA_PREFIX "trusted.glusterfs.quota."
+
+#define EC_MISSING_DATA ((data_t *)1ULL)
+
+struct _ec_dict_info;
+typedef struct _ec_dict_info ec_dict_info_t;
+
+struct _ec_dict_combine;
+typedef struct _ec_dict_combine ec_dict_combine_t;
+
+struct _ec_dict_info {
+    dict_t *dict;
+    int32_t count;
+};
+
+struct _ec_dict_combine {
+    ec_cbk_data_t *cbk;
+    int32_t which;
+};
+
+int32_t
+ec_combine_write(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    int valid = 0;
+
+    if (!fop || !dst || !src)
+        return 0;
+
+    switch (fop->id) {
+        case GF_FOP_REMOVEXATTR:
+        case GF_FOP_FREMOVEXATTR:
+        case GF_FOP_SETXATTR:
+        case GF_FOP_FSETXATTR:
+            return 1;
+
+        case GF_FOP_SYMLINK:
+        case GF_FOP_LINK:
+        case GF_FOP_CREATE:
+        case GF_FOP_MKNOD:
+        case GF_FOP_MKDIR:
+            valid = 3;
+            break;
+        case GF_FOP_UNLINK:
+        case GF_FOP_RMDIR:
+        case GF_FOP_SETATTR:
+        case GF_FOP_FSETATTR:
+        case GF_FOP_TRUNCATE:
+        case GF_FOP_FTRUNCATE:
+        case GF_FOP_WRITE:
+        case GF_FOP_FALLOCATE:
+        case GF_FOP_DISCARD:
+        case GF_FOP_ZEROFILL:
+            valid = 2;
+            break;
+        case GF_FOP_RENAME:
+            valid = 5;
+            break;
+        default:
+            gf_msg_callingfn(fop->xl->name, GF_LOG_WARNING, EINVAL,
+                             EC_MSG_INVALID_FOP, "Invalid fop %d", fop->id);
+            return 0;
+            break;
+    }
+
+    if (!ec_iatt_combine(fop, dst->iatt, src->iatt, valid)) {
+        gf_msg(fop->xl->name, GF_LOG_NOTICE, 0, EC_MSG_IATT_MISMATCH,
+               "Mismatching iatt in "
+               "answers of '%s'",
+               gf_fop_list[fop->id]);
+        return 0;
+    }
+    return 1;
+}
+
+void
+ec_iatt_time_merge(int64_t *dst_sec, uint32_t *dst_nsec, int64_t src_sec,
+                   uint32_t src_nsec)
+{
+    if ((*dst_sec < src_sec) ||
+        ((*dst_sec == src_sec) && (*dst_nsec < src_nsec))) {
+        *dst_sec = src_sec;
+        *dst_nsec = src_nsec;
+    }
+}
+
+static gf_boolean_t
+ec_iatt_is_trusted(ec_fop_data_t *fop, struct iatt *iatt)
+{
+    uint64_t ino;
+    int32_t i;
+
+    /* Only the top level fop will have fop->locks filled. */
+    while (fop->parent != NULL) {
+        fop = fop->parent;
+    }
+
+    /* Lookups are special requests always done without locks taken but they
+     * require to be able to identify differences between bricks. Special
+     * handling of these differences is already done in lookup specific code
+     * so we shouldn't ignore any difference here and consider all iatt
+     * structures as trusted. */
+    if (fop->id == GF_FOP_LOOKUP) {
+        return _gf_true;
+    }
+
+    /* Check if the iatt references an inode locked by the current fop */
+    for (i = 0; i < fop->lock_count; i++) {
+        ino = gfid_to_ino(fop->locks[i].lock->loc.inode->gfid);
+        if (iatt->ia_ino == ino) {
+            return _gf_true;
+        }
+    }
+
+    return _gf_false;
+}
+
+int32_t
+ec_iatt_combine(ec_fop_data_t *fop, struct iatt *dst, struct iatt *src,
+                int32_t count)
+{
+    int32_t i;
+    gf_boolean_t failed = _gf_false;
+
+    for (i = 0; i < count; i++) {
+        /* Check for basic fields. These fields must be equal always, even if
+         * the inode is not locked because in these cases the parent inode
+         * will be locked and differences in these fields require changes in
+         * the parent directory. */
+        if ((dst[i].ia_ino != src[i].ia_ino) ||
+            (((dst[i].ia_type == IA_IFBLK) || (dst[i].ia_type == IA_IFCHR)) &&
+             (dst[i].ia_rdev != src[i].ia_rdev)) ||
+            (gf_uuid_compare(dst[i].ia_gfid, src[i].ia_gfid) != 0)) {
+            failed = _gf_true;
+        }
+        /* Check for not so stable fields. These fields can change if the
+         * inode is not locked. */
+        if (!failed && ((dst[i].ia_uid != src[i].ia_uid) ||
+                        (dst[i].ia_gid != src[i].ia_gid) ||
+                        (st_mode_from_ia(dst[i].ia_prot, dst[i].ia_type) !=
+                         st_mode_from_ia(src[i].ia_prot, src[i].ia_type)))) {
+            if (ec_iatt_is_trusted(fop, dst)) {
+                /* If the iatt contains information from an inode that is
+                 * locked, these differences are real problems, so we need to
+                 * report them. Otherwise we ignore them and don't care which
+                 * data is returned. */
+                failed = _gf_true;
+            } else {
+                gf_msg_debug(fop->xl->name, 0,
+                             "Ignoring iatt differences because inode is not "
+                             "locked");
+            }
+        }
+        if (failed) {
+            gf_msg(fop->xl->name, GF_LOG_WARNING, 0, EC_MSG_IATT_COMBINE_FAIL,
+                   "Failed to combine iatt (inode: %" PRIu64 "-%" PRIu64
+                   ", "
+                   "links: %u-%u, uid: %u-%u, gid: %u-%u, "
+                   "rdev: %" PRIu64 "-%" PRIu64 ", size: %" PRIu64 "-%" PRIu64
+                   ", "
+                   "mode: %o-%o), %s",
+                   dst[i].ia_ino, src[i].ia_ino, dst[i].ia_nlink,
+                   src[i].ia_nlink, dst[i].ia_uid, src[i].ia_uid, dst[i].ia_gid,
+                   src[i].ia_gid, dst[i].ia_rdev, src[i].ia_rdev,
+                   dst[i].ia_size, src[i].ia_size,
+                   st_mode_from_ia(dst[i].ia_prot, dst[i].ia_type),
+                   st_mode_from_ia(src[i].ia_prot, dst[i].ia_type),
+                   ec_msg_str(fop));
+
+            return 0;
+        }
+    }
+
+    while (count-- > 0) {
+        dst[count].ia_blocks += src[count].ia_blocks;
+        if (dst[count].ia_blksize < src[count].ia_blksize) {
+            dst[count].ia_blksize = src[count].ia_blksize;
+        }
+
+        ec_iatt_time_merge(&dst[count].ia_atime, &dst[count].ia_atime_nsec,
+                           src[count].ia_atime, src[count].ia_atime_nsec);
+        ec_iatt_time_merge(&dst[count].ia_mtime, &dst[count].ia_mtime_nsec,
+                           src[count].ia_mtime, src[count].ia_mtime_nsec);
+        ec_iatt_time_merge(&dst[count].ia_ctime, &dst[count].ia_ctime_nsec,
+                           src[count].ia_ctime, src[count].ia_ctime_nsec);
+    }
+
+    return 1;
+}
+
+void
+ec_iatt_rebuild(ec_t *ec, struct iatt *iatt, int32_t count, int32_t answers)
+{
+    uint64_t blocks;
+
+    while (count-- > 0) {
+        blocks = iatt[count].ia_blocks * ec->fragments + answers - 1;
+        blocks /= answers;
+        iatt[count].ia_blocks = blocks;
+    }
+}
+
+gf_boolean_t
+ec_xattr_match(dict_t *dict, char *key, data_t *value, void *arg)
+{
+    if ((fnmatch(GF_XATTR_STIME_PATTERN, key, 0) == 0) ||
+        (strcmp(key, GET_LINK_COUNT) == 0) ||
+        (strcmp(key, GLUSTERFS_INODELK_COUNT) == 0) ||
+        (strcmp(key, GLUSTERFS_ENTRYLK_COUNT) == 0) ||
+        (strcmp(key, GLUSTERFS_OPEN_FD_COUNT) == 0)) {
+        return _gf_false;
+    }
+
+    return _gf_true;
+}
+
+gf_boolean_t
+ec_value_ignore(char *key)
+{
+    if ((strcmp(key, GF_CONTENT_KEY) == 0) ||
+        (strcmp(key, GF_XATTR_PATHINFO_KEY) == 0) ||
+        (strcmp(key, GF_XATTR_USER_PATHINFO_KEY) == 0) ||
+        (strcmp(key, GF_XATTR_LOCKINFO_KEY) == 0) ||
+        (strcmp(key, GLUSTERFS_OPEN_FD_COUNT) == 0) ||
+        (strcmp(key, GLUSTERFS_INODELK_COUNT) == 0) ||
+        (strcmp(key, GLUSTERFS_ENTRYLK_COUNT) == 0) ||
+        (strncmp(key, GF_XATTR_CLRLK_CMD, SLEN(GF_XATTR_CLRLK_CMD)) == 0) ||
+        (strcmp(key, DHT_IATT_IN_XDATA_KEY) == 0) ||
+        (strncmp(key, EC_QUOTA_PREFIX, SLEN(EC_QUOTA_PREFIX)) == 0) ||
+        (fnmatch(MARKER_XATTR_PREFIX ".*." XTIME, key, 0) == 0) ||
+        (fnmatch(GF_XATTR_MARKER_KEY ".*", key, 0) == 0) ||
+        (XATTR_IS_NODE_UUID(key))) {
+        return _gf_true;
+    }
+
+    return _gf_false;
+}
+
+int32_t
+ec_dict_compare(dict_t *dict1, dict_t *dict2)
+{
+    if (are_dicts_equal(dict1, dict2, ec_xattr_match, ec_value_ignore))
+        return 1;
+    return 0;
+}
+
+static uint32_t
+ec_dict_list(data_t **list, ec_cbk_data_t *cbk, int32_t which, char *key,
+             gf_boolean_t global)
+{
+    ec_t *ec = cbk->fop->xl->private;
+    ec_cbk_data_t *ans = NULL;
+    dict_t *dict = NULL;
+    data_t *data;
+    uint32_t count;
+    int32_t i;
+
+    for (i = 0; i < ec->nodes; i++) {
+        /* We initialize the list with EC_MISSING_DATA if we are
+         * returning a global list or the current subvolume belongs
+         * to the group of the accepted answer. Note that if some
+         * subvolume is known to be down before issuing the request,
+         * we won't have any answer from it, so we set here the
+         * appropriate default value. */
+        if (global || ((cbk->mask & (1ULL << i)) != 0)) {
+            list[i] = EC_MISSING_DATA;
+        } else {
+            list[i] = NULL;
+        }
+    }
+
+    count = 0;
+    list_for_each_entry(ans, &cbk->fop->answer_list, answer_list)
+    {
+        if (global || ((cbk->mask & ans->mask) != 0)) {
+            dict = (which == EC_COMBINE_XDATA) ? ans->xdata : ans->dict;
+            data = dict_get(dict, key);
+            if (data != NULL) {
+                list[ans->idx] = data;
+                count++;
+            }
+        }
+    }
+
+    return count;
+}
+
+int32_t
+ec_concat_prepare(xlator_t *xl, char **str, char **sep, char **post,
+                  const char *fmt, va_list args)
+{
+    char *tmp;
+    int32_t len;
+
+    len = gf_vasprintf(str, fmt, args);
+    if (len < 0) {
+        return -ENOMEM;
+    }
+
+    tmp = strchr(*str, '{');
+    if (tmp == NULL) {
+        goto out;
+    }
+    *tmp++ = 0;
+    *sep = tmp;
+    tmp = strchr(tmp, '}');
+    if (tmp == NULL) {
+        goto out;
+    }
+    *tmp++ = 0;
+    *post = tmp;
+
+    return 0;
+
+out:
+    gf_msg(xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_INVALID_FORMAT,
+           "Invalid concat format");
+
+    GF_FREE(*str);
+
+    return -EINVAL;
+}
+
+static int32_t
+ec_dict_data_concat(ec_cbk_data_t *cbk, int32_t which, char *key, char *new_key,
+                    const char *def, gf_boolean_t global, const char *fmt, ...)
+{
+    ec_t *ec = cbk->fop->xl->private;
+    data_t *data[ec->nodes];
+    char *str = NULL, *pre = NULL, *sep, *post;
+    dict_t *dict;
+    va_list args;
+    int32_t i, num, len, deflen, prelen, postlen, seplen, tmp;
+    int32_t err;
+
+    ec_dict_list(data, cbk, which, key, global);
+
+    va_start(args, fmt);
+    err = ec_concat_prepare(cbk->fop->xl, &pre, &sep, &post, fmt, args);
+    va_end(args);
+
+    if (err != 0) {
+        return err;
+    }
+
+    prelen = strlen(pre);
+    seplen = strlen(sep);
+    postlen = strlen(post);
+
+    deflen = 0;
+    if (def != NULL) {
+        deflen = strlen(def);
+    }
+
+    len = prelen + postlen + 1;
+    num = -1;
+    for (i = 0; i < ec->nodes; i++) {
+        if (data[i] == NULL) {
+            continue;
+        }
+        if (data[i] == EC_MISSING_DATA) {
+            if (def == NULL) {
+                continue;
+            }
+            len += deflen;
+        } else {
+            len += data[i]->len - 1;
+        }
+        if (num >= 0) {
+            len += seplen;
+        }
+        num++;
+    }
+
+    err = -ENOMEM;
+
+    str = GF_MALLOC(len, gf_common_mt_char);
+    if (str == NULL) {
+        goto out;
+    }
+
+    memcpy(str, pre, prelen);
+    len = prelen;
+    for (i = 0; i < ec->nodes; i++) {
+        if (data[i] == NULL) {
+            continue;
+        }
+        if (data[i] == EC_MISSING_DATA) {
+            if (deflen == 0) {
+                continue;
+            }
+            tmp = deflen;
+            memcpy(str + len, def, tmp);
+        } else {
+            tmp = data[i]->len - 1;
+            memcpy(str + len, data[i]->data, tmp);
+        }
+        len += tmp;
+        if (i < num) {
+            memcpy(str + len, sep, seplen);
+            len += seplen;
+        }
+    }
+    memcpy(str + len, post, postlen + 1);
+
+    dict = (which == EC_COMBINE_XDATA) ? cbk->xdata : cbk->dict;
+    if (new_key) {
+        key = new_key;
+    }
+    err = dict_set_dynstr(dict, key, str);
+    if (err != 0) {
+        goto out;
+    }
+
+    str = NULL;
+
+out:
+    GF_FREE(str);
+    GF_FREE(pre);
+
+    return err;
+}
+
+int32_t
+ec_dict_data_merge(ec_cbk_data_t *cbk, int32_t which, char *key)
+{
+    ec_t *ec = cbk->fop->xl->private;
+    data_t *data[ec->nodes];
+    dict_t *dict, *lockinfo, *tmp = NULL;
+    char *ptr = NULL;
+    int32_t i, len;
+    int32_t err;
+
+    ec_dict_list(data, cbk, which, key, _gf_false);
+
+    lockinfo = dict_new();
+    if (lockinfo == NULL) {
+        return -ENOMEM;
+    }
+
+    for (i = 0; i < ec->nodes; i++) {
+        if ((data[i] == NULL) || (data[i] == EC_MISSING_DATA)) {
+            continue;
+        }
+
+        tmp = dict_new();
+        if (tmp == NULL) {
+            err = -ENOMEM;
+
+            goto out;
+        }
+        err = dict_unserialize(data[i]->data, data[i]->len, &tmp);
+        if (err != 0) {
+            goto out;
+        }
+        if (dict_copy(tmp, lockinfo) == NULL) {
+            err = -ENOMEM;
+
+            goto out;
+        }
+
+        dict_unref(tmp);
+    }
+
+    tmp = NULL;
+
+    err = dict_allocate_and_serialize(lockinfo, (char **)&ptr,
+                                      (unsigned int *)&len);
+    if (err != 0) {
+        goto out;
+    }
+
+    dict = (which == EC_COMBINE_XDATA) ? cbk->xdata : cbk->dict;
+    err = dict_set_dynptr(dict, key, ptr, len);
+    if (err != 0) {
+        goto out;
+    }
+
+    ptr = NULL;
+
+out:
+    GF_FREE(ptr);
+    dict_unref(lockinfo);
+    if (tmp != NULL) {
+        dict_unref(tmp);
+    }
+
+    return err;
+}
+
+int32_t
+ec_dict_data_uuid(ec_cbk_data_t *cbk, int32_t which, char *key)
+{
+    ec_cbk_data_t *ans, *min;
+    dict_t *src, *dst;
+    data_t *data;
+
+    min = cbk;
+    for (ans = cbk->next; ans != NULL; ans = ans->next) {
+        if (ans->idx < min->idx) {
+            min = ans;
+        }
+    }
+
+    if (min != cbk) {
+        src = (which == EC_COMBINE_XDATA) ? min->xdata : min->dict;
+        dst = (which == EC_COMBINE_XDATA) ? cbk->xdata : cbk->dict;
+
+        data = dict_get(src, key);
+        if (data == NULL) {
+            return -ENOENT;
+        }
+        if (dict_set(dst, key, data) != 0) {
+            return -ENOMEM;
+        }
+    }
+
+    return 0;
+}
+
+int32_t
+ec_dict_data_iatt(ec_cbk_data_t *cbk, int32_t which, char *key)
+{
+    ec_t *ec = cbk->fop->xl->private;
+    data_t *data[ec->nodes];
+    dict_t *dict;
+    struct iatt *stbuf, *tmp;
+    int32_t i, ret;
+
+    ec_dict_list(data, cbk, which, key, _gf_false);
+
+    stbuf = NULL;
+    for (i = 0; i < ec->nodes; i++) {
+        if ((data[i] == NULL) || (data[i] == EC_MISSING_DATA)) {
+            continue;
+        }
+        tmp = data_to_iatt(data[i], key);
+        if (tmp == NULL) {
+            ret = -EINVAL;
+            goto out;
+        }
+        if (stbuf == NULL) {
+            stbuf = GF_MALLOC(sizeof(struct iatt), gf_common_mt_char);
+            if (stbuf == NULL) {
+                ret = -ENOMEM;
+                goto out;
+            }
+            *stbuf = *tmp;
+        } else {
+            if (!ec_iatt_combine(cbk->fop, stbuf, tmp, 1)) {
+                ret = -EINVAL;
+                goto out;
+            }
+        }
+    }
+
+    if ((stbuf != NULL) && (stbuf->ia_type == IA_IFREG)) {
+        ec_iatt_rebuild(ec, stbuf, 1, cbk->count);
+        /* TODO: not sure if an iatt could come in xdata from a fop that takes
+         *       no locks. */
+        if (!ec_get_inode_size(cbk->fop, cbk->fop->locks[0].lock->loc.inode,
+                               &stbuf->ia_size)) {
+            ret = -EINVAL;
+            goto out;
+        }
+    }
+
+    dict = (which == EC_COMBINE_XDATA) ? cbk->xdata : cbk->dict;
+    ret = dict_set_iatt(dict, key, stbuf, false);
+    if (ret >= 0) {
+        stbuf = NULL;
+    }
+
+out:
+    GF_FREE(stbuf);
+
+    return ret;
+}
+
+int32_t
+ec_dict_data_max32(ec_cbk_data_t *cbk, int32_t which, char *key)
+{
+    ec_t *ec = cbk->fop->xl->private;
+    data_t *data[ec->nodes];
+    dict_t *dict;
+    int32_t i;
+    uint32_t max, tmp;
+
+    ec_dict_list(data, cbk, which, key, _gf_false);
+
+    max = 0;
+    for (i = 0; i < ec->nodes; i++) {
+        if ((data[i] == NULL) || (data[i] == EC_MISSING_DATA)) {
+            continue;
+        }
+
+        tmp = data_to_uint32(data[i]);
+        if (max < tmp) {
+            max = tmp;
+        }
+    }
+
+    dict = (which == EC_COMBINE_XDATA) ? cbk->xdata : cbk->dict;
+    return dict_set_uint32(dict, key, max);
+}
+
+int32_t
+ec_dict_data_max64(ec_cbk_data_t *cbk, int32_t which, char *key)
+{
+    ec_t *ec = cbk->fop->xl->private;
+    data_t *data[ec->nodes];
+    dict_t *dict;
+    int32_t i;
+    uint64_t max, tmp;
+
+    ec_dict_list(data, cbk, which, key, _gf_false);
+
+    max = 0;
+    for (i = 0; i < ec->nodes; i++) {
+        if ((data[i] == NULL) || (data[i] == EC_MISSING_DATA)) {
+            continue;
+        }
+
+        tmp = data_to_uint64(data[i]);
+        if (max < tmp) {
+            max = tmp;
+        }
+    }
+
+    dict = (which == EC_COMBINE_XDATA) ? cbk->xdata : cbk->dict;
+    return dict_set_uint64(dict, key, max);
+}
+
+int32_t
+ec_dict_data_quota(ec_cbk_data_t *cbk, int32_t which, char *key)
+{
+    ec_t *ec = cbk->fop->xl->private;
+    data_t *data[ec->nodes];
+    dict_t *dict = NULL;
+    int32_t i = 0;
+    quota_meta_t size = {
+        0,
+    };
+    quota_meta_t max_size = {
+        0,
+    };
+
+    if (ec_dict_list(data, cbk, which, key, _gf_false) == 0) {
+        return 0;
+    }
+
+    /* Quota size xattr is managed outside of the control of the ec xlator.
+     * This means that it might not be updated at the same time on all
+     * bricks and we can receive slightly different values. If that's the
+     * case, we take the maximum of all received values.
+     */
+    for (i = 0; i < ec->nodes; i++) {
+        if ((data[i] == NULL) || (data[i] == EC_MISSING_DATA) ||
+            (quota_data_to_meta(data[i], &size) < 0)) {
+            continue;
+        }
+
+        if (size.size > max_size.size)
+            max_size.size = size.size;
+        if (size.file_count > max_size.file_count)
+            max_size.file_count = size.file_count;
+        if (size.dir_count > max_size.dir_count)
+            max_size.dir_count = size.dir_count;
+    }
+
+    max_size.size *= ec->fragments;
+
+    dict = (which == EC_COMBINE_XDATA) ? cbk->xdata : cbk->dict;
+    return quota_dict_set_meta(dict, key, &max_size, IA_IFDIR);
+}
+
+int32_t
+ec_dict_data_stime(ec_cbk_data_t *cbk, int32_t which, char *key)
+{
+    ec_t *ec = cbk->fop->xl->private;
+    data_t *data[ec->nodes];
+    dict_t *dict;
+    int32_t i, err;
+
+    ec_dict_list(data, cbk, which, key, _gf_false);
+
+    dict = (which == EC_COMBINE_XDATA) ? cbk->xdata : cbk->dict;
+    for (i = 0; i < ec->nodes; i++) {
+        if ((data[i] == NULL) || (data[i] == EC_MISSING_DATA)) {
+            continue;
+        }
+        err = gf_get_max_stime(cbk->fop->xl, dict, key, data[i]);
+        if (err != 0) {
+            gf_msg(cbk->fop->xl->name, GF_LOG_ERROR, -err,
+                   EC_MSG_STIME_COMBINE_FAIL, "STIME combination failed");
+
+            return err;
+        }
+    }
+
+    return 0;
+}
+
+int32_t
+ec_dict_data_combine(dict_t *dict, char *key, data_t *value, void *arg)
+{
+    ec_dict_combine_t *data = arg;
+
+    if ((strcmp(key, GF_XATTR_PATHINFO_KEY) == 0) ||
+        (strcmp(key, GF_XATTR_USER_PATHINFO_KEY) == 0)) {
+        return ec_dict_data_concat(data->cbk, data->which, key, NULL, NULL,
+                                   _gf_false, _gf_false, "(<EC:%s> { })",
+                                   data->cbk->fop->xl->name);
+    }
+
+    if (strncmp(key, GF_XATTR_CLRLK_CMD, SLEN(GF_XATTR_CLRLK_CMD)) == 0) {
+        return ec_dict_data_concat(data->cbk, data->which, key, NULL, NULL,
+                                   _gf_false, "{\n}");
+    }
+
+    if (strncmp(key, GF_XATTR_LOCKINFO_KEY, SLEN(GF_XATTR_LOCKINFO_KEY)) == 0) {
+        return ec_dict_data_merge(data->cbk, data->which, key);
+    }
+
+    if (strcmp(key, GET_LINK_COUNT) == 0) {
+        return ec_dict_data_max32(data->cbk, data->which, key);
+    }
+
+    if (strcmp(key, GLUSTERFS_OPEN_FD_COUNT) == 0) {
+        return ec_dict_data_max32(data->cbk, data->which, key);
+    }
+    if ((strcmp(key, GLUSTERFS_INODELK_COUNT) == 0) ||
+        (strcmp(key, GLUSTERFS_ENTRYLK_COUNT) == 0)) {
+        return ec_dict_data_max32(data->cbk, data->which, key);
+    }
+
+    if (strcmp(key, QUOTA_SIZE_KEY) == 0) {
+        return ec_dict_data_quota(data->cbk, data->which, key);
+    }
+    /* Ignore all other quota attributes */
+    if (strncmp(key, EC_QUOTA_PREFIX, SLEN(EC_QUOTA_PREFIX)) == 0) {
+        return 0;
+    }
+
+    if (XATTR_IS_NODE_UUID(key)) {
+        if (data->cbk->fop->int32) {
+            /* List of node uuid is requested */
+            return ec_dict_data_concat(data->cbk, data->which, key,
+                                       GF_XATTR_LIST_NODE_UUIDS_KEY, UUID0_STR,
+                                       _gf_true, "{ }");
+        } else {
+            return ec_dict_data_uuid(data->cbk, data->which, key);
+        }
+    }
+
+    if (fnmatch(GF_XATTR_STIME_PATTERN, key, FNM_NOESCAPE) == 0) {
+        return ec_dict_data_stime(data->cbk, data->which, key);
+    }
+
+    if (fnmatch(MARKER_XATTR_PREFIX ".*." XTIME, key, FNM_NOESCAPE) == 0) {
+        return ec_dict_data_max64(data->cbk, data->which, key);
+    }
+
+    if (strcmp(key, GF_PRESTAT) == 0 || strcmp(key, GF_POSTSTAT) == 0) {
+        return ec_dict_data_iatt(data->cbk, data->which, key);
+    }
+
+    return 0;
+}
+
+int32_t
+ec_dict_combine(ec_cbk_data_t *cbk, int32_t which)
+{
+    dict_t *dict = NULL;
+    ec_dict_combine_t data;
+    int32_t err = 0;
+
+    data.cbk = cbk;
+    data.which = which;
+
+    dict = (which == EC_COMBINE_XDATA) ? cbk->xdata : cbk->dict;
+    if (dict != NULL) {
+        err = dict_foreach(dict, ec_dict_data_combine, &data);
+        if (err != 0) {
+            gf_msg(cbk->fop->xl->name, GF_LOG_ERROR, -err,
+                   EC_MSG_DICT_COMBINE_FAIL, "Dictionary combination failed");
+
+            return err;
+        }
+    }
+
+    return 0;
+}
+
+int32_t
+ec_vector_compare(struct iovec *dst_vector, int32_t dst_count,
+                  struct iovec *src_vector, int32_t src_count)
+{
+    int32_t dst_size = 0, src_size = 0;
+
+    if (dst_count > 0) {
+        dst_size = iov_length(dst_vector, dst_count);
+    }
+    if (src_count > 0) {
+        src_size = iov_length(src_vector, src_count);
+    }
+
+    return (dst_size == src_size);
+}
+
+int32_t
+ec_flock_compare(struct gf_flock *dst, struct gf_flock *src)
+{
+    if ((dst->l_type != src->l_type) || (dst->l_whence != src->l_whence) ||
+        (dst->l_start != src->l_start) || (dst->l_len != src->l_len) ||
+        (dst->l_pid != src->l_pid) ||
+        !is_same_lkowner(&dst->l_owner, &src->l_owner)) {
+        return 0;
+    }
+
+    return 1;
+}
+
+void
+ec_statvfs_combine(struct statvfs *dst, struct statvfs *src)
+{
+    if (dst->f_bsize < src->f_bsize) {
+        dst->f_bsize = src->f_bsize;
+    }
+
+    if (dst->f_frsize < src->f_frsize) {
+        dst->f_blocks *= dst->f_frsize;
+        dst->f_blocks /= src->f_frsize;
+
+        dst->f_bfree *= dst->f_frsize;
+        dst->f_bfree /= src->f_frsize;
+
+        dst->f_bavail *= dst->f_frsize;
+        dst->f_bavail /= src->f_frsize;
+
+        dst->f_frsize = src->f_frsize;
+    } else if (dst->f_frsize > src->f_frsize) {
+        src->f_blocks *= src->f_frsize;
+        src->f_blocks /= dst->f_frsize;
+
+        src->f_bfree *= src->f_frsize;
+        src->f_bfree /= dst->f_frsize;
+
+        src->f_bavail *= src->f_frsize;
+        src->f_bavail /= dst->f_frsize;
+    }
+    if (dst->f_blocks > src->f_blocks) {
+        dst->f_blocks = src->f_blocks;
+    }
+    if (dst->f_bfree > src->f_bfree) {
+        dst->f_bfree = src->f_bfree;
+    }
+    if (dst->f_bavail > src->f_bavail) {
+        dst->f_bavail = src->f_bavail;
+    }
+
+    if (dst->f_files < src->f_files) {
+        dst->f_files = src->f_files;
+    }
+    if (dst->f_ffree > src->f_ffree) {
+        dst->f_ffree = src->f_ffree;
+    }
+    if (dst->f_favail > src->f_favail) {
+        dst->f_favail = src->f_favail;
+    }
+    if (dst->f_namemax > src->f_namemax) {
+        dst->f_namemax = src->f_namemax;
+    }
+
+    if (dst->f_flag != src->f_flag) {
+        gf_msg_debug(THIS->name, 0,
+                     "Mismatching file system flags "
+                     "(%lX, %lX)",
+                     dst->f_flag, src->f_flag);
+    }
+    dst->f_flag &= src->f_flag;
+}
+
+int32_t
+ec_combine_check(ec_cbk_data_t *dst, ec_cbk_data_t *src, ec_combine_f combine)
+{
+    ec_fop_data_t *fop = dst->fop;
+
+    if (dst->op_ret != src->op_ret) {
+        gf_msg_debug(fop->xl->name, 0,
+                     "Mismatching return code in "
+                     "answers of '%s': %d <-> %d",
+                     ec_fop_name(fop->id), dst->op_ret, src->op_ret);
+
+        return 0;
+    }
+    if (dst->op_ret < 0) {
+        if (dst->op_errno != src->op_errno) {
+            gf_msg_debug(fop->xl->name, 0,
+                         "Mismatching errno code in "
+                         "answers of '%s': %d <-> %d",
+                         ec_fop_name(fop->id), dst->op_errno, src->op_errno);
+
+            return 0;
+        }
+    }
+
+    if (!ec_dict_compare(dst->xdata, src->xdata)) {
+        gf_msg(fop->xl->name, GF_LOG_DEBUG, 0, EC_MSG_XDATA_MISMATCH,
+               "Mismatching xdata in answers "
+               "of '%s'",
+               ec_fop_name(fop->id));
+
+        return 0;
+    }
+
+    if ((dst->op_ret >= 0) && (combine != NULL)) {
+        return combine(fop, dst, src);
+    }
+
+    return 1;
+}
+
+void
+ec_combine(ec_cbk_data_t *newcbk, ec_combine_f combine)
+{
+    ec_fop_data_t *fop = newcbk->fop;
+    ec_cbk_data_t *cbk = NULL, *tmp = NULL;
+    struct list_head *item = NULL;
+    int32_t needed = 0;
+    char str[32];
+
+    LOCK(&fop->lock);
+
+    fop->received |= newcbk->mask;
+
+    item = fop->cbk_list.prev;
+    list_for_each_entry(cbk, &fop->cbk_list, list)
+    {
+        if (ec_combine_check(newcbk, cbk, combine)) {
+            newcbk->count += cbk->count;
+            newcbk->mask |= cbk->mask;
+
+            item = cbk->list.prev;
+            while (item != &fop->cbk_list) {
+                tmp = list_entry(item, ec_cbk_data_t, list);
+                if (tmp->count >= newcbk->count) {
+                    break;
+                }
+                item = item->prev;
+            }
+            list_del(&cbk->list);
+
+            newcbk->next = cbk;
+
+            break;
+        }
+    }
+    list_add(&newcbk->list, item);
+
+    ec_trace("ANSWER", fop, "combine=%s[%d]",
+             ec_bin(str, sizeof(str), newcbk->mask, 0), newcbk->count);
+
+    cbk = list_entry(fop->cbk_list.next, ec_cbk_data_t, list);
+    if ((fop->mask ^ fop->remaining) == fop->received) {
+        needed = fop->minimum - cbk->count;
+    }
+
+    UNLOCK(&fop->lock);
+
+    if (needed > 0) {
+        ec_dispatch_next(fop, newcbk->idx);
+    }
+}
diff --git a/xlators/cluster/ec/src/ec-combine.h b/xlators/cluster/ec/src/ec-combine.h
new file mode 100644
index 00000000000..1010cc3be26
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-combine.h
@@ -0,0 +1,44 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_COMBINE_H__
+#define __EC_COMBINE_H__
+
+#define EC_COMBINE_DICT 0
+#define EC_COMBINE_XDATA 1
+
+typedef int32_t (*ec_combine_f)(ec_fop_data_t *fop, ec_cbk_data_t *dst,
+                                ec_cbk_data_t *src);
+
+void
+ec_iatt_rebuild(ec_t *ec, struct iatt *iatt, int32_t count, int32_t answers);
+
+int32_t
+ec_iatt_combine(ec_fop_data_t *fop, struct iatt *dst, struct iatt *src,
+                int32_t count);
+int32_t
+ec_dict_compare(dict_t *dict1, dict_t *dict2);
+int32_t
+ec_vector_compare(struct iovec *dst_vector, int32_t dst_count,
+                  struct iovec *src_vector, int32_t src_count);
+int32_t
+ec_flock_compare(struct gf_flock *dst, struct gf_flock *src);
+void
+ec_statvfs_combine(struct statvfs *dst, struct statvfs *src);
+
+int32_t
+ec_dict_combine(ec_cbk_data_t *cbk, int32_t which);
+
+void
+ec_combine(ec_cbk_data_t *cbk, ec_combine_f combine);
+
+int32_t
+ec_combine_write(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src);
+#endif /* __EC_COMBINE_H__ */
diff --git a/xlators/cluster/ec/src/ec-common.c b/xlators/cluster/ec/src/ec-common.c
new file mode 100644
index 00000000000..b955efd8c2d
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-common.c
@@ -0,0 +1,3042 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/byte-order.h>
+#include <glusterfs/hashfn.h>
+
+#include "ec-mem-types.h"
+#include "ec-types.h"
+#include "ec-helpers.h"
+#include "ec-combine.h"
+#include "ec-common.h"
+#include "ec-fops.h"
+#include "ec-method.h"
+#include "ec.h"
+#include "ec-messages.h"
+
+#define EC_INVALID_INDEX UINT32_MAX
+
+void
+ec_update_fd_status(fd_t *fd, xlator_t *xl, int idx, int32_t ret_status)
+{
+    ec_fd_t *fd_ctx;
+
+    if (fd == NULL)
+        return;
+
+    LOCK(&fd->lock);
+    {
+        fd_ctx = __ec_fd_get(fd, xl);
+        if (fd_ctx) {
+            if (ret_status >= 0)
+                fd_ctx->fd_status[idx] = EC_FD_OPENED;
+            else
+                fd_ctx->fd_status[idx] = EC_FD_NOT_OPENED;
+        }
+    }
+    UNLOCK(&fd->lock);
+}
+
+static uintptr_t
+ec_fd_ctx_need_open(fd_t *fd, xlator_t *this, uintptr_t mask)
+{
+    int i = 0;
+    int count = 0;
+    ec_t *ec = NULL;
+    ec_fd_t *fd_ctx = NULL;
+    uintptr_t need_open = 0;
+
+    ec = this->private;
+
+    fd_ctx = ec_fd_get(fd, this);
+    if (!fd_ctx)
+        return count;
+
+    LOCK(&fd->lock);
+    {
+        for (i = 0; i < ec->nodes; i++) {
+            if ((fd_ctx->fd_status[i] == EC_FD_NOT_OPENED) &&
+                ((ec->xl_up & (1 << i)) != 0) && ((mask & (1 << i)) != 0)) {
+                fd_ctx->fd_status[i] = EC_FD_OPENING;
+                need_open |= (1 << i);
+                count++;
+            }
+        }
+    }
+    UNLOCK(&fd->lock);
+
+    /* If fd needs to open on minimum number of nodes
+     * then ignore fixing the fd as it has been
+     * requested from heal operation.
+     */
+    if (count >= ec->fragments) {
+        need_open = 0;
+    }
+
+    return need_open;
+}
+
+static gf_boolean_t
+ec_is_fd_fixable(fd_t *fd)
+{
+    if (!fd || !fd->inode)
+        return _gf_false;
+    else if (fd_is_anonymous(fd))
+        return _gf_false;
+    else if (gf_uuid_is_null(fd->inode->gfid))
+        return _gf_false;
+
+    return _gf_true;
+}
+
+static void
+ec_fix_open(ec_fop_data_t *fop, uintptr_t mask)
+{
+    uintptr_t need_open = 0;
+    int ret = 0;
+    int32_t flags = 0;
+    loc_t loc = {
+        0,
+    };
+
+    if (!ec_is_fd_fixable(fop->fd))
+        goto out;
+
+    /* Evaluate how many remote fd's to be opened */
+    need_open = ec_fd_ctx_need_open(fop->fd, fop->xl, mask);
+    if (need_open == 0) {
+        goto out;
+    }
+
+    loc.inode = inode_ref(fop->fd->inode);
+    gf_uuid_copy(loc.gfid, fop->fd->inode->gfid);
+    ret = loc_path(&loc, NULL);
+    if (ret < 0) {
+        goto out;
+    }
+
+    flags = fop->fd->flags & (~(O_TRUNC | O_APPEND | O_CREAT | O_EXCL));
+    if (IA_IFDIR == fop->fd->inode->ia_type) {
+        ec_opendir(fop->frame, fop->xl, need_open,
+                   EC_MINIMUM_ONE | EC_FOP_NO_PROPAGATE_ERROR, NULL, NULL,
+                   &fop->loc[0], fop->fd, NULL);
+    } else {
+        ec_open(fop->frame, fop->xl, need_open,
+                EC_MINIMUM_ONE | EC_FOP_NO_PROPAGATE_ERROR, NULL, NULL, &loc,
+                flags, fop->fd, NULL);
+    }
+
+out:
+    loc_wipe(&loc);
+}
+
+static off_t
+ec_range_end_get(off_t fl_start, uint64_t fl_size)
+{
+    if (fl_size > 0) {
+        if (fl_size >= EC_RANGE_FULL) {
+            /* Infinity */
+            fl_start = LLONG_MAX;
+        } else {
+            fl_start += fl_size - 1;
+            if (fl_start < 0) {
+                /* Overflow */
+                fl_start = LLONG_MAX;
+            }
+        }
+    }
+
+    return fl_start;
+}
+
+static gf_boolean_t
+ec_is_range_conflict(ec_lock_link_t *l1, ec_lock_link_t *l2)
+{
+    return ((l1->fl_end >= l2->fl_start) && (l2->fl_end >= l1->fl_start));
+}
+
+static gf_boolean_t
+ec_lock_conflict(ec_lock_link_t *l1, ec_lock_link_t *l2)
+{
+    ec_t *ec = l1->fop->xl->private;
+
+    /* Fops like access/stat won't have to worry what the other fops are
+     * modifying as the fop is wound only to one brick. So it can be
+     * executed in parallel*/
+    if (l1->fop->minimum == EC_MINIMUM_ONE ||
+        l2->fop->minimum == EC_MINIMUM_ONE)
+        return _gf_false;
+
+    if ((l1->fop->flags & EC_FLAG_LOCK_SHARED) &&
+        (l2->fop->flags & EC_FLAG_LOCK_SHARED))
+        return _gf_false;
+
+    if (!ec->parallel_writes) {
+        return _gf_true;
+    }
+
+    return ec_is_range_conflict(l1, l2);
+}
+
+uint32_t
+ec_select_first_by_read_policy(ec_t *ec, ec_fop_data_t *fop)
+{
+    if (ec->read_policy == EC_ROUND_ROBIN) {
+        return ec->idx;
+    } else if (ec->read_policy == EC_GFID_HASH) {
+        if (fop->use_fd) {
+            return SuperFastHash((char *)fop->fd->inode->gfid,
+                                 sizeof(fop->fd->inode->gfid)) %
+                   ec->nodes;
+        } else {
+            if (gf_uuid_is_null(fop->loc[0].gfid))
+                loc_gfid(&fop->loc[0], fop->loc[0].gfid);
+            return SuperFastHash((char *)fop->loc[0].gfid,
+                                 sizeof(fop->loc[0].gfid)) %
+                   ec->nodes;
+        }
+    }
+    return 0;
+}
+
+static gf_boolean_t
+ec_child_valid(ec_t *ec, ec_fop_data_t *fop, uint32_t idx)
+{
+    return (idx < ec->nodes) && (((fop->remaining >> idx) & 1) == 1);
+}
+
+static uint32_t
+ec_child_next(ec_t *ec, ec_fop_data_t *fop, uint32_t idx)
+{
+    while (!ec_child_valid(ec, fop, idx)) {
+        if (++idx >= ec->nodes) {
+            idx = 0;
+        }
+        if (idx == fop->first) {
+            return EC_INVALID_INDEX;
+        }
+    }
+
+    return idx;
+}
+
+int32_t
+ec_heal_report(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, uintptr_t mask, uintptr_t good,
+               uintptr_t bad, uint32_t pending, dict_t *xdata)
+{
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_DEBUG, op_errno, EC_MSG_HEAL_FAIL,
+               "Heal failed");
+    } else {
+        if ((mask & ~good) != 0) {
+            gf_msg(this->name, GF_LOG_DEBUG, 0, EC_MSG_HEAL_SUCCESS,
+                   "Heal succeeded on %d/%d "
+                   "subvolumes",
+                   gf_bits_count(mask & ~(good | bad)),
+                   gf_bits_count(mask & ~good));
+        }
+    }
+
+    return 0;
+}
+
+static uintptr_t
+ec_fop_needs_name_heal(ec_fop_data_t *fop)
+{
+    ec_t *ec = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    ec_cbk_data_t *enoent_cbk = NULL;
+
+    ec = fop->xl->private;
+    if (fop->id != GF_FOP_LOOKUP)
+        return 0;
+
+    if (!fop->loc[0].name || strlen(fop->loc[0].name) == 0)
+        return 0;
+
+    list_for_each_entry(cbk, &fop->cbk_list, list)
+    {
+        if (cbk->op_ret < 0 && cbk->op_errno == ENOENT) {
+            enoent_cbk = cbk;
+            break;
+        }
+    }
+
+    if (!enoent_cbk)
+        return 0;
+
+    return ec->xl_up & ~enoent_cbk->mask;
+}
+
+int32_t
+ec_fop_needs_heal(ec_fop_data_t *fop)
+{
+    ec_t *ec = fop->xl->private;
+
+    if (fop->lock_count == 0) {
+        /*
+         * if fop->lock_count is zero that means it saw version mismatch
+         * without any locks so it can't be trusted. If we launch a heal
+         * based on this it will lead to INODELKs which will affect I/O
+         * performance. Considering self-heal-daemon and operations on
+         * the inode from client which take locks can still trigger the
+         * heal we can choose to not attempt a heal when fop->lock_count
+         * is zero.
+         */
+        return 0;
+    }
+    return (ec->xl_up & ~(fop->remaining | fop->good)) != 0;
+}
+
+void
+ec_check_status(ec_fop_data_t *fop)
+{
+    ec_t *ec = fop->xl->private;
+    int32_t partial = 0;
+    char str1[32], str2[32], str3[32], str4[32], str5[32];
+
+    if (!ec_fop_needs_name_heal(fop) && !ec_fop_needs_heal(fop)) {
+        return;
+    }
+
+    if (fop->answer && fop->answer->op_ret >= 0) {
+        if ((fop->id == GF_FOP_LOOKUP) || (fop->id == GF_FOP_STAT) ||
+            (fop->id == GF_FOP_FSTAT)) {
+            partial = fop->answer->iatt[0].ia_type == IA_IFDIR;
+        } else if (fop->id == GF_FOP_OPENDIR) {
+            partial = 1;
+        }
+    }
+
+    gf_msg(
+        fop->xl->name, GF_LOG_WARNING, 0, EC_MSG_OP_FAIL_ON_SUBVOLS,
+        "Operation failed on %d of %d subvolumes.(up=%s, mask=%s, "
+        "remaining=%s, good=%s, bad=%s,"
+        "(Least significant bit represents first client/brick of subvol), %s)",
+        gf_bits_count(ec->xl_up & ~(fop->remaining | fop->good)), ec->nodes,
+        ec_bin(str1, sizeof(str1), ec->xl_up, ec->nodes),
+        ec_bin(str2, sizeof(str2), fop->mask, ec->nodes),
+        ec_bin(str3, sizeof(str3), fop->remaining, ec->nodes),
+        ec_bin(str4, sizeof(str4), fop->good, ec->nodes),
+        ec_bin(str5, sizeof(str5), ec->xl_up & ~(fop->remaining | fop->good),
+               ec->nodes),
+        ec_msg_str(fop));
+    if (fop->use_fd) {
+        if (fop->fd != NULL) {
+            ec_fheal(NULL, fop->xl, -1, EC_MINIMUM_ONE, ec_heal_report, NULL,
+                     fop->fd, partial, NULL);
+        }
+    } else {
+        ec_heal(NULL, fop->xl, -1, EC_MINIMUM_ONE, ec_heal_report, NULL,
+                &fop->loc[0], partial, NULL);
+
+        if (fop->loc[1].inode != NULL) {
+            ec_heal(NULL, fop->xl, -1, EC_MINIMUM_ONE, ec_heal_report, NULL,
+                    &fop->loc[1], partial, NULL);
+        }
+    }
+}
+
+void
+ec_update_good(ec_fop_data_t *fop, uintptr_t good)
+{
+    fop->good = good;
+
+    /* Fops that are executed only on one brick do not have enough information
+     * to decide if healing is needed or not. */
+    if ((fop->expected != 1) && (fop->parent == NULL)) {
+        ec_check_status(fop);
+    }
+}
+
+void
+ec_lock_update_good(ec_lock_t *lock, ec_fop_data_t *fop)
+{
+    /* Fops that are executed only on one brick do not have enough information
+     * to update the global mask of good bricks. */
+    if (fop->expected == 1) {
+        return;
+    }
+
+    /* When updating the good mask of the lock, we only take into consideration
+     * those bits corresponding to the bricks where the fop has been executed.
+     * Bad bricks are removed from good_mask, but once marked as bad it's never
+     * set to good until the lock is released and reacquired */
+
+    lock->good_mask &= fop->good | fop->remaining;
+}
+
+void
+__ec_fop_set_error(ec_fop_data_t *fop, int32_t error)
+{
+    if ((error != 0) && (fop->error == 0)) {
+        fop->error = error;
+    }
+}
+
+void
+ec_fop_set_error(ec_fop_data_t *fop, int32_t error)
+{
+    LOCK(&fop->lock);
+
+    __ec_fop_set_error(fop, error);
+
+    UNLOCK(&fop->lock);
+}
+
+gf_boolean_t
+ec_cbk_set_error(ec_cbk_data_t *cbk, int32_t error, gf_boolean_t ro)
+{
+    if ((error != 0) && (cbk->op_ret >= 0)) {
+        /* If cbk->op_errno was 0, it means that the fop succeeded and this
+         * error has happened while processing the answer. If the operation was
+         * read-only, there's no problem (i.e. we simply return the generated
+         * error code). However if it caused a modification, we must return EIO
+         * to indicate that the operation has been partially executed. */
+        cbk->op_errno = ro ? error : EIO;
+        cbk->op_ret = -1;
+
+        ec_fop_set_error(cbk->fop, cbk->op_errno);
+    }
+
+    return (cbk->op_ret < 0);
+}
+
+ec_cbk_data_t *
+ec_fop_prepare_answer(ec_fop_data_t *fop, gf_boolean_t ro)
+{
+    ec_cbk_data_t *cbk;
+    int32_t err;
+
+    cbk = fop->answer;
+    if (cbk == NULL) {
+        ec_fop_set_error(fop, EIO);
+
+        return NULL;
+    }
+
+    if (cbk->op_ret < 0) {
+        ec_fop_set_error(fop, cbk->op_errno);
+    }
+
+    err = ec_dict_combine(cbk, EC_COMBINE_XDATA);
+    if (ec_cbk_set_error(cbk, -err, ro)) {
+        return NULL;
+    }
+
+    return cbk;
+}
+
+void
+ec_sleep(ec_fop_data_t *fop)
+{
+    LOCK(&fop->lock);
+
+    GF_ASSERT(fop->refs > 0);
+    fop->refs++;
+    fop->jobs++;
+
+    UNLOCK(&fop->lock);
+}
+
+int32_t
+ec_check_complete(ec_fop_data_t *fop, ec_resume_f resume)
+{
+    int32_t error = -1;
+
+    LOCK(&fop->lock);
+
+    GF_ASSERT(fop->resume == NULL);
+
+    if (--fop->jobs != 0) {
+        ec_trace("WAIT", fop, "resume=%p", resume);
+
+        fop->resume = resume;
+    } else {
+        error = fop->error;
+        fop->error = 0;
+    }
+
+    UNLOCK(&fop->lock);
+
+    return error;
+}
+
+void
+ec_resume(ec_fop_data_t *fop, int32_t error)
+{
+    ec_resume_f resume = NULL;
+
+    LOCK(&fop->lock);
+
+    __ec_fop_set_error(fop, error);
+
+    if (--fop->jobs == 0) {
+        resume = fop->resume;
+        fop->resume = NULL;
+        if (resume != NULL) {
+            ec_trace("RESUME", fop, "error=%d", error);
+
+            if (fop->error != 0) {
+                error = fop->error;
+            }
+            fop->error = 0;
+        }
+    }
+
+    UNLOCK(&fop->lock);
+
+    if (resume != NULL) {
+        resume(fop, error);
+    }
+
+    ec_fop_data_release(fop);
+}
+
+void
+ec_resume_parent(ec_fop_data_t *fop)
+{
+    ec_fop_data_t *parent;
+    int32_t error = 0;
+
+    parent = fop->parent;
+    if (parent != NULL) {
+        if ((fop->fop_flags & EC_FOP_NO_PROPAGATE_ERROR) == 0) {
+            error = fop->error;
+        }
+        ec_trace("RESUME_PARENT", fop, "error=%u", error);
+        fop->parent = NULL;
+        ec_resume(parent, error);
+    }
+}
+
+gf_boolean_t
+ec_is_recoverable_error(int32_t op_errno)
+{
+    switch (op_errno) {
+        case ENOTCONN:
+        case ESTALE:
+        case ENOENT:
+        case EBADFD: /*Opened fd but brick is disconnected*/
+        case EIO:    /*Backend-fs crash like XFS/ext4 etc*/
+            return _gf_true;
+    }
+    return _gf_false;
+}
+
+void
+ec_complete(ec_fop_data_t *fop)
+{
+    ec_cbk_data_t *cbk = NULL;
+    int32_t resume = 0, update = 0;
+    int healing_count = 0;
+
+    LOCK(&fop->lock);
+
+    ec_trace("COMPLETE", fop, "");
+
+    if (--fop->winds == 0) {
+        if (fop->answer == NULL) {
+            if (!list_empty(&fop->cbk_list)) {
+                cbk = list_entry(fop->cbk_list.next, ec_cbk_data_t, list);
+                healing_count = gf_bits_count(cbk->mask & fop->healing);
+                /* fop shouldn't be treated as success if it is not
+                 * successful on at least fop->minimum good copies*/
+                if ((cbk->count - healing_count) >= fop->minimum) {
+                    fop->answer = cbk;
+
+                    update = 1;
+                }
+            }
+
+            resume = 1;
+        }
+    }
+
+    UNLOCK(&fop->lock);
+
+    /* ec_update_good() locks inode->lock. This may cause deadlocks with
+       fop->lock when used in another order. Since ec_update_good() will not
+       be called more than once for each fop, it can be called from outside
+       the fop->lock locked region. */
+    if (update) {
+        ec_update_good(fop, cbk->mask);
+    }
+
+    if (resume) {
+        ec_resume(fop, 0);
+    }
+
+    ec_fop_data_release(fop);
+}
+
+/* There could be already granted locks sitting on the bricks, unlock for which
+ * must be wound at all costs*/
+static gf_boolean_t
+ec_must_wind(ec_fop_data_t *fop)
+{
+    if ((fop->id == GF_FOP_INODELK) || (fop->id == GF_FOP_FINODELK) ||
+        (fop->id == GF_FOP_LK)) {
+        if (fop->flock.l_type == F_UNLCK)
+            return _gf_true;
+    } else if ((fop->id == GF_FOP_ENTRYLK) || (fop->id == GF_FOP_FENTRYLK)) {
+        if (fop->entrylk_cmd == ENTRYLK_UNLOCK)
+            return _gf_true;
+    }
+
+    return _gf_false;
+}
+
+static gf_boolean_t
+ec_internal_op(ec_fop_data_t *fop)
+{
+    if (ec_must_wind(fop))
+        return _gf_true;
+    if (fop->id == GF_FOP_XATTROP)
+        return _gf_true;
+    if (fop->id == GF_FOP_FXATTROP)
+        return _gf_true;
+    if (fop->id == GF_FOP_OPEN)
+        return _gf_true;
+    return _gf_false;
+}
+
+char *
+ec_msg_str(ec_fop_data_t *fop)
+{
+    loc_t *loc1 = NULL;
+    loc_t *loc2 = NULL;
+    char gfid1[64] = {0};
+    char gfid2[64] = {0};
+    ec_fop_data_t *parent = fop->parent;
+
+    if (fop->errstr)
+        return fop->errstr;
+    if (!fop->use_fd) {
+        loc1 = &fop->loc[0];
+        loc2 = &fop->loc[1];
+
+        if (fop->id == GF_FOP_RENAME) {
+            gf_asprintf(&fop->errstr,
+                        "FOP : '%s' failed on '%s' and '%s' with gfids "
+                        "%s and %s respectively. Parent FOP: %s",
+                        ec_fop_name(fop->id), loc1->path, loc2->path,
+                        uuid_utoa_r(loc1->gfid, gfid1),
+                        uuid_utoa_r(loc2->gfid, gfid2),
+                        parent ? ec_fop_name(parent->id) : "No Parent");
+        } else {
+            gf_asprintf(
+                &fop->errstr,
+                "FOP : '%s' failed on '%s' with gfid %s. Parent FOP: %s",
+                ec_fop_name(fop->id), loc1->path,
+                uuid_utoa_r(loc1->gfid, gfid1),
+                parent ? ec_fop_name(parent->id) : "No Parent");
+        }
+    } else {
+        gf_asprintf(
+            &fop->errstr, "FOP : '%s' failed on gfid %s. Parent FOP: %s",
+            ec_fop_name(fop->id), uuid_utoa_r(fop->fd->inode->gfid, gfid1),
+            parent ? ec_fop_name(parent->id) : "No Parent");
+    }
+    return fop->errstr;
+}
+
+static void
+ec_log_insufficient_vol(ec_fop_data_t *fop, int32_t have, uint32_t need,
+                        int32_t loglevel)
+{
+    ec_t *ec = fop->xl->private;
+    char str1[32], str2[32], str3[32];
+
+    gf_msg(ec->xl->name, loglevel, 0, EC_MSG_CHILDS_INSUFFICIENT,
+           "Insufficient available children for this request: "
+           "Have : %d, Need : %u : Child UP : %s "
+           "Mask: %s, Healing : %s : %s ",
+           have, need, ec_bin(str1, sizeof(str1), ec->xl_up, ec->nodes),
+           ec_bin(str2, sizeof(str2), fop->mask, ec->nodes),
+           ec_bin(str3, sizeof(str3), fop->healing, ec->nodes),
+           ec_msg_str(fop));
+}
+
+static int32_t
+ec_child_select(ec_fop_data_t *fop)
+{
+    ec_t *ec = fop->xl->private;
+    int32_t first = 0, num = 0;
+
+    ec_fop_cleanup(fop);
+
+    fop->mask &= ec->node_mask;
+    /* Wind the fop on same subvols as parent for any internal extra fops like
+     * head/tail read in case of writev fop. Unlocks shouldn't do this because
+     * unlock should go on all subvols where lock is performed*/
+    if (fop->parent && !ec_internal_op(fop)) {
+        fop->mask &= (fop->parent->mask & ~fop->parent->healing);
+        if (ec_is_data_fop(fop->id)) {
+            fop->healing |= fop->parent->healing;
+        }
+    }
+
+    if ((fop->mask & ~ec->xl_up) != 0) {
+        gf_msg(fop->xl->name, GF_LOG_WARNING, 0, EC_MSG_OP_EXEC_UNAVAIL,
+               "Executing operation with "
+               "some subvolumes unavailable. (%" PRIXPTR "). %s ",
+               fop->mask & ~ec->xl_up, ec_msg_str(fop));
+        fop->mask &= ec->xl_up;
+    }
+
+    switch (fop->minimum) {
+        case EC_MINIMUM_ALL:
+            fop->minimum = gf_bits_count(fop->mask);
+            if (fop->minimum >= ec->fragments) {
+                break;
+            }
+        case EC_MINIMUM_MIN:
+            fop->minimum = ec->fragments;
+            break;
+        case EC_MINIMUM_ONE:
+            fop->minimum = 1;
+    }
+
+    if (ec->read_policy == EC_ROUND_ROBIN) {
+        first = ec->idx;
+        if (++first >= ec->nodes) {
+            first = 0;
+        }
+        ec->idx = first;
+    }
+
+    num = gf_bits_count(fop->mask);
+    /*Unconditionally wind on healing subvolumes*/
+    fop->mask |= fop->healing;
+    fop->remaining = fop->mask;
+    fop->received = 0;
+
+    ec_trace("SELECT", fop, "");
+
+    if ((num < fop->minimum) && (num < ec->fragments)) {
+        ec_log_insufficient_vol(fop, num, fop->minimum, GF_LOG_ERROR);
+        return 0;
+    }
+
+    if (!fop->parent && fop->lock_count &&
+        (fop->locks[0].update[EC_DATA_TXN] ||
+         fop->locks[0].update[EC_METADATA_TXN])) {
+        if (ec->quorum_count && (num < ec->quorum_count)) {
+            ec_log_insufficient_vol(fop, num, ec->quorum_count, GF_LOG_ERROR);
+            return 0;
+        }
+    }
+
+    return 1;
+}
+
+void
+ec_dispatch_next(ec_fop_data_t *fop, uint32_t idx)
+{
+    uint32_t i = EC_INVALID_INDEX;
+    ec_t *ec = fop->xl->private;
+
+    LOCK(&fop->lock);
+
+    i = ec_child_next(ec, fop, idx);
+    if (i < EC_MAX_NODES) {
+        idx = i;
+
+        fop->remaining ^= 1ULL << idx;
+
+        ec_trace("EXECUTE", fop, "idx=%d", idx);
+
+        fop->winds++;
+        fop->refs++;
+    }
+
+    UNLOCK(&fop->lock);
+
+    if (i < EC_MAX_NODES) {
+        fop->wind(ec, fop, idx);
+    }
+}
+
+void
+ec_dispatch_mask(ec_fop_data_t *fop, uintptr_t mask)
+{
+    ec_t *ec = fop->xl->private;
+    int32_t count, idx;
+
+    count = gf_bits_count(mask);
+
+    LOCK(&fop->lock);
+
+    ec_trace("EXECUTE", fop, "mask=%lX", mask);
+
+    fop->remaining ^= mask;
+
+    fop->winds += count;
+    fop->refs += count;
+
+    UNLOCK(&fop->lock);
+
+    idx = 0;
+    while (mask != 0) {
+        if ((mask & 1) != 0) {
+            fop->wind(ec, fop, idx);
+        }
+        idx++;
+        mask >>= 1;
+    }
+}
+
+void
+ec_dispatch_start(ec_fop_data_t *fop)
+{
+    fop->answer = NULL;
+    fop->good = 0;
+
+    INIT_LIST_HEAD(&fop->cbk_list);
+
+    if (fop->lock_count > 0) {
+        ec_owner_copy(fop->frame, &fop->req_frame->root->lk_owner);
+    }
+}
+
+void
+ec_dispatch_one(ec_fop_data_t *fop)
+{
+    ec_dispatch_start(fop);
+
+    if (ec_child_select(fop)) {
+        ec_sleep(fop);
+
+        fop->expected = 1;
+        fop->first = ec_select_first_by_read_policy(fop->xl->private, fop);
+
+        ec_dispatch_next(fop, fop->first);
+    }
+}
+
+gf_boolean_t
+ec_dispatch_one_retry(ec_fop_data_t *fop, ec_cbk_data_t **cbk)
+{
+    ec_cbk_data_t *tmp;
+
+    tmp = ec_fop_prepare_answer(fop, _gf_true);
+    if (cbk != NULL) {
+        *cbk = tmp;
+    }
+    if ((tmp != NULL) && (tmp->op_ret < 0) &&
+        ec_is_recoverable_error(tmp->op_errno)) {
+        GF_ASSERT(fop->mask & (1ULL << tmp->idx));
+        fop->mask ^= (1ULL << tmp->idx);
+        if (fop->mask) {
+            return _gf_true;
+        }
+    }
+
+    return _gf_false;
+}
+
+void
+ec_dispatch_inc(ec_fop_data_t *fop)
+{
+    ec_dispatch_start(fop);
+
+    if (ec_child_select(fop)) {
+        ec_sleep(fop);
+
+        fop->expected = gf_bits_count(fop->remaining);
+        fop->first = 0;
+
+        ec_dispatch_next(fop, 0);
+    }
+}
+
+void
+ec_dispatch_all(ec_fop_data_t *fop)
+{
+    ec_dispatch_start(fop);
+
+    if (ec_child_select(fop)) {
+        ec_sleep(fop);
+
+        fop->expected = gf_bits_count(fop->remaining);
+        fop->first = 0;
+
+        ec_dispatch_mask(fop, fop->remaining);
+    }
+}
+
+void
+ec_dispatch_min(ec_fop_data_t *fop)
+{
+    ec_t *ec = fop->xl->private;
+    uintptr_t mask;
+    uint32_t idx;
+    int32_t count;
+
+    ec_dispatch_start(fop);
+
+    if (ec_child_select(fop)) {
+        ec_sleep(fop);
+
+        fop->expected = count = ec->fragments;
+        fop->first = ec_select_first_by_read_policy(fop->xl->private, fop);
+        idx = fop->first - 1;
+        mask = 0;
+        while (count-- > 0) {
+            idx = ec_child_next(ec, fop, idx + 1);
+            if (idx < EC_MAX_NODES)
+                mask |= 1ULL << idx;
+        }
+
+        ec_dispatch_mask(fop, mask);
+    }
+}
+
+void
+ec_succeed_all(ec_fop_data_t *fop)
+{
+    ec_dispatch_start(fop);
+
+    if (ec_child_select(fop)) {
+        fop->expected = gf_bits_count(fop->remaining);
+        fop->first = 0;
+
+        /* Simulate a successful execution on all bricks */
+        ec_trace("SUCCEED", fop, "");
+
+        fop->good = fop->remaining;
+        fop->remaining = 0;
+    }
+}
+
+ec_lock_t *
+ec_lock_allocate(ec_fop_data_t *fop, loc_t *loc)
+{
+    ec_t *ec = fop->xl->private;
+    ec_lock_t *lock;
+    int32_t err;
+
+    if ((loc->inode == NULL) ||
+        (gf_uuid_is_null(loc->gfid) && gf_uuid_is_null(loc->inode->gfid))) {
+        gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_INVALID_INODE,
+               "Trying to lock based on an invalid "
+               "inode");
+
+        __ec_fop_set_error(fop, EINVAL);
+
+        return NULL;
+    }
+
+    lock = mem_get0(ec->lock_pool);
+    if (lock != NULL) {
+        lock->good_mask = UINTPTR_MAX;
+        INIT_LIST_HEAD(&lock->owners);
+        INIT_LIST_HEAD(&lock->waiting);
+        INIT_LIST_HEAD(&lock->frozen);
+        err = ec_loc_from_loc(fop->xl, &lock->loc, loc);
+        if (err != 0) {
+            mem_put(lock);
+            lock = NULL;
+
+            __ec_fop_set_error(fop, -err);
+        }
+    }
+
+    return lock;
+}
+
+void
+ec_lock_destroy(ec_lock_t *lock)
+{
+    loc_wipe(&lock->loc);
+    if (lock->fd != NULL) {
+        fd_unref(lock->fd);
+    }
+
+    mem_put(lock);
+}
+
+int32_t
+ec_lock_compare(ec_lock_t *lock1, ec_lock_t *lock2)
+{
+    return gf_uuid_compare(lock1->loc.gfid, lock2->loc.gfid);
+}
+
+static void
+ec_lock_insert(ec_fop_data_t *fop, ec_lock_t *lock, uint32_t flags, loc_t *base,
+               off_t fl_start, uint64_t fl_size)
+{
+    ec_lock_link_t *link;
+
+    /* This check is only prepared for up to 2 locks per fop. If more locks
+     * are needed this must be changed. */
+    if ((fop->lock_count > 0) &&
+        (ec_lock_compare(fop->locks[0].lock, lock) < 0)) {
+        fop->first_lock = fop->lock_count;
+    } else {
+        /* When the first lock is added to the current fop, request lock
+         * counts from locks xlator to be able to determine if there is
+         * contention and release the lock sooner. */
+        if (fop->xdata == NULL) {
+            fop->xdata = dict_new();
+            if (fop->xdata == NULL) {
+                ec_fop_set_error(fop, ENOMEM);
+                return;
+            }
+        }
+        if (dict_set_str(fop->xdata, GLUSTERFS_INODELK_DOM_COUNT,
+                         fop->xl->name) != 0) {
+            ec_fop_set_error(fop, ENOMEM);
+            return;
+        }
+    }
+
+    link = &fop->locks[fop->lock_count++];
+
+    link->lock = lock;
+    link->fop = fop;
+    link->update[EC_DATA_TXN] = (flags & EC_UPDATE_DATA) != 0;
+    link->update[EC_METADATA_TXN] = (flags & EC_UPDATE_META) != 0;
+    link->base = base;
+    link->fl_start = fl_start;
+    link->fl_end = ec_range_end_get(fl_start, fl_size);
+
+    lock->refs_pending++;
+}
+
+static void
+ec_lock_prepare_inode_internal(ec_fop_data_t *fop, loc_t *loc, uint32_t flags,
+                               loc_t *base, off_t fl_start, uint64_t fl_size)
+{
+    ec_lock_t *lock = NULL;
+    ec_inode_t *ctx;
+
+    if ((fop->parent != NULL) || (fop->error != 0) || (loc->inode == NULL)) {
+        return;
+    }
+
+    LOCK(&loc->inode->lock);
+
+    ctx = __ec_inode_get(loc->inode, fop->xl);
+    if (ctx == NULL) {
+        __ec_fop_set_error(fop, ENOMEM);
+
+        goto unlock;
+    }
+
+    if (ctx->inode_lock != NULL) {
+        lock = ctx->inode_lock;
+
+        /* If there's another lock, make sure that it's not the same. Otherwise
+         * do not insert it.
+         *
+         * This can only happen on renames where source and target names are
+         * in the same directory. */
+        if ((fop->lock_count > 0) && (fop->locks[0].lock == lock)) {
+            /* Combine data/meta updates */
+            fop->locks[0].update[EC_DATA_TXN] |= (flags & EC_UPDATE_DATA) != 0;
+            fop->locks[0].update[EC_METADATA_TXN] |= (flags & EC_UPDATE_META) !=
+                                                     0;
+
+            /* Only one base inode is allowed per fop, so there shouldn't be
+             * overwrites here. */
+            if (base != NULL) {
+                fop->locks[0].base = base;
+            }
+
+            goto update_query;
+        }
+
+        ec_trace("LOCK_INODELK", fop,
+                 "lock=%p, inode=%p. Lock already "
+                 "acquired",
+                 lock, loc->inode);
+
+        goto insert;
+    }
+
+    lock = ec_lock_allocate(fop, loc);
+    if (lock == NULL) {
+        goto unlock;
+    }
+
+    ec_trace("LOCK_CREATE", fop, "lock=%p", lock);
+
+    lock->flock.l_type = F_WRLCK;
+    lock->flock.l_whence = SEEK_SET;
+
+    lock->ctx = ctx;
+    ctx->inode_lock = lock;
+
+insert:
+    ec_lock_insert(fop, lock, flags, base, fl_start, fl_size);
+update_query:
+    lock->query |= (flags & EC_QUERY_INFO) != 0;
+unlock:
+    UNLOCK(&loc->inode->lock);
+}
+
+void
+ec_lock_prepare_inode(ec_fop_data_t *fop, loc_t *loc, uint32_t flags,
+                      off_t fl_start, uint64_t fl_size)
+{
+    ec_lock_prepare_inode_internal(fop, loc, flags, NULL, fl_start, fl_size);
+}
+
+void
+ec_lock_prepare_parent_inode(ec_fop_data_t *fop, loc_t *loc, loc_t *base,
+                             uint32_t flags)
+{
+    loc_t tmp;
+    int32_t err;
+
+    if (fop->error != 0) {
+        return;
+    }
+
+    err = ec_loc_parent(fop->xl, loc, &tmp);
+    if (err != 0) {
+        ec_fop_set_error(fop, -err);
+
+        return;
+    }
+
+    if ((flags & EC_INODE_SIZE) != 0) {
+        flags ^= EC_INODE_SIZE;
+    } else {
+        base = NULL;
+    }
+
+    ec_lock_prepare_inode_internal(fop, &tmp, flags, base, 0, EC_RANGE_FULL);
+
+    loc_wipe(&tmp);
+}
+
+void
+ec_lock_prepare_fd(ec_fop_data_t *fop, fd_t *fd, uint32_t flags, off_t fl_start,
+                   uint64_t fl_size)
+{
+    loc_t loc;
+    int32_t err;
+
+    if (fop->error != 0) {
+        return;
+    }
+
+    err = ec_loc_from_fd(fop->xl, &loc, fd);
+    if (err != 0) {
+        ec_fop_set_error(fop, -err);
+
+        return;
+    }
+
+    ec_lock_prepare_inode_internal(fop, &loc, flags, NULL, fl_start, fl_size);
+
+    loc_wipe(&loc);
+}
+
+gf_boolean_t
+ec_config_check(xlator_t *xl, ec_config_t *config)
+{
+    ec_t *ec;
+
+    ec = xl->private;
+    if ((config->version != EC_CONFIG_VERSION) ||
+        (config->algorithm != EC_CONFIG_ALGORITHM) ||
+        (config->gf_word_size != EC_GF_BITS) || (config->bricks != ec->nodes) ||
+        (config->redundancy != ec->redundancy) ||
+        (config->chunk_size != EC_METHOD_CHUNK_SIZE)) {
+        uint32_t data_bricks;
+
+        /* This combination of version/algorithm requires the following
+           values. Incorrect values for these fields are a sign of
+           corruption:
+
+             redundancy > 0
+             redundancy * 2 < bricks
+             gf_word_size must be a power of 2
+             chunk_size (in bits) must be a multiple of gf_word_size *
+                 (bricks - redundancy) */
+
+        data_bricks = config->bricks - config->redundancy;
+        if ((config->redundancy < 1) ||
+            (config->redundancy * 2 >= config->bricks) ||
+            !ec_is_power_of_2(config->gf_word_size) ||
+            ((config->chunk_size * 8) % (config->gf_word_size * data_bricks) !=
+             0)) {
+            gf_msg(xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_INVALID_CONFIG,
+                   "Invalid or corrupted config");
+        } else {
+            gf_msg(xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_INVALID_CONFIG,
+                   "Unsupported config "
+                   "(V=%u, A=%u, W=%u, "
+                   "N=%u, R=%u, S=%u)",
+                   config->version, config->algorithm, config->gf_word_size,
+                   config->bricks, config->redundancy, config->chunk_size);
+        }
+
+        return _gf_false;
+    }
+
+    return _gf_true;
+}
+
+gf_boolean_t
+ec_set_dirty_flag(ec_lock_link_t *link, ec_inode_t *ctx, uint64_t *dirty)
+{
+    gf_boolean_t set_dirty = _gf_false;
+
+    if (link->update[EC_DATA_TXN] && !ctx->dirty[EC_DATA_TXN]) {
+        if (!link->optimistic_changelog)
+            dirty[EC_DATA_TXN] = 1;
+    }
+
+    if (link->update[EC_METADATA_TXN] && !ctx->dirty[EC_METADATA_TXN]) {
+        if (!link->optimistic_changelog)
+            dirty[EC_METADATA_TXN] = 1;
+    }
+
+    if (dirty[EC_METADATA_TXN] || dirty[EC_DATA_TXN]) {
+        set_dirty = _gf_true;
+    }
+
+    return set_dirty;
+}
+
+int32_t
+ec_prepare_update_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *dict,
+                      dict_t *xdata)
+{
+    struct list_head list;
+    ec_fop_data_t *fop = cookie, *parent, *tmp;
+    ec_lock_link_t *parent_link = fop->data;
+    ec_lock_link_t *link = NULL;
+    ec_lock_t *lock = NULL;
+    ec_inode_t *ctx;
+    gf_boolean_t release = _gf_false;
+    uint64_t provided_flags = 0;
+    uint64_t dirty[EC_VERSION_SIZE] = {0, 0};
+    lock = parent_link->lock;
+    parent = parent_link->fop;
+    ctx = lock->ctx;
+
+    INIT_LIST_HEAD(&list);
+    provided_flags = EC_PROVIDED_FLAGS(parent_link->waiting_flags);
+
+    LOCK(&lock->loc.inode->lock);
+
+    list_for_each_entry(link, &lock->owners, owner_list)
+    {
+        if ((link->waiting_flags & provided_flags) != 0) {
+            link->waiting_flags ^= (link->waiting_flags & provided_flags);
+            if (EC_NEEDED_FLAGS(link->waiting_flags) == 0)
+                list_add_tail(&link->fop->cbk_list, &list);
+        }
+    }
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, EC_MSG_SIZE_VERS_GET_FAIL,
+               "Failed to get size and version :  %s", ec_msg_str(fop));
+
+        goto unlock;
+    }
+
+    if (EC_FLAGS_HAVE(provided_flags, EC_FLAG_XATTROP)) {
+        op_errno = -ec_dict_del_array(dict, EC_XATTR_VERSION, ctx->pre_version,
+                                      EC_VERSION_SIZE);
+        if (op_errno != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                   EC_MSG_VER_XATTR_GET_FAIL, "Unable to get version xattr. %s",
+                   ec_msg_str(fop));
+            goto unlock;
+        }
+        ctx->post_version[0] += ctx->pre_version[0];
+        ctx->post_version[1] += ctx->pre_version[1];
+
+        ctx->have_version = _gf_true;
+
+        if (lock->loc.inode->ia_type == IA_IFREG ||
+            lock->loc.inode->ia_type == IA_INVAL) {
+            op_errno = -ec_dict_del_number(dict, EC_XATTR_SIZE, &ctx->pre_size);
+            if (op_errno != 0) {
+                if (lock->loc.inode->ia_type == IA_IFREG) {
+                    gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                           EC_MSG_SIZE_XATTR_GET_FAIL,
+                           "Unable to get size xattr. %s", ec_msg_str(fop));
+                    goto unlock;
+                }
+            } else {
+                ctx->post_size = ctx->pre_size;
+
+                ctx->have_size = _gf_true;
+            }
+
+            op_errno = -ec_dict_del_config(dict, EC_XATTR_CONFIG, &ctx->config);
+            if (op_errno != 0) {
+                if ((lock->loc.inode->ia_type == IA_IFREG) ||
+                    (op_errno != ENODATA)) {
+                    gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                           EC_MSG_CONFIG_XATTR_GET_FAIL,
+                           "Unable to get config xattr. %s", ec_msg_str(fop));
+
+                    goto unlock;
+                }
+            } else {
+                if (!ec_config_check(parent->xl, &ctx->config)) {
+                    gf_msg(this->name, GF_LOG_ERROR, EINVAL,
+                           EC_MSG_CONFIG_XATTR_INVALID, "Invalid config xattr");
+
+                    op_errno = EINVAL;
+
+                    goto unlock;
+                }
+                ctx->have_config = _gf_true;
+            }
+        }
+        ctx->have_info = _gf_true;
+    }
+
+    ec_set_dirty_flag(fop->data, ctx, dirty);
+    if (dirty[EC_METADATA_TXN] &&
+        (EC_FLAGS_HAVE(provided_flags, EC_FLAG_METADATA_DIRTY))) {
+        GF_ASSERT(!ctx->dirty[EC_METADATA_TXN]);
+        ctx->dirty[EC_METADATA_TXN] = 1;
+    }
+
+    if (dirty[EC_DATA_TXN] &&
+        (EC_FLAGS_HAVE(provided_flags, EC_FLAG_DATA_DIRTY))) {
+        GF_ASSERT(!ctx->dirty[EC_DATA_TXN]);
+        ctx->dirty[EC_DATA_TXN] = 1;
+    }
+    op_errno = 0;
+unlock:
+
+    lock->waiting_flags ^= provided_flags;
+
+    if (op_errno == 0) {
+        /* If the fop fails on any of the good bricks, it is important to mark
+         * it dirty and update versions right away if dirty was not set before.
+         */
+        if (lock->good_mask & ~(fop->good | fop->remaining)) {
+            release = _gf_true;
+        }
+
+        if (parent_link->update[0] && !parent_link->dirty[0]) {
+            lock->release |= release;
+        }
+
+        if (parent_link->update[1] && !parent_link->dirty[1]) {
+            lock->release |= release;
+        }
+
+        /* We don't allow the main fop to be executed on bricks that have not
+         * succeeded the initial xattrop. */
+        ec_lock_update_good(lock, fop);
+
+        /*As of now only data healing marks bricks as healing*/
+        lock->healing |= fop->healing;
+    }
+
+    UNLOCK(&lock->loc.inode->lock);
+
+    while (!list_empty(&list)) {
+        tmp = list_entry(list.next, ec_fop_data_t, cbk_list);
+        list_del_init(&tmp->cbk_list);
+
+        if (op_errno == 0) {
+            tmp->mask &= fop->good;
+
+            /*As of now only data healing marks bricks as healing*/
+            if (ec_is_data_fop(tmp->id)) {
+                tmp->healing |= fop->healing;
+            }
+        }
+
+        ec_resume(tmp, op_errno);
+    }
+
+    return 0;
+}
+
+static gf_boolean_t
+ec_set_needed_flag(ec_lock_t *lock, ec_lock_link_t *link, uint64_t flag)
+{
+    uint64_t current;
+
+    link->waiting_flags |= EC_FLAG_NEEDS(flag);
+
+    current = EC_NEEDED_FLAGS(lock->waiting_flags);
+    if (!EC_FLAGS_HAVE(current, flag)) {
+        lock->waiting_flags |= EC_FLAG_NEEDS(flag);
+        link->waiting_flags |= EC_FLAG_PROVIDES(flag);
+
+        return _gf_true;
+    }
+
+    return _gf_false;
+}
+
+static uint64_t
+ec_set_xattrop_flags_and_params(ec_lock_t *lock, ec_lock_link_t *link,
+                                uint64_t *dirty)
+{
+    uint64_t oldflags = 0;
+    uint64_t newflags = 0;
+    ec_inode_t *ctx = lock->ctx;
+
+    oldflags = EC_NEEDED_FLAGS(lock->waiting_flags);
+
+    if (lock->query && !ctx->have_info) {
+        ec_set_needed_flag(lock, link, EC_FLAG_XATTROP);
+    }
+
+    if (dirty[EC_DATA_TXN]) {
+        if (!ec_set_needed_flag(lock, link, EC_FLAG_DATA_DIRTY)) {
+            dirty[EC_DATA_TXN] = 0;
+        }
+    }
+
+    if (dirty[EC_METADATA_TXN]) {
+        if (!ec_set_needed_flag(lock, link, EC_FLAG_METADATA_DIRTY)) {
+            dirty[EC_METADATA_TXN] = 0;
+        }
+    }
+    newflags = EC_NEEDED_FLAGS(lock->waiting_flags);
+
+    return oldflags ^ newflags;
+}
+
+void
+ec_get_size_version(ec_lock_link_t *link)
+{
+    loc_t loc;
+    ec_lock_t *lock;
+    ec_inode_t *ctx;
+    ec_fop_data_t *fop;
+    dict_t *dict = NULL;
+    dict_t *xdata = NULL;
+    ec_t *ec = NULL;
+    int32_t error = 0;
+    gf_boolean_t set_dirty = _gf_false;
+    uint64_t allzero[EC_VERSION_SIZE] = {0, 0};
+    uint64_t dirty[EC_VERSION_SIZE] = {0, 0};
+    lock = link->lock;
+    ctx = lock->ctx;
+    fop = link->fop;
+    ec = fop->xl->private;
+    uint64_t changed_flags = 0;
+
+    if (ec->optimistic_changelog && !(ec->node_mask & ~link->lock->good_mask) &&
+        !ec_is_data_fop(fop->id))
+        link->optimistic_changelog = _gf_true;
+
+    memset(&loc, 0, sizeof(loc));
+
+    LOCK(&lock->loc.inode->lock);
+
+    set_dirty = ec_set_dirty_flag(link, ctx, dirty);
+
+    /* If ec metadata has already been retrieved, do not try again. */
+    if (ctx->have_info) {
+        if (ec_is_data_fop(fop->id)) {
+            fop->healing |= lock->healing;
+        }
+        if (!set_dirty)
+            goto unlock;
+    }
+
+    /* Determine if there's something we need to retrieve for the current
+     * operation. */
+    if (!set_dirty && !lock->query && (lock->loc.inode->ia_type != IA_IFREG) &&
+        (lock->loc.inode->ia_type != IA_INVAL)) {
+        goto unlock;
+    }
+
+    changed_flags = ec_set_xattrop_flags_and_params(lock, link, dirty);
+    if (link->waiting_flags) {
+        /* This fop needs to wait until all its flags are cleared which
+         * potentially can be cleared by other xattrops that are already
+         * wound*/
+        ec_sleep(fop);
+    } else {
+        GF_ASSERT(!changed_flags);
+    }
+
+unlock:
+    UNLOCK(&lock->loc.inode->lock);
+
+    if (!changed_flags)
+        goto out;
+
+    dict = dict_new();
+    if (dict == NULL) {
+        error = -ENOMEM;
+        goto out;
+    }
+
+    if (EC_FLAGS_HAVE(changed_flags, EC_FLAG_XATTROP)) {
+        /* Once we know that an xattrop will be needed,
+         * we try to get all available information in a
+         * single call. */
+        error = ec_dict_set_array(dict, EC_XATTR_VERSION, allzero,
+                                  EC_VERSION_SIZE);
+        if (error != 0) {
+            goto out;
+        }
+
+        if (lock->loc.inode->ia_type == IA_IFREG ||
+            lock->loc.inode->ia_type == IA_INVAL) {
+            error = ec_dict_set_number(dict, EC_XATTR_SIZE, 0);
+            if (error == 0) {
+                error = ec_dict_set_number(dict, EC_XATTR_CONFIG, 0);
+            }
+            if (error != 0) {
+                goto out;
+            }
+
+            xdata = dict_new();
+            if (xdata == NULL || dict_set_int32(xdata, GF_GET_SIZE, 1)) {
+                error = -ENOMEM;
+                goto out;
+            }
+        }
+    }
+
+    if (memcmp(allzero, dirty, sizeof(allzero))) {
+        error = ec_dict_set_array(dict, EC_XATTR_DIRTY, dirty, EC_VERSION_SIZE);
+        if (error != 0) {
+            goto out;
+        }
+    }
+
+    fop->frame->root->uid = 0;
+    fop->frame->root->gid = 0;
+
+    /* For normal fops, ec_[f]xattrop() must succeed on at least
+     * EC_MINIMUM_MIN bricks, however when this is called as part of a
+     * self-heal operation the mask of target bricks (fop->mask) could
+     * contain less than EC_MINIMUM_MIN bricks, causing the xattrop to
+     * always fail. Thus we always use the same minimum used for the main
+     * fop.
+     */
+    if (lock->fd == NULL) {
+        error = ec_loc_from_loc(fop->xl, &loc, &lock->loc);
+        if (error != 0) {
+            goto out;
+        }
+        if (gf_uuid_is_null(loc.pargfid)) {
+            if (loc.parent != NULL) {
+                inode_unref(loc.parent);
+                loc.parent = NULL;
+            }
+            GF_FREE((char *)loc.path);
+            loc.path = NULL;
+            loc.name = NULL;
+        }
+
+        ec_xattrop(fop->frame, fop->xl, fop->mask, fop->minimum,
+                   ec_prepare_update_cbk, link, &loc, GF_XATTROP_ADD_ARRAY64,
+                   dict, xdata);
+    } else {
+        ec_fxattrop(fop->frame, fop->xl, fop->mask, fop->minimum,
+                    ec_prepare_update_cbk, link, lock->fd,
+                    GF_XATTROP_ADD_ARRAY64, dict, xdata);
+    }
+
+    error = 0;
+
+out:
+    fop->frame->root->uid = fop->uid;
+    fop->frame->root->gid = fop->gid;
+
+    loc_wipe(&loc);
+
+    if (dict != NULL) {
+        dict_unref(dict);
+    }
+
+    if (xdata != NULL) {
+        dict_unref(xdata);
+    }
+
+    if (error != 0) {
+        ec_fop_set_error(fop, -error);
+    }
+}
+
+gf_boolean_t
+__ec_get_inode_size(ec_fop_data_t *fop, inode_t *inode, uint64_t *size)
+{
+    ec_inode_t *ctx;
+    gf_boolean_t found = _gf_false;
+
+    ctx = __ec_inode_get(inode, fop->xl);
+    if (ctx == NULL) {
+        goto out;
+    }
+
+    if (ctx->have_size) {
+        *size = ctx->post_size;
+        found = _gf_true;
+    }
+
+out:
+    return found;
+}
+
+gf_boolean_t
+ec_get_inode_size(ec_fop_data_t *fop, inode_t *inode, uint64_t *size)
+{
+    gf_boolean_t found = _gf_false;
+
+    LOCK(&inode->lock);
+    {
+        found = __ec_get_inode_size(fop, inode, size);
+    }
+    UNLOCK(&inode->lock);
+
+    return found;
+}
+
+gf_boolean_t
+__ec_set_inode_size(ec_fop_data_t *fop, inode_t *inode, uint64_t size)
+{
+    ec_inode_t *ctx;
+    gf_boolean_t found = _gf_false;
+
+    ctx = __ec_inode_get(inode, fop->xl);
+    if (ctx == NULL) {
+        goto out;
+    }
+
+    /* Normal fops always have ctx->have_size set. However self-heal calls this
+     * to prepare the inode, so ctx->have_size will be false. In this case we
+     * prepare both pre_size and post_size, and set have_size and have_info to
+     * true. */
+    if (!ctx->have_size) {
+        ctx->pre_size = size;
+        ctx->have_size = ctx->have_info = _gf_true;
+    }
+    ctx->post_size = size;
+
+    found = _gf_true;
+
+out:
+    return found;
+}
+
+gf_boolean_t
+ec_set_inode_size(ec_fop_data_t *fop, inode_t *inode, uint64_t size)
+{
+    gf_boolean_t found = _gf_false;
+
+    LOCK(&inode->lock);
+    {
+        found = __ec_set_inode_size(fop, inode, size);
+    }
+    UNLOCK(&inode->lock);
+
+    return found;
+}
+
+static void
+ec_release_stripe_cache(ec_inode_t *ctx)
+{
+    ec_stripe_list_t *stripe_cache = NULL;
+    ec_stripe_t *stripe = NULL;
+
+    stripe_cache = &ctx->stripe_cache;
+    while (!list_empty(&stripe_cache->lru)) {
+        stripe = list_first_entry(&stripe_cache->lru, ec_stripe_t, lru);
+        list_del(&stripe->lru);
+        GF_FREE(stripe);
+    }
+    stripe_cache->count = 0;
+    stripe_cache->max = 0;
+}
+
+void
+ec_clear_inode_info(ec_fop_data_t *fop, inode_t *inode)
+{
+    ec_inode_t *ctx;
+
+    LOCK(&inode->lock);
+
+    ctx = __ec_inode_get(inode, fop->xl);
+    if (ctx == NULL) {
+        goto unlock;
+    }
+
+    ec_release_stripe_cache(ctx);
+    ctx->have_info = _gf_false;
+    ctx->have_config = _gf_false;
+    ctx->have_version = _gf_false;
+    ctx->have_size = _gf_false;
+
+    memset(&ctx->config, 0, sizeof(ctx->config));
+    memset(ctx->pre_version, 0, sizeof(ctx->pre_version));
+    memset(ctx->post_version, 0, sizeof(ctx->post_version));
+    ctx->pre_size = ctx->post_size = 0;
+    memset(ctx->dirty, 0, sizeof(ctx->dirty));
+
+unlock:
+    UNLOCK(&inode->lock);
+}
+
+int32_t
+ec_get_real_size_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, inode_t *inode,
+                     struct iatt *buf, dict_t *xdata, struct iatt *postparent)
+{
+    ec_fop_data_t *fop = cookie;
+    ec_lock_link_t *link;
+
+    if (op_ret >= 0) {
+        link = fop->data;
+        link->size = buf->ia_size;
+    } else {
+        /* Prevent failure of parent fop. */
+        fop->error = 0;
+    }
+
+    return 0;
+}
+
+/* This function is used to get the trusted.ec.size xattr from a file when
+ * no lock is needed on the inode. This is only required to maintain iatt
+ * structs on fops that manipulate directory entries but do not operate
+ * directly on the inode, like link, rename, ...
+ *
+ * Any error processing this request is ignored. In the worst case, an invalid
+ * or not up to date value in the iatt could cause some cache invalidation.
+ */
+void
+ec_get_real_size(ec_lock_link_t *link)
+{
+    ec_fop_data_t *fop;
+    dict_t *xdata;
+
+    if (link->base == NULL || link->base->inode == NULL) {
+        return;
+    }
+
+    if (link->base->inode->ia_type != IA_IFREG) {
+        return;
+    }
+
+    fop = link->fop;
+
+    if (ec_get_inode_size(fop, link->base->inode, &link->size)) {
+        return;
+    }
+
+    xdata = dict_new();
+    if (xdata == NULL) {
+        return;
+    }
+    if (ec_dict_set_number(xdata, EC_XATTR_SIZE, 0) != 0) {
+        goto out;
+    }
+
+    /* Send a simple lookup. A single answer is considered ok since this value
+     * is only used to return an iatt struct related to an inode that is not
+     * locked and have not suffered any operation. */
+    ec_lookup(fop->frame, fop->xl, fop->mask, 1, ec_get_real_size_cbk, link,
+              link->base, xdata);
+
+out:
+    if (xdata != NULL) {
+        dict_unref(xdata);
+    }
+}
+
+static void
+ec_lock_update_fd(ec_lock_t *lock, ec_fop_data_t *fop)
+{
+    /* If the fop has an fd available, attach it to the lock structure to be
+     * able to do fxattrop calls instead of xattrop. */
+    if (fop->use_fd && (lock->fd == NULL)) {
+        lock->fd = __fd_ref(fop->fd);
+    }
+}
+
+static gf_boolean_t
+ec_link_has_lock_conflict(ec_lock_link_t *link, gf_boolean_t waitlist_check)
+{
+    ec_lock_link_t *trav_link = NULL;
+
+    list_for_each_entry(trav_link, &link->lock->owners, owner_list)
+    {
+        if (ec_lock_conflict(trav_link, link))
+            return _gf_true;
+    }
+
+    if (!waitlist_check)
+        return _gf_false;
+
+    list_for_each_entry(trav_link, &link->lock->waiting, wait_list)
+    {
+        if (ec_lock_conflict(trav_link, link))
+            return _gf_true;
+    }
+
+    return _gf_false;
+}
+
+static void
+ec_lock_wake_shared(ec_lock_t *lock, struct list_head *list)
+{
+    ec_fop_data_t *fop;
+    ec_lock_link_t *link;
+    gf_boolean_t conflict = _gf_false;
+
+    while (!conflict && !list_empty(&lock->waiting)) {
+        link = list_entry(lock->waiting.next, ec_lock_link_t, wait_list);
+        fop = link->fop;
+
+        /* If lock is not acquired, at most one fop can be assigned as owner.
+         * The following fops will need to wait in the lock->waiting queue
+         * until the lock has been fully acquired. */
+        conflict = !lock->acquired;
+
+        /* If the fop is not shareable, only this fop can be assigned as owner.
+         * Other fops will need to wait until this one finishes. */
+        if (ec_link_has_lock_conflict(link, _gf_false)) {
+            conflict = _gf_true;
+        }
+
+        /* If only one fop is allowed, it can be assigned as the owner of the
+         * lock only if there weren't any other owner. */
+        if (conflict && !list_empty(&lock->owners)) {
+            break;
+        }
+
+        list_move_tail(&link->wait_list, list);
+
+        list_add_tail(&link->owner_list, &lock->owners);
+        lock->refs_owners++;
+
+        ec_lock_update_fd(lock, fop);
+    }
+}
+
+static void
+ec_lock_apply(ec_lock_link_t *link)
+{
+    ec_fop_data_t *fop = link->fop;
+
+    fop->mask &= link->lock->good_mask;
+    fop->locked++;
+
+    ec_get_size_version(link);
+    ec_get_real_size(link);
+}
+
+gf_boolean_t
+ec_lock_acquire(ec_lock_link_t *link);
+
+static void
+ec_lock_resume_shared(struct list_head *list)
+{
+    ec_lock_link_t *link;
+
+    while (!list_empty(list)) {
+        link = list_entry(list->next, ec_lock_link_t, wait_list);
+        list_del_init(&link->wait_list);
+
+        if (link->lock->acquired) {
+            ec_lock_apply(link);
+            ec_lock(link->fop);
+        } else {
+            GF_ASSERT(list_empty(list));
+
+            ec_lock_acquire(link);
+        }
+
+        ec_resume(link->fop, 0);
+    }
+}
+
+void
+ec_lock_acquired(ec_lock_link_t *link)
+{
+    struct list_head list;
+    ec_lock_t *lock;
+    ec_fop_data_t *fop;
+
+    lock = link->lock;
+    fop = link->fop;
+
+    ec_trace("LOCKED", fop, "lock=%p", lock);
+
+    INIT_LIST_HEAD(&list);
+
+    LOCK(&lock->loc.inode->lock);
+
+    lock->acquired = _gf_true;
+    if (lock->contention) {
+        lock->release = _gf_true;
+        lock->contention = _gf_false;
+    }
+
+    ec_lock_update_fd(lock, fop);
+    ec_lock_wake_shared(lock, &list);
+
+    UNLOCK(&lock->loc.inode->lock);
+
+    ec_lock_apply(link);
+
+    if (fop->use_fd &&
+        (link->update[EC_DATA_TXN] || link->update[EC_METADATA_TXN])) {
+        /* Try to reopen closed fd's only if lock has succeeded. */
+        ec_fix_open(fop, lock->mask);
+    }
+
+    ec_lock_resume_shared(&list);
+}
+
+int32_t
+ec_locked(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+          int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = cookie;
+    ec_lock_link_t *link = NULL;
+    ec_lock_t *lock = NULL;
+
+    link = fop->data;
+    lock = link->lock;
+    if (op_ret >= 0) {
+        lock->mask = lock->good_mask = fop->good;
+        lock->healing = 0;
+
+        ec_lock_acquired(link);
+        ec_lock(fop->parent);
+    } else {
+        LOCK(&lock->loc.inode->lock);
+        {
+            lock->contention = _gf_false;
+        }
+        UNLOCK(&lock->loc.inode->lock);
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, EC_MSG_PREOP_LOCK_FAILED,
+               "Failed to complete preop lock");
+    }
+
+    return 0;
+}
+
+gf_boolean_t
+ec_lock_acquire(ec_lock_link_t *link)
+{
+    ec_lock_t *lock;
+    ec_fop_data_t *fop;
+    gf_lkowner_t lk_owner;
+
+    lock = link->lock;
+    fop = link->fop;
+
+    if (!lock->acquired) {
+        set_lk_owner_from_ptr(&lk_owner, lock);
+
+        ec_trace("LOCK_ACQUIRE", fop, "lock=%p, inode=%p", lock,
+                 lock->loc.inode);
+
+        lock->flock.l_type = F_WRLCK;
+        ec_inodelk(fop->frame, fop->xl, &lk_owner, -1, EC_MINIMUM_ALL,
+                   ec_locked, link, fop->xl->name, &lock->loc, F_SETLKW,
+                   &lock->flock, NULL);
+
+        return _gf_false;
+    }
+
+    ec_trace("LOCK_REUSE", fop, "lock=%p", lock);
+
+    ec_lock_acquired(link);
+
+    return _gf_true;
+}
+
+static ec_lock_link_t *
+ec_lock_timer_cancel(xlator_t *xl, ec_lock_t *lock)
+{
+    ec_lock_link_t *timer_link;
+
+    /* If we don't have any timer, there's nothing to cancel. */
+    if (lock->timer == NULL) {
+        return NULL;
+    }
+
+    /* We are trying to access a lock that has an unlock timer active.
+     * This means that the lock must be idle, i.e. no fop can be in the
+     * owner, waiting or frozen lists. It also means that the lock cannot
+     * have been marked as being released (this is done without timers).
+     * There should only be one owner reference, but it's possible that
+     * some fops are being prepared to use this lock. */
+    GF_ASSERT((lock->refs_owners == 1) && list_empty(&lock->owners) &&
+              list_empty(&lock->waiting));
+
+    /* We take the timer_link before cancelling the timer, since a
+     * successful cancellation will destroy it. It must not be NULL
+     * because it references the fop responsible for the delayed unlock
+     * that we are currently trying to cancel. */
+    timer_link = lock->timer->data;
+    GF_ASSERT(timer_link != NULL);
+
+    if (gf_timer_call_cancel(xl->ctx, lock->timer) < 0) {
+        /* It's too late to avoid the execution of the timer callback.
+         * Since we need to be sure that the callback has access to all
+         * needed resources, we cannot resume the execution of the
+         * timer fop now. This will be done in the callback. */
+        timer_link = NULL;
+    } else {
+        /* The timer has been cancelled. The fop referenced by
+         * timer_link holds the last reference. The caller is
+         * responsible to release it when not needed anymore. */
+        ec_trace("UNLOCK_CANCELLED", timer_link->fop, "lock=%p", lock);
+    }
+
+    /* We have two options here:
+     *
+     * 1. The timer has been successfully cancelled.
+     *
+     *    This is the easiest case and we can continue with the currently
+     *    acquired lock.
+     *
+     * 2. The timer callback has already been fired.
+     *
+     *    In this case we have not been able to cancel the timer before
+     *    the timer callback has been fired, but we also know that
+     *    lock->timer != NULL. This means that the timer callback is still
+     *    trying to acquire the inode mutex that we currently own. We are
+     *    safe until we release it. In this case we can safely clear
+     *    lock->timer. This will cause that the timer callback does nothing
+     *    once it acquires the mutex.
+     */
+    lock->timer = NULL;
+
+    return timer_link;
+}
+
+static gf_boolean_t
+ec_lock_assign_owner(ec_lock_link_t *link)
+{
+    ec_fop_data_t *fop;
+    ec_lock_t *lock;
+    ec_lock_link_t *timer_link = NULL;
+    gf_boolean_t assigned = _gf_false;
+
+    /* The link cannot be in any list because we have just finished preparing
+     * it. */
+    GF_ASSERT(list_empty(&link->wait_list));
+
+    fop = link->fop;
+    lock = link->lock;
+
+    LOCK(&lock->loc.inode->lock);
+
+    /* Since the link has just been prepared but it's not active yet, the
+     * refs_pending must be one at least (the ref owned by this link). */
+    GF_ASSERT(lock->refs_pending > 0);
+    /* The link is not pending any more. It will be assigned to the owner,
+     * waiting or frozen list. */
+    lock->refs_pending--;
+
+    if (lock->release) {
+        ec_trace("LOCK_QUEUE_FREEZE", fop, "lock=%p", lock);
+
+        /* When lock->release is set, we'll unlock the lock as soon as
+         * possible, meaning that we won't use a timer. */
+        GF_ASSERT(lock->timer == NULL);
+
+        /* The lock is marked to be released. We can still have owners and fops
+         * in the waiting ilist f they have been added before the lock has been
+         * marked to be released. However new fops are put into the frozen list
+         * to wait for the next unlock/lock cycle. */
+        list_add_tail(&link->wait_list, &lock->frozen);
+
+        goto unlock;
+    }
+
+    /* The lock is not marked to be released, so the frozen list should be
+     * empty. */
+    GF_ASSERT(list_empty(&lock->frozen));
+
+    timer_link = ec_lock_timer_cancel(fop->xl, lock);
+
+    if (!list_empty(&lock->owners)) {
+        /* There are other owners of this lock. We can only take ownership if
+         * the lock is already acquired and doesn't have conflict with existing
+         * owners, or waiters(to prevent starvation).
+         * Otherwise we need to wait.
+         */
+        if (!lock->acquired || ec_link_has_lock_conflict(link, _gf_true)) {
+            ec_trace("LOCK_QUEUE_WAIT", fop, "lock=%p", lock);
+
+            list_add_tail(&link->wait_list, &lock->waiting);
+
+            goto unlock;
+        }
+    }
+
+    list_add_tail(&link->owner_list, &lock->owners);
+
+    /* If timer_link is not NULL, it means that we have inherited the owner
+     * reference assigned to the timer fop. In this case we simply reuse it.
+     * Otherwise we need to increase the number of owners. */
+    if (timer_link == NULL) {
+        lock->refs_owners++;
+    }
+
+    assigned = _gf_true;
+
+unlock:
+    if (!assigned) {
+        /* We have not been able to take ownership of this lock. The fop must
+         * be put to sleep. */
+        ec_sleep(fop);
+    }
+
+    UNLOCK(&lock->loc.inode->lock);
+
+    /* If we have cancelled the timer, we need to resume the fop that was
+     * waiting for it. */
+    if (timer_link != NULL) {
+        ec_resume(timer_link->fop, 0);
+    }
+
+    return assigned;
+}
+
+static void
+ec_lock_next_owner(ec_lock_link_t *link, ec_cbk_data_t *cbk,
+                   gf_boolean_t release)
+{
+    struct list_head list;
+    ec_lock_t *lock = link->lock;
+    ec_fop_data_t *fop = link->fop;
+    ec_inode_t *ctx = lock->ctx;
+
+    INIT_LIST_HEAD(&list);
+
+    LOCK(&lock->loc.inode->lock);
+
+    ec_trace("LOCK_DONE", fop, "lock=%p", lock);
+
+    /* Current link must belong to the owner list of the lock. We don't
+     * decrement lock->refs_owners here because the inode mutex is released
+     * before ec_unlock() is called and we need to know when the last owner
+     * unlocks the lock to do proper cleanup. lock->refs_owners is used for
+     * this task. */
+    GF_ASSERT((lock->refs_owners > 0) && !list_empty(&link->owner_list));
+    list_del_init(&link->owner_list);
+
+    lock->release |= release;
+
+    if ((fop->error == 0) && (cbk != NULL) && (cbk->op_ret >= 0)) {
+        if (link->update[0]) {
+            ctx->post_version[0]++;
+        }
+        if (link->update[1]) {
+            ctx->post_version[1]++;
+        }
+        /* If the fop fails on any of the good bricks, it is important to mark
+         * it dirty and update versions right away. */
+        if (link->update[0] || link->update[1]) {
+            if (lock->good_mask & ~(fop->good | fop->remaining)) {
+                lock->release = _gf_true;
+            }
+        }
+    }
+
+    if (fop->healing) {
+        lock->healing = fop->healing & (fop->good | fop->remaining);
+    }
+    ec_lock_update_good(lock, fop);
+
+    ec_lock_wake_shared(lock, &list);
+
+    UNLOCK(&lock->loc.inode->lock);
+
+    ec_lock_resume_shared(&list);
+}
+
+void
+ec_lock(ec_fop_data_t *fop)
+{
+    ec_lock_link_t *link;
+
+    /* There is a chance that ec_resume is called on fop even before ec_sleep.
+     * Which can result in refs == 0 for fop leading to use after free in this
+     * function when it calls ec_sleep so do ec_sleep at start and ec_resume at
+     * the end of this function.*/
+    ec_sleep(fop);
+
+    while (fop->locked < fop->lock_count) {
+        /* Since there are only up to 2 locks per fop, this xor will change
+         * the order of the locks if fop->first_lock is 1. */
+        link = &fop->locks[fop->locked ^ fop->first_lock];
+
+        if (!ec_lock_assign_owner(link) || !ec_lock_acquire(link)) {
+            break;
+        }
+    }
+
+    ec_resume(fop, 0);
+}
+
+void
+ec_lock_unfreeze(ec_lock_link_t *link)
+{
+    struct list_head list;
+    ec_lock_t *lock;
+    gf_boolean_t destroy = _gf_false;
+
+    lock = link->lock;
+
+    INIT_LIST_HEAD(&list);
+
+    LOCK(&lock->loc.inode->lock);
+
+    /* The lock must be marked to be released here, since we have just released
+     * it and any attempt to assign it to more fops must have added them to the
+     * frozen list. We can only have one active reference here: the one that
+     * is processing this unfreeze. */
+    GF_ASSERT(lock->release && (lock->refs_owners == 1));
+    lock->release = _gf_false;
+    lock->refs_owners = 0;
+
+    lock->acquired = _gf_false;
+
+    /* We are unfreezing a lock. This means that the lock has already been
+     * released. In this state it shouldn't have a pending timer nor have any
+     * owner, and the waiting list should be empty. Only the frozen list can
+     * contain some fop. */
+    GF_ASSERT((lock->timer == NULL) && list_empty(&lock->waiting) &&
+              list_empty(&lock->owners));
+
+    /* We move all frozen fops to the waiting list. */
+    list_splice_init(&lock->frozen, &lock->waiting);
+
+    /* If we don't have any fop waiting nor there are any prepared fops using
+     * this lock, we can finally dispose it. */
+    destroy = list_empty(&lock->waiting) && (lock->refs_pending == 0);
+    if (destroy) {
+        ec_trace("LOCK_DESTROY", link->fop, "lock=%p", lock);
+
+        lock->ctx->inode_lock = NULL;
+    } else {
+        ec_trace("LOCK_UNFREEZE", link->fop, "lock=%p", lock);
+
+        ec_lock_wake_shared(lock, &list);
+    }
+
+    UNLOCK(&lock->loc.inode->lock);
+
+    ec_lock_resume_shared(&list);
+
+    if (destroy) {
+        ec_lock_destroy(lock);
+    }
+}
+
+int32_t
+ec_unlocked(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = cookie;
+    ec_lock_link_t *link = fop->data;
+
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, EC_MSG_UNLOCK_FAILED,
+               "entry/inode unlocking failed :(%s)", ec_msg_str(link->fop));
+    } else {
+        ec_trace("UNLOCKED", link->fop, "lock=%p", link->lock);
+    }
+
+    ec_lock_unfreeze(link);
+
+    return 0;
+}
+
+void
+ec_unlock_lock(ec_lock_link_t *link)
+{
+    ec_lock_t *lock;
+    ec_fop_data_t *fop;
+    gf_lkowner_t lk_owner;
+
+    lock = link->lock;
+    fop = link->fop;
+
+    lock->unlock_now = _gf_false;
+    ec_clear_inode_info(fop, lock->loc.inode);
+
+    if ((lock->mask != 0) && lock->acquired) {
+        set_lk_owner_from_ptr(&lk_owner, lock);
+        lock->flock.l_type = F_UNLCK;
+        ec_trace("UNLOCK_INODELK", fop, "lock=%p, inode=%p", lock,
+                 lock->loc.inode);
+
+        ec_inodelk(fop->frame, fop->xl, &lk_owner, lock->mask, EC_MINIMUM_ONE,
+                   ec_unlocked, link, fop->xl->name, &lock->loc, F_SETLK,
+                   &lock->flock, NULL);
+    } else {
+        ec_lock_unfreeze(link);
+    }
+}
+
+void
+ec_inode_bad_inc(inode_t *inode, xlator_t *xl)
+{
+    ec_inode_t *ctx = NULL;
+
+    LOCK(&inode->lock);
+    {
+        ctx = __ec_inode_get(inode, xl);
+        if (ctx == NULL) {
+            goto unlock;
+        }
+        ctx->bad_version++;
+    }
+unlock:
+    UNLOCK(&inode->lock);
+}
+
+int32_t
+ec_update_size_version_done(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int32_t op_ret, int32_t op_errno, dict_t *xattr,
+                            dict_t *xdata)
+{
+    ec_fop_data_t *fop = cookie;
+    ec_lock_link_t *link;
+    ec_lock_t *lock;
+    ec_inode_t *ctx;
+
+    link = fop->data;
+    lock = link->lock;
+    ctx = lock->ctx;
+
+    if (op_ret < 0) {
+        if (link->lock->fd == NULL) {
+            ec_inode_bad_inc(link->lock->loc.inode, this);
+        } else {
+            ec_inode_bad_inc(link->lock->fd->inode, this);
+        }
+
+        gf_msg(fop->xl->name, fop_log_level(fop->id, op_errno), op_errno,
+               EC_MSG_SIZE_VERS_UPDATE_FAIL,
+               "Failed to update version and size. %s", ec_msg_str(fop));
+    } else {
+        fop->parent->good &= fop->good;
+
+        ec_lock_update_good(lock, fop);
+
+        if (ec_dict_del_array(xattr, EC_XATTR_VERSION, ctx->post_version,
+                              EC_VERSION_SIZE) == 0) {
+            ctx->pre_version[0] = ctx->post_version[0];
+            ctx->pre_version[1] = ctx->post_version[1];
+
+            ctx->have_version = _gf_true;
+        }
+        if (ec_dict_del_number(xattr, EC_XATTR_SIZE, &ctx->post_size) == 0) {
+            ctx->pre_size = ctx->post_size;
+
+            ctx->have_size = _gf_true;
+        }
+        if ((ec_dict_del_config(xdata, EC_XATTR_CONFIG, &ctx->config) == 0) &&
+            ec_config_check(fop->xl, &ctx->config)) {
+            ctx->have_config = _gf_true;
+        }
+
+        ctx->have_info = _gf_true;
+    }
+    /* If we are here because of fop's and other than unlock request,
+     * that means we are still holding a lock. That make sure
+     * lock->unlock_now can not be modified.
+     */
+    if (lock->unlock_now) {
+        ec_unlock_lock(fop->data);
+    }
+
+    return 0;
+}
+
+void
+ec_update_size_version(ec_lock_link_t *link, uint64_t *version, uint64_t size,
+                       uint64_t *dirty)
+{
+    ec_fop_data_t *fop;
+    ec_lock_t *lock;
+    ec_inode_t *ctx;
+    dict_t *dict = NULL;
+    uintptr_t update_on = 0;
+    int32_t err = -ENOMEM;
+
+    fop = link->fop;
+    lock = link->lock;
+    ctx = lock->ctx;
+
+    ec_trace("UPDATE", fop, "version=%ld/%ld, size=%ld, dirty=%ld/%ld",
+             version[0], version[1], size, dirty[0], dirty[1]);
+
+    dict = dict_new();
+    if (dict == NULL) {
+        goto out;
+    }
+
+    /* If we don't have version information or it has been modified, we
+     * update it. */
+    if (!ctx->have_version || (version[0] != 0) || (version[1] != 0)) {
+        err = ec_dict_set_array(dict, EC_XATTR_VERSION, version,
+                                EC_VERSION_SIZE);
+        if (err != 0) {
+            goto out;
+        }
+    }
+
+    if (size != 0) {
+        /* If size has been changed, we should already
+         * know the previous size of the file. */
+        GF_ASSERT(ctx->have_size);
+
+        err = ec_dict_set_number(dict, EC_XATTR_SIZE, size);
+        if (err != 0) {
+            goto out;
+        }
+    }
+
+    if (dirty[0] || dirty[1]) {
+        err = ec_dict_set_array(dict, EC_XATTR_DIRTY, dirty, EC_VERSION_SIZE);
+        if (err != 0) {
+            goto out;
+        }
+    }
+
+    /* If config information is not known, we request it now. */
+    if ((lock->loc.inode->ia_type == IA_IFREG) && !ctx->have_config) {
+        /* A failure requesting this xattr is ignored because it's not
+         * absolutely required right now. */
+        (void)ec_dict_set_number(dict, EC_XATTR_CONFIG, 0);
+    }
+
+    fop->frame->root->uid = 0;
+    fop->frame->root->gid = 0;
+
+    update_on = lock->good_mask | lock->healing;
+
+    if (link->lock->fd == NULL) {
+        ec_xattrop(fop->frame, fop->xl, update_on, EC_MINIMUM_MIN,
+                   ec_update_size_version_done, link, &link->lock->loc,
+                   GF_XATTROP_ADD_ARRAY64, dict, NULL);
+    } else {
+        ec_fxattrop(fop->frame, fop->xl, update_on, EC_MINIMUM_MIN,
+                    ec_update_size_version_done, link, link->lock->fd,
+                    GF_XATTROP_ADD_ARRAY64, dict, NULL);
+    }
+
+    fop->frame->root->uid = fop->uid;
+    fop->frame->root->gid = fop->gid;
+
+    dict_unref(dict);
+
+    return;
+
+out:
+    if (dict != NULL) {
+        dict_unref(dict);
+    }
+
+    ec_fop_set_error(fop, -err);
+
+    gf_msg(fop->xl->name, GF_LOG_ERROR, -err, EC_MSG_SIZE_VERS_UPDATE_FAIL,
+           "Unable to update version and size. %s", ec_msg_str(fop));
+
+    if (lock->unlock_now) {
+        ec_unlock_lock(fop->data);
+    }
+}
+
+gf_boolean_t
+ec_update_info(ec_lock_link_t *link)
+{
+    ec_lock_t *lock;
+    ec_inode_t *ctx;
+    uint64_t version[2] = {0, 0};
+    uint64_t dirty[2] = {0, 0};
+    uint64_t size;
+    ec_t *ec = NULL;
+    uintptr_t mask;
+
+    lock = link->lock;
+    ctx = lock->ctx;
+    ec = link->fop->xl->private;
+
+    /* pre_version[*] will be 0 if have_version is false */
+    version[EC_DATA_TXN] = ctx->post_version[EC_DATA_TXN] -
+                           ctx->pre_version[EC_DATA_TXN];
+    version[EC_METADATA_TXN] = ctx->post_version[EC_METADATA_TXN] -
+                               ctx->pre_version[EC_METADATA_TXN];
+
+    size = ctx->post_size - ctx->pre_size;
+    /* If we set the dirty flag for update fop, we have to unset it.
+     * If fop has failed on some bricks, leave the dirty as marked. */
+
+    if (lock->unlock_now) {
+        if (version[EC_DATA_TXN]) {
+            /*A data fop will have difference in post and pre version
+             *and for data fop we send writes on healing bricks also */
+            mask = lock->good_mask | lock->healing;
+        } else {
+            mask = lock->good_mask;
+        }
+        /* Ensure that nodes are up while doing final
+         * metadata update.*/
+        if (!(ec->node_mask & ~(mask)) && !(ec->node_mask & ~ec->xl_up)) {
+            if (ctx->dirty[EC_DATA_TXN] != 0) {
+                dirty[EC_DATA_TXN] = -1;
+            }
+            if (ctx->dirty[EC_METADATA_TXN] != 0) {
+                dirty[EC_METADATA_TXN] = -1;
+            }
+            /*If everything is fine and we already
+             *have version xattr set on entry, there
+             *is no need to update version again*/
+            if (ctx->pre_version[EC_DATA_TXN]) {
+                version[EC_DATA_TXN] = 0;
+            }
+            if (ctx->pre_version[EC_METADATA_TXN]) {
+                version[EC_METADATA_TXN] = 0;
+            }
+        } else {
+            link->optimistic_changelog = _gf_false;
+            ec_set_dirty_flag(link, ctx, dirty);
+        }
+        memset(ctx->dirty, 0, sizeof(ctx->dirty));
+    }
+
+    if ((version[EC_DATA_TXN] != 0) || (version[EC_METADATA_TXN] != 0) ||
+        (dirty[EC_DATA_TXN] != 0) || (dirty[EC_METADATA_TXN] != 0)) {
+        ec_update_size_version(link, version, size, dirty);
+        return _gf_true;
+    }
+
+    return _gf_false;
+}
+
+void
+ec_unlock_now(ec_lock_link_t *link)
+{
+    ec_lock_t *lock;
+    lock = link->lock;
+
+    ec_trace("UNLOCK_NOW", link->fop, "lock=%p", link->lock);
+    /*At this point, lock is not being used by any fop and
+     *can not be reused by any fop as it is going to be released.
+     *lock->unlock_now can not be modified at any other place.
+     */
+    lock->unlock_now = _gf_true;
+
+    if (!ec_update_info(link)) {
+        ec_unlock_lock(link);
+    }
+
+    ec_resume(link->fop, 0);
+}
+
+void
+ec_lock_release(ec_t *ec, inode_t *inode)
+{
+    ec_lock_t *lock;
+    ec_inode_t *ctx;
+    ec_lock_link_t *timer_link = NULL;
+
+    LOCK(&inode->lock);
+
+    ctx = __ec_inode_get(inode, ec->xl);
+    if (ctx == NULL) {
+        goto done;
+    }
+    lock = ctx->inode_lock;
+    if ((lock == NULL) || lock->release) {
+        goto done;
+    }
+
+    gf_msg_debug(ec->xl->name, 0, "Releasing inode %p due to lock contention",
+                 inode);
+
+    if (!lock->acquired) {
+        /* This happens if some bricks already got the lock while inodelk is in
+         * progress.  Set release to true after lock is acquired*/
+        lock->contention = _gf_true;
+        goto done;
+    }
+
+    /* The lock is not marked to be released, so the frozen list should be
+     * empty. */
+    GF_ASSERT(list_empty(&lock->frozen));
+
+    timer_link = ec_lock_timer_cancel(ec->xl, lock);
+
+    /* We mark the lock to be released as soon as possible. */
+    lock->release = _gf_true;
+
+done:
+    UNLOCK(&inode->lock);
+
+    /* If we have cancelled the timer, we need to start the unlock of the
+     * inode. If there was a timer but we have been unable to cancel it
+     * because it was just triggered, the timer callback will take care
+     * of releasing the inode. */
+    if (timer_link != NULL) {
+        ec_unlock_now(timer_link);
+    }
+}
+
+void
+ec_unlock_timer_add(ec_lock_link_t *link);
+
+void
+ec_unlock_timer_del(ec_lock_link_t *link)
+{
+    ec_lock_t *lock;
+    inode_t *inode;
+    gf_boolean_t now = _gf_false;
+
+    /* If we are here, it means that the timer has expired before having
+     * been cancelled. This guarantees that 'link' is still valid because
+     * the fop that contains it must be pending (if timer cancellation in
+     * ec_lock_assign_owner() fails, the fop is left sleeping).
+     *
+     * At the same time, the fop still has a reference to the lock, so
+     * it must also be valid.
+     */
+    lock = link->lock;
+
+    /* 'lock' must have a valid inode since it can only be destroyed
+     * when the lock itself is destroyed, but we have a reference to the
+     * lock to avoid this.
+     */
+    inode = lock->loc.inode;
+
+    LOCK(&inode->lock);
+
+    if (lock->timer != NULL) {
+        ec_trace("UNLOCK_DELAYED", link->fop, "lock=%p", lock);
+
+        /* The unlock timer has expired without anyone cancelling it.
+         * This means that it shouldn't have any owner, and the waiting
+         * and frozen lists should be empty.  It must have only one
+         * owner reference, but there can be fops being prepared
+         * though.
+         * */
+        GF_ASSERT(!lock->release && (lock->refs_owners == 1) &&
+                  list_empty(&lock->owners) && list_empty(&lock->waiting) &&
+                  list_empty(&lock->frozen));
+
+        gf_timer_call_cancel(link->fop->xl->ctx, lock->timer);
+        lock->timer = NULL;
+
+        /* Any fop being processed from now on, will need to wait
+         * until the next unlock/lock cycle. */
+        lock->release = now = _gf_true;
+    }
+
+    UNLOCK(&inode->lock);
+
+    if (now) {
+        ec_unlock_now(link);
+    } else {
+        /* The timer has been cancelled just after firing it but before
+         * getting here. This means that another fop has used the lock
+         * and everything should be handled as if this callback were
+         * have not been executed. However we still have an owner
+         * reference.
+         *
+         * We need to release our reference. If this is not the last
+         * reference (the most common case because another fop has
+         * taken another ref) we only need to decrement the counter.
+         * Otherwise we have been delayed enough so that the other fop
+         * has had time to acquire the reference, do its operation and
+         * release it. At the time of releasing it, the fop did found
+         * that the ref counter was > 1 (our reference), so the delayed
+         * unlock timer wasn't started. We need to start it again if we
+         * are the last reference.
+         *
+         * ec_unlock_timer_add() handles both cases.
+         */
+        ec_unlock_timer_add(link);
+
+        /* We need to resume the fop that was waiting for the delayed
+         * unlock.
+         */
+        ec_resume(link->fop, 0);
+    }
+}
+
+void
+ec_unlock_timer_cbk(void *data)
+{
+    ec_unlock_timer_del(data);
+}
+
+static gf_boolean_t
+ec_eager_lock_used(ec_t *ec, ec_fop_data_t *fop)
+{
+    /* Fops with no locks at this point mean that they are sent as sub-fops
+     * of other higher level fops. In this case we simply assume that the
+     * parent fop will take correct care of the eager lock. */
+    if (fop->lock_count == 0) {
+        return _gf_true;
+    }
+
+    /* We may have more than one lock, but this only happens in the rename
+     * fop, and both locks will reference an inode of the same type (a
+     * directory in this case), so we only need to check the first lock. */
+    if (fop->locks[0].lock->loc.inode->ia_type == IA_IFREG) {
+        return ec->eager_lock;
+    }
+
+    return ec->other_eager_lock;
+}
+
+static uint32_t
+ec_eager_lock_timeout(ec_t *ec, ec_lock_t *lock)
+{
+    if (lock->loc.inode->ia_type == IA_IFREG) {
+        return ec->eager_lock_timeout;
+    }
+
+    return ec->other_eager_lock_timeout;
+}
+
+static gf_boolean_t
+ec_lock_delay_create(ec_lock_link_t *link)
+{
+    struct timespec delay;
+    ec_fop_data_t *fop = link->fop;
+    ec_lock_t *lock = link->lock;
+
+    delay.tv_sec = ec_eager_lock_timeout(fop->xl->private, lock);
+    delay.tv_nsec = 0;
+    lock->timer = gf_timer_call_after(fop->xl->ctx, delay, ec_unlock_timer_cbk,
+                                      link);
+    if (lock->timer == NULL) {
+        gf_msg(fop->xl->name, GF_LOG_WARNING, ENOMEM,
+               EC_MSG_UNLOCK_DELAY_FAILED, "Unable to delay an unlock");
+
+        return _gf_false;
+    }
+
+    return _gf_true;
+}
+
+void
+ec_unlock_timer_add(ec_lock_link_t *link)
+{
+    ec_fop_data_t *fop = link->fop;
+    ec_lock_t *lock = link->lock;
+    gf_boolean_t now = _gf_false;
+
+    LOCK(&lock->loc.inode->lock);
+
+    /* We are trying to unlock the lock. We can have multiple scenarios here,
+     * but all of them need to have lock->timer == NULL:
+     *
+     * 1. There are other owners currently running that can call ec_unlock().
+     *
+     *    None of them can have started the timer until the last one. But this
+     *    call should be the consequence of this lastest one.
+     *
+     * 2. There are fops in the waiting or frozen lists.
+     *
+     *    These fops cannot call ec_unlock(). So we should be here.
+     *
+     * We must reach here with at least one owner reference.
+     */
+    GF_ASSERT((lock->timer == NULL) && (lock->refs_owners > 0));
+
+    /* If the fop detects that a heal is needed, we mark the lock to be
+     * released as soon as possible. */
+    lock->release |= ec_fop_needs_heal(fop);
+
+    if (lock->refs_owners > 1) {
+        ec_trace("UNLOCK_SKIP", fop, "lock=%p", lock);
+
+        /* If there are other owners we cannot do anything else with the lock.
+         * Note that the current fop has already been removed from the owners
+         * list in ec_lock_reuse(). */
+        lock->refs_owners--;
+
+        UNLOCK(&lock->loc.inode->lock);
+    } else if (lock->acquired) {
+        /* There are no other owners and the lock is acquired. If there were
+         * fops waiting, at least one of them should have been promoted to an
+         * owner, so the waiting list should be empty. */
+        GF_ASSERT(list_empty(&lock->owners) && list_empty(&lock->waiting));
+
+        ec_t *ec = fop->xl->private;
+
+        /* If everything goes as expected this fop will be put to sleep until
+         * the timer callback is executed. */
+        ec_sleep(fop);
+
+        /* If the lock needs to be released, or ec is shutting down, do not
+         * delay lock release. */
+        if (!lock->release && !ec->shutdown) {
+            ec_trace("UNLOCK_DELAY", fop, "lock=%p, release=%d", lock,
+                     lock->release);
+
+            if (!ec_lock_delay_create(link)) {
+                /* We are unable to create a new timer. We immediately release
+                 * the lock. */
+                lock->release = now = _gf_true;
+            }
+
+        } else {
+            ec_trace("UNLOCK_FORCE", fop, "lock=%p, release=%d", lock,
+                     lock->release);
+            lock->release = now = _gf_true;
+        }
+
+        UNLOCK(&lock->loc.inode->lock);
+
+        if (now) {
+            ec_unlock_now(link);
+        }
+    } else {
+        /* There are no owners and the lock is not acquired. This can only
+         * happen if a lock attempt has failed and we get to the unlock step
+         * of the fop. As in the previous case, the waiting list must be
+         * empty. */
+        GF_ASSERT(list_empty(&lock->owners) && list_empty(&lock->waiting));
+
+        /* We need to mark the lock to be released to correctly handle fops
+         * that may get in after we release the inode mutex but before
+         * ec_lock_unfreeze() is processed. */
+        lock->release = _gf_true;
+
+        UNLOCK(&lock->loc.inode->lock);
+
+        ec_lock_unfreeze(link);
+    }
+}
+
+void
+ec_unlock(ec_fop_data_t *fop)
+{
+    int32_t i;
+
+    for (i = 0; i < fop->lock_count; i++) {
+        ec_unlock_timer_add(&fop->locks[i]);
+    }
+}
+
+void
+ec_flush_size_version(ec_fop_data_t *fop)
+{
+    GF_ASSERT(fop->lock_count == 1);
+    ec_update_info(&fop->locks[0]);
+}
+
+static void
+ec_update_stripe(ec_t *ec, ec_stripe_list_t *stripe_cache, ec_stripe_t *stripe,
+                 ec_fop_data_t *fop)
+{
+    off_t base;
+
+    /* On write fops, we only update existing fragments if the write has
+     * succeeded. Otherwise, we remove them from the cache. */
+    if ((fop->id == GF_FOP_WRITE) && (fop->answer != NULL) &&
+        (fop->answer->op_ret >= 0)) {
+        base = stripe->frag_offset - fop->frag_range.first;
+        base *= ec->fragments;
+
+        /* We check if the stripe offset falls inside the real region
+         * modified by the write fop (a write request is allowed,
+         * though uncommon, to write less bytes than requested). The
+         * current write fop implementation doesn't allow partial
+         * writes of fragments, so if there's no error, we are sure
+         * that a full stripe has been completely modified or not
+         * touched at all. The value of op_ret may not be a multiple
+         * of the stripe size because it depends on the requested
+         * size by the user, so we update the stripe if the write has
+         * modified at least one byte (meaning ec has written the full
+         * stripe). */
+        if (base < fop->answer->op_ret + fop->head) {
+            memcpy(stripe->data, fop->vector[0].iov_base + base,
+                   ec->stripe_size);
+            list_move_tail(&stripe->lru, &stripe_cache->lru);
+
+            GF_ATOMIC_INC(ec->stats.stripe_cache.updates);
+        }
+    } else {
+        stripe->frag_offset = -1;
+        list_move(&stripe->lru, &stripe_cache->lru);
+
+        GF_ATOMIC_INC(ec->stats.stripe_cache.invals);
+    }
+}
+
+static void
+ec_update_cached_stripes(ec_fop_data_t *fop)
+{
+    uint64_t first;
+    uint64_t last;
+    ec_stripe_t *stripe = NULL;
+    ec_inode_t *ctx = NULL;
+    ec_stripe_list_t *stripe_cache = NULL;
+    inode_t *inode = NULL;
+    struct list_head *temp;
+    struct list_head sentinel;
+
+    first = fop->frag_range.first;
+    /* 'last' represents the first stripe not touched by the operation */
+    last = fop->frag_range.last;
+
+    /* If there are no modified stripes, we don't need to do anything
+     * else. */
+    if (last <= first) {
+        return;
+    }
+
+    if (!fop->use_fd) {
+        inode = fop->loc[0].inode;
+    } else {
+        inode = fop->fd->inode;
+    }
+
+    LOCK(&inode->lock);
+
+    ctx = __ec_inode_get(inode, fop->xl);
+    if (ctx == NULL) {
+        goto out;
+    }
+    stripe_cache = &ctx->stripe_cache;
+
+    /* Since we'll be moving elements of the list to the tail, we might
+     * end in an infinite loop. To avoid it, we insert a sentinel element
+     * into the list, so that it will be used to detect when we have
+     * traversed all existing elements once. */
+    list_add_tail(&sentinel, &stripe_cache->lru);
+    temp = stripe_cache->lru.next;
+    while (temp != &sentinel) {
+        stripe = list_entry(temp, ec_stripe_t, lru);
+        temp = temp->next;
+        if ((first <= stripe->frag_offset) && (stripe->frag_offset < last)) {
+            ec_update_stripe(fop->xl->private, stripe_cache, stripe, fop);
+        }
+    }
+    list_del(&sentinel);
+
+out:
+    UNLOCK(&inode->lock);
+}
+
+void
+ec_lock_reuse(ec_fop_data_t *fop)
+{
+    ec_cbk_data_t *cbk;
+    ec_t *ec = NULL;
+    int32_t i, count;
+    gf_boolean_t release = _gf_false;
+    ec = fop->xl->private;
+    cbk = fop->answer;
+
+    if (ec_eager_lock_used(ec, fop) && cbk != NULL) {
+        if (cbk->xdata != NULL) {
+            if ((dict_get_int32(cbk->xdata, GLUSTERFS_INODELK_COUNT, &count) ==
+                 0) &&
+                (count > 1)) {
+                release = _gf_true;
+            }
+            if (release) {
+                gf_msg_debug(fop->xl->name, 0, "Lock contention detected");
+            }
+        }
+    } else {
+        /* If eager lock is disabled or if we haven't get
+         * an answer with enough quorum, we always release
+         * the lock. */
+        release = _gf_true;
+    }
+    ec_update_cached_stripes(fop);
+
+    for (i = 0; i < fop->lock_count; i++) {
+        ec_lock_next_owner(&fop->locks[i], cbk, release);
+    }
+}
+
+void
+__ec_manager(ec_fop_data_t *fop, int32_t error)
+{
+    ec_t *ec = fop->xl->private;
+
+    do {
+        ec_trace("MANAGER", fop, "error=%d", error);
+
+        if (!ec_must_wind(fop)) {
+            if (ec->xl_up_count < ec->fragments) {
+                error = ENOTCONN;
+            }
+        }
+
+        if (error != 0) {
+            fop->error = error;
+            fop->state = -fop->state;
+        }
+
+        if ((fop->state == EC_STATE_END) || (fop->state == -EC_STATE_END)) {
+            ec_fop_data_release(fop);
+
+            break;
+        }
+
+        /* At each state, fop must not be used anywhere else and there
+         * shouldn't be any pending subfop going on. */
+        GF_ASSERT(fop->jobs == 0);
+
+        /* While the manager is running we need to avoid that subfops launched
+         * from it could finish and call ec_resume() before the fop->handler
+         * has completed. This could lead to the same manager being executed
+         * by two threads concurrently. ec_check_complete() will take care of
+         * this reference. */
+        fop->jobs = 1;
+
+        fop->state = fop->handler(fop, fop->state);
+        GF_ASSERT(fop->state >= 0);
+
+        error = ec_check_complete(fop, __ec_manager);
+    } while (error >= 0);
+}
+
+void
+ec_manager(ec_fop_data_t *fop, int32_t error)
+{
+    GF_ASSERT(fop->jobs == 0);
+    GF_ASSERT(fop->winds == 0);
+    GF_ASSERT(fop->error == 0);
+
+    if (fop->state == EC_STATE_START) {
+        fop->state = EC_STATE_INIT;
+    }
+
+    __ec_manager(fop, error);
+}
+
+gf_boolean_t
+__ec_is_last_fop(ec_t *ec)
+{
+    if ((list_empty(&ec->pending_fops)) &&
+        (GF_ATOMIC_GET(ec->async_fop_count) == 0)) {
+        return _gf_true;
+    }
+    return _gf_false;
+}
diff --git a/xlators/cluster/ec/src/ec-common.h b/xlators/cluster/ec/src/ec-common.h
new file mode 100644
index 00000000000..51493612ac6
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-common.h
@@ -0,0 +1,234 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_COMMON_H__
+#define __EC_COMMON_H__
+
+#include "glusterfs/compat-errno.h"  // for ENODATA on BSD
+#include "ec-data.h"
+
+typedef enum { EC_DATA_TXN, EC_METADATA_TXN } ec_txn_t;
+
+#define EC_FOP_HEAL -1
+#define EC_FOP_FHEAL -2
+
+#define EC_CONFIG_VERSION 0
+
+#define EC_CONFIG_ALGORITHM 0
+
+#define EC_FLAG_LOCK_SHARED 0x0001
+
+#define QUORUM_CBK(fn, fop, frame, cookie, this, op_ret, op_errno, params...)  \
+    do {                                                                       \
+        ec_t *__ec = fop->xl->private;                                         \
+        int32_t __op_ret = 0;                                                  \
+        int32_t __op_errno = 0;                                                \
+        int32_t __success_count = gf_bits_count(fop->good);                    \
+                                                                               \
+        __op_ret = op_ret;                                                     \
+        __op_errno = op_errno;                                                 \
+        if (!fop->parent && frame &&                                           \
+            (GF_CLIENT_PID_SELF_HEALD != frame->root->pid) &&                  \
+            __ec->quorum_count && (__success_count < __ec->quorum_count) &&    \
+            op_ret >= 0) {                                                     \
+            __op_ret = -1;                                                     \
+            __op_errno = EIO;                                                  \
+            gf_msg(__ec->xl->name, GF_LOG_ERROR, 0,                            \
+                   EC_MSG_CHILDS_INSUFFICIENT,                                 \
+                   "Insufficient available children for this request "         \
+                   "(have %d, need %d). %s",                                   \
+                   __success_count, __ec->quorum_count, ec_msg_str(fop));      \
+        }                                                                      \
+        fn(frame, cookie, this, __op_ret, __op_errno, params);                 \
+    } while (0)
+
+enum _ec_xattrop_flags {
+    EC_FLAG_XATTROP,
+    EC_FLAG_DATA_DIRTY,
+    EC_FLAG_METADATA_DIRTY,
+
+    /* Add any new flag here, before EC_FLAG_MAX. The maximum number of
+     * flags that can be defined is 16. */
+
+    EC_FLAG_MAX
+};
+
+/* We keep two sets of flags. One to determine what's really providing the
+ * current xattrop and the other to know what the parent fop of the xattrop
+ * needs to proceed. It might happen that a fop needs some information that
+ * is being already requested by a previous fop. The two sets are stored
+ * contiguously. */
+
+#define EC_FLAG_NEEDS(_flag) (1 << (_flag))
+#define EC_FLAG_PROVIDES(_flag) (1 << ((_flag) + EC_FLAG_MAX))
+
+#define EC_NEEDED_FLAGS(_flags) ((_flags) & ((1 << EC_FLAG_MAX) - 1))
+
+#define EC_PROVIDED_FLAGS(_flags) EC_NEEDED_FLAGS((_flags) >> EC_FLAG_MAX)
+
+#define EC_FLAGS_HAVE(_flags, _flag) (((_flags) & (1 << (_flag))) != 0)
+
+#define EC_SELFHEAL_BIT 62
+
+#define EC_MINIMUM_ONE (1 << 6)
+#define EC_MINIMUM_MIN (2 << 6)
+#define EC_MINIMUM_ALL (3 << 6)
+#define EC_FOP_NO_PROPAGATE_ERROR (1 << 8)
+#define EC_FOP_MINIMUM(_flags) ((_flags)&255)
+#define EC_FOP_FLAGS(_flags) ((_flags) & ~255)
+
+#define EC_UPDATE_DATA 1
+#define EC_UPDATE_META 2
+#define EC_QUERY_INFO 4
+#define EC_INODE_SIZE 8
+
+#define EC_STATE_START 0
+#define EC_STATE_END 0
+#define EC_STATE_INIT 1
+#define EC_STATE_LOCK 2
+#define EC_STATE_DISPATCH 3
+#define EC_STATE_PREPARE_ANSWER 4
+#define EC_STATE_REPORT 5
+#define EC_STATE_LOCK_REUSE 6
+#define EC_STATE_UNLOCK 7
+
+#define EC_STATE_DELAYED_START 100
+
+#define EC_STATE_HEAL_ENTRY_LOOKUP 200
+#define EC_STATE_HEAL_ENTRY_PREPARE 201
+#define EC_STATE_HEAL_PRE_INODELK_LOCK 202
+#define EC_STATE_HEAL_PRE_INODE_LOOKUP 203
+#define EC_STATE_HEAL_XATTRIBUTES_REMOVE 204
+#define EC_STATE_HEAL_XATTRIBUTES_SET 205
+#define EC_STATE_HEAL_ATTRIBUTES 206
+#define EC_STATE_HEAL_OPEN 207
+#define EC_STATE_HEAL_REOPEN_FD 208
+#define EC_STATE_HEAL_UNLOCK 209
+#define EC_STATE_HEAL_UNLOCK_ENTRY 210
+#define EC_STATE_HEAL_DATA_LOCK 211
+#define EC_STATE_HEAL_DATA_COPY 212
+#define EC_STATE_HEAL_DATA_UNLOCK 213
+#define EC_STATE_HEAL_POST_INODELK_LOCK 214
+#define EC_STATE_HEAL_POST_INODE_LOOKUP 215
+#define EC_STATE_HEAL_SETATTR 216
+#define EC_STATE_HEAL_POST_INODELK_UNLOCK 217
+#define EC_STATE_HEAL_DISPATCH 218
+
+/* Value to cover the full range of a file */
+#define EC_RANGE_FULL ((uint64_t)LLONG_MAX + 1)
+
+gf_boolean_t
+ec_dispatch_one_retry(ec_fop_data_t *fop, ec_cbk_data_t **cbk);
+void
+ec_dispatch_next(ec_fop_data_t *fop, uint32_t idx);
+
+void
+ec_complete(ec_fop_data_t *fop);
+
+void
+ec_update_good(ec_fop_data_t *fop, uintptr_t good);
+
+void
+ec_fop_set_error(ec_fop_data_t *fop, int32_t error);
+
+void
+__ec_fop_set_error(ec_fop_data_t *fop, int32_t error);
+
+ec_cbk_data_t *
+ec_fop_prepare_answer(ec_fop_data_t *fop, gf_boolean_t ro);
+
+gf_boolean_t
+ec_cbk_set_error(ec_cbk_data_t *cbk, int32_t error, gf_boolean_t ro);
+
+void
+ec_lock_prepare_inode(ec_fop_data_t *fop, loc_t *loc, uint32_t flags,
+                      off_t fl_start, uint64_t fl_size);
+void
+ec_lock_prepare_parent_inode(ec_fop_data_t *fop, loc_t *loc, loc_t *base,
+                             uint32_t flags);
+void
+ec_lock_prepare_fd(ec_fop_data_t *fop, fd_t *fd, uint32_t flags, off_t fl_start,
+                   uint64_t fl_size);
+void
+ec_lock(ec_fop_data_t *fop);
+void
+ec_lock_reuse(ec_fop_data_t *fop);
+void
+ec_unlock(ec_fop_data_t *fop);
+void
+ec_lock_release(ec_t *ec, inode_t *inode);
+
+gf_boolean_t
+ec_get_inode_size(ec_fop_data_t *fop, inode_t *inode, uint64_t *size);
+gf_boolean_t
+__ec_get_inode_size(ec_fop_data_t *fop, inode_t *inode, uint64_t *size);
+gf_boolean_t
+ec_set_inode_size(ec_fop_data_t *fop, inode_t *inode, uint64_t size);
+gf_boolean_t
+__ec_set_inode_size(ec_fop_data_t *fop, inode_t *inode, uint64_t size);
+void
+ec_clear_inode_info(ec_fop_data_t *fop, inode_t *inode);
+
+void
+ec_flush_size_version(ec_fop_data_t *fop);
+
+void
+ec_dispatch_all(ec_fop_data_t *fop);
+void
+ec_dispatch_inc(ec_fop_data_t *fop);
+void
+ec_dispatch_min(ec_fop_data_t *fop);
+void
+ec_dispatch_one(ec_fop_data_t *fop);
+
+void
+ec_succeed_all(ec_fop_data_t *fop);
+
+void
+ec_sleep(ec_fop_data_t *fop);
+void
+ec_resume(ec_fop_data_t *fop, int32_t error);
+void
+ec_resume_parent(ec_fop_data_t *fop);
+
+void
+ec_manager(ec_fop_data_t *fop, int32_t error);
+gf_boolean_t
+ec_is_recoverable_error(int32_t op_errno);
+void
+ec_handle_healers_done(ec_fop_data_t *fop);
+
+int32_t
+ec_heal_inspect(call_frame_t *frame, ec_t *ec, inode_t *inode,
+                unsigned char *locked_on, gf_boolean_t self_locked,
+                gf_boolean_t thorough, ec_heal_need_t *need_heal);
+int32_t
+ec_get_heal_info(xlator_t *this, loc_t *loc, dict_t **dict);
+
+int32_t
+ec_lock_unlocked(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata);
+
+void
+ec_update_fd_status(fd_t *fd, xlator_t *xl, int child_index,
+                    int32_t ret_status);
+gf_boolean_t
+ec_is_entry_healing(ec_fop_data_t *fop);
+void
+ec_set_entry_healing(ec_fop_data_t *fop);
+void
+ec_reset_entry_healing(ec_fop_data_t *fop);
+char *
+ec_msg_str(ec_fop_data_t *fop);
+gf_boolean_t
+__ec_is_last_fop(ec_t *ec);
+void
+ec_lock_update_good(ec_lock_t *lock, ec_fop_data_t *fop);
+#endif /* __EC_COMMON_H__ */
diff --git a/xlators/cluster/ec/src/ec-data.c b/xlators/cluster/ec/src/ec-data.c
new file mode 100644
index 00000000000..06388833546
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-data.c
@@ -0,0 +1,288 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "ec-helpers.h"
+#include "ec-common.h"
+#include "ec-data.h"
+#include "ec-messages.h"
+
+ec_cbk_data_t *
+ec_cbk_data_allocate(call_frame_t *frame, xlator_t *this, ec_fop_data_t *fop,
+                     int32_t id, int32_t idx, int32_t op_ret, int32_t op_errno)
+{
+    ec_cbk_data_t *cbk;
+    ec_t *ec = this->private;
+
+    if (fop->xl != this) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, EC_MSG_XLATOR_MISMATCH,
+               "Mismatching xlators between request "
+               "and answer (req=%s, ans=%s).",
+               fop->xl->name, this->name);
+
+        return NULL;
+    }
+    if (fop->frame != frame) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, EC_MSG_FRAME_MISMATCH,
+               "Mismatching frames between request "
+               "and answer (req=%p, ans=%p).",
+               fop->frame, frame);
+
+        return NULL;
+    }
+    if (fop->id != id) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, EC_MSG_FOP_MISMATCH,
+               "Mismatching fops between request "
+               "and answer (req=%d, ans=%d).",
+               fop->id, id);
+
+        return NULL;
+    }
+
+    cbk = mem_get0(ec->cbk_pool);
+    if (cbk == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+               "Failed to allocate memory for an "
+               "answer.");
+        return NULL;
+    }
+
+    cbk->fop = fop;
+    cbk->idx = idx;
+    cbk->mask = 1ULL << idx;
+    cbk->count = 1;
+    cbk->op_ret = op_ret;
+    cbk->op_errno = op_errno;
+    INIT_LIST_HEAD(&cbk->entries.list);
+
+    LOCK(&fop->lock);
+
+    list_add_tail(&cbk->answer_list, &fop->answer_list);
+
+    UNLOCK(&fop->lock);
+
+    return cbk;
+}
+
+void
+ec_cbk_data_destroy(ec_cbk_data_t *cbk)
+{
+    if (cbk->xdata != NULL) {
+        dict_unref(cbk->xdata);
+    }
+    if (cbk->dict != NULL) {
+        dict_unref(cbk->dict);
+    }
+    if (cbk->inode != NULL) {
+        inode_unref(cbk->inode);
+    }
+    if (cbk->fd != NULL) {
+        fd_unref(cbk->fd);
+    }
+    if (cbk->buffers != NULL) {
+        iobref_unref(cbk->buffers);
+    }
+    GF_FREE(cbk->vector);
+    gf_dirent_free(&cbk->entries);
+    GF_FREE(cbk->str);
+
+    mem_put(cbk);
+}
+
+ec_fop_data_t *
+ec_fop_data_allocate(call_frame_t *frame, xlator_t *this, int32_t id,
+                     uint32_t flags, uintptr_t target, uint32_t fop_flags,
+                     ec_wind_f wind, ec_handler_f handler, ec_cbk_t cbks,
+                     void *data)
+{
+    ec_fop_data_t *fop, *parent;
+    ec_t *ec = this->private;
+
+    fop = mem_get0(ec->fop_pool);
+    if (fop == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+               "Failed to allocate memory for a "
+               "request.");
+
+        return NULL;
+    }
+
+    INIT_LIST_HEAD(&fop->cbk_list);
+    INIT_LIST_HEAD(&fop->healer);
+    INIT_LIST_HEAD(&fop->answer_list);
+    INIT_LIST_HEAD(&fop->pending_list);
+    INIT_LIST_HEAD(&fop->locks[0].owner_list);
+    INIT_LIST_HEAD(&fop->locks[0].wait_list);
+    INIT_LIST_HEAD(&fop->locks[1].owner_list);
+    INIT_LIST_HEAD(&fop->locks[1].wait_list);
+
+    fop->xl = this;
+    fop->req_frame = frame;
+
+    /* fops need a private frame to be able to execute some postop operations
+     * even if the original fop has completed and reported back to the upper
+     * xlator and it has destroyed the base frame.
+     *
+     * TODO: minimize usage of private frames. Reuse req_frame as much as
+     *       possible.
+     */
+    if (frame != NULL) {
+        fop->frame = copy_frame(frame);
+    } else {
+        fop->frame = create_frame(this, this->ctx->pool);
+    }
+    if (fop->frame == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+               "Failed to create a private frame "
+               "for a request");
+
+        mem_put(fop);
+
+        return NULL;
+    }
+    fop->id = id;
+    fop->refs = 1;
+
+    fop->flags = flags;
+    fop->minimum = EC_FOP_MINIMUM(fop_flags);
+    fop->fop_flags = EC_FOP_FLAGS(fop_flags);
+    fop->mask = target;
+
+    fop->wind = wind;
+    fop->handler = handler;
+    fop->cbks = cbks;
+    fop->data = data;
+
+    fop->uid = fop->frame->root->uid;
+    fop->gid = fop->frame->root->gid;
+
+    LOCK_INIT(&fop->lock);
+
+    fop->frame->local = fop;
+
+    if (frame != NULL) {
+        parent = frame->local;
+        if (parent != NULL) {
+            ec_sleep(parent);
+        }
+
+        fop->parent = parent;
+    }
+
+    LOCK(&ec->lock);
+
+    list_add_tail(&fop->pending_list, &ec->pending_fops);
+
+    UNLOCK(&ec->lock);
+
+    return fop;
+}
+
+void
+ec_fop_data_acquire(ec_fop_data_t *fop)
+{
+    LOCK(&fop->lock);
+
+    ec_trace("ACQUIRE", fop, "");
+
+    fop->refs++;
+
+    UNLOCK(&fop->lock);
+}
+
+static void
+ec_handle_last_pending_fop_completion(ec_fop_data_t *fop, gf_boolean_t *notify)
+{
+    ec_t *ec = fop->xl->private;
+
+    *notify = _gf_false;
+
+    if (!list_empty(&fop->pending_list)) {
+        LOCK(&ec->lock);
+        {
+            list_del_init(&fop->pending_list);
+            *notify = __ec_is_last_fop(ec);
+        }
+        UNLOCK(&ec->lock);
+    }
+}
+
+void
+ec_fop_cleanup(ec_fop_data_t *fop)
+{
+    ec_cbk_data_t *cbk, *tmp;
+
+    list_for_each_entry_safe(cbk, tmp, &fop->answer_list, answer_list)
+    {
+        list_del_init(&cbk->answer_list);
+
+        ec_cbk_data_destroy(cbk);
+    }
+    INIT_LIST_HEAD(&fop->cbk_list);
+
+    fop->answer = NULL;
+}
+
+void
+ec_fop_data_release(ec_fop_data_t *fop)
+{
+    ec_t *ec = NULL;
+    int32_t refs;
+    gf_boolean_t notify = _gf_false;
+
+    LOCK(&fop->lock);
+
+    ec_trace("RELEASE", fop, "");
+
+    GF_ASSERT(fop->refs > 0);
+    refs = --fop->refs;
+
+    UNLOCK(&fop->lock);
+
+    if (refs == 0) {
+        fop->frame->local = NULL;
+        STACK_DESTROY(fop->frame->root);
+
+        LOCK_DESTROY(&fop->lock);
+
+        if (fop->xdata != NULL) {
+            dict_unref(fop->xdata);
+        }
+        if (fop->dict != NULL) {
+            dict_unref(fop->dict);
+        }
+        if (fop->inode != NULL) {
+            inode_unref(fop->inode);
+        }
+        if (fop->fd != NULL) {
+            fd_unref(fop->fd);
+        }
+        if (fop->buffers != NULL) {
+            iobref_unref(fop->buffers);
+        }
+        GF_FREE(fop->vector);
+        GF_FREE(fop->str[0]);
+        GF_FREE(fop->str[1]);
+        loc_wipe(&fop->loc[0]);
+        loc_wipe(&fop->loc[1]);
+        GF_FREE(fop->errstr);
+
+        ec_resume_parent(fop);
+
+        ec_fop_cleanup(fop);
+
+        ec = fop->xl->private;
+        ec_handle_last_pending_fop_completion(fop, &notify);
+        ec_handle_healers_done(fop);
+        mem_put(fop);
+        if (notify) {
+            ec_pending_fops_completed(ec);
+        }
+    }
+}
diff --git a/xlators/cluster/ec/src/ec-data.h b/xlators/cluster/ec/src/ec-data.h
new file mode 100644
index 00000000000..c8a74ffe1ed
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-data.h
@@ -0,0 +1,35 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_DATA_H__
+#define __EC_DATA_H__
+
+#include "ec-types.h"
+
+ec_cbk_data_t *
+ec_cbk_data_allocate(call_frame_t *frame, xlator_t *this, ec_fop_data_t *fop,
+                     int32_t id, int32_t idx, int32_t op_ret, int32_t op_errno);
+ec_fop_data_t *
+ec_fop_data_allocate(call_frame_t *frame, xlator_t *this, int32_t id,
+                     uint32_t flags, uintptr_t target, uint32_t fop_flags,
+                     ec_wind_f wind, ec_handler_f handler, ec_cbk_t cbks,
+                     void *data);
+void
+ec_fop_data_acquire(ec_fop_data_t *fop);
+void
+ec_fop_data_release(ec_fop_data_t *fop);
+
+void
+ec_fop_cleanup(ec_fop_data_t *fop);
+
+void
+ec_pending_fops_completed(ec_t *ec);
+
+#endif /* __EC_DATA_H__ */
diff --git a/xlators/cluster/ec/src/ec-dir-read.c b/xlators/cluster/ec/src/ec-dir-read.c
new file mode 100644
index 00000000000..f71dcfac293
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-dir-read.c
@@ -0,0 +1,647 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "ec.h"
+#include "ec-messages.h"
+#include "ec-helpers.h"
+#include "ec-common.h"
+#include "ec-combine.h"
+#include "ec-fops.h"
+
+/****************************************************************
+ *
+ * File Operation: opendir
+ *
+ ***************************************************************/
+
+int32_t
+ec_combine_opendir(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    if (dst->fd != src->fd) {
+        gf_msg(fop->xl->name, GF_LOG_NOTICE, 0, EC_MSG_FD_MISMATCH,
+               "Mismatching fd in answers "
+               "of 'GF_FOP_OPENDIR': %p <-> %p",
+               dst->fd, src->fd);
+
+        return 0;
+    }
+
+    return 1;
+}
+
+int32_t
+ec_opendir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_OPENDIR, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            if (fd != NULL) {
+                cbk->fd = fd_ref(fd);
+                if (cbk->fd == NULL) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0,
+                           EC_MSG_FILE_DESC_REF_FAIL,
+                           "Failed to reference a "
+                           "file descriptor.");
+
+                    goto out;
+                }
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, ec_combine_opendir);
+
+        ec_update_fd_status(fd, this, idx, op_ret);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_opendir(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_opendir_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->opendir,
+                      &fop->loc[0], fop->fd, fop->xdata);
+}
+
+int32_t
+ec_manager_opendir(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+    ec_fd_t *ctx;
+    int32_t err;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            LOCK(&fop->fd->lock);
+
+            ctx = __ec_fd_get(fop->fd, fop->xl);
+            if (ctx == NULL) {
+                UNLOCK(&fop->fd->lock);
+
+                fop->error = ENOMEM;
+
+                return EC_STATE_REPORT;
+            }
+            if (!ctx->loc.inode) {
+                err = ec_loc_from_loc(fop->xl, &ctx->loc, &fop->loc[0]);
+                if (err != 0) {
+                    UNLOCK(&fop->fd->lock);
+
+                    fop->error = -err;
+
+                    return EC_STATE_REPORT;
+                }
+            }
+
+            UNLOCK(&fop->fd->lock);
+
+            /* Fall through */
+
+        case EC_STATE_LOCK:
+            ec_lock_prepare_inode(fop, &fop->loc[0], EC_QUERY_INFO, 0,
+                                  EC_RANGE_FULL);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_true);
+            if (cbk != NULL) {
+                /* Save which subvolumes successfully opened the directory.
+                 * If ctx->open is 0, it means that readdir cannot be
+                 * processed in this directory.
+                 */
+                LOCK(&fop->fd->lock);
+
+                ctx = __ec_fd_get(fop->fd, fop->xl);
+                if (ctx != NULL) {
+                    ctx->open |= cbk->mask;
+                }
+
+                UNLOCK(&fop->fd->lock);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.opendir != NULL) {
+                fop->cbks.opendir(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                  cbk->op_errno, cbk->fd, cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.opendir != NULL) {
+                fop->cbks.opendir(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                  NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_opendir(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_opendir_cbk_t func, void *data, loc_t *loc,
+           fd_t *fd, dict_t *xdata)
+{
+    ec_cbk_t callback = {.opendir = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(OPENDIR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_OPENDIR, EC_FLAG_LOCK_SHARED,
+                               target, fop_flags, ec_wind_opendir,
+                               ec_manager_opendir, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
+
+/* Returns -1 if client_id is invalid else index of child subvol in xl_list */
+int
+ec_deitransform(xlator_t *this, off_t offset)
+{
+    int idx = -1;
+    int client_id = -1;
+    ec_t *ec = this->private;
+    char id[32] = {0};
+    int err;
+
+    client_id = gf_deitransform(this, offset);
+    sprintf(id, "%d", client_id);
+    err = dict_get_int32(ec->leaf_to_subvolid, id, &idx);
+    if (err < 0) {
+        idx = err;
+        goto out;
+    }
+
+out:
+    if (idx < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, EC_MSG_INVALID_REQUEST,
+               "Invalid index %d in readdirp request", client_id);
+        idx = -EINVAL;
+    }
+    return idx;
+}
+
+/* FOP: readdir */
+
+void
+ec_adjust_readdirp(ec_t *ec, int32_t idx, gf_dirent_t *entries)
+{
+    gf_dirent_t *entry;
+
+    list_for_each_entry(entry, &entries->list, list)
+    {
+        if (!entry->inode)
+            continue;
+
+        if (entry->d_stat.ia_type == IA_IFREG) {
+            if ((entry->dict == NULL) ||
+                (ec_dict_del_number(entry->dict, EC_XATTR_SIZE,
+                                    &entry->d_stat.ia_size) != 0)) {
+                inode_unref(entry->inode);
+                entry->inode = NULL;
+            } else {
+                ec_iatt_rebuild(ec, &entry->d_stat, 1, 1);
+            }
+        }
+    }
+}
+
+int32_t
+ec_common_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, gf_dirent_t *entries,
+                      dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, fop->id, idx, op_ret,
+                               op_errno);
+    if (cbk) {
+        if (xdata)
+            cbk->xdata = dict_ref(xdata);
+        if (cbk->op_ret >= 0)
+            list_splice_init(&entries->list, &cbk->entries.list);
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_readdir(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_common_readdir_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->readdir,
+                      fop->fd, fop->size, fop->offset, fop->xdata);
+}
+
+int32_t
+ec_manager_readdir(ec_fop_data_t *fop, int32_t state)
+{
+    ec_fd_t *ctx = NULL;
+    ec_cbk_data_t *cbk = NULL;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            /* Return error if opendir has not been successfully called on
+             * any subvolume. */
+            ctx = ec_fd_get(fop->fd, fop->xl);
+            if (ctx == NULL) {
+                fop->error = ENOMEM;
+            } else if (ctx->open == 0) {
+                fop->error = EBADFD;
+            }
+
+            if (fop->error) {
+                gf_msg(fop->xl->name, GF_LOG_ERROR, fop->error,
+                       EC_MSG_INVALID_REQUEST, "EC is not winding readdir: %s",
+                       ec_msg_str(fop));
+                return EC_STATE_REPORT;
+            }
+
+            if (fop->id == GF_FOP_READDIRP) {
+                int32_t err;
+
+                if (fop->xdata == NULL) {
+                    fop->xdata = dict_new();
+                    if (fop->xdata == NULL) {
+                        fop->error = ENOMEM;
+
+                        return EC_STATE_REPORT;
+                    }
+                }
+
+                err = dict_set_uint64(fop->xdata, EC_XATTR_SIZE, 0);
+                if (err != 0) {
+                    fop->error = -err;
+
+                    return EC_STATE_REPORT;
+                }
+            }
+
+            if (fop->offset != 0) {
+                /* Non-zero offset is irrecoverable error as the offset may not
+                 * be valid on other bricks*/
+                int32_t idx = -1;
+
+                idx = ec_deitransform(fop->xl, fop->offset);
+
+                if (idx < 0) {
+                    fop->error = -idx;
+                    return EC_STATE_REPORT;
+                }
+                fop->mask &= 1ULL << idx;
+            } else {
+                ec_lock_prepare_fd(fop, fop->fd, EC_QUERY_INFO, 0,
+                                   EC_RANGE_FULL);
+                ec_lock(fop);
+            }
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_one(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            if (ec_dispatch_one_retry(fop, &cbk)) {
+                return EC_STATE_DISPATCH;
+            }
+
+            if ((cbk != NULL) && (cbk->op_ret > 0) &&
+                (fop->id == GF_FOP_READDIRP)) {
+                ec_adjust_readdirp(fop->xl->private, cbk->idx, &cbk->entries);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+            GF_ASSERT(cbk);
+            if (fop->id == GF_FOP_READDIR) {
+                if (fop->cbks.readdir != NULL) {
+                    fop->cbks.readdir(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                      cbk->op_errno, &cbk->entries, cbk->xdata);
+                }
+            } else {
+                if (fop->cbks.readdirp != NULL) {
+                    fop->cbks.readdirp(fop->req_frame, fop, fop->xl,
+                                       cbk->op_ret, cbk->op_errno,
+                                       &cbk->entries, cbk->xdata);
+                }
+            }
+            if (fop->offset == 0)
+                return EC_STATE_LOCK_REUSE;
+            else
+                return EC_STATE_END;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            if (fop->id == GF_FOP_READDIR) {
+                if (fop->cbks.readdir != NULL) {
+                    fop->cbks.readdir(fop->req_frame, fop, fop->xl, -1,
+                                      fop->error, NULL, NULL);
+                }
+            } else {
+                if (fop->cbks.readdirp != NULL) {
+                    fop->cbks.readdirp(fop->req_frame, fop, fop->xl, -1,
+                                       fop->error, NULL, NULL);
+                }
+            }
+            if (fop->offset == 0)
+                return EC_STATE_LOCK_REUSE;
+            else
+                return EC_STATE_END;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            GF_ASSERT(fop->offset == 0);
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            GF_ASSERT(fop->offset == 0);
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_readdir(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_readdir_cbk_t func, void *data, fd_t *fd,
+           size_t size, off_t offset, dict_t *xdata)
+{
+    ec_cbk_t callback = {.readdir = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(READDIR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_READDIR, EC_FLAG_LOCK_SHARED,
+                               target, fop_flags, ec_wind_readdir,
+                               ec_manager_readdir, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->size = size;
+    fop->offset = offset;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
+
+/* FOP: readdirp */
+
+void
+ec_wind_readdirp(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_common_readdir_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->readdirp,
+                      fop->fd, fop->size, fop->offset, fop->xdata);
+}
+
+void
+ec_readdirp(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_readdirp_cbk_t func, void *data, fd_t *fd,
+            size_t size, off_t offset, dict_t *xdata)
+{
+    ec_cbk_t callback = {.readdirp = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(READDIRP) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(
+        frame, this, GF_FOP_READDIRP, EC_FLAG_LOCK_SHARED, target, fop_flags,
+        ec_wind_readdirp, ec_manager_readdir, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->size = size;
+    fop->offset = offset;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
diff --git a/xlators/cluster/ec/src/ec-dir-write.c b/xlators/cluster/ec/src/ec-dir-write.c
new file mode 100644
index 00000000000..53d27d895c3
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-dir-write.c
@@ -0,0 +1,1487 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "ec.h"
+#include "ec-messages.h"
+#include "ec-helpers.h"
+#include "ec-common.h"
+#include "ec-combine.h"
+#include "ec-method.h"
+#include "ec-fops.h"
+
+int
+ec_dir_write_cbk(call_frame_t *frame, xlator_t *this, void *cookie, int op_ret,
+                 int op_errno, struct iatt *poststat, struct iatt *preparent,
+                 struct iatt *postparent, struct iatt *preparent2,
+                 struct iatt *postparent2, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int i = 0;
+    int idx = 0;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+    idx = (long)cookie;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, fop->id, idx, op_ret,
+                               op_errno);
+    if (!cbk)
+        goto out;
+
+    if (xdata)
+        cbk->xdata = dict_ref(xdata);
+
+    if (op_ret < 0)
+        goto out;
+
+    if (poststat)
+        cbk->iatt[i++] = *poststat;
+
+    if (preparent)
+        cbk->iatt[i++] = *preparent;
+
+    if (postparent)
+        cbk->iatt[i++] = *postparent;
+
+    if (preparent2)
+        cbk->iatt[i++] = *preparent2;
+
+    if (postparent2)
+        cbk->iatt[i++] = *postparent2;
+
+out:
+    if (cbk)
+        ec_combine(cbk, ec_combine_write);
+
+    if (fop)
+        ec_complete(fop);
+    return 0;
+}
+
+/* FOP: create */
+
+int32_t
+ec_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, fd_t *fd, inode_t *inode, struct iatt *buf,
+              struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    return ec_dir_write_cbk(frame, this, cookie, op_ret, op_errno, buf,
+                            preparent, postparent, NULL, NULL, xdata);
+}
+
+void
+ec_wind_create(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_create_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->create,
+                      &fop->loc[0], fop->int32, fop->mode[0], fop->mode[1],
+                      fop->fd, fop->xdata);
+}
+
+int32_t
+ec_manager_create(ec_fop_data_t *fop, int32_t state)
+{
+    ec_config_t config;
+    ec_t *ec;
+    ec_cbk_data_t *cbk;
+    ec_fd_t *ctx;
+    uint64_t version[2] = {0, 0};
+    int32_t err;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            LOCK(&fop->fd->lock);
+
+            ctx = __ec_fd_get(fop->fd, fop->xl);
+            if (ctx == NULL) {
+                UNLOCK(&fop->fd->lock);
+
+                fop->error = ENOMEM;
+
+                return EC_STATE_REPORT;
+            }
+            err = ec_loc_from_loc(fop->xl, &ctx->loc, &fop->loc[0]);
+            if (err != 0) {
+                UNLOCK(&fop->fd->lock);
+
+                fop->error = -err;
+
+                return EC_STATE_REPORT;
+            }
+
+            ctx->flags = fop->int32;
+
+            UNLOCK(&fop->fd->lock);
+
+            if (fop->xdata == NULL) {
+                fop->xdata = dict_new();
+                if (fop->xdata == NULL) {
+                    fop->error = ENOMEM;
+
+                    return EC_STATE_REPORT;
+                }
+            }
+
+            ec = fop->xl->private;
+
+            config.version = EC_CONFIG_VERSION;
+            config.algorithm = EC_CONFIG_ALGORITHM;
+            config.gf_word_size = EC_GF_BITS;
+            config.bricks = ec->nodes;
+            config.redundancy = ec->redundancy;
+            config.chunk_size = EC_METHOD_CHUNK_SIZE;
+
+            err = ec_dict_set_config(fop->xdata, EC_XATTR_CONFIG, &config);
+            if (err != 0) {
+                fop->error = -err;
+
+                return EC_STATE_REPORT;
+            }
+            err = ec_dict_set_array(fop->xdata, EC_XATTR_VERSION, version,
+                                    EC_VERSION_SIZE);
+            if (err != 0) {
+                fop->error = -err;
+
+                return EC_STATE_REPORT;
+            }
+            err = ec_dict_set_number(fop->xdata, EC_XATTR_SIZE, 0);
+            if (err != 0) {
+                fop->error = -err;
+
+                return EC_STATE_REPORT;
+            }
+
+            /* We need to write to specific offsets on the bricks, so we
+             * need to remove O_APPEND from flags (if present) */
+            fop->int32 &= ~O_APPEND;
+
+            /* Fall through */
+
+        case EC_STATE_LOCK:
+            ec_lock_prepare_parent_inode(fop, &fop->loc[0], NULL,
+                                         EC_UPDATE_DATA | EC_UPDATE_META);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                int32_t err;
+
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 3, cbk->count);
+
+                err = ec_loc_update(fop->xl, &fop->loc[0], cbk->inode,
+                                    &cbk->iatt[0]);
+                if (!ec_cbk_set_error(cbk, -err, _gf_false)) {
+                    LOCK(&fop->fd->lock);
+
+                    ctx = __ec_fd_get(fop->fd, fop->xl);
+                    if (ctx != NULL) {
+                        ctx->open |= cbk->mask;
+                    }
+
+                    UNLOCK(&fop->fd->lock);
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.create != NULL) {
+                QUORUM_CBK(fop->cbks.create, fop, fop->req_frame, fop, fop->xl,
+                           cbk->op_ret, cbk->op_errno, fop->fd,
+                           fop->loc[0].inode, &cbk->iatt[0], &cbk->iatt[1],
+                           &cbk->iatt[2], cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.create != NULL) {
+                fop->cbks.create(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                 NULL, NULL, NULL, NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_create(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_create_cbk_t func, void *data, loc_t *loc,
+          int32_t flags, mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+    ec_cbk_t callback = {.create = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(CREATE) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_CREATE, 0, target, fop_flags,
+                               ec_wind_create, ec_manager_create, callback,
+                               data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->int32 = flags;
+    fop->mode[0] = mode;
+    fop->mode[1] = umask;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: link */
+
+int32_t
+ec_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, inode_t *inode, struct iatt *buf,
+            struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    return ec_dir_write_cbk(frame, this, cookie, op_ret, op_errno, buf,
+                            preparent, postparent, NULL, NULL, xdata);
+}
+
+void
+ec_wind_link(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_link_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->link,
+                      &fop->loc[0], &fop->loc[1], fop->xdata);
+}
+
+int32_t
+ec_manager_link(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ec_lock_prepare_parent_inode(
+                fop, &fop->loc[1], &fop->loc[0],
+                EC_UPDATE_DATA | EC_UPDATE_META | EC_INODE_SIZE);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                int32_t err;
+
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 3, cbk->count);
+
+                if (cbk->iatt[0].ia_type == IA_IFREG) {
+                    cbk->iatt[0].ia_size = fop->locks[0].size;
+                }
+
+                err = ec_loc_update(fop->xl, &fop->loc[0], cbk->inode,
+                                    &cbk->iatt[0]);
+                ec_cbk_set_error(cbk, -err, _gf_false);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.link != NULL) {
+                QUORUM_CBK(fop->cbks.link, fop, fop->req_frame, fop, fop->xl,
+                           cbk->op_ret, cbk->op_errno, fop->loc[0].inode,
+                           &cbk->iatt[0], &cbk->iatt[1], &cbk->iatt[2],
+                           cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.link != NULL) {
+                fop->cbks.link(fop->req_frame, fop, fop->xl, -1, fop->error,
+                               NULL, NULL, NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_link(call_frame_t *frame, xlator_t *this, uintptr_t target,
+        uint32_t fop_flags, fop_link_cbk_t func, void *data, loc_t *oldloc,
+        loc_t *newloc, dict_t *xdata)
+{
+    ec_cbk_t callback = {.link = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(LINK) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_LINK, 0, target, fop_flags,
+                               ec_wind_link, ec_manager_link, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    if (oldloc != NULL) {
+        if (loc_copy(&fop->loc[0], oldloc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (newloc != NULL) {
+        if (loc_copy(&fop->loc[1], newloc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: mkdir */
+
+int32_t
+ec_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, inode_t *inode, struct iatt *buf,
+             struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    return ec_dir_write_cbk(frame, this, cookie, op_ret, op_errno, buf,
+                            preparent, postparent, NULL, NULL, xdata);
+}
+
+void
+ec_wind_mkdir(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_mkdir_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->mkdir,
+                      &fop->loc[0], fop->mode[0], fop->mode[1], fop->xdata);
+}
+
+int32_t
+ec_manager_mkdir(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+    uint64_t version[2] = {0, 0};
+    int32_t err;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            if (fop->xdata == NULL) {
+                fop->xdata = dict_new();
+                if (fop->xdata == NULL) {
+                    fop->error = ENOMEM;
+
+                    return EC_STATE_REPORT;
+                }
+            }
+
+            err = ec_dict_set_array(fop->xdata, EC_XATTR_VERSION, version,
+                                    EC_VERSION_SIZE);
+            if (err != 0) {
+                fop->error = -err;
+                return EC_STATE_REPORT;
+            }
+
+            /* Fall through */
+
+        case EC_STATE_LOCK:
+            ec_lock_prepare_parent_inode(fop, &fop->loc[0], NULL,
+                                         EC_UPDATE_DATA | EC_UPDATE_META);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                int32_t err;
+
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 3, cbk->count);
+
+                err = ec_loc_update(fop->xl, &fop->loc[0], cbk->inode,
+                                    &cbk->iatt[0]);
+                ec_cbk_set_error(cbk, -err, _gf_false);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.mkdir != NULL) {
+                QUORUM_CBK(fop->cbks.mkdir, fop, fop->req_frame, fop, fop->xl,
+                           cbk->op_ret, cbk->op_errno, fop->loc[0].inode,
+                           &cbk->iatt[0], &cbk->iatt[1], &cbk->iatt[2],
+                           cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            cbk = fop->answer;
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.mkdir != NULL) {
+                fop->cbks.mkdir(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                NULL, NULL, NULL, NULL,
+                                ((cbk) ? cbk->xdata : NULL));
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_mkdir(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_mkdir_cbk_t func, void *data, loc_t *loc,
+         mode_t mode, mode_t umask, dict_t *xdata)
+{
+    ec_cbk_t callback = {.mkdir = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(MKDIR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_MKDIR, 0, target, fop_flags,
+                               ec_wind_mkdir, ec_manager_mkdir, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->mode[0] = mode;
+    fop->mode[1] = umask;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: mknod */
+
+int32_t
+ec_mknod_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, inode_t *inode, struct iatt *buf,
+             struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    return ec_dir_write_cbk(frame, this, cookie, op_ret, op_errno, buf,
+                            preparent, postparent, NULL, NULL, xdata);
+}
+
+void
+ec_wind_mknod(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_mknod_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->mknod,
+                      &fop->loc[0], fop->mode[0], fop->dev, fop->mode[1],
+                      fop->xdata);
+}
+
+int32_t
+ec_manager_mknod(ec_fop_data_t *fop, int32_t state)
+{
+    ec_config_t config;
+    ec_t *ec;
+    ec_cbk_data_t *cbk;
+    uint64_t version[2] = {0, 0};
+
+    switch (state) {
+        case EC_STATE_INIT:
+            if (S_ISREG(fop->mode[0])) {
+                int32_t err;
+
+                if (fop->xdata == NULL) {
+                    fop->xdata = dict_new();
+                    if (fop->xdata == NULL) {
+                        fop->error = ENOMEM;
+
+                        return EC_STATE_REPORT;
+                    }
+                }
+
+                ec = fop->xl->private;
+
+                config.version = EC_CONFIG_VERSION;
+                config.algorithm = EC_CONFIG_ALGORITHM;
+                config.gf_word_size = EC_GF_BITS;
+                config.bricks = ec->nodes;
+                config.redundancy = ec->redundancy;
+                config.chunk_size = EC_METHOD_CHUNK_SIZE;
+
+                err = ec_dict_set_config(fop->xdata, EC_XATTR_CONFIG, &config);
+                if (err != 0) {
+                    fop->error = -err;
+
+                    return EC_STATE_REPORT;
+                }
+                err = ec_dict_set_array(fop->xdata, EC_XATTR_VERSION, version,
+                                        EC_VERSION_SIZE);
+                if (err != 0) {
+                    fop->error = -err;
+
+                    return EC_STATE_REPORT;
+                }
+                err = ec_dict_set_number(fop->xdata, EC_XATTR_SIZE, 0);
+                if (err != 0) {
+                    fop->error = -err;
+
+                    return EC_STATE_REPORT;
+                }
+            }
+
+            /* Fall through */
+
+        case EC_STATE_LOCK:
+            ec_lock_prepare_parent_inode(fop, &fop->loc[0], NULL,
+                                         EC_UPDATE_DATA | EC_UPDATE_META);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                int32_t err;
+
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 3, cbk->count);
+
+                err = ec_loc_update(fop->xl, &fop->loc[0], cbk->inode,
+                                    &cbk->iatt[0]);
+                ec_cbk_set_error(cbk, -err, _gf_false);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.mknod != NULL) {
+                QUORUM_CBK(fop->cbks.mknod, fop, fop->req_frame, fop, fop->xl,
+                           cbk->op_ret, cbk->op_errno, fop->loc[0].inode,
+                           &cbk->iatt[0], &cbk->iatt[1], &cbk->iatt[2],
+                           cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.mknod != NULL) {
+                fop->cbks.mknod(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                NULL, NULL, NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_mknod(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_mknod_cbk_t func, void *data, loc_t *loc,
+         mode_t mode, dev_t rdev, mode_t umask, dict_t *xdata)
+{
+    ec_cbk_t callback = {.mknod = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(MKNOD) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_MKNOD, 0, target, fop_flags,
+                               ec_wind_mknod, ec_manager_mknod, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->mode[0] = mode;
+    fop->dev = rdev;
+    fop->mode[1] = umask;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: rename */
+
+int32_t
+ec_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct iatt *buf, struct iatt *preoldparent,
+              struct iatt *postoldparent, struct iatt *prenewparent,
+              struct iatt *postnewparent, dict_t *xdata)
+{
+    return ec_dir_write_cbk(frame, this, cookie, op_ret, op_errno, buf,
+                            preoldparent, postoldparent, prenewparent,
+                            postnewparent, xdata);
+}
+
+void
+ec_wind_rename(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_rename_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->rename,
+                      &fop->loc[0], &fop->loc[1], fop->xdata);
+}
+
+int32_t
+ec_manager_rename(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ec_lock_prepare_parent_inode(
+                fop, &fop->loc[0], &fop->loc[0],
+                EC_UPDATE_DATA | EC_UPDATE_META | EC_INODE_SIZE);
+            ec_lock_prepare_parent_inode(fop, &fop->loc[1], NULL,
+                                         EC_UPDATE_DATA | EC_UPDATE_META);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 5, cbk->count);
+
+                if (cbk->iatt[0].ia_type == IA_IFREG) {
+                    cbk->iatt[0].ia_size = fop->locks[0].size;
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.rename != NULL) {
+                QUORUM_CBK(fop->cbks.rename, fop, fop->req_frame, fop, fop->xl,
+                           cbk->op_ret, cbk->op_errno, &cbk->iatt[0],
+                           &cbk->iatt[1], &cbk->iatt[2], &cbk->iatt[3],
+                           &cbk->iatt[4], cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.rename != NULL) {
+                fop->cbks.rename(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                 NULL, NULL, NULL, NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_rename(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_rename_cbk_t func, void *data, loc_t *oldloc,
+          loc_t *newloc, dict_t *xdata)
+{
+    ec_cbk_t callback = {.rename = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(RENAME) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_RENAME, 0, target, fop_flags,
+                               ec_wind_rename, ec_manager_rename, callback,
+                               data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    if (oldloc != NULL) {
+        if (loc_copy(&fop->loc[0], oldloc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (newloc != NULL) {
+        if (loc_copy(&fop->loc[1], newloc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: rmdir */
+
+int32_t
+ec_rmdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, struct iatt *preparent, struct iatt *postparent,
+             dict_t *xdata)
+{
+    return ec_dir_write_cbk(frame, this, cookie, op_ret, op_errno, NULL,
+                            preparent, postparent, NULL, NULL, xdata);
+}
+
+void
+ec_wind_rmdir(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_rmdir_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->rmdir,
+                      &fop->loc[0], fop->int32, fop->xdata);
+}
+
+int32_t
+ec_manager_rmdir(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ec_lock_prepare_parent_inode(fop, &fop->loc[0], NULL,
+                                         EC_UPDATE_DATA | EC_UPDATE_META);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            ec_fop_prepare_answer(fop, _gf_false);
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.rmdir != NULL) {
+                QUORUM_CBK(fop->cbks.rmdir, fop, fop->req_frame, fop, fop->xl,
+                           cbk->op_ret, cbk->op_errno, &cbk->iatt[0],
+                           &cbk->iatt[1], cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.rmdir != NULL) {
+                fop->cbks.rmdir(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_rmdir(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_rmdir_cbk_t func, void *data, loc_t *loc,
+         int xflags, dict_t *xdata)
+{
+    ec_cbk_t callback = {.rmdir = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(RMDIR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_RMDIR, 0, target, fop_flags,
+                               ec_wind_rmdir, ec_manager_rmdir, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->int32 = xflags;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: symlink */
+
+int32_t
+ec_symlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, inode_t *inode,
+               struct iatt *buf, struct iatt *preparent,
+               struct iatt *postparent, dict_t *xdata)
+{
+    return ec_dir_write_cbk(frame, this, cookie, op_ret, op_errno, buf,
+                            preparent, postparent, NULL, NULL, xdata);
+}
+
+void
+ec_wind_symlink(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_symlink_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->symlink,
+                      fop->str[0], &fop->loc[0], fop->mode[0], fop->xdata);
+}
+
+int32_t
+ec_manager_symlink(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ec_lock_prepare_parent_inode(fop, &fop->loc[0], NULL,
+                                         EC_UPDATE_DATA | EC_UPDATE_META);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                int32_t err;
+
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 3, cbk->count);
+
+                err = ec_loc_update(fop->xl, &fop->loc[0], cbk->inode,
+                                    &cbk->iatt[0]);
+                ec_cbk_set_error(cbk, -err, _gf_false);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.symlink != NULL) {
+                QUORUM_CBK(fop->cbks.symlink, fop, fop->req_frame, fop, fop->xl,
+                           cbk->op_ret, cbk->op_errno, fop->loc[0].inode,
+                           &cbk->iatt[0], &cbk->iatt[1], &cbk->iatt[2],
+                           cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.symlink != NULL) {
+                fop->cbks.symlink(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                  NULL, NULL, NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_symlink(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_symlink_cbk_t func, void *data,
+           const char *linkname, loc_t *loc, mode_t umask, dict_t *xdata)
+{
+    ec_cbk_t callback = {.symlink = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(SYMLINK) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_SYMLINK, 0, target,
+                               fop_flags, ec_wind_symlink, ec_manager_symlink,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->mode[0] = umask;
+
+    if (linkname != NULL) {
+        fop->str[0] = gf_strdup(linkname);
+        if (fop->str[0] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: unlink */
+
+int32_t
+ec_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct iatt *preparent, struct iatt *postparent,
+              dict_t *xdata)
+{
+    return ec_dir_write_cbk(frame, this, cookie, op_ret, op_errno, NULL,
+                            preparent, postparent, NULL, NULL, xdata);
+}
+
+void
+ec_wind_unlink(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_unlink_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->unlink,
+                      &fop->loc[0], fop->int32, fop->xdata);
+}
+
+int32_t
+ec_manager_unlink(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ec_lock_prepare_parent_inode(fop, &fop->loc[0], NULL,
+                                         EC_UPDATE_DATA | EC_UPDATE_META);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            ec_fop_prepare_answer(fop, _gf_false);
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.unlink != NULL) {
+                QUORUM_CBK(fop->cbks.unlink, fop, fop->req_frame, fop, fop->xl,
+                           cbk->op_ret, cbk->op_errno, &cbk->iatt[0],
+                           &cbk->iatt[1], cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.unlink != NULL) {
+                fop->cbks.unlink(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                 NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_unlink(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_unlink_cbk_t func, void *data, loc_t *loc,
+          int xflags, dict_t *xdata)
+{
+    ec_cbk_t callback = {.unlink = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(UNLINK) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_UNLINK, 0, target, fop_flags,
+                               ec_wind_unlink, ec_manager_unlink, callback,
+                               data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->int32 = xflags;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
diff --git a/xlators/cluster/ec/src/ec-fops.h b/xlators/cluster/ec/src/ec-fops.h
new file mode 100644
index 00000000000..07edf8a7fec
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-fops.h
@@ -0,0 +1,254 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_FOPS_H__
+#define __EC_FOPS_H__
+
+#include <glusterfs/xlator.h>
+
+#include "ec-types.h"
+#include "ec-common.h"
+
+void
+ec_access(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_access_cbk_t func, void *data, loc_t *loc,
+          int32_t mask, dict_t *xdata);
+
+void
+ec_create(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_create_cbk_t func, void *data, loc_t *loc,
+          int32_t flags, mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata);
+
+void
+ec_entrylk(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_entrylk_cbk_t func, void *data,
+           const char *volume, loc_t *loc, const char *basename,
+           entrylk_cmd cmd, entrylk_type type, dict_t *xdata);
+
+void
+ec_fentrylk(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_fentrylk_cbk_t func, void *data,
+            const char *volume, fd_t *fd, const char *basename, entrylk_cmd cmd,
+            entrylk_type type, dict_t *xdata);
+
+void
+ec_flush(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_flush_cbk_t func, void *data, fd_t *fd,
+         dict_t *xdata);
+
+void
+ec_fsync(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_fsync_cbk_t func, void *data, fd_t *fd,
+         int32_t datasync, dict_t *xdata);
+
+void
+ec_fsyncdir(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_fsyncdir_cbk_t func, void *data, fd_t *fd,
+            int32_t datasync, dict_t *xdata);
+
+void
+ec_getxattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_getxattr_cbk_t func, void *data, loc_t *loc,
+            const char *name, dict_t *xdata);
+
+void
+ec_fgetxattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+             uint32_t fop_flags, fop_fgetxattr_cbk_t func, void *data, fd_t *fd,
+             const char *name, dict_t *xdata);
+
+void
+ec_heal(call_frame_t *frame, xlator_t *this, uintptr_t target,
+        uint32_t fop_flags, fop_heal_cbk_t func, void *data, loc_t *loc,
+        int32_t partial, dict_t *xdata);
+
+void
+ec_fheal(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_fheal_cbk_t func, void *data, fd_t *fd,
+         int32_t partial, dict_t *xdata);
+
+void
+ec_inodelk(call_frame_t *frame, xlator_t *this, gf_lkowner_t *owner,
+           uintptr_t target, uint32_t fop_flags, fop_inodelk_cbk_t func,
+           void *data, const char *volume, loc_t *loc, int32_t cmd,
+           struct gf_flock *flock, dict_t *xdata);
+
+void
+ec_finodelk(call_frame_t *frame, xlator_t *this, gf_lkowner_t *owner,
+            uintptr_t target, uint32_t fop_flags, fop_finodelk_cbk_t func,
+            void *data, const char *volume, fd_t *fd, int32_t cmd,
+            struct gf_flock *flock, dict_t *xdata);
+
+void
+ec_link(call_frame_t *frame, xlator_t *this, uintptr_t target,
+        uint32_t fop_flags, fop_link_cbk_t func, void *data, loc_t *oldloc,
+        loc_t *newloc, dict_t *xdata);
+
+void
+ec_lk(call_frame_t *frame, xlator_t *this, uintptr_t target, uint32_t fop_flags,
+      fop_lk_cbk_t func, void *data, fd_t *fd, int32_t cmd,
+      struct gf_flock *flock, dict_t *xdata);
+
+void
+ec_lookup(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_lookup_cbk_t func, void *data, loc_t *loc,
+          dict_t *xdata);
+
+void
+ec_mkdir(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_mkdir_cbk_t func, void *data, loc_t *loc,
+         mode_t mode, mode_t umask, dict_t *xdata);
+
+void
+ec_mknod(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_mknod_cbk_t func, void *data, loc_t *loc,
+         mode_t mode, dev_t rdev, mode_t umask, dict_t *xdata);
+
+void
+ec_open(call_frame_t *frame, xlator_t *this, uintptr_t target,
+        uint32_t fop_flags, fop_open_cbk_t func, void *data, loc_t *loc,
+        int32_t flags, fd_t *fd, dict_t *xdata);
+
+void
+ec_opendir(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_opendir_cbk_t func, void *data, loc_t *loc,
+           fd_t *fd, dict_t *xdata);
+
+void
+ec_readdir(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_readdir_cbk_t func, void *data, fd_t *fd,
+           size_t size, off_t offset, dict_t *xdata);
+
+void
+ec_readdirp(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_readdirp_cbk_t func, void *data, fd_t *fd,
+            size_t size, off_t offset, dict_t *xdata);
+
+void
+ec_readlink(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_readlink_cbk_t func, void *data, loc_t *loc,
+            size_t size, dict_t *xdata);
+
+void
+ec_readv(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_readv_cbk_t func, void *data, fd_t *fd,
+         size_t size, off_t offset, uint32_t flags, dict_t *xdata);
+
+void
+ec_removexattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+               uint32_t fop_flags, fop_removexattr_cbk_t func, void *data,
+               loc_t *loc, const char *name, dict_t *xdata);
+
+void
+ec_fremovexattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+                uint32_t fop_flags, fop_fremovexattr_cbk_t func, void *data,
+                fd_t *fd, const char *name, dict_t *xdata);
+
+void
+ec_rename(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_rename_cbk_t func, void *data, loc_t *oldloc,
+          loc_t *newloc, dict_t *xdata);
+
+void
+ec_rmdir(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_rmdir_cbk_t func, void *data, loc_t *loc,
+         int xflags, dict_t *xdata);
+
+void
+ec_setattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_setattr_cbk_t func, void *data, loc_t *loc,
+           struct iatt *stbuf, int32_t valid, dict_t *xdata);
+
+void
+ec_fsetattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_fsetattr_cbk_t func, void *data, fd_t *fd,
+            struct iatt *stbuf, int32_t valid, dict_t *xdata);
+
+void
+ec_setxattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_setxattr_cbk_t func, void *data, loc_t *loc,
+            dict_t *dict, int32_t flags, dict_t *xdata);
+
+void
+ec_fsetxattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+             uint32_t fop_flags, fop_fsetxattr_cbk_t func, void *data, fd_t *fd,
+             dict_t *dict, int32_t flags, dict_t *xdata);
+
+void
+ec_stat(call_frame_t *frame, xlator_t *this, uintptr_t target,
+        uint32_t fop_flags, fop_stat_cbk_t func, void *data, loc_t *loc,
+        dict_t *xdata);
+
+void
+ec_fstat(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_fstat_cbk_t func, void *data, fd_t *fd,
+         dict_t *xdata);
+
+void
+ec_statfs(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_statfs_cbk_t func, void *data, loc_t *loc,
+          dict_t *xdata);
+
+void
+ec_symlink(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_symlink_cbk_t func, void *data,
+           const char *linkname, loc_t *loc, mode_t umask, dict_t *xdata);
+
+void
+ec_fallocate(call_frame_t *frame, xlator_t *this, uintptr_t target,
+             uint32_t fop_flags, fop_fallocate_cbk_t func, void *data, fd_t *fd,
+             int32_t mode, off_t offset, size_t len, dict_t *xdata);
+
+void
+ec_discard(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_discard_cbk_t func, void *data, fd_t *fd,
+           off_t offset, size_t len, dict_t *xdata);
+
+void
+ec_truncate(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_truncate_cbk_t func, void *data, loc_t *loc,
+            off_t offset, dict_t *xdata);
+
+void
+ec_ftruncate(call_frame_t *frame, xlator_t *this, uintptr_t target,
+             uint32_t fop_flags, fop_ftruncate_cbk_t func, void *data, fd_t *fd,
+             off_t offset, dict_t *xdata);
+
+void
+ec_unlink(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_unlink_cbk_t func, void *data, loc_t *loc,
+          int xflags, dict_t *xdata);
+
+void
+ec_writev(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_writev_cbk_t func, void *data, fd_t *fd,
+          struct iovec *vector, int32_t count, off_t offset, uint32_t flags,
+          struct iobref *iobref, dict_t *xdata);
+
+void
+ec_xattrop(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_xattrop_cbk_t func, void *data, loc_t *loc,
+           gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata);
+
+void
+ec_fxattrop(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_fxattrop_cbk_t func, void *data, fd_t *fd,
+            gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata);
+
+void
+ec_seek(call_frame_t *frame, xlator_t *this, uintptr_t target,
+        uint32_t fop_flags, fop_seek_cbk_t func, void *data, fd_t *fd,
+        off_t offset, gf_seek_what_t what, dict_t *xdata);
+
+void
+ec_ipc(call_frame_t *frame, xlator_t *this, uintptr_t target,
+       uint32_t fop_flags, fop_ipc_cbk_t func, void *data, int32_t op,
+       dict_t *xdata);
+
+#endif /* __EC_FOPS_H__ */
diff --git a/xlators/cluster/ec/src/ec-galois.c b/xlators/cluster/ec/src/ec-galois.c
new file mode 100644
index 00000000000..6e4990c71f5
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-galois.c
@@ -0,0 +1,183 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <string.h>
+
+#include "ec-mem-types.h"
+#include "ec-gf8.h"
+#include "ec-helpers.h"
+
+static ec_gf_t *
+ec_gf_alloc(uint32_t bits, uint32_t mod)
+{
+    ec_gf_t *gf;
+
+    gf = GF_MALLOC(sizeof(ec_gf_t), ec_mt_ec_gf_t);
+    if (gf == NULL) {
+        goto failed;
+    }
+
+    gf->bits = bits;
+    gf->size = 1 << bits;
+    gf->mod = mod;
+
+    gf->log = GF_MALLOC(sizeof(uint32_t) * (gf->size * 2 - 1),
+                        gf_common_mt_int);
+    if (gf->log == NULL) {
+        goto failed_gf;
+    }
+    gf->pow = GF_MALLOC(sizeof(uint32_t) * (gf->size * 2 - 1),
+                        gf_common_mt_int);
+    if (gf->pow == NULL) {
+        goto failed_log;
+    }
+
+    return gf;
+
+failed_log:
+    GF_FREE(gf->log);
+failed_gf:
+    GF_FREE(gf);
+failed:
+    return EC_ERR(ENOMEM);
+}
+
+static void
+ec_gf_init_tables(ec_gf_t *gf)
+{
+    uint32_t i, tmp;
+
+    memset(gf->log, -1, sizeof(uint32_t) * gf->size);
+
+    gf->pow[0] = 1;
+    gf->log[0] = gf->size;
+    gf->log[1] = 0;
+    for (i = 1; i < gf->size; i++) {
+        tmp = gf->pow[i - 1] << 1;
+        if (tmp >= gf->size) {
+            tmp ^= gf->mod;
+        }
+        gf->pow[i + gf->size - 1] = gf->pow[i] = tmp;
+        gf->log[tmp + gf->size - 1] = gf->log[tmp] = i;
+    }
+}
+
+ec_gf_t *
+ec_gf_prepare(uint32_t bits, uint32_t mod)
+{
+    ec_gf_mul_t **tbl;
+    ec_gf_t *gf;
+    uint32_t i, j;
+
+    if (bits != 8) {
+        return EC_ERR(EINVAL);
+    }
+
+    tbl = ec_gf8_mul;
+    if (mod == 0) {
+        mod = 0x11d;
+    }
+
+    gf = ec_gf_alloc(bits, mod);
+    if (EC_IS_ERR(gf)) {
+        return gf;
+    }
+    ec_gf_init_tables(gf);
+
+    gf->table = tbl;
+    gf->min_ops = bits * bits;
+    gf->max_ops = 0;
+    gf->avg_ops = 0;
+    for (i = 1; i < gf->size; i++) {
+        for (j = 0; tbl[i]->ops[j].op != EC_GF_OP_END; j++) {
+        }
+        if (gf->max_ops < j) {
+            gf->max_ops = j;
+        }
+        if (gf->min_ops > j) {
+            gf->min_ops = j;
+        }
+        gf->avg_ops += j;
+    }
+    gf->avg_ops /= gf->size;
+
+    return gf;
+}
+
+void
+ec_gf_destroy(ec_gf_t *gf)
+{
+    GF_FREE(gf->pow);
+    GF_FREE(gf->log);
+    GF_FREE(gf);
+}
+
+uint32_t
+ec_gf_add(ec_gf_t *gf, uint32_t a, uint32_t b)
+{
+    if ((a >= gf->size) || (b >= gf->size)) {
+        return gf->size;
+    }
+
+    return a ^ b;
+}
+
+uint32_t
+ec_gf_mul(ec_gf_t *gf, uint32_t a, uint32_t b)
+{
+    if ((a >= gf->size) || (b >= gf->size)) {
+        return gf->size;
+    }
+
+    if ((a != 0) && (b != 0)) {
+        return gf->pow[gf->log[a] + gf->log[b]];
+    }
+
+    return 0;
+}
+
+uint32_t
+ec_gf_div(ec_gf_t *gf, uint32_t a, uint32_t b)
+{
+    if ((a >= gf->size) || (b >= gf->size)) {
+        return gf->size;
+    }
+
+    if (b != 0) {
+        if (a != 0) {
+            return gf->pow[gf->size - 1 + gf->log[a] - gf->log[b]];
+        }
+
+        return 0;
+    }
+
+    return gf->size;
+}
+
+uint32_t
+ec_gf_exp(ec_gf_t *gf, uint32_t a, uint32_t b)
+{
+    uint32_t r;
+
+    if ((a >= gf->size) || ((a == 0) && (b == 0))) {
+        return gf->size;
+    }
+
+    r = 1;
+    while (b != 0) {
+        if ((b & 1) != 0) {
+            r = ec_gf_mul(gf, r, a);
+        }
+        a = ec_gf_mul(gf, a, a);
+        b >>= 1;
+    }
+
+    return r;
+}
diff --git a/xlators/cluster/ec/src/ec-galois.h b/xlators/cluster/ec/src/ec-galois.h
new file mode 100644
index 00000000000..ed55d53e419
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-galois.h
@@ -0,0 +1,32 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_GALOIS_H__
+#define __EC_GALOIS_H__
+
+#include <inttypes.h>
+
+#include "ec-types.h"
+
+ec_gf_t *
+ec_gf_prepare(uint32_t bits, uint32_t mod);
+void
+ec_gf_destroy(ec_gf_t *gf);
+
+uint32_t
+ec_gf_add(ec_gf_t *gf, uint32_t a, uint32_t b);
+uint32_t
+ec_gf_mul(ec_gf_t *gf, uint32_t a, uint32_t b);
+uint32_t
+ec_gf_div(ec_gf_t *gf, uint32_t a, uint32_t b);
+uint32_t
+ec_gf_exp(ec_gf_t *gf, uint32_t a, uint32_t b);
+
+#endif /* __EC_GALOIS_H__ */
diff --git a/xlators/cluster/ec/src/ec-generic.c b/xlators/cluster/ec/src/ec-generic.c
new file mode 100644
index 00000000000..884deb93669
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-generic.c
@@ -0,0 +1,1591 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/byte-order.h>
+
+#include "ec.h"
+#include "ec-messages.h"
+#include "ec-helpers.h"
+#include "ec-common.h"
+#include "ec-combine.h"
+#include "ec-fops.h"
+
+/* FOP: flush */
+
+int32_t
+ec_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_FLUSH, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_flush(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_flush_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->flush, fop->fd,
+                      fop->xdata);
+}
+
+int32_t
+ec_manager_flush(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ec_lock_prepare_fd(fop, fop->fd, 0, 0, EC_RANGE_FULL);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_flush_size_version(fop);
+
+            return EC_STATE_DELAYED_START;
+
+        case EC_STATE_DELAYED_START:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            ec_fop_prepare_answer(fop, _gf_false);
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.flush != NULL) {
+                fop->cbks.flush(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                cbk->op_errno, cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DELAYED_START:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.flush != NULL) {
+                fop->cbks.flush(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+static int32_t
+ec_validate_fd(fd_t *fd, xlator_t *xl)
+{
+    uint64_t iversion = 0;
+    uint64_t fversion = 0;
+    ec_inode_t *inode_ctx = NULL;
+    ec_fd_t *fd_ctx = NULL;
+
+    LOCK(&fd->lock);
+    {
+        fd_ctx = __ec_fd_get(fd, xl);
+        if (fd_ctx) {
+            fversion = fd_ctx->bad_version;
+        }
+    }
+    UNLOCK(&fd->lock);
+
+    LOCK(&fd->inode->lock);
+    {
+        inode_ctx = __ec_inode_get(fd->inode, xl);
+        if (inode_ctx) {
+            iversion = inode_ctx->bad_version;
+        }
+    }
+    UNLOCK(&fd->inode->lock);
+    if (fversion < iversion) {
+        return EBADF;
+    }
+    return 0;
+}
+
+void
+ec_flush(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_flush_cbk_t func, void *data, fd_t *fd,
+         dict_t *xdata)
+{
+    ec_cbk_t callback = {.flush = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FLUSH) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    if (fd) {
+        error = ec_validate_fd(fd, this);
+        if (error) {
+            gf_msg(this->name, GF_LOG_ERROR, EBADF, EC_MSG_FD_BAD,
+                   "Failing %s on %s", gf_fop_list[GF_FOP_FLUSH],
+                   fd->inode ? uuid_utoa(fd->inode->gfid) : "");
+            goto out;
+        }
+    }
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FLUSH, 0, target, fop_flags,
+                               ec_wind_flush, ec_manager_flush, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/* FOP: fsync */
+
+int32_t
+ec_combine_fsync(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    if (!ec_iatt_combine(fop, dst->iatt, src->iatt, 2)) {
+        gf_msg(fop->xl->name, GF_LOG_NOTICE, 0, EC_MSG_IATT_MISMATCH,
+               "Mismatching iatt in "
+               "answers of 'GF_FOP_FSYNC'");
+
+        return 0;
+    }
+
+    return 1;
+}
+
+int32_t
+ec_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, struct iatt *prebuf, struct iatt *postbuf,
+             dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_FSYNC, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            if (prebuf != NULL) {
+                cbk->iatt[0] = *prebuf;
+            }
+            if (postbuf != NULL) {
+                cbk->iatt[1] = *postbuf;
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, ec_combine_fsync);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_fsync(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_fsync_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->fsync, fop->fd,
+                      fop->int32, fop->xdata);
+}
+
+int32_t
+ec_manager_fsync(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ec_lock_prepare_fd(fop, fop->fd, EC_QUERY_INFO, 0, EC_RANGE_FULL);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_flush_size_version(fop);
+
+            return EC_STATE_DELAYED_START;
+
+        case EC_STATE_DELAYED_START:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 2, cbk->count);
+
+                /* This shouldn't fail because we have the inode locked. */
+                GF_ASSERT(ec_get_inode_size(fop, fop->fd->inode,
+                                            &cbk->iatt[0].ia_size));
+                cbk->iatt[1].ia_size = cbk->iatt[0].ia_size;
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.fsync != NULL) {
+                fop->cbks.fsync(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                cbk->op_errno, &cbk->iatt[0], &cbk->iatt[1],
+                                cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+        case -EC_STATE_DELAYED_START:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.fsync != NULL) {
+                fop->cbks.fsync(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_fsync(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_fsync_cbk_t func, void *data, fd_t *fd,
+         int32_t datasync, dict_t *xdata)
+{
+    ec_cbk_t callback = {.fsync = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FSYNC) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    if (fd) {
+        error = ec_validate_fd(fd, this);
+        if (error) {
+            gf_msg(this->name, GF_LOG_ERROR, EBADF, EC_MSG_FD_BAD,
+                   "Failing %s on %s", gf_fop_list[GF_FOP_FSYNC],
+                   fd->inode ? uuid_utoa(fd->inode->gfid) : "");
+            goto out;
+        }
+    }
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FSYNC, 0, target, fop_flags,
+                               ec_wind_fsync, ec_manager_fsync, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->int32 = datasync;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: fsyncdir */
+
+int32_t
+ec_fsyncdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_FSYNCDIR, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_fsyncdir(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_fsyncdir_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->fsyncdir,
+                      fop->fd, fop->int32, fop->xdata);
+}
+
+int32_t
+ec_manager_fsyncdir(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ec_lock_prepare_fd(fop, fop->fd, 0, 0, EC_RANGE_FULL);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_flush_size_version(fop);
+
+            return EC_STATE_DELAYED_START;
+
+        case EC_STATE_DELAYED_START:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            ec_fop_prepare_answer(fop, _gf_false);
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.fsyncdir != NULL) {
+                fop->cbks.fsyncdir(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                   cbk->op_errno, cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+        case -EC_STATE_DELAYED_START:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.fsyncdir != NULL) {
+                fop->cbks.fsyncdir(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                   NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_fsyncdir(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_fsyncdir_cbk_t func, void *data, fd_t *fd,
+            int32_t datasync, dict_t *xdata)
+{
+    ec_cbk_t callback = {.fsyncdir = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FSYNCDIR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FSYNCDIR, 0, target,
+                               fop_flags, ec_wind_fsyncdir, ec_manager_fsyncdir,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->int32 = datasync;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/* FOP: lookup */
+
+void
+ec_lookup_rebuild(ec_t *ec, ec_fop_data_t *fop, ec_cbk_data_t *cbk)
+{
+    ec_inode_t *ctx = NULL;
+    uint64_t size = 0;
+    int32_t have_size = 0, err;
+
+    if (cbk->op_ret < 0) {
+        return;
+    }
+
+    ec_dict_del_array(cbk->xdata, EC_XATTR_VERSION, cbk->version,
+                      EC_VERSION_SIZE);
+
+    err = ec_loc_update(fop->xl, &fop->loc[0], cbk->inode, &cbk->iatt[0]);
+    if (ec_cbk_set_error(cbk, -err, _gf_true)) {
+        return;
+    }
+
+    LOCK(&cbk->inode->lock);
+
+    ctx = __ec_inode_get(cbk->inode, fop->xl);
+    if (ctx != NULL) {
+        if (ctx->have_version) {
+            cbk->version[0] = ctx->post_version[0];
+            cbk->version[1] = ctx->post_version[1];
+        }
+        if (ctx->have_size) {
+            size = ctx->post_size;
+            have_size = 1;
+        }
+    }
+
+    UNLOCK(&cbk->inode->lock);
+
+    if (cbk->iatt[0].ia_type == IA_IFREG) {
+        cbk->size = cbk->iatt[0].ia_size;
+        ec_dict_del_number(cbk->xdata, EC_XATTR_SIZE, &cbk->iatt[0].ia_size);
+        if (have_size) {
+            cbk->iatt[0].ia_size = size;
+        }
+    }
+}
+
+int32_t
+ec_combine_lookup(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    if (!ec_iatt_combine(fop, dst->iatt, src->iatt, 2)) {
+        gf_msg(fop->xl->name, GF_LOG_DEBUG, 0, EC_MSG_IATT_MISMATCH,
+               "Mismatching iatt in "
+               "answers of 'GF_FOP_LOOKUP'");
+
+        return 0;
+    }
+
+    return 1;
+}
+
+int32_t
+ec_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, inode_t *inode, struct iatt *buf, dict_t *xdata,
+              struct iatt *postparent)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+    uint64_t dirty[2] = {0};
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_LOOKUP, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            if (inode != NULL) {
+                cbk->inode = inode_ref(inode);
+                if (cbk->inode == NULL) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_INODE_REF_FAIL,
+                           "Failed to reference an inode.");
+
+                    goto out;
+                }
+            }
+            if (buf != NULL) {
+                cbk->iatt[0] = *buf;
+            }
+            if (postparent != NULL) {
+                cbk->iatt[1] = *postparent;
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+            ec_dict_del_array(xdata, EC_XATTR_DIRTY, dirty, EC_VERSION_SIZE);
+        }
+
+        ec_combine(cbk, ec_combine_lookup);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_lookup(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_lookup_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->lookup,
+                      &fop->loc[0], fop->xdata);
+}
+
+int32_t
+ec_manager_lookup(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+    int32_t err;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            if (fop->xdata == NULL) {
+                fop->xdata = dict_new();
+                if (fop->xdata == NULL) {
+                    gf_msg(fop->xl->name, GF_LOG_ERROR, ENOMEM,
+                           EC_MSG_LOOKUP_REQ_PREP_FAIL,
+                           "Unable to prepare "
+                           "lookup request");
+
+                    fop->error = ENOMEM;
+
+                    return EC_STATE_REPORT;
+                }
+            } else {
+                /*TODO: To be handled once we have 'syndromes' */
+                dict_del(fop->xdata, GF_CONTENT_KEY);
+            }
+            err = dict_set_uint64(fop->xdata, EC_XATTR_SIZE, 0);
+            if (err == 0) {
+                err = dict_set_uint64(fop->xdata, EC_XATTR_VERSION, 0);
+            }
+            if (err == 0) {
+                err = dict_set_uint64(fop->xdata, EC_XATTR_DIRTY, 0);
+            }
+            if (err != 0) {
+                gf_msg(fop->xl->name, GF_LOG_ERROR, -err,
+                       EC_MSG_LOOKUP_REQ_PREP_FAIL,
+                       "Unable to prepare lookup "
+                       "request");
+
+                fop->error = -err;
+
+                return EC_STATE_REPORT;
+            }
+
+            /* Fall through */
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            /*
+             * Lookup happens without any lock, so there is a chance that it
+             * will have answers before modification happened and after
+             * modification happened in the same response. So choose the next
+             * best answer when the answers don't match for EC_MINIMUM_MIN
+             */
+
+            if (!fop->answer && !list_empty(&fop->cbk_list)) {
+                fop->answer = list_entry(fop->cbk_list.next, ec_cbk_data_t,
+                                         list);
+            }
+
+            cbk = ec_fop_prepare_answer(fop, _gf_true);
+            if (cbk != NULL) {
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 2, cbk->count);
+
+                ec_lookup_rebuild(fop->xl->private, fop, cbk);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.lookup != NULL) {
+                fop->cbks.lookup(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                 cbk->op_errno, cbk->inode, &cbk->iatt[0],
+                                 cbk->xdata, &cbk->iatt[1]);
+            }
+
+            return EC_STATE_END;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.lookup != NULL) {
+                fop->cbks.lookup(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                 NULL, NULL, NULL, NULL);
+            }
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_lookup(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_lookup_cbk_t func, void *data, loc_t *loc,
+          dict_t *xdata)
+{
+    ec_cbk_t callback = {.lookup = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(LOOKUP) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_LOOKUP, EC_FLAG_LOCK_SHARED,
+                               target, fop_flags, ec_wind_lookup,
+                               ec_manager_lookup, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        /* Do not log failures here as a memory problem would have already
+         * been logged by the corresponding alloc functions */
+        if (fop->xdata == NULL)
+            goto out;
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: statfs */
+
+int32_t
+ec_combine_statfs(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    ec_statvfs_combine(&dst->statvfs, &src->statvfs);
+
+    return 1;
+}
+
+int32_t
+ec_statfs_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct statvfs *buf, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_STATFS, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            if (buf != NULL) {
+                cbk->statvfs = *buf;
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, ec_combine_statfs);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_statfs(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_statfs_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->statfs,
+                      &fop->loc[0], fop->xdata);
+}
+
+int32_t
+ec_manager_statfs(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk = NULL;
+    gf_boolean_t deem_statfs_enabled = _gf_false;
+    int32_t err = 0;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_true);
+            if (cbk != NULL) {
+                ec_t *ec = fop->xl->private;
+
+                if (cbk->xdata) {
+                    err = dict_get_int8(cbk->xdata, "quota-deem-statfs",
+                                        (int8_t *)&deem_statfs_enabled);
+                    if (err != -ENOENT) {
+                        ec_cbk_set_error(cbk, -err, _gf_true);
+                    }
+                }
+
+                if (err != 0 || deem_statfs_enabled == _gf_false) {
+                    cbk->statvfs.f_blocks *= ec->fragments;
+                    cbk->statvfs.f_bfree *= ec->fragments;
+                    cbk->statvfs.f_bavail *= ec->fragments;
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.statfs != NULL) {
+                fop->cbks.statfs(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                 cbk->op_errno, &cbk->statvfs, cbk->xdata);
+            }
+
+            return EC_STATE_END;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.statfs != NULL) {
+                fop->cbks.statfs(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                 NULL, NULL);
+            }
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_statfs(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_statfs_cbk_t func, void *data, loc_t *loc,
+          dict_t *xdata)
+{
+    ec_cbk_t callback = {.statfs = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(STATFS) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_STATFS, EC_FLAG_LOCK_SHARED,
+                               target, fop_flags, ec_wind_statfs,
+                               ec_manager_statfs, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
+
+/* FOP: xattrop */
+
+int32_t
+ec_combine_xattrop(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    if (!ec_dict_compare(dst->dict, src->dict)) {
+        gf_msg(fop->xl->name, GF_LOG_DEBUG, 0, EC_MSG_DICT_MISMATCH,
+               "Mismatching dictionary in "
+               "answers of 'GF_FOP_XATTROP'");
+
+        return 0;
+    }
+
+    return 1;
+}
+
+int32_t
+ec_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, dict_t *xattr, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_lock_link_t *link = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    uint64_t dirty[2] = {0};
+    data_t *data;
+    uint64_t *version;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, fop->id, idx, op_ret,
+                               op_errno);
+    if (!cbk)
+        goto out;
+
+    if (op_ret >= 0) {
+        cbk->dict = dict_ref(xattr);
+
+        data = dict_get(cbk->dict, EC_XATTR_VERSION);
+        if ((data != NULL) && (data->len >= sizeof(uint64_t))) {
+            version = (uint64_t *)data->data;
+
+            if (((ntoh64(version[0]) >> EC_SELFHEAL_BIT) & 1) != 0) {
+                LOCK(&fop->lock);
+
+                fop->healing |= 1ULL << idx;
+
+                UNLOCK(&fop->lock);
+            }
+        }
+
+        ec_dict_del_array(xattr, EC_XATTR_DIRTY, dirty, EC_VERSION_SIZE);
+        link = fop->data;
+        if (link) {
+            /*Keep a note of if the dirty is already set or not*/
+            link->dirty[0] |= (dirty[0] != 0);
+            link->dirty[1] |= (dirty[1] != 0);
+        }
+    }
+
+    if (xdata)
+        cbk->xdata = dict_ref(xdata);
+
+    ec_combine(cbk, ec_combine_xattrop);
+
+out:
+    if (fop)
+        ec_complete(fop);
+
+    return 0;
+}
+
+void
+ec_wind_xattrop(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_xattrop_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->xattrop,
+                      &fop->loc[0], fop->xattrop_flags, fop->dict, fop->xdata);
+}
+
+int32_t
+ec_manager_xattrop(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            if (fop->fd == NULL) {
+                ec_lock_prepare_inode(fop, &fop->loc[0], EC_UPDATE_META, 0,
+                                      EC_RANGE_FULL);
+            } else {
+                ec_lock_prepare_fd(fop, fop->fd, EC_UPDATE_META, 0,
+                                   EC_RANGE_FULL);
+            }
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                int32_t err;
+
+                err = ec_dict_combine(cbk, EC_COMBINE_DICT);
+                ec_cbk_set_error(cbk, -err, _gf_false);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->id == GF_FOP_XATTROP) {
+                if (fop->cbks.xattrop != NULL) {
+                    fop->cbks.xattrop(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                      cbk->op_errno, cbk->dict, cbk->xdata);
+                }
+            } else {
+                if (fop->cbks.fxattrop != NULL) {
+                    fop->cbks.fxattrop(fop->req_frame, fop, fop->xl,
+                                       cbk->op_ret, cbk->op_errno, cbk->dict,
+                                       cbk->xdata);
+                }
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->id == GF_FOP_XATTROP) {
+                if (fop->cbks.xattrop != NULL) {
+                    fop->cbks.xattrop(fop->req_frame, fop, fop->xl, -1,
+                                      fop->error, NULL, NULL);
+                }
+            } else {
+                if (fop->cbks.fxattrop != NULL) {
+                    fop->cbks.fxattrop(fop->req_frame, fop, fop->xl, -1,
+                                       fop->error, NULL, NULL);
+                }
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_xattrop(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_xattrop_cbk_t func, void *data, loc_t *loc,
+           gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+    ec_cbk_t callback = {.xattrop = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(XATTROP) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_XATTROP, 0, target,
+                               fop_flags, ec_wind_xattrop, ec_manager_xattrop,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->xattrop_flags = optype;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xattr != NULL) {
+        fop->dict = dict_ref(xattr);
+        if (fop->dict == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
+
+void
+ec_wind_fxattrop(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_xattrop_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->fxattrop,
+                      fop->fd, fop->xattrop_flags, fop->dict, fop->xdata);
+}
+
+void
+ec_fxattrop(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_fxattrop_cbk_t func, void *data, fd_t *fd,
+            gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+    ec_cbk_t callback = {.fxattrop = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FXATTROP) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FXATTROP, 0, target,
+                               fop_flags, ec_wind_fxattrop, ec_manager_xattrop,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->xattrop_flags = optype;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xattr != NULL) {
+        fop->dict = dict_ref(xattr);
+        if (fop->dict == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
+
+/* FOP: IPC */
+
+int32_t
+ec_ipc_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+           int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_IPC, idx, op_ret,
+                               op_errno);
+
+    if (cbk != NULL) {
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+        }
+
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_ipc(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_ipc_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->ipc, fop->int32,
+                      fop->xdata);
+}
+
+int32_t
+ec_manager_ipc(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            ec_fop_prepare_answer(fop, _gf_true);
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+            if (fop->cbks.ipc != NULL) {
+                fop->cbks.ipc(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                              cbk->op_errno, cbk->xdata);
+            }
+
+            return EC_STATE_END;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.ipc != NULL) {
+                fop->cbks.ipc(fop->req_frame, fop, fop->xl, -1, fop->error,
+                              NULL);
+            }
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_ipc(call_frame_t *frame, xlator_t *this, uintptr_t target,
+       uint32_t fop_flags, fop_ipc_cbk_t func, void *data, int32_t op,
+       dict_t *xdata)
+{
+    ec_cbk_t callback = {.ipc = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(IPC) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_IPC, 0, target, fop_flags,
+                               ec_wind_ipc, ec_manager_ipc, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+    }
+    fop->int32 = op;
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
diff --git a/xlators/cluster/ec/src/ec-gf8.c b/xlators/cluster/ec/src/ec-gf8.c
new file mode 100644
index 00000000000..039adae5929
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-gf8.c
@@ -0,0 +1,5882 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "ec-gf8.h"
+
+static ec_gf_op_t ec_gf8_mul_00_ops[] = {{EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_00 = {0,
+                                    {
+                                        0,
+                                    },
+                                    ec_gf8_mul_00_ops};
+
+static ec_gf_op_t ec_gf8_mul_01_ops[] = {{EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_01 = {8,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_01_ops};
+
+static ec_gf_op_t ec_gf8_mul_02_ops[] = {{EC_GF_OP_XOR2, 1, 7, 0},
+                                         {EC_GF_OP_XOR2, 2, 7, 0},
+                                         {EC_GF_OP_XOR2, 3, 7, 0},
+                                         {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_02 = {8,
+                                    {
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_02_ops};
+
+static ec_gf_op_t ec_gf8_mul_03_ops[] = {
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_COPY, 8, 3, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 4, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_03 = {9,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        8,
+                                    },
+                                    ec_gf8_mul_03_ops};
+
+static ec_gf_op_t ec_gf8_mul_04_ops[] = {
+    {EC_GF_OP_XOR3, 8, 6, 7}, {EC_GF_OP_XOR2, 2, 8, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 1, 8, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_04 = {9,
+                                    {
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        8,
+                                    },
+                                    ec_gf8_mul_04_ops};
+
+static ec_gf_op_t ec_gf8_mul_05_ops[] = {
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_05 = {8,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        6,
+                                        7,
+                                        3,
+                                        4,
+                                        5,
+                                    },
+                                    ec_gf8_mul_05_ops};
+
+static ec_gf_op_t ec_gf8_mul_06_ops[] = {
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_COPY, 8, 2, 0},
+    {EC_GF_OP_XOR2, 8, 3, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_06 = {9,
+                                    {
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        8,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_06_ops};
+
+static ec_gf_op_t ec_gf8_mul_07_ops[] = {
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_07 = {8,
+                                    {
+                                        6,
+                                        0,
+                                        1,
+                                        3,
+                                        2,
+                                        4,
+                                        5,
+                                        7,
+                                    },
+                                    ec_gf8_mul_07_ops};
+
+static ec_gf_op_t ec_gf8_mul_08_ops[] = {
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR3, 8, 6, 7},
+    {EC_GF_OP_XOR2, 1, 8, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 2, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_08 = {9,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        8,
+                                    },
+                                    ec_gf8_mul_08_ops};
+
+static ec_gf_op_t ec_gf8_mul_09_ops[] = {
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_09 = {8,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        5,
+                                        6,
+                                        7,
+                                        4,
+                                    },
+                                    ec_gf8_mul_09_ops};
+
+static ec_gf_op_t ec_gf8_mul_0A_ops[] = {
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_0A = {8,
+                                    {
+                                        5,
+                                        0,
+                                        1,
+                                        2,
+                                        6,
+                                        7,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_0A_ops};
+
+static ec_gf_op_t ec_gf8_mul_0B_ops[] = {
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_COPY, 9, 3, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_COPY, 8, 5, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR3, 3, 8, 6}, {EC_GF_OP_XOR2, 1, 9, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_0B = {10,
+                                    {
+                                        7,
+                                        1,
+                                        5,
+                                        2,
+                                        4,
+                                        3,
+                                        0,
+                                        6,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_0B_ops};
+
+static ec_gf_op_t ec_gf8_mul_0C_ops[] = {
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_COPY, 8, 1, 0},
+    {EC_GF_OP_XOR2, 8, 2, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_0C = {9,
+                                    {
+                                        5,
+                                        7,
+                                        0,
+                                        1,
+                                        8,
+                                        2,
+                                        3,
+                                        4,
+                                        6,
+                                    },
+                                    ec_gf8_mul_0C_ops};
+
+static ec_gf_op_t ec_gf8_mul_0D_ops[] = {
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR3, 8, 2, 4}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 1, 8, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR3, 2, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_0D = {9,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        3,
+                                        1,
+                                        0,
+                                        2,
+                                        4,
+                                        8,
+                                    },
+                                    ec_gf8_mul_0D_ops};
+
+static ec_gf_op_t ec_gf8_mul_0E_ops[] = {
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_0E = {8,
+                                    {
+                                        7,
+                                        0,
+                                        6,
+                                        1,
+                                        3,
+                                        2,
+                                        4,
+                                        5,
+                                    },
+                                    ec_gf8_mul_0E_ops};
+
+static ec_gf_op_t ec_gf8_mul_0F_ops[] = {
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_0F = {8,
+                                    {
+                                        1,
+                                        0,
+                                        5,
+                                        6,
+                                        7,
+                                        2,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_0F_ops};
+
+static ec_gf_op_t ec_gf8_mul_10_ops[] = {
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_10 = {8,
+                                    {
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                    },
+                                    ec_gf8_mul_10_ops};
+
+static ec_gf_op_t ec_gf8_mul_11_ops[] = {
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_11 = {8,
+                                    {
+                                        4,
+                                        1,
+                                        2,
+                                        6,
+                                        0,
+                                        5,
+                                        7,
+                                        3,
+                                    },
+                                    ec_gf8_mul_11_ops};
+
+static ec_gf_op_t ec_gf8_mul_12_ops[] = {
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_12 = {8,
+                                    {
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        5,
+                                        6,
+                                        4,
+                                    },
+                                    ec_gf8_mul_12_ops};
+
+static ec_gf_op_t ec_gf8_mul_13_ops[] = {
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR3, 8, 3, 7},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 6, 8, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 0, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_13 = {9,
+                                    {
+                                        4,
+                                        5,
+                                        2,
+                                        6,
+                                        0,
+                                        1,
+                                        7,
+                                        3,
+                                        8,
+                                    },
+                                    ec_gf8_mul_13_ops};
+
+static ec_gf_op_t ec_gf8_mul_14_ops[] = {
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_14 = {8,
+                                    {
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        4,
+                                        5,
+                                        3,
+                                    },
+                                    ec_gf8_mul_14_ops};
+
+static ec_gf_op_t ec_gf8_mul_15_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR3, 5, 8, 7},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_15 = {9,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        4,
+                                        7,
+                                        6,
+                                        5,
+                                        3,
+                                        8,
+                                    },
+                                    ec_gf8_mul_15_ops};
+
+static ec_gf_op_t ec_gf8_mul_16_ops[] = {
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_16 = {8,
+                                    {
+                                        6,
+                                        7,
+                                        4,
+                                        1,
+                                        2,
+                                        3,
+                                        5,
+                                        0,
+                                    },
+                                    ec_gf8_mul_16_ops};
+
+static ec_gf_op_t ec_gf8_mul_17_ops[] = {
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_17 = {8,
+                                    {
+                                        5,
+                                        7,
+                                        0,
+                                        1,
+                                        3,
+                                        2,
+                                        4,
+                                        6,
+                                    },
+                                    ec_gf8_mul_17_ops};
+
+static ec_gf_op_t ec_gf8_mul_18_ops[] = {
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 6, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_18 = {9,
+                                    {
+                                        4,
+                                        5,
+                                        7,
+                                        6,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        8,
+                                    },
+                                    ec_gf8_mul_18_ops};
+
+static ec_gf_op_t ec_gf8_mul_19_ops[] = {
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_19 = {8,
+                                    {
+                                        0,
+                                        5,
+                                        2,
+                                        6,
+                                        7,
+                                        1,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_19_ops};
+
+static ec_gf_op_t ec_gf8_mul_1A_ops[] = {
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_1A = {8,
+                                    {
+                                        7,
+                                        0,
+                                        4,
+                                        5,
+                                        3,
+                                        1,
+                                        2,
+                                        6,
+                                    },
+                                    ec_gf8_mul_1A_ops};
+
+static ec_gf_op_t ec_gf8_mul_1B_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_1B = {8,
+                                    {
+                                        7,
+                                        4,
+                                        5,
+                                        6,
+                                        3,
+                                        1,
+                                        2,
+                                        0,
+                                    },
+                                    ec_gf8_mul_1B_ops};
+
+static ec_gf_op_t ec_gf8_mul_1C_ops[] = {
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_1C = {8,
+                                    {
+                                        5,
+                                        4,
+                                        3,
+                                        0,
+                                        1,
+                                        7,
+                                        2,
+                                        6,
+                                    },
+                                    ec_gf8_mul_1C_ops};
+
+static ec_gf_op_t ec_gf8_mul_1D_ops[] = {
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR3, 8, 4, 2},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 5, 8, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_1D = {9,
+                                    {
+                                        0,
+                                        7,
+                                        5,
+                                        8,
+                                        2,
+                                        3,
+                                        4,
+                                        1,
+                                        6,
+                                    },
+                                    ec_gf8_mul_1D_ops};
+
+static ec_gf_op_t ec_gf8_mul_1E_ops[] = {
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_1E = {8,
+                                    {
+                                        4,
+                                        7,
+                                        5,
+                                        1,
+                                        6,
+                                        0,
+                                        2,
+                                        3,
+                                    },
+                                    ec_gf8_mul_1E_ops};
+
+static ec_gf_op_t ec_gf8_mul_1F_ops[] = {
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR3, 8, 3, 7},
+    {EC_GF_OP_XOR2, 4, 8, 0}, {EC_GF_OP_XOR2, 1, 8, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_1F = {9,
+                                    {
+                                        1,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        3,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_1F_ops};
+
+static ec_gf_op_t ec_gf8_mul_20_ops[] = {
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_20 = {8,
+                                    {
+                                        7,
+                                        4,
+                                        5,
+                                        6,
+                                        3,
+                                        0,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_20_ops};
+
+static ec_gf_op_t ec_gf8_mul_21_ops[] = {
+    {EC_GF_OP_COPY, 9, 0, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR3, 8, 7, 5}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 3, 8, 0}, {EC_GF_OP_XOR2, 2, 8, 0},
+    {EC_GF_OP_XOR2, 4, 9, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_21 = {10,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        7,
+                                        5,
+                                        4,
+                                        3,
+                                        6,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_21_ops};
+
+static ec_gf_op_t ec_gf8_mul_22_ops[] = {
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_22 = {8,
+                                    {
+                                        3,
+                                        0,
+                                        5,
+                                        2,
+                                        6,
+                                        4,
+                                        1,
+                                        7,
+                                    },
+                                    ec_gf8_mul_22_ops};
+
+static ec_gf_op_t ec_gf8_mul_23_ops[] = {
+    {EC_GF_OP_COPY, 8, 2, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 3, 8, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_23 = {9,
+                                    {
+                                        0,
+                                        4,
+                                        3,
+                                        2,
+                                        5,
+                                        6,
+                                        1,
+                                        8,
+                                        7,
+                                    },
+                                    ec_gf8_mul_23_ops};
+
+static ec_gf_op_t ec_gf8_mul_24_ops[] = {
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_24 = {8,
+                                    {
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        4,
+                                        5,
+                                        3,
+                                    },
+                                    ec_gf8_mul_24_ops};
+
+static ec_gf_op_t ec_gf8_mul_25_ops[] = {
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_25 = {8,
+                                    {
+                                        2,
+                                        7,
+                                        0,
+                                        1,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_25_ops};
+
+static ec_gf_op_t ec_gf8_mul_26_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_26 = {8,
+                                    {
+                                        3,
+                                        4,
+                                        1,
+                                        2,
+                                        0,
+                                        5,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_26_ops};
+
+static ec_gf_op_t ec_gf8_mul_27_ops[] = {
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_27 = {8,
+                                    {
+                                        3,
+                                        0,
+                                        1,
+                                        2,
+                                        6,
+                                        7,
+                                        4,
+                                        5,
+                                    },
+                                    ec_gf8_mul_27_ops};
+
+static ec_gf_op_t ec_gf8_mul_28_ops[] = {
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_28 = {8,
+                                    {
+                                        5,
+                                        6,
+                                        3,
+                                        0,
+                                        1,
+                                        2,
+                                        4,
+                                        7,
+                                    },
+                                    ec_gf8_mul_28_ops};
+
+static ec_gf_op_t ec_gf8_mul_29_ops[] = {
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_29 = {8,
+                                    {
+                                        4,
+                                        6,
+                                        3,
+                                        5,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_29_ops};
+
+static ec_gf_op_t ec_gf8_mul_2A_ops[] = {
+    {EC_GF_OP_COPY, 8, 1, 0}, {EC_GF_OP_XOR2, 8, 0, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR3, 6, 8, 4}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_2A = {9,
+                                    {
+                                        3,
+                                        4,
+                                        7,
+                                        2,
+                                        6,
+                                        5,
+                                        1,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_2A_ops};
+
+static ec_gf_op_t ec_gf8_mul_2B_ops[] = {
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_2B = {8,
+                                    {
+                                        3,
+                                        4,
+                                        7,
+                                        5,
+                                        6,
+                                        0,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_2B_ops};
+
+static ec_gf_op_t ec_gf8_mul_2C_ops[] = {
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_2C = {8,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        2,
+                                        3,
+                                        4,
+                                        1,
+                                    },
+                                    ec_gf8_mul_2C_ops};
+
+static ec_gf_op_t ec_gf8_mul_2D_ops[] = {
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR3, 8, 4, 6},
+    {EC_GF_OP_XOR2, 5, 8, 0}, {EC_GF_OP_XOR2, 7, 8, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_2D = {9,
+                                    {
+                                        7,
+                                        0,
+                                        3,
+                                        5,
+                                        1,
+                                        4,
+                                        2,
+                                        6,
+                                        8,
+                                    },
+                                    ec_gf8_mul_2D_ops};
+
+static ec_gf_op_t ec_gf8_mul_2E_ops[] = {
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_COPY, 8, 4, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 8, 7, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 2, 8, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 6, 8, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_2E = {9,
+                                    {
+                                        5,
+                                        0,
+                                        7,
+                                        3,
+                                        2,
+                                        6,
+                                        4,
+                                        1,
+                                        8,
+                                    },
+                                    ec_gf8_mul_2E_ops};
+
+static ec_gf_op_t ec_gf8_mul_2F_ops[] = {
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR3, 8, 7, 6}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 3, 8, 0},
+    {EC_GF_OP_XOR2, 2, 8, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_2F = {9,
+                                    {
+                                        6,
+                                        3,
+                                        2,
+                                        5,
+                                        7,
+                                        0,
+                                        1,
+                                        4,
+                                        8,
+                                    },
+                                    ec_gf8_mul_2F_ops};
+
+static ec_gf_op_t ec_gf8_mul_30_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 8, 1, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR3, 6, 8, 7},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_30 = {9,
+                                    {
+                                        3,
+                                        4,
+                                        7,
+                                        5,
+                                        0,
+                                        6,
+                                        1,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_30_ops};
+
+static ec_gf_op_t ec_gf8_mul_31_ops[] = {
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_31 = {8,
+                                    {
+                                        7,
+                                        1,
+                                        4,
+                                        5,
+                                        6,
+                                        0,
+                                        2,
+                                        3,
+                                    },
+                                    ec_gf8_mul_31_ops};
+
+static ec_gf_op_t ec_gf8_mul_32_ops[] = {
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_32 = {8,
+                                    {
+                                        3,
+                                        4,
+                                        6,
+                                        7,
+                                        5,
+                                        0,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_32_ops};
+
+static ec_gf_op_t ec_gf8_mul_33_ops[] = {
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_33 = {8,
+                                    {
+                                        5,
+                                        4,
+                                        3,
+                                        0,
+                                        2,
+                                        1,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_33_ops};
+
+static ec_gf_op_t ec_gf8_mul_34_ops[] = {
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_34 = {8,
+                                    {
+                                        7,
+                                        5,
+                                        3,
+                                        0,
+                                        2,
+                                        4,
+                                        1,
+                                        6,
+                                    },
+                                    ec_gf8_mul_34_ops};
+
+static ec_gf_op_t ec_gf8_mul_35_ops[] = {
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_35 = {8,
+                                    {
+                                        6,
+                                        7,
+                                        5,
+                                        4,
+                                        2,
+                                        0,
+                                        1,
+                                        3,
+                                    },
+                                    ec_gf8_mul_35_ops};
+
+static ec_gf_op_t ec_gf8_mul_36_ops[] = {
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_36 = {8,
+                                    {
+                                        6,
+                                        7,
+                                        4,
+                                        1,
+                                        2,
+                                        3,
+                                        0,
+                                        5,
+                                    },
+                                    ec_gf8_mul_36_ops};
+
+static ec_gf_op_t ec_gf8_mul_37_ops[] = {
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR3, 8, 0, 1},
+    {EC_GF_OP_XOR2, 3, 8, 0}, {EC_GF_OP_XOR2, 7, 8, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_37 = {9,
+                                    {
+                                        6,
+                                        7,
+                                        2,
+                                        1,
+                                        0,
+                                        3,
+                                        4,
+                                        5,
+                                        8,
+                                    },
+                                    ec_gf8_mul_37_ops};
+
+static ec_gf_op_t ec_gf8_mul_38_ops[] = {
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR3, 8, 6, 7},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 0, 8, 0}, {EC_GF_OP_XOR2, 4, 8, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_38 = {9,
+                                    {
+                                        4,
+                                        5,
+                                        6,
+                                        3,
+                                        0,
+                                        1,
+                                        7,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_38_ops};
+
+static ec_gf_op_t ec_gf8_mul_39_ops[] = {
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_39 = {8,
+                                    {
+                                        1,
+                                        6,
+                                        3,
+                                        0,
+                                        5,
+                                        2,
+                                        4,
+                                        7,
+                                    },
+                                    ec_gf8_mul_39_ops};
+
+static ec_gf_op_t ec_gf8_mul_3A_ops[] = {
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_3A = {8,
+                                    {
+                                        3,
+                                        4,
+                                        7,
+                                        0,
+                                        5,
+                                        6,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_3A_ops};
+
+static ec_gf_op_t ec_gf8_mul_3B_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR3, 8, 7, 3}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 1, 8, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_3B = {9,
+                                    {
+                                        3,
+                                        0,
+                                        1,
+                                        7,
+                                        6,
+                                        2,
+                                        4,
+                                        8,
+                                        5,
+                                    },
+                                    ec_gf8_mul_3B_ops};
+
+static ec_gf_op_t ec_gf8_mul_3C_ops[] = {
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_3C = {8,
+                                    {
+                                        3,
+                                        6,
+                                        4,
+                                        1,
+                                        7,
+                                        2,
+                                        0,
+                                        5,
+                                    },
+                                    ec_gf8_mul_3C_ops};
+
+static ec_gf_op_t ec_gf8_mul_3D_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_3D = {8,
+                                    {
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                    },
+                                    ec_gf8_mul_3D_ops};
+
+static ec_gf_op_t ec_gf8_mul_3E_ops[] = {
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_3E = {8,
+                                    {
+                                        6,
+                                        1,
+                                        2,
+                                        7,
+                                        0,
+                                        3,
+                                        5,
+                                        4,
+                                    },
+                                    ec_gf8_mul_3E_ops};
+
+static ec_gf_op_t ec_gf8_mul_3F_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0},  {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_COPY, 10, 4, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0},  {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0},  {EC_GF_OP_COPY, 9, 2, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0},  {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0},  {EC_GF_OP_XOR3, 4, 9, 7},
+    {EC_GF_OP_XOR2, 3, 4, 0},  {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0},  {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 3, 10, 0}, {EC_GF_OP_XOR2, 5, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_3F = {11,
+                                    {
+                                        1,
+                                        7,
+                                        6,
+                                        2,
+                                        4,
+                                        3,
+                                        5,
+                                        0,
+                                        8,
+                                        9,
+                                        10,
+                                    },
+                                    ec_gf8_mul_3F_ops};
+
+static ec_gf_op_t ec_gf8_mul_40_ops[] = {
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR3, 8, 7, 6},
+    {EC_GF_OP_XOR2, 1, 8, 0}, {EC_GF_OP_XOR2, 5, 8, 0},
+    {EC_GF_OP_XOR2, 4, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_40 = {9,
+                                    {
+                                        5,
+                                        7,
+                                        4,
+                                        6,
+                                        2,
+                                        3,
+                                        0,
+                                        1,
+                                        8,
+                                    },
+                                    ec_gf8_mul_40_ops};
+
+static ec_gf_op_t ec_gf8_mul_41_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 8, 4, 0},
+    {EC_GF_OP_XOR2, 8, 5, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_41 = {9,
+                                    {
+                                        0,
+                                        7,
+                                        6,
+                                        5,
+                                        3,
+                                        4,
+                                        8,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_41_ops};
+
+static ec_gf_op_t ec_gf8_mul_42_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 8, 3, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 7, 8, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_42 = {9,
+                                    {
+                                        2,
+                                        7,
+                                        1,
+                                        6,
+                                        4,
+                                        3,
+                                        0,
+                                        5,
+                                        8,
+                                    },
+                                    ec_gf8_mul_42_ops};
+
+static ec_gf_op_t ec_gf8_mul_43_ops[] = {
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_43 = {8,
+                                    {
+                                        2,
+                                        6,
+                                        4,
+                                        1,
+                                        7,
+                                        3,
+                                        0,
+                                        5,
+                                    },
+                                    ec_gf8_mul_43_ops};
+
+static ec_gf_op_t ec_gf8_mul_44_ops[] = {
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_44 = {8,
+                                    {
+                                        2,
+                                        3,
+                                        4,
+                                        1,
+                                        6,
+                                        5,
+                                        0,
+                                        7,
+                                    },
+                                    ec_gf8_mul_44_ops};
+
+static ec_gf_op_t ec_gf8_mul_45_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_45 = {8,
+                                    {
+                                        2,
+                                        3,
+                                        0,
+                                        1,
+                                        7,
+                                        4,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_45_ops};
+
+static ec_gf_op_t ec_gf8_mul_46_ops[] = {
+    {EC_GF_OP_XOR3, 8, 2, 4}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 8, 0, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 1, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_46 = {9,
+                                    {
+                                        2,
+                                        0,
+                                        1,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        8,
+                                    },
+                                    ec_gf8_mul_46_ops};
+
+static ec_gf_op_t ec_gf8_mul_47_ops[] = {
+    {EC_GF_OP_XOR3, 8, 0, 1}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 3, 8, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 4, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_47 = {9,
+                                    {
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        8,
+                                    },
+                                    ec_gf8_mul_47_ops};
+
+static ec_gf_op_t ec_gf8_mul_48_ops[] = {
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_48 = {8,
+                                    {
+                                        4,
+                                        5,
+                                        6,
+                                        0,
+                                        1,
+                                        3,
+                                        7,
+                                        2,
+                                    },
+                                    ec_gf8_mul_48_ops};
+
+static ec_gf_op_t ec_gf8_mul_49_ops[] = {
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR3, 8, 0, 6},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 7, 8, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR3, 1, 8, 5},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_49 = {9,
+                                    {
+                                        7,
+                                        2,
+                                        4,
+                                        0,
+                                        3,
+                                        5,
+                                        1,
+                                        6,
+                                        8,
+                                    },
+                                    ec_gf8_mul_49_ops};
+
+static ec_gf_op_t ec_gf8_mul_4A_ops[] = {
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_4A = {8,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        3,
+                                        4,
+                                        2,
+                                    },
+                                    ec_gf8_mul_4A_ops};
+
+static ec_gf_op_t ec_gf8_mul_4B_ops[] = {
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR3, 8, 3, 7}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 4, 8, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 5, 8, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_4B = {9,
+                                    {
+                                        5,
+                                        3,
+                                        6,
+                                        7,
+                                        0,
+                                        2,
+                                        4,
+                                        1,
+                                        8,
+                                    },
+                                    ec_gf8_mul_4B_ops};
+
+static ec_gf_op_t ec_gf8_mul_4C_ops[] = {
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_4C = {8,
+                                    {
+                                        5,
+                                        3,
+                                        4,
+                                        7,
+                                        0,
+                                        6,
+                                        2,
+                                        1,
+                                    },
+                                    ec_gf8_mul_4C_ops};
+
+static ec_gf_op_t ec_gf8_mul_4D_ops[] = {
+    {EC_GF_OP_COPY, 8, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR3, 9, 3, 1},
+    {EC_GF_OP_XOR2, 5, 9, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR3, 0, 8, 2},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_4D = {10,
+                                    {
+                                        0,
+                                        9,
+                                        3,
+                                        5,
+                                        6,
+                                        4,
+                                        7,
+                                        1,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_4D_ops};
+
+static ec_gf_op_t ec_gf8_mul_4E_ops[] = {
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_4E = {8,
+                                    {
+                                        2,
+                                        3,
+                                        0,
+                                        1,
+                                        5,
+                                        6,
+                                        7,
+                                        4,
+                                    },
+                                    ec_gf8_mul_4E_ops};
+
+static ec_gf_op_t ec_gf8_mul_4F_ops[] = {
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_4F = {8,
+                                    {
+                                        0,
+                                        3,
+                                        5,
+                                        6,
+                                        1,
+                                        2,
+                                        7,
+                                        4,
+                                    },
+                                    ec_gf8_mul_4F_ops};
+
+static ec_gf_op_t ec_gf8_mul_50_ops[] = {
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_50 = {8,
+                                    {
+                                        4,
+                                        5,
+                                        7,
+                                        3,
+                                        0,
+                                        1,
+                                        2,
+                                        6,
+                                    },
+                                    ec_gf8_mul_50_ops};
+
+static ec_gf_op_t ec_gf8_mul_51_ops[] = {
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_51 = {8,
+                                    {
+                                        0,
+                                        1,
+                                        7,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_51_ops};
+
+static ec_gf_op_t ec_gf8_mul_52_ops[] = {
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_COPY, 9, 4, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR3, 3, 5, 8},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 2, 9, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_52 = {10,
+                                    {
+                                        2,
+                                        3,
+                                        1,
+                                        4,
+                                        6,
+                                        7,
+                                        0,
+                                        5,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_52_ops};
+
+static ec_gf_op_t ec_gf8_mul_53_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_53 = {8,
+                                    {
+                                        2,
+                                        0,
+                                        1,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        3,
+                                    },
+                                    ec_gf8_mul_53_ops};
+
+static ec_gf_op_t ec_gf8_mul_54_ops[] = {
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_54 = {8,
+                                    {
+                                        7,
+                                        3,
+                                        0,
+                                        4,
+                                        2,
+                                        6,
+                                        5,
+                                        1,
+                                    },
+                                    ec_gf8_mul_54_ops};
+
+static ec_gf_op_t ec_gf8_mul_55_ops[] = {
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_55 = {8,
+                                    {
+                                        1,
+                                        5,
+                                        6,
+                                        4,
+                                        3,
+                                        7,
+                                        2,
+                                        0,
+                                    },
+                                    ec_gf8_mul_55_ops};
+
+static ec_gf_op_t ec_gf8_mul_56_ops[] = {
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_56 = {8,
+                                    {
+                                        2,
+                                        3,
+                                        0,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        1,
+                                    },
+                                    ec_gf8_mul_56_ops};
+
+static ec_gf_op_t ec_gf8_mul_57_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_57 = {8,
+                                    {
+                                        2,
+                                        3,
+                                        0,
+                                        1,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_57_ops};
+
+static ec_gf_op_t ec_gf8_mul_58_ops[] = {
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_58 = {8,
+                                    {
+                                        4,
+                                        3,
+                                        2,
+                                        7,
+                                        0,
+                                        1,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_58_ops};
+
+static ec_gf_op_t ec_gf8_mul_59_ops[] = {
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_59 = {8,
+                                    {
+                                        7,
+                                        3,
+                                        5,
+                                        6,
+                                        1,
+                                        2,
+                                        0,
+                                        4,
+                                    },
+                                    ec_gf8_mul_59_ops};
+
+static ec_gf_op_t ec_gf8_mul_5A_ops[] = {
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_5A = {8,
+                                    {
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        5,
+                                        4,
+                                    },
+                                    ec_gf8_mul_5A_ops};
+
+static ec_gf_op_t ec_gf8_mul_5B_ops[] = {
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_5B = {8,
+                                    {
+                                        6,
+                                        0,
+                                        7,
+                                        5,
+                                        2,
+                                        1,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_5B_ops};
+
+static ec_gf_op_t ec_gf8_mul_5C_ops[] = {
+    {EC_GF_OP_COPY, 8, 3, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 2, 8, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_5C = {9,
+                                    {
+                                        7,
+                                        5,
+                                        2,
+                                        4,
+                                        1,
+                                        0,
+                                        6,
+                                        3,
+                                        8,
+                                    },
+                                    ec_gf8_mul_5C_ops};
+
+static ec_gf_op_t ec_gf8_mul_5D_ops[] = {
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_5D = {8,
+                                    {
+                                        1,
+                                        3,
+                                        5,
+                                        4,
+                                        6,
+                                        7,
+                                        2,
+                                        0,
+                                    },
+                                    ec_gf8_mul_5D_ops};
+
+static ec_gf_op_t ec_gf8_mul_5E_ops[] = {
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_5E = {8,
+                                    {
+                                        4,
+                                        3,
+                                        6,
+                                        2,
+                                        5,
+                                        7,
+                                        0,
+                                        1,
+                                    },
+                                    ec_gf8_mul_5E_ops};
+
+static ec_gf_op_t ec_gf8_mul_5F_ops[] = {
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_5F = {8,
+                                    {
+                                        6,
+                                        1,
+                                        3,
+                                        4,
+                                        5,
+                                        7,
+                                        2,
+                                        0,
+                                    },
+                                    ec_gf8_mul_5F_ops};
+
+static ec_gf_op_t ec_gf8_mul_60_ops[] = {
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_60 = {8,
+                                    {
+                                        2,
+                                        3,
+                                        4,
+                                        7,
+                                        5,
+                                        6,
+                                        0,
+                                        1,
+                                    },
+                                    ec_gf8_mul_60_ops};
+
+static ec_gf_op_t ec_gf8_mul_61_ops[] = {
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_61 = {8,
+                                    {
+                                        0,
+                                        5,
+                                        6,
+                                        7,
+                                        4,
+                                        2,
+                                        1,
+                                        3,
+                                    },
+                                    ec_gf8_mul_61_ops};
+
+static ec_gf_op_t ec_gf8_mul_62_ops[] = {
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_62 = {8,
+                                    {
+                                        2,
+                                        0,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        1,
+                                    },
+                                    ec_gf8_mul_62_ops};
+
+static ec_gf_op_t ec_gf8_mul_63_ops[] = {
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_63 = {8,
+                                    {
+                                        3,
+                                        4,
+                                        6,
+                                        5,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_63_ops};
+
+static ec_gf_op_t ec_gf8_mul_64_ops[] = {
+    {EC_GF_OP_COPY, 8, 1, 0}, {EC_GF_OP_XOR2, 8, 0, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 8, 7, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_64 = {9,
+                                    {
+                                        2,
+                                        3,
+                                        4,
+                                        6,
+                                        5,
+                                        7,
+                                        8,
+                                        1,
+                                        0,
+                                    },
+                                    ec_gf8_mul_64_ops};
+
+static ec_gf_op_t ec_gf8_mul_65_ops[] = {
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_65 = {8,
+                                    {
+                                        2,
+                                        5,
+                                        1,
+                                        3,
+                                        4,
+                                        0,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_65_ops};
+
+static ec_gf_op_t ec_gf8_mul_66_ops[] = {
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_66 = {8,
+                                    {
+                                        2,
+                                        3,
+                                        1,
+                                        4,
+                                        5,
+                                        7,
+                                        0,
+                                        6,
+                                    },
+                                    ec_gf8_mul_66_ops};
+
+static ec_gf_op_t ec_gf8_mul_67_ops[] = {
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_67 = {8,
+                                    {
+                                        2,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        3,
+                                        1,
+                                        0,
+                                    },
+                                    ec_gf8_mul_67_ops};
+
+static ec_gf_op_t ec_gf8_mul_68_ops[] = {
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_68 = {8,
+                                    {
+                                        5,
+                                        7,
+                                        2,
+                                        3,
+                                        0,
+                                        6,
+                                        4,
+                                        1,
+                                    },
+                                    ec_gf8_mul_68_ops};
+
+static ec_gf_op_t ec_gf8_mul_69_ops[] = {
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_69 = {8,
+                                    {
+                                        0,
+                                        1,
+                                        3,
+                                        2,
+                                        4,
+                                        5,
+                                        7,
+                                        6,
+                                    },
+                                    ec_gf8_mul_69_ops};
+
+static ec_gf_op_t ec_gf8_mul_6A_ops[] = {
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_6A = {8,
+                                    {
+                                        5,
+                                        7,
+                                        4,
+                                        6,
+                                        1,
+                                        2,
+                                        0,
+                                        3,
+                                    },
+                                    ec_gf8_mul_6A_ops};
+
+static ec_gf_op_t ec_gf8_mul_6B_ops[] = {
+    {EC_GF_OP_COPY, 8, 1, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 0, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_6B = {9,
+                                    {
+                                        6,
+                                        7,
+                                        2,
+                                        0,
+                                        3,
+                                        1,
+                                        5,
+                                        4,
+                                        8,
+                                    },
+                                    ec_gf8_mul_6B_ops};
+
+static ec_gf_op_t ec_gf8_mul_6C_ops[] = {
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_6C = {8,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_6C_ops};
+
+static ec_gf_op_t ec_gf8_mul_6D_ops[] = {
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR3, 8, 3, 4}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 0, 8, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 6, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_6D = {9,
+                                    {
+                                        3,
+                                        6,
+                                        7,
+                                        0,
+                                        4,
+                                        5,
+                                        1,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_6D_ops};
+
+static ec_gf_op_t ec_gf8_mul_6E_ops[] = {
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_6E = {8,
+                                    {
+                                        5,
+                                        6,
+                                        3,
+                                        1,
+                                        7,
+                                        2,
+                                        0,
+                                        4,
+                                    },
+                                    ec_gf8_mul_6E_ops};
+
+static ec_gf_op_t ec_gf8_mul_6F_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR3, 0, 8, 7}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_6F = {9,
+                                    {
+                                        2,
+                                        6,
+                                        3,
+                                        7,
+                                        0,
+                                        1,
+                                        4,
+                                        5,
+                                        8,
+                                    },
+                                    ec_gf8_mul_6F_ops};
+
+static ec_gf_op_t ec_gf8_mul_70_ops[] = {
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_70 = {8,
+                                    {
+                                        3,
+                                        4,
+                                        5,
+                                        2,
+                                        6,
+                                        0,
+                                        1,
+                                        7,
+                                    },
+                                    ec_gf8_mul_70_ops};
+
+static ec_gf_op_t ec_gf8_mul_71_ops[] = {
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_71 = {8,
+                                    {
+                                        4,
+                                        7,
+                                        5,
+                                        3,
+                                        6,
+                                        0,
+                                        2,
+                                        1,
+                                    },
+                                    ec_gf8_mul_71_ops};
+
+static ec_gf_op_t ec_gf8_mul_72_ops[] = {
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_72 = {8,
+                                    {
+                                        0,
+                                        5,
+                                        2,
+                                        7,
+                                        4,
+                                        1,
+                                        3,
+                                        6,
+                                    },
+                                    ec_gf8_mul_72_ops};
+
+static ec_gf_op_t ec_gf8_mul_73_ops[] = {
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_73 = {8,
+                                    {
+                                        6,
+                                        0,
+                                        1,
+                                        7,
+                                        4,
+                                        5,
+                                        2,
+                                        3,
+                                    },
+                                    ec_gf8_mul_73_ops};
+
+static ec_gf_op_t ec_gf8_mul_74_ops[] = {
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_74 = {8,
+                                    {
+                                        3,
+                                        2,
+                                        1,
+                                        0,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_74_ops};
+
+static ec_gf_op_t ec_gf8_mul_75_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_75 = {8,
+                                    {
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                    },
+                                    ec_gf8_mul_75_ops};
+
+static ec_gf_op_t ec_gf8_mul_76_ops[] = {
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR3, 8, 6, 2},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 0, 8, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_76 = {9,
+                                    {
+                                        2,
+                                        3,
+                                        0,
+                                        6,
+                                        5,
+                                        1,
+                                        7,
+                                        8,
+                                        4,
+                                    },
+                                    ec_gf8_mul_76_ops};
+
+static ec_gf_op_t ec_gf8_mul_77_ops[] = {
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_77 = {8,
+                                    {
+                                        7,
+                                        4,
+                                        3,
+                                        6,
+                                        0,
+                                        1,
+                                        5,
+                                        2,
+                                    },
+                                    ec_gf8_mul_77_ops};
+
+static ec_gf_op_t ec_gf8_mul_78_ops[] = {
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR3, 8, 0, 2},
+    {EC_GF_OP_XOR2, 4, 8, 0}, {EC_GF_OP_XOR2, 1, 8, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_78 = {9,
+                                    {
+                                        4,
+                                        7,
+                                        3,
+                                        2,
+                                        5,
+                                        1,
+                                        6,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_78_ops};
+
+static ec_gf_op_t ec_gf8_mul_79_ops[] = {
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR3, 8, 4, 7},
+    {EC_GF_OP_XOR2, 0, 8, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 6, 8, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_79 = {9,
+                                    {
+                                        4,
+                                        5,
+                                        7,
+                                        3,
+                                        1,
+                                        6,
+                                        2,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_79_ops};
+
+static ec_gf_op_t ec_gf8_mul_7A_ops[] = {
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_7A = {8,
+                                    {
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                    },
+                                    ec_gf8_mul_7A_ops};
+
+static ec_gf_op_t ec_gf8_mul_7B_ops[] = {
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR3, 8, 5, 3},
+    {EC_GF_OP_XOR2, 8, 0, 0}, {EC_GF_OP_COPY, 9, 4, 0},
+    {EC_GF_OP_XOR2, 8, 2, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 4, 8, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR3, 4, 1, 9},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_7B = {10,
+                                    {
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        8,
+                                        5,
+                                        6,
+                                        0,
+                                        7,
+                                        9,
+                                    },
+                                    ec_gf8_mul_7B_ops};
+
+static ec_gf_op_t ec_gf8_mul_7C_ops[] = {
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_7C = {8,
+                                    {
+                                        2,
+                                        4,
+                                        1,
+                                        6,
+                                        3,
+                                        5,
+                                        7,
+                                        0,
+                                    },
+                                    ec_gf8_mul_7C_ops};
+
+static ec_gf_op_t ec_gf8_mul_7D_ops[] = {
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_7D = {8,
+                                    {
+                                        1,
+                                        0,
+                                        3,
+                                        5,
+                                        6,
+                                        7,
+                                        2,
+                                        4,
+                                    },
+                                    ec_gf8_mul_7D_ops};
+
+static ec_gf_op_t ec_gf8_mul_7E_ops[] = {
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_COPY, 8, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR3, 6, 2, 7},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 6, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_7E = {9,
+                                    {
+                                        5,
+                                        1,
+                                        2,
+                                        0,
+                                        7,
+                                        3,
+                                        4,
+                                        6,
+                                        8,
+                                    },
+                                    ec_gf8_mul_7E_ops};
+
+static ec_gf_op_t ec_gf8_mul_7F_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR3, 9, 7, 5}, {EC_GF_OP_XOR2, 2, 9, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 6, 9, 0},
+    {EC_GF_OP_XOR3, 9, 6, 4}, {EC_GF_OP_XOR2, 7, 9, 0},
+    {EC_GF_OP_XOR2, 3, 9, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 7, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_7F = {10,
+                                    {
+                                        4,
+                                        1,
+                                        0,
+                                        5,
+                                        6,
+                                        7,
+                                        2,
+                                        3,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_7F_ops};
+
+static ec_gf_op_t ec_gf8_mul_80_ops[] = {
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_80 = {8,
+                                    {
+                                        7,
+                                        5,
+                                        6,
+                                        4,
+                                        1,
+                                        2,
+                                        3,
+                                        0,
+                                    },
+                                    ec_gf8_mul_80_ops};
+
+static ec_gf_op_t ec_gf8_mul_81_ops[] = {
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_81 = {8,
+                                    {
+                                        2,
+                                        7,
+                                        4,
+                                        1,
+                                        5,
+                                        6,
+                                        3,
+                                        0,
+                                    },
+                                    ec_gf8_mul_81_ops};
+
+static ec_gf_op_t ec_gf8_mul_82_ops[] = {
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_COPY, 8, 6, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR3, 5, 8, 7}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_82 = {9,
+                                    {
+                                        6,
+                                        2,
+                                        7,
+                                        5,
+                                        1,
+                                        3,
+                                        4,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_82_ops};
+
+static ec_gf_op_t ec_gf8_mul_83_ops[] = {
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_83 = {8,
+                                    {
+                                        3,
+                                        5,
+                                        6,
+                                        7,
+                                        1,
+                                        2,
+                                        4,
+                                        0,
+                                    },
+                                    ec_gf8_mul_83_ops};
+
+static ec_gf_op_t ec_gf8_mul_84_ops[] = {
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_84 = {8,
+                                    {
+                                        7,
+                                        6,
+                                        0,
+                                        4,
+                                        1,
+                                        5,
+                                        3,
+                                        2,
+                                    },
+                                    ec_gf8_mul_84_ops};
+
+static ec_gf_op_t ec_gf8_mul_85_ops[] = {
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_85 = {8,
+                                    {
+                                        7,
+                                        6,
+                                        0,
+                                        3,
+                                        2,
+                                        4,
+                                        5,
+                                        1,
+                                    },
+                                    ec_gf8_mul_85_ops};
+
+static ec_gf_op_t ec_gf8_mul_86_ops[] = {
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_86 = {8,
+                                    {
+                                        1,
+                                        2,
+                                        6,
+                                        4,
+                                        5,
+                                        7,
+                                        3,
+                                        0,
+                                    },
+                                    ec_gf8_mul_86_ops};
+
+static ec_gf_op_t ec_gf8_mul_87_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_COPY, 8, 1, 0},
+    {EC_GF_OP_XOR2, 8, 6, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR3, 5, 8, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 2, 8, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_87 = {9,
+                                    {
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        7,
+                                        6,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_87_ops};
+
+static ec_gf_op_t ec_gf8_mul_88_ops[] = {
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_88 = {8,
+                                    {
+                                        6,
+                                        7,
+                                        3,
+                                        1,
+                                        2,
+                                        4,
+                                        5,
+                                        0,
+                                    },
+                                    ec_gf8_mul_88_ops};
+
+static ec_gf_op_t ec_gf8_mul_89_ops[] = {
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR3, 8, 5, 2},
+    {EC_GF_OP_XOR2, 4, 8, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 0, 8, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_89 = {9,
+                                    {
+                                        2,
+                                        1,
+                                        6,
+                                        5,
+                                        7,
+                                        3,
+                                        4,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_89_ops};
+
+static ec_gf_op_t ec_gf8_mul_8A_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_8A = {8,
+                                    {
+                                        1,
+                                        2,
+                                        3,
+                                        0,
+                                        6,
+                                        7,
+                                        4,
+                                        5,
+                                    },
+                                    ec_gf8_mul_8A_ops};
+
+static ec_gf_op_t ec_gf8_mul_8B_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_8B = {8,
+                                    {
+                                        6,
+                                        1,
+                                        2,
+                                        3,
+                                        5,
+                                        7,
+                                        4,
+                                        0,
+                                    },
+                                    ec_gf8_mul_8B_ops};
+
+static ec_gf_op_t ec_gf8_mul_8C_ops[] = {
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_8C = {8,
+                                    {
+                                        1,
+                                        2,
+                                        0,
+                                        7,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_8C_ops};
+
+static ec_gf_op_t ec_gf8_mul_8D_ops[] = {
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_8D = {8,
+                                    {
+                                        7,
+                                        1,
+                                        3,
+                                        2,
+                                        4,
+                                        5,
+                                        0,
+                                        6,
+                                    },
+                                    ec_gf8_mul_8D_ops};
+
+static ec_gf_op_t ec_gf8_mul_8E_ops[] = {{EC_GF_OP_XOR2, 2, 0, 0},
+                                         {EC_GF_OP_XOR2, 3, 0, 0},
+                                         {EC_GF_OP_XOR2, 4, 0, 0},
+                                         {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_8E = {8,
+                                    {
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                    },
+                                    ec_gf8_mul_8E_ops};
+
+static ec_gf_op_t ec_gf8_mul_8F_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_8F = {8,
+                                    {
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                    },
+                                    ec_gf8_mul_8F_ops};
+
+static ec_gf_op_t ec_gf8_mul_90_ops[] = {
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_90 = {8,
+                                    {
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        3,
+                                        2,
+                                    },
+                                    ec_gf8_mul_90_ops};
+
+static ec_gf_op_t ec_gf8_mul_91_ops[] = {
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_COPY, 9, 1, 0}, {EC_GF_OP_COPY, 8, 3, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 7, 9, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR3, 5, 8, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_91 = {10,
+                                    {
+                                        2,
+                                        3,
+                                        1,
+                                        4,
+                                        0,
+                                        6,
+                                        7,
+                                        5,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_91_ops};
+
+static ec_gf_op_t ec_gf8_mul_92_ops[] = {
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_92 = {8,
+                                    {
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        5,
+                                        4,
+                                    },
+                                    ec_gf8_mul_92_ops};
+
+static ec_gf_op_t ec_gf8_mul_93_ops[] = {
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_93 = {8,
+                                    {
+                                        6,
+                                        4,
+                                        5,
+                                        1,
+                                        7,
+                                        2,
+                                        3,
+                                        0,
+                                    },
+                                    ec_gf8_mul_93_ops};
+
+static ec_gf_op_t ec_gf8_mul_94_ops[] = {
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_94 = {8,
+                                    {
+                                        7,
+                                        5,
+                                        0,
+                                        2,
+                                        6,
+                                        1,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_94_ops};
+
+static ec_gf_op_t ec_gf8_mul_95_ops[] = {
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_95 = {8,
+                                    {
+                                        7,
+                                        6,
+                                        1,
+                                        3,
+                                        0,
+                                        4,
+                                        5,
+                                        2,
+                                    },
+                                    ec_gf8_mul_95_ops};
+
+static ec_gf_op_t ec_gf8_mul_96_ops[] = {
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR3, 8, 0, 4}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 7, 8, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 8, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 1, 8, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 5, 8, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_96 = {9,
+                                    {
+                                        4,
+                                        0,
+                                        1,
+                                        6,
+                                        7,
+                                        2,
+                                        3,
+                                        5,
+                                        8,
+                                    },
+                                    ec_gf8_mul_96_ops};
+
+static ec_gf_op_t ec_gf8_mul_97_ops[] = {
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_COPY, 8, 2, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 8, 6, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 1, 8, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 5, 8, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_97 = {9,
+                                    {
+                                        4,
+                                        5,
+                                        3,
+                                        6,
+                                        7,
+                                        1,
+                                        2,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_97_ops};
+
+static ec_gf_op_t ec_gf8_mul_98_ops[] = {
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_98 = {8,
+                                    {
+                                        4,
+                                        2,
+                                        3,
+                                        6,
+                                        7,
+                                        5,
+                                        1,
+                                        0,
+                                    },
+                                    ec_gf8_mul_98_ops};
+
+static ec_gf_op_t ec_gf8_mul_99_ops[] = {
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_99 = {8,
+                                    {
+                                        6,
+                                        5,
+                                        3,
+                                        7,
+                                        0,
+                                        1,
+                                        4,
+                                        2,
+                                    },
+                                    ec_gf8_mul_99_ops};
+
+static ec_gf_op_t ec_gf8_mul_9A_ops[] = {
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR3, 8, 4, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 7, 8, 0}, {EC_GF_OP_XOR2, 1, 8, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_9A = {9,
+                                    {
+                                        6,
+                                        3,
+                                        4,
+                                        0,
+                                        5,
+                                        1,
+                                        2,
+                                        7,
+                                        8,
+                                    },
+                                    ec_gf8_mul_9A_ops};
+
+static ec_gf_op_t ec_gf8_mul_9B_ops[] = {
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_COPY, 9, 5, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR3, 8, 3, 2}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 3, 9, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_9B = {10,
+                                    {
+                                        4,
+                                        5,
+                                        8,
+                                        6,
+                                        7,
+                                        1,
+                                        2,
+                                        0,
+                                        3,
+                                        9,
+                                    },
+                                    ec_gf8_mul_9B_ops};
+
+static ec_gf_op_t ec_gf8_mul_9C_ops[] = {
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_9C = {8,
+                                    {
+                                        3,
+                                        2,
+                                        1,
+                                        0,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_9C_ops};
+
+static ec_gf_op_t ec_gf8_mul_9D_ops[] = {
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_9D = {8,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        7,
+                                        4,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_9D_ops};
+
+static ec_gf_op_t ec_gf8_mul_9E_ops[] = {
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_COPY, 8, 7, 0},
+    {EC_GF_OP_XOR2, 8, 5, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 0, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_9E = {9,
+                                    {
+                                        4,
+                                        5,
+                                        3,
+                                        8,
+                                        6,
+                                        0,
+                                        2,
+                                        7,
+                                        1,
+                                    },
+                                    ec_gf8_mul_9E_ops};
+
+static ec_gf_op_t ec_gf8_mul_9F_ops[] = {
+    {EC_GF_OP_XOR3, 8, 1, 2}, {EC_GF_OP_XOR2, 8, 3, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 6, 8, 0}, {EC_GF_OP_XOR2, 5, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_9F = {9,
+                                    {
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        8,
+                                    },
+                                    ec_gf8_mul_9F_ops};
+
+static ec_gf_op_t ec_gf8_mul_A0_ops[] = {
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_A0 = {8,
+                                    {
+                                        3,
+                                        1,
+                                        6,
+                                        7,
+                                        5,
+                                        2,
+                                        4,
+                                        0,
+                                    },
+                                    ec_gf8_mul_A0_ops};
+
+static ec_gf_op_t ec_gf8_mul_A1_ops[] = {
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR3, 8, 0, 6},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 7, 8, 0},
+    {EC_GF_OP_XOR2, 3, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_A1 = {9,
+                                    {
+                                        7,
+                                        4,
+                                        1,
+                                        5,
+                                        6,
+                                        0,
+                                        2,
+                                        3,
+                                        8,
+                                    },
+                                    ec_gf8_mul_A1_ops};
+
+static ec_gf_op_t ec_gf8_mul_A2_ops[] = {
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_A2 = {8,
+                                    {
+                                        7,
+                                        0,
+                                        6,
+                                        3,
+                                        2,
+                                        1,
+                                        4,
+                                        5,
+                                    },
+                                    ec_gf8_mul_A2_ops};
+
+static ec_gf_op_t ec_gf8_mul_A3_ops[] = {
+    {EC_GF_OP_COPY, 8, 2, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 3, 8, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_A3 = {9,
+                                    {
+                                        3,
+                                        7,
+                                        2,
+                                        6,
+                                        1,
+                                        4,
+                                        0,
+                                        5,
+                                        8,
+                                    },
+                                    ec_gf8_mul_A3_ops};
+
+static ec_gf_op_t ec_gf8_mul_A4_ops[] = {
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_A4 = {8,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        2,
+                                        4,
+                                        3,
+                                        0,
+                                        1,
+                                    },
+                                    ec_gf8_mul_A4_ops};
+
+static ec_gf_op_t ec_gf8_mul_A5_ops[] = {
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR3, 8, 5, 6}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 4, 8, 0}, {EC_GF_OP_XOR2, 7, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_A5 = {9,
+                                    {
+                                        1,
+                                        4,
+                                        2,
+                                        5,
+                                        6,
+                                        7,
+                                        3,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_A5_ops};
+
+static ec_gf_op_t ec_gf8_mul_A6_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_A6 = {8,
+                                    {
+                                        1,
+                                        2,
+                                        0,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_A6_ops};
+
+static ec_gf_op_t ec_gf8_mul_A7_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_A7 = {8,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        5,
+                                        6,
+                                        7,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_A7_ops};
+
+static ec_gf_op_t ec_gf8_mul_A8_ops[] = {
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 8, 1, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_COPY, 9, 4, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 8, 3, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 2, 9, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_A8 = {10,
+                                    {
+                                        1,
+                                        7,
+                                        5,
+                                        8,
+                                        6,
+                                        3,
+                                        4,
+                                        0,
+                                        2,
+                                        9,
+                                    },
+                                    ec_gf8_mul_A8_ops};
+
+static ec_gf_op_t ec_gf8_mul_A9_ops[] = {
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_A9 = {8,
+                                    {
+                                        3,
+                                        7,
+                                        6,
+                                        1,
+                                        2,
+                                        0,
+                                        4,
+                                        5,
+                                    },
+                                    ec_gf8_mul_A9_ops};
+
+static ec_gf_op_t ec_gf8_mul_AA_ops[] = {
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_AA = {8,
+                                    {
+                                        0,
+                                        4,
+                                        5,
+                                        3,
+                                        6,
+                                        7,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_AA_ops};
+
+static ec_gf_op_t ec_gf8_mul_AB_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_COPY, 9, 6, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 8, 7, 0}, {EC_GF_OP_XOR2, 3, 8, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR3, 3, 9, 7},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_AB = {10,
+                                    {
+                                        2,
+                                        3,
+                                        8,
+                                        0,
+                                        5,
+                                        6,
+                                        1,
+                                        4,
+                                        7,
+                                        9,
+                                    },
+                                    ec_gf8_mul_AB_ops};
+
+static ec_gf_op_t ec_gf8_mul_AC_ops[] = {
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_AC = {8,
+                                    {
+                                        3,
+                                        2,
+                                        1,
+                                        0,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_AC_ops};
+
+static ec_gf_op_t ec_gf8_mul_AD_ops[] = {
+    {EC_GF_OP_XOR3, 8, 1, 2}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 4, 8, 0},
+    {EC_GF_OP_XOR2, 5, 8, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_AD = {9,
+                                    {
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_AD_ops};
+
+static ec_gf_op_t ec_gf8_mul_AE_ops[] = {
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_COPY, 8, 5, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 4, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_AE = {9,
+                                    {
+                                        7,
+                                        0,
+                                        5,
+                                        6,
+                                        3,
+                                        4,
+                                        1,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_AE_ops};
+
+static ec_gf_op_t ec_gf8_mul_AF_ops[] = {
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_AF = {8,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        7,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_AF_ops};
+
+static ec_gf_op_t ec_gf8_mul_B0_ops[] = {
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_B0 = {8,
+                                    {
+                                        4,
+                                        0,
+                                        7,
+                                        2,
+                                        3,
+                                        1,
+                                        6,
+                                        5,
+                                    },
+                                    ec_gf8_mul_B0_ops};
+
+static ec_gf_op_t ec_gf8_mul_B1_ops[] = {
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_COPY, 8, 4, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR3, 5, 8, 1}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_B1 = {9,
+                                    {
+                                        2,
+                                        6,
+                                        4,
+                                        7,
+                                        0,
+                                        1,
+                                        3,
+                                        5,
+                                        8,
+                                    },
+                                    ec_gf8_mul_B1_ops};
+
+static ec_gf_op_t ec_gf8_mul_B2_ops[] = {
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR3, 8, 4, 5},
+    {EC_GF_OP_XOR2, 2, 8, 0}, {EC_GF_OP_XOR2, 8, 1, 0},
+    {EC_GF_OP_XOR2, 7, 8, 0}, {EC_GF_OP_XOR2, 3, 8, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_B2 = {9,
+                                    {
+                                        0,
+                                        7,
+                                        4,
+                                        5,
+                                        6,
+                                        1,
+                                        2,
+                                        3,
+                                        8,
+                                    },
+                                    ec_gf8_mul_B2_ops};
+
+static ec_gf_op_t ec_gf8_mul_B3_ops[] = {
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_COPY, 9, 5, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR3, 8, 6, 4},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 8, 5, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 7, 8, 0},
+    {EC_GF_OP_XOR2, 0, 8, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR3, 1, 9, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_B3 = {10,
+                                    {
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        1,
+                                        6,
+                                        0,
+                                        7,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_B3_ops};
+
+static ec_gf_op_t ec_gf8_mul_B4_ops[] = {
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_B4 = {8,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_B4_ops};
+
+static ec_gf_op_t ec_gf8_mul_B5_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_COPY, 8, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR3, 4, 8, 3}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_B5 = {9,
+                                    {
+                                        3,
+                                        4,
+                                        0,
+                                        7,
+                                        1,
+                                        5,
+                                        6,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_B5_ops};
+
+static ec_gf_op_t ec_gf8_mul_B6_ops[] = {
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_B6 = {8,
+                                    {
+                                        5,
+                                        3,
+                                        6,
+                                        4,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_B6_ops};
+
+static ec_gf_op_t ec_gf8_mul_B7_ops[] = {
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_B7 = {8,
+                                    {
+                                        5,
+                                        0,
+                                        1,
+                                        4,
+                                        2,
+                                        6,
+                                        7,
+                                        3,
+                                    },
+                                    ec_gf8_mul_B7_ops};
+
+static ec_gf_op_t ec_gf8_mul_B8_ops[] = {
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_B8 = {8,
+                                    {
+                                        6,
+                                        4,
+                                        5,
+                                        1,
+                                        2,
+                                        0,
+                                        7,
+                                        3,
+                                    },
+                                    ec_gf8_mul_B8_ops};
+
+static ec_gf_op_t ec_gf8_mul_B9_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR3, 0, 8, 2}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_B9 = {9,
+                                    {
+                                        6,
+                                        7,
+                                        0,
+                                        2,
+                                        1,
+                                        4,
+                                        5,
+                                        3,
+                                        8,
+                                    },
+                                    ec_gf8_mul_B9_ops};
+
+static ec_gf_op_t ec_gf8_mul_BA_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_BA = {8,
+                                    {
+                                        1,
+                                        2,
+                                        4,
+                                        3,
+                                        5,
+                                        6,
+                                        0,
+                                        7,
+                                    },
+                                    ec_gf8_mul_BA_ops};
+
+static ec_gf_op_t ec_gf8_mul_BB_ops[] = {
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_COPY, 8, 3, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 8, 5, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 8, 7, 0}, {EC_GF_OP_XOR2, 2, 8, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_BB = {9,
+                                    {
+                                        7,
+                                        2,
+                                        1,
+                                        8,
+                                        3,
+                                        5,
+                                        6,
+                                        4,
+                                        0,
+                                    },
+                                    ec_gf8_mul_BB_ops};
+
+static ec_gf_op_t ec_gf8_mul_BC_ops[] = {
+    {EC_GF_OP_COPY, 8, 1, 0}, {EC_GF_OP_XOR2, 8, 2, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 7, 8, 0}, {EC_GF_OP_XOR3, 2, 8, 4},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_BC = {9,
+                                    {
+                                        2,
+                                        6,
+                                        3,
+                                        4,
+                                        5,
+                                        1,
+                                        7,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_BC_ops};
+
+static ec_gf_op_t ec_gf8_mul_BD_ops[] = {
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_BD = {8,
+                                    {
+                                        4,
+                                        5,
+                                        0,
+                                        2,
+                                        7,
+                                        1,
+                                        6,
+                                        3,
+                                    },
+                                    ec_gf8_mul_BD_ops};
+
+static ec_gf_op_t ec_gf8_mul_BE_ops[] = {
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_BE = {8,
+                                    {
+                                        0,
+                                        6,
+                                        7,
+                                        4,
+                                        5,
+                                        1,
+                                        3,
+                                        2,
+                                    },
+                                    ec_gf8_mul_BE_ops};
+
+static ec_gf_op_t ec_gf8_mul_BF_ops[] = {
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_BF = {8,
+                                    {
+                                        5,
+                                        6,
+                                        1,
+                                        7,
+                                        3,
+                                        0,
+                                        2,
+                                        4,
+                                    },
+                                    ec_gf8_mul_BF_ops};
+
+static ec_gf_op_t ec_gf8_mul_C0_ops[] = {
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_C0 = {8,
+                                    {
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        7,
+                                        5,
+                                        6,
+                                        0,
+                                    },
+                                    ec_gf8_mul_C0_ops};
+
+static ec_gf_op_t ec_gf8_mul_C1_ops[] = {
+    {EC_GF_OP_XOR3, 8, 1, 2}, {EC_GF_OP_XOR2, 8, 3, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 6, 8, 0}, {EC_GF_OP_XOR2, 5, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_C1 = {9,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        4,
+                                        1,
+                                        2,
+                                        3,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_C1_ops};
+
+static ec_gf_op_t ec_gf8_mul_C2_ops[] = {
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_C2 = {8,
+                                    {
+                                        7,
+                                        6,
+                                        3,
+                                        0,
+                                        1,
+                                        4,
+                                        5,
+                                        2,
+                                    },
+                                    ec_gf8_mul_C2_ops};
+
+static ec_gf_op_t ec_gf8_mul_C3_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR3, 0, 2, 6}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR3, 9, 1, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 7, 9, 0},
+    {EC_GF_OP_XOR2, 3, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_C3 = {10,
+                                    {
+                                        5,
+                                        6,
+                                        4,
+                                        7,
+                                        1,
+                                        2,
+                                        3,
+                                        0,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_C3_ops};
+
+static ec_gf_op_t ec_gf8_mul_C4_ops[] = {
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 1, 0, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_C4 = {8,
+                                    {
+                                        0,
+                                        2,
+                                        1,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_C4_ops};
+
+static ec_gf_op_t ec_gf8_mul_C5_ops[] = {
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_C5 = {8,
+                                    {
+                                        4,
+                                        3,
+                                        5,
+                                        7,
+                                        6,
+                                        2,
+                                        0,
+                                        1,
+                                    },
+                                    ec_gf8_mul_C5_ops};
+
+static ec_gf_op_t ec_gf8_mul_C6_ops[] = {
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_COPY, 8, 4, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR3, 9, 5, 4},
+    {EC_GF_OP_XOR2, 6, 9, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 7, 9, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 6, 8, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_C6 = {10,
+                                    {
+                                        6,
+                                        3,
+                                        0,
+                                        4,
+                                        5,
+                                        7,
+                                        2,
+                                        1,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_C6_ops};
+
+static ec_gf_op_t ec_gf8_mul_C7_ops[] = {
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_C7 = {8,
+                                    {
+                                        7,
+                                        0,
+                                        6,
+                                        2,
+                                        5,
+                                        3,
+                                        4,
+                                        1,
+                                    },
+                                    ec_gf8_mul_C7_ops};
+
+static ec_gf_op_t ec_gf8_mul_C8_ops[] = {
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_C8 = {8,
+                                    {
+                                        1,
+                                        3,
+                                        2,
+                                        4,
+                                        6,
+                                        7,
+                                        5,
+                                        0,
+                                    },
+                                    ec_gf8_mul_C8_ops};
+
+static ec_gf_op_t ec_gf8_mul_C9_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_C9 = {8,
+                                    {
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                    },
+                                    ec_gf8_mul_C9_ops};
+
+static ec_gf_op_t ec_gf8_mul_CA_ops[] = {
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 4, 5, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_CA = {8,
+                                    {
+                                        1,
+                                        2,
+                                        5,
+                                        7,
+                                        3,
+                                        4,
+                                        0,
+                                        6,
+                                    },
+                                    ec_gf8_mul_CA_ops};
+
+static ec_gf_op_t ec_gf8_mul_CB_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_CB = {8,
+                                    {
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        7,
+                                        6,
+                                        0,
+                                        1,
+                                    },
+                                    ec_gf8_mul_CB_ops};
+
+static ec_gf_op_t ec_gf8_mul_CC_ops[] = {
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_CC = {8,
+                                    {
+                                        2,
+                                        7,
+                                        1,
+                                        0,
+                                        5,
+                                        6,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_CC_ops};
+
+static ec_gf_op_t ec_gf8_mul_CD_ops[] = {
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_CD = {8,
+                                    {
+                                        0,
+                                        6,
+                                        1,
+                                        2,
+                                        7,
+                                        3,
+                                        4,
+                                        5,
+                                    },
+                                    ec_gf8_mul_CD_ops};
+
+static ec_gf_op_t ec_gf8_mul_CE_ops[] = {
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_COPY, 8, 7, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR3, 3, 6, 8},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR3, 8, 2, 3},
+    {EC_GF_OP_XOR2, 1, 8, 0}, {EC_GF_OP_XOR2, 4, 8, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_CE = {9,
+                                    {
+                                        5,
+                                        7,
+                                        3,
+                                        0,
+                                        2,
+                                        6,
+                                        4,
+                                        1,
+                                        8,
+                                    },
+                                    ec_gf8_mul_CE_ops};
+
+static ec_gf_op_t ec_gf8_mul_CF_ops[] = {
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_CF = {8,
+                                    {
+                                        3,
+                                        6,
+                                        7,
+                                        0,
+                                        2,
+                                        4,
+                                        5,
+                                        1,
+                                    },
+                                    ec_gf8_mul_CF_ops};
+
+static ec_gf_op_t ec_gf8_mul_D0_ops[] = {
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_D0 = {8,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        2,
+                                        0,
+                                        3,
+                                        1,
+                                        4,
+                                    },
+                                    ec_gf8_mul_D0_ops};
+
+static ec_gf_op_t ec_gf8_mul_D1_ops[] = {
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR3, 8, 6, 0},
+    {EC_GF_OP_XOR2, 4, 8, 0}, {EC_GF_OP_XOR2, 1, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_D1 = {9,
+                                    {
+                                        5,
+                                        6,
+                                        3,
+                                        2,
+                                        0,
+                                        7,
+                                        4,
+                                        1,
+                                        8,
+                                    },
+                                    ec_gf8_mul_D1_ops};
+
+static ec_gf_op_t ec_gf8_mul_D2_ops[] = {
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_D2 = {8,
+                                    {
+                                        7,
+                                        0,
+                                        2,
+                                        1,
+                                        3,
+                                        4,
+                                        6,
+                                        5,
+                                    },
+                                    ec_gf8_mul_D2_ops};
+
+static ec_gf_op_t ec_gf8_mul_D3_ops[] = {
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_COPY, 8, 4, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 8, 6, 0}, {EC_GF_OP_XOR2, 3, 8, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 1, 3, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_D3 = {9,
+                                    {
+                                        0,
+                                        3,
+                                        2,
+                                        8,
+                                        4,
+                                        6,
+                                        7,
+                                        1,
+                                        5,
+                                    },
+                                    ec_gf8_mul_D3_ops};
+
+static ec_gf_op_t ec_gf8_mul_D4_ops[] = {
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_COPY, 8, 1, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR3, 1, 7, 8},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_D4 = {9,
+                                    {
+                                        4,
+                                        1,
+                                        7,
+                                        5,
+                                        0,
+                                        6,
+                                        3,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_D4_ops};
+
+static ec_gf_op_t ec_gf8_mul_D5_ops[] = {
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_D5 = {8,
+                                    {
+                                        6,
+                                        7,
+                                        4,
+                                        5,
+                                        2,
+                                        3,
+                                        1,
+                                        0,
+                                    },
+                                    ec_gf8_mul_D5_ops};
+
+static ec_gf_op_t ec_gf8_mul_D6_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_D6 = {9,
+                                    {
+                                        0,
+                                        6,
+                                        2,
+                                        7,
+                                        1,
+                                        3,
+                                        4,
+                                        5,
+                                        8,
+                                    },
+                                    ec_gf8_mul_D6_ops};
+
+static ec_gf_op_t ec_gf8_mul_D7_ops[] = {
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR3, 8, 3, 5}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 0, 8, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR3, 6, 7, 8}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_D7 = {9,
+                                    {
+                                        3,
+                                        4,
+                                        6,
+                                        5,
+                                        0,
+                                        7,
+                                        1,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_D7_ops};
+
+static ec_gf_op_t ec_gf8_mul_D8_ops[] = {
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_D8 = {8,
+                                    {
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                    },
+                                    ec_gf8_mul_D8_ops};
+
+static ec_gf_op_t ec_gf8_mul_D9_ops[] = {
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 7, 0, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_D9 = {8,
+                                    {
+                                        1,
+                                        2,
+                                        6,
+                                        7,
+                                        4,
+                                        5,
+                                        0,
+                                        3,
+                                    },
+                                    ec_gf8_mul_D9_ops};
+
+static ec_gf_op_t ec_gf8_mul_DA_ops[] = {
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR3, 8, 2, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 1, 8, 0},
+    {EC_GF_OP_XOR2, 5, 8, 0}, {EC_GF_OP_XOR2, 2, 4, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_DA = {9,
+                                    {
+                                        2,
+                                        5,
+                                        7,
+                                        1,
+                                        0,
+                                        4,
+                                        3,
+                                        6,
+                                        8,
+                                    },
+                                    ec_gf8_mul_DA_ops};
+
+static ec_gf_op_t ec_gf8_mul_DB_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 8, 4, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 6, 3, 0}, {EC_GF_OP_XOR2, 3, 8, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_DB = {9,
+                                    {
+                                        7,
+                                        5,
+                                        6,
+                                        2,
+                                        3,
+                                        4,
+                                        1,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_DB_ops};
+
+static ec_gf_op_t ec_gf8_mul_DC_ops[] = {
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_DC = {8,
+                                    {
+                                        4,
+                                        5,
+                                        2,
+                                        6,
+                                        7,
+                                        1,
+                                        0,
+                                        3,
+                                    },
+                                    ec_gf8_mul_DC_ops};
+
+static ec_gf_op_t ec_gf8_mul_DD_ops[] = {
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_DD = {8,
+                                    {
+                                        1,
+                                        2,
+                                        3,
+                                        6,
+                                        7,
+                                        0,
+                                        4,
+                                        5,
+                                    },
+                                    ec_gf8_mul_DD_ops};
+
+static ec_gf_op_t ec_gf8_mul_DE_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_DE = {8,
+                                    {
+                                        0,
+                                        5,
+                                        2,
+                                        6,
+                                        7,
+                                        1,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_DE_ops};
+
+static ec_gf_op_t ec_gf8_mul_DF_ops[] = {
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 8, 3, 0},
+    {EC_GF_OP_COPY, 9, 0, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 8, 7, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR2, 5, 8, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR3, 1, 9, 2}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_DF = {10,
+                                    {
+                                        7,
+                                        2,
+                                        8,
+                                        4,
+                                        3,
+                                        1,
+                                        0,
+                                        6,
+                                        5,
+                                        9,
+                                    },
+                                    ec_gf8_mul_DF_ops};
+
+static ec_gf_op_t ec_gf8_mul_E0_ops[] = {
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 4, 1, 0}, {EC_GF_OP_XOR2, 7, 1, 0},
+    {EC_GF_OP_XOR2, 5, 7, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_E0 = {8,
+                                    {
+                                        2,
+                                        3,
+                                        4,
+                                        7,
+                                        5,
+                                        6,
+                                        0,
+                                        1,
+                                    },
+                                    ec_gf8_mul_E0_ops};
+
+static ec_gf_op_t ec_gf8_mul_E1_ops[] = {
+    {EC_GF_OP_COPY, 8, 1, 0}, {EC_GF_OP_XOR2, 8, 7, 0},
+    {EC_GF_OP_XOR2, 3, 8, 0}, {EC_GF_OP_XOR3, 9, 5, 3},
+    {EC_GF_OP_XOR2, 0, 9, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 4, 9, 0}, {EC_GF_OP_XOR2, 0, 2, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 2, 8, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_E1 = {10,
+                                    {
+                                        0,
+                                        7,
+                                        1,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        2,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_E1_ops};
+
+static ec_gf_op_t ec_gf8_mul_E2_ops[] = {
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_E2 = {8,
+                                    {
+                                        2,
+                                        3,
+                                        7,
+                                        1,
+                                        5,
+                                        6,
+                                        0,
+                                        4,
+                                    },
+                                    ec_gf8_mul_E2_ops};
+
+static ec_gf_op_t ec_gf8_mul_E3_ops[] = {
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 3, 1, 0},
+    {EC_GF_OP_XOR3, 8, 2, 7}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 0, 1, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_XOR2, 0, 8, 0}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR3, 6, 8, 4},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_E3 = {9,
+                                    {
+                                        5,
+                                        4,
+                                        7,
+                                        2,
+                                        1,
+                                        3,
+                                        6,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_E3_ops};
+
+static ec_gf_op_t ec_gf8_mul_E4_ops[] = {
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 4, 5, 0},
+    {EC_GF_OP_XOR2, 3, 4, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_E4 = {8,
+                                    {
+                                        7,
+                                        0,
+                                        1,
+                                        6,
+                                        3,
+                                        4,
+                                        2,
+                                        5,
+                                    },
+                                    ec_gf8_mul_E4_ops};
+
+static ec_gf_op_t ec_gf8_mul_E5_ops[] = {
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_COPY, 8, 0, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 3, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_E5 = {9,
+                                    {
+                                        4,
+                                        5,
+                                        3,
+                                        6,
+                                        7,
+                                        1,
+                                        0,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_E5_ops};
+
+static ec_gf_op_t ec_gf8_mul_E6_ops[] = {
+    {EC_GF_OP_XOR2, 6, 2, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 1, 4, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_E6 = {8,
+                                    {
+                                        5,
+                                        4,
+                                        3,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_E6_ops};
+
+static ec_gf_op_t ec_gf8_mul_E7_ops[] = {
+    {EC_GF_OP_COPY, 8, 6, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR3, 9, 0, 6}, {EC_GF_OP_XOR2, 4, 9, 0},
+    {EC_GF_OP_XOR2, 5, 9, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_E7 = {10,
+                                    {
+                                        1,
+                                        4,
+                                        3,
+                                        6,
+                                        7,
+                                        5,
+                                        2,
+                                        0,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_E7_ops};
+
+static ec_gf_op_t ec_gf8_mul_E8_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 2, 5, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 1, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_E8 = {8,
+                                    {
+                                        1,
+                                        4,
+                                        2,
+                                        7,
+                                        3,
+                                        0,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_E8_ops};
+
+static ec_gf_op_t ec_gf8_mul_E9_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_COPY, 8, 1, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 6, 3, 0},
+    {EC_GF_OP_XOR2, 4, 6, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR3, 1, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_E9 = {9,
+                                    {
+                                        6,
+                                        2,
+                                        0,
+                                        3,
+                                        4,
+                                        1,
+                                        5,
+                                        7,
+                                        8,
+                                    },
+                                    ec_gf8_mul_E9_ops};
+
+static ec_gf_op_t ec_gf8_mul_EA_ops[] = {
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_EA = {8,
+                                    {
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_EA_ops};
+
+static ec_gf_op_t ec_gf8_mul_EB_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_EB = {8,
+                                    {
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                    },
+                                    ec_gf8_mul_EB_ops};
+
+static ec_gf_op_t ec_gf8_mul_EC_ops[] = {
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR3, 8, 4, 0}, {EC_GF_OP_XOR2, 1, 8, 0},
+    {EC_GF_OP_XOR2, 7, 3, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 3, 8, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_EC = {9,
+                                    {
+                                        7,
+                                        4,
+                                        3,
+                                        0,
+                                        2,
+                                        5,
+                                        1,
+                                        6,
+                                        8,
+                                    },
+                                    ec_gf8_mul_EC_ops};
+
+static ec_gf_op_t ec_gf8_mul_ED_ops[] = {
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 4, 0, 0},
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 3, 6, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 5, 2, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_ED = {8,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        4,
+                                        3,
+                                        2,
+                                    },
+                                    ec_gf8_mul_ED_ops};
+
+static ec_gf_op_t ec_gf8_mul_EE_ops[] = {
+    {EC_GF_OP_XOR2, 5, 3, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR3, 8, 2, 3},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_XOR2, 4, 8, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 8, 5, 0},
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 1, 8, 0}, {EC_GF_OP_XOR2, 7, 8, 0},
+    {EC_GF_OP_XOR2, 6, 0, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_EE = {9,
+                                    {
+                                        6,
+                                        4,
+                                        5,
+                                        7,
+                                        2,
+                                        3,
+                                        0,
+                                        1,
+                                        8,
+                                    },
+                                    ec_gf8_mul_EE_ops};
+
+static ec_gf_op_t ec_gf8_mul_EF_ops[] = {
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_COPY, 8, 0, 0},
+    {EC_GF_OP_XOR2, 8, 2, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 7, 8, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 6, 8, 0},
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 7, 5, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_EF = {9,
+                                    {
+                                        6,
+                                        4,
+                                        5,
+                                        7,
+                                        2,
+                                        0,
+                                        3,
+                                        1,
+                                        8,
+                                    },
+                                    ec_gf8_mul_EF_ops};
+
+static ec_gf_op_t ec_gf8_mul_F0_ops[] = {
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR3, 8, 3, 6},
+    {EC_GF_OP_XOR2, 5, 8, 0}, {EC_GF_OP_XOR2, 8, 4, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 7, 8, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 1, 8, 0},
+    {EC_GF_OP_XOR2, 0, 2, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_F0 = {9,
+                                    {
+                                        3,
+                                        4,
+                                        6,
+                                        1,
+                                        2,
+                                        0,
+                                        5,
+                                        7,
+                                        8,
+                                    },
+                                    ec_gf8_mul_F0_ops};
+
+static ec_gf_op_t ec_gf8_mul_F1_ops[] = {
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_COPY, 8, 3, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 2, 3, 0}, {EC_GF_OP_COPY, 9, 2, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 9, 0, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 5, 2, 0},
+    {EC_GF_OP_XOR2, 7, 9, 0}, {EC_GF_OP_XOR2, 4, 9, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR3, 9, 8, 7},
+    {EC_GF_OP_XOR2, 1, 9, 0}, {EC_GF_OP_XOR2, 5, 9, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_F1 = {10,
+                                    {
+                                        7,
+                                        2,
+                                        6,
+                                        3,
+                                        5,
+                                        1,
+                                        4,
+                                        0,
+                                        8,
+                                        9,
+                                    },
+                                    ec_gf8_mul_F1_ops};
+
+static ec_gf_op_t ec_gf8_mul_F2_ops[] = {
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 7, 2, 0},
+    {EC_GF_OP_XOR2, 0, 6, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 2, 3, 0},
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_XOR3, 8, 6, 4},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 3, 8, 0}, {EC_GF_OP_XOR2, 5, 8, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_F2 = {9,
+                                    {
+                                        1,
+                                        0,
+                                        6,
+                                        7,
+                                        4,
+                                        5,
+                                        2,
+                                        3,
+                                        8,
+                                    },
+                                    ec_gf8_mul_F2_ops};
+
+static ec_gf_op_t ec_gf8_mul_F3_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 6, 5, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_F3 = {8,
+                                    {
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                    },
+                                    ec_gf8_mul_F3_ops};
+
+static ec_gf_op_t ec_gf8_mul_F4_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 1, 7, 0}, {EC_GF_OP_XOR2, 3, 7, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_F4 = {8,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                    },
+                                    ec_gf8_mul_F4_ops};
+
+static ec_gf_op_t ec_gf8_mul_F5_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_F5 = {8,
+                                    {
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                        6,
+                                    },
+                                    ec_gf8_mul_F5_ops};
+
+static ec_gf_op_t ec_gf8_mul_F6_ops[] = {
+    {EC_GF_OP_XOR2, 3, 1, 0}, {EC_GF_OP_COPY, 8, 3, 0},
+    {EC_GF_OP_XOR2, 3, 5, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_COPY, 9, 3, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 2, 7, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 9, 4, 0}, {EC_GF_OP_XOR2, 4, 1, 0},
+    {EC_GF_OP_XOR2, 6, 9, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 5, 7, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR3, 7, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_F6 = {10,
+                                    {
+                                        0,
+                                        6,
+                                        2,
+                                        7,
+                                        4,
+                                        3,
+                                        5,
+                                        9,
+                                        1,
+                                        8,
+                                    },
+                                    ec_gf8_mul_F6_ops};
+
+static ec_gf_op_t ec_gf8_mul_F7_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 3, 2, 0}, {EC_GF_OP_XOR2, 4, 3, 0},
+    {EC_GF_OP_XOR2, 5, 4, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_F7 = {8,
+                                    {
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                        4,
+                                        5,
+                                    },
+                                    ec_gf8_mul_F7_ops};
+
+static ec_gf_op_t ec_gf8_mul_F8_ops[] = {
+    {EC_GF_OP_XOR2, 4, 0, 0}, {EC_GF_OP_XOR2, 3, 5, 0},
+    {EC_GF_OP_XOR2, 6, 4, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 1, 6, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 5, 1, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 6, 7, 0},
+    {EC_GF_OP_XOR2, 0, 3, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_F8 = {8,
+                                    {
+                                        6,
+                                        2,
+                                        0,
+                                        1,
+                                        4,
+                                        5,
+                                        3,
+                                        7,
+                                    },
+                                    ec_gf8_mul_F8_ops};
+
+static ec_gf_op_t ec_gf8_mul_F9_ops[] = {
+    {EC_GF_OP_XOR2, 1, 5, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 0, 5, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 6, 4, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR3, 8, 7, 1}, {EC_GF_OP_XOR2, 1, 3, 0},
+    {EC_GF_OP_XOR2, 4, 8, 0}, {EC_GF_OP_XOR2, 5, 8, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_F9 = {9,
+                                    {
+                                        4,
+                                        1,
+                                        7,
+                                        6,
+                                        0,
+                                        3,
+                                        5,
+                                        2,
+                                        8,
+                                    },
+                                    ec_gf8_mul_F9_ops};
+
+static ec_gf_op_t ec_gf8_mul_FA_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 2, 1, 0}, {EC_GF_OP_XOR2, 0, 7, 0},
+    {EC_GF_OP_XOR2, 7, 2, 0}, {EC_GF_OP_XOR2, 1, 5, 0},
+    {EC_GF_OP_XOR2, 3, 7, 0}, {EC_GF_OP_XOR2, 5, 0, 0},
+    {EC_GF_OP_XOR2, 7, 6, 0}, {EC_GF_OP_XOR2, 0, 3, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_FA = {8,
+                                    {
+                                        0,
+                                        1,
+                                        2,
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        3,
+                                    },
+                                    ec_gf8_mul_FA_ops};
+
+static ec_gf_op_t ec_gf8_mul_FB_ops[] = {
+    {EC_GF_OP_XOR2, 1, 0, 0}, {EC_GF_OP_XOR2, 2, 1, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 3, 2, 0},
+    {EC_GF_OP_XOR2, 0, 7, 0}, {EC_GF_OP_XOR2, 2, 7, 0},
+    {EC_GF_OP_XOR2, 1, 6, 0}, {EC_GF_OP_XOR2, 7, 6, 0},
+    {EC_GF_OP_XOR2, 4, 3, 0}, {EC_GF_OP_XOR2, 6, 5, 0},
+    {EC_GF_OP_XOR2, 7, 4, 0}, {EC_GF_OP_XOR2, 5, 4, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_FB = {8,
+                                    {
+                                        4,
+                                        5,
+                                        6,
+                                        7,
+                                        0,
+                                        1,
+                                        2,
+                                        3,
+                                    },
+                                    ec_gf8_mul_FB_ops};
+
+static ec_gf_op_t ec_gf8_mul_FC_ops[] = {
+    {EC_GF_OP_XOR2, 7, 0, 0}, {EC_GF_OP_XOR2, 7, 4, 0},
+    {EC_GF_OP_XOR2, 5, 1, 0}, {EC_GF_OP_COPY, 9, 3, 0},
+    {EC_GF_OP_XOR3, 8, 5, 7}, {EC_GF_OP_XOR2, 3, 6, 0},
+    {EC_GF_OP_XOR2, 8, 3, 0}, {EC_GF_OP_XOR2, 2, 8, 0},
+    {EC_GF_OP_XOR2, 1, 2, 0}, {EC_GF_OP_XOR2, 4, 2, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 3, 4, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 6, 0, 0},
+    {EC_GF_OP_XOR3, 0, 9, 2}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_FC = {10,
+                                    {
+                                        5,
+                                        6,
+                                        3,
+                                        7,
+                                        1,
+                                        8,
+                                        0,
+                                        4,
+                                        2,
+                                        9,
+                                    },
+                                    ec_gf8_mul_FC_ops};
+
+static ec_gf_op_t ec_gf8_mul_FD_ops[] = {
+    {EC_GF_OP_XOR2, 7, 1, 0}, {EC_GF_OP_COPY, 8, 7, 0},
+    {EC_GF_OP_XOR2, 5, 0, 0}, {EC_GF_OP_XOR2, 7, 5, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 4, 7, 0},
+    {EC_GF_OP_XOR2, 5, 6, 0}, {EC_GF_OP_XOR2, 0, 4, 0},
+    {EC_GF_OP_XOR2, 3, 0, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 2, 5, 0}, {EC_GF_OP_XOR2, 1, 2, 0},
+    {EC_GF_OP_XOR2, 0, 1, 0}, {EC_GF_OP_XOR2, 6, 1, 0},
+    {EC_GF_OP_XOR3, 1, 8, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_FD = {9,
+                                    {
+                                        5,
+                                        3,
+                                        7,
+                                        6,
+                                        1,
+                                        2,
+                                        4,
+                                        0,
+                                        8,
+                                    },
+                                    ec_gf8_mul_FD_ops};
+
+static ec_gf_op_t ec_gf8_mul_FE_ops[] = {
+    {EC_GF_OP_XOR2, 2, 0, 0}, {EC_GF_OP_COPY, 8, 2, 0},
+    {EC_GF_OP_XOR2, 2, 4, 0}, {EC_GF_OP_XOR2, 6, 2, 0},
+    {EC_GF_OP_XOR2, 8, 5, 0}, {EC_GF_OP_XOR2, 5, 6, 0},
+    {EC_GF_OP_XOR2, 6, 1, 0}, {EC_GF_OP_XOR2, 0, 6, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 7, 8, 0}, {EC_GF_OP_XOR2, 3, 0, 0},
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR2, 0, 4, 0}, {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_FE = {9,
+                                    {
+                                        3,
+                                        4,
+                                        8,
+                                        2,
+                                        5,
+                                        0,
+                                        6,
+                                        1,
+                                        7,
+                                    },
+                                    ec_gf8_mul_FE_ops};
+
+static ec_gf_op_t ec_gf8_mul_FF_ops[] = {
+    {EC_GF_OP_XOR2, 4, 7, 0}, {EC_GF_OP_COPY, 9, 0, 0},
+    {EC_GF_OP_COPY, 8, 4, 0}, {EC_GF_OP_XOR2, 9, 1, 0},
+    {EC_GF_OP_XOR2, 4, 2, 0}, {EC_GF_OP_XOR2, 9, 4, 0},
+    {EC_GF_OP_XOR2, 0, 5, 0}, {EC_GF_OP_XOR2, 2, 0, 0},
+    {EC_GF_OP_XOR2, 3, 9, 0}, {EC_GF_OP_XOR2, 7, 3, 0},
+    {EC_GF_OP_XOR2, 2, 6, 0}, {EC_GF_OP_XOR2, 5, 3, 0},
+    {EC_GF_OP_XOR2, 6, 7, 0}, {EC_GF_OP_XOR2, 1, 7, 0},
+    {EC_GF_OP_XOR3, 3, 8, 5}, {EC_GF_OP_XOR2, 4, 6, 0},
+    {EC_GF_OP_END, 0, 0, 0}};
+
+static ec_gf_mul_t ec_gf8_mul_FF = {10,
+                                    {
+                                        6,
+                                        5,
+                                        0,
+                                        1,
+                                        2,
+                                        4,
+                                        9,
+                                        3,
+                                        7,
+                                        8,
+                                    },
+                                    ec_gf8_mul_FF_ops};
+
+ec_gf_mul_t *ec_gf8_mul[] = {
+    &ec_gf8_mul_00, &ec_gf8_mul_01, &ec_gf8_mul_02, &ec_gf8_mul_03,
+    &ec_gf8_mul_04, &ec_gf8_mul_05, &ec_gf8_mul_06, &ec_gf8_mul_07,
+    &ec_gf8_mul_08, &ec_gf8_mul_09, &ec_gf8_mul_0A, &ec_gf8_mul_0B,
+    &ec_gf8_mul_0C, &ec_gf8_mul_0D, &ec_gf8_mul_0E, &ec_gf8_mul_0F,
+    &ec_gf8_mul_10, &ec_gf8_mul_11, &ec_gf8_mul_12, &ec_gf8_mul_13,
+    &ec_gf8_mul_14, &ec_gf8_mul_15, &ec_gf8_mul_16, &ec_gf8_mul_17,
+    &ec_gf8_mul_18, &ec_gf8_mul_19, &ec_gf8_mul_1A, &ec_gf8_mul_1B,
+    &ec_gf8_mul_1C, &ec_gf8_mul_1D, &ec_gf8_mul_1E, &ec_gf8_mul_1F,
+    &ec_gf8_mul_20, &ec_gf8_mul_21, &ec_gf8_mul_22, &ec_gf8_mul_23,
+    &ec_gf8_mul_24, &ec_gf8_mul_25, &ec_gf8_mul_26, &ec_gf8_mul_27,
+    &ec_gf8_mul_28, &ec_gf8_mul_29, &ec_gf8_mul_2A, &ec_gf8_mul_2B,
+    &ec_gf8_mul_2C, &ec_gf8_mul_2D, &ec_gf8_mul_2E, &ec_gf8_mul_2F,
+    &ec_gf8_mul_30, &ec_gf8_mul_31, &ec_gf8_mul_32, &ec_gf8_mul_33,
+    &ec_gf8_mul_34, &ec_gf8_mul_35, &ec_gf8_mul_36, &ec_gf8_mul_37,
+    &ec_gf8_mul_38, &ec_gf8_mul_39, &ec_gf8_mul_3A, &ec_gf8_mul_3B,
+    &ec_gf8_mul_3C, &ec_gf8_mul_3D, &ec_gf8_mul_3E, &ec_gf8_mul_3F,
+    &ec_gf8_mul_40, &ec_gf8_mul_41, &ec_gf8_mul_42, &ec_gf8_mul_43,
+    &ec_gf8_mul_44, &ec_gf8_mul_45, &ec_gf8_mul_46, &ec_gf8_mul_47,
+    &ec_gf8_mul_48, &ec_gf8_mul_49, &ec_gf8_mul_4A, &ec_gf8_mul_4B,
+    &ec_gf8_mul_4C, &ec_gf8_mul_4D, &ec_gf8_mul_4E, &ec_gf8_mul_4F,
+    &ec_gf8_mul_50, &ec_gf8_mul_51, &ec_gf8_mul_52, &ec_gf8_mul_53,
+    &ec_gf8_mul_54, &ec_gf8_mul_55, &ec_gf8_mul_56, &ec_gf8_mul_57,
+    &ec_gf8_mul_58, &ec_gf8_mul_59, &ec_gf8_mul_5A, &ec_gf8_mul_5B,
+    &ec_gf8_mul_5C, &ec_gf8_mul_5D, &ec_gf8_mul_5E, &ec_gf8_mul_5F,
+    &ec_gf8_mul_60, &ec_gf8_mul_61, &ec_gf8_mul_62, &ec_gf8_mul_63,
+    &ec_gf8_mul_64, &ec_gf8_mul_65, &ec_gf8_mul_66, &ec_gf8_mul_67,
+    &ec_gf8_mul_68, &ec_gf8_mul_69, &ec_gf8_mul_6A, &ec_gf8_mul_6B,
+    &ec_gf8_mul_6C, &ec_gf8_mul_6D, &ec_gf8_mul_6E, &ec_gf8_mul_6F,
+    &ec_gf8_mul_70, &ec_gf8_mul_71, &ec_gf8_mul_72, &ec_gf8_mul_73,
+    &ec_gf8_mul_74, &ec_gf8_mul_75, &ec_gf8_mul_76, &ec_gf8_mul_77,
+    &ec_gf8_mul_78, &ec_gf8_mul_79, &ec_gf8_mul_7A, &ec_gf8_mul_7B,
+    &ec_gf8_mul_7C, &ec_gf8_mul_7D, &ec_gf8_mul_7E, &ec_gf8_mul_7F,
+    &ec_gf8_mul_80, &ec_gf8_mul_81, &ec_gf8_mul_82, &ec_gf8_mul_83,
+    &ec_gf8_mul_84, &ec_gf8_mul_85, &ec_gf8_mul_86, &ec_gf8_mul_87,
+    &ec_gf8_mul_88, &ec_gf8_mul_89, &ec_gf8_mul_8A, &ec_gf8_mul_8B,
+    &ec_gf8_mul_8C, &ec_gf8_mul_8D, &ec_gf8_mul_8E, &ec_gf8_mul_8F,
+    &ec_gf8_mul_90, &ec_gf8_mul_91, &ec_gf8_mul_92, &ec_gf8_mul_93,
+    &ec_gf8_mul_94, &ec_gf8_mul_95, &ec_gf8_mul_96, &ec_gf8_mul_97,
+    &ec_gf8_mul_98, &ec_gf8_mul_99, &ec_gf8_mul_9A, &ec_gf8_mul_9B,
+    &ec_gf8_mul_9C, &ec_gf8_mul_9D, &ec_gf8_mul_9E, &ec_gf8_mul_9F,
+    &ec_gf8_mul_A0, &ec_gf8_mul_A1, &ec_gf8_mul_A2, &ec_gf8_mul_A3,
+    &ec_gf8_mul_A4, &ec_gf8_mul_A5, &ec_gf8_mul_A6, &ec_gf8_mul_A7,
+    &ec_gf8_mul_A8, &ec_gf8_mul_A9, &ec_gf8_mul_AA, &ec_gf8_mul_AB,
+    &ec_gf8_mul_AC, &ec_gf8_mul_AD, &ec_gf8_mul_AE, &ec_gf8_mul_AF,
+    &ec_gf8_mul_B0, &ec_gf8_mul_B1, &ec_gf8_mul_B2, &ec_gf8_mul_B3,
+    &ec_gf8_mul_B4, &ec_gf8_mul_B5, &ec_gf8_mul_B6, &ec_gf8_mul_B7,
+    &ec_gf8_mul_B8, &ec_gf8_mul_B9, &ec_gf8_mul_BA, &ec_gf8_mul_BB,
+    &ec_gf8_mul_BC, &ec_gf8_mul_BD, &ec_gf8_mul_BE, &ec_gf8_mul_BF,
+    &ec_gf8_mul_C0, &ec_gf8_mul_C1, &ec_gf8_mul_C2, &ec_gf8_mul_C3,
+    &ec_gf8_mul_C4, &ec_gf8_mul_C5, &ec_gf8_mul_C6, &ec_gf8_mul_C7,
+    &ec_gf8_mul_C8, &ec_gf8_mul_C9, &ec_gf8_mul_CA, &ec_gf8_mul_CB,
+    &ec_gf8_mul_CC, &ec_gf8_mul_CD, &ec_gf8_mul_CE, &ec_gf8_mul_CF,
+    &ec_gf8_mul_D0, &ec_gf8_mul_D1, &ec_gf8_mul_D2, &ec_gf8_mul_D3,
+    &ec_gf8_mul_D4, &ec_gf8_mul_D5, &ec_gf8_mul_D6, &ec_gf8_mul_D7,
+    &ec_gf8_mul_D8, &ec_gf8_mul_D9, &ec_gf8_mul_DA, &ec_gf8_mul_DB,
+    &ec_gf8_mul_DC, &ec_gf8_mul_DD, &ec_gf8_mul_DE, &ec_gf8_mul_DF,
+    &ec_gf8_mul_E0, &ec_gf8_mul_E1, &ec_gf8_mul_E2, &ec_gf8_mul_E3,
+    &ec_gf8_mul_E4, &ec_gf8_mul_E5, &ec_gf8_mul_E6, &ec_gf8_mul_E7,
+    &ec_gf8_mul_E8, &ec_gf8_mul_E9, &ec_gf8_mul_EA, &ec_gf8_mul_EB,
+    &ec_gf8_mul_EC, &ec_gf8_mul_ED, &ec_gf8_mul_EE, &ec_gf8_mul_EF,
+    &ec_gf8_mul_F0, &ec_gf8_mul_F1, &ec_gf8_mul_F2, &ec_gf8_mul_F3,
+    &ec_gf8_mul_F4, &ec_gf8_mul_F5, &ec_gf8_mul_F6, &ec_gf8_mul_F7,
+    &ec_gf8_mul_F8, &ec_gf8_mul_F9, &ec_gf8_mul_FA, &ec_gf8_mul_FB,
+    &ec_gf8_mul_FC, &ec_gf8_mul_FD, &ec_gf8_mul_FE, &ec_gf8_mul_FF};
diff --git a/xlators/cluster/ec/src/ec-gf8.h b/xlators/cluster/ec/src/ec-gf8.h
new file mode 100644
index 00000000000..4aca91127fc
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-gf8.h
@@ -0,0 +1,18 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_GF8_H__
+#define __EC_GF8_H__
+
+#include "ec-galois.h"
+
+extern ec_gf_mul_t *ec_gf8_mul[];
+
+#endif /* __EC_GF8_H__ */
diff --git a/xlators/cluster/ec/src/ec-heal.c b/xlators/cluster/ec/src/ec-heal.c
new file mode 100644
index 00000000000..7d991f04aac
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-heal.c
@@ -0,0 +1,3367 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/defaults.h>
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/byte-order.h>
+#include <glusterfs/syncop.h>
+#include <glusterfs/syncop-utils.h>
+#include <glusterfs/cluster-syncop.h>
+
+#include "ec.h"
+#include "ec-types.h"
+#include "ec-messages.h"
+#include "ec-helpers.h"
+#include "ec-common.h"
+#include "ec-combine.h"
+#include "ec-method.h"
+#include "ec-fops.h"
+#include "ec-heald.h"
+
+#define EC_COUNT(array, max)                                                   \
+    ({                                                                         \
+        int __i;                                                               \
+        int __res = 0;                                                         \
+        for (__i = 0; __i < max; __i++)                                        \
+            if (array[__i])                                                    \
+                __res++;                                                       \
+        __res;                                                                 \
+    })
+#define EC_INTERSECT(dst, src1, src2, max)                                     \
+    ({                                                                         \
+        int __i;                                                               \
+        for (__i = 0; __i < max; __i++)                                        \
+            dst[__i] = src1[__i] && src2[__i];                                 \
+    })
+#define EC_ADJUST_SOURCE(source, sources, max)                                 \
+    ({                                                                         \
+        int __i;                                                               \
+        if (sources[source] == 0) {                                            \
+            source = -1;                                                       \
+            for (__i = 0; __i < max; __i++)                                    \
+                if (sources[__i])                                              \
+                    source = __i;                                              \
+        }                                                                      \
+    })
+#define IA_EQUAL(f, s, field)                                                  \
+    (memcmp(&(f.ia_##field), &(s.ia_##field), sizeof(s.ia_##field)) == 0)
+#define EC_REPLIES_ALLOC(replies, numsubvols)                                  \
+    do {                                                                       \
+        int __i = 0;                                                           \
+        replies = alloca0(numsubvols * sizeof(*replies));                      \
+        for (__i = 0; __i < numsubvols; __i++)                                 \
+            INIT_LIST_HEAD(&replies[__i].entries.list);                        \
+    } while (0)
+
+struct ec_name_data {
+    call_frame_t *frame;
+    unsigned char *participants;
+    unsigned char *failed_on;
+    unsigned char *gfidless;
+    unsigned char *enoent;
+    unsigned char *same;
+    char *name;
+    inode_t *parent;
+    default_args_cbk_t *replies;
+    uint32_t heal_pending;
+};
+
+static char *ec_ignore_xattrs[] = {GF_SELINUX_XATTR_KEY, QUOTA_SIZE_KEY, NULL};
+
+static gf_boolean_t
+ec_ignorable_key_match(dict_t *dict, char *key, data_t *val, void *mdata)
+{
+    int i = 0;
+
+    if (!key)
+        goto out;
+
+    if (strncmp(key, EC_XATTR_PREFIX, SLEN(EC_XATTR_PREFIX)) == 0)
+        return _gf_true;
+
+    for (i = 0; ec_ignore_xattrs[i]; i++) {
+        if (!strcmp(key, ec_ignore_xattrs[i]))
+            return _gf_true;
+    }
+
+out:
+    return _gf_false;
+}
+
+static gf_boolean_t
+ec_sh_key_match(dict_t *dict, char *key, data_t *val, void *mdata)
+{
+    return !ec_ignorable_key_match(dict, key, val, mdata);
+}
+/* FOP: heal */
+
+void
+ec_set_entry_healing(ec_fop_data_t *fop)
+{
+    ec_inode_t *ctx = NULL;
+    loc_t *loc = NULL;
+
+    if (!fop)
+        return;
+
+    loc = &fop->loc[0];
+    LOCK(&loc->inode->lock);
+    {
+        ctx = __ec_inode_get(loc->inode, fop->xl);
+        if (ctx) {
+            ctx->heal_count += 1;
+        }
+    }
+    UNLOCK(&loc->inode->lock);
+}
+
+void
+ec_reset_entry_healing(ec_fop_data_t *fop)
+{
+    ec_inode_t *ctx = NULL;
+    loc_t *loc = NULL;
+    int32_t heal_count = 0;
+    if (!fop)
+        return;
+
+    loc = &fop->loc[0];
+    LOCK(&loc->inode->lock);
+    {
+        ctx = __ec_inode_get(loc->inode, fop->xl);
+        if (ctx) {
+            ctx->heal_count += -1;
+            heal_count = ctx->heal_count;
+        }
+    }
+    UNLOCK(&loc->inode->lock);
+    GF_ASSERT(heal_count >= 0);
+}
+
+uintptr_t
+ec_heal_check(ec_fop_data_t *fop, uintptr_t *pgood)
+{
+    ec_cbk_data_t *cbk;
+    uintptr_t mask[2] = {0, 0};
+
+    list_for_each_entry(cbk, &fop->cbk_list, list)
+    {
+        mask[cbk->op_ret >= 0] |= cbk->mask;
+    }
+
+    if (pgood != NULL) {
+        *pgood = mask[1];
+    }
+
+    return mask[0];
+}
+
+void
+ec_heal_update(ec_fop_data_t *fop, int32_t is_open)
+{
+    ec_heal_t *heal = fop->data;
+    uintptr_t good, bad;
+
+    bad = ec_heal_check(fop, &good);
+
+    LOCK(&heal->lock);
+
+    heal->bad &= ~bad;
+    if (is_open) {
+        heal->open |= good;
+    }
+
+    UNLOCK(&heal->lock);
+
+    fop->error = 0;
+}
+
+void
+ec_heal_avoid(ec_fop_data_t *fop)
+{
+    ec_heal_t *heal = fop->data;
+    uintptr_t bad;
+
+    bad = ec_heal_check(fop, NULL);
+
+    LOCK(&heal->lock);
+
+    heal->good &= ~bad;
+
+    UNLOCK(&heal->lock);
+}
+
+int32_t
+ec_heal_lock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = cookie;
+    ec_heal_t *heal = fop->data;
+
+    if (op_ret >= 0) {
+        GF_ASSERT(
+            ec_set_inode_size(heal->fop, heal->fd->inode, heal->total_size));
+    }
+
+    return 0;
+}
+
+void
+ec_heal_lock(ec_heal_t *heal, int32_t type, fd_t *fd, loc_t *loc, off_t offset,
+             size_t size)
+{
+    struct gf_flock flock;
+    fop_inodelk_cbk_t cbk = NULL;
+
+    flock.l_type = type;
+    flock.l_whence = SEEK_SET;
+    flock.l_start = offset;
+    flock.l_len = size;
+    flock.l_pid = 0;
+    flock.l_owner.len = 0;
+
+    if (type == F_UNLCK) {
+        /* Remove inode size information before unlocking it. */
+        if (fd == NULL) {
+            ec_clear_inode_info(heal->fop, heal->loc.inode);
+        } else {
+            ec_clear_inode_info(heal->fop, heal->fd->inode);
+        }
+        cbk = ec_lock_unlocked;
+    } else {
+        /* Otherwise use the callback to update size information. */
+        cbk = ec_heal_lock_cbk;
+    }
+
+    if (fd != NULL) {
+        ec_finodelk(heal->fop->frame, heal->xl,
+                    &heal->fop->frame->root->lk_owner, heal->fop->mask,
+                    EC_MINIMUM_ALL, cbk, heal, heal->xl->name, fd, F_SETLKW,
+                    &flock, NULL);
+    } else {
+        ec_inodelk(heal->fop->frame, heal->xl,
+                   &heal->fop->frame->root->lk_owner, heal->fop->mask,
+                   EC_MINIMUM_ALL, cbk, heal, heal->xl->name, loc, F_SETLKW,
+                   &flock, NULL);
+    }
+}
+
+void
+ec_heal_inodelk(ec_heal_t *heal, int32_t type, int32_t use_fd, off_t offset,
+                size_t size)
+{
+    ec_heal_lock(heal, type, use_fd ? heal->fd : NULL, &heal->loc, offset,
+                 size);
+}
+
+int32_t
+ec_heal_xattr_clean(dict_t *dict, char *key, data_t *data, void *arg)
+{
+    dict_t *base = arg;
+
+    if (ec_ignorable_key_match(NULL, key, NULL, NULL)) {
+        dict_del(dict, key);
+        return 0;
+    }
+
+    if (dict_get(base, key) != NULL)
+        dict_del(dict, key);
+
+    return 0;
+}
+
+/********************************************************************
+ * ec_wind_xattrop_parallel:
+ *              Helper function to update the extended attributes
+ *    in parallel.
+ *
+ *******************************************************************/
+void
+ec_wind_xattrop_parallel(call_frame_t *frame, xlator_t *subvol, int child_index,
+                         loc_t *loc, gf_xattrop_flags_t flags, dict_t **dict,
+                         dict_t *xdata)
+{
+    gf_msg_debug("EC", 0, "WIND: on child %d ", child_index);
+    STACK_WIND_COOKIE(
+        frame, cluster_xattrop_cbk, (void *)(uintptr_t)child_index, subvol,
+        subvol->fops->xattrop, loc, flags, dict[child_index], xdata);
+}
+
+int32_t
+ec_heal_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                   struct iatt *postbuf, dict_t *xdata)
+{
+    ec_fop_data_t *fop = cookie;
+    ec_heal_t *heal = fop->data;
+
+    ec_trace("WRITE_CBK", cookie, "ret=%d, errno=%d", op_ret, op_errno);
+
+    gf_msg_debug(fop->xl->name, 0,
+                 "%s: write op_ret %d, op_errno %s"
+                 " at %" PRIu64,
+                 uuid_utoa(heal->fd->inode->gfid), op_ret, strerror(op_errno),
+                 heal->offset);
+
+    ec_heal_update(cookie, 0);
+
+    return 0;
+}
+
+int32_t
+ec_heal_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct iovec *vector,
+                  int32_t count, struct iatt *stbuf, struct iobref *iobref,
+                  dict_t *xdata)
+{
+    ec_fop_data_t *fop = cookie;
+    ec_heal_t *heal = fop->data;
+
+    ec_trace("READ_CBK", fop, "ret=%d, errno=%d", op_ret, op_errno);
+
+    ec_heal_avoid(fop);
+
+    if (op_ret > 0) {
+        gf_msg_debug(fop->xl->name, 0,
+                     "%s: read succeeded, proceeding "
+                     "to write at %" PRIu64,
+                     uuid_utoa(heal->fd->inode->gfid), heal->offset);
+        ec_writev(heal->fop->frame, heal->xl, heal->bad, EC_MINIMUM_ONE,
+                  ec_heal_writev_cbk, heal, heal->fd, vector, count,
+                  heal->offset, 0, iobref, NULL);
+    } else {
+        if (op_ret < 0) {
+            gf_msg_debug(fop->xl->name, 0,
+                         "%s: read failed %s, failing "
+                         "to heal block at %" PRIu64,
+                         uuid_utoa(heal->fd->inode->gfid), strerror(op_errno),
+                         heal->offset);
+            heal->bad = 0;
+        }
+        heal->done = 1;
+    }
+
+    return 0;
+}
+
+void
+ec_heal_data_block(ec_heal_t *heal)
+{
+    ec_trace("DATA", heal->fop, "good=%lX, bad=%lX", heal->good, heal->bad);
+
+    if ((heal->good != 0) && (heal->bad != 0) &&
+        (heal->iatt.ia_type == IA_IFREG)) {
+        ec_readv(heal->fop->frame, heal->xl, heal->good, EC_MINIMUM_MIN,
+                 ec_heal_readv_cbk, heal, heal->fd, heal->size, heal->offset, 0,
+                 NULL);
+    }
+}
+
+/* FOP: fheal */
+
+void
+ec_fheal(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_fheal_cbk_t func, void *data, fd_t *fd,
+         int32_t partial, dict_t *xdata)
+{
+    ec_fd_t *ctx = ec_fd_get(fd, this);
+
+    if (ctx != NULL) {
+        gf_msg_trace("ec", 0, "FHEAL ctx: flags=%X, open=%" PRIXPTR, ctx->flags,
+                     ctx->open);
+        ec_heal(frame, this, target, fop_flags, func, data, &ctx->loc, partial,
+                xdata);
+    }
+}
+
+/* Common heal code */
+void
+ec_mask_to_char_array(uintptr_t mask, unsigned char *array, int numsubvols)
+{
+    int i = 0;
+
+    for (i = 0; i < numsubvols; i++)
+        array[i] = ((mask >> i) & 1);
+}
+
+uintptr_t
+ec_char_array_to_mask(unsigned char *array, int numsubvols)
+{
+    int i = 0;
+    uintptr_t mask = 0;
+
+    if (array == NULL)
+        goto out;
+
+    for (i = 0; i < numsubvols; i++)
+        if (array[i])
+            mask |= (1ULL << i);
+out:
+    return mask;
+}
+
+int
+ec_heal_entry_find_direction(ec_t *ec, default_args_cbk_t *replies,
+                             uint64_t *versions, uint64_t *dirty,
+                             unsigned char *sources,
+                             unsigned char *healed_sinks)
+{
+    uint64_t xattr[EC_VERSION_SIZE] = {0};
+    int source = -1;
+    uint64_t max_version = 0;
+    int ret = 0;
+    int i = 0;
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (!replies[i].valid)
+            continue;
+
+        if (replies[i].op_ret == -1)
+            continue;
+
+        if (source == -1)
+            source = i;
+
+        ret = ec_dict_get_array(replies[i].xdata, EC_XATTR_VERSION, xattr,
+                                EC_VERSION_SIZE);
+        if (ret == 0) {
+            versions[i] = xattr[EC_DATA_TXN];
+            if (max_version < versions[i]) {
+                max_version = versions[i];
+                source = i;
+            }
+        }
+
+        memset(xattr, 0, sizeof(xattr));
+        ret = ec_dict_get_array(replies[i].xdata, EC_XATTR_DIRTY, xattr,
+                                EC_VERSION_SIZE);
+        if (ret == 0) {
+            dirty[i] = xattr[EC_DATA_TXN];
+        }
+    }
+
+    if (source < 0)
+        goto out;
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (!replies[i].valid)
+            continue;
+
+        if (replies[i].op_ret == -1)
+            continue;
+
+        if (versions[i] == versions[source])
+            sources[i] = 1;
+        else
+            healed_sinks[i] = 1;
+    }
+
+out:
+    return source;
+}
+
+int
+ec_adjust_versions(call_frame_t *frame, ec_t *ec, ec_txn_t type, inode_t *inode,
+                   int source, unsigned char *sources,
+                   unsigned char *healed_sinks, uint64_t *versions,
+                   uint64_t *dirty)
+{
+    int i = 0;
+    int ret = 0;
+    int call_count = 0;
+    dict_t **xattr = NULL;
+    int op_ret = 0;
+    loc_t loc = {0};
+    gf_boolean_t erase_dirty = _gf_false;
+    uint64_t *versions_xattr = NULL;
+    uint64_t *dirty_xattr = NULL;
+    uint64_t allzero[2] = {0};
+    unsigned char *on = NULL;
+    unsigned char *output = NULL;
+    default_args_cbk_t *replies = NULL;
+
+    /* Allocate the required memory */
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+    on = alloca0(ec->nodes);
+    output = alloca0(ec->nodes);
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    xattr = GF_CALLOC(ec->nodes, sizeof(*xattr), gf_common_mt_pointer);
+    if (!xattr) {
+        op_ret = -ENOMEM;
+        goto out;
+    }
+    for (i = 0; i < ec->nodes; i++) {
+        xattr[i] = dict_new();
+        if (!xattr[i]) {
+            op_ret = -ENOMEM;
+            goto out;
+        }
+    }
+
+    /* dirty xattr represents if the file/dir needs heal. Unless all the
+     * copies are healed, don't erase it */
+    if (EC_COUNT(sources, ec->nodes) + EC_COUNT(healed_sinks, ec->nodes) ==
+        ec->nodes)
+        erase_dirty = _gf_true;
+    else
+        op_ret = -ENOTCONN;
+
+    /* Populate the xattr array */
+    for (i = 0; i < ec->nodes; i++) {
+        if (!sources[i] && !healed_sinks[i])
+            continue;
+        versions_xattr = GF_CALLOC(EC_VERSION_SIZE, sizeof(*versions_xattr),
+                                   gf_common_mt_pointer);
+        if (!versions_xattr) {
+            op_ret = -ENOMEM;
+            continue;
+        }
+
+        versions_xattr[type] = hton64(versions[source] - versions[i]);
+        ret = dict_set_bin(xattr[i], EC_XATTR_VERSION, versions_xattr,
+                           (sizeof(*versions_xattr) * EC_VERSION_SIZE));
+        if (ret < 0) {
+            op_ret = -ENOMEM;
+            continue;
+        }
+
+        if (erase_dirty) {
+            dirty_xattr = GF_CALLOC(EC_VERSION_SIZE, sizeof(*dirty_xattr),
+                                    gf_common_mt_pointer);
+            if (!dirty_xattr) {
+                op_ret = -ENOMEM;
+                continue;
+            }
+
+            dirty_xattr[type] = hton64(-dirty[i]);
+            ret = dict_set_bin(xattr[i], EC_XATTR_DIRTY, dirty_xattr,
+                               (sizeof(*dirty_xattr) * EC_VERSION_SIZE));
+            if (ret < 0) {
+                op_ret = -ENOMEM;
+                continue;
+            }
+        }
+
+        if (memcmp(versions_xattr, allzero,
+                   (sizeof(*versions_xattr) * EC_VERSION_SIZE)) == 0) {
+            if (!erase_dirty) {
+                continue;
+            }
+
+            if (memcmp(dirty_xattr, allzero,
+                       (sizeof(*dirty_xattr) * EC_VERSION_SIZE)) == 0) {
+                continue;
+            }
+        }
+
+        on[i] = 1;
+        call_count++;
+    }
+
+    /* Update the bricks with xattr */
+    if (call_count) {
+        PARALLEL_FOP_ONLIST(ec->xl_list, on, ec->nodes, replies, frame,
+                            ec_wind_xattrop_parallel, &loc,
+                            GF_XATTROP_ADD_ARRAY64, xattr, NULL);
+        ret = cluster_fop_success_fill(replies, ec->nodes, output);
+    }
+
+    if (ret < call_count) {
+        op_ret = -ENOTCONN;
+        goto out;
+    }
+
+out:
+    /* Cleanup */
+    if (xattr) {
+        for (i = 0; i < ec->nodes; i++) {
+            if (xattr[i])
+                dict_unref(xattr[i]);
+        }
+        GF_FREE(xattr);
+    }
+    cluster_replies_wipe(replies, ec->nodes);
+    loc_wipe(&loc);
+    return op_ret;
+}
+
+int
+ec_heal_metadata_find_direction(ec_t *ec, default_args_cbk_t *replies,
+                                uint64_t *versions, uint64_t *dirty,
+                                unsigned char *sources,
+                                unsigned char *healed_sinks)
+{
+    uint64_t xattr[EC_VERSION_SIZE] = {0};
+    uint64_t max_version = 0;
+    int same_count = 0;
+    int max_same_count = 0;
+    int same_source = -1;
+    int ret = 0;
+    int i = 0;
+    int j = 0;
+    int *groups = NULL;
+    struct iatt source_ia = {0};
+    struct iatt child_ia = {0};
+
+    groups = alloca0(ec->nodes * sizeof(*groups));
+    for (i = 0; i < ec->nodes; i++)
+        groups[i] = -1;
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (!replies[i].valid)
+            continue;
+        if (replies[i].op_ret < 0)
+            continue;
+        ret = ec_dict_get_array(replies[i].xdata, EC_XATTR_VERSION, xattr,
+                                EC_VERSION_SIZE);
+        if (ret == 0) {
+            versions[i] = xattr[EC_METADATA_TXN];
+        }
+
+        memset(xattr, 0, sizeof(xattr));
+        ret = ec_dict_get_array(replies[i].xdata, EC_XATTR_DIRTY, xattr,
+                                EC_VERSION_SIZE);
+        if (ret == 0) {
+            dirty[i] = xattr[EC_METADATA_TXN];
+        }
+        if (groups[i] >= 0) /*Already part of group*/
+            continue;
+        groups[i] = i;
+        same_count = 1;
+        source_ia = replies[i].stat;
+        for (j = i + 1; j < ec->nodes; j++) {
+            if (!replies[j].valid || replies[j].op_ret < 0)
+                continue;
+            child_ia = replies[j].stat;
+            if (!IA_EQUAL(source_ia, child_ia, gfid) ||
+                !IA_EQUAL(source_ia, child_ia, type) ||
+                !IA_EQUAL(source_ia, child_ia, prot) ||
+                !IA_EQUAL(source_ia, child_ia, uid) ||
+                !IA_EQUAL(source_ia, child_ia, gid))
+                continue;
+            if (!are_dicts_equal(replies[i].xdata, replies[j].xdata,
+                                 ec_sh_key_match, NULL))
+                continue;
+            groups[j] = i;
+            same_count++;
+        }
+
+        if (max_same_count < same_count) {
+            max_same_count = same_count;
+            same_source = i;
+        }
+    }
+
+    if (max_same_count < ec->fragments) {
+        ret = -EIO;
+        goto out;
+    }
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (groups[i] == groups[same_source])
+            sources[i] = 1;
+        else if (replies[i].valid && replies[i].op_ret >= 0)
+            healed_sinks[i] = 1;
+    }
+    for (i = 0; i < ec->nodes; i++) {
+        if (sources[i] && (versions[i] > max_version)) {
+            same_source = i;
+            max_version = versions[i];
+        }
+    }
+    ret = same_source;
+out:
+    return ret;
+}
+
+int
+__ec_heal_metadata_prepare(call_frame_t *frame, ec_t *ec, inode_t *inode,
+                           unsigned char *locked_on,
+                           default_args_cbk_t *replies, uint64_t *versions,
+                           uint64_t *dirty, unsigned char *sources,
+                           unsigned char *healed_sinks)
+{
+    loc_t loc = {0};
+    unsigned char *output = NULL;
+    unsigned char *lookup_on = NULL;
+    int ret = 0;
+    int source = 0;
+    default_args_cbk_t *greplies = NULL;
+    int i = 0;
+    EC_REPLIES_ALLOC(greplies, ec->nodes);
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+    output = alloca0(ec->nodes);
+    lookup_on = alloca0(ec->nodes);
+    ret = cluster_lookup(ec->xl_list, locked_on, ec->nodes, replies, output,
+                         frame, ec->xl, &loc, NULL);
+    if (ret <= ec->fragments) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+
+    memcpy(lookup_on, output, ec->nodes);
+    /*Use getxattr to get the filtered xattrs which filter internal xattrs*/
+    ret = cluster_getxattr(ec->xl_list, lookup_on, ec->nodes, greplies, output,
+                           frame, ec->xl, &loc, NULL, NULL);
+    for (i = 0; i < ec->nodes; i++) {
+        if (lookup_on[i] && !output[i]) {
+            replies[i].valid = 0;
+            continue;
+        }
+        if (replies[i].xdata) {
+            dict_unref(replies[i].xdata);
+            replies[i].xdata = NULL;
+            if (greplies[i].xattr)
+                replies[i].xdata = dict_ref(greplies[i].xattr);
+        }
+    }
+
+    source = ec_heal_metadata_find_direction(ec, replies, versions, dirty,
+                                             sources, healed_sinks);
+    if (source < 0) {
+        ret = -EIO;
+        goto out;
+    }
+    ret = source;
+out:
+    cluster_replies_wipe(greplies, ec->nodes);
+    loc_wipe(&loc);
+    return ret;
+}
+
+/* Metadata heal */
+int
+__ec_removexattr_sinks(call_frame_t *frame, ec_t *ec, inode_t *inode,
+                       int source, unsigned char *sources,
+                       unsigned char *healed_sinks, default_args_cbk_t *replies)
+{
+    int i = 0;
+    int ret = 0;
+    loc_t loc = {0};
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (i == source)
+            continue;
+        if (!sources[i] && !healed_sinks[i])
+            continue;
+        ret = dict_foreach(replies[i].xdata, ec_heal_xattr_clean,
+                           replies[source].xdata);
+        if (ret < 0) {
+            sources[i] = 0;
+            healed_sinks[i] = 0;
+            continue;
+        }
+
+        if (replies[i].xdata->count == 0) {
+            continue;
+        } else if (sources[i]) {
+            /* This can happen if setxattr/removexattr succeeds on
+             * the bricks but fails to update the version. This
+             * will make sure that the xattrs are made equal after
+             * heal*/
+            sources[i] = 0;
+            healed_sinks[i] = 1;
+        }
+
+        ret = syncop_removexattr(ec->xl_list[i], &loc, "", replies[i].xdata,
+                                 NULL);
+        if (ret < 0)
+            healed_sinks[i] = 0;
+    }
+
+    loc_wipe(&loc);
+    if (EC_COUNT(healed_sinks, ec->nodes) == 0)
+        return -ENOTCONN;
+    return 0;
+}
+
+int
+__ec_heal_metadata(call_frame_t *frame, ec_t *ec, inode_t *inode,
+                   unsigned char *locked_on, unsigned char *sources,
+                   unsigned char *healed_sinks)
+{
+    loc_t loc = {0};
+    int ret = 0;
+    int source = 0;
+    default_args_cbk_t *replies = NULL;
+    default_args_cbk_t *sreplies = NULL;
+    uint64_t *versions = NULL;
+    uint64_t *dirty = NULL;
+    unsigned char *output = NULL;
+    dict_t *source_dict = NULL;
+    struct iatt source_buf = {0};
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    EC_REPLIES_ALLOC(sreplies, ec->nodes);
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+    output = alloca0(ec->nodes);
+    versions = alloca0(ec->nodes * sizeof(*versions));
+    dirty = alloca0(ec->nodes * sizeof(*dirty));
+    source = __ec_heal_metadata_prepare(frame, ec, inode, locked_on, replies,
+                                        versions, dirty, sources, healed_sinks);
+    if (source < 0) {
+        ret = -EIO;
+        goto out;
+    }
+
+    if ((EC_COUNT(sources, ec->nodes) == ec->nodes) ||
+        (EC_COUNT(healed_sinks, ec->nodes) == 0)) {
+        ret = 0;
+        goto erase_dirty;
+    }
+
+    source_buf = replies[source].stat;
+    ret = cluster_setattr(ec->xl_list, healed_sinks, ec->nodes, sreplies,
+                          output, frame, ec->xl, &loc, &source_buf,
+                          GF_SET_ATTR_MODE | GF_SET_ATTR_UID | GF_SET_ATTR_GID,
+                          NULL);
+    /*In case the operation fails on some of the subvols*/
+    memcpy(healed_sinks, output, ec->nodes);
+    if (EC_COUNT(healed_sinks, ec->nodes) == 0) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+
+    ret = __ec_removexattr_sinks(frame, ec, inode, source, sources,
+                                 healed_sinks, replies);
+    if (ret < 0)
+        goto out;
+
+    source_dict = dict_ref(replies[source].xdata);
+    if (dict_foreach_match(source_dict, ec_ignorable_key_match, NULL,
+                           dict_remove_foreach_fn, NULL) == -1) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ret = cluster_setxattr(ec->xl_list, healed_sinks, ec->nodes, replies,
+                           output, frame, ec->xl, &loc, source_dict, 0, NULL);
+
+    EC_INTERSECT(healed_sinks, healed_sinks, output, ec->nodes);
+    if (EC_COUNT(healed_sinks, ec->nodes) == 0) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+
+erase_dirty:
+    ret = ec_adjust_versions(frame, ec, EC_METADATA_TXN, inode, source, sources,
+                             healed_sinks, versions, dirty);
+out:
+    if (source_dict)
+        dict_unref(source_dict);
+
+    loc_wipe(&loc);
+    cluster_replies_wipe(replies, ec->nodes);
+    cluster_replies_wipe(sreplies, ec->nodes);
+    return ret;
+}
+
+int
+ec_heal_metadata(call_frame_t *frame, ec_t *ec, inode_t *inode,
+                 unsigned char *sources, unsigned char *healed_sinks)
+{
+    unsigned char *locked_on = NULL;
+    unsigned char *up_subvols = NULL;
+    unsigned char *output = NULL;
+    int ret = 0;
+    default_args_cbk_t *replies = NULL;
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    locked_on = alloca0(ec->nodes);
+    output = alloca0(ec->nodes);
+    up_subvols = alloca0(ec->nodes);
+    ec_mask_to_char_array(ec->xl_up, up_subvols, ec->nodes);
+    ret = cluster_inodelk(ec->xl_list, up_subvols, ec->nodes, replies,
+                          locked_on, frame, ec->xl, ec->xl->name, inode, 0, 0);
+    {
+        if (ret <= ec->fragments) {
+            gf_msg_debug(ec->xl->name, 0,
+                         "%s: Skipping heal "
+                         "as only %d number of subvolumes could "
+                         "be locked",
+                         uuid_utoa(inode->gfid), ret);
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+        ret = __ec_heal_metadata(frame, ec, inode, locked_on, sources,
+                                 healed_sinks);
+    }
+unlock:
+    cluster_uninodelk(ec->xl_list, locked_on, ec->nodes, replies, output, frame,
+                      ec->xl, ec->xl->name, inode, 0, 0);
+    cluster_replies_wipe(replies, ec->nodes);
+    return ret;
+}
+
+/*entry heal*/
+int
+__ec_heal_entry_prepare(call_frame_t *frame, ec_t *ec, inode_t *inode,
+                        unsigned char *locked_on, uint64_t *versions,
+                        uint64_t *dirty, unsigned char *sources,
+                        unsigned char *healed_sinks)
+{
+    loc_t loc = {0};
+    int source = 0;
+    int ret = 0;
+    default_args_cbk_t *replies = NULL;
+    unsigned char *output = NULL;
+    dict_t *xdata = NULL;
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+    xdata = dict_new();
+    if (!xdata) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    if (dict_set_uint64(xdata, EC_XATTR_VERSION, 0) ||
+        dict_set_uint64(xdata, EC_XATTR_DIRTY, 0)) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    output = alloca0(ec->nodes);
+    ret = cluster_lookup(ec->xl_list, locked_on, ec->nodes, replies, output,
+                         frame, ec->xl, &loc, xdata);
+    if (ret <= ec->fragments) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+
+    source = ec_heal_entry_find_direction(ec, replies, versions, dirty, sources,
+                                          healed_sinks);
+    if (source < 0) {
+        ret = -EIO;
+        goto out;
+    }
+    ret = source;
+out:
+    if (xdata)
+        dict_unref(xdata);
+    loc_wipe(&loc);
+    cluster_replies_wipe(replies, ec->nodes);
+    return ret;
+}
+int32_t
+ec_set_new_entry_dirty(ec_t *ec, loc_t *loc, struct iatt *ia,
+                       call_frame_t *frame, xlator_t *this, unsigned char *on)
+{
+    dict_t *xattr = NULL;
+    int32_t ret = -1;
+    default_args_cbk_t *replies = NULL;
+    unsigned char *output = NULL;
+    uint64_t dirty[EC_VERSION_SIZE] = {1, 1};
+    loc_t newloc = {0};
+
+    /*Symlinks don't have any data to be healed*/
+    if (ia->ia_type == IA_IFLNK)
+        dirty[EC_DATA_TXN] = 0;
+
+    newloc.inode = inode_ref(loc->inode);
+    gf_uuid_copy(newloc.gfid, ia->ia_gfid);
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    output = alloca0(ec->nodes);
+    xattr = dict_new();
+    if (!xattr) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ret = ec_dict_set_array(xattr, EC_XATTR_DIRTY, dirty, EC_VERSION_SIZE);
+    if (ret)
+        goto out;
+
+    ret = cluster_xattrop(ec->xl_list, on, ec->nodes, replies, output, frame,
+                          ec->xl, &newloc, GF_XATTROP_ADD_ARRAY64, xattr, NULL);
+
+    if (ret < ec->fragments) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+
+out:
+    if (xattr)
+        dict_unref(xattr);
+    cluster_replies_wipe(replies, ec->nodes);
+    loc_wipe(&newloc);
+    return ret;
+}
+
+/*Name heal*/
+int
+ec_delete_stale_name(dict_t *gfid_db, char *key, data_t *d, void *data)
+{
+    struct ec_name_data *name_data = data;
+    struct iatt *ia = NULL;
+    ec_t *ec = NULL;
+    loc_t loc = {0};
+    unsigned char *same = data_to_bin(d);
+    default_args_cbk_t *replies = NULL;
+    unsigned char *output = NULL;
+    int ret = 0;
+    int estale_count = 0;
+    int i = 0;
+    call_frame_t *frame = name_data->frame;
+    uuid_t gfid;
+
+    ec = name_data->frame->this->private;
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    if (EC_COUNT(same, ec->nodes) >= ec->fragments) {
+        ret = 0;
+        goto out;
+    }
+
+    loc.parent = inode_ref(name_data->parent);
+    loc.inode = inode_new(name_data->parent->table);
+    if (!loc.inode) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    gf_uuid_parse(key, gfid);
+    gf_uuid_copy(loc.pargfid, name_data->parent->gfid);
+    loc.name = name_data->name;
+    output = alloca0(ec->nodes);
+    ret = cluster_lookup(ec->xl_list, name_data->participants, ec->nodes,
+                         replies, output, name_data->frame, ec->xl, &loc, NULL);
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (!replies[i].valid)
+            continue;
+        if (replies[i].op_ret == -1) {
+            if (replies[i].op_errno == ESTALE || replies[i].op_errno == ENOENT)
+                estale_count++;
+            else
+                name_data->participants[i] = 0;
+        } else if (gf_uuid_compare(gfid, replies[i].stat.ia_gfid)) {
+            estale_count++;
+            gf_msg_debug(ec->xl->name, 0, "%s/%s: different gfid as %s",
+                         uuid_utoa(name_data->parent->gfid), name_data->name,
+                         key);
+        }
+    }
+
+    if (estale_count <= ec->redundancy) {
+        /* We have at least ec->fragments number of fragments, so the
+         * file is recoverable, so don't delete it*/
+
+        /* Please note that the lookup call above could fail with
+         * ENOTCONN on all subvoumes and still this branch will be
+         * true, but in those cases conservatively we decide to not
+         * delete the file until we are sure*/
+        ret = 0;
+        goto out;
+    }
+
+    /*Noway to recover, delete the name*/
+    loc_wipe(&loc);
+    loc.parent = inode_ref(name_data->parent);
+    gf_uuid_copy(loc.pargfid, loc.parent->gfid);
+    loc.name = name_data->name;
+    for (i = 0; i < ec->nodes; i++) {
+        if (same[i] && replies[i].valid && (replies[i].op_ret == 0)) {
+            ia = &replies[i].stat;
+            break;
+        }
+    }
+
+    if (!ia) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+
+    if (IA_ISDIR(ia->ia_type)) {
+        ret = cluster_rmdir(ec->xl_list, same, ec->nodes, replies, output,
+                            frame, ec->xl, &loc, 1, NULL);
+        gf_msg_debug(ec->xl->name, 0,
+                     "cluster rmdir succeeded on %d "
+                     "nodes",
+                     ret);
+    } else {
+        ret = cluster_unlink(ec->xl_list, same, ec->nodes, replies, output,
+                             frame, ec->xl, &loc, 0, NULL);
+        gf_msg_debug(ec->xl->name, 0,
+                     "cluster unlink succeeded on %d "
+                     "nodes",
+                     ret);
+    }
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (output[i]) {
+            same[i] = 0;
+            name_data->enoent[i] = 1;
+        } else {
+            /*op failed*/
+            if (same[i])
+                name_data->participants[i] = 0;
+        }
+    }
+    ret = 0;
+    /*This will help in making decisions about creating names*/
+    dict_del(gfid_db, key);
+out:
+    if (ret < 0) {
+        gf_msg_debug(ec->xl->name, 0, "%s/%s: heal failed %s",
+                     uuid_utoa(name_data->parent->gfid), name_data->name,
+                     strerror(-ret));
+    }
+    cluster_replies_wipe(replies, ec->nodes);
+    loc_wipe(&loc);
+    return ret;
+}
+
+int
+ec_delete_stale_names(call_frame_t *frame, ec_t *ec, inode_t *parent,
+                      char *name, default_args_cbk_t *replies, dict_t *gfid_db,
+                      unsigned char *enoent, unsigned char *gfidless,
+                      unsigned char *participants)
+{
+    struct ec_name_data name_data = {0};
+
+    name_data.enoent = enoent;
+    name_data.gfidless = gfidless;
+    name_data.participants = participants;
+    name_data.name = name;
+    name_data.parent = parent;
+    name_data.frame = frame;
+    name_data.replies = replies;
+    return dict_foreach(gfid_db, ec_delete_stale_name, &name_data);
+}
+
+int
+_assign_same(dict_t *dict, char *key, data_t *value, void *data)
+{
+    struct ec_name_data *name_data = data;
+
+    name_data->same = data_to_bin(value);
+    return 0;
+}
+
+int
+ec_create_name(call_frame_t *frame, ec_t *ec, inode_t *parent, char *name,
+               default_args_cbk_t *lookup_replies, dict_t *gfid_db,
+               unsigned char *enoent, unsigned char *participants)
+{
+    int ret = 0;
+    int i = 0;
+    struct ec_name_data name_data = {0};
+    struct iatt *ia = NULL;
+    unsigned char *output = 0;
+    unsigned char *output1 = 0;
+    unsigned char *on = NULL;
+    default_args_cbk_t *replies = NULL;
+    loc_t loc = {0};
+    loc_t srcloc = {0};
+    unsigned char *link = NULL;
+    unsigned char *create = NULL;
+    dict_t *xdata = NULL;
+    char *linkname = NULL;
+    ec_config_t config;
+
+    /* There should be just one gfid key */
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    if (gfid_db->count != 1) {
+        ret = -EINVAL;
+        goto out;
+    }
+
+    ret = dict_foreach(gfid_db, _assign_same, &name_data);
+    if (ret < 0)
+        goto out;
+    /*There should at least be one valid success reply with gfid*/
+    for (i = 0; i < ec->nodes; i++)
+        if (name_data.same[i])
+            break;
+
+    if (i == ec->nodes) {
+        ret = -EINVAL;
+        goto out;
+    }
+
+    ia = &lookup_replies[i].stat;
+    xdata = dict_new();
+    loc.parent = inode_ref(parent);
+    gf_uuid_copy(loc.pargfid, parent->gfid);
+    loc.inode = inode_new(parent->table);
+    if (loc.inode)
+        srcloc.inode = inode_ref(loc.inode);
+    gf_uuid_copy(srcloc.gfid, ia->ia_gfid);
+    if (!loc.inode || !xdata ||
+        dict_set_static_bin(xdata, "gfid-req", ia->ia_gfid,
+                            sizeof(ia->ia_gfid))) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    loc.name = name;
+    link = alloca0(ec->nodes);
+    create = alloca0(ec->nodes);
+    on = alloca0(ec->nodes);
+    output = alloca0(ec->nodes);
+    output1 = alloca0(ec->nodes);
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (!lookup_replies[i].valid)
+            continue;
+        if (lookup_replies[i].op_ret)
+            continue;
+        on[i] = 1;
+    }
+    switch (ia->ia_type) {
+        case IA_IFDIR:
+            ec_set_new_entry_dirty(ec, &loc, ia, frame, ec->xl, on);
+            (void)cluster_mkdir(
+                ec->xl_list, enoent, ec->nodes, replies, output, frame, ec->xl,
+                &loc, st_mode_from_ia(ia->ia_prot, ia->ia_type), 0, xdata);
+            break;
+
+        case IA_IFLNK:
+            /*Check for hard links and create/link*/
+            ret = cluster_lookup(ec->xl_list, enoent, ec->nodes, replies,
+                                 output, frame, ec->xl, &srcloc, NULL);
+            for (i = 0; i < ec->nodes; i++) {
+                if (output[i]) {
+                    link[i] = 1;
+                } else {
+                    if (replies[i].op_errno == ENOENT ||
+                        replies[i].op_errno == ESTALE) {
+                        create[i] = 1;
+                    }
+                }
+            }
+
+            if (EC_COUNT(link, ec->nodes)) {
+                cluster_link(ec->xl_list, link, ec->nodes, replies, output1,
+                             frame, ec->xl, &srcloc, &loc, NULL);
+            }
+
+            if (EC_COUNT(create, ec->nodes)) {
+                cluster_readlink(ec->xl_list, name_data.same, ec->nodes,
+                                 replies, output, frame, ec->xl, &srcloc, 4096,
+                                 NULL);
+                if (EC_COUNT(output, ec->nodes) == 0) {
+                    ret = -ENOTCONN;
+                    goto out;
+                }
+
+                for (i = 0; i < ec->nodes; i++) {
+                    if (output[i])
+                        break;
+                }
+                linkname = alloca0(strlen(replies[i].buf) + 1);
+                strcpy(linkname, replies[i].buf);
+                ec_set_new_entry_dirty(ec, &loc, ia, frame, ec->xl, on);
+                cluster_symlink(ec->xl_list, create, ec->nodes, replies, output,
+                                frame, ec->xl, linkname, &loc, 0, xdata);
+            }
+            for (i = 0; i < ec->nodes; i++)
+                if (output1[i])
+                    output[i] = 1;
+            break;
+        case IA_IFREG:
+            ec_set_new_entry_dirty(ec, &loc, ia, frame, ec->xl, on);
+            config.version = EC_CONFIG_VERSION;
+            config.algorithm = EC_CONFIG_ALGORITHM;
+            config.gf_word_size = EC_GF_BITS;
+            config.bricks = ec->nodes;
+            config.redundancy = ec->redundancy;
+            config.chunk_size = EC_METHOD_CHUNK_SIZE;
+
+            ret = ec_dict_set_config(xdata, EC_XATTR_CONFIG, &config);
+            if (ret != 0) {
+                goto out;
+            }
+
+            /* Fall through */
+
+        default:
+            ret = dict_set_int32(xdata, GLUSTERFS_INTERNAL_FOP_KEY, 1);
+            if (ret)
+                goto out;
+            ret = cluster_mknod(
+                ec->xl_list, enoent, ec->nodes, replies, output, frame, ec->xl,
+                &loc, st_mode_from_ia(ia->ia_prot, ia->ia_type),
+                makedev(ia_major(ia->ia_rdev), ia_minor(ia->ia_rdev)), 0,
+                xdata);
+            break;
+    }
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (enoent[i] && !output[i])
+            participants[i] = 0;
+    }
+
+    ret = 0;
+out:
+    if (ret < 0)
+        gf_msg_debug(ec->xl->name, 0, "%s/%s: heal failed %s",
+                     uuid_utoa(parent->gfid), name, strerror(-ret));
+    cluster_replies_wipe(replies, ec->nodes);
+    loc_wipe(&loc);
+    loc_wipe(&srcloc);
+    if (xdata)
+        dict_unref(xdata);
+    return ret;
+}
+
+int
+__ec_heal_name(call_frame_t *frame, ec_t *ec, inode_t *parent, char *name,
+               unsigned char *participants)
+{
+    unsigned char *output = NULL;
+    unsigned char *enoent = NULL;
+    default_args_cbk_t *replies = NULL;
+    dict_t *xdata = NULL;
+    dict_t *gfid_db = NULL;
+    int ret = 0;
+    loc_t loc = {0};
+    int i = 0;
+    struct iatt *ia = NULL;
+    char gfid[64] = {0};
+    unsigned char *same = NULL;
+    unsigned char *gfidless = NULL;
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    loc.parent = inode_ref(parent);
+    loc.inode = inode_new(parent->table);
+    gf_uuid_copy(loc.pargfid, parent->gfid);
+    loc.name = name;
+    xdata = dict_new();
+    gfid_db = dict_new();
+    if (!xdata || !gfid_db || !loc.inode) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ret = dict_set_int32(xdata, GF_GFIDLESS_LOOKUP, 1);
+    if (ret) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    output = alloca0(ec->nodes);
+    gfidless = alloca0(ec->nodes);
+    enoent = alloca0(ec->nodes);
+    ret = cluster_lookup(ec->xl_list, participants, ec->nodes, replies, output,
+                         frame, ec->xl, &loc, NULL);
+    for (i = 0; i < ec->nodes; i++) {
+        if (!replies[i].valid)
+            continue;
+
+        if (replies[i].op_ret == -1) {
+            /*If ESTALE comes here, that means parent dir is not
+             * present, nothing to do there, so reset participants
+             * for that brick*/
+            if (replies[i].op_errno == ENOENT)
+                enoent[i] = 1;
+            else
+                participants[i] = 0;
+            continue;
+        }
+        ia = &replies[i].stat;
+        if (gf_uuid_is_null(ia->ia_gfid)) {
+            if (IA_ISDIR(ia->ia_type) || ia->ia_size == 0)
+                gfidless[i] = 1;
+            else
+                participants[i] = 0;
+        } else {
+            uuid_utoa_r(ia->ia_gfid, gfid);
+            ret = dict_get_bin(gfid_db, gfid, (void **)&same);
+            if (ret < 0) {
+                same = alloca0(ec->nodes);
+            }
+            same[i] = 1;
+            if (ret < 0) {
+                ret = dict_set_static_bin(gfid_db, gfid, same, ec->nodes);
+            }
+            if (ret < 0)
+                goto out;
+        }
+    }
+
+    ret = ec_delete_stale_names(frame, ec, parent, name, replies, gfid_db,
+                                enoent, gfidless, participants);
+
+    if (gfid_db->count == 0) {
+        /* All entries seem to be stale entries and deleted,
+         * nothing more to do.*/
+        goto out;
+    }
+
+    if (gfid_db->count > 1) {
+        gf_msg(ec->xl->name, GF_LOG_INFO, 0, EC_MSG_HEAL_FAIL,
+               "%s/%s: Not able to heal", uuid_utoa(parent->gfid), name);
+        memset(participants, 0, ec->nodes);
+        goto out;
+    }
+
+    EC_INTERSECT(enoent, enoent, participants, ec->nodes);
+    if (EC_COUNT(enoent, ec->nodes) == 0) {
+        ret = 0;
+        goto out;
+    }
+
+    ret = ec_create_name(frame, ec, parent, name, replies, gfid_db, enoent,
+                         participants);
+    if (ret >= 0) {
+        /* If ec_create_name() succeeded we return 1 to indicate that a new
+         * file has been created and it will need to be healed. */
+        ret = 1;
+    }
+out:
+    cluster_replies_wipe(replies, ec->nodes);
+    loc_wipe(&loc);
+    if (xdata)
+        dict_unref(xdata);
+    if (gfid_db)
+        dict_unref(gfid_db);
+    return ret;
+}
+
+int
+ec_heal_name(call_frame_t *frame, ec_t *ec, inode_t *parent, char *name,
+             unsigned char *participants)
+{
+    int ret = 0;
+    default_args_cbk_t *replies = NULL;
+    unsigned char *output = NULL;
+    unsigned char *locked_on = NULL;
+    loc_t loc = {0};
+
+    loc.parent = inode_ref(parent);
+    loc.name = name;
+    loc.inode = inode_new(parent->table);
+    if (!loc.inode) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    output = alloca0(ec->nodes);
+    locked_on = alloca0(ec->nodes);
+    ret = cluster_inodelk(ec->xl_list, participants, ec->nodes, replies,
+                          locked_on, frame, ec->xl, ec->xl->name, parent, 0, 0);
+    {
+        if (ret <= ec->fragments) {
+            gf_msg_debug(ec->xl->name, 0,
+                         "%s/%s: Skipping "
+                         "heal as only %d number of subvolumes could "
+                         "be locked",
+                         uuid_utoa(parent->gfid), name, ret);
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+        EC_INTERSECT(participants, participants, locked_on, ec->nodes);
+        ret = __ec_heal_name(frame, ec, parent, name, participants);
+    }
+unlock:
+    cluster_uninodelk(ec->xl_list, locked_on, ec->nodes, replies, output, frame,
+                      ec->xl, ec->xl->name, parent, 0, 0);
+out:
+    cluster_replies_wipe(replies, ec->nodes);
+    loc_wipe(&loc);
+    return ret;
+}
+
+int
+ec_name_heal_handler(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent,
+                     void *data)
+{
+    struct ec_name_data *name_data = data;
+    xlator_t *this = THIS;
+    ec_t *ec = this->private;
+    unsigned char *name_on = alloca0(ec->nodes);
+    int i = 0;
+    int ret = 0;
+
+    if (ec->shutdown) {
+        gf_msg_debug(this->name, 0,
+                     "Cancelling directory heal "
+                     "because EC is stopping.");
+        return -ENOTCONN;
+    }
+
+    memcpy(name_on, name_data->participants, ec->nodes);
+    ret = ec_heal_name(name_data->frame, ec, parent->inode, entry->d_name,
+                       name_on);
+
+    if (ret < 0) {
+        memset(name_on, 0, ec->nodes);
+    } else {
+        name_data->heal_pending += ret;
+    }
+
+    for (i = 0; i < ec->nodes; i++)
+        if (name_data->participants[i] && !name_on[i])
+            name_data->failed_on[i] = 1;
+
+    return 0;
+}
+
+int
+ec_heal_names(call_frame_t *frame, ec_t *ec, inode_t *inode,
+              unsigned char *participants, uint32_t *pending)
+{
+    int i = 0;
+    int j = 0;
+    loc_t loc = {0};
+    struct ec_name_data name_data = {0};
+    int ret = 0;
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+    name_data.frame = frame;
+    name_data.participants = participants;
+    name_data.failed_on = alloca0(ec->nodes);
+    name_data.heal_pending = 0;
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (!participants[i])
+            continue;
+        ret = syncop_dir_scan(ec->xl_list[i], &loc, GF_CLIENT_PID_SELF_HEALD,
+                              &name_data, ec_name_heal_handler);
+        if (ret < 0) {
+            break;
+        }
+        for (j = 0; j < ec->nodes; j++)
+            if (name_data.failed_on[j])
+                participants[j] = 0;
+
+        if (EC_COUNT(participants, ec->nodes) <= ec->fragments) {
+            ret = -ENOTCONN;
+            break;
+        }
+    }
+    *pending += name_data.heal_pending;
+
+    loc_wipe(&loc);
+    return ret;
+}
+
+int
+__ec_heal_entry(call_frame_t *frame, ec_t *ec, inode_t *inode,
+                unsigned char *heal_on, unsigned char *sources,
+                unsigned char *healed_sinks, uint32_t *pending)
+{
+    unsigned char *locked_on = NULL;
+    unsigned char *output = NULL;
+    uint64_t *versions = NULL;
+    uint64_t *dirty = NULL;
+    unsigned char *participants = NULL;
+    default_args_cbk_t *replies = NULL;
+    int ret = 0;
+    int source = 0;
+    int i = 0;
+
+    locked_on = alloca0(ec->nodes);
+    output = alloca0(ec->nodes);
+    versions = alloca0(ec->nodes * sizeof(*versions));
+    dirty = alloca0(ec->nodes * sizeof(*dirty));
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    ret = cluster_inodelk(ec->xl_list, heal_on, ec->nodes, replies, locked_on,
+                          frame, ec->xl, ec->xl->name, inode, 0, 0);
+    {
+        if (ret <= ec->fragments) {
+            gf_msg_debug(ec->xl->name, 0,
+                         "%s: Skipping heal "
+                         "as only %d number of subvolumes could "
+                         "be locked",
+                         uuid_utoa(inode->gfid), ret);
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+        ret = __ec_heal_entry_prepare(frame, ec, inode, locked_on, versions,
+                                      dirty, sources, healed_sinks);
+        source = ret;
+    }
+unlock:
+    cluster_uninodelk(ec->xl_list, locked_on, ec->nodes, replies, output, frame,
+                      ec->xl, ec->xl->name, inode, 0, 0);
+    if (ret < 0)
+        goto out;
+
+    participants = alloca0(ec->nodes);
+    for (i = 0; i < ec->nodes; i++) {
+        if (sources[i] || healed_sinks[i])
+            participants[i] = 1;
+    }
+    ret = ec_heal_names(frame, ec, inode, participants, pending);
+
+    if (EC_COUNT(participants, ec->nodes) <= ec->fragments)
+        goto out;
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (!participants[i]) {
+            sources[i] = 0;
+            healed_sinks[i] = 0;
+        }
+    }
+
+    ec_adjust_versions(frame, ec, EC_DATA_TXN, inode, source, sources,
+                       healed_sinks, versions, dirty);
+out:
+    cluster_replies_wipe(replies, ec->nodes);
+    return ret;
+}
+
+int
+ec_heal_entry(call_frame_t *frame, ec_t *ec, inode_t *inode,
+              unsigned char *sources, unsigned char *healed_sinks,
+              uint32_t *pending)
+{
+    unsigned char *locked_on = NULL;
+    unsigned char *up_subvols = NULL;
+    unsigned char *output = NULL;
+    char selfheal_domain[1024] = {0};
+    int ret = 0;
+    default_args_cbk_t *replies = NULL;
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    locked_on = alloca0(ec->nodes);
+    output = alloca0(ec->nodes);
+    up_subvols = alloca0(ec->nodes);
+
+    sprintf(selfheal_domain, "%s:self-heal", ec->xl->name);
+    ec_mask_to_char_array(ec->xl_up, up_subvols, ec->nodes);
+    /*If other processes are already doing the heal, don't block*/
+    ret = cluster_tiebreaker_inodelk(ec->xl_list, up_subvols, ec->nodes,
+                                     replies, locked_on, frame, ec->xl,
+                                     selfheal_domain, inode, 0, 0);
+    {
+        if (ret <= ec->fragments) {
+            gf_msg_debug(ec->xl->name, 0,
+                         "%s: Skipping heal "
+                         "as only %d number of subvolumes could "
+                         "be locked",
+                         uuid_utoa(inode->gfid), ret);
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+        ret = __ec_heal_entry(frame, ec, inode, locked_on, sources,
+                              healed_sinks, pending);
+    }
+unlock:
+    cluster_uninodelk(ec->xl_list, locked_on, ec->nodes, replies, output, frame,
+                      ec->xl, selfheal_domain, inode, 0, 0);
+    cluster_replies_wipe(replies, ec->nodes);
+    return ret;
+}
+
+/*Find direction for data heal and heal info*/
+int
+ec_heal_data_find_direction(ec_t *ec, default_args_cbk_t *replies,
+                            uint64_t *data_versions, uint64_t *dirty,
+                            uint64_t *size, unsigned char *sources,
+                            unsigned char *healed_sinks,
+                            gf_boolean_t check_ondisksize, int which)
+{
+    uint64_t xattr[EC_VERSION_SIZE] = {0};
+    char version_size[128] = {0};
+    dict_t *version_size_db = NULL;
+    unsigned char *same = NULL;
+    int max_same_count = 0;
+    int source = 0;
+    int i = 0;
+    int ret = 0;
+    dict_t *dict = NULL;
+    uint64_t source_size = 0;
+
+    version_size_db = dict_new();
+    if (!version_size_db) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (!replies[i].valid)
+            continue;
+        if (replies[i].op_ret < 0)
+            continue;
+        dict = (which == EC_COMBINE_XDATA) ? replies[i].xdata
+                                           : replies[i].xattr;
+
+        ret = ec_dict_get_array(dict, EC_XATTR_VERSION, xattr, EC_VERSION_SIZE);
+        if (ret == 0) {
+            data_versions[i] = xattr[EC_DATA_TXN];
+        }
+
+        memset(xattr, 0, sizeof(xattr));
+        ret = ec_dict_get_array(dict, EC_XATTR_DIRTY, xattr, EC_VERSION_SIZE);
+        if (ret == 0) {
+            dirty[i] = xattr[EC_DATA_TXN];
+        }
+        ret = ec_dict_del_number(dict, EC_XATTR_SIZE, &size[i]);
+        /*Build a db of same metadata and data version and size*/
+        snprintf(version_size, sizeof(version_size), "%" PRIu64 "-%" PRIu64,
+                 data_versions[i], size[i]);
+
+        ret = dict_get_bin(version_size_db, version_size, (void **)&same);
+        if (ret < 0) {
+            same = alloca0(ec->nodes);
+        }
+
+        same[i] = 1;
+        if (max_same_count < EC_COUNT(same, ec->nodes)) {
+            max_same_count = EC_COUNT(same, ec->nodes);
+            source = i;
+        }
+
+        if (ret < 0) {
+            ret = dict_set_static_bin(version_size_db, version_size, same,
+                                      ec->nodes);
+        }
+
+        if (ret < 0) {
+            ret = -ENOMEM;
+            goto out;
+        }
+    }
+    /* If we don't have ec->fragments number of same version,size it is not
+     * recoverable*/
+    if (max_same_count < ec->fragments) {
+        ret = -EIO;
+        goto out;
+    } else {
+        snprintf(version_size, sizeof(version_size), "%" PRIu64 "-%" PRIu64,
+                 data_versions[source], size[source]);
+
+        ret = dict_get_bin(version_size_db, version_size, (void **)&same);
+        if (ret < 0)
+            goto out;
+        memcpy(sources, same, ec->nodes);
+        for (i = 0; i < ec->nodes; i++) {
+            if (replies[i].valid && (replies[i].op_ret == 0) && !sources[i])
+                healed_sinks[i] = 1;
+        }
+    }
+
+    /* There could be files with versions, size same but on disk ia_size
+     * could be different because of disk crashes, mark them as sinks as
+     * well*/
+
+    if (check_ondisksize) {
+        source_size = size[source];
+        ec_adjust_size_up(ec, &source_size, _gf_true);
+
+        for (i = 0; i < ec->nodes; i++) {
+            if (sources[i]) {
+                if (replies[i].stat.ia_size != source_size) {
+                    sources[i] = 0;
+                    healed_sinks[i] = 1;
+                    max_same_count--;
+                } else {
+                    source = i;
+                }
+            }
+        }
+        if (max_same_count < ec->fragments) {
+            ret = -EIO;
+            goto out;
+        }
+    }
+
+    ret = source;
+out:
+    if (version_size_db)
+        dict_unref(version_size_db);
+    return ret;
+}
+
+int
+__ec_heal_data_prepare(call_frame_t *frame, ec_t *ec, fd_t *fd,
+                       unsigned char *locked_on, uint64_t *versions,
+                       uint64_t *dirty, uint64_t *size, unsigned char *sources,
+                       unsigned char *healed_sinks, unsigned char *trim,
+                       struct iatt *stbuf)
+{
+    default_args_cbk_t *replies = NULL;
+    default_args_cbk_t *fstat_replies = NULL;
+    unsigned char *output = NULL;
+    unsigned char *fstat_output = NULL;
+    dict_t *xattrs = NULL;
+    uint64_t zero_array[2] = {0};
+    int source = 0;
+    int ret = 0;
+    uint64_t zero_value = 0;
+    int i = 0;
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    EC_REPLIES_ALLOC(fstat_replies, ec->nodes);
+    output = alloca0(ec->nodes);
+    fstat_output = alloca0(ec->nodes);
+    xattrs = dict_new();
+    if (!xattrs ||
+        dict_set_static_bin(xattrs, EC_XATTR_VERSION, zero_array,
+                            sizeof(zero_array)) ||
+        dict_set_static_bin(xattrs, EC_XATTR_DIRTY, zero_array,
+                            sizeof(zero_array)) ||
+        dict_set_static_bin(xattrs, EC_XATTR_SIZE, &zero_value,
+                            sizeof(zero_value))) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ret = cluster_fxattrop(ec->xl_list, locked_on, ec->nodes, replies, output,
+                           frame, ec->xl, fd, GF_XATTROP_ADD_ARRAY64, xattrs,
+                           NULL);
+
+    ret = cluster_fstat(ec->xl_list, locked_on, ec->nodes, fstat_replies,
+                        fstat_output, frame, ec->xl, fd, NULL);
+
+    for (i = 0; i < ec->nodes; i++) {
+        output[i] = output[i] && fstat_output[i];
+        replies[i].valid = output[i];
+        if (output[i])
+            replies[i].stat = fstat_replies[i].stat;
+    }
+
+    if (EC_COUNT(output, ec->nodes) <= ec->fragments) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+
+    source = ec_heal_data_find_direction(ec, replies, versions, dirty, size,
+                                         sources, healed_sinks, _gf_true,
+                                         EC_COMBINE_DICT);
+    ret = source;
+    if (ret < 0)
+        goto out;
+
+    if (stbuf)
+        *stbuf = replies[source].stat;
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (healed_sinks[i]) {
+            if (replies[i].stat.ia_size)
+                trim[i] = 1;
+        }
+    }
+
+    if (EC_COUNT(sources, ec->nodes) < ec->fragments) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+
+    ret = source;
+out:
+    if (xattrs)
+        dict_unref(xattrs);
+    cluster_replies_wipe(replies, ec->nodes);
+    cluster_replies_wipe(fstat_replies, ec->nodes);
+    if (ret < 0) {
+        gf_msg_debug(ec->xl->name, 0, "%s: heal failed %s",
+                     uuid_utoa(fd->inode->gfid), strerror(-ret));
+    } else {
+        gf_msg_debug(ec->xl->name, 0,
+                     "%s: sources: %d, sinks: "
+                     "%d",
+                     uuid_utoa(fd->inode->gfid), EC_COUNT(sources, ec->nodes),
+                     EC_COUNT(healed_sinks, ec->nodes));
+    }
+    return ret;
+}
+
+int
+__ec_heal_mark_sinks(call_frame_t *frame, ec_t *ec, fd_t *fd,
+                     uint64_t *versions, unsigned char *healed_sinks)
+{
+    int i = 0;
+    int ret = 0;
+    unsigned char *mark = NULL;
+    dict_t *xattrs = NULL;
+    default_args_cbk_t *replies = NULL;
+    unsigned char *output = NULL;
+    uint64_t versions_xattr[2] = {0};
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    xattrs = dict_new();
+    if (!xattrs) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    mark = alloca0(ec->nodes);
+    for (i = 0; i < ec->nodes; i++) {
+        if (!healed_sinks[i])
+            continue;
+        if ((versions[i] >> EC_SELFHEAL_BIT) & 1)
+            continue;
+        mark[i] = 1;
+    }
+
+    if (EC_COUNT(mark, ec->nodes) == 0)
+        return 0;
+
+    versions_xattr[EC_DATA_TXN] = hton64(1ULL << EC_SELFHEAL_BIT);
+    if (dict_set_static_bin(xattrs, EC_XATTR_VERSION, versions_xattr,
+                            sizeof(versions_xattr))) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    output = alloca0(ec->nodes);
+    ret = cluster_fxattrop(ec->xl_list, mark, ec->nodes, replies, output, frame,
+                           ec->xl, fd, GF_XATTROP_ADD_ARRAY64, xattrs, NULL);
+    for (i = 0; i < ec->nodes; i++) {
+        if (!output[i]) {
+            if (mark[i])
+                healed_sinks[i] = 0;
+            continue;
+        }
+        versions[i] |= (1ULL << EC_SELFHEAL_BIT);
+    }
+
+    if (EC_COUNT(healed_sinks, ec->nodes) == 0) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+    ret = 0;
+
+out:
+    cluster_replies_wipe(replies, ec->nodes);
+    if (xattrs)
+        dict_unref(xattrs);
+    if (ret < 0)
+        gf_msg_debug(ec->xl->name, 0, "%s: heal failed %s",
+                     uuid_utoa(fd->inode->gfid), strerror(-ret));
+    return ret;
+}
+
+int32_t
+ec_manager_heal_block(ec_fop_data_t *fop, int32_t state)
+{
+    ec_heal_t *heal = fop->data;
+    heal->fop = fop;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            ec_owner_set(fop->frame, fop->frame->root);
+
+            ec_heal_inodelk(heal, F_WRLCK, 1, 0, 0);
+
+            return EC_STATE_HEAL_DATA_COPY;
+
+        case EC_STATE_HEAL_DATA_COPY:
+            gf_msg_debug(fop->xl->name, 0, "%s: read/write starting",
+                         uuid_utoa(heal->fd->inode->gfid));
+            ec_heal_data_block(heal);
+
+            return EC_STATE_HEAL_DATA_UNLOCK;
+
+        case -EC_STATE_HEAL_DATA_COPY:
+        case -EC_STATE_HEAL_DATA_UNLOCK:
+        case EC_STATE_HEAL_DATA_UNLOCK:
+            ec_heal_inodelk(heal, F_UNLCK, 1, 0, 0);
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            if (fop->cbks.heal) {
+                fop->cbks.heal(fop->req_frame, fop->data, fop->xl, 0, 0,
+                               (heal->good | heal->bad), heal->good, heal->bad,
+                               0, NULL);
+            }
+
+            return EC_STATE_END;
+        case -EC_STATE_REPORT:
+            if (fop->cbks.heal) {
+                fop->cbks.heal(fop->req_frame, fop->data, fop->xl, -1,
+                               fop->error, 0, 0, 0, 0, NULL);
+            }
+
+            return EC_STATE_END;
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, 0, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+/*Takes lock */
+void
+ec_heal_block(call_frame_t *frame, xlator_t *this, uintptr_t target,
+              uint32_t fop_flags, fop_heal_cbk_t func, ec_heal_t *heal)
+{
+    ec_cbk_t callback = {.heal = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(HEAL) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, EC_FOP_HEAL, 0, target, fop_flags,
+                               NULL, ec_manager_heal_block, callback, heal);
+    if (fop == NULL)
+        goto out;
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, heal, this, -1, error, 0, 0, 0, 0, NULL);
+    }
+}
+
+int32_t
+ec_heal_block_done(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, uintptr_t mask,
+                   uintptr_t good, uintptr_t bad, uint32_t pending,
+                   dict_t *xdata)
+{
+    ec_heal_t *heal = cookie;
+
+    if (heal->fop) {
+        heal->fop->heal = NULL;
+    }
+    heal->fop = NULL;
+    heal->error = op_ret < 0 ? op_errno : 0;
+    syncbarrier_wake(heal->data);
+    return 0;
+}
+
+int
+ec_sync_heal_block(call_frame_t *frame, xlator_t *this, ec_heal_t *heal)
+{
+    ec_heal_block(frame, this, heal->bad | heal->good, EC_MINIMUM_ONE,
+                  ec_heal_block_done, heal);
+    syncbarrier_wait(heal->data, 1);
+    if (heal->error != 0) {
+        return -heal->error;
+    }
+    if (heal->bad == 0)
+        return -ENOTCONN;
+    return 0;
+}
+
+int
+ec_rebuild_data(call_frame_t *frame, ec_t *ec, fd_t *fd, uint64_t size,
+                unsigned char *sources, unsigned char *healed_sinks)
+{
+    ec_heal_t *heal = NULL;
+    int ret = 0;
+    syncbarrier_t barrier;
+
+    if (syncbarrier_init(&barrier))
+        return -ENOMEM;
+
+    heal = alloca0(sizeof(*heal));
+    heal->fd = fd_ref(fd);
+    heal->xl = ec->xl;
+    heal->data = &barrier;
+    ec_adjust_size_up(ec, &size, _gf_false);
+    heal->total_size = size;
+    heal->size = (128 * GF_UNIT_KB * (ec->self_heal_window_size));
+    /* We need to adjust the size to a multiple of the stripe size of the
+     * volume. Otherwise writes would need to fill gaps (head and/or tail)
+     * with existent data from the bad bricks. This could be garbage on a
+     * damaged file or it could fail if there aren't enough bricks. */
+    heal->size -= heal->size % ec->stripe_size;
+    heal->bad = ec_char_array_to_mask(healed_sinks, ec->nodes);
+    heal->good = ec_char_array_to_mask(sources, ec->nodes);
+    heal->iatt.ia_type = IA_IFREG;
+    LOCK_INIT(&heal->lock);
+
+    for (heal->offset = 0; (heal->offset < size) && !heal->done;
+         heal->offset += heal->size) {
+        /* We immediately abort any heal if a shutdown request has been
+         * received to avoid delays. The healing of this file will be
+         * restarted by another SHD or other client that accesses the
+         * file. */
+        if (ec->shutdown) {
+            gf_msg_debug(ec->xl->name, 0,
+                         "Cancelling heal because "
+                         "EC is stopping.");
+            ret = -ENOTCONN;
+            break;
+        }
+
+        gf_msg_debug(ec->xl->name, 0,
+                     "%s: sources: %d, sinks: "
+                     "%d, offset: %" PRIu64 " bsize: %" PRIu64,
+                     uuid_utoa(fd->inode->gfid), EC_COUNT(sources, ec->nodes),
+                     EC_COUNT(healed_sinks, ec->nodes), heal->offset,
+                     heal->size);
+        ret = ec_sync_heal_block(frame, ec->xl, heal);
+        if (ret < 0)
+            break;
+    }
+    memset(healed_sinks, 0, ec->nodes);
+    ec_mask_to_char_array(heal->bad, healed_sinks, ec->nodes);
+    fd_unref(heal->fd);
+    LOCK_DESTROY(&heal->lock);
+    syncbarrier_destroy(heal->data);
+    if (ret < 0)
+        gf_msg_debug(ec->xl->name, 0, "%s: heal failed %s",
+                     uuid_utoa(fd->inode->gfid), strerror(-ret));
+    return ret;
+}
+
+int
+__ec_heal_trim_sinks(call_frame_t *frame, ec_t *ec, fd_t *fd,
+                     unsigned char *healed_sinks, unsigned char *trim,
+                     uint64_t size)
+{
+    default_args_cbk_t *replies = NULL;
+    unsigned char *output = NULL;
+    int ret = 0;
+    int i = 0;
+    off_t trim_offset = 0;
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    output = alloca0(ec->nodes);
+
+    if (EC_COUNT(trim, ec->nodes) == 0) {
+        ret = 0;
+        goto out;
+    }
+    trim_offset = size;
+    ec_adjust_offset_up(ec, &trim_offset, _gf_true);
+    ret = cluster_ftruncate(ec->xl_list, trim, ec->nodes, replies, output,
+                            frame, ec->xl, fd, trim_offset, NULL);
+    for (i = 0; i < ec->nodes; i++) {
+        if (!output[i] && trim[i])
+            healed_sinks[i] = 0;
+    }
+
+    if (EC_COUNT(healed_sinks, ec->nodes) == 0) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+
+out:
+    cluster_replies_wipe(replies, ec->nodes);
+    if (ret < 0)
+        gf_msg_debug(ec->xl->name, 0, "%s: heal failed %s",
+                     uuid_utoa(fd->inode->gfid), strerror(-ret));
+    return ret;
+}
+
+int
+ec_data_undo_pending(call_frame_t *frame, ec_t *ec, fd_t *fd, dict_t *xattr,
+                     uint64_t *versions, uint64_t *dirty, uint64_t *size,
+                     int source, gf_boolean_t erase_dirty, int idx)
+{
+    uint64_t versions_xattr[2] = {0};
+    uint64_t dirty_xattr[2] = {0};
+    uint64_t allzero[2] = {0};
+    uint64_t size_xattr = 0;
+    int ret = 0;
+
+    versions_xattr[EC_DATA_TXN] = hton64(versions[source] - versions[idx]);
+    ret = dict_set_static_bin(xattr, EC_XATTR_VERSION, versions_xattr,
+                              sizeof(versions_xattr));
+    if (ret < 0)
+        goto out;
+
+    size_xattr = hton64(size[source] - size[idx]);
+    ret = dict_set_static_bin(xattr, EC_XATTR_SIZE, &size_xattr,
+                              sizeof(size_xattr));
+    if (ret < 0)
+        goto out;
+
+    if (erase_dirty) {
+        dirty_xattr[EC_DATA_TXN] = hton64(-dirty[idx]);
+        ret = dict_set_static_bin(xattr, EC_XATTR_DIRTY, dirty_xattr,
+                                  sizeof(dirty_xattr));
+        if (ret < 0)
+            goto out;
+    }
+
+    if ((memcmp(versions_xattr, allzero, sizeof(allzero)) == 0) &&
+        (memcmp(dirty_xattr, allzero, sizeof(allzero)) == 0) &&
+        (size_xattr == 0)) {
+        ret = 0;
+        goto out;
+    }
+
+    ret = syncop_fxattrop(ec->xl_list[idx], fd, GF_XATTROP_ADD_ARRAY64, xattr,
+                          NULL, NULL, NULL);
+out:
+    return ret;
+}
+
+int
+__ec_fd_data_adjust_versions(call_frame_t *frame, ec_t *ec, fd_t *fd,
+                             unsigned char *sources,
+                             unsigned char *healed_sinks, uint64_t *versions,
+                             uint64_t *dirty, uint64_t *size)
+{
+    dict_t *xattr = NULL;
+    int i = 0;
+    int ret = 0;
+    int op_ret = 0;
+    int source = -1;
+    gf_boolean_t erase_dirty = _gf_false;
+
+    xattr = dict_new();
+    if (!xattr) {
+        op_ret = -ENOMEM;
+        goto out;
+    }
+
+    /* dirty xattr represents if the file needs heal. Unless all the
+     * copies are healed, don't erase it */
+    if (EC_COUNT(sources, ec->nodes) + EC_COUNT(healed_sinks, ec->nodes) ==
+        ec->nodes)
+        erase_dirty = _gf_true;
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (sources[i]) {
+            source = i;
+            break;
+        }
+    }
+
+    if (source == -1) {
+        op_ret = -ENOTCONN;
+        goto out;
+    }
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (healed_sinks[i]) {
+            ret = ec_data_undo_pending(frame, ec, fd, xattr, versions, dirty,
+                                       size, source, erase_dirty, i);
+            if (ret < 0)
+                goto out;
+        }
+    }
+
+    if (!erase_dirty)
+        goto out;
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (sources[i]) {
+            ret = ec_data_undo_pending(frame, ec, fd, xattr, versions, dirty,
+                                       size, source, erase_dirty, i);
+            if (ret < 0)
+                continue;
+        }
+    }
+out:
+    if (xattr)
+        dict_unref(xattr);
+    return op_ret;
+}
+
+int
+ec_restore_time_and_adjust_versions(call_frame_t *frame, ec_t *ec, fd_t *fd,
+                                    unsigned char *sources,
+                                    unsigned char *healed_sinks,
+                                    uint64_t *versions, uint64_t *dirty,
+                                    uint64_t *size)
+{
+    unsigned char *locked_on = NULL;
+    unsigned char *participants = NULL;
+    unsigned char *output = NULL;
+    default_args_cbk_t *replies = NULL;
+    unsigned char *postsh_sources = NULL;
+    unsigned char *postsh_healed_sinks = NULL;
+    unsigned char *postsh_trim = NULL;
+    uint64_t *postsh_versions = NULL;
+    uint64_t *postsh_dirty = NULL;
+    uint64_t *postsh_size = NULL;
+    int ret = 0;
+    int i = 0;
+    struct iatt source_buf = {0};
+    loc_t loc = {0};
+
+    locked_on = alloca0(ec->nodes);
+    output = alloca0(ec->nodes);
+    participants = alloca0(ec->nodes);
+    postsh_sources = alloca0(ec->nodes);
+    postsh_healed_sinks = alloca0(ec->nodes);
+    postsh_trim = alloca0(ec->nodes);
+    postsh_versions = alloca0(ec->nodes * sizeof(*postsh_versions));
+    postsh_dirty = alloca0(ec->nodes * sizeof(*postsh_dirty));
+    postsh_size = alloca0(ec->nodes * sizeof(*postsh_size));
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (healed_sinks[i] || sources[i])
+            participants[i] = 1;
+    }
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    ret = cluster_inodelk(ec->xl_list, participants, ec->nodes, replies,
+                          locked_on, frame, ec->xl, ec->xl->name, fd->inode, 0,
+                          0);
+    {
+        if (ret <= ec->fragments) {
+            gf_msg_debug(ec->xl->name, 0,
+                         "%s: Skipping heal "
+                         "as only %d number of subvolumes could "
+                         "be locked",
+                         uuid_utoa(fd->inode->gfid), ret);
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+
+        ret = __ec_heal_data_prepare(frame, ec, fd, locked_on, postsh_versions,
+                                     postsh_dirty, postsh_size, postsh_sources,
+                                     postsh_healed_sinks, postsh_trim,
+                                     &source_buf);
+        if (ret < 0)
+            goto unlock;
+
+        loc.inode = inode_ref(fd->inode);
+        gf_uuid_copy(loc.gfid, fd->inode->gfid);
+        ret = cluster_setattr(
+            ec->xl_list, healed_sinks, ec->nodes, replies, output, frame,
+            ec->xl, &loc, &source_buf,
+            GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME | GF_SET_ATTR_CTIME, NULL);
+        EC_INTERSECT(healed_sinks, healed_sinks, output, ec->nodes);
+        if (EC_COUNT(healed_sinks, ec->nodes) == 0) {
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+        ret = __ec_fd_data_adjust_versions(frame, ec, fd, sources, healed_sinks,
+                                           versions, dirty, size);
+    }
+unlock:
+    cluster_uninodelk(ec->xl_list, locked_on, ec->nodes, replies, output, frame,
+                      ec->xl, ec->xl->name, fd->inode, 0, 0);
+    cluster_replies_wipe(replies, ec->nodes);
+    loc_wipe(&loc);
+    return ret;
+}
+
+int
+__ec_heal_data(call_frame_t *frame, ec_t *ec, fd_t *fd, unsigned char *heal_on,
+               unsigned char *sources, unsigned char *healed_sinks)
+{
+    unsigned char *locked_on = NULL;
+    unsigned char *output = NULL;
+    uint64_t *versions = NULL;
+    uint64_t *dirty = NULL;
+    uint64_t *size = NULL;
+    unsigned char *trim = NULL;
+    default_args_cbk_t *replies = NULL;
+    int ret = 0;
+    int source = 0;
+
+    locked_on = alloca0(ec->nodes);
+    output = alloca0(ec->nodes);
+    trim = alloca0(ec->nodes);
+    versions = alloca0(ec->nodes * sizeof(*versions));
+    dirty = alloca0(ec->nodes * sizeof(*dirty));
+    size = alloca0(ec->nodes * sizeof(*size));
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    ret = cluster_inodelk(ec->xl_list, heal_on, ec->nodes, replies, locked_on,
+                          frame, ec->xl, ec->xl->name, fd->inode, 0, 0);
+    {
+        if (ret <= ec->fragments) {
+            gf_msg_debug(ec->xl->name, 0,
+                         "%s: Skipping heal "
+                         "as only %d number of subvolumes could "
+                         "be locked",
+                         uuid_utoa(fd->inode->gfid), ret);
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+
+        ret = __ec_heal_data_prepare(frame, ec, fd, locked_on, versions, dirty,
+                                     size, sources, healed_sinks, trim, NULL);
+        if (ret < 0)
+            goto unlock;
+
+        if (EC_COUNT(healed_sinks, ec->nodes) == 0) {
+            ret = __ec_fd_data_adjust_versions(
+                frame, ec, fd, sources, healed_sinks, versions, dirty, size);
+            goto unlock;
+        }
+
+        source = ret;
+        ret = __ec_heal_mark_sinks(frame, ec, fd, versions, healed_sinks);
+        if (ret < 0)
+            goto unlock;
+
+        ret = __ec_heal_trim_sinks(frame, ec, fd, healed_sinks, trim,
+                                   size[source]);
+    }
+unlock:
+    cluster_uninodelk(ec->xl_list, locked_on, ec->nodes, replies, output, frame,
+                      ec->xl, ec->xl->name, fd->inode, 0, 0);
+    if (ret < 0)
+        goto out;
+
+    if (EC_COUNT(healed_sinks, ec->nodes) == 0)
+        goto out;
+
+    gf_msg_debug(ec->xl->name, 0,
+                 "%s: sources: %d, sinks: "
+                 "%d",
+                 uuid_utoa(fd->inode->gfid), EC_COUNT(sources, ec->nodes),
+                 EC_COUNT(healed_sinks, ec->nodes));
+
+    ret = ec_rebuild_data(frame, ec, fd, size[source], sources, healed_sinks);
+    if (ret < 0)
+        goto out;
+
+    ret = ec_restore_time_and_adjust_versions(
+        frame, ec, fd, sources, healed_sinks, versions, dirty, size);
+out:
+    cluster_replies_wipe(replies, ec->nodes);
+    return ret;
+}
+
+int
+ec_heal_data(call_frame_t *frame, ec_t *ec, gf_boolean_t block, inode_t *inode,
+             unsigned char *sources, unsigned char *healed_sinks)
+{
+    unsigned char *locked_on = NULL;
+    unsigned char *up_subvols = NULL;
+    unsigned char *output = NULL;
+    default_args_cbk_t *replies = NULL;
+    fd_t *fd = NULL;
+    loc_t loc = {0};
+    char selfheal_domain[1024] = {0};
+    int ret = 0;
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+
+    locked_on = alloca0(ec->nodes);
+    output = alloca0(ec->nodes);
+    up_subvols = alloca0(ec->nodes);
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    fd = fd_create(inode, 0);
+    if (!fd) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ec_mask_to_char_array(ec->xl_up, up_subvols, ec->nodes);
+
+    ret = cluster_open(ec->xl_list, up_subvols, ec->nodes, replies, output,
+                       frame, ec->xl, &loc, O_RDWR | O_LARGEFILE, fd, NULL);
+    if (ret <= ec->fragments) {
+        ret = -ENOTCONN;
+        goto out;
+    }
+
+    fd_bind(fd);
+    sprintf(selfheal_domain, "%s:self-heal", ec->xl->name);
+    /*If other processes are already doing the heal, don't block*/
+    if (block) {
+        ret = cluster_inodelk(ec->xl_list, output, ec->nodes, replies,
+                              locked_on, frame, ec->xl, selfheal_domain, inode,
+                              0, 0);
+    } else {
+        ret = cluster_tiebreaker_inodelk(ec->xl_list, output, ec->nodes,
+                                         replies, locked_on, frame, ec->xl,
+                                         selfheal_domain, inode, 0, 0);
+    }
+    {
+        if (ret <= ec->fragments) {
+            gf_msg_debug(ec->xl->name, 0,
+                         "%s: Skipping heal "
+                         "as only %d number of subvolumes could "
+                         "be locked",
+                         uuid_utoa(inode->gfid), ret);
+            ret = -ENOTCONN;
+            goto unlock;
+        }
+        ret = __ec_heal_data(frame, ec, fd, locked_on, sources, healed_sinks);
+    }
+unlock:
+    cluster_uninodelk(ec->xl_list, locked_on, ec->nodes, replies, output, frame,
+                      ec->xl, selfheal_domain, inode, 0, 0);
+out:
+    if (fd)
+        fd_unref(fd);
+    loc_wipe(&loc);
+    cluster_replies_wipe(replies, ec->nodes);
+    return ret;
+}
+
+int
+ec_heal_purge_stale_index(call_frame_t *frame, ec_t *ec, inode_t *inode)
+{
+    int i = 0;
+    int ret = 0;
+    dict_t **xattr = NULL;
+    loc_t loc = {0};
+    uint64_t dirty_xattr[EC_VERSION_SIZE] = {0};
+    unsigned char *on = NULL;
+    default_args_cbk_t *replies = NULL;
+    dict_t *dict = NULL;
+
+    /* Allocate the required memory */
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+    on = alloca0(ec->nodes);
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    xattr = GF_CALLOC(ec->nodes, sizeof(*xattr), gf_common_mt_pointer);
+    if (!xattr) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    dict = dict_new();
+    if (!dict) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    for (i = 0; i < ec->nodes; i++) {
+        xattr[i] = dict;
+        on[i] = 1;
+    }
+    ret = dict_set_static_bin(dict, EC_XATTR_DIRTY, dirty_xattr,
+                              (sizeof(*dirty_xattr) * EC_VERSION_SIZE));
+    if (ret < 0) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    PARALLEL_FOP_ONLIST(ec->xl_list, on, ec->nodes, replies, frame,
+                        ec_wind_xattrop_parallel, &loc, GF_XATTROP_ADD_ARRAY64,
+                        xattr, NULL);
+out:
+    if (dict) {
+        dict_unref(dict);
+    }
+    if (xattr) {
+        GF_FREE(xattr);
+    }
+    cluster_replies_wipe(replies, ec->nodes);
+    loc_wipe(&loc);
+    return ret;
+}
+
+void
+ec_heal_do(xlator_t *this, void *data, loc_t *loc, int32_t partial)
+{
+    call_frame_t *frame = NULL;
+    unsigned char *participants = NULL;
+    unsigned char *msources = NULL;
+    unsigned char *mhealed_sinks = NULL;
+    unsigned char *sources = NULL;
+    unsigned char *healed_sinks = NULL;
+    ec_t *ec = NULL;
+    int ret = 0;
+    int op_ret = 0;
+    int op_errno = 0;
+    intptr_t mgood = 0;
+    intptr_t mbad = 0;
+    intptr_t good = 0;
+    intptr_t bad = 0;
+    uint32_t pending = 0;
+    ec_fop_data_t *fop = data;
+    gf_boolean_t blocking = _gf_false;
+    ec_heal_need_t need_heal = EC_HEAL_NONEED;
+    unsigned char *up_subvols = NULL;
+    char up_bricks[32];
+
+    ec = this->private;
+
+    /* If it is heal request from getxattr, complete the heal and then
+     * unwind, if it is ec_heal with NULL as frame then no need to block
+     * the heal as the caller doesn't care about its completion. In case
+     * of heald whichever gets tiebreaking inodelk will take care of the
+     * heal, so no need to block*/
+    if (fop->req_frame && !ec->shd.iamshd)
+        blocking = _gf_true;
+
+    frame = create_frame(this, this->ctx->pool);
+    if (!frame)
+        goto out;
+
+    ec_owner_set(frame, frame->root);
+    /*Do heal as root*/
+    frame->root->uid = 0;
+    frame->root->gid = 0;
+    /*Mark the fops as internal*/
+    frame->root->pid = GF_CLIENT_PID_SELF_HEALD;
+    participants = alloca0(ec->nodes);
+    ec_mask_to_char_array(ec->xl_up, participants, ec->nodes);
+
+    up_subvols = alloca0(ec->nodes);
+    ec_mask_to_char_array(ec->xl_up, up_subvols, ec->nodes);
+
+    if (loc->name && strlen(loc->name)) {
+        ret = ec_heal_name(frame, ec, loc->parent, (char *)loc->name,
+                           participants);
+        if (ret >= 0) {
+            gf_msg_debug(this->name, 0,
+                         "%s: name heal "
+                         "successful on %" PRIXPTR,
+                         loc->path,
+                         ec_char_array_to_mask(participants, ec->nodes));
+        } else {
+            gf_msg_debug(
+                this->name, 0,
+                "%s: name heal "
+                "failed. ret = %d, subvolumes up = %s",
+                loc->path, ret,
+                ec_bin(up_bricks, sizeof(up_bricks), ec->xl_up, ec->nodes));
+        }
+    }
+
+    /* Mount triggers heal only when it detects that it must need heal, shd
+     * triggers heals periodically which need not be thorough*/
+    if (ec->shd.iamshd && (ret <= 0)) {
+        ec_heal_inspect(frame, ec, loc->inode, up_subvols, _gf_false, _gf_false,
+                        &need_heal);
+
+        if (need_heal == EC_HEAL_PURGE_INDEX) {
+            gf_msg(ec->xl->name, GF_LOG_INFO, 0, EC_MSG_HEAL_FAIL,
+                   "Index entry needs to be purged for: %s ",
+                   uuid_utoa(loc->gfid));
+            /* We need to send zero-xattrop so that stale index entry could be
+             * removed. We need not take lock on this entry to do so as
+             * xattrop on a brick is atomic. */
+            ec_heal_purge_stale_index(frame, ec, loc->inode);
+            goto out;
+        } else if (need_heal == EC_HEAL_NONEED) {
+            gf_msg(ec->xl->name, GF_LOG_DEBUG, 0, EC_MSG_HEAL_FAIL,
+                   "Heal is not required for : %s ", uuid_utoa(loc->gfid));
+            goto out;
+        }
+    }
+
+    sources = alloca0(ec->nodes);
+    healed_sinks = alloca0(ec->nodes);
+    if (IA_ISREG(loc->inode->ia_type)) {
+        ret = ec_heal_data(frame, ec, blocking, loc->inode, sources,
+                           healed_sinks);
+    } else if (IA_ISDIR(loc->inode->ia_type) && !partial) {
+        ret = ec_heal_entry(frame, ec, loc->inode, sources, healed_sinks,
+                            &pending);
+    } else {
+        ret = 0;
+        memcpy(sources, participants, ec->nodes);
+        memcpy(healed_sinks, participants, ec->nodes);
+    }
+
+    if (ret == 0) {
+        good = ec_char_array_to_mask(sources, ec->nodes);
+        bad = ec_char_array_to_mask(healed_sinks, ec->nodes);
+    } else {
+        op_ret = -1;
+        op_errno = -ret;
+    }
+    msources = alloca0(ec->nodes);
+    mhealed_sinks = alloca0(ec->nodes);
+    ret = ec_heal_metadata(frame, ec, loc->inode, msources, mhealed_sinks);
+    if (ret == 0) {
+        mgood = ec_char_array_to_mask(msources, ec->nodes);
+        mbad = ec_char_array_to_mask(mhealed_sinks, ec->nodes);
+    } else {
+        op_ret = -1;
+        op_errno = -ret;
+    }
+
+out:
+    ec_reset_entry_healing(fop);
+    if (fop->cbks.heal) {
+        fop->cbks.heal(fop->req_frame, fop->data, fop->xl, op_ret, op_errno,
+                       ec_char_array_to_mask(participants, ec->nodes),
+                       mgood & good, mbad & bad, pending, NULL);
+    }
+    if (frame)
+        STACK_DESTROY(frame->root);
+
+    return;
+}
+
+int
+ec_synctask_heal_wrap(void *opaque)
+{
+    ec_fop_data_t *fop = opaque;
+    ec_heal_do(fop->xl, fop, &fop->loc[0], fop->int32);
+    return 0;
+}
+
+int
+ec_heal_done(int ret, call_frame_t *heal, void *opaque)
+{
+    if (opaque)
+        ec_fop_data_release(opaque);
+    return 0;
+}
+
+ec_fop_data_t *
+__ec_dequeue_heals(ec_t *ec)
+{
+    ec_fop_data_t *fop = NULL;
+
+    if (list_empty(&ec->heal_waiting))
+        goto none;
+
+    if ((ec->background_heals > 0) && (ec->healers >= ec->background_heals))
+        goto none;
+
+    fop = list_entry(ec->heal_waiting.next, ec_fop_data_t, healer);
+    ec->heal_waiters--;
+    list_del_init(&fop->healer);
+    list_add(&fop->healer, &ec->healing);
+    ec->healers++;
+    return fop;
+none:
+    gf_msg_debug(ec->xl->name, 0, "Num healers: %d, Num Waiters: %d",
+                 ec->healers, ec->heal_waiters);
+    return NULL;
+}
+
+void
+ec_heal_fail(ec_t *ec, ec_fop_data_t *fop)
+{
+    if (fop->cbks.heal) {
+        fop->cbks.heal(fop->req_frame, fop->data, ec->xl, -1, fop->error, 0, 0,
+                       0, 0, NULL);
+    }
+    ec_fop_data_release(fop);
+}
+
+void
+ec_launch_heal(ec_t *ec, ec_fop_data_t *fop)
+{
+    int ret = 0;
+    call_frame_t *frame = NULL;
+
+    frame = create_frame(ec->xl, ec->xl->ctx->pool);
+    if (!frame) {
+        ret = -1;
+        goto out;
+    }
+
+    ec_owner_set(frame, frame->root);
+    /*Do heal as root*/
+    frame->root->uid = 0;
+    frame->root->gid = 0;
+    /*Mark the fops as internal*/
+    frame->root->pid = GF_CLIENT_PID_SELF_HEALD;
+
+    ret = synctask_new(ec->xl->ctx->env, ec_synctask_heal_wrap, ec_heal_done,
+                       frame, fop);
+out:
+    if (ret < 0) {
+        ec_fop_set_error(fop, ENOMEM);
+        ec_heal_fail(ec, fop);
+    }
+
+    if (frame)
+        STACK_DESTROY(frame->root);
+}
+
+void
+ec_handle_healers_done(ec_fop_data_t *fop)
+{
+    ec_t *ec = fop->xl->private;
+    ec_fop_data_t *heal_fop = NULL;
+
+    if (list_empty(&fop->healer))
+        return;
+
+    LOCK(&ec->lock);
+
+    list_del_init(&fop->healer);
+
+    do {
+        ec->healers--;
+        heal_fop = __ec_dequeue_heals(ec);
+
+        if ((heal_fop != NULL) && ec->shutdown) {
+            /* This will prevent ec_handle_healers_done() to be
+             * called recursively. That would be problematic if
+             * the queue is too big. */
+            list_del_init(&heal_fop->healer);
+
+            UNLOCK(&ec->lock);
+
+            ec_fop_set_error(fop, ENOTCONN);
+            ec_heal_fail(ec, heal_fop);
+
+            LOCK(&ec->lock);
+        }
+    } while ((heal_fop != NULL) && ec->shutdown);
+
+    UNLOCK(&ec->lock);
+
+    if (heal_fop)
+        ec_launch_heal(ec, heal_fop);
+}
+
+gf_boolean_t
+ec_is_entry_healing(ec_fop_data_t *fop)
+{
+    ec_inode_t *ctx = NULL;
+    int32_t heal_count = 0;
+    loc_t *loc = NULL;
+
+    loc = &fop->loc[0];
+
+    LOCK(&loc->inode->lock);
+    {
+        ctx = __ec_inode_get(loc->inode, fop->xl);
+        if (ctx) {
+            heal_count = ctx->heal_count;
+        }
+    }
+    UNLOCK(&loc->inode->lock);
+    GF_ASSERT(heal_count >= 0);
+    return heal_count;
+}
+
+void
+ec_heal_throttle(xlator_t *this, ec_fop_data_t *fop)
+{
+    gf_boolean_t can_heal = _gf_true;
+    ec_t *ec = this->private;
+    ec_fop_data_t *fop_rel = NULL;
+
+    if (fop->req_frame == NULL) {
+        LOCK(&ec->lock);
+        {
+            if ((ec->background_heals > 0) &&
+                (ec->heal_wait_qlen + ec->background_heals) >
+                    (ec->heal_waiters + ec->healers)) {
+                if (!ec_is_entry_healing(fop)) {
+                    list_add_tail(&fop->healer, &ec->heal_waiting);
+                    ec->heal_waiters++;
+                    ec_set_entry_healing(fop);
+                } else {
+                    fop_rel = fop;
+                }
+                fop = __ec_dequeue_heals(ec);
+            } else {
+                can_heal = _gf_false;
+            }
+        }
+        UNLOCK(&ec->lock);
+    }
+
+    if (can_heal) {
+        if (fop) {
+            if (fop->req_frame != NULL) {
+                ec_set_entry_healing(fop);
+            }
+            ec_launch_heal(ec, fop);
+        }
+    } else {
+        gf_msg_debug(this->name, 0,
+                     "Max number of heals are "
+                     "pending, background self-heal rejected");
+        ec_fop_set_error(fop, EBUSY);
+        ec_heal_fail(ec, fop);
+    }
+    if (fop_rel) {
+        ec_heal_done(0, NULL, fop_rel);
+    }
+}
+
+void
+ec_heal(call_frame_t *frame, xlator_t *this, uintptr_t target,
+        uint32_t fop_flags, fop_heal_cbk_t func, void *data, loc_t *loc,
+        int32_t partial, dict_t *xdata)
+{
+    ec_cbk_t callback = {.heal = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t err = EINVAL;
+
+    gf_msg_trace("ec", 0, "EC(HEAL) %p", frame);
+
+    VALIDATE_OR_GOTO(this, fail);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, fail);
+
+    if (!loc || !loc->inode || gf_uuid_is_null(loc->inode->gfid))
+        goto fail;
+
+    if (frame && frame->local)
+        goto fail;
+    fop = ec_fop_data_allocate(frame, this, EC_FOP_HEAL, 0, target, fop_flags,
+                               NULL, NULL, callback, data);
+
+    err = ENOMEM;
+
+    if (fop == NULL)
+        goto fail;
+
+    fop->int32 = partial;
+
+    if (loc) {
+        if (loc_copy(&fop->loc[0], loc) != 0)
+            goto fail;
+    }
+
+    if (xdata)
+        fop->xdata = dict_ref(xdata);
+
+    ec_heal_throttle(this, fop);
+
+    return;
+
+fail:
+    if (fop)
+        ec_fop_data_release(fop);
+    if (func)
+        func(frame, data, this, -1, err, 0, 0, 0, 0, NULL);
+}
+
+int
+ec_replace_heal_done(int ret, call_frame_t *heal, void *opaque)
+{
+    ec_t *ec = opaque;
+    gf_boolean_t last_fop = _gf_false;
+
+    if (GF_ATOMIC_DEC(ec->async_fop_count) == 0) {
+        LOCK(&ec->lock);
+        {
+            last_fop = __ec_is_last_fop(ec);
+        }
+        UNLOCK(&ec->lock);
+    }
+    gf_msg_debug(ec->xl->name, 0, "getxattr on bricks is done ret %d", ret);
+
+    if (last_fop)
+        ec_pending_fops_completed(ec);
+
+    return 0;
+}
+
+int32_t
+ec_replace_heal(ec_t *ec, inode_t *inode)
+{
+    loc_t loc = {0};
+    int ret = 0;
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+    ret = syncop_getxattr(ec->xl, &loc, NULL, EC_XATTR_HEAL, NULL, NULL);
+    if (ret < 0)
+        gf_msg_debug(ec->xl->name, 0, "Heal failed for replace brick ret = %d",
+                     ret);
+
+    /* Once the root inode has been checked, it might have triggered a
+     * self-heal on it after a replace brick command or for some other
+     * reason. It can also happen that the volume already had damaged
+     * files in the index, even if the heal on the root directory failed.
+     * In both cases we need to wake all index healers to continue
+     * healing remaining entries that are marked as dirty. */
+    ec_shd_index_healer_wake(ec);
+
+    loc_wipe(&loc);
+    return ret;
+}
+
+int32_t
+ec_replace_brick_heal_wrap(void *opaque)
+{
+    ec_t *ec = opaque;
+    inode_table_t *itable = NULL;
+    int32_t ret = -1;
+
+    if (ec->xl->itable)
+        itable = ec->xl->itable;
+    else
+        goto out;
+
+    if (xlator_is_cleanup_starting(ec->xl))
+        goto out;
+
+    ret = ec_replace_heal(ec, itable->root);
+out:
+    return ret;
+}
+
+int32_t
+ec_launch_replace_heal(ec_t *ec)
+{
+    int ret = -1;
+
+    ret = synctask_new(ec->xl->ctx->env, ec_replace_brick_heal_wrap,
+                       ec_replace_heal_done, NULL, ec);
+
+    if (ret < 0) {
+        gf_msg_debug(ec->xl->name, 0, "Heal failed for replace brick ret = %d",
+                     ret);
+        ec_replace_heal_done(-1, NULL, ec);
+    }
+
+    return ret;
+}
+
+int32_t
+ec_set_heal_info(dict_t **dict_rsp, char *status)
+{
+    dict_t *dict = NULL;
+    int ret = 0;
+
+    dict = dict_new();
+    if (!dict) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    ret = dict_set_str(dict, "heal-info", status);
+    if (ret) {
+        gf_msg(THIS->name, GF_LOG_WARNING, -ret, EC_MSG_HEAL_FAIL,
+               "Failed to set heal-info key to "
+               "%s",
+               status);
+        dict_unref(dict);
+        dict = NULL;
+    }
+    *dict_rsp = dict;
+out:
+    return ret;
+}
+
+static int32_t
+_need_heal_calculate(ec_t *ec, uint64_t *dirty, unsigned char *sources,
+                     gf_boolean_t self_locked, int32_t lock_count,
+                     ec_heal_need_t *need_heal, uint64_t *versions)
+{
+    int i = 0;
+    int source_count = 0;
+
+    source_count = EC_COUNT(sources, ec->nodes);
+    if (source_count == ec->nodes) {
+        *need_heal = EC_HEAL_NONEED;
+        if (self_locked || lock_count == 0) {
+            for (i = 0; i < ec->nodes; i++) {
+                if (dirty[i] || (versions[i] != versions[0])) {
+                    *need_heal = EC_HEAL_MUST;
+                    goto out;
+                }
+            }
+            /* If lock count is 0, all dirty flags are 0 and all the
+             * versions are macthing then why are we here. It looks
+             * like something went wrong while removing the index entries
+             * after completing a successful heal or fop. In this case
+             * we need to remove this index entry to avoid triggering heal
+             * in a loop and causing lookups again and again*/
+            *need_heal = EC_HEAL_PURGE_INDEX;
+        } else {
+            for (i = 0; i < ec->nodes; i++) {
+                /* Since each lock can only increment the dirty
+                 * count once, if dirty is > 1 it means that
+                 * another operation has left the dirty count
+                 * set and this indicates a problem in the
+                 * inode.*/
+                if (dirty[i] > 1) {
+                    *need_heal = EC_HEAL_MUST;
+                    goto out;
+                }
+                if (dirty[i] != dirty[0] || (versions[i] != versions[0])) {
+                    *need_heal = EC_HEAL_MAYBE;
+                }
+            }
+        }
+    } else {
+        *need_heal = EC_HEAL_MUST;
+    }
+
+out:
+    return source_count;
+}
+
+static int32_t
+ec_need_metadata_heal(ec_t *ec, inode_t *inode, default_args_cbk_t *replies,
+                      int32_t lock_count, gf_boolean_t self_locked,
+                      gf_boolean_t thorough, ec_heal_need_t *need_heal)
+{
+    uint64_t *dirty = NULL;
+    unsigned char *sources = NULL;
+    unsigned char *healed_sinks = NULL;
+    uint64_t *meta_versions = NULL;
+    int ret = 0;
+
+    sources = alloca0(ec->nodes);
+    healed_sinks = alloca0(ec->nodes);
+    dirty = alloca0(ec->nodes * sizeof(*dirty));
+    meta_versions = alloca0(ec->nodes * sizeof(*meta_versions));
+    ret = ec_heal_metadata_find_direction(ec, replies, meta_versions, dirty,
+                                          sources, healed_sinks);
+    if (ret < 0 && ret != -EIO) {
+        goto out;
+    }
+
+    ret = _need_heal_calculate(ec, dirty, sources, self_locked, lock_count,
+                               need_heal, meta_versions);
+out:
+    return ret;
+}
+
+static int32_t
+ec_need_data_heal(ec_t *ec, inode_t *inode, default_args_cbk_t *replies,
+                  int32_t lock_count, gf_boolean_t self_locked,
+                  gf_boolean_t thorough, ec_heal_need_t *need_heal)
+{
+    uint64_t *dirty = NULL;
+    unsigned char *sources = NULL;
+    unsigned char *healed_sinks = NULL;
+    uint64_t *data_versions = NULL;
+    uint64_t *size = NULL;
+    int ret = 0;
+
+    sources = alloca0(ec->nodes);
+    healed_sinks = alloca0(ec->nodes);
+    dirty = alloca0(ec->nodes * sizeof(*dirty));
+    data_versions = alloca0(ec->nodes * sizeof(*data_versions));
+    size = alloca0(ec->nodes * sizeof(*size));
+
+    /* When dd is going on and heal info is called there is a very good
+     * chance for on disk sizes to mismatch even though nothing is wrong
+     * we don't need ondisk size check there. But if the file is either
+     * self-locked or the caller wants a thorough check then make sure to
+     * perform on disk check also. */
+    ret = ec_heal_data_find_direction(
+        ec, replies, data_versions, dirty, size, sources, healed_sinks,
+        self_locked || thorough, EC_COMBINE_XDATA);
+    if (ret < 0 && ret != -EIO) {
+        goto out;
+    }
+
+    ret = _need_heal_calculate(ec, dirty, sources, self_locked, lock_count,
+                               need_heal, data_versions);
+out:
+    return ret;
+}
+
+static int32_t
+ec_need_entry_heal(ec_t *ec, inode_t *inode, default_args_cbk_t *replies,
+                   int32_t lock_count, gf_boolean_t self_locked,
+                   gf_boolean_t thorough, ec_heal_need_t *need_heal)
+{
+    uint64_t *dirty = NULL;
+    unsigned char *sources = NULL;
+    unsigned char *healed_sinks = NULL;
+    uint64_t *data_versions = NULL;
+    int ret = 0;
+
+    sources = alloca0(ec->nodes);
+    healed_sinks = alloca0(ec->nodes);
+    dirty = alloca0(ec->nodes * sizeof(*dirty));
+    data_versions = alloca0(ec->nodes * sizeof(*data_versions));
+
+    ret = ec_heal_entry_find_direction(ec, replies, data_versions, dirty,
+                                       sources, healed_sinks);
+    if (ret < 0 && ret != -EIO) {
+        goto out;
+    }
+
+    ret = _need_heal_calculate(ec, dirty, sources, self_locked, lock_count,
+                               need_heal, data_versions);
+out:
+    return ret;
+}
+
+static int32_t
+ec_need_heal(ec_t *ec, inode_t *inode, default_args_cbk_t *replies,
+             int32_t lock_count, gf_boolean_t self_locked,
+             gf_boolean_t thorough, ec_heal_need_t *need_heal)
+{
+    int ret = 0;
+
+    ret = ec_need_metadata_heal(ec, inode, replies, lock_count, self_locked,
+                                thorough, need_heal);
+    if (ret < 0)
+        goto out;
+
+    if (*need_heal == EC_HEAL_MUST)
+        goto out;
+
+    if (inode->ia_type == IA_IFREG) {
+        ret = ec_need_data_heal(ec, inode, replies, lock_count, self_locked,
+                                thorough, need_heal);
+    } else if (inode->ia_type == IA_IFDIR) {
+        ret = ec_need_entry_heal(ec, inode, replies, lock_count, self_locked,
+                                 thorough, need_heal);
+    }
+
+out:
+    return ret;
+}
+
+int32_t
+ec_heal_inspect(call_frame_t *frame, ec_t *ec, inode_t *inode,
+                unsigned char *locked_on, gf_boolean_t self_locked,
+                gf_boolean_t thorough, ec_heal_need_t *need_heal)
+{
+    loc_t loc = {0};
+    int i = 0;
+    int ret = 0;
+    dict_t *xdata = NULL;
+    uint64_t zero_array[2] = {0};
+    uint64_t zero_value = 0;
+    unsigned char *output = NULL;
+    default_args_cbk_t *replies = NULL;
+    int32_t lock_count = 0;
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    output = alloca0(ec->nodes);
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    xdata = dict_new();
+    if (!xdata ||
+        dict_set_static_bin(xdata, EC_XATTR_VERSION, zero_array,
+                            sizeof(zero_array)) ||
+        dict_set_static_bin(xdata, EC_XATTR_DIRTY, zero_array,
+                            sizeof(zero_array)) ||
+        dict_set_static_bin(xdata, EC_XATTR_SIZE, &zero_value,
+                            sizeof(zero_value))) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    if (!self_locked) {
+        ret = dict_set_str(xdata, GLUSTERFS_INODELK_DOM_COUNT, ec->xl->name);
+        if (ret) {
+            ret = -ENOMEM;
+            goto out;
+        }
+    }
+
+    ret = cluster_lookup(ec->xl_list, locked_on, ec->nodes, replies, output,
+                         frame, ec->xl, &loc, xdata);
+
+    if (ret != ec->nodes) {
+        ret = ec->nodes;
+        *need_heal = EC_HEAL_MUST;
+        goto out;
+    }
+
+    if (self_locked)
+        goto need_heal;
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (!output[i] || !replies[i].xdata) {
+            continue;
+        }
+        if ((dict_get_int32(replies[i].xdata, GLUSTERFS_INODELK_COUNT,
+                            &lock_count) == 0) &&
+            lock_count > 0) {
+            break;
+        }
+    }
+need_heal:
+    ret = ec_need_heal(ec, inode, replies, lock_count, self_locked, thorough,
+                       need_heal);
+out:
+    cluster_replies_wipe(replies, ec->nodes);
+    loc_wipe(&loc);
+    if (xdata) {
+        dict_unref(xdata);
+    }
+    return ret;
+}
+
+int32_t
+ec_heal_locked_inspect(call_frame_t *frame, ec_t *ec, inode_t *inode,
+                       ec_heal_need_t *need_heal)
+{
+    unsigned char *locked_on = NULL;
+    unsigned char *up_subvols = NULL;
+    unsigned char *output = NULL;
+    default_args_cbk_t *replies = NULL;
+    int ret = 0;
+
+    EC_REPLIES_ALLOC(replies, ec->nodes);
+    locked_on = alloca0(ec->nodes);
+    output = alloca0(ec->nodes);
+    up_subvols = alloca0(ec->nodes);
+    ec_mask_to_char_array(ec->xl_up, up_subvols, ec->nodes);
+
+    ret = cluster_inodelk(ec->xl_list, up_subvols, ec->nodes, replies,
+                          locked_on, frame, ec->xl, ec->xl->name, inode, 0, 0);
+    if (ret != ec->nodes) {
+        *need_heal = EC_HEAL_MUST;
+        goto unlock;
+    }
+    ret = ec_heal_inspect(frame, ec, inode, locked_on, _gf_true, _gf_true,
+                          need_heal);
+unlock:
+    cluster_uninodelk(ec->xl_list, locked_on, ec->nodes, replies, output, frame,
+                      ec->xl, ec->xl->name, inode, 0, 0);
+    cluster_replies_wipe(replies, ec->nodes);
+    return ret;
+}
+
+int32_t
+ec_get_heal_info(xlator_t *this, loc_t *entry_loc, dict_t **dict_rsp)
+{
+    int ret = -ENOMEM;
+    ec_heal_need_t need_heal = EC_HEAL_NONEED;
+    call_frame_t *frame = NULL;
+    ec_t *ec = NULL;
+    unsigned char *up_subvols = NULL;
+    loc_t loc = {
+        0,
+    };
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, entry_loc, out);
+
+    ec = this->private;
+    up_subvols = alloca0(ec->nodes);
+    ec_mask_to_char_array(ec->xl_up, up_subvols, ec->nodes);
+
+    if (EC_COUNT(up_subvols, ec->nodes) != ec->nodes) {
+        need_heal = EC_HEAL_MUST;
+        goto set_heal;
+    }
+    frame = create_frame(this, this->ctx->pool);
+    if (!frame) {
+        goto out;
+    }
+    ec_owner_set(frame, frame->root);
+    frame->root->uid = 0;
+    frame->root->gid = 0;
+    frame->root->pid = GF_CLIENT_PID_SELF_HEALD;
+
+    if (loc_copy(&loc, entry_loc) != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+               "Failed to copy a location.");
+        goto out;
+    }
+    if (!loc.inode) {
+        ret = syncop_inode_find(this, this, loc.gfid, &loc.inode, NULL, NULL);
+        if (ret < 0)
+            goto out;
+    }
+
+    ret = ec_heal_inspect(frame, ec, loc.inode, up_subvols, _gf_false,
+                          _gf_false, &need_heal);
+    if (ret == ec->nodes && need_heal != EC_HEAL_MAYBE) {
+        goto set_heal;
+    }
+    need_heal = EC_HEAL_NONEED;
+    ret = ec_heal_locked_inspect(frame, ec, loc.inode, &need_heal);
+    if (ret < 0)
+        goto out;
+set_heal:
+    if (need_heal == EC_HEAL_MUST) {
+        ret = ec_set_heal_info(dict_rsp, "heal");
+    } else {
+        ret = ec_set_heal_info(dict_rsp, "no-heal");
+    }
+out:
+    if (frame) {
+        STACK_DESTROY(frame->root);
+    }
+    loc_wipe(&loc);
+    return ret;
+}
diff --git a/xlators/cluster/ec/src/ec-heald.c b/xlators/cluster/ec/src/ec-heald.c
new file mode 100644
index 00000000000..5c1586bc9c5
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-heald.c
@@ -0,0 +1,681 @@
+/*
+  Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/defaults.h>
+#include <glusterfs/compat-errno.h>
+#include "ec.h"
+#include "ec-messages.h"
+#include "ec-heald.h"
+#include "ec-mem-types.h"
+#include <glusterfs/syncop.h>
+#include <glusterfs/syncop-utils.h>
+#include "protocol-common.h"
+
+#define NTH_INDEX_HEALER(this, n)                                              \
+    (&((((ec_t *)this->private))->shd.index_healers[n]))
+#define NTH_FULL_HEALER(this, n)                                               \
+    (&((((ec_t *)this->private))->shd.full_healers[n]))
+
+gf_boolean_t
+ec_shd_is_subvol_local(xlator_t *this, int subvol)
+{
+    ec_t *ec = NULL;
+    gf_boolean_t is_local = _gf_false;
+    loc_t loc = {
+        0,
+    };
+
+    ec = this->private;
+    loc.inode = this->itable->root;
+    syncop_is_subvol_local(ec->xl_list[subvol], &loc, &is_local);
+    return is_local;
+}
+
+char *
+ec_subvol_name(xlator_t *this, int subvol)
+{
+    ec_t *ec = NULL;
+
+    ec = this->private;
+    if (subvol < 0 || subvol > ec->nodes)
+        return NULL;
+
+    return ec->xl_list[subvol]->name;
+}
+
+int
+__ec_shd_healer_wait(struct subvol_healer *healer)
+{
+    ec_t *ec = NULL;
+    struct timespec wait_till = {
+        0,
+    };
+    int ret = 0;
+
+    ec = healer->this->private;
+
+disabled_loop:
+    wait_till.tv_sec = gf_time() + ec->shd.timeout;
+
+    while (!healer->rerun) {
+        ret = pthread_cond_timedwait(&healer->cond, &healer->mutex, &wait_till);
+        if (ret == ETIMEDOUT)
+            break;
+    }
+
+    if (ec->shutdown) {
+        healer->running = _gf_false;
+        return -1;
+    }
+
+    ret = healer->rerun;
+    healer->rerun = 0;
+
+    if (!ec->shd.enabled || !ec->up)
+        goto disabled_loop;
+
+    return ret;
+}
+
+int
+ec_shd_healer_wait(struct subvol_healer *healer)
+{
+    int ret = 0;
+
+    pthread_mutex_lock(&healer->mutex);
+    {
+        ret = __ec_shd_healer_wait(healer);
+    }
+    pthread_mutex_unlock(&healer->mutex);
+
+    return ret;
+}
+
+int
+ec_shd_index_inode(xlator_t *this, xlator_t *subvol, inode_t **inode)
+{
+    loc_t rootloc = {
+        0,
+    };
+    int ret = 0;
+    dict_t *xattr = NULL;
+    void *index_gfid = NULL;
+
+    *inode = NULL;
+    rootloc.inode = inode_ref(this->itable->root);
+    gf_uuid_copy(rootloc.gfid, rootloc.inode->gfid);
+
+    ret = syncop_getxattr(subvol, &rootloc, &xattr, GF_XATTROP_INDEX_GFID, NULL,
+                          NULL);
+    if (ret < 0)
+        goto out;
+    if (!xattr) {
+        ret = -EINVAL;
+        goto out;
+    }
+
+    ret = dict_get_ptr(xattr, GF_XATTROP_INDEX_GFID, &index_gfid);
+    if (ret)
+        goto out;
+
+    gf_msg_debug(this->name, 0, "index-dir gfid for %s: %s", subvol->name,
+                 uuid_utoa(index_gfid));
+
+    ret = syncop_inode_find(this, subvol, index_gfid, inode, NULL, NULL);
+
+out:
+    loc_wipe(&rootloc);
+
+    if (xattr)
+        dict_unref(xattr);
+
+    return ret;
+}
+
+int
+ec_shd_index_purge(xlator_t *subvol, inode_t *inode, char *name)
+{
+    loc_t loc = {
+        0,
+    };
+    int ret = 0;
+
+    loc.parent = inode_ref(inode);
+    loc.name = name;
+
+    ret = syncop_unlink(subvol, &loc, NULL, NULL);
+
+    loc_wipe(&loc);
+    return ret;
+}
+
+static gf_boolean_t
+ec_is_heal_completed(char *status)
+{
+    char *bad_pos = NULL;
+    char *zero_pos = NULL;
+
+    if (!status) {
+        return _gf_false;
+    }
+
+    /*Logic:
+     * Status will be of the form Good: <binary>, Bad: <binary>
+     * If heal completes, if we do strchr for '0' it should be present after
+     * 'Bad:' i.e. strRchr for ':'
+     * */
+
+    zero_pos = strchr(status, '0');
+    bad_pos = strrchr(status, ':');
+    if (!zero_pos || !bad_pos) {
+        /*malformed status*/
+        return _gf_false;
+    }
+
+    if (zero_pos > bad_pos) {
+        return _gf_true;
+    }
+
+    return _gf_false;
+}
+
+int
+ec_shd_selfheal(struct subvol_healer *healer, int child, loc_t *loc,
+                gf_boolean_t full)
+{
+    dict_t *xdata = NULL;
+    dict_t *dict = NULL;
+    uint32_t count;
+    int32_t ret;
+    char *heal_status = NULL;
+    ec_t *ec = healer->this->private;
+
+    GF_ATOMIC_INC(ec->stats.shd.attempted);
+    ret = syncop_getxattr(healer->this, loc, &dict, EC_XATTR_HEAL, NULL,
+                          &xdata);
+    if (ret == 0) {
+        if (dict && (dict_get_str(dict, EC_XATTR_HEAL, &heal_status) == 0)) {
+            if (ec_is_heal_completed(heal_status)) {
+                GF_ATOMIC_INC(ec->stats.shd.completed);
+            }
+        }
+    }
+
+    if (!full && (loc->inode->ia_type == IA_IFDIR)) {
+        /* If we have just healed a directory, it's possible that
+         * other index entries have appeared to be healed. */
+        if ((xdata != NULL) &&
+            (dict_get_uint32(xdata, EC_XATTR_HEAL_NEW, &count) == 0) &&
+            (count > 0)) {
+            /* Force a rerun of the index healer. */
+            gf_msg_debug(healer->this->name, 0, "%d more entries to heal",
+                         count);
+
+            healer->rerun = _gf_true;
+        }
+    }
+
+    if (xdata != NULL) {
+        dict_unref(xdata);
+    }
+
+    if (dict) {
+        dict_unref(dict);
+    }
+
+    return ret;
+}
+
+int
+ec_shd_index_heal(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent,
+                  void *data)
+{
+    struct subvol_healer *healer = data;
+    ec_t *ec = NULL;
+    loc_t loc = {0};
+    int ret = 0;
+
+    ec = healer->this->private;
+    if (ec->xl_up_count <= ec->fragments) {
+        return -ENOTCONN;
+    }
+    if (!ec->shd.enabled)
+        return -EBUSY;
+
+    gf_msg_debug(healer->this->name, 0, "got entry: %s", entry->d_name);
+
+    ret = gf_uuid_parse(entry->d_name, loc.gfid);
+    if (ret)
+        return 0;
+
+    /* If this fails with ENOENT/ESTALE index is stale */
+    ret = syncop_gfid_to_path(healer->this->itable, subvol, loc.gfid,
+                              (char **)&loc.path);
+    if (ret < 0)
+        goto out;
+
+    ret = syncop_inode_find(healer->this, healer->this, loc.gfid, &loc.inode,
+                            NULL, NULL);
+    if (ret < 0)
+        goto out;
+
+    ec_shd_selfheal(healer, healer->subvol, &loc, _gf_false);
+out:
+    if (ret == -ENOENT || ret == -ESTALE) {
+        gf_msg(healer->this->name, GF_LOG_DEBUG, 0, EC_MSG_HEAL_FAIL,
+               "Purging index for gfid %s:", uuid_utoa(loc.gfid));
+        ec_shd_index_purge(subvol, parent->inode, entry->d_name);
+    }
+    loc_wipe(&loc);
+
+    return 0;
+}
+
+int
+ec_shd_index_sweep(struct subvol_healer *healer)
+{
+    loc_t loc = {0};
+    ec_t *ec = NULL;
+    int ret = 0;
+    xlator_t *subvol = NULL;
+    dict_t *xdata = NULL;
+
+    ec = healer->this->private;
+    subvol = ec->xl_list[healer->subvol];
+
+    ret = ec_shd_index_inode(healer->this, subvol, &loc.inode);
+    if (ret < 0) {
+        gf_msg(healer->this->name, GF_LOG_WARNING, errno,
+               EC_MSG_INDEX_DIR_GET_FAIL, "unable to get index-dir on %s",
+               subvol->name);
+        goto out;
+    }
+
+    xdata = dict_new();
+    if (!xdata || dict_set_int32(xdata, "get-gfid-type", 1)) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    _mask_cancellation();
+    ret = syncop_mt_dir_scan(NULL, subvol, &loc, GF_CLIENT_PID_SELF_HEALD,
+                             healer, ec_shd_index_heal, xdata,
+                             ec->shd.max_threads, ec->shd.wait_qlength);
+    _unmask_cancellation();
+out:
+    if (xdata)
+        dict_unref(xdata);
+    loc_wipe(&loc);
+
+    return ret;
+}
+
+int
+ec_shd_full_heal(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent,
+                 void *data)
+{
+    struct subvol_healer *healer = data;
+    xlator_t *this = healer->this;
+    ec_t *ec = NULL;
+    loc_t loc = {0};
+    int ret = 0;
+
+    ec = this->private;
+
+    if (this->cleanup_starting) {
+        return -ENOTCONN;
+    }
+
+    if (ec->xl_up_count <= ec->fragments) {
+        return -ENOTCONN;
+    }
+    if (!ec->shd.enabled)
+        return -EBUSY;
+
+    if (gf_uuid_is_null(entry->d_stat.ia_gfid)) {
+        /* It's possible that an entry has been removed just after
+         * being seen in a directory but before getting its stat info.
+         * In this case we'll receive a NULL gfid here. Since the file
+         * doesn't exist anymore, we can safely ignore it. */
+        return 0;
+    }
+
+    loc.parent = inode_ref(parent->inode);
+    loc.name = entry->d_name;
+    gf_uuid_copy(loc.gfid, entry->d_stat.ia_gfid);
+
+    /* If this fails with ENOENT/ESTALE index is stale */
+    ret = syncop_gfid_to_path(this->itable, subvol, loc.gfid,
+                              (char **)&loc.path);
+    if (ret < 0)
+        goto out;
+
+    ret = syncop_inode_find(this, this, loc.gfid, &loc.inode, NULL, NULL);
+    if (ret < 0)
+        goto out;
+
+    ec_shd_selfheal(healer, healer->subvol, &loc, _gf_true);
+
+    ret = 0;
+
+out:
+    loc_wipe(&loc);
+    return ret;
+}
+
+int
+ec_shd_full_sweep(struct subvol_healer *healer, inode_t *inode)
+{
+    ec_t *ec = NULL;
+    loc_t loc = {0};
+    int ret = -1;
+
+    ec = healer->this->private;
+    loc.inode = inode;
+    _mask_cancellation();
+    ret = syncop_ftw(ec->xl_list[healer->subvol], &loc,
+                     GF_CLIENT_PID_SELF_HEALD, healer, ec_shd_full_heal);
+    _unmask_cancellation();
+    return ret;
+}
+
+void *
+ec_shd_index_healer(void *data)
+{
+    struct subvol_healer *healer = NULL;
+    xlator_t *this = NULL;
+    int run = 0;
+
+    healer = data;
+    THIS = this = healer->this;
+    ec_t *ec = this->private;
+
+    for (;;) {
+        run = ec_shd_healer_wait(healer);
+        if (run == -1)
+            break;
+
+        if (ec->xl_up_count > ec->fragments) {
+            gf_msg_debug(this->name, 0, "starting index sweep on subvol %s",
+                         ec_subvol_name(this, healer->subvol));
+            ec_shd_index_sweep(healer);
+        }
+        gf_msg_debug(this->name, 0, "finished index sweep on subvol %s",
+                     ec_subvol_name(this, healer->subvol));
+    }
+
+    return NULL;
+}
+
+void *
+ec_shd_full_healer(void *data)
+{
+    struct subvol_healer *healer = NULL;
+    xlator_t *this = NULL;
+    loc_t rootloc = {0};
+
+    int run = 0;
+
+    healer = data;
+    THIS = this = healer->this;
+    ec_t *ec = this->private;
+
+    rootloc.inode = this->itable->root;
+    for (;;) {
+        run = ec_shd_healer_wait(healer);
+        if (run < 0) {
+            break;
+        } else if (run == 0) {
+            continue;
+        }
+
+        if (ec->xl_up_count > ec->fragments) {
+            gf_msg(this->name, GF_LOG_INFO, 0, EC_MSG_FULL_SWEEP_START,
+                   "starting full sweep on subvol %s",
+                   ec_subvol_name(this, healer->subvol));
+
+            ec_shd_selfheal(healer, healer->subvol, &rootloc, _gf_true);
+            ec_shd_full_sweep(healer, this->itable->root);
+        }
+
+        gf_msg(this->name, GF_LOG_INFO, 0, EC_MSG_FULL_SWEEP_STOP,
+               "finished full sweep on subvol %s",
+               ec_subvol_name(this, healer->subvol));
+    }
+
+    return NULL;
+}
+
+int
+ec_shd_healer_init(xlator_t *this, struct subvol_healer *healer)
+{
+    int ret = 0;
+
+    ret = pthread_mutex_init(&healer->mutex, NULL);
+    if (ret)
+        goto out;
+
+    ret = pthread_cond_init(&healer->cond, NULL);
+    if (ret)
+        goto out;
+
+    healer->this = this;
+    healer->running = _gf_false;
+    healer->rerun = _gf_false;
+out:
+    return ret;
+}
+
+int
+ec_shd_healer_spawn(xlator_t *this, struct subvol_healer *healer,
+                    void *(threadfn)(void *))
+{
+    int ret = 0;
+
+    pthread_mutex_lock(&healer->mutex);
+    {
+        if (healer->running) {
+            pthread_cond_signal(&healer->cond);
+        } else {
+            ret = gf_thread_create(&healer->thread, NULL, threadfn, healer,
+                                   "ecshd");
+            if (ret)
+                goto unlock;
+            healer->running = 1;
+        }
+
+        healer->rerun = 1;
+    }
+unlock:
+    pthread_mutex_unlock(&healer->mutex);
+
+    return ret;
+}
+
+int
+ec_shd_full_healer_spawn(xlator_t *this, int subvol)
+{
+    if (xlator_is_cleanup_starting(this))
+        return -1;
+
+    return ec_shd_healer_spawn(this, NTH_FULL_HEALER(this, subvol),
+                               ec_shd_full_healer);
+}
+
+int
+ec_shd_index_healer_spawn(xlator_t *this, int subvol)
+{
+    if (xlator_is_cleanup_starting(this))
+        return -1;
+
+    return ec_shd_healer_spawn(this, NTH_INDEX_HEALER(this, subvol),
+                               ec_shd_index_healer);
+}
+
+void
+ec_shd_index_healer_wake(ec_t *ec)
+{
+    int32_t i;
+
+    for (i = 0; i < ec->nodes; i++) {
+        if (((ec->xl_up >> i) & 1) != 0) {
+            ec_shd_index_healer_spawn(ec->xl, i);
+        }
+    }
+}
+
+int
+ec_selfheal_daemon_init(xlator_t *this)
+{
+    ec_t *ec = NULL;
+    ec_self_heald_t *shd = NULL;
+    int ret = -1;
+    int i = 0;
+
+    ec = this->private;
+    shd = &ec->shd;
+
+    shd->index_healers = GF_CALLOC(sizeof(*shd->index_healers), ec->nodes,
+                                   ec_mt_subvol_healer_t);
+    if (!shd->index_healers)
+        goto out;
+
+    for (i = 0; i < ec->nodes; i++) {
+        shd->index_healers[i].subvol = i;
+        ret = ec_shd_healer_init(this, &shd->index_healers[i]);
+        if (ret)
+            goto out;
+    }
+
+    shd->full_healers = GF_CALLOC(sizeof(*shd->full_healers), ec->nodes,
+                                  ec_mt_subvol_healer_t);
+    if (!shd->full_healers)
+        goto out;
+
+    for (i = 0; i < ec->nodes; i++) {
+        shd->full_healers[i].subvol = i;
+        ret = ec_shd_healer_init(this, &shd->full_healers[i]);
+        if (ret)
+            goto out;
+    }
+
+    ret = 0;
+out:
+    return ret;
+}
+
+int
+ec_heal_op(xlator_t *this, dict_t *output, gf_xl_afr_op_t op, int xl_id)
+{
+    char key[64] = {0};
+    int op_ret = 0;
+    ec_t *ec = NULL;
+    int i = 0;
+    GF_UNUSED int ret = 0;
+
+    ec = this->private;
+
+    op_ret = -1;
+    for (i = 0; i < ec->nodes; i++) {
+        snprintf(key, sizeof(key), "%d-%d-status", xl_id, i);
+
+        if (((ec->xl_up >> i) & 1) == 0) {
+            ret = dict_set_str(output, key, "Brick is not connected");
+        } else if (!ec->up) {
+            ret = dict_set_str(output, key, "Disperse subvolume is not up");
+        } else if (!ec_shd_is_subvol_local(this, i)) {
+            ret = dict_set_str(output, key, "Brick is remote");
+        } else {
+            ret = dict_set_str(output, key, "Started self-heal");
+            if (op == GF_SHD_OP_HEAL_FULL) {
+                ec_shd_full_healer_spawn(this, i);
+            } else if (op == GF_SHD_OP_HEAL_INDEX) {
+                ec_shd_index_healer_spawn(this, i);
+            }
+            op_ret = 0;
+        }
+    }
+    return op_ret;
+}
+
+int
+ec_xl_op(xlator_t *this, dict_t *input, dict_t *output)
+{
+    gf_xl_afr_op_t op = GF_SHD_OP_INVALID;
+    int ret = 0;
+    int xl_id = 0;
+
+    ret = dict_get_int32(input, "xl-op", (int32_t *)&op);
+    if (ret)
+        goto out;
+
+    ret = dict_get_int32(input, this->name, &xl_id);
+    if (ret)
+        goto out;
+
+    ret = dict_set_int32(output, this->name, xl_id);
+    if (ret)
+        goto out;
+
+    switch (op) {
+        case GF_SHD_OP_HEAL_FULL:
+            ret = ec_heal_op(this, output, op, xl_id);
+            break;
+
+        case GF_SHD_OP_HEAL_INDEX:
+            ret = ec_heal_op(this, output, op, xl_id);
+            break;
+
+        default:
+            ret = -1;
+            break;
+    }
+out:
+    dict_del(output, this->name);
+    return ret;
+}
+
+void
+ec_destroy_healer_object(xlator_t *this, struct subvol_healer *healer)
+{
+    if (!healer)
+        return;
+
+    pthread_cond_destroy(&healer->cond);
+    pthread_mutex_destroy(&healer->mutex);
+}
+
+void
+ec_selfheal_daemon_fini(xlator_t *this)
+{
+    struct subvol_healer *healer = NULL;
+    ec_self_heald_t *shd = NULL;
+    ec_t *priv = NULL;
+    int i = 0;
+
+    priv = this->private;
+    if (!priv)
+        return;
+
+    shd = &priv->shd;
+    if (!shd->iamshd)
+        return;
+
+    for (i = 0; i < priv->nodes; i++) {
+        healer = &shd->index_healers[i];
+        ec_destroy_healer_object(this, healer);
+
+        healer = &shd->full_healers[i];
+        ec_destroy_healer_object(this, healer);
+    }
+
+    GF_FREE(shd->index_healers);
+    GF_FREE(shd->full_healers);
+}
diff --git a/xlators/cluster/ec/src/ec-heald.h b/xlators/cluster/ec/src/ec-heald.h
new file mode 100644
index 00000000000..6c7da4edc10
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-heald.h
@@ -0,0 +1,30 @@
+/*
+  Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_HEALD_H__
+#define __EC_HEALD_H__
+
+#include "ec-types.h"           // for ec_t
+#include "glusterfs/dict.h"     // for dict_t
+#include "glusterfs/globals.h"  // for xlator_t
+
+int
+ec_xl_op(xlator_t *this, dict_t *input, dict_t *output);
+
+int
+ec_selfheal_daemon_init(xlator_t *this);
+
+void
+ec_shd_index_healer_wake(ec_t *ec);
+
+void
+ec_selfheal_daemon_fini(xlator_t *this);
+
+#endif /* __EC_HEALD_H__ */
diff --git a/xlators/cluster/ec/src/ec-helpers.c b/xlators/cluster/ec/src/ec-helpers.c
new file mode 100644
index 00000000000..48f54475e01
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-helpers.c
@@ -0,0 +1,867 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <libgen.h>
+
+#include <glusterfs/byte-order.h>
+
+#include "ec.h"
+#include "ec-mem-types.h"
+#include "ec-messages.h"
+#include "ec-fops.h"
+#include "ec-method.h"
+#include "ec-helpers.h"
+
+static const char *ec_fop_list[] = {[-EC_FOP_HEAL] = "HEAL"};
+
+const char *
+ec_bin(char *str, size_t size, uint64_t value, int32_t digits)
+{
+    str += size;
+
+    if (size-- < 1) {
+        goto failed;
+    }
+    *--str = 0;
+
+    while ((value != 0) || (digits > 0)) {
+        if (size-- < 1) {
+            goto failed;
+        }
+        *--str = '0' + (value & 1);
+        digits--;
+        value >>= 1;
+    }
+
+    return str;
+
+failed:
+    return "<buffer too small>";
+}
+
+const char *
+ec_fop_name(int32_t id)
+{
+    if (id >= 0) {
+        return gf_fop_list[id];
+    }
+
+    return ec_fop_list[-id];
+}
+
+void
+ec_trace(const char *event, ec_fop_data_t *fop, const char *fmt, ...)
+{
+    char str1[32], str2[32], str3[32];
+    char *msg;
+    ec_t *ec = fop->xl->private;
+    va_list args;
+    int32_t ret;
+
+    va_start(args, fmt);
+    ret = vasprintf(&msg, fmt, args);
+    va_end(args);
+
+    if (ret < 0) {
+        msg = "<memory allocation error>";
+    }
+
+    gf_msg_trace("ec", 0,
+                 "%s(%s) %p(%p) [refs=%d, winds=%d, jobs=%d] "
+                 "frame=%p/%p, min/exp=%d/%d, err=%d state=%d "
+                 "{%s:%s:%s} %s",
+                 event, ec_fop_name(fop->id), fop, fop->parent, fop->refs,
+                 fop->winds, fop->jobs, fop->req_frame, fop->frame,
+                 fop->minimum, fop->expected, fop->error, fop->state,
+                 ec_bin(str1, sizeof(str1), fop->mask, ec->nodes),
+                 ec_bin(str2, sizeof(str2), fop->remaining, ec->nodes),
+                 ec_bin(str3, sizeof(str3), fop->good, ec->nodes), msg);
+
+    if (ret >= 0) {
+        free(msg);
+    }
+}
+
+int32_t
+ec_bits_consume(uint64_t *n)
+{
+    uint64_t tmp;
+
+    tmp = *n;
+    tmp &= -tmp;
+    *n ^= tmp;
+
+    return gf_bits_index(tmp);
+}
+
+size_t
+ec_iov_copy_to(void *dst, struct iovec *vector, int32_t count, off_t offset,
+               size_t size)
+{
+    int32_t i = 0;
+    size_t total = 0, len = 0;
+
+    while (i < count) {
+        if (offset < vector[i].iov_len) {
+            while ((i < count) && (size > 0)) {
+                len = size;
+                if (len > vector[i].iov_len - offset) {
+                    len = vector[i].iov_len - offset;
+                }
+                memcpy(dst, vector[i++].iov_base + offset, len);
+                offset = 0;
+                dst += len;
+                total += len;
+                size -= len;
+            }
+
+            break;
+        }
+
+        offset -= vector[i].iov_len;
+        i++;
+    }
+
+    return total;
+}
+
+int32_t
+ec_buffer_alloc(xlator_t *xl, size_t size, struct iobref **piobref, void **ptr)
+{
+    struct iobref *iobref = NULL;
+    struct iobuf *iobuf = NULL;
+    int32_t ret = -ENOMEM;
+
+    iobuf = iobuf_get_page_aligned(xl->ctx->iobuf_pool, size,
+                                   EC_METHOD_WORD_SIZE);
+    if (iobuf == NULL) {
+        goto out;
+    }
+
+    iobref = *piobref;
+    if (iobref == NULL) {
+        iobref = iobref_new();
+        if (iobref == NULL) {
+            goto out;
+        }
+    }
+
+    ret = iobref_add(iobref, iobuf);
+    if (ret != 0) {
+        if (iobref != *piobref) {
+            iobref_unref(iobref);
+        }
+        iobref = NULL;
+
+        goto out;
+    }
+
+    GF_ASSERT(EC_ALIGN_CHECK(iobuf->ptr, EC_METHOD_WORD_SIZE));
+
+    *ptr = iobuf->ptr;
+
+out:
+    if (iobuf != NULL) {
+        iobuf_unref(iobuf);
+    }
+
+    if (iobref != NULL) {
+        *piobref = iobref;
+    }
+
+    return ret;
+}
+
+int32_t
+ec_dict_set_array(dict_t *dict, char *key, uint64_t value[], int32_t size)
+{
+    int ret = -1;
+    uint64_t *ptr = NULL;
+    int32_t vindex;
+
+    if (value == NULL) {
+        return -EINVAL;
+    }
+
+    ptr = GF_MALLOC(sizeof(uint64_t) * size, gf_common_mt_char);
+    if (ptr == NULL) {
+        return -ENOMEM;
+    }
+    for (vindex = 0; vindex < size; vindex++) {
+        ptr[vindex] = hton64(value[vindex]);
+    }
+    ret = dict_set_bin(dict, key, ptr, sizeof(uint64_t) * size);
+    if (ret)
+        GF_FREE(ptr);
+    return ret;
+}
+
+int32_t
+ec_dict_get_array(dict_t *dict, char *key, uint64_t value[], int32_t size)
+{
+    void *ptr;
+    int32_t len;
+    int32_t vindex;
+    int32_t old_size = 0;
+    int32_t err;
+
+    if (dict == NULL) {
+        return -EINVAL;
+    }
+    err = dict_get_ptr_and_len(dict, key, &ptr, &len);
+    if (err != 0) {
+        return err;
+    }
+
+    if (len > (size * sizeof(uint64_t)) || (len % sizeof(uint64_t))) {
+        return -EINVAL;
+    }
+
+    /* 3.6 version ec would have stored version in 64 bit. In that case treat
+     * metadata versions same as data*/
+    old_size = min(size, len / sizeof(uint64_t));
+    for (vindex = 0; vindex < old_size; vindex++) {
+        value[vindex] = ntoh64(*((uint64_t *)ptr + vindex));
+    }
+
+    if (old_size < size) {
+        for (vindex = old_size; vindex < size; vindex++) {
+            value[vindex] = value[old_size - 1];
+        }
+    }
+
+    return 0;
+}
+
+int32_t
+ec_dict_del_array(dict_t *dict, char *key, uint64_t value[], int32_t size)
+{
+    int ret = 0;
+
+    ret = ec_dict_get_array(dict, key, value, size);
+    if (ret == 0)
+        dict_del(dict, key);
+
+    return ret;
+}
+
+int32_t
+ec_dict_set_number(dict_t *dict, char *key, uint64_t value)
+{
+    int ret = -1;
+    uint64_t *ptr;
+
+    ptr = GF_MALLOC(sizeof(value), gf_common_mt_char);
+    if (ptr == NULL) {
+        return -ENOMEM;
+    }
+
+    *ptr = hton64(value);
+
+    ret = dict_set_bin(dict, key, ptr, sizeof(value));
+    if (ret)
+        GF_FREE(ptr);
+
+    return ret;
+}
+
+int32_t
+ec_dict_del_number(dict_t *dict, char *key, uint64_t *value)
+{
+    void *ptr;
+    int32_t len, err;
+
+    if (dict == NULL) {
+        return -EINVAL;
+    }
+    err = dict_get_ptr_and_len(dict, key, &ptr, &len);
+    if (err != 0) {
+        return err;
+    }
+    if (len != sizeof(uint64_t)) {
+        return -EINVAL;
+    }
+
+    *value = ntoh64(*(uint64_t *)ptr);
+
+    dict_del(dict, key);
+
+    return 0;
+}
+
+int32_t
+ec_dict_set_config(dict_t *dict, char *key, ec_config_t *config)
+{
+    int ret = -1;
+    uint64_t *ptr, data;
+
+    if (config->version > EC_CONFIG_VERSION) {
+        gf_msg("ec", GF_LOG_ERROR, EINVAL, EC_MSG_UNSUPPORTED_VERSION,
+               "Trying to store an unsupported config "
+               "version (%u)",
+               config->version);
+
+        return -EINVAL;
+    }
+
+    ptr = GF_MALLOC(sizeof(uint64_t), gf_common_mt_char);
+    if (ptr == NULL) {
+        return -ENOMEM;
+    }
+
+    data = ((uint64_t)config->version) << 56;
+    data |= ((uint64_t)config->algorithm) << 48;
+    data |= ((uint64_t)config->gf_word_size) << 40;
+    data |= ((uint64_t)config->bricks) << 32;
+    data |= ((uint64_t)config->redundancy) << 24;
+    data |= config->chunk_size;
+
+    *ptr = hton64(data);
+
+    ret = dict_set_bin(dict, key, ptr, sizeof(uint64_t));
+    if (ret)
+        GF_FREE(ptr);
+
+    return ret;
+}
+
+int32_t
+ec_dict_del_config(dict_t *dict, char *key, ec_config_t *config)
+{
+    void *ptr;
+    uint64_t data;
+    int32_t len, err;
+
+    if (dict == NULL) {
+        return -EINVAL;
+    }
+    err = dict_get_ptr_and_len(dict, key, &ptr, &len);
+    if (err != 0) {
+        return err;
+    }
+    if (len != sizeof(uint64_t)) {
+        return -EINVAL;
+    }
+
+    data = ntoh64(*(uint64_t *)ptr);
+    /* Currently we need to get the config xattr for entries of type IA_INVAL.
+     * These entries can later become IA_DIR entries (after inode_link()),
+     * which don't have a config xattr. However, since the xattr is requested
+     * using an xattrop() fop, it will always return a config full of 0's
+     * instead of saying that it doesn't exist.
+     *
+     * We need to filter out this case and consider that a config xattr == 0 is
+     * the same as a non-existent xattr. Otherwise ec_config_check() will fail.
+     */
+    if (data == 0) {
+        return -ENODATA;
+    }
+
+    config->version = (data >> 56) & 0xff;
+    if (config->version > EC_CONFIG_VERSION) {
+        gf_msg("ec", GF_LOG_ERROR, EINVAL, EC_MSG_UNSUPPORTED_VERSION,
+               "Found an unsupported config version (%u)", config->version);
+
+        return -EINVAL;
+    }
+
+    config->algorithm = (data >> 48) & 0xff;
+    config->gf_word_size = (data >> 40) & 0xff;
+    config->bricks = (data >> 32) & 0xff;
+    config->redundancy = (data >> 24) & 0xff;
+    config->chunk_size = data & 0xffffff;
+
+    dict_del(dict, key);
+
+    return 0;
+}
+
+gf_boolean_t
+ec_loc_gfid_check(xlator_t *xl, uuid_t dst, uuid_t src)
+{
+    if (gf_uuid_is_null(src)) {
+        return _gf_true;
+    }
+
+    if (gf_uuid_is_null(dst)) {
+        gf_uuid_copy(dst, src);
+
+        return _gf_true;
+    }
+
+    if (gf_uuid_compare(dst, src) != 0) {
+        gf_msg(xl->name, GF_LOG_WARNING, 0, EC_MSG_GFID_MISMATCH,
+               "Mismatching GFID's in loc");
+
+        return _gf_false;
+    }
+
+    return _gf_true;
+}
+
+int32_t
+ec_loc_setup_inode(xlator_t *xl, inode_table_t *table, loc_t *loc)
+{
+    int32_t ret = -EINVAL;
+
+    if (loc->inode != NULL) {
+        if (!ec_loc_gfid_check(xl, loc->gfid, loc->inode->gfid)) {
+            goto out;
+        }
+    } else if (table != NULL) {
+        if (!gf_uuid_is_null(loc->gfid)) {
+            loc->inode = inode_find(table, loc->gfid);
+        } else if (loc->path && strchr(loc->path, '/')) {
+            loc->inode = inode_resolve(table, (char *)loc->path);
+        }
+    }
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+int32_t
+ec_loc_setup_parent(xlator_t *xl, inode_table_t *table, loc_t *loc)
+{
+    char *path, *parent;
+    int32_t ret = -EINVAL;
+
+    if (loc->parent != NULL) {
+        if (!ec_loc_gfid_check(xl, loc->pargfid, loc->parent->gfid)) {
+            goto out;
+        }
+    } else if (table != NULL) {
+        if (!gf_uuid_is_null(loc->pargfid)) {
+            loc->parent = inode_find(table, loc->pargfid);
+        } else if (loc->path && strchr(loc->path, '/')) {
+            path = gf_strdup(loc->path);
+            if (path == NULL) {
+                gf_msg(xl->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                       "Unable to duplicate path '%s'", loc->path);
+
+                ret = -ENOMEM;
+
+                goto out;
+            }
+            parent = dirname(path);
+            loc->parent = inode_resolve(table, parent);
+            if (loc->parent != NULL) {
+                gf_uuid_copy(loc->pargfid, loc->parent->gfid);
+            }
+            GF_FREE(path);
+        }
+    }
+
+    /* If 'pargfid' has not been determined, clear 'name' to avoid resolutions
+       based on <gfid:pargfid>/name. */
+    if (gf_uuid_is_null(loc->pargfid)) {
+        loc->name = NULL;
+    }
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+int32_t
+ec_loc_setup_path(xlator_t *xl, loc_t *loc)
+{
+    static uuid_t root = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1};
+    char *name;
+    int32_t ret = -EINVAL;
+
+    if (loc->path != NULL) {
+        name = strrchr(loc->path, '/');
+        if (name == NULL) {
+            /* Allow gfid paths: <gfid:...> */
+            if (strncmp(loc->path, "<gfid:", 6) == 0) {
+                ret = 0;
+            }
+            goto out;
+        }
+        if (name == loc->path) {
+            if (name[1] == 0) {
+                if (!ec_loc_gfid_check(xl, loc->gfid, root)) {
+                    goto out;
+                }
+            } else {
+                if (!ec_loc_gfid_check(xl, loc->pargfid, root)) {
+                    goto out;
+                }
+            }
+        }
+        name++;
+
+        if (loc->name != NULL) {
+            if (strcmp(loc->name, name) != 0) {
+                gf_msg(xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_INVALID_LOC_NAME,
+                       "Invalid name '%s' in loc", loc->name);
+
+                goto out;
+            }
+        } else {
+            loc->name = name;
+        }
+    }
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+int32_t
+ec_loc_parent(xlator_t *xl, loc_t *loc, loc_t *parent)
+{
+    inode_table_t *table = NULL;
+    char *str = NULL;
+    int32_t ret = -ENOMEM;
+
+    memset(parent, 0, sizeof(loc_t));
+
+    if (loc->parent != NULL) {
+        table = loc->parent->table;
+        parent->inode = inode_ref(loc->parent);
+    } else if (loc->inode != NULL) {
+        table = loc->inode->table;
+    }
+    if (!gf_uuid_is_null(loc->pargfid)) {
+        gf_uuid_copy(parent->gfid, loc->pargfid);
+    }
+    if (loc->path && strchr(loc->path, '/')) {
+        str = gf_strdup(loc->path);
+        if (str == NULL) {
+            gf_msg(xl->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Unable to duplicate path '%s'", loc->path);
+
+            goto out;
+        }
+        parent->path = gf_strdup(dirname(str));
+        if (parent->path == NULL) {
+            gf_msg(xl->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Unable to duplicate path '%s'", dirname(str));
+
+            goto out;
+        }
+    }
+
+    ret = ec_loc_setup_path(xl, parent);
+    if (ret == 0) {
+        ret = ec_loc_setup_inode(xl, table, parent);
+    }
+    if (ret == 0) {
+        ret = ec_loc_setup_parent(xl, table, parent);
+    }
+    if (ret != 0) {
+        goto out;
+    }
+
+    if ((parent->inode == NULL) && (parent->path == NULL) &&
+        gf_uuid_is_null(parent->gfid)) {
+        gf_msg(xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_LOC_PARENT_INODE_MISSING,
+               "Parent inode missing for loc_t");
+
+        ret = -EINVAL;
+
+        goto out;
+    }
+
+    ret = 0;
+
+out:
+    GF_FREE(str);
+
+    if (ret != 0) {
+        loc_wipe(parent);
+    }
+
+    return ret;
+}
+
+int32_t
+ec_loc_update(xlator_t *xl, loc_t *loc, inode_t *inode, struct iatt *iatt)
+{
+    inode_table_t *table = NULL;
+    int32_t ret = -EINVAL;
+
+    if (inode != NULL) {
+        table = inode->table;
+        if (loc->inode != inode) {
+            if (loc->inode != NULL) {
+                inode_unref(loc->inode);
+            }
+            loc->inode = inode_ref(inode);
+            gf_uuid_copy(loc->gfid, inode->gfid);
+        }
+    } else if (loc->inode != NULL) {
+        table = loc->inode->table;
+    } else if (loc->parent != NULL) {
+        table = loc->parent->table;
+    }
+
+    if (iatt != NULL) {
+        if (!ec_loc_gfid_check(xl, loc->gfid, iatt->ia_gfid)) {
+            goto out;
+        }
+    }
+
+    ret = ec_loc_setup_path(xl, loc);
+    if (ret == 0) {
+        ret = ec_loc_setup_inode(xl, table, loc);
+    }
+    if (ret == 0) {
+        ret = ec_loc_setup_parent(xl, table, loc);
+    }
+    if (ret != 0) {
+        goto out;
+    }
+
+out:
+    return ret;
+}
+
+int32_t
+ec_loc_from_fd(xlator_t *xl, loc_t *loc, fd_t *fd)
+{
+    ec_fd_t *ctx;
+    int32_t ret = -ENOMEM;
+
+    memset(loc, 0, sizeof(*loc));
+
+    ctx = ec_fd_get(fd, xl);
+    if (ctx != NULL) {
+        if (loc_copy(loc, &ctx->loc) != 0) {
+            goto out;
+        }
+    }
+
+    ret = ec_loc_update(xl, loc, fd->inode, NULL);
+    if (ret != 0) {
+        goto out;
+    }
+
+out:
+    if (ret != 0) {
+        loc_wipe(loc);
+    }
+
+    return ret;
+}
+
+int32_t
+ec_loc_from_loc(xlator_t *xl, loc_t *dst, loc_t *src)
+{
+    int32_t ret = -ENOMEM;
+
+    memset(dst, 0, sizeof(*dst));
+
+    if (loc_copy(dst, src) != 0) {
+        goto out;
+    }
+
+    ret = ec_loc_update(xl, dst, NULL, NULL);
+    if (ret != 0) {
+        goto out;
+    }
+
+out:
+    if (ret != 0) {
+        loc_wipe(dst);
+    }
+
+    return ret;
+}
+
+void
+ec_owner_set(call_frame_t *frame, void *owner)
+{
+    set_lk_owner_from_ptr(&frame->root->lk_owner, owner);
+}
+
+void
+ec_owner_copy(call_frame_t *frame, gf_lkowner_t *owner)
+{
+    lk_owner_copy(&frame->root->lk_owner, owner);
+}
+
+static void
+ec_stripe_cache_init(ec_t *ec, ec_inode_t *ctx)
+{
+    ec_stripe_list_t *stripe_cache = NULL;
+
+    stripe_cache = &(ctx->stripe_cache);
+    if (stripe_cache->max == 0) {
+        stripe_cache->max = ec->stripe_cache;
+    }
+}
+
+ec_inode_t *
+__ec_inode_get(inode_t *inode, xlator_t *xl)
+{
+    ec_inode_t *ctx = NULL;
+    uint64_t value = 0;
+
+    if ((__inode_ctx_get(inode, xl, &value) != 0) || (value == 0)) {
+        ctx = GF_MALLOC(sizeof(*ctx), ec_mt_ec_inode_t);
+        if (ctx != NULL) {
+            memset(ctx, 0, sizeof(*ctx));
+            INIT_LIST_HEAD(&ctx->heal);
+            INIT_LIST_HEAD(&ctx->stripe_cache.lru);
+            ctx->heal_count = 0;
+            value = (uint64_t)(uintptr_t)ctx;
+            if (__inode_ctx_set(inode, xl, &value) != 0) {
+                GF_FREE(ctx);
+
+                return NULL;
+            }
+        }
+    } else {
+        ctx = (ec_inode_t *)(uintptr_t)value;
+    }
+    if (ctx)
+        ec_stripe_cache_init(xl->private, ctx);
+
+    return ctx;
+}
+
+ec_inode_t *
+ec_inode_get(inode_t *inode, xlator_t *xl)
+{
+    ec_inode_t *ctx = NULL;
+
+    LOCK(&inode->lock);
+
+    ctx = __ec_inode_get(inode, xl);
+
+    UNLOCK(&inode->lock);
+
+    return ctx;
+}
+
+ec_fd_t *
+__ec_fd_get(fd_t *fd, xlator_t *xl)
+{
+    int i = 0;
+    ec_fd_t *ctx = NULL;
+    ec_inode_t *ictx = NULL;
+    uint64_t value = 0;
+    ec_t *ec = xl->private;
+
+    if ((__fd_ctx_get(fd, xl, &value) != 0) || (value == 0)) {
+        ctx = GF_MALLOC(sizeof(*ctx) + (sizeof(ec_fd_status_t) * ec->nodes),
+                        ec_mt_ec_fd_t);
+        if (ctx != NULL) {
+            memset(ctx, 0, sizeof(*ctx));
+
+            for (i = 0; i < ec->nodes; i++) {
+                if (fd_is_anonymous(fd)) {
+                    ctx->fd_status[i] = EC_FD_OPENED;
+                } else {
+                    ctx->fd_status[i] = EC_FD_NOT_OPENED;
+                }
+            }
+
+            value = (uint64_t)(uintptr_t)ctx;
+            if (__fd_ctx_set(fd, xl, value) != 0) {
+                GF_FREE(ctx);
+                return NULL;
+            }
+            /* Only refering bad-version so no need for lock
+             * */
+            ictx = __ec_inode_get(fd->inode, xl);
+            if (ictx) {
+                ctx->bad_version = ictx->bad_version;
+            }
+        }
+    } else {
+        ctx = (ec_fd_t *)(uintptr_t)value;
+    }
+
+    /* Treat anonymous fd specially */
+    if (fd->anonymous && ctx) {
+        /* Mark the fd open for all subvolumes. */
+        ctx->open = -1;
+        /* Try to populate ctx->loc with fd->inode information. */
+        ec_loc_update(xl, &ctx->loc, fd->inode, NULL);
+    }
+
+    return ctx;
+}
+
+ec_fd_t *
+ec_fd_get(fd_t *fd, xlator_t *xl)
+{
+    ec_fd_t *ctx = NULL;
+
+    LOCK(&fd->lock);
+
+    ctx = __ec_fd_get(fd, xl);
+
+    UNLOCK(&fd->lock);
+
+    return ctx;
+}
+
+gf_boolean_t
+ec_is_internal_xattr(dict_t *dict, char *key, data_t *value, void *data)
+{
+    if (key && (strncmp(key, EC_XATTR_PREFIX, SLEN(EC_XATTR_PREFIX)) == 0))
+        return _gf_true;
+
+    return _gf_false;
+}
+
+void
+ec_filter_internal_xattrs(dict_t *xattr)
+{
+    dict_foreach_match(xattr, ec_is_internal_xattr, NULL,
+                       dict_remove_foreach_fn, NULL);
+}
+
+gf_boolean_t
+ec_is_data_fop(glusterfs_fop_t fop)
+{
+    switch (fop) {
+        case GF_FOP_WRITE:
+        case GF_FOP_TRUNCATE:
+        case GF_FOP_FTRUNCATE:
+        case GF_FOP_FALLOCATE:
+        case GF_FOP_DISCARD:
+        case GF_FOP_ZEROFILL:
+            return _gf_true;
+        default:
+            return _gf_false;
+    }
+    return _gf_false;
+}
+/*
+gf_boolean_t
+ec_is_metadata_fop (int32_t lock_kind, glusterfs_fop_t fop)
+{
+        if (lock_kind == EC_LOCK_ENTRY) {
+                return _gf_false;
+        }
+
+        switch (fop) {
+        case GF_FOP_SETATTR:
+        case GF_FOP_FSETATTR:
+        case GF_FOP_SETXATTR:
+        case GF_FOP_FSETXATTR:
+        case GF_FOP_REMOVEXATTR:
+        case GF_FOP_FREMOVEXATTR:
+                return _gf_true;
+        default:
+                return _gf_false;
+        }
+        return _gf_false;
+}*/
diff --git a/xlators/cluster/ec/src/ec-helpers.h b/xlators/cluster/ec/src/ec-helpers.h
new file mode 100644
index 00000000000..015db514e05
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-helpers.h
@@ -0,0 +1,200 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_HELPERS_H__
+#define __EC_HELPERS_H__
+
+#include "ec-types.h"
+
+#define EC_ERR(_x) ((void *)-(intptr_t)(_x))
+#define EC_IS_ERR(_x) (((uintptr_t)(_x) & ~0xfffULL) == ~0xfffULL)
+#define EC_GET_ERR(_x) ((int32_t)(intptr_t)(_x))
+
+#define EC_ALIGN_CHECK(_ptr, _align) ((((uintptr_t)(_ptr)) & ((_align)-1)) == 0)
+
+const char *
+ec_bin(char *str, size_t size, uint64_t value, int32_t digits);
+const char *
+ec_fop_name(int32_t id);
+void
+ec_trace(const char *event, ec_fop_data_t *fop, const char *fmt, ...);
+int32_t
+ec_bits_consume(uint64_t *n);
+size_t
+ec_iov_copy_to(void *dst, struct iovec *vector, int32_t count, off_t offset,
+               size_t size);
+int32_t
+ec_buffer_alloc(xlator_t *xl, size_t size, struct iobref **piobref, void **ptr);
+int32_t
+ec_dict_set_array(dict_t *dict, char *key, uint64_t *value, int32_t size);
+int32_t
+ec_dict_get_array(dict_t *dict, char *key, uint64_t value[], int32_t size);
+
+int32_t
+ec_dict_del_array(dict_t *dict, char *key, uint64_t *value, int32_t size);
+int32_t
+ec_dict_set_number(dict_t *dict, char *key, uint64_t value);
+int32_t
+ec_dict_del_number(dict_t *dict, char *key, uint64_t *value);
+int32_t
+ec_dict_set_config(dict_t *dict, char *key, ec_config_t *config);
+int32_t
+ec_dict_del_config(dict_t *dict, char *key, ec_config_t *config);
+
+int32_t
+ec_loc_parent(xlator_t *xl, loc_t *loc, loc_t *parent);
+int32_t
+ec_loc_update(xlator_t *xl, loc_t *loc, inode_t *inode, struct iatt *iatt);
+
+int32_t
+ec_loc_from_fd(xlator_t *xl, loc_t *loc, fd_t *fd);
+int32_t
+ec_loc_from_loc(xlator_t *xl, loc_t *dst, loc_t *src);
+
+void
+ec_owner_set(call_frame_t *frame, void *owner);
+void
+ec_owner_copy(call_frame_t *frame, gf_lkowner_t *owner);
+
+ec_inode_t *
+__ec_inode_get(inode_t *inode, xlator_t *xl);
+ec_inode_t *
+ec_inode_get(inode_t *inode, xlator_t *xl);
+ec_fd_t *
+__ec_fd_get(fd_t *fd, xlator_t *xl);
+ec_fd_t *
+ec_fd_get(fd_t *fd, xlator_t *xl);
+
+static inline uint32_t
+ec_adjust_size_down(ec_t *ec, uint64_t *value, gf_boolean_t scale)
+{
+    uint64_t head, tmp;
+
+    tmp = *value;
+    head = tmp % ec->stripe_size;
+    tmp -= head;
+
+    if (scale) {
+        tmp /= ec->fragments;
+    }
+
+    *value = tmp;
+
+    return (uint32_t)head;
+}
+
+/* This function can cause an overflow if the passed value is too near to the
+ * uint64_t limit. If this happens, it returns the tail in negative form and
+ * the value is set to UINT64_MAX. */
+static inline int32_t
+ec_adjust_size_up(ec_t *ec, uint64_t *value, gf_boolean_t scale)
+{
+    uint64_t tmp;
+    int32_t tail;
+
+    tmp = *value;
+    /* We first adjust the value down. This never causes overflow. */
+    tail = ec_adjust_size_down(ec, &tmp, scale);
+
+    /* If the value was already aligned, tail will be 0 and nothing else
+     * needs to be done. */
+    if (tail != 0) {
+        /* Otherwise, we need to compute the real tail and adjust the
+         * returned value to the next stripe. */
+        tail = ec->stripe_size - tail;
+        if (scale) {
+            tmp += ec->fragment_size;
+        } else {
+            tmp += ec->stripe_size;
+            /* If no scaling is requested there's a possibility of
+             * overflow. */
+            if (tmp < ec->stripe_size) {
+                tmp = UINT64_MAX;
+                tail = -tail;
+            }
+        }
+    }
+
+    *value = tmp;
+
+    return tail;
+}
+
+/* This function is equivalent to ec_adjust_size_down() but with a potentially
+ * different parameter size (off_t vs uint64_t). */
+static inline uint32_t
+ec_adjust_offset_down(ec_t *ec, off_t *value, gf_boolean_t scale)
+{
+    off_t head, tmp;
+
+    tmp = *value;
+    head = tmp % ec->stripe_size;
+    tmp -= head;
+
+    if (scale) {
+        tmp /= ec->fragments;
+    }
+
+    *value = tmp;
+
+    return (uint32_t)head;
+}
+
+/* This function is equivalent to ec_adjust_size_up() but with a potentially
+ * different parameter size (off_t vs uint64_t). */
+static inline int32_t
+ec_adjust_offset_up(ec_t *ec, off_t *value, gf_boolean_t scale)
+{
+    uint64_t tail, tmp;
+
+    /* An offset is a signed type that can only have positive values, so
+     * we take advantage of this to avoid overflows. We simply convert it
+     * to an unsigned integer and operate normally. This won't cause an
+     * overflow. Overflow is only checked when converting back to an
+     * off_t. */
+    tmp = *value;
+    tail = ec->stripe_size;
+    tail -= (tmp + tail - 1) % tail + 1;
+    tmp += tail;
+    if (scale) {
+        /* If we are scaling, we'll never get an overflow. */
+        tmp /= ec->fragments;
+    } else {
+        /* Check if there has been an overflow. */
+        if ((off_t)tmp < 0) {
+            tmp = GF_OFF_MAX;
+            tail = -tail;
+        }
+    }
+
+    *value = (off_t)tmp;
+
+    return (int32_t)tail;
+}
+
+static inline int32_t
+ec_is_power_of_2(uint32_t value)
+{
+    return (value != 0) && ((value & (value - 1)) == 0);
+}
+
+gf_boolean_t
+ec_is_internal_xattr(dict_t *dict, char *key, data_t *value, void *data);
+
+void
+ec_filter_internal_xattrs(dict_t *xattr);
+
+gf_boolean_t
+ec_is_data_fop(glusterfs_fop_t fop);
+
+int32_t
+ec_launch_replace_heal(ec_t *ec);
+
+#endif /* __EC_HELPERS_H__ */
diff --git a/xlators/cluster/ec/src/ec-inode-read.c b/xlators/cluster/ec/src/ec-inode-read.c
new file mode 100644
index 00000000000..dad5f4d7018
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-inode-read.c
@@ -0,0 +1,2046 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "ec.h"
+#include "ec-messages.h"
+#include "ec-helpers.h"
+#include "ec-common.h"
+#include "ec-combine.h"
+#include "ec-method.h"
+#include "ec-fops.h"
+
+/* FOP: access */
+
+int32_t
+ec_access_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_ACCESS, idx, op_ret,
+                               op_errno);
+    if (cbk) {
+        if (xdata)
+            cbk->xdata = dict_ref(xdata);
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_access(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_access_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->access,
+                      &fop->loc[0], fop->int32, fop->xdata);
+}
+
+int32_t
+ec_manager_access(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk = NULL;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ec_lock_prepare_inode(fop, &fop->loc[0], EC_QUERY_INFO, 0,
+                                  EC_RANGE_FULL);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_one(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            if (ec_dispatch_one_retry(fop, NULL)) {
+                return EC_STATE_DISPATCH;
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+            GF_ASSERT(cbk);
+            if (fop->cbks.access != NULL) {
+                if (cbk) {
+                    fop->cbks.access(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                     cbk->op_errno, cbk->xdata);
+                }
+            }
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            if (fop->cbks.access != NULL) {
+                fop->cbks.access(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                 NULL);
+            }
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_access(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_access_cbk_t func, void *data, loc_t *loc,
+          int32_t mask, dict_t *xdata)
+{
+    ec_cbk_t callback = {.access = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(ACCESS) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_ACCESS, EC_FLAG_LOCK_SHARED,
+                               target, fop_flags, ec_wind_access,
+                               ec_manager_access, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->int32 = mask;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/* FOP: getxattr */
+
+int32_t
+ec_combine_getxattr(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    if (!ec_dict_compare(dst->dict, src->dict)) {
+        gf_msg(fop->xl->name, GF_LOG_NOTICE, 0, EC_MSG_DICT_MISMATCH,
+               "Mismatching dictionary in "
+               "answers of 'GF_FOP_GETXATTR'");
+
+        return 0;
+    }
+
+    return 1;
+}
+
+int32_t
+ec_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_GETXATTR, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            if (dict != NULL) {
+                cbk->dict = dict_ref(dict);
+                if (cbk->dict == NULL) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                           "Failed to reference a "
+                           "dictionary.");
+
+                    goto out;
+                }
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, ec_combine_getxattr);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_getxattr(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_getxattr_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->getxattr,
+                      &fop->loc[0], fop->str[0], fop->xdata);
+}
+
+void
+ec_handle_special_xattrs(ec_fop_data_t *fop)
+{
+    ec_cbk_data_t *cbk = NULL;
+    /* Stime may not be available on all the bricks, so even if some of the
+     * subvols succeed the operation, treat it as answer.*/
+    if (fop->str[0] && fnmatch(GF_XATTR_STIME_PATTERN, fop->str[0], 0) == 0) {
+        if (!fop->answer || (fop->answer->op_ret < 0)) {
+            list_for_each_entry(cbk, &fop->cbk_list, list)
+            {
+                if (cbk->op_ret >= 0) {
+                    fop->answer = cbk;
+                    break;
+                }
+            }
+        }
+    }
+}
+
+int32_t
+ec_manager_getxattr(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            /* clear-locks commands must be done without any locks acquired
+               to avoid interferences. */
+            if ((fop->str[0] == NULL) ||
+                (strncmp(fop->str[0], GF_XATTR_CLRLK_CMD,
+                         SLEN(GF_XATTR_CLRLK_CMD)) != 0)) {
+                if (fop->fd == NULL) {
+                    ec_lock_prepare_inode(fop, &fop->loc[0], EC_QUERY_INFO, 0,
+                                          EC_RANGE_FULL);
+                } else {
+                    ec_lock_prepare_fd(fop, fop->fd, EC_QUERY_INFO, 0,
+                                       EC_RANGE_FULL);
+                }
+                ec_lock(fop);
+            }
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            if (fop->minimum == EC_MINIMUM_ALL) {
+                ec_dispatch_all(fop);
+            } else {
+                ec_dispatch_one(fop);
+            }
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            ec_handle_special_xattrs(fop);
+            if (fop->minimum == EC_MINIMUM_ALL) {
+                cbk = ec_fop_prepare_answer(fop, _gf_true);
+            } else {
+                if (ec_dispatch_one_retry(fop, &cbk)) {
+                    return EC_STATE_DISPATCH;
+                }
+            }
+            if (cbk != NULL) {
+                int32_t err;
+
+                err = ec_dict_combine(cbk, EC_COMBINE_DICT);
+                if (!ec_cbk_set_error(cbk, -err, _gf_true)) {
+                    if (cbk->xdata != NULL)
+                        ec_filter_internal_xattrs(cbk->xdata);
+
+                    if (cbk->dict != NULL)
+                        ec_filter_internal_xattrs(cbk->dict);
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.getxattr != NULL) {
+                fop->cbks.getxattr(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                   cbk->op_errno, cbk->dict, cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.getxattr != NULL) {
+                fop->cbks.getxattr(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                   NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+int32_t
+ec_getxattr_heal_cbk(call_frame_t *frame, void *cookie, xlator_t *xl,
+                     int32_t op_ret, int32_t op_errno, uintptr_t mask,
+                     uintptr_t good, uintptr_t bad, uint32_t pending,
+                     dict_t *xdata)
+{
+    fop_getxattr_cbk_t func = cookie;
+    ec_t *ec = xl->private;
+    dict_t *dict = NULL;
+    char *str;
+    char bin1[65], bin2[65];
+
+    /* We try to return the 'pending' information in xdata, but if this cannot
+     * be set, we will ignore it silently. We prefer to report the success or
+     * failure of the heal itself. */
+    if (xdata == NULL) {
+        xdata = dict_new();
+    } else {
+        dict_ref(xdata);
+    }
+    if (xdata != NULL) {
+        if (dict_set_uint32(xdata, EC_XATTR_HEAL_NEW, pending) != 0) {
+            /* dict_set_uint32() is marked as 'warn_unused_result' and gcc
+             * enforces to check the result in this case. However we don't
+             * really care if it succeeded or not. We'll just do the same.
+             *
+             * This empty 'if' avoids the warning, and it will be removed by
+             * the optimizer. */
+        }
+    }
+
+    if (op_ret >= 0) {
+        dict = dict_new();
+        if (dict == NULL) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+        } else {
+            if (gf_asprintf(&str, "Good: %s, Bad: %s",
+                            ec_bin(bin1, sizeof(bin1), good, ec->nodes),
+                            ec_bin(bin2, sizeof(bin2), mask & ~(good | bad),
+                                   ec->nodes)) < 0) {
+                dict_unref(dict);
+                dict = NULL;
+
+                op_ret = -1;
+                op_errno = ENOMEM;
+
+                goto out;
+            }
+
+            if (dict_set_dynstr(dict, EC_XATTR_HEAL, str) != 0) {
+                GF_FREE(str);
+                dict_unref(dict);
+                dict = NULL;
+
+                op_ret = -1;
+                op_errno = ENOMEM;
+
+                goto out;
+            }
+        }
+    }
+
+out:
+    func(frame, NULL, xl, op_ret, op_errno, dict, xdata);
+
+    if (dict != NULL) {
+        dict_unref(dict);
+    }
+    if (xdata != NULL) {
+        dict_unref(xdata);
+    }
+
+    return 0;
+}
+
+void
+ec_getxattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_getxattr_cbk_t func, void *data, loc_t *loc,
+            const char *name, dict_t *xdata)
+{
+    ec_cbk_t callback = {.getxattr = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(GETXATTR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    /* Special handling of an explicit self-heal request */
+    if ((name != NULL) && (strcmp(name, EC_XATTR_HEAL) == 0)) {
+        ec_heal(frame, this, target, EC_MINIMUM_ONE, ec_getxattr_heal_cbk, func,
+                loc, 0, NULL);
+
+        return;
+    }
+
+    fop = ec_fop_data_allocate(
+        frame, this, GF_FOP_GETXATTR, EC_FLAG_LOCK_SHARED, target, fop_flags,
+        ec_wind_getxattr, ec_manager_getxattr, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (name != NULL) {
+        /* In case of list-node-uuids xattr, set flag to indicate
+         * the same and use node-uuid xattr for winding fop */
+        if (XATTR_IS_NODE_UUID_LIST(name)) {
+            fop->int32 = 1;
+            fop->str[0] = gf_strdup(GF_XATTR_NODE_UUID_KEY);
+        } else {
+            fop->str[0] = gf_strdup(name);
+        }
+        if (fop->str[0] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
+
+/* FOP: fgetxattr */
+
+int32_t
+ec_fgetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_FGETXATTR, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            if (dict != NULL) {
+                cbk->dict = dict_ref(dict);
+                if (cbk->dict == NULL) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                           "Failed to reference a "
+                           "dictionary.");
+
+                    goto out;
+                }
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, ec_combine_getxattr);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_fgetxattr(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_fgetxattr_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->fgetxattr,
+                      fop->fd, fop->str[0], fop->xdata);
+}
+
+void
+ec_fgetxattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+             uint32_t fop_flags, fop_fgetxattr_cbk_t func, void *data, fd_t *fd,
+             const char *name, dict_t *xdata)
+{
+    ec_cbk_t callback = {.fgetxattr = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FGETXATTR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(
+        frame, this, GF_FOP_FGETXATTR, EC_FLAG_LOCK_SHARED, target, fop_flags,
+        ec_wind_fgetxattr, ec_manager_getxattr, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (name != NULL) {
+        fop->str[0] = gf_strdup(name);
+        if (fop->str[0] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
+
+/* FOP: open */
+
+int32_t
+ec_combine_open(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    if (dst->fd != src->fd) {
+        gf_msg(fop->xl->name, GF_LOG_NOTICE, 0, EC_MSG_FD_MISMATCH,
+               "Mismatching fd in answers "
+               "of 'GF_FOP_OPEN': %p <-> %p",
+               dst->fd, src->fd);
+
+        return 0;
+    }
+
+    return 1;
+}
+
+int32_t
+ec_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, fd_t *fd, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_OPEN, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            if (fd != NULL) {
+                cbk->fd = fd_ref(fd);
+                if (cbk->fd == NULL) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0,
+                           EC_MSG_FILE_DESC_REF_FAIL,
+                           "Failed to reference a "
+                           "file descriptor.");
+
+                    goto out;
+                }
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, ec_combine_open);
+
+        ec_update_fd_status(fd, this, idx, op_ret);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_open(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_open_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->open,
+                      &fop->loc[0], fop->int32, fop->fd, fop->xdata);
+}
+
+int32_t
+ec_open_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+    ec_fop_data_t *fop = cookie;
+    int32_t error = 0;
+
+    fop = fop->data;
+    if (op_ret >= 0) {
+        fop->answer->iatt[0] = *postbuf;
+    } else {
+        error = op_errno;
+    }
+
+    ec_resume(fop, error);
+
+    return 0;
+}
+
+int32_t
+ec_manager_open(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+    ec_fd_t *ctx;
+    int32_t err;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            LOCK(&fop->fd->lock);
+
+            ctx = __ec_fd_get(fop->fd, fop->xl);
+            if (ctx == NULL) {
+                UNLOCK(&fop->fd->lock);
+
+                fop->error = ENOMEM;
+
+                return EC_STATE_REPORT;
+            }
+            if (!ctx->loc.inode) {
+                err = ec_loc_from_loc(fop->xl, &ctx->loc, &fop->loc[0]);
+                if (err != 0) {
+                    UNLOCK(&fop->fd->lock);
+
+                    fop->error = -err;
+
+                    return EC_STATE_REPORT;
+                }
+            }
+
+            ctx->flags = fop->int32;
+
+            UNLOCK(&fop->fd->lock);
+
+            /* We need to write to specific offsets on the bricks, so we
+               need to remove O_APPEND from flags (if present).
+               If O_TRUNC is specified, we remove it from open and an
+               ftruncate will be executed later, which will correctly update
+               the file size taking appropriate locks. O_TRUNC flag is saved
+               into fop->uint32 to use it later.*/
+            fop->uint32 = fop->int32 & O_TRUNC;
+            fop->int32 &= ~(O_APPEND | O_TRUNC);
+
+            /* Fall through */
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_true);
+            if (cbk != NULL) {
+                int32_t err;
+
+                err = ec_loc_update(fop->xl, &fop->loc[0], cbk->fd->inode,
+                                    NULL);
+                if (!ec_cbk_set_error(cbk, -err, _gf_true)) {
+                    LOCK(&fop->fd->lock);
+
+                    ctx = __ec_fd_get(fop->fd, fop->xl);
+                    if (ctx != NULL) {
+                        ctx->open |= cbk->mask;
+                    }
+
+                    UNLOCK(&fop->fd->lock);
+
+                    /* If O_TRUNC was specified, call ftruncate to
+                       effectively trunc the file with appropriate locks
+                       acquired. We don't use ctx->flags because self-heal
+                       can use the same fd with different flags. */
+                    if (fop->uint32 != 0) {
+                        ec_sleep(fop);
+                        ec_ftruncate(fop->req_frame, fop->xl, cbk->mask,
+                                     fop->minimum, ec_open_truncate_cbk, fop,
+                                     cbk->fd, 0, NULL);
+                    }
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.open != NULL) {
+                fop->cbks.open(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                               cbk->op_errno, cbk->fd, cbk->xdata);
+            }
+
+            return EC_STATE_END;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.open != NULL) {
+                fop->cbks.open(fop->req_frame, fop, fop->xl, -1, fop->error,
+                               NULL, NULL);
+            }
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_open(call_frame_t *frame, xlator_t *this, uintptr_t target,
+        uint32_t fop_flags, fop_open_cbk_t func, void *data, loc_t *loc,
+        int32_t flags, fd_t *fd, dict_t *xdata)
+{
+    ec_cbk_t callback = {.open = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(OPEN) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_OPEN, EC_FLAG_LOCK_SHARED,
+                               target, fop_flags, ec_wind_open, ec_manager_open,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->int32 = flags;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
+
+/* FOP: readlink */
+
+int32_t
+ec_combine_readlink(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    if (!ec_iatt_combine(fop, dst->iatt, src->iatt, 1)) {
+        gf_msg(fop->xl->name, GF_LOG_NOTICE, 0, EC_MSG_IATT_MISMATCH,
+               "Mismatching iatt in "
+               "answers of 'GF_FOP_READLINK'");
+
+        return 0;
+    }
+
+    return 1;
+}
+
+int32_t
+ec_readlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, const char *path,
+                struct iatt *buf, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, fop->id, idx, op_ret,
+                               op_errno);
+    if (cbk) {
+        if (xdata)
+            cbk->xdata = dict_ref(xdata);
+
+        if (cbk->op_ret >= 0) {
+            cbk->iatt[0] = *buf;
+            cbk->str = gf_strdup(path);
+            if (!cbk->str) {
+                ec_cbk_set_error(cbk, ENOMEM, _gf_true);
+            }
+        }
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL)
+        ec_complete(fop);
+
+    return 0;
+}
+
+void
+ec_wind_readlink(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_readlink_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->readlink,
+                      &fop->loc[0], fop->size, fop->xdata);
+}
+
+int32_t
+ec_manager_readlink(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk = NULL;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ec_lock_prepare_inode(fop, &fop->loc[0], EC_QUERY_INFO, 0,
+                                  EC_RANGE_FULL);
+            ec_lock(fop);
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_one(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            if (ec_dispatch_one_retry(fop, &cbk)) {
+                return EC_STATE_DISPATCH;
+            }
+
+            if ((cbk != NULL) && (cbk->op_ret >= 0)) {
+                ec_iatt_rebuild(fop->xl->private, &cbk->iatt[0], 1, 1);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+            GF_ASSERT(cbk);
+            if (fop->cbks.readlink != NULL) {
+                fop->cbks.readlink(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                   cbk->op_errno, cbk->str, &cbk->iatt[0],
+                                   cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            if (fop->cbks.readlink != NULL) {
+                fop->cbks.readlink(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                   NULL, NULL, NULL);
+            }
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_readlink(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_readlink_cbk_t func, void *data, loc_t *loc,
+            size_t size, dict_t *xdata)
+{
+    ec_cbk_t callback = {.readlink = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(READLINK) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(
+        frame, this, GF_FOP_READLINK, EC_FLAG_LOCK_SHARED, target, fop_flags,
+        ec_wind_readlink, ec_manager_readlink, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->size = size;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: readv */
+
+int32_t
+ec_readv_rebuild(ec_t *ec, ec_fop_data_t *fop, ec_cbk_data_t *cbk)
+{
+    struct iovec vector[1];
+    ec_cbk_data_t *ans = NULL;
+    struct iobref *iobref = NULL;
+    void *ptr;
+    uint64_t fsize = 0, size = 0, max = 0;
+    int32_t pos, err = -ENOMEM;
+
+    if (cbk->op_ret < 0) {
+        err = -cbk->op_errno;
+
+        goto out;
+    }
+
+    /* This shouldn't fail because we have the inode locked. */
+    GF_ASSERT(ec_get_inode_size(fop, fop->fd->inode, &cbk->iatt[0].ia_size));
+
+    if (cbk->op_ret > 0) {
+        void *blocks[cbk->count];
+        uint32_t values[cbk->count];
+
+        fsize = cbk->op_ret;
+        size = fsize * ec->fragments;
+        for (ans = cbk; ans != NULL; ans = ans->next) {
+            pos = gf_bits_count(cbk->mask & ((1 << ans->idx) - 1));
+            values[pos] = ans->idx + 1;
+            blocks[pos] = ans->vector[0].iov_base;
+            if ((ans->int32 != 1) ||
+                !EC_ALIGN_CHECK(blocks[pos], EC_METHOD_WORD_SIZE)) {
+                if (iobref == NULL) {
+                    err = ec_buffer_alloc(ec->xl, size, &iobref, &ptr);
+                    if (err != 0) {
+                        goto out;
+                    }
+                }
+                ec_iov_copy_to(ptr, ans->vector, ans->int32, 0, fsize);
+                blocks[pos] = ptr;
+                ptr += fsize;
+            }
+        }
+
+        err = ec_buffer_alloc(ec->xl, size, &iobref, &ptr);
+        if (err != 0) {
+            goto out;
+        }
+
+        err = ec_method_decode(&ec->matrix, fsize, cbk->mask, values, blocks,
+                               ptr);
+        if (err != 0) {
+            goto out;
+        }
+
+        vector[0].iov_base = ptr + fop->head;
+        vector[0].iov_len = size - fop->head;
+
+        max = fop->offset * ec->fragments + size;
+        if (max > cbk->iatt[0].ia_size) {
+            max = cbk->iatt[0].ia_size;
+        }
+        max -= fop->offset * ec->fragments + fop->head;
+        if (max > fop->user_size) {
+            max = fop->user_size;
+        }
+        size -= fop->head;
+        if (size > max) {
+            vector[0].iov_len -= size - max;
+            size = max;
+        }
+
+        cbk->op_ret = size;
+        cbk->int32 = 1;
+
+        iobref_unref(cbk->buffers);
+        cbk->buffers = iobref;
+
+        GF_FREE(cbk->vector);
+        cbk->vector = iov_dup(vector, 1);
+        if (cbk->vector == NULL) {
+            return -ENOMEM;
+        }
+    }
+
+    return 0;
+
+out:
+    if (iobref != NULL) {
+        iobref_unref(iobref);
+    }
+
+    return err;
+}
+
+int32_t
+ec_combine_readv(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    if (!ec_vector_compare(dst->vector, dst->int32, src->vector, src->int32)) {
+        gf_msg(fop->xl->name, GF_LOG_NOTICE, 0, EC_MSG_VECTOR_MISMATCH,
+               "Mismatching vector in "
+               "answers of 'GF_FOP_READ'");
+
+        return 0;
+    }
+
+    if (!ec_iatt_combine(fop, dst->iatt, src->iatt, 1)) {
+        gf_msg(fop->xl->name, GF_LOG_NOTICE, 0, EC_MSG_IATT_MISMATCH,
+               "Mismatching iatt in "
+               "answers of 'GF_FOP_READ'");
+
+        return 0;
+    }
+
+    return 1;
+}
+
+int32_t
+ec_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, struct iovec *vector, int32_t count,
+             struct iatt *stbuf, struct iobref *iobref, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    ec_t *ec = this->private;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_READ, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            cbk->int32 = count;
+
+            if (count > 0) {
+                cbk->vector = iov_dup(vector, count);
+                if (cbk->vector == NULL) {
+                    gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                           "Failed to duplicate a "
+                           "vector list.");
+
+                    goto out;
+                }
+                cbk->int32 = count;
+            }
+            if (stbuf != NULL) {
+                cbk->iatt[0] = *stbuf;
+            }
+            if (iobref != NULL) {
+                cbk->buffers = iobref_ref(iobref);
+                if (cbk->buffers == NULL) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_BUF_REF_FAIL,
+                           "Failed to reference a "
+                           "buffer.");
+
+                    goto out;
+                }
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        if ((op_ret > 0) && ((op_ret % ec->fragment_size) != 0)) {
+            ec_cbk_set_error(cbk, EIO, _gf_true);
+        }
+
+        ec_combine(cbk, ec_combine_readv);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_readv(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_readv_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->readv, fop->fd,
+                      fop->size, fop->offset, fop->uint32, fop->xdata);
+}
+
+int32_t
+ec_manager_readv(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+    ec_t *ec = fop->xl->private;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            fop->user_size = fop->size;
+            fop->head = ec_adjust_offset_down(fop->xl->private, &fop->offset,
+                                              _gf_true);
+            fop->size += fop->head;
+            ec_adjust_size_up(fop->xl->private, &fop->size, _gf_true);
+
+            /* Fall through */
+
+        case EC_STATE_LOCK:
+            ec_lock_prepare_fd(fop, fop->fd, EC_QUERY_INFO, fop->offset,
+                               fop->size);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            if (ec->read_mask) {
+                fop->mask &= ec->read_mask;
+            }
+            ec_dispatch_min(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_true);
+            if (cbk != NULL) {
+                int32_t err;
+
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 1, cbk->count);
+
+                err = ec_readv_rebuild(fop->xl->private, fop, cbk);
+                if (err != 0) {
+                    ec_cbk_set_error(cbk, -err, _gf_true);
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.readv != NULL) {
+                fop->cbks.readv(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                cbk->op_errno, cbk->vector, cbk->int32,
+                                &cbk->iatt[0], cbk->buffers, cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.readv != NULL) {
+                fop->cbks.readv(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                NULL, 0, NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_readv(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_readv_cbk_t func, void *data, fd_t *fd,
+         size_t size, off_t offset, uint32_t flags, dict_t *xdata)
+{
+    ec_cbk_t callback = {.readv = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(READ) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_READ, EC_FLAG_LOCK_SHARED,
+                               target, fop_flags, ec_wind_readv,
+                               ec_manager_readv, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->size = size;
+    fop->offset = offset;
+    fop->uint32 = flags;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, 0, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: seek */
+
+int32_t
+ec_seek_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, off_t offset, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    ec_t *ec = this->private;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_SEEK, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            cbk->offset = offset;
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+        }
+
+        if ((op_ret > 0) && ((cbk->offset % ec->fragment_size) != 0)) {
+            cbk->op_ret = -1;
+            cbk->op_errno = EIO;
+        }
+
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_seek(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_seek_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->seek, fop->fd,
+                      fop->offset, fop->seek, fop->xdata);
+}
+
+int32_t
+ec_manager_seek(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+    uint64_t size;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            fop->user_size = fop->offset;
+            fop->head = ec_adjust_offset_down(fop->xl->private, &fop->offset,
+                                              _gf_true);
+
+            /* Fall through */
+
+        case EC_STATE_LOCK:
+            ec_lock_prepare_fd(fop, fop->fd, EC_QUERY_INFO, fop->offset,
+                               EC_RANGE_FULL);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            /* This shouldn't fail because we have the inode locked. */
+            GF_ASSERT(
+                ec_get_inode_size(fop, fop->locks[0].lock->loc.inode, &size));
+
+            if (fop->user_size >= size) {
+                ec_fop_set_error(fop, ENXIO);
+
+                return EC_STATE_REPORT;
+            }
+
+            ec_dispatch_one(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            if (ec_dispatch_one_retry(fop, &cbk)) {
+                return EC_STATE_DISPATCH;
+            }
+            if ((cbk != NULL) && (cbk->op_ret >= 0)) {
+                ec_t *ec = fop->xl->private;
+
+                /* This shouldn't fail because we have the inode locked. */
+                GF_ASSERT(ec_get_inode_size(fop, fop->locks[0].lock->loc.inode,
+                                            &size));
+
+                cbk->offset *= ec->fragments;
+                if (cbk->offset < fop->user_size) {
+                    cbk->offset = fop->user_size;
+                }
+                if (cbk->offset > size) {
+                    cbk->offset = size;
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.seek != NULL) {
+                fop->cbks.seek(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                               cbk->op_errno, cbk->offset, cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.seek != NULL) {
+                fop->cbks.seek(fop->req_frame, fop, fop->xl, -1, fop->error, 0,
+                               NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, 0, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_seek(call_frame_t *frame, xlator_t *this, uintptr_t target,
+        uint32_t fop_flags, fop_seek_cbk_t func, void *data, fd_t *fd,
+        off_t offset, gf_seek_what_t what, dict_t *xdata)
+{
+    ec_cbk_t callback = {.seek = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = EIO;
+
+    gf_msg_trace("ec", 0, "EC(SEEK) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_SEEK, EC_FLAG_LOCK_SHARED,
+                               target, fop_flags, ec_wind_seek, ec_manager_seek,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->offset = offset;
+    fop->seek = what;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, EIO, 0, NULL);
+    }
+}
+
+/* FOP: stat */
+
+int32_t
+ec_combine_stat(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    if (!ec_iatt_combine(fop, dst->iatt, src->iatt, 1)) {
+        gf_msg(fop->xl->name, GF_LOG_NOTICE, 0, EC_MSG_IATT_MISMATCH,
+               "Mismatching iatt in "
+               "answers of 'GF_FOP_STAT'");
+
+        return 0;
+    }
+
+    return 1;
+}
+
+int32_t
+ec_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, struct iatt *buf, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_STAT, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            if (buf != NULL) {
+                cbk->iatt[0] = *buf;
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, ec_combine_stat);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_stat(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_stat_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->stat,
+                      &fop->loc[0], fop->xdata);
+}
+
+int32_t
+ec_manager_stat(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            if (fop->fd == NULL) {
+                ec_lock_prepare_inode(fop, &fop->loc[0], EC_QUERY_INFO, 0,
+                                      EC_RANGE_FULL);
+            } else {
+                ec_lock_prepare_fd(fop, fop->fd, EC_QUERY_INFO, 0,
+                                   EC_RANGE_FULL);
+            }
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_true);
+
+            if (cbk != NULL) {
+                if (cbk->iatt[0].ia_type == IA_IFREG) {
+                    ec_iatt_rebuild(fop->xl->private, cbk->iatt, 1, cbk->count);
+
+                    /* This shouldn't fail because we have the inode locked. */
+                    GF_ASSERT(ec_get_inode_size(fop,
+                                                fop->locks[0].lock->loc.inode,
+                                                &cbk->iatt[0].ia_size));
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->id == GF_FOP_STAT) {
+                if (fop->cbks.stat != NULL) {
+                    fop->cbks.stat(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                   cbk->op_errno, &cbk->iatt[0], cbk->xdata);
+                }
+            } else {
+                if (fop->cbks.fstat != NULL) {
+                    fop->cbks.fstat(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                    cbk->op_errno, &cbk->iatt[0], cbk->xdata);
+                }
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->id == GF_FOP_STAT) {
+                if (fop->cbks.stat != NULL) {
+                    fop->cbks.stat(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                   NULL, NULL);
+                }
+            } else {
+                if (fop->cbks.fstat != NULL) {
+                    fop->cbks.fstat(fop->req_frame, fop, fop->xl, -1,
+                                    fop->error, NULL, NULL);
+                }
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_stat(call_frame_t *frame, xlator_t *this, uintptr_t target,
+        uint32_t fop_flags, fop_stat_cbk_t func, void *data, loc_t *loc,
+        dict_t *xdata)
+{
+    ec_cbk_t callback = {.stat = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(STAT) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_STAT, EC_FLAG_LOCK_SHARED,
+                               target, fop_flags, ec_wind_stat, ec_manager_stat,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
+
+/* FOP: fstat */
+
+int32_t
+ec_fstat_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, struct iatt *buf, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_FSTAT, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            if (buf != NULL) {
+                cbk->iatt[0] = *buf;
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, ec_combine_stat);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_fstat(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_fstat_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->fstat, fop->fd,
+                      fop->xdata);
+}
+
+void
+ec_fstat(call_frame_t *frame, xlator_t *this, uintptr_t target,
+         uint32_t fop_flags, fop_fstat_cbk_t func, void *data, fd_t *fd,
+         dict_t *xdata)
+{
+    ec_cbk_t callback = {.fstat = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FSTAT) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FSTAT, EC_FLAG_LOCK_SHARED,
+                               target, fop_flags, ec_wind_fstat,
+                               ec_manager_stat, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
diff --git a/xlators/cluster/ec/src/ec-inode-write.c b/xlators/cluster/ec/src/ec-inode-write.c
new file mode 100644
index 00000000000..9b5fe2a7fdc
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-inode-write.c
@@ -0,0 +1,2369 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "ec-messages.h"
+#include "ec-helpers.h"
+#include "ec-common.h"
+#include "ec-combine.h"
+#include "ec-method.h"
+#include "ec-fops.h"
+#include "ec-mem-types.h"
+
+int32_t
+ec_update_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+    ec_fop_data_t *fop = cookie;
+    ec_cbk_data_t *cbk = NULL;
+    ec_fop_data_t *parent = fop->parent;
+    int i = 0;
+
+    ec_trace("UPDATE_WRITEV_CBK", cookie, "ret=%d, errno=%d, parent-fop=%s",
+             op_ret, op_errno, ec_fop_name(parent->id));
+
+    if (op_ret < 0) {
+        ec_fop_set_error(parent, op_errno);
+        goto out;
+    }
+    cbk = ec_cbk_data_allocate(parent->frame, this, parent, parent->id, 0,
+                               op_ret, op_errno);
+    if (!cbk) {
+        ec_fop_set_error(parent, ENOMEM);
+        goto out;
+    }
+
+    if (xdata)
+        cbk->xdata = dict_ref(xdata);
+
+    if (prebuf)
+        cbk->iatt[i++] = *prebuf;
+
+    if (postbuf)
+        cbk->iatt[i++] = *postbuf;
+
+    LOCK(&parent->lock);
+    {
+        parent->good &= fop->good;
+
+        if (gf_bits_count(parent->good) < parent->minimum) {
+            __ec_fop_set_error(parent, EIO);
+        } else if (fop->error == 0 && parent->answer == NULL) {
+            parent->answer = cbk;
+        }
+    }
+    UNLOCK(&parent->lock);
+out:
+    return 0;
+}
+
+static int32_t
+ec_update_write(ec_fop_data_t *fop, uintptr_t mask, off_t offset, uint64_t size)
+{
+    struct iobref *iobref = NULL;
+    struct iobuf *iobuf = NULL;
+    struct iovec vector;
+    int32_t err = -ENOMEM;
+
+    iobref = iobref_new();
+    if (iobref == NULL) {
+        goto out;
+    }
+    iobuf = iobuf_get(fop->xl->ctx->iobuf_pool);
+    if (iobuf == NULL) {
+        goto out;
+    }
+    err = iobref_add(iobref, iobuf);
+    if (err != 0) {
+        goto out;
+    }
+
+    if (fop->locks[0].lock)
+        ec_lock_update_good(fop->locks[0].lock, fop);
+    vector.iov_base = iobuf->ptr;
+    vector.iov_len = size;
+    memset(vector.iov_base, 0, vector.iov_len);
+
+    ec_writev(fop->frame, fop->xl, mask, fop->minimum, ec_update_writev_cbk,
+              NULL, fop->fd, &vector, 1, offset, 0, iobref, NULL);
+
+    err = 0;
+
+out:
+    if (iobuf != NULL) {
+        iobuf_unref(iobuf);
+    }
+    if (iobref != NULL) {
+        iobref_unref(iobref);
+    }
+
+    return err;
+}
+
+int
+ec_inode_write_cbk(call_frame_t *frame, xlator_t *this, void *cookie,
+                   int op_ret, int op_errno, struct iatt *prestat,
+                   struct iatt *poststat, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int i = 0;
+    int idx = 0;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+    idx = (int32_t)(uintptr_t)cookie;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, fop->id, idx, op_ret,
+                               op_errno);
+    if (!cbk)
+        goto out;
+
+    if (op_ret < 0)
+        goto out;
+
+    if (xdata)
+        cbk->xdata = dict_ref(xdata);
+
+    if (prestat)
+        cbk->iatt[i++] = *prestat;
+
+    if (poststat)
+        cbk->iatt[i++] = *poststat;
+
+out:
+    if (cbk)
+        ec_combine(cbk, ec_combine_write);
+
+    if (fop)
+        ec_complete(fop);
+    return 0;
+}
+/* FOP: removexattr */
+
+int32_t
+ec_removexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    return ec_inode_write_cbk(frame, this, cookie, op_ret, op_errno, NULL, NULL,
+                              xdata);
+}
+
+void
+ec_wind_removexattr(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_removexattr_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->removexattr,
+                      &fop->loc[0], fop->str[0], fop->xdata);
+}
+
+void
+ec_xattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = cookie;
+    switch (fop->id) {
+        case GF_FOP_SETXATTR:
+            if (fop->cbks.setxattr) {
+                QUORUM_CBK(fop->cbks.setxattr, fop, frame, cookie, this, op_ret,
+                           op_errno, xdata);
+            }
+            break;
+        case GF_FOP_REMOVEXATTR:
+            if (fop->cbks.removexattr) {
+                QUORUM_CBK(fop->cbks.removexattr, fop, frame, cookie, this,
+                           op_ret, op_errno, xdata);
+            }
+            break;
+        case GF_FOP_FSETXATTR:
+            if (fop->cbks.fsetxattr) {
+                QUORUM_CBK(fop->cbks.fsetxattr, fop, frame, cookie, this,
+                           op_ret, op_errno, xdata);
+            }
+            break;
+        case GF_FOP_FREMOVEXATTR:
+            if (fop->cbks.fremovexattr) {
+                QUORUM_CBK(fop->cbks.fremovexattr, fop, frame, cookie, this,
+                           op_ret, op_errno, xdata);
+            }
+            break;
+    }
+}
+
+int32_t
+ec_manager_xattr(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            if (fop->fd == NULL) {
+                ec_lock_prepare_inode(fop, &fop->loc[0],
+                                      EC_UPDATE_META | EC_QUERY_INFO, 0,
+                                      EC_RANGE_FULL);
+            } else {
+                ec_lock_prepare_fd(fop, fop->fd, EC_UPDATE_META | EC_QUERY_INFO,
+                                   0, EC_RANGE_FULL);
+            }
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            ec_fop_prepare_answer(fop, _gf_false);
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            ec_xattr_cbk(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                         cbk->op_errno, cbk->xdata);
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            ec_xattr_cbk(fop->req_frame, fop, fop->xl, -1, fop->error, NULL);
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_removexattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+               uint32_t fop_flags, fop_removexattr_cbk_t func, void *data,
+               loc_t *loc, const char *name, dict_t *xdata)
+{
+    ec_cbk_t callback = {.removexattr = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(REMOVEXATTR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_REMOVEXATTR, 0, target,
+                               fop_flags, ec_wind_removexattr, ec_manager_xattr,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (name != NULL) {
+        fop->str[0] = gf_strdup(name);
+        if (fop->str[0] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/* FOP: fremovexattr */
+
+int32_t
+ec_fremovexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    return ec_inode_write_cbk(frame, this, cookie, op_ret, op_errno, NULL, NULL,
+                              xdata);
+}
+
+void
+ec_wind_fremovexattr(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_fremovexattr_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->fremovexattr,
+                      fop->fd, fop->str[0], fop->xdata);
+}
+
+void
+ec_fremovexattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+                uint32_t fop_flags, fop_fremovexattr_cbk_t func, void *data,
+                fd_t *fd, const char *name, dict_t *xdata)
+{
+    ec_cbk_t callback = {.fremovexattr = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FREMOVEXATTR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FREMOVEXATTR, 0, target,
+                               fop_flags, ec_wind_fremovexattr,
+                               ec_manager_xattr, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (name != NULL) {
+        fop->str[0] = gf_strdup(name);
+        if (fop->str[0] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/* FOP: setattr */
+
+int32_t
+ec_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, struct iatt *prestat,
+               struct iatt *poststat, dict_t *xdata)
+{
+    return ec_inode_write_cbk(frame, this, cookie, op_ret, op_errno, prestat,
+                              poststat, xdata);
+}
+
+void
+ec_wind_setattr(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_setattr_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->setattr,
+                      &fop->loc[0], &fop->iatt, fop->int32, fop->xdata);
+}
+
+int32_t
+ec_manager_setattr(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            if (fop->fd == NULL) {
+                ec_lock_prepare_inode(fop, &fop->loc[0],
+                                      EC_UPDATE_META | EC_QUERY_INFO, 0,
+                                      EC_RANGE_FULL);
+            } else {
+                ec_lock_prepare_fd(fop, fop->fd, EC_UPDATE_META | EC_QUERY_INFO,
+                                   0, EC_RANGE_FULL);
+            }
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                if (cbk->iatt[0].ia_type == IA_IFREG) {
+                    ec_iatt_rebuild(fop->xl->private, cbk->iatt, 2, cbk->count);
+
+                    /* This shouldn't fail because we have the inode locked. */
+                    GF_ASSERT(ec_get_inode_size(fop,
+                                                fop->locks[0].lock->loc.inode,
+                                                &cbk->iatt[0].ia_size));
+                    cbk->iatt[1].ia_size = cbk->iatt[0].ia_size;
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->id == GF_FOP_SETATTR) {
+                if (fop->cbks.setattr != NULL) {
+                    QUORUM_CBK(fop->cbks.setattr, fop, fop->req_frame, fop,
+                               fop->xl, cbk->op_ret, cbk->op_errno,
+                               &cbk->iatt[0], &cbk->iatt[1], cbk->xdata);
+                }
+            } else {
+                if (fop->cbks.fsetattr != NULL) {
+                    QUORUM_CBK(fop->cbks.fsetattr, fop, fop->req_frame, fop,
+                               fop->xl, cbk->op_ret, cbk->op_errno,
+                               &cbk->iatt[0], &cbk->iatt[1], cbk->xdata);
+                }
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->id == GF_FOP_SETATTR) {
+                if (fop->cbks.setattr != NULL) {
+                    fop->cbks.setattr(fop->req_frame, fop, fop->xl, -1,
+                                      fop->error, NULL, NULL, NULL);
+                }
+            } else {
+                if (fop->cbks.fsetattr != NULL) {
+                    fop->cbks.fsetattr(fop->req_frame, fop, fop->xl, -1,
+                                       fop->error, NULL, NULL, NULL);
+                }
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_setattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_setattr_cbk_t func, void *data, loc_t *loc,
+           struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    ec_cbk_t callback = {.setattr = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(SETATTR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_SETATTR, 0, target,
+                               fop_flags, ec_wind_setattr, ec_manager_setattr,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->int32 = valid;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (stbuf != NULL) {
+        fop->iatt = *stbuf;
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: fsetattr */
+
+int32_t
+ec_fsetattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *prestat,
+                struct iatt *poststat, dict_t *xdata)
+{
+    return ec_inode_write_cbk(frame, this, cookie, op_ret, op_errno, prestat,
+                              poststat, xdata);
+}
+
+void
+ec_wind_fsetattr(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_fsetattr_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->fsetattr,
+                      fop->fd, &fop->iatt, fop->int32, fop->xdata);
+}
+
+void
+ec_fsetattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_fsetattr_cbk_t func, void *data, fd_t *fd,
+            struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    ec_cbk_t callback = {.fsetattr = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FSETATTR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FSETATTR, 0, target,
+                               fop_flags, ec_wind_fsetattr, ec_manager_setattr,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->int32 = valid;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (stbuf != NULL) {
+        fop->iatt = *stbuf;
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: setxattr */
+
+int32_t
+ec_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    return ec_inode_write_cbk(frame, this, cookie, op_ret, op_errno, NULL, NULL,
+                              xdata);
+}
+
+void
+ec_wind_setxattr(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_setxattr_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->setxattr,
+                      &fop->loc[0], fop->dict, fop->int32, fop->xdata);
+}
+
+void
+ec_setxattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_setxattr_cbk_t func, void *data, loc_t *loc,
+            dict_t *dict, int32_t flags, dict_t *xdata)
+{
+    ec_cbk_t callback = {.setxattr = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(SETXATTR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_SETXATTR, 0, target,
+                               fop_flags, ec_wind_setxattr, ec_manager_xattr,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->int32 = flags;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (dict != NULL) {
+        fop->dict = dict_copy_with_ref(dict, NULL);
+        if (fop->dict == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/* FOP: fsetxattr */
+
+int32_t
+ec_fsetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_FSETXATTR, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_fsetxattr(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_fsetxattr_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->fsetxattr,
+                      fop->fd, fop->dict, fop->int32, fop->xdata);
+}
+
+void
+ec_fsetxattr(call_frame_t *frame, xlator_t *this, uintptr_t target,
+             uint32_t fop_flags, fop_fsetxattr_cbk_t func, void *data, fd_t *fd,
+             dict_t *dict, int32_t flags, dict_t *xdata)
+{
+    ec_cbk_t callback = {.fsetxattr = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FSETXATTR) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FSETXATTR, 0, target,
+                               fop_flags, ec_wind_fsetxattr, ec_manager_xattr,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->int32 = flags;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (dict != NULL) {
+        fop->dict = dict_copy_with_ref(dict, NULL);
+        if (fop->dict == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/*********************************************************************
+ *
+ * File Operation : fallocate
+ *
+ *********************************************************************/
+
+int32_t
+ec_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                 struct iatt *postbuf, dict_t *xdata)
+{
+    return ec_inode_write_cbk(frame, this, cookie, op_ret, op_errno, prebuf,
+                              postbuf, xdata);
+}
+
+void
+ec_wind_fallocate(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_fallocate_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->fallocate,
+                      fop->fd, fop->int32, fop->offset, fop->size, fop->xdata);
+}
+
+int32_t
+ec_manager_fallocate(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk = NULL;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            if (fop->size == 0) {
+                ec_fop_set_error(fop, EINVAL);
+                return EC_STATE_REPORT;
+            }
+            if (fop->int32 &
+                (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
+                 FALLOC_FL_ZERO_RANGE | FALLOC_FL_PUNCH_HOLE)) {
+                ec_fop_set_error(fop, ENOTSUP);
+                return EC_STATE_REPORT;
+            }
+            fop->user_size = fop->offset + fop->size;
+            fop->head = ec_adjust_offset_down(fop->xl->private, &fop->offset,
+                                              _gf_true);
+            fop->size += fop->head;
+            ec_adjust_size_up(fop->xl->private, &fop->size, _gf_true);
+
+            /* Fall through */
+
+        case EC_STATE_LOCK:
+            ec_lock_prepare_fd(fop, fop->fd,
+                               EC_UPDATE_DATA | EC_UPDATE_META | EC_QUERY_INFO,
+                               fop->offset, fop->size);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 2, cbk->count);
+
+                /* This shouldn't fail because we have the inode locked. */
+                LOCK(&fop->locks[0].lock->loc.inode->lock);
+                {
+                    GF_ASSERT(__ec_get_inode_size(fop,
+                                                  fop->locks[0].lock->loc.inode,
+                                                  &cbk->iatt[0].ia_size));
+
+                    /*If mode has FALLOC_FL_KEEP_SIZE keep the size */
+                    if (fop->int32 & FALLOC_FL_KEEP_SIZE) {
+                        cbk->iatt[1].ia_size = cbk->iatt[0].ia_size;
+                    } else if (fop->user_size > cbk->iatt[0].ia_size) {
+                        cbk->iatt[1].ia_size = fop->user_size;
+
+                        /* This shouldn't fail because we have the inode
+                         * locked. */
+                        GF_ASSERT(__ec_set_inode_size(
+                            fop, fop->locks[0].lock->loc.inode,
+                            cbk->iatt[1].ia_size));
+                    } else {
+                        cbk->iatt[1].ia_size = cbk->iatt[0].ia_size;
+                    }
+                }
+                UNLOCK(&fop->locks[0].lock->loc.inode->lock);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.fallocate != NULL) {
+                QUORUM_CBK(fop->cbks.fallocate, fop, fop->req_frame, fop,
+                           fop->xl, cbk->op_ret, cbk->op_errno, &cbk->iatt[0],
+                           &cbk->iatt[1], cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.fallocate != NULL) {
+                fop->cbks.fallocate(fop->req_frame, fop, fop->xl, -1,
+                                    fop->error, NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_fallocate(call_frame_t *frame, xlator_t *this, uintptr_t target,
+             uint32_t fop_flags, fop_fallocate_cbk_t func, void *data, fd_t *fd,
+             int32_t mode, off_t offset, size_t len, dict_t *xdata)
+{
+    ec_cbk_t callback = {.fallocate = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FALLOCATE) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FALLOCATE, 0, target,
+                               fop_flags, ec_wind_fallocate,
+                               ec_manager_fallocate, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+    fop->int32 = mode;
+    fop->offset = offset;
+    fop->size = len;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+            goto out;
+        }
+    }
+
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
+
+/*********************************************************************
+ *
+ * File Operation : Discard
+ *
+ *********************************************************************/
+void
+ec_update_discard_write(ec_fop_data_t *fop, uintptr_t mask)
+{
+    ec_t *ec = fop->xl->private;
+    off_t off_head = 0;
+    off_t off_tail = 0;
+    uint64_t size_head = 0;
+    uint64_t size_tail = 0;
+    int error = 0;
+
+    off_head = fop->offset * ec->fragments - fop->int32;
+    if (fop->size == 0) {
+        error = ec_update_write(fop, mask, off_head, fop->user_size);
+    } else {
+        size_head = fop->int32;
+        size_tail = (off_head + fop->user_size) % ec->stripe_size;
+        off_tail = off_head + fop->user_size - size_tail;
+        if (size_head) {
+            error = ec_update_write(fop, mask, off_head, size_head);
+            if (error) {
+                goto out;
+            }
+        }
+        if (size_tail) {
+            error = ec_update_write(fop, mask, off_tail, size_tail);
+        }
+    }
+out:
+    if (error)
+        ec_fop_set_error(fop, -error);
+}
+
+void
+ec_discard_adjust_offset_size(ec_fop_data_t *fop)
+{
+    ec_t *ec = fop->xl->private;
+
+    fop->user_size = fop->size;
+    /* If discard length covers at least a fragment on brick, we will
+     * perform discard operation(when fop->size is non-zero) else we just
+     * write zeros.
+     */
+    fop->int32 = ec_adjust_offset_up(ec, &fop->offset, _gf_true);
+    fop->frag_range.first = fop->offset;
+    if (fop->size < fop->int32) {
+        fop->size = 0;
+    } else {
+        fop->size -= fop->int32;
+        ec_adjust_size_down(ec, &fop->size, _gf_true);
+    }
+    fop->frag_range.last = fop->offset + fop->size;
+}
+
+int32_t
+ec_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+               struct iatt *postbuf, dict_t *xdata)
+{
+    return ec_inode_write_cbk(frame, this, cookie, op_ret, op_errno, prebuf,
+                              postbuf, xdata);
+}
+
+void
+ec_wind_discard(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_discard_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->discard,
+                      fop->fd, fop->offset, fop->size, fop->xdata);
+}
+
+int32_t
+ec_manager_discard(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk = NULL;
+    off_t fl_start = 0;
+    uint64_t fl_size = 0;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            if ((fop->size <= 0) || (fop->offset < 0)) {
+                ec_fop_set_error(fop, EINVAL);
+                return EC_STATE_REPORT;
+            }
+            /* Because of the head/tail writes, "discard" happens on the
+             * remaining regions, but we need to compute region including
+             * head/tail writes so compute them separately*/
+            fl_start = fop->offset;
+            fl_size = fop->size;
+            fl_size += ec_adjust_offset_down(fop->xl->private, &fl_start,
+                                             _gf_true);
+            ec_adjust_size_up(fop->xl->private, &fl_size, _gf_true);
+
+            ec_discard_adjust_offset_size(fop);
+
+            /* Fall through */
+
+        case EC_STATE_LOCK:
+            ec_lock_prepare_fd(fop, fop->fd,
+                               EC_UPDATE_DATA | EC_UPDATE_META | EC_QUERY_INFO,
+                               fl_start, fl_size);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+
+            /* Dispatch discard fop only if we have whole fragment
+             * to deallocate */
+            if (fop->size) {
+                ec_dispatch_all(fop);
+                return EC_STATE_DELAYED_START;
+            } else {
+                /* Assume discard to have succeeded on all bricks */
+                ec_succeed_all(fop);
+            }
+
+            /* Fall through */
+
+        case EC_STATE_DELAYED_START:
+
+            if (fop->size) {
+                if (fop->answer && fop->answer->op_ret == 0)
+                    ec_update_discard_write(fop, fop->answer->mask);
+            } else {
+                ec_update_discard_write(fop, fop->mask);
+            }
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 2, cbk->count);
+
+                /* This shouldn't fail because we have the inode locked. */
+                GF_ASSERT(ec_get_inode_size(fop, fop->locks[0].lock->loc.inode,
+                                            &cbk->iatt[0].ia_size));
+
+                cbk->iatt[1].ia_size = cbk->iatt[0].ia_size;
+            }
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.discard != NULL) {
+                QUORUM_CBK(fop->cbks.discard, fop, fop->req_frame, fop, fop->xl,
+                           cbk->op_ret, cbk->op_errno, &cbk->iatt[0],
+                           &cbk->iatt[1], cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_DELAYED_START:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.discard != NULL) {
+                fop->cbks.discard(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                  NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_discard(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_discard_cbk_t func, void *data, fd_t *fd,
+           off_t offset, size_t len, dict_t *xdata)
+{
+    ec_cbk_t callback = {.discard = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(DISCARD) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_DISCARD, 0, target,
+                               fop_flags, ec_wind_discard, ec_manager_discard,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+    fop->offset = offset;
+    fop->size = len;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+    }
+
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
+
+/*********************************************************************
+ *
+ * File Operation : truncate
+ *
+ *********************************************************************/
+
+int32_t
+ec_update_truncate_write(ec_fop_data_t *fop, uintptr_t mask)
+{
+    ec_t *ec = fop->xl->private;
+    uint64_t size = fop->offset * ec->fragments - fop->user_size;
+    return ec_update_write(fop, mask, fop->user_size, size);
+}
+
+int32_t
+ec_truncate_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
+{
+    ec_fop_data_t *fop = cookie;
+    int32_t err;
+
+    fop->parent->good &= fop->good;
+    if (op_ret >= 0) {
+        fd_bind(fd);
+        err = ec_update_truncate_write(fop->parent, fop->answer->mask);
+        if (err != 0) {
+            ec_fop_set_error(fop->parent, -err);
+        }
+    }
+
+    return 0;
+}
+
+int32_t
+ec_truncate_clean(ec_fop_data_t *fop)
+{
+    if (fop->fd == NULL) {
+        fop->fd = fd_create(fop->loc[0].inode, fop->frame->root->pid);
+        if (fop->fd == NULL) {
+            return -ENOMEM;
+        }
+
+        ec_open(fop->frame, fop->xl, fop->answer->mask, fop->minimum,
+                ec_truncate_open_cbk, fop, &fop->loc[0], O_RDWR, fop->fd, NULL);
+
+        return 0;
+    } else {
+        return ec_update_truncate_write(fop, fop->answer->mask);
+    }
+}
+
+int32_t
+ec_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *prestat,
+                struct iatt *poststat, dict_t *xdata)
+{
+    return ec_inode_write_cbk(frame, this, cookie, op_ret, op_errno, prestat,
+                              poststat, xdata);
+}
+
+void
+ec_wind_truncate(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_truncate_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->truncate,
+                      &fop->loc[0], fop->offset, fop->xdata);
+}
+
+int32_t
+ec_manager_truncate(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+    off_t offset_down;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            fop->user_size = fop->offset;
+            ec_adjust_offset_up(fop->xl->private, &fop->offset, _gf_true);
+            fop->frag_range.first = fop->offset;
+            fop->frag_range.last = UINT64_MAX;
+
+            /* Fall through */
+
+        case EC_STATE_LOCK:
+            offset_down = fop->user_size;
+            ec_adjust_offset_down(fop->xl->private, &offset_down, _gf_true);
+
+            if (fop->id == GF_FOP_TRUNCATE) {
+                ec_lock_prepare_inode(
+                    fop, &fop->loc[0],
+                    EC_UPDATE_DATA | EC_UPDATE_META | EC_QUERY_INFO,
+                    offset_down, EC_RANGE_FULL);
+            } else {
+                ec_lock_prepare_fd(
+                    fop, fop->fd,
+                    EC_UPDATE_DATA | EC_UPDATE_META | EC_QUERY_INFO,
+                    offset_down, EC_RANGE_FULL);
+            }
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                int32_t err;
+
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 2, cbk->count);
+
+                /* This shouldn't fail because we have the inode locked. */
+                /* Inode size doesn't need to be updated under locks, because
+                 * conflicting operations won't be in-flight
+                 */
+                GF_ASSERT(ec_get_inode_size(fop, fop->locks[0].lock->loc.inode,
+                                            &cbk->iatt[0].ia_size));
+                cbk->iatt[1].ia_size = fop->user_size;
+                /* This shouldn't fail because we have the inode locked. */
+                GF_ASSERT(ec_set_inode_size(fop, fop->locks[0].lock->loc.inode,
+                                            fop->user_size));
+                if ((cbk->iatt[0].ia_size > cbk->iatt[1].ia_size) &&
+                    (fop->user_size != fop->offset)) {
+                    err = ec_truncate_clean(fop);
+                    if (err != 0) {
+                        ec_cbk_set_error(cbk, -err, _gf_false);
+                    }
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->id == GF_FOP_TRUNCATE) {
+                if (fop->cbks.truncate != NULL) {
+                    QUORUM_CBK(fop->cbks.truncate, fop, fop->req_frame, fop,
+                               fop->xl, cbk->op_ret, cbk->op_errno,
+                               &cbk->iatt[0], &cbk->iatt[1], cbk->xdata);
+                }
+            } else {
+                if (fop->cbks.ftruncate != NULL) {
+                    QUORUM_CBK(fop->cbks.ftruncate, fop, fop->req_frame, fop,
+                               fop->xl, cbk->op_ret, cbk->op_errno,
+                               &cbk->iatt[0], &cbk->iatt[1], cbk->xdata);
+                }
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->id == GF_FOP_TRUNCATE) {
+                if (fop->cbks.truncate != NULL) {
+                    fop->cbks.truncate(fop->req_frame, fop, fop->xl, -1,
+                                       fop->error, NULL, NULL, NULL);
+                }
+            } else {
+                if (fop->cbks.ftruncate != NULL) {
+                    fop->cbks.ftruncate(fop->req_frame, fop, fop->xl, -1,
+                                        fop->error, NULL, NULL, NULL);
+                }
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_truncate(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_truncate_cbk_t func, void *data, loc_t *loc,
+            off_t offset, dict_t *xdata)
+{
+    ec_cbk_t callback = {.truncate = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(TRUNCATE) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_TRUNCATE, 0, target,
+                               fop_flags, ec_wind_truncate, ec_manager_truncate,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->offset = offset;
+
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: ftruncate */
+
+int32_t
+ec_ftruncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct iatt *prestat,
+                 struct iatt *poststat, dict_t *xdata)
+{
+    return ec_inode_write_cbk(frame, this, cookie, op_ret, op_errno, prestat,
+                              poststat, xdata);
+}
+
+void
+ec_wind_ftruncate(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_ftruncate_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->ftruncate,
+                      fop->fd, fop->offset, fop->xdata);
+}
+
+void
+ec_ftruncate(call_frame_t *frame, xlator_t *this, uintptr_t target,
+             uint32_t fop_flags, fop_ftruncate_cbk_t func, void *data, fd_t *fd,
+             off_t offset, dict_t *xdata)
+{
+    ec_cbk_t callback = {.ftruncate = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FTRUNCATE) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FTRUNCATE, 0, target,
+                               fop_flags, ec_wind_ftruncate,
+                               ec_manager_truncate, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->offset = offset;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
+
+/* FOP: writev */
+static ec_stripe_t *
+ec_allocate_stripe(ec_t *ec, ec_stripe_list_t *stripe_cache)
+{
+    ec_stripe_t *stripe = NULL;
+
+    if (stripe_cache->count >= stripe_cache->max) {
+        GF_ASSERT(!list_empty(&stripe_cache->lru));
+        stripe = list_first_entry(&stripe_cache->lru, ec_stripe_t, lru);
+        list_move_tail(&stripe->lru, &stripe_cache->lru);
+        GF_ATOMIC_INC(ec->stats.stripe_cache.evicts);
+    } else {
+        stripe = GF_MALLOC(sizeof(ec_stripe_t) + ec->stripe_size,
+                           ec_mt_ec_stripe_t);
+        if (stripe != NULL) {
+            stripe_cache->count++;
+            list_add_tail(&stripe->lru, &stripe_cache->lru);
+            GF_ATOMIC_INC(ec->stats.stripe_cache.allocs);
+        } else {
+            GF_ATOMIC_INC(ec->stats.stripe_cache.errors);
+        }
+    }
+
+    return stripe;
+}
+
+static void
+ec_write_stripe_data(ec_t *ec, ec_fop_data_t *fop, ec_stripe_t *stripe)
+{
+    off_t base;
+
+    base = fop->size - ec->stripe_size;
+    memcpy(stripe->data, fop->vector[0].iov_base + base, ec->stripe_size);
+    stripe->frag_offset = fop->frag_range.last - ec->fragment_size;
+}
+
+static void
+ec_add_stripe_in_cache(ec_t *ec, ec_fop_data_t *fop)
+{
+    ec_inode_t *ctx = NULL;
+    ec_stripe_t *stripe = NULL;
+    ec_stripe_list_t *stripe_cache = NULL;
+    gf_boolean_t failed = _gf_true;
+
+    LOCK(&fop->fd->inode->lock);
+
+    ctx = __ec_inode_get(fop->fd->inode, fop->xl);
+    if (ctx == NULL) {
+        goto out;
+    }
+
+    stripe_cache = &ctx->stripe_cache;
+    if (stripe_cache->max > 0) {
+        stripe = ec_allocate_stripe(ec, stripe_cache);
+        if (stripe == NULL) {
+            goto out;
+        }
+
+        ec_write_stripe_data(ec, fop, stripe);
+    }
+
+    failed = _gf_false;
+
+out:
+    UNLOCK(&fop->fd->inode->lock);
+
+    if (failed) {
+        gf_msg(ec->xl->name, GF_LOG_DEBUG, ENOMEM, EC_MSG_FILE_DESC_REF_FAIL,
+               "Failed to create and add stripe in cache");
+    }
+}
+
+int32_t
+ec_writev_merge_tail(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iovec *vector,
+                     int32_t count, struct iatt *stbuf, struct iobref *iobref,
+                     dict_t *xdata)
+{
+    ec_t *ec = this->private;
+    ec_fop_data_t *fop = frame->local;
+    uint64_t size, base, tmp;
+
+    if (op_ret >= 0) {
+        tmp = 0;
+        size = fop->size - fop->user_size - fop->head;
+        base = ec->stripe_size - size;
+        if (op_ret > base) {
+            tmp = min(op_ret - base, size);
+            ec_iov_copy_to(fop->vector[0].iov_base + fop->size - size, vector,
+                           count, base, tmp);
+
+            size -= tmp;
+        }
+
+        if (size > 0) {
+            memset(fop->vector[0].iov_base + fop->size - size, 0, size);
+        }
+
+        if (ec->stripe_cache) {
+            ec_add_stripe_in_cache(ec, fop);
+        }
+    }
+    return 0;
+}
+
+int32_t
+ec_writev_merge_head(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iovec *vector,
+                     int32_t count, struct iatt *stbuf, struct iobref *iobref,
+                     dict_t *xdata)
+{
+    ec_t *ec = this->private;
+    ec_fop_data_t *fop = frame->local;
+    uint64_t size, base;
+
+    if (op_ret >= 0) {
+        size = fop->head;
+        base = 0;
+
+        if (op_ret > 0) {
+            base = min(op_ret, size);
+            ec_iov_copy_to(fop->vector[0].iov_base, vector, count, 0, base);
+
+            size -= base;
+        }
+
+        if (size > 0) {
+            memset(fop->vector[0].iov_base + base, 0, size);
+        }
+
+        size = fop->size - fop->user_size - fop->head;
+        if ((size > 0) && (fop->size == ec->stripe_size)) {
+            ec_writev_merge_tail(frame, cookie, this, op_ret, op_errno, vector,
+                                 count, stbuf, iobref, xdata);
+        }
+    }
+
+    return 0;
+}
+
+static int
+ec_make_internal_fop_xdata(dict_t **xdata)
+{
+    dict_t *dict = NULL;
+
+    if (*xdata)
+        return 0;
+
+    dict = dict_new();
+    if (!dict)
+        goto out;
+
+    if (dict_set_str(dict, GLUSTERFS_INTERNAL_FOP_KEY, "yes"))
+        goto out;
+
+    *xdata = dict;
+    return 0;
+out:
+    if (dict)
+        dict_unref(dict);
+    return -1;
+}
+
+static int32_t
+ec_writev_prepare_buffers(ec_t *ec, ec_fop_data_t *fop)
+{
+    struct iobref *iobref = NULL;
+    struct iovec *iov;
+    void *ptr;
+    int32_t err;
+
+    fop->user_size = iov_length(fop->vector, fop->int32);
+    fop->head = ec_adjust_offset_down(ec, &fop->offset, _gf_false);
+    fop->frag_range.first = fop->offset / ec->fragments;
+    fop->size = fop->user_size + fop->head;
+    ec_adjust_size_up(ec, &fop->size, _gf_false);
+    fop->frag_range.last = fop->frag_range.first + fop->size / ec->fragments;
+
+    if ((fop->int32 != 1) || (fop->head != 0) || (fop->size > fop->user_size) ||
+        !EC_ALIGN_CHECK(fop->vector[0].iov_base, EC_METHOD_WORD_SIZE)) {
+        err = ec_buffer_alloc(ec->xl, fop->size, &iobref, &ptr);
+        if (err != 0) {
+            goto out;
+        }
+
+        ec_iov_copy_to(ptr + fop->head, fop->vector, fop->int32, 0,
+                       fop->user_size);
+
+        fop->vector[0].iov_base = ptr;
+        fop->vector[0].iov_len = fop->size;
+
+        iobref_unref(fop->buffers);
+        fop->buffers = iobref;
+    }
+
+    if (fop->int32 != 2) {
+        iov = GF_MALLOC(VECTORSIZE(2), gf_common_mt_iovec);
+        if (iov == NULL) {
+            err = -ENOMEM;
+
+            goto out;
+        }
+        iov[0].iov_base = fop->vector[0].iov_base;
+        iov[0].iov_len = fop->vector[0].iov_len;
+
+        GF_FREE(fop->vector);
+        fop->vector = iov;
+    }
+
+    fop->vector[1].iov_len = fop->size / ec->fragments;
+    err = ec_buffer_alloc(ec->xl, fop->vector[1].iov_len * ec->nodes,
+                          &fop->buffers, &fop->vector[1].iov_base);
+    if (err != 0) {
+        goto out;
+    }
+
+    err = 0;
+
+out:
+    return err;
+}
+
+static void
+ec_merge_stripe_head_locked(ec_t *ec, ec_fop_data_t *fop, ec_stripe_t *stripe)
+{
+    uint32_t head, size;
+
+    head = fop->head;
+    memcpy(fop->vector[0].iov_base, stripe->data, head);
+
+    size = ec->stripe_size - head;
+    if (size > fop->user_size) {
+        head += fop->user_size;
+        size = ec->stripe_size - head;
+        memcpy(fop->vector[0].iov_base + head, stripe->data + head, size);
+    }
+}
+
+static void
+ec_merge_stripe_tail_locked(ec_t *ec, ec_fop_data_t *fop, ec_stripe_t *stripe)
+{
+    uint32_t head, tail;
+    off_t offset;
+
+    offset = fop->user_size + fop->head;
+    tail = fop->size - offset;
+    head = ec->stripe_size - tail;
+
+    memcpy(fop->vector[0].iov_base + offset, stripe->data + head, tail);
+}
+
+static ec_stripe_t *
+ec_get_stripe_from_cache_locked(ec_t *ec, ec_fop_data_t *fop,
+                                uint64_t frag_offset)
+{
+    ec_inode_t *ctx = NULL;
+    ec_stripe_t *stripe = NULL;
+    ec_stripe_list_t *stripe_cache = NULL;
+
+    ctx = __ec_inode_get(fop->fd->inode, fop->xl);
+    if (ctx == NULL) {
+        GF_ATOMIC_INC(ec->stats.stripe_cache.errors);
+        return NULL;
+    }
+
+    stripe_cache = &ctx->stripe_cache;
+    list_for_each_entry(stripe, &stripe_cache->lru, lru)
+    {
+        if (stripe->frag_offset == frag_offset) {
+            list_move_tail(&stripe->lru, &stripe_cache->lru);
+            GF_ATOMIC_INC(ec->stats.stripe_cache.hits);
+            return stripe;
+        }
+    }
+
+    GF_ATOMIC_INC(ec->stats.stripe_cache.misses);
+
+    return NULL;
+}
+
+static gf_boolean_t
+ec_get_and_merge_stripe(ec_t *ec, ec_fop_data_t *fop, ec_stripe_part_t which)
+{
+    uint64_t frag_offset;
+    ec_stripe_t *stripe = NULL;
+    gf_boolean_t found = _gf_false;
+
+    if (!ec->stripe_cache) {
+        return found;
+    }
+
+    LOCK(&fop->fd->inode->lock);
+    if (which == EC_STRIPE_HEAD) {
+        frag_offset = fop->frag_range.first;
+        stripe = ec_get_stripe_from_cache_locked(ec, fop, frag_offset);
+        if (stripe) {
+            ec_merge_stripe_head_locked(ec, fop, stripe);
+            found = _gf_true;
+        }
+    }
+
+    if (which == EC_STRIPE_TAIL) {
+        frag_offset = fop->frag_range.last - ec->fragment_size;
+        stripe = ec_get_stripe_from_cache_locked(ec, fop, frag_offset);
+        if (stripe) {
+            ec_merge_stripe_tail_locked(ec, fop, stripe);
+            found = _gf_true;
+        }
+    }
+    UNLOCK(&fop->fd->inode->lock);
+
+    return found;
+}
+
+static uintptr_t
+ec_get_lock_good_mask(inode_t *inode, xlator_t *xl)
+{
+    ec_lock_t *lock = NULL;
+    ec_inode_t *ictx = NULL;
+    LOCK(&inode->lock);
+    {
+        ictx = __ec_inode_get(inode, xl);
+        if (ictx)
+            lock = ictx->inode_lock;
+    }
+    UNLOCK(&inode->lock);
+    if (lock)
+        return lock->good_mask;
+    return 0;
+}
+
+void
+ec_writev_start(ec_fop_data_t *fop)
+{
+    ec_t *ec = fop->xl->private;
+    ec_fd_t *ctx;
+    fd_t *fd;
+    dict_t *xdata = NULL;
+    uint64_t tail, current;
+    int32_t err = -ENOMEM;
+    gf_boolean_t found_stripe = _gf_false;
+
+    /* This shouldn't fail because we have the inode locked. */
+    GF_ASSERT(ec_get_inode_size(fop, fop->fd->inode, &current));
+
+    fd = fd_anonymous(fop->fd->inode);
+    if (fd == NULL) {
+        goto failed;
+    }
+
+    fop->frame->root->uid = 0;
+    fop->frame->root->gid = 0;
+
+    ctx = ec_fd_get(fop->fd, fop->xl);
+    if (ctx != NULL) {
+        if ((ctx->flags & O_APPEND) != 0) {
+            /* Appending writes take full locks so size won't change because
+             * of any parallel operations
+             */
+            fop->offset = current;
+        }
+    }
+
+    err = ec_writev_prepare_buffers(ec, fop);
+    if (err != 0) {
+        goto failed_fd;
+    }
+    tail = fop->size - fop->user_size - fop->head;
+    if (fop->head > 0) {
+        if (current > fop->offset) {
+            found_stripe = ec_get_and_merge_stripe(ec, fop, EC_STRIPE_HEAD);
+            if (!found_stripe) {
+                if (ec_make_internal_fop_xdata(&xdata)) {
+                    err = -ENOMEM;
+                    goto failed_xdata;
+                }
+                ec_readv(fop->frame, fop->xl,
+                         ec_get_lock_good_mask(fop->fd->inode, fop->xl),
+                         EC_MINIMUM_MIN, ec_writev_merge_head, NULL, fd,
+                         ec->stripe_size, fop->offset, 0, xdata);
+            }
+        } else {
+            memset(fop->vector[0].iov_base, 0, fop->head);
+            memset(fop->vector[0].iov_base + fop->size - tail, 0, tail);
+            if (ec->stripe_cache && (fop->size <= ec->stripe_size)) {
+                ec_add_stripe_in_cache(ec, fop);
+            }
+        }
+    }
+
+    if ((tail > 0) && ((fop->head == 0) || (fop->size > ec->stripe_size))) {
+        /* Current locking scheme will make sure the 'current' below will
+         * never decrease while the fop is in progress, so the checks will
+         * work as expected
+         */
+        if (current > fop->offset + fop->head + fop->user_size) {
+            found_stripe = ec_get_and_merge_stripe(ec, fop, EC_STRIPE_TAIL);
+            if (!found_stripe) {
+                if (ec_make_internal_fop_xdata(&xdata)) {
+                    err = -ENOMEM;
+                    goto failed_xdata;
+                }
+                ec_readv(fop->frame, fop->xl,
+                         ec_get_lock_good_mask(fop->fd->inode, fop->xl),
+                         EC_MINIMUM_MIN, ec_writev_merge_tail, NULL, fd,
+                         ec->stripe_size,
+                         fop->offset + fop->size - ec->stripe_size, 0, xdata);
+            }
+        } else {
+            memset(fop->vector[0].iov_base + fop->size - tail, 0, tail);
+            if (ec->stripe_cache) {
+                ec_add_stripe_in_cache(ec, fop);
+            }
+        }
+    }
+
+    err = 0;
+
+failed_xdata:
+    if (xdata) {
+        dict_unref(xdata);
+    }
+failed_fd:
+    fd_unref(fd);
+failed:
+    ec_fop_set_error(fop, -err);
+}
+
+int32_t
+ec_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct iatt *prestat, struct iatt *poststat,
+              dict_t *xdata)
+{
+    ec_t *ec = NULL;
+    if (this && this->private) {
+        ec = this->private;
+        if ((op_ret > 0) && ((op_ret % ec->fragment_size) != 0)) {
+            op_ret = -1;
+            op_errno = EIO;
+        }
+    }
+    return ec_inode_write_cbk(frame, this, cookie, op_ret, op_errno, prestat,
+                              poststat, xdata);
+}
+
+void
+ec_wind_writev(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    struct iovec vector[1];
+    size_t size;
+
+    size = fop->vector[1].iov_len;
+
+    vector[0].iov_base = fop->vector[1].iov_base + idx * size;
+    vector[0].iov_len = size;
+
+    STACK_WIND_COOKIE(fop->frame, ec_writev_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->writev, fop->fd,
+                      vector, 1, fop->offset / ec->fragments, fop->uint32,
+                      fop->buffers, fop->xdata);
+}
+
+static void
+ec_writev_encode(ec_fop_data_t *fop)
+{
+    ec_t *ec = fop->xl->private;
+    void *blocks[ec->nodes];
+    uint32_t i;
+
+    blocks[0] = fop->vector[1].iov_base;
+    for (i = 1; i < ec->nodes; i++) {
+        blocks[i] = blocks[i - 1] + fop->vector[1].iov_len;
+    }
+    ec_method_encode(&ec->matrix, fop->vector[0].iov_len,
+                     fop->vector[0].iov_base, blocks);
+}
+
+int32_t
+ec_manager_writev(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+    ec_fd_t *ctx = NULL;
+    ec_t *ec = fop->xl->private;
+    off_t fl_start = 0;
+    uint64_t fl_size = LONG_MAX;
+
+    switch (state) {
+        case EC_STATE_INIT:
+        case EC_STATE_LOCK:
+            ctx = ec_fd_get(fop->fd, fop->xl);
+            if (ctx != NULL) {
+                if ((ctx->flags & O_APPEND) == 0) {
+                    off_t user_size = 0;
+                    off_t head = 0;
+
+                    fl_start = fop->offset;
+                    user_size = iov_length(fop->vector, fop->int32);
+                    head = ec_adjust_offset_down(ec, &fl_start, _gf_true);
+                    fl_size = user_size + head;
+                    ec_adjust_size_up(ec, &fl_size, _gf_true);
+                }
+            }
+            ec_lock_prepare_fd(fop, fop->fd,
+                               EC_UPDATE_DATA | EC_UPDATE_META | EC_QUERY_INFO,
+                               fl_start, fl_size);
+            ec_lock(fop);
+
+            return EC_STATE_DISPATCH;
+
+        case EC_STATE_DISPATCH:
+            ec_writev_start(fop);
+
+            return EC_STATE_DELAYED_START;
+
+        case EC_STATE_DELAYED_START:
+            /* Restore uid, gid if they were changed to do some partial
+             * reads. */
+            fop->frame->root->uid = fop->uid;
+            fop->frame->root->gid = fop->gid;
+
+            ec_writev_encode(fop);
+
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+            cbk = ec_fop_prepare_answer(fop, _gf_false);
+            if (cbk != NULL) {
+                ec_t *ec = fop->xl->private;
+                uint64_t size;
+
+                ec_iatt_rebuild(fop->xl->private, cbk->iatt, 2, cbk->count);
+
+                /* This shouldn't fail because we have the inode locked. */
+                LOCK(&fop->fd->inode->lock);
+                {
+                    GF_ASSERT(__ec_get_inode_size(fop, fop->fd->inode,
+                                                  &cbk->iatt[0].ia_size));
+                    cbk->iatt[1].ia_size = cbk->iatt[0].ia_size;
+                    size = fop->offset + fop->head + fop->user_size;
+                    if (size > cbk->iatt[0].ia_size) {
+                        /* Only update inode size if this is a top level fop.
+                         * Otherwise this is an internal write and the top
+                         * level fop should take care of the real inode size.
+                         */
+                        if (fop->parent == NULL) {
+                            /* This shouldn't fail because we have the inode
+                             * locked. */
+                            GF_ASSERT(
+                                __ec_set_inode_size(fop, fop->fd->inode, size));
+                        }
+                        cbk->iatt[1].ia_size = size;
+                    }
+                }
+                UNLOCK(&fop->fd->inode->lock);
+
+                if (fop->error == 0) {
+                    cbk->op_ret *= ec->fragments;
+                    if (cbk->op_ret < fop->head) {
+                        cbk->op_ret = 0;
+                    } else {
+                        cbk->op_ret -= fop->head;
+                    }
+                    if (cbk->op_ret > fop->user_size) {
+                        cbk->op_ret = fop->user_size;
+                    }
+                }
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.writev != NULL) {
+                QUORUM_CBK(fop->cbks.writev, fop, fop->req_frame, fop, fop->xl,
+                           cbk->op_ret, cbk->op_errno, &cbk->iatt[0],
+                           &cbk->iatt[1], cbk->xdata);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_DELAYED_START:
+            /* We have failed while doing partial reads. We need to restore
+             * original uid, gid. */
+            fop->frame->root->uid = fop->uid;
+            fop->frame->root->gid = fop->gid;
+
+            /* Fall through */
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_LOCK:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.writev != NULL) {
+                fop->cbks.writev(fop->req_frame, fop, fop->xl, -1, fop->error,
+                                 NULL, NULL, NULL);
+            }
+
+            return EC_STATE_LOCK_REUSE;
+
+        case -EC_STATE_LOCK_REUSE:
+        case EC_STATE_LOCK_REUSE:
+            ec_lock_reuse(fop);
+
+            return EC_STATE_UNLOCK;
+
+        case -EC_STATE_UNLOCK:
+        case EC_STATE_UNLOCK:
+            ec_unlock(fop);
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_writev(call_frame_t *frame, xlator_t *this, uintptr_t target,
+          uint32_t fop_flags, fop_writev_cbk_t func, void *data, fd_t *fd,
+          struct iovec *vector, int32_t count, off_t offset, uint32_t flags,
+          struct iobref *iobref, dict_t *xdata)
+{
+    ec_cbk_t callback = {.writev = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(WRITE) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_WRITE, 0, target, fop_flags,
+                               ec_wind_writev, ec_manager_writev, callback,
+                               data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->int32 = count;
+    fop->offset = offset;
+    fop->uint32 = flags;
+
+    fop->use_fd = 1;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (count > 0) {
+        fop->vector = iov_dup(vector, count);
+        if (fop->vector == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a "
+                   "vector list.");
+
+            goto out;
+        }
+        fop->int32 = count;
+    }
+    if (iobref != NULL) {
+        fop->buffers = iobref_ref(iobref);
+        if (fop->buffers == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_BUF_REF_FAIL,
+                   "Failed to reference a "
+                   "buffer.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_copy_with_ref(xdata, NULL);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL, NULL);
+    }
+}
diff --git a/xlators/cluster/ec/src/ec-locks.c b/xlators/cluster/ec/src/ec-locks.c
new file mode 100644
index 00000000000..601960d6154
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-locks.c
@@ -0,0 +1,1128 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "ec-helpers.h"
+#include "ec-common.h"
+#include "ec-combine.h"
+#include "ec-fops.h"
+#include "ec-messages.h"
+
+#define EC_LOCK_MODE_NONE 0
+#define EC_LOCK_MODE_INC 1
+#define EC_LOCK_MODE_ALL 2
+
+int32_t
+ec_lock_check(ec_fop_data_t *fop, uintptr_t *mask)
+{
+    ec_t *ec = fop->xl->private;
+    ec_cbk_data_t *ans = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    uintptr_t locked = 0;
+    int32_t good = 0;
+    int32_t eagain = 0;
+    int32_t estale = 0;
+    int32_t error = -1;
+
+    /* There are some errors that we'll handle in an special way while trying
+     * to acquire a lock.
+     *
+     *   EAGAIN:  If it's found during a parallel non-blocking lock request, we
+     *            consider that there's contention on the inode, so we consider
+     *            the acquisition a failure and try again with a sequential
+     *            blocking lock request. This will ensure that we get a lock on
+     *            as many bricks as possible (ignoring EAGAIN here would cause
+     *            unnecessary triggers of self-healing).
+     *
+     *            If it's found during a sequential blocking lock request, it's
+     *            considered an error. Lock will only succeed if there are
+     *            enough other bricks locked.
+     *
+     *   ESTALE:  This can appear during parallel or sequential lock request if
+     *            the inode has just been unlinked. We consider this error is
+     *            not recoverable, but we also don't consider it as fatal. So,
+     *            if it happens during parallel lock, we won't attempt a
+     *            sequential one unless there are EAGAIN errors on other
+     *            bricks (and are enough to form a quorum), but if we reach
+     *            quorum counting the ESTALE bricks, we consider the whole
+     *            result of the operation is ESTALE instead of EIO.
+     */
+
+    list_for_each_entry(ans, &fop->cbk_list, list)
+    {
+        if (ans->op_ret >= 0) {
+            if (locked != 0) {
+                error = EIO;
+            }
+            locked |= ans->mask;
+            good = ans->count;
+            cbk = ans;
+        } else if (ans->op_errno == ESTALE) {
+            estale += ans->count;
+        } else if ((ans->op_errno == EAGAIN) &&
+                   (fop->uint32 != EC_LOCK_MODE_INC)) {
+            eagain += ans->count;
+        }
+    }
+
+    if (error == -1) {
+        /* If we have enough quorum with succeeded and EAGAIN answers, we
+         * ignore for now any ESTALE answer. If there are EAGAIN answers,
+         * we retry with a sequential blocking lock request if needed.
+         * Otherwise we succeed. */
+        if ((good + eagain) >= ec->fragments) {
+            if (eagain == 0) {
+                if (fop->answer == NULL) {
+                    fop->answer = cbk;
+                }
+
+                ec_update_good(fop, locked);
+
+                error = 0;
+            } else {
+                switch (fop->uint32) {
+                    case EC_LOCK_MODE_NONE:
+                        error = EAGAIN;
+                        break;
+                    case EC_LOCK_MODE_ALL:
+                        fop->uint32 = EC_LOCK_MODE_INC;
+                        break;
+                    default:
+                        /* This shouldn't happen because eagain cannot be > 0
+                         * when fop->uint32 is EC_LOCK_MODE_INC. */
+                        error = EIO;
+                        break;
+                }
+            }
+        } else {
+            /* We have been unable to find enough candidates that will be able
+             * to take the lock. If we have quorum on some answer, we return
+             * it. Otherwise we check if ESTALE answers allow us to reach
+             * quorum. If so, we return ESTALE. */
+            if (fop->answer && fop->answer->op_ret < 0) {
+                error = fop->answer->op_errno;
+            } else if ((good + eagain + estale) >= ec->fragments) {
+                error = ESTALE;
+            } else {
+                error = EIO;
+            }
+        }
+    }
+
+    *mask = locked;
+
+    return error;
+}
+
+int32_t
+ec_lock_unlocked(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, EC_MSG_UNLOCK_FAILED,
+               "Failed to unlock an entry/inode");
+    }
+
+    return 0;
+}
+
+int32_t
+ec_lock_lk_unlocked(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct gf_flock *flock,
+                    dict_t *xdata)
+{
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, EC_MSG_LK_UNLOCK_FAILED,
+               "Failed to unlock an lk");
+    }
+
+    return 0;
+}
+
+/* FOP: entrylk */
+
+int32_t
+ec_entrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_ENTRYLK, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_entrylk(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_entrylk_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->entrylk,
+                      fop->str[0], &fop->loc[0], fop->str[1], fop->entrylk_cmd,
+                      fop->entrylk_type, fop->xdata);
+}
+
+int32_t
+ec_manager_entrylk(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            if (fop->entrylk_cmd == ENTRYLK_LOCK) {
+                fop->uint32 = EC_LOCK_MODE_ALL;
+                fop->entrylk_cmd = ENTRYLK_LOCK_NB;
+            }
+
+            /* Fall through */
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_PREPARE_ANSWER:
+            if (fop->entrylk_cmd != ENTRYLK_UNLOCK) {
+                uintptr_t mask;
+
+                ec_fop_set_error(fop, ec_lock_check(fop, &mask));
+                if (fop->error != 0) {
+                    if (mask != 0) {
+                        if (fop->id == GF_FOP_ENTRYLK) {
+                            ec_entrylk(
+                                fop->frame, fop->xl, mask, 1, ec_lock_unlocked,
+                                NULL, fop->str[0], &fop->loc[0], fop->str[1],
+                                ENTRYLK_UNLOCK, fop->entrylk_type, fop->xdata);
+                        } else {
+                            ec_fentrylk(fop->frame, fop->xl, mask, 1,
+                                        ec_lock_unlocked, NULL, fop->str[0],
+                                        fop->fd, fop->str[1], ENTRYLK_UNLOCK,
+                                        fop->entrylk_type, fop->xdata);
+                        }
+                    }
+                    if (fop->error < 0) {
+                        fop->error = 0;
+
+                        fop->entrylk_cmd = ENTRYLK_LOCK;
+
+                        ec_dispatch_inc(fop);
+
+                        return EC_STATE_PREPARE_ANSWER;
+                    }
+                }
+            } else {
+                ec_fop_prepare_answer(fop, _gf_true);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->id == GF_FOP_ENTRYLK) {
+                if (fop->cbks.entrylk != NULL) {
+                    fop->cbks.entrylk(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                      cbk->op_errno, cbk->xdata);
+                }
+            } else {
+                if (fop->cbks.fentrylk != NULL) {
+                    fop->cbks.fentrylk(fop->req_frame, fop, fop->xl,
+                                       cbk->op_ret, cbk->op_errno, cbk->xdata);
+                }
+            }
+
+            return EC_STATE_END;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->id == GF_FOP_ENTRYLK) {
+                if (fop->cbks.entrylk != NULL) {
+                    fop->cbks.entrylk(fop->req_frame, fop, fop->xl, -1,
+                                      fop->error, NULL);
+                }
+            } else {
+                if (fop->cbks.fentrylk != NULL) {
+                    fop->cbks.fentrylk(fop->req_frame, fop, fop->xl, -1,
+                                       fop->error, NULL);
+                }
+            }
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_entrylk(call_frame_t *frame, xlator_t *this, uintptr_t target,
+           uint32_t fop_flags, fop_entrylk_cbk_t func, void *data,
+           const char *volume, loc_t *loc, const char *basename,
+           entrylk_cmd cmd, entrylk_type type, dict_t *xdata)
+{
+    ec_cbk_t callback = {.entrylk = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(ENTRYLK) %p", frame);
+
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_ENTRYLK, 0, target,
+                               fop_flags, ec_wind_entrylk, ec_manager_entrylk,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->entrylk_cmd = cmd;
+    fop->entrylk_type = type;
+
+    if (volume != NULL) {
+        fop->str[0] = gf_strdup(volume);
+        if (fop->str[0] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (basename != NULL) {
+        fop->str[1] = gf_strdup(basename);
+        if (fop->str[1] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/* FOP: fentrylk */
+
+int32_t
+ec_fentrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_FENTRYLK, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_fentrylk(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_fentrylk_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->fentrylk,
+                      fop->str[0], fop->fd, fop->str[1], fop->entrylk_cmd,
+                      fop->entrylk_type, fop->xdata);
+}
+
+void
+ec_fentrylk(call_frame_t *frame, xlator_t *this, uintptr_t target,
+            uint32_t fop_flags, fop_fentrylk_cbk_t func, void *data,
+            const char *volume, fd_t *fd, const char *basename, entrylk_cmd cmd,
+            entrylk_type type, dict_t *xdata)
+{
+    ec_cbk_t callback = {.fentrylk = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FENTRYLK) %p", frame);
+
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FENTRYLK, 0, target,
+                               fop_flags, ec_wind_fentrylk, ec_manager_entrylk,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->entrylk_cmd = cmd;
+    fop->entrylk_type = type;
+
+    if (volume != NULL) {
+        fop->str[0] = gf_strdup(volume);
+        if (fop->str[0] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (basename != NULL) {
+        fop->str[1] = gf_strdup(basename);
+        if (fop->str[1] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/* FOP: inodelk */
+
+int32_t
+ec_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_INODELK, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_inodelk(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_inodelk_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->inodelk,
+                      fop->str[0], &fop->loc[0], fop->int32, &fop->flock,
+                      fop->xdata);
+}
+
+int32_t
+ec_manager_inodelk(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            fop->flock.l_len += ec_adjust_offset_down(
+                fop->xl->private, &fop->flock.l_start, _gf_true);
+            ec_adjust_offset_up(fop->xl->private, &fop->flock.l_len, _gf_true);
+            if ((fop->int32 == F_SETLKW) && (fop->flock.l_type != F_UNLCK)) {
+                fop->uint32 = EC_LOCK_MODE_ALL;
+                fop->int32 = F_SETLK;
+            }
+
+            /* Fall through */
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_PREPARE_ANSWER:
+            if (fop->flock.l_type != F_UNLCK) {
+                uintptr_t mask;
+
+                ec_fop_set_error(fop, ec_lock_check(fop, &mask));
+                if (fop->error != 0) {
+                    if (mask != 0) {
+                        ec_t *ec = fop->xl->private;
+                        struct gf_flock flock;
+
+                        flock.l_type = F_UNLCK;
+                        flock.l_whence = fop->flock.l_whence;
+                        flock.l_start = fop->flock.l_start * ec->fragments;
+                        flock.l_len = fop->flock.l_len * ec->fragments;
+                        flock.l_pid = 0;
+                        flock.l_owner.len = 0;
+
+                        if (fop->id == GF_FOP_INODELK) {
+                            ec_inodelk(fop->frame, fop->xl,
+                                       &fop->frame->root->lk_owner, mask, 1,
+                                       ec_lock_unlocked, NULL, fop->str[0],
+                                       &fop->loc[0], F_SETLK, &flock,
+                                       fop->xdata);
+                        } else {
+                            ec_finodelk(fop->frame, fop->xl,
+                                        &fop->frame->root->lk_owner, mask, 1,
+                                        ec_lock_unlocked, NULL, fop->str[0],
+                                        fop->fd, F_SETLK, &flock, fop->xdata);
+                        }
+                    }
+                    if (fop->error < 0) {
+                        fop->error = 0;
+
+                        fop->int32 = F_SETLKW;
+
+                        ec_dispatch_inc(fop);
+
+                        return EC_STATE_PREPARE_ANSWER;
+                    }
+                }
+            } else {
+                ec_fop_prepare_answer(fop, _gf_true);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->id == GF_FOP_INODELK) {
+                if (fop->cbks.inodelk != NULL) {
+                    fop->cbks.inodelk(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                                      cbk->op_errno, cbk->xdata);
+                }
+            } else {
+                if (fop->cbks.finodelk != NULL) {
+                    fop->cbks.finodelk(fop->req_frame, fop, fop->xl,
+                                       cbk->op_ret, cbk->op_errno, cbk->xdata);
+                }
+            }
+
+            return EC_STATE_END;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->id == GF_FOP_INODELK) {
+                if (fop->cbks.inodelk != NULL) {
+                    fop->cbks.inodelk(fop->req_frame, fop, fop->xl, -1,
+                                      fop->error, NULL);
+                }
+            } else {
+                if (fop->cbks.finodelk != NULL) {
+                    fop->cbks.finodelk(fop->req_frame, fop, fop->xl, -1,
+                                       fop->error, NULL);
+                }
+            }
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_inodelk(call_frame_t *frame, xlator_t *this, gf_lkowner_t *owner,
+           uintptr_t target, uint32_t fop_flags, fop_inodelk_cbk_t func,
+           void *data, const char *volume, loc_t *loc, int32_t cmd,
+           struct gf_flock *flock, dict_t *xdata)
+{
+    ec_cbk_t callback = {.inodelk = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(INODELK) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_INODELK, 0, target,
+                               fop_flags, ec_wind_inodelk, ec_manager_inodelk,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->int32 = cmd;
+    ec_owner_copy(fop->frame, owner);
+
+    if (volume != NULL) {
+        fop->str[0] = gf_strdup(volume);
+        if (fop->str[0] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (loc != NULL) {
+        if (loc_copy(&fop->loc[0], loc) != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_LOC_COPY_FAIL,
+                   "Failed to copy a location.");
+
+            goto out;
+        }
+    }
+    if (flock != NULL) {
+        fop->flock.l_type = flock->l_type;
+        fop->flock.l_whence = flock->l_whence;
+        fop->flock.l_start = flock->l_start;
+        fop->flock.l_len = flock->l_len;
+        fop->flock.l_pid = flock->l_pid;
+        fop->flock.l_owner.len = flock->l_owner.len;
+        if (flock->l_owner.len > 0) {
+            memcpy(fop->flock.l_owner.data, flock->l_owner.data,
+                   flock->l_owner.len);
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/* FOP: finodelk */
+
+int32_t
+ec_finodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_FINODELK, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, NULL);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_finodelk(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_finodelk_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->finodelk,
+                      fop->str[0], fop->fd, fop->int32, &fop->flock,
+                      fop->xdata);
+}
+
+void
+ec_finodelk(call_frame_t *frame, xlator_t *this, gf_lkowner_t *owner,
+            uintptr_t target, uint32_t fop_flags, fop_finodelk_cbk_t func,
+            void *data, const char *volume, fd_t *fd, int32_t cmd,
+            struct gf_flock *flock, dict_t *xdata)
+{
+    ec_cbk_t callback = {.finodelk = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(FINODELK) %p", frame);
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_FINODELK, 0, target,
+                               fop_flags, ec_wind_finodelk, ec_manager_inodelk,
+                               callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->int32 = cmd;
+    ec_owner_copy(fop->frame, owner);
+
+    if (volume != NULL) {
+        fop->str[0] = gf_strdup(volume);
+        if (fop->str[0] == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+                   "Failed to duplicate a string.");
+
+            goto out;
+        }
+    }
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (flock != NULL) {
+        fop->flock.l_type = flock->l_type;
+        fop->flock.l_whence = flock->l_whence;
+        fop->flock.l_start = flock->l_start;
+        fop->flock.l_len = flock->l_len;
+        fop->flock.l_pid = flock->l_pid;
+        fop->flock.l_owner.len = flock->l_owner.len;
+        if (flock->l_owner.len > 0) {
+            memcpy(fop->flock.l_owner.data, flock->l_owner.data,
+                   flock->l_owner.len);
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL);
+    }
+}
+
+/* FOP: lk */
+
+int32_t
+ec_combine_lk(ec_fop_data_t *fop, ec_cbk_data_t *dst, ec_cbk_data_t *src)
+{
+    if (!ec_flock_compare(&dst->flock, &src->flock)) {
+        gf_msg(fop->xl->name, GF_LOG_NOTICE, 0, EC_MSG_LOCK_MISMATCH,
+               "Mismatching lock in "
+               "answers of 'GF_FOP_LK'");
+
+        return 0;
+    }
+
+    return 1;
+}
+
+int32_t
+ec_lk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+          int32_t op_errno, struct gf_flock *flock, dict_t *xdata)
+{
+    ec_fop_data_t *fop = NULL;
+    ec_cbk_data_t *cbk = NULL;
+    int32_t idx = (int32_t)(uintptr_t)cookie;
+
+    VALIDATE_OR_GOTO(this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = frame->local;
+
+    ec_trace("CBK", fop, "idx=%d, frame=%p, op_ret=%d, op_errno=%d", idx, frame,
+             op_ret, op_errno);
+
+    cbk = ec_cbk_data_allocate(frame, this, fop, GF_FOP_LK, idx, op_ret,
+                               op_errno);
+    if (cbk != NULL) {
+        if (op_ret >= 0) {
+            if (flock != NULL) {
+                cbk->flock.l_type = flock->l_type;
+                cbk->flock.l_whence = flock->l_whence;
+                cbk->flock.l_start = flock->l_start;
+                cbk->flock.l_len = flock->l_len;
+                cbk->flock.l_pid = flock->l_pid;
+                cbk->flock.l_owner.len = flock->l_owner.len;
+                if (flock->l_owner.len > 0) {
+                    memcpy(cbk->flock.l_owner.data, flock->l_owner.data,
+                           flock->l_owner.len);
+                }
+            }
+        }
+        if (xdata != NULL) {
+            cbk->xdata = dict_ref(xdata);
+            if (cbk->xdata == NULL) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                       "Failed to reference a "
+                       "dictionary.");
+
+                goto out;
+            }
+        }
+
+        ec_combine(cbk, ec_combine_lk);
+    }
+
+out:
+    if (fop != NULL) {
+        ec_complete(fop);
+    }
+
+    return 0;
+}
+
+void
+ec_wind_lk(ec_t *ec, ec_fop_data_t *fop, int32_t idx)
+{
+    ec_trace("WIND", fop, "idx=%d", idx);
+
+    STACK_WIND_COOKIE(fop->frame, ec_lk_cbk, (void *)(uintptr_t)idx,
+                      ec->xl_list[idx], ec->xl_list[idx]->fops->lk, fop->fd,
+                      fop->int32, &fop->flock, fop->xdata);
+}
+
+int32_t
+ec_manager_lk(ec_fop_data_t *fop, int32_t state)
+{
+    ec_cbk_data_t *cbk;
+
+    switch (state) {
+        case EC_STATE_INIT:
+            if ((fop->int32 == F_SETLKW) && (fop->flock.l_type != F_UNLCK)) {
+                fop->uint32 = EC_LOCK_MODE_ALL;
+                fop->int32 = F_SETLK;
+            }
+
+            /* Fall through */
+
+        case EC_STATE_DISPATCH:
+            ec_dispatch_all(fop);
+
+            return EC_STATE_PREPARE_ANSWER;
+
+        case EC_STATE_PREPARE_ANSWER:
+        case -EC_STATE_PREPARE_ANSWER:
+            if (fop->flock.l_type != F_UNLCK) {
+                uintptr_t mask;
+
+                ec_fop_set_error(fop, ec_lock_check(fop, &mask));
+                if (fop->error != 0) {
+                    if (mask != 0) {
+                        struct gf_flock flock = {0};
+
+                        flock.l_type = F_UNLCK;
+                        flock.l_whence = fop->flock.l_whence;
+                        flock.l_start = fop->flock.l_start;
+                        flock.l_len = fop->flock.l_len;
+                        flock.l_pid = fop->flock.l_pid;
+                        lk_owner_copy(&flock.l_owner, &fop->flock.l_owner);
+
+                        ec_lk(fop->frame, fop->xl, mask, 1, ec_lock_lk_unlocked,
+                              NULL, fop->fd, F_SETLK, &flock, fop->xdata);
+                    }
+
+                    if (fop->error < 0) {
+                        fop->error = 0;
+
+                        fop->int32 = F_SETLKW;
+
+                        ec_dispatch_inc(fop);
+
+                        return EC_STATE_PREPARE_ANSWER;
+                    }
+                }
+            } else {
+                ec_fop_prepare_answer(fop, _gf_true);
+            }
+
+            return EC_STATE_REPORT;
+
+        case EC_STATE_REPORT:
+            cbk = fop->answer;
+
+            GF_ASSERT(cbk != NULL);
+
+            if (fop->cbks.lk != NULL) {
+                fop->cbks.lk(fop->req_frame, fop, fop->xl, cbk->op_ret,
+                             cbk->op_errno, &cbk->flock, cbk->xdata);
+            }
+
+            return EC_STATE_END;
+
+        case -EC_STATE_INIT:
+        case -EC_STATE_DISPATCH:
+        case -EC_STATE_REPORT:
+            GF_ASSERT(fop->error != 0);
+
+            if (fop->cbks.lk != NULL) {
+                fop->cbks.lk(fop->req_frame, fop, fop->xl, -1, fop->error, NULL,
+                             NULL);
+            }
+
+            return EC_STATE_END;
+
+        default:
+            gf_msg(fop->xl->name, GF_LOG_ERROR, EINVAL, EC_MSG_UNHANDLED_STATE,
+                   "Unhandled state %d for %s", state, ec_fop_name(fop->id));
+
+            return EC_STATE_END;
+    }
+}
+
+void
+ec_lk(call_frame_t *frame, xlator_t *this, uintptr_t target, uint32_t fop_flags,
+      fop_lk_cbk_t func, void *data, fd_t *fd, int32_t cmd,
+      struct gf_flock *flock, dict_t *xdata)
+{
+    ec_cbk_t callback = {.lk = func};
+    ec_fop_data_t *fop = NULL;
+    int32_t error = ENOMEM;
+
+    gf_msg_trace("ec", 0, "EC(LK) %p", frame);
+
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    fop = ec_fop_data_allocate(frame, this, GF_FOP_LK, 0, target, fop_flags,
+                               ec_wind_lk, ec_manager_lk, callback, data);
+    if (fop == NULL) {
+        goto out;
+    }
+
+    fop->use_fd = 1;
+
+    fop->int32 = cmd;
+
+    if (fd != NULL) {
+        fop->fd = fd_ref(fd);
+        if (fop->fd == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_FILE_DESC_REF_FAIL,
+                   "Failed to reference a "
+                   "file descriptor.");
+
+            goto out;
+        }
+    }
+    if (flock != NULL) {
+        fop->flock.l_type = flock->l_type;
+        fop->flock.l_whence = flock->l_whence;
+        fop->flock.l_start = flock->l_start;
+        fop->flock.l_len = flock->l_len;
+        fop->flock.l_pid = flock->l_pid;
+        fop->flock.l_owner.len = flock->l_owner.len;
+        if (flock->l_owner.len > 0) {
+            memcpy(fop->flock.l_owner.data, flock->l_owner.data,
+                   flock->l_owner.len);
+        }
+    }
+    if (xdata != NULL) {
+        fop->xdata = dict_ref(xdata);
+        if (fop->xdata == NULL) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_DICT_REF_FAIL,
+                   "Failed to reference a "
+                   "dictionary.");
+
+            goto out;
+        }
+    }
+
+    error = 0;
+
+out:
+    if (fop != NULL) {
+        ec_manager(fop, error);
+    } else {
+        func(frame, NULL, this, -1, error, NULL, NULL);
+    }
+}
diff --git a/xlators/cluster/ec/src/ec-mem-types.h b/xlators/cluster/ec/src/ec-mem-types.h
new file mode 100644
index 00000000000..3252c4c1c58
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-mem-types.h
@@ -0,0 +1,30 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_MEM_TYPES_H__
+#define __EC_MEM_TYPES_H__
+
+#include <glusterfs/mem-types.h>
+
+enum gf_ec_mem_types_ {
+    ec_mt_ec_t = gf_common_mt_end + 1,
+    ec_mt_xlator_t,
+    ec_mt_ec_inode_t,
+    ec_mt_ec_fd_t,
+    ec_mt_subvol_healer_t,
+    ec_mt_ec_gf_t,
+    ec_mt_ec_code_t,
+    ec_mt_ec_code_builder_t,
+    ec_mt_ec_matrix_t,
+    ec_mt_ec_stripe_t,
+    ec_mt_end
+};
+
+#endif /* __EC_MEM_TYPES_H__ */
diff --git a/xlators/cluster/ec/src/ec-messages.h b/xlators/cluster/ec/src/ec-messages.h
new file mode 100644
index 00000000000..72e98f11286
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-messages.h
@@ -0,0 +1,61 @@
+/*
+  Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _EC_MESSAGES_H_
+#define _EC_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(EC, EC_MSG_INVALID_CONFIG, EC_MSG_HEAL_FAIL,
+           EC_MSG_DICT_COMBINE_FAIL, EC_MSG_STIME_COMBINE_FAIL,
+           EC_MSG_INVALID_DICT_NUMS, EC_MSG_IATT_COMBINE_FAIL,
+           EC_MSG_INVALID_FORMAT, EC_MSG_DICT_GET_FAILED,
+           EC_MSG_UNHANDLED_STATE, EC_MSG_FILE_DESC_REF_FAIL,
+           EC_MSG_LOC_COPY_FAIL, EC_MSG_BUF_REF_FAIL, EC_MSG_DICT_REF_FAIL,
+           EC_MSG_LK_UNLOCK_FAILED, EC_MSG_UNLOCK_FAILED,
+           EC_MSG_LOC_PARENT_INODE_MISSING, EC_MSG_INVALID_LOC_NAME,
+           EC_MSG_NO_MEMORY, EC_MSG_GFID_MISMATCH, EC_MSG_UNSUPPORTED_VERSION,
+           EC_MSG_FD_CREATE_FAIL, EC_MSG_READDIRP_REQ_PREP_FAIL,
+           EC_MSG_LOOKUP_REQ_PREP_FAIL, EC_MSG_INODE_REF_FAIL,
+           EC_MSG_LOOKUP_READAHEAD_FAIL, EC_MSG_FRAME_MISMATCH,
+           EC_MSG_XLATOR_MISMATCH, EC_MSG_VECTOR_MISMATCH, EC_MSG_IATT_MISMATCH,
+           EC_MSG_FD_MISMATCH, EC_MSG_DICT_MISMATCH, EC_MSG_INDEX_DIR_GET_FAIL,
+           EC_MSG_PREOP_LOCK_FAILED, EC_MSG_CHILDS_INSUFFICIENT,
+           EC_MSG_OP_EXEC_UNAVAIL, EC_MSG_UNLOCK_DELAY_FAILED,
+           EC_MSG_SIZE_VERS_UPDATE_FAIL, EC_MSG_INVALID_REQUEST,
+           EC_MSG_INVALID_LOCK_TYPE, EC_MSG_SIZE_VERS_GET_FAIL,
+           EC_MSG_FILE_SIZE_GET_FAIL, EC_MSG_FOP_MISMATCH,
+           EC_MSG_SUBVOL_ID_DICT_SET_FAIL, EC_MSG_SUBVOL_BUILD_FAIL,
+           EC_MSG_XLATOR_INIT_FAIL, EC_MSG_NO_PARENTS, EC_MSG_TIMER_CREATE_FAIL,
+           EC_MSG_TOO_MANY_SUBVOLS, EC_MSG_DATA_UNAVAILABLE,
+           EC_MSG_INODE_REMOVE_FAIL, EC_MSG_INVALID_REDUNDANCY,
+           EC_MSG_XLATOR_PARSE_OPT_FAIL, EC_MSG_OP_FAIL_ON_SUBVOLS,
+           EC_MSG_INVALID_INODE, EC_MSG_LOCK_MISMATCH, EC_MSG_XDATA_MISMATCH,
+           EC_MSG_HEALING_INFO, EC_MSG_HEAL_SUCCESS, EC_MSG_FULL_SWEEP_START,
+           EC_MSG_FULL_SWEEP_STOP, EC_MSG_INVALID_FOP, EC_MSG_EC_UP,
+           EC_MSG_EC_DOWN, EC_MSG_SIZE_XATTR_GET_FAIL,
+           EC_MSG_VER_XATTR_GET_FAIL, EC_MSG_CONFIG_XATTR_GET_FAIL,
+           EC_MSG_CONFIG_XATTR_INVALID, EC_MSG_EXTENSION, EC_MSG_EXTENSION_NONE,
+           EC_MSG_EXTENSION_UNKNOWN, EC_MSG_EXTENSION_UNSUPPORTED,
+           EC_MSG_EXTENSION_FAILED, EC_MSG_NO_GF, EC_MSG_MATRIX_FAILED,
+           EC_MSG_DYN_CREATE_FAILED, EC_MSG_DYN_CODEGEN_FAILED,
+           EC_MSG_THREAD_CLEANUP_FAILED, EC_MSG_FD_BAD);
+
+#endif /* !_EC_MESSAGES_H_ */
diff --git a/xlators/cluster/ec/src/ec-method.c b/xlators/cluster/ec/src/ec-method.c
new file mode 100644
index 00000000000..55faed0b193
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-method.c
@@ -0,0 +1,433 @@
+/*
+  Copyright (c) 2012-2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <string.h>
+#include <inttypes.h>
+
+#include "ec-types.h"
+#include "ec-mem-types.h"
+#include "ec-galois.h"
+#include "ec-code.h"
+#include "ec-method.h"
+#include "ec-helpers.h"
+
+static void
+ec_method_matrix_normal(ec_gf_t *gf, uint32_t *matrix, uint32_t columns,
+                        uint32_t *values, uint32_t count)
+{
+    uint32_t i, j, v, tmp;
+
+    columns--;
+    for (i = 0; i < count; i++) {
+        v = *values++;
+        *matrix++ = tmp = ec_gf_exp(gf, v, columns);
+        for (j = 0; j < columns; j++) {
+            *matrix++ = tmp = ec_gf_div(gf, tmp, v);
+        }
+    }
+}
+
+static void
+ec_method_matrix_inverse(ec_gf_t *gf, uint32_t *matrix, uint32_t *values,
+                         uint32_t count)
+{
+    uint32_t a[count];
+    uint32_t i, j, p, last, tmp;
+
+    last = count - 1;
+    for (i = 0; i < last; i++) {
+        a[i] = 1;
+    }
+    a[i] = values[0];
+    for (i = last; i > 0; i--) {
+        for (j = i - 1; j < last; j++) {
+            a[j] = a[j + 1] ^ ec_gf_mul(gf, values[i], a[j]);
+        }
+        a[j] = ec_gf_mul(gf, values[i], a[j]);
+    }
+    for (i = 0; i < count; i++) {
+        p = a[0];
+        matrix += count;
+        *matrix = tmp = p ^ values[i];
+        for (j = 1; j < last; j++) {
+            matrix += count;
+            *matrix = tmp = a[j] ^ ec_gf_mul(gf, values[i], tmp);
+            p = tmp ^ ec_gf_mul(gf, values[i], p);
+        }
+        for (j = 0; j < last; j++) {
+            *matrix = ec_gf_div(gf, *matrix, p);
+            matrix -= count;
+        }
+        *matrix = ec_gf_div(gf, 1, p);
+        matrix++;
+    }
+}
+
+static void
+ec_method_matrix_init(ec_matrix_list_t *list, ec_matrix_t *matrix,
+                      uintptr_t mask, uint32_t *rows, gf_boolean_t inverse)
+{
+    uint32_t i;
+
+    matrix->refs = 1;
+    matrix->mask = mask;
+    matrix->code = list->code;
+    matrix->columns = list->columns;
+    INIT_LIST_HEAD(&matrix->lru);
+
+    if (inverse) {
+        matrix->rows = list->columns;
+        ec_method_matrix_inverse(matrix->code->gf, matrix->values, rows,
+                                 matrix->rows);
+        for (i = 0; i < matrix->rows; i++) {
+            matrix->row_data[i].values = matrix->values + i * matrix->columns;
+            matrix->row_data[i].func.interleaved = ec_code_build_interleaved(
+                matrix->code, EC_METHOD_WORD_SIZE, matrix->row_data[i].values,
+                matrix->columns);
+        }
+    } else {
+        matrix->rows = list->rows;
+        ec_method_matrix_normal(matrix->code->gf, matrix->values,
+                                matrix->columns, rows, matrix->rows);
+        for (i = 0; i < matrix->rows; i++) {
+            matrix->row_data[i].values = matrix->values + i * matrix->columns;
+            matrix->row_data[i].func.linear = ec_code_build_linear(
+                matrix->code, EC_METHOD_WORD_SIZE, matrix->row_data[i].values,
+                matrix->columns);
+        }
+    }
+}
+
+static void
+ec_method_matrix_release(ec_matrix_t *matrix)
+{
+    uint32_t i;
+
+    for (i = 0; i < matrix->rows; i++) {
+        if (matrix->row_data[i].func.linear != NULL) {
+            ec_code_release(matrix->code, &matrix->row_data[i].func);
+            matrix->row_data[i].func.linear = NULL;
+        }
+    }
+}
+
+static void
+ec_method_matrix_destroy(ec_matrix_list_t *list, ec_matrix_t *matrix)
+{
+    list_del_init(&matrix->lru);
+
+    ec_method_matrix_release(matrix);
+
+    mem_put(matrix);
+
+    list->count--;
+}
+
+static void
+ec_method_matrix_unref(ec_matrix_list_t *list, ec_matrix_t *matrix)
+{
+    if (--matrix->refs == 0) {
+        list_add_tail(&matrix->lru, &list->lru);
+        if (list->count > list->max) {
+            matrix = list_first_entry(&list->lru, ec_matrix_t, lru);
+            ec_method_matrix_destroy(list, matrix);
+        }
+    }
+}
+
+static ec_matrix_t *
+ec_method_matrix_lookup(ec_matrix_list_t *list, uintptr_t mask, uint32_t *pos)
+{
+    ec_matrix_t *matrix;
+    uint32_t i, j, k;
+
+    i = 0;
+    j = list->count;
+    while (i < j) {
+        k = (i + j) >> 1;
+        matrix = list->objects[k];
+        if (matrix->mask == mask) {
+            *pos = k;
+            return matrix;
+        }
+        if (matrix->mask < mask) {
+            i = k + 1;
+        } else {
+            j = k;
+        }
+    }
+    *pos = i;
+
+    return NULL;
+}
+
+static void
+ec_method_matrix_remove(ec_matrix_list_t *list, uintptr_t mask)
+{
+    uint32_t pos;
+
+    if (ec_method_matrix_lookup(list, mask, &pos) != NULL) {
+        list->count--;
+        if (pos < list->count) {
+            memmove(list->objects + pos, list->objects + pos + 1,
+                    sizeof(ec_matrix_t *) * (list->count - pos));
+        }
+    }
+}
+
+static void
+ec_method_matrix_insert(ec_matrix_list_t *list, ec_matrix_t *matrix)
+{
+    uint32_t pos;
+
+    GF_ASSERT(ec_method_matrix_lookup(list, matrix->mask, &pos) == NULL);
+
+    if (pos < list->count) {
+        memmove(list->objects + pos + 1, list->objects + pos,
+                sizeof(ec_matrix_t *) * (list->count - pos));
+    }
+    list->objects[pos] = matrix;
+    list->count++;
+}
+
+static ec_matrix_t *
+ec_method_matrix_get(ec_matrix_list_t *list, uintptr_t mask, uint32_t *rows)
+{
+    ec_matrix_t *matrix;
+    uint32_t pos;
+
+    LOCK(&list->lock);
+
+    matrix = ec_method_matrix_lookup(list, mask, &pos);
+    if (matrix != NULL) {
+        list_del_init(&matrix->lru);
+        matrix->refs++;
+
+        goto out;
+    }
+
+    if ((list->count >= list->max) && !list_empty(&list->lru)) {
+        matrix = list_first_entry(&list->lru, ec_matrix_t, lru);
+        list_del_init(&matrix->lru);
+
+        ec_method_matrix_remove(list, matrix->mask);
+
+        ec_method_matrix_release(matrix);
+    } else {
+        matrix = mem_get0(list->pool);
+        if (matrix == NULL) {
+            matrix = EC_ERR(ENOMEM);
+            goto out;
+        }
+        matrix->values = (uint32_t *)((uintptr_t)matrix + sizeof(ec_matrix_t) +
+                                      sizeof(ec_matrix_row_t) * list->columns);
+    }
+
+    ec_method_matrix_init(list, matrix, mask, rows, _gf_true);
+
+    if (list->count < list->max) {
+        ec_method_matrix_insert(list, matrix);
+    } else {
+        matrix->mask = 0;
+    }
+
+out:
+    UNLOCK(&list->lock);
+
+    return matrix;
+}
+
+static void
+ec_method_matrix_put(ec_matrix_list_t *list, ec_matrix_t *matrix)
+{
+    LOCK(&list->lock);
+
+    ec_method_matrix_unref(list, matrix);
+
+    UNLOCK(&list->lock);
+}
+
+static int32_t
+ec_method_setup(xlator_t *xl, ec_matrix_list_t *list, const char *gen)
+{
+    ec_matrix_t *matrix;
+    uint32_t values[list->rows];
+    uint32_t i;
+    int32_t err;
+
+    matrix = GF_MALLOC(sizeof(ec_matrix_t) +
+                           sizeof(ec_matrix_row_t) * list->rows +
+                           sizeof(uint32_t) * list->columns * list->rows,
+                       ec_mt_ec_matrix_t);
+    if (matrix == NULL) {
+        err = -ENOMEM;
+        goto failed;
+    }
+    memset(matrix, 0, sizeof(ec_matrix_t));
+    matrix->values = (uint32_t *)((uintptr_t)matrix + sizeof(ec_matrix_t) +
+                                  sizeof(ec_matrix_row_t) * list->rows);
+
+    list->code = ec_code_create(list->gf, ec_code_detect(xl, gen));
+    if (EC_IS_ERR(list->code)) {
+        err = EC_GET_ERR(list->code);
+        list->code = NULL;
+        goto failed_matrix;
+    }
+
+    for (i = 0; i < list->rows; i++) {
+        values[i] = i + 1;
+    }
+    ec_method_matrix_init(list, matrix, 0, values, _gf_false);
+
+    list->encode = matrix;
+
+    return 0;
+
+failed_matrix:
+    GF_FREE(matrix);
+failed:
+    return err;
+}
+
+int32_t
+ec_method_init(xlator_t *xl, ec_matrix_list_t *list, uint32_t columns,
+               uint32_t rows, uint32_t max, const char *gen)
+{
+    list->columns = columns;
+    list->rows = rows;
+    list->max = max;
+    list->stripe = EC_METHOD_CHUNK_SIZE * list->columns;
+    INIT_LIST_HEAD(&list->lru);
+    int32_t err;
+
+    list->pool = mem_pool_new_fn(xl->ctx,
+                                 sizeof(ec_matrix_t) +
+                                     sizeof(ec_matrix_row_t) * columns +
+                                     sizeof(uint32_t) * columns * columns,
+                                 128, "ec_matrix_t");
+    if (list->pool == NULL) {
+        err = -ENOMEM;
+        goto failed;
+    }
+
+    list->objects = GF_MALLOC(sizeof(ec_matrix_t *) * max, ec_mt_ec_matrix_t);
+    if (list->objects == NULL) {
+        err = -ENOMEM;
+        goto failed_pool;
+    }
+
+    list->gf = ec_gf_prepare(EC_GF_BITS, EC_GF_MOD);
+    if (EC_IS_ERR(list->gf)) {
+        err = EC_GET_ERR(list->gf);
+        goto failed_objects;
+    }
+
+    err = ec_method_setup(xl, list, gen);
+    if (err != 0) {
+        goto failed_gf;
+    }
+
+    LOCK_INIT(&list->lock);
+
+    return 0;
+
+failed_gf:
+    ec_gf_destroy(list->gf);
+failed_objects:
+    GF_FREE(list->objects);
+failed_pool:
+    mem_pool_destroy(list->pool);
+failed:
+    list->pool = NULL;
+    list->objects = NULL;
+    list->gf = NULL;
+
+    return err;
+}
+
+void
+ec_method_fini(ec_matrix_list_t *list)
+{
+    ec_matrix_t *matrix;
+
+    if (list->encode == NULL) {
+        return;
+    }
+
+    while (!list_empty(&list->lru)) {
+        matrix = list_first_entry(&list->lru, ec_matrix_t, lru);
+        ec_method_matrix_destroy(list, matrix);
+    }
+
+    GF_ASSERT(list->count == 0);
+
+    if (list->pool) /*Init was successful*/
+        LOCK_DESTROY(&list->lock);
+
+    ec_method_matrix_release(list->encode);
+    GF_FREE(list->encode);
+
+    ec_code_destroy(list->code);
+    ec_gf_destroy(list->gf);
+    GF_FREE(list->objects);
+
+    if (list->pool)
+        mem_pool_destroy(list->pool);
+}
+
+int32_t
+ec_method_update(xlator_t *xl, ec_matrix_list_t *list, const char *gen)
+{
+    /* TODO: Allow changing code generator */
+
+    return 0;
+}
+
+void
+ec_method_encode(ec_matrix_list_t *list, uint64_t size, void *in, void **out)
+{
+    ec_matrix_t *matrix;
+    uint64_t pos;
+    uint32_t i;
+
+    matrix = list->encode;
+    for (pos = 0; pos < size; pos += list->stripe) {
+        for (i = 0; i < matrix->rows; i++) {
+            matrix->row_data[i].func.linear(
+                out[i], in, pos, matrix->row_data[i].values, list->columns);
+            out[i] += EC_METHOD_CHUNK_SIZE;
+        }
+    }
+}
+
+int32_t
+ec_method_decode(ec_matrix_list_t *list, uint64_t size, uintptr_t mask,
+                 uint32_t *rows, void **in, void *out)
+{
+    ec_matrix_t *matrix;
+    uint64_t pos;
+    uint32_t i;
+
+    matrix = ec_method_matrix_get(list, mask, rows);
+    if (EC_IS_ERR(matrix)) {
+        return EC_GET_ERR(matrix);
+    }
+    for (pos = 0; pos < size; pos += EC_METHOD_CHUNK_SIZE) {
+        for (i = 0; i < matrix->rows; i++) {
+            matrix->row_data[i].func.interleaved(
+                out, in, pos, matrix->row_data[i].values, list->columns);
+            out += EC_METHOD_CHUNK_SIZE;
+        }
+    }
+
+    ec_method_matrix_put(list, matrix);
+
+    return 0;
+}
diff --git a/xlators/cluster/ec/src/ec-method.h b/xlators/cluster/ec/src/ec-method.h
new file mode 100644
index 00000000000..f91233b2f88
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-method.h
@@ -0,0 +1,48 @@
+/*
+  Copyright (c) 2012-2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_METHOD_H__
+#define __EC_METHOD_H__
+
+#include "ec-types.h"
+#include "ec-galois.h"
+
+#define EC_GF_BITS 8
+#define EC_GF_MOD 0x11D
+
+#define EC_GF_SIZE (1 << EC_GF_BITS)
+
+/* Determines the maximum size of the matrix used to encode/decode data */
+#define EC_METHOD_MAX_FRAGMENTS 16
+/* Determines the maximum number of usable elements in the Galois Field */
+#define EC_METHOD_MAX_NODES (EC_GF_SIZE - 1)
+
+#define EC_METHOD_WORD_SIZE 64
+
+#define EC_METHOD_CHUNK_SIZE (EC_METHOD_WORD_SIZE * EC_GF_BITS)
+
+int32_t
+ec_method_init(xlator_t *xl, ec_matrix_list_t *list, uint32_t columns,
+               uint32_t rows, uint32_t max, const char *gen);
+
+void
+ec_method_fini(ec_matrix_list_t *list);
+
+int32_t
+ec_method_update(xlator_t *xl, ec_matrix_list_t *list, const char *gen);
+
+void
+ec_method_encode(ec_matrix_list_t *list, uint64_t size, void *in, void **out);
+
+int32_t
+ec_method_decode(ec_matrix_list_t *list, uint64_t size, uintptr_t mask,
+                 uint32_t *rows, void **in, void *out);
+
+#endif /* __EC_METHOD_H__ */
diff --git a/xlators/cluster/ec/src/ec-types.h b/xlators/cluster/ec/src/ec-types.h
new file mode 100644
index 00000000000..de9b89bb2c9
--- /dev/null
+++ b/xlators/cluster/ec/src/ec-types.h
@@ -0,0 +1,690 @@
+/*
+  Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_TYPES_H__
+#define __EC_TYPES_H__
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/timer.h>
+#include "libxlator.h"
+#include <glusterfs/atomic.h>
+
+#define EC_GF_MAX_REGS 16
+
+enum _ec_heal_need;
+typedef enum _ec_heal_need ec_heal_need_t;
+
+enum _ec_stripe_part;
+typedef enum _ec_stripe_part ec_stripe_part_t;
+
+enum _ec_read_policy;
+typedef enum _ec_read_policy ec_read_policy_t;
+
+struct _ec_config;
+typedef struct _ec_config ec_config_t;
+
+struct _ec_fd;
+typedef struct _ec_fd ec_fd_t;
+
+struct _ec_fragment_range;
+typedef struct _ec_fragment_range ec_fragment_range_t;
+
+struct _ec_inode;
+typedef struct _ec_inode ec_inode_t;
+
+union _ec_cbk;
+typedef union _ec_cbk ec_cbk_t;
+
+struct _ec_lock;
+typedef struct _ec_lock ec_lock_t;
+
+struct _ec_lock_link;
+typedef struct _ec_lock_link ec_lock_link_t;
+
+struct _ec_fop_data;
+typedef struct _ec_fop_data ec_fop_data_t;
+
+struct _ec_cbk_data;
+typedef struct _ec_cbk_data ec_cbk_data_t;
+
+enum _ec_gf_opcode;
+typedef enum _ec_gf_opcode ec_gf_opcode_t;
+
+struct _ec_gf_op;
+typedef struct _ec_gf_op ec_gf_op_t;
+
+struct _ec_gf_mul;
+typedef struct _ec_gf_mul ec_gf_mul_t;
+
+struct _ec_gf;
+typedef struct _ec_gf ec_gf_t;
+
+struct _ec_code_gen;
+typedef struct _ec_code_gen ec_code_gen_t;
+
+struct _ec_code;
+typedef struct _ec_code ec_code_t;
+
+struct _ec_code_arg;
+typedef struct _ec_code_arg ec_code_arg_t;
+
+struct _ec_code_op;
+typedef struct _ec_code_op ec_code_op_t;
+
+struct _ec_code_builder;
+typedef struct _ec_code_builder ec_code_builder_t;
+
+struct _ec_code_chunk;
+typedef struct _ec_code_chunk ec_code_chunk_t;
+
+struct _ec_stripe;
+typedef struct _ec_stripe ec_stripe_t;
+
+struct _ec_stripe_list;
+typedef struct _ec_stripe_list ec_stripe_list_t;
+
+struct _ec_code_space;
+typedef struct _ec_code_space ec_code_space_t;
+
+typedef void (*ec_code_func_linear_t)(void *dst, void *src, uint64_t offset,
+                                      uint32_t *values, uint32_t count);
+
+typedef void (*ec_code_func_interleaved_t)(void *dst, void **src,
+                                           uint64_t offset, uint32_t *values,
+                                           uint32_t count);
+
+union _ec_code_func;
+typedef union _ec_code_func ec_code_func_t;
+
+struct _ec_matrix_row;
+typedef struct _ec_matrix_row ec_matrix_row_t;
+
+struct _ec_matrix;
+typedef struct _ec_matrix ec_matrix_t;
+
+struct _ec_matrix_list;
+typedef struct _ec_matrix_list ec_matrix_list_t;
+
+struct _ec_heal;
+typedef struct _ec_heal ec_heal_t;
+
+struct _ec_self_heald;
+typedef struct _ec_self_heald ec_self_heald_t;
+
+struct _ec_statistics;
+typedef struct _ec_statistics ec_statistics_t;
+
+struct _ec;
+typedef struct _ec ec_t;
+
+typedef void (*ec_wind_f)(ec_t *, ec_fop_data_t *, int32_t);
+typedef int32_t (*ec_handler_f)(ec_fop_data_t *, int32_t);
+typedef void (*ec_resume_f)(ec_fop_data_t *, int32_t);
+
+enum _ec_read_policy { EC_ROUND_ROBIN, EC_GFID_HASH, EC_READ_POLICY_MAX };
+
+enum _ec_heal_need {
+    EC_HEAL_NONEED,
+    EC_HEAL_MAYBE,
+    EC_HEAL_MUST,
+    EC_HEAL_PURGE_INDEX
+};
+
+enum _ec_stripe_part { EC_STRIPE_HEAD, EC_STRIPE_TAIL };
+
+/* Enumartions to indicate FD status. */
+typedef enum { EC_FD_NOT_OPENED, EC_FD_OPENED, EC_FD_OPENING } ec_fd_status_t;
+
+struct _ec_config {
+    uint32_t version;
+    uint8_t algorithm;
+    uint8_t gf_word_size;
+    uint8_t bricks;
+    uint8_t redundancy;
+    uint32_t chunk_size;
+};
+
+struct _ec_fd {
+    loc_t loc;
+    uintptr_t open;
+    int32_t flags;
+    uint64_t bad_version;
+    ec_fd_status_t fd_status[0];
+};
+
+struct _ec_stripe {
+    struct list_head lru; /* LRU list member */
+    uint64_t frag_offset; /* Fragment offset of this stripe */
+    char data[];          /* Contents of the stripe */
+};
+
+struct _ec_stripe_list {
+    struct list_head lru;
+    uint32_t count;
+    uint32_t max;
+};
+
+struct _ec_inode {
+    ec_lock_t *inode_lock;
+    gf_boolean_t have_info;
+    gf_boolean_t have_config;
+    gf_boolean_t have_version;
+    gf_boolean_t have_size;
+    int32_t heal_count;
+    ec_config_t config;
+    uint64_t pre_version[2];
+    uint64_t post_version[2];
+    uint64_t pre_size;
+    uint64_t post_size;
+    uint64_t dirty[2];
+    struct list_head heal;
+    ec_stripe_list_t stripe_cache;
+    uint64_t bad_version;
+};
+
+typedef int32_t (*fop_heal_cbk_t)(call_frame_t *, void *, xlator_t *, int32_t,
+                                  int32_t, uintptr_t, uintptr_t, uintptr_t,
+                                  uint32_t, dict_t *);
+typedef int32_t (*fop_fheal_cbk_t)(call_frame_t *, void *, xlator_t *, int32_t,
+                                   int32_t, uintptr_t, uintptr_t, uintptr_t,
+                                   uint32_t, dict_t *);
+
+union _ec_cbk {
+    fop_access_cbk_t access;
+    fop_create_cbk_t create;
+    fop_discard_cbk_t discard;
+    fop_entrylk_cbk_t entrylk;
+    fop_fentrylk_cbk_t fentrylk;
+    fop_fallocate_cbk_t fallocate;
+    fop_flush_cbk_t flush;
+    fop_fsync_cbk_t fsync;
+    fop_fsyncdir_cbk_t fsyncdir;
+    fop_getxattr_cbk_t getxattr;
+    fop_fgetxattr_cbk_t fgetxattr;
+    fop_heal_cbk_t heal;
+    fop_fheal_cbk_t fheal;
+    fop_inodelk_cbk_t inodelk;
+    fop_finodelk_cbk_t finodelk;
+    fop_link_cbk_t link;
+    fop_lk_cbk_t lk;
+    fop_lookup_cbk_t lookup;
+    fop_mkdir_cbk_t mkdir;
+    fop_mknod_cbk_t mknod;
+    fop_open_cbk_t open;
+    fop_opendir_cbk_t opendir;
+    fop_readdir_cbk_t readdir;
+    fop_readdirp_cbk_t readdirp;
+    fop_readlink_cbk_t readlink;
+    fop_readv_cbk_t readv;
+    fop_removexattr_cbk_t removexattr;
+    fop_fremovexattr_cbk_t fremovexattr;
+    fop_rename_cbk_t rename;
+    fop_rmdir_cbk_t rmdir;
+    fop_setattr_cbk_t setattr;
+    fop_fsetattr_cbk_t fsetattr;
+    fop_setxattr_cbk_t setxattr;
+    fop_fsetxattr_cbk_t fsetxattr;
+    fop_stat_cbk_t stat;
+    fop_fstat_cbk_t fstat;
+    fop_statfs_cbk_t statfs;
+    fop_symlink_cbk_t symlink;
+    fop_truncate_cbk_t truncate;
+    fop_ftruncate_cbk_t ftruncate;
+    fop_unlink_cbk_t unlink;
+    fop_writev_cbk_t writev;
+    fop_xattrop_cbk_t xattrop;
+    fop_fxattrop_cbk_t fxattrop;
+    fop_zerofill_cbk_t zerofill;
+    fop_seek_cbk_t seek;
+    fop_ipc_cbk_t ipc;
+};
+
+struct _ec_lock {
+    ec_inode_t *ctx;
+    gf_timer_t *timer;
+
+    /* List of owners of this lock. All fops added to this list are running
+     * concurrently. */
+    struct list_head owners;
+
+    /* List of fops waiting to be an owner of the lock. Fops are added to this
+     * list when the current owner has an incompatible access (conflicting lock)
+     * or the lock is not acquired yet. */
+    struct list_head waiting;
+
+    /* List of fops that will wait until the next unlock/lock cycle. This
+     * happens when the currently acquired lock is decided to be released as
+     * soon as possible. In this case, all frozen fops will be continued only
+     * after the lock is reacquired. */
+    struct list_head frozen;
+
+    uintptr_t mask;
+    uintptr_t good_mask;
+    uintptr_t healing;
+    uint32_t refs_owners;   /* Refs for fops owning the lock */
+    uint32_t refs_pending;  /* Refs assigned to fops being prepared */
+    uint32_t waiting_flags; /*Track xattrop/dirty marking*/
+    gf_boolean_t acquired;
+    gf_boolean_t contention;
+    gf_boolean_t unlock_now;
+    gf_boolean_t release;
+    gf_boolean_t query;
+    fd_t *fd;
+    loc_t loc;
+    union {
+        entrylk_type type;
+        struct gf_flock flock;
+    };
+};
+
+struct _ec_lock_link {
+    ec_lock_t *lock;
+    ec_fop_data_t *fop;
+    struct list_head owner_list;
+    struct list_head wait_list;
+    gf_boolean_t update[2];
+    gf_boolean_t dirty[2];
+    gf_boolean_t optimistic_changelog;
+    loc_t *base;
+    uint64_t size;
+    uint32_t waiting_flags;
+    off_t fl_start;
+    off_t fl_end;
+};
+
+/* This structure keeps a range of fragment offsets affected by a fop. Since
+ * real file offsets can be difficult to handle correctly because of overflows,
+ * we use the 'scaled' offset, which corresponds to the offset of the fragment
+ * seen by the bricks, which is always smaller and cannot overflow. */
+struct _ec_fragment_range {
+    uint64_t first; /* Address of the first affected fragment as seen by the
+                       bricks (offset on brick) */
+    uint64_t last;  /* Address of the first non affected fragment as seen by
+                       the bricks (offset on brick) */
+};
+
+/* EC xlator data structure to collect all the data required to perform
+ * the file operation.*/
+struct _ec_fop_data {
+    int32_t id; /* ID of the file operation */
+    int32_t refs;
+    int32_t state;
+    uint32_t minimum; /* Minimum number of successful
+                         operation required to conclude a
+                         fop as successful */
+    int32_t expected;
+    int32_t winds;
+    int32_t jobs;
+    int32_t error;
+    ec_fop_data_t *parent;
+    xlator_t *xl;                  /* points to EC xlator */
+    call_frame_t *req_frame;       /* frame of the calling xlator */
+    call_frame_t *frame;           /* frame used by this fop */
+    struct list_head cbk_list;     /* sorted list of groups of answers */
+    struct list_head answer_list;  /* list of answers */
+    struct list_head pending_list; /* member of ec_t.pending_fops */
+    ec_cbk_data_t *answer;         /* accepted answer */
+    int32_t lock_count;
+    int32_t locked;
+    gf_lock_t lock;
+    ec_lock_link_t locks[2];
+    int32_t first_lock;
+
+    uint32_t fop_flags; /* Flags passed by the caller. */
+    uint32_t flags;     /* Internal flags. */
+    uint32_t first;
+    uintptr_t mask;
+    uintptr_t healing; /*Dispatch is done but call is successful only
+                         if fop->minimum number of subvolumes succeed
+                         which are not healing*/
+    uintptr_t remaining;
+    uintptr_t received; /* Mask of responses */
+    uintptr_t good;
+
+    uid_t uid;
+    gid_t gid;
+
+    ec_wind_f wind;       /* Function to wind to */
+    ec_handler_f handler; /* FOP manager function */
+    ec_resume_f resume;
+    ec_cbk_t cbks; /* Callback function for this FOP */
+    void *data;
+    ec_heal_t *heal;
+    struct list_head healer;
+
+    uint64_t user_size;
+    uint32_t head;
+
+    int32_t use_fd; /* Indicates whether this FOP uses FD or
+                       not */
+
+    dict_t *xdata;
+    dict_t *dict;
+    int32_t int32;
+    uint32_t uint32;
+    uint64_t size;
+    off_t offset;
+    mode_t mode[2];
+    entrylk_cmd entrylk_cmd;
+    entrylk_type entrylk_type;
+    gf_xattrop_flags_t xattrop_flags;
+    dev_t dev;
+    inode_t *inode;
+    fd_t *fd; /* FD of the file on which FOP is
+                 being carried upon */
+    struct iatt iatt;
+    char *str[2];
+    loc_t loc[2]; /* Holds the location details for
+                     the file */
+    struct gf_flock flock;
+    struct iovec *vector;
+    struct iobref *buffers;
+    gf_seek_what_t seek;
+    ec_fragment_range_t frag_range; /* This will hold the range of stripes
+                                        affected by the fop. */
+    char *errstr;                   /*String of fop name, path and gfid
+                                     to be used in gf_msg. */
+};
+
+struct _ec_cbk_data {
+    struct list_head list;        /* item in the sorted list of groups */
+    struct list_head answer_list; /* item in the list of answers */
+    ec_fop_data_t *fop;
+    ec_cbk_data_t *next; /* next answer in the same group */
+    uint32_t idx;
+    int32_t op_ret;
+    int32_t op_errno;
+    int32_t count;
+    uintptr_t mask;
+
+    dict_t *xdata;
+    dict_t *dict;
+    int32_t int32;
+    uintptr_t uintptr[3];
+    uint64_t size;
+    uint64_t version[2];
+    inode_t *inode;
+    fd_t *fd;
+    struct statvfs statvfs;
+    struct iatt iatt[5];
+    struct gf_flock flock;
+    struct iovec *vector;
+    struct iobref *buffers;
+    char *str;
+    gf_dirent_t entries;
+    off_t offset;
+    gf_seek_what_t what;
+};
+
+enum _ec_gf_opcode {
+    EC_GF_OP_LOAD,
+    EC_GF_OP_STORE,
+    EC_GF_OP_COPY,
+    EC_GF_OP_XOR2,
+    EC_GF_OP_XOR3,
+    EC_GF_OP_XORM,
+    EC_GF_OP_END
+};
+
+struct _ec_gf_op {
+    ec_gf_opcode_t op;
+    uint32_t arg1;
+    uint32_t arg2;
+    uint32_t arg3;
+};
+
+struct _ec_gf_mul {
+    uint32_t regs;
+    uint32_t map[EC_GF_MAX_REGS];
+    ec_gf_op_t *ops;
+};
+
+struct _ec_gf {
+    uint32_t bits;
+    uint32_t size;
+    uint32_t mod;
+    uint32_t min_ops;
+    uint32_t max_ops;
+    uint32_t avg_ops;
+    uint32_t *log;
+    uint32_t *pow;
+    ec_gf_mul_t **table;
+};
+
+struct _ec_code_gen {
+    char *name;
+    char **flags;
+    uint32_t width;
+
+    void (*prolog)(ec_code_builder_t *builder);
+    void (*epilog)(ec_code_builder_t *builder);
+    void (*load)(ec_code_builder_t *builder, uint32_t reg, uint32_t offset,
+                 uint32_t bit);
+    void (*store)(ec_code_builder_t *builder, uint32_t reg, uint32_t bit);
+    void (*copy)(ec_code_builder_t *builder, uint32_t dst, uint32_t src);
+    void (*xor2)(ec_code_builder_t *builder, uint32_t dst, uint32_t src);
+    void (*xor3)(ec_code_builder_t *builder, uint32_t dst, uint32_t src1,
+                 uint32_t src2);
+    void (*xorm)(ec_code_builder_t *builder, uint32_t dst, uint32_t offset,
+                 uint32_t bit);
+};
+
+struct _ec_code {
+    gf_lock_t lock;
+    struct list_head spaces;
+    ec_gf_t *gf;
+    ec_code_gen_t *gen;
+};
+
+struct _ec_code_arg {
+    uint32_t value;
+};
+
+struct _ec_code_op {
+    ec_gf_opcode_t op;
+    ec_code_arg_t arg1;
+    ec_code_arg_t arg2;
+    ec_code_arg_t arg3;
+};
+
+struct _ec_code_builder {
+    ec_code_t *code;
+    uint64_t address;
+    uint8_t *data;
+    uint32_t size;
+    int32_t error;
+    uint32_t regs;
+    uint32_t bits;
+    uint32_t width;
+    uint32_t count;
+    uint32_t base;
+    uint32_t map[EC_GF_MAX_REGS];
+    gf_boolean_t linear;
+    uint64_t loop;
+    ec_code_op_t ops[0];
+};
+
+struct _ec_code_chunk {
+    struct list_head list;
+    size_t size;
+    ec_code_space_t *space;
+};
+
+struct _ec_code_space {
+    struct list_head list;
+    struct list_head chunks;
+    ec_code_t *code;
+    void *exec;
+    size_t size;
+};
+
+union _ec_code_func {
+    ec_code_func_linear_t linear;
+    ec_code_func_interleaved_t interleaved;
+};
+
+struct _ec_matrix_row {
+    ec_code_func_t func;
+    uint32_t *values;
+};
+
+struct _ec_matrix {
+    struct list_head lru;
+    uint32_t refs;
+    uint32_t columns;
+    uint32_t rows;
+    uintptr_t mask;
+    ec_code_t *code;
+    uint32_t *values;
+    ec_matrix_row_t row_data[0];
+};
+
+struct _ec_matrix_list {
+    struct list_head lru;
+    gf_lock_t lock;
+    uint32_t columns;
+    uint32_t rows;
+    uint32_t max;
+    uint32_t count;
+    uint32_t stripe;
+    struct mem_pool *pool;
+    ec_gf_t *gf;
+    ec_code_t *code;
+    ec_matrix_t *encode;
+    ec_matrix_t **objects;
+};
+
+struct _ec_heal {
+    struct list_head list;
+    gf_lock_t lock;
+    xlator_t *xl;
+    ec_fop_data_t *fop;
+    void *data;
+    ec_fop_data_t *lookup;
+    loc_t loc;
+    struct iatt iatt;
+    char *symlink;
+    fd_t *fd;
+    int32_t partial;
+    int32_t done;
+    int32_t error;
+    gf_boolean_t nameheal;
+    uintptr_t available;
+    uintptr_t good;
+    uintptr_t bad;
+    uintptr_t open;
+    uintptr_t fixed;
+    uint64_t offset;
+    uint64_t size;
+    uint64_t total_size;
+    uint64_t version[2];
+    uint64_t raw_size;
+};
+
+struct subvol_healer {
+    xlator_t *this;
+    int subvol;
+    gf_boolean_t running;
+    gf_boolean_t rerun;
+    pthread_mutex_t mutex;
+    pthread_cond_t cond;
+    pthread_t thread;
+};
+
+struct _ec_self_heald {
+    gf_boolean_t iamshd;
+    gf_boolean_t enabled;
+    int timeout;
+    uint32_t max_threads;
+    uint32_t wait_qlength;
+    struct subvol_healer *index_healers;
+    struct subvol_healer *full_healers;
+};
+
+struct _ec_statistics {
+    struct {
+        gf_atomic_t hits;    /* Cache hits. */
+        gf_atomic_t misses;  /* Cache misses. */
+        gf_atomic_t updates; /* Number of times an existing stripe has
+                                been updated with new content. */
+        gf_atomic_t invals;  /* Number of times an existing stripe has
+                                been invalidated because of truncates
+                                or discards. */
+        gf_atomic_t evicts;  /* Number of times that an existing entry
+                                has been evicted to make room for newer
+                                entries. */
+        gf_atomic_t allocs;  /* Number of memory allocations made to
+                                store stripes. */
+        gf_atomic_t errors;  /* Number of errors that have caused extra
+                                requests. (Basically memory allocation
+                                errors). */
+    } stripe_cache;
+    struct {
+        gf_atomic_t attempted; /*Number of heals attempted on
+                                files/directories*/
+        gf_atomic_t completed; /*Number of heals complted on files/directories*/
+    } shd;
+};
+
+struct _ec {
+    xlator_t *xl;
+    int32_t healers;
+    int32_t heal_waiters;
+    int32_t nodes; /* Total number of bricks(n) */
+    int32_t bits_for_nodes;
+    int32_t fragments;      /* Data bricks(k) */
+    int32_t redundancy;     /* Redundant bricks(m) */
+    uint32_t fragment_size; /* Size of fragment/chunk on a
+                               brick. */
+    uint32_t stripe_size;   /* (fragment_size * fragments)
+                               maximum size of user data
+                               stored in one stripe. */
+    int32_t up;             /* Represents whether EC volume is
+                               up or not. */
+    uint32_t idx;
+    uint32_t xl_up_count;     /* Number of UP bricks. */
+    uintptr_t xl_up;          /* Bit flag representing UP
+                                 bricks */
+    uint32_t xl_notify_count; /* Number of notifications. */
+    uintptr_t xl_notify;      /* Bit flag representing
+                                 notification for bricks. */
+    uintptr_t node_mask;
+    uintptr_t read_mask;         /*Stores user defined read-mask*/
+    gf_atomic_t async_fop_count; /* Number of on going asynchronous fops. */
+    xlator_t **xl_list;
+    gf_lock_t lock;
+    gf_timer_t *timer;
+    gf_boolean_t shutdown;
+    gf_boolean_t eager_lock;
+    gf_boolean_t other_eager_lock;
+    gf_boolean_t optimistic_changelog;
+    gf_boolean_t parallel_writes;
+    uint32_t stripe_cache;
+    uint32_t quorum_count;
+    uint32_t background_heals;
+    uint32_t heal_wait_qlen;
+    uint32_t self_heal_window_size; /* max size of read/writes */
+    uint32_t eager_lock_timeout;
+    uint32_t other_eager_lock_timeout;
+    struct list_head pending_fops;
+    struct list_head heal_waiting;
+    struct list_head healing;
+    struct mem_pool *fop_pool;
+    struct mem_pool *cbk_pool;
+    struct mem_pool *lock_pool;
+    ec_self_heald_t shd;
+    char vol_uuid[UUID_SIZE + 1];
+    dict_t *leaf_to_subvolid;
+    ec_read_policy_t read_policy;
+    ec_matrix_list_t matrix;
+    ec_statistics_t stats;
+};
+
+#endif /* __EC_TYPES_H__ */
diff --git a/xlators/cluster/ec/src/ec.c b/xlators/cluster/ec/src/ec.c
new file mode 100644
index 00000000000..7344be4968d
--- /dev/null
+++ b/xlators/cluster/ec/src/ec.c
@@ -0,0 +1,1873 @@
+/*
+  Copyright (c) 2012-2015 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/defaults.h>
+#include <glusterfs/statedump.h>
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/upcall-utils.h>
+
+#include "ec.h"
+#include "ec-messages.h"
+#include "ec-mem-types.h"
+#include "ec-types.h"
+#include "ec-helpers.h"
+#include "ec-common.h"
+#include "ec-fops.h"
+#include "ec-method.h"
+#include "ec-code.h"
+#include "ec-heald.h"
+#include <glusterfs/events.h>
+
+static char *ec_read_policies[EC_READ_POLICY_MAX + 1] = {
+    [EC_ROUND_ROBIN] = "round-robin",
+    [EC_GFID_HASH] = "gfid-hash",
+    [EC_READ_POLICY_MAX] = NULL};
+
+#define EC_INTERNAL_XATTR_OR_GOTO(name, xattr, op_errno, label)                \
+    do {                                                                       \
+        if (ec_is_internal_xattr(NULL, (char *)name, NULL, NULL)) {            \
+            op_errno = EPERM;                                                  \
+            goto label;                                                        \
+        }                                                                      \
+        if (name && (strlen(name) == 0) && xattr) {                            \
+            /* Bulk [f]removexattr/[f]setxattr */                              \
+            GF_IF_INTERNAL_XATTR_GOTO(EC_XATTR_PREFIX "*", xattr, op_errno,    \
+                                      label);                                  \
+        }                                                                      \
+    } while (0)
+
+int32_t
+ec_parse_options(xlator_t *this)
+{
+    ec_t *ec = this->private;
+    int32_t error = EINVAL;
+    uintptr_t mask;
+
+    GF_OPTION_INIT("redundancy", ec->redundancy, int32, out);
+    ec->fragments = ec->nodes - ec->redundancy;
+    if ((ec->redundancy < 1) || (ec->redundancy >= ec->fragments) ||
+        (ec->fragments > EC_MAX_FRAGMENTS)) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, EC_MSG_INVALID_REDUNDANCY,
+               "Invalid redundancy (must be between "
+               "1 and %d)",
+               (ec->nodes - 1) / 2);
+
+        goto out;
+    }
+
+    ec->bits_for_nodes = 1;
+    mask = 2;
+    while (ec->nodes > mask) {
+        ec->bits_for_nodes++;
+        mask <<= 1;
+    }
+    ec->node_mask = (1ULL << ec->nodes) - 1ULL;
+    ec->fragment_size = EC_METHOD_CHUNK_SIZE;
+    ec->stripe_size = ec->fragment_size * ec->fragments;
+
+    gf_msg_debug("ec", 0,
+                 "Initialized with: nodes=%u, fragments=%u, "
+                 "stripe_size=%u, node_mask=%" PRIxFAST32,
+                 ec->nodes, ec->fragments, ec->stripe_size, ec->node_mask);
+
+    error = 0;
+
+out:
+    return error;
+}
+
+int32_t
+ec_prepare_childs(xlator_t *this)
+{
+    ec_t *ec = this->private;
+    xlator_list_t *child = NULL;
+    int32_t count = 0;
+
+    for (child = this->children; child != NULL; child = child->next) {
+        count++;
+    }
+    if (count > EC_MAX_NODES) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, EC_MSG_TOO_MANY_SUBVOLS,
+               "Too many subvolumes");
+
+        return EINVAL;
+    }
+    ec->nodes = count;
+
+    ec->xl_list = GF_CALLOC(count, sizeof(ec->xl_list[0]), ec_mt_xlator_t);
+    if (ec->xl_list == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+               "Allocation of xlator list failed");
+
+        return ENOMEM;
+    }
+    ec->xl_up = 0;
+    ec->xl_up_count = 0;
+
+    count = 0;
+    for (child = this->children; child != NULL; child = child->next) {
+        ec->xl_list[count++] = child->xlator;
+    }
+
+    return 0;
+}
+
+/* This function transforms the subvol to subvol-id*/
+static int
+_subvol_to_subvolid(dict_t *this, char *key, data_t *value, void *data)
+{
+    ec_t *ec = data;
+    xlator_t *subvol = NULL;
+    int i = 0;
+    int ret = -1;
+
+    subvol = data_to_ptr(value);
+    for (i = 0; i < ec->nodes; i++) {
+        if (ec->xl_list[i] == subvol) {
+            ret = dict_set_int32(this, key, i);
+            /* -1 stops dict_foreach and returns -1*/
+            if (ret < 0)
+                ret = -1;
+            goto out;
+        }
+    }
+out:
+    return ret;
+}
+
+int
+ec_subvol_to_subvol_id_transform(ec_t *ec, dict_t *leaf_to_subvolid)
+{
+    return dict_foreach(leaf_to_subvolid, _subvol_to_subvolid, ec);
+}
+
+void
+__ec_destroy_private(xlator_t *this)
+{
+    ec_t *ec = this->private;
+
+    if (ec != NULL) {
+        LOCK(&ec->lock);
+
+        if (ec->timer != NULL) {
+            gf_timer_call_cancel(this->ctx, ec->timer);
+            ec->timer = NULL;
+        }
+
+        UNLOCK(&ec->lock);
+
+        /* There is a race with timer because there is no way to know if
+         * timer callback has really been cancelled or it has been scheduled
+         * for execution. If it has been scheduled, it will crash if we
+         * destroy ec too fast.
+         *
+         * Not sure how this can be solved without using global variables or
+         * having support from gf_timer_call_cancel()
+         */
+        sleep(2);
+
+        this->private = NULL;
+        if (ec->xl_list != NULL) {
+            GF_FREE(ec->xl_list);
+            ec->xl_list = NULL;
+        }
+
+        if (ec->fop_pool != NULL) {
+            mem_pool_destroy(ec->fop_pool);
+        }
+
+        if (ec->cbk_pool != NULL) {
+            mem_pool_destroy(ec->cbk_pool);
+        }
+
+        if (ec->lock_pool != NULL) {
+            mem_pool_destroy(ec->lock_pool);
+        }
+
+        LOCK_DESTROY(&ec->lock);
+
+        if (ec->leaf_to_subvolid)
+            dict_unref(ec->leaf_to_subvolid);
+
+        ec_method_fini(&ec->matrix);
+
+        GF_FREE(ec);
+    }
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    if (xlator_mem_acct_init(this, ec_mt_end + 1) != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+               "Memory accounting initialization "
+               "failed.");
+
+        return -1;
+    }
+
+    return 0;
+}
+
+void
+ec_configure_background_heal_opts(ec_t *ec, int background_heals,
+                                  int heal_wait_qlen)
+{
+    if (background_heals == 0) {
+        ec->heal_wait_qlen = 0;
+    } else {
+        ec->heal_wait_qlen = heal_wait_qlen;
+    }
+    ec->background_heals = background_heals;
+}
+
+int
+ec_assign_read_policy(ec_t *ec, char *read_policy)
+{
+    int read_policy_idx = -1;
+
+    read_policy_idx = gf_get_index_by_elem(ec_read_policies, read_policy);
+    if (read_policy_idx < 0 || read_policy_idx >= EC_READ_POLICY_MAX)
+        return -1;
+
+    ec->read_policy = read_policy_idx;
+    return 0;
+}
+
+int32_t
+reconfigure(xlator_t *this, dict_t *options)
+{
+    ec_t *ec = this->private;
+    char *read_policy = NULL;
+    char *extensions = NULL;
+    uint32_t heal_wait_qlen = 0;
+    uint32_t background_heals = 0;
+    int32_t ret = -1;
+    int32_t err;
+
+    GF_OPTION_RECONF("cpu-extensions", extensions, options, str, failed);
+
+    GF_OPTION_RECONF("self-heal-daemon", ec->shd.enabled, options, bool,
+                     failed);
+    GF_OPTION_RECONF("iam-self-heal-daemon", ec->shd.iamshd, options, bool,
+                     failed);
+    GF_OPTION_RECONF("eager-lock", ec->eager_lock, options, bool, failed);
+    GF_OPTION_RECONF("other-eager-lock", ec->other_eager_lock, options, bool,
+                     failed);
+    GF_OPTION_RECONF("eager-lock-timeout", ec->eager_lock_timeout, options,
+                     uint32, failed);
+    GF_OPTION_RECONF("other-eager-lock-timeout", ec->other_eager_lock_timeout,
+                     options, uint32, failed);
+    GF_OPTION_RECONF("background-heals", background_heals, options, uint32,
+                     failed);
+    GF_OPTION_RECONF("heal-wait-qlength", heal_wait_qlen, options, uint32,
+                     failed);
+    GF_OPTION_RECONF("self-heal-window-size", ec->self_heal_window_size,
+                     options, uint32, failed);
+    GF_OPTION_RECONF("heal-timeout", ec->shd.timeout, options, int32, failed);
+    ec_configure_background_heal_opts(ec, background_heals, heal_wait_qlen);
+    GF_OPTION_RECONF("shd-max-threads", ec->shd.max_threads, options, uint32,
+                     failed);
+    GF_OPTION_RECONF("shd-wait-qlength", ec->shd.wait_qlength, options, uint32,
+                     failed);
+
+    GF_OPTION_RECONF("read-policy", read_policy, options, str, failed);
+
+    GF_OPTION_RECONF("optimistic-change-log", ec->optimistic_changelog, options,
+                     bool, failed);
+    GF_OPTION_RECONF("parallel-writes", ec->parallel_writes, options, bool,
+                     failed);
+    GF_OPTION_RECONF("stripe-cache", ec->stripe_cache, options, uint32, failed);
+    GF_OPTION_RECONF("quorum-count", ec->quorum_count, options, uint32, failed);
+    ret = 0;
+    if (ec_assign_read_policy(ec, read_policy)) {
+        ret = -1;
+    }
+
+    err = ec_method_update(this, &ec->matrix, extensions);
+    if (err != 0) {
+        ret = -1;
+    }
+
+failed:
+    return ret;
+}
+
+glusterfs_event_t
+ec_get_event_from_state(ec_t *ec)
+{
+    int down_count = 0;
+
+    if (ec->xl_up_count >= ec->fragments) {
+        /* If ec is up but some subvolumes are yet to notify, give
+         * grace time for other subvols to notify to prevent start of
+         * I/O which may result in self-heals */
+        if (ec->xl_notify_count < ec->nodes)
+            return GF_EVENT_MAXVAL;
+
+        return GF_EVENT_CHILD_UP;
+    } else {
+        down_count = ec->xl_notify_count - ec->xl_up_count;
+        if (down_count > ec->redundancy)
+            return GF_EVENT_CHILD_DOWN;
+    }
+
+    return GF_EVENT_MAXVAL;
+}
+
+void
+ec_up(xlator_t *this, ec_t *ec)
+{
+    char str1[32], str2[32];
+
+    if (ec->timer != NULL) {
+        gf_timer_call_cancel(this->ctx, ec->timer);
+        ec->timer = NULL;
+    }
+
+    ec->up = 1;
+    gf_msg(this->name, GF_LOG_INFO, 0, EC_MSG_EC_UP,
+           "Going UP : Child UP = %s Child Notify = %s",
+           ec_bin(str1, sizeof(str1), ec->xl_up, ec->nodes),
+           ec_bin(str2, sizeof(str2), ec->xl_notify, ec->nodes));
+
+    gf_event(EVENT_EC_MIN_BRICKS_UP, "subvol=%s", this->name);
+}
+
+void
+ec_down(xlator_t *this, ec_t *ec)
+{
+    char str1[32], str2[32];
+
+    if (ec->timer != NULL) {
+        gf_timer_call_cancel(this->ctx, ec->timer);
+        ec->timer = NULL;
+    }
+
+    ec->up = 0;
+    gf_msg(this->name, GF_LOG_INFO, 0, EC_MSG_EC_DOWN,
+           "Going DOWN : Child UP = %s Child Notify = %s",
+           ec_bin(str1, sizeof(str1), ec->xl_up, ec->nodes),
+           ec_bin(str2, sizeof(str2), ec->xl_notify, ec->nodes));
+
+    gf_event(EVENT_EC_MIN_BRICKS_NOT_UP, "subvol=%s", this->name);
+}
+
+void
+ec_notify_cbk(void *data)
+{
+    ec_t *ec = data;
+    glusterfs_event_t event = GF_EVENT_MAXVAL;
+    gf_boolean_t propagate = _gf_false;
+    gf_boolean_t launch_heal = _gf_false;
+
+    LOCK(&ec->lock);
+    {
+        if (!ec->timer) {
+            /*
+             * Either child_up/child_down is already sent to parent
+             * This is a spurious wake up.
+             */
+            goto unlock;
+        }
+
+        gf_timer_call_cancel(ec->xl->ctx, ec->timer);
+        ec->timer = NULL;
+
+        /* The timeout has expired, so any subvolume that has not
+         * already reported its state, will be considered to be down.
+         * We mark as if all bricks had reported. */
+        ec->xl_notify = (1ULL << ec->nodes) - 1ULL;
+        ec->xl_notify_count = ec->nodes;
+
+        /* Since we have marked all subvolumes as notified, it's
+         * guaranteed that ec_get_event_from_state() will return
+         * CHILD_UP or CHILD_DOWN, but not MAXVAL. */
+        event = ec_get_event_from_state(ec);
+        if (event == GF_EVENT_CHILD_UP) {
+            /* We are ready to bring the volume up. If there are
+             * still bricks DOWN, they will be healed when they
+             * come up. */
+            ec_up(ec->xl, ec);
+
+            if (ec->shd.iamshd && !ec->shutdown) {
+                launch_heal = _gf_true;
+                GF_ATOMIC_INC(ec->async_fop_count);
+            }
+        }
+
+        propagate = _gf_true;
+    }
+unlock:
+    UNLOCK(&ec->lock);
+
+    if (launch_heal) {
+        /* We have just brought the volume UP, so we trigger
+         * a self-heal check on the root directory. */
+        ec_launch_replace_heal(ec);
+    }
+    if (propagate) {
+        default_notify(ec->xl, event, NULL);
+    }
+}
+
+void
+ec_launch_notify_timer(xlator_t *this, ec_t *ec)
+{
+    struct timespec delay = {
+        0,
+    };
+
+    gf_msg_debug(this->name, 0, "Initiating child-down timer");
+    delay.tv_sec = 10;
+    delay.tv_nsec = 0;
+    ec->timer = gf_timer_call_after(this->ctx, delay, ec_notify_cbk, ec);
+    if (ec->timer == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_TIMER_CREATE_FAIL,
+               "Cannot create timer "
+               "for delayed initialization");
+    }
+}
+
+gf_boolean_t
+ec_disable_delays(ec_t *ec)
+{
+    ec->shutdown = _gf_true;
+
+    return __ec_is_last_fop(ec);
+}
+
+void
+ec_cleanup_healer_object(ec_t *ec)
+{
+    struct subvol_healer *healer = NULL;
+    ec_self_heald_t *shd = NULL;
+    void *res = NULL;
+    int i = 0;
+    gf_boolean_t is_join = _gf_false;
+
+    shd = &ec->shd;
+    if (!shd->iamshd)
+        return;
+
+    for (i = 0; i < ec->nodes; i++) {
+        healer = &shd->index_healers[i];
+        pthread_mutex_lock(&healer->mutex);
+        {
+            healer->rerun = 1;
+            if (healer->running) {
+                pthread_cond_signal(&healer->cond);
+                is_join = _gf_true;
+            }
+        }
+        pthread_mutex_unlock(&healer->mutex);
+        if (is_join) {
+            pthread_join(healer->thread, &res);
+            is_join = _gf_false;
+        }
+
+        healer = &shd->full_healers[i];
+        pthread_mutex_lock(&healer->mutex);
+        {
+            healer->rerun = 1;
+            if (healer->running) {
+                pthread_cond_signal(&healer->cond);
+                is_join = _gf_true;
+            }
+        }
+        pthread_mutex_unlock(&healer->mutex);
+        if (is_join) {
+            pthread_join(healer->thread, &res);
+            is_join = _gf_false;
+        }
+    }
+}
+void
+ec_pending_fops_completed(ec_t *ec)
+{
+    if (ec->shutdown) {
+        default_notify(ec->xl, GF_EVENT_PARENT_DOWN, NULL);
+    }
+}
+
+static gf_boolean_t
+ec_set_up_state(ec_t *ec, uintptr_t index_mask, uintptr_t new_state)
+{
+    uintptr_t current_state = 0;
+
+    if (xlator_is_cleanup_starting(ec->xl))
+        return _gf_false;
+
+    if ((ec->xl_notify & index_mask) == 0) {
+        ec->xl_notify |= index_mask;
+        ec->xl_notify_count++;
+    }
+    current_state = ec->xl_up & index_mask;
+    if (current_state != new_state) {
+        ec->xl_up ^= index_mask;
+        ec->xl_up_count += (current_state ? -1 : 1);
+
+        return _gf_true;
+    }
+
+    return _gf_false;
+}
+
+static gf_boolean_t
+ec_upcall(ec_t *ec, struct gf_upcall *upcall)
+{
+    struct gf_upcall_cache_invalidation *ci = NULL;
+    struct gf_upcall_inodelk_contention *lc = NULL;
+    inode_t *inode;
+    inode_table_t *table;
+
+    switch (upcall->event_type) {
+        case GF_UPCALL_CACHE_INVALIDATION:
+            ci = upcall->data;
+            ci->flags |= UP_INVAL_ATTR;
+            return _gf_true;
+
+        case GF_UPCALL_INODELK_CONTENTION:
+            lc = upcall->data;
+            if (strcmp(lc->domain, ec->xl->name) != 0) {
+                /* The lock is not owned by EC, ignore it. */
+                return _gf_true;
+            }
+            table = ((xlator_t *)ec->xl->graph->top)->itable;
+            if (table == NULL) {
+                /* Self-heal daemon doesn't have an inode table on the top
+                 * xlator because it doesn't need it. In this case we should
+                 * use the inode table managed by EC itself where all inodes
+                 * being healed should be present. However self-heal doesn't
+                 * use eager-locking and inodelk's are already released as
+                 * soon as possible. In this case we can safely ignore these
+                 * notifications. */
+                return _gf_false;
+            }
+            inode = inode_find(table, upcall->gfid);
+            /* If inode is not found, it means that it's already released,
+             * so we can ignore it. Probably it has been released and
+             * destroyed while the contention notification was being sent.
+             */
+            if (inode != NULL) {
+                ec_lock_release(ec, inode);
+                inode_unref(inode);
+            }
+
+            return _gf_false;
+
+        default:
+            return _gf_true;
+    }
+}
+
+int32_t
+ec_notify(xlator_t *this, int32_t event, void *data, void *data2)
+{
+    ec_t *ec = this->private;
+    int32_t idx = 0;
+    int32_t error = 0;
+    glusterfs_event_t old_event = GF_EVENT_MAXVAL;
+    dict_t *input = NULL;
+    dict_t *output = NULL;
+    gf_boolean_t propagate = _gf_true;
+    gf_boolean_t needs_shd_check = _gf_false;
+    int32_t orig_event = event;
+    uintptr_t mask = 0;
+
+    gf_msg_trace(this->name, 0, "NOTIFY(%d): %p, %p", event, data, data2);
+
+    if (event == GF_EVENT_UPCALL) {
+        propagate = ec_upcall(ec, data);
+        goto done;
+    }
+
+    if (event == GF_EVENT_TRANSLATOR_OP) {
+        if (!ec->up) {
+            error = -1;
+        } else {
+            input = data;
+            output = data2;
+            error = ec_xl_op(this, input, output);
+        }
+        goto out;
+    }
+
+    for (idx = 0; idx < ec->nodes; idx++) {
+        if (ec->xl_list[idx] == data) {
+            break;
+        }
+    }
+
+    LOCK(&ec->lock);
+
+    if (event == GF_EVENT_PARENT_UP) {
+        /*
+         * Start a timer which sends appropriate event to parent
+         * xlator to prevent the 'mount' syscall from hanging.
+         */
+        ec_launch_notify_timer(this, ec);
+        goto unlock;
+    } else if (event == GF_EVENT_PARENT_DOWN) {
+        /* If there aren't pending fops running after we have waken up
+         * them, we immediately propagate the notification. */
+        propagate = ec_disable_delays(ec);
+        ec_cleanup_healer_object(ec);
+        goto unlock;
+    }
+
+    if (idx < ec->nodes) { /* CHILD_* events */
+        old_event = ec_get_event_from_state(ec);
+
+        mask = 1ULL << idx;
+        if (event == GF_EVENT_CHILD_UP) {
+            /* We need to trigger a selfheal if a brick changes
+             * to UP state. */
+            if (ec_set_up_state(ec, mask, mask) && ec->shd.iamshd &&
+                !ec->shutdown) {
+                needs_shd_check = _gf_true;
+            }
+        } else if (event == GF_EVENT_CHILD_DOWN) {
+            ec_set_up_state(ec, mask, 0);
+        }
+
+        event = ec_get_event_from_state(ec);
+
+        if (event == GF_EVENT_CHILD_UP) {
+            if (!ec->up) {
+                ec_up(this, ec);
+            }
+        } else {
+            /* If the volume is not UP, it's irrelevant if one
+             * brick has come up. We cannot heal anything. */
+            needs_shd_check = _gf_false;
+
+            if ((event == GF_EVENT_CHILD_DOWN) && ec->up) {
+                ec_down(this, ec);
+            }
+        }
+
+        if (event != GF_EVENT_MAXVAL) {
+            if (event == old_event) {
+                if (orig_event == GF_EVENT_CHILD_UP)
+                    event = GF_EVENT_SOME_DESCENDENT_UP;
+                else /* orig_event has to be GF_EVENT_CHILD_DOWN */
+                    event = GF_EVENT_SOME_DESCENDENT_DOWN;
+            }
+        } else {
+            propagate = _gf_false;
+            needs_shd_check = _gf_false;
+        }
+
+        if (needs_shd_check) {
+            GF_ATOMIC_INC(ec->async_fop_count);
+        }
+    }
+unlock:
+    UNLOCK(&ec->lock);
+
+done:
+    if (needs_shd_check) {
+        ec_launch_replace_heal(ec);
+    }
+    if (propagate) {
+        error = default_notify(this, event, data);
+    }
+
+out:
+    return error;
+}
+
+int32_t
+notify(xlator_t *this, int32_t event, void *data, ...)
+{
+    int ret = -1;
+    va_list ap;
+    void *data2 = NULL;
+
+    va_start(ap, data);
+    data2 = va_arg(ap, dict_t *);
+    va_end(ap);
+    ret = ec_notify(this, event, data, data2);
+
+    return ret;
+}
+
+static void
+ec_statistics_init(ec_t *ec)
+{
+    GF_ATOMIC_INIT(ec->stats.stripe_cache.hits, 0);
+    GF_ATOMIC_INIT(ec->stats.stripe_cache.misses, 0);
+    GF_ATOMIC_INIT(ec->stats.stripe_cache.updates, 0);
+    GF_ATOMIC_INIT(ec->stats.stripe_cache.invals, 0);
+    GF_ATOMIC_INIT(ec->stats.stripe_cache.evicts, 0);
+    GF_ATOMIC_INIT(ec->stats.stripe_cache.allocs, 0);
+    GF_ATOMIC_INIT(ec->stats.stripe_cache.errors, 0);
+    GF_ATOMIC_INIT(ec->stats.shd.attempted, 0);
+    GF_ATOMIC_INIT(ec->stats.shd.completed, 0);
+}
+
+static int
+ec_assign_read_mask(ec_t *ec, char *read_mask_str)
+{
+    char *mask = NULL;
+    char *maskptr = NULL;
+    char *saveptr = NULL;
+    char *id_str = NULL;
+    int id = 0;
+    int ret = 0;
+    uintptr_t read_mask = 0;
+
+    if (!read_mask_str) {
+        ec->read_mask = 0;
+        ret = 0;
+        goto out;
+    }
+
+    mask = gf_strdup(read_mask_str);
+    if (!mask) {
+        ret = -1;
+        goto out;
+    }
+    maskptr = mask;
+
+    for (;;) {
+        id_str = strtok_r(maskptr, ":", &saveptr);
+        if (id_str == NULL)
+            break;
+        if (gf_string2int(id_str, &id)) {
+            gf_msg(ec->xl->name, GF_LOG_ERROR, 0, EC_MSG_XLATOR_INIT_FAIL,
+                   "In read-mask \"%s\" id %s is not a valid integer",
+                   read_mask_str, id_str);
+            ret = -1;
+            goto out;
+        }
+
+        if ((id < 0) || (id >= ec->nodes)) {
+            gf_msg(ec->xl->name, GF_LOG_ERROR, 0, EC_MSG_XLATOR_INIT_FAIL,
+                   "In read-mask \"%s\" id %d is not in range [0 - %d]",
+                   read_mask_str, id, ec->nodes - 1);
+            ret = -1;
+            goto out;
+        }
+        read_mask |= (1UL << id);
+        maskptr = NULL;
+    }
+
+    if (gf_bits_count(read_mask) < ec->fragments) {
+        gf_msg(ec->xl->name, GF_LOG_ERROR, 0, EC_MSG_XLATOR_INIT_FAIL,
+               "read-mask \"%s\" should contain at least %d ids", read_mask_str,
+               ec->fragments);
+        ret = -1;
+        goto out;
+    }
+    ec->read_mask = read_mask;
+    ret = 0;
+out:
+    GF_FREE(mask);
+    return ret;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    ec_t *ec = NULL;
+    char *read_policy = NULL;
+    char *extensions = NULL;
+    int32_t err;
+    char *read_mask_str = NULL;
+
+    if (this->parents == NULL) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, EC_MSG_NO_PARENTS,
+               "Volume does not have parents.");
+    }
+
+    ec = GF_MALLOC(sizeof(*ec), ec_mt_ec_t);
+    if (ec == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+               "Failed to allocate private memory.");
+
+        return -1;
+    }
+    memset(ec, 0, sizeof(*ec));
+
+    this->private = ec;
+
+    ec->xl = this;
+    LOCK_INIT(&ec->lock);
+
+    GF_ATOMIC_INIT(ec->async_fop_count, 0);
+    INIT_LIST_HEAD(&ec->pending_fops);
+    INIT_LIST_HEAD(&ec->heal_waiting);
+    INIT_LIST_HEAD(&ec->healing);
+
+    ec->fop_pool = mem_pool_new(ec_fop_data_t, 1024);
+    ec->cbk_pool = mem_pool_new(ec_cbk_data_t, 4096);
+    ec->lock_pool = mem_pool_new(ec_lock_t, 1024);
+    if ((ec->fop_pool == NULL) || (ec->cbk_pool == NULL) ||
+        (ec->lock_pool == NULL)) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, EC_MSG_NO_MEMORY,
+               "Failed to create memory pools.");
+
+        goto failed;
+    }
+
+    if (ec_prepare_childs(this) != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_XLATOR_INIT_FAIL,
+               "Failed to initialize xlator");
+
+        goto failed;
+    }
+
+    if (ec_parse_options(this) != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, EC_MSG_XLATOR_PARSE_OPT_FAIL,
+               "Failed to parse xlator options");
+
+        goto failed;
+    }
+
+    GF_OPTION_INIT("cpu-extensions", extensions, str, failed);
+
+    err = ec_method_init(this, &ec->matrix, ec->fragments, ec->nodes,
+                         ec->nodes * 2, extensions);
+    if (err != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, -err, EC_MSG_MATRIX_FAILED,
+               "Failed to initialize matrix management");
+
+        goto failed;
+    }
+
+    GF_OPTION_INIT("self-heal-daemon", ec->shd.enabled, bool, failed);
+    GF_OPTION_INIT("iam-self-heal-daemon", ec->shd.iamshd, bool, failed);
+    GF_OPTION_INIT("eager-lock", ec->eager_lock, bool, failed);
+    GF_OPTION_INIT("other-eager-lock", ec->other_eager_lock, bool, failed);
+    GF_OPTION_INIT("eager-lock-timeout", ec->eager_lock_timeout, uint32,
+                   failed);
+    GF_OPTION_INIT("other-eager-lock-timeout", ec->other_eager_lock_timeout,
+                   uint32, failed);
+    GF_OPTION_INIT("background-heals", ec->background_heals, uint32, failed);
+    GF_OPTION_INIT("heal-wait-qlength", ec->heal_wait_qlen, uint32, failed);
+    GF_OPTION_INIT("self-heal-window-size", ec->self_heal_window_size, uint32,
+                   failed);
+    ec_configure_background_heal_opts(ec, ec->background_heals,
+                                      ec->heal_wait_qlen);
+    GF_OPTION_INIT("read-policy", read_policy, str, failed);
+    if (ec_assign_read_policy(ec, read_policy))
+        goto failed;
+
+    GF_OPTION_INIT("heal-timeout", ec->shd.timeout, int32, failed);
+    GF_OPTION_INIT("shd-max-threads", ec->shd.max_threads, uint32, failed);
+    GF_OPTION_INIT("shd-wait-qlength", ec->shd.wait_qlength, uint32, failed);
+    GF_OPTION_INIT("optimistic-change-log", ec->optimistic_changelog, bool,
+                   failed);
+    GF_OPTION_INIT("parallel-writes", ec->parallel_writes, bool, failed);
+    GF_OPTION_INIT("stripe-cache", ec->stripe_cache, uint32, failed);
+    GF_OPTION_INIT("quorum-count", ec->quorum_count, uint32, failed);
+    GF_OPTION_INIT("ec-read-mask", read_mask_str, str, failed);
+
+    if (ec_assign_read_mask(ec, read_mask_str))
+        goto failed;
+
+    this->itable = inode_table_new(EC_SHD_INODE_LRU_LIMIT, this);
+    if (!this->itable)
+        goto failed;
+
+    if (ec->shd.iamshd)
+        ec_selfheal_daemon_init(this);
+    gf_msg_debug(this->name, 0, "Disperse translator initialized.");
+
+    ec->leaf_to_subvolid = dict_new();
+    if (!ec->leaf_to_subvolid)
+        goto failed;
+    if (glusterfs_reachable_leaves(this, ec->leaf_to_subvolid)) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_SUBVOL_BUILD_FAIL,
+               "Failed to build subvol "
+               "dictionary");
+        goto failed;
+    }
+
+    if (ec_subvol_to_subvol_id_transform(ec, ec->leaf_to_subvolid) < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, EC_MSG_SUBVOL_ID_DICT_SET_FAIL,
+               "Failed to build subvol-id "
+               "dictionary");
+        goto failed;
+    }
+
+    ec_statistics_init(ec);
+
+    return 0;
+
+failed:
+    __ec_destroy_private(this);
+
+    return -1;
+}
+
+void
+fini(xlator_t *this)
+{
+    ec_selfheal_daemon_fini(this);
+    __ec_destroy_private(this);
+}
+
+int32_t
+ec_gf_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+             dict_t *xdata)
+{
+    ec_access(frame, this, -1, EC_MINIMUM_ONE, default_access_cbk, NULL, loc,
+              mask, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+             mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+    ec_create(frame, this, -1, EC_MINIMUM_MIN, default_create_cbk, NULL, loc,
+              flags, mode, umask, fd, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+              size_t len, dict_t *xdata)
+{
+    ec_discard(frame, this, -1, EC_MINIMUM_MIN, default_discard_cbk, NULL, fd,
+               offset, len, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_entrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+              loc_t *loc, const char *basename, entrylk_cmd cmd,
+              entrylk_type type, dict_t *xdata)
+{
+    uint32_t fop_flags = EC_MINIMUM_ALL;
+
+    if (cmd == ENTRYLK_UNLOCK)
+        fop_flags = EC_MINIMUM_ONE;
+    ec_entrylk(frame, this, -1, fop_flags, default_entrylk_cbk, NULL, volume,
+               loc, basename, cmd, type, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+               fd_t *fd, const char *basename, entrylk_cmd cmd,
+               entrylk_type type, dict_t *xdata)
+{
+    uint32_t fop_flags = EC_MINIMUM_ALL;
+
+    if (cmd == ENTRYLK_UNLOCK)
+        fop_flags = EC_MINIMUM_ONE;
+    ec_fentrylk(frame, this, -1, fop_flags, default_fentrylk_cbk, NULL, volume,
+                fd, basename, cmd, type, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+                off_t offset, size_t len, dict_t *xdata)
+{
+    ec_fallocate(frame, this, -1, EC_MINIMUM_MIN, default_fallocate_cbk, NULL,
+                 fd, mode, offset, len, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    ec_flush(frame, this, -1, EC_MINIMUM_MIN, default_flush_cbk, NULL, fd,
+             xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+            dict_t *xdata)
+{
+    ec_fsync(frame, this, -1, EC_MINIMUM_MIN, default_fsync_cbk, NULL, fd,
+             datasync, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+               dict_t *xdata)
+{
+    ec_fsyncdir(frame, this, -1, EC_MINIMUM_MIN, default_fsyncdir_cbk, NULL, fd,
+                datasync, xdata);
+
+    return 0;
+}
+
+int
+ec_marker_populate_args(call_frame_t *frame, int type, int *gauge,
+                        xlator_t **subvols)
+{
+    xlator_t *this = frame->this;
+    ec_t *ec = this->private;
+
+    memcpy(subvols, ec->xl_list, sizeof(*subvols) * ec->nodes);
+
+    if (type == MARKER_XTIME_TYPE) {
+        /*Don't error out on ENOENT/ENOTCONN */
+        gauge[MCNT_NOTFOUND] = 0;
+        gauge[MCNT_ENOTCONN] = 0;
+    }
+
+    return ec->nodes;
+}
+
+int32_t
+ec_handle_heal_commands(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                        const char *name, dict_t *xdata)
+{
+    dict_t *dict_rsp = NULL;
+    int op_ret = -1;
+    int op_errno = ENOMEM;
+
+    if (!name || strcmp(name, GF_HEAL_INFO))
+        return -1;
+
+    op_errno = -ec_get_heal_info(this, loc, &dict_rsp);
+    if (op_errno <= 0) {
+        op_errno = op_ret = 0;
+    }
+
+    STACK_UNWIND_STRICT(getxattr, frame, op_ret, op_errno, dict_rsp, NULL);
+    if (dict_rsp)
+        dict_unref(dict_rsp);
+    return 0;
+}
+
+int32_t
+ec_gf_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               const char *name, dict_t *xdata)
+{
+    int error = 0;
+    ec_t *ec = this->private;
+    int32_t fop_flags = EC_MINIMUM_ONE;
+
+    if (name && strcmp(name, EC_XATTR_HEAL) != 0) {
+        EC_INTERNAL_XATTR_OR_GOTO(name, NULL, error, out);
+    }
+
+    if (ec_handle_heal_commands(frame, this, loc, name, xdata) == 0)
+        return 0;
+
+    if (cluster_handle_marker_getxattr(frame, loc, name, ec->vol_uuid, NULL,
+                                       ec_marker_populate_args) == 0)
+        return 0;
+
+    if (name && ((fnmatch(GF_XATTR_STIME_PATTERN, name, 0) == 0) ||
+                 XATTR_IS_NODE_UUID(name) || XATTR_IS_NODE_UUID_LIST(name))) {
+        fop_flags = EC_MINIMUM_ALL;
+    }
+
+    ec_getxattr(frame, this, -1, fop_flags, default_getxattr_cbk, NULL, loc,
+                name, xdata);
+
+    return 0;
+out:
+    error = ENODATA;
+    STACK_UNWIND_STRICT(getxattr, frame, -1, error, NULL, NULL);
+    return 0;
+}
+
+int32_t
+ec_gf_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+                dict_t *xdata)
+{
+    int error = 0;
+
+    EC_INTERNAL_XATTR_OR_GOTO(name, NULL, error, out);
+
+    ec_fgetxattr(frame, this, -1, EC_MINIMUM_ONE, default_fgetxattr_cbk, NULL,
+                 fd, name, xdata);
+    return 0;
+out:
+    error = ENODATA;
+    STACK_UNWIND_STRICT(fgetxattr, frame, -1, error, NULL, NULL);
+    return 0;
+}
+
+int32_t
+ec_gf_inodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+              loc_t *loc, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
+{
+    int32_t fop_flags = EC_MINIMUM_ALL;
+
+    if (flock->l_type == F_UNLCK)
+        fop_flags = EC_MINIMUM_ONE;
+
+    ec_inodelk(frame, this, &frame->root->lk_owner, -1, fop_flags,
+               default_inodelk_cbk, NULL, volume, loc, cmd, flock, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_finodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+               fd_t *fd, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
+{
+    int32_t fop_flags = EC_MINIMUM_ALL;
+
+    if (flock->l_type == F_UNLCK)
+        fop_flags = EC_MINIMUM_ONE;
+    ec_finodelk(frame, this, &frame->root->lk_owner, -1, fop_flags,
+                default_finodelk_cbk, NULL, volume, fd, cmd, flock, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+           dict_t *xdata)
+{
+    ec_link(frame, this, -1, EC_MINIMUM_MIN, default_link_cbk, NULL, oldloc,
+            newloc, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+         struct gf_flock *flock, dict_t *xdata)
+{
+    int32_t fop_flags = EC_MINIMUM_ALL;
+
+    if (flock->l_type == F_UNLCK)
+        fop_flags = EC_MINIMUM_ONE;
+    ec_lk(frame, this, -1, fop_flags, default_lk_cbk, NULL, fd, cmd, flock,
+          xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    ec_lookup(frame, this, -1, EC_MINIMUM_MIN, default_lookup_cbk, NULL, loc,
+              xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+            mode_t umask, dict_t *xdata)
+{
+    ec_mkdir(frame, this, -1, EC_MINIMUM_MIN, default_mkdir_cbk, NULL, loc,
+             mode, umask, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+            dev_t rdev, mode_t umask, dict_t *xdata)
+{
+    ec_mknod(frame, this, -1, EC_MINIMUM_MIN, default_mknod_cbk, NULL, loc,
+             mode, rdev, umask, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+           fd_t *fd, dict_t *xdata)
+{
+    ec_open(frame, this, -1, EC_MINIMUM_MIN, default_open_cbk, NULL, loc, flags,
+            fd, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+              dict_t *xdata)
+{
+    ec_opendir(frame, this, -1, EC_MINIMUM_MIN, default_opendir_cbk, NULL, loc,
+               fd, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+              off_t offset, dict_t *xdata)
+{
+    ec_readdir(frame, this, -1, EC_MINIMUM_ONE, default_readdir_cbk, NULL, fd,
+               size, offset, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+               off_t offset, dict_t *xdata)
+{
+    ec_readdirp(frame, this, -1, EC_MINIMUM_ONE, default_readdirp_cbk, NULL, fd,
+                size, offset, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+               dict_t *xdata)
+{
+    ec_readlink(frame, this, -1, EC_MINIMUM_ONE, default_readlink_cbk, NULL,
+                loc, size, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t offset, uint32_t flags, dict_t *xdata)
+{
+    ec_readv(frame, this, -1, EC_MINIMUM_MIN, default_readv_cbk, NULL, fd, size,
+             offset, flags, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                  const char *name, dict_t *xdata)
+{
+    int error = 0;
+
+    EC_INTERNAL_XATTR_OR_GOTO(name, xdata, error, out);
+
+    ec_removexattr(frame, this, -1, EC_MINIMUM_MIN, default_removexattr_cbk,
+                   NULL, loc, name, xdata);
+
+    return 0;
+out:
+    STACK_UNWIND_STRICT(removexattr, frame, -1, error, NULL);
+    return 0;
+}
+
+int32_t
+ec_gf_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                   const char *name, dict_t *xdata)
+{
+    int error = 0;
+
+    EC_INTERNAL_XATTR_OR_GOTO(name, xdata, error, out);
+
+    ec_fremovexattr(frame, this, -1, EC_MINIMUM_MIN, default_fremovexattr_cbk,
+                    NULL, fd, name, xdata);
+
+    return 0;
+out:
+    STACK_UNWIND_STRICT(fremovexattr, frame, -1, error, NULL);
+    return 0;
+}
+
+int32_t
+ec_gf_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+             dict_t *xdata)
+{
+    ec_rename(frame, this, -1, EC_MINIMUM_MIN, default_rename_cbk, NULL, oldloc,
+              newloc, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflags,
+            dict_t *xdata)
+{
+    ec_rmdir(frame, this, -1, EC_MINIMUM_MIN, default_rmdir_cbk, NULL, loc,
+             xflags, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+              struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    ec_setattr(frame, this, -1, EC_MINIMUM_MIN, default_setattr_cbk, NULL, loc,
+               stbuf, valid, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+               struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    ec_fsetattr(frame, this, -1, EC_MINIMUM_MIN, default_fsetattr_cbk, NULL, fd,
+                stbuf, valid, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+               int32_t flags, dict_t *xdata)
+{
+    int error = 0;
+
+    EC_INTERNAL_XATTR_OR_GOTO("", dict, error, out);
+
+    ec_setxattr(frame, this, -1, EC_MINIMUM_MIN, default_setxattr_cbk, NULL,
+                loc, dict, flags, xdata);
+
+    return 0;
+out:
+    STACK_UNWIND_STRICT(setxattr, frame, -1, error, NULL);
+    return 0;
+}
+
+int32_t
+ec_gf_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+                int32_t flags, dict_t *xdata)
+{
+    int error = 0;
+
+    EC_INTERNAL_XATTR_OR_GOTO("", dict, error, out);
+
+    ec_fsetxattr(frame, this, -1, EC_MINIMUM_MIN, default_fsetxattr_cbk, NULL,
+                 fd, dict, flags, xdata);
+
+    return 0;
+out:
+    STACK_UNWIND_STRICT(fsetxattr, frame, -1, error, NULL);
+    return 0;
+}
+
+int32_t
+ec_gf_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    ec_stat(frame, this, -1, EC_MINIMUM_MIN, default_stat_cbk, NULL, loc,
+            xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    ec_fstat(frame, this, -1, EC_MINIMUM_MIN, default_fstat_cbk, NULL, fd,
+             xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    ec_statfs(frame, this, -1, EC_MINIMUM_MIN, default_statfs_cbk, NULL, loc,
+              xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_symlink(call_frame_t *frame, xlator_t *this, const char *linkname,
+              loc_t *loc, mode_t umask, dict_t *xdata)
+{
+    ec_symlink(frame, this, -1, EC_MINIMUM_MIN, default_symlink_cbk, NULL,
+               linkname, loc, umask, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+               dict_t *xdata)
+{
+    ec_truncate(frame, this, -1, EC_MINIMUM_MIN, default_truncate_cbk, NULL,
+                loc, offset, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                dict_t *xdata)
+{
+    ec_ftruncate(frame, this, -1, EC_MINIMUM_MIN, default_ftruncate_cbk, NULL,
+                 fd, offset, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflags,
+             dict_t *xdata)
+{
+    ec_unlink(frame, this, -1, EC_MINIMUM_MIN, default_unlink_cbk, NULL, loc,
+              xflags, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_writev(call_frame_t *frame, xlator_t *this, fd_t *fd,
+             struct iovec *vector, int32_t count, off_t offset, uint32_t flags,
+             struct iobref *iobref, dict_t *xdata)
+{
+    ec_writev(frame, this, -1, EC_MINIMUM_MIN, default_writev_cbk, NULL, fd,
+              vector, count, offset, flags, iobref, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+              gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+    ec_xattrop(frame, this, -1, EC_MINIMUM_MIN, default_xattrop_cbk, NULL, loc,
+               optype, xattr, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+               gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+    ec_fxattrop(frame, this, -1, EC_MINIMUM_MIN, default_fxattrop_cbk, NULL, fd,
+                optype, xattr, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+               off_t len, dict_t *xdata)
+{
+    default_zerofill_failure_cbk(frame, ENOTSUP);
+
+    return 0;
+}
+
+int32_t
+ec_gf_seek(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+           gf_seek_what_t what, dict_t *xdata)
+{
+    ec_seek(frame, this, -1, EC_MINIMUM_ONE, default_seek_cbk, NULL, fd, offset,
+            what, xdata);
+
+    return 0;
+}
+
+int32_t
+ec_gf_ipc(call_frame_t *frame, xlator_t *this, int32_t op, dict_t *xdata)
+{
+    ec_ipc(frame, this, -1, EC_MINIMUM_MIN, default_ipc_cbk, NULL, op, xdata);
+    return 0;
+}
+
+int32_t
+ec_gf_forget(xlator_t *this, inode_t *inode)
+{
+    uint64_t value = 0;
+    ec_inode_t *ctx = NULL;
+
+    if ((inode_ctx_del(inode, this, &value) == 0) && (value != 0)) {
+        ctx = (ec_inode_t *)(uintptr_t)value;
+        /* We can only forget an inode if it has been unlocked, so the stripe
+         * cache should also be empty. */
+        GF_ASSERT(list_empty(&ctx->stripe_cache.lru));
+        GF_FREE(ctx);
+    }
+
+    return 0;
+}
+
+void
+ec_gf_release_fd(xlator_t *this, fd_t *fd)
+{
+    uint64_t value = 0;
+    ec_fd_t *ctx = NULL;
+
+    if ((fd_ctx_del(fd, this, &value) == 0) && (value != 0)) {
+        ctx = (ec_fd_t *)(uintptr_t)value;
+        loc_wipe(&ctx->loc);
+        GF_FREE(ctx);
+    }
+}
+
+int32_t
+ec_gf_release(xlator_t *this, fd_t *fd)
+{
+    ec_gf_release_fd(this, fd);
+
+    return 0;
+}
+
+int32_t
+ec_gf_releasedir(xlator_t *this, fd_t *fd)
+{
+    ec_gf_release_fd(this, fd);
+
+    return 0;
+}
+
+int32_t
+ec_dump_private(xlator_t *this)
+{
+    ec_t *ec = NULL;
+    char key_prefix[GF_DUMP_MAX_BUF_LEN];
+    char tmp[65];
+
+    GF_ASSERT(this);
+
+    ec = this->private;
+    GF_ASSERT(ec);
+
+    snprintf(key_prefix, GF_DUMP_MAX_BUF_LEN, "%s.%s", this->type, this->name);
+    gf_proc_dump_add_section("%s", key_prefix);
+    gf_proc_dump_write("up", "%u", ec->up);
+    gf_proc_dump_write("nodes", "%u", ec->nodes);
+    gf_proc_dump_write("redundancy", "%u", ec->redundancy);
+    gf_proc_dump_write("fragment_size", "%u", ec->fragment_size);
+    gf_proc_dump_write("stripe_size", "%u", ec->stripe_size);
+    gf_proc_dump_write("childs_up", "%u", ec->xl_up_count);
+    gf_proc_dump_write("childs_up_mask", "%s",
+                       ec_bin(tmp, sizeof(tmp), ec->xl_up, ec->nodes));
+    if (ec->read_mask) {
+        gf_proc_dump_write("read-mask", "%s",
+                           ec_bin(tmp, sizeof(tmp), ec->read_mask, ec->nodes));
+    }
+    gf_proc_dump_write("background-heals", "%d", ec->background_heals);
+    gf_proc_dump_write("heal-wait-qlength", "%d", ec->heal_wait_qlen);
+    gf_proc_dump_write("self-heal-window-size", "%" PRIu32,
+                       ec->self_heal_window_size);
+    gf_proc_dump_write("healers", "%d", ec->healers);
+    gf_proc_dump_write("heal-waiters", "%d", ec->heal_waiters);
+    gf_proc_dump_write("read-policy", "%s", ec_read_policies[ec->read_policy]);
+    gf_proc_dump_write("parallel-writes", "%d", ec->parallel_writes);
+    gf_proc_dump_write("quorum-count", "%u", ec->quorum_count);
+
+    snprintf(key_prefix, GF_DUMP_MAX_BUF_LEN, "%s.%s.stats.stripe_cache",
+             this->type, this->name);
+    gf_proc_dump_add_section("%s", key_prefix);
+
+    gf_proc_dump_write("hits", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(ec->stats.stripe_cache.hits));
+    gf_proc_dump_write("misses", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(ec->stats.stripe_cache.misses));
+    gf_proc_dump_write("updates", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(ec->stats.stripe_cache.updates));
+    gf_proc_dump_write("invalidations", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(ec->stats.stripe_cache.invals));
+    gf_proc_dump_write("evicts", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(ec->stats.stripe_cache.evicts));
+    gf_proc_dump_write("allocations", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(ec->stats.stripe_cache.allocs));
+    gf_proc_dump_write("errors", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(ec->stats.stripe_cache.errors));
+    gf_proc_dump_write("heals-attempted", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(ec->stats.shd.attempted));
+    gf_proc_dump_write("heals-completed", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(ec->stats.shd.completed));
+
+    return 0;
+}
+
+struct xlator_fops fops = {.lookup = ec_gf_lookup,
+                           .stat = ec_gf_stat,
+                           .fstat = ec_gf_fstat,
+                           .truncate = ec_gf_truncate,
+                           .ftruncate = ec_gf_ftruncate,
+                           .access = ec_gf_access,
+                           .readlink = ec_gf_readlink,
+                           .mknod = ec_gf_mknod,
+                           .mkdir = ec_gf_mkdir,
+                           .unlink = ec_gf_unlink,
+                           .rmdir = ec_gf_rmdir,
+                           .symlink = ec_gf_symlink,
+                           .rename = ec_gf_rename,
+                           .link = ec_gf_link,
+                           .create = ec_gf_create,
+                           .open = ec_gf_open,
+                           .readv = ec_gf_readv,
+                           .writev = ec_gf_writev,
+                           .flush = ec_gf_flush,
+                           .fsync = ec_gf_fsync,
+                           .opendir = ec_gf_opendir,
+                           .readdir = ec_gf_readdir,
+                           .readdirp = ec_gf_readdirp,
+                           .fsyncdir = ec_gf_fsyncdir,
+                           .statfs = ec_gf_statfs,
+                           .setxattr = ec_gf_setxattr,
+                           .getxattr = ec_gf_getxattr,
+                           .fsetxattr = ec_gf_fsetxattr,
+                           .fgetxattr = ec_gf_fgetxattr,
+                           .removexattr = ec_gf_removexattr,
+                           .fremovexattr = ec_gf_fremovexattr,
+                           .lk = ec_gf_lk,
+                           .inodelk = ec_gf_inodelk,
+                           .finodelk = ec_gf_finodelk,
+                           .entrylk = ec_gf_entrylk,
+                           .fentrylk = ec_gf_fentrylk,
+                           .xattrop = ec_gf_xattrop,
+                           .fxattrop = ec_gf_fxattrop,
+                           .setattr = ec_gf_setattr,
+                           .fsetattr = ec_gf_fsetattr,
+                           .fallocate = ec_gf_fallocate,
+                           .discard = ec_gf_discard,
+                           .zerofill = ec_gf_zerofill,
+                           .seek = ec_gf_seek,
+                           .ipc = ec_gf_ipc};
+
+struct xlator_cbks cbks = {.forget = ec_gf_forget,
+                           .release = ec_gf_release,
+                           .releasedir = ec_gf_releasedir};
+
+struct xlator_dumpops dumpops = {.priv = ec_dump_private};
+
+struct volume_options options[] = {
+    {.key = {"redundancy"},
+     .type = GF_OPTION_TYPE_INT,
+     .default_value = "{{ volume.redundancy }}",
+     .description = "Maximum number of bricks that can fail "
+                    "simultaneously without losing data."},
+    {
+        .key = {"self-heal-daemon"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .description = "self-heal daemon enable/disable",
+        .default_value = "enable",
+        .op_version = {GD_OP_VERSION_3_7_0},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"disperse"},
+    },
+    {.key = {"iam-self-heal-daemon"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = "This option differentiates if the disperse "
+                    "translator is running as part of self-heal-daemon "
+                    "or not."},
+    {.key = {"eager-lock"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .op_version = {GD_OP_VERSION_3_7_10},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"disperse"},
+     .description = "Enable/Disable eager lock for regular files on a "
+                    "disperse volume. If a fop takes a lock and completes "
+                    "its operation, it waits for next 1 second before "
+                    "releasing the lock, to see if the lock can be reused "
+                    "for next fop from the same client. If ec finds any lock "
+                    "contention within 1 second it releases the lock "
+                    "immediately before time expires. This improves the "
+                    "performance of file operations. However, as it takes "
+                    "lock on first brick, for few operations like read, "
+                    "discovery of lock contention might take long time and "
+                    "can actually degrade the performance. If eager lock is "
+                    "disabled, lock will be released as soon as fop "
+                    "completes."},
+    {.key = {"other-eager-lock"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .op_version = {GD_OP_VERSION_3_13_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"disperse"},
+     .description = "It's equivalent to the eager-lock option but for non "
+                    "regular files."},
+    {.key = {"eager-lock-timeout"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 60,
+     .default_value = "1",
+     .op_version = {GD_OP_VERSION_4_0_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"disperse", "locks", "timeout"},
+     .description = "Maximum time (in seconds) that a lock on an inode is "
+                    "kept held if no new operations on the inode are "
+                    "received."},
+    {.key = {"other-eager-lock-timeout"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 60,
+     .default_value = "1",
+     .op_version = {GD_OP_VERSION_4_0_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"disperse", "locks", "timeout"},
+     .description = "It's equivalent to eager-lock-timeout option but for "
+                    "non regular files."},
+    {
+        .key = {"background-heals"},
+        .type = GF_OPTION_TYPE_INT,
+        .min = 0, /*Disabling background heals*/
+        .max = 256,
+        .default_value = "8",
+        .op_version = {GD_OP_VERSION_3_7_3},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+        .tags = {"disperse"},
+        .description = "This option can be used to control number of parallel"
+                       " heals",
+    },
+    {
+        .key = {"heal-wait-qlength"},
+        .type = GF_OPTION_TYPE_INT,
+        .min = 0,
+        .max =
+            65536, /*Around 100MB as of now with sizeof(ec_fop_data_t) at 1800*/
+        .default_value = "128",
+        .op_version = {GD_OP_VERSION_3_7_3},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+        .tags = {"disperse"},
+        .description = "This option can be used to control number of heals"
+                       " that can wait",
+    },
+    {.key = {"heal-timeout"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 60,
+     .max = INT_MAX,
+     .default_value = "600",
+     .op_version = {GD_OP_VERSION_3_7_3},
+     .flags = OPT_FLAG_SETTABLE,
+     .tags = {"disperse"},
+     .description = "time interval for checking the need to self-heal "
+                    "in self-heal-daemon"},
+    {
+        .key = {"read-policy"},
+        .type = GF_OPTION_TYPE_STR,
+        .value = {"round-robin", "gfid-hash"},
+        .default_value = "gfid-hash",
+        .op_version = {GD_OP_VERSION_3_7_6},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+        .tags = {"disperse"},
+        .description =
+            "inode-read fops happen only on 'k' number of bricks in"
+            " n=k+m disperse subvolume. 'round-robin' selects the read"
+            " subvolume using round-robin algo. 'gfid-hash' selects read"
+            " subvolume based on hash of the gfid of that file/directory.",
+    },
+    {.key = {"shd-max-threads"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 64,
+     .default_value = "1",
+     .op_version = {GD_OP_VERSION_3_9_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"disperse"},
+     .description = "Maximum number of parallel heals SHD can do per local "
+                    "brick.  This can substantially lower heal times, "
+                    "but can also crush your bricks if you don't have "
+                    "the storage hardware to support this."},
+    {.key = {"shd-wait-qlength"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 65536,
+     .default_value = "1024",
+     .op_version = {GD_OP_VERSION_3_9_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"disperse"},
+     .description = "This option can be used to control number of heals"
+                    " that can wait in SHD per subvolume"},
+    {.key = {"cpu-extensions"},
+     .type = GF_OPTION_TYPE_STR,
+     .value = {"none", "auto", "x64", "sse", "avx"},
+     .default_value = "auto",
+     .op_version = {GD_OP_VERSION_3_9_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"disperse"},
+     .description = "force the cpu extensions to be used to accelerate the "
+                    "galois field computations."},
+    {.key = {"self-heal-window-size"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 1,
+     .max = 1024,
+     .default_value = "1",
+     .op_version = {GD_OP_VERSION_3_11_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"disperse"},
+     .description = "Maximum number blocks(128KB) per file for which "
+                    "self-heal process would be applied simultaneously."},
+    {.key = {"optimistic-change-log"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .op_version = {GD_OP_VERSION_3_10_1},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT,
+     .tags = {"disperse"},
+     .description = "Set/Unset dirty flag for every update fop at the start"
+                    "of the fop. If OFF, this option impacts performance of"
+                    "entry  operations or metadata operations as it will"
+                    "set dirty flag at the start and unset it at the end of"
+                    "ALL update fop. If ON and all the bricks are good,"
+                    "dirty flag will be set at the start only for file fops"
+                    "For metadata and entry fops dirty flag will not be set"
+                    "at the start, if all the bricks are good. This does"
+                    "not impact performance for metadata operations and"
+                    "entry operation but has a very small window to miss"
+                    "marking entry as dirty in case it is required to be"
+                    "healed"},
+    {.key = {"parallel-writes"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .description = "This controls if writes can be wound in parallel as long"
+                    "as it doesn't modify same stripes"},
+    {.key = {"stripe-cache"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 0, /*Disabling stripe_cache*/
+     .max = EC_STRIPE_CACHE_MAX_SIZE,
+     .default_value = "4",
+     .description = "This option will keep the last stripe of write fop"
+                    "in memory. If next write falls in this stripe, we need"
+                    "not to read it again from backend and we can save READ"
+                    "fop going over the network. This will improve performance,"
+                    "specially for sequential writes. However, this will also"
+                    "lead to extra memory consumption, maximum "
+                    "(cache size * stripe size) Bytes per open file."},
+    {
+        .key = {"quorum-count"},
+        .type = GF_OPTION_TYPE_INT,
+        .default_value = "0",
+        .description =
+            "This option can be used to define how many successes on"
+            "the bricks constitute a success to the application. This"
+            " count should be in the range"
+            "[disperse-data-count,  disperse-count] (inclusive)",
+    },
+    {
+        .key = {"ec-read-mask"},
+        .type = GF_OPTION_TYPE_STR,
+        .default_value = NULL,
+        .description = "This option can be used to choose which bricks can be"
+                       " used for reading data/metadata of a file/directory",
+    },
+    {
+        .key = {NULL},
+    },
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1},
+    .dumpops = &dumpops,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "disperse",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/cluster/ec/src/ec.h b/xlators/cluster/ec/src/ec.h
new file mode 100644
index 00000000000..6f6de6d5981
--- /dev/null
+++ b/xlators/cluster/ec/src/ec.h
@@ -0,0 +1,34 @@
+/*
+  Copyright (c) 2012-2014 DataLab, s.l. <http://www.datalab.es>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __EC_H__
+#define __EC_H__
+
+#include "ec-method.h"
+
+#define EC_XATTR_PREFIX "trusted.ec."
+#define EC_XATTR_CONFIG EC_XATTR_PREFIX "config"
+#define EC_XATTR_SIZE EC_XATTR_PREFIX "size"
+#define EC_XATTR_VERSION EC_XATTR_PREFIX "version"
+#define EC_XATTR_HEAL EC_XATTR_PREFIX "heal"
+#define EC_XATTR_HEAL_NEW EC_XATTR_PREFIX "heal-new"
+#define EC_XATTR_DIRTY EC_XATTR_PREFIX "dirty"
+#define EC_STRIPE_CACHE_MAX_SIZE 10
+#define EC_VERSION_SIZE 2
+#define EC_SHD_INODE_LRU_LIMIT 10
+
+#define EC_MAX_FRAGMENTS EC_METHOD_MAX_FRAGMENTS
+/* The maximum number of nodes is derived from the maximum allowed fragments
+ * using the rule that redundancy cannot be equal or greater than the number
+ * of fragments.
+ */
+#define EC_MAX_NODES min(EC_MAX_FRAGMENTS * 2 - 1, EC_METHOD_MAX_NODES)
+
+#endif /* __EC_H__ */
diff --git a/xlators/cluster/ha/src/Makefile.am b/xlators/cluster/ha/src/Makefile.am
deleted file mode 100644
index 5f78a296533..00000000000
--- a/xlators/cluster/ha/src/Makefile.am
+++ /dev/null
@@ -1,15 +0,0 @@
-xlator_LTLIBRARIES = ha.la
-xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/testing/cluster
-
-ha_la_LDFLAGS = -module -avoidversion 
-
-ha_la_SOURCES = ha-helpers.c ha.c
-ha_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
-
-noinst_HEADERS = ha.h
-
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS) \
-	    -I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
-
-CLEANFILES = 
-
diff --git a/xlators/cluster/ha/src/ha-helpers.c b/xlators/cluster/ha/src/ha-helpers.c
deleted file mode 100644
index 759501f9f84..00000000000
--- a/xlators/cluster/ha/src/ha-helpers.c
+++ /dev/null
@@ -1,204 +0,0 @@
-/*
-  Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-#include "xlator.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "dict.h"
-#include "compat-errno.h"
-#include "ha.h"
-
-#define HA_TRANSPORT_NOTCONN(_ret, _errno, _fd) \
-	((_ret == -1) && (_fd ? (_errno == EBADFD):(_errno == ENOTCONN)))
-
-int ha_alloc_init_fd (call_frame_t *frame, fd_t *fd)
-{
-	ha_local_t *local = NULL;
-	int i = -1;
-	ha_private_t *pvt = NULL;
-	int child_count = 0;
-	int ret = -1;
-	hafd_t *hafdp = NULL;
-	xlator_t *this = NULL;
-	uint64_t tmp_hafdp = 0;
-
-	this = frame->this;
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-
-	if (local == NULL) {
-		ret = fd_ctx_get (fd, this, &tmp_hafdp);
-		if (ret < 0) {
-			goto out;
-		}
-		hafdp = (hafd_t *)(long)tmp_hafdp;
-                local = frame->local = GF_CALLOC (1, sizeof (*local), 
-                                                  gf_ha_mt_ha_local_t);
-		if (local == NULL) {
-			ret = -ENOMEM;
-			goto out;
-		}
-                local->state = GF_CALLOC (1, child_count, 
-                                          gf_ha_mt_child_count);
-		if (local->state == NULL) {
-			ret = -ENOMEM;
-			goto out;
-		}
-
-		/* take care of the preferred subvolume */
-		if (pvt->pref_subvol == -1)
-			local->active = hafdp->active;
-		else
-			local->active = pvt->pref_subvol;
-
-		LOCK (&hafdp->lock);
-		memcpy (local->state, hafdp->fdstate, child_count);
-		UNLOCK (&hafdp->lock);
-
-		/* in case the preferred subvolume is down */
-		if ((local->active != -1) && (local->state[local->active] == 0))
-			local->active = -1;
-
-		for (i = 0; i < child_count; i++) {
-			if (local->state[i]) {
-				if (local->active == -1)
-					local->active = i;
-				local->tries++;
-			}
-		}
-		if (local->active == -1) {
-			ret = -ENOTCONN;
-			goto out;
-		}
-		local->fd = fd_ref (fd);
-	}
-	ret = 0;
-out:
-	return ret;
-}
-
-int ha_handle_cbk (call_frame_t *frame, void *cookie, int op_ret, int op_errno) 
-{
-	xlator_t *xl = NULL;
-	ha_private_t *pvt = NULL;
-	xlator_t **children = NULL;
-	int prev_child = -1;
-	hafd_t *hafdp = NULL;
-	int ret = -1;
-	call_stub_t *stub = NULL;
-	ha_local_t *local = NULL;
-	uint64_t tmp_hafdp = 0;
-
-	xl = frame->this;
-	pvt = xl->private;
-	children = pvt->children;
-	prev_child = (long) cookie;
-	local = frame->local;
-
-	if (op_ret == -1) {
-		gf_log (xl->name, GF_LOG_ERROR ,"(child=%s) (op_ret=%d op_errno=%s)",
-			children[prev_child]->name, op_ret, strerror (op_errno));
-	}
-
-	if (HA_TRANSPORT_NOTCONN (op_ret, op_errno, (local->fd))) {
-		ret = 0;
-		if (local->fd) {
-			ret = fd_ctx_get (local->fd, xl, &tmp_hafdp);
-		}
-		hafdp = (hafd_t *)(long)tmp_hafdp;		
-		if (ret == 0) {
-			if (local->fd) {
-				LOCK(&hafdp->lock);
-				hafdp->fdstate[prev_child] = 0;
-				UNLOCK(&hafdp->lock);
-			}
-			local->tries--;
-			if (local->tries != 0) {
-				while (1) {
-					local->active = (local->active + 1) % pvt->child_count;
-					if (local->state[local->active])
-						break;
-				}
-				stub = local->stub;
-				local->stub = NULL;
-				call_resume (stub);
-				return -1;
-			}
-		}
-	}
-	if (local->stub) {
-		call_stub_destroy (local->stub);
-                local->stub = NULL;
-        }
-
-	if (local->fd) {
-		GF_FREE (local->state);
-                local->state = NULL;
-
-		fd_unref (local->fd);
-                local->fd = NULL;
-	}
-	return 0;
-}
-
-int ha_alloc_init_inode (call_frame_t *frame, inode_t *inode)
-{
-	int i = -1;
-	ha_private_t *pvt = NULL;
-	xlator_t *xl = NULL;
-	int ret = -1;
-	ha_local_t *local = NULL;
-	uint64_t tmp_state = 0;
-
-	xl = frame->this;
-	pvt = xl->private;
-	local = frame->local;
-
-	if (local == NULL) {
-                local = frame->local = GF_CALLOC (1, sizeof (*local), 
-                                                  gf_ha_mt_ha_local_t);
-		if (local == NULL) {
-			ret = -ENOMEM;
-			goto out;
-		}
-		local->active = pvt->pref_subvol;
-		ret = inode_ctx_get (inode, xl, &tmp_state);
-		if (ret < 0) {
-			goto out;
-		}
-		local->state = (char *)(long)tmp_state;
-		if (local->active != -1 && local->state[local->active] == 0)
-			local->active = -1;
-		for (i = 0; i < pvt->child_count; i++) {
-			if (local->state[i]) {
-				if (local->active == -1)
-					local->active = i;
-				local->tries++;
-			}
-		}
-		if (local->active == -1) {
-			ret = -ENOTCONN;
-			goto out;
-		}
-	}
-	ret = 0;
-out:
-	return ret;
-}
diff --git a/xlators/cluster/ha/src/ha-mem-types.h b/xlators/cluster/ha/src/ha-mem-types.h
deleted file mode 100644
index 4d3b5747acc..00000000000
--- a/xlators/cluster/ha/src/ha-mem-types.h
+++ /dev/null
@@ -1,37 +0,0 @@
-
-/*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-
-#ifndef __HA_MEM_TYPES_H__
-#define __HA_MEM_TYPES_H__
-
-#include "mem-types.h"
-
-enum gf_ha_mem_types_ {
-        gf_ha_mt_ha_local_t = gf_common_mt_end + 1,
-        gf_ha_mt_hafd_t,
-        gf_ha_mt_char,
-        gf_ha_mt_child_count,
-        gf_ha_mt_xlator_t,
-        gf_ha_mt_ha_private_t,
-        gf_ha_mt_end
-};
-#endif
-
diff --git a/xlators/cluster/ha/src/ha.c b/xlators/cluster/ha/src/ha.c
deleted file mode 100644
index 7310b5e951e..00000000000
--- a/xlators/cluster/ha/src/ha.c
+++ /dev/null
@@ -1,4023 +0,0 @@
-/*
-  Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-/* generate errors randomly, code is simple now, better alogorithm
- * can be written to decide what error to be returned and when
- */
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "xlator.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "dict.h"
-#include "compat-errno.h"
-#include "ha.h"
-
-/*
- * TODO:
- * - dbench fails if ha over server side afr
- * - lock calls - lock on all subvols.
- * - support preferred-subvolume option. code already there.
- * - do not alloc the call-stub in case only one subvol is up.
- */
-
-void
-ha_local_wipe (ha_local_t *local)
-{
-        if (local->stub) {
-                call_stub_destroy (local->stub);
-                local->stub = NULL;
-        }
-
-        if (local->state) {
-                GF_FREE (local->state);
-                local->state = NULL;
-        }
-
-        if (local->dict) {
-                dict_unref (local->dict);
-                local->dict = NULL;
-        }
-
-        loc_wipe (&local->loc);
-
-        if (local->fd) {
-                fd_unref (local->fd);
-                local->fd = NULL;
-        }
-
-        if (local->inode) {
-                inode_unref (local->inode);
-                local->inode = NULL;
-        }
-
-        GF_FREE (local);
-        return;
-}
-
-
-int
-ha_forget (xlator_t *this,
-	   inode_t *inode)
-{
-	uint64_t stateino = 0;
-	char *state = NULL;
-	if (!inode_ctx_del (inode, this, &stateino)) {
-		state =  ((char *)(long)stateino);
-		GF_FREE (state);
-	}
-
-	return 0;
-
-}
-
-int32_t 
-ha_lookup_cbk (call_frame_t *frame,
-	       void *cookie,
-	       xlator_t *this,
-	       int32_t op_ret,
-	       int32_t op_errno,
-	       inode_t *inode,
-	       struct iatt *buf,
-               dict_t *dict,
-               struct iatt *postparent)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	int child_count = 0, i = 0, callcnt = 0;
-	char *state = NULL;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	uint64_t tmp_state = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	for (i = 0; i < child_count; i++) {
-		if (pvt->children[i] == prev_frame->this)
-			break;
-	}
-	if ((op_ret == -1) && (op_errno != ENOENT)) {
-		gf_log (this->name, GF_LOG_ERROR, "(child=%s) (op_ret=%d op_errno=%s)", 
-			  children[i]->name, op_ret, strerror (op_errno));
-	}
-	inode_ctx_get (local->inode, this, &tmp_state);
-	state = (char *)(long)tmp_state;
-
-	LOCK (&frame->lock);
-	if (local->revalidate == 1) {
-		if ((!op_ret) != state[i]) {
-			local->revalidate_error = 1;
-			gf_log (this->name, GF_LOG_DEBUG, "revalidate error on %s", 
-				pvt->children[i]->name);
-		}
-	} else {
-		if (op_ret == 0) {
-			state[i] = 1;
-		}
-	}
-	if (local->op_ret == -1 && op_ret == 0) {
-		local->op_ret = 0;
-		local->buf = *buf;
-                local->postparent = *postparent;
-		if (dict)
-			local->dict = dict_ref (dict);
-	}
-	if (op_ret == -1 && op_ret != ENOTCONN)
-		local->op_errno = op_errno;
-	callcnt = --local->call_count;
-	UNLOCK (&frame->lock);
-
-	if (callcnt == 0) {
-		dict_t *ctx = local->dict;
-		inode_t *inode = local->inode;
-		if (local->revalidate_error == 1) {
-			local->op_ret = -1;
-			local->op_errno = EIO;
-			gf_log (this->name, GF_LOG_DEBUG, "revalidate error, returning EIO");
-		}
-		STACK_UNWIND (frame,
-			      local->op_ret,
-			      local->op_errno,
-			      inode,
-			      &local->buf,
-			      ctx,
-                              &local->postparent);
-		if (inode)
-			inode_unref (inode);
-		if (ctx)
-			dict_unref (ctx);
-	}
-	return 0;
-}
-
-int32_t
-ha_lookup (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *loc,
-	   dict_t *xattr_req)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	int child_count = 0, i = 0;
-	char *state = NULL;
-	xlator_t **children = NULL;
-	int ret = -1;
-        int32_t op_errno = EINVAL;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	children = pvt->children;
-
-	frame->local = local = GF_CALLOC (1, sizeof (*local), 
-                                          gf_ha_mt_ha_local_t);
-        if (!local) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto unwind;
-        }
-
-	child_count = pvt->child_count;
-	local->inode = inode_ref (loc->inode);
-
-	ret = inode_ctx_get (loc->inode, this, NULL);
-	if (ret) {
-		state = GF_CALLOC (1, child_count, gf_ha_mt_child_count);
-                if (state == NULL) {
-                        gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                        op_errno = ENOMEM;
-                        goto unwind;
-                }
-
-		inode_ctx_put (loc->inode, this, (uint64_t)(long)state);
-	} else
-		local->revalidate = 1;
-
-	local->op_ret = -1;
-	local->op_errno = ENOTCONN;
-	local->call_count = child_count;
-
-	for (i = 0; i < child_count; i++) {
-		STACK_WIND (frame,
-			    ha_lookup_cbk,
-			    children[i],
-			    children[i]->fops->lookup,
-			    loc,
-			    xattr_req);
-	}
-	return 0;
-
-unwind:
-        local = frame->local;
-        frame->local = NULL;
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL, NULL, NULL);
-
-        ha_local_wipe (local);
-        return 0;
-}
-
- int32_t
-ha_stat_cbk (call_frame_t *frame,
-	     void *cookie,
-	     xlator_t *this,
-	     int32_t op_ret,
-	     int32_t op_errno,
-	     struct iatt *buf)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-			      buf);
-	}
-	return 0;
-}
-
-int32_t
-ha_stat (call_frame_t *frame,
-	 xlator_t *this,
-	 loc_t *loc)
-{
-	ha_local_t *local = NULL;
-	int op_errno = ENOTCONN;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_stat_stub (frame, ha_stat, loc);
-
-	STACK_WIND_COOKIE (frame,
-			   ha_stat_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->stat,
-			   loc);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno, NULL);
-	return 0;	
-}
-
-int32_t
-ha_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                int32_t op_ret, int32_t op_errno, struct iatt *statpre,
-                struct iatt *statpost)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame, op_ret, op_errno, statpre, statpost);
-	}
-	return 0;
-}
-
-
-int32_t
-ha_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *stbuf,
-            int32_t valid)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_setattr_stub (frame, ha_setattr, loc, stbuf, valid);
-
-	STACK_WIND_COOKIE (frame,
-			   ha_setattr_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->setattr,
-			   loc, stbuf, valid);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-	return 0;
-}
-
-
-int32_t
-ha_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
-             int32_t valid)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_fsetattr_stub (frame, ha_fsetattr, fd, stbuf, valid);
-
-	STACK_WIND_COOKIE (frame,	      
-			   ha_setattr_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->fsetattr,
-			   fd, stbuf, valid);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-	return 0;
-}
-
-
-int32_t
-ha_truncate_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 struct iatt *prebuf,
-                 struct iatt *postbuf)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-			      prebuf,
-                              postbuf);
-	}
-	return 0;
-}
-
-int32_t
-ha_truncate (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     off_t offset)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_truncate_stub (frame, ha_truncate, loc, offset);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_truncate_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->truncate,
-			   loc,
-			   offset);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-	return 0;
-}
-
- int32_t
-ha_ftruncate_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  struct iatt *prebuf,
-                  struct iatt *postbuf)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-			      prebuf, postbuf);
-	}
-	return 0;
-}
-
-int32_t
-ha_ftruncate (call_frame_t *frame,
-	      xlator_t *this,
-	      fd_t *fd,
-	      off_t offset)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_ftruncate_stub (frame, ha_ftruncate, fd, offset);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_ftruncate_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->ftruncate,
-			   fd,
-			   offset);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
-int32_t
-ha_access_cbk (call_frame_t *frame,
-	       void *cookie,
-	       xlator_t *this,
-	       int32_t op_ret,
-	       int32_t op_errno)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno);
-	}
-	return 0;
-}
-
-int32_t
-ha_access (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *loc,
-	   int32_t mask)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_access_stub (frame, ha_access, loc, mask);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_access_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->access,
-			   loc,
-			   mask);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
-
- int32_t
-ha_readlink_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 const char *path,
-                 struct iatt *sbuf)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-			      path,
-                              sbuf);
-	}
-	return 0;
-}
-
-int32_t
-ha_readlink (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     size_t size)
-{
-	ha_local_t *local = frame->local;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_readlink_stub (frame, ha_readlink, loc, size);
-
-	STACK_WIND_COOKIE (frame,
-			   ha_readlink_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->readlink,
-			   loc,
-			   size);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-	return 0;
-}
-
-int
-ha_mknod_lookup_cbk (call_frame_t *frame,
-		     void *cookie,
-		     xlator_t *this,
-		     int32_t op_ret,
-		     int32_t op_errno,
-		     inode_t *inode,
-		     struct iatt *buf,
-                     dict_t *dict,
-                     struct iatt *postparent)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	char *stateino = NULL;
-	int child_count = 0, i = 0, cnt = 0, ret = 0;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	uint64_t tmp_stateino = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	for (i = 0; i < child_count; i++)
-		if (prev_frame->this == children[i])
-			break;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"(path=%s) (op_ret=%d op_errno=%d)", 
-			local->stub->args.mknod.loc.path, op_ret, op_errno);
-	}
-	ret = inode_ctx_get (local->stub->args.mknod.loc.inode, 
-			     this, &tmp_stateino);
-	stateino = (char *)(long)tmp_stateino;
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"unwind(-1), inode_ctx_get() error");
-		/* It is difficult to handle this error at this stage
-		 * as we still expect more cbks, we can't return as
-		 * of now
-		 */
-	} else if (op_ret == 0) {
-		stateino[i] = 1;
-	}
-	LOCK (&frame->lock);
-	cnt = --local->call_count;
-	UNLOCK (&frame->lock);
-
-	if (cnt == 0) {
-		call_stub_t *stub = local->stub;
-		GF_FREE (local->state);
-		STACK_UNWIND (frame,
-			      local->op_ret,
-			      local->op_errno,
-			      local->stub->args.mknod.loc.inode,
-			      &local->buf, &local->preparent,
-                              &local->postparent);
-		call_stub_destroy (stub);
-	}
-	return 0;
-}
-
-int32_t
-ha_mknod_cbk (call_frame_t *frame,
-	      void *cookie,
-	      xlator_t *this,
-	      int32_t op_ret,
-	      int32_t op_errno,
-	      inode_t *inode,
-              struct iatt *buf,
-              struct iatt *preparent,
-              struct iatt *postparent)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	char *stateino = NULL;
-	int child_count = 0, i = 0, cnt = 0, ret = 0;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	uint64_t tmp_stateino = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	for (i = 0; i < child_count; i++)
-		if (prev_frame->this == children[i])
-			break;
-
-	if (op_ret == -1) {
-		local->op_errno = op_errno;
-		gf_log (this->name, GF_LOG_ERROR, "(path=%s) (op_ret=%d op_errno=%d)", local->stub->args.mknod.loc.path, op_ret, op_errno);
-	}
-
-	ret = inode_ctx_get (local->stub->args.mknod.loc.inode, 
-			     this, &tmp_stateino);
-	stateino = (char *)(long)tmp_stateino;
-
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR, "inode_ctx_get() error");
-		/* FIXME: handle the case */
-	}
-	if (op_ret == 0) {
-		stateino[i] = 1;
-		local->op_ret = 0;
-		local->first_success = 1;
-		local->buf = *buf;
-                local->preparent = *preparent;
-                local->postparent = *postparent;
-	}
-	cnt = --local->call_count;
-	for (i = local->active + 1; i < child_count; i++) {
-		if (local->state[i])
-			break;
-	}
-
-	if (cnt == 0 || i == child_count) {
-		call_stub_t *stub = local->stub;
-		GF_FREE (local->state);
-		stub = local->stub;
-		STACK_UNWIND (frame, local->op_ret, local->op_errno,
-                              local->stub->args.mknod.loc.inode, &local->buf,
-                              &local->preparent, &local->postparent);
-		call_stub_destroy (stub);
-		return 0;
-	}
-
-	local->active = i;
-
-	if (local->first_success == 0) {
-		STACK_WIND (frame,
-			    ha_mknod_cbk,
-			    children[i],
-			    children[i]->fops->mknod,
-			    &local->stub->args.mknod.loc,
-			    local->stub->args.mknod.mode,
-			    local->stub->args.mknod.rdev);
-		return 0;
-	}
-	cnt = local->call_count;
-
-	for (; i < child_count; i++) {
-		if (local->state[i]) {
-			STACK_WIND (frame,
-				    ha_mknod_lookup_cbk,
-				    children[i],
-				    children[i]->fops->lookup,
-				    &local->stub->args.mknod.loc,
-				    0);
-			if (--cnt == 0)
-				break;
-		}
-	}
-	return 0;
-}
-
-int32_t
-ha_mknod (call_frame_t *frame,
-	  xlator_t *this,
-	  loc_t *loc,
-	  mode_t mode,
-	  dev_t rdev)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	int child_count = 0, i = 0;
-	char *stateino = NULL;
-        int32_t op_errno = EINVAL;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-
-	frame->local = local = GF_CALLOC (1, sizeof (*local), 
-                                          gf_ha_mt_ha_local_t);
-        if (!local) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	local->stub = fop_mknod_stub (frame, ha_mknod, loc, mode, rdev);
-        if (!local->stub) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	local->op_ret = -1;
-	local->op_errno = ENOTCONN;
-	local->state = GF_CALLOC (1, child_count, gf_ha_mt_char);
-        if (!local->state) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	memcpy (local->state, pvt->state, child_count);
-	local->active = -1;
-
-	stateino = GF_CALLOC (1, child_count, gf_ha_mt_char);
-        if (!stateino) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	inode_ctx_put (loc->inode, this, (uint64_t)(long)stateino);
-
-	for (i = 0; i < child_count; i++) {
-		if (local->state[i]) {
-			local->call_count++;
-			if (local->active == -1) 
-				local->active = i;
-		}
-	}
-
-	STACK_WIND (frame,
-		    ha_mknod_cbk,
-		    HA_ACTIVE_CHILD(this, local),
-		    HA_ACTIVE_CHILD(this, local)->fops->mknod,
-		    loc, mode, rdev);
-	return 0;
-
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL, NULL, NULL);
-        ha_local_wipe (local);
-        return 0;
-}
-
-
-int
-ha_mkdir_lookup_cbk (call_frame_t *frame,
-		     void *cookie,
-		     xlator_t *this,
-		     int32_t op_ret,
-		     int32_t op_errno,
-		     inode_t *inode,
-		     struct iatt *buf,
-                     dict_t *dict,
-                     struct iatt *postparent)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	char *stateino = NULL;
-	int child_count = 0, i = 0, cnt = 0;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	uint64_t tmp_stateino = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	for (i = 0; i < child_count; i++)
-		if (prev_frame->this == children[i])
-			break;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR, "(path=%s) (op_ret=%d op_errno=%d)", local->stub->args.mkdir.loc.path, op_ret, op_errno);
-	}
-	inode_ctx_get (local->stub->args.mkdir.loc.inode, 
-		       this, &tmp_stateino);  
-	stateino = (char *)(long)tmp_stateino;
-
-	if (op_ret == 0)
-		stateino[i] = 1;
-
-	LOCK (&frame->lock);
-	cnt = --local->call_count;
-	UNLOCK (&frame->lock);
-
-	if (cnt == 0) {
-		call_stub_t *stub = local->stub;
-		GF_FREE (local->state);
-		STACK_UNWIND (frame,
-			      local->op_ret,
-			      local->op_errno,
-			      local->stub->args.mkdir.loc.inode, &local->buf,
-			      &local->preparent, &local->postparent);
-		call_stub_destroy (stub);
-	}
-	return 0;
-}
-
-int32_t
-ha_mkdir_cbk (call_frame_t *frame,
-	      void *cookie,
-	      xlator_t *this,
-	      int32_t op_ret,
-	      int32_t op_errno,
-	      inode_t *inode,
-              struct iatt *buf,
-              struct iatt *preparent,
-              struct iatt *postparent)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	char *stateino = NULL;
-	int child_count = 0, i = 0, cnt = 0;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	uint64_t tmp_stateino = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	children = pvt->children;
-	
-	for (i = 0; i < child_count; i++)
-		if (prev_frame->this == children[i])
-			break;
-
-	if (op_ret == -1) {
-		local->op_errno = op_errno;
-		gf_log (this->name, GF_LOG_ERROR, "(path=%s) (op_ret=%d op_errno=%d)", local->stub->args.mkdir.loc.path, op_ret, op_errno);
-	}
-
-	inode_ctx_get (local->stub->args.mkdir.loc.inode, 
-		       this, &tmp_stateino);
-	stateino = (char *)(long)tmp_stateino;
-
-	if (op_ret == 0) {
-		stateino[i] = 1;
-		local->op_ret = 0;
-		local->first_success = 1;
-		local->buf = *buf;
-                local->preparent = *preparent;
-                local->postparent = *postparent;
-	}
-	cnt = --local->call_count;
-	for (i = local->active + 1; i < child_count; i++) {
-		if (local->state[i])
-			break;
-	}
-
-	if (cnt == 0 || i == child_count) {
-		call_stub_t *stub = local->stub;
-		GF_FREE (local->state);
-		stub = local->stub;
-		STACK_UNWIND (frame, local->op_ret, local->op_errno,
-                              local->stub->args.mkdir.loc.inode, &local->buf,
-                              &local->preparent, &local->postparent);
-		call_stub_destroy (stub);
-		return 0;
-	}
-
-	local->active = i;
-
-	if (local->first_success == 0) {
-		STACK_WIND (frame,
-			    ha_mkdir_cbk,
-			    children[i],
-			    children[i]->fops->mkdir,
-			    &local->stub->args.mkdir.loc,
-			    local->stub->args.mkdir.mode);
-		return 0;
-	}
-	cnt = local->call_count;
-
-	for (; i < child_count; i++) {
-		if (local->state[i]) {
-			STACK_WIND (frame,
-				    ha_mkdir_lookup_cbk,
-				    children[i],
-				    children[i]->fops->lookup,
-				    &local->stub->args.mkdir.loc,
-				    0);
-			if (--cnt == 0)
-				break;
-		}
-	}
-	return 0;
-}
-
-int32_t
-ha_mkdir (call_frame_t *frame,
-	  xlator_t *this,
-	  loc_t *loc,
-	  mode_t mode)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	int child_count = 0, i = 0;
-	char *stateino = NULL;
-        int32_t op_errno = EINVAL;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-
-	frame->local = local = GF_CALLOC (1, sizeof (*local), 
-                                          gf_ha_mt_ha_local_t);
-        if (!frame->local) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	local->stub = fop_mkdir_stub (frame, ha_mkdir, loc, mode);
-        if (!local->stub) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	local->op_ret = -1;
-	local->op_errno = ENOTCONN;
-	local->state = GF_CALLOC (1, child_count, gf_ha_mt_char);
-        if (!local->state) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	memcpy (local->state, pvt->state, child_count);
-	local->active = -1;
-
-	stateino = GF_CALLOC (1, child_count, gf_ha_mt_char);
-        if (!stateino) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	inode_ctx_put (loc->inode, this, (uint64_t)(long)stateino);
-	for (i = 0; i < child_count; i++) {
-		if (local->state[i]) {
-			local->call_count++;
-			if (local->active == -1)
-				local->active = i;
-		}
-	}
-
-	STACK_WIND (frame,
-		    ha_mkdir_cbk,
-		    HA_ACTIVE_CHILD(this, local),
-		    HA_ACTIVE_CHILD(this, local)->fops->mkdir,
-		    loc, mode);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL, NULL, NULL);
-        ha_local_wipe (local);
-        return 0;
-}
-
- int32_t
-ha_unlink_cbk (call_frame_t *frame,
-	       void *cookie,
-	       xlator_t *this,
-	       int32_t op_ret,
-	       int32_t op_errno,
-               struct iatt *preparent,
-               struct iatt *postparent)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-	if (ret == 0) {
-		STACK_UNWIND (frame, op_ret, op_errno, preparent, postparent);
-	}
-	return 0;
-}
-
-int32_t
-ha_unlink (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *loc)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_unlink_stub (frame, ha_unlink, loc);
-        if (!local->stub) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_unlink_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->unlink,
-			   loc);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-	return 0;
-}
-
- int32_t
-ha_rmdir_cbk (call_frame_t *frame,
-	      void *cookie,
-	      xlator_t *this,
-	      int32_t op_ret,
-	      int32_t op_errno,
-              struct iatt *preparent,
-              struct iatt *postparent)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-                              preparent,
-                              postparent);
-	}
-	return 0;
-}
-
-int32_t
-ha_rmdir (call_frame_t *frame,
-	  xlator_t *this,
-	  loc_t *loc)
-{
-	ha_local_t *local = frame->local;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_rmdir_stub (frame, ha_rmdir, loc);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_rmdir_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->rmdir,
-			   loc);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-	return 0;
-}
-
-
-int
-ha_symlink_lookup_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno,
-		       inode_t *inode,
-		       struct iatt *buf,
-                       dict_t *dict,
-                       struct iatt *postparent)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	char *stateino = NULL;
-	int child_count = 0, i = 0, cnt = 0;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	uint64_t tmp_stateino = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	for (i = 0; i < child_count; i++)
-		if (prev_frame->this == children[i])
-			break;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR, "(path=%s) (op_ret=%d op_errno=%d)", local->stub->args.symlink.loc.path, op_ret, op_errno);
-	}
-	inode_ctx_get (local->stub->args.symlink.loc.inode,
-		       this, &tmp_stateino);  
-	stateino = (char *)(long)tmp_stateino;
-
-	if (op_ret == 0)
-		stateino[i] = 1;
-
-	LOCK (&frame->lock);
-	cnt = --local->call_count;
-	UNLOCK (&frame->lock);
-
-	if (cnt == 0) {
-		call_stub_t *stub = local->stub;
-		GF_FREE (local->state);
-		STACK_UNWIND (frame,
-			      local->op_ret,
-			      local->op_errno,
-			      local->stub->args.symlink.loc.inode, &local->buf,
-                              &local->preparent, &local->postparent);
-		call_stub_destroy (stub);
-	}
-	return 0;
-}
-
-int32_t
-ha_symlink_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		inode_t *inode,
-                struct iatt *buf,
-                struct iatt *preparent,
-                struct iatt *postparent)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	char *stateino = NULL;
-	int child_count = 0, i = 0, cnt = 0;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	uint64_t tmp_stateino = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	for (i = 0; i < child_count; i++)
-		if (prev_frame->this == children[i])
-			break;
-
-	if (op_ret == -1) {
-		local->op_errno = op_errno;
-		gf_log (this->name, GF_LOG_ERROR, "(path=%s) (op_ret=%d op_errno=%d)", local->stub->args.symlink.loc.path, op_ret, op_errno);
-	}
-	inode_ctx_get (local->stub->args.symlink.loc.inode, 
-		       this, &tmp_stateino);
-	stateino = (char *)(long)tmp_stateino;
-
-	if (op_ret == 0) {
-		stateino[i] = 1;
-		local->op_ret = 0;
-		local->first_success = 1;
-		local->buf = *buf;
-                local->preparent = *preparent;
-                local->postparent = *postparent;
-	}
-	cnt = --local->call_count;
-	for (i = local->active + 1; i < child_count; i++) {
-		if (local->state[i])
-			break;
-	}
-
-	if (cnt == 0 || i == child_count) {
-		call_stub_t *stub = local->stub;
-		GF_FREE (local->state);
-		stub = local->stub;
-		STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-			      local->stub->args.symlink.loc.inode, &local->buf,
-                              &local->preparent, &local->postparent);
-		call_stub_destroy (stub);
-		return 0;
-	}
-
-	local->active = i;
-
-	if (local->first_success == 0) {
-		STACK_WIND (frame,
-			    ha_symlink_cbk,
-			    children[i],
-			    children[i]->fops->symlink,
-			    local->stub->args.symlink.linkname,
-			    &local->stub->args.symlink.loc);
-		return 0;
-	}
-	cnt = local->call_count;
-
-	for (; i < child_count; i++) {
-		if (local->state[i]) {
-			STACK_WIND (frame,
-				    ha_symlink_lookup_cbk,
-				    children[i],
-				    children[i]->fops->lookup,
-				    &local->stub->args.symlink.loc,
-				    0);
-			if (--cnt == 0)
-				break;
-		}
-	}
-	return 0;
-}
-
-int32_t
-ha_symlink (call_frame_t *frame,
-	    xlator_t *this,
-	    const char *linkname,
-	    loc_t *loc)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	int child_count = 0, i = 0;
-	char *stateino = NULL;
-        int32_t op_errno = EINVAL;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-
-	frame->local = local = GF_CALLOC (1, sizeof (*local), 
-                                          gf_ha_mt_ha_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	local->stub = fop_symlink_stub (frame, ha_symlink, linkname, loc);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	local->op_ret = -1;
-	local->op_errno = ENOTCONN;
-	local->state = GF_CALLOC (1, child_count, gf_ha_mt_char);
-        if (!local->state) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	memcpy (local->state, pvt->state, child_count);
-	local->active = -1;
-
-	stateino = GF_CALLOC (1, child_count, gf_ha_mt_char);
-        if (!stateino) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	inode_ctx_put (loc->inode, this, (uint64_t)(long)stateino);
-
-	for (i = 0; i < child_count; i++) {
-		if (local->state[i]) {
-			local->call_count++;
-			if (local->active == -1) {
-				local->active = i;
-			}
-		}
-	}
-
-	STACK_WIND (frame,
-		    ha_symlink_cbk,
-		    HA_ACTIVE_CHILD(this, local),
-		    HA_ACTIVE_CHILD(this, local)->fops->symlink,
-		    linkname, loc);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-        STACK_UNWIND (frame, -1, op_errno, NULL, NULL, NULL, NULL);
-        ha_local_wipe (local);
-        return 0;
-}
-
- int32_t
-ha_rename_cbk (call_frame_t *frame,
-	       void *cookie,
-	       xlator_t *this,
-	       int32_t op_ret,
-	       int32_t op_errno,
-	       struct iatt *buf,
-               struct iatt *preoldparent,
-               struct iatt *postoldparent,
-               struct iatt *prenewparent,
-               struct iatt *postnewparent)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame, op_ret, op_errno, buf, preoldparent,
-                              postoldparent, prenewparent, postnewparent);
-	}
-	return 0;
-}
-
-int32_t
-ha_rename (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *oldloc,
-	   loc_t *newloc)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, oldloc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_rename_stub (frame, ha_rename, oldloc, newloc);
-        if (!local->stub) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_rename_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->rename,
-			   oldloc, newloc);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
-	return 0;
-}
-
-int
-ha_link_lookup_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    inode_t *inode,
-		    struct iatt *buf,
-                    dict_t *dict,
-                    struct iatt *postparent)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	char *stateino = NULL;
-	int child_count = 0, i = 0, cnt = 0;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	uint64_t tmp_stateino = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	for (i = 0; i < child_count; i++)
-		if (prev_frame->this == children[i])
-			break;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR, "(path=%s) (op_ret=%d op_errno=%d)", local->stub->args.link.newloc.path, op_ret, op_errno);
-	}
-	inode_ctx_get (local->stub->args.link.newloc.inode, 
-		       this, &tmp_stateino);  
-	stateino = (char *)(long)tmp_stateino;
-
-	if (op_ret == 0)
-		stateino[i] = 1;
-
-	LOCK (&frame->lock);
-	cnt = --local->call_count;
-	UNLOCK (&frame->lock);
-
-	if (cnt == 0) {
-		call_stub_t *stub = local->stub;
-		GF_FREE (local->state);
-		STACK_UNWIND (frame,
-			      local->op_ret,
-			      local->op_errno,
-			      local->stub->args.link.oldloc.inode, &local->buf,
-                              &local->preparent, &local->postparent);
-		call_stub_destroy (stub);
-	}
-	return 0;
-}
-
-int32_t
-ha_link_cbk (call_frame_t *frame,
-	     void *cookie,
-	     xlator_t *this,
-	     int32_t op_ret,
-	     int32_t op_errno,
-	     inode_t *inode,
-             struct iatt *buf,
-             struct iatt *preparent,
-             struct iatt *postparent)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	char *stateino = NULL;
-	int child_count = 0, i = 0, cnt = 0;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	uint64_t tmp_stateino = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	for (i = 0; i < child_count; i++)
-		if (prev_frame->this == children[i])
-			break;
-
-	if (op_ret == -1) {
-		local->op_errno = op_errno;
-		gf_log (this->name, GF_LOG_ERROR, "(path=%s) (op_ret=%d op_errno=%d)", local->stub->args.link.newloc.path, op_ret, op_errno);
-	}
-	inode_ctx_get (local->stub->args.link.newloc.inode, 
-		       this, &tmp_stateino);
-	stateino = (char *)(long)tmp_stateino;
-
-	if (op_ret == 0) {
-		stateino[i] = 1;
-		local->op_ret = 0;
-		local->first_success = 1;
-		local->buf = *buf;
-                local->preparent = *preparent;
-                local->postparent = *postparent;
-	}
-	cnt = --local->call_count;
-	for (i = local->active + 1; i < child_count; i++) {
-		if (local->state[i])
-			break;
-	}
-
-	if (cnt == 0 || i == child_count) {
-		call_stub_t *stub = local->stub;
-		GF_FREE (local->state);
-		stub = local->stub;
-		STACK_UNWIND (frame, local->op_ret, local->op_errno,
-                              local->stub->args.link.oldloc.inode, &local->buf,
-                              &local->preparent, &local->postparent);
-		call_stub_destroy (stub);
-		return 0;
-	}
-
-	local->active = i;
-
-	if (local->first_success == 0) {
-		STACK_WIND (frame,
-			    ha_link_cbk,
-			    children[i],
-			    children[i]->fops->link,
-			    &local->stub->args.link.oldloc,
-			    &local->stub->args.link.newloc);
-		return 0;
-	}
-	cnt = local->call_count;
-
-	for (; i < child_count; i++) {
-		if (local->state[i]) {
-			STACK_WIND (frame,
-				    ha_link_lookup_cbk,
-				    children[i],
-				    children[i]->fops->lookup,
-				    &local->stub->args.link.newloc,
-				    0);
-			if (--cnt == 0)
-				break;
-		}
-	}
-	return 0;
-}
-
-int32_t
-ha_link (call_frame_t *frame,
-	 xlator_t *this,
-	 loc_t *oldloc,
-	 loc_t *newloc)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	int child_count = 0, i = 0;
-	char *stateino = NULL;
-	int32_t ret = 0, op_errno = 0;
-	uint64_t tmp_stateino = 0;
-
-	ret = inode_ctx_get (newloc->inode, this, &tmp_stateino);
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR, "dict_ptr_error()");
-	}
-	stateino = (char *)(long)tmp_stateino;
-
-	if (stateino == NULL) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"newloc->inode's ctx is NULL, returning EINVAL");
-		STACK_UNWIND (frame, -1, EINVAL, oldloc->inode, NULL, NULL,
-                              NULL);
-		return 0;
-	}
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-
-	frame->local = local = GF_CALLOC (1, sizeof (*local), 
-                                          gf_ha_mt_ha_local_t);
-        if (!frame->local) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	local->stub = fop_link_stub (frame, ha_link, oldloc, newloc);
-        if (!local->stub) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	local->op_ret = -1;
-	local->op_errno = ENOTCONN;
-	local->state = GF_CALLOC (1, child_count, gf_ha_mt_char);
-        if (!local->state) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	memcpy (local->state, pvt->state, child_count);
-	local->active = -1;
-
-	for (i = 0; i < child_count; i++) {
-		if (local->state[i]) {
-			local->call_count++;
-			if (local->active == -1)
-				local->active = i;
-		}
-	}
-
-	STACK_WIND (frame,
-		    ha_link_cbk,
-		    HA_ACTIVE_CHILD(this, local),
-		    HA_ACTIVE_CHILD(this, local)->fops->link,
-		    oldloc,
-		    newloc);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-        STACK_UNWIND (frame, -1, op_errno, NULL, NULL, NULL, NULL);
-        return 0;
-}
-
-int32_t
-ha_create_cbk (call_frame_t *frame,
-	       void *cookie,
-	       xlator_t *this,
-	       int32_t op_ret,
-	       int32_t op_errno,
-	       fd_t *fd,
-	       inode_t *inode,
-	       struct iatt *buf,
-               struct iatt *preparent,
-               struct iatt *postparent)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	int i, child_count = 0, cnt = 0, ret = 0;
-	char *stateino = NULL;
-	hafd_t *hafdp = NULL;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	uint64_t tmp_stateino = 0;
-	uint64_t tmp_hafdp = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	ret = inode_ctx_get (local->stub->args.create.loc.inode, 
-			     this, &tmp_stateino);
-	stateino = (char *)(long)tmp_stateino;
-
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR, "dict_to_ptr() error");
-		/* FIXME: handle */
-	}
-	ret = fd_ctx_get (local->stub->args.create.fd, this, &tmp_hafdp);
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR, "dict_to_ptr() error");
-		/* FIXME: handle */
-	}
-	hafdp = (hafd_t *)(long)tmp_hafdp;
-
-	for (i = 0; i < child_count; i++) {
-		if (prev_frame->this == children[i])
-			break;
-	}
-
-	if (op_ret == -1) {
-		local->op_errno = op_errno;
-		gf_log (this->name, GF_LOG_ERROR, "(path=%s) (op_ret=%d op_errno=%d)", local->stub->args.create.loc.path, op_ret, op_errno);
-	}
-	if (op_ret != -1) {
-		stateino[i] = 1;
-		hafdp->fdstate[i] = 1;
-		if (local->op_ret == -1) {
-			local->op_ret = 0;
-			local->buf = *buf;
-                        local->preparent = *preparent;
-                        local->postparent = *postparent;
-			local->first_success = 1;
-		}
-		local->stub->args.create.flags &= (~O_EXCL);
-	}
-	LOCK (&frame->lock);
-	cnt = --local->call_count;
-	UNLOCK (&frame->lock);
-
-	for (i = local->active + 1; i < child_count; i++) {
-		if (local->state[i])
-			break;
-	}
-
-	if (cnt == 0 || i == child_count) {
-		char *state = local->state;
-		call_stub_t *stub = local->stub;
-		STACK_UNWIND (frame, local->op_ret, local->op_errno,
-			      stub->args.create.fd,
-			      stub->args.create.loc.inode, &local->buf,
-                              &local->preparent, &local->postparent);
-		GF_FREE (state);
-		call_stub_destroy (stub);
-		return 0;
-	}
-	local->active = i;
-	cnt = local->call_count;
-	for (; i < child_count; i++) {
-		if (local->state[i]) {
-			STACK_WIND (frame,
-				    ha_create_cbk,
-				    children[i],
-				    children[i]->fops->create,
-				    &local->stub->args.create.loc,
-				    local->stub->args.create.flags,
-				    local->stub->args.create.mode,
-				    local->stub->args.create.fd);
-			if ((local->first_success == 0) || (cnt == 0))
-				break;
-		}
-	}
-	return 0;
-}
-
-int32_t
-ha_create (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *loc,
-	   int32_t flags,
-	   mode_t mode, fd_t *fd)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	int i, child_count = 0;
-	char *stateino = NULL;
-	xlator_t **children = NULL;
-	hafd_t *hafdp = NULL;
-        int32_t op_errno = EINVAL;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	children = pvt->children;
-
-	if (local == NULL) {
-                frame->local = local = GF_CALLOC (1, sizeof (*local), 
-                                                  gf_ha_mt_ha_local_t);
-                if (!local) {
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                        goto err;
-                }
-
-		local->stub = fop_create_stub (frame, ha_create, loc, flags, mode, fd);
-                if (!local->stub) {
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                        goto err;
-                }
-
-                local->state = GF_CALLOC (1, child_count, gf_ha_mt_char);
-                if (!local->state) {
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                        goto err;
-                }
-
-		local->active = -1;
-		local->op_ret = -1;
-		local->op_errno = ENOTCONN;
-		memcpy (local->state, pvt->state, child_count);
-
-		for (i = 0; i < pvt->child_count; i++) {
-			if (local->state[i]) {
-				local->call_count++;
-				if (local->active == -1)
-					local->active = i;
-			}
-		}
-		/* FIXME handle active -1 */
-		stateino = GF_CALLOC (1, child_count, gf_ha_mt_char);
-                if (!stateino) {
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                        goto err;
-                }
-
-		hafdp = GF_CALLOC (1, sizeof (*hafdp), gf_ha_mt_hafd_t);
-                if (!hafdp) {
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                        goto err;
-                }
-
-		hafdp->fdstate = GF_CALLOC (1, child_count, gf_ha_mt_char);
-                if (!hafdp->fdstate) {
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                        goto err;
-                }
-
-		hafdp->path = gf_strdup(loc->path);
-                if (!hafdp->path) {
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                        goto err;
-                }
-
-		LOCK_INIT (&hafdp->lock);
-		fd_ctx_set (fd, this, (uint64_t)(long)hafdp);
-		inode_ctx_put (loc->inode, this, (uint64_t)(long)stateino);
-	}
-
-	STACK_WIND (frame,
-		    ha_create_cbk,
-		    children[local->active],
-		    children[local->active]->fops->create,
-		    loc, flags, mode, fd);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-        STACK_UNWIND (frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
-        ha_local_wipe (local);
-
-        if (stateino) {
-                GF_FREE (stateino);
-                stateino = NULL;
-        }
-
-        if (hafdp) {
-                if (hafdp->fdstate) {
-                        GF_FREE (hafdp->fdstate);
-                }
-
-                if (hafdp->path) {
-                        GF_FREE (hafdp->path);
-                }
-
-                GF_FREE (hafdp);
-        }
-
-        return 0;
-}
-
- int32_t
-ha_open_cbk (call_frame_t *frame,
-	     void *cookie,
-	     xlator_t *this,
-	     int32_t op_ret,
-	     int32_t op_errno,
-	     fd_t *fd)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	xlator_t **children = NULL;
-	int i = 0, child_count = 0, callcnt = 0, ret = 0;
-	call_frame_t *prev_frame = NULL;
-	hafd_t *hafdp = NULL;
-	uint64_t tmp_hafdp = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	children = pvt->children;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-
-	ret = fd_ctx_get (local->fd, this, &tmp_hafdp);
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR, "dict_ptr_error()");
-	}
-	hafdp = (hafd_t *)(long)tmp_hafdp;
-
-	for (i = 0; i < child_count; i++)
-		if (children[i] == prev_frame->this)
-			break;
-	LOCK (&frame->lock);
-	if (op_ret != -1) {
-		hafdp->fdstate[i] = 1;
-		local->op_ret = 0;
-	}
-	if (op_ret == -1 && op_errno != ENOTCONN)
-		local->op_errno = op_errno;
-	callcnt = --local->call_count;
-	UNLOCK (&frame->lock);
-
-	if (callcnt == 0) {
-		STACK_UNWIND (frame,
-			      local->op_ret,
-			      local->op_errno,
-			      local->fd);
-	}
-	return 0;
-}
-
-int32_t
-ha_open (call_frame_t *frame,
-	 xlator_t *this,
-	 loc_t *loc,
-	 int32_t flags, fd_t *fd, int wbflags)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	char *stateino = NULL;
-	xlator_t **children = NULL;
-	int cnt = 0, i, child_count = 0, ret = 0;
-	hafd_t *hafdp = NULL;
-	uint64_t tmp_stateino = 0;
-        int32_t op_errno = ENOMEM;
-
-	local = frame->local;
-	pvt = this->private;
-	children = pvt->children;
-	child_count = pvt->child_count;
-
-
-	frame->local = local = GF_CALLOC (1, sizeof (*local), 
-                                          gf_ha_mt_ha_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-                
-	local->op_ret = -1;
-	local->op_errno = ENOTCONN;
-	local->fd = fd;
-
-	hafdp = GF_CALLOC (1, sizeof (*hafdp), gf_ha_mt_hafd_t);
-        if (!hafdp) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	hafdp->fdstate = GF_CALLOC (1, child_count, gf_ha_mt_char);
-        if (!hafdp->fdstate) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	hafdp->path = gf_strdup (loc->path);
-        if (!hafdp->path) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	hafdp->active = -1;
-	if (pvt->pref_subvol == -1) {
-		hafdp->active = fd->inode->ino % child_count;
-	}
-
-	LOCK_INIT (&hafdp->lock);
-	fd_ctx_set (fd, this, (uint64_t)(long)hafdp);
-	ret = inode_ctx_get (loc->inode, this, &tmp_stateino);
-	stateino = (char *)(long)tmp_stateino;
-
-	for (i = 0; i < child_count; i++)
-		if (stateino[i])
-			cnt++;
-	local->call_count = cnt;
-	for (i = 0; i < child_count; i++) {
-		if (stateino[i]) {
-			STACK_WIND (frame,
-				    ha_open_cbk,
-				    children[i],
-				    children[i]->fops->open,
-				    loc, flags, fd, wbflags);
-			if (--cnt == 0)
-				break;
-		}
-	}
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, fd);
-        if (hafdp) {
-                if (hafdp->fdstate) {
-                        GF_FREE (hafdp->fdstate);
-                        hafdp->fdstate = NULL;
-                }
-
-                if (hafdp->path) {
-                        GF_FREE (hafdp->path);
-                        hafdp->path = NULL;
-                }
-
-                GF_FREE (hafdp);
-        }
-
-        ha_local_wipe (local);
-        return 0;
-}
-
- int32_t
-ha_readv_cbk (call_frame_t *frame,
-	      void *cookie,
-	      xlator_t *this,
-	      int32_t op_ret,
-	      int32_t op_errno,
-	      struct iovec *vector,
-	      int32_t count,
-	      struct iatt *stbuf,
-              struct iobref *iobref)
-{
-	int ret = 0;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-			      vector,
-			      count,
-			      stbuf,
-                              iobref);
-	}
-	return 0;
-}
-
-int32_t
-ha_readv (call_frame_t *frame,
-	  xlator_t *this,
-	  fd_t *fd,
-	  size_t size,
-	  off_t offset)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_readv_stub (frame, ha_readv, fd, size, offset);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_readv_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->readv,
-			   fd,
-			   size,
-			   offset);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, NULL, 0, NULL, NULL);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
- int32_t
-ha_writev_cbk (call_frame_t *frame,
-	       void *cookie,
-	       xlator_t *this,
-	       int32_t op_ret,
-	       int32_t op_errno,
-               struct iatt *prebuf,
-	       struct iatt *postbuf)
-{
-	int ret = 0;
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-                              prebuf,
-			      postbuf);
-	}
-	return 0;
-}
-
-int32_t
-ha_writev (call_frame_t *frame,
-	   xlator_t *this,
-	   fd_t *fd,
-	   struct iovec *vector,
-	   int32_t count,
-	   off_t off,
-           struct iobref *iobref)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_writev_stub (frame, ha_writev, fd, vector, count, off,
-                                       iobref);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_writev_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->writev,
-			   fd,
-			   vector,
-			   count,
-			   off,
-                           iobref);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-        ha_local_wipe (local);
-	return 0;	
-}
-
- int32_t
-ha_flush_cbk (call_frame_t *frame,
-	      void *cookie,
-	      xlator_t *this,
-	      int32_t op_ret,
-	      int32_t op_errno)
-{
-	int ret = 0;
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno);
-	}
-	return 0;
-}
-
-int32_t
-ha_flush (call_frame_t *frame,
-	  xlator_t *this,
-	  fd_t *fd)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_flush_stub (frame, ha_flush, fd);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-                
-	STACK_WIND_COOKIE (frame,
-			   ha_flush_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->flush,
-			   fd);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
-
- int32_t
-ha_fsync_cbk (call_frame_t *frame,
-	      void *cookie,
-	      xlator_t *this,
-	      int32_t op_ret,
-	      int32_t op_errno,
-              struct iatt *prebuf,
-              struct iatt *postbuf)
-{
-	int ret = 0;
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno);
-	}
-	return 0;
-}
-
-int32_t
-ha_fsync (call_frame_t *frame,
-	  xlator_t *this,
-	  fd_t *fd,
-	  int32_t flags)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_fsync_stub (frame, ha_fsync, fd, flags);
-        if (!local->stub) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_fsync_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->fsync,
-			   fd,
-			   flags);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
- int32_t
-ha_fstat_cbk (call_frame_t *frame,
-	      void *cookie,
-	      xlator_t *this,
-	      int32_t op_ret,
-	      int32_t op_errno,
-	      struct iatt *buf)
-{
-	int ret = 0;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-			      buf);
-	}
-	return 0;
-}
-
-int32_t
-ha_fstat (call_frame_t *frame,
-	  xlator_t *this,
-	  fd_t *fd)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_fstat_stub (frame, ha_fstat, fd);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_fstat_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->fstat,
-			   fd);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, NULL);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
-int32_t
-ha_opendir_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		fd_t *fd)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	xlator_t **children = NULL;
-	int i = 0, child_count = 0, callcnt = 0, ret = 0;
-	call_frame_t *prev_frame = NULL;
-	hafd_t *hafdp = NULL;
-	uint64_t tmp_hafdp = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	children = pvt->children;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-
-	ret = fd_ctx_get (local->fd, this, &tmp_hafdp);
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR, "dict_ptr_error()");
-	}
-	hafdp = (hafd_t *)(long)tmp_hafdp;
-
-	for (i = 0; i < child_count; i++)
-		if (children[i] == prev_frame->this)
-			break;
-	LOCK (&frame->lock);
-	if (op_ret != -1) {
-		hafdp->fdstate[i] = 1;
-		local->op_ret = 0;
-	}
-	if (op_ret == -1 && op_errno != ENOTCONN)
-		local->op_errno = op_errno;
-	callcnt = --local->call_count;
-	UNLOCK (&frame->lock);
-
-	if (callcnt == 0) {
-		STACK_UNWIND (frame,
-			      local->op_ret,
-			      local->op_errno,
-			      local->fd);
-	}
-	return 0;
-}
-
-int32_t
-ha_opendir (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc, fd_t *fd)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	char *stateino = NULL;
-	xlator_t **children = NULL;
-	int cnt = 0, i, child_count = 0, ret = 0;
-	hafd_t *hafdp = NULL;
-	uint64_t tmp_stateino = 0;
-        int32_t op_errno = EINVAL;
-
-	local = frame->local;
-	pvt = this->private;
-	children = pvt->children;
-	child_count = pvt->child_count;
-
-	frame->local = local = GF_CALLOC (1, sizeof (*local), 
-                                          gf_ha_mt_ha_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	local->op_ret = -1;
-	local->op_errno = ENOTCONN;
-	local->fd = fd;
-
-	hafdp = GF_CALLOC (1, sizeof (*hafdp), gf_ha_mt_hafd_t);
-        if (!hafdp) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	hafdp->fdstate = GF_CALLOC (1, child_count, gf_ha_mt_char);
-        if (!hafdp->fdstate) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	hafdp->path = gf_strdup (loc->path);
-        if (!hafdp->path) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	LOCK_INIT (&hafdp->lock);
-	fd_ctx_set (fd, this, (uint64_t)(long)hafdp);
-	ret = inode_ctx_get (loc->inode, this, &tmp_stateino);
-	stateino = (char *)(long)tmp_stateino;
-	
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR, "inode_ctx_get() error");
-	}
-	for (i = 0; i < child_count; i++)
-		if (stateino[i])
-			cnt++;
-	local->call_count = cnt;
-	for (i = 0; i < child_count; i++) {
-		if (stateino[i]) {
-			STACK_WIND (frame,
-				    ha_opendir_cbk,
-				    children[i],
-				    children[i]->fops->opendir,
-				    loc, fd);
-			if (--cnt == 0)
-				break;
-		}
-	}
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, NULL);
-        ha_local_wipe (local);
-        if (hafdp) {
-                if (hafdp->fdstate) {
-                        GF_FREE (hafdp->fdstate);
-                        hafdp->fdstate = NULL;
-                }
-
-                if (hafdp->path) {
-                        GF_FREE (hafdp->path);
-                        hafdp->path = NULL;
-                }
-
-                GF_FREE (hafdp);
-        }
-        return 0;
-}
-
- int32_t
-ha_getdents_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 dir_entry_t *entries,
-		 int32_t count)
-{
-	int ret = 0;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-			      entries,
-			      count);
-	}
-	return 0;
-}
-
-int32_t
-ha_getdents (call_frame_t *frame,
-	     xlator_t *this,
-	     fd_t *fd,
-	     size_t size,
-	     off_t offset,
-	     int32_t flag)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_getdents_stub (frame, ha_getdents, fd, size, offset,
-                                         flag);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_getdents_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->getdents,
-			   fd,
-			   size,
-			   offset,
-			   flag);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, NULL, 0);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
- int32_t
-ha_setdents_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno)
-{
-	int ret = 0;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno);
-	}
-	return 0;
-}
-
-int32_t
-ha_setdents (call_frame_t *frame,
-	     xlator_t *this,
-	     fd_t *fd,
-	     int32_t flags,
-	     dir_entry_t *entries,
-	     int32_t count)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-
-	local->stub = fop_setdents_stub (frame, ha_setdents, fd, flags, entries,
-                                         count);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_setdents_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->setdents,
-			   fd,
-			   flags,
-			   entries,
-			   count);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
- int32_t
-ha_fsyncdir_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno)
-{
-	int ret = 0;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno);
-	}
-	return 0;
-}
-
-int32_t
-ha_fsyncdir (call_frame_t *frame,
-	     xlator_t *this,
-	     fd_t *fd,
-	     int32_t flags)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_fsyncdir_stub (frame, ha_fsyncdir, fd, flags);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_fsyncdir_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->fsyncdir,
-			   fd,
-			   flags);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
- 
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
-
- int32_t
-ha_statfs_cbk (call_frame_t *frame,
-	       void *cookie,
-	       xlator_t *this,
-	       int32_t op_ret,
-	       int32_t op_errno,
-	       struct statvfs *buf)
-{
-        ha_local_t   *local = NULL;
-        ha_private_t *priv  = NULL;
-
-        local = frame->local;
-        if (-1 == op_ret) {
-                local->active = (local->active + 1) % priv->child_count;
-                local->tries--;
-                if (!local->tries)
-                        goto out;
-                
-                STACK_WIND (frame, ha_statfs_cbk,
-                            HA_ACTIVE_CHILD(this, local),
-                            HA_ACTIVE_CHILD(this, local)->fops->statfs, 
-                            &local->loc);
-                return 0;
-        }
-
- out:
-        loc_wipe (&local->loc);
-        STACK_UNWIND (frame, op_ret, op_errno, buf);
-        
-	return 0;
-}
-
-int32_t
-ha_statfs (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *loc)
-{
-        ha_private_t *priv = NULL;
-	ha_local_t   *local = NULL;
-	int           op_errno = 0;
-
-        /* The normal way of handling failover doesn't work here
-         * as loc->inode may be null in this case.
-         */
-        local = GF_CALLOC (1, sizeof (*local), 
-                           gf_ha_mt_ha_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        priv = this->private;
-        frame->local  = local;
-        local->active = priv->pref_subvol;
-        if (-1 == local->active)
-                local->active = 0;
-        local->tries  = priv->child_count;
-        loc_copy (&local->loc, loc);
-
-	STACK_WIND (frame, ha_statfs_cbk, HA_ACTIVE_CHILD(this, local),
-                    HA_ACTIVE_CHILD(this, local)->fops->statfs, loc);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno, NULL);
-	return 0;
-}
-
- int32_t
-ha_setxattr_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno);
-	}
-	return 0;
-}
-
-int32_t
-ha_setxattr (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     dict_t *dict,
-	     int32_t flags)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_setxattr_stub (frame, ha_setxattr, loc, dict, flags);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_setxattr_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->setxattr,
-			   loc,
-			   dict,
-			   flags);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
- int32_t
-ha_getxattr_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 dict_t *dict)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-			      dict);
-	}
-	return 0;
-}
-
-int32_t
-ha_getxattr (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     const char *name)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_getxattr_stub (frame, ha_getxattr, loc, name);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_getxattr_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->getxattr,
-			   loc,
-			   name);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, NULL);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
-int32_t
-ha_xattrop_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		dict_t *dict)
-{
-	int ret = -1;
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-	if (ret == 0) {
-		STACK_UNWIND (frame, op_ret, op_errno, dict);
-	}
-	return 0;
-}
-
-
-int32_t
-ha_xattrop (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc,
-	    gf_xattrop_flags_t flags,
-	    dict_t *dict)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-
-	local->stub = fop_xattrop_stub (frame, ha_xattrop, loc, flags, dict);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_xattrop_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->xattrop,
-			   loc,
-			   flags,
-			   dict);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, dict);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
-int32_t
-ha_fxattrop_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 dict_t *dict)
-{
-	int ret = -1;
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-	if (ret == 0)
-		STACK_UNWIND (frame, op_ret, op_errno, dict);
-	return 0;
-}
-
-int32_t
-ha_fxattrop (call_frame_t *frame,
-	     xlator_t *this,
-	     fd_t *fd,
-	     gf_xattrop_flags_t flags,
-	     dict_t *dict)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_fxattrop_stub (frame, ha_fxattrop, fd, flags, dict);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_fxattrop_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->fxattrop,
-			   fd,
-			   flags,
-			   dict);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, dict);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
- int32_t
-ha_removexattr_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno)
-{
-	int ret = -1;
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno);
-	}
-	return 0;
-}
-
-int32_t
-ha_removexattr (call_frame_t *frame,
-		xlator_t *this,
-		loc_t *loc,
-		const char *name)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	
-	local->stub = fop_removexattr_stub (frame, ha_removexattr, loc, name);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_removexattr_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->removexattr,
-			   loc,
-			   name);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
-int32_t
-ha_lk_setlk_unlck_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno,
-		       struct gf_flock *lock)
-{
-	ha_local_t *local = NULL;
-	int cnt = 0;
-	call_stub_t *stub = NULL;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	cnt = --local->call_count;
-	if (op_ret == 0)
-		local->op_ret = 0;
-	UNLOCK (&frame->lock);
-
-	if (cnt == 0) {
-		stub = local->stub;
-		GF_FREE (local->state);
-		if (stub->args.lk.lock.l_type == F_UNLCK) {
-			STACK_UNWIND (frame, local->op_ret, local->op_errno, &stub->args.lk.lock);
-		} else {
-			STACK_UNWIND (frame, -1, EIO, NULL);
-		}
-		call_stub_destroy (stub);
-	}
-	return 0;
-}
-
-int32_t
-ha_lk_setlk_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 struct gf_flock *lock)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	xlator_t **children = NULL;
-	int i = 0, cnt = 0, j = 0;
-	int child_count = 0;
-	call_frame_t *prev_frame = NULL;
-	char *state = NULL;
-
-	local = frame->local;
-	pvt = this->private;
-	children = pvt->children;
-	child_count = pvt->child_count;
-	prev_frame = cookie;
-	state = local->state;
-
-	if (op_ret == 0)
-		local->op_ret = 0;
-
-	if ((op_ret == 0) || (op_ret == -1 && op_errno == ENOTCONN)) {
-		for (i = 0; i < child_count; i++) {
-			if (prev_frame->this == cookie)
-				break;
-		}
-		i++;
-		for (; i < child_count; i++) {
-			if (local->state[i])
-				break;
-		}
-		if (i == child_count) {
-			call_stub_t *stub = local->stub;
-			GF_FREE (local->state);
-			STACK_UNWIND (frame, 0, op_errno, &stub->args.lk.lock);
-			call_stub_destroy (stub);
-			return 0;
-		}
-		STACK_WIND (frame,
-			    ha_lk_setlk_cbk,
-			    children[i],
-			    children[i]->fops->lk,
-			    local->stub->args.lk.fd,
-			    local->stub->args.lk.cmd,
-			    &local->stub->args.lk.lock);
-		return 0;
-	} else {
-		for (i = 0; i < child_count; i++) {
-			if (prev_frame->this == cookie)
-				break;
-		}
-		cnt = 0;
-		for (j = 0; j < i; j++) {
-			if (state[i])
-				cnt++;
-		}
-		if (cnt) {
-			struct gf_flock lock;
-			lock = local->stub->args.lk.lock;
-			for (i = 0; i < child_count; i++) {
-				if (state[i]) {
-					STACK_WIND (frame,
-						    ha_lk_setlk_unlck_cbk,
-						    children[i],
-						    children[i]->fops->lk,
-						    local->stub->args.lk.fd,
-						    local->stub->args.lk.cmd,
-						    &lock);
-					if (--cnt == 0)
-						break;
-				}
-			}
-			return 0;
-		} else {
-			GF_FREE (local->state);
-			call_stub_destroy (local->stub);
-			STACK_UNWIND (frame,
-				      op_ret,
-				      op_errno,
-				      lock);
-			return 0;
-		}
-	}
-}
-
-int32_t
-ha_lk_getlk_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 struct gf_flock *lock)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	fd_t *fd = NULL;
-	int child_count = 0, i = 0;
-	xlator_t **children = NULL;
-	call_frame_t *prev_frame = NULL;
-
-	local = frame->local;
-	pvt = this->private;
-	fd = local->stub->args.lk.fd;
-	child_count = pvt->child_count;
-	children = pvt->children;
-	prev_frame = cookie;
-
-	if (op_ret == 0) {
-		GF_FREE (local->state);
-		call_stub_destroy (local->stub);
-		STACK_UNWIND (frame, 0, 0, lock);
-		return 0;
-	}
-
-	for (i = 0; i < child_count; i++) {
-		if (prev_frame->this == children[i])
-			break;
-	}
-
-	for (; i < child_count; i++) {
-		if (local->state[i])
-			break;
-	}
-
-	if (i == child_count) {
-		GF_FREE (local->state);
-		call_stub_destroy (local->stub);
-		STACK_UNWIND (frame, op_ret, op_errno, lock);
-		return 0;
-	}
-
-	STACK_WIND (frame,
-		    ha_lk_getlk_cbk,
-		    children[i],
-		    children[i]->fops->lk,
-		    fd,
-		    local->stub->args.lk.cmd,
-		    &local->stub->args.lk.lock);
-	return 0;
-}
-
-int32_t
-ha_lk (call_frame_t *frame,
-       xlator_t *this,
-       fd_t *fd,
-       int32_t cmd,
-       struct gf_flock *lock)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	hafd_t *hafdp = NULL;
-	char *state = NULL;
-	int child_count = 0, i = 0, cnt = 0, ret = 0;
-	xlator_t **children = NULL;
-	uint64_t tmp_hafdp = 0;
-        int32_t op_errno = EINVAL;
-
-	local = frame->local;
-	pvt = this->private;
-	child_count = pvt->child_count;
-	children = pvt->children;
-	ret = fd_ctx_get (fd, this, &tmp_hafdp);
-	if (ret < 0)
-		gf_log (this->name, GF_LOG_ERROR, "fd_ctx_get failed");
-
-	if (local == NULL) {
-                local = frame->local = GF_CALLOC (1, sizeof (*local),
-                                                  gf_ha_mt_ha_local_t);
-                if (!local) {
-                        op_errno = ENOMEM;
-                        gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                        goto err;
-                }
-                        
-		local->active = -1;
-		local->op_ret = -1;
-		local->op_errno = ENOTCONN;
-	}
-	hafdp = (hafd_t *)(long)tmp_hafdp;
-
-	if (local->active == -1) {
-                op_errno = ENOTCONN;
-                goto err;
-	}
-
-	local->stub = fop_lk_stub (frame, ha_lk, fd, cmd, lock);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	local->state = GF_CALLOC (1, child_count, gf_ha_mt_char);
-        if (!local->state) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	state = hafdp->fdstate;
-	LOCK (&hafdp->lock);
-	memcpy (local->state, state, child_count);
-	UNLOCK (&hafdp->lock);
-	if (cmd == F_GETLK) {
-		for (i = 0; i < child_count; i++) {
-			if (local->state[i])
-				break;
-		}
-		STACK_WIND (frame,
-			    ha_lk_getlk_cbk,
-			    children[i],
-			    children[i]->fops->lk,
-			    fd,
-			    cmd,
-			    lock);
-	} else if (cmd == F_SETLK && lock->l_type == F_UNLCK) {
-		for (i = 0; i < child_count; i++) {
-			if (local->state[i])
-				local->call_count++;
-		}
-		cnt = local->call_count;
-		for (i = 0; i < child_count; i++) {
-			if (local->state[i]) {
-				STACK_WIND (frame,
-					    ha_lk_setlk_unlck_cbk,
-					    children[i],
-					    children[i]->fops->lk,
-					    fd, cmd, lock);
-				if (--cnt == 0)
-					break;
-			}
-		}
-	} else {
-		for (i = 0; i < child_count; i++) {
-			if (local->state[i])
-				break;
-		}
-		STACK_WIND (frame,
-			    ha_lk_setlk_cbk,
-			    children[i],
-			    children[i]->fops->lk,
-			    fd,
-			    cmd,
-			    lock);
-	}
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-        STACK_UNWIND (frame, -1, op_errno, NULL);
-
-        ha_local_wipe (local);
-        return 0;
-}
-
- int32_t
-ha_inode_entry_lk_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno);
-	}
-	return 0;
-}
-
-int32_t
-ha_inodelk (call_frame_t *frame,
-	    xlator_t *this,
-            const char *volume,
-	    loc_t *loc,
-	    int32_t cmd,
-	    struct gf_flock *lock)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_inodelk_stub (frame, ha_inodelk, volume,
-                                        loc, cmd, lock);
-	STACK_WIND_COOKIE (frame,
-			   ha_inode_entry_lk_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->inodelk,
-			   volume,
-                           loc,
-			   cmd,
-			   lock);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno);
-	return 0;
-}
-
-int32_t
-ha_entrylk (call_frame_t *frame,
-	    xlator_t *this,
-            const char *volume,
-	    loc_t *loc,
-	    const char *basename,
-	    entrylk_cmd cmd,
-	    entrylk_type type)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_entrylk_stub (frame, ha_entrylk, volume,
-                                        loc, basename, cmd, type);
-	STACK_WIND_COOKIE (frame,
-			   ha_inode_entry_lk_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->entrylk,
-			   volume, loc, basename, cmd, type);
-	return 0;
-err:
-	STACK_UNWIND (frame, -1, op_errno);
-	return 0;
-}
-
- int32_t
-ha_checksum_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 uint8_t *file_checksum,
-		 uint8_t *dir_checksum)
-{
-	int ret = -1;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-	if (ret == 0) {
-		STACK_UNWIND (frame,
-			      op_ret,
-			      op_errno,
-			      file_checksum,
-			      dir_checksum);
-	}
-	return 0;
-}
-
-int32_t
-ha_checksum (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     int32_t flag)
-{
-	int op_errno = 0;
-	ha_local_t *local = NULL;
-
-	op_errno = ha_alloc_init_inode (frame, loc->inode);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-	local->stub = fop_checksum_stub (frame, ha_checksum, loc, flag);
-        if (!local->stub) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	STACK_WIND_COOKIE (frame,
-			   ha_checksum_cbk,
-			   (void *)(long)local->active,
-			   HA_ACTIVE_CHILD(this, local),
-			   HA_ACTIVE_CHILD(this, local)->fops->checksum,
-			   loc,
-			   flag);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
-int32_t
-ha_common_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, gf_dirent_t *entries)
-{
-	int ret = 0;
-
-	ret = ha_handle_cbk (frame, cookie, op_ret, op_errno);
-	if (ret == 0)
-		STACK_UNWIND (frame, op_ret, op_errno, entries);
-	return 0;
-}
-
-int32_t
-ha_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-            off_t off);
-
-int32_t
-ha_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-             off_t off);
-int32_t
-ha_do_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-               off_t off, int whichop)
-{
-	ha_local_t *local = NULL;
-	int op_errno = 0;
-
-	op_errno = ha_alloc_init_fd (frame, fd);
-	if (op_errno < 0) {
-		op_errno = -op_errno;
-		goto err;
-	}
-	local = frame->local;
-        if (whichop == GF_FOP_READDIR)
-                local->stub = fop_readdir_stub (frame, ha_readdir, fd, size,
-                                                off);
-        else
-                local->stub = fop_readdirp_stub (frame, ha_readdirp, fd, size,
-                                                 off);
-        if (!local->stub) {
-                op_errno = ENOMEM; 
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-        if (whichop == GF_FOP_READDIR)
-                STACK_WIND_COOKIE (frame, ha_common_readdir_cbk,
-                                   (void *)(long)local->active,
-                                   HA_ACTIVE_CHILD(this, local),
-                                   HA_ACTIVE_CHILD(this, local)->fops->readdir,
-                                   fd, size, off);
-        else
-                STACK_WIND_COOKIE (frame, ha_common_readdir_cbk,
-                                   (void *)(long)local->active,
-                                   HA_ACTIVE_CHILD(this, local),
-                                   HA_ACTIVE_CHILD(this, local)->fops->readdirp,
-                                   fd, size, off);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-	STACK_UNWIND (frame, -1, ENOTCONN, NULL);
-
-        ha_local_wipe (local);
-	return 0;
-}
-
-
-int32_t
-ha_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-            off_t off)
-{
-        ha_do_readdir (frame, this, fd, size, off, GF_FOP_READDIR);
-        return 0;
-}
-
-int32_t
-ha_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-             off_t off)
-{
-        ha_do_readdir (frame, this, fd, size, off, GF_FOP_READDIRP);
-        return 0;
-}
-
-/* Management operations */
-
- int32_t
-ha_stats_cbk (call_frame_t *frame,
-	      void *cookie,
-	      xlator_t *this,
-	      int32_t op_ret,
-	      int32_t op_errno,
-	      struct xlator_stats *stats)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	int i = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	if (op_ret == -1 && op_errno == ENOTCONN) {
-		for (i = 0; i < pvt->child_count; i++) {
-			if (prev_frame->this == children[i])
-				break;
-		}
-		i++;
-		for (; i < pvt->child_count; i++) {
-			if (pvt->state[i])
-				break;
-		}
-
-		if (i == pvt->child_count) {
-			STACK_UNWIND (frame, -1, ENOTCONN, NULL);
-			return 0;
-		}
-		STACK_WIND (frame,
-			    ha_stats_cbk,
-			    children[i],
-			    children[i]->mops->stats,
-			    local->flags);
-		return 0;
-	}
-
-	STACK_UNWIND (frame,
-		      op_ret,
-		      op_errno,
-		      stats);
-	return 0;
-}
-
-int32_t
-ha_stats (call_frame_t *frame,
-	  xlator_t *this,
-	  int32_t flags)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	xlator_t **children = NULL;
-	int i = 0;
-        int32_t op_errno = EINVAL;
-
-	local = frame->local = GF_CALLOC (1, sizeof (*local),
-                                          gf_ha_mt_ha_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	pvt = this->private;
-	children = pvt->children;
-	for (i = 0; i < pvt->child_count; i++) {
-		if (pvt->state[i])
-			break;
-	}
-
-	if (i == pvt->child_count) {
-                op_errno = ENOTCONN;
-                goto err;
-	}
-	local->flags = flags;
-
-	STACK_WIND (frame,
-		    ha_stats_cbk,
-		    children[i],
-		    children[i]->mops->stats,
-		    flags);
-	return 0;
-
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-        STACK_UNWIND (frame, -1, ENOTCONN, NULL);
-
-        ha_local_wipe (local);
-        return 0;
-
-}
-
-
-int32_t
-ha_getspec_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		char *spec_data)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	call_frame_t *prev_frame = NULL;
-	xlator_t **children = NULL;
-	int i = 0;
-
-	local = frame->local;
-	pvt = this->private;
-	prev_frame = cookie;
-	children = pvt->children;
-
-	if (op_ret == -1 && op_errno == ENOTCONN) {
-		for (i = 0; i < pvt->child_count; i++) {
-			if (prev_frame->this == children[i])
-				break;
-		}
-		i++;
-		for (; i < pvt->child_count; i++) {
-			if (pvt->state[i])
-				break;
-		}
-
-		if (i == pvt->child_count) {
-			STACK_UNWIND (frame, -1, ENOTCONN, NULL);
-			return 0;
-		}
-		STACK_WIND (frame,
-			    ha_getspec_cbk,
-			    children[i],
-			    children[i]->mops->getspec,
-			    local->pattern,
-			    local->flags);
-		return 0;
-	}
-
-	STACK_UNWIND (frame,
-		      op_ret,
-		      op_errno,
-		      spec_data);
-	return 0;
-}
-
-int32_t
-ha_getspec (call_frame_t *frame,
-	    xlator_t *this,
-	    const char *key,
-	    int32_t flags)
-{
-	ha_local_t *local = NULL;
-	ha_private_t *pvt = NULL;
-	xlator_t **children = NULL;
-	int i = 0;
-        int32_t op_errno = EINVAL;
-
-	local = frame->local = GF_CALLOC (1, sizeof (*local), 
-                                          gf_ha_mt_ha_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto err;
-        }
-
-	pvt = this->private;
-	children = pvt->children;
-
-	for (i = 0; i < pvt->child_count; i++) {
-		if (pvt->state[i])
-			break;
-	}
-
-	if (i == pvt->child_count) {
-                op_errno = ENOTCONN;
-                goto err;
-	}
-	local->flags = flags;
-	local->pattern = (char *)key;
-
-	STACK_WIND (frame,
-		    ha_getspec_cbk,
-		    children[i],
-		    children[i]->mops->getspec,
-		    key, flags);
-	return 0;
-err:
-        local = frame->local;
-        frame->local = NULL;
-
-        STACK_UNWIND (frame, -1, ENOTCONN, NULL);
-
-        ha_local_wipe (local);
-        return 0;
-
-}
-
-int32_t
-ha_closedir (xlator_t *this,
-	     fd_t *fd)
-{
-	hafd_t *hafdp = NULL;
-	int op_errno = 0;
-	uint64_t tmp_hafdp = 0;
-
-	op_errno = fd_ctx_del (fd, this, &tmp_hafdp);
-	if (op_errno != 0) {
-		gf_log (this->name, GF_LOG_ERROR, "fd_ctx_del() error");
-		return 0;
-	}
-	hafdp = (hafd_t *)(long)tmp_hafdp;
-
-	GF_FREE (hafdp->fdstate);
-	GF_FREE (hafdp->path);
-	LOCK_DESTROY (&hafdp->lock);
-	return 0;
-}
-
-int32_t
-ha_close (xlator_t *this,
-	  fd_t *fd)
-{
-	hafd_t *hafdp = NULL;
-	int op_errno = 0;
-	uint64_t tmp_hafdp = 0;
-
-	op_errno = fd_ctx_del (fd, this, &tmp_hafdp);
-	if (op_errno != 0) {
-		gf_log (this->name, GF_LOG_ERROR, "fd_ctx_del() error");
-		return 0;
-	}
-	hafdp = (hafd_t *)(long)tmp_hafdp;
-
-	GF_FREE (hafdp->fdstate);
-	GF_FREE (hafdp->path);
-	LOCK_DESTROY (&hafdp->lock);
-	return 0;
-}
-
-/* notify */
-int32_t
-notify (xlator_t *this,
-	int32_t event,
-	void *data,
-	...)
-{
-	ha_private_t *pvt = NULL;
-	int32_t i = 0, upcnt = 0;
-
-	pvt = this->private;
-	if (pvt == NULL) {
-		gf_log (this->name, GF_LOG_DEBUG, "got notify before init()");
-		return 0;
-	}
-
-	switch (event)
-	{
-	case GF_EVENT_CHILD_DOWN:
-	{
-		for (i = 0; i < pvt->child_count; i++) {
-			if (data == pvt->children[i])
-				break;
-		}
-	        gf_log (this->name, GF_LOG_DEBUG, "GF_EVENT_CHILD_DOWN from %s", pvt->children[i]->name);
-		pvt->state[i] = 0;
-		for (i = 0; i < pvt->child_count; i++) {
-			if (pvt->state[i])
-				break;
-		}
-		if (i == pvt->child_count) {
-			default_notify (this, event, data);
-		}
-	}
-	break;
-	case GF_EVENT_CHILD_UP:
-	{
-		for (i = 0; i < pvt->child_count; i++) {
-			if (data == pvt->children[i])
-				break;
-		}
-
-		gf_log (this->name, GF_LOG_DEBUG, "GF_EVENT_CHILD_UP from %s", pvt->children[i]->name);
-
-		pvt->state[i] = 1;
-
-		for (i = 0; i < pvt->child_count; i++) {
-			if (pvt->state[i])
-				upcnt++;
-		}
-
-		if (upcnt == 1) {
-			default_notify (this, event, data);
-		}
-	}
-	break;
-
-	default:
-	{
-		default_notify (this, event, data);
-	}
-	}
-
-	return 0;
-}
-
-int32_t
-mem_acct_init (xlator_t *this)
-{
-        int     ret = -1;
-
-        if (!this)
-                return ret;
-
-        ret = xlator_mem_acct_init (this, gf_ha_mt_end + 1);
-        
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
-                                "failed");
-                return ret;
-        }
-
-        return ret;
-}
-
-int
-init (xlator_t *this)
-{
-	ha_private_t *pvt = NULL;
-	xlator_list_t *trav = NULL;
-	int count = 0, ret = 0;
-
-
-	if (!this->children) {
-		gf_log (this->name,GF_LOG_ERROR, 
-			"FATAL: ha should have one or more child defined");
-		return -1;
-	}
-
-	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"dangling volume. check volfile ");
-	}
-  
-	trav = this->children;
-	pvt = GF_CALLOC (1, sizeof (ha_private_t), gf_ha_mt_ha_private_t);
-
-	ret = dict_get_int32 (this->options, "preferred-subvolume", 
-			      &pvt->pref_subvol);
-	if (ret < 0) {
-		pvt->pref_subvol = -1;
-	}
-
-	trav = this->children;
-	while (trav) {
-		count++;
-		trav = trav->next;
-	}
-
-	pvt->child_count = count;
-	pvt->children = GF_CALLOC (count, sizeof (xlator_t*), 
-                                   gf_ha_mt_xlator_t);
-
-	trav = this->children;
-	count = 0;
-	while (trav) {
-		pvt->children[count] = trav->xlator;
-		count++;
-		trav = trav->next;
-	}
-
-	pvt->state = GF_CALLOC (1, count, gf_ha_mt_char);
-	this->private = pvt;
-	return 0;
-}
-
-void
-fini (xlator_t *this)
-{
-	ha_private_t *priv = NULL;
-	priv = this->private;
-	GF_FREE (priv);
-	return;
-}
-
-
-struct xlator_fops fops = {
-	.lookup      = ha_lookup,
-	.stat        = ha_stat,
-	.readlink    = ha_readlink,
-	.mknod       = ha_mknod,
-	.mkdir       = ha_mkdir,
-	.unlink      = ha_unlink,
-	.rmdir       = ha_rmdir,
-	.symlink     = ha_symlink,
-	.rename      = ha_rename,
-	.link        = ha_link,
-	.truncate    = ha_truncate,
-	.create      = ha_create,
-	.open        = ha_open,
-	.readv       = ha_readv,
-	.writev      = ha_writev,
-	.statfs      = ha_statfs,
-	.flush       = ha_flush,
-	.fsync       = ha_fsync,
-	.setxattr    = ha_setxattr,
-	.getxattr    = ha_getxattr,
-	.removexattr = ha_removexattr,
-	.opendir     = ha_opendir,
-	.readdir     = ha_readdir,
-	.readdirp    = ha_readdirp,
-	.getdents    = ha_getdents,
-	.fsyncdir    = ha_fsyncdir,
-	.access      = ha_access,
-	.ftruncate   = ha_ftruncate,
-	.fstat       = ha_fstat,
-	.lk          = ha_lk,
-	.setdents    = ha_setdents,
-	.lookup_cbk  = ha_lookup_cbk,
-	.checksum    = ha_checksum,
-	.xattrop     = ha_xattrop,
-	.fxattrop    = ha_fxattrop,
-        .setattr     = ha_setattr,
-        .fsetattr    = ha_fsetattr,
-};
-
-struct xlator_cbks cbks = {
-	.release    = ha_close,
-	.releasedir = ha_closedir,
-	.forget     = ha_forget,
-};
diff --git a/xlators/cluster/ha/src/ha.h b/xlators/cluster/ha/src/ha.h
deleted file mode 100644
index 5fb7b258579..00000000000
--- a/xlators/cluster/ha/src/ha.h
+++ /dev/null
@@ -1,63 +0,0 @@
-/*
-  Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef __HA_H_
-#define __HA_H_
-
-#include "ha-mem-types.h"
-
-typedef struct {
-	call_stub_t *stub;
-	int32_t op_ret, op_errno;
-	int32_t active, tries, revalidate, revalidate_error;
-	int32_t call_count;
-	char *state, *pattern;
-	dict_t *dict;
-	loc_t loc;
-	struct iatt buf;
-        struct iatt postparent;
-        struct iatt preparent;
-	fd_t *fd;
-	inode_t *inode;
-	int32_t flags;
-	int32_t first_success;
-} ha_local_t;
-
-typedef struct {
-	char *state;
-	xlator_t **children;
-	int child_count, pref_subvol;
-} ha_private_t;
-
-typedef struct {
-	char *fdstate;
-	char *path;
-	gf_lock_t lock;
-	int active;
-} hafd_t;
-
-#define HA_ACTIVE_CHILD(this, local) (((ha_private_t *)this->private)->children[local->active])
-
-extern int ha_alloc_init_fd (call_frame_t *frame, fd_t *fd);
-
-extern int ha_handle_cbk (call_frame_t *frame, void *cookie, int op_ret, int op_errno) ;
-
-extern int ha_alloc_init_inode (call_frame_t *frame, inode_t *inode);
-
-#endif
diff --git a/xlators/cluster/map/src/Makefile.am b/xlators/cluster/map/src/Makefile.am
deleted file mode 100644
index 26e19137a8b..00000000000
--- a/xlators/cluster/map/src/Makefile.am
+++ /dev/null
@@ -1,15 +0,0 @@
-xlator_LTLIBRARIES = map.la
-xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/testing/cluster
-
-map_la_LDFLAGS = -module -avoidversion 
-
-map_la_SOURCES = map.c map-helper.c
-map_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
-
-noinst_HEADERS = map.h
-
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS) \
-	    -I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
-
-CLEANFILES = 
-
diff --git a/xlators/cluster/map/src/map-helper.c b/xlators/cluster/map/src/map-helper.c
deleted file mode 100644
index 41a81b617f9..00000000000
--- a/xlators/cluster/map/src/map-helper.c
+++ /dev/null
@@ -1,358 +0,0 @@
-/*
-  Copyright (c) 2009-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "xlator.h"
-#include "map.h"
-
-
-xlator_t *
-map_subvol_next (xlator_t *this, xlator_t *prev)
-{
-	map_private_t *priv = NULL;
-	xlator_t      *next = NULL;
-	int            i = 0;
-
-	priv = this->private;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (priv->xlarray[i].xl == prev) {
-			if ((i + 1) < priv->child_count)
-				next = priv->xlarray[i + 1].xl;
-			break;
-		}
-	}
-
-	return next;
-}
-
-int
-map_subvol_cnt (xlator_t *this, xlator_t *subvol)
-{
-	int i = 0;
-	int ret = -1;
-	map_private_t *priv = NULL;
-
-	priv = this->private;
-
-	for (i = 0; i < priv->child_count; i++) {
-		if (subvol == priv->xlarray[i].xl) {
-			ret = i;
-			break;
-		}
-	}
-
-	return ret;
-}
-
-int
-map_itransform (xlator_t *this, xlator_t *subvol, uint64_t x, uint64_t *y_p)
-{
-	map_private_t *priv = NULL;
-	int         cnt = 0;
-	int         max = 0;
-	uint64_t    y = 0;
-
-	if (x == ((uint64_t) -1)) {
-		y = (uint64_t) -1;
-		goto out;
-	}
-
-	priv = this->private;
-
-	max = priv->child_count;
-	cnt = map_subvol_cnt (this, subvol);
-
-	y = ((x * max) + cnt);
-
-out:
-	if (y_p)
-		*y_p = y;
-
-	return 0;
-}
-
-
-int
-map_deitransform (xlator_t *this, uint64_t y, xlator_t **subvol_p,
-		  uint64_t *x_p)
-{
-	int         cnt = 0;
-	int         max = 0;
-	uint64_t    x = 0;
-	xlator_t   *subvol = 0;
-	map_private_t *priv = NULL;
-
-	priv = this->private;
-	max = priv->child_count;
-
-	cnt = y % max;
-	x   = y / max;
-
-	subvol = priv->xlarray[cnt].xl;
-
-	if (subvol_p)
-		*subvol_p = subvol;
-
-	if (x_p)
-		*x_p = x;
-
-	return 0;
-}
-
-
-xlator_t *
-get_mapping_subvol_from_path (xlator_t *this, const char *path) 
-{
-	map_private_t      *priv = NULL;
-	struct map_pattern *map = NULL;
-
-	/* To make sure we handle '/' properly */
-	if (!strcmp (path, "/"))
-		return NULL;
-
-	priv = this->private;
-
-	map = priv->map;
-	while (map) {
-		if (!strncmp (map->directory, path, map->dir_len)) {
-			if ((path[map->dir_len] == '/') ||
-			    (path[map->dir_len] == '\0')) {
-				return map->xl;
-			}
-		}
-		
-		map = map->next;
-	}
-
-	return priv->default_xl;
-}
-
-xlator_t *
-get_mapping_subvol_from_ctx (xlator_t *this, inode_t *inode)
-{
-	uint64_t subvol = 0;
-	int      ret    = -1;
-
-	ret = inode_ctx_get (inode, this, &subvol);
-	if (ret != 0) 
-		return NULL;
-
-	return (xlator_t *)(long)subvol;
-}
-
-int
-check_multiple_volume_entry (xlator_t *this, 
-			     xlator_t *subvol)
-{
-	int ret = -1;
-	int idx = 0;
-	map_private_t *priv = NULL;
-
-	priv = this->private;
-	
-	for (idx = 0; idx < priv->child_count; idx++) {
-		if (priv->xlarray[idx].xl == subvol) {
-			if (priv->xlarray[idx].mapped) {
-				gf_log (this->name, GF_LOG_ERROR,
-					"subvolume '%s' is already mapped",
-					subvol->name);
-				goto out;
-			}
-			priv->xlarray[idx].mapped = 1;
-			ret = 0;
-			goto out;
-		}
-	}
-
-	gf_log (this->name, GF_LOG_ERROR,
-		"subvolume '%s' is not found",
-		subvol->name);
-	
- out:
-	return ret;
-}
-
-int
-verify_dir_and_assign_subvol (xlator_t *this, 
-			      const char *directory, 
-			      const char *subvol)
-{
-	int            default_flag = 0;
-	int            ret  = -1;
-	int            idx  = 0;
-	map_private_t *priv = NULL;
-	xlator_list_t *trav = NULL;
-	struct map_pattern *tmp_map = NULL;
-
-	priv = this->private;
-
-	/* check if directory is valid, ie, its a top level dir, and 
-	 * not includes a '*' in it.
-	 */
-	if (!strcmp ("*", directory)) {
-		default_flag = 1;
-	} else {
-		if (directory[0] != '/') {
-			gf_log (this->name, GF_LOG_ERROR,
-				"map takes absolute path, starting with '/'. "
-				"not '%s'", directory);
-			goto out;
-		}
-		for (idx = 1; idx < (strlen (directory) - 1); idx++) {
-			if (directory[idx] == '/') {
-				gf_log (this->name, GF_LOG_ERROR,
-					"map takes only top level directory, "
-					"not '%s'", directory);
-				goto out;
-			}
-		}
-	}
-
-	/* Assign proper subvolume */
-	trav = this->children;
-	while (trav) {
-		if (!strcmp (trav->xlator->name, subvol)) {
-			
-			/* Check if there is another directory for 
-			 * same volume, if yes, return error.
-			 */
-			ret = check_multiple_volume_entry (this, 
-							   trav->xlator);
-			if (ret != 0) {
-				goto out;
-			}
-
-			ret = 0;
-			if (default_flag) {
-				if (priv->default_xl) {
-					ret = -1;
-					gf_log (this->name, GF_LOG_ERROR,
-						"'*' specified more than "
-						"once. don't confuse me!!!");
-				}
-
-				priv->default_xl = trav->xlator;
-				goto out;
-			}
-
-			tmp_map = GF_CALLOC (1, sizeof (struct map_pattern),
-                                             gf_map_mt_map_pattern);
-			tmp_map->xl = trav->xlator;
-			tmp_map->dir_len = strlen (directory);
-
-			/* make sure that the top level directory starts 
-			 * with '/' and ends without '/'
-			 */
-			tmp_map->directory = gf_strdup (directory);
-			if (directory[tmp_map->dir_len - 1] == '/') {
-				tmp_map->dir_len--;
-			}
-
-			if (!priv->map) 
-				priv->map = tmp_map;
-			else {
-				struct map_pattern *trav_map = NULL;
-				trav_map = priv->map;
-				while (trav_map->next)
-					trav_map = trav_map->next;
-				trav_map->next = tmp_map;
-			}
-			
-			goto out;
-		}
-
-		trav = trav->next;
-	}
-
-	gf_log (this->name, GF_LOG_ERROR, 
-		"map volume '%s' is not proper subvolume", subvol);
-
- out:
-	return ret;
-}
-
-int 
-assign_default_subvol (xlator_t *this, const char *default_xl)
-{
-	int ret = -1;
-	map_private_t *priv = NULL;
-	xlator_list_t *trav = NULL;
-
-	priv = this->private;
-	trav = this->children;
-
-	while (trav) {
-		if (!strcmp (trav->xlator->name, default_xl)) {
-			ret = check_multiple_volume_entry (this, 
-							   trav->xlator);
-			if (ret != 0) {
-				goto out;
-			}
-			if (priv->default_xl)
-				gf_log (this->name, GF_LOG_WARNING,
-					"default-volume option provided, "
-					"overriding earlier '*' option");
-			priv->default_xl = trav->xlator;
-			return 0;
-		}
-		trav = trav->next;
-	}
-
-	gf_log (this->name, GF_LOG_ERROR,
-		"default-volume value is not an valid subvolume. check again");
- out:
-	return -1;
-}
-
-void
-verify_if_all_subvolumes_got_used (xlator_t *this)
-{
-	int idx = 0;
-	map_private_t *priv = NULL;
-
-	priv = this->private;
-	
-	for (idx = 0; idx < priv->child_count; idx++) {
-		if (!priv->xlarray[idx].mapped) {
-			if (!priv->default_xl) {
-				priv->default_xl = priv->xlarray[idx].xl;
-				priv->xlarray[idx].mapped = 1;
-			} else {
-				gf_log (this->name, GF_LOG_WARNING,
-					"subvolume '%s' is not mapped to "
-					"any directory",
-					priv->xlarray[idx].xl->name);
-			}
-		}
-	}
-
-	if (!priv->default_xl) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"default subvolume not specified, filesystem "
-			"may not work properly. Check 'map' translator "
-			"documentation for more info");
-	}
-
-	return ;
-}
diff --git a/xlators/cluster/map/src/map-mem-types.h b/xlators/cluster/map/src/map-mem-types.h
deleted file mode 100644
index 10f0dc60dd4..00000000000
--- a/xlators/cluster/map/src/map-mem-types.h
+++ /dev/null
@@ -1,35 +0,0 @@
-
-/*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-
-#ifndef __MAP_MEM_TYPES_H__
-#define __MAP_MEM_TYPES_H__
-
-#include "mem-types.h"
-
-enum gf_map_mem_types_ {
-        gf_map_mt_map_private_t = gf_common_mt_end + 1,
-        gf_map_mt_map_local_t,
-        gf_map_mt_map_xlator_array,
-        gf_map_mt_map_pattern,
-        gf_map_mt_end
-};
-#endif
-
diff --git a/xlators/cluster/map/src/map.c b/xlators/cluster/map/src/map.c
deleted file mode 100644
index 62987fc2c26..00000000000
--- a/xlators/cluster/map/src/map.c
+++ /dev/null
@@ -1,2577 +0,0 @@
-/*
-  Copyright (c) 2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "xlator.h"
-#include "map.h"
-
-/* TODO : 
- *     -> support for 'get' 'put' API in through xattrs.
- *     -> define the behavior of notify()
- */
-
-static int32_t
-map_stat_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  struct iatt *buf)
-
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, buf->ia_ino, &buf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, buf);
-	return 0;
-}
-
-static int32_t
-map_setattr_cbk (call_frame_t *frame,
-                 void *cookie,
-                 xlator_t *this,
-                 int32_t op_ret,
-                 int32_t op_errno,
-                 struct iatt *statpre,
-                 struct iatt *statpost)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-
-	map_itransform (this, prev->this, statpre->ia_ino, &statpre->ia_ino);
-	map_itransform (this, prev->this, statpost->ia_ino, &statpost->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, statpre, statpost);
-	return 0;
-}
-
-static int32_t
-map_fsetattr_cbk (call_frame_t *frame,
-                  void *cookie,
-                  xlator_t *this,
-                  int32_t op_ret,
-                  int32_t op_errno,
-                  struct iatt *statpre,
-                  struct iatt *statpost)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-
-	map_itransform (this, prev->this, statpre->ia_ino, &statpre->ia_ino);
-	map_itransform (this, prev->this, statpost->ia_ino, &statpost->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, statpre, statpost);
-	return 0;
-}
-
-static int32_t
-map_truncate_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno,
-		      struct iatt *prebuf,
-                      struct iatt *postbuf)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, postbuf->ia_ino, &postbuf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, prebuf, postbuf);
-	return 0;
-}
-
-static int32_t
-map_ftruncate_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno,
-		       struct iatt *prebuf,
-                       struct iatt *postbuf)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, postbuf->ia_ino, &postbuf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, prebuf, postbuf);
-	return 0;
-}
-
-
-static int32_t
-map_access_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-static int32_t
-map_readlink_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno,
-		      const char *path,
-                      struct iatt *sbuf)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, path, sbuf);
-	return 0;
-}
-
-static int32_t
-map_unlink_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-                    struct iatt *preparent,
-                    struct iatt *postparent)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, preparent, postparent);
-	return 0;
-}
-
-static int32_t
-map_rmdir_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-                   struct iatt *preparent,
-                   struct iatt *postparent)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, preparent, postparent);
-	return 0;
-}
-
-
-static int32_t
-map_rename_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    struct iatt *buf,
-                    struct iatt *preoldparent,
-                    struct iatt *postoldparent,
-                    struct iatt *prenewparent,
-                    struct iatt *postnewparent)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, buf->ia_ino, &buf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, buf);
-	return 0;
-}
-
-static int32_t
-map_link_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  inode_t *inode,
-                  struct iatt *buf,
-                  struct iatt *preparent,
-                  struct iatt *postparent)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, buf->ia_ino, &buf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, inode,	buf);
-	return 0;
-}
-
-static int32_t
-map_open_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  fd_t *fd)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, fd);
-	return 0;
-}
-
-static int32_t
-map_readv_cbk (call_frame_t *frame,
-               void *cookie,
-               xlator_t *this,
-               int32_t op_ret,
-               int32_t op_errno,
-               struct iovec *vector,
-               int32_t count,
-               struct iatt *stbuf,
-               struct iobref *iobref)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, stbuf->ia_ino, &stbuf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, vector, count, stbuf, iobref);
-	return 0;
-}
-
-static int32_t
-map_writev_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-                    struct iatt *prebuf,
-		    struct iatt *postbuf)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, postbuf->ia_ino, &postbuf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, prebuf, postbuf);
-	return 0;
-}
-
-static int32_t
-map_flush_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-
-static int32_t
-map_fsync_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-                   struct iatt *prebuf,
-                   struct iatt *postbuf)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, prebuf, postbuf);
-	return 0;
-}
-
-
-static int32_t
-map_fstat_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-		   struct iatt *buf)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, buf->ia_ino, &buf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, buf);
-	return 0;
-}
-
-
-static int32_t
-map_getdents_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno,
-		      dir_entry_t *entries,
-		      int32_t count)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, entries, count);
-	return 0;
-}
-
-
-static int32_t
-map_setdents_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-static int32_t
-map_fsyncdir_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-
-static int32_t
-map_setxattr_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-
-static int32_t
-map_fsetxattr_cbk (call_frame_t *frame,
-                       void *cookie,
-                       xlator_t *this,
-                       int32_t op_ret,
-                       int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-
-static int32_t
-map_fgetxattr_cbk (call_frame_t *frame,
-                       void *cookie,
-                       xlator_t *this,
-                       int32_t op_ret,
-                       int32_t op_errno,
-                       dict_t *dict)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, dict);
-	return 0;
-}
-
-
-
-static int32_t
-map_getxattr_cbk (call_frame_t *frame,
-                      void *cookie,
-                      xlator_t *this,
-                      int32_t op_ret,
-                      int32_t op_errno,
-                      dict_t *dict)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, dict);
-	return 0;
-}
-
-int32_t
-map_xattrop_cbk (call_frame_t *frame,
-		     void *cookie,
-		     xlator_t *this,
-		     int32_t op_ret,
-		     int32_t op_errno,
-		     dict_t *dict)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, dict);
-	return 0;
-}
-
-int32_t
-map_fxattrop_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno,
-		      dict_t *dict)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, dict);
-	return 0;
-}
-
-static int32_t
-map_removexattr_cbk (call_frame_t *frame,
-			 void *cookie,
-			 xlator_t *this,
-			 int32_t op_ret,
-			 int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-
-static int32_t
-map_lk_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		struct gf_flock *lock)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, lock);
-	return 0;
-}
-
-
-static int32_t
-map_inodelk_cbk (call_frame_t *frame, void *cookie,
-		     xlator_t *this, int32_t op_ret, int32_t op_errno)
-
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-
-
-static int32_t
-map_finodelk_cbk (call_frame_t *frame, void *cookie,
-		      xlator_t *this, int32_t op_ret, int32_t op_errno)
-
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-
-static int32_t
-map_entrylk_cbk (call_frame_t *frame, void *cookie,
-		     xlator_t *this, int32_t op_ret, int32_t op_errno)
-
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-
-static int32_t
-map_fentrylk_cbk (call_frame_t *frame, void *cookie,
-		      xlator_t *this, int32_t op_ret, int32_t op_errno)
-
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-static int32_t
-map_newentry_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  inode_t *inode,
-                  struct iatt *buf,
-                  struct iatt *preparent,
-                  struct iatt *postparent)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, buf->ia_ino, &buf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, inode, buf);
-	return 0;
-
-}
-
-
-static int32_t
-map_create_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		fd_t *fd,
-		inode_t *inode,
-		struct iatt *buf,
-                struct iatt *preparent,
-                struct iatt *postparent)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, buf->ia_ino, &buf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, fd, inode, buf);
-	return 0;
-}
-
-
-/*
- * map_normalize_stats -
- */
-void
-map_normalize_stats (struct statvfs *buf,
-		     unsigned long bsize,
-		     unsigned long frsize)
-{
-	double factor;
-
-	if (buf->f_bsize != bsize) {
-		factor = ((double) buf->f_bsize) / bsize;
-		buf->f_bsize  = bsize;
-		buf->f_bfree  = (fsblkcnt_t) (factor * buf->f_bfree);
-		buf->f_bavail = (fsblkcnt_t) (factor * buf->f_bavail);
-	}
-  
-	if (buf->f_frsize != frsize) {
-		factor = ((double) buf->f_frsize) / frsize;
-		buf->f_frsize = frsize;
-		buf->f_blocks = (fsblkcnt_t) (factor * buf->f_blocks);
-	}
-}
-
-
-int32_t
-map_statfs_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		struct statvfs *stbuf)
-{
-	struct statvfs *dict_buf = NULL;
-	map_local_t  *local = NULL;
-	int           this_call_cnt = 0;
-	unsigned long bsize;
-	unsigned long frsize;
-
-	local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		this_call_cnt = --local->call_count;
-
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			goto unlock;
-		}
-		local->op_ret = 0;
-		
-		/* when a call is successfull, add it to local->dict */
-		dict_buf = &local->statvfs;
-		
-		if (dict_buf->f_bsize != 0) {
-			bsize  = max (dict_buf->f_bsize, 
-				      stbuf->f_bsize);
-			
-			frsize = max (dict_buf->f_frsize, 
-				      stbuf->f_frsize);
-			map_normalize_stats(dict_buf, bsize, frsize);
-			map_normalize_stats(stbuf, bsize, frsize);
-		} else {
-			dict_buf->f_bsize   = stbuf->f_bsize;
-			dict_buf->f_frsize  = stbuf->f_frsize;
-		}
-		
-		dict_buf->f_blocks += stbuf->f_blocks;
-		dict_buf->f_bfree  += stbuf->f_bfree;
-		dict_buf->f_bavail += stbuf->f_bavail;
-		dict_buf->f_files  += stbuf->f_files;
-		dict_buf->f_ffree  += stbuf->f_ffree;
-		dict_buf->f_favail += stbuf->f_favail;
-		dict_buf->f_fsid    = stbuf->f_fsid;
-		dict_buf->f_flag    = stbuf->f_flag;
-		dict_buf->f_namemax = stbuf->f_namemax;
-	}
-unlock:
-	UNLOCK (&frame->lock);
-
-	if (!this_call_cnt) {
-		STACK_UNWIND (frame, local->op_ret, local->op_errno,
-			      &local->statvfs);
-	}
-
-	return 0;
-}
-
-int32_t
-map_single_lookup_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno,
-		       inode_t *inode,
-		       struct iatt *buf,
-                       dict_t *dict,
-                       struct iatt *postparent)
-{
-        call_frame_t *prev = NULL;
-        prev  = cookie;
-	
-	map_itransform (this, prev->this, buf->ia_ino, &buf->ia_ino);
-
-	STACK_UNWIND (frame, op_ret, op_errno, inode, buf, dict);
-
-	return 0;
-}
-
-int32_t
-map_root_lookup_cbk (call_frame_t *frame,
-                     void *cookie,
-                     xlator_t *this,
-                     int32_t op_ret,
-                     int32_t op_errno,
-                     inode_t *inode,
-                     struct iatt *buf,
-                     dict_t *dict,
-                     struct iatt *postparent)
-{
-	int          callcnt = 0;
-	map_local_t *local = NULL;
-	inode_t     *tmp_inode = NULL;
-	dict_t      *tmp_dict = NULL;
-
-	local = frame->local;
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-		if ((op_ret == 0) && (local->op_ret == -1)) {
-			local->op_ret = 0;
-			local->stbuf = *buf;
-			if (dict)
-				local->dict = dict_ref (dict);
-			local->inode = inode_ref (inode);
-		}
-		if (op_ret == -1)
-			local->op_errno = op_errno;
-		
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		tmp_dict = local->dict;
-		tmp_inode = local->inode;
-
-		STACK_UNWIND (frame, local->op_ret, 
-			      local->op_errno, local->inode, 
-			      &local->stbuf, local->dict);
-
-		inode_unref (local->inode);
-		if (tmp_dict)
-			dict_unref (tmp_dict);
-	}
-
-	return 0;
-}
-
-
-int32_t
-map_opendir_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 fd_t *fd)
-{
-	int callcnt = 0;
-	map_local_t *local = NULL;
-	fd_t *local_fd = NULL;
-
-	local = frame->local;
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			goto unlock;
-		}
-
-		local->op_ret = 0;
-	}
- unlock:
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		local_fd = local->fd;
-		local->fd = NULL;
-
-		STACK_UNWIND (frame, local->op_ret, 
-			      local->op_errno, local_fd);
-
-		fd_unref (local_fd);
-	}
-	return 0;
-}
-
-int32_t
-map_single_readdir_cbk (call_frame_t *frame,
-			void *cookie,
-			xlator_t *this,
-			int32_t op_ret,
-			int32_t op_errno,
-			gf_dirent_t *entries)
-{
-        call_frame_t *prev = NULL;
-	gf_dirent_t  *orig_entry = NULL;
-
-        prev  = cookie;
-
-	list_for_each_entry (orig_entry, &entries->list, list) {
-		map_itransform (this, prev->this, orig_entry->d_ino, 
-				&orig_entry->d_ino);
-	}
-	STACK_UNWIND (frame, op_ret, op_errno, entries);
-	
-	return 0;
-}
-
-
-int32_t
-map_single_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, gf_dirent_t *entries)
-{
-        call_frame_t *prev = NULL;
-	gf_dirent_t  *orig_entry = NULL;
-
-        prev  = cookie;
-
-	list_for_each_entry (orig_entry, &entries->list, list) {
-		map_itransform (this, prev->this, orig_entry->d_ino,
-				&orig_entry->d_ino);
-                orig_entry->d_stat.ia_ino = orig_entry->d_ino;
-	}
-	STACK_UNWIND (frame, op_ret, op_errno, entries);
-	return 0;
-}
-
-int
-map_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int op_ret, int op_errno, gf_dirent_t *orig_entries);
-
-int
-map_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int op_ret, int op_errno, gf_dirent_t *orig_entries);
-
-int
-map_generic_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int op_ret, int op_errno, gf_dirent_t *orig_entries,
-                         int whichop)
-{
-	map_local_t  *local = NULL;
-	gf_dirent_t   entries;
-	gf_dirent_t  *orig_entry = NULL;
-	gf_dirent_t  *entry = NULL;
-	call_frame_t *prev = NULL;
-	xlator_t     *subvol = NULL;
-	xlator_t     *next_subvol = NULL;
-        off_t         next_offset = 0;
-	int           count = 0;
-	fd_t         *local_fd = NULL;
-
-	INIT_LIST_HEAD (&entries.list);
-	prev = cookie;
-	local = frame->local;
-
-	if (op_ret < 0)
-		goto done;
-
-	list_for_each_entry (orig_entry, &orig_entries->list, list) {
-		subvol = prev->this;
-
-		entry = gf_dirent_for_name (orig_entry->d_name);
-		if (!entry) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"memory allocation failed :(");
-			goto unwind;
-		}
-		
-		map_itransform (this, subvol, orig_entry->d_ino,
-				&entry->d_ino);
-		map_itransform (this, subvol, orig_entry->d_off,
-				&entry->d_off);
-		
-                if (whichop == GF_FOP_READDIRP)
-                        entry->d_stat.ia_ino = entry->d_ino;
-		entry->d_type = orig_entry->d_type;
-		entry->d_len  = orig_entry->d_len;
-		
-		list_add_tail (&entry->list, &entries.list);
-		count++;
-                next_offset = orig_entry->d_off;
-	}
-
-	op_ret = count;
-
-done:
-	if (count == 0) {
-                /* non-zero next_offset means that
-                   EOF is not yet hit on the current subvol
-                */
-                if (next_offset == 0) {
-                        next_subvol = map_subvol_next (this, prev->this);
-                } else {
-                        next_subvol = prev->this;
-                }
-
-		if (!next_subvol) {
-			goto unwind;
-		}
-
-                if (whichop == GF_FOP_READDIR)
-                        STACK_WIND (frame, map_readdir_cbk, next_subvol,
-                                    next_subvol->fops->readdir, local->fd,
-                                    local->size, 0);
-                else
-                        STACK_WIND (frame, map_readdirp_cbk, next_subvol,
-                                    next_subvol->fops->readdirp, local->fd,
-                                    local->size, 0);
-		return 0;
-	}
-
-unwind:
-	if (op_ret < 0)
-		op_ret = 0;
-
-	local_fd = local->fd;
-	local->fd = NULL;
-
-	STACK_UNWIND (frame, op_ret, op_errno, &entries);
-
-	fd_unref (local_fd);
-
-	gf_dirent_free (&entries);
-
-        return 0;
-}
-
-
-int
-map_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int op_ret, int op_errno, gf_dirent_t *orig_entries)
-{
-        map_generic_readdir_cbk (frame, cookie, this, op_ret, op_errno,
-                                 orig_entries, GF_FOP_READDIR);
-        return 0;
-}
-
-
-int
-map_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int op_ret, int op_errno, gf_dirent_t *orig_entries)
-{
-        map_generic_readdir_cbk (frame, cookie, this, op_ret, op_errno,
-                                 orig_entries, GF_FOP_READDIRP);
-        return 0;
-}
-
-
-/* Management operations */
-
-static int32_t
-map_checksum_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno,
-		      uint8_t *file_checksum,
-		      uint8_t *dir_checksum)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, file_checksum, dir_checksum);
-	return 0;
-}
-
-
-/* Fops starts here */
-
-int32_t
-map_stat (call_frame_t *frame,
-	  xlator_t *this,
-	  loc_t *loc)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_stat_cbk, subvol, subvol->fops->stat, loc);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_setattr (call_frame_t *frame,
-             xlator_t *this,
-             loc_t *loc,
-             struct iatt *stbuf,
-             int32_t valid)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        GF_VALIDATE_OR_GOTO ("map", this, err);
-        GF_VALIDATE_OR_GOTO (this->name, frame, err);
-        GF_VALIDATE_OR_GOTO (this->name, loc, err);
-        GF_VALIDATE_OR_GOTO (this->name, loc->inode, err);
-        GF_VALIDATE_OR_GOTO (this->name, loc->path, err);
-        GF_VALIDATE_OR_GOTO (this->name, stbuf, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_setattr_cbk, subvol,
-		    subvol->fops->setattr, loc, stbuf, valid);
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_fsetattr (call_frame_t *frame,
-              xlator_t *this,
-              fd_t *fd,
-              struct iatt *stbuf,
-              int32_t valid)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        GF_VALIDATE_OR_GOTO ("map", this, err);
-        GF_VALIDATE_OR_GOTO (this->name, frame, err);
-        GF_VALIDATE_OR_GOTO (this->name, fd, err);
-        GF_VALIDATE_OR_GOTO (this->name, stbuf, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_fsetattr_cbk, subvol,
-		    subvol->fops->fsetattr, fd, stbuf, valid);
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_truncate (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      off_t offset)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_truncate_cbk, subvol, 
-                    subvol->fops->truncate, loc, offset);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_ftruncate (call_frame_t *frame,
-	       xlator_t *this,
-	       fd_t *fd,
-	       off_t offset)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_ftruncate_cbk, subvol,
-		    subvol->fops->ftruncate, fd, offset);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_access (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc,
-	    int32_t mask)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_access_cbk, subvol,
-		    subvol->fops->access, loc, mask);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_readlink (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      size_t size)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_readlink_cbk, subvol,
-		    subvol->fops->readlink, loc, size);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_unlink (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_unlink_cbk, subvol, subvol->fops->unlink, loc);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_rmdir (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *loc)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_rmdir_cbk, subvol, subvol->fops->rmdir, loc);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_rename (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *oldloc,
-	    loc_t *newloc)
-{
-	int32_t op_errno = 1;
-	xlator_t *old_subvol = NULL;
-	xlator_t *new_subvol = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (oldloc, err);
-        VALIDATE_OR_GOTO (oldloc->inode, err);
-        VALIDATE_OR_GOTO (oldloc->path, err);
-        VALIDATE_OR_GOTO (newloc, err);
-
-	old_subvol = get_mapping_subvol_from_ctx (this, oldloc->inode);
-	if (!old_subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	if (newloc->path) {
-		new_subvol = get_mapping_subvol_from_path (this, newloc->path);
-		if (new_subvol && (new_subvol != old_subvol)) {
-			op_errno = EXDEV;
-			goto err;
-		}
-	}
-
-	STACK_WIND (frame, map_rename_cbk, old_subvol,
-		    old_subvol->fops->rename, oldloc, newloc);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_link (call_frame_t *frame,
-	  xlator_t *this,
-	  loc_t *oldloc,
-	  loc_t *newloc)
-{
-	int32_t op_errno = 1;
-	xlator_t *old_subvol = NULL;
-	xlator_t *new_subvol = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (oldloc, err);
-        VALIDATE_OR_GOTO (oldloc->inode, err);
-        VALIDATE_OR_GOTO (oldloc->path, err);
-        VALIDATE_OR_GOTO (newloc, err);
-
-	old_subvol = get_mapping_subvol_from_ctx (this, oldloc->inode);
-	if (!old_subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	if (newloc->path) {
-		new_subvol = get_mapping_subvol_from_path (this, newloc->path);
-		if (new_subvol && (new_subvol != old_subvol)) {
-			op_errno = EXDEV;
-			goto err;
-		}
-	}
-
-	STACK_WIND (frame, map_link_cbk, old_subvol,
-		    old_subvol->fops->link, oldloc, newloc);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_open (call_frame_t *frame,
-	  xlator_t *this,
-	  loc_t *loc,
-	  int32_t flags, fd_t *fd, int wbflags)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_open_cbk, subvol, 
-                    subvol->fops->open, loc, flags, fd, wbflags);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_readv (call_frame_t *frame,
-	   xlator_t *this,
-	   fd_t *fd,
-	   size_t size,
-	   off_t offset)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_readv_cbk, subvol,
-		    subvol->fops->readv, fd, size, offset);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, 0, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_writev (call_frame_t *frame,
-	    xlator_t *this,
-	    fd_t *fd,
-	    struct iovec *vector,
-	    int32_t count,
-	    off_t off,
-            struct iobref *iobref)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_writev_cbk, subvol,
-		    subvol->fops->writev, fd, vector, count, off, iobref);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_flush (call_frame_t *frame,
-	   xlator_t *this,
-	   fd_t *fd)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_flush_cbk, subvol, subvol->fops->flush, fd);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_fsync (call_frame_t *frame,
-	   xlator_t *this,
-	   fd_t *fd,
-	   int32_t flags)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_fsync_cbk, subvol,
-		    subvol->fops->fsync, fd, flags);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_fstat (call_frame_t *frame,
-	   xlator_t *this,
-	   fd_t *fd)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_fstat_cbk, subvol, subvol->fops->fstat, fd);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_getdents (call_frame_t *frame,
-	      xlator_t *this,
-	      fd_t *fd,
-	      size_t size,
-	      off_t offset,
-	      int32_t flag)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_getdents_cbk, subvol,
-		    subvol->fops->getdents, fd, size, offset, flag);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_setdents (call_frame_t *frame,
-	      xlator_t *this,
-	      fd_t *fd,
-	      int32_t flags,
-	      dir_entry_t *entries,
-	      int32_t count)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_setdents_cbk, subvol,
-		    subvol->fops->setdents, fd, flags, entries, count);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_fsyncdir (call_frame_t *frame,
-	      xlator_t *this,
-	      fd_t *fd,
-	      int32_t flags)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_fsyncdir_cbk, subvol,
-		    subvol->fops->fsyncdir, fd, flags);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-
-
-int32_t
-map_setxattr (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      dict_t *dict,
-	      int32_t flags)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_setxattr_cbk, subvol,
-		    subvol->fops->setxattr, loc, dict, flags);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_getxattr (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      const char *name)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_getxattr_cbk, subvol,
-		    subvol->fops->getxattr, loc, name);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_fsetxattr (call_frame_t *frame,
-               xlator_t *this,
-               fd_t *fd,
-               dict_t *dict,
-               int32_t flags)
-{
-	int32_t   op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_fsetxattr_cbk, subvol, 
-                    subvol->fops->fsetxattr, fd, dict, flags);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_fgetxattr (call_frame_t *frame,
-               xlator_t *this,
-               fd_t *fd,
-               const char *name)
-{
-	int32_t   op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_fgetxattr_cbk, subvol,
-		    subvol->fops->fgetxattr, fd, name);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_xattrop (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     gf_xattrop_flags_t flags,
-	     dict_t *dict)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_xattrop_cbk, subvol,
-		    subvol->fops->xattrop, loc, flags, dict);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_fxattrop (call_frame_t *frame,
-	      xlator_t *this,
-	      fd_t *fd,
-	      gf_xattrop_flags_t flags,
-	      dict_t *dict)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_fxattrop_cbk, subvol,
-		    subvol->fops->fxattrop, fd, flags, dict);
-
-        return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_removexattr (call_frame_t *frame,
-		 xlator_t *this,
-		 loc_t *loc,
-		 const char *name)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_removexattr_cbk, subvol,
-		    subvol->fops->removexattr, loc, name);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_lk (call_frame_t *frame,
-	xlator_t *this,
-	fd_t *fd,
-	int32_t cmd,
-	struct gf_flock *lock)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_lk_cbk, subvol,
-		    subvol->fops->lk, fd, cmd, lock);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_inodelk (call_frame_t *frame, xlator_t *this,
-	     const char *volume, loc_t *loc, int32_t cmd, struct gf_flock *lock)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_inodelk_cbk, subvol,
-		    subvol->fops->inodelk, volume, loc, cmd, lock);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_finodelk (call_frame_t *frame, xlator_t *this,
-	      const char *volume, fd_t *fd, int32_t cmd, struct gf_flock *lock)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_finodelk_cbk, subvol,
-		    subvol->fops->finodelk, volume, fd, cmd, lock);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_entrylk (call_frame_t *frame, xlator_t *this,
-	     const char *volume, loc_t *loc, const char *basename,
-	     entrylk_cmd cmd, entrylk_type type)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_entrylk_cbk, subvol,
-		    subvol->fops->entrylk, volume, loc, basename, cmd, type);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_fentrylk (call_frame_t *frame, xlator_t *this,
-	      const char *volume, fd_t *fd, const char *basename,
-	      entrylk_cmd cmd, entrylk_type type)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_fentrylk_cbk, subvol,
-		    subvol->fops->fentrylk, volume, fd, basename, cmd, type);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_checksum (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      int32_t flag)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	STACK_WIND (frame, map_checksum_cbk, subvol,
-		    subvol->fops->checksum, loc, flag);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_mknod (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *loc,
-	   mode_t mode,
-	   dev_t rdev)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-	subvol = get_mapping_subvol_from_path (this, loc->path);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-	
-	op_errno = inode_ctx_put (loc->inode, this, (uint64_t)(long)subvol);
-	if (op_errno != 0) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"%s: failed to set subvolume ptr in inode ctx",
-			loc->path);
-	}
-
-	STACK_WIND (frame, map_newentry_cbk, subvol,
-		    subvol->fops->mknod, loc, mode, rdev);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_mkdir (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *loc,
-	   mode_t mode)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-	subvol = get_mapping_subvol_from_path (this, loc->path);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	op_errno = inode_ctx_put (loc->inode, this, (uint64_t)(long)subvol);
-	if (op_errno != 0) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"%s: failed to set subvolume ptr in inode ctx",
-			loc->path);
-	}
-
-	STACK_WIND (frame, map_newentry_cbk, subvol,
-		    subvol->fops->mkdir, loc, mode);
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_symlink (call_frame_t *frame,
-	     xlator_t *this,
-	     const char *linkpath,
-	     loc_t *loc)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-	subvol = get_mapping_subvol_from_path (this, loc->path);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-	
-	op_errno = inode_ctx_put (loc->inode, this, (uint64_t)(long)subvol);
-	if (op_errno != 0) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"%s: failed to set subvolume ptr in inode ctx",
-			loc->path);
-	}
-
-	STACK_WIND (frame, map_newentry_cbk, subvol,
-		    subvol->fops->symlink, linkpath, loc);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-int32_t
-map_create (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc,
-	    int32_t flags,
-	    mode_t mode, fd_t *fd)
-{
-	int32_t op_errno = 1;
-	xlator_t *subvol   = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-	subvol = get_mapping_subvol_from_path (this, loc->path);
-	if (!subvol) {
-		op_errno = EINVAL;
-		goto err;
-	}
-
-	op_errno = inode_ctx_put (loc->inode, this, (uint64_t)(long)subvol);
-	if (op_errno != 0) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"%s: failed to set subvolume ptr in inode ctx",
-			loc->path);
-	}
-
-	STACK_WIND (frame, map_create_cbk, subvol,
-		    subvol->fops->create, loc, flags, mode, fd);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_lookup (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc,
-	    dict_t *xattr_req)
-{
-	int32_t op_errno = EINVAL;
-	xlator_t *subvol   = NULL;
-	map_local_t *local = NULL;
-	map_private_t *priv = NULL;
-	xlator_list_t *trav = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-	priv = this->private;
-
-	if (loc->inode->ino == 1)
-		goto root_inode;
-
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		subvol = get_mapping_subvol_from_path (this, loc->path);
-		if (!subvol) {
-			goto err;
-		}
-
-		op_errno = inode_ctx_put (loc->inode, this, 
-					  (uint64_t)(long)subvol);
-		if (op_errno != 0) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"%s: failed to set subvolume in inode ctx",
-				loc->path);
-		}
-	}
-
-	/* Just one callback */
-	STACK_WIND (frame, map_single_lookup_cbk, subvol,
-		    subvol->fops->lookup, loc, xattr_req);
-
-	return 0;
-
- root_inode:
-	local = GF_CALLOC (1, sizeof (map_local_t),
-                           gf_map_mt_map_local_t);
-
-	frame->local = local;
-	local->call_count = priv->child_count;
-	local->op_ret = -1;
-
-	trav = this->children;
-	while (trav) {
-		STACK_WIND (frame, map_root_lookup_cbk, trav->xlator,
-			    trav->xlator->fops->lookup, loc, xattr_req);
-		trav = trav->next;
-	}
-
-	return 0;
-
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_statfs (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc)
-{
-	int32_t op_errno = EINVAL;
-	xlator_t *subvol   = NULL;
-	map_local_t *local = NULL;
-	map_private_t *priv = NULL;
-	xlator_list_t *trav = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-	if (loc->inode->ino == 1)
-		goto root_inode;
-	subvol = get_mapping_subvol_from_ctx (this, loc->inode);
-	if (!subvol) {
-		goto err;
-	}
-	
-	/* Just one callback */
-	STACK_WIND (frame, map_statfs_cbk, subvol, subvol->fops->statfs, loc);
-
-	return 0;
-
- root_inode:
-	local = GF_CALLOC (1, sizeof (map_local_t),
-                           gf_map_mt_map_local_t);
-
-	priv = this->private;
-	frame->local = local;
-	local->call_count = priv->child_count;
-	local->op_ret = -1;
-
-	trav = this->children;
-	while (trav) {
-		STACK_WIND (frame, map_statfs_cbk, trav->xlator,
-			    trav->xlator->fops->statfs, loc);
-		trav = trav->next;
-	}
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL);
-
-	return 0;
-}
-
-int32_t
-map_opendir (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc, fd_t *fd)
-{
-	int32_t op_errno = EINVAL;
-	xlator_t *subvol   = NULL;
-	map_local_t *local = NULL;
-	map_private_t *priv = NULL;
-	xlator_list_t *trav = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	if (loc->inode->ino == 1) 
-		goto root_inode;
-	
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		goto err;
-	}
-	
-	/* Just one callback */
-	STACK_WIND (frame, map_opendir_cbk, subvol,
-		    subvol->fops->opendir, loc, fd);
-
-	return 0;
-
- root_inode:
-	local = GF_CALLOC (1, sizeof (map_local_t),
-                           gf_map_mt_map_local_t);
-
-	priv = this->private;
-	frame->local = local;
-	local->call_count = priv->child_count;
-	local->op_ret = -1;
-	local->fd = fd_ref (fd);
-
-	trav = this->children;
-	while (trav) {
-		STACK_WIND (frame, map_opendir_cbk, trav->xlator,
-			    trav->xlator->fops->opendir, loc, fd);
-		trav = trav->next;
-	}
-	
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL);
-
-	return 0;
-}
-
-
-int32_t
-map_do_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-                off_t yoff, int whichop)
-{
-	int32_t        op_errno = EINVAL;
-	xlator_t      *subvol = NULL;
-	map_local_t   *local = NULL;
-	map_private_t *priv = NULL;
-	xlator_t      *xvol = NULL;
-	off_t          xoff = 0;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-	if (fd->inode->ino == 1) 
-		goto root_inode;
-
-	subvol = get_mapping_subvol_from_ctx (this, fd->inode);
-	if (!subvol) {
-		goto err;
-	}
-	
-	/* Just one callback */
-        if (whichop == GF_FOP_READDIR)
-                STACK_WIND (frame, map_single_readdir_cbk, subvol,
-                            subvol->fops->readdir, fd, size, yoff);
-        else
-                STACK_WIND (frame, map_single_readdirp_cbk, subvol,
-                            subvol->fops->readdirp, fd, size, yoff);
-
-	return 0;
-
- root_inode:
-	/* readdir on '/' */
-	local = GF_CALLOC (1, sizeof (map_local_t),
-                           gf_map_mt_map_local_t);
-	if (!local) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"memory allocation failed :(");
-		op_errno = ENOMEM;
-		goto err;
-	}
-	
-	priv = this->private;
-	frame->local = local;
-	local->op_errno = ENOENT;
-	local->op_ret = -1;
-
-	local->fd = fd_ref (fd);
-	local->size = size;
-
-	map_deitransform (this, yoff, &xvol, (uint64_t *)&xoff);
-
-        if (whichop == GF_FOP_READDIR)
-                STACK_WIND (frame, map_readdir_cbk, xvol, xvol->fops->readdir,
-                            fd, size, xoff);
-        else
-                STACK_WIND (frame, map_readdirp_cbk, xvol, xvol->fops->readdirp,
-                            fd, size, xoff);
-
-	return 0;
- err:
-	STACK_UNWIND (frame, -1, op_errno, NULL);
-
-	return 0;
-}
-
-
-
-int32_t
-map_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-             off_t yoff)
-{
-        map_do_readdir (frame, this, fd, size, yoff, GF_FOP_READDIR);
-        return 0;
-}
-
-int32_t
-map_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-              off_t yoff)
-{
-        map_do_readdir (frame, this, fd, size, yoff, GF_FOP_READDIRP);
-        return 0;
-}
-
-
-void
-fini (xlator_t *this)
-{
-	map_private_t *priv = NULL;
-	struct map_pattern *trav_map = NULL;
-	struct map_pattern *tmp_map  = NULL;
-
-	priv = this->private;
-
-	if (priv) {
-		if (priv->xlarray)
-			GF_FREE (priv->xlarray);
-
-		trav_map = priv->map;
-		while (trav_map) {
-			tmp_map = trav_map;
-			trav_map = trav_map->next;
-			GF_FREE (tmp_map);
-		}
-
-		GF_FREE(priv);
-	}
-
-	return;
-}
-
-int32_t
-mem_acct_init (xlator_t *this)
-{
-        int     ret = -1;
-
-        if (!this)
-                return ret;
-
-        ret = xlator_mem_acct_init (this, gf_map_mt_end + 1);
-        
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
-                                "failed");
-                return ret;
-        }
-
-        return ret;
-}
-
-int
-init (xlator_t *this)
-{
-	map_private_t *priv = NULL;
-	xlator_list_t *trav = NULL;
-	int   count = 0;
-	int   ret = -1;
-	char *pattern_string = NULL;
-	char *map_pair_str = NULL;
-	char *tmp_str = NULL;
-	char *tmp_str1 = NULL;
-	char *dup_map_pair = NULL;
-	char *dir_str = NULL;
-	char *subvol_str = NULL;
-	char *map_xl = NULL;
-
-
-	if (!this->children) {
-		gf_log (this->name,GF_LOG_ERROR,
-			"FATAL: map should have one or more child defined");
-		return -1;
-	}
-
-	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"dangling volume. check volfile ");
-	}
-
-	priv = GF_CALLOC (1, sizeof (map_private_t),
-                          gf_map_mt_map_private_t);
-	this->private = priv;
-
-	/* allocate xlator array */
-	trav = this->children;
-	while (trav) {
-		count++;
-		trav = trav->next;
-	}
-	priv->xlarray = GF_CALLOC (1, sizeof (struct map_xlator_array) * count,
-                                   gf_map_mt_map_xlator_array);
-	priv->child_count = count;
-
-	/* build xlator array */
-	count = 0;
-	trav = this->children;
-	while (trav) {
-		priv->xlarray[count++].xl = trav->xlator;
-		trav = trav->next;
-	}
-
-	/* map dir1:brick1;dir2:brick2;dir3:brick3;*:brick4 */
-	ret = dict_get_str (this->options, "map-directory", &pattern_string);
-	if (ret != 0) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"map.pattern not given, can't continue");
-		goto err;
-	}
-	map_pair_str = strtok_r (pattern_string, ";", &tmp_str);
-	while (map_pair_str) {
-		dup_map_pair = gf_strdup (map_pair_str);
-		dir_str = strtok_r (dup_map_pair, ":", &tmp_str1);
-		if (!dir_str) {
-			gf_log (this->name, GF_LOG_ERROR, 
-				"directory string invalid");
-			goto err;
-		}
-		subvol_str = strtok_r (NULL, ":", &tmp_str1);
-		if (!subvol_str) {
-			gf_log (this->name, GF_LOG_ERROR, 
-				"mapping subvolume string invalid");
-			goto err;
-		}
-		ret = verify_dir_and_assign_subvol (this, 
-						    dir_str, 
-						    subvol_str);
-		if (ret != 0) {
-			gf_log (this->name, GF_LOG_ERROR, 
-				"verification failed");
-			goto err;
-		}
-		
-		GF_FREE (dup_map_pair);
-
-		map_pair_str = strtok_r (NULL, ";", &tmp_str);
-	}
-
-	/* default-volume brick4 */
-	ret = dict_get_str (this->options, "default-volume", &map_xl);
-	if (ret == 0) {
-		ret = assign_default_subvol (this, map_xl);
-		if (ret != 0) {
-			gf_log (this->name, GF_LOG_ERROR, 
-				"assigning default failed");
-			goto err;
-		}
-	}
-
-	verify_if_all_subvolumes_got_used (this);
-	
-	return 0;
- err:
-	fini (this);
-	return -1;
-}
-
-
-struct xlator_fops fops = {
-	.lookup      = map_lookup,
-	.mknod       = map_mknod,
-	.create      = map_create,
-
-	.stat        = map_stat,
-	.fstat       = map_fstat,
-	.truncate    = map_truncate,
-	.ftruncate   = map_ftruncate,
-	.access      = map_access,
-	.readlink    = map_readlink,
-	.setxattr    = map_setxattr,
-	.getxattr    = map_getxattr,
-	.fsetxattr   = map_fsetxattr,
-	.fgetxattr   = map_fgetxattr,
-	.removexattr = map_removexattr,
-	.open        = map_open,
-	.readv       = map_readv,
-	.writev      = map_writev,
-	.flush       = map_flush,
-	.fsync       = map_fsync,
-	.statfs      = map_statfs,
-	.lk          = map_lk,
-	.opendir     = map_opendir,
-	.readdir     = map_readdir,
-	.readdirp    = map_readdirp,
-	.fsyncdir    = map_fsyncdir,
-	.symlink     = map_symlink,
-	.unlink      = map_unlink,
-	.link        = map_link,
-	.mkdir       = map_mkdir,
-	.rmdir       = map_rmdir,
-	.rename      = map_rename,
-	.inodelk     = map_inodelk,
-	.finodelk    = map_finodelk,
-	.entrylk     = map_entrylk,
-	.fentrylk    = map_fentrylk,
-	.xattrop     = map_xattrop,
-	.fxattrop    = map_fxattrop,
-	.setdents    = map_setdents,
-	.getdents    = map_getdents,
-	.checksum    = map_checksum,
-        .setattr     = map_setattr,
-        .fsetattr    = map_fsetattr,
-};
-
-struct xlator_cbks cbks = {
-};
-
-struct volume_options options[] = {
-	{ .key   = {"map-directory"},  
-	  .type  = GF_OPTION_TYPE_ANY 
-	},
-	{ .key   = {"default-volume"},  
-	  .type  = GF_OPTION_TYPE_XLATOR 
-	},
-	
-	{ .key = {NULL} }
-};
diff --git a/xlators/cluster/map/src/map.h b/xlators/cluster/map/src/map.h
deleted file mode 100644
index a4543776bee..00000000000
--- a/xlators/cluster/map/src/map.h
+++ /dev/null
@@ -1,77 +0,0 @@
-/*
-  Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef __MAP_H__
-#define __MAP_H__
-
-#include "xlator.h"
-#include "map-mem-types.h"
-
-struct map_pattern {
-	struct map_pattern *next;
-	xlator_t           *xl;
-	char               *directory;
-	int                 dir_len;
-};
-
-struct map_xlator_array {
-	xlator_t *xl;
-	int       mapped; /* yes/no */
-};
-
-typedef struct {
-	struct map_pattern      *map;
-	xlator_t                *default_xl;
-	struct map_xlator_array *xlarray;
-	int                      child_count;
-} map_private_t;
-
-typedef struct {
-	int32_t        op_ret;
-	int32_t        op_errno;
-	int            call_count;
-	struct statvfs statvfs;
-	struct iatt    stbuf;
-	inode_t       *inode;
-	dict_t        *dict;
-	fd_t          *fd;
-
-	size_t        size;
-} map_local_t;
-
-xlator_t *map_subvol_next (xlator_t *this, xlator_t *prev);
-int map_subvol_cnt (xlator_t *this, xlator_t *subvol);
-
-int map_itransform (xlator_t *this, xlator_t *subvol, 
-		    uint64_t x, uint64_t *y_p);
-int map_deitransform (xlator_t *this, uint64_t y, 
-		      xlator_t **subvol_p, uint64_t *x_p);
-
-
-xlator_t *get_mapping_subvol_from_path (xlator_t *this, const char *path);
-xlator_t *get_mapping_subvol_from_ctx (xlator_t *this, inode_t *inode);
-
-int check_multiple_volume_entry (xlator_t *this, xlator_t *subvol);
-int verify_dir_and_assign_subvol (xlator_t *this, 
-				  const char *directory, const char *subvol);
-int assign_default_subvol (xlator_t *this, const char *default_xl);
-void verify_if_all_subvolumes_got_used (xlator_t *this);
-
-
-#endif /* __MAP_H__ */
diff --git a/xlators/cluster/stripe/src/Makefile.am b/xlators/cluster/stripe/src/Makefile.am
deleted file mode 100644
index 6d4fae4d268..00000000000
--- a/xlators/cluster/stripe/src/Makefile.am
+++ /dev/null
@@ -1,16 +0,0 @@
-
-xlator_LTLIBRARIES = stripe.la
-xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/cluster
-
-stripe_la_LDFLAGS = -module -avoidversion
-
-stripe_la_SOURCES = stripe.c
-stripe_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
-
-noinst_HEADERS = stripe.h stripe-mem-types.h
-
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS)\
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
-
-CLEANFILES = 
-
diff --git a/xlators/cluster/stripe/src/stripe-mem-types.h b/xlators/cluster/stripe/src/stripe-mem-types.h
deleted file mode 100644
index 9479fdc800f..00000000000
--- a/xlators/cluster/stripe/src/stripe-mem-types.h
+++ /dev/null
@@ -1,40 +0,0 @@
-
-/*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-
-#ifndef __STRIPE_MEM_TYPES_H__
-#define __STRIPE_MEM_TYPES_H__
-
-#include "mem-types.h"
-
-enum gf_stripe_mem_types_ {
-        gf_stripe_mt_stripe_local_t = gf_common_mt_end + 1,
-        gf_stripe_mt_iovec,
-        gf_stripe_mt_readv_replies,
-        gf_stripe_mt_stripe_fd_ctx_t,
-        gf_stripe_mt_char,
-        gf_stripe_mt_int8_t,
-        gf_stripe_mt_xlator_t,
-        gf_stripe_mt_stripe_private_t,
-        gf_stripe_mt_stripe_options,
-        gf_stripe_mt_end
-};
-#endif
-
diff --git a/xlators/cluster/stripe/src/stripe.c b/xlators/cluster/stripe/src/stripe.c
deleted file mode 100644
index a8ac55aa4a2..00000000000
--- a/xlators/cluster/stripe/src/stripe.c
+++ /dev/null
@@ -1,4001 +0,0 @@
-/*
-  Copyright (c) 2007-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-/**
- * xlators/cluster/stripe:
- *    Stripe translator, stripes the data accross its child nodes,
- *    as per the options given in the volfile. The striping works
- *    fairly simple. It writes files at different offset as per
- *    calculation. So, 'ls -l' output at the real posix level will
- *    show file size bigger than the actual size. But when one does
- *    'df' or 'du <file>', real size of the file on the server is shown.
- *
- * WARNING:
- *  Stripe translator can't regenerate data if a child node gets disconnected.
- *  So, no 'self-heal' for stripe. Hence the advice, use stripe only when its
- *  very much necessary, or else, use it in combination with AFR, to have a
- *  backup copy.
- */
-
-#include "stripe.h"
-
-void
-stripe_local_wipe (stripe_local_t *local)
-{
-        if (!local)
-                goto out;
-
-        if (local->loc.path)
-                loc_wipe (&local->loc);
-        if (local->loc2.path)
-                loc_wipe (&local->loc2);
-out:
-        return;
-}
-
-/**
- * stripe_get_matching_bs - Get the matching block size for the given path.
- */
-int32_t
-stripe_get_matching_bs (const char *path, struct stripe_options *opts,
-                        uint64_t default_bs)
-{
-        struct stripe_options *trav       = NULL;
-        char                  *pathname   = NULL;
-        uint64_t               block_size = 0;
-
-        block_size = default_bs;
-
-        if (!path || !opts)
-                goto out;
-
-        /* FIXME: is a strdup really necessary? */
-        pathname = gf_strdup (path);
-        if (!pathname)
-                goto out;
-
-        trav = opts;
-        while (trav) {
-                if (!fnmatch (trav->path_pattern, pathname, FNM_NOESCAPE)) {
-                        block_size = trav->block_size;
-                        break;
-                }
-                trav = trav->next;
-        }
-
-        GF_FREE (pathname);
-
-out:
-        return block_size;
-}
-
-
-
-int32_t
-stripe_sh_chown_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno,
-                     struct iatt *preop, struct iatt *postop)
-{
-        int             callcnt = -1;
-        stripe_local_t *local   = NULL;
-
-        if (!this || !frame || !frame->local) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                stripe_local_wipe (local);
-                STACK_DESTROY (frame->root);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_sh_make_entry_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t op_ret, int32_t op_errno, inode_t *inode,
-                          struct iatt *buf, struct iatt *preparent,
-                          struct iatt *postparent)
-{
-        stripe_local_t *local = NULL;
-        call_frame_t    *prev = NULL;
-
-        if (!frame || !frame->local || !cookie || !this) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        STACK_WIND (frame, stripe_sh_chown_cbk, prev->this,
-                    prev->this->fops->setattr, &local->loc,
-                    &local->stbuf, (GF_SET_ATTR_UID | GF_SET_ATTR_GID));
-
-out:
-        return 0;
-}
-
-int32_t
-stripe_entry_self_heal (call_frame_t *frame, xlator_t *this,
-                        stripe_local_t *local)
-{
-        xlator_list_t    *trav   = NULL;
-        call_frame_t     *rframe = NULL;
-        stripe_local_t   *rlocal = NULL;
-        stripe_private_t *priv   = NULL;
-        dict_t           *dict   = NULL;
-        int               ret    = 0;
-
-        if (!local || !this || !frame) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        if (!(IA_ISREG (local->stbuf.ia_type) ||
-              IA_ISDIR (local->stbuf.ia_type)))
-                return 0;
-
-        priv = this->private;
-        trav = this->children;
-        rframe = copy_frame (frame);
-        if (!rframe) {
-                goto out;
-        }
-        rlocal = GF_CALLOC (1, sizeof (stripe_local_t),
-                            gf_stripe_mt_stripe_local_t);
-        if (!rlocal) {
-                goto out;
-        }
-        rframe->local = rlocal;
-        rlocal->call_count = priv->child_count;
-        loc_copy (&rlocal->loc, &local->loc);
-        memcpy (&rlocal->stbuf, &local->stbuf, sizeof (struct iatt));
-
-        dict = dict_new ();
-        if (!dict)
-                goto out;
-
-        ret = dict_set_static_bin (dict, "gfid-req", local->stbuf.ia_gfid, 16);
-        if (ret)
-                gf_log (this->name, GF_LOG_WARNING,
-                        "failed to set gfid-req");
-
-        while (trav) {
-                if (IA_ISREG (local->stbuf.ia_type)) {
-                        STACK_WIND (rframe, stripe_sh_make_entry_cbk,
-                                    trav->xlator, trav->xlator->fops->mknod,
-                                    &local->loc,
-                                    st_mode_from_ia (local->stbuf.ia_prot,
-                                                     local->stbuf.ia_type), 0,
-                                    dict);
-                }
-                if (IA_ISDIR (local->stbuf.ia_type)) {
-                        STACK_WIND (rframe, stripe_sh_make_entry_cbk,
-                                    trav->xlator, trav->xlator->fops->mkdir,
-                                    &local->loc, st_mode_from_ia (local->stbuf.ia_prot,
-                                                                  local->stbuf.ia_type),
-                                    dict);
-                }
-                trav = trav->next;
-        }
-
-out:
-        if (dict)
-                dict_unref (dict);
-
-        return 0;
-}
-
-int32_t
-stripe_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, inode_t *inode,
-                   struct iatt *buf, dict_t *dict, struct iatt *postparent)
-{
-        int32_t         callcnt = 0;
-        dict_t         *tmp_dict = NULL;
-        inode_t        *tmp_inode = NULL;
-        stripe_local_t *local = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        if (op_errno != ENOENT)
-                                gf_log (this->name, GF_LOG_DEBUG,
-                                        "%s returned error %s",
-                                        prev->this->name,
-                                        strerror (op_errno));
-                        if (local->op_errno != ESTALE)
-                                local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                        if (op_errno == ENOENT)
-                                local->entry_self_heal_needed = 1;
-                }
-
-                if (op_ret >= 0) {
-                        local->op_ret = 0;
-
-                        if (FIRST_CHILD(this) == prev->this) {
-                                local->stbuf      = *buf;
-                                local->postparent = *postparent;
-                                local->inode = inode_ref (inode);
-                                local->dict = dict_ref (dict);
-                        }
-                        local->stbuf_blocks      += buf->ia_blocks;
-                        local->postparent_blocks += postparent->ia_blocks;
-
-                        if (local->stbuf_size < buf->ia_size)
-                                local->stbuf_size = buf->ia_size;
-                        if (local->postparent_size < postparent->ia_size)
-                                local->postparent_size = postparent->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->op_ret == 0 && local->entry_self_heal_needed)
-                        stripe_entry_self_heal (frame, this, local);
-
-                if (local->failed)
-                        local->op_ret = -1;
-
-                tmp_dict  = local->dict;
-                tmp_inode = local->inode;
-
-                if (local->op_ret != -1) {
-                        local->stbuf.ia_blocks      = local->stbuf_blocks;
-                        local->stbuf.ia_size        = local->stbuf_size;
-                        local->postparent.ia_blocks = local->postparent_blocks;
-                        local->postparent.ia_size   = local->postparent_size;
-                }
-
-                stripe_local_wipe (local);
-                STACK_UNWIND_STRICT (lookup, frame, local->op_ret,
-                                     local->op_errno, local->inode,
-                                     &local->stbuf, local->dict,
-                                     &local->postparent);
-
-                if (tmp_inode)
-                        inode_unref (tmp_inode);
-                if (tmp_dict)
-                        dict_unref (tmp_dict);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc,
-               dict_t *xattr_req)
-{
-        stripe_local_t   *local = NULL;
-        xlator_list_t    *trav = NULL;
-        stripe_private_t *priv = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        loc_copy (&local->loc, loc);
-
-        /* Everytime in stripe lookup, all child nodes
-           should be looked up */
-        local->call_count = priv->child_count;
-        while (trav) {
-                STACK_WIND (frame, stripe_lookup_cbk, trav->xlator,
-                            trav->xlator->fops->lookup,
-                            loc, xattr_req);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-
-                if (op_ret == 0) {
-                        local->op_ret = 0;
-
-                        if (FIRST_CHILD(this) == prev->this) {
-                                local->stbuf = *buf;
-                        }
-
-                        local->stbuf_blocks += buf->ia_blocks;
-                        if (local->stbuf_size < buf->ia_size)
-                                local->stbuf_size = buf->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret != -1) {
-                        local->stbuf.ia_size   = local->stbuf_size;
-                        local->stbuf.ia_blocks = local->stbuf_blocks;
-                }
-
-                stripe_local_wipe (local);
-                STACK_UNWIND_STRICT (stat, frame, local->op_ret,
-                                     local->op_errno, &local->stbuf);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_stat (call_frame_t *frame, xlator_t *this, loc_t *loc)
-{
-        xlator_list_t    *trav = NULL;
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        if (priv->first_child_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-        while (trav) {
-                STACK_WIND (frame, stripe_stat_cbk, trav->xlator,
-                            trav->xlator->fops->stat, loc);
-                trav = trav->next;
-        }
-
-        return 0;
-
-err:
-        STACK_UNWIND_STRICT (stat, frame, -1, op_errno, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct statvfs *stbuf)
-{
-        stripe_local_t *local = NULL;
-        int32_t         callcnt = 0;
-
-        if (!this || !frame || !frame->local) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-        local = frame->local;
-
-        LOCK(&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret && (op_errno != ENOTCONN)) {
-                        local->op_errno = op_errno;
-                }
-                if (op_ret == 0) {
-                        struct statvfs *dict_buf = &local->statvfs_buf;
-                        dict_buf->f_bsize   = stbuf->f_bsize;
-                        dict_buf->f_frsize  = stbuf->f_frsize;
-                        dict_buf->f_blocks += stbuf->f_blocks;
-                        dict_buf->f_bfree  += stbuf->f_bfree;
-                        dict_buf->f_bavail += stbuf->f_bavail;
-                        dict_buf->f_files  += stbuf->f_files;
-                        dict_buf->f_ffree  += stbuf->f_ffree;
-                        dict_buf->f_favail += stbuf->f_favail;
-                        dict_buf->f_fsid    = stbuf->f_fsid;
-                        dict_buf->f_flag    = stbuf->f_flag;
-                        dict_buf->f_namemax = stbuf->f_namemax;
-                        local->op_ret = 0;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                STACK_UNWIND_STRICT (statfs, frame, local->op_ret,
-                                     local->op_errno, &local->statvfs_buf);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc)
-{
-        stripe_local_t   *local = NULL;
-        xlator_list_t    *trav = NULL;
-        stripe_private_t *priv = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-
-        trav = this->children;
-        priv = this->private;
-
-        /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        local->op_errno = ENOTCONN;
-        frame->local = local;
-
-        local->call_count = priv->child_count;
-        while (trav) {
-                STACK_WIND (frame, stripe_statfs_cbk, trav->xlator,
-                            trav->xlator->fops->statfs, loc);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (statfs, frame, -1, op_errno, NULL);
-        return 0;
-}
-
-
-
-int32_t
-stripe_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                     struct iatt *postbuf)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-
-                if (op_ret == 0) {
-                        local->op_ret = 0;
-                        if (FIRST_CHILD(this) == prev->this) {
-                                local->pre_buf  = *prebuf;
-                                local->post_buf = *postbuf;
-                        }
-
-                        local->prebuf_blocks  += prebuf->ia_blocks;
-                        local->postbuf_blocks += postbuf->ia_blocks;
-
-                        if (local->prebuf_size < prebuf->ia_size)
-                                local->prebuf_size = prebuf->ia_size;
-
-                        if (local->postbuf_size < postbuf->ia_size)
-                                local->postbuf_size = postbuf->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret != -1) {
-                        local->pre_buf.ia_blocks  = local->prebuf_blocks;
-                        local->pre_buf.ia_size    = local->prebuf_size;
-                        local->post_buf.ia_blocks = local->postbuf_blocks;
-                        local->post_buf.ia_size   = local->postbuf_size;
-                }
-
-                stripe_local_wipe (local);
-                STACK_UNWIND_STRICT (truncate, frame, local->op_ret,
-                                     local->op_errno, &local->pre_buf,
-                                     &local->post_buf);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset)
-{
-        xlator_list_t    *trav = NULL;
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        if (priv->first_child_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-        while (trav) {
-                STACK_WIND (frame, stripe_truncate_cbk, trav->xlator,
-                            trav->xlator->fops->truncate, loc, offset);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (truncate, frame, -1, op_errno, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno,
-                    struct iatt *preop, struct iatt *postop)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-
-                if (op_ret == 0) {
-                        local->op_ret = 0;
-
-                        if (FIRST_CHILD(this) == prev->this) {
-                                local->pre_buf  = *preop;
-                                local->post_buf = *postop;
-                        }
-
-                        local->prebuf_blocks  += preop->ia_blocks;
-                        local->postbuf_blocks += postop->ia_blocks;
-
-                        if (local->prebuf_size < preop->ia_size)
-                                local->prebuf_size = preop->ia_size;
-                        if (local->postbuf_size < postop->ia_size)
-                                local->postbuf_size = postop->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret != -1) {
-                        local->pre_buf.ia_blocks  = local->prebuf_blocks;
-                        local->pre_buf.ia_size    = local->prebuf_size;
-                        local->post_buf.ia_blocks = local->postbuf_blocks;
-                        local->post_buf.ia_size   = local->postbuf_size;
-                }
-
-                stripe_local_wipe (local);
-                STACK_UNWIND_STRICT (setattr, frame, local->op_ret,
-                                     local->op_errno, &local->pre_buf,
-                                     &local->post_buf);
-        }
-out:
-        return 0;
-}
-
-
-int32_t
-stripe_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                struct iatt *stbuf, int32_t valid)
-{
-        xlator_list_t    *trav = NULL;
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        if (priv->first_child_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-        while (trav) {
-                STACK_WIND (frame, stripe_setattr_cbk,
-                            trav->xlator, trav->xlator->fops->setattr,
-                            loc, stbuf, valid);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (setattr, frame, -1, op_errno, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                 struct iatt *stbuf, int32_t valid)
-{
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        xlator_list_t    *trav = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-        while (trav) {
-                STACK_WIND (frame, stripe_setattr_cbk, trav->xlator,
-                            trav->xlator->fops->fsetattr, fd, stbuf, valid);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (fsetattr, frame, -1, op_errno, NULL, NULL);
-        return 0;
-}
-
-int32_t
-stripe_stack_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, struct iatt *buf,
-                         struct iatt *preoldparent, struct iatt *postoldparent,
-                         struct iatt *prenewparent, struct iatt *postnewparent)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-
-                if (op_ret == 0) {
-                        local->op_ret = 0;
-
-                        local->stbuf.ia_blocks      += buf->ia_blocks;
-                        local->preparent.ia_blocks  += preoldparent->ia_blocks;
-                        local->postparent.ia_blocks += postoldparent->ia_blocks;
-                        local->pre_buf.ia_blocks    += prenewparent->ia_blocks;
-                        local->post_buf.ia_blocks   += postnewparent->ia_blocks;
-
-                        if (local->stbuf.ia_size < buf->ia_size)
-                                local->stbuf.ia_size =  buf->ia_size;
-
-                        if (local->preparent.ia_size < preoldparent->ia_size)
-                                local->preparent.ia_size = preoldparent->ia_size;
-
-                        if (local->postparent.ia_size < postoldparent->ia_size)
-                                local->postparent.ia_size = postoldparent->ia_size;
-
-                        if (local->pre_buf.ia_size < prenewparent->ia_size)
-                                local->pre_buf.ia_size = prenewparent->ia_size;
-
-                        if (local->post_buf.ia_size < postnewparent->ia_size)
-                                local->post_buf.ia_size = postnewparent->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                stripe_local_wipe (local);
-                STACK_UNWIND_STRICT (rename, frame, local->op_ret, local->op_errno,
-                                     &local->stbuf, &local->preparent,
-                                     &local->postparent,  &local->pre_buf,
-                                     &local->post_buf);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_first_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, struct iatt *buf,
-                         struct iatt *preoldparent, struct iatt *postoldparent,
-                         struct iatt *prenewparent, struct iatt *postnewparent)
-{
-        stripe_local_t *local = NULL;
-        xlator_list_t  *trav = NULL;
-
-        if (!this || !frame || !frame->local) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                op_errno = EINVAL;
-                goto unwind;
-        }
-
-        if (op_ret == -1) {
-                goto unwind;
-        }
-
-        local = frame->local;
-        trav = this->children;
-
-        local->stbuf      = *buf;
-        local->preparent  = *preoldparent;
-        local->postparent = *postoldparent;
-        local->pre_buf    = *prenewparent;
-        local->post_buf   = *postnewparent;
-
-        local->op_ret = 0;
-        local->call_count--;
-
-        trav = trav->next; /* Skip first child */
-        while (trav) {
-                STACK_WIND (frame, stripe_stack_rename_cbk,
-                            trav->xlator, trav->xlator->fops->rename,
-                            &local->loc, &local->loc2);
-                trav = trav->next;
-        }
-        return 0;
-
-unwind:
-        stripe_local_wipe (local);
-        STACK_UNWIND_STRICT (rename, frame, -1, op_errno, buf, preoldparent,
-                             postoldparent, prenewparent, postnewparent);
-        return 0;
-}
-
-int32_t
-stripe_rename (call_frame_t *frame, xlator_t *this, loc_t *oldloc,
-               loc_t *newloc)
-{
-        stripe_private_t *priv = NULL;
-        stripe_local_t   *local = NULL;
-        xlator_list_t    *trav = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (oldloc, err);
-        VALIDATE_OR_GOTO (oldloc->path, err);
-        VALIDATE_OR_GOTO (oldloc->inode, err);
-        VALIDATE_OR_GOTO (newloc, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        /* If any one node is down, don't allow rename */
-        if (priv->nodes_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        loc_copy (&local->loc, oldloc);
-        loc_copy (&local->loc2, newloc);
-
-        local->call_count = priv->child_count;
-
-        frame->local = local;
-
-        STACK_WIND (frame, stripe_first_rename_cbk, trav->xlator,
-                    trav->xlator->fops->rename, oldloc, newloc);
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (rename, frame, -1, op_errno, NULL, NULL, NULL,
-                             NULL, NULL);
-        return 0;
-}
-
-
-
-int32_t
-stripe_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct iatt *preparent,
-                   struct iatt *postparent)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local   = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG, "%s returned %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-                if (op_ret >= 0) {
-                        local->op_ret = op_ret;
-                        if (FIRST_CHILD(this) == prev->this) {
-                                local->preparent  = *preparent;
-                                local->postparent = *postparent;
-                        }
-                        local->preparent_blocks  += preparent->ia_blocks;
-                        local->postparent_blocks += postparent->ia_blocks;
-
-                        if (local->preparent_size < preparent->ia_size)
-                                local->preparent_size = preparent->ia_size;
-
-                        if (local->postparent_size < postparent->ia_size)
-                                local->postparent_size = postparent->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret != -1) {
-                        local->preparent.ia_blocks  = local->preparent_blocks;
-                        local->preparent.ia_size    = local->preparent_size;
-                        local->postparent.ia_blocks = local->postparent_blocks;
-                        local->postparent.ia_size   = local->postparent_size;
-                }
-
-                stripe_local_wipe (local);
-                STACK_UNWIND_STRICT (unlink, frame, local->op_ret,
-                                     local->op_errno, &local->preparent,
-                                     &local->postparent);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc)
-{
-        xlator_list_t    *trav = NULL;
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        if (priv->first_child_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Don't unlink a file if a node is down */
-        if (priv->nodes_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-        while (trav) {
-                STACK_WIND (frame, stripe_unlink_cbk,
-                            trav->xlator, trav->xlator->fops->unlink,
-                            loc);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (unlink, frame, -1, op_errno, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_first_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno,struct iatt *preparent,
-                        struct iatt *postparent)
-
-{
-        xlator_list_t *trav = NULL;
-        stripe_local_t *local = NULL;
-
-        if (!this || !frame || !frame->local) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                op_errno = EINVAL;
-                goto err;
-        }
-
-        if (op_ret == -1) {
-                goto err;
-        }
-
-        trav = this->children;
-        local = frame->local;
-
-        local->call_count--; /* First child successful */
-        trav = trav->next; /* Skip first child */
-
-        local->preparent  = *preparent;
-        local->postparent = *postparent;
-        local->preparent_size  = preparent->ia_size;
-        local->postparent_size = postparent->ia_size;
-        local->preparent_blocks  += preparent->ia_blocks;
-        local->postparent_blocks += postparent->ia_blocks;
-
-        while (trav) {
-                STACK_WIND (frame, stripe_unlink_cbk, trav->xlator,
-                            trav->xlator->fops->rmdir, &local->loc,
-                            local->flags);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (rmdir, frame, op_ret, op_errno, NULL, NULL);
-        return 0;
-
-}
-
-int32_t
-stripe_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags)
-{
-        xlator_list_t    *trav = NULL;
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        /* don't delete a directory if any of the subvolume is down */
-        if (priv->nodes_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        loc_copy (&local->loc, loc);
-        local->flags = flags;
-        local->call_count = priv->child_count;
-
-        STACK_WIND (frame, stripe_first_rmdir_cbk,  trav->xlator,
-                    trav->xlator->fops->rmdir, loc, flags);
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (rmdir, frame, -1, op_errno, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_mknod_ifreg_fail_unlink_cbk (call_frame_t *frame, void *cookie,
-                                    xlator_t *this, int32_t op_ret,
-                                    int32_t op_errno, struct iatt *preparent,
-                                    struct iatt *postparent)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-
-        if (!this || !frame || !frame->local) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                stripe_local_wipe (local);
-                STACK_UNWIND_STRICT (mknod, frame, local->op_ret, local->op_errno,
-                                     local->inode, &local->stbuf,
-                                     &local->preparent, &local->postparent);
-        }
-out:
-        return 0;
-}
-
-
-/**
- */
-int32_t
-stripe_mknod_ifreg_setxattr_cbk (call_frame_t *frame, void *cookie,
-                                 xlator_t *this, int32_t op_ret,
-                                 int32_t op_errno)
-{
-        int32_t           callcnt = 0;
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        xlator_list_t    *trav = NULL;
-        call_frame_t     *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        priv  = this->private;
-        local = frame->local;
-
-	LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_ret = -1;
-                        local->op_errno = op_errno;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->op_ret == -1) {
-                        local->call_count = priv->child_count;
-                        while (trav) {
-                                STACK_WIND (frame,
-                                            stripe_mknod_ifreg_fail_unlink_cbk,
-                                            trav->xlator,
-                                            trav->xlator->fops->unlink,
-                                            &local->loc);
-                                trav = trav->next;
-                        }
-                        return 0;
-                }
-
-                stripe_local_wipe (local);
-                STACK_UNWIND_STRICT (mknod, frame, local->op_ret, local->op_errno,
-                                     local->inode, &local->stbuf,
-                                     &local->preparent, &local->postparent);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_mknod_ifreg_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, inode_t *inode,
-                        struct iatt *buf, struct iatt *preparent,
-                        struct iatt *postparent)
-{
-        int               ret = 0;
-        int32_t           callcnt = 0;
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        call_frame_t     *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-	priv  = this->private;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                        local->op_errno = op_errno;
-                }
-
-                if (op_ret >= 0) {
-                        local->op_ret = op_ret;
-
-                        if (FIRST_CHILD(this) == prev->this) {
-                                local->stbuf      = *buf;
-                                local->preparent  = *preparent;
-                                local->postparent = *postparent;
-                        }
-
-                        local->stbuf_blocks += buf->ia_blocks;
-                        local->preparent_blocks  += preparent->ia_blocks;
-                        local->postparent_blocks += postparent->ia_blocks;
-
-                        if (local->stbuf_size < buf->ia_size)
-                                local->stbuf_size = buf->ia_size;
-                        if (local->preparent_size < preparent->ia_size)
-                                local->preparent_size = preparent->ia_size;
-                        if (local->postparent_size < postparent->ia_size)
-                                local->postparent_size = postparent->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret != -1) {
-                        local->preparent.ia_blocks  = local->preparent_blocks;
-                        local->preparent.ia_size    = local->preparent_size;
-                        local->postparent.ia_blocks = local->postparent_blocks;
-                        local->postparent.ia_size   = local->postparent_size;
-                        local->stbuf.ia_size        = local->stbuf_size;
-                        local->stbuf.ia_blocks      = local->stbuf_blocks;
-                }
-
-                if ((local->op_ret != -1) && priv->xattr_supported) {
-                        /* Send a setxattr request to nodes where the
-                           files are created */
-                        int32_t  i              = 0;
-                        char     size_key[256]  = {0,};
-                        char     index_key[256] = {0,};
-                        char     count_key[256] = {0,};
-                        dict_t  *dict           = NULL;
-
-                        sprintf (size_key,
-                                 "trusted.%s.stripe-size", this->name);
-                        sprintf (count_key,
-                                 "trusted.%s.stripe-count", this->name);
-                        sprintf (index_key,
-                                 "trusted.%s.stripe-index", this->name);
-
-                        local->call_count = priv->child_count;
-                        memcpy (local->loc.inode->gfid, local->stbuf.ia_gfid, 16);
-                        for (i = 0; i < priv->child_count; i++) {
-                                dict = get_new_dict ();
-                                if (!dict) {
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                "failed to allocate dict");
-                                }
-
-                                dict_ref (dict);
-                                /* TODO: check return value */
-                                ret = dict_set_int64 (dict, size_key,
-                                                      local->stripe_size);
-                                if (ret)
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                "%s: set stripe-size failed",
-                                                local->loc.path);
-                                ret = dict_set_int32 (dict, count_key,
-                                                      priv->child_count);
-                                if (ret)
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                "%s: set child_count failed",
-                                                local->loc.path);
-                                ret = dict_set_int32 (dict, index_key, i);
-                                if (ret)
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                "%s: set stripe-index failed",
-                                                local->loc.path);
-
-                                STACK_WIND (frame,
-                                            stripe_mknod_ifreg_setxattr_cbk,
-                                            priv->xl_array[i],
-                                            priv->xl_array[i]->fops->setxattr,
-                                            &local->loc, dict, 0);
-
-                                dict_unref (dict);
-                        }
-                        return 0;
-                }
-
-                /* Create itself has failed.. so return
-                   without setxattring */
-                stripe_local_wipe (local);
-                STACK_UNWIND_STRICT (mknod, frame, local->op_ret, local->op_errno,
-                                     local->inode, &local->stbuf,
-                                     &local->preparent, &local->postparent);
-        }
-out:
-        return 0;
-}
-
-
-int32_t
-stripe_single_mknod_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, inode_t *inode,
-                         struct iatt *buf, struct iatt *preparent,
-                         struct iatt *postparent)
-{
-        STACK_UNWIND_STRICT (mknod, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent);
-        return 0;
-}
-
-
-int
-stripe_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
-              dev_t rdev, dict_t *params)
-{
-        stripe_private_t *priv = NULL;
-        stripe_local_t   *local = NULL;
-        xlator_list_t    *trav = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        if (priv->first_child_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        if (S_ISREG(mode)) {
-                /* NOTE: on older kernels (older than 2.6.9),
-                   creat() fops is sent as mknod() + open(). Hence handling
-                   S_IFREG files is necessary */
-                if (priv->nodes_down) {
-                        gf_log (this->name, GF_LOG_WARNING,
-                                "Some node down, returning EIO");
-                        op_errno = EIO;
-                        goto err;
-                }
-
-                /* Initialization */
-                local = GF_CALLOC (1, sizeof (stripe_local_t),
-                                   gf_stripe_mt_stripe_local_t);
-                if (!local) {
-                        op_errno = ENOMEM;
-                        goto err;
-                }
-                local->op_ret = -1;
-                local->op_errno = ENOTCONN;
-                local->stripe_size = stripe_get_matching_bs (loc->path,
-                                                             priv->pattern,
-                                                             priv->block_size);
-                frame->local = local;
-                local->inode = loc->inode;
-                loc_copy (&local->loc, loc);
-
-                /* Everytime in stripe lookup, all child nodes should
-                   be looked up */
-                local->call_count = priv->child_count;
-
-                while (trav) {
-                        STACK_WIND (frame, stripe_mknod_ifreg_cbk,
-                                    trav->xlator, trav->xlator->fops->mknod,
-                                    loc, mode, rdev, params);
-                        trav = trav->next;
-                }
-
-                /* This case is handled, no need to continue further. */
-                return 0;
-        }
-
-        STACK_WIND (frame, stripe_single_mknod_cbk,
-                    FIRST_CHILD(this), FIRST_CHILD(this)->fops->mknod,
-                    loc, mode, rdev, params);
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, inode_t *inode,
-                  struct iatt *buf, struct iatt *preparent,
-                  struct iatt *postparent)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t  *local   = NULL;
-        inode_t         *local_inode = NULL;
-        call_frame_t    *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-
-                if (op_ret >= 0) {
-                        local->op_ret = 0;
-
-                        if (FIRST_CHILD(this) == prev->this) {
-                                local->inode      = inode_ref (inode);
-                                local->stbuf      = *buf;
-                                local->postparent = *postparent;
-                                local->preparent  = *preparent;
-                        }
-                        local->stbuf_blocks      += buf->ia_blocks;
-                        local->preparent_blocks  += preparent->ia_blocks;
-                        local->postparent_blocks += postparent->ia_blocks;
-
-                        if (local->stbuf_size < buf->ia_size)
-                                local->stbuf_size = buf->ia_size;
-                        if (local->preparent_size < preparent->ia_size)
-                                local->preparent_size = preparent->ia_size;
-                        if (local->postparent_size < postparent->ia_size)
-                                local->postparent_size = postparent->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                local_inode = local->inode;
-
-                if (local->op_ret != -1) {
-                        local->preparent.ia_blocks  = local->preparent_blocks;
-                        local->preparent.ia_size    = local->preparent_size;
-                        local->postparent.ia_blocks = local->postparent_blocks;
-                        local->postparent.ia_size   = local->postparent_size;
-                        local->stbuf.ia_size        = local->stbuf_size;
-                        local->stbuf.ia_blocks      = local->stbuf_blocks;
-                }
-                STACK_UNWIND_STRICT (mkdir, frame, local->op_ret,
-                                     local->op_errno, local->inode,
-                                     &local->stbuf, &local->preparent,
-                                     &local->postparent);
-
-                if (local_inode)
-                        inode_unref (local_inode);
-        }
-out:
-        return 0;
-}
-
-
-int
-stripe_mkdir (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
-              dict_t *params)
-{
-        stripe_private_t *priv = NULL;
-        stripe_local_t   *local = NULL;
-        xlator_list_t    *trav = NULL;
-        int32_t           op_errno = 1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        if (priv->first_child_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        local->call_count = priv->child_count;
-        frame->local = local;
-
-        /* Everytime in stripe lookup, all child nodes should be looked up */
-        while (trav) {
-                STACK_WIND (frame, stripe_mkdir_cbk,
-                            trav->xlator, trav->xlator->fops->mkdir,
-                            loc, mode, params);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno, inode_t *inode,
-                 struct iatt *buf, struct iatt *preparent,
-                 struct iatt *postparent)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t  *local   = NULL;
-        inode_t         *local_inode = NULL;
-        call_frame_t    *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-
-                if (op_ret >= 0) {
-                        local->op_ret = 0;
-
-                        if (FIRST_CHILD(this) == prev->this) {
-                                local->inode      = inode_ref (inode);
-                                local->stbuf      = *buf;
-                                local->postparent = *postparent;
-                                local->preparent  = *preparent;
-                        }
-                        local->stbuf_blocks      += buf->ia_blocks;
-                        local->preparent_blocks  += preparent->ia_blocks;
-                        local->postparent_blocks += postparent->ia_blocks;
-
-                        if (local->stbuf_size < buf->ia_size)
-                                local->stbuf_size = buf->ia_size;
-                        if (local->preparent_size < preparent->ia_size)
-                                local->preparent_size = preparent->ia_size;
-                        if (local->postparent_size < postparent->ia_size)
-                                local->postparent_size = postparent->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                local_inode = local->inode;
-
-                if (local->op_ret != -1) {
-                        local->preparent.ia_blocks  = local->preparent_blocks;
-                        local->preparent.ia_size    = local->preparent_size;
-                        local->postparent.ia_blocks = local->postparent_blocks;
-                        local->postparent.ia_size   = local->postparent_size;
-                        local->stbuf.ia_size        = local->stbuf_size;
-                        local->stbuf.ia_blocks      = local->stbuf_blocks;
-                }
-                STACK_UNWIND_STRICT (link, frame, local->op_ret,
-                                     local->op_errno, local->inode,
-                                     &local->stbuf, &local->preparent,
-                                     &local->postparent);
-
-                if (local_inode)
-                        inode_unref (local_inode);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_link (call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc)
-{
-        xlator_list_t    *trav = NULL;
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        int32_t           op_errno = 1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (oldloc, err);
-        VALIDATE_OR_GOTO (oldloc->path, err);
-        VALIDATE_OR_GOTO (oldloc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        /* If any one node is down, don't allow link operation */
-        if (priv->nodes_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-        /* Everytime in stripe lookup, all child
-           nodes should be looked up */
-        while (trav) {
-                STACK_WIND (frame, stripe_link_cbk,
-                            trav->xlator, trav->xlator->fops->link,
-                            oldloc, newloc);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (link, frame, -1, op_errno, NULL, NULL, NULL, NULL);
-        return 0;
-}
-
-int32_t
-stripe_create_fail_unlink_cbk (call_frame_t *frame, void *cookie,
-                               xlator_t *this, int32_t op_ret,
-                               int32_t op_errno, struct iatt *preparent,
-                               struct iatt *postparent)
-{
-        int32_t         callcnt = 0;
-        fd_t           *lfd = NULL;
-        stripe_local_t *local = NULL;
-        inode_t        *local_inode = NULL;
-
-        if (!this || !frame || !frame->local) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                local_inode = local->inode;
-                lfd = local->fd;
-
-                stripe_local_wipe (local);
-                STACK_UNWIND_STRICT (create, frame, local->op_ret, local->op_errno,
-                                     local->fd, local->inode, &local->stbuf,
-                                     &local->preparent, &local->postparent);
-
-                if (local_inode)
-                        inode_unref (local_inode);
-                if (lfd)
-                        fd_unref (lfd);
-        }
-out:
-        return 0;
-}
-
-
-int32_t
-stripe_create_setxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                            int32_t op_ret, int32_t op_errno)
-{
-        inode_t          *local_inode = NULL;
-        fd_t             *lfd = NULL;
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        xlator_list_t    *trav = NULL;
-        int32_t           callcnt = 0;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-	priv  = this->private;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_ret = -1;
-                        local->op_errno = op_errno;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->op_ret == -1) {
-                        local->call_count = priv->child_count;
-                        trav = this->children;
-                        while (trav) {
-                                STACK_WIND (frame,
-                                            stripe_create_fail_unlink_cbk,
-                                            trav->xlator,
-                                            trav->xlator->fops->unlink,
-                                            &local->loc);
-                                trav = trav->next;
-                        }
-
-                        return 0;
-                }
-
-                lfd = local->fd;
-                local_inode = local->inode;
-
-                stripe_local_wipe (local);
-                STACK_UNWIND_STRICT (create, frame, local->op_ret, local->op_errno,
-                                     local->fd, local->inode, &local->stbuf,
-                                     &local->preparent, &local->postparent);
-
-                if (local_inode)
-                        inode_unref (local_inode);
-                if (lfd)
-                        fd_unref (lfd);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, fd_t *fd,
-                   inode_t *inode, struct iatt *buf, struct iatt *preparent,
-                   struct iatt *postparent)
-{
-        int32_t           callcnt = 0;
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        fd_t             *lfd = NULL;
-        stripe_fd_ctx_t  *fctx = NULL;
-        inode_t          *local_inode = NULL;
-        call_frame_t     *prev = NULL;
-        int               ret = 0;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        priv  = this->private;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                        local->op_errno = op_errno;
-                }
-
-                if (op_ret >= 0) {
-                        local->op_ret = op_ret;
-                        /* Get the mapping in inode private */
-                        /* Get the stat buf right */
-                        if (FIRST_CHILD(this) == prev->this) {
-                                local->stbuf      = *buf;
-                                local->preparent  = *preparent;
-                                local->postparent = *postparent;
-                        }
-
-                        local->stbuf_blocks += buf->ia_blocks;
-                        local->preparent_blocks  += preparent->ia_blocks;
-                        local->postparent_blocks += postparent->ia_blocks;
-
-                        if (local->stbuf_size < buf->ia_size)
-                                local->stbuf_size = buf->ia_size;
-                        if (local->preparent_size < preparent->ia_size)
-                                local->preparent_size = preparent->ia_size;
-                        if (local->postparent_size < postparent->ia_size)
-                                local->postparent_size = postparent->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret != -1) {
-                        local->preparent.ia_blocks  = local->preparent_blocks;
-                        local->preparent.ia_size    = local->preparent_size;
-                        local->postparent.ia_blocks = local->postparent_blocks;
-                        local->postparent.ia_size   = local->postparent_size;
-                        local->stbuf.ia_size        = local->stbuf_size;
-                        local->stbuf.ia_blocks      = local->stbuf_blocks;
-                }
-
-                /* */
-                if (local->op_ret >= 0) {
-                        fctx = GF_CALLOC (1, sizeof (stripe_fd_ctx_t),
-                                          gf_stripe_mt_stripe_fd_ctx_t);
-                        if (!fctx) {
-                                local->op_ret = -1;
-                                local->op_errno = ENOMEM;
-                                goto unwind;
-                        }
-
-                        fctx->stripe_size  = local->stripe_size;
-                        fctx->stripe_count = priv->child_count;
-                        fctx->static_array = 1;
-                        fctx->xl_array = priv->xl_array;
-                        fd_ctx_set (local->fd, this,
-                                    (uint64_t)(long)fctx);
-                }
-
-                if ((local->op_ret != -1) &&
-                    local->stripe_size && priv->xattr_supported) {
-                        /* Send a setxattr request to nodes where
-                           the files are created */
-                        int32_t        i = 0;
-                        char           size_key[256] = {0,};
-                        char           index_key[256] = {0,};
-                        char           count_key[256] = {0,};
-                        dict_t        *dict = NULL;
-
-                        sprintf (size_key,
-                                 "trusted.%s.stripe-size", this->name);
-                        sprintf (count_key,
-                                 "trusted.%s.stripe-count", this->name);
-                        sprintf (index_key,
-                                 "trusted.%s.stripe-index", this->name);
-
-                        local->call_count = priv->child_count;
-                        memcpy (local->loc.inode->gfid, local->stbuf.ia_gfid, 16);
-                        for (i = 0; i < priv->child_count; i++) {
-                                dict = get_new_dict ();
-                                if (!dict) {
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                "error allocating dict");
-                                }
-                                dict_ref (dict);
-
-                                /* TODO: check return values */
-                                ret = dict_set_int64 (dict, size_key,
-                                                      local->stripe_size);
-                                if (ret)
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                "%s: set stripe-size failed",
-                                                local->loc.path);
-
-                                ret = dict_set_int32 (dict, count_key,
-                                                      priv->child_count);
-                                if (ret)
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                "%s: set stripe-size failed",
-                                                local->loc.path);
-
-                                ret = dict_set_int32 (dict, index_key, i);
-                                if (ret)
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                "%s: set stripe-size failed",
-                                                local->loc.path);
-
-                                STACK_WIND (frame, stripe_create_setxattr_cbk,
-                                            priv->xl_array[i],
-                                            priv->xl_array[i]->fops->setxattr,
-                                            &local->loc, dict, 0);
-
-                                dict_unref (dict);
-                        }
-                        return 0;
-                }
-
-unwind:
-                /* Create itself has failed.. so return
-                   without setxattring */
-                lfd = local->fd;
-                local_inode = local->inode;
-
-                stripe_local_wipe (local);
-                STACK_UNWIND_STRICT (create, frame, local->op_ret,
-                                     local->op_errno, local->fd,
-                                     local->inode, &local->stbuf,
-                                     &local->preparent, &local->postparent);
-
-                if (local_inode)
-                        inode_unref (local_inode);
-                if (lfd)
-                        fd_unref (lfd);
-        }
-
-out:
-        return 0;
-}
-
-
-/**
- * stripe_create - If a block-size is specified for the 'name', create the
- *    file in all the child nodes. If not, create it in only first child.
- *
- * @name- complete path of the file to be created.
- */
-int32_t
-stripe_create (call_frame_t *frame, xlator_t *this, loc_t *loc,
-               int32_t flags, mode_t mode, fd_t *fd, dict_t *params)
-{
-        stripe_private_t *priv = NULL;
-        stripe_local_t   *local = NULL;
-        xlator_list_t    *trav = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-
-        /* files created in O_APPEND mode does not allow lseek() on fd */
-        flags &= ~O_APPEND;
-
-        if (priv->first_child_down || priv->nodes_down) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "First node down, returning EIO");
-                op_errno = EIO;
-                goto err;
-        }
-
-        /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        local->op_errno = ENOTCONN;
-        local->stripe_size = stripe_get_matching_bs (loc->path,
-                                                     priv->pattern,
-                                                     priv->block_size);
-        frame->local = local;
-        local->inode = inode_ref (loc->inode);
-        loc_copy (&local->loc, loc);
-        local->fd = fd_ref (fd);
-
-        local->call_count = priv->child_count;
-
-        trav = this->children;
-        while (trav) {
-                STACK_WIND (frame, stripe_create_cbk, trav->xlator,
-                            trav->xlator->fops->create, loc, flags,
-                            mode, fd, params);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (create, frame, -1, op_errno, NULL, NULL, NULL,
-                             NULL, NULL);
-        return 0;
-}
-
-int32_t
-stripe_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno, fd_t *fd)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-        fd_t           *lfd = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                        local->op_errno = op_errno;
-                }
-
-                if (op_ret >= 0)
-                        local->op_ret = op_ret;
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret == -1) {
-                        if (local->fctx) {
-                                if (!local->fctx->static_array)
-                                        GF_FREE (local->fctx->xl_array);
-                                GF_FREE (local->fctx);
-                        }
-                } else {
-                        fd_ctx_set (local->fd, this,
-                                    (uint64_t)(long)local->fctx);
-                }
-
-                lfd = local->fd;
-
-                stripe_local_wipe (local);
-                STACK_UNWIND_STRICT (open, frame, local->op_ret,
-                                     local->op_errno, local->fd);
-                if (lfd)
-                        fd_unref (lfd);
-
-        }
-out:
-        return 0;
-}
-
-
-int32_t
-stripe_open_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t op_ret, int32_t op_errno, dict_t *dict)
-{
-        int32_t           index = 0;
-        int32_t           callcnt = 0;
-        char              key[256] = {0,};
-        stripe_local_t   *local = NULL;
-        xlator_list_t    *trav = NULL;
-        stripe_private_t *priv = NULL;
-        data_t           *data = NULL;
-        call_frame_t     *prev = NULL;
-        fd_t             *lfd = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = (call_frame_t *)cookie;
-        priv  = this->private;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_ret = -1;
-                        if (local->op_errno != EIO)
-                                local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                        goto unlock;
-                }
-
-                if (!dict)
-                        goto unlock;
-
-                if (!local->fctx) {
-                        local->fctx =  GF_CALLOC (1, sizeof (stripe_fd_ctx_t),
-                                                  gf_stripe_mt_stripe_fd_ctx_t);
-                        if (!local->fctx) {
-                                local->op_errno = ENOMEM;
-                                local->op_ret = -1;
-                                goto unlock;
-                        }
-
-                        local->fctx->static_array = 0;
-                }
-                /* Stripe block size */
-                sprintf (key, "trusted.%s.stripe-size", this->name);
-                data = dict_get (dict, key);
-                if (!data) {
-                        local->xattr_self_heal_needed = 1;
-                } else {
-                        if (!local->fctx->stripe_size) {
-                                local->fctx->stripe_size =
-                                        data_to_int64 (data);
-                        }
-
-                        if (local->fctx->stripe_size != data_to_int64 (data)) {
-                                gf_log (this->name, GF_LOG_WARNING,
-                                        "stripe-size mismatch in blocks");
-                                local->xattr_self_heal_needed = 1;
-                        }
-                }
-                /* Stripe count */
-                sprintf (key, "trusted.%s.stripe-count", this->name);
-                data = dict_get (dict, key);
-                if (!data) {
-                        local->xattr_self_heal_needed = 1;
-                        goto unlock;
-                }
-                if (!local->fctx->xl_array) {
-                        local->fctx->stripe_count = data_to_int32 (data);
-                        if (!local->fctx->stripe_count) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "error with stripe-count xattr");
-                                local->op_ret   = -1;
-                                local->op_errno = EIO;
-                                goto unlock;
-                        }
-
-                        local->fctx->xl_array =
-                                GF_CALLOC (local->fctx->stripe_count,
-                                           sizeof (xlator_t *),
-                                           gf_stripe_mt_xlator_t);
-                        if (!local->fctx->xl_array) {
-                                local->op_errno = ENOMEM;
-                                local->op_ret   = -1;
-                                goto unlock;
-                        }
-                }
-                if (local->fctx->stripe_count != data_to_int32 (data)) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "error with stripe-count xattr (%d != %d)",
-                                local->fctx->stripe_count, data_to_int32 (data));
-                        local->op_ret   = -1;
-                        local->op_errno = EIO;
-                        goto unlock;
-                }
-
-                /* index */
-                sprintf (key, "trusted.%s.stripe-index", this->name);
-                data = dict_get (dict, key);
-                if (!data) {
-                        local->xattr_self_heal_needed = 1;
-                        goto unlock;
-                }
-                index = data_to_int32 (data);
-                if (index > priv->child_count) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "error with stripe-index xattr (%d)", index);
-                        local->op_ret   = -1;
-                        local->op_errno = EIO;
-                        goto unlock;
-                }
-                if (local->fctx->xl_array) {
-                        if (local->fctx->xl_array[index]) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "duplicate entry @ index (%d)", index);
-                                local->op_ret   = -1;
-                                local->op_errno = EIO;
-                                goto unlock;
-                        }
-                        local->fctx->xl_array[index] = prev->this;
-                }
-                local->entry_count++;
-                local->op_ret = 0;
-        }
-unlock:
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                /* TODO: if self-heal flag is set, do it */
-                if (local->xattr_self_heal_needed) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s: stripe info need to be healed",
-                                local->loc.path);
-                }
-
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret)
-                        goto err;
-
-                if (local->entry_count != local->fctx->stripe_count) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "entry-count (%d) != stripe-count (%d)",
-                                local->entry_count, local->fctx->stripe_count);
-                        local->op_ret = -1;
-                        local->op_errno = EIO;
-                        goto err;
-                }
-                if (!local->fctx->stripe_size) {
-                        gf_log (this->name, GF_LOG_ERROR, "stripe size not set");
-                        local->op_ret = -1;
-                        local->op_errno = EIO;
-                        goto err;
-                }
-
-                local->call_count = local->fctx->stripe_count;
-
-                trav = this->children;
-                while (trav) {
-                        STACK_WIND (frame, stripe_open_cbk, trav->xlator,
-                                    trav->xlator->fops->open, &local->loc,
-                                    local->flags, local->fd, 0);
-                        trav = trav->next;
-                }
-        }
-
-        return 0;
-err:
-        lfd = local->fd;
-
-        stripe_local_wipe (local);
-        STACK_UNWIND_STRICT (open, frame, local->op_ret, local->op_errno,
-                             local->fd);
-        if (lfd)
-                fd_unref (lfd);
-out:
-        return 0;
-}
-
-/**
- * stripe_open -
- */
-int32_t
-stripe_open (call_frame_t *frame, xlator_t *this, loc_t *loc,
-             int32_t flags, fd_t *fd, int32_t wbflags)
-{
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        xlator_list_t    *trav = NULL;
-        int32_t           op_errno = 1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        if (priv->first_child_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        /* files opened in O_APPEND mode does not allow lseek() on fd */
-        flags &= ~O_APPEND;
-
-        local->fd = fd_ref (fd);
-        frame->local = local;
-        loc_copy (&local->loc, loc);
-
-        /* Striped files */
-        local->flags = flags;
-        local->call_count = priv->child_count;
-        local->stripe_size = stripe_get_matching_bs (loc->path,
-                                                     priv->pattern,
-                                                     priv->block_size);
-
-        if (priv->xattr_supported) {
-                while (trav) {
-                        STACK_WIND (frame, stripe_open_getxattr_cbk,
-                                    trav->xlator, trav->xlator->fops->getxattr,
-                                    loc, NULL);
-                        trav = trav->next;
-                }
-                return 0;
-        }
-        local->fctx =  GF_CALLOC (1, sizeof (stripe_fd_ctx_t),
-                                  gf_stripe_mt_stripe_fd_ctx_t);
-        if (!local->fctx) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        local->fctx->static_array = 1;
-        local->fctx->stripe_size  = local->stripe_size;
-        local->fctx->stripe_count = priv->child_count;
-        local->fctx->xl_array     = priv->xl_array;
-
-        while (trav) {
-                STACK_WIND (frame, stripe_open_cbk, trav->xlator,
-                            trav->xlator->fops->open,
-                            &local->loc, local->flags, local->fd,
-                            wbflags);
-                trav = trav->next;
-        }
-        return 0;
-err:
-        STACK_UNWIND_STRICT (open, frame, -1, op_errno, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, fd_t *fd)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-        fd_t           *local_fd = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_ret = -1;
-                        local->op_errno = op_errno;
-                }
-
-                if (op_ret >= 0)
-                        local->op_ret = op_ret;
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                local_fd = local->fd;
-                STACK_UNWIND_STRICT (opendir, frame, local->op_ret,
-                                     local->op_errno, local->fd);
-                if (local_fd)
-                        fd_unref (local_fd);
-        }
-out:
-        return 0;
-}
-
-
-int32_t
-stripe_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd)
-{
-        xlator_list_t    *trav = NULL;
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (loc, err);
-        VALIDATE_OR_GOTO (loc->path, err);
-        VALIDATE_OR_GOTO (loc->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        if (priv->first_child_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-
-        /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        frame->local = local;
-        local->call_count = priv->child_count;
-        local->fd = fd_ref (fd);
-
-        while (trav) {
-                STACK_WIND (frame, stripe_opendir_cbk, trav->xlator,
-                            trav->xlator->fops->opendir, loc, fd);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (opendir, frame, -1, op_errno, NULL);
-        return 0;
-}
-
-int32_t
-stripe_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-               int32_t op_ret, int32_t op_errno, struct gf_flock *lock)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-                if (op_ret >= 0) {
-                        if (FIRST_CHILD(this) == prev->this) {
-                                /* First successful call, copy the *lock */
-                                local->op_ret = op_ret;
-                                local->lock = *lock;
-                        }
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-                STACK_UNWIND_STRICT (lk, frame, local->op_ret,
-                                     local->op_errno, &local->lock);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_lk (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
-           struct gf_flock *lock)
-{
-        stripe_local_t   *local = NULL;
-        xlator_list_t    *trav = NULL;
-        stripe_private_t *priv = NULL;
-        int32_t           op_errno = EINVAL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-        trav = this->children;
-        priv = this->private;
-
-        /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-        while (trav) {
-                STACK_WIND (frame, stripe_lk_cbk, trav->xlator,
-                            trav->xlator->fops->lk, fd, cmd, lock);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (lk, frame, -1, op_errno, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local   = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-                if (op_ret >= 0)
-                        local->op_ret = op_ret;
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                stripe_local_wipe (local);
-                STACK_UNWIND_STRICT (flush, frame, local->op_ret,
-                                     local->op_errno);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_flush (call_frame_t *frame, xlator_t *this, fd_t *fd)
-{
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        xlator_list_t    *trav = NULL;
-        int32_t           op_errno = 1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        if (priv->first_child_down) {
-                op_errno = ENOTCONN;
-                goto err;
-        }
-        /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-        while (trav) {
-                STACK_WIND (frame, stripe_flush_cbk,  trav->xlator,
-                            trav->xlator->fops->flush, fd);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (flush, frame, -1, op_errno);
-        return 0;
-}
-
-
-
-int32_t
-stripe_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                  struct iatt *postbuf)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local   = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-                if (op_ret >= 0) {
-                        local->op_ret = op_ret;
-                        if (FIRST_CHILD(this) == prev->this) {
-                                local->pre_buf  = *prebuf;
-                                local->post_buf = *postbuf;
-                        }
-                        local->prebuf_blocks  += prebuf->ia_blocks;
-                        local->postbuf_blocks += postbuf->ia_blocks;
-
-                        if (local->prebuf_size < prebuf->ia_size)
-                                local->prebuf_size = prebuf->ia_size;
-
-                        if (local->postbuf_size < postbuf->ia_size)
-                                local->postbuf_size = postbuf->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret != -1) {
-                        local->pre_buf.ia_blocks  = local->prebuf_blocks;
-                        local->pre_buf.ia_size    = local->prebuf_size;
-                        local->post_buf.ia_blocks = local->postbuf_blocks;
-                        local->post_buf.ia_size   = local->postbuf_size;
-                }
-
-                stripe_local_wipe (local);
-                STACK_UNWIND_STRICT (fsync, frame, local->op_ret,
-                                     local->op_errno, &local->pre_buf,
-                                     &local->post_buf);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags)
-{
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        xlator_list_t    *trav = NULL;
-        int32_t           op_errno = 1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-        while (trav) {
-                STACK_WIND (frame, stripe_fsync_cbk, trav->xlator,
-                            trav->xlator->fops->fsync, fd, flags);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (fsync, frame, -1, op_errno, NULL, NULL);
-        return 0;
-}
-
-int32_t
-stripe_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-
-                if (op_ret == 0) {
-                        local->op_ret = 0;
-
-                        if (FIRST_CHILD(this) == prev->this)
-                                local->stbuf = *buf;
-
-                        local->stbuf_blocks += buf->ia_blocks;
-                        if (local->stbuf_size < buf->ia_size)
-                                local->stbuf_size = buf->ia_size;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                if (local->op_ret != -1) {
-                        local->stbuf.ia_size   = local->stbuf_size;
-                        local->stbuf.ia_blocks = local->stbuf_blocks;
-                }
-
-                stripe_local_wipe (local);
-                STACK_UNWIND_STRICT (fstat, frame, local->op_ret,
-                                     local->op_errno, &local->stbuf);
-        }
-
-out:
-        return 0;
-}
-
-int32_t
-stripe_fstat (call_frame_t *frame,
-              xlator_t *this,
-              fd_t *fd)
-{
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        xlator_list_t    *trav = NULL;
-        int32_t           op_errno = 1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-        while (trav) {
-                STACK_WIND (frame, stripe_fstat_cbk, trav->xlator,
-                            trav->xlator->fops->fstat, fd);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (fstat, frame, -1, op_errno, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset)
-{
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        xlator_list_t    *trav = NULL;
-        int32_t           op_errno = 1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-        while (trav) {
-                STACK_WIND (frame, stripe_truncate_cbk, trav->xlator,
-                            trav->xlator->fops->ftruncate, fd, offset);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (ftruncate, frame, -1, op_errno, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_fsyncdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local   = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        if ((op_errno != ENOENT) ||
-                            (prev->this == FIRST_CHILD (this)))
-                                local->failed = 1;
-                }
-                if (op_ret >= 0)
-                        local->op_ret = op_ret;
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                if (local->failed)
-                        local->op_ret = -1;
-
-                stripe_local_wipe (local);
-                STACK_UNWIND_STRICT (fsyncdir, frame, local->op_ret,
-                                     local->op_errno);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags)
-{
-        stripe_local_t   *local = NULL;
-        stripe_private_t *priv = NULL;
-        xlator_list_t    *trav = NULL;
-        int32_t           op_errno = 1;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-        priv = this->private;
-        trav = this->children;
-
-        /* Initialization */
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        local->op_ret = -1;
-        frame->local = local;
-        local->call_count = priv->child_count;
-
-        while (trav) {
-                STACK_WIND (frame, stripe_fsyncdir_cbk, trav->xlator,
-                            trav->xlator->fops->fsyncdir, fd, flags);
-                trav = trav->next;
-        }
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (fsyncdir, frame, -1, op_errno);
-        return 0;
-}
-
-
-int32_t
-stripe_readv_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        int32_t         i = 0;
-        int32_t         callcnt = 0;
-        int32_t         count = 0;
-        stripe_local_t *local = NULL;
-        struct iovec   *vec = NULL;
-        struct iatt     tmp_stbuf = {0,};
-        struct iobref  *tmp_iobref = NULL;
-        struct iobuf   *iobuf = NULL;
-
-        if (!this || !frame || !frame->local) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        local = frame->local;
-
-        LOCK (&frame->lock);
-        {
-                callcnt = --local->call_count;
-                if (op_ret != -1)
-                        if (local->stbuf_size < buf->ia_size)
-                                local->stbuf_size = buf->ia_size;
-        }
-        UNLOCK (&frame->lock);
-
-        if (!callcnt) {
-                op_ret = 0;
-
-                /* Keep extra space for filling in '\0's */
-                vec = GF_CALLOC ((local->count * 2), sizeof (struct iovec),
-                                 gf_stripe_mt_iovec);
-                if (!vec) {
-                        op_ret = -1;
-                        goto done;
-                }
-
-                for (i = 0; i < local->wind_count; i++) {
-                        if (local->replies[i].op_ret) {
-                                memcpy ((vec + count), local->replies[i].vector,
-                                        (local->replies[i].count * sizeof (struct iovec)));
-                                count +=  local->replies[i].count;
-                                op_ret += local->replies[i].op_ret;
-                        }
-                        if ((local->replies[i].op_ret <
-                             local->replies[i].requested_size) &&
-                            (local->stbuf_size > (local->offset + op_ret))) {
-                                /* Fill in 0s here */
-                                vec[count].iov_len  =
-                                        (local->replies[i].requested_size -
-                                         local->replies[i].op_ret);
-                                iobuf = iobuf_get (this->ctx->iobuf_pool);
-                                if (!iobuf) {
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                "Out of memory.");
-                                        op_ret = -1;
-                                        op_errno = ENOMEM;
-                                        goto done;
-                                }
-                                memset (iobuf->ptr, 0, vec[count].iov_len);
-                                iobref_add (local->iobref, iobuf);
-                                vec[count].iov_base = iobuf->ptr;
-
-                                op_ret += vec[count].iov_len;
-                                count++;
-                        }
-                        GF_FREE (local->replies[i].vector);
-                }
-
-                /* FIXME: notice that st_ino, and st_dev (gen) will be
-                 * different than what inode will have. Make sure this doesn't
-                 * cause any bugs at higher levels */
-                memcpy (&tmp_stbuf, &local->replies[0].stbuf,
-                        sizeof (struct iatt));
-                tmp_stbuf.ia_size = local->stbuf_size;
-
-        done:
-                GF_FREE (local->replies);
-                tmp_iobref = local->iobref;
-                fd_unref (local->fd);
-                STACK_UNWIND_STRICT (readv, frame, op_ret, op_errno, vec,
-                                     count, &tmp_stbuf, tmp_iobref);
-
-                iobref_unref (tmp_iobref);
-                if (vec)
-                        GF_FREE (vec);
-        }
-out:
-        return 0;
-}
-
-/**
- * stripe_readv_cbk - get all the striped reads, and order it properly, send it
- *        to above layer after putting it in a single vector.
- */
-int32_t
-stripe_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, struct iovec *vector,
-                  int32_t count, struct iatt *stbuf, struct iobref *iobref)
-{
-        int32_t         index = 0;
-        int32_t         callcnt = 0;
-        int32_t         final_count = 0;
-        int32_t         need_to_check_proper_size = 0;
-        call_frame_t   *mframe = NULL;
-        stripe_local_t *mlocal = NULL;
-        stripe_local_t *local = NULL;
-        struct iovec   *final_vec = NULL;
-        struct iatt     tmp_stbuf = {0,};
-        struct iobref  *tmp_iobref = NULL;
-        stripe_fd_ctx_t  *fctx = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto end;
-        }
-
-        local  = frame->local;
-        index  = local->node_index;
-        mframe = local->orig_frame;
-        if (!mframe)
-                goto out;
-
-        mlocal = mframe->local;
-        if (!mlocal)
-                goto out;
-
-        fctx = mlocal->fctx;
-
-        LOCK (&mframe->lock);
-        {
-                mlocal->replies[index].op_ret = op_ret;
-                mlocal->replies[index].op_errno = op_errno;
-                mlocal->replies[index].requested_size = local->readv_size;
-                if (op_ret >= 0) {
-                        mlocal->replies[index].stbuf  = *stbuf;
-                        mlocal->replies[index].count  = count;
-                        mlocal->replies[index].vector = iov_dup (vector, count);
-
-                        if (!mlocal->iobref)
-                                mlocal->iobref = iobref_new ();
-                        iobref_merge (mlocal->iobref, iobref);
-                }
-                callcnt = ++mlocal->call_count;
-        }
-        UNLOCK(&mframe->lock);
-
-        if (callcnt == mlocal->wind_count) {
-                op_ret = 0;
-
-                for (index=0; index < mlocal->wind_count; index++) {
-                        /* check whether each stripe returned
-                         * 'expected' number of bytes */
-                        if (mlocal->replies[index].op_ret == -1) {
-                                op_ret = -1;
-                                op_errno = mlocal->replies[index].op_errno;
-                                break;
-                        }
-                        /* TODO: handle the 'holes' within the read range
-                           properly */
-                        if (mlocal->replies[index].op_ret <
-                            mlocal->replies[index].requested_size) {
-                                need_to_check_proper_size = 1;
-                        }
-
-                        op_ret       += mlocal->replies[index].op_ret;
-                        mlocal->count += mlocal->replies[index].count;
-                }
-                if (op_ret == -1)
-                        goto done;
-                if (need_to_check_proper_size)
-                        goto check_size;
-
-                final_vec = GF_CALLOC (mlocal->count, sizeof (struct iovec),
-                                       gf_stripe_mt_iovec);
-
-                if (!final_vec) {
-                        op_ret = -1;
-                        goto done;
-                }
-
-                for (index = 0; index < mlocal->wind_count; index++) {
-                        memcpy ((final_vec + final_count),
-                                mlocal->replies[index].vector,
-                                (mlocal->replies[index].count *
-                                 sizeof (struct iovec)));
-                        final_count +=  mlocal->replies[index].count;
-                        GF_FREE (mlocal->replies[index].vector);
-                }
-
-                /* FIXME: notice that st_ino, and st_dev (gen) will be
-                 * different than what inode will have. Make sure this doesn't
-                 * cause any bugs at higher levels */
-                memcpy (&tmp_stbuf, &mlocal->replies[0].stbuf,
-                        sizeof (struct iatt));
-
-        done:
-                /* */
-                GF_FREE (mlocal->replies);
-                tmp_iobref = mlocal->iobref;
-                fd_unref (mlocal->fd);
-                STACK_UNWIND_STRICT (readv, mframe, op_ret, op_errno, final_vec,
-                                     final_count, &tmp_stbuf, tmp_iobref);
-
-                iobref_unref (tmp_iobref);
-                if (final_vec)
-                        GF_FREE (final_vec);
-        }
-
-        goto out;
-
-check_size:
-        mlocal->call_count = fctx->stripe_count;
-
-        for (index = 0; index < fctx->stripe_count; index++) {
-                STACK_WIND (mframe, stripe_readv_fstat_cbk,
-                            (fctx->xl_array[index]),
-                            (fctx->xl_array[index])->fops->fstat,
-                            mlocal->fd);
-        }
-
-out:
-        STACK_DESTROY (frame->root);
-end:
-        return 0;
-}
-
-
-int32_t
-stripe_readv (call_frame_t *frame, xlator_t *this, fd_t *fd,
-              size_t size, off_t offset)
-{
-        int32_t           op_errno = EINVAL;
-        int32_t           idx = 0;
-        int32_t           index = 0;
-        int32_t           num_stripe = 0;
-        int32_t           off_index = 0;
-        size_t            frame_size = 0;
-        off_t             rounded_end = 0;
-        uint64_t          tmp_fctx = 0;
-        uint64_t          stripe_size = 0;
-        off_t             rounded_start = 0;
-        off_t             frame_offset = offset;
-        stripe_local_t   *local = NULL;
-        call_frame_t     *rframe = NULL;
-        stripe_local_t   *rlocal = NULL;
-        stripe_fd_ctx_t  *fctx = NULL;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-        fd_ctx_get (fd, this, &tmp_fctx);
-        if (!tmp_fctx) {
-                op_errno = EBADFD;
-                goto err;
-        }
-        fctx = (stripe_fd_ctx_t *)(long)tmp_fctx;
-        stripe_size = fctx->stripe_size;
-
-        if (!stripe_size) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Wrong stripe size for the file");
-                goto err;
-        }
-        /* The file is stripe across the child nodes. Send the read request
-         * to the child nodes appropriately after checking which region of
-         * the file is in which child node. Always '0-<stripe_size>' part of
-         * the file resides in the first child.
-         */
-        rounded_start = floor (offset, stripe_size);
-        rounded_end = roof (offset+size, stripe_size);
-        num_stripe = (rounded_end- rounded_start)/stripe_size;
-
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        frame->local = local;
-
-        /* This is where all the vectors should be copied. */
-        local->replies = GF_CALLOC (num_stripe, sizeof (struct readv_replies),
-                                    gf_stripe_mt_readv_replies);
-        if (!local->replies) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-
-        off_index = (offset / stripe_size) % fctx->stripe_count;
-        local->wind_count = num_stripe;
-        local->readv_size = size;
-        local->offset     = offset;
-        local->fd         = fd_ref (fd);
-        local->fctx       = fctx;
-
-        for (index = off_index; index < (num_stripe + off_index); index++) {
-                rframe = copy_frame (frame);
-                rlocal = GF_CALLOC (1, sizeof (stripe_local_t),
-                                    gf_stripe_mt_stripe_local_t);
-                if (!rlocal) {
-                        op_errno = ENOMEM;
-                        goto err;
-                }
-
-                frame_size = min (roof (frame_offset+1, stripe_size),
-                                  (offset + size)) - frame_offset;
-
-                rlocal->node_index = index - off_index;
-                rlocal->orig_frame = frame;
-                rlocal->readv_size = frame_size;
-                rframe->local = rlocal;
-                idx = (index % fctx->stripe_count);
-                STACK_WIND (rframe, stripe_readv_cbk, fctx->xl_array[idx],
-                            fctx->xl_array[idx]->fops->readv,
-                            fd, frame_size, frame_offset);
-
-                frame_offset += frame_size;
-        }
-
-        return 0;
-err:
-        if (local && local->fd)
-                fd_unref (local->fd);
-
-        STACK_UNWIND_STRICT (readv, frame, -1, op_errno, NULL, 0, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                   struct iatt *postbuf)
-{
-        int32_t         callcnt = 0;
-        stripe_local_t *local = NULL;
-        call_frame_t   *prev = NULL;
-
-        if (!this || !frame || !frame->local || !cookie) {
-                gf_log ("stripe", GF_LOG_DEBUG, "possible NULL deref");
-                goto out;
-        }
-
-        prev  = cookie;
-        local = frame->local;
-
-        LOCK(&frame->lock);
-        {
-                callcnt = ++local->call_count;
-
-                if (op_ret == -1) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s returned error %s",
-                                prev->this->name, strerror (op_errno));
-                        local->op_errno = op_errno;
-                        local->op_ret = -1;
-                }
-                if (op_ret >= 0) {
-                        local->op_ret += op_ret;
-                        local->post_buf = *postbuf;
-                        local->pre_buf = *prebuf;
-                }
-        }
-        UNLOCK (&frame->lock);
-
-        if ((callcnt == local->wind_count) && local->unwind) {
-                STACK_UNWIND_STRICT (writev, frame, local->op_ret,
-                                     local->op_errno, &local->pre_buf,
-                                     &local->post_buf);
-        }
-out:
-        return 0;
-}
-
-int32_t
-stripe_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
-               struct iovec *vector, int32_t count, off_t offset,
-               struct iobref *iobref)
-{
-        struct iovec     *tmp_vec = NULL;
-        stripe_local_t   *local = NULL;
-        stripe_fd_ctx_t  *fctx = NULL;
-        int32_t           op_errno = 1;
-        int32_t           idx = 0;
-        int32_t           total_size = 0;
-        int32_t           offset_offset = 0;
-        int32_t           remaining_size = 0;
-        int32_t           tmp_count = count;
-        off_t             fill_size = 0;
-        uint64_t          stripe_size = 0;
-        uint64_t          tmp_fctx = 0;
-
-        VALIDATE_OR_GOTO (frame, err);
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-        VALIDATE_OR_GOTO (fd->inode, err);
-
-        fd_ctx_get (fd, this, &tmp_fctx);
-        if (!tmp_fctx) {
-                op_errno = EINVAL;
-                goto err;
-        }
-        fctx = (stripe_fd_ctx_t *)(long)tmp_fctx;
-        stripe_size = fctx->stripe_size;
-
-        /* File has to be stripped across the child nodes */
-        for (idx = 0; idx< count; idx ++) {
-                total_size += vector[idx].iov_len;
-        }
-        remaining_size = total_size;
-
-        local = GF_CALLOC (1, sizeof (stripe_local_t),
-                           gf_stripe_mt_stripe_local_t);
-        if (!local) {
-                op_errno = ENOMEM;
-                goto err;
-        }
-        frame->local = local;
-        local->stripe_size = stripe_size;
-
-        while (1) {
-                /* Send striped chunk of the vector to child
-                   nodes appropriately. */
-                idx = (((offset + offset_offset) /
-                        local->stripe_size) % fctx->stripe_count);
-
-                fill_size = (local->stripe_size -
-                             ((offset + offset_offset) % local->stripe_size));
-                if (fill_size > remaining_size)
-                        fill_size = remaining_size;
-
-                remaining_size -= fill_size;
-
-                tmp_count = iov_subset (vector, count, offset_offset,
-                                        offset_offset + fill_size, NULL);
-                tmp_vec = GF_CALLOC (tmp_count, sizeof (struct iovec),
-                                     gf_stripe_mt_iovec);
-                if (!tmp_vec) {
-                        op_errno = ENOMEM;
-                        goto err;
-                }
-                tmp_count = iov_subset (vector, count, offset_offset,
-                                        offset_offset + fill_size, tmp_vec);
-
-                local->wind_count++;
-                if (remaining_size == 0)
-                        local->unwind = 1;
-
-                STACK_WIND (frame, stripe_writev_cbk, fctx->xl_array[idx],
-                            fctx->xl_array[idx]->fops->writev, fd, tmp_vec,
-                            tmp_count, offset + offset_offset, iobref);
-                GF_FREE (tmp_vec);
-                offset_offset += fill_size;
-                if (remaining_size == 0)
-                        break;
-        }
-
-        return 0;
-err:
-        STACK_UNWIND_STRICT (writev, frame, -1, op_errno, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-stripe_release (xlator_t *this, fd_t *fd)
-{
-        uint64_t          tmp_fctx = 0;
-        stripe_fd_ctx_t  *fctx = NULL;
-
-        VALIDATE_OR_GOTO (this, err);
-        VALIDATE_OR_GOTO (fd, err);
-
-        fd_ctx_del (fd, this, &tmp_fctx);
-        if (!tmp_fctx) {
-                goto err;
-        }
-
-        fctx = (stripe_fd_ctx_t *)(long)tmp_fctx;
-
-        if (!fctx->static_array)
-                GF_FREE (fctx->xl_array);
-
-        GF_FREE (fctx);
-
-err:
-	return 0;
-}
-
-
-int32_t
-notify (xlator_t *this, int32_t event, void *data, ...)
-{
-        stripe_private_t *priv = NULL;
-        int               down_client = 0;
-        int               i = 0;
-
-        if (!this)
-                return 0;
-
-        priv = this->private;
-        if (!priv)
-                return 0;
-
-        switch (event)
-        {
-        case GF_EVENT_CHILD_UP:
-        {
-                /* get an index number to set */
-                for (i = 0; i < priv->child_count; i++) {
-                        if (data == priv->xl_array[i])
-                                break;
-                }
-                priv->state[i] = 1;
-                for (i = 0; i < priv->child_count; i++) {
-                        if (!priv->state[i])
-                                down_client++;
-                }
-
-                LOCK (&priv->lock);
-                {
-                        priv->nodes_down = down_client;
-                        if (data == FIRST_CHILD (this))
-                                priv->first_child_down = 0;
-                        if (!priv->nodes_down)
-                                default_notify (this, event, data);
-                }
-                UNLOCK (&priv->lock);
-        }
-        break;
-        case GF_EVENT_CHILD_DOWN:
-        {
-                /* get an index number to set */
-                for (i = 0; i < priv->child_count; i++) {
-                        if (data == priv->xl_array[i])
-                                break;
-                }
-                priv->state[i] = 0;
-                for (i = 0; i < priv->child_count; i++) {
-                        if (!priv->state[i])
-                                down_client++;
-                }
-
-                LOCK (&priv->lock);
-                {
-                        priv->nodes_down = down_client;
-
-                        if (data == FIRST_CHILD (this))
-                                priv->first_child_down = 1;
-                        if (priv->nodes_down)
-                                default_notify (this, event, data);
-                }
-                UNLOCK (&priv->lock);
-        }
-        break;
-
-        default:
-        {
-                /* */
-                default_notify (this, event, data);
-        }
-        break;
-        }
-
-        return 0;
-}
-
-int
-set_stripe_block_size (xlator_t *this, stripe_private_t *priv, char *data)
-{
-        int                    ret = -1;
-        char                  *tmp_str = NULL;
-        char                  *tmp_str1 = NULL;
-        char                  *dup_str = NULL;
-        char                  *stripe_str = NULL;
-        char                  *pattern = NULL;
-        char                  *num = NULL;
-        struct stripe_options *temp_stripeopt = NULL;
-        struct stripe_options *stripe_opt = NULL;
-
-        if (!this || !priv || !data)
-                goto out;
-
-        /* Get the pattern for striping.
-           "option block-size *avi:10MB" etc */
-        stripe_str = strtok_r (data, ",", &tmp_str);
-        while (stripe_str) {
-                dup_str = gf_strdup (stripe_str);
-                stripe_opt = CALLOC (1, sizeof (struct stripe_options));
-                if (!stripe_opt) {
-                        GF_FREE (dup_str);
-                        goto out;
-                }
-
-                pattern = strtok_r (dup_str, ":", &tmp_str1);
-                num = strtok_r (NULL, ":", &tmp_str1);
-                if (!num) {
-                        num = pattern;
-                        pattern = "*";
-                }
-                if (gf_string2bytesize (num, &stripe_opt->block_size) != 0) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "invalid number format \"%s\"", num);
-                        goto out;
-                }
-                memcpy (stripe_opt->path_pattern, pattern, strlen (pattern));
-
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "block-size : pattern %s : size %"PRId64,
-                        stripe_opt->path_pattern, stripe_opt->block_size);
-
-                if (!priv->pattern) {
-                        priv->pattern = stripe_opt;
-                } else {
-                        temp_stripeopt = priv->pattern;
-                        while (temp_stripeopt->next)
-                                temp_stripeopt = temp_stripeopt->next;
-                        temp_stripeopt->next = stripe_opt;
-                }
-                stripe_str = strtok_r (NULL, ",", &tmp_str);
-                GF_FREE (dup_str);
-        }
-
-        ret = 0;
-out:
-        return ret;
-}
-
-int32_t
-mem_acct_init (xlator_t *this)
-{
-        int     ret = -1;
-
-        if (!this)
-                goto out;
-
-        ret = xlator_mem_acct_init (this, gf_stripe_mt_end + 1);
-
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
-                        "failed");
-                goto out;
-        }
-
-out:
-        return ret;
-}
-int
-validate_options (xlator_t *this, dict_t *options, char **op_errstr)
-{
-
-
-        data_t           *data = NULL;
-        int               ret = 0;
-        stripe_private_t *priv = NULL;
-
-        data = dict_get (options, "block-size");
-        if (data) {
-                gf_log (this->name, GF_LOG_TRACE,"Reconfiguring Stripe"
-                                " Block-size");
-                priv = GF_CALLOC (1, sizeof (stripe_private_t),
-                                  gf_stripe_mt_stripe_private_t);
-                if (!priv) {
-                        gf_log ("",GF_LOG_ERROR, "Unable to allocate memory");
-                        ret = -1;
-                        goto out;
-                }
-
-                ret = set_stripe_block_size (this, priv, data->data);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Reconfigue: Block-Size reconfiguration failed");
-                        *op_errstr = gf_strdup ("Error, could not parse list");
-                        ret = -1;
-                        goto out;
-                }
-                gf_log (this->name, GF_LOG_TRACE,
-                        "Reconfigue: Block-Size reconfigured Successfully");
-        }
-
-out:
-                if (priv)
-                GF_FREE (priv);
-        return ret;
-
-}
-
-int
-reconfigure (xlator_t *this, dict_t *options)
-{
-
-	stripe_private_t *priv = NULL;
-	data_t           *data = NULL;
-	int 		  ret = 0;
-
-
-	priv = this->private;
-
-	data = dict_get (options, "block-size");
-        if (data) {
-		gf_log (this->name, GF_LOG_TRACE,"Reconfiguring Stripe"
-			" Block-size");
-                ret = set_stripe_block_size (this, priv, data->data);
-                if (ret) {
-			gf_log (this->name, GF_LOG_ERROR,
-                        "Reconfigue: Block-Size reconfiguration failed");
-                        ret = -1;
-			goto out;
-		}
-		gf_log (this->name, GF_LOG_TRACE,
-                        "Reconfigue: Block-Size reconfigured Successfully");
-	}
-        else {
-                priv->block_size = (128 * GF_UNIT_KB);
-        }
-                        
-
-out:
-	return ret;
-
-}
-
-/**
- * init - This function is called when xlator-graph gets initialized.
- *     The option given in volfiles are parsed here.
- * @this -
- */
-int32_t
-init (xlator_t *this)
-{
-        stripe_private_t *priv = NULL;
-        xlator_list_t    *trav = NULL;
-        data_t           *data = NULL;
-        int32_t           count = 0;
-        int               ret = -1;
-
-        if (!this)
-                goto out;
-
-        trav = this->children;
-        while (trav) {
-                count++;
-                trav = trav->next;
-        }
-
-        if (!count) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "stripe configured without \"subvolumes\" option. "
-                        "exiting");
-                goto out;
-        }
-
-        if (!this->parents) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "dangling volume. check volfile ");
-        }
-
-        if (count == 1) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "stripe configured with only one \"subvolumes\" option."
-                        " please check the volume. exiting");
-                goto out;
-        }
-
-        priv = GF_CALLOC (1, sizeof (stripe_private_t),
-                          gf_stripe_mt_stripe_private_t);
-
-        if (!priv)
-                goto out;
-        priv->xl_array = GF_CALLOC (count, sizeof (xlator_t *),
-                                    gf_stripe_mt_xlator_t);
-        if (!priv->xl_array)
-                goto out;
-
-        priv->state = GF_CALLOC (count, sizeof (int8_t),
-                                 gf_stripe_mt_int8_t);
-        if (!priv->state)
-                goto out;
-
-        priv->child_count = count;
-        LOCK_INIT (&priv->lock);
-
-        trav = this->children;
-        count = 0;
-        while (trav) {
-                priv->xl_array[count++] = trav->xlator;
-                trav = trav->next;
-        }
-
-        if (count > 256) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "maximum number of stripe subvolumes supported "
-                        "is 256");
-                goto out;
-        }
-
-        priv->block_size = (128 * GF_UNIT_KB);
-        /* option stripe-pattern *avi:1GB,*pdf:4096 */
-        data = dict_get (this->options, "block-size");
-        if (!data) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "No \"option block-size <x>\" given, defaulting "
-                        "to 128KB");
-        } else {
-                ret = set_stripe_block_size (this, priv, data->data);
-                if (ret)
-                        goto out;
-        }
-
-        priv->xattr_supported = 1;
-        data = dict_get (this->options, "use-xattr");
-        if (data) {
-                if (gf_string2boolean (data->data,
-                                       &priv->xattr_supported) == -1) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "error setting hard check for extended "
-                                "attribute");
-                        //return -1;
-                }
-        }
-
-        /* notify related */
-        priv->nodes_down = priv->child_count;
-        this->private = priv;
-
-        ret = 0;
-out:
-        if (ret) {
-                if (priv) {
-                        if (priv->xl_array)
-                                GF_FREE (priv->xl_array);
-                        GF_FREE (priv);
-                }
-        }
-        return ret;
-}
-
-/**
- * fini -   Free all the private variables
- * @this -
- */
-void
-fini (xlator_t *this)
-{
-        stripe_private_t      *priv = NULL;
-        struct stripe_options *prev = NULL;
-        struct stripe_options *trav = NULL;
-
-        if (!this)
-                goto out;
-
-        priv = this->private;
-        if (priv) {
-                if (priv->xl_array)
-                        GF_FREE (priv->xl_array);
-
-                trav = priv->pattern;
-                while (trav) {
-                        prev = trav;
-                        trav = trav->next;
-                        FREE (prev);
-                }
-                LOCK_DESTROY (&priv->lock);
-                GF_FREE (priv);
-        }
-
-out:
-        return;
-}
-
-
-struct xlator_fops fops = {
-        .stat        = stripe_stat,
-        .unlink      = stripe_unlink,
-        .rename      = stripe_rename,
-        .link        = stripe_link,
-        .truncate    = stripe_truncate,
-        .create      = stripe_create,
-        .open        = stripe_open,
-        .readv       = stripe_readv,
-        .writev      = stripe_writev,
-        .statfs      = stripe_statfs,
-        .flush       = stripe_flush,
-        .fsync       = stripe_fsync,
-        .ftruncate   = stripe_ftruncate,
-        .fstat       = stripe_fstat,
-        .mkdir       = stripe_mkdir,
-        .rmdir       = stripe_rmdir,
-        .lk          = stripe_lk,
-        .opendir     = stripe_opendir,
-        .fsyncdir    = stripe_fsyncdir,
-        .setattr     = stripe_setattr,
-        .fsetattr    = stripe_fsetattr,
-        .lookup      = stripe_lookup,
-        .mknod       = stripe_mknod,
-};
-
-struct xlator_cbks cbks = {
-        .release = stripe_release,
-};
-
-
-struct volume_options options[] = {
-        { .key  = {"block-size"},
-          .type = GF_OPTION_TYPE_ANY
-        },
-        { .key  = {"use-xattr"},
-          .type = GF_OPTION_TYPE_BOOL
-        },
-        { .key  = {NULL} },
-};
diff --git a/xlators/cluster/stripe/src/stripe.h b/xlators/cluster/stripe/src/stripe.h
deleted file mode 100644
index 18fa158c9ef..00000000000
--- a/xlators/cluster/stripe/src/stripe.h
+++ /dev/null
@@ -1,159 +0,0 @@
-/*
-  Copyright (c) 2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-
-#ifndef _STRIPE_H_
-#define _STRIPE_H_
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "xlator.h"
-#include "logging.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat.h"
-#include "compat-errno.h"
-#include "stripe-mem-types.h"
-#include <fnmatch.h>
-#include <signal.h>
-
-
-/**
- * struct stripe_options : This keeps the pattern and the block-size
- *     information, which is used for striping on a file.
- */
-struct stripe_options {
-        struct stripe_options *next;
-        char                   path_pattern[256];
-        uint64_t               block_size;
-};
-
-/**
- * Private structure for stripe translator
- */
-struct stripe_private {
-        struct stripe_options  *pattern;
-        xlator_t              **xl_array;
-        uint64_t                block_size;
-        gf_lock_t               lock;
-        uint8_t                 nodes_down;
-        int8_t                  first_child_down;
-        int8_t                  child_count;
-        int8_t                 *state; /* Current state of child node */
-        gf_boolean_t            xattr_supported;  /* default yes */
-};
-
-/**
- * Used to keep info about the replies received from fops->readv calls
- */
-struct readv_replies {
-        struct iovec *vector;
-        int32_t       count;    //count of vector
-        int32_t       op_ret;   //op_ret of readv
-        int32_t       op_errno;
-        int32_t       requested_size;
-        struct iatt   stbuf;    /* 'stbuf' is also a part of reply */
-};
-
-typedef struct _stripe_fd_ctx {
-        off_t      stripe_size;
-        int        stripe_count;
-        int        static_array;
-        xlator_t **xl_array;
-} stripe_fd_ctx_t;
-
-
-/**
- * Local structure to be passed with all the frames in case of STACK_WIND
- */
-struct stripe_local; /* this itself is used inside the structure; */
-
-struct stripe_local {
-        struct stripe_local *next;
-        call_frame_t        *orig_frame;
-
-        stripe_fd_ctx_t     *fctx;
-
-        /* Used by _cbk functions */
-        struct iatt          stbuf;
-        struct iatt          pre_buf;
-        struct iatt          post_buf;
-        struct iatt          preparent;
-        struct iatt          postparent;
-
-        off_t                stbuf_size;
-        off_t                prebuf_size;
-        off_t                postbuf_size;
-        off_t                preparent_size;
-        off_t                postparent_size;
-
-        blkcnt_t             stbuf_blocks;
-        blkcnt_t             prebuf_blocks;
-        blkcnt_t             postbuf_blocks;
-        blkcnt_t             preparent_blocks;
-        blkcnt_t             postparent_blocks;
-
-        struct readv_replies *replies;
-        struct statvfs       statvfs_buf;
-        dir_entry_t         *entry;
-
-        int8_t               revalidate;
-        int8_t               failed;
-        int8_t               unwind;
-
-        size_t               readv_size;
-        int32_t              entry_count;
-        int32_t              node_index;
-        int32_t              call_count;
-        int32_t              wind_count; /* used instead of child_cound
-                                            in case of read and write */
-        int32_t              op_ret;
-        int32_t              op_errno;
-        int32_t              count;
-        int32_t              flags;
-        char                *name;
-        inode_t             *inode;
-
-        loc_t                loc;
-        loc_t                loc2;
-
-        /* For File I/O fops */
-        dict_t              *dict;
-
-        /* General usage */
-        off_t                offset;
-        off_t                stripe_size;
-
-        int xattr_self_heal_needed;
-        int entry_self_heal_needed;
-
-        int8_t              *list;
-        struct gf_flock         lock;
-        fd_t                *fd;
-        void                *value;
-        struct iobref       *iobref;
-};
-
-typedef struct stripe_local   stripe_local_t;
-typedef struct stripe_private stripe_private_t;
-
-#endif /* _STRIPE_H_ */
diff --git a/xlators/cluster/unify/src/Makefile.am b/xlators/cluster/unify/src/Makefile.am
deleted file mode 100644
index 2a1fe837263..00000000000
--- a/xlators/cluster/unify/src/Makefile.am
+++ /dev/null
@@ -1,16 +0,0 @@
-
-xlator_LTLIBRARIES = unify.la
-xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/legacy/cluster
-
-unify_la_LDFLAGS = -module -avoidversion
-
-unify_la_SOURCES = unify.c unify-self-heal.c
-unify_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
-
-noinst_HEADERS = unify.h
-
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS) \
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
-
-CLEANFILES = 
-
diff --git a/xlators/cluster/unify/src/unify-mem-types.h b/xlators/cluster/unify/src/unify-mem-types.h
deleted file mode 100644
index 9da061cfb0c..00000000000
--- a/xlators/cluster/unify/src/unify-mem-types.h
+++ /dev/null
@@ -1,41 +0,0 @@
-
-/*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-
-#ifndef __UNIFY_MEM_TYPES_H__
-#define __UNIFY_MEM_TYPES_H__
-
-#include "mem-types.h"
-
-enum gf_unify_mem_types_ {
-        gf_unify_mt_char = gf_common_mt_end + 1,
-        gf_unify_mt_int16_t,
-        gf_unify_mt_xlator_t,
-        gf_unify_mt_unify_private_t,
-        gf_unify_mt_xlator_list_t,
-        gf_unify_mt_dir_entry_t,
-        gf_unify_mt_off_t,
-        gf_unify_mt_int,
-        gf_unify_mt_unify_self_heal_struct,
-        gf_unify_mt_unify_local_t,
-        gf_unify_mt_end
-};
-#endif
-
diff --git a/xlators/cluster/unify/src/unify-self-heal.c b/xlators/cluster/unify/src/unify-self-heal.c
deleted file mode 100644
index b3013d363c4..00000000000
--- a/xlators/cluster/unify/src/unify-self-heal.c
+++ /dev/null
@@ -1,1239 +0,0 @@
-/*
-  Copyright (c) 2007-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-/**
- * unify-self-heal.c : 
- *   This file implements few functions which enables 'unify' translator 
- *  to be consistent in its behaviour when 
- *     > a node fails, 
- *     > a node gets added, 
- *     > a failed node comes back
- *     > a new namespace server is added (ie, an fresh namespace server).
- * 
- *  This functionality of 'unify' will enable glusterfs to support storage
- *  system failure, and maintain consistancy. This works both ways, ie, when
- *  an entry (either file or directory) is found on namespace server, and not
- *  on storage nodes, its created in storage nodes and vica-versa.
- * 
- *  The two fops, where it can be implemented are 'getdents ()' and 'lookup ()'
- *
- */
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "unify.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "common-utils.h"
-
-int32_t
-unify_sh_getdents_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno,
-		       dir_entry_t *entry,
-		       int32_t count);
-
-int32_t
-unify_sh_ns_getdents_cbk (call_frame_t *frame,
-			  void *cookie,
-			  xlator_t *this,
-			  int32_t op_ret,
-			  int32_t op_errno,
-			  dir_entry_t *entry,
-			  int32_t count);
-
-int32_t
-unify_bgsh_getdents_cbk (call_frame_t *frame,
-			 void *cookie,
-			 xlator_t *this,
-			 int32_t op_ret,
-			 int32_t op_errno,
-			 dir_entry_t *entry,
-			 int32_t count);
-
-int32_t
-unify_bgsh_ns_getdents_cbk (call_frame_t *frame,
-			    void *cookie,
-			    xlator_t *this,
-			    int32_t op_ret,
-			    int32_t op_errno,
-			    dir_entry_t *entry,
-			    int32_t count);
-
-/**
- * unify_local_wipe - free all the extra allocation of local->* here.
- */
-static void 
-unify_local_wipe (unify_local_t *local)
-{
-	/* Free the strdup'd variables in the local structure */
-	if (local->name) {
-		GF_FREE (local->name);
-	}
-
-	if (local->sh_struct) {
-		if (local->sh_struct->offset_list)
-			GF_FREE (local->sh_struct->offset_list);
-
-		if (local->sh_struct->entry_list)
-			GF_FREE (local->sh_struct->entry_list);
-
-		if (local->sh_struct->count_list)
-			GF_FREE (local->sh_struct->count_list);
-
-		GF_FREE (local->sh_struct);
-	}
-
-	loc_wipe (&local->loc1);
-	loc_wipe (&local->loc2);
-}
-
-int32_t 
-unify_sh_setdents_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno)
-{
-	int32_t callcnt = -1;
-	unify_local_t *local = frame->local;
-	inode_t *inode = NULL;
-	dict_t *tmp_dict = NULL;
-	dir_entry_t *prev, *entry, *trav;
-
-	LOCK (&frame->lock);
-	{
-		/* if local->call_count == 0, that means, setdents on 
-		 * storagenodes is still pending.
-		 */
-		if (local->call_count)
-			callcnt = --local->call_count;
-	}
-	UNLOCK (&frame->lock);
-
-	if (callcnt == 0) {
-		if (local->sh_struct->entry_list[0]) {
-			prev = entry = local->sh_struct->entry_list[0];
-			if (!entry)
-				return 0;
-			trav = entry->next;
-			while (trav) {
-				prev->next = trav->next;
-				GF_FREE (trav->name);
-				if (IA_ISLNK (trav->buf.ia_type))
-					GF_FREE (trav->link);
-				GF_FREE (trav);
-				trav = prev->next;
-			}
-			GF_FREE (entry);
-		}
-
-		if (!local->flags) {
-			if (local->sh_struct->count_list[0] >= 
-			    UNIFY_SELF_HEAL_GETDENTS_COUNT) {
-				/* count == size, that means, there are more entries
-				   to read from */
-				//local->call_count = 0;
-				local->sh_struct->offset_list[0] += 
-					UNIFY_SELF_HEAL_GETDENTS_COUNT;
-				STACK_WIND (frame,
-					    unify_sh_ns_getdents_cbk,
-					    NS(this),
-					    NS(this)->fops->getdents,
-					    local->fd,
-					    UNIFY_SELF_HEAL_GETDENTS_COUNT,
-					    local->sh_struct->offset_list[0],
-					    GF_GET_DIR_ONLY);
-			}		
-		} else {
-			inode = local->loc1.inode;
-			fd_unref (local->fd);
-			tmp_dict = local->dict;
-
-			unify_local_wipe (local);
-			
-			STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-				      inode, &local->stbuf, local->dict,
-                                      &local->oldpostparent);
-			if (tmp_dict)
-				dict_unref (tmp_dict);
-		}
-	}
-  
-	return 0;
-}
-
-
-int32_t
-unify_sh_ns_getdents_cbk (call_frame_t *frame,
-			  void *cookie,
-			  xlator_t *this,
-			  int32_t op_ret,
-			  int32_t op_errno,
-			  dir_entry_t *entry,
-			  int32_t count)
-{
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	long index = 0;
-	unsigned long final = 0;
-	dir_entry_t *tmp = GF_CALLOC (1, sizeof (dir_entry_t),
-                                      gf_unify_mt_dir_entry_t);
-	
-	local->sh_struct->entry_list[0] = tmp;
-	local->sh_struct->count_list[0] = count;
-	if (entry) {
-		tmp->next = entry->next;
-		entry->next = NULL;
-	}
-
-	if ((count < UNIFY_SELF_HEAL_GETDENTS_COUNT) || !entry) {
-		final = 1;
-	}
-
-	LOCK (&frame->lock);
-	{
-		/* local->call_count will be '0' till now. make it 1 so, it 
-		   can be UNWIND'ed for the last call. */
-		local->call_count = priv->child_count;
-		if (final)
-			local->flags = 1;
-	}
-	UNLOCK (&frame->lock);
-
-	for (index = 0; index < priv->child_count; index++) 
-	{
-		STACK_WIND_COOKIE (frame,
-				   unify_sh_setdents_cbk, 
-				   (void *)index,
-				   priv->xl_array[index],
-				   priv->xl_array[index]->fops->setdents,
-				   local->fd, GF_SET_DIR_ONLY,
-				   local->sh_struct->entry_list[0], count);
-	}
-
-	return 0;
-}
-
-int32_t 
-unify_sh_ns_setdents_cbk (call_frame_t *frame,
-			  void *cookie,
-			  xlator_t *this,
-			  int32_t op_ret,
-			  int32_t op_errno)
-{
-	int32_t callcnt = -1;
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	long index = (long)cookie;
-	dir_entry_t *prev, *entry, *trav;
-
-	LOCK (&frame->lock);
-	{
-		if (local->sh_struct->entry_list[index]) {
-			prev = entry = local->sh_struct->entry_list[index];
-			trav = entry->next;
-			while (trav) {
-				prev->next = trav->next;
-				GF_FREE (trav->name);
-				if (IA_ISLNK (trav->buf.ia_type))
-					GF_FREE (trav->link);
-				GF_FREE (trav);
-				trav = prev->next;
-			}
-			GF_FREE (entry);
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	if (local->sh_struct->count_list[index] < 
-	    UNIFY_SELF_HEAL_GETDENTS_COUNT) {
-		LOCK (&frame->lock);
-		{
-			callcnt = --local->call_count;
-		}
-		UNLOCK (&frame->lock);
-	} else {
-		/* count == size, that means, there are more entries 
-		   to read from */
-		local->sh_struct->offset_list[index] += 
-			UNIFY_SELF_HEAL_GETDENTS_COUNT;
-		STACK_WIND_COOKIE (frame,
-				   unify_sh_getdents_cbk,
-				   cookie,
-				   priv->xl_array[index],
-				   priv->xl_array[index]->fops->getdents,
-				   local->fd,
-				   UNIFY_SELF_HEAL_GETDENTS_COUNT,
-				   local->sh_struct->offset_list[index],
-				   GF_GET_ALL);
-    
-		gf_log (this->name, GF_LOG_DEBUG, 
-			"readdir on (%s) with offset %"PRId64"", 
-			priv->xl_array[index]->name, 
-			local->sh_struct->offset_list[index]);
-	}
-
-	if (!callcnt) {
-		/* All storage nodes have done unified setdents on NS node.
-		 * Now, do getdents from NS and do setdents on storage nodes.
-		 */
-    
-		/* sh_struct->offset_list is no longer required for
-		   storage nodes now */
-		local->sh_struct->offset_list[0] = 0; /* reset */
-
-		STACK_WIND (frame,
-			    unify_sh_ns_getdents_cbk,
-			    NS(this),
-			    NS(this)->fops->getdents,
-			    local->fd,
-			    UNIFY_SELF_HEAL_GETDENTS_COUNT,
-			    0, /* In this call, do send '0' as offset */
-			    GF_GET_DIR_ONLY);
-	}
-
-	return 0;
-}
-
-
-/**
- * unify_sh_getdents_cbk -
- */
-int32_t
-unify_sh_getdents_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno,
-		       dir_entry_t *entry,
-		       int32_t count)
-{
-	int32_t callcnt = -1;
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	long index = (long)cookie;
-	dir_entry_t *tmp = NULL; 
-
-	if (op_ret >= 0 && count > 0) {
-		/* There is some dentry found, just send the dentry to NS */
-		tmp = GF_CALLOC (1, sizeof (dir_entry_t),
-                                 gf_unify_mt_dir_entry_t);
-		local->sh_struct->entry_list[index] = tmp;
-		local->sh_struct->count_list[index] = count;
-		if (entry) {
-			tmp->next = entry->next;
-			entry->next = NULL;
-		}
-		STACK_WIND_COOKIE (frame,
-				   unify_sh_ns_setdents_cbk,
-				   cookie,
-				   NS(this),
-				   NS(this)->fops->setdents,
-				   local->fd,
-				   GF_SET_IF_NOT_PRESENT,
-				   local->sh_struct->entry_list[index],
-				   count);
-		return 0;
-	}
-  
-	if (count < UNIFY_SELF_HEAL_GETDENTS_COUNT) {
-		LOCK (&frame->lock);
-		{
-			callcnt = --local->call_count;
-		}
-		UNLOCK (&frame->lock);
-	} else {
-		/* count == size, that means, there are more entries 
-		   to read from */
-		local->sh_struct->offset_list[index] += 
-			UNIFY_SELF_HEAL_GETDENTS_COUNT;
-		STACK_WIND_COOKIE (frame,
-				   unify_sh_getdents_cbk,
-				   cookie,
-				   priv->xl_array[index],
-				   priv->xl_array[index]->fops->getdents,
-				   local->fd,
-				   UNIFY_SELF_HEAL_GETDENTS_COUNT,
-				   local->sh_struct->offset_list[index],
-				   GF_GET_ALL);
-    
-		gf_log (this->name, GF_LOG_DEBUG, 
-			"readdir on (%s) with offset %"PRId64"", 
-			priv->xl_array[index]->name, 
-			local->sh_struct->offset_list[index]);
-	}
-
-	if (!callcnt) {
-		/* All storage nodes have done unified setdents on NS node.
-		 * Now, do getdents from NS and do setdents on storage nodes.
-		 */
-    
-		/* sh_struct->offset_list is no longer required for
-		   storage nodes now */
-		local->sh_struct->offset_list[0] = 0; /* reset */
-
-		STACK_WIND (frame,
-			    unify_sh_ns_getdents_cbk,
-			    NS(this),
-			    NS(this)->fops->getdents,
-			    local->fd,
-			    UNIFY_SELF_HEAL_GETDENTS_COUNT,
-			    0, /* In this call, do send '0' as offset */
-			    GF_GET_DIR_ONLY);
-	}
-
-	return 0;
-}
-
-/**
- * unify_sh_opendir_cbk -
- *
- * @cookie: 
- */
-int32_t 
-unify_sh_opendir_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno,
-		      fd_t *fd)
-{
-	int32_t callcnt = 0;
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	int16_t index = 0;
-	inode_t *inode = NULL;
-	dict_t *tmp_dict = NULL;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-    
-		if (op_ret >= 0) {
-			local->op_ret = op_ret;
-		} else {
-			gf_log (this->name, GF_LOG_WARNING, "failed");
-			local->failed = 1;
-		}
-	}
-	UNLOCK (&frame->lock);
-  
-	if (!callcnt) {
-		local->call_count = priv->child_count + 1;
-    
-		if (!local->failed) {
-			/* send getdents() namespace after finishing
-			   storage nodes */
-			local->call_count--; 
-      
-			fd_bind (fd);
-
-			if (local->call_count) {
-				/* Used as the offset index. This list keeps
-				 * track of offset sent to each node during
-				 * STACK_WIND.
-				 */
-				local->sh_struct->offset_list = 
-					GF_CALLOC (priv->child_count, 
-					           sizeof (off_t),
-                                                   gf_unify_mt_off_t);
-				ERR_ABORT (local->sh_struct->offset_list);
-	
-				local->sh_struct->entry_list = 
-					GF_CALLOC (priv->child_count, 
-						   sizeof (dir_entry_t *),
-                                                   gf_unify_mt_dir_entry_t);
-				ERR_ABORT (local->sh_struct->entry_list);
-
-				local->sh_struct->count_list = 
-					GF_CALLOC (priv->child_count, 
-						   sizeof (int),
-                                                   gf_unify_mt_int);
-				ERR_ABORT (local->sh_struct->count_list);
-
-				/* Send getdents on all the fds */
-				for (index = 0; 
-				     index < priv->child_count; index++) {
-					STACK_WIND_COOKIE (frame,
-							   unify_sh_getdents_cbk,
-							   (void *)(long)index,
-							   priv->xl_array[index],
-							   priv->xl_array[index]->fops->getdents,
-							   local->fd,
-							   UNIFY_SELF_HEAL_GETDENTS_COUNT,
-							   0, /* In this call, do send '0' as offset */
-							   GF_GET_ALL);
-				}
-
-				/* did stack wind, so no need to unwind here */
-				return 0;
-			} /* (local->call_count) */
-		} /* (!local->failed) */
-
-		/* Opendir failed on one node. */ 
-		inode = local->loc1.inode;
-		fd_unref (local->fd);
-		tmp_dict = local->dict;
-
-		unify_local_wipe (local);
-		/* Only 'self-heal' failed, lookup() was successful. */
-		local->op_ret = 0;
-
-		/* This is lookup_cbk ()'s UNWIND. */
-		STACK_UNWIND (frame, local->op_ret, local->op_errno, inode,
-			      &local->stbuf, local->dict, &local->oldpostparent);
-		if (tmp_dict)
-			dict_unref (tmp_dict);
-	}
-
-	return 0;
-}
-
-/**
- * gf_sh_checksum_cbk - 
- * 
- * @frame: frame used in lookup. get a copy of it, and use that copy.
- * @this: pointer to unify xlator.
- * @inode: pointer to inode, for which the consistency check is required.
- *
- */
-int32_t 
-unify_sh_checksum_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno,
-		       uint8_t *file_checksum,
-		       uint8_t *dir_checksum)
-{
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	int16_t index = 0;
-	int32_t callcnt = 0;
-	inode_t *inode = NULL;
-	dict_t *tmp_dict = NULL;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-		if (op_ret >= 0) {
-			if (NS(this) == (xlator_t *)cookie) {
-				memcpy (local->sh_struct->ns_file_checksum, 
-					file_checksum, NAME_MAX);
-				memcpy (local->sh_struct->ns_dir_checksum, 
-					dir_checksum, NAME_MAX);
-			} else {
-				if (local->entry_count == 0) {
-					/* Initialize the dir_checksum to be 
-					 * used for comparision with other
-					 * storage nodes. Should be done for
-					 * the first successful call *only*. 
-					 */
-                                        /* Using 'entry_count' as a flag */
-					local->entry_count = 1;
-					memcpy (local->sh_struct->dir_checksum,
-						dir_checksum, NAME_MAX);
-				}
-
-				/* Reply from the storage nodes */
-				for (index = 0; 
-				     index < NAME_MAX; index++) {
-					/* Files should be present in
-					   only one node */
-					local->sh_struct->file_checksum[index] ^= file_checksum[index];
-	  
-					/* directory structure should be
-					   same accross */
-					if (local->sh_struct->dir_checksum[index] != dir_checksum[index])
-						local->failed = 1;
-				}
-			}
-		} 
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		for (index = 0; index < NAME_MAX ; index++) {
-			if (local->sh_struct->file_checksum[index] != 
-			    local->sh_struct->ns_file_checksum[index]) {
-				local->failed = 1;
-				break;
-			}
-			if (local->sh_struct->dir_checksum[index] != 
-			    local->sh_struct->ns_dir_checksum[index]) {
-				local->failed = 1;
-				break;
-			}
-		}
-	
-		if (local->failed) {
-			/* Log it, it should be a rare event */
-			gf_log (this->name, GF_LOG_WARNING, 
-				"Self-heal triggered on directory %s", 
-				local->loc1.path);
-
-			/* Any self heal will be done at directory level */
-			local->call_count = 0;
-			local->op_ret = -1;
-			local->failed = 0;
-      
-			local->fd = fd_create (local->loc1.inode, 
-					       frame->root->pid);
-
-			local->call_count = priv->child_count + 1;
-	
-			for (index = 0; 
-			     index < (priv->child_count + 1); index++) {
-				STACK_WIND_COOKIE (frame,
-						   unify_sh_opendir_cbk,
-						   priv->xl_array[index]->name,
-						   priv->xl_array[index],
-						   priv->xl_array[index]->fops->opendir,
-						   &local->loc1,
-						   local->fd);
-			}
-			/* opendir can be done on the directory */
-			return 0;
-		}
-
-		/* no mismatch */
-		inode = local->loc1.inode;
-		tmp_dict = local->dict;
-
-		unify_local_wipe (local);
-
-		/* This is lookup_cbk ()'s UNWIND. */
-		STACK_UNWIND (frame,
-			      local->op_ret,
-			      local->op_errno,
-			      inode,
-			      &local->stbuf,
-			      local->dict, &local->oldpostparent);
-		if (tmp_dict)
-			dict_unref (tmp_dict);
-	}
-
-	return 0;
-}
-
-/* Foreground self-heal part over */
-
-/* Background self-heal part */
-
-int32_t 
-unify_bgsh_setdents_cbk (call_frame_t *frame,
-			 void *cookie,
-			 xlator_t *this,
-			 int32_t op_ret,
-			 int32_t op_errno)
-{
-	int32_t callcnt = -1;
-	unify_local_t *local = frame->local;
-	dir_entry_t *prev, *entry, *trav;
-
-	LOCK (&frame->lock);
-	{
-		/* if local->call_count == 0, that means, setdents 
-		   on storagenodes is still pending. */
-		if (local->call_count)
-			callcnt = --local->call_count;
-	}
-	UNLOCK (&frame->lock);
-
-
-	if (callcnt == 0) {
-		if (local->sh_struct->entry_list[0]) {
-			prev = entry = local->sh_struct->entry_list[0];
-			trav = entry->next;
-			while (trav) {
-				prev->next = trav->next;
-				GF_FREE (trav->name);
-				if (IA_ISLNK (trav->buf.ia_type))
-					GF_FREE (trav->link);
-				GF_FREE (trav);
-				trav = prev->next;
-			}
-			GF_FREE (entry);
-		}
-
-		if (!local->flags) {
-			if (local->sh_struct->count_list[0] >= 
-			    UNIFY_SELF_HEAL_GETDENTS_COUNT) {
-				/* count == size, that means, there are more
-				   entries to read from */
-				//local->call_count = 0;
-				local->sh_struct->offset_list[0] += 
-					UNIFY_SELF_HEAL_GETDENTS_COUNT;
-				STACK_WIND (frame,
-					    unify_bgsh_ns_getdents_cbk,
-					    NS(this),
-					    NS(this)->fops->getdents,
-					    local->fd,
-					    UNIFY_SELF_HEAL_GETDENTS_COUNT,
-					    local->sh_struct->offset_list[0],
-					    GF_GET_DIR_ONLY);
-			}		
-		} else {
-			fd_unref (local->fd);
-			unify_local_wipe (local);
-			STACK_DESTROY (frame->root);
-		}
-	}
-
-	return 0;
-}
-
-
-int32_t
-unify_bgsh_ns_getdents_cbk (call_frame_t *frame,
-			    void *cookie,
-			    xlator_t *this,
-			    int32_t op_ret,
-			    int32_t op_errno,
-			    dir_entry_t *entry,
-			    int32_t count)
-{
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	long index = 0;
-	unsigned long final = 0;
-	dir_entry_t *tmp = GF_CALLOC (1, sizeof (dir_entry_t),
-                                      gf_unify_mt_dir_entry_t);
-
-	local->sh_struct->entry_list[0] = tmp;
-	local->sh_struct->count_list[0] = count;
-	if (entry) {
-		tmp->next = entry->next;
-		entry->next = NULL;
-	}
-
-	if ((count < UNIFY_SELF_HEAL_GETDENTS_COUNT) || !entry) {
-		final = 1;
-	}
-
-	LOCK (&frame->lock);
-	{
-		/* local->call_count will be '0' till now. make it 1 so, 
-		   it can be UNWIND'ed for the last call. */
-		local->call_count = priv->child_count;
-		if (final)
-			local->flags = 1;
-	}
-	UNLOCK (&frame->lock);
-
-	for (index = 0; index < priv->child_count; index++) 
-	{
-		STACK_WIND_COOKIE (frame,
-				   unify_bgsh_setdents_cbk, 
-				   (void *)index,
-				   priv->xl_array[index],
-				   priv->xl_array[index]->fops->setdents,
-				   local->fd, GF_SET_DIR_ONLY,
-				   local->sh_struct->entry_list[0], count);
-	}
-
-	return 0;
-}
-
-int32_t 
-unify_bgsh_ns_setdents_cbk (call_frame_t *frame,
-			    void *cookie,
-			    xlator_t *this,
-			    int32_t op_ret,
-			    int32_t op_errno)
-{
-	int32_t callcnt = -1;
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	long index = (long)cookie;
-	dir_entry_t *prev, *entry, *trav;
-
-	if (local->sh_struct->entry_list[index]) {
-		prev = entry = local->sh_struct->entry_list[index];
-		if (!entry)
-			return 0;
-		trav = entry->next;
-		while (trav) {
-			prev->next = trav->next;
-			GF_FREE (trav->name);
-			if (IA_ISLNK (trav->buf.ia_type))
-				GF_FREE (trav->link);
-			GF_FREE (trav);
-			trav = prev->next;
-		}
-		GF_FREE (entry);
-	}
-
-	if (local->sh_struct->count_list[index] < 
-	    UNIFY_SELF_HEAL_GETDENTS_COUNT) {
-		LOCK (&frame->lock);
-		{
-			callcnt = --local->call_count;
-		}
-		UNLOCK (&frame->lock);
-	} else {
-		/* count == size, that means, there are more entries 
-		   to read from */
-		local->sh_struct->offset_list[index] += 
-			UNIFY_SELF_HEAL_GETDENTS_COUNT;
-		STACK_WIND_COOKIE (frame,
-				   unify_bgsh_getdents_cbk,
-				   cookie,
-				   priv->xl_array[index],
-				   priv->xl_array[index]->fops->getdents,
-				   local->fd,
-				   UNIFY_SELF_HEAL_GETDENTS_COUNT,
-				   local->sh_struct->offset_list[index],
-				   GF_GET_ALL);
-    
-		gf_log (this->name, GF_LOG_DEBUG, 
-			"readdir on (%s) with offset %"PRId64"", 
-			priv->xl_array[index]->name, 
-			local->sh_struct->offset_list[index]);
-	}
-
-	if (!callcnt) {
-		/* All storage nodes have done unified setdents on NS node.
-		 * Now, do getdents from NS and do setdents on storage nodes.
-		 */
-    
-		/* sh_struct->offset_list is no longer required for
-		   storage nodes now */
-		local->sh_struct->offset_list[0] = 0; /* reset */
-
-		STACK_WIND (frame,
-			    unify_bgsh_ns_getdents_cbk,
-			    NS(this),
-			    NS(this)->fops->getdents,
-			    local->fd,
-			    UNIFY_SELF_HEAL_GETDENTS_COUNT,
-			    0, /* In this call, do send '0' as offset */
-			    GF_GET_DIR_ONLY);
-	}
-
-	return 0;
-}
-
-
-/**
- * unify_bgsh_getdents_cbk -
- */
-int32_t
-unify_bgsh_getdents_cbk (call_frame_t *frame,
-			 void *cookie,
-			 xlator_t *this,
-			 int32_t op_ret,
-			 int32_t op_errno,
-			 dir_entry_t *entry,
-			 int32_t count)
-{
-	int32_t callcnt = -1;
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	long index = (long)cookie;
-	dir_entry_t *tmp = NULL; 
-
-	if (op_ret >= 0 && count > 0) {
-		/* There is some dentry found, just send the dentry to NS */
-		tmp = GF_CALLOC (1, sizeof (dir_entry_t),
-                                 gf_unify_mt_dir_entry_t);
-		local->sh_struct->entry_list[index] = tmp;
-		local->sh_struct->count_list[index] = count;
-		if (entry) {
-			tmp->next = entry->next;
-			entry->next = NULL;
-		}
-		STACK_WIND_COOKIE (frame,
-				   unify_bgsh_ns_setdents_cbk,
-				   cookie,
-				   NS(this),
-				   NS(this)->fops->setdents,
-				   local->fd,
-				   GF_SET_IF_NOT_PRESENT,
-				   local->sh_struct->entry_list[index],
-				   count);
-		return 0;
-	}
-  
-	if (count < UNIFY_SELF_HEAL_GETDENTS_COUNT) {
-		LOCK (&frame->lock);
-		{
-			callcnt = --local->call_count;
-		}
-		UNLOCK (&frame->lock);
-	} else {
-		/* count == size, that means, there are more entries to read from */
-		local->sh_struct->offset_list[index] += 
-			UNIFY_SELF_HEAL_GETDENTS_COUNT;
-
-		STACK_WIND_COOKIE (frame,
-				   unify_bgsh_getdents_cbk,
-				   cookie,
-				   priv->xl_array[index],
-				   priv->xl_array[index]->fops->getdents,
-				   local->fd,
-				   UNIFY_SELF_HEAL_GETDENTS_COUNT,
-				   local->sh_struct->offset_list[index],
-				   GF_GET_ALL);
-    
-		gf_log (this->name, GF_LOG_DEBUG, 
-			"readdir on (%s) with offset %"PRId64"", 
-			priv->xl_array[index]->name, 
-			local->sh_struct->offset_list[index]);
-	}
-
-	if (!callcnt) {
-		/* All storage nodes have done unified setdents on NS node.
-		 * Now, do getdents from NS and do setdents on storage nodes.
-		 */
-    
-		/* sh_struct->offset_list is no longer required for 
-		   storage nodes now */
-		local->sh_struct->offset_list[0] = 0; /* reset */
-
-		STACK_WIND (frame,
-			    unify_bgsh_ns_getdents_cbk,
-			    NS(this),
-			    NS(this)->fops->getdents,
-			    local->fd,
-			    UNIFY_SELF_HEAL_GETDENTS_COUNT,
-			    0, /* In this call, do send '0' as offset */
-			    GF_GET_DIR_ONLY);
-	}
-
-	return 0;
-}
-
-/**
- * unify_bgsh_opendir_cbk -
- *
- * @cookie: 
- */
-int32_t 
-unify_bgsh_opendir_cbk (call_frame_t *frame,
-			void *cookie,
-			xlator_t *this,
-			int32_t op_ret,
-			int32_t op_errno,
-			fd_t *fd)
-{
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	int32_t callcnt = 0;
-	int16_t index = 0;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-    
-		if (op_ret >= 0) {
-			local->op_ret = op_ret;
-		} else {
-			local->failed = 1;
-		}
-	}
-	UNLOCK (&frame->lock);
-  
-	if (!callcnt) {
-		local->call_count = priv->child_count + 1;
-    
-		if (!local->failed) {
-			/* send getdents() namespace after finishing 
-			   storage nodes */
-			local->call_count--; 
-			callcnt = local->call_count;
-      
-			fd_bind (fd);
-
-			if (local->call_count) {
-				/* Used as the offset index. This list keeps 
-				   track of offset sent to each node during 
-				   STACK_WIND. */
-				local->sh_struct->offset_list = 
-					GF_CALLOC (priv->child_count, 
-					           sizeof (off_t),
-                                                   gf_unify_mt_off_t);
-				ERR_ABORT (local->sh_struct->offset_list);
-	
-				local->sh_struct->entry_list = 
-					GF_CALLOC (priv->child_count, 
-						   sizeof (dir_entry_t *),
-                                                   gf_unify_mt_dir_entry_t);
-				ERR_ABORT (local->sh_struct->entry_list);
-
-				local->sh_struct->count_list = 
-					GF_CALLOC (priv->child_count, 
-						   sizeof (int),
-                                                   gf_unify_mt_int);
-				ERR_ABORT (local->sh_struct->count_list);
-
-				/* Send getdents on all the fds */
-				for (index = 0; 
-				     index < priv->child_count; index++) {
-					STACK_WIND_COOKIE (frame,
-							   unify_bgsh_getdents_cbk,
-							   (void *)(long)index,
-							   priv->xl_array[index],
-							   priv->xl_array[index]->fops->getdents,
-							   local->fd,
-							   UNIFY_SELF_HEAL_GETDENTS_COUNT,
-							   0, /* In this call, do send '0' as offset */
-							   GF_GET_ALL);
-				}
-				/* did a stack wind, so no need to unwind here */
-				return 0;
-			} /* (local->call_count) */
-		} /* (!local->failed) */
-
-		/* Opendir failed on one node. 	 */
-		fd_unref (local->fd);
-		
-		unify_local_wipe (local);
-		STACK_DESTROY (frame->root);
-	}
-
-	return 0;
-}
-
-/**
- * gf_bgsh_checksum_cbk - 
- * 
- * @frame: frame used in lookup. get a copy of it, and use that copy.
- * @this: pointer to unify xlator.
- * @inode: pointer to inode, for which the consistency check is required.
- *
- */
-int32_t 
-unify_bgsh_checksum_cbk (call_frame_t *frame,
-			 void *cookie,
-			 xlator_t *this,
-			 int32_t op_ret,
-			 int32_t op_errno,
-			 uint8_t *file_checksum,
-			 uint8_t *dir_checksum)
-{
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	int16_t index = 0;
-	int32_t callcnt = 0;
-  
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-		if (op_ret >= 0) {
-			if (NS(this) == (xlator_t *)cookie) {
-				memcpy (local->sh_struct->ns_file_checksum, 
-					file_checksum, NAME_MAX);
-				memcpy (local->sh_struct->ns_dir_checksum, 
-					dir_checksum, NAME_MAX);
-			} else {
-				if (local->entry_count == 0) {
-					/* Initialize the dir_checksum to be 
-					 * used for comparision with other 
-					 * storage nodes. Should be done for
-					 * the first successful call *only*. 
-					 */
-					/* Using 'entry_count' as a flag */
-					local->entry_count = 1; 
-					memcpy (local->sh_struct->dir_checksum,
-						dir_checksum, NAME_MAX);
-				}
-
-				/* Reply from the storage nodes */
-				for (index = 0; 
-				     index < NAME_MAX; index++) {
-					/* Files should be present in only 
-					   one node */
-					local->sh_struct->file_checksum[index] ^= file_checksum[index];
-	  
-					/* directory structure should be same 
-					   accross */
-					if (local->sh_struct->dir_checksum[index] != dir_checksum[index])
-						local->failed = 1;
-				}
-			}
-		} 
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		for (index = 0; index < NAME_MAX ; index++) {
-			if (local->sh_struct->file_checksum[index] != 
-			    local->sh_struct->ns_file_checksum[index]) {
-				local->failed = 1;
-				break;
-			}
-			if (local->sh_struct->dir_checksum[index] != 
-			    local->sh_struct->ns_dir_checksum[index]) {
-				local->failed = 1;
-				break;
-			}
-		}
-	
-		if (local->failed) {
-			/* Log it, it should be a rare event */
-			gf_log (this->name, GF_LOG_WARNING, 
-				"Self-heal triggered on directory %s", 
-				local->loc1.path);
-
-			/* Any self heal will be done at the directory level */
-			local->op_ret = -1;
-			local->failed = 0;
-      
-			local->fd = fd_create (local->loc1.inode, 
-					       frame->root->pid);
-			local->call_count = priv->child_count + 1;
-	
-			for (index = 0; 
-			     index < (priv->child_count + 1); index++) {
-				STACK_WIND_COOKIE (frame,
-						   unify_bgsh_opendir_cbk,
-						   priv->xl_array[index]->name,
-						   priv->xl_array[index],
-						   priv->xl_array[index]->fops->opendir,
-						   &local->loc1,
-						   local->fd);
-			}
-      
-			/* opendir can be done on the directory */
-			return 0;
-		}
-
-		/* no mismatch */
-		unify_local_wipe (local);
-		STACK_DESTROY (frame->root);
-	}
-
-	return 0;
-}
-
-/* Background self-heal part over */
-
-
-
-
-/**
- * zr_unify_self_heal - 
- * 
- * @frame: frame used in lookup. get a copy of it, and use that copy.
- * @this: pointer to unify xlator.
- * @inode: pointer to inode, for which the consistency check is required.
- *
- */
-int32_t 
-zr_unify_self_heal (call_frame_t *frame,
-		    xlator_t *this,
-		    unify_local_t *local)
-{
-	unify_private_t *priv = this->private;
-	call_frame_t *bg_frame = NULL;
-	unify_local_t *bg_local = NULL;
-	inode_t *tmp_inode = NULL;
-	dict_t *tmp_dict = NULL;
-	int16_t index = 0;
-  
-	if (local->inode_generation < priv->inode_generation) {
-		/* Any self heal will be done at the directory level */
-		/* Update the inode's generation to the current generation
-		   value. */
-		local->inode_generation = priv->inode_generation;
-		inode_ctx_put (local->loc1.inode, this, 
-			  (uint64_t)(long)local->inode_generation);
-
-		if (priv->self_heal == ZR_UNIFY_FG_SELF_HEAL) {
-			local->op_ret = 0;
-			local->failed = 0;
-			local->call_count = priv->child_count + 1;
-			local->sh_struct = 
-	         	     GF_CALLOC (1, sizeof (struct unify_self_heal_struct),
-                                        gf_unify_mt_unify_self_heal_struct);
-      
-			/* +1 is for NS */
-			for (index = 0; 
-			     index < (priv->child_count + 1); index++) {
-				STACK_WIND_COOKIE (frame,
-						   unify_sh_checksum_cbk,
-						   priv->xl_array[index],
-						   priv->xl_array[index],
-						   priv->xl_array[index]->fops->checksum,
-						   &local->loc1,
-						   0);
-			}
-
-			/* Self-heal in foreground, hence no need 
-			   to UNWIND here */
-			return 0;
-		}
-
-		/* Self Heal done in background */
-		bg_frame = copy_frame (frame);
-		INIT_LOCAL (bg_frame, bg_local);
-		loc_copy (&bg_local->loc1, &local->loc1);
-		bg_local->op_ret = 0;
-		bg_local->failed = 0;
-		bg_local->call_count = priv->child_count + 1;
-		bg_local->sh_struct = 
-			GF_CALLOC (1, sizeof (struct unify_self_heal_struct),
-                                   gf_unify_mt_unify_self_heal_struct);
-    
-		/* +1 is for NS */
-		for (index = 0; index < (priv->child_count + 1); index++) {
-			STACK_WIND_COOKIE (bg_frame,
-					   unify_bgsh_checksum_cbk,
-					   priv->xl_array[index],
-					   priv->xl_array[index],
-					   priv->xl_array[index]->fops->checksum,
-					   &bg_local->loc1,
-					   0);
-		}
-	}
-
-	/* generation number matches, self heal already done or
-	 * self heal done in background: just do STACK_UNWIND 
-	 */
-	tmp_inode = local->loc1.inode;
-	tmp_dict = local->dict;
-
-	unify_local_wipe (local);
-
-	/* This is lookup_cbk ()'s UNWIND. */
-	STACK_UNWIND (frame,
-		      local->op_ret,
-		      local->op_errno,
-		      tmp_inode,
-		      &local->stbuf,
-		      local->dict,
-                      &local->oldpostparent);
-
-	if (tmp_dict)
-		dict_unref (tmp_dict);
-
-	return 0;
-}
-
diff --git a/xlators/cluster/unify/src/unify.c b/xlators/cluster/unify/src/unify.c
deleted file mode 100644
index cec7f38c6f7..00000000000
--- a/xlators/cluster/unify/src/unify.c
+++ /dev/null
@@ -1,4589 +0,0 @@
-/*
-  Copyright (c) 2006-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-/**
- * xlators/cluster/unify:
- *     - This xlator is one of the main translator in GlusterFS, which
- *   actually does the clustering work of the file system. One need to 
- *   understand that, unify assumes file to be existing in only one of 
- *   the child node, and directories to be present on all the nodes. 
- *
- * NOTE:
- *   Now, unify has support for global namespace, which is used to keep a 
- * global view of fs's namespace tree. The stat for directories are taken
- * just from the namespace, where as for files, just 'ia_ino' is taken from
- * Namespace node, and other stat info is taken from the actual storage node.
- * Also Namespace node helps to keep consistant inode for files across 
- * glusterfs (re-)mounts.
- */
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "unify.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include <signal.h>
-#include <libgen.h>
-#include "compat-errno.h"
-#include "compat.h"
-
-#define UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR(_loc) do { \
-  if (!(_loc && _loc->inode)) {                            \
-    STACK_UNWIND (frame, -1, EINVAL, NULL, NULL, NULL);    \
-    return 0;                                              \
-  }                                                        \
-} while(0)
-
-
-#define UNIFY_CHECK_FD_CTX_AND_UNWIND_ON_ERR(_fd) do { \
-  if (!(_fd && !fd_ctx_get (_fd, this, NULL))) {       \
-    STACK_UNWIND (frame, -1, EBADFD, NULL, NULL);      \
-    return 0;                                          \
-  }                                                    \
-} while(0)
-
-#define UNIFY_CHECK_FD_AND_UNWIND_ON_ERR(_fd) do { \
-  if (!_fd) {                                      \
-    STACK_UNWIND (frame, -1, EBADFD, NULL, NULL);  \
-    return 0;                                      \
-  }                                                \
-} while(0)
-
-/**
- * unify_local_wipe - free all the extra allocation of local->* here.
- */
-static void 
-unify_local_wipe (unify_local_t *local)
-{
-	/* Free the strdup'd variables in the local structure */
-	if (local->name) {
-		GF_FREE (local->name);
-	}
-	loc_wipe (&local->loc1);
-	loc_wipe (&local->loc2);
-}
-
-
-
-/*
- * unify_normalize_stats -
- */
-void
-unify_normalize_stats (struct statvfs *buf,
-		       unsigned long bsize,
-		       unsigned long frsize)
-{
-	double factor;
-
-	if (buf->f_bsize != bsize) {
-		factor = ((double) buf->f_bsize) / bsize;
-		buf->f_bsize  = bsize;
-		buf->f_bfree  = (fsblkcnt_t) (factor * buf->f_bfree);
-		buf->f_bavail = (fsblkcnt_t) (factor * buf->f_bavail);
-	}
-  
-	if (buf->f_frsize != frsize) {
-		factor = ((double) buf->f_frsize) / frsize;
-		buf->f_frsize = frsize;
-		buf->f_blocks = (fsblkcnt_t) (factor * buf->f_blocks);
-	}
-}
-
-
-xlator_t *
-unify_loc_subvol (loc_t *loc, xlator_t *this)
-{
-	unify_private_t *priv = NULL;
-	xlator_t        *subvol = NULL;
-	int16_t         *list = NULL;
-	long             index = 0;
-	xlator_t        *subvol_i = NULL;
-	int              ret = 0;
-	uint64_t         tmp_list = 0;
-
-	priv   = this->private;
-	subvol = NS (this);
-
-	if (!IA_ISDIR (loc->inode->ia_type)) {
-		ret = inode_ctx_get (loc->inode, this, &tmp_list);
-		list = (int16_t *)(long)tmp_list;
-		if (!list)
-			goto out;
-
-		for (index = 0; list[index] != -1; index++) {
-			subvol_i = priv->xl_array[list[index]];
-			if (subvol_i != NS (this)) {
-				subvol = subvol_i;
-				break;
-			}
-		}
-	}
-out:
-	return subvol;
-}
-
-
-
-/**
- * unify_statfs_cbk -
- */
-int32_t
-unify_statfs_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  struct statvfs *stbuf)
-{
-	int32_t callcnt = 0;
-	struct statvfs *dict_buf = NULL;
-	unsigned long bsize;
-	unsigned long frsize;
-	unify_local_t *local = (unify_local_t *)frame->local;
-	call_frame_t *prev_frame = cookie;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret >= 0) {
-			/* when a call is successfull, add it to local->dict */
-			dict_buf = &local->statvfs_buf;
-
-			if (dict_buf->f_bsize != 0) {
-				bsize  = max (dict_buf->f_bsize, 
-					      stbuf->f_bsize);
-
-				frsize = max (dict_buf->f_frsize, 
-					      stbuf->f_frsize);
-				unify_normalize_stats(dict_buf, bsize, frsize);
-				unify_normalize_stats(stbuf, bsize, frsize);
-			} else {
-				dict_buf->f_bsize   = stbuf->f_bsize;
-				dict_buf->f_frsize  = stbuf->f_frsize;
-			}
-      
-			dict_buf->f_blocks += stbuf->f_blocks;
-			dict_buf->f_bfree  += stbuf->f_bfree;
-			dict_buf->f_bavail += stbuf->f_bavail;
-			dict_buf->f_files  += stbuf->f_files;
-			dict_buf->f_ffree  += stbuf->f_ffree;
-			dict_buf->f_favail += stbuf->f_favail;
-			dict_buf->f_fsid    = stbuf->f_fsid;
-			dict_buf->f_flag    = stbuf->f_flag;
-			dict_buf->f_namemax = stbuf->f_namemax;
-			local->op_ret = op_ret;
-		} else {
-			/* fop on storage node has failed due to some error */
-			if (op_errno != ENOTCONN) {
-				gf_log (this->name, GF_LOG_ERROR, 
-					"child(%s): %s", 
-					prev_frame->this->name, 
-					strerror (op_errno));
-			}
-			local->op_errno = op_errno;
-		}
-		callcnt = --local->call_count;
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-			      &local->statvfs_buf);
-	}
-
-	return 0;
-}
-
-/**
- * unify_statfs -
- */
-int32_t
-unify_statfs (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc)
-{
-	unify_local_t *local = NULL;
-	xlator_list_t *trav = this->children;
-
-	INIT_LOCAL (frame, local);
-	local->call_count = ((unify_private_t *)this->private)->child_count;
-
-	while(trav) {
-		STACK_WIND (frame,
-			    unify_statfs_cbk,
-			    trav->xlator,
-			    trav->xlator->fops->statfs,
-			    loc);
-		trav = trav->next;
-	}
-
-	return 0;
-}
-
-/**
- * unify_buf_cbk - 
- */
-int32_t
-unify_buf_cbk (call_frame_t *frame,
-	       void *cookie,
-	       xlator_t *this,
-	       int32_t op_ret,
-	       int32_t op_errno,
-	       struct iatt *buf)
-{
-	int32_t callcnt = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-	call_frame_t *prev_frame = cookie;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-    
-		if (op_ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"%s(): child(%s): path(%s): %s", 
-				gf_fop_list[frame->root->op],
-				prev_frame->this->name, 
-				(local->loc1.path)?local->loc1.path:"", 
-				strerror (op_errno));
-
-			local->op_errno = op_errno;
-			if ((op_errno == ENOENT) && priv->optimist) 
-				local->op_ret = 0;
-		}
-
-		if (op_ret >= 0) {
-			local->op_ret = 0;
-
-			if (NS (this) == prev_frame->this) {
-				local->ia_ino = buf->ia_ino;
-				/* If the entry is directory, get the stat
-				   from NS node */
-				if (IA_ISDIR (buf->ia_type) || 
-				    !local->stbuf.ia_blksize) {
-					local->stbuf = *buf;
-				}
-			}
-
-			if ((!IA_ISDIR (buf->ia_type)) && 
-			    (NS (this) != prev_frame->this)) {
-				/* If file, take the stat info from Storage 
-				   node. */
-				local->stbuf = *buf;
-			}
-		}
-	}
-	UNLOCK (&frame->lock);
-    
-	if (!callcnt) {
-		/* If the inode number is not filled, operation should
-		   fail */
-		if (!local->ia_ino)
-			local->op_ret = -1;
-
-		local->stbuf.ia_ino = local->ia_ino;
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-			      &local->stbuf);
-	}
-
-	return 0;
-}
-
-#define check_if_dht_linkfile(s) \
-        ((st_mode_from_ia (s->ia_prot, s->ia_type) & ~S_IFMT) == S_ISVTX)
-
-/**
- * unify_lookup_cbk - 
- */
-int32_t 
-unify_lookup_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  inode_t *inode,
-		  struct iatt *buf,
-		  dict_t *dict,
-                  struct iatt *postparent)
-{
-	int32_t callcnt = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-	inode_t *tmp_inode = NULL;
-	dict_t *local_dict = NULL;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
- 
-		if (op_ret == -1) {
-                        if (local->revalidate &&
-                            (op_errno == ESTALE)) {
-                                /* ESTALE takes priority */
-                                local->op_errno = op_errno;
-                                local->failed = 1;
-                        }
-
-			if ((op_errno != ENOTCONN) 
-                            && (op_errno != ENOENT)
-                            && (local->op_errno != ESTALE)) {
-                                /* if local->op_errno is already ESTALE, then
-                                 * ESTALE has to propogated to the parent first.
-                                 * do not enter here.
-                                 */
-				gf_log (this->name, GF_LOG_ERROR,
-					"child(%s): path(%s): %s", 
-					priv->xl_array[(long)cookie]->name, 
-					local->loc1.path, strerror (op_errno));
-                                local->op_errno = op_errno;
-				local->failed = 1;
-
-			} else if (local->revalidate && 
-                                   (local->op_errno != ESTALE) &&
-				   !(priv->optimist && (op_errno == ENOENT))) {
-
-				gf_log (this->name, 
-					(op_errno == ENOTCONN) ? 
-					GF_LOG_DEBUG:GF_LOG_ERROR,
-					"child(%s): path(%s): %s", 
-					priv->xl_array[(long)cookie]->name, 
-					local->loc1.path, strerror (op_errno));
-				local->op_errno = op_errno;
-				local->failed = 1;
-			}
-		}
-
-		if (op_ret == 0) {
-			local->op_ret = 0; 
-			
-			if (check_if_dht_linkfile(buf)) {
-				gf_log (this->name, GF_LOG_CRITICAL,
-					"file %s may be DHT link file on %s, "
-					"make sure the backend is not shared "
-					"between unify and DHT", 
-					local->loc1.path, 
-					priv->xl_array[(long)cookie]->name);
-			}
-
-			if (local->stbuf.ia_type && local->stbuf.ia_blksize) {
-				/* make sure we already have a stbuf
-				   stored in local->stbuf */
-				if (IA_ISDIR (local->stbuf.ia_type) && 
-				    !IA_ISDIR (buf->ia_type)) {
-					gf_log (this->name, GF_LOG_CRITICAL, 
-						"[CRITICAL] '%s' is directory "
-						"on namespace, non-directory "
-						"on node '%s', returning EIO",
-						local->loc1.path, 
-						priv->xl_array[(long)cookie]->name);
-					local->return_eio = 1;
-				}
-				if (!IA_ISDIR (local->stbuf.ia_type) && 
-				    IA_ISDIR (buf->ia_type)) {
-					gf_log (this->name, GF_LOG_CRITICAL, 
-						"[CRITICAL] '%s' is directory "
-						"on node '%s', non-directory "
-						"on namespace, returning EIO",
-						local->loc1.path, 
-						priv->xl_array[(long)cookie]->name);
-					local->return_eio = 1;
-				}
-			}
-	
-			if (!local->revalidate && !IA_ISDIR (buf->ia_type)) {
-				/* This is the first time lookup on file*/
-				if (!local->list) {
-					/* list is not allocated, allocate 
-					   the max possible range */
-					local->list = GF_CALLOC (1, 2 * (priv->child_count + 2),
-                                                                 gf_unify_mt_int16_t);
-					if (!local->list) {
-						gf_log (this->name, 
-							GF_LOG_CRITICAL, 
-							"Not enough memory");
-						STACK_UNWIND (frame, -1, 
-							      ENOMEM, inode, 
-							      NULL, NULL, NULL);
-						return 0;
-					}
-				}
-				/* update the index of the list */
-				local->list [local->index++] = 
-					(int16_t)(long)cookie;
-			}
-                        
-                        if (!local->revalidate && IA_ISDIR (buf->ia_type)) {
-                                /* fresh lookup of a directory */
-                                inode_ctx_put (local->loc1.inode, this, 
-                                               priv->inode_generation);
-                        }
-
-			if ((!local->dict) && dict &&
-			    (priv->xl_array[(long)cookie] != NS(this)))	{
-				local->dict = dict_ref (dict);
-			}
-
-			/* index of NS node is == total child count */
-			if (priv->child_count == (int16_t)(long)cookie) {
-				/* Take the inode number from namespace */
-				local->ia_ino = buf->ia_ino;
-				if (IA_ISDIR (buf->ia_type) || 
-				    !(local->stbuf.ia_blksize)) {
-					local->stbuf = *buf;
-                                        local->oldpostparent = *postparent;
-				}
-			} else if (!IA_ISDIR (buf->ia_type)) {
-				/* If file, then get the stat from 
-				   storage node */
-				local->stbuf = *buf;
-			}
-
-			if (local->ia_nlink < buf->ia_nlink) {
-				local->ia_nlink = buf->ia_nlink;
-			}
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		local_dict = local->dict;
-		if (local->return_eio) {
-			gf_log (this->name, GF_LOG_CRITICAL, 
-				"[CRITICAL] Unable to fix the path (%s) with "
-				"self-heal, try manual verification. "
-				"returning EIO.", local->loc1.path);
-			unify_local_wipe (local);
-			STACK_UNWIND (frame, -1, EIO, inode, NULL, NULL);
-			if (local_dict)	{
-				dict_unref (local_dict);
-			}
-			return 0;
-		}
-
-		if (!local->stbuf.ia_blksize) {
-			/* Inode not present */
-			local->op_ret = -1;
-		} else {
-			if (!local->revalidate && 
-			    !IA_ISDIR (local->stbuf.ia_type)) { 
-				/* If its a file, big array is useless, 
-				   allocate the smaller one */
-				int16_t *list = NULL;
-				list = GF_CALLOC (1, 2 * (local->index + 1),
-                                                  gf_unify_mt_int16_t);
-				ERR_ABORT (list);
-				memcpy (list, local->list, 2 * local->index);
-				/* Make the end of the list as -1 */
-				GF_FREE (local->list);
-				local->list = list;
-				local->list [local->index] = -1;
-				/* Update the inode's ctx with proper array */
-				/* TODO: log on failure */
-				inode_ctx_put (local->loc1.inode, this, 
-					       (uint64_t)(long)local->list);
-			}
-
-			if (IA_ISDIR(local->loc1.inode->ia_type)) {
-				/* lookup is done for directory */
-				if (local->failed && priv->self_heal) {
-					/* Triggering self-heal */
-                                        /* means, self-heal required for this 
-					   inode */
-					local->inode_generation = 0; 
-					priv->inode_generation++;
-				}
-			} else {
-				local->stbuf.ia_ino = local->ia_ino;
-			}
-	  
-			local->stbuf.ia_nlink = local->ia_nlink;
-		}
-		if (local->op_ret == -1) {
-			if (!local->revalidate && local->list)
-				GF_FREE (local->list);
-		}
-
-		if ((local->op_ret >= 0) && local->failed && 
-		    local->revalidate) {
-			/* Done revalidate, but it failed */
-			if ((op_errno != ENOTCONN) 
-                            && (local->op_errno != ESTALE)) {
-				gf_log (this->name, GF_LOG_ERROR, 
-					"Revalidate failed for path(%s): %s", 
-					local->loc1.path, strerror (op_errno));
-			}
-			local->op_ret = -1;
-		}
-
-		if ((priv->self_heal && !priv->optimist) && 
-		    (!local->revalidate && (local->op_ret == 0) && 
-		     IA_ISDIR(local->stbuf.ia_type))) {
-			/* Let the self heal be done here */
-			zr_unify_self_heal (frame, this, local);
-			local_dict = NULL;
-		} else {
-                        if (local->failed) {
-                                /* NOTE: directory lookup is sent to all 
-                                 * subvolumes and success from a subvolume
-                                 * might set local->op_ret to 0 (zero) */
-                                local->op_ret = -1;
-                        }
-
-			/* either no self heal, or op_ret == -1 (failure) */
-			tmp_inode = local->loc1.inode;
-			unify_local_wipe (local);
-			STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-				      tmp_inode, &local->stbuf, local->dict,
-                                      &local->oldpostparent);
-		}
-		if (local_dict) {
-			dict_unref (local_dict);
-		}
-	}
-  
-	return 0;
-}
-
-/**
- * unify_lookup - 
- */
-int32_t 
-unify_lookup (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      dict_t *xattr_req)
-{
-	unify_local_t *local = NULL;
-	unify_private_t *priv = this->private;
-	int16_t *list = NULL;
-	long index = 0;
-
-	if (!(loc && loc->inode)) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"%s: Argument not right", loc?loc->path:"(null)");
-		STACK_UNWIND (frame, -1, EINVAL, NULL, NULL, NULL, NULL);
-		return 0;
-	}
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	loc_copy (&local->loc1, loc);
-	if (local->loc1.path == NULL) {
-		gf_log (this->name, GF_LOG_CRITICAL, "Not enough memory :O");
-		STACK_UNWIND (frame, -1, ENOMEM, loc->inode, NULL, NULL, NULL);
-		return 0;
-	}
-        
-        if (inode_ctx_get (loc->inode, this, NULL)
-            && IA_ISDIR (loc->inode->ia_type)) {
-                local->revalidate = 1;
-        }
-
-	if (!inode_ctx_get (loc->inode, this, NULL) && 
-	    loc->inode->ia_type && 
-	    !IA_ISDIR (loc->inode->ia_type)) {
-		uint64_t tmp_list = 0;
-		/* check if revalidate or fresh lookup */
-		inode_ctx_get (loc->inode, this, &tmp_list);
-		local->list = (int16_t *)(long)tmp_list;
-	}
-
-	if (local->list) {
-		list = local->list;
-		for (index = 0; list[index] != -1; index++);
-		if (index != 2) {
-			if (index < 2) {
-				gf_log (this->name, GF_LOG_ERROR,
-					"returning ESTALE for %s: file "
-					"count is %ld", loc->path, index);
-				/* Print where all the file is present */
-				for (index = 0; 
-				     local->list[index] != -1; index++) {
-					gf_log (this->name, GF_LOG_ERROR, 
-						"%s: found on %s", loc->path, 
-						priv->xl_array[list[index]]->name);
-				}
-				unify_local_wipe (local);
-				STACK_UNWIND (frame, -1, ESTALE, 
-					      NULL, NULL, NULL, NULL);
-				return 0;  
-			} else {
-				/* There are more than 2 presences */
-				/* Just log and continue */
-				gf_log (this->name, GF_LOG_ERROR,
-					"%s: file count is %ld", 
-					loc->path, index);
-				/* Print where all the file is present */
-				for (index = 0; 
-				     local->list[index] != -1; index++) {
-					gf_log (this->name, GF_LOG_ERROR, 
-						"%s: found on %s", loc->path, 
-						priv->xl_array[list[index]]->name);
-				}
-			}
-		}
-      
-		/* is revalidate */
-		local->revalidate = 1;
-      
-		for (index = 0; list[index] != -1; index++)
-			local->call_count++;
-      
-		for (index = 0; list[index] != -1; index++) {
-			char need_break = (list[index+1] == -1);
-			STACK_WIND_COOKIE (frame,
-					   unify_lookup_cbk,
-					   (void *)(long)list[index], //cookie
-					   priv->xl_array [list[index]],
-					   priv->xl_array [list[index]]->fops->lookup,
-					   loc,
-					   xattr_req);
-			if (need_break)
-				break;
-		}
-	} else {
-		if (loc->inode->ia_type) {
-			if (inode_ctx_get (loc->inode, this, NULL)) {
-				inode_ctx_get (loc->inode, this, 
-					       &local->inode_generation);
-			}
-		}
-		/* This is first call, there is no list */
-		/* call count should be all child + 1 namespace */
-		local->call_count = priv->child_count + 1;
-      
-		for (index = 0; index <= priv->child_count; index++) {
-			STACK_WIND_COOKIE (frame,
-					   unify_lookup_cbk,
-					   (void *)index, //cookie
-					   priv->xl_array[index],
-					   priv->xl_array[index]->fops->lookup,
-					   loc,
-					   xattr_req);
-		}
-	}
-
-	return 0;
-}
-
-/**
- * unify_stat - if directory, get the stat directly from NameSpace child.
- *     if file, check for a hint and send it only there (also to NS).
- *     if its a fresh stat, then do it on all the nodes.
- *
- * NOTE: for all the call, sending cookie as xlator pointer, which will be 
- *       used in cbk.
- */
-int32_t
-unify_stat (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc)
-{
-	unify_local_t *local = NULL;
-	unify_private_t *priv = this->private;
-	int16_t index = 0;
-	int16_t *list = NULL;
-	uint64_t tmp_list = 0;
-
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	loc_copy (&local->loc1, loc);
-	if (local->loc1.path == NULL) {
-		gf_log (this->name, GF_LOG_CRITICAL, "Not enough memory :O");
-		STACK_UNWIND (frame, -1, ENOMEM, NULL);
-		return 0;
-	}
-	local->ia_ino = loc->inode->ino;
-	if (IA_ISDIR (loc->inode->ia_type)) {
-		/* Directory */
-		local->call_count = 1;
-		STACK_WIND (frame, unify_buf_cbk, NS(this),
-			    NS(this)->fops->stat, loc);
-	} else {
-		/* File */
-		inode_ctx_get (loc->inode, this, &tmp_list);
-    		list = (int16_t *)(long)tmp_list;
-
-		for (index = 0; list[index] != -1; index++)
-			local->call_count++;
-    
-		for (index = 0; list[index] != -1; index++) {
-			char need_break = (list[index+1] == -1);
-			STACK_WIND (frame,
-				    unify_buf_cbk,
-				    priv->xl_array[list[index]],
-				    priv->xl_array[list[index]]->fops->stat,
-				    loc);
-			if (need_break)
-				break;
-		}
-	}
-
-	return 0;
-}
-
-/**
- * unify_access_cbk -
- */
-int32_t
-unify_access_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-
-/**
- * unify_access - Send request to only namespace, which has all the 
- *      attributes set for the file.
- */
-int32_t
-unify_access (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      int32_t mask)
-{
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-
-	STACK_WIND (frame,
-		    unify_access_cbk,
-		    NS(this),
-		    NS(this)->fops->access,
-		    loc,
-		    mask);
-
-	return 0;
-}
-
-int32_t
-unify_mkdir_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 inode_t *inode,
-                 struct iatt *buf,
-                 struct iatt *preparent,
-                 struct iatt *postparent)
-{
-	int32_t callcnt = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-	inode_t *tmp_inode = NULL;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-  
-		if ((op_ret == -1) && !(priv->optimist && 
-					(op_errno == ENOENT || 
-					 op_errno == EEXIST))) {
-			/* TODO: Decrement the inode_generation of 
-			 * this->inode's parent inode, hence the missing 
-			 * directory is created properly by self-heal. 
-			 * Currently, there is no way to get the parent 
-			 * inode directly.
-			 */
-			gf_log (this->name, GF_LOG_ERROR,
-				"child(%s): path(%s): %s", 
-				priv->xl_array[(long)cookie]->name, 
-				local->loc1.path, strerror (op_errno));
-			if (op_errno != EEXIST)
-				local->failed = 1;
-			local->op_errno = op_errno;
-		}
-  
-		if (op_ret >= 0)
-			local->op_ret = 0;
-
-	}
-	UNLOCK (&frame->lock);
-  
-	if (!callcnt) {
-		if (!local->failed) {
-			inode_ctx_put (local->loc1.inode, this, 
-				       priv->inode_generation);
-		}
-		
-		tmp_inode = local->loc1.inode;
-		unify_local_wipe (local);
-
-		STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-			      tmp_inode, &local->stbuf,
-                              &local->oldpreparent, &local->oldpostparent);
-	}
-
-	return 0;
-}
-
-/**
- * unify_ns_mkdir_cbk -
- */
-int32_t
-unify_ns_mkdir_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    inode_t *inode,
-                    struct iatt *buf,
-                    struct iatt *preparent,
-                    struct iatt *postparent)
-{
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-	long index = 0;
-
-	if (op_ret == -1) {
-		/* No need to send mkdir request to other servers, 
-		 * as namespace action failed 
-		 */
-		gf_log (this->name, GF_LOG_ERROR,
-			"namespace: path(%s): %s", 
-			local->name, strerror (op_errno));
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, op_ret, op_errno, inode, NULL,
-                              NULL, NULL);
-		return 0;
-	}
-  
-	/* Create one inode for this entry */
-	local->op_ret = 0;
-	local->stbuf = *buf;
-
-        local->oldpreparent = *preparent;
-        local->oldpostparent = *postparent;
-
-	local->call_count = priv->child_count;
-
-	/* Send mkdir request to all the nodes now */
-	for (index = 0; index < priv->child_count; index++) {
-		STACK_WIND_COOKIE (frame,
-				   unify_mkdir_cbk,
-				   (void *)index, //cookie
-				   priv->xl_array[index],
-				   priv->xl_array[index]->fops->mkdir,
-				   &local->loc1,
-				   local->mode);
-	}
-  
-	return 0;
-}
-
-
-/**
- * unify_mkdir -
- */
-int32_t
-unify_mkdir (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     mode_t mode)
-{
-	unify_local_t *local = NULL;
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	local->mode = mode;
-
-	loc_copy (&local->loc1, loc);
-
-	if (local->loc1.path == NULL) {
-		gf_log (this->name, GF_LOG_CRITICAL, "Not enough memory :O");
-		STACK_UNWIND (frame, -1, ENOMEM, NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame,
-		    unify_ns_mkdir_cbk,
-		    NS(this),
-		    NS(this)->fops->mkdir,
-		    loc,
-		    mode);
-	return 0;
-}
-
-/**
- * unify_rmdir_cbk -
- */
-int32_t
-unify_rmdir_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-                 struct iatt *preparent,
-                 struct iatt *postparent)
-{
-	int32_t callcnt = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-		if (op_ret == 0 || (priv->optimist && (op_errno == ENOENT)))
-			local->op_ret = 0;
-		if (op_ret == -1)
-			local->op_errno = op_errno;
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, local->op_ret, local->op_errno,
-                              &local->oldpreparent, &local->oldpostparent);
-	}
-
-	return 0;
-}
-
-/**
- * unify_ns_rmdir_cbk -
- */
-int32_t
-unify_ns_rmdir_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-                    struct iatt *preparent,
-                    struct iatt *postparent)
-{
-	int16_t index = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-  
-	if (op_ret == -1) {
-		/* No need to send rmdir request to other servers, 
-		 * as namespace action failed 
-		 */
-		gf_log (this->name, 
-			((op_errno != ENOTEMPTY) ? 
-			 GF_LOG_ERROR : GF_LOG_DEBUG),
-			"namespace: path(%s): %s", 
-			local->loc1.path, strerror (op_errno));
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, op_ret, op_errno, NULL, NULL);
-		return 0;
-	}
-
-	local->call_count = priv->child_count;
-
-        local->oldpreparent = *preparent;
-        local->oldpostparent = *postparent;
-
-	for (index = 0; index < priv->child_count; index++) {
-		STACK_WIND (frame,
-			    unify_rmdir_cbk,
-			    priv->xl_array[index],
-			    priv->xl_array[index]->fops->rmdir,
-			    &local->loc1);
-	}
-
-	return 0;
-}
-
-/**
- * unify_rmdir -
- */
-int32_t
-unify_rmdir (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc)
-{
-	unify_local_t *local = NULL;
-
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-
-	loc_copy (&local->loc1, loc);
-	if (local->loc1.path == NULL) {
-		gf_log (this->name, GF_LOG_CRITICAL, "Not enough memory :O");
-		STACK_UNWIND (frame, -1, ENOMEM, NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame,
-		    unify_ns_rmdir_cbk,
-		    NS(this),
-		    NS(this)->fops->rmdir,
-		    loc);
-
-	return 0;
-}
-
-/**
- * unify_open_cbk -
- */
-int32_t
-unify_open_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		fd_t *fd)
-{
-	int32_t callcnt = 0;
-	unify_local_t *local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret >= 0) {
-			local->op_ret = op_ret;
-			if (NS(this) != (xlator_t *)cookie) {
-				/* Store child node's ptr, used in 
-				   all the f*** / FileIO calls */
-				fd_ctx_set (fd, this, (uint64_t)(long)cookie);
-			}
-		}
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			local->failed = 1;
-		}
-		callcnt = --local->call_count;
-	}
-	UNLOCK (&frame->lock);
-  
-	if (!callcnt) {
-		if ((local->failed == 1) && (local->op_ret >= 0)) {
-			local->call_count = 1;
-			/* return -1 to user */
-			local->op_ret = -1;
-			//local->op_errno = EIO; 
-      
-			if (!fd_ctx_get (local->fd, this, NULL)) {
-				gf_log (this->name, GF_LOG_ERROR, 
-					"Open success on child node, "
-					"failed on namespace");
-			} else {
-				gf_log (this->name, GF_LOG_ERROR, 
-					"Open success on namespace, "
-					"failed on child node");
-			}
-		}
-
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, local->op_ret, 
-			      local->op_errno, local->fd);
-	}
-
-	return 0;
-}
-
-#ifdef GF_DARWIN_HOST_OS
-/**
- * unify_create_lookup_cbk - 
- */
-int32_t 
-unify_open_lookup_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno,
-		       inode_t *inode,
-		       struct iatt *buf,
-		       dict_t *dict,
-		       struct iatt *postparent)
-{
-	int32_t callcnt = 0;
-	int16_t index = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-		if ((op_ret == -1) && (op_errno != ENOENT)) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"child(%s): path(%s): %s", 
-				priv->xl_array[(long)cookie]->name, 
-				local->loc1.path, strerror (op_errno));
-			local->op_errno = op_errno;
-		}
-    
-		if (op_ret >= 0) {
-			local->op_ret = op_ret; 
-			local->index++;
-			if (NS(this) == priv->xl_array[(long)cookie]) {
-				local->list[0] = (int16_t)(long)cookie;
-			} else {
-				local->list[1] = (int16_t)(long)cookie;
-			}
-			if (IA_ISDIR (buf->ia_type))
-				local->failed = 1;
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		int16_t file_list[3] = {0,};
-		local->op_ret = -1;
-
-		file_list[0] = local->list[0];
-		file_list[1] = local->list[1];
-		file_list[2] = -1;
-
-		if (local->index != 2) {
-			/* Lookup failed, can't do open */
-			gf_log (this->name, GF_LOG_ERROR,
-				"%s: present on %d nodes", 
-				local->name, local->index);
-
-			if (local->index < 2) {
-				unify_local_wipe (local);
-				gf_log (this->name, GF_LOG_ERROR,
-					"returning as file found on less "
-					"than 2 nodes");
-				STACK_UNWIND (frame, local->op_ret, 
-					      local->op_errno, local->fd);
-				return 0;
-			}
-		}
-
-		if (local->failed) {
-			/* Open on directory, return EISDIR */
-			unify_local_wipe (local);
-			STACK_UNWIND (frame, -1, EISDIR, local->fd);
-			return 0;
-		}
-
-		/* Everything is perfect :) */    
-		local->call_count = 2;
-    
-		for (index = 0; file_list[index] != -1; index++) {
-			char need_break = (file_list[index+1] == -1);
-			STACK_WIND_COOKIE (frame,
-					   unify_open_cbk,
-					   priv->xl_array[file_list[index]],
-					   priv->xl_array[file_list[index]],
-					   priv->xl_array[file_list[index]]->fops->open,
-					   &local->loc1,
-					   local->flags,
-					   local->fd, local->wbflags);
-			if (need_break)
-				break;
-		}
-	}
-
-	return 0;
-}
-
-
-int32_t
-unify_open_readlink_cbk (call_frame_t *frame,
-			 void *cookie,
-			 xlator_t *this,
-			 int32_t op_ret,
-			 int32_t op_errno,
-			 const char *path,
-			 struct iatt *sbuf)
-{
-	int16_t index = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-
-	if (op_ret == -1) {
-		STACK_UNWIND (frame, -1, ENOENT);
-		return 0;
-	}
-
-	if (path[0] == '/') {
-		local->name = gf_strdup (path);
-		ERR_ABORT (local->name);
-	} else {
-		char *tmp_str = gf_strdup (local->loc1.path);
-		char *tmp_base = dirname (tmp_str);
-		local->name = GF_CALLOC (1, ZR_PATH_MAX, gf_unify_mt_char);
-		strcpy (local->name, tmp_base);
-		strncat (local->name, "/", 1);
-		strcat (local->name, path);
-		GF_FREE (tmp_str);
-	}
-  
-	local->list = GF_CALLOC (1, sizeof (int16_t) * 3,
-                                 gf_unify_mt_int16_t);
-	ERR_ABORT (local->list);
-	local->call_count = priv->child_count + 1;
-	local->op_ret = -1;
-	for (index = 0; index <= priv->child_count; index++) {
-		/* Send the lookup to all the nodes including namespace */
-		STACK_WIND_COOKIE (frame,
-				   unify_open_lookup_cbk,
-				   (void *)(long)index,
-				   priv->xl_array[index],
-				   priv->xl_array[index]->fops->lookup,
-				   &local->loc1,
-				   NULL);
-	}
-
-	return 0;
-}
-#endif /* GF_DARWIN_HOST_OS */
-
-/**
- * unify_open - 
- */
-int32_t
-unify_open (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc,
-	    int32_t flags,
-	    fd_t *fd,
-            int32_t wbflags)
-{
-	unify_private_t *priv = this->private;
-	unify_local_t *local = NULL;
-	int16_t *list = NULL;
-	int16_t index = 0;
-	int16_t file_list[3] = {0,};
-	uint64_t tmp_list = 0;
-
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-
-	/* Init */
-	INIT_LOCAL (frame, local);
-	loc_copy (&local->loc1, loc);
-	local->fd    = fd;
-	local->flags = flags;
-        local->wbflags = wbflags;
-	inode_ctx_get (loc->inode, this, &tmp_list);
-	list = (int16_t *)(long)tmp_list;
-
-	local->list = list;
-	file_list[0] = priv->child_count; /* Thats namespace */
-	file_list[2] = -1;
-	for (index = 0; list[index] != -1; index++) {
-		local->call_count++;
-		if (list[index] != priv->child_count)
-			file_list[1] = list[index];
-	}
-
-	if (local->call_count != 2) {
-		/* If the lookup was done for file */
-		gf_log (this->name, GF_LOG_ERROR,
-			"%s: entry_count is %d",
-			loc->path, local->call_count);
-		for (index = 0; local->list[index] != -1; index++)
-			gf_log (this->name, GF_LOG_ERROR, "%s: found on %s",
-				loc->path, priv->xl_array[list[index]]->name);
-
-		if (local->call_count < 2) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"returning EIO as file found on onlyone node");
-			STACK_UNWIND (frame, -1, EIO, fd);
-			return 0;
-		}
-	}
-
-#ifdef GF_DARWIN_HOST_OS
-	/* Handle symlink here */
-	if (IA_ISLNK (loc->inode->ia_type)) {
-		/* Callcount doesn't matter here */
-		STACK_WIND (frame,
-			    unify_open_readlink_cbk,
-			    NS(this),
-			    NS(this)->fops->readlink,
-			    loc, ZR_PATH_MAX);
-		return 0;
-	}
-#endif /* GF_DARWIN_HOST_OS */
-
-	local->call_count = 2;
-	for (index = 0; file_list[index] != -1; index++) {
-		char need_break = (file_list[index+1] == -1);
-		STACK_WIND_COOKIE (frame,
-				   unify_open_cbk,
-				   priv->xl_array[file_list[index]], //cookie
-				   priv->xl_array[file_list[index]],
-				   priv->xl_array[file_list[index]]->fops->open,
-				   loc,
-				   flags,
-				   fd, wbflags);
-		if (need_break)
-			break;
-	}
-
-	return 0;
-}
-
-
-int32_t 
-unify_create_unlink_cbk (call_frame_t *frame,
-			 void *cookie,
-			 xlator_t *this,
-			 int32_t op_ret,
-			 int32_t op_errno,
-                         struct iatt *preparent,
-                         struct iatt *postparent)
-{
-	unify_local_t *local = frame->local;
-	inode_t *inode = local->loc1.inode;
-
-	unify_local_wipe (local);
-
-	STACK_UNWIND (frame, local->op_ret, local->op_errno, local->fd, 
-		      inode, &local->stbuf,
-                      &local->oldpreparent, &local->oldpostparent);
-  
-	return 0;
-}
-
-/**
- * unify_create_open_cbk -
- */
-int32_t
-unify_create_open_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno,
-		       fd_t *fd)
-{
-	int ret = 0;
-	int32_t callcnt = 0;
-	unify_local_t *local = frame->local;
-	inode_t *inode = NULL;
-	xlator_t *child = NULL;
-	uint64_t tmp_value = 0;
-
-	LOCK (&frame->lock);
-	{
-		if (op_ret >= 0) {
-			local->op_ret = op_ret;
-			if (NS(this) != (xlator_t *)cookie) {
-				/* Store child node's ptr, used in all 
-				   the f*** / FileIO calls */
-				/* TODO: log on failure */
-				ret = fd_ctx_get (fd, this, &tmp_value);
-				cookie = (void *)(long)tmp_value;
-			} else {
-				/* NOTE: open successful on namespace.
-				 *       fd's ctx can be used to identify open 
-				 *       failure on storage subvolume. cool 
-				 *       ide ;) */
-				local->failed = 0;
-			}
-		} else {
-			gf_log (this->name, GF_LOG_ERROR,
-				"child(%s): path(%s): %s", 
-				((xlator_t *)cookie)->name,
-				local->loc1.path, strerror (op_errno));
-			local->op_errno = op_errno;
-			local->failed = 1;
-		}
-		callcnt = --local->call_count;
-	}
-	UNLOCK (&frame->lock);
-  
-	if (!callcnt) {
-		if (local->failed == 1 && (local->op_ret >= 0)) {
-			local->call_count = 1;
-			/* return -1 to user */
-			local->op_ret = -1;
-			local->op_errno = EIO;
-			local->fd = fd;
-			local->call_count = 1;
-
-			if (!fd_ctx_get (local->fd, this, &tmp_value)) {
-				child = (xlator_t *)(long)tmp_value;
-
-				gf_log (this->name, GF_LOG_ERROR, 
-					"Create success on child node, "
-					"failed on namespace");
-
-				STACK_WIND (frame,
-					    unify_create_unlink_cbk,
-					    child,
-					    child->fops->unlink,
-					    &local->loc1);
-			} else {
-				gf_log (this->name, GF_LOG_ERROR, 
-					"Create success on namespace, "
-					"failed on child node");
-
-				STACK_WIND (frame,
-					    unify_create_unlink_cbk,
-					    NS(this),
-					    NS(this)->fops->unlink,
-					    &local->loc1);
-			}
-			return 0;
-		}
-		inode = local->loc1.inode;
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, local->op_ret, local->op_errno, fd,
-			      inode, &local->stbuf,
-                              &local->oldpreparent, &local->oldpostparent);
-	}
-	return 0;
-}
-
-/**
- * unify_create_lookup_cbk - 
- */
-int32_t 
-unify_create_lookup_cbk (call_frame_t *frame,
-			 void *cookie,
-			 xlator_t *this,
-			 int32_t op_ret,
-			 int32_t op_errno,
-			 inode_t *inode,
-			 struct iatt *buf,
-			 dict_t *dict,
-                         struct iatt *postparent)
-{
-	int32_t callcnt = 0;
-	int16_t index = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-		if (op_ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"child(%s): path(%s): %s", 
-				priv->xl_array[(long)cookie]->name, 
-				local->loc1.path, strerror (op_errno));
-			local->op_errno = op_errno;
-			local->failed = 1;
-		}
-
-		if (op_ret >= 0) {
-			local->op_ret = op_ret; 
-			local->list[local->index++] = (int16_t)(long)cookie;
-			if (NS(this) == priv->xl_array[(long)cookie]) {
-				local->ia_ino = buf->ia_ino;
-			} else {
-				local->stbuf = *buf;
-			}
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		int16_t *list = local->list;
-		int16_t file_list[3] = {0,};
-		local->op_ret = -1;
-
-		local->list [local->index] = -1;
-		file_list[0] = list[0];
-		file_list[1] = list[1];
-		file_list[2] = -1;
-
-		local->stbuf.ia_ino = local->ia_ino;
-		/* TODO: log on failure */
-		inode_ctx_put (local->loc1.inode, this, 
-			       (uint64_t)(long)local->list);
-
-		if (local->index != 2) {
-			/* Lookup failed, can't do open */
-			gf_log (this->name, GF_LOG_ERROR,
-				"%s: present on %d nodes", 
-				local->loc1.path, local->index);
-			file_list[0] = priv->child_count;
-			for (index = 0; list[index] != -1; index++) {
-				gf_log (this->name, GF_LOG_ERROR, 
-					"%s: found on %s", local->loc1.path, 
-					priv->xl_array[list[index]]->name);
-				if (list[index] != priv->child_count)
-					file_list[1] = list[index];
-			}
-
-			if (local->index < 2) {
-				unify_local_wipe (local);
-				gf_log (this->name, GF_LOG_ERROR,
-					"returning EIO as file found on "
-					"only one node");
-				STACK_UNWIND (frame, -1, EIO, 
-					      local->fd, inode, NULL,
-                                              NULL, NULL);
-				return 0;
-			}
-		}
-		/* Everything is perfect :) */    
-		local->call_count = 2;
-    
-		for (index = 0; file_list[index] != -1; index++) {
-			char need_break = (file_list[index+1] == -1);
-			STACK_WIND_COOKIE (frame,
-					   unify_create_open_cbk,
-					   priv->xl_array[file_list[index]],
-					   priv->xl_array[file_list[index]],
-					   priv->xl_array[file_list[index]]->fops->open,
-					   &local->loc1,
-					   local->flags,
-					   local->fd, 0);
-			if (need_break)
-				break;
-		}
-	}
-
-	return 0;
-}
-
-
-/**
- * unify_create_cbk -
- */
-int32_t
-unify_create_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  fd_t *fd,
-		  inode_t *inode,
-		  struct iatt *buf,
-                  struct iatt *preparent,
-                  struct iatt *postparent)
-{
-	int ret = 0;
-	unify_local_t *local = frame->local;
-	call_frame_t *prev_frame = cookie;
-	inode_t *tmp_inode = NULL;
-
-	if (op_ret == -1) {
-		/* send unlink () on Namespace */
-		local->op_errno = op_errno;
-		local->op_ret = -1;
-		local->call_count = 1;
-		gf_log (this->name, GF_LOG_ERROR,
-			"create failed on %s (file %s, error %s), "
-			"sending unlink to namespace", 
-			prev_frame->this->name, 
-			local->loc1.path, strerror (op_errno));
-
-		STACK_WIND (frame,
-			    unify_create_unlink_cbk,
-			    NS(this),
-			    NS(this)->fops->unlink,
-			    &local->loc1);
-
-		return 0;
-	}
-
-	if (op_ret >= 0) {
-		local->op_ret = op_ret;
-		local->stbuf = *buf;
-		/* Just inode number should be from NS node */
-		local->stbuf.ia_ino = local->ia_ino;
-
-		/* TODO: log on failure */
-		ret = fd_ctx_set (fd, this, (uint64_t)(long)prev_frame->this);
-	}
-  
-	tmp_inode = local->loc1.inode;
-	unify_local_wipe (local);
-	STACK_UNWIND (frame, local->op_ret, local->op_errno, local->fd, 
-		      tmp_inode, &local->stbuf,
-                      &local->oldpreparent, &local->oldpostparent);
-
-	return 0;
-}
-
-/**
- * unify_ns_create_cbk -
- * 
- */
-int32_t
-unify_ns_create_cbk (call_frame_t *frame,
-		     void *cookie,
-		     xlator_t *this,
-		     int32_t op_ret,
-		     int32_t op_errno,
-		     fd_t *fd,
-		     inode_t *inode,
-		     struct iatt *buf,
-                     struct iatt *preparent,
-                     struct iatt *postparent)
-{
-	struct sched_ops *sched_ops = NULL;
-	xlator_t *sched_xl = NULL;
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	int16_t *list = NULL;
-	int16_t index = 0;
-
-	if (op_ret == -1) {
-		/* No need to send create request to other servers, as 
-		   namespace action failed. Handle exclusive create here. */
-		if ((op_errno != EEXIST) || 
-		    ((op_errno == EEXIST) && 
-		     ((local->flags & O_EXCL) == O_EXCL))) {
-			/* If its just a create call without O_EXCL, 
-			   don't do this */
-			gf_log (this->name, GF_LOG_ERROR,
-				"namespace: path(%s): %s", 
-				local->loc1.path, strerror (op_errno));
-			unify_local_wipe (local);
-			STACK_UNWIND (frame, op_ret, op_errno, fd, inode, buf,
-                                      preparent, postparent);
-			return 0;
-		}
-	}
-  
-	if (op_ret >= 0) {
-		/* Get the inode number from the NS node */
-		local->ia_ino = buf->ia_ino;
-  
-                local->oldpreparent = *preparent;
-                local->oldpostparent = *postparent;
-
-		local->op_ret = -1;
-
-		/* Start the mapping list */
-		list = GF_CALLOC (1, sizeof (int16_t) * 3,
-                                  gf_unify_mt_int16_t);
-		ERR_ABORT (list);
-		inode_ctx_put (inode, this, (uint64_t)(long)list);
-		list[0] = priv->child_count;
-		list[2] = -1;
-
-		/* This means, file doesn't exist anywhere in the Filesystem */
-		sched_ops = priv->sched_ops;
-
-		/* Send create request to the scheduled node now */
-		sched_xl = sched_ops->schedule (this, local->loc1.path);
-		if (sched_xl == NULL)
-		{
-			/* send unlink () on Namespace */
-			local->op_errno = ENOTCONN;
-			local->op_ret = -1;
-			local->call_count = 1;
-			gf_log (this->name, GF_LOG_ERROR,
-				"no node online to schedule create:(file %s) "
-				"sending unlink to namespace", 
-				(local->loc1.path)?local->loc1.path:"");
-
-			STACK_WIND (frame,
-				    unify_create_unlink_cbk,
-				    NS(this),
-				    NS(this)->fops->unlink,
-				    &local->loc1);
-	
-			return 0;
-		}
-
-		for (index = 0; index < priv->child_count; index++)
-			if (sched_xl == priv->xl_array[index])
-				break;
-		list[1] = index;
-
-		STACK_WIND (frame, unify_create_cbk,
-			    sched_xl, sched_xl->fops->create,
-			    &local->loc1, local->flags, local->mode, fd);
-	} else {
-		/* File already exists, and there is no O_EXCL flag */
-
-		gf_log (this->name, GF_LOG_DEBUG, 
-			"File(%s) already exists on namespace, sending "
-			"open instead", local->loc1.path);
-
-		local->list = GF_CALLOC (1, sizeof (int16_t) * 3,
-                                         gf_unify_mt_int16_t);
-		ERR_ABORT (local->list);
-		local->call_count = priv->child_count + 1;
-		local->op_ret = -1;
-		for (index = 0; index <= priv->child_count; index++) {
-			/* Send lookup() to all nodes including namespace */
-			STACK_WIND_COOKIE (frame,
-					   unify_create_lookup_cbk,
-					   (void *)(long)index,
-					   priv->xl_array[index],
-					   priv->xl_array[index]->fops->lookup,
-					   &local->loc1,
-					   NULL);
-		}
-	}
-	return 0;
-}
-
-/**
- * unify_create - create a file in global namespace first, so other 
- *    clients can see them. Create the file in storage nodes in background.
- */
-int32_t
-unify_create (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      int32_t flags,
-	      mode_t mode,
-	      fd_t *fd)
-{
-	unify_local_t *local = NULL;
-  
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	local->mode = mode;
-	local->flags = flags;
-	local->fd = fd;
-
-	loc_copy (&local->loc1, loc);
-	if (local->loc1.path == NULL) {
-		gf_log (this->name, GF_LOG_CRITICAL, "Not enough memory :O");
-		STACK_UNWIND (frame, -1, ENOMEM, fd, loc->inode, NULL,
-                              NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame,
-		    unify_ns_create_cbk,
-		    NS(this),
-		    NS(this)->fops->create,
-		    loc,
-		    flags | O_EXCL,
-		    mode,
-		    fd);
-  
-	return 0;
-}
-
-
-/**
- * unify_opendir_cbk - 
- */
-int32_t
-unify_opendir_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-		   fd_t *fd)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, fd);
-
-	return 0;
-}
-
-/** 
- * unify_opendir -
- */
-int32_t
-unify_opendir (call_frame_t *frame,
-	       xlator_t *this,
-	       loc_t *loc,
-	       fd_t *fd)
-{
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-
-	STACK_WIND (frame, unify_opendir_cbk,
-		    NS(this), NS(this)->fops->opendir, loc, fd);
-
-	return 0;
-}
-
-
-int32_t
-unify_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct iatt *statpre,
-                   struct iatt *statpost)
-{
-	int32_t callcnt = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-	call_frame_t *prev_frame = cookie;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-    
-		if (op_ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"%s(): child(%s): path(%s): %s", 
-				gf_fop_list[frame->root->op],
-				prev_frame->this->name, 
-				(local->loc1.path)?local->loc1.path:"", 
-				strerror (op_errno));
-
-			local->op_errno = op_errno;
-			if ((op_errno == ENOENT) && priv->optimist) 
-				local->op_ret = 0;
-		}
-
-		if (op_ret >= 0) {
-			local->op_ret = 0;
-
-			if (NS (this) == prev_frame->this) {
-				local->ia_ino = statpost->ia_ino;
-				/* If the entry is directory, get the stat
-				   from NS node */
-				if (IA_ISDIR (statpost->ia_type) || 
-				    !local->stpost.ia_blksize) {
-					local->stpre = *statpre;
-                                        local->stpost = *statpost;
-				}
-			}
-
-			if ((!IA_ISDIR (statpost->ia_type)) && 
-			    (NS (this) != prev_frame->this)) {
-				/* If file, take the stat info from Storage 
-				   node. */
-				local->stpre = *statpre;
-                                local->stpost = *statpost;
-			}
-		}
-	}
-	UNLOCK (&frame->lock);
-    
-	if (!callcnt) {
-		/* If the inode number is not filled, operation should
-		   fail */
-		if (!local->ia_ino)
-			local->op_ret = -1;
-
-                local->stpre.ia_ino = local->ia_ino;
-		local->stpost.ia_ino = local->ia_ino;
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-			      &local->stpre, &local->stpost);
-	}
-
-	return 0;
-}
-
-
-int32_t
-unify_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-               struct iatt *stbuf, int32_t valid)
-{
-	unify_local_t *local = NULL;
-	unify_private_t *priv = this->private;
-	int32_t index = 0;
-	int32_t callcnt = 0;
-  	uint64_t tmp_list = 0;
-
-        if (!(loc && loc->inode)) {                     
-                STACK_UNWIND (frame, -1, EINVAL, NULL, NULL);
-                return 0;
-        }                                                    
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	loc_copy (&local->loc1, loc);
-
-	if (IA_ISDIR (loc->inode->ia_type)) {
-		local->call_count = 1;
-      
-                STACK_WIND (frame,
-                            unify_setattr_cbk,
-                            NS (this),
-                            NS (this)->fops->setattr,
-                            loc, stbuf, valid);
-        } else {
-		inode_ctx_get (loc->inode, this, &tmp_list);
-		local->list = (int16_t *)(long)tmp_list;
-
-		for (index = 0; local->list[index] != -1; index++) {
-			local->call_count++;
-			callcnt++;
-		}
-      
-		for (index = 0; local->list[index] != -1; index++) {
-			STACK_WIND (frame,
-				    unify_setattr_cbk,
-				    priv->xl_array[local->list[index]],
-				    priv->xl_array[local->list[index]]->fops->setattr,
-				    loc, stbuf, valid);
-
-			if (!--callcnt)
-				break;
-		}
-	}
-
-	return 0;
-}
-
-
-int32_t
-unify_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                struct iatt *stbuf, int32_t valid)
-{
-	unify_local_t *local = NULL;
-	xlator_t *child = NULL;
-	uint64_t tmp_child = 0;
-
-	UNIFY_CHECK_FD_AND_UNWIND_ON_ERR(fd);
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-
-	if (!fd_ctx_get (fd, this, &tmp_child)) {
-		/* If its set, then its file */
-		child = (xlator_t *)(long)tmp_child;		     
-
-		local->call_count = 2;
-
-		STACK_WIND (frame, unify_setattr_cbk, child,
-			    child->fops->fsetattr, fd, stbuf, valid);
-
-		STACK_WIND (frame, unify_setattr_cbk, NS(this),
-			    NS(this)->fops->fsetattr, fd, stbuf, valid);
-	} else {
-		local->call_count = 1;
-    
-		STACK_WIND (frame, unify_setattr_cbk,
-			    NS(this), NS(this)->fops->fsetattr,
-			    fd, stbuf, valid);
-	}
-  
-	return 0;
-}
-
-
-/**
- * unify_truncate_cbk - 
- */
-int32_t
-unify_truncate_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    struct iatt *prebuf,
-                    struct iatt *postbuf)
-{
-	int32_t callcnt = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-	call_frame_t *prev_frame = cookie;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-    
-		if (op_ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"child(%s): path(%s): %s", 
-				prev_frame->this->name, 
-				(local->loc1.path)?local->loc1.path:"", 
-				strerror (op_errno));
-			local->op_errno = op_errno;
-			if (!((op_errno == ENOENT) && priv->optimist))
-				local->op_ret = -1;
-		}
-
-		if (op_ret >= 0) {
-			if (NS (this) == prev_frame->this) {
-				local->ia_ino = postbuf->ia_ino;
-				/* If the entry is directory, get the 
-				   stat from NS node */
-				if (IA_ISDIR (postbuf->ia_type) ||
-				    !local->stbuf.ia_blksize) {
-					local->stbuf = *prebuf;
-                                        local->poststbuf = *postbuf;
-				}
-			}
-
-			if ((!IA_ISDIR (postbuf->ia_type)) &&
-			    (NS (this) != prev_frame->this)) {
-				/* If file, take the stat info from 
-				   Storage node. */
-				local->stbuf = *prebuf;
-                                local->poststbuf = *postbuf;
-			}
-		}
-	}
-	UNLOCK (&frame->lock);
-    
-	if (!callcnt) {
-		if (local->ia_ino) {
-			local->stbuf.ia_ino = local->ia_ino;
-                        local->poststbuf.ia_ino = local->ia_ino;
-                } else {
-			local->op_ret = -1;
-                }
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-			      &local->stbuf, &local->poststbuf);
-	}
-
-	return 0;
-}
-
-
-/**
- * unify_truncate - 
- */
-int32_t
-unify_truncate (call_frame_t *frame,
-		xlator_t *this,
-		loc_t *loc,
-		off_t offset)
-{
-	unify_local_t *local = NULL;
-	unify_private_t *priv = this->private;
-	int32_t index = 0;
-	int32_t callcnt = 0;
-  	uint64_t tmp_list = 0;
-
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	loc_copy (&local->loc1, loc);
-	local->ia_ino = loc->inode->ino;
-
-	if (IA_ISDIR (loc->inode->ia_type)) {
-		local->call_count = 1;
-      
-		STACK_WIND (frame,
-			    unify_truncate_cbk,
-			    NS(this),
-			    NS(this)->fops->truncate,
-			    loc,
-                            0);
-	} else {
-		local->op_ret = 0;
-		inode_ctx_get (loc->inode, this, &tmp_list);
-		local->list = (int16_t *)(long)tmp_list;
-
-		for (index = 0; local->list[index] != -1; index++) {
-			local->call_count++;
-			callcnt++;
-		}
-      
-		/* Don't send offset to NS truncate */
-		STACK_WIND (frame, unify_truncate_cbk, NS(this),
-			    NS(this)->fops->truncate, loc, 0);
-		callcnt--;
-
-		for (index = 0; local->list[index] != -1; index++) {
-			if (NS(this) != priv->xl_array[local->list[index]]) {
-				STACK_WIND (frame,
-					    unify_truncate_cbk,
-					    priv->xl_array[local->list[index]],
-					    priv->xl_array[local->list[index]]->fops->truncate,
-					    loc,
-					    offset);
-				if (!--callcnt)
-					break;
-			}
-		}
-	}
-
-	return 0;
-}
-
-/**
- * unify_readlink_cbk - 
- */
-int32_t
-unify_readlink_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    const char *path,
-                    struct iatt *sbuf)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, path, sbuf);
-	return 0;
-}
-
-/**
- * unify_readlink - Read the link only from the storage node.
- */
-int32_t
-unify_readlink (call_frame_t *frame,
-		xlator_t *this,
-		loc_t *loc,
-		size_t size)
-{
-	unify_private_t *priv = this->private;
-	int32_t entry_count = 0;
-	int16_t *list = NULL;
-	int16_t index = 0;
-  	uint64_t tmp_list = 0;
-  
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-	
-	inode_ctx_get (loc->inode, this, &tmp_list);
-	list = (int16_t *)(long)tmp_list;
-
-	for (index = 0; list[index] != -1; index++)
-		entry_count++;
-
-	if (entry_count >= 2) {
-		for (index = 0; list[index] != -1; index++) {
-			if (priv->xl_array[list[index]] != NS(this)) {
-				STACK_WIND (frame,
-					    unify_readlink_cbk,
-					    priv->xl_array[list[index]],
-					    priv->xl_array[list[index]]->fops->readlink,
-					    loc,
-					    size);
-				break;
-			}
-		}
-	} else {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"returning ENOENT, no softlink files found "
-			"on storage node");
-		STACK_UNWIND (frame, -1, ENOENT, NULL);
-	}
-
-	return 0;
-}
-
-
-/**
- * unify_unlink_cbk - 
- */
-int32_t
-unify_unlink_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-                  struct iatt *preparent,
-                  struct iatt *postparent)
-{
-	int32_t callcnt = 0;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-		if (op_ret == 0  || ((op_errno == ENOENT) && priv->optimist))
-			local->op_ret = 0;
-		if (op_ret == -1)
-			local->op_errno = op_errno;
-
-                if (((call_frame_t *)cookie)->this == NS(this)) {
-                        local->oldpreparent = *preparent;
-                        local->oldpostparent = *postparent;
-                }
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, local->op_ret, local->op_errno,
-                              &local->oldpreparent, &local->oldpostparent);
-	}
-
-	return 0;
-}
-
-
-/**
- * unify_unlink - 
- */
-int32_t
-unify_unlink (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc)
-{
-	unify_private_t *priv = this->private;
-	unify_local_t *local = NULL;
-	int16_t *list = NULL;
-	int16_t index = 0;
-  	uint64_t tmp_list = 0;
-
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	loc_copy (&local->loc1, loc);
-
-	inode_ctx_get (loc->inode, this, &tmp_list);
-	list = (int16_t *)(long)tmp_list;
-
-	for (index = 0; list[index] != -1; index++)
-		local->call_count++;
-
-	if (local->call_count) {
-		for (index = 0; list[index] != -1; index++) {
-			char need_break = (list[index+1] == -1);
-			STACK_WIND (frame,
-				    unify_unlink_cbk,
-				    priv->xl_array[list[index]],
-				    priv->xl_array[list[index]]->fops->unlink,
-				    loc);
-			if (need_break)
-				break;
-		}
-	} else {
-		gf_log (this->name, GF_LOG_ERROR,
-			"%s: returning ENOENT", loc->path);
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-	}
-
-	return 0;
-}
-
-
-/**
- * unify_readv_cbk - 
- */
-int32_t
-unify_readv_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 struct iovec *vector,
-		 int32_t count,
-		 struct iatt *stbuf,
-                 struct iobref *iobref)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, vector, count, stbuf, iobref);
-	return 0;
-}
-
-/**
- * unify_readv - 
- */
-int32_t
-unify_readv (call_frame_t *frame,
-	     xlator_t *this,
-	     fd_t *fd,
-	     size_t size,
-	     off_t offset)
-{
-	UNIFY_CHECK_FD_CTX_AND_UNWIND_ON_ERR (fd);
-	xlator_t *child = NULL;
-	uint64_t tmp_child = 0;
-
-	fd_ctx_get (fd, this, &tmp_child);
-	child = (xlator_t *)(long)tmp_child;		     
-
-	STACK_WIND (frame,
-		    unify_readv_cbk,
-		    child,
-		    child->fops->readv,
-		    fd,
-		    size,
-		    offset);
-
-
-	return 0;
-}
-
-/**
- * unify_writev_cbk - 
- */
-int32_t
-unify_writev_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-                  struct iatt *prebuf,
-		  struct iatt *postbuf)
-{
-	unify_local_t *local = NULL;
-
-        local = frame->local;
-
-        local->stbuf = *prebuf;
-        local->stbuf.ia_ino = local->ia_ino;
-
-        local->poststbuf = *postbuf;
-        local->poststbuf.ia_ino = local->ia_ino;
-
-	STACK_UNWIND (frame, op_ret, op_errno,
-                      &local->stbuf, &local->poststbuf);
-	return 0;
-}
-
-/**
- * unify_writev - 
- */
-int32_t
-unify_writev (call_frame_t *frame,
-	      xlator_t *this,
-	      fd_t *fd,
-	      struct iovec *vector,
-	      int32_t count,
-	      off_t off,
-              struct iobref *iobref)
-{
-	UNIFY_CHECK_FD_CTX_AND_UNWIND_ON_ERR (fd);
-	xlator_t *child = NULL;
-	uint64_t tmp_child = 0;
-	unify_local_t *local = NULL;
-
-	INIT_LOCAL (frame, local);
-        local->ia_ino = fd->inode->ino;
-
-	fd_ctx_get (fd, this, &tmp_child);
-	child = (xlator_t *)(long)tmp_child;		     
-
-	STACK_WIND (frame,
-		    unify_writev_cbk,
-		    child,
-		    child->fops->writev,
-		    fd,
-		    vector,
-		    count,
-		    off,
-                    iobref);
-
-	return 0;
-}
-
-/**
- * unify_ftruncate -
- */
-int32_t
-unify_ftruncate (call_frame_t *frame,
-		 xlator_t *this,
-		 fd_t *fd,
-		 off_t offset)
-{
-	xlator_t *child = NULL;
-	unify_local_t *local = NULL;
-	uint64_t tmp_child = 0;
-
-	UNIFY_CHECK_FD_CTX_AND_UNWIND_ON_ERR(fd);
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	local->op_ret = 0;
-
-	fd_ctx_get (fd, this, &tmp_child);
-	child = (xlator_t *)(long)tmp_child;		     
-
-	local->call_count = 2;
-  
-	STACK_WIND (frame, unify_truncate_cbk,
-		    child, child->fops->ftruncate,
-		    fd, offset);
-  
-	STACK_WIND (frame, unify_truncate_cbk,
-		    NS(this), NS(this)->fops->ftruncate,
-		    fd, 0);
-  
-	return 0;
-}
-
-
-/**
- * unify_flush_cbk - 
- */
-int32_t
-unify_flush_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-/**
- * unify_flush -
- */
-int32_t
-unify_flush (call_frame_t *frame,
-	     xlator_t *this,
-	     fd_t *fd)
-{
-	UNIFY_CHECK_FD_CTX_AND_UNWIND_ON_ERR (fd);
-	xlator_t *child = NULL;
-	uint64_t tmp_child = 0;
-
-	fd_ctx_get (fd, this, &tmp_child);
-	child = (xlator_t *)(long)tmp_child;		     
-
-	STACK_WIND (frame, unify_flush_cbk, child, 
-		    child->fops->flush, fd);
-
-	return 0;
-}
-
-
-/**
- * unify_fsync_cbk - 
- */
-int32_t
-unify_fsync_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-                 struct iatt *prebuf,
-                 struct iatt *postbuf)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, prebuf, postbuf);
-	return 0;
-}
-
-/**
- * unify_fsync - 
- */
-int32_t
-unify_fsync (call_frame_t *frame,
-	     xlator_t *this,
-	     fd_t *fd,
-	     int32_t flags)
-{
-	UNIFY_CHECK_FD_CTX_AND_UNWIND_ON_ERR (fd);
-	xlator_t *child = NULL;
-	uint64_t tmp_child = 0;
-
-	fd_ctx_get (fd, this, &tmp_child);
-	child = (xlator_t *)(long)tmp_child;		     
-
-	STACK_WIND (frame, unify_fsync_cbk, child,
-		    child->fops->fsync, fd, flags);
-
-	return 0;
-}
-
-/**
- * unify_fstat - Send fstat FOP to Namespace only if its directory, and to 
- *     both namespace and the storage node if its a file.
- */
-int32_t
-unify_fstat (call_frame_t *frame,
-	     xlator_t *this,
-	     fd_t *fd)
-{
-	unify_local_t *local = NULL;
-	xlator_t *child = NULL;
-	uint64_t tmp_child = 0;
-
-	UNIFY_CHECK_FD_AND_UNWIND_ON_ERR(fd);
-
-	INIT_LOCAL (frame, local);
-	local->ia_ino = fd->inode->ino;
-
-	if (!fd_ctx_get (fd, this, &tmp_child)) {
-		/* If its set, then its file */
-		child = (xlator_t *)(long)tmp_child;		     
-		local->call_count = 2;
-
-		STACK_WIND (frame, unify_buf_cbk, child,
-			    child->fops->fstat, fd);
-
-		STACK_WIND (frame, unify_buf_cbk, NS(this),
-			    NS(this)->fops->fstat, fd);
-
-	} else {
-		/* this is an directory */
-		local->call_count = 1;
-		STACK_WIND (frame, unify_buf_cbk, NS(this),
-			    NS(this)->fops->fstat, fd);
-	}
-
-	return 0;
-}
-
-/**
- * unify_getdents_cbk - 
- */
-int32_t
-unify_getdents_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    dir_entry_t *entry,
-		    int32_t count)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, entry, count);
-	return 0;
-}
-
-/**
- * unify_getdents - send the FOP request to all the nodes.
- */
-int32_t
-unify_getdents (call_frame_t *frame,
-		xlator_t *this,
-		fd_t *fd,
-		size_t size,
-		off_t offset,
-		int32_t flag)
-{
-	UNIFY_CHECK_FD_AND_UNWIND_ON_ERR (fd);
-
-	STACK_WIND (frame, unify_getdents_cbk, NS(this),
-		    NS(this)->fops->getdents, fd, size, offset, flag);
-
-	return 0;
-}
-
-
-/**
- * unify_readdir_cbk - 
- */
-int32_t
-unify_readdir_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-		   gf_dirent_t *buf)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, buf);
-
-	return 0;
-}
-
-/**
- * unify_readdir - send the FOP request to all the nodes.
- */
-int32_t
-unify_readdir (call_frame_t *frame,
-	       xlator_t *this,
-	       fd_t *fd,
-	       size_t size,
-	       off_t offset)
-{
-	UNIFY_CHECK_FD_AND_UNWIND_ON_ERR (fd);
-
-	STACK_WIND (frame, unify_readdir_cbk, NS(this),
-		    NS(this)->fops->readdir, fd, size, offset);
-
-	return 0;
-}
-
-
-int32_t
-unify_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, gf_dirent_t *buf)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, buf);
-
-	return 0;
-}
-
-
-int32_t
-unify_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-                off_t offset)
-{
-	UNIFY_CHECK_FD_AND_UNWIND_ON_ERR (fd);
-
-	STACK_WIND (frame, unify_readdirp_cbk, NS(this),
-		    NS(this)->fops->readdirp, fd, size, offset);
-
-	return 0;
-}
-
-
-/**
- * unify_fsyncdir_cbk - 
- */
-int32_t
-unify_fsyncdir_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-
-	return 0;
-}
-
-/**
- * unify_fsyncdir -
- */
-int32_t
-unify_fsyncdir (call_frame_t *frame,
-		xlator_t *this,
-		fd_t *fd,
-		int32_t flags)
-{
-	UNIFY_CHECK_FD_AND_UNWIND_ON_ERR (fd);
-
-	STACK_WIND (frame, unify_fsyncdir_cbk,
-		    NS(this), NS(this)->fops->fsyncdir, fd, flags);
-
-	return 0;
-}
-
-/**
- * unify_lk_cbk - UNWIND frame with the proper return arguments.
- */
-int32_t
-unify_lk_cbk (call_frame_t *frame,
-	      void *cookie,
-	      xlator_t *this,
-	      int32_t op_ret,
-	      int32_t op_errno,
-	      struct gf_flock *lock)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, lock);
-	return 0;
-}
-
-/**
- * unify_lk - Send it to all the storage nodes, (should be 1) which has file.
- */
-int32_t
-unify_lk (call_frame_t *frame,
-	  xlator_t *this,
-	  fd_t *fd,
-	  int32_t cmd,
-	  struct gf_flock *lock)
-{
-	UNIFY_CHECK_FD_CTX_AND_UNWIND_ON_ERR (fd);
-	xlator_t *child = NULL;
-	uint64_t tmp_child = 0;
-
-	fd_ctx_get (fd, this, &tmp_child);
-	child = (xlator_t *)(long)tmp_child;		     
-
-	STACK_WIND (frame, unify_lk_cbk, child,
-		    child->fops->lk, fd, cmd, lock);
-
-	return 0;
-}
-
-
-int32_t
-unify_setxattr_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno);
-
-static int32_t
-unify_setxattr_file_cbk (call_frame_t *frame,
-			 void *cookie,
-			 xlator_t *this,
-			 int32_t op_ret,
-			 int32_t op_errno)
-{
-	unify_private_t *private = this->private;
-	unify_local_t *local = frame->local;
-	xlator_t *sched_xl = NULL;
-	struct sched_ops *sched_ops = NULL;
-
-	if (op_ret == -1) {
-		if (!ENOTSUP)
-			gf_log (this->name, GF_LOG_ERROR,
-				"setxattr with XATTR_CREATE on ns: "
-				"path(%s) key(%s): %s",
-				local->loc1.path, local->name, 
-				strerror (op_errno));
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, op_ret, op_errno);
-		return 0;
-	} 
-
-	LOCK (&frame->lock);
-	{
-		local->failed = 0;
-		local->op_ret = 0;
-		local->op_errno = 0;
-		local->call_count = 1;
-	}
-	UNLOCK (&frame->lock);
-
-	/* schedule XATTR_CREATE on one of the child node */
-	sched_ops = private->sched_ops;
-    
-	/* Send create request to the scheduled node now */
-	sched_xl = sched_ops->schedule (this, local->name); 
-	if (!sched_xl) {
-		STACK_UNWIND (frame, -1, ENOTCONN);
-		return 0;
-	}
-
-	STACK_WIND (frame,
-		    unify_setxattr_cbk,
-		    sched_xl,
-		    sched_xl->fops->setxattr,
-		    &local->loc1,
-		    local->dict,
-		    local->flags);
-	return 0;
-}
-
-/**
- * unify_setxattr_cbk - When all the child nodes return, UNWIND frame.
- */
-int32_t
-unify_setxattr_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno)
-{
-	int32_t callcnt = 0;
-	unify_local_t *local = frame->local;
-	call_frame_t *prev_frame = cookie;
-	dict_t *dict = NULL;
-
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-    
-		if (op_ret == -1) {
-			gf_log (this->name, (((op_errno == ENOENT) || 
-					      (op_errno == ENOTSUP))? 
-					     GF_LOG_DEBUG : GF_LOG_ERROR), 
-				"child(%s): path(%s): %s", 
-				prev_frame->this->name, 
-				(local->loc1.path)?local->loc1.path:"", 
-				strerror (op_errno));
-			if (local->failed == -1) {
-				local->failed = 1;
-			}
-			local->op_errno = op_errno;
-		} else {
-			local->failed = 0;
-			local->op_ret = op_ret;
-		}
-	}
-	UNLOCK (&frame->lock);
-  
-	if (!callcnt) {
-		if (local->failed && local->name && 
-		    ZR_FILE_CONTENT_REQUEST(local->name)) {      
-			dict = get_new_dict ();
-			dict_set (dict, local->dict->members_list->key, 
-				  data_from_dynptr(NULL, 0));
-			dict_ref (dict);
-
-			local->call_count = 1;
-
-			STACK_WIND (frame,
-				    unify_setxattr_file_cbk,
-				    NS(this),
-				    NS(this)->fops->setxattr,
-				    &local->loc1,
-				    dict,
-				    XATTR_CREATE);
-
-			dict_unref (dict);
-			return 0;
-		}
-    
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, local->op_ret, local->op_errno);
-	}
-
-	return 0;
-}
-
-/**
- * unify_sexattr - This function should be sent to all the storage nodes, 
- *       which contains the file, (excluding namespace).
- */
-int32_t
-unify_setxattr (call_frame_t *frame,
-		xlator_t *this,
-		loc_t *loc,
-		dict_t *dict,
-		int32_t flags)
-{
-	unify_private_t *priv = this->private;
-	unify_local_t *local = NULL;
-	int16_t *list = NULL;
-	int16_t index = 0;
-	int32_t call_count = 0;
-  	uint64_t tmp_list = 0;
-	data_pair_t *trav = dict->members_list;
-
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	local->failed = -1;
-	loc_copy (&local->loc1, loc);
-
-	if (IA_ISDIR (loc->inode->ia_type)) {
-
-		if (trav && trav->key && ZR_FILE_CONTENT_REQUEST(trav->key)) {
-			/* direct the storage xlators to change file 
-			   content only if file exists */
-			local->flags = flags;
-			local->dict = dict;
-			local->name = gf_strdup (trav->key);
-			flags |= XATTR_REPLACE;
-		}
-
-		local->call_count = priv->child_count;
-		for (index = 0; index < priv->child_count; index++) {
-			STACK_WIND (frame,
-				    unify_setxattr_cbk,
-				    priv->xl_array[index],
-				    priv->xl_array[index]->fops->setxattr,
-				    loc, dict, flags);
-		}
-		return 0;
-	}
-
-	inode_ctx_get (loc->inode, this, &tmp_list);
-	list = (int16_t *)(long)tmp_list;
-
-	for (index = 0; list[index] != -1; index++) {
-		if (NS(this) != priv->xl_array[list[index]]) {
-			local->call_count++;
-			call_count++;
-		}
-	}
-  
-	if (local->call_count) {
-		for (index = 0; list[index] != -1; index++) {
-			if (priv->xl_array[list[index]] != NS(this)) {
-				STACK_WIND (frame,
-					    unify_setxattr_cbk,
-					    priv->xl_array[list[index]],
-					    priv->xl_array[list[index]]->fops->setxattr,
-					    loc,
-					    dict,
-					    flags);
-				if (!--call_count)
-					break;
-			}
-		}
-		return 0;
-	}
-
-	/* No entry in storage nodes */
-	gf_log (this->name, GF_LOG_DEBUG, 
-		"returning ENOENT, file not found on storage node.");
-	STACK_UNWIND (frame, -1, ENOENT);
-
-	return 0;
-}
-
-
-/**
- * unify_getxattr_cbk - This function is called from only one child, so, no
- *     need of any lock or anything else, just send it to above layer 
- */
-int32_t
-unify_getxattr_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    dict_t *value)
-{
-	int32_t callcnt = 0;
-	dict_t *local_value = NULL;
-	unify_local_t *local = frame->local;
-	call_frame_t *prev_frame = cookie;
-  
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-    
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			gf_log (this->name, 
-				(((op_errno == ENOENT) || 
-				  (op_errno == ENODATA) || 
-				  (op_errno == ENOTSUP)) ? 
-				 GF_LOG_DEBUG : GF_LOG_ERROR), 
-				"child(%s): path(%s): %s", 
-				prev_frame->this->name, 
-				(local->loc1.path)?local->loc1.path:"", 
-				strerror (op_errno));
-		} else {
-			if (!local->dict)
-				local->dict = dict_ref (value);
-			local->op_ret = op_ret;
-		}
-	}
-	UNLOCK (&frame->lock);
-  
-	if (!callcnt) {
-		local_value = local->dict;
-		local->dict = NULL;
-      
-		STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-			      local_value);
-      
-		if (local_value)
-			dict_unref (local_value);
-	} 
-
-	return 0;
-}
-
-
-/** 
- * unify_getxattr - This FOP is sent to only the storage node.
- */
-int32_t
-unify_getxattr (call_frame_t *frame,
-		xlator_t *this,
-		loc_t *loc,
-		const char *name)
-{
-	unify_private_t *priv = this->private;
-	int16_t *list = NULL;
-	int16_t index = 0;
-	int16_t count = 0;
-	unify_local_t *local = NULL;
-  	uint64_t tmp_list = 0;
-
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-	INIT_LOCAL (frame, local);
-
-	if (IA_ISDIR (loc->inode->ia_type)) {
-		local->call_count = priv->child_count;
-		for (index = 0; index < priv->child_count; index++)
-			STACK_WIND (frame,
-				    unify_getxattr_cbk,
-				    priv->xl_array[index],
-				    priv->xl_array[index]->fops->getxattr,
-				    loc,
-				    name);
-		return 0;
-	}
-
-	inode_ctx_get (loc->inode, this, &tmp_list);
-	list = (int16_t *)(long)tmp_list;
-
-	for (index = 0; list[index] != -1; index++) {
-		if (NS(this) != priv->xl_array[list[index]]) {
-			local->call_count++;
-			count++;
-		}
-	}
-
-	if (count) {
-		for (index = 0; list[index] != -1; index++) {
-			if (priv->xl_array[list[index]] != NS(this)) {
-				STACK_WIND (frame,
-					    unify_getxattr_cbk,
-					    priv->xl_array[list[index]],
-					    priv->xl_array[list[index]]->fops->getxattr,
-					    loc,
-					    name);
-				if (!--count)
-					break;
-			}
-		}
-	} else {
-		dict_t *tmp_dict = get_new_dict ();
-		gf_log (this->name, GF_LOG_DEBUG, 
-			"%s: returning ENODATA, no file found on storage node",
-			loc->path);
-		STACK_UNWIND (frame, -1, ENODATA, tmp_dict);
-		dict_destroy (tmp_dict);
-	}
-
-	return 0;
-}
-
-/**
- * unify_removexattr_cbk - Wait till all the child node returns the call
- *      and then UNWIND to above layer.
- */
-int32_t
-unify_removexattr_cbk (call_frame_t *frame,
-		       void *cookie,
-		       xlator_t *this,
-		       int32_t op_ret,
-		       int32_t op_errno)
-{
-	int32_t callcnt = 0;
-	unify_local_t *local = frame->local;
-	call_frame_t *prev_frame = cookie;
-
-	LOCK (&frame->lock);
-	{ 
-		callcnt = --local->call_count;
-		if (op_ret == -1) {
-			local->op_errno = op_errno;
-			if (op_errno != ENOTSUP)
-				gf_log (this->name, GF_LOG_ERROR, 
-					"child(%s): path(%s): %s", 
-					prev_frame->this->name, 
-					local->loc1.path, strerror (op_errno));
-		} else {
-			local->op_ret = op_ret;
-		}
-	}
-	UNLOCK (&frame->lock);  
-
-	if (!callcnt) {
-		STACK_UNWIND (frame, local->op_ret, local->op_errno);
-	}
-
-	return 0;
-}
-
-/**
- * unify_removexattr - Send it to all the child nodes which has the files.
- */
-int32_t
-unify_removexattr (call_frame_t *frame,
-		   xlator_t *this,
-		   loc_t *loc,
-		   const char *name)
-{
-	unify_private_t *priv = this->private;
-	unify_local_t *local = NULL;
-	int16_t *list = NULL;
-	int16_t index = 0;
-	int32_t call_count = 0;
-  	uint64_t tmp_list = 0;
-
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (loc);
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-
-	if (IA_ISDIR (loc->inode->ia_type)) {
-		local->call_count = priv->child_count;
-		for (index = 0; index < priv->child_count; index++)
-			STACK_WIND (frame, 		    
-				    unify_removexattr_cbk,
-				    priv->xl_array[index],
-				    priv->xl_array[index]->fops->removexattr,
-				    loc,
-				    name);
-
-		return 0;
-	}
-
-	inode_ctx_get (loc->inode, this, &tmp_list);
-	list = (int16_t *)(long)tmp_list;
-
-	for (index = 0; list[index] != -1; index++) {
-		if (NS(this) != priv->xl_array[list[index]]) {
-			local->call_count++;
-			call_count++;
-		}
-	}
-
-	if (local->call_count) {
-		for (index = 0; list[index] != -1; index++) {
-			if (priv->xl_array[list[index]] != NS(this)) {
-				STACK_WIND (frame,
-					    unify_removexattr_cbk,
-					    priv->xl_array[list[index]],
-					    priv->xl_array[list[index]]->fops->removexattr,
-					    loc,
-					    name);
-				if (!--call_count)
-					break;
-			}
-		}
-		return 0;
-	} 
-
-	gf_log (this->name, GF_LOG_DEBUG, 
-		"%s: returning ENOENT, not found on storage node.", loc->path);
-	STACK_UNWIND (frame, -1, ENOENT);
-
-	return 0;
-}
-
-
-int32_t 
-unify_mknod_unlink_cbk (call_frame_t *frame,
-			void *cookie,
-			xlator_t *this,
-			int32_t op_ret,
-			int32_t op_errno,
-                        struct iatt *preparent,
-                        struct iatt *postparent)
-{
-	unify_local_t *local = frame->local;
-
-	if (op_ret == -1)
-		gf_log (this->name, GF_LOG_ERROR, 
-			"%s: %s", local->loc1.path, strerror (op_errno));
-  
-	unify_local_wipe (local);
-	/* No log required here as this -1 is for mknod call */
-	STACK_UNWIND (frame, -1, local->op_errno, NULL, NULL);
-	return 0;
-}
-
-/**
- * unify_mknod_cbk - 
- */
-int32_t
-unify_mknod_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 inode_t *inode,
-                 struct iatt *buf,
-                 struct iatt *preparent,
-                 struct iatt *postparent)
-{
-	unify_local_t *local = frame->local;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"mknod failed on storage node, sending unlink to "
-			"namespace");
-		local->op_errno = op_errno;
-		STACK_WIND (frame,
-			    unify_mknod_unlink_cbk,
-			    NS(this),
-			    NS(this)->fops->unlink,
-			    &local->loc1);
-		return 0;
-	}
-  
-	local->stbuf = *buf;
-	local->stbuf.ia_ino = local->ia_ino;
-	unify_local_wipe (local);
-	STACK_UNWIND (frame, op_ret, op_errno, inode, &local->stbuf,
-                      &local->oldpreparent, &local->oldpostparent);
-	return 0;
-}
-
-/**
- * unify_ns_mknod_cbk - 
- */
-int32_t
-unify_ns_mknod_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    inode_t *inode,
-                    struct iatt *buf,
-                    struct iatt *preparent,
-                    struct iatt *postparent)
-{
-	struct sched_ops *sched_ops = NULL;
-	xlator_t *sched_xl = NULL;
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	int16_t *list = NULL;
-	int16_t index = 0;
-	call_frame_t *prev_frame = cookie;
-
-	if (op_ret == -1) {
-		/* No need to send mknod request to other servers, 
-		 * as namespace action failed 
-		 */
-		gf_log (this->name, GF_LOG_ERROR, 
-			"child(%s): path(%s): %s", 
-			prev_frame->this->name, local->loc1.path, 
-			strerror (op_errno));
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, op_ret, op_errno, inode, buf,
-                              preparent, postparent);
-		return 0;
-	}
-  
-	/* Create one inode for this entry */
-	local->op_ret = 0;
-	local->stbuf = *buf;
-	local->ia_ino = buf->ia_ino;
-
-        local->oldpreparent = *preparent;
-        local->oldpostparent = *postparent;
-
-	list = GF_CALLOC (1, sizeof (int16_t) * 3, gf_unify_mt_int16_t);
-	ERR_ABORT (list);
-	list[0] = priv->child_count;
-	list[2] = -1;
-	inode_ctx_put (inode, this, (uint64_t)(long)list);
-
-	sched_ops = priv->sched_ops;
-
-	/* Send mknod request to scheduled node now */
-	sched_xl = sched_ops->schedule (this, local->loc1.path); 
-	if (!sched_xl) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"mknod failed on storage node, no node online "
-			"at the moment, sending unlink to NS");
-		local->op_errno = ENOTCONN;
-		STACK_WIND (frame,
-			    unify_mknod_unlink_cbk,
-			    NS(this),
-			    NS(this)->fops->unlink,
-			    &local->loc1);
-      
-		return 0;
-	}
-
-	for (index = 0; index < priv->child_count; index++)
-		if (sched_xl == priv->xl_array[index])
-			break;
-	list[1] = index;
-  
-	STACK_WIND (frame,  unify_mknod_cbk,
-		    sched_xl,  sched_xl->fops->mknod,
-		    &local->loc1, local->mode, local->dev);
-
-	return 0;
-}
-
-/**
- * unify_mknod - Create a device on namespace first, and later create on 
- *       the storage node.
- */
-int32_t
-unify_mknod (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     mode_t mode,
-	     dev_t rdev)
-{
-	unify_local_t *local = NULL;
-  
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	local->mode = mode;
-	local->dev = rdev;
-	loc_copy (&local->loc1, loc);
-	if (local->loc1.path == NULL) {
-		gf_log (this->name, GF_LOG_CRITICAL, "Not enough memory :O");
-		STACK_UNWIND (frame, -1, ENOMEM, loc->inode, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame,
-		    unify_ns_mknod_cbk,
-		    NS(this),
-		    NS(this)->fops->mknod,
-		    loc,
-		    mode,
-		    rdev);
-
-	return 0;
-}
-
-int32_t 
-unify_symlink_unlink_cbk (call_frame_t *frame,
-			  void *cookie,
-			  xlator_t *this,
-			  int32_t op_ret,
-			  int32_t op_errno,
-                          struct iatt *preparent,
-                          struct iatt *postparent)
-{
-	unify_local_t *local = frame->local;
-	if (op_ret == -1)
-		gf_log (this->name, GF_LOG_ERROR, 
-			"%s: %s", local->loc1.path, strerror (op_errno));
-
-	unify_local_wipe (local);
-	STACK_UNWIND (frame, -1, local->op_errno, NULL, NULL);
-	return 0;
-}
-
-/**
- * unify_symlink_cbk - 
- */
-int32_t
-unify_symlink_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-		   inode_t *inode,
-                   struct iatt *buf,
-                   struct iatt *preparent,
-                   struct iatt *postparent)
-{
-	unify_local_t *local = frame->local;
-
-	if (op_ret == -1) {
-		/* Symlink on storage node failed, hence send unlink 
-		   to the NS node */
-		local->op_errno = op_errno;
-		gf_log (this->name, GF_LOG_ERROR, 
-			"symlink on storage node failed, sending unlink "
-			"to namespace");
-
-		STACK_WIND (frame,
-			    unify_symlink_unlink_cbk,
-			    NS(this),
-			    NS(this)->fops->unlink,
-			    &local->loc1);
-    
-		return 0;
-	}
-  
-	local->stbuf = *buf;
-	local->stbuf.ia_ino = local->ia_ino;
-	unify_local_wipe (local);
-	STACK_UNWIND (frame, op_ret, op_errno, inode, &local->stbuf,
-                      &local->oldpreparent, &local->oldpostparent);
-
-	return 0;
-}
-
-/**
- * unify_ns_symlink_cbk - 
- */
-int32_t
-unify_ns_symlink_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno,
-		      inode_t *inode,
-                      struct iatt *buf,
-                      struct iatt *preparent,
-                      struct iatt *postparent)
-{
-
-	struct sched_ops *sched_ops = NULL;
-	xlator_t *sched_xl = NULL;
-	int16_t *list = NULL;
-	unify_local_t *local = frame->local;
-	unify_private_t *priv = this->private;
-	int16_t index = 0;
-
-	if (op_ret == -1) {
-		/* No need to send symlink request to other servers, 
-		 * as namespace action failed 
-		 */
-		gf_log (this->name, GF_LOG_ERROR, 
-			"namespace: path(%s): %s", 
-			local->loc1.path, strerror (op_errno));
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, op_ret, op_errno, NULL, buf,
-                              preparent, postparent);
-		return 0;
-	}
-  
-	/* Create one inode for this entry */
-	local->op_ret = 0;
-	local->ia_ino = buf->ia_ino;
-  
-        local->oldpreparent = *preparent;
-        local->oldpostparent = *postparent;
-
-	/* Start the mapping list */
-
-	list = GF_CALLOC (1, sizeof (int16_t) * 3, gf_unify_mt_int16_t);
-	ERR_ABORT (list);
-	list[0] = priv->child_count; //namespace's index
-	list[2] = -1;
-	inode_ctx_put (inode, this, (uint64_t)(long)list);
-
-	sched_ops = priv->sched_ops;
-
-	/* Send symlink request to all the nodes now */
-	sched_xl = sched_ops->schedule (this, local->loc1.path); 
-	if (!sched_xl) {
-		/* Symlink on storage node failed, hence send unlink 
-		   to the NS node */
-		local->op_errno = ENOTCONN;
-		gf_log (this->name, GF_LOG_ERROR, 
-			"symlink on storage node failed, no node online, "
-			"sending unlink to namespace");
-      
-		STACK_WIND (frame,
-			    unify_symlink_unlink_cbk,
-			    NS(this),
-			    NS(this)->fops->unlink,
-			    &local->loc1);
-      
-		return 0;
-	}
-
-	for (index = 0; index < priv->child_count; index++)
-		if (sched_xl == priv->xl_array[index])
-			break;
-	list[1] = index;
-	
-	STACK_WIND (frame,
-		    unify_symlink_cbk,
-		    sched_xl,
-		    sched_xl->fops->symlink,
-		    local->name,
-		    &local->loc1);
-
-	return 0;
-}
-
-/**
- * unify_symlink - 
- */
-int32_t
-unify_symlink (call_frame_t *frame,
-	       xlator_t *this,
-	       const char *linkpath,
-	       loc_t *loc)
-{
-	unify_local_t *local = NULL;
-  
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	loc_copy (&local->loc1, loc);
-	local->name = gf_strdup (linkpath);
-
-	if ((local->name == NULL) || 
-	    (local->loc1.path == NULL)) {
-		gf_log (this->name, GF_LOG_CRITICAL, "Not enough memory :O");
-		STACK_UNWIND (frame, -1, ENOMEM, loc->inode, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame,
-		    unify_ns_symlink_cbk,
-		    NS(this),
-		    NS(this)->fops->symlink,
-		    linkpath,
-		    loc);
-
-	return 0;
-}
-
-
-int32_t 
-unify_rename_unlink_cbk (call_frame_t *frame,
-			 void *cookie,
-			 xlator_t *this,
-			 int32_t op_ret,
-			 int32_t op_errno,
-                         struct iatt *preparent,
-                         struct iatt *postparent)
-{
-	int32_t callcnt = 0;
-	unify_local_t *local = frame->local;
-	call_frame_t *prev_frame = cookie;
-  
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"child(%s): path(%s -> %s): %s", 
-			prev_frame->this->name, 
-			local->loc1.path, local->loc2.path, 
-			strerror (op_errno));
-      
-	}
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		local->stbuf.ia_ino = local->ia_ino;
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, local->op_ret, local->op_errno, 
-			      &local->stbuf);
-	}
-	return 0;
-}
-
-int32_t 
-unify_ns_rename_undo_cbk (call_frame_t *frame,
-			  void *cookie,
-			  xlator_t *this,
-			  int32_t op_ret,
-			  int32_t op_errno,
-			  struct iatt *buf,
-                          struct iatt *preoldparent,
-                          struct iatt *postoldparent,
-                          struct iatt *prenewparent,
-                          struct iatt *postnewparent)
-{
-	unify_local_t *local = frame->local;
-
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"namespace: path(%s -> %s): %s", 
-			local->loc1.path, local->loc2.path, 
-			strerror (op_errno));
-	}
-
-	local->stbuf.ia_ino = local->ia_ino;
-	unify_local_wipe (local);
-	STACK_UNWIND (frame, local->op_ret, local->op_errno, &local->stbuf);
-	return 0;
-}
-
-int32_t 
-unify_rename_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  struct iatt *buf,
-                  struct iatt *preoldparent,
-                  struct iatt *postoldparent,
-                  struct iatt *prenewparent,
-                  struct iatt *postnewparent)
-{
-	int32_t index = 0;
-	int32_t callcnt = 0;
-	int16_t *list = NULL;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-	call_frame_t *prev_frame = cookie;
-  
-	LOCK (&frame->lock);
-	{
-		callcnt = --local->call_count;
-		if (op_ret >= 0) {
-			if (!IA_ISDIR (buf->ia_type))
-				local->stbuf = *buf;
-			local->op_ret = op_ret;
-		} else {
-			gf_log (this->name, GF_LOG_ERROR, 
-				"child(%s): path(%s -> %s): %s", 
-				prev_frame->this->name, 
-				local->loc1.path, local->loc2.path, 
-				strerror (op_errno));
-			local->op_errno = op_errno;
-		}
-	}
-	UNLOCK (&frame->lock);
-
-	if (!callcnt) {
-		local->stbuf.ia_ino = local->ia_ino;
-		if (IA_ISDIR (local->loc1.inode->ia_type)) {
-			unify_local_wipe (local);
-			STACK_UNWIND (frame, local->op_ret, local->op_errno,
-                                      &local->stbuf, &local->oldpreparent,
-                                      &local->oldpostparent, &local->newpreparent,
-                                      &local->newpostparent);
-			return 0;
-		}
-
-		if (local->op_ret == -1) {
-			/* TODO: check this logic */
-
-			/* Rename failed in storage node, successful on NS, 
-			 * hence, rename back the entries in NS */
-			/* NOTE: this will be done only if the destination 
-			 * doesn't exists, if  the destination exists, the 
-			 * job of correcting NS is left to self-heal
-			 */
-			if (!local->index) {
-				loc_t tmp_oldloc = {
-                                        /* its actual 'newloc->path' */
-					.path = local->loc2.path, 
-					.inode = local->loc1.inode,
-					.parent = local->loc2.parent
-				};
-	
-				loc_t tmp_newloc = {
-					/* Actual 'oldloc->path' */
-					.path = local->loc1.path,
-					.parent = local->loc1.parent
-				};
-
-				gf_log (this->name, GF_LOG_ERROR, 
-					"rename succussful on namespace, on "
-					"stroage node failed, reverting back");
-
-				STACK_WIND (frame,
-					    unify_ns_rename_undo_cbk,
-					    NS(this),
-					    NS(this)->fops->rename,
-					    &tmp_oldloc,
-					    &tmp_newloc);
-				return 0;
-			}
-		} else {
-			/* Rename successful on storage nodes */
-
-			int32_t idx = 0;
-			int16_t *tmp_list = NULL;
-			uint64_t tmp_list_int64 = 0;
-			if (local->loc2.inode) {
-				inode_ctx_get (local->loc2.inode, 
-					       this, &tmp_list_int64);
-				list = (int16_t *)(long)tmp_list_int64;
-
-			}
-
-			if (list) {				
-				for (index = 0; list[index] != -1; index++);
-				tmp_list = GF_CALLOC (1, index * 2, 
-                                                      gf_unify_mt_int16_t);
-				memcpy (tmp_list, list, index * 2);
-
-				for (index = 0; list[index] != -1; index++) {
-					/* TODO: Check this logic. */
-					/* If the destination file exists in 
-					 * the same storage node where we sent
-					 * 'rename' call, no need to send 
-					 * unlink 
-					 */
-					for (idx = 0; 
-					     local->list[idx] != -1; idx++) {
-						if (tmp_list[index] == local->list[idx]) {
-							tmp_list[index] = priv->child_count;
-							continue;
-						}
-					}
-	  
-					if (NS(this) != priv->xl_array[tmp_list[index]]) {
-						local->call_count++;
-						callcnt++;
-					}
-				}
-
-				if (local->call_count) {
-					if (callcnt > 1)
-						gf_log (this->name, 
-							GF_LOG_ERROR, 
-							"%s->%s: more (%d) "
-							"subvolumes have the "
-							"newloc entry", 
-							local->loc1.path, 
-							local->loc2.path, 
-							callcnt);
-
-					for (index=0; 
-					     tmp_list[index] != -1; index++) {
-						if (NS(this) != priv->xl_array[tmp_list[index]]) {		    
-							STACK_WIND (frame,
-								    unify_rename_unlink_cbk,
-								    priv->xl_array[tmp_list[index]],
-								    priv->xl_array[tmp_list[index]]->fops->unlink,
-								    &local->loc2);
-							if (!--callcnt)
-								break;
-						}
-					}
-
-					GF_FREE (tmp_list);
-					return 0;
-				}
-				if (tmp_list)
-					GF_FREE (tmp_list);
-			}
-		}
-    
-		/* Need not send 'unlink' to storage node */
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, local->op_ret, 
-			      local->op_errno, &local->stbuf,
-                              &local->oldpreparent, &local->oldpostparent,
-                              &local->newpreparent, &local->newpostparent);
-	}
-
-	return 0;
-}
-
-int32_t 
-unify_ns_rename_cbk (call_frame_t *frame,
-		     void *cookie,
-		     xlator_t *this,
-		     int32_t op_ret,
-		     int32_t op_errno,
-		     struct iatt *buf,
-                     struct iatt *preoldparent,
-                     struct iatt *postoldparent,
-                     struct iatt *prenewparent,
-                     struct iatt *postnewparent)
-{
-	int32_t index = 0;
-	int32_t callcnt = 0;
-	int16_t *list = NULL;
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-
-	if (op_ret == -1) {
-		/* Free local->new_inode */
-		gf_log (this->name, GF_LOG_ERROR, 
-			"namespace: path(%s -> %s): %s", 
-			local->loc1.path, local->loc2.path, 
-			strerror (op_errno));
-
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, op_ret, op_errno, buf,
-                              preoldparent, postoldparent,
-                              prenewparent, postnewparent);
-		return 0;
-	}
-
-	local->stbuf = *buf;
-	local->ia_ino = buf->ia_ino;
-
-        local->oldpreparent = *preoldparent;
-        local->oldpostparent = *postoldparent;
-        local->newpreparent = *prenewparent;
-        local->newpostparent = *postnewparent;
-
-	/* Everything is fine. */
-	if (IA_ISDIR (buf->ia_type)) {
-		local->call_count = priv->child_count;
-		for (index=0; index < priv->child_count; index++) {
-			STACK_WIND (frame,
-				    unify_rename_cbk,
-				    priv->xl_array[index],
-				    priv->xl_array[index]->fops->rename,
-				    &local->loc1,
-				    &local->loc2);
-		}
-
-		return 0;
-	}
-
-	local->call_count = 0;  
-	/* send rename */
-	list = local->list;
-	for (index=0; list[index] != -1; index++) {
-		if (NS(this) != priv->xl_array[list[index]]) {
-			local->call_count++;
-			callcnt++;
-		}
-	}
-
-	if (local->call_count) {
-		for (index=0; list[index] != -1; index++) {
-			if (NS(this) != priv->xl_array[list[index]]) {
-				STACK_WIND (frame,
-					    unify_rename_cbk,
-					    priv->xl_array[list[index]],
-					    priv->xl_array[list[index]]->fops->rename,
-					    &local->loc1,
-					    &local->loc2);
-				if (!--callcnt)
-					break;
-			}
-		}
-	} else {
-		/* file doesn't seem to be present in storage nodes */
-		gf_log (this->name, GF_LOG_CRITICAL,
-			"CRITICAL: source file not in storage node, "
-			"rename successful on namespace :O");
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, -1, EIO, NULL,
-                              NULL, NULL, /* preoldparent, postoldparent */
-                              NULL, NULL); /* prenewparent, postnewparent */
-	}
-	return 0;
-}
-
-
-/**
- * unify_rename - One of the tricky function. The deadliest of all :O
- */
-int32_t
-unify_rename (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *oldloc,
-	      loc_t *newloc)
-{
-	unify_local_t *local = NULL;
-  	uint64_t tmp_list = 0;
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-	loc_copy (&local->loc1, oldloc);
-	loc_copy (&local->loc2, newloc);
-
-	if ((local->loc1.path == NULL) || 
-	    (local->loc2.path == NULL)) {
-		gf_log (this->name, GF_LOG_CRITICAL, "Not enough memory :O");
-		STACK_UNWIND (frame, -1, ENOMEM, NULL,
-                              NULL, NULL, /* preoldparent, postoldparent */
-                              NULL, NULL); /* prenewparent, postnewparent */
-		return 0;
-	}
-  
-	inode_ctx_get (oldloc->inode, this, &tmp_list);
-	local->list = (int16_t *)(long)tmp_list;
-
-	STACK_WIND (frame,
-		    unify_ns_rename_cbk,
-		    NS(this),
-		    NS(this)->fops->rename,
-		    oldloc,
-		    newloc);
-	return 0;
-}
-
-/**
- * unify_link_cbk -
- */
-int32_t
-unify_link_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		inode_t *inode,
-                struct iatt *buf,
-                struct iatt *preparent,
-                struct iatt *postparent)
-{
-	unify_local_t *local = frame->local;
-
-	if (op_ret >= 0) 
-		local->stbuf = *buf;
-	local->stbuf.ia_ino = local->ia_ino;
-
-	unify_local_wipe (local);
-	STACK_UNWIND (frame, op_ret, op_errno, inode, &local->stbuf,
-                      &local->oldpreparent, &local->oldpostparent);
-
-	return 0;
-}
-
-/**
- * unify_ns_link_cbk - 
- */
-int32_t
-unify_ns_link_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-		   inode_t *inode,
-                   struct iatt *buf,
-                   struct iatt *preparent,
-                   struct iatt *postparent)
-{
-	unify_private_t *priv = this->private;
-	unify_local_t *local = frame->local;
-	int16_t *list = local->list;
-	int16_t index = 0;
-
-	if (op_ret == -1) {
-		/* No need to send link request to other servers, 
-		 * as namespace action failed 
-		 */
-		gf_log (this->name, GF_LOG_ERROR, 
-			"namespace: path(%s -> %s): %s", 
-			local->loc1.path, local->loc2.path, 
-			strerror (op_errno));
-		unify_local_wipe (local);
-		STACK_UNWIND (frame, op_ret, op_errno, inode, buf,
-                              preparent, postparent);
-		return 0;
-	}
-
-	/* Update inode for this entry */
-	local->op_ret = 0;
-	local->ia_ino = buf->ia_ino;
-
-        local->oldpreparent = *preparent;
-        local->oldpostparent = *postparent;
-
-	/* Send link request to the node now */
-	for (index = 0; list[index] != -1; index++) {
-		char need_break = (list[index+1] == -1);
-		if (priv->xl_array[list[index]] != NS (this)) {
-			STACK_WIND (frame,
-				    unify_link_cbk,
-				    priv->xl_array[list[index]],
-				    priv->xl_array[list[index]]->fops->link,
-				    &local->loc1,
-				    &local->loc2);
-                        break;
-		}
-		if (need_break)
-			break;
-	}
-
-	return 0;
-}
-
-/**
- * unify_link - 
- */
-int32_t
-unify_link (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *oldloc,
-	    loc_t *newloc)
-{
-	unify_local_t *local = NULL;
-  	uint64_t tmp_list = 0;
-
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (oldloc);
-	UNIFY_CHECK_INODE_CTX_AND_UNWIND_ON_ERR (newloc);
-
-	/* Initialization */
-	INIT_LOCAL (frame, local);
-
-	loc_copy (&local->loc1, oldloc);
-	loc_copy (&local->loc2, newloc);
-
-	inode_ctx_get (oldloc->inode, this, &tmp_list);
-	local->list = (int16_t *)(long)tmp_list;
-
-	STACK_WIND (frame,
-		    unify_ns_link_cbk,
-		    NS(this),
-		    NS(this)->fops->link,
-		    oldloc,
-		    newloc);
-
-	return 0;
-}
-
-
-/**
- * unify_checksum_cbk - 
- */
-int32_t
-unify_checksum_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    uint8_t *fchecksum,
-		    uint8_t *dchecksum)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, fchecksum, dchecksum);
-
-	return 0;
-}
-
-/**
- * unify_checksum - 
- */
-int32_t
-unify_checksum (call_frame_t *frame,
-		xlator_t *this,
-		loc_t *loc,
-		int32_t flag)
-{
-	STACK_WIND (frame,
-		    unify_checksum_cbk,
-		    NS(this),
-		    NS(this)->fops->checksum,
-		    loc,
-		    flag);
-
-	return 0;
-}
-
-
-/**
- * unify_finodelk_cbk - 
- */
-int
-unify_finodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		    int32_t op_ret, int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-/**
- * unify_finodelk
- */
-int
-unify_finodelk (call_frame_t *frame, xlator_t *this,
-		const char *volume, fd_t *fd, int cmd, struct gf_flock *flock)
-{
-	UNIFY_CHECK_FD_CTX_AND_UNWIND_ON_ERR (fd);
-	xlator_t *child = NULL;
-	uint64_t tmp_child = 0;
-
-	fd_ctx_get (fd, this, &tmp_child);
-	child = (xlator_t *)(long)tmp_child;		     
-
-	STACK_WIND (frame, unify_finodelk_cbk,
-		    child, child->fops->finodelk,
-		    volume, fd, cmd, flock);
-
-	return 0;
-}
-
-
-
-/**
- * unify_fentrylk_cbk - 
- */
-int
-unify_fentrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		    int32_t op_ret, int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-/**
- * unify_fentrylk
- */
-int
-unify_fentrylk (call_frame_t *frame, xlator_t *this,
-		const char *volume, fd_t *fd, const char *basename,
-		entrylk_cmd cmd, entrylk_type type)
-		
-{
-	UNIFY_CHECK_FD_CTX_AND_UNWIND_ON_ERR (fd);
-	xlator_t *child = NULL;
-	uint64_t tmp_child = 0;
-
-	fd_ctx_get (fd, this, &tmp_child);
-	child = (xlator_t *)(long)tmp_child;		     
-
-	STACK_WIND (frame, unify_fentrylk_cbk,
-		    child, child->fops->fentrylk,
-		    volume, fd, basename, cmd, type);
-
-	return 0;
-}
-
-
-
-/**
- * unify_fxattrop_cbk - 
- */
-int
-unify_fxattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		    int32_t op_ret, int32_t op_errno, dict_t *xattr)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, xattr);
-	return 0;
-}
-
-/**
- * unify_fxattrop
- */
-int
-unify_fxattrop (call_frame_t *frame, xlator_t *this,
-		fd_t *fd, gf_xattrop_flags_t optype, dict_t *xattr)
-{
-	UNIFY_CHECK_FD_CTX_AND_UNWIND_ON_ERR (fd);
-	xlator_t *child = NULL;
-	uint64_t tmp_child = 0;
-
-	fd_ctx_get (fd, this, &tmp_child);
-	child = (xlator_t *)(long)tmp_child;		     
-
-	STACK_WIND (frame, unify_fxattrop_cbk,
-		    child, child->fops->fxattrop,
-		    fd, optype, xattr);
-
-	return 0;
-}
-
-
-/**
- * unify_inodelk_cbk - 
- */
-int
-unify_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		   int32_t op_ret, int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-
-/**
- * unify_inodelk
- */
-int
-unify_inodelk (call_frame_t *frame, xlator_t *this,
-	       const char *volume, loc_t *loc, int cmd, struct gf_flock *flock)
-{
-	xlator_t *child = NULL;
-
-	child = unify_loc_subvol (loc, this);
-
-	STACK_WIND (frame, unify_inodelk_cbk,
-		    child, child->fops->inodelk,
-		    volume, loc, cmd, flock);
-
-	return 0;
-}
-
-
-
-/**
- * unify_entrylk_cbk - 
- */
-int
-unify_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		   int32_t op_ret, int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-/**
- * unify_entrylk
- */
-int
-unify_entrylk (call_frame_t *frame, xlator_t *this,
-	       const char *volume, loc_t *loc, const char *basename,
-	       entrylk_cmd cmd, entrylk_type type)
-		
-{
-	xlator_t *child = NULL;
-
-	child = unify_loc_subvol (loc, this);
-
-	STACK_WIND (frame, unify_entrylk_cbk,
-		    child, child->fops->entrylk,
-		    volume, loc, basename, cmd, type);
-
-	return 0;
-}
-
-
-
-/**
- * unify_xattrop_cbk - 
- */
-int
-unify_xattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		   int32_t op_ret, int32_t op_errno, dict_t *xattr)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, xattr);
-	return 0;
-}
-
-/**
- * unify_xattrop
- */
-int
-unify_xattrop (call_frame_t *frame, xlator_t *this,
-		loc_t *loc, gf_xattrop_flags_t optype, dict_t *xattr)
-{
-	xlator_t *child = NULL;
-
-	child = unify_loc_subvol (loc, this);
-
-	STACK_WIND (frame, unify_xattrop_cbk,
-		    child, child->fops->xattrop,
-		    loc, optype, xattr);
-
-	return 0;
-}
-
-int
-unify_forget (xlator_t *this,
-              inode_t *inode)
-{
-        int16_t *list = NULL;
-        uint64_t tmp_list = 0;
-
-        if (inode->ia_type && (!IA_ISDIR(inode->ia_type))) {
-                inode_ctx_get (inode, this, &tmp_list);
-                if (tmp_list) {
-                        list = (int16_t *)(long)tmp_list;
-                        GF_FREE (list);
-                }
-        }
-
-        return 0;
-}
-
-/**
- * notify
- */
-int32_t
-notify (xlator_t *this,
-        int32_t event,
-        void *data,
-        ...)
-{
-	unify_private_t *priv = this->private;
-	struct sched_ops *sched = NULL;
-
-	if (!priv) {
-		return 0;
-	}
-
-	sched = priv->sched_ops;    
-	if (!sched) {
-		gf_log (this->name, GF_LOG_CRITICAL, "No scheduler :O");
-		raise (SIGTERM);
-		return 0;
-	}
-	if (priv->namespace == data) {
-		if (event == GF_EVENT_CHILD_UP) {
-			sched->notify (this, event, data);
-		}
-		return 0;
-	}
-
-	switch (event)
-	{
-	case GF_EVENT_CHILD_UP:
-	{
-		/* Call scheduler's update () to enable it for scheduling */
-		sched->notify (this, event, data);
-	
-		LOCK (&priv->lock);
-		{
-			/* Increment the inode's generation, which is 
-			   used for self_heal */
-			++priv->inode_generation;
-			++priv->num_child_up;
-		}
-		UNLOCK (&priv->lock);
-
-		if (!priv->is_up) {
-			default_notify (this, event, data);
-			priv->is_up = 1;
-		}
-	}
-	break;
-	case GF_EVENT_CHILD_DOWN:
-	{
-		/* Call scheduler's update () to disable the child node 
-		 * for scheduling
-		 */
-		sched->notify (this, event, data);
-		LOCK (&priv->lock);
-		{
-			--priv->num_child_up;
-		}
-		UNLOCK (&priv->lock);
-
-		if (priv->num_child_up == 0) {
-			/* Send CHILD_DOWN to upper layer */
-			default_notify (this, event, data);
-			priv->is_up = 0;
-		}
-	}
-	break;
-
-	default:
-	{
-		default_notify (this, event, data);
-	}
-	break;
-	}
-
-	return 0;
-}
-
-int32_t
-mem_acct_init (xlator_t *this)
-{
-        int     ret = -1;
-
-        if (!this)
-                return ret;
-
-        ret = xlator_mem_acct_init (this, gf_unify_mt_end + 1);
-        
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
-                                "failed");
-                return ret;
-        }
-
-        return ret;
-}
-
-/** 
- * init - This function is called first in the xlator, while initializing.
- *   All the config file options are checked and appropriate flags are set.
- *
- * @this - 
- */
-int32_t 
-init (xlator_t *this)
-{
-	int32_t          ret       = 0;
-	int32_t          count     = 0;
-	data_t          *scheduler = NULL;
-	data_t          *data      = NULL;
-	xlator_t        *ns_xl     = NULL;
-	xlator_list_t   *trav      = NULL;
-	xlator_list_t   *xlparent  = NULL;
-	xlator_list_t   *parent    = NULL;
-	unify_private_t *_private  = NULL; 
-
-
-	/* Check for number of child nodes, if there is no child nodes, exit */
-	if (!this->children) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"No child nodes specified. check \"subvolumes \" "
-			"option in volfile");
-		return -1;
-	}
-
-  	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"dangling volume. check volfile ");
-	}
-  
-	/* Check for 'scheduler' in volume */
-	scheduler = dict_get (this->options, "scheduler");
-	if (!scheduler) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"\"option scheduler <x>\" is missing in volfile");
-		return -1;
-	}
-
-	/* Setting "option namespace <node>" */
-	data = dict_get (this->options, "namespace");
-	if(!data) {
-		gf_log (this->name, GF_LOG_CRITICAL, 
-			"namespace option not specified, Exiting");
-		return -1;
-	}
-	/* Search namespace in the child node, if found, exit */
-	trav = this->children;
-	while (trav) {
-		if (strcmp (trav->xlator->name, data->data) == 0)
-			break;
-		trav = trav->next;
-	}
-	if (trav) {
-		gf_log (this->name, GF_LOG_CRITICAL, 
-			"namespace node used as a subvolume, Exiting");
-		return -1;
-	}
-	
-	/* Search for the namespace node, if found, continue */
-	ns_xl = this->next;
-	while (ns_xl) {
-		if (strcmp (ns_xl->name, data->data) == 0)
-			break;
-		ns_xl = ns_xl->next;
-	}
-	if (!ns_xl) {
-		gf_log (this->name, GF_LOG_CRITICAL, 
-			"namespace node not found in volfile, Exiting");
-		return -1;
-	}
-	
-	gf_log (this->name, GF_LOG_DEBUG, 
-		"namespace node specified as %s", data->data);
-	
-	_private = GF_CALLOC (1, sizeof (*_private), 
-                              gf_unify_mt_unify_private_t);
-	ERR_ABORT (_private);
-	_private->sched_ops = get_scheduler (this, scheduler->data);
-	if (!_private->sched_ops) {
-		gf_log (this->name, GF_LOG_CRITICAL, 
-			"Error while loading scheduler. Exiting");
-		GF_FREE (_private);
-		return -1;
-	}
-	
-	if (ns_xl->parents) {
-		gf_log (this->name, GF_LOG_CRITICAL,
-			"Namespace node should not be a child of any other node. Exiting");
-		GF_FREE (_private);
-		return -1;
-	}
-
-	_private->namespace = ns_xl;
-	
-	/* update _private structure */
-	{
-		count = 0;
-		trav = this->children;
-		/* Get the number of child count */
-		while (trav) {
-			count++;
-			trav = trav->next;
-		}
-		
-		gf_log (this->name, GF_LOG_DEBUG, 
-			"Child node count is %d", count);    
-
-		_private->child_count = count;
-		if (count == 1) {
-			/* TODO: Should I error out here? */
-			gf_log (this->name, GF_LOG_CRITICAL, 
-				"WARNING: You have defined only one "
-				"\"subvolumes\" for unify volume. It may not "
-				"be the desired config, review your volume "
-				"volfile. If this is how you are testing it,"
-				" you may hit some performance penalty");
-		}
-		
-		_private->xl_array = GF_CALLOC (1, 
-					        sizeof (xlator_t) * (count + 1),
-                                                gf_unify_mt_xlator_t);
-		ERR_ABORT (_private->xl_array);
-		
-		count = 0;
-		trav = this->children;
-		while (trav) {
-			_private->xl_array[count++] = trav->xlator;
-			trav = trav->next;
-		}
-		_private->xl_array[count] = _private->namespace;
-		
-		/* self-heal part, start with generation '1' */
-		_private->inode_generation = 1; 
-                /* Because, Foreground part is tested well */
-		_private->self_heal = ZR_UNIFY_FG_SELF_HEAL; 
-		data = dict_get (this->options, "self-heal");
-		if (data) {
-			if (strcasecmp (data->data, "off") == 0) 
-				_private->self_heal = ZR_UNIFY_SELF_HEAL_OFF;
-
-			if (strcasecmp (data->data, "foreground") == 0)
-				_private->self_heal = ZR_UNIFY_FG_SELF_HEAL;
-
-			if (strcasecmp (data->data, "background") == 0)
-				_private->self_heal = ZR_UNIFY_BG_SELF_HEAL;
-		}
-    
-		/* optimist - ask bulde for more about it */
-		data = dict_get (this->options, "optimist");
-		if (data) {
-			if (gf_string2boolean (data->data, 
-					       &_private->optimist) == -1) {
-				gf_log (this->name, GF_LOG_ERROR, 
-					"optimist excepts only boolean "
-					"options");
-			}
-		}
-
-		LOCK_INIT (&_private->lock);
-	}
-
-	/* Now that everything is fine. */
-	this->private = (void *)_private;
-	{
-                ret = _private->sched_ops->mem_acct_init (this);
-
-                if (ret == -1) {
-                        return -1;
-                }
-
-		/* Initialize scheduler, if everything else is successful */
-		ret = _private->sched_ops->init (this); 
-		if (ret == -1) {
-			gf_log (this->name, GF_LOG_CRITICAL,
-				"Initializing scheduler failed, Exiting");
-			GF_FREE (_private);
-			return -1;
-		}
-
-
-		ret = 0;
-
-		/* This section is required because some fops may look 
-		 * for 'xl->parent' variable 
-		 */
-		xlparent = GF_CALLOC (1, sizeof (*xlparent),
-                                      gf_unify_mt_xlator_list_t);
-		xlparent->xlator = this;
-		if (!ns_xl->parents) {
-			ns_xl->parents = xlparent;
-		} else {
-			parent = ns_xl->parents;
-			while (parent->next)
-				parent = parent->next;
-			parent->next = xlparent;
-		}
-	}
-
-	/* Tell namespace node that init is done */
-	xlator_notify (ns_xl, GF_EVENT_PARENT_UP, this);
-
-	return 0;
-}
-
-/** 
- * fini  - Free all the allocated memory 
- */
-void
-fini (xlator_t *this)
-{
-	unify_private_t *priv = this->private;
-	priv->sched_ops->fini (this);
-	this->private = NULL;
-	LOCK_DESTROY (&priv->lock);
-	GF_FREE (priv->xl_array);
-	GF_FREE (priv);
-	return;
-}
-
-
-struct xlator_fops fops = {
-	.stat        = unify_stat,
-	.readlink    = unify_readlink,
-	.mknod       = unify_mknod,
-	.mkdir       = unify_mkdir,
-	.unlink      = unify_unlink,
-	.rmdir       = unify_rmdir,
-	.symlink     = unify_symlink,
-	.rename      = unify_rename,
-	.link        = unify_link,
-	.truncate    = unify_truncate,
-	.create      = unify_create,
-	.open        = unify_open,
-	.readv       = unify_readv,
-	.writev      = unify_writev,
-	.statfs      = unify_statfs,
-	.flush       = unify_flush,
-	.fsync       = unify_fsync,
-	.setxattr    = unify_setxattr,
-	.getxattr    = unify_getxattr,
-	.removexattr = unify_removexattr,
-	.opendir     = unify_opendir,
-	.readdir     = unify_readdir,
-	.readdirp    = unify_readdirp,
-	.fsyncdir    = unify_fsyncdir,
-	.access      = unify_access,
-	.ftruncate   = unify_ftruncate,
-	.fstat       = unify_fstat,
-	.lk          = unify_lk,
-	.lookup      = unify_lookup,
-	.getdents    = unify_getdents,
-	.checksum    = unify_checksum,
-	.inodelk     = unify_inodelk,
-	.finodelk    = unify_finodelk,
-	.entrylk     = unify_entrylk,
-	.fentrylk    = unify_fentrylk,
-	.xattrop     = unify_xattrop,
-	.fxattrop    = unify_fxattrop,
-        .setattr     = unify_setattr, 
-        .fsetattr    = unify_fsetattr,
-};
-
-
-struct xlator_cbks cbks = {
-        .forget  = unify_forget,
-};
-
-struct volume_options options[] = {
-	{ .key   = { "namespace" },  
-	  .type  = GF_OPTION_TYPE_XLATOR 
-	},
-	{ .key   = { "scheduler" },  
-	  .value = { "alu", "rr", "random", "nufa", "switch" },
-	  .type  = GF_OPTION_TYPE_STR
-	},
-	{ .key   = {"self-heal"},  
-	  .value = { "foreground", "background", "off" },
-	  .type  = GF_OPTION_TYPE_STR
-	},
-	/* TODO: remove it some time later */
-	{ .key   = {"optimist"},  
-	  .type  = GF_OPTION_TYPE_BOOL 
-	},
-
-	{ .key   = {NULL} },
-};
diff --git a/xlators/cluster/unify/src/unify.h b/xlators/cluster/unify/src/unify.h
deleted file mode 100644
index 18e3db11f71..00000000000
--- a/xlators/cluster/unify/src/unify.h
+++ /dev/null
@@ -1,146 +0,0 @@
-/*
-  Copyright (c) 2006-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#ifndef _UNIFY_H
-#define _UNIFY_H
-
-#include "scheduler.h"
-#include "list.h"
-#include "unify-mem-types.h"
-
-#define MAX_DIR_ENTRY_STRING     (32 * 1024)
-
-#define ZR_UNIFY_SELF_HEAL_OFF 0
-#define ZR_UNIFY_FG_SELF_HEAL  1
-#define ZR_UNIFY_BG_SELF_HEAL  2
-
-/* Sometimes one should use completely random numbers.. its good :p */
-#define UNIFY_SELF_HEAL_GETDENTS_COUNT 512
-
-#define NS(xl)          (((unify_private_t *)xl->private)->namespace)
-
-/* This is used to allocate memory for local structure */
-#define INIT_LOCAL(fr, loc)                   \
-do {                                          \
-  loc = GF_CALLOC (1, sizeof (unify_local_t), gf_unify_mt_unify_local_t);   \
-  ERR_ABORT (loc);			      \
-  if (!loc) {                                 \
-    STACK_UNWIND (fr, -1, ENOMEM);            \
-    return 0;                                 \
-  }                                           \
-  fr->local = loc;                            \
-  loc->op_ret = -1;                           \
-  loc->op_errno = ENOENT;                     \
-} while (0)
-
-
-
-struct unify_private {
-	/* Update this structure depending on requirement */
-	void *scheduler;               /* THIS SHOULD BE THE FIRST VARIABLE, 
-					  if xlator is using scheduler */
-	struct sched_ops *sched_ops;   /* Scheduler options  */
-	xlator_t *namespace;           /* ptr to namespace xlator */
-	xlator_t **xl_array;
-	gf_boolean_t optimist;
-	int16_t child_count;
-	int16_t num_child_up;
-	uint8_t self_heal;
-	uint8_t is_up;
-	uint64_t inode_generation;
-	gf_lock_t lock;
-};
-typedef struct unify_private unify_private_t;
-
-struct unify_self_heal_struct {
-	uint8_t dir_checksum[NAME_MAX];
-	uint8_t ns_dir_checksum[NAME_MAX];
-	uint8_t file_checksum[NAME_MAX];
-	uint8_t ns_file_checksum[NAME_MAX];
-	off_t *offset_list;
-	int   *count_list;
-	dir_entry_t **entry_list;
-};
-
-
-struct _unify_local_t {
-	int32_t call_count;
-	int32_t op_ret;
-	int32_t op_errno;
-	mode_t mode;
-	off_t offset;
-	dev_t dev;
-	uid_t uid;
-	gid_t gid;
-	int32_t flags;
-	int32_t entry_count;
-	int32_t count;    // dir_entry_t count;
-	fd_t *fd;
-	struct iatt stbuf;
-        struct iatt stpre;
-        struct iatt stpost;
-	struct statvfs statvfs_buf;
-	struct timespec tv[2];
-	char *name;
-	int32_t revalidate;
-
-	ino_t ia_ino;
-	nlink_t ia_nlink;
-  
-	dict_t *dict;
-
-	int16_t *list;
-	int16_t *new_list; /* Used only in case of rename */
-	int16_t index;
-
-	int32_t failed;
-	int32_t return_eio;  /* Used in case of different st-mode 
-				present for a given path */
-
-	uint64_t inode_generation; /* used to store the per directory 
-				    * inode_generation. Got from inode's ctx 
-				    * of directory inodes
-				    */
-
-	struct unify_self_heal_struct *sh_struct;
-	loc_t loc1, loc2;
-
-        struct iatt poststbuf;
-        /* When not used for rename, old*
-         * are used as the attrs for the current
-         * parent directory.
-         */
-        struct iatt oldpreparent;
-        struct iatt oldpostparent;
-        struct iatt newpreparent;
-        struct iatt newpostparent;
-        int32_t wbflags;
-};
-typedef struct _unify_local_t unify_local_t;
-
-int32_t zr_unify_self_heal (call_frame_t *frame,
-			    xlator_t *this,
-			    unify_local_t *local);
-
-#endif /* _UNIFY_H */
diff --git a/xlators/debug/Makefile.am b/xlators/debug/Makefile.am
index b655554efec..88fac1c6d9e 100644
--- a/xlators/debug/Makefile.am
+++ b/xlators/debug/Makefile.am
@@ -1,3 +1,3 @@
-SUBDIRS = trace error-gen io-stats
+SUBDIRS = error-gen io-stats sink trace delay-gen
 
 CLEANFILES = 
diff --git a/xlators/features/access-control/Makefile.am b/xlators/debug/delay-gen/Makefile.am
index a985f42a877..a985f42a877 100644
--- a/xlators/features/access-control/Makefile.am
+++ b/xlators/debug/delay-gen/Makefile.am
diff --git a/xlators/debug/delay-gen/src/Makefile.am b/xlators/debug/delay-gen/src/Makefile.am
new file mode 100644
index 00000000000..8f758dec199
--- /dev/null
+++ b/xlators/debug/delay-gen/src/Makefile.am
@@ -0,0 +1,11 @@
+
+xlator_LTLIBRARIES = delay-gen.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/debug
+delay_gen_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+delay_gen_la_SOURCES = delay-gen.c
+delay_gen_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+noinst_HEADERS = delay-gen.h delay-gen-mem-types.h delay-gen-messages.h
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	   -I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+AM_CFLAGS = -Wall -fno-strict-aliasing $(GF_CFLAGS)
+CLEANFILES =
diff --git a/xlators/debug/delay-gen/src/delay-gen-mem-types.h b/xlators/debug/delay-gen/src/delay-gen-mem-types.h
new file mode 100644
index 00000000000..c89a9217193
--- /dev/null
+++ b/xlators/debug/delay-gen/src/delay-gen-mem-types.h
@@ -0,0 +1,21 @@
+/*
+ *   Copyright (c) 2017 Red Hat, Inc. <http://www.redhat.com>
+ *   This file is part of GlusterFS.
+ *
+ *   This file is licensed to you under your choice of the GNU Lesser
+ *   General Public License, version 3 or any later version (LGPLv3 or
+ *   later), or the GNU General Public License, version 2 (GPLv2), in all
+ *   cases as published by the Free Software Foundation.
+ */
+
+#ifndef __DELAY_GEN_MEM_TYPES_H__
+#define __DELAY_GEN_MEM_TYPES_H__
+
+#include <glusterfs/mem-types.h>
+
+enum gf_delay_gen_mem_types_ {
+    gf_delay_gen_mt_dg_t = gf_common_mt_end + 1,
+    gf_delay_gen_mt_end
+};
+
+#endif /* __DELAY_GEN_MEM_TYPES_H__ */
diff --git a/xlators/debug/delay-gen/src/delay-gen-messages.h b/xlators/debug/delay-gen/src/delay-gen-messages.h
new file mode 100644
index 00000000000..bc98cec2885
--- /dev/null
+++ b/xlators/debug/delay-gen/src/delay-gen-messages.h
@@ -0,0 +1,26 @@
+/*
+ *   Copyright (c) 2017 Red Hat, Inc. <http://www.redhat.com>
+ *   This file is part of GlusterFS.
+ *
+ *   This file is licensed to you under your choice of the GNU Lesser
+ *   General Public License, version 3 or any later version (LGPLv3 or
+ *   later), or the GNU General Public License, version 2 (GPLv2), in all
+ *   cases as published by the Free Software Foundation.
+ */
+
+#ifndef __DELAY_GEN_MESSAGES_H__
+#define __DELAY_GEN_MESSAGES_H__
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+#endif /* __DELAY_GEN_MESSAGES_H__ */
diff --git a/xlators/debug/delay-gen/src/delay-gen.c b/xlators/debug/delay-gen/src/delay-gen.c
new file mode 100644
index 00000000000..4698f1fd785
--- /dev/null
+++ b/xlators/debug/delay-gen/src/delay-gen.c
@@ -0,0 +1,697 @@
+/*
+ *  Copyright (c) 2017 Red Hat, Inc. <http://www.redhat.com>
+ *  This file is part of GlusterFS.
+ *
+ *  This file is licensed to you under your choice of the GNU Lesser
+ *  General Public License, version 3 or any later version (LGPLv3 or
+ *  later), or the GNU General Public License, version 2 (GPLv2), in all
+ *  cases as published by the Free Software Foundation.
+ */
+
+#include "delay-gen.h"
+
+#define DELAY_GRANULARITY (1 << 20)
+
+#define DG_FOP(fop, name, frame, this, args...)                                \
+    do {                                                                       \
+        delay_gen(this, fop);                                                  \
+        default_##name(frame, this, args);                                     \
+    } while (0)
+
+int
+delay_gen(xlator_t *this, int fop)
+{
+    dg_t *dg = this->private;
+
+    if (!dg->enable[fop] || !dg->delay_ppm)
+        return 0;
+
+    if ((rand() % DELAY_GRANULARITY) < dg->delay_ppm)
+        gf_nanosleep(dg->delay_duration * GF_US_IN_NS);
+
+    return 0;
+}
+
+int32_t
+dg_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+          dict_t *xdata)
+{
+    DG_FOP(GF_FOP_RENAME, rename, frame, this, oldloc, newloc, xdata);
+    return 0;
+}
+
+int32_t
+dg_ipc(call_frame_t *frame, xlator_t *this, int32_t op, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_IPC, ipc, frame, this, op, xdata);
+    return 0;
+}
+
+int32_t
+dg_setactivelk(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               lock_migration_info_t *locklist, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_SETACTIVELK, setactivelk, frame, this, loc, locklist, xdata);
+    return 0;
+}
+
+int32_t
+dg_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FLUSH, flush, frame, this, fd, xdata);
+    return 0;
+}
+
+int32_t
+dg_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+           off_t off, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_READDIR, readdir, frame, this, fd, size, off, xdata);
+    return 0;
+}
+
+int32_t
+dg_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+            int32_t flags, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_SETXATTR, setxattr, frame, this, loc, dict, flags, xdata);
+    return 0;
+}
+
+int32_t
+dg_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+         dev_t rdev, mode_t umask, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_MKNOD, mknod, frame, this, loc, mode, rdev, umask, xdata);
+    return 0;
+}
+
+int32_t
+dg_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+             int32_t flags, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FSETXATTR, fsetxattr, frame, this, fd, dict, flags, xdata);
+    return 0;
+}
+
+int32_t
+dg_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+         off_t offset, uint32_t flags, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_READ, readv, frame, this, fd, size, offset, flags, xdata);
+    return 0;
+}
+
+int32_t
+dg_inodelk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+           int32_t cmd, struct gf_flock *lock, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_INODELK, inodelk, frame, this, volume, loc, cmd, lock, xdata);
+    return 0;
+}
+
+int32_t
+dg_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+                dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FREMOVEXATTR, fremovexattr, frame, this, fd, name, xdata);
+    return 0;
+}
+
+int32_t
+dg_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+        fd_t *fd, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_OPEN, open, frame, this, loc, flags, fd, xdata);
+    return 0;
+}
+
+int32_t
+dg_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+           gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_XATTROP, xattrop, frame, this, loc, flags, dict, xdata);
+    return 0;
+}
+
+int32_t
+dg_entrylk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+           const char *basename, entrylk_cmd cmd, entrylk_type type,
+           dict_t *xdata)
+{
+    DG_FOP(GF_FOP_ENTRYLK, entrylk, frame, this, volume, loc, basename, cmd,
+           type, xdata);
+    return 0;
+}
+
+int32_t
+dg_getactivelk(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_GETACTIVELK, getactivelk, frame, this, loc, xdata);
+    return 0;
+}
+
+int32_t
+dg_finodelk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+            int32_t cmd, struct gf_flock *lock, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FINODELK, finodelk, frame, this, volume, fd, cmd, lock,
+           xdata);
+    return 0;
+}
+
+int32_t
+dg_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+          mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_CREATE, create, frame, this, loc, flags, mode, umask, fd,
+           xdata);
+    return 0;
+}
+
+int32_t
+dg_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+           size_t len, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_DISCARD, discard, frame, this, fd, offset, len, xdata);
+    return 0;
+}
+
+int32_t
+dg_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+         mode_t umask, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_MKDIR, mkdir, frame, this, loc, mode, umask, xdata);
+    return 0;
+}
+
+int32_t
+dg_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+      struct gf_flock *lock, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_LK, lk, frame, this, fd, cmd, lock, xdata);
+    return 0;
+}
+
+int32_t
+dg_writev(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
+          int32_t count, off_t off, uint32_t flags, struct iobref *iobref,
+          dict_t *xdata)
+{
+    DG_FOP(GF_FOP_WRITE, writev, frame, this, fd, vector, count, off, flags,
+           iobref, xdata);
+    return 0;
+}
+
+int32_t
+dg_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+          dict_t *xdata)
+{
+    DG_FOP(GF_FOP_ACCESS, access, frame, this, loc, mask, xdata);
+    return 0;
+}
+
+int32_t
+dg_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_LOOKUP, lookup, frame, this, loc, xdata);
+    return 0;
+}
+
+int32_t
+dg_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+         dict_t *xdata)
+{
+    DG_FOP(GF_FOP_RMDIR, rmdir, frame, this, loc, flags, xdata);
+    return 0;
+}
+
+int32_t
+dg_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t keep_size,
+             off_t offset, size_t len, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FALLOCATE, fallocate, frame, this, fd, keep_size, offset, len,
+           xdata);
+    return 0;
+}
+
+int32_t
+dg_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FSTAT, fstat, frame, this, fd, xdata);
+    return 0;
+}
+
+int32_t
+dg_lease(call_frame_t *frame, xlator_t *this, loc_t *loc,
+         struct gf_lease *lease, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_LEASE, lease, frame, this, loc, lease, xdata);
+    return 0;
+}
+
+int32_t
+dg_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_STAT, stat, frame, this, loc, xdata);
+    return 0;
+}
+
+int32_t
+dg_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+            dict_t *xdata)
+{
+    DG_FOP(GF_FOP_TRUNCATE, truncate, frame, this, loc, offset, xdata);
+    return 0;
+}
+
+int32_t
+dg_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, const char *name,
+            dict_t *xdata)
+{
+    DG_FOP(GF_FOP_GETXATTR, getxattr, frame, this, loc, name, xdata);
+    return 0;
+}
+
+int32_t
+dg_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+           loc_t *loc, mode_t umask, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_SYMLINK, symlink, frame, this, linkpath, loc, umask, xdata);
+    return 0;
+}
+
+int32_t
+dg_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+            off_t len, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_ZEROFILL, zerofill, frame, this, fd, offset, len, xdata);
+    return 0;
+}
+
+int32_t
+dg_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+            dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FSYNCDIR, fsyncdir, frame, this, fd, flags, xdata);
+    return 0;
+}
+
+int32_t
+dg_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+             dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FGETXATTR, fgetxattr, frame, this, fd, name, xdata);
+    return 0;
+}
+
+int32_t
+dg_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t off, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_READDIRP, readdirp, frame, this, fd, size, off, xdata);
+    return 0;
+}
+
+int32_t
+dg_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+        dict_t *xdata)
+{
+    DG_FOP(GF_FOP_LINK, link, frame, this, oldloc, newloc, xdata);
+    return 0;
+}
+
+int32_t
+dg_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+            gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FXATTROP, fxattrop, frame, this, fd, flags, dict, xdata);
+    return 0;
+}
+
+int32_t
+dg_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FTRUNCATE, ftruncate, frame, this, fd, offset, xdata);
+    return 0;
+}
+
+int32_t
+dg_rchecksum(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             int32_t len, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_RCHECKSUM, rchecksum, frame, this, fd, offset, len, xdata);
+    return 0;
+}
+
+int32_t
+dg_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+          dict_t *xdata)
+{
+    DG_FOP(GF_FOP_UNLINK, unlink, frame, this, loc, flags, xdata);
+    return 0;
+}
+
+int32_t
+dg_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+            const char *basename, entrylk_cmd cmd, entrylk_type type,
+            dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FENTRYLK, fentrylk, frame, this, volume, fd, basename, cmd,
+           type, xdata);
+    return 0;
+}
+
+int32_t
+dg_getspec(call_frame_t *frame, xlator_t *this, const char *key, int32_t flags)
+{
+    DG_FOP(GF_FOP_GETSPEC, getspec, frame, this, key, flags);
+    return 0;
+}
+
+int32_t
+dg_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *stbuf,
+           int32_t valid, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_SETATTR, setattr, frame, this, loc, stbuf, valid, xdata);
+    return 0;
+}
+
+int32_t
+dg_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+         dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FSYNC, fsync, frame, this, fd, flags, xdata);
+    return 0;
+}
+
+int32_t
+dg_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_STATFS, statfs, frame, this, loc, xdata);
+    return 0;
+}
+
+int32_t
+dg_seek(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+        gf_seek_what_t what, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_SEEK, seek, frame, this, fd, offset, what, xdata);
+    return 0;
+}
+
+int32_t
+dg_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
+            int32_t valid, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_FSETATTR, fsetattr, frame, this, fd, stbuf, valid, xdata);
+    return 0;
+}
+
+int32_t
+dg_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+           dict_t *xdata)
+{
+    DG_FOP(GF_FOP_OPENDIR, opendir, frame, this, loc, fd, xdata);
+    return 0;
+}
+
+int32_t
+dg_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+            dict_t *xdata)
+{
+    DG_FOP(GF_FOP_READLINK, readlink, frame, this, loc, size, xdata);
+    return 0;
+}
+
+int32_t
+dg_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               const char *name, dict_t *xdata)
+{
+    DG_FOP(GF_FOP_REMOVEXATTR, removexattr, frame, this, loc, name, xdata);
+    return 0;
+}
+
+int32_t
+dg_forget(xlator_t *this, inode_t *inode)
+{
+    return 0;
+}
+
+int32_t
+dg_release(xlator_t *this, fd_t *fd)
+{
+    return 0;
+}
+
+int32_t
+dg_releasedir(xlator_t *this, fd_t *fd)
+{
+    return 0;
+}
+
+static int
+delay_gen_parse_fill_fops(dg_t *dg, char *enable_fops)
+{
+    char *op_no_str = NULL;
+    int op_no = -1;
+    int i = 0;
+    int ret = 0;
+    xlator_t *this = THIS;
+    char *saveptr = NULL;
+    char *dup_enable_fops = NULL;
+
+    if (strlen(enable_fops) == 0) {
+        for (i = GF_FOP_NULL + 1; i < GF_FOP_MAXVALUE; i++)
+            dg->enable[i] = 1;
+    } else {
+        dup_enable_fops = gf_strdup(enable_fops);
+        if (!dup_enable_fops) {
+            ret = -1;
+            goto out;
+        }
+        op_no_str = strtok_r(dup_enable_fops, ",", &saveptr);
+        while (op_no_str) {
+            op_no = gf_fop_int(op_no_str);
+            if (op_no == -1) {
+                gf_log(this->name, GF_LOG_WARNING, "Wrong option value %s",
+                       op_no_str);
+                ret = -1;
+                goto out;
+            } else {
+                dg->enable[op_no] = 1;
+            }
+
+            op_no_str = strtok_r(NULL, ",", &saveptr);
+        }
+    }
+out:
+    GF_FREE(dup_enable_fops);
+    return ret;
+}
+
+void
+delay_gen_set_delay_ppm(dg_t *dg, double percent)
+{
+    double ppm;
+
+    ppm = (percent / 100.0) * (double)DELAY_GRANULARITY;
+    dg->delay_ppm = ppm;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    dg_t *dg = NULL;
+    int32_t ret = 0;
+    double delay_percent = 0;
+    char *delay_enable_fops = NULL;
+
+    if (!this->children || this->children->next) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "delay-gen not configured with one subvolume");
+        ret = -1;
+        goto out;
+    }
+
+    if (!this->parents) {
+        gf_log(this->name, GF_LOG_WARNING, "dangling volume. check volfile ");
+    }
+
+    dg = GF_CALLOC(1, sizeof(*dg), gf_delay_gen_mt_dg_t);
+
+    if (!dg) {
+        ret = -1;
+        goto out;
+    }
+
+    ret = -1;
+
+    GF_OPTION_INIT("delay-percentage", delay_percent, percent, out);
+    GF_OPTION_INIT("enable", delay_enable_fops, str, out);
+    GF_OPTION_INIT("delay-duration", dg->delay_duration, int32, out);
+
+    delay_gen_set_delay_ppm(dg, delay_percent);
+
+    ret = delay_gen_parse_fill_fops(dg, delay_enable_fops);
+    if (ret)
+        goto out;
+
+    this->private = dg;
+
+    ret = 0;
+out:
+    if (ret)
+        GF_FREE(dg);
+    return ret;
+}
+
+void
+fini(xlator_t *this)
+{
+    GF_FREE(this->private);
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    if (!this)
+        return ret;
+
+    ret = xlator_mem_acct_init(this, gf_delay_gen_mt_end + 1);
+
+    if (ret != 0) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Memory accounting init"
+               " failed");
+        return ret;
+    }
+
+    return ret;
+}
+
+int32_t
+reconfigure(xlator_t *this, dict_t *dict)
+{
+    /*At the moment I don't see any need to implement this. In future
+     *if this is needed we can add code here.
+     */
+    return 0;
+}
+
+int
+notify(xlator_t *this, int event, void *data, ...)
+{
+    return default_notify(this, event, data);
+}
+
+struct xlator_fops fops = {
+    .rename = dg_rename,
+    .ipc = dg_ipc,
+    .setactivelk = dg_setactivelk,
+    .flush = dg_flush,
+    .readdir = dg_readdir,
+    .setxattr = dg_setxattr,
+    .mknod = dg_mknod,
+    .fsetxattr = dg_fsetxattr,
+    .readv = dg_readv,
+    .inodelk = dg_inodelk,
+    .fremovexattr = dg_fremovexattr,
+    .open = dg_open,
+    .xattrop = dg_xattrop,
+    .entrylk = dg_entrylk,
+    .getactivelk = dg_getactivelk,
+    .finodelk = dg_finodelk,
+    .create = dg_create,
+    .discard = dg_discard,
+    .mkdir = dg_mkdir,
+    .lk = dg_lk,
+    .writev = dg_writev,
+    .access = dg_access,
+    .lookup = dg_lookup,
+    .rmdir = dg_rmdir,
+    .fallocate = dg_fallocate,
+    .fstat = dg_fstat,
+    .lease = dg_lease,
+    .stat = dg_stat,
+    .truncate = dg_truncate,
+    .getxattr = dg_getxattr,
+    .symlink = dg_symlink,
+    .zerofill = dg_zerofill,
+    .fsyncdir = dg_fsyncdir,
+    .fgetxattr = dg_fgetxattr,
+    .readdirp = dg_readdirp,
+    .link = dg_link,
+    .fxattrop = dg_fxattrop,
+    .ftruncate = dg_ftruncate,
+    .rchecksum = dg_rchecksum,
+    .unlink = dg_unlink,
+    .fentrylk = dg_fentrylk,
+    .getspec = dg_getspec,
+    .setattr = dg_setattr,
+    .fsync = dg_fsync,
+    .statfs = dg_statfs,
+    .seek = dg_seek,
+    .fsetattr = dg_fsetattr,
+    .opendir = dg_opendir,
+    .readlink = dg_readlink,
+    .removexattr = dg_removexattr,
+};
+
+struct xlator_cbks cbks = {
+    .forget = dg_forget,
+    .release = dg_release,
+    .releasedir = dg_releasedir,
+};
+
+struct volume_options options[] = {
+    {
+        .key = {"delay-percentage"},
+        .type = GF_OPTION_TYPE_PERCENT,
+        .default_value = "10%",
+        .description = "Percentage delay of operations when enabled.",
+        .op_version = {GD_OP_VERSION_3_13_0},
+        .flags = OPT_FLAG_SETTABLE,
+        .tags = {"delay-gen"},
+    },
+
+    {
+        .key = {"delay-duration"},
+        .type = GF_OPTION_TYPE_INT,
+        .description = "Delay duration in micro seconds",
+        .default_value = "100000",
+        .op_version = {GD_OP_VERSION_3_13_0},
+        .flags = OPT_FLAG_SETTABLE,
+        .tags = {"delay-gen"},
+    },
+
+    {
+        .key = {"enable"},
+        .type = GF_OPTION_TYPE_STR,
+        .description = "Accepts a string which takes ',' separated fop "
+                       "strings to denote which fops are enabled for delay",
+        .op_version = {GD_OP_VERSION_3_13_0},
+        .flags = OPT_FLAG_SETTABLE,
+        .tags = {"delay-gen"},
+        .default_value = "",
+    },
+
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {GD_OP_VERSION_3_12_0},
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "delay-gen",
+    .category = GF_TECH_PREVIEW,
+};
diff --git a/xlators/debug/delay-gen/src/delay-gen.h b/xlators/debug/delay-gen/src/delay-gen.h
new file mode 100644
index 00000000000..afa95e5eb2d
--- /dev/null
+++ b/xlators/debug/delay-gen/src/delay-gen.h
@@ -0,0 +1,27 @@
+/*
+ *   Copyright (c) 2017 Red Hat, Inc. <http://www.redhat.com>
+ *   This file is part of GlusterFS.
+ *
+ *   This file is licensed to you under your choice of the GNU Lesser
+ *   General Public License, version 3 or any later version (LGPLv3 or
+ *   later), or the GNU General Public License, version 2 (GPLv2), in all
+ *   cases as published by the Free Software Foundation.
+ */
+
+#ifndef __DELAY_GEN_H__
+#define __DELAY_GEN_H__
+
+#include "delay-gen-mem-types.h"
+#include "delay-gen-messages.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+typedef struct {
+    int enable[GF_FOP_MAXVALUE];
+    int op_count;
+    int delay_ppm;
+    int delay_duration;
+} dg_t;
+
+#endif /* __DELAY_GEN_H__ */
diff --git a/xlators/debug/error-gen/src/Makefile.am b/xlators/debug/error-gen/src/Makefile.am
index f353b61e69c..038d2e8e66d 100644
--- a/xlators/debug/error-gen/src/Makefile.am
+++ b/xlators/debug/error-gen/src/Makefile.am
@@ -2,15 +2,17 @@
 xlator_LTLIBRARIES = error-gen.la
 xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/debug
 
-error_gen_la_LDFLAGS = -module -avoidversion
+error_gen_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
 
 error_gen_la_SOURCES = error-gen.c
 error_gen_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
-noinst_HEADERS = error-gen.h
+noinst_HEADERS = error-gen.h error-gen-mem-types.h
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS)\
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
 
 CLEANFILES = 
 
diff --git a/xlators/debug/error-gen/src/error-gen-mem-types.h b/xlators/debug/error-gen/src/error-gen-mem-types.h
new file mode 100644
index 00000000000..b9b713af8fc
--- /dev/null
+++ b/xlators/debug/error-gen/src/error-gen-mem-types.h
@@ -0,0 +1,20 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __ERROR_GEN_MEM_TYPES_H__
+#define __ERROR_GEN_MEM_TYPES_H__
+
+#include <glusterfs/mem-types.h>
+
+enum gf_error_gen_mem_types_ {
+    gf_error_gen_mt_eg_t = gf_common_mt_end + 1,
+    gf_error_gen_mt_end
+};
+#endif
diff --git a/xlators/debug/error-gen/src/error-gen.c b/xlators/debug/error-gen/src/error-gen.c
index 8945de43ee5..d45655ef4c3 100644
--- a/xlators/debug/error-gen/src/error-gen.c
+++ b/xlators/debug/error-gen/src/error-gen.c
@@ -1,2024 +1,1663 @@
 /*
-  Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
-#include "xlator.h"
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include <glusterfs/xlator.h>
 #include "error-gen.h"
+#include <glusterfs/statedump.h>
+#include <glusterfs/defaults.h>
+
+/*
+ * The user can specify an error probability as a float percentage, but we
+ * store it internally as a numerator with this as the denominator.  When it's
+ * used, it's like this:
+ *
+ *    (rand() % FAILURE_GRANULARITY) < error_rate
+ *
+ * To minimize rounding errors from the modulo operation, it's good for this to
+ * be a power of two.
+ *
+ * (BTW this is just the normal case.  If "random-failure" is set, that does
+ * something completely different and this number is irrelevant.  See error_gen
+ * for the legacy code.)
+ */
+#define FAILURE_GRANULARITY (1 << 20)
 
 sys_error_t error_no_list[] = {
-        [GF_FOP_LOOKUP]            = { .error_no_count = 4,
-                                    .error_no = {ENOENT,ENOTDIR,
-                                                 ENAMETOOLONG,EAGAIN}},
-        [GF_FOP_STAT]              = { .error_no_count = 7,
-                                    .error_no = {EACCES,EBADF,EFAULT,
-                                                 ENAMETOOLONG,ENOENT,
-                                                 ENOMEM,ENOTDIR}},
-        [GF_FOP_READLINK]          = { .error_no_count = 8,
-                                    .error_no = {EACCES,EFAULT,EINVAL,EIO,
-                                                 ENAMETOOLONG,ENOENT,ENOMEM,
-                                                 ENOTDIR}},
-        [GF_FOP_MKNOD]             = { .error_no_count = 11,
-                                    .error_no = {EACCES,EEXIST,EFAULT,
-                                                 EINVAL,ENAMETOOLONG,
-                                                 ENOENT,ENOMEM,ENOSPC,
-                                                 ENOTDIR,EPERM,EROFS}},
-        [GF_FOP_MKDIR]             = { .error_no_count = 10,
-                                    .error_no = {EACCES,EEXIST,EFAULT,
-                                                 ENAMETOOLONG,ENOENT,
-                                                 ENOMEM,ENOSPC,ENOTDIR,
-						 EPERM,EROFS}},
-        [GF_FOP_UNLINK]            = { .error_no_count = 10,
-                                    .error_no = {EACCES,EBUSY,EFAULT,EIO,
-                                                 EISDIR,ENAMETOOLONG,
-                                                 ENOENT,ENOMEM,ENOTDIR,
-                                                 EPERM,EROFS}},
-        [GF_FOP_RMDIR]             = { .error_no_count = 8,
-                                    .error_no = {EACCES,EBUSY,EFAULT,
-                                                 ENOMEM,ENOTDIR,ENOTEMPTY,
-                                                 EPERM,EROFS}},
-        [GF_FOP_SYMLINK]           = { .error_no_count = 11,
-                                    .error_no = {EACCES,EEXIST,EFAULT,EIO,
-                                                 ENAMETOOLONG,ENOENT,ENOMEM,
-                                                 ENOSPC,ENOTDIR,EPERM,
-                                                 EROFS}},
-        [GF_FOP_RENAME]            = { .error_no_count = 13,
-                                    .error_no = {EACCES,EBUSY,EFAULT,
-                                                 EINVAL,EISDIR,EMLINK,
-                                                 ENAMETOOLONG,ENOENT,ENOMEM,
-                                                 ENOSPC,ENOTDIR,EEXIST,
-                                                 EXDEV}},
-        [GF_FOP_LINK]              = { .error_no_count = 13,
-                                    .error_no = {EACCES,EFAULT,EEXIST,EIO,
-                                                 EMLINK,ENAMETOOLONG,
-                                                 ENOENT,ENOMEM,ENOSPC,
-                                                 ENOTDIR,EPERM,EROFS,
-                                                 EXDEV}},
-        [GF_FOP_TRUNCATE]          = { .error_no_count = 10,
-                                    .error_no = {EACCES,EFAULT,EFBIG,
-                                                 EINTR,EINVAL,EIO,EISDIR,
-                                                 ENAMETOOLONG,ENOENT,
-                                                 EISDIR}},
-        [GF_FOP_CREATE]            = {.error_no_count = 10,
-                                   .error_no = {EACCES,EEXIST,EFAULT,
-                                                EISDIR,EMFILE,ENAMETOOLONG,
-                                                ENFILE,ENODEV,ENOENT,
-                                                ENODEV}},
-        [GF_FOP_OPEN]              = { .error_no_count = 10,
-                                    .error_no = {EACCES,EEXIST,EFAULT,
-                                                 EISDIR,EMFILE,
-                                                 ENAMETOOLONG,ENFILE,
-                                                 ENODEV,ENOENT,ENOMEM}},
-        [GF_FOP_READ]              = { .error_no_count = 5,
-                                    .error_no = {EINVAL,EBADF,EFAULT,EISDIR,
-                                                 ENAMETOOLONG}},
-        [GF_FOP_WRITE]             = { .error_no_count = 5,
-                                    .error_no = {EINVAL,EBADF,EFAULT,EISDIR,
-                                                 ENAMETOOLONG}},
-        [GF_FOP_STATFS]            = {.error_no_count = 10,
-                                   .error_no = {EACCES,EBADF,EFAULT,EINTR,
-                                                EIO,ENAMETOOLONG,ENOENT,
-                                                ENOMEM,ENOSYS,ENOTDIR}},
-        [GF_FOP_FLUSH]             = { .error_no_count = 5,
-                                    .error_no = {EACCES,EFAULT,
-                                                 ENAMETOOLONG,ENOSYS,
-                                                 ENOENT}},
-        [GF_FOP_FSYNC]             = { .error_no_count = 4,
-                                    .error_no = {EBADF,EIO,EROFS,EINVAL}},
-        [GF_FOP_SETXATTR]          = { .error_no_count = 4,
-                                    .error_no = {EACCES,EBADF,EINTR,
-                                                 ENAMETOOLONG}},
-        [GF_FOP_GETXATTR]          = { .error_no_count = 4,
-                                    .error_no = {EACCES,EBADF,ENAMETOOLONG,
-                                                 EINTR}},
-        [GF_FOP_REMOVEXATTR]       = { .error_no_count = 4,
-                                    .error_no = {EACCES,EBADF,ENAMETOOLONG,
-                                                 EINTR}},
-        [GF_FOP_OPENDIR]           = { .error_no_count = 8,
-                                    .error_no = {EACCES,EEXIST,EFAULT,
-                                                 EISDIR,EMFILE,
-                                                 ENAMETOOLONG,ENFILE,
-                                                 ENODEV}},
-        [GF_FOP_READDIR]           = { .error_no_count = 5,
-                                    .error_no = {EINVAL,EACCES,EBADF,
-                                                 EMFILE,ENOENT}},
-        [GF_FOP_READDIRP]          = { .error_no_count = 5,
-                                    .error_no = {EINVAL,EACCES,EBADF,
-                                                 EMFILE,ENOENT}},
-        [GF_FOP_FSYNCDIR]          = { .error_no_count = 4,
-                                    .error_no = {EBADF,EIO,EROFS,EINVAL}},
-        [GF_FOP_ACCESS]            = { .error_no_count = 8,
-                                    .error_no = {EACCES,ENAMETOOLONG,
-                                                 ENOENT,ENOTDIR,EROFS,
-                                                 EFAULT,EINVAL,EIO}},
-        [GF_FOP_FTRUNCATE]         = { .error_no_count = 9,
-                                    .error_no = {EACCES,EFAULT,EFBIG,
-                                                 EINTR,EINVAL,EIO,EISDIR,
-                                                 ENAMETOOLONG,ENOENT}},
-        [GF_FOP_FSTAT]             = { .error_no_count = 7,
-                                    .error_no = {EACCES,EBADF,EFAULT,
-                                                 ENAMETOOLONG,ENOENT,
-                                                 ENOMEM,ENOTDIR}},
-        [GF_FOP_LK]                = { .error_no_count = 4,
-                                    .error_no = {EACCES,EFAULT,ENOENT,
-                                                 EINTR}},
-        [GF_FOP_XATTROP]           = { .error_no_count = 5,
-                                    .error_no = {EACCES,EFAULT,
-                                                 ENAMETOOLONG,ENOSYS,
-                                                 ENOENT}},
-        [GF_FOP_FXATTROP]          = { .error_no_count = 4,
-                                    .error_no = {EBADF,EIO,EROFS,EINVAL}},
-        [GF_FOP_INODELK]           = { .error_no_count = 4,
-                                    .error_no = {EACCES,EBADF,EINTR,
-                                                 ENAMETOOLONG}},
-        [GF_FOP_FINODELK]          = { .error_no_count = 4,
-                                    .error_no = {EACCES,EBADF,EINTR,
-                                                 ENAMETOOLONG}},
-        [GF_FOP_ENTRYLK]           = { .error_no_count = 4,
-                                    .error_no = {EACCES,EBADF,
-                                                 ENAMETOOLONG,EINTR}},
-        [GF_FOP_FENTRYLK]          = { .error_no_count = 10,
-                                    .error_no = {EACCES,EEXIST,EFAULT,
-                                                 EISDIR,EMFILE,
-                                                 ENAMETOOLONG,ENFILE,
-                                                 ENODEV,ENOENT,ENOMEM}},
-        [GF_FOP_SETATTR]           = {.error_no_count = 11,
-                                    .error_no = {EACCES,EFAULT,EIO,
-                                                 ENAMETOOLONG,ENOENT,
-                                                 ENOMEM,ENOTDIR,EPERM,
-                                                 EROFS,EBADF,EIO}},
-        [GF_FOP_FSETATTR]          = { .error_no_count = 11,
-                                    .error_no = {EACCES,EFAULT,EIO,
-                                                 ENAMETOOLONG,ENOENT,
-                                                 ENOMEM,ENOTDIR,EPERM,
-                                                 EROFS,EBADF,EIO}},
-        [GF_FOP_GETSPEC]           = { .error_no_count = 4,
-                                    .error_no = {EACCES,EBADF,ENAMETOOLONG,
-                                                 EINTR}}
-};
+    [GF_FOP_LOOKUP] = {.error_no_count = 4,
+                       .error_no = {ENOENT, ENOTDIR, ENAMETOOLONG, EAGAIN}},
+    [GF_FOP_STAT] = {.error_no_count = 6,
+                     .error_no = {EACCES, EFAULT, ENAMETOOLONG, ENOENT, ENOMEM,
+                                  ENOTDIR}},
+    [GF_FOP_READLINK] = {.error_no_count = 8,
+                         .error_no = {EACCES, EFAULT, EINVAL, EIO, ENAMETOOLONG,
+                                      ENOENT, ENOMEM, ENOTDIR}},
+    [GF_FOP_MKNOD] = {.error_no_count = 11,
+                      .error_no = {EACCES, EEXIST, EFAULT, EINVAL, ENAMETOOLONG,
+                                   ENOENT, ENOMEM, ENOSPC, ENOTDIR, EPERM,
+                                   EROFS}},
+    [GF_FOP_MKDIR] = {.error_no_count = 10,
+                      .error_no = {EACCES, EEXIST, EFAULT, ENAMETOOLONG, ENOENT,
+                                   ENOMEM, ENOSPC, ENOTDIR, EPERM, EROFS}},
+    [GF_FOP_UNLINK] = {.error_no_count = 10,
+                       .error_no = {EACCES, EBUSY, EFAULT, EIO, EISDIR,
+                                    ENAMETOOLONG, ENOENT, ENOMEM, ENOTDIR,
+                                    EPERM, EROFS}},
+    [GF_FOP_RMDIR] = {.error_no_count = 8,
+                      .error_no = {EACCES, EBUSY, EFAULT, ENOMEM, ENOTDIR,
+                                   ENOTEMPTY, EPERM, EROFS}},
+    [GF_FOP_SYMLINK] = {.error_no_count = 11,
+                        .error_no = {EACCES, EEXIST, EFAULT, EIO, ENAMETOOLONG,
+                                     ENOENT, ENOMEM, ENOSPC, ENOTDIR, EPERM,
+                                     EROFS}},
+    [GF_FOP_RENAME] = {.error_no_count = 13,
+                       .error_no = {EACCES, EBUSY, EFAULT, EINVAL, EISDIR,
+                                    EMLINK, ENAMETOOLONG, ENOENT, ENOMEM,
+                                    ENOSPC, ENOTDIR, EEXIST, EXDEV}},
+    [GF_FOP_LINK] = {.error_no_count = 13,
+                     .error_no = {EACCES, EFAULT, EEXIST, EIO, EMLINK,
+                                  ENAMETOOLONG, ENOENT, ENOMEM, ENOSPC, ENOTDIR,
+                                  EPERM, EROFS, EXDEV}},
+    [GF_FOP_TRUNCATE] = {.error_no_count = 10,
+                         .error_no = {EACCES, EFAULT, EFBIG, EINTR, EINVAL, EIO,
+                                      EISDIR, ENAMETOOLONG, ENOENT, EISDIR}},
+    [GF_FOP_CREATE] = {.error_no_count = 10,
+                       .error_no = {EACCES, EEXIST, EFAULT, EISDIR, EMFILE,
+                                    ENAMETOOLONG, ENFILE, ENODEV, ENOENT,
+                                    ENODEV}},
+    [GF_FOP_OPEN] = {.error_no_count = 10,
+                     .error_no = {EACCES, EEXIST, EFAULT, EISDIR, EMFILE,
+                                  ENAMETOOLONG, ENFILE, ENODEV, ENOENT,
+                                  ENOMEM}},
+    [GF_FOP_READ] = {.error_no_count = 5,
+                     .error_no = {EINVAL, EBADF, EFAULT, EISDIR, ENAMETOOLONG}},
+    [GF_FOP_WRITE] = {.error_no_count = 7,
+                      .error_no = {EINVAL, EBADF, EFAULT, EISDIR, ENAMETOOLONG,
+                                   ENOSPC, GF_ERROR_SHORT_WRITE}},
+    [GF_FOP_STATFS] = {.error_no_count = 9,
+                       .error_no = {EACCES, EFAULT, EINTR, EIO, ENAMETOOLONG,
+                                    ENOENT, ENOMEM, ENOSYS, ENOTDIR}},
+    [GF_FOP_FLUSH] = {.error_no_count = 5,
+                      .error_no = {EACCES, EFAULT, ENAMETOOLONG, ENOSYS,
+                                   ENOENT}},
+    [GF_FOP_FSYNC] = {.error_no_count = 4,
+                      .error_no = {EBADF, EIO, EROFS, EINVAL}},
+    [GF_FOP_SETXATTR] = {.error_no_count = 3,
+                         .error_no = {EACCES, EINTR, ENAMETOOLONG}},
+    [GF_FOP_GETXATTR] = {.error_no_count = 3,
+                         .error_no = {EACCES, ENAMETOOLONG, EINTR}},
+    [GF_FOP_REMOVEXATTR] = {.error_no_count = 3,
+                            .error_no = {EACCES, ENAMETOOLONG, EINTR}},
+    [GF_FOP_FSETXATTR] = {.error_no_count = 4,
+                          .error_no = {EACCES, EBADF, EINTR, ENAMETOOLONG}},
+    [GF_FOP_FGETXATTR] = {.error_no_count = 4,
+                          .error_no = {EACCES, EBADF, ENAMETOOLONG, EINTR}},
+    [GF_FOP_FREMOVEXATTR] = {.error_no_count = 4,
+                             .error_no = {EACCES, EBADF, ENAMETOOLONG, EINTR}},
+    [GF_FOP_OPENDIR] = {.error_no_count = 8,
+                        .error_no = {EACCES, EEXIST, EFAULT, EISDIR, EMFILE,
+                                     ENAMETOOLONG, ENFILE, ENODEV}},
+    [GF_FOP_READDIR] = {.error_no_count = 5,
+                        .error_no = {EINVAL, EACCES, EBADF, EMFILE, ENOENT}},
+    [GF_FOP_READDIRP] = {.error_no_count = 5,
+                         .error_no = {EINVAL, EACCES, EBADF, EMFILE, ENOENT}},
+    [GF_FOP_FSYNCDIR] = {.error_no_count = 4,
+                         .error_no = {EBADF, EIO, EROFS, EINVAL}},
+    [GF_FOP_ACCESS] = {.error_no_count = 8,
+                       .error_no = {EACCES, ENAMETOOLONG, ENOENT, ENOTDIR,
+                                    EROFS, EFAULT, EINVAL, EIO}},
+    [GF_FOP_FTRUNCATE] = {.error_no_count = 9,
+                          .error_no = {EACCES, EFAULT, EFBIG, EINTR, EINVAL,
+                                       EIO, EISDIR, ENAMETOOLONG, ENOENT}},
+    [GF_FOP_FSTAT] = {.error_no_count = 7,
+                      .error_no = {EACCES, EBADF, EFAULT, ENAMETOOLONG, ENOENT,
+                                   ENOMEM, ENOTDIR}},
+    [GF_FOP_LK] = {.error_no_count = 4,
+                   .error_no = {EACCES, EFAULT, ENOENT, EINTR}},
+    [GF_FOP_XATTROP] = {.error_no_count = 5,
+                        .error_no = {EACCES, EFAULT, ENAMETOOLONG, ENOSYS,
+                                     ENOENT}},
+    [GF_FOP_FXATTROP] = {.error_no_count = 4,
+                         .error_no = {EBADF, EIO, EROFS, EINVAL}},
+    [GF_FOP_INODELK] = {.error_no_count = 3,
+                        .error_no = {EACCES, EINTR, ENAMETOOLONG}},
+    [GF_FOP_FINODELK] = {.error_no_count = 4,
+                         .error_no = {EACCES, EBADF, EINTR, ENAMETOOLONG}},
+    [GF_FOP_ENTRYLK] = {.error_no_count = 3,
+                        .error_no = {EACCES, ENAMETOOLONG, EINTR}},
+    [GF_FOP_FENTRYLK] = {.error_no_count = 10,
+                         .error_no = {EACCES, EEXIST, EFAULT, EISDIR, EMFILE,
+                                      ENAMETOOLONG, ENFILE, ENODEV, ENOENT,
+                                      ENOMEM}},
+    [GF_FOP_SETATTR] = {.error_no_count = 10,
+                        .error_no = {EACCES, EFAULT, EIO, ENAMETOOLONG, ENOENT,
+                                     ENOMEM, ENOTDIR, EPERM, EROFS, EIO}},
+    [GF_FOP_FSETATTR] = {.error_no_count = 11,
+                         .error_no = {EACCES, EFAULT, EIO, ENAMETOOLONG, ENOENT,
+                                      ENOMEM, ENOTDIR, EPERM, EROFS, EBADF,
+                                      EIO}},
+    [GF_FOP_GETSPEC] = {.error_no_count = 3,
+                        .error_no = {EACCES, ENAMETOOLONG, EINTR}}};
+
+int
+generate_rand_no(int op_no)
+{
+    int rand_no = 0;
+    int error_no_list_size = 0;
+
+    error_no_list_size = sizeof(error_no_list) / sizeof(error_no_list[0]);
+
+    if (op_no < error_no_list_size)
+        /* coverity[DC.WEAK_CRYPTO] */
+        rand_no = rand() % error_no_list[op_no].error_no_count;
+    return rand_no;
+}
+
+int
+conv_errno_to_int(char **error_no)
+{
+    if (!strcmp((*error_no), "ENOENT"))
+        return ENOENT;
+    else if (!strcmp((*error_no), "ENOTDIR"))
+        return ENOTDIR;
+    else if (!strcmp((*error_no), "ENAMETOOLONG"))
+        return ENAMETOOLONG;
+    else if (!strcmp((*error_no), "EACCES"))
+        return EACCES;
+    else if (!strcmp((*error_no), "EBADF"))
+        return EBADF;
+    else if (!strcmp((*error_no), "EFAULT"))
+        return EFAULT;
+    else if (!strcmp((*error_no), "ENOMEM"))
+        return ENOMEM;
+    else if (!strcmp((*error_no), "EINVAL"))
+        return EINVAL;
+    else if (!strcmp((*error_no), "EIO"))
+        return EIO;
+    else if (!strcmp((*error_no), "EEXIST"))
+        return EEXIST;
+    else if (!strcmp((*error_no), "ENOSPC"))
+        return ENOSPC;
+    else if (!strcmp((*error_no), "EPERM"))
+        return EPERM;
+    else if (!strcmp((*error_no), "EROFS"))
+        return EROFS;
+    else if (!strcmp((*error_no), "EBUSY"))
+        return EBUSY;
+    else if (!strcmp((*error_no), "EISDIR"))
+        return EISDIR;
+    else if (!strcmp((*error_no), "ENOTEMPTY"))
+        return ENOTEMPTY;
+    else if (!strcmp((*error_no), "EMLINK"))
+        return EMLINK;
+    else if (!strcmp((*error_no), "ENODEV"))
+        return ENODEV;
+    else if (!strcmp((*error_no), "EXDEV"))
+        return EXDEV;
+    else if (!strcmp((*error_no), "EMFILE"))
+        return EMFILE;
+    else if (!strcmp((*error_no), "ENFILE"))
+        return ENFILE;
+    else if (!strcmp((*error_no), "ENOSYS"))
+        return ENOSYS;
+    else if (!strcmp((*error_no), "EINTR"))
+        return EINTR;
+    else if (!strcmp((*error_no), "EFBIG"))
+        return EFBIG;
+    else if (!strcmp((*error_no), "GF_ERROR_SHORT_WRITE"))
+        return GF_ERROR_SHORT_WRITE;
+    else
+        return EAGAIN;
+}
+
+int
+error_gen(xlator_t *this, int op_no)
+{
+    eg_t *egp = NULL;
+    int count = 0;
+    int error_no_int = 0;
+    int rand_no = 0;
+    int ret = 0;
+    gf_boolean_t should_err = _gf_false;
+    int error_no_list_size = 0;
+
+    egp = this->private;
+
+    if (egp->random_failure) {
+        /*
+         * I honestly don't know why anyone would use this "feature"
+         * but I'll try to preserve its functionality anyway.  Without
+         * locking twice to update failure_iter_no and egp->op_count
+         * separately, then not locking at all to update
+         * egp->failure_iter_no.  That's not needed for compatibility,
+         * and it's abhorrently wrong.  I have *some* standards.
+         */
+        LOCK(&egp->lock);
+        {
+            count = ++(egp->op_count);
+            error_no_int = egp->error_no_int;
+            if ((count % egp->failure_iter_no) == 0) {
+                egp->op_count = 0;
+                /* coverity[DC.WEAK_CRYPTO] */
+                egp->failure_iter_no = 3 + (rand() % GF_UNIVERSAL_ANSWER);
+                should_err = _gf_true;
+            }
+        }
+        UNLOCK(&egp->lock);
+    } else {
+        /*
+         * It turns out that rand() is almost universally implemented
+         * as a linear congruential PRNG, which is about as cheap as
+         * it gets.  This gets us real random behavior, including
+         * phenomena like streaks and dry spells, with controllable
+         * long-term probability, cheaply.
+         */
+        if ((rand() % FAILURE_GRANULARITY) < egp->failure_iter_no) {
+            should_err = _gf_true;
+        }
+    }
+
+    error_no_list_size = sizeof(error_no_list) / sizeof(error_no_list[0]);
+    if (should_err) {
+        if (error_no_int)
+            ret = error_no_int;
+        else {
+            rand_no = generate_rand_no(op_no);
+            if (op_no >= error_no_list_size)
+                op_no = 0;
+            if (rand_no >= error_no_list[op_no].error_no_count)
+                rand_no = 0;
+            ret = error_no_list[op_no].error_no[rand_no];
+        }
+    }
+
+    return ret;
+}
 
 int
-generate_rand_no (int op_no)
+error_gen_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-        int             rand_no = 0;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
+
+    egp = this->private;
+    enable = egp->enable[GF_FOP_LOOKUP];
+
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_LOOKUP);
+
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(lookup, frame, -1, op_errno, NULL, NULL, NULL,
+                            NULL);
+        return 0;
+    }
 
-        if (op_no < GF_FOP_MAXVALUE)
-                rand_no = rand () % error_no_list[op_no].error_no_count;
-        return rand_no;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->lookup,
+                    loc, xdata);
+    return 0;
 }
 
 int
-conv_errno_to_int (char **error_no)
+error_gen_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-        if (!strcmp ((*error_no), "ENOENT"))
-                return ENOENT;
-        else if (!strcmp ((*error_no), "ENOTDIR"))
-                return ENOTDIR;
-        else if (!strcmp ((*error_no), "ENAMETOOLONG"))
-                return ENAMETOOLONG;
-        else if (!strcmp ((*error_no), "EACCES"))
-                return EACCES;
-        else if (!strcmp ((*error_no), "EBADF"))
-                return EBADF;
-        else if (!strcmp ((*error_no), "EFAULT"))
-                return EFAULT;
-        else if (!strcmp ((*error_no), "ENOMEM"))
-                return ENOMEM;
-        else if (!strcmp ((*error_no), "EINVAL"))
-                return EINVAL;
-        else if (!strcmp ((*error_no), "EIO"))
-                return EIO;
-        else if (!strcmp ((*error_no), "EEXIST"))
-                return EEXIST;
-        else if (!strcmp ((*error_no), "ENOSPC"))
-                return ENOSPC;
-        else if (!strcmp ((*error_no), "EPERM"))
-                return EPERM;
-        else if (!strcmp ((*error_no), "EROFS"))
-                return EROFS;
-        else if (!strcmp ((*error_no), "EBUSY"))
-                return EBUSY;
-        else if (!strcmp ((*error_no), "EISDIR"))
-                return EISDIR;
-        else if (!strcmp ((*error_no), "ENOTEMPTY"))
-                return ENOTEMPTY;
-        else if (!strcmp ((*error_no), "EMLINK"))
-                return EMLINK;
-        else if (!strcmp ((*error_no), "ENODEV"))
-                return ENODEV;
-        else if (!strcmp ((*error_no), "EXDEV"))
-                return EXDEV;
-        else if (!strcmp ((*error_no), "EMFILE"))
-                return EMFILE;
-        else if (!strcmp ((*error_no), "ENFILE"))
-                return ENFILE;
-        else if (!strcmp ((*error_no), "ENOSYS"))
-                return ENOSYS;
-        else if (!strcmp ((*error_no), "EINTR"))
-                return EINTR;
-        else if (!strcmp ((*error_no), "EFBIG"))
-                return EFBIG;
-        else
-                return EAGAIN;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
+
+    egp = this->private;
+    enable = egp->enable[GF_FOP_STAT];
+
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_STAT);
+
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(stat, frame, -1, op_errno, NULL, xdata);
+        return 0;
+    }
+
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->stat,
+                    loc, xdata);
+    return 0;
 }
 
 int
-get_fop_int (char **op_no_str)
+error_gen_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                  struct iatt *stbuf, int32_t valid, dict_t *xdata)
 {
-        if (!strcmp ((*op_no_str), "lookup"))
-                return GF_FOP_LOOKUP;
-        else if (!strcmp ((*op_no_str), "stat"))
-                return GF_FOP_STAT;
-        else if (!strcmp ((*op_no_str), "readlink"))
-                return GF_FOP_READLINK;
-        else if (!strcmp ((*op_no_str), "mknod"))
-                return GF_FOP_MKNOD;
-        else if (!strcmp ((*op_no_str), "mkdir"))
-                return GF_FOP_MKDIR;
-        else if (!strcmp ((*op_no_str), "unlink"))
-                return GF_FOP_UNLINK;
-        else if (!strcmp ((*op_no_str), "rmdir"))
-                return GF_FOP_RMDIR;
-        else if (!strcmp ((*op_no_str), "symlink"))
-                return GF_FOP_SYMLINK;
-        else if (!strcmp ((*op_no_str), "rename"))
-                return GF_FOP_RENAME;
-        else if (!strcmp ((*op_no_str), "link"))
-                return GF_FOP_LINK;
-        else if (!strcmp ((*op_no_str), "truncate"))
-                return GF_FOP_TRUNCATE;
-        else if (!strcmp ((*op_no_str), "create"))
-                return GF_FOP_CREATE;
-        else if (!strcmp ((*op_no_str), "open"))
-                return GF_FOP_OPEN;
-        else if (!strcmp ((*op_no_str), "readv"))
-                return GF_FOP_READ;
-        else if (!strcmp ((*op_no_str), "writev"))
-                return GF_FOP_WRITE;
-        else if (!strcmp ((*op_no_str), "statfs"))
-                return GF_FOP_STATFS;
-        else if (!strcmp ((*op_no_str), "flush"))
-                return GF_FOP_FLUSH;
-        else if (!strcmp ((*op_no_str), "fsync"))
-                return GF_FOP_FSYNC;
-        else if (!strcmp ((*op_no_str), "setxattr"))
-                return GF_FOP_SETXATTR;
-        else if (!strcmp ((*op_no_str), "getxattr"))
-                return GF_FOP_GETXATTR;
-        else if (!strcmp ((*op_no_str), "removexattr"))
-                return GF_FOP_REMOVEXATTR;
-        else if (!strcmp ((*op_no_str), "opendir"))
-                return GF_FOP_OPENDIR;
-        else if (!strcmp ((*op_no_str), "readdir"))
-                return GF_FOP_READDIR;
-        else if (!strcmp ((*op_no_str), "readdirp"))
-                return GF_FOP_READDIRP;
-        else if (!strcmp ((*op_no_str), "fsyncdir"))
-                return GF_FOP_FSYNCDIR;
-        else if (!strcmp ((*op_no_str), "access"))
-                return GF_FOP_ACCESS;
-        else if (!strcmp ((*op_no_str), "ftruncate"))
-                return GF_FOP_FTRUNCATE;
-        else if (!strcmp ((*op_no_str), "fstat"))
-                return GF_FOP_FSTAT;
-        else if (!strcmp ((*op_no_str), "lk"))
-                return GF_FOP_LK;
-        else if (!strcmp ((*op_no_str), "xattrop"))
-                return GF_FOP_XATTROP;
-        else if (!strcmp ((*op_no_str), "fxattrop"))
-                return GF_FOP_FXATTROP;
-        else if (!strcmp ((*op_no_str), "inodelk"))
-                return GF_FOP_INODELK;
-        else if (!strcmp ((*op_no_str), "finodelk"))
-                return GF_FOP_FINODELK;
-        else if (!strcmp ((*op_no_str), "etrylk"))
-                return GF_FOP_ENTRYLK;
-        else if (!strcmp ((*op_no_str), "fentrylk"))
-                return GF_FOP_FENTRYLK;
-        else if (!strcmp ((*op_no_str), "setattr"))
-                return GF_FOP_SETATTR;
-        else if (!strcmp ((*op_no_str), "fsetattr"))
-                return GF_FOP_FSETATTR;
-        else if (!strcmp ((*op_no_str), "getspec"))
-                return GF_FOP_GETSPEC;
-	else
-                return -1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
+
+    egp = this->private;
+    enable = egp->enable[GF_FOP_SETATTR];
+
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_SETATTR);
+
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(setattr, frame, -1, op_errno, NULL, NULL, xdata);
+        return 0;
+    }
+
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->setattr,
+                    loc, stbuf, valid, xdata);
+    return 0;
 }
 
 int
-error_gen (xlator_t *this, int op_no)
+error_gen_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                   struct iatt *stbuf, int32_t valid, dict_t *xdata)
 {
-        eg_t             *egp = NULL;
-        int              count = 0;
-        int              failure_iter_no = GF_FAILURE_DEFAULT;
-        char             *error_no = NULL;
-        int              rand_no = 0;
-        int              ret = 0;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-        egp = this->private;
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FSETATTR];
 
-        LOCK (&egp->lock);
-        {
-                count = ++egp->op_count;
-                failure_iter_no = egp->failure_iter_no;
-                error_no = egp->error_no;
-        }
-        UNLOCK (&egp->lock);
-
-        if((count % failure_iter_no) == 0) {
-                LOCK (&egp->lock);
-                {
-                        egp->op_count = 0;
-                }
-                UNLOCK (&egp->lock);
-
-                if (error_no)
-                        ret = conv_errno_to_int (&error_no);
-                else {
-
-                        rand_no = generate_rand_no (op_no);
-                        if (op_no >= GF_FOP_MAXVALUE)
-                                op_no = 0;
-                        if (rand_no >= error_no_list[op_no].error_no_count)
-                                rand_no = 0;
-                        ret = error_no_list[op_no].error_no[rand_no];
-                }
-                egp->failure_iter_no = 3 + (rand () % GF_UNIVERSAL_ANSWER);
-        }
-        return ret;
-}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FSETATTR);
 
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(fsetattr, frame, -1, op_errno, NULL, NULL, xdata);
+        return 0;
+    }
 
-int
-error_gen_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno, inode_t *inode,
-		      struct iatt *buf, dict_t *dict, struct iatt *postparent)
-{
-	STACK_UNWIND_STRICT (lookup, frame, op_ret, op_errno, inode,
-                             buf, dict, postparent);
-	return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->fsetattr,
+                    fd, stbuf, valid, xdata);
+    return 0;
 }
 
-
 int
-error_gen_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                  dict_t *xattr_req)
+error_gen_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   off_t offset, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_LOOKUP];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_LOOKUP);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (lookup, frame, -1, op_errno, NULL, NULL, NULL,
-                              NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_lookup_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->lookup,
-		    loc, xattr_req);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_TRUNCATE];
 
-int
-error_gen_forget (xlator_t *this, inode_t *inode)
-{
-	return 0;
-}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_TRUNCATE);
 
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(truncate, frame, -1, op_errno, NULL, NULL, xdata);
+        return 0;
+    }
+
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->truncate,
+                    loc, offset, xdata);
+    return 0;
+}
 
 int
-error_gen_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		    int32_t op_ret, int32_t op_errno, struct iatt *buf)
+error_gen_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                    dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (stat, frame, op_ret, op_errno, buf);
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-	return 0;
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FTRUNCATE];
+
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FTRUNCATE);
+
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(ftruncate, frame, -1, op_errno, NULL, NULL, xdata);
+        return 0;
+    }
+
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+    return 0;
 }
 
 int
-error_gen_stat (call_frame_t *frame, xlator_t *this, loc_t *loc)
+error_gen_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+                 dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_STAT];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_STAT);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (stat, frame, -1, op_errno, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_stat_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->stat,
-		    loc);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_ACCESS];
 
-int
-error_gen_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno,
-                       struct iatt *preop, struct iatt *postop)
-{
-	STACK_UNWIND_STRICT (setattr, frame, op_ret, op_errno, preop, postop);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_ACCESS);
 
-	return 0;
-}
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(access, frame, -1, op_errno, xdata);
+        return 0;
+    }
 
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->access,
+                    loc, mask, xdata);
+    return 0;
+}
 
 int
-error_gen_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                   struct iatt *stbuf, int32_t valid)
+error_gen_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+                   dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_SETATTR];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_SETATTR);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (setattr, frame, -1, op_errno, NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_setattr_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->setattr,
-		    loc, stbuf, valid);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_READLINK];
 
-int
-error_gen_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                    struct iatt *stbuf, int32_t valid)
-{
-	int             op_errno = 0;
-        eg_t            *egp = NULL;
-        int             enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FSETATTR];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FSETATTR);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fsetattr, frame, -1, op_errno, NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_setattr_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->fsetattr,
-		    fd, stbuf, valid);
-	return 0;
-}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_READLINK);
 
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(readlink, frame, -1, op_errno, NULL, NULL, xdata);
+        return 0;
+    }
 
-int
-error_gen_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int32_t op_ret, int32_t op_errno,
-			struct iatt *prebuf, struct iatt *postbuf)
-{
-	STACK_UNWIND_STRICT (truncate, frame, op_ret, op_errno,
-                             prebuf, postbuf);
-	return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->readlink,
+                    loc, size, xdata);
+    return 0;
 }
 
-
 int
-error_gen_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                    off_t offset)
+error_gen_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+                dev_t rdev, mode_t umask, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_TRUNCATE];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_TRUNCATE);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (truncate, frame, -1, op_errno,
-                                     NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_truncate_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->truncate,
-		    loc, offset);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_MKNOD];
 
-int
-error_gen_ftruncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			 int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                         struct iatt *postbuf)
-{
-	STACK_UNWIND_STRICT (ftruncate, frame, op_ret, op_errno,
-                             prebuf, postbuf);
-	return 0;
-}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_MKNOD);
 
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                            xdata);
+        return 0;
+    }
 
-int
-error_gen_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd,
-		     off_t offset)
-{
-	int              op_errno = 0;
-        eg_t            *egp =NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FTRUNCATE];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FTRUNCATE);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (ftruncate, frame, -1, op_errno,
-                                     NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_ftruncate_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->ftruncate,
-		    fd, offset);
-	return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->mknod,
+                    loc, mode, rdev, umask, xdata);
+    return 0;
 }
 
-
 int
-error_gen_access_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno)
+error_gen_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+                mode_t umask, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (access, frame, op_ret, op_errno);
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-	return 0;
-}
+    egp = this->private;
+    enable = egp->enable[GF_FOP_MKDIR];
 
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_MKDIR);
 
-int
-error_gen_access (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		  int32_t mask)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_ACCESS];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_ACCESS);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (access, frame, -1, op_errno);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_access_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->access,
-		    loc, mask);
-	return 0;
-}
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                            xdata);
+        return 0;
+    }
 
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->mkdir,
+                    loc, mode, umask, xdata);
+    return 0;
+}
 
 int
-error_gen_readlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int32_t op_ret, int32_t op_errno,
-			const char *path, struct iatt *sbuf)
+error_gen_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+                 dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (readlink, frame, op_ret, op_errno, path, sbuf);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_UNLINK];
 
-int
-error_gen_readlink (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                    size_t size)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_READLINK];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_READLINK);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (readlink, frame, -1, op_errno, NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_readlink_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->readlink,
-		    loc, size);
-	return 0;
-}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_UNLINK);
 
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(unlink, frame, -1, op_errno, NULL, NULL, xdata);
+        return 0;
+    }
 
-int
-error_gen_mknod_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		     int32_t op_ret, int32_t op_errno, inode_t *inode,
-                     struct iatt *buf, struct iatt *preparent,
-                     struct iatt *postparent)
-{
-	STACK_UNWIND_STRICT (mknod, frame, op_ret, op_errno,
-                             inode, buf,
-                             preparent, postparent);
-	return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->unlink,
+                    loc, xflag, xdata);
+    return 0;
 }
 
-
 int
-error_gen_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		 mode_t mode, dev_t rdev, dict_t *params)
+error_gen_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+                dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_MKNOD];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_MKNOD);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (mknod, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_mknod_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->mknod,
-		    loc, mode, rdev, params);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_RMDIR];
 
-int
-error_gen_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		     int32_t op_ret, int32_t op_errno, inode_t *inode,
-                     struct iatt *buf, struct iatt *preparent,
-                     struct iatt *postparent)
-{
-	STACK_UNWIND_STRICT (mkdir, frame, op_ret, op_errno,
-                             inode, buf,
-                             preparent, postparent);
-	return 0;
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_RMDIR);
+
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(rmdir, frame, -1, op_errno, NULL, NULL, xdata);
+        return 0;
+    }
+
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->rmdir,
+                    loc, flags, xdata);
+    return 0;
 }
 
 int
-error_gen_mkdir (call_frame_t *frame, xlator_t *this,
-		 loc_t *loc, mode_t mode, dict_t *params)
+error_gen_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+                  loc_t *loc, mode_t umask, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_MKDIR];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_MKDIR);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (mkdir, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_mkdir_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->mkdir,
-		    loc, mode, params);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_SYMLINK];
 
-int
-error_gen_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno,
-                      struct iatt *preparent, struct iatt *postparent)
-{
-	STACK_UNWIND_STRICT (unlink, frame, op_ret, op_errno,
-                             preparent, postparent);
-	return 0;
-}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_SYMLINK);
 
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(symlink, frame, -1, op_errno, NULL, NULL, NULL,
+                            NULL, NULL); /* pre & post parent attr */
+        return 0;
+    }
 
-int
-error_gen_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_UNLINK];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_UNLINK);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (unlink, frame, -1, op_errno, NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_unlink_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->unlink,
-		    loc);
-	return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->symlink,
+                    linkpath, loc, umask, xdata);
+    return 0;
 }
 
-
 int
-error_gen_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		     int32_t op_ret, int32_t op_errno,
-                     struct iatt *preparent, struct iatt *postparent)
+error_gen_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc,
+                 loc_t *newloc, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (rmdir, frame, op_ret, op_errno,
-                             preparent, postparent);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_RENAME];
 
-int
-error_gen_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_RMDIR];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_RMDIR);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (rmdir, frame, -1, op_errno, NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_rmdir_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->rmdir,
-		    loc, flags);
-	return 0;
-}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_RENAME);
 
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(rename, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                            NULL, NULL);
+        return 0;
+    }
 
-int
-error_gen_symlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno, inode_t *inode,
-                       struct iatt *buf, struct iatt *preparent,
-                       struct iatt *postparent)
-{
-	STACK_UNWIND_STRICT (symlink, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent);
-	return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->rename,
+                    oldloc, newloc, xdata);
+    return 0;
 }
 
-
 int
-error_gen_symlink (call_frame_t *frame, xlator_t *this, const char *linkpath,
-		   loc_t *loc, dict_t *params)
+error_gen_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc,
+               loc_t *newloc, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_SYMLINK];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_SYMLINK);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (symlink, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL); /* pre & post parent attr */
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_symlink_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->symlink,
-		    linkpath, loc, params);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_LINK];
 
-int
-error_gen_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno, struct iatt *buf,
-                      struct iatt *preoldparent, struct iatt *postoldparent,
-                      struct iatt *prenewparent, struct iatt *postnewparent)
-{
-	STACK_UNWIND_STRICT (rename, frame, op_ret, op_errno, buf,
-                             preoldparent, postoldparent,
-                             prenewparent, postnewparent);
-	return 0;
-}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_LINK);
 
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(link, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                            NULL);
+        return 0;
+    }
 
-int
-error_gen_rename (call_frame_t *frame, xlator_t *this,
-		  loc_t *oldloc, loc_t *newloc)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_RENAME];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_RENAME);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (rename, frame, -1, op_errno, NULL,
-                                     NULL, NULL, NULL, NULL); /* pre & post parent attr */
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_rename_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->rename,
-		    oldloc, newloc);
-	return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->link,
+                    oldloc, newloc, xdata);
+    return 0;
 }
 
-
 int
-error_gen_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		    int32_t op_ret, int32_t op_errno, inode_t *inode,
-                    struct iatt *buf, struct iatt *preparent,
-                    struct iatt *postparent)
+error_gen_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+                 mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (link, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_CREATE];
 
-int
-error_gen_link (call_frame_t *frame, xlator_t *this,
-		loc_t *oldloc, loc_t *newloc)
-{
-	int             op_errno = 0;
-        eg_t            *egp = NULL;
-        int             enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_LINK];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_LINK);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (link, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL); /* pre & post parent attr */
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_link_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->link,
-		    oldloc, newloc);
-	return 0;
-}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_CREATE);
 
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(create, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                            NULL, NULL);
+        return 0;
+    }
 
-int
-error_gen_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno,
-		      fd_t *fd, inode_t *inode, struct iatt *buf,
-                      struct iatt *preparent, struct iatt *postparent)
-{
-	STACK_UNWIND_STRICT (create, frame, op_ret, op_errno, fd, inode, buf,
-                             preparent, postparent);
-	return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->create,
+                    loc, flags, mode, umask, fd, xdata);
+    return 0;
 }
 
-
 int
-error_gen_create (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		  int32_t flags, mode_t mode, fd_t *fd, dict_t *params)
+error_gen_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+               fd_t *fd, dict_t *xdata)
 {
-	int             op_errno = 0;
-        eg_t            *egp = NULL;
-        int             enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_CREATE];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_CREATE);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (create, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL, NULL); /* pre & post attr */
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_create_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->create,
-		    loc, flags, mode, fd, params);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_OPEN];
 
-int
-error_gen_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		    int32_t op_ret, int32_t op_errno, fd_t *fd)
-{
-	STACK_UNWIND_STRICT (open, frame, op_ret, op_errno, fd);
-	return 0;
-}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_OPEN);
 
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(open, frame, -1, op_errno, NULL, xdata);
+        return 0;
+    }
 
-int
-error_gen_open (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		int32_t flags, fd_t *fd, int32_t wbflags)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_OPEN];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_OPEN);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (open, frame, -1, op_errno, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_open_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->open,
-		    loc, flags, fd, wbflags);
-	return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->open,
+                    loc, flags, fd, xdata);
+    return 0;
 }
 
-
 int
-error_gen_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		     int32_t op_ret, int32_t op_errno,
-		     struct iovec *vector, int32_t count,
-		     struct iatt *stbuf, struct iobref *iobref)
+error_gen_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                off_t offset, uint32_t flags, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (readv, frame, op_ret, op_errno,
-                             vector, count, stbuf, iobref);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
+
+    egp = this->private;
+    enable = egp->enable[GF_FOP_READ];
+
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_READ);
 
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(readv, frame, -1, op_errno, NULL, 0, NULL, NULL,
+                            xdata);
+        return 0;
+    }
+
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->readv,
+                    fd, size, offset, flags, xdata);
+    return 0;
+}
 
 int
-error_gen_readv (call_frame_t *frame, xlator_t *this,
-		 fd_t *fd, size_t size, off_t offset)
+error_gen_writev(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                 struct iovec *vector, int32_t count, off_t off, uint32_t flags,
+                 struct iobref *iobref, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
+    struct iovec *shortvec = NULL;
 
-        egp = this->private;
-        enable = egp->enable[GF_FOP_READ];
+    egp = this->private;
+    enable = egp->enable[GF_FOP_WRITE];
 
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_READ);
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_WRITE);
 
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (readv, frame, -1, op_errno, NULL, 0,
-                                     NULL, NULL);
-		return 0;
-	}
+    if (op_errno == GF_ERROR_SHORT_WRITE) {
+        /*
+         * A short write error returns some value less than what was
+         * requested from a write. To simulate this, replace the vector
+         * with one half the size;
+         */
+        shortvec = iov_dup(vector, 1);
+        shortvec->iov_len /= 2;
+        count = 1;
+        goto wind;
+    } else if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(writev, frame, -1, op_errno, NULL, NULL, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->writev,
+                    fd, shortvec ? shortvec : vector, count, off, flags, iobref,
+                    xdata);
 
-
-	STACK_WIND (frame, error_gen_readv_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->readv,
-		    fd, size, offset);
-	return 0;
+    if (shortvec)
+        GF_FREE(shortvec);
+    return 0;
 }
 
-
 int
-error_gen_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno,
-                      struct iatt *prebuf, struct iatt *postbuf)
+error_gen_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, prebuf, postbuf);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FLUSH];
 
-int
-error_gen_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
-		  struct iovec *vector, int32_t count,
-		  off_t off, struct iobref *iobref)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_WRITE];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_WRITE);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (writev, frame, -1, op_errno, NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_writev_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->writev,
-		    fd, vector, count, off, iobref);
-	return 0;
-}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FLUSH);
 
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(flush, frame, -1, op_errno, xdata);
+        return 0;
+    }
 
-int
-error_gen_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		     int32_t op_ret, int32_t op_errno)
-{
-	STACK_UNWIND_STRICT (flush, frame, op_ret, op_errno);
-	return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->flush,
+                    fd, xdata);
+    return 0;
 }
 
-
 int
-error_gen_flush (call_frame_t *frame, xlator_t *this, fd_t *fd)
+error_gen_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+                dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FLUSH];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FLUSH);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (flush, frame, -1, op_errno);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_flush_cbk,
-                    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->flush,
-		    fd);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FSYNC];
 
-int
-error_gen_fsync_cbk (call_frame_t *frame, void *cookie,
-		     xlator_t *this, int32_t op_ret,
-		     int32_t op_errno, struct iatt *prebuf,
-                     struct iatt *postbuf)
-{
-	STACK_UNWIND_STRICT (fsync, frame, op_ret, op_errno, prebuf, postbuf);
-	return 0;
-}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FSYNC);
 
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(fsync, frame, -1, op_errno, NULL, NULL, xdata);
+        return 0;
+    }
 
-int
-error_gen_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FSYNC];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FSYNC);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fsync, frame, -1, op_errno, NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_fsync_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->fsync,
-		    fd, flags);
-	return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->fsync,
+                    fd, flags, xdata);
+    return 0;
 }
 
-
 int
-error_gen_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, struct iatt *buf)
+error_gen_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (fstat, frame, op_ret, op_errno, buf);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FSTAT];
 
-int
-error_gen_fstat (call_frame_t *frame, xlator_t *this, fd_t *fd)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FSTAT];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FSTAT);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fstat, frame, -1, op_errno, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_fstat_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->fstat,
-		    fd);
-	return 0;
-}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FSTAT);
 
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(fstat, frame, -1, op_errno, NULL, xdata);
+        return 0;
+    }
 
-int
-error_gen_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno, fd_t *fd)
-{
-	STACK_UNWIND_STRICT (opendir, frame, op_ret, op_errno, fd);
-	return 0;
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->fstat,
+                    fd, xdata);
+    return 0;
 }
 
-
 int
-error_gen_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd)
+error_gen_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+                  dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_OPENDIR];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_OPENDIR);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (opendir, frame, -1, op_errno, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_opendir_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->opendir,
-		    loc, fd);
-	return 0;
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
+
+    egp = this->private;
+    enable = egp->enable[GF_FOP_OPENDIR];
+
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_OPENDIR);
+
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(opendir, frame, -1, op_errno, NULL, xdata);
+        return 0;
+    }
+
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->opendir,
+                    loc, fd, xdata);
+    return 0;
 }
 
 int
-error_gen_fsyncdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int32_t op_ret, int32_t op_errno)
+error_gen_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+                   dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (fsyncdir, frame, op_ret, op_errno);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FSYNCDIR];
 
-int
-error_gen_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                    int32_t flags)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FSYNCDIR];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FSYNCDIR);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fsyncdir, frame, -1, op_errno);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_fsyncdir_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->fsyncdir,
-		    fd, flags);
-	return 0;
-}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FSYNCDIR);
 
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(fsyncdir, frame, -1, op_errno, xdata);
+        return 0;
+    }
+
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->fsyncdir,
+                    fd, flags, xdata);
+    return 0;
+}
 
 int
-error_gen_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno, struct statvfs *buf)
+error_gen_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (statfs, frame, op_ret, op_errno, buf);
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-	return 0;
-}
+    egp = this->private;
+    enable = egp->enable[GF_FOP_STATFS];
 
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_STATFS);
 
-int
-error_gen_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_STATFS];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_STATFS);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (statfs, frame, -1, op_errno, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_statfs_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->statfs,
-		    loc);
-	return 0;
-}
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(statfs, frame, -1, op_errno, NULL, xdata);
+        return 0;
+    }
 
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->statfs,
+                    loc, xdata);
+    return 0;
+}
 
 int
-error_gen_setxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int32_t op_ret, int32_t op_errno)
+error_gen_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   dict_t *dict, int32_t flags, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (setxattr, frame, op_ret, op_errno);
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-	return 0;
-}
+    egp = this->private;
+    enable = egp->enable[GF_FOP_SETXATTR];
 
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_SETXATTR);
 
-int
-error_gen_setxattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		    dict_t *dict, int32_t flags)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_SETXATTR];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_SETXATTR);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (setxattr, frame, -1, op_errno);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_setxattr_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->setxattr,
-		    loc, dict, flags);
-	return 0;
-}
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(setxattr, frame, -1, op_errno, xdata);
+        return 0;
+    }
 
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->setxattr,
+                    loc, dict, flags, xdata);
+    return 0;
+}
 
 int
-error_gen_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int32_t op_ret, int32_t op_errno, dict_t *dict)
+error_gen_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   const char *name, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (getxattr, frame, op_ret, op_errno, dict);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_GETXATTR];
 
-int
-error_gen_getxattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		    const char *name)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_GETXATTR];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_GETXATTR);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (getxattr, frame, -1, op_errno, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_getxattr_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->getxattr,
-		    loc, name);
-	return 0;
-}
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_GETXATTR);
+
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(getxattr, frame, -1, op_errno, NULL, xdata);
+        return 0;
+    }
 
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->getxattr,
+                    loc, name, xdata);
+    return 0;
+}
 
 int
-error_gen_xattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, dict_t *dict)
+error_gen_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+                    int32_t flags, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (xattrop, frame, op_ret, op_errno, dict);
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-	return 0;
-}
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FSETXATTR];
 
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FSETXATTR);
 
-int
-error_gen_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		   gf_xattrop_flags_t flags, dict_t *dict)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_XATTROP];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_XATTROP);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (xattrop, frame, -1, op_errno, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_xattrop_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->xattrop,
-		    loc, flags, dict);
-	return 0;
-}
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(fsetxattr, frame, -1, op_errno, xdata);
+        return 0;
+    }
 
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
+    return 0;
+}
 
 int
-error_gen_fxattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int32_t op_ret, int32_t op_errno, dict_t *dict)
+error_gen_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                    const char *name, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (fxattrop, frame, op_ret, op_errno, dict);
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-	return 0;
-}
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FGETXATTR];
 
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FGETXATTR);
 
-int
-error_gen_fxattrop (call_frame_t *frame, xlator_t *this, fd_t *fd,
-		    gf_xattrop_flags_t flags, dict_t *dict)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FXATTROP];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FXATTROP);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fxattrop, frame, -1, op_errno, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_fxattrop_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->fxattrop,
-		    fd, flags, dict);
-	return 0;
-}
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(fgetxattr, frame, -1, op_errno, NULL, xdata);
+        return 0;
+    }
 
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->fgetxattr, fd, name, xdata);
+    return 0;
+}
 
 int
-error_gen_removexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			   int32_t op_ret, int32_t op_errno)
+error_gen_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                  gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (removexattr, frame, op_ret, op_errno);
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-	return 0;
-}
+    egp = this->private;
+    enable = egp->enable[GF_FOP_XATTROP];
 
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_XATTROP);
 
-int
-error_gen_removexattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-		       const char *name)
-{
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_REMOVEXATTR];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_REMOVEXATTR);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (removexattr, frame, -1, op_errno);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_removexattr_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->removexattr,
-		    loc, name);
-	return 0;
-}
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(xattrop, frame, -1, op_errno, NULL, xdata);
+        return 0;
+    }
 
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->xattrop,
+                    loc, flags, dict, xdata);
+    return 0;
+}
 
 int
-error_gen_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		  int32_t op_ret, int32_t op_errno, struct gf_flock *lock)
+error_gen_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                   gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (lk, frame, op_ret, op_errno, lock);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
+
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FXATTROP];
+
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FXATTROP);
+
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(fxattrop, frame, -1, op_errno, NULL, xdata);
+        return 0;
+    }
 
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->fxattrop,
+                    fd, flags, dict, xdata);
+    return 0;
+}
 
 int
-error_gen_lk (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
-	      struct gf_flock *lock)
+error_gen_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                      const char *name, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_LK];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_LK);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (lk, frame, -1, op_errno, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_lk_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->lk,
-		    fd, cmd, lock);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_REMOVEXATTR];
 
-int
-error_gen_inodelk_cbk (call_frame_t *frame, void *cookie,
-		       xlator_t *this, int32_t op_ret, int32_t op_errno)
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_REMOVEXATTR);
 
-{
-	STACK_UNWIND_STRICT (inodelk, frame, op_ret, op_errno);
-	return 0;
-}
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(removexattr, frame, -1, op_errno, xdata);
+        return 0;
+    }
 
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->removexattr, loc, name, xdata);
+    return 0;
+}
 
 int
-error_gen_inodelk (call_frame_t *frame, xlator_t *this,
-		   const char *volume, loc_t *loc, int32_t cmd,
-                   struct gf_flock *lock)
+error_gen_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                       const char *name, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_INODELK];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_INODELK);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (inodelk, frame, -1, op_errno);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_inodelk_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->inodelk,
-		    volume, loc, cmd, lock);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FREMOVEXATTR];
 
-int
-error_gen_finodelk_cbk (call_frame_t *frame, void *cookie,
-			xlator_t *this, int32_t op_ret, int32_t op_errno)
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FREMOVEXATTR);
 
-{
-	STACK_UNWIND_STRICT (finodelk, frame, op_ret, op_errno);
-	return 0;
-}
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(fremovexattr, frame, -1, op_errno, xdata);
+        return 0;
+    }
 
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->fremovexattr, fd, name, xdata);
+    return 0;
+}
 
 int
-error_gen_finodelk (call_frame_t *frame, xlator_t *this,
-		    const char *volume, fd_t *fd, int32_t cmd,
-                    struct gf_flock *lock)
+error_gen_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+             struct gf_flock *lock, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FINODELK];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FINODELK);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (finodelk, frame, -1, op_errno);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_finodelk_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->finodelk,
-		    volume, fd, cmd, lock);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_LK];
 
-int
-error_gen_entrylk_cbk (call_frame_t *frame, void *cookie,
-		       xlator_t *this, int32_t op_ret, int32_t op_errno)
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_LK);
 
-{
-	STACK_UNWIND_STRICT (entrylk, frame, op_ret, op_errno);
-	return 0;
-}
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(lk, frame, -1, op_errno, NULL, xdata);
+        return 0;
+    }
 
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->lk, fd,
+                    cmd, lock, xdata);
+    return 0;
+}
 
 int
-error_gen_entrylk (call_frame_t *frame, xlator_t *this,
-		   const char *volume, loc_t *loc, const char *basename,
-		   entrylk_cmd cmd, entrylk_type type)
+error_gen_inodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+                  loc_t *loc, int32_t cmd, struct gf_flock *lock, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_ENTRYLK];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_ENTRYLK);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (entrylk, frame, -1, op_errno);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_entrylk_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->entrylk,
-		    volume, loc, basename, cmd, type);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_INODELK];
 
-int
-error_gen_fentrylk_cbk (call_frame_t *frame, void *cookie,
-			xlator_t *this, int32_t op_ret, int32_t op_errno)
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_INODELK);
 
-{
-	STACK_UNWIND_STRICT (fentrylk, frame, op_ret, op_errno);
-	return 0;
-}
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(inodelk, frame, -1, op_errno, xdata);
+        return 0;
+    }
 
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->inodelk,
+                    volume, loc, cmd, lock, xdata);
+    return 0;
+}
 
 int
-error_gen_fentrylk (call_frame_t *frame, xlator_t *this,
-		    const char *volume, fd_t *fd, const char *basename,
-		    entrylk_cmd cmd, entrylk_type type)
+error_gen_finodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+                   fd_t *fd, int32_t cmd, struct gf_flock *lock, dict_t *xdata)
 {
-	int             op_errno = 0;
-        eg_t            *egp = NULL;
-        int             enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_FENTRYLK];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_FENTRYLK);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (fentrylk, frame, -1, op_errno);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_fentrylk_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->fentrylk,
-		    volume, fd, basename, cmd, type);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FINODELK];
 
-/* Management operations */
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FINODELK);
 
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(finodelk, frame, -1, op_errno, xdata);
+        return 0;
+    }
+
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->finodelk,
+                    volume, fd, cmd, lock, xdata);
+    return 0;
+}
 
 int
-error_gen_getspec_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno, char *spec_data)
+error_gen_entrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+                  loc_t *loc, const char *basename, entrylk_cmd cmd,
+                  entrylk_type type, dict_t *xdata)
 {
-	STACK_UNWIND_STRICT (getspec, frame, op_ret, op_errno, spec_data);
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
-	return 0;
-}
+    egp = this->private;
+    enable = egp->enable[GF_FOP_ENTRYLK];
+
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_ENTRYLK);
 
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(entrylk, frame, -1, op_errno, xdata);
+        return 0;
+    }
+
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->entrylk,
+                    volume, loc, basename, cmd, type, xdata);
+    return 0;
+}
 
 int
-error_gen_getspec (call_frame_t *frame, xlator_t *this, const char *key,
-		   int32_t flags)
+error_gen_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+                   fd_t *fd, const char *basename, entrylk_cmd cmd,
+                   entrylk_type type, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_GETSPEC];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_GETSPEC);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (getspec, frame, -1, op_errno, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_getspec_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->getspec,
-		    key, flags);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
+
+    egp = this->private;
+    enable = egp->enable[GF_FOP_FENTRYLK];
+
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_FENTRYLK);
 
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(fentrylk, frame, -1, op_errno, xdata);
+        return 0;
+    }
+
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->fentrylk,
+                    volume, fd, basename, cmd, type, xdata);
+    return 0;
+}
 
 int
-error_gen_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno, gf_dirent_t *entries)
+error_gen_getspec(call_frame_t *frame, xlator_t *this, const char *key,
+                  int32_t flags)
 {
-	STACK_UNWIND_STRICT (readdir, frame, op_ret, op_errno, entries);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
+
+    egp = this->private;
+    enable = egp->enable[GF_FOP_GETSPEC];
 
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_GETSPEC);
+
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(getspec, frame, -1, op_errno, NULL);
+        return 0;
+    }
+
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->getspec,
+                    key, flags);
+    return 0;
+}
 
 int
-error_gen_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd,
-		   size_t size, off_t off)
+error_gen_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                  off_t off, dict_t *xdata)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_READDIR];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_READDIR);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (readdir, frame, -1, op_errno, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_readdir_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->readdir,
-		    fd, size, off);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
+
+    egp = this->private;
+    enable = egp->enable[GF_FOP_READDIR];
+
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_READDIR);
+
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(readdir, frame, -1, op_errno, NULL, xdata);
+        return 0;
+    }
 
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->readdir,
+                    fd, size, off, xdata);
+    return 0;
+}
 
 int
-error_gen_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, gf_dirent_t *entries)
+error_gen_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                   off_t off, dict_t *dict)
 {
-	STACK_UNWIND_STRICT (readdirp, frame, op_ret, op_errno, entries);
-	return 0;
-}
+    int op_errno = 0;
+    eg_t *egp = NULL;
+    int enable = 1;
 
+    egp = this->private;
+    enable = egp->enable[GF_FOP_READDIRP];
 
-int
-error_gen_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-                    off_t off)
+    if (enable)
+        op_errno = error_gen(this, GF_FOP_READDIRP);
+
+    if (op_errno) {
+        GF_ERROR(this, "unwind(-1, %s)", strerror(op_errno));
+        STACK_UNWIND_STRICT(readdirp, frame, -1, op_errno, NULL, NULL);
+        return 0;
+    }
+
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->readdirp,
+                    fd, size, off, dict);
+    return 0;
+}
+
+static void
+error_gen_set_failure(eg_t *pvt, double percent)
 {
-	int              op_errno = 0;
-        eg_t            *egp = NULL;
-        int              enable = 1;
-
-        egp = this->private;
-        enable = egp->enable[GF_FOP_READDIRP];
-
-        if (enable)
-                op_errno = error_gen (this, GF_FOP_READDIRP);
-
-	if (op_errno) {
-		GF_ERROR(this, "unwind(-1, %s)", strerror (op_errno));
-		STACK_UNWIND_STRICT (readdirp, frame, -1, op_errno, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame, error_gen_readdirp_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->readdirp,
-                    fd, size, off);
-	return 0;
+    double ppm;
+
+    GF_ASSERT(pvt);
+
+    ppm = (percent / 100.0) * (double)FAILURE_GRANULARITY;
+    pvt->failure_iter_no = (int)ppm;
 }
 
+static void
+error_gen_parse_fill_fops(eg_t *pvt, char *enable_fops)
+{
+    char *op_no_str = NULL;
+    int op_no = -1;
+    int i = 0;
+    xlator_t *this = THIS;
+    char *saveptr = NULL;
 
-int
-error_gen_closedir (xlator_t *this, fd_t *fd)
+    GF_ASSERT(pvt);
+    GF_ASSERT(this);
+
+    for (i = 0; i < GF_FOP_MAXVALUE; i++)
+        pvt->enable[i] = 0;
+
+    if (!enable_fops) {
+        gf_log(this->name, GF_LOG_WARNING, "All fops are enabled.");
+        for (i = 0; i < GF_FOP_MAXVALUE; i++)
+            pvt->enable[i] = 1;
+    } else {
+        op_no_str = strtok_r(enable_fops, ",", &saveptr);
+        while (op_no_str) {
+            op_no = gf_fop_int(op_no_str);
+            if (op_no == -1) {
+                gf_log(this->name, GF_LOG_WARNING, "Wrong option value %s",
+                       op_no_str);
+            } else
+                pvt->enable[op_no] = 1;
+
+            op_no_str = strtok_r(NULL, ",", &saveptr);
+        }
+    }
+}
+
+int32_t
+error_gen_priv_dump(xlator_t *this)
 {
-	return 0;
+    char key_prefix[GF_DUMP_MAX_BUF_LEN];
+    int ret = -1;
+    eg_t *conf = NULL;
+
+    if (!this)
+        goto out;
+
+    conf = this->private;
+    if (!conf)
+        goto out;
+
+    ret = TRY_LOCK(&conf->lock);
+    if (ret != 0) {
+        return ret;
+    }
+
+    gf_proc_dump_add_section("xlator.debug.error-gen.%s.priv", this->name);
+    gf_proc_dump_build_key(key_prefix, "xlator.debug.error-gen", "%s.priv",
+                           this->name);
+
+    gf_proc_dump_write("op_count", "%d", conf->op_count);
+    gf_proc_dump_write("failure_iter_no", "%d", conf->failure_iter_no);
+    gf_proc_dump_write("error_no", "%d", conf->error_no_int);
+    gf_proc_dump_write("random_failure", "%d", conf->random_failure);
+
+    UNLOCK(&conf->lock);
+out:
+    return ret;
 }
 
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    if (!this)
+        return ret;
+
+    ret = xlator_mem_acct_init(this, gf_error_gen_mt_end + 1);
+
+    if (ret != 0) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Memory accounting init"
+               " failed");
+        return ret;
+    }
+
+    return ret;
+}
 
 int
-error_gen_close (xlator_t *this, fd_t *fd)
+reconfigure(xlator_t *this, dict_t *options)
 {
-	return 0;
-}
+    eg_t *pvt = NULL;
+    int32_t ret = 0;
+    char *error_enable_fops = NULL;
+    char *error_no = NULL;
+    double failure_percent_dbl = 0.0;
+
+    if (!this || !this->private)
+        goto out;
+
+    pvt = this->private;
+
+    ret = -1;
+
+    GF_OPTION_RECONF("error-no", error_no, options, str, out);
+
+    if (error_no)
+        pvt->error_no_int = conv_errno_to_int(&error_no);
+
+    GF_OPTION_RECONF("failure", failure_percent_dbl, options, percent, out);
+
+    GF_OPTION_RECONF("enable", error_enable_fops, options, str, out);
 
+    GF_OPTION_RECONF("random-failure", pvt->random_failure, options, bool, out);
+
+    error_gen_parse_fill_fops(pvt, error_enable_fops);
+    error_gen_set_failure(pvt, failure_percent_dbl);
+
+    ret = 0;
+out:
+    gf_log(this ? this->name : "error-gen", GF_LOG_DEBUG,
+           "reconfigure returning %d", ret);
+    return ret;
+}
 
 int
-init (xlator_t *this)
+init(xlator_t *this)
 {
-        eg_t            *pvt = NULL;
-        data_t          *error_no = NULL;
-        data_t          *failure_percent = NULL;
-        data_t          *enable = NULL;
-        int32_t          ret = 0;
-        char            *error_enable_fops = NULL;
-        char            *op_no_str = NULL;
-        int              op_no = -1;
-        int              i = 0;
-        int32_t          failure_percent_int = 0;
-
-        if (!this->children || this->children->next) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "error-gen not configured with one subvolume");
-                ret = -1;
-                goto out;
-        }
+    eg_t *pvt = NULL;
+    int32_t ret = 0;
+    char *error_enable_fops = NULL;
+    char *error_no = NULL;
+    double failure_percent_dbl = 0.0;
 
-        if (!this->parents) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "dangling volume. check volfile ");
-        }
+    if (!this->children || this->children->next) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "error-gen not configured with one subvolume");
+        ret = -1;
+        goto out;
+    }
 
-        error_no = dict_get (this->options, "error-no");
-        failure_percent = dict_get (this->options, "failure");
-        enable = dict_get (this->options, "enable");
+    if (!this->parents) {
+        gf_log(this->name, GF_LOG_WARNING, "dangling volume. check volfile ");
+    }
 
-        pvt = CALLOC (1, sizeof (eg_t));
+    pvt = GF_CALLOC(1, sizeof(eg_t), gf_error_gen_mt_eg_t);
 
-        if (!pvt) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "out of memory.");
-                ret = -1;
-                goto out;
-        }
+    if (!pvt) {
+        ret = -1;
+        goto out;
+    }
 
-        LOCK_INIT (&pvt->lock);
+    LOCK_INIT(&pvt->lock);
 
-        for (i = 0; i < GF_FOP_MAXVALUE; i++)
-                pvt->enable[i] = 0;
-        if (!error_no) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "error-no not specified.");
-        } else {
-                pvt->error_no = data_to_str (error_no);
-        }
+    ret = -1;
 
-        if (!failure_percent) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "failure percent not specified.");
-                pvt->failure_iter_no = 100/GF_FAILURE_DEFAULT;
-        } else {
-                failure_percent_int = data_to_int32 (failure_percent);
-                if (failure_percent_int)
-                        pvt->failure_iter_no = 100/failure_percent_int;
-                else
-                        pvt->failure_iter_no = 100/GF_FAILURE_DEFAULT;
-        }
+    GF_OPTION_INIT("error-no", error_no, str, out);
 
-        if (!enable) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "All fops are enabled.");
-                for (i = 0; i < GF_FOP_MAXVALUE; i++)
-                        pvt->enable[i] = 1;
-        } else {
-                error_enable_fops = data_to_str (enable);
-                op_no_str = error_enable_fops;
-                while ((*error_enable_fops) != '\0') {
-                        error_enable_fops++;
-                        if (((*error_enable_fops) == ',') ||
-                            ((*error_enable_fops) == '\0')) {
-                                if ((*error_enable_fops) != '\0') {
-                                        (*error_enable_fops) = '\0';
-                                        error_enable_fops++;
-                                }
-                                op_no = get_fop_int (&op_no_str);
-                                if (op_no == -1) {
-                                        gf_log (this->name, GF_LOG_WARNING,
-                                                "Wrong option value %s",
-                                                op_no_str);
-                                } else
-                                        pvt->enable[op_no] = 1;
-                                op_no_str = error_enable_fops;
-                        }
-                }
-        }
-        this->private = pvt;
+    if (error_no)
+        pvt->error_no_int = conv_errno_to_int(&error_no);
+
+    GF_OPTION_INIT("failure", failure_percent_dbl, percent, out);
+
+    GF_OPTION_INIT("enable", error_enable_fops, str, out);
+
+    GF_OPTION_INIT("random-failure", pvt->random_failure, bool, out);
 
-        /* Give some seed value here */
-        srand (time(NULL));
+    error_gen_parse_fill_fops(pvt, error_enable_fops);
+    error_gen_set_failure(pvt, failure_percent_dbl);
+
+    this->private = pvt;
+
+    /* Give some seed value here. */
+    srand(gf_time());
+
+    ret = 0;
 out:
-        return ret;
+    if (ret)
+        GF_FREE(pvt);
+    return ret;
 }
 
-
 void
-fini (xlator_t *this)
+fini(xlator_t *this)
 {
-        eg_t            *pvt = NULL;
-
-        if (!this)
-                 return;
-        pvt = this->private;
+    eg_t *pvt = NULL;
 
-        if (pvt) {
-                LOCK_DESTROY (&pvt->lock);
-                FREE (pvt);
-                gf_log (this->name, GF_LOG_DEBUG, "fini called");
-        }
+    if (!this)
         return;
+    pvt = this->private;
+
+    if (pvt) {
+        LOCK_DESTROY(&pvt->lock);
+        GF_FREE(pvt);
+        gf_log(this->name, GF_LOG_DEBUG, "fini called");
+    }
+    return;
 }
 
-struct xlator_fops fops = {
-	.lookup      = error_gen_lookup,
-	.stat        = error_gen_stat,
-	.readlink    = error_gen_readlink,
-	.mknod       = error_gen_mknod,
-	.mkdir       = error_gen_mkdir,
-	.unlink      = error_gen_unlink,
-	.rmdir       = error_gen_rmdir,
-	.symlink     = error_gen_symlink,
-	.rename      = error_gen_rename,
-	.link        = error_gen_link,
-	.truncate    = error_gen_truncate,
-	.create      = error_gen_create,
-	.open        = error_gen_open,
-	.readv       = error_gen_readv,
-	.writev      = error_gen_writev,
-	.statfs      = error_gen_statfs,
-	.flush       = error_gen_flush,
-	.fsync       = error_gen_fsync,
-	.setxattr    = error_gen_setxattr,
-	.getxattr    = error_gen_getxattr,
-	.removexattr = error_gen_removexattr,
-	.opendir     = error_gen_opendir,
-	.readdir     = error_gen_readdir,
-	.readdirp    = error_gen_readdirp,
-	.fsyncdir    = error_gen_fsyncdir,
-	.access      = error_gen_access,
-	.ftruncate   = error_gen_ftruncate,
-	.fstat       = error_gen_fstat,
-	.lk          = error_gen_lk,
-	.lookup_cbk  = error_gen_lookup_cbk,
-	.xattrop     = error_gen_xattrop,
-	.fxattrop    = error_gen_fxattrop,
-	.inodelk     = error_gen_inodelk,
-	.finodelk    = error_gen_finodelk,
-	.entrylk     = error_gen_entrylk,
-	.fentrylk    = error_gen_fentrylk,
-        .setattr     = error_gen_setattr,
-        .fsetattr    = error_gen_fsetattr,
-	.getspec     = error_gen_getspec,
+struct xlator_dumpops dumpops = {
+    .priv = error_gen_priv_dump,
 };
 
-struct xlator_cbks cbks = {
-	.release = error_gen_close,
-	.releasedir = error_gen_closedir,
+struct xlator_cbks cbks;
+
+struct xlator_fops fops = {
+    .lookup = error_gen_lookup,
+    .stat = error_gen_stat,
+    .readlink = error_gen_readlink,
+    .mknod = error_gen_mknod,
+    .mkdir = error_gen_mkdir,
+    .unlink = error_gen_unlink,
+    .rmdir = error_gen_rmdir,
+    .symlink = error_gen_symlink,
+    .rename = error_gen_rename,
+    .link = error_gen_link,
+    .truncate = error_gen_truncate,
+    .create = error_gen_create,
+    .open = error_gen_open,
+    .readv = error_gen_readv,
+    .writev = error_gen_writev,
+    .statfs = error_gen_statfs,
+    .flush = error_gen_flush,
+    .fsync = error_gen_fsync,
+    .setxattr = error_gen_setxattr,
+    .getxattr = error_gen_getxattr,
+    .removexattr = error_gen_removexattr,
+    .fsetxattr = error_gen_fsetxattr,
+    .fgetxattr = error_gen_fgetxattr,
+    .fremovexattr = error_gen_fremovexattr,
+    .opendir = error_gen_opendir,
+    .readdir = error_gen_readdir,
+    .readdirp = error_gen_readdirp,
+    .fsyncdir = error_gen_fsyncdir,
+    .access = error_gen_access,
+    .ftruncate = error_gen_ftruncate,
+    .fstat = error_gen_fstat,
+    .lk = error_gen_lk,
+    .xattrop = error_gen_xattrop,
+    .fxattrop = error_gen_fxattrop,
+    .inodelk = error_gen_inodelk,
+    .finodelk = error_gen_finodelk,
+    .entrylk = error_gen_entrylk,
+    .fentrylk = error_gen_fentrylk,
+    .setattr = error_gen_setattr,
+    .fsetattr = error_gen_fsetattr,
+    .getspec = error_gen_getspec,
 };
 
 struct volume_options options[] = {
-        { .key  = {"failure"},
-          .type = GF_OPTION_TYPE_INT },
-        { .key  = {"error-no"},
-          .value = {"ENOENT","ENOTDIR","ENAMETOOLONG","EACCES","EBADF",
-                    "EFAULT","ENOMEM","EINVAL","EIO","EEXIST","ENOSPC",
-                    "EPERM","EROFS","EBUSY","EISDIR","ENOTEMPTY","EMLINK"
-                    "ENODEV","EXDEV","EMFILE","ENFILE","ENOSYS","EINTR",
-                    "EFBIG","EAGAIN"},
-          .type = GF_OPTION_TYPE_STR },
-        { .key  = {"enable"},
-          .type = GF_OPTION_TYPE_STR },
-        { .key  = {NULL} }
+    {
+        .key = {"failure"},
+        .type = GF_OPTION_TYPE_PERCENT,
+        .description = "Percentage failure of operations when enabled.",
+    },
+
+    {
+        .key = {"error-no"},
+        .value = {"ENOENT",
+                  "ENOTDIR",
+                  "ENAMETOOLONG",
+                  "EACCES",
+                  "EBADF",
+                  "EFAULT",
+                  "ENOMEM",
+                  "EINVAL",
+                  "EIO",
+                  "EEXIST",
+                  "ENOSPC",
+                  "EPERM",
+                  "EROFS",
+                  "EBUSY",
+                  "EISDIR",
+                  "ENOTEMPTY",
+                  "EMLINK"
+                  "ENODEV",
+                  "EXDEV",
+                  "EMFILE",
+                  "ENFILE",
+                  "ENOSYS",
+                  "EINTR",
+                  "EFBIG",
+                  "EAGAIN",
+                  "GF_ERROR_SHORT_WRITE"},
+        .type = GF_OPTION_TYPE_STR,
+        .op_version = {3},
+        .tags = {"error-gen"},
+        .flags = OPT_FLAG_SETTABLE,
+
+    },
+
+    {
+        .key = {"random-failure"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "off",
+        .op_version = {3},
+        .tags = {"error-gen"},
+        .flags = OPT_FLAG_SETTABLE,
+    },
+
+    {
+        .key = {"enable", "error-fops"},
+        .type = GF_OPTION_TYPE_STR,
+        .description = "Accepts a string which takes ',' separated fop "
+                       "strings to denote which fops are enabled for error",
+        .op_version = {3},
+        .tags = {"error-gen"},
+        .flags = OPT_FLAG_SETTABLE,
+    },
+
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1},
+    .dumpops = &dumpops,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "error-gen",
+    .category = GF_TECH_PREVIEW,
 };
diff --git a/xlators/debug/error-gen/src/error-gen.h b/xlators/debug/error-gen/src/error-gen.h
index a27c2de5045..2478cd5b21c 100644
--- a/xlators/debug/error-gen/src/error-gen.h
+++ b/xlators/debug/error-gen/src/error-gen.h
@@ -1,43 +1,49 @@
 /*
-  Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
 */
-
 #ifndef _ERROR_GEN_H
 #define _ERROR_GEN_H
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+#include "error-gen-mem-types.h"
 
 #define GF_FAILURE_DEFAULT 10
 
+/*
+ * Pseudo-errors refer to errors beyond the scope of traditional <-1, op_errno>
+ * returns. This facilitates the ability to return unexpected, but not -1 values
+ * and/or to inject operations that lead to implicit error conditions. The range
+ * for pseudo errors resides at a high value to avoid conflicts with the errno
+ * range.
+ */
+enum GF_PSEUDO_ERRORS {
+    GF_ERROR_SHORT_WRITE = 1000, /* short writev return value */
+    GF_ERROR_MAX
+};
+
 typedef struct {
-        int enable[GF_FOP_MAXVALUE];
-        int op_count;
-        int failure_iter_no;
-        char *error_no;
-        gf_lock_t lock;
+    int enable[GF_FOP_MAXVALUE];
+    int op_count;
+    /*
+     * This is only an iteration number in the random-failure case.  For
+     * the normal controlled-probability case, it's actually a numerator
+     * for the failure probability (see FAILURE_GRANULARITY declaration).
+     * It's just not worth blowing up the diff by changing it.
+     */
+    int failure_iter_no;
+    int error_no_int;
+    gf_boolean_t random_failure;
+    gf_lock_t lock;
 } eg_t;
 
 typedef struct {
-        int error_no_count;
-	int error_no[20];
+    int error_no_count;
+    int error_no[20];
 } sys_error_t;
 
 #endif
diff --git a/xlators/debug/io-stats/src/Makefile.am b/xlators/debug/io-stats/src/Makefile.am
index b894e79c3fe..c69f3caf0fe 100644
--- a/xlators/debug/io-stats/src/Makefile.am
+++ b/xlators/debug/io-stats/src/Makefile.am
@@ -2,14 +2,18 @@
 xlator_LTLIBRARIES = io-stats.la
 xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/debug
 
-io_stats_la_LDFLAGS = -module -avoidversion
+io_stats_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
 
 io_stats_la_SOURCES = io-stats.c
 io_stats_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
 noinst_HEADERS = io-stats-mem-types.h
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS)\
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src \
+	-I$(top_srcdir)/rpc/rpc-lib/src \
+	-DDATADIR=\"$(localstatedir)\"
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
 
 CLEANFILES =
diff --git a/xlators/debug/io-stats/src/io-stats-mem-types.h b/xlators/debug/io-stats/src/io-stats-mem-types.h
index e53d2d22a61..51d38d8b97c 100644
--- a/xlators/debug/io-stats/src/io-stats-mem-types.h
+++ b/xlators/debug/io-stats/src/io-stats-mem-types.h
@@ -1,33 +1,27 @@
-
 /*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
    This file is part of GlusterFS.
 
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
 */
 
-
 #ifndef __IO_STATS_MEM_TYPES_H__
 #define __IO_STATS_MEM_TYPES_H__
 
-#include "mem-types.h"
+#include <glusterfs/mem-types.h>
+
+extern const char *__progname;
 
 enum gf_io_stats_mem_types_ {
-        gf_io_stats_mt_ios_conf = gf_common_mt_end + 1,
-        gf_io_stats_mt_ios_fd,
-        gf_io_stats_mt_end
+    gf_io_stats_mt_ios_conf = gf_common_mt_end + 1,
+    gf_io_stats_mt_ios_fd,
+    gf_io_stats_mt_ios_stat,
+    gf_io_stats_mt_ios_stat_list,
+    gf_io_stats_mt_ios_sample_buf,
+    gf_io_stats_mt_ios_sample,
+    gf_io_stats_mt_end
 };
 #endif
-
diff --git a/xlators/debug/io-stats/src/io-stats.c b/xlators/debug/io-stats/src/io-stats.c
index e8733436466..aa00c446e5a 100644
--- a/xlators/debug/io-stats/src/io-stats.c
+++ b/xlators/debug/io-stats/src/io-stats.c
@@ -1,27 +1,14 @@
 /*
-  Copyright (c) 2006-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
+   Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#include "xlator.h"
-#endif
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include <glusterfs/xlator.h>
+#include <glusterfs/syscall.h>
 
 /**
  * xlators/debug/io_stats :
@@ -30,1736 +17,4464 @@
  *
  *  a) total read data - since process start, last interval and per fd
  *  b) total write data - since process start, last interval and per fd
- *  c) counts of read IO block size - since process start, last interval and per fd
- *  d) counts of write IO block size - since process start, last interval and per fd
- *  e) counts of all FOP types passing through it
+ *  c) counts of read IO block size - since process start, last interval and per
+ * fd d) counts of write IO block size - since process start, last interval and
+ * per fd e) counts of all FOP types passing through it
  *
- *  Usage: setfattr -n io-stats-dump /tmp/filename /mnt/gluster
+ *  Usage: setfattr -n trusted.io-stats-dump /tmp/filename /mnt/gluster
+ *      output is written to /tmp/filename.<iostats xlator instance name>
  *
  */
 
 #include <fnmatch.h>
 #include <errno.h>
-#include "glusterfs.h"
-#include "xlator.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/xlator.h>
 #include "io-stats-mem-types.h"
+#include <stdarg.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/statedump.h>
+#include <glusterfs/syncop.h>
+#include <pwd.h>
+#include <grp.h>
+#include <glusterfs/upcall-utils.h>
+#include <glusterfs/async.h>
+
+#define MAX_LIST_MEMBERS 100
+#define DEFAULT_PWD_BUF_SZ 16384
+#define DEFAULT_GRP_BUF_SZ 16384
+#define IOS_BLOCK_COUNT_SIZE 32
+
+#define IOS_STATS_DUMP_DIR DEFAULT_VAR_RUN_DIRECTORY
+
+typedef enum {
+    IOS_STATS_TYPE_NONE,
+    IOS_STATS_TYPE_OPEN,
+    IOS_STATS_TYPE_READ,
+    IOS_STATS_TYPE_WRITE,
+    IOS_STATS_TYPE_OPENDIR,
+    IOS_STATS_TYPE_READDIRP,
+    IOS_STATS_TYPE_READ_THROUGHPUT,
+    IOS_STATS_TYPE_WRITE_THROUGHPUT,
+    IOS_STATS_TYPE_MAX
+} ios_stats_type_t;
+
+typedef enum {
+    IOS_STATS_THRU_READ,
+    IOS_STATS_THRU_WRITE,
+    IOS_STATS_THRU_MAX,
+} ios_stats_thru_t;
+
+/* This is same as gf1_cli_info_op */
+/* had to be defined here again, so we have modularity between
+ xdr, xlator, and library functions */
+typedef enum ios_info_op {
+    GF_IOS_INFO_NONE = 0,
+    GF_IOS_INFO_ALL = 1,
+    GF_IOS_INFO_INCREMENTAL = 2,
+    GF_IOS_INFO_CUMULATIVE = 3,
+    GF_IOS_INFO_CLEAR = 4,
+} ios_info_op_t;
+
+struct ios_stat_lat {
+    struct timeval time;
+    double throughput;
+};
+
+struct ios_stat {
+    gf_lock_t lock;
+    uuid_t gfid;
+    char *filename;
+    gf_atomic_t counters[IOS_STATS_TYPE_MAX];
+    struct ios_stat_lat thru_counters[IOS_STATS_THRU_MAX];
+    gf_atomic_t refcnt;
+};
+
+struct ios_stat_list {
+    struct list_head list;
+    struct ios_stat *iosstat;
+    double value;
+};
+
+struct ios_stat_head {
+    gf_lock_t lock;
+    double min_cnt;
+    uint64_t members;
+    struct ios_stat_list *iosstats;
+};
+
+typedef struct _ios_sample_t {
+    uid_t uid;
+    gid_t gid;
+    char identifier[UNIX_PATH_MAX];
+    glusterfs_fop_t fop_type;
+    struct timeval timestamp;
+    double elapsed;
+} ios_sample_t;
+
+typedef struct _ios_sample_buf_t {
+    uint64_t pos;              /* Position in write buffer */
+    uint64_t size;             /* Size of ring buffer */
+    uint64_t collected;        /* Number of samples we've collected */
+    uint64_t observed;         /* Number of FOPs we've observed */
+    ios_sample_t *ios_samples; /* Our list of samples */
+} ios_sample_buf_t;
 
 struct ios_lat {
-        double  min;
-        double  max;
-        double  avg;
+    double min;
+    double max;
+    double avg;
+    uint64_t total;
 };
 
 struct ios_global_stats {
-        uint64_t        data_written;
-        uint64_t        data_read;
-        uint64_t        block_count_write[32];
-        uint64_t        block_count_read[32];
-        uint64_t        fop_hits[GF_FOP_MAXVALUE];
-        struct timeval  started_at;
-        struct ios_lat  latency[GF_FOP_MAXVALUE];
+    gf_atomic_t data_written;
+    gf_atomic_t data_read;
+    gf_atomic_t block_count_write[IOS_BLOCK_COUNT_SIZE];
+    gf_atomic_t block_count_read[IOS_BLOCK_COUNT_SIZE];
+    gf_atomic_t fop_hits[GF_FOP_MAXVALUE];
+    gf_atomic_t upcall_hits[GF_UPCALL_FLAGS_MAXVALUE];
+    time_t started_at;
+    struct ios_lat latency[GF_FOP_MAXVALUE];
+    uint64_t nr_opens;
+    uint64_t max_nr_opens;
+    struct timeval max_openfd_time;
 };
 
+typedef enum {
+    IOS_DUMP_TYPE_NONE = 0,
+    IOS_DUMP_TYPE_FILE = 1,
+    IOS_DUMP_TYPE_DICT = 2,
+    IOS_DUMP_TYPE_JSON_FILE = 3,
+    IOS_DUMP_TYPE_SAMPLES = 4,
+    IOS_DUMP_TYPE_MAX = 5
+} ios_dump_type_t;
 
 struct ios_conf {
-        gf_lock_t                 lock;
-        struct ios_global_stats   cumulative;
-        uint64_t                  increment;
-        struct ios_global_stats   incremental;
-        gf_boolean_t              dump_fd_stats;
-        int                       measure_latency;
+    gf_lock_t lock;
+    struct ios_global_stats cumulative;
+    uint64_t increment;
+    struct ios_global_stats incremental;
+    gf_boolean_t dump_fd_stats;
+    gf_boolean_t count_fop_hits;
+    gf_boolean_t measure_latency;
+    struct ios_stat_head list[IOS_STATS_TYPE_MAX];
+    struct ios_stat_head thru_list[IOS_STATS_THRU_MAX];
+    int32_t ios_dump_interval;
+    pthread_t dump_thread;
+    gf_boolean_t dump_thread_should_die;
+    gf_boolean_t dump_thread_running;
+    gf_lock_t ios_sampling_lock;
+    int32_t ios_sample_interval;
+    int32_t ios_sample_buf_size;
+    ios_sample_buf_t *ios_sample_buf;
+    struct dnscache *dnscache;
+    int32_t ios_dnscache_ttl_sec;
+    /*
+     * What we really need here is just a unique value to keep files
+     * created by this instance distinct from those created by any other.
+     * On the client side this isn't a problem, so we just use the
+     * translator name.  On the server side conflicts can occur, so the
+     * volfile-generation code automatically sets this (via an option)
+     * to be the brick path.
+     *
+     * NB While the *field* name has changed, it didn't seem worth changing
+     * all of the cases where "xlator_name" is used as a *variable* name.
+     */
+    char *unique_id;
+    ios_dump_type_t dump_format;
 };
 
-
 struct ios_fd {
-        char           *filename;
-        uint64_t        data_written;
-        uint64_t        data_read;
-        uint64_t        block_count_write[32];
-        uint64_t        block_count_read[32];
-        struct timeval  opened_at;
+    char *filename;
+    gf_atomic_t data_written;
+    gf_atomic_t data_read;
+    gf_atomic_t block_count_write[IOS_BLOCK_COUNT_SIZE];
+    gf_atomic_t block_count_read[IOS_BLOCK_COUNT_SIZE];
+    struct timeval opened_at;
+};
+
+struct ios_dump_args {
+    ios_dump_type_t type;
+    union {
+        FILE *logfp;
+        dict_t *dict;
+    } u;
 };
 
+typedef int (*block_dump_func)(xlator_t *, struct ios_dump_args *, int, int,
+                               uint64_t);
 
 struct ios_local {
-        struct timeval  wind_at;
-        struct timeval  unwind_at;
+    struct timeval wind_at;
+    struct timeval unwind_at;
 };
 
-#define END_FOP_LATENCY(frame, op)                                      \
-        do {                                                            \
-                struct ios_conf  *conf = NULL;                          \
-                                                                        \
-                conf = this->private;                                   \
-                if (conf->measure_latency) {                            \
-                        gettimeofday (&frame->end, NULL);        \
-                        update_ios_latency (conf, frame, GF_FOP_##op);  \
-                }                                                       \
-        } while (0)
-
-#define START_FOP_LATENCY(frame)                                        \
-        do {                                                            \
-                struct ios_conf  *conf = NULL;                          \
-                                                                        \
-                conf = this->private;                                   \
-                if (conf->measure_latency) {                            \
-                        gettimeofday (&frame->begin, NULL);      \
-                }                                                       \
-        } while (0)
-
-
-#define BUMP_FOP(op)                                                    \
-        do {                                                            \
-                struct ios_conf  *conf = NULL;                          \
-                                                                        \
-                conf = this->private;                                   \
-                LOCK (&conf->lock);                                     \
-                {                                                       \
-                        conf->cumulative.fop_hits[GF_FOP_##op]++;       \
-                        conf->incremental.fop_hits[GF_FOP_##op]++;      \
-                }                                                       \
-                UNLOCK (&conf->lock);                                   \
-        } while (0)
-
-
-#define BUMP_READ(fd, len)                                              \
-        do {                                                            \
-                struct ios_conf  *conf = NULL;                          \
-                struct ios_fd    *iosfd = NULL;                         \
-                int               lb2 = 0;                              \
-                                                                        \
-                conf = this->private;                                   \
-                lb2 = log_base2 (len);                                  \
-                ios_fd_ctx_get (fd, this, &iosfd);                      \
-                                                                        \
-                LOCK (&conf->lock);                                     \
-                {                                                       \
-                        conf->cumulative.data_read += len;              \
-                        conf->incremental.data_read += len;             \
-                        conf->cumulative.block_count_read[lb2]++;       \
-                        conf->incremental.block_count_read[lb2]++;      \
-                                                                        \
-                        if (iosfd) {                                    \
-                                iosfd->data_read += len;                \
-                                iosfd->block_count_read[lb2]++;         \
-                        }                                               \
-                }                                                       \
-                UNLOCK (&conf->lock);                                   \
-        } while (0)
-
-
-#define BUMP_WRITE(fd, len)                                             \
-        do {                                                            \
-                struct ios_conf  *conf = NULL;                          \
-                struct ios_fd    *iosfd = NULL;                         \
-                int               lb2 = 0;                              \
-                                                                        \
-                conf = this->private;                                   \
-                lb2 = log_base2 (len);                                  \
-                ios_fd_ctx_get (fd, this, &iosfd);                      \
-                                                                        \
-                LOCK (&conf->lock);                                     \
-                {                                                       \
-                        conf->cumulative.data_written += len;           \
-                        conf->incremental.data_written += len;          \
-                        conf->cumulative.block_count_write[lb2]++;      \
-                        conf->incremental.block_count_write[lb2]++;     \
-                                                                        \
-                        if (iosfd) {                                    \
-                                iosfd->data_written += len;             \
-                                iosfd->block_count_write[lb2]++;        \
-                        }                                               \
-                }                                                       \
-                UNLOCK (&conf->lock);                                   \
-        } while (0)
-
-
-int
-ios_fd_ctx_get (fd_t *fd, xlator_t *this, struct ios_fd **iosfd)
-{
-        uint64_t      iosfd64 = 0;
-        unsigned long iosfdlong = 0;
-        int           ret = 0;
-
-        ret = fd_ctx_get (fd, this, &iosfd64);
-        iosfdlong = iosfd64;
-        if (ret != -1)
-                *iosfd = (void *) iosfdlong;
+struct volume_options options[];
 
-        return ret;
+static int
+is_fop_latency_started(call_frame_t *frame)
+{
+    GF_ASSERT(frame);
+    struct timeval epoch = {
+        0,
+    };
+    return memcmp(&frame->begin, &epoch, sizeof(epoch));
 }
 
+#define _IOS_SAMP_DIR DEFAULT_LOG_FILE_DIRECTORY "/samples"
+#ifdef GF_LINUX_HOST_OS
+#define _IOS_DUMP_DIR DATADIR "/lib/glusterd/stats"
+#else
+#define _IOS_DUMP_DIR DATADIR "/db/glusterd/stats"
+#endif
 
-
-int
-ios_fd_ctx_set (fd_t *fd, xlator_t *this, struct ios_fd *iosfd)
+#define END_FOP_LATENCY(frame, op)                                             \
+    do {                                                                       \
+        struct ios_conf *conf = NULL;                                          \
+                                                                               \
+        conf = this->private;                                                  \
+        if (conf && conf->measure_latency) {                                   \
+            timespec_now(&frame->end);                                         \
+            update_ios_latency(conf, frame, GF_FOP_##op);                      \
+        }                                                                      \
+    } while (0)
+
+#define START_FOP_LATENCY(frame)                                               \
+    do {                                                                       \
+        struct ios_conf *conf = NULL;                                          \
+                                                                               \
+        conf = this->private;                                                  \
+        if (conf && conf->measure_latency) {                                   \
+            timespec_now(&frame->begin);                                       \
+        } else {                                                               \
+            memset(&frame->begin, 0, sizeof(frame->begin));                    \
+        }                                                                      \
+    } while (0)
+
+#define BUMP_FOP(op)                                                           \
+    do {                                                                       \
+        struct ios_conf *conf = NULL;                                          \
+                                                                               \
+        conf = this->private;                                                  \
+        if (!conf)                                                             \
+            break;                                                             \
+        GF_ATOMIC_INC(conf->cumulative.fop_hits[GF_FOP_##op]);                 \
+        GF_ATOMIC_INC(conf->incremental.fop_hits[GF_FOP_##op]);                \
+    } while (0)
+
+#define UPDATE_PROFILE_STATS(frame, op)                                        \
+    do {                                                                       \
+        struct ios_conf *conf = NULL;                                          \
+                                                                               \
+        if (!is_fop_latency_started(frame))                                    \
+            break;                                                             \
+        conf = this->private;                                                  \
+        if (conf && conf->measure_latency && conf->count_fop_hits) {           \
+            BUMP_FOP(op);                                                      \
+            timespec_now(&frame->end);                                         \
+            update_ios_latency(conf, frame, GF_FOP_##op);                      \
+        }                                                                      \
+    } while (0)
+
+#define BUMP_THROUGHPUT(iosstat, type)                                         \
+    do {                                                                       \
+        struct ios_conf *conf = NULL;                                          \
+        double elapsed;                                                        \
+        struct timespec *begin, *end;                                          \
+        double throughput;                                                     \
+        int flag = 0;                                                          \
+        struct timeval tv = {                                                  \
+            0,                                                                 \
+        };                                                                     \
+                                                                               \
+        begin = &frame->begin;                                                 \
+        end = &frame->end;                                                     \
+                                                                               \
+        elapsed = gf_tsdiff(begin, end) / 1000.0;                              \
+        throughput = op_ret / elapsed;                                         \
+                                                                               \
+        conf = this->private;                                                  \
+        gettimeofday(&tv, NULL);                                               \
+        LOCK(&iosstat->lock);                                                  \
+        {                                                                      \
+            if (iosstat->thru_counters[type].throughput <= throughput) {       \
+                iosstat->thru_counters[type].throughput = throughput;          \
+                memcpy(&iosstat->thru_counters[type].time, &tv,                \
+                       sizeof(struct timeval));                                \
+                flag = 1;                                                      \
+            }                                                                  \
+        }                                                                      \
+        UNLOCK(&iosstat->lock);                                                \
+        if (flag)                                                              \
+            ios_stat_add_to_list(&conf->thru_list[type], throughput, iosstat); \
+    } while (0)
+
+static int
+ios_fd_ctx_get(fd_t *fd, xlator_t *this, struct ios_fd **iosfd)
 {
-        uint64_t   iosfd64 = 0;
-        int        ret = 0;
+    uint64_t iosfd64 = 0;
+    unsigned long iosfdlong = 0;
+    int ret = 0;
 
-        iosfd64 = (unsigned long) iosfd;
-        ret = fd_ctx_set (fd, this, iosfd64);
+    ret = fd_ctx_get(fd, this, &iosfd64);
+    iosfdlong = iosfd64;
+    if (ret != -1)
+        *iosfd = (void *)iosfdlong;
 
-        return ret;
+    return ret;
 }
 
+static int
+ios_fd_ctx_set(fd_t *fd, xlator_t *this, struct ios_fd *iosfd)
+{
+    uint64_t iosfd64 = 0;
+    int ret = 0;
 
-#define ios_log(this, logfp, fmt ...)                           \
-        do {                                                    \
-                if (logfp) {                                    \
-                        fprintf (logfp, fmt);                   \
-                        fprintf (logfp, "\n");                  \
-                }                                               \
-                gf_log (this->name, GF_LOG_NORMAL, fmt);        \
-        } while (0)
+    iosfd64 = (unsigned long)iosfd;
+    ret = fd_ctx_set(fd, this, iosfd64);
 
+    return ret;
+}
 
-int
-io_stats_dump_global (xlator_t *this, struct ios_global_stats *stats,
-                      struct timeval *now, int interval, FILE *logfp)
+static int
+ios_stat_ref(struct ios_stat *iosstat)
 {
-        int    i = 0;
+    uint64_t refcnt = 0;
+    refcnt = GF_ATOMIC_INC(iosstat->refcnt);
 
-        if (interval == -1)
-                ios_log (this, logfp, "=== Cumulative stats ===");
-        else
-                ios_log (this, logfp, "=== Interval %d stats ===",
-                         interval);
-        ios_log (this, logfp, "      Duration : %"PRId64"secs",
-                 (uint64_t) (now->tv_sec - stats->started_at.tv_sec));
-        ios_log (this, logfp, "     BytesRead : %"PRId64,
-                 stats->data_read);
-        ios_log (this, logfp, "  BytesWritten : %"PRId64,
-                 stats->data_written);
-
-        for (i = 0; i < 32; i++) {
-                if (stats->block_count_read[i])
-                        ios_log (this, logfp, " Read %06db+ : %"PRId64,
-                                 (1 << i), stats->block_count_read[i]);
-        }
+    return refcnt;
+}
 
-        for (i = 0; i < 32; i++) {
-                if (stats->block_count_write[i])
-                        ios_log (this, logfp, "Write %06db+ : %"PRId64,
-                                 (1 << i), stats->block_count_write[i]);
+static int
+ios_stat_unref(struct ios_stat *iosstat)
+{
+    int cleanup = 0;
+    uint64_t refcnt = 0;
+
+    refcnt = GF_ATOMIC_DEC(iosstat->refcnt);
+    if (refcnt == 0) {
+        if (iosstat->filename) {
+            GF_FREE(iosstat->filename);
+            iosstat->filename = NULL;
         }
+        cleanup = 1;
+    }
+
+    if (cleanup) {
+        LOCK_DESTROY(&iosstat->lock);
+        GF_FREE(iosstat);
+        iosstat = NULL;
+    }
+
+    return 0;
+}
 
-        for (i = 0; i < GF_FOP_MAXVALUE; i++) {
-                if (stats->fop_hits[i] && !stats->latency[i].avg)
-                        ios_log (this, logfp, "%14s : %"PRId64,
-                                 gf_fop_list[i], stats->fop_hits[i]);
-                else if (stats->fop_hits[i] && stats->latency[i].avg)
-                        ios_log (this, logfp, "%14s : %"PRId64 ", latency"
-                                 "(avg: %f, min: %f, max: %f)",
-                                 gf_fop_list[i], stats->fop_hits[i],
-                                 stats->latency[i].avg, stats->latency[i].min,
-                                 stats->latency[i].max);
+static int
+ios_stat_add_to_list(struct ios_stat_head *list_head, uint64_t value,
+                     struct ios_stat *iosstat)
+{
+    struct ios_stat_list *new = NULL;
+    struct ios_stat_list *entry = NULL;
+    struct ios_stat_list *t = NULL;
+    struct ios_stat_list *list_entry = NULL;
+    struct ios_stat_list *tmp = NULL;
+    struct ios_stat_list *last = NULL;
+    struct ios_stat *stat = NULL;
+    int cnt = 0;
+    int found = 0;
+    int reposition = 0;
+    double min_count = 0;
+
+    LOCK(&list_head->lock);
+    {
+        if (list_head->min_cnt == 0)
+            list_head->min_cnt = value;
+        if ((list_head->members == MAX_LIST_MEMBERS) &&
+            (list_head->min_cnt > value))
+            goto out;
+
+        list_for_each_entry_safe(entry, t, &list_head->iosstats->list, list)
+        {
+            cnt++;
+            if (cnt == list_head->members)
+                last = entry;
+
+            if (!gf_uuid_compare(iosstat->gfid, entry->iosstat->gfid)) {
+                list_entry = entry;
+                found = cnt;
+                entry->value = value;
+                if (!reposition) {
+                    if (cnt == list_head->members)
+                        list_head->min_cnt = value;
+                    goto out;
+                }
+                break;
+            } else if (entry->value <= value && !reposition) {
+                reposition = cnt;
+                tmp = entry;
+                if (cnt == list_head->members - 1)
+                    min_count = entry->value;
+            }
+        }
+        if (found) {
+            list_del(&list_entry->list);
+            list_add_tail(&list_entry->list, &tmp->list);
+            if (min_count)
+                list_head->min_cnt = min_count;
+            goto out;
+        } else if (list_head->members == MAX_LIST_MEMBERS && reposition) {
+            new = GF_CALLOC(1, sizeof(*new), gf_io_stats_mt_ios_stat_list);
+            new->iosstat = iosstat;
+            new->value = value;
+            ios_stat_ref(iosstat);
+            list_add_tail(&new->list, &tmp->list);
+            if (last) {
+                stat = last->iosstat;
+                last->iosstat = NULL;
+                ios_stat_unref(stat);
+                list_del(&last->list);
+                GF_FREE(last);
+            }
+            if (reposition == MAX_LIST_MEMBERS)
+                list_head->min_cnt = value;
+            else if (min_count) {
+                list_head->min_cnt = min_count;
+            }
+        } else if (list_head->members < MAX_LIST_MEMBERS) {
+            new = GF_CALLOC(1, sizeof(*new), gf_io_stats_mt_ios_stat_list);
+            new->iosstat = iosstat;
+            new->value = value;
+            ios_stat_ref(iosstat);
+            if (reposition) {
+                list_add_tail(&new->list, &tmp->list);
+            } else {
+                list_add_tail(&new->list, &entry->list);
+            }
+            list_head->members++;
+            if (list_head->min_cnt > value)
+                list_head->min_cnt = value;
         }
+    }
+out:
+    UNLOCK(&list_head->lock);
+    return 0;
+}
 
-        return 0;
+static void
+ios_bump_read(xlator_t *this, fd_t *fd, size_t len)
+{
+    struct ios_conf *conf = NULL;
+    struct ios_fd *iosfd = NULL;
+    int lb2 = 0;
+
+    conf = this->private;
+    lb2 = log_base2(len);
+    ios_fd_ctx_get(fd, this, &iosfd);
+    if (!conf)
+        return;
+
+    GF_ATOMIC_ADD(conf->cumulative.data_read, len);
+    GF_ATOMIC_ADD(conf->incremental.data_read, len);
+    GF_ATOMIC_INC(conf->cumulative.block_count_read[lb2]);
+    GF_ATOMIC_INC(conf->incremental.block_count_read[lb2]);
+
+    if (iosfd) {
+        GF_ATOMIC_ADD(iosfd->data_read, len);
+        GF_ATOMIC_INC(iosfd->block_count_read[lb2]);
+    }
+}
+
+static void
+ios_bump_write(xlator_t *this, fd_t *fd, size_t len)
+{
+    struct ios_conf *conf = NULL;
+    struct ios_fd *iosfd = NULL;
+    int lb2 = 0;
+
+    conf = this->private;
+    lb2 = log_base2(len);
+    ios_fd_ctx_get(fd, this, &iosfd);
+    if (!conf)
+        return;
+
+    GF_ATOMIC_ADD(conf->cumulative.data_written, len);
+    GF_ATOMIC_ADD(conf->incremental.data_written, len);
+    GF_ATOMIC_INC(conf->cumulative.block_count_write[lb2]);
+    GF_ATOMIC_INC(conf->incremental.block_count_write[lb2]);
+
+    if (iosfd) {
+        GF_ATOMIC_ADD(iosfd->data_written, len);
+        GF_ATOMIC_INC(iosfd->block_count_write[lb2]);
+    }
+}
+
+static void
+ios_bump_upcall(xlator_t *this, gf_upcall_flags_t event)
+{
+    struct ios_conf *conf = NULL;
+
+    conf = this->private;
+    if (!conf)
+        return;
+    if (conf->count_fop_hits) {
+        GF_ATOMIC_INC(conf->cumulative.upcall_hits[event]);
+        GF_ATOMIC_INC(conf->incremental.upcall_hits[event]);
+    }
 }
 
+static void
+ios_bump_stats(xlator_t *this, struct ios_stat *iosstat, ios_stats_type_t type)
+{
+    struct ios_conf *conf = NULL;
+    uint64_t value = 0;
+
+    conf = this->private;
+
+    value = GF_ATOMIC_INC(iosstat->counters[type]);
+    ios_stat_add_to_list(&conf->list[type], value, iosstat);
+}
 
 int
-io_stats_dump (xlator_t *this, char *filename, inode_t *inode,
-               const char *path)
+ios_inode_ctx_set(inode_t *inode, xlator_t *this, struct ios_stat *iosstat)
 {
-        struct ios_conf         *conf = NULL;
-        struct ios_global_stats  cumulative = {0, };
-        struct ios_global_stats  incremental = {0, };
-        int                      increment = 0;
-        struct timeval           now;
-        FILE                    *logfp = NULL;
+    uint64_t iosstat64 = 0;
+    int ret = 0;
 
-        conf = this->private;
+    ios_stat_ref(iosstat);
+    iosstat64 = (unsigned long)iosstat;
+    ret = inode_ctx_put(inode, this, iosstat64);
+    return ret;
+}
 
-        gettimeofday (&now, NULL);
-        LOCK (&conf->lock);
-        {
-                cumulative  = conf->cumulative;
-                incremental = conf->incremental;
+int
+ios_inode_ctx_get(inode_t *inode, xlator_t *this, struct ios_stat **iosstat)
+{
+    uint64_t iosstat64 = 0;
+    unsigned long iosstatlong = 0;
+    int ret = 0;
 
-                increment = conf->increment++;
+    ret = inode_ctx_get(inode, this, &iosstat64);
+    iosstatlong = iosstat64;
+    if (ret != -1)
+        *iosstat = (void *)iosstatlong;
 
-                memset (&conf->incremental, 0, sizeof (conf->incremental));
-                conf->incremental.started_at = now;
-        }
-        UNLOCK (&conf->lock);
+    return ret;
+}
+
+/*
+ * So why goto all this trouble?  Why not just queue up some samples in
+ * a big list and malloc away?  Well malloc is expensive relative
+ * to what we are measuring, so cannot have any malloc's (or worse
+ * callocs) in our measurement code paths.  Instead, we are going to
+ * pre-allocate a circular buffer and collect a maximum number of samples.
+ * Prior to dumping them all we'll create a new buffer and swap the
+ * old buffer with the new, and then proceed to dump the statistics
+ * in our dump thread.
+ *
+ */
+ios_sample_buf_t *
+ios_create_sample_buf(size_t buf_size)
+{
+    ios_sample_buf_t *ios_sample_buf = NULL;
+    ios_sample_t *ios_samples = NULL;
+
+    ios_sample_buf = GF_CALLOC(1, sizeof(*ios_sample_buf),
+                               gf_io_stats_mt_ios_sample_buf);
+    if (!ios_sample_buf)
+        goto err;
+
+    ios_samples = GF_CALLOC(buf_size, sizeof(*ios_samples),
+                            gf_io_stats_mt_ios_sample);
+
+    if (!ios_samples)
+        goto err;
+
+    ios_sample_buf->ios_samples = ios_samples;
+    ios_sample_buf->size = buf_size;
+    ios_sample_buf->pos = 0;
+    ios_sample_buf->observed = 0;
+    ios_sample_buf->collected = 0;
+
+    return ios_sample_buf;
+err:
+    GF_FREE(ios_sample_buf);
+    return NULL;
+}
 
-        logfp = fopen (filename, "w+");
-        io_stats_dump_global (this, &cumulative, &now, -1, logfp);
-        io_stats_dump_global (this, &incremental, &now, increment, logfp);
+void
+ios_destroy_sample_buf(ios_sample_buf_t *ios_sample_buf)
+{
+    GF_FREE(ios_sample_buf->ios_samples);
+    GF_FREE(ios_sample_buf);
+}
 
-        if (logfp)
-                fclose (logfp);
+static int
+ios_init_sample_buf(struct ios_conf *conf)
+{
+    int32_t ret = -1;
+
+    GF_ASSERT(conf);
+    LOCK(&conf->lock);
+    conf->ios_sample_buf = ios_create_sample_buf(conf->ios_sample_buf_size);
+    if (!conf->ios_sample_buf)
+        goto out;
+    ret = 0;
+out:
+    UNLOCK(&conf->lock);
+    return ret;
+}
+
+static int
+ios_stats_cleanup(xlator_t *this, inode_t *inode)
+{
+    struct ios_stat *iosstat = NULL;
+    uint64_t iosstat64 = 0;
+
+    inode_ctx_del(inode, this, &iosstat64);
+    if (!iosstat64) {
+        gf_log(this->name, GF_LOG_WARNING, "could not get inode ctx");
         return 0;
+    }
+    iosstat = (void *)(long)iosstat64;
+    if (iosstat) {
+        ios_stat_unref(iosstat);
+    }
+    return 0;
 }
 
+#define ios_log(this, logfp, fmt...)                                           \
+    do {                                                                       \
+        if (logfp) {                                                           \
+            fprintf(logfp, fmt);                                               \
+            fprintf(logfp, "\n");                                              \
+        }                                                                      \
+        gf_log(this->name, GF_LOG_DEBUG, fmt);                                 \
+    } while (0)
 
 int
-io_stats_dump_fd (xlator_t *this, struct ios_fd *iosfd)
+ios_dump_file_stats(struct ios_stat_head *list_head, xlator_t *this,
+                    FILE *logfp)
 {
-        struct ios_conf         *conf = NULL;
-        struct timeval           now;
-        uint64_t                 sec = 0;
-        uint64_t                 usec = 0;
-        int                      i = 0;
+    struct ios_stat_list *entry = NULL;
 
-        conf = this->private;
+    LOCK(&list_head->lock);
+    {
+        list_for_each_entry(entry, &list_head->iosstats->list, list)
+        {
+            ios_log(this, logfp, "%-12.0f %s", entry->value,
+                    entry->iosstat->filename);
+        }
+    }
+    UNLOCK(&list_head->lock);
+    return 0;
+}
+
+int
+ios_dump_throughput_stats(struct ios_stat_head *list_head, xlator_t *this,
+                          FILE *logfp, ios_stats_thru_t type)
+{
+    struct ios_stat_list *entry = NULL;
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
+
+    LOCK(&list_head->lock);
+    {
+        list_for_each_entry(entry, &list_head->iosstats->list, list)
+        {
+            gf_time_fmt_tv(timestr, sizeof timestr,
+                           &entry->iosstat->thru_counters[type].time,
+                           gf_timefmt_FT);
+
+            ios_log(this, logfp, "%s \t %-10.2f  \t  %s", timestr, entry->value,
+                    entry->iosstat->filename);
+        }
+    }
+    UNLOCK(&list_head->lock);
+    return 0;
+}
 
-        if (!conf->dump_fd_stats)
-                return 0;
+int
+_io_stats_get_key_prefix(xlator_t *this, char **key_prefix)
+{
+    char *key_root = "gluster";
+    char *xlator_name = NULL;
+    char *instance_name = NULL;
+    size_t key_len = 0;
+    int bytes_written = 0;
+    int i = 0;
+    int ret = 0;
+    struct ios_conf *conf = this->private;
+
+    xlator_name = strdupa(conf->unique_id);
+    for (i = 0; i < strlen(xlator_name); i++) {
+        if (xlator_name[i] == '/')
+            xlator_name[i] = '_';
+    }
+
+    instance_name = this->instance_name;
+    if (this->name && strcmp(this->name, "glustershd") == 0) {
+        xlator_name = "shd";
+    } else if (this->prev && strcmp(this->prev->name, "nfs-server") == 0) {
+        xlator_name = "nfsd";
+        if (this->prev->instance_name)
+            instance_name = strdupa(this->prev->instance_name);
+    }
+
+    if (strcmp(__progname, "glusterfsd") == 0)
+        key_root = "gluster.brick";
+
+    if (instance_name) {
+        /* +3 for 2 x "." + NULL */
+        key_len = strlen(key_root) + strlen(xlator_name) +
+                  strlen(instance_name) + 3;
+        *key_prefix = GF_CALLOC(key_len, sizeof(char), gf_common_mt_char);
+        if (!*key_prefix) {
+            ret = -ENOMEM;
+            goto err;
+        }
+        bytes_written = snprintf(*key_prefix, key_len, "%s.%s.%s", key_root,
+                                 xlator_name, instance_name);
+        if (bytes_written != key_len - 1) {
+            ret = -EINVAL;
+            goto err;
+        }
+    } else {
+        /* +2 for 1 x "." + NULL */
+        key_len = strlen(key_root) + strlen(xlator_name) + 2;
+        *key_prefix = GF_CALLOC(key_len, sizeof(char), gf_common_mt_char);
+        if (!*key_prefix) {
+            ret = -ENOMEM;
+            goto err;
+        }
+        bytes_written = snprintf(*key_prefix, key_len, "%s.%s", key_root,
+                                 xlator_name);
+        if (bytes_written != key_len - 1) {
+            ret = -EINVAL;
+            goto err;
+        }
+    }
+    return 0;
+err:
+    GF_FREE(*key_prefix);
+    *key_prefix = NULL;
+    return ret;
+}
 
-        if (!iosfd)
-                return 0;
+int
+io_stats_dump_global_to_json_logfp(xlator_t *this,
+                                   struct ios_global_stats *stats, time_t now,
+                                   int interval, FILE *logfp)
+{
+    int i = 0;
+    int j = 0;
+    struct ios_conf *conf = NULL;
+    char *key_prefix = NULL;
+    char *str_prefix = NULL;
+    char *lc_fop_name = NULL;
+    int ret = 1; /* Default to error */
+    int rw_size;
+    char *rw_unit = NULL;
+    uint64_t fop_hits;
+    float fop_lat_ave;
+    float fop_lat_min;
+    float fop_lat_max;
+    double interval_sec;
+    double fop_ave_usec = 0.0;
+    double fop_ave_usec_sum = 0.0;
+    double weighted_fop_ave_usec = 0.0;
+    double weighted_fop_ave_usec_sum = 0.0;
+    long total_fop_hits = 0;
+    loc_t unused_loc = {
+        0,
+    };
+    dict_t *xattr = NULL;
+
+    interval_sec = (double)(now - stats->started_at);
+
+    conf = this->private;
+
+    ret = _io_stats_get_key_prefix(this, &key_prefix);
+    if (ret) {
+        goto out;
+    }
+
+    if (interval == -1) {
+        str_prefix = "aggr";
+
+    } else {
+        str_prefix = "inter";
+    }
+    ios_log(this, logfp, "{");
+
+    for (i = 0; i < 31; i++) {
+        rw_size = (1 << i);
+        if (rw_size >= 1024 * 1024) {
+            rw_size = rw_size / (1024 * 1024);
+            rw_unit = "mb";
+        } else if (rw_size >= 1024) {
+            rw_size = rw_size / 1024;
+            rw_unit = "kb";
+        } else {
+            rw_unit = "b";
+        }
 
-        gettimeofday (&now, NULL);
+        if (interval == -1) {
+            ios_log(this, logfp, "\"%s.%s.read_%d%s\": %" GF_PRI_ATOMIC ",",
+                    key_prefix, str_prefix, rw_size, rw_unit,
+                    GF_ATOMIC_GET(stats->block_count_read[i]));
+            ios_log(this, logfp, "\"%s.%s.write_%d%s\": %" GF_PRI_ATOMIC ",",
+                    key_prefix, str_prefix, rw_size, rw_unit,
+                    GF_ATOMIC_GET(stats->block_count_write[i]));
+        } else {
+            ios_log(this, logfp, "\"%s.%s.read_%d%s_per_sec\": %0.2lf,",
+                    key_prefix, str_prefix, rw_size, rw_unit,
+                    (double)(GF_ATOMIC_GET(stats->block_count_read[i]) /
+                             interval_sec));
+            ios_log(this, logfp, "\"%s.%s.write_%d%s_per_sec\": %0.2lf,",
+                    key_prefix, str_prefix, rw_size, rw_unit,
+                    (double)(GF_ATOMIC_GET(stats->block_count_write[i]) /
+                             interval_sec));
+        }
+    }
+
+    if (interval == -1) {
+        ios_log(this, logfp, "\"%s.%s.fds.open_count\": %" PRId64 ",",
+                key_prefix, str_prefix, conf->cumulative.nr_opens);
+        ios_log(this, logfp, "\"%s.%s.fds.max_open_count\": %" PRId64 ",",
+                key_prefix, str_prefix, conf->cumulative.max_nr_opens);
+    }
+
+    for (i = 0; i < GF_FOP_MAXVALUE; i++) {
+        lc_fop_name = strdupa(gf_fop_list[i]);
+        for (j = 0; lc_fop_name[j]; j++) {
+            lc_fop_name[j] = tolower(lc_fop_name[j]);
+        }
 
-        if (iosfd->opened_at.tv_usec > now.tv_usec) {
-                now.tv_usec += 1000000;
-                now.tv_usec--;
+        fop_hits = GF_ATOMIC_GET(stats->fop_hits[i]);
+        fop_lat_ave = 0.0;
+        fop_lat_min = 0.0;
+        fop_lat_max = 0.0;
+        if (fop_hits) {
+            if (stats->latency[i].avg) {
+                fop_lat_ave = stats->latency[i].avg;
+                fop_lat_min = stats->latency[i].min;
+                fop_lat_max = stats->latency[i].max;
+            }
+        }
+        if (interval == -1) {
+            ios_log(this, logfp, "\"%s.%s.fop.%s.count\": %" GF_PRI_ATOMIC ",",
+                    key_prefix, str_prefix, lc_fop_name, fop_hits);
+        } else {
+            ios_log(this, logfp, "\"%s.%s.fop.%s.per_sec\": %0.2lf,",
+                    key_prefix, str_prefix, lc_fop_name,
+                    (double)(fop_hits / interval_sec));
         }
 
-        sec = now.tv_sec - iosfd->opened_at.tv_sec;
-        usec = now.tv_usec - iosfd->opened_at.tv_usec;
-
-        gf_log (this->name, GF_LOG_NORMAL,
-                "--- fd stats ---");
-
-        if (iosfd->filename)
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "      Filename : %s",
-                        iosfd->filename);
-
-        if (sec)
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "      Lifetime : %"PRId64"secs, %"PRId64"usecs",
-                        sec, usec);
-
-        if (iosfd->data_read)
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "     BytesRead : %"PRId64" bytes",
-                        iosfd->data_read);
-
-        if (iosfd->data_written)
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "  BytesWritten : %"PRId64" bytes",
-                        iosfd->data_written);
-
-        for (i = 0; i < 32; i++) {
-                if (iosfd->block_count_read[i])
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                " Read %06db+ : %"PRId64,
-                                (1 << i), iosfd->block_count_read[i]);
+        ios_log(this, logfp, "\"%s.%s.fop.%s.latency_ave_usec\": %0.2lf,",
+                key_prefix, str_prefix, lc_fop_name, fop_lat_ave);
+        ios_log(this, logfp, "\"%s.%s.fop.%s.latency_min_usec\": %0.2lf,",
+                key_prefix, str_prefix, lc_fop_name, fop_lat_min);
+        ios_log(this, logfp, "\"%s.%s.fop.%s.latency_max_usec\": %0.2lf,",
+                key_prefix, str_prefix, lc_fop_name, fop_lat_max);
+
+        fop_ave_usec_sum += fop_lat_ave;
+        weighted_fop_ave_usec_sum += fop_hits * fop_lat_ave;
+        total_fop_hits += fop_hits;
+    }
+
+    if (total_fop_hits) {
+        weighted_fop_ave_usec = weighted_fop_ave_usec_sum / total_fop_hits;
+        /* Extra key that does not print out an entry w/ 0.00 for
+         * intervals with no data
+         */
+        ios_log(this, logfp,
+                "\"%s.%s.fop.weighted_latency_ave_usec_nozerofill\": "
+                "%0.4lf,",
+                key_prefix, str_prefix, weighted_fop_ave_usec);
+    }
+    ios_log(this, logfp, "\"%s.%s.fop.weighted_latency_ave_usec\": %0.4lf,",
+            key_prefix, str_prefix, weighted_fop_ave_usec);
+    ios_log(this, logfp, "\"%s.%s.fop.weighted_fop_count\": %ld,", key_prefix,
+            str_prefix, total_fop_hits);
+
+    fop_ave_usec = fop_ave_usec_sum / GF_FOP_MAXVALUE;
+    ios_log(this, logfp, "\"%s.%s.fop.unweighted_latency_ave_usec\":%0.4lf,",
+            key_prefix, str_prefix, fop_ave_usec);
+
+    for (i = 0; i < GF_UPCALL_FLAGS_MAXVALUE; i++) {
+        lc_fop_name = strdupa(gf_upcall_list[i]);
+        for (j = 0; lc_fop_name[j]; j++) {
+            lc_fop_name[j] = tolower(lc_fop_name[j]);
         }
-        for (i = 0; i < 32; i++) {
-                if (iosfd->block_count_write[i])
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "Write %06db+ : %"PRId64,
-                                (1 << i), iosfd->block_count_write[i]);
+        fop_hits = GF_ATOMIC_GET(stats->upcall_hits[i]);
+        if (interval == -1) {
+            ios_log(this, logfp, "\"%s.%s.fop.%s.count\": %" GF_PRI_ATOMIC ",",
+                    key_prefix, str_prefix, lc_fop_name, fop_hits);
+        } else {
+            ios_log(this, logfp, "\"%s.%s.fop.%s.per_sec\": %0.2lf,",
+                    key_prefix, str_prefix, lc_fop_name,
+                    (double)(fop_hits / interval_sec));
         }
-        return 0;
+    }
+
+    ret = syncop_getxattr(this, &unused_loc, &xattr, IO_THREADS_QUEUE_SIZE_KEY,
+                          NULL, NULL);
+    if (xattr) {
+        /*
+         * Iterate over the dictionary returned to us by io-threads and
+         * dump the results to the stats file.
+         */
+        data_pair_t *curr = NULL;
+
+        dict_foreach_inline(xattr, curr)
+        {
+            ios_log(this, logfp, "\"%s.%s.%s.queue_size\": %d,", key_prefix,
+                    str_prefix, curr->key, data_to_int32(curr->value));
+        }
+
+        /* Free the dictionary */
+        dict_unref(xattr);
+    } else {
+        gf_log(this->name, GF_LOG_WARNING,
+               "Unable to get queue size counts from "
+               "the io-threads translator!");
+    }
+
+    if (interval == -1) {
+        ios_log(this, logfp, "\"%s.%s.uptime\": %" PRIu64 ",", key_prefix,
+                str_prefix, (uint64_t)(now - stats->started_at));
+        ios_log(this, logfp,
+                "\"%s.%s.bytes_read\": "
+                "%" GF_PRI_ATOMIC ",",
+                key_prefix, str_prefix, GF_ATOMIC_GET(stats->data_read));
+        ios_log(this, logfp,
+                "\"%s.%s.bytes_written\": "
+                "%" GF_PRI_ATOMIC "",
+                key_prefix, str_prefix, GF_ATOMIC_GET(stats->data_written));
+    } else {
+        ios_log(this, logfp, "\"%s.%s.sample_interval_sec\": %0.2lf,",
+                key_prefix, str_prefix, interval_sec);
+        ios_log(this, logfp, "\"%s.%s.bytes_read_per_sec\": %0.2lf,",
+                key_prefix, str_prefix,
+                (double)(GF_ATOMIC_GET(stats->data_read) / interval_sec));
+        ios_log(this, logfp, "\"%s.%s.bytes_written_per_sec\": %0.2lf",
+                key_prefix, str_prefix,
+                (double)(GF_ATOMIC_GET(stats->data_written) / interval_sec));
+    }
+
+    ios_log(this, logfp, "}");
+    ret = 0;
+out:
+    GF_FREE(key_prefix);
+    return ret;
 }
 
-int
-update_ios_latency (struct ios_conf *conf, call_frame_t *frame,
-                    glusterfs_fop_t op)
+char *
+_resolve_username(xlator_t *this, uid_t uid)
 {
-        double elapsed;
-        struct timeval *begin, *end;
-        double avg;
+    struct passwd pwd;
+    struct passwd *pwd_result = NULL;
+    size_t pwd_buf_len;
+    char *pwd_buf = NULL;
+    char *ret = NULL;
+
+    /* Prepare our buffer for the uid->username translation */
+#ifdef _SC_GETGR_R_SIZE_MAX
+    pwd_buf_len = sysconf(_SC_GETGR_R_SIZE_MAX);
+#else
+    pwd_buf_len = -1;
+#endif
+    if (pwd_buf_len == -1) {
+        pwd_buf_len = DEFAULT_PWD_BUF_SZ; /* per the man page */
+    }
 
-        begin = &frame->begin;
-        end   = &frame->end;
+    pwd_buf = alloca(pwd_buf_len);
+    if (!pwd_buf)
+        goto err;
 
-        elapsed = (end->tv_sec - begin->tv_sec) * 1e6
-                + (end->tv_usec - begin->tv_usec);
+    getpwuid_r(uid, &pwd, pwd_buf, pwd_buf_len, &pwd_result);
+    if (!pwd_result)
+        goto err;
 
-        /* Cumulative */
-        if (!conf->cumulative.latency[op].min)
-                conf->cumulative.latency[op].min = elapsed;
-        if (conf->cumulative.latency[op].min > elapsed)
-                conf->cumulative.latency[op].min = elapsed;
-        if (conf->cumulative.latency[op].max < elapsed)
-                conf->cumulative.latency[op].max = elapsed;
+    ret = gf_strdup(pwd.pw_name);
+    if (ret)
+        return ret;
+    else
+        gf_log(this->name, GF_LOG_ERROR,
+               "gf_strdup failed, failing username "
+               "resolution.");
+err:
+    return ret;
+}
 
-        avg = conf->cumulative.latency[op].avg;
+char *
+_resolve_group_name(xlator_t *this, gid_t gid)
+{
+    struct group grp;
+    struct group *grp_result = NULL;
+    size_t grp_buf_len;
+    char *grp_buf = NULL;
+    char *ret = NULL;
+
+    /* Prepare our buffer for the gid->group name translation */
+#ifdef _SC_GETGR_R_SIZE_MAX
+    grp_buf_len = sysconf(_SC_GETGR_R_SIZE_MAX);
+#else
+    grp_buf_len = -1;
+#endif
+    if (grp_buf_len == -1) {
+        grp_buf_len = DEFAULT_GRP_BUF_SZ; /* per the man page */
+    }
 
-        conf->cumulative.latency[op].avg =
-                avg + (elapsed - avg) / conf->cumulative.fop_hits[op];
+    grp_buf = alloca(grp_buf_len);
+    if (!grp_buf) {
+        goto err;
+    }
 
-        /* Incremental */
-        if (!conf->incremental.latency[op].min)
-                conf->incremental.latency[op].min = elapsed;
-        if (conf->incremental.latency[op].min > elapsed)
-                conf->incremental.latency[op].min = elapsed;
-        if (conf->incremental.latency[op].max < elapsed)
-                conf->incremental.latency[op].max = elapsed;
+    if (getgrgid_r(gid, &grp, grp_buf, grp_buf_len, &grp_result) != 0)
+        goto err;
 
-        avg = conf->incremental.latency[op].avg;
+    if (!grp_result)
+        goto err;
 
-        conf->incremental.latency[op].avg =
-                avg + (elapsed - avg) / conf->incremental.fop_hits[op];
+    ret = gf_strdup(grp.gr_name);
+    if (ret)
+        return ret;
+    else
+        gf_log(this->name, GF_LOG_ERROR,
+               "gf_strdup failed, failing username "
+               "resolution.");
+err:
+    return ret;
+}
 
-        return 0;
+/*
+ * This function writes out a latency sample to a given file descriptor
+ * and beautifies the output in the process.
+ */
+void
+_io_stats_write_latency_sample(xlator_t *this, ios_sample_t *sample,
+                               FILE *logfp)
+{
+    double epoch_time = 0.00;
+    char *xlator_name = NULL;
+    char *instance_name = NULL;
+    char *hostname = NULL;
+    char *identifier = NULL;
+    char *port = NULL;
+    char *port_pos = NULL;
+    char *group_name = NULL;
+    char *username = NULL;
+    struct ios_conf *conf = NULL;
+
+    conf = this->private;
+
+    epoch_time = (sample->timestamp).tv_sec +
+                 ((sample->timestamp).tv_usec / 1000000.0);
+
+    if (strlen(sample->identifier) == 0) {
+        hostname = "Unknown";
+        port = "Unknown";
+    } else {
+        identifier = strdupa(sample->identifier);
+        port_pos = strrchr(identifier, ':');
+        if (!port_pos || strlen(port_pos) < 2)
+            goto err;
+        port = strdupa(port_pos + 1);
+        if (!port)
+            goto err;
+        *port_pos = '\0';
+        hostname = gf_rev_dns_lookup_cached(identifier, conf->dnscache);
+        if (!hostname)
+            hostname = "Unknown";
+    }
+
+    xlator_name = conf->unique_id;
+    if (!xlator_name || strlen(xlator_name) == 0)
+        xlator_name = "Unknown";
+
+    instance_name = this->instance_name;
+    if (!instance_name || strlen(instance_name) == 0)
+        instance_name = "N/A";
+
+    /* Resolve the UID to a string username */
+    username = _resolve_username(this, sample->uid);
+    if (!username) {
+        username = GF_MALLOC(30, gf_common_mt_char);
+        if (!username) {
+            goto out;
+        }
+        sprintf(username, "%d", (int32_t)sample->uid);
+    }
+
+    /* Resolve the GID to a string group name */
+    group_name = _resolve_group_name(this, sample->gid);
+    if (!group_name) {
+        group_name = GF_MALLOC(30, gf_common_mt_char);
+        if (!group_name) {
+            goto out;
+        }
+        sprintf(group_name, "%d", (int32_t)sample->gid);
+    }
+
+    ios_log(this, logfp, "%0.6lf,%s,%s,%0.4lf,%s,%s,%s,%s,%s,%s", epoch_time,
+            fop_enum_to_pri_string(sample->fop_type),
+            gf_fop_string(sample->fop_type), sample->elapsed, xlator_name,
+            instance_name, username, group_name, hostname, port);
+    goto out;
+err:
+    gf_log(this->name, GF_LOG_ERROR, "Error parsing socket identifier");
+out:
+    GF_FREE(group_name);
+    GF_FREE(username);
 }
 
+/*
+ * Takes our current sample buffer in conf->io_sample_buf, and saves
+ * a reference to this, init's a new buffer, and then dumps out the
+ * contents of the saved reference.
+ */
 int
-io_stats_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, fd_t *fd,
-                     inode_t *inode, struct iatt *buf,
-                     struct iatt *preparent, struct iatt *postparent)
+io_stats_dump_latency_samples_logfp(xlator_t *this, FILE *logfp)
 {
-        struct ios_fd *iosfd = NULL;
-        char          *path = NULL;
+    uint64_t i = 0;
+    struct ios_conf *conf = NULL;
+    ios_sample_buf_t *sample_buf = NULL;
+    int ret = 1; /* Default to error */
+
+    conf = this->private;
+
+    /* Save pointer to old buffer; the CS equivalent of
+     * Indiana Jones: https://www.youtube.com/watch?v=Pr-8AP0To4k,
+     * though ours will end better I hope!
+     */
+    sample_buf = conf->ios_sample_buf;
+    if (!sample_buf) {
+        gf_log(this->name, GF_LOG_WARNING, "Sampling buffer is null, bailing!");
+        goto out;
+    }
+
+    /* Empty case, nothing to do, exit. */
+    if (sample_buf->collected == 0) {
+        gf_log(this->name, GF_LOG_DEBUG, "No samples, dump not required.");
+        ret = 0;
+        goto out;
+    }
+
+    /* Init a new buffer, so we are free to work on the one we saved a
+     * reference to above.
+     */
+    if (ios_init_sample_buf(conf) != 0) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "Failed to init new sampling buffer, out of memory?");
+        goto out;
+    }
+
+    /* Wrap-around case, dump from pos to sample_buf->size -1
+     * and then from 0 to sample_buf->pos (covered off by
+     * "simple case")
+     */
+    if (sample_buf->collected > sample_buf->pos + 1) {
+        for (i = sample_buf->pos; i < sample_buf->size; i++) {
+            _io_stats_write_latency_sample(this, &(sample_buf->ios_samples[i]),
+                                           logfp);
+        }
+    }
 
-        path = frame->local;
-        frame->local = NULL;
+    /* Simple case: Dump from 0 to sample_buf->pos */
+    for (i = 0; i < sample_buf->pos; i++) {
+        _io_stats_write_latency_sample(this, &(sample_buf->ios_samples[i]),
+                                       logfp);
+    }
+    ios_destroy_sample_buf(sample_buf);
 
-        if (!path)
-                goto unwind;
+out:
+    return ret;
+}
 
-        if (op_ret < 0) {
-                GF_FREE (path);
-                goto unwind;
+int
+io_stats_dump_global_to_logfp(xlator_t *this, struct ios_global_stats *stats,
+                              time_t now, int interval, FILE *logfp)
+{
+    int i = 0;
+    int per_line = 0;
+    int index = 0;
+    struct ios_stat_head *list_head = NULL;
+    struct ios_conf *conf = NULL;
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    char str_header[128] = {0};
+    char str_read[128] = {0};
+    char str_write[128] = {0};
+    uint64_t fop_hits = 0;
+    uint64_t block_count_read = 0;
+    uint64_t block_count_write = 0;
+
+    conf = this->private;
+
+    if (interval == -1)
+        ios_log(this, logfp, "\n=== Cumulative stats ===");
+    else
+        ios_log(this, logfp, "\n=== Interval %d stats ===", interval);
+    ios_log(this, logfp, "      Duration : %" PRIu64 " secs",
+            (uint64_t)(now - stats->started_at));
+    ios_log(this, logfp, "     BytesRead : %" GF_PRI_ATOMIC,
+            GF_ATOMIC_GET(stats->data_read));
+    ios_log(this, logfp, "  BytesWritten : %" GF_PRI_ATOMIC "\n",
+            GF_ATOMIC_GET(stats->data_written));
+
+    snprintf(str_header, sizeof(str_header), "%-12s %c", "Block Size", ':');
+    snprintf(str_read, sizeof(str_read), "%-12s %c", "Read Count", ':');
+    snprintf(str_write, sizeof(str_write), "%-12s %c", "Write Count", ':');
+    index = 14;
+    for (i = 0; i < IOS_BLOCK_COUNT_SIZE; i++) {
+        block_count_read = GF_ATOMIC_GET(stats->block_count_read[i]);
+        block_count_write = GF_ATOMIC_GET(stats->block_count_write[i]);
+        if ((block_count_read == 0) && (block_count_write == 0))
+            continue;
+        per_line++;
+
+        snprintf(str_header + index, sizeof(str_header) - index, "%16dB+",
+                 (1 << i));
+        if (block_count_read)
+            snprintf(str_read + index, sizeof(str_read) - index, "%18" PRId64,
+                     block_count_read);
+        else
+            snprintf(str_read + index, sizeof(str_read) - index, "%18s", "0");
+        if (block_count_write)
+            snprintf(str_write + index, sizeof(str_write) - index,
+                     "%18" GF_PRI_ATOMIC, block_count_write);
+        else
+            snprintf(str_write + index, sizeof(str_write) - index, "%18s", "0");
+
+        index += 18;
+        if (per_line == 3) {
+            ios_log(this, logfp, "%s", str_header);
+            ios_log(this, logfp, "%s", str_read);
+            ios_log(this, logfp, "%s\n", str_write);
+
+            snprintf(str_header, sizeof(str_header), "%-12s %c", "Block Size",
+                     ':');
+            snprintf(str_read, sizeof(str_read), "%-12s %c", "Read Count", ':');
+            snprintf(str_write, sizeof(str_write), "%-12s %c", "Write Count",
+                     ':');
+
+            index = 14;
+            per_line = 0;
         }
-
-        iosfd = GF_CALLOC (1, sizeof (*iosfd), gf_io_stats_mt_ios_fd);
-        if (!iosfd) {
-                GF_FREE (path);
-                goto unwind;
+    }
+
+    if (per_line != 0) {
+        ios_log(this, logfp, "%s", str_header);
+        ios_log(this, logfp, "%s", str_read);
+        ios_log(this, logfp, "%s\n", str_write);
+    }
+
+    ios_log(this, logfp, "%-13s %10s %14s %14s %14s", "Fop", "Call Count",
+            "Avg-Latency", "Min-Latency", "Max-Latency");
+    ios_log(this, logfp, "%-13s %10s %14s %14s %14s", "---", "----------",
+            "-----------", "-----------", "-----------");
+
+    for (i = 0; i < GF_FOP_MAXVALUE; i++) {
+        fop_hits = GF_ATOMIC_GET(stats->fop_hits[i]);
+        if (fop_hits && !stats->latency[i].avg)
+            ios_log(this, logfp,
+                    "%-13s %10" GF_PRI_ATOMIC
+                    " %11s "
+                    "us %11s us %11s us",
+                    gf_fop_list[i], fop_hits, "0", "0", "0");
+        else if (fop_hits && stats->latency[i].avg)
+            ios_log(this, logfp,
+                    "%-13s %10" GF_PRI_ATOMIC
+                    " "
+                    "%11.2lf us %11.2lf us %11.2lf us",
+                    gf_fop_list[i], fop_hits, stats->latency[i].avg,
+                    stats->latency[i].min, stats->latency[i].max);
+    }
+
+    for (i = 0; i < GF_UPCALL_FLAGS_MAXVALUE; i++) {
+        fop_hits = GF_ATOMIC_GET(stats->upcall_hits[i]);
+        if (fop_hits)
+            ios_log(this, logfp,
+                    "%-13s %10" PRId64
+                    " %11s "
+                    "us %11s us %11s us",
+                    gf_upcall_list[i], fop_hits, "0", "0", "0");
+    }
+
+    ios_log(this, logfp,
+            "------ ----- ----- ----- ----- ----- ----- ----- "
+            " ----- ----- ----- -----\n");
+
+    if (interval == -1) {
+        LOCK(&conf->lock);
+        {
+            gf_time_fmt_tv(timestr, sizeof timestr,
+                           &conf->cumulative.max_openfd_time, gf_timefmt_FT);
+            ios_log(this, logfp,
+                    "Current open fd's: %" PRId64 " Max open fd's: %" PRId64
+                    " time %s",
+                    conf->cumulative.nr_opens, conf->cumulative.max_nr_opens,
+                    timestr);
         }
+        UNLOCK(&conf->lock);
+        ios_log(this, logfp, "\n==========Open File Stats========");
+        ios_log(this, logfp, "\nCOUNT:  \t  FILE NAME");
+        list_head = &conf->list[IOS_STATS_TYPE_OPEN];
+        ios_dump_file_stats(list_head, this, logfp);
+
+        ios_log(this, logfp, "\n==========Read File Stats========");
+        ios_log(this, logfp, "\nCOUNT:  \t  FILE NAME");
+        list_head = &conf->list[IOS_STATS_TYPE_READ];
+        ios_dump_file_stats(list_head, this, logfp);
+
+        ios_log(this, logfp, "\n==========Write File Stats========");
+        ios_log(this, logfp, "\nCOUNT:  \t  FILE NAME");
+        list_head = &conf->list[IOS_STATS_TYPE_WRITE];
+        ios_dump_file_stats(list_head, this, logfp);
+
+        ios_log(this, logfp, "\n==========Directory open stats========");
+        ios_log(this, logfp, "\nCOUNT:  \t  DIRECTORY NAME");
+        list_head = &conf->list[IOS_STATS_TYPE_OPENDIR];
+        ios_dump_file_stats(list_head, this, logfp);
+
+        ios_log(this, logfp, "\n========Directory readdirp Stats=======");
+        ios_log(this, logfp, "\nCOUNT:  \t  DIRECTORY NAME");
+        list_head = &conf->list[IOS_STATS_TYPE_READDIRP];
+        ios_dump_file_stats(list_head, this, logfp);
+
+        ios_log(this, logfp, "\n========Read Throughput File Stats=====");
+        ios_log(this, logfp,
+                "\nTIMESTAMP \t\t\t THROUGHPUT(KBPS)"
+                "\tFILE NAME");
+        list_head = &conf->thru_list[IOS_STATS_THRU_READ];
+        ios_dump_throughput_stats(list_head, this, logfp, IOS_STATS_THRU_READ);
+
+        ios_log(this, logfp, "\n======Write Throughput File Stats======");
+        ios_log(this, logfp,
+                "\nTIMESTAMP \t\t\t THROUGHPUT(KBPS)"
+                "\tFILE NAME");
+        list_head = &conf->thru_list[IOS_STATS_THRU_WRITE];
+        ios_dump_throughput_stats(list_head, this, logfp, IOS_STATS_THRU_WRITE);
+    }
+    return 0;
+}
 
-        iosfd->filename = path;
-        gettimeofday (&iosfd->opened_at, NULL);
-
-        ios_fd_ctx_set (fd, this, iosfd);
+int
+io_stats_dump_global_to_dict(xlator_t *this, struct ios_global_stats *stats,
+                             time_t now, int interval, dict_t *dict)
+{
+    int ret = 0;
+    char key[64] = {0};
+    uint64_t sec = 0;
+    int i = 0;
+    uint64_t count = 0;
+    uint64_t fop_hits = 0;
+
+    GF_ASSERT(stats);
+    GF_ASSERT(now);
+    GF_ASSERT(dict);
+    GF_ASSERT(this);
+
+    if (interval == -1)
+        snprintf(key, sizeof(key), "cumulative");
+    else
+        snprintf(key, sizeof(key), "interval");
+    ret = dict_set_int32(dict, key, interval);
+    if (ret)
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to set "
+               "interval %d",
+               interval);
+
+    snprintf(key, sizeof(key), "%d-duration", interval);
+    sec = now - stats->started_at;
+    ret = dict_set_uint64(dict, key, sec);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to set "
+               "duration(%d) - %" PRId64,
+               interval, sec);
+        goto out;
+    }
+
+    snprintf(key, sizeof(key), "%d-total-read", interval);
+    ret = dict_set_uint64(dict, key, GF_ATOMIC_GET(stats->data_read));
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to set total "
+               "read(%d) - %" GF_PRI_ATOMIC,
+               interval, GF_ATOMIC_GET(stats->data_read));
+        goto out;
+    }
+
+    snprintf(key, sizeof(key), "%d-total-write", interval);
+    ret = dict_set_uint64(dict, key, GF_ATOMIC_GET(stats->data_written));
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to set total "
+               "write(%d) - %" GF_PRI_ATOMIC,
+               interval, GF_ATOMIC_GET(stats->data_written));
+        goto out;
+    }
+    for (i = 0; i < 32; i++) {
+        count = GF_ATOMIC_GET(stats->block_count_read[i]);
+        if (count) {
+            snprintf(key, sizeof(key), "%d-read-%d", interval, (1 << i));
+            ret = dict_set_uint64(dict, key, count);
+            if (ret) {
+                gf_log(this->name, GF_LOG_ERROR,
+                       "failed to "
+                       "set read-%db+, with: %" PRId64,
+                       (1 << i), count);
+                goto out;
+            }
+        }
+    }
+
+    for (i = 0; i < IOS_BLOCK_COUNT_SIZE; i++) {
+        count = GF_ATOMIC_GET(stats->block_count_write[i]);
+        if (count) {
+            snprintf(key, sizeof(key), "%d-write-%d", interval, (1 << i));
+            ret = dict_set_uint64(dict, key, count);
+            if (ret) {
+                gf_log(this->name, GF_LOG_ERROR,
+                       "failed to "
+                       "set write-%db+, with: %" PRId64,
+                       (1 << i), count);
+                goto out;
+            }
+        }
+    }
+
+    for (i = 0; i < GF_FOP_MAXVALUE; i++) {
+        fop_hits = GF_ATOMIC_GET(stats->fop_hits[i]);
+        if (fop_hits == 0)
+            continue;
+        snprintf(key, sizeof(key), "%d-%d-hits", interval, i);
+        ret = dict_set_uint64(dict, key, fop_hits);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "failed to set "
+                   "%s-fop-hits: %" GF_PRI_ATOMIC,
+                   gf_fop_list[i], fop_hits);
+            goto out;
+        }
 
-unwind:
-        END_FOP_LATENCY (frame, CREATE);
-        STACK_UNWIND_STRICT (create, frame, op_ret, op_errno, fd, inode, buf,
-                             preparent, postparent);
-        return 0;
+        if (stats->latency[i].avg == 0)
+            continue;
+        snprintf(key, sizeof(key), "%d-%d-avglatency", interval, i);
+        ret = dict_set_double(dict, key, stats->latency[i].avg);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "failed to set %s "
+                   "avglatency(%d) with %f",
+                   gf_fop_list[i], interval, stats->latency[i].avg);
+            goto out;
+        }
+        snprintf(key, sizeof(key), "%d-%d-minlatency", interval, i);
+        ret = dict_set_double(dict, key, stats->latency[i].min);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "failed to set %s "
+                   "minlatency(%d) with %f",
+                   gf_fop_list[i], interval, stats->latency[i].min);
+            goto out;
+        }
+        snprintf(key, sizeof(key), "%d-%d-maxlatency", interval, i);
+        ret = dict_set_double(dict, key, stats->latency[i].max);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "failed to set %s "
+                   "maxlatency(%d) with %f",
+                   gf_fop_list[i], interval, stats->latency[i].max);
+            goto out;
+        }
+    }
+    for (i = 0; i < GF_UPCALL_FLAGS_MAXVALUE; i++) {
+        fop_hits = GF_ATOMIC_GET(stats->upcall_hits[i]);
+        if (fop_hits == 0)
+            continue;
+        snprintf(key, sizeof(key), "%d-%d-upcall-hits", interval, i);
+        ret = dict_set_uint64(dict, key, fop_hits);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "failed to "
+                   "set %s-upcall-hits: %" PRIu64,
+                   gf_upcall_list[i], fop_hits);
+            goto out;
+        }
+    }
+out:
+    gf_log(this->name, GF_LOG_DEBUG, "returning %d", ret);
+    return ret;
 }
 
+int
+io_stats_dump_global(xlator_t *this, struct ios_global_stats *stats, time_t now,
+                     int interval, struct ios_dump_args *args)
+{
+    int ret = -1;
+
+    GF_ASSERT(args);
+    GF_ASSERT(now);
+    GF_ASSERT(stats);
+    GF_ASSERT(this);
+
+    switch (args->type) {
+        case IOS_DUMP_TYPE_JSON_FILE:
+            ret = io_stats_dump_global_to_json_logfp(this, stats, now, interval,
+                                                     args->u.logfp);
+            break;
+        case IOS_DUMP_TYPE_FILE:
+            ret = io_stats_dump_global_to_logfp(this, stats, now, interval,
+                                                args->u.logfp);
+            break;
+        case IOS_DUMP_TYPE_DICT:
+            ret = io_stats_dump_global_to_dict(this, stats, now, interval,
+                                               args->u.dict);
+            break;
+        default:
+            GF_ASSERT(0);
+            ret = -1;
+            break;
+    }
+    return ret;
+}
 
 int
-io_stats_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, fd_t *fd)
+ios_dump_args_init(struct ios_dump_args *args, ios_dump_type_t type,
+                   void *output)
 {
-        struct ios_fd *iosfd = NULL;
-        char          *path = NULL;
+    int ret = 0;
+
+    GF_ASSERT(args);
+    GF_ASSERT(type > IOS_DUMP_TYPE_NONE && type < IOS_DUMP_TYPE_MAX);
+    GF_ASSERT(output);
+
+    args->type = type;
+    switch (args->type) {
+        case IOS_DUMP_TYPE_JSON_FILE:
+        case IOS_DUMP_TYPE_FILE:
+            args->u.logfp = output;
+            break;
+        case IOS_DUMP_TYPE_DICT:
+            args->u.dict = output;
+            break;
+        default:
+            GF_ASSERT(0);
+            ret = -1;
+    }
+
+    return ret;
+}
 
-        path = frame->local;
-        frame->local = NULL;
+static void
+ios_global_stats_clear(struct ios_global_stats *stats, time_t now)
+{
+    GF_ASSERT(stats);
+    GF_ASSERT(now);
 
-        if (!path)
-                goto unwind;
+    memset(stats, 0, sizeof(*stats));
+    stats->started_at = now;
+}
 
-        if (op_ret < 0) {
-                GF_FREE (path);
-                goto unwind;
-        }
+int
+io_stats_dump(xlator_t *this, struct ios_dump_args *args, ios_info_op_t op,
+              gf_boolean_t is_peek)
+{
+    struct ios_conf *conf = NULL;
+    struct ios_global_stats cumulative = {};
+    struct ios_global_stats incremental = {};
+    int increment = 0;
+    time_t now = 0;
+
+    GF_ASSERT(this);
+    GF_ASSERT(args);
+    GF_ASSERT(args->type > IOS_DUMP_TYPE_NONE);
+    GF_ASSERT(args->type < IOS_DUMP_TYPE_MAX);
+
+    conf = this->private;
+    now = gf_time();
+
+    LOCK(&conf->lock);
+    {
+        if (op == GF_IOS_INFO_ALL || op == GF_IOS_INFO_CUMULATIVE)
+            cumulative = conf->cumulative;
+
+        if (op == GF_IOS_INFO_ALL || op == GF_IOS_INFO_INCREMENTAL) {
+            incremental = conf->incremental;
+            increment = conf->increment;
+
+            if (!is_peek) {
+                increment = conf->increment++;
 
-        iosfd = GF_CALLOC (1, sizeof (*iosfd), gf_io_stats_mt_ios_fd);
-        if (!iosfd) {
-                GF_FREE (path);
-                goto unwind;
+                ios_global_stats_clear(&conf->incremental, now);
+            }
         }
+    }
+    UNLOCK(&conf->lock);
 
-        iosfd->filename = path;
-        gettimeofday (&iosfd->opened_at, NULL);
+    if (op == GF_IOS_INFO_ALL || op == GF_IOS_INFO_CUMULATIVE)
+        io_stats_dump_global(this, &cumulative, now, -1, args);
 
-        ios_fd_ctx_set (fd, this, iosfd);
+    if (op == GF_IOS_INFO_ALL || op == GF_IOS_INFO_INCREMENTAL)
+        io_stats_dump_global(this, &incremental, now, increment, args);
 
-unwind:
-        END_FOP_LATENCY (frame, OPEN);
-        STACK_UNWIND_STRICT (open, frame, op_ret, op_errno, fd);
-        return 0;
+    return 0;
 }
 
-
 int
-io_stats_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct iatt *buf)
+io_stats_dump_fd(xlator_t *this, struct ios_fd *iosfd)
 {
-        END_FOP_LATENCY (frame, STAT);
-        STACK_UNWIND_STRICT (stat, frame, op_ret, op_errno, buf);
+    struct ios_conf *conf = NULL;
+    struct timeval now;
+    int i = 0;
+    double usecs = 0;
+    uint64_t data_read = 0;
+    uint64_t data_written = 0;
+    uint64_t block_count_read = 0;
+    uint64_t block_count_write = 0;
+
+    conf = this->private;
+
+    if (!conf->dump_fd_stats)
+        return 0;
+
+    if (!iosfd)
         return 0;
+
+    gettimeofday(&now, NULL);
+    usecs = gf_tvdiff(&iosfd->opened_at, &now);
+
+    gf_log(this->name, GF_LOG_INFO, "--- fd stats ---");
+
+    if (iosfd->filename)
+        gf_log(this->name, GF_LOG_INFO, "      Filename : %s", iosfd->filename);
+
+    if (usecs)
+        gf_log(this->name, GF_LOG_INFO, "      Lifetime : %lf secs", usecs);
+
+    data_read = GF_ATOMIC_GET(iosfd->data_read);
+    if (data_read)
+        gf_log(this->name, GF_LOG_INFO, "     BytesRead : %" PRId64 " bytes",
+               data_read);
+
+    data_written = GF_ATOMIC_GET(iosfd->data_written);
+    if (data_written)
+        gf_log(this->name, GF_LOG_INFO, "  BytesWritten : %" PRId64 " bytes",
+               data_written);
+
+    for (i = 0; i < 32; i++) {
+        block_count_read = GF_ATOMIC_GET(iosfd->block_count_read[i]);
+        if (block_count_read)
+            gf_log(this->name, GF_LOG_INFO,
+                   " Read %06db+ :"
+                   "%" PRId64,
+                   (1 << i), block_count_read);
+    }
+    for (i = 0; i < IOS_BLOCK_COUNT_SIZE; i++) {
+        block_count_write = GF_ATOMIC_GET(iosfd->block_count_write[i]);
+        if (block_count_write)
+            gf_log(this->name, GF_LOG_INFO, "Write %06db+ : %" PRId64, (1 << i),
+                   block_count_write);
+    }
+    return 0;
 }
 
+void
+collect_ios_latency_sample(struct ios_conf *conf, glusterfs_fop_t fop_type,
+                           double elapsed, call_frame_t *frame)
+{
+    ios_sample_buf_t *ios_sample_buf = NULL;
+    ios_sample_t *ios_sample = NULL;
+    struct timespec *timestamp = NULL;
+    call_stack_t *root = NULL;
+
+    ios_sample_buf = conf->ios_sample_buf;
+    LOCK(&conf->ios_sampling_lock);
+    if (conf->ios_sample_interval == 0 ||
+        ios_sample_buf->observed % conf->ios_sample_interval != 0)
+        goto out;
+
+    timestamp = &frame->begin;
+    root = frame->root;
+
+    ios_sample = &(ios_sample_buf->ios_samples[ios_sample_buf->pos]);
+    ios_sample->elapsed = elapsed;
+    ios_sample->fop_type = fop_type;
+    ios_sample->uid = root->uid;
+    ios_sample->gid = root->gid;
+    (ios_sample->timestamp).tv_sec = timestamp->tv_sec;
+    (ios_sample->timestamp).tv_usec = timestamp->tv_nsec / 1000;
+    memcpy(&ios_sample->identifier, &root->identifier,
+           sizeof(root->identifier));
+
+    /* We've reached the end of the circular buffer, start from the
+     * beginning. */
+    if (ios_sample_buf->pos == (ios_sample_buf->size - 1))
+        ios_sample_buf->pos = 0;
+    else
+        ios_sample_buf->pos++;
+    ios_sample_buf->collected++;
+out:
+    ios_sample_buf->observed++;
+    UNLOCK(&conf->ios_sampling_lock);
+    return;
+}
 
-int
-io_stats_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno,
-                    struct iovec *vector, int32_t count,
-                    struct iatt *buf, struct iobref *iobref)
+static void
+update_ios_latency_stats(struct ios_global_stats *stats, double elapsed,
+                         glusterfs_fop_t op)
 {
-        int              len = 0;
-        fd_t            *fd = NULL;
+    double avg;
 
-        fd = frame->local;
-        frame->local = NULL;
+    GF_ASSERT(stats);
 
-        if (op_ret > 0) {
-                len = iov_length (vector, count);
-                BUMP_READ (fd, len);
-        }
+    stats->latency[op].total += elapsed;
 
-        END_FOP_LATENCY (frame, READ);
-        STACK_UNWIND_STRICT (readv, frame, op_ret, op_errno,
-                             vector, count, buf, iobref);
-        return 0;
-}
+    if (!stats->latency[op].min)
+        stats->latency[op].min = elapsed;
+    if (stats->latency[op].min > elapsed)
+        stats->latency[op].min = elapsed;
+    if (stats->latency[op].max < elapsed)
+        stats->latency[op].max = elapsed;
+
+    avg = stats->latency[op].avg;
 
+    stats->latency[op].avg = avg + (elapsed - avg) /
+                                       GF_ATOMIC_GET(stats->fop_hits[op]);
+}
 
 int
-io_stats_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno,
-                     struct iatt *prebuf, struct iatt *postbuf)
+update_ios_latency(struct ios_conf *conf, call_frame_t *frame,
+                   glusterfs_fop_t op)
 {
-        END_FOP_LATENCY (frame, WRITE);
-        STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, prebuf, postbuf);
-        return 0;
-}
+    double elapsed;
+    struct timespec *begin, *end;
 
+    begin = &frame->begin;
+    end = &frame->end;
 
+    elapsed = gf_tsdiff(begin, end) / 1000.0;
 
+    update_ios_latency_stats(&conf->cumulative, elapsed, op);
+    update_ios_latency_stats(&conf->incremental, elapsed, op);
+    collect_ios_latency_sample(conf, op, elapsed, frame);
 
-int
-io_stats_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, gf_dirent_t *buf)
+    return 0;
+}
+
+int32_t
+io_stats_dump_stats_to_dict(xlator_t *this, dict_t *resp,
+                            ios_stats_type_t flags, int32_t list_cnt)
 {
-        END_FOP_LATENCY (frame, READDIRP);
-        STACK_UNWIND_STRICT (readdirp, frame, op_ret, op_errno, buf);
-        return 0;
+    struct ios_conf *conf = NULL;
+    int cnt = 0;
+    char key[32];
+    int keylen;
+    struct ios_stat_head *list_head = NULL;
+    struct ios_stat_list *entry = NULL;
+    int ret = -1;
+    ios_stats_thru_t index = IOS_STATS_THRU_MAX;
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    char *dict_timestr = NULL;
+
+    conf = this->private;
+
+    switch (flags) {
+        case IOS_STATS_TYPE_OPEN:
+            list_head = &conf->list[IOS_STATS_TYPE_OPEN];
+            LOCK(&conf->lock);
+            {
+                ret = dict_set_uint64(resp, "current-open",
+                                      conf->cumulative.nr_opens);
+                if (ret)
+                    goto unlock;
+                ret = dict_set_uint64(resp, "max-open",
+                                      conf->cumulative.max_nr_opens);
+
+                gf_time_fmt_tv(timestr, sizeof timestr,
+                               &conf->cumulative.max_openfd_time,
+                               gf_timefmt_FT);
+
+                dict_timestr = gf_strdup(timestr);
+                if (!dict_timestr)
+                    goto unlock;
+                ret = dict_set_dynstr(resp, "max-openfd-time", dict_timestr);
+                if (ret)
+                    goto unlock;
+            }
+        unlock:
+            UNLOCK(&conf->lock);
+            /* Do not proceed if we came here because of some error
+             * during the dict operation */
+            if (ret)
+                goto out;
+            break;
+        case IOS_STATS_TYPE_READ:
+            list_head = &conf->list[IOS_STATS_TYPE_READ];
+            break;
+        case IOS_STATS_TYPE_WRITE:
+            list_head = &conf->list[IOS_STATS_TYPE_WRITE];
+            break;
+        case IOS_STATS_TYPE_OPENDIR:
+            list_head = &conf->list[IOS_STATS_TYPE_OPENDIR];
+            break;
+        case IOS_STATS_TYPE_READDIRP:
+            list_head = &conf->list[IOS_STATS_TYPE_READDIRP];
+            break;
+        case IOS_STATS_TYPE_READ_THROUGHPUT:
+            list_head = &conf->thru_list[IOS_STATS_THRU_READ];
+            index = IOS_STATS_THRU_READ;
+            break;
+        case IOS_STATS_TYPE_WRITE_THROUGHPUT:
+            list_head = &conf->thru_list[IOS_STATS_THRU_WRITE];
+            index = IOS_STATS_THRU_WRITE;
+            break;
+
+        default:
+            goto out;
+    }
+    ret = dict_set_int32_sizen(resp, "top-op", flags);
+    if (!list_cnt)
+        goto out;
+    LOCK(&list_head->lock);
+    {
+        list_for_each_entry(entry, &list_head->iosstats->list, list)
+        {
+            cnt++;
+            keylen = snprintf(key, sizeof(key), "filename-%d", cnt);
+            ret = dict_set_strn(resp, key, keylen, entry->iosstat->filename);
+            if (ret)
+                goto unlock_list_head;
+            snprintf(key, sizeof(key), "value-%d", cnt);
+            ret = dict_set_uint64(resp, key, entry->value);
+            if (ret)
+                goto unlock_list_head;
+            if (index != IOS_STATS_THRU_MAX) {
+                keylen = snprintf(key, sizeof(key), "time-sec-%d", cnt);
+                ret = dict_set_int32n(
+                    resp, key, keylen,
+                    entry->iosstat->thru_counters[index].time.tv_sec);
+                if (ret)
+                    goto unlock_list_head;
+                keylen = snprintf(key, sizeof(key), "time-usec-%d", cnt);
+                ret = dict_set_int32n(
+                    resp, key, keylen,
+                    entry->iosstat->thru_counters[index].time.tv_usec);
+                if (ret)
+                    goto unlock_list_head;
+            }
+            if (cnt == list_cnt)
+                break;
+        }
+    }
+unlock_list_head:
+    UNLOCK(&list_head->lock);
+    /* ret is !=0 if some dict operation in the above critical region
+     * failed. */
+    if (ret)
+        goto out;
+    ret = dict_set_int32_sizen(resp, "members", cnt);
+out:
+    return ret;
 }
 
+static struct ios_stat *
+ios_init_iosstat(xlator_t *this, char *path, uuid_t gfid, inode_t *inode)
+{
+    struct ios_stat *iosstat = NULL;
+    int i = 0;
+
+    iosstat = GF_CALLOC(1, sizeof(*iosstat), gf_io_stats_mt_ios_stat);
+    if (!iosstat)
+        goto out;
+
+    iosstat->filename = gf_strdup(path);
+    gf_uuid_copy(iosstat->gfid, gfid);
+    LOCK_INIT(&iosstat->lock);
+
+    for (i = 0; i < IOS_STATS_TYPE_MAX; i++)
+        GF_ATOMIC_INIT(iosstat->counters[i], 0);
+
+    ios_inode_ctx_set(inode, this, iosstat);
+
+out:
+    return iosstat;
+}
 
 int
-io_stats_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, gf_dirent_t *buf)
+io_stats_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, fd_t *fd, inode_t *inode,
+                    struct iatt *buf, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, READDIR);
-        STACK_UNWIND_STRICT (readdir, frame, op_ret, op_errno, buf);
-        return 0;
-}
+    struct ios_fd *iosfd = NULL;
+    char *path = NULL;
+    struct ios_stat *iosstat = NULL;
+    struct ios_conf *conf = NULL;
+
+    conf = this->private;
+
+    path = frame->local;
+    frame->local = NULL;
+
+    if (!path)
+        goto unwind;
+
+    if (op_ret < 0) {
+        GF_FREE(path);
+        goto unwind;
+    }
+
+    iosfd = GF_CALLOC(1, sizeof(*iosfd), gf_io_stats_mt_ios_fd);
+    if (!iosfd) {
+        GF_FREE(path);
+        goto unwind;
+    }
+
+    iosfd->filename = path;
+    gettimeofday(&iosfd->opened_at, NULL);
+
+    ios_fd_ctx_set(fd, this, iosfd);
+    LOCK(&conf->lock);
+    {
+        conf->cumulative.nr_opens++;
+        if (conf->cumulative.nr_opens > conf->cumulative.max_nr_opens) {
+            conf->cumulative.max_nr_opens = conf->cumulative.nr_opens;
+            conf->cumulative.max_openfd_time = iosfd->opened_at;
+        }
+    }
+    UNLOCK(&conf->lock);
 
+    iosstat = ios_init_iosstat(this, path, buf->ia_gfid, inode);
+    if (!iosstat)
+        GF_FREE(path);
+
+unwind:
+    UPDATE_PROFILE_STATS(frame, CREATE);
+    STACK_UNWIND_STRICT(create, frame, op_ret, op_errno, fd, inode, buf,
+                        preparent, postparent, xdata);
+    return 0;
+}
 
 int
-io_stats_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno,
-                    struct iatt *prebuf, struct iatt *postbuf)
+io_stats_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, FSYNC);
-        STACK_UNWIND_STRICT (fsync, frame, op_ret, op_errno, prebuf, postbuf);
-        return 0;
-}
+    struct ios_fd *iosfd = NULL;
+    char *path = NULL;
+    struct ios_stat *iosstat = NULL;
+    struct ios_conf *conf = NULL;
+    int i = 0;
+
+    conf = this->private;
+    path = frame->local;
+    frame->local = NULL;
+
+    if (!path)
+        goto unwind;
+
+    if (op_ret < 0) {
+        GF_FREE(path);
+        goto unwind;
+    }
+
+    iosfd = GF_CALLOC(1, sizeof(*iosfd), gf_io_stats_mt_ios_fd);
+    if (!iosfd) {
+        GF_FREE(path);
+        goto unwind;
+    }
+
+    iosfd->filename = path;
+    GF_ATOMIC_INIT(iosfd->data_read, 0);
+    GF_ATOMIC_INIT(iosfd->data_written, 0);
+    for (i = 0; i < IOS_BLOCK_COUNT_SIZE; i++) {
+        GF_ATOMIC_INIT(iosfd->block_count_write[i], 0);
+        GF_ATOMIC_INIT(iosfd->block_count_read[i], 0);
+    }
+    gettimeofday(&iosfd->opened_at, NULL);
+
+    ios_fd_ctx_set(fd, this, iosfd);
+
+    ios_inode_ctx_get(fd->inode, this, &iosstat);
+    if (!iosstat) {
+        iosstat = ios_init_iosstat(this, path, fd->inode->gfid, fd->inode);
+    }
+
+    LOCK(&conf->lock);
+    {
+        conf->cumulative.nr_opens++;
+        if (conf->cumulative.nr_opens > conf->cumulative.max_nr_opens) {
+            conf->cumulative.max_nr_opens = conf->cumulative.nr_opens;
+            conf->cumulative.max_openfd_time = iosfd->opened_at;
+        }
+    }
+    UNLOCK(&conf->lock);
+    if (iosstat) {
+        ios_bump_stats(this, iosstat, IOS_STATS_TYPE_OPEN);
+        iosstat = NULL;
+    }
+unwind:
+    UPDATE_PROFILE_STATS(frame, OPEN);
 
+    STACK_UNWIND_STRICT(open, frame, op_ret, op_errno, fd, xdata);
+    return 0;
+}
 
 int
-io_stats_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno,
-                      struct iatt *preop, struct iatt *postop)
+io_stats_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                  dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, SETATTR);
-        STACK_UNWIND_STRICT (setattr, frame, op_ret, op_errno, preop, postop);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, STAT);
+    STACK_UNWIND_STRICT(stat, frame, op_ret, op_errno, buf, xdata);
+    return 0;
 }
 
-
 int
-io_stats_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno,
-                     struct iatt *preparent, struct iatt *postparent)
+io_stats_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iovec *vector,
+                   int32_t count, struct iatt *buf, struct iobref *iobref,
+                   dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, UNLINK);
-        STACK_UNWIND_STRICT (unlink, frame, op_ret, op_errno,
-                             preparent, postparent);
-        return 0;
+    int len = 0;
+    fd_t *fd = NULL;
+    struct ios_stat *iosstat = NULL;
+
+    fd = frame->local;
+    frame->local = NULL;
+
+    if (op_ret > 0) {
+        len = iov_length(vector, count);
+        ios_bump_read(this, fd, len);
+    }
+
+    UPDATE_PROFILE_STATS(frame, READ);
+    ios_inode_ctx_get(fd->inode, this, &iosstat);
+
+    if (iosstat) {
+        ios_bump_stats(this, iosstat, IOS_STATS_TYPE_READ);
+        BUMP_THROUGHPUT(iosstat, IOS_STATS_THRU_READ);
+        iosstat = NULL;
+    }
+
+    STACK_UNWIND_STRICT(readv, frame, op_ret, op_errno, vector, count, buf,
+                        iobref, xdata);
+    return 0;
 }
 
-
 int
-io_stats_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, struct iatt *buf,
-                     struct iatt *preoldparent, struct iatt *postoldparent,
-                     struct iatt *prenewparent, struct iatt *postnewparent)
+io_stats_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                    struct iatt *postbuf, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, RENAME);
-        STACK_UNWIND_STRICT (rename, frame, op_ret, op_errno, buf,
-                             preoldparent, postoldparent,
-                             prenewparent, postnewparent);
-        return 0;
-}
+    struct ios_stat *iosstat = NULL;
+    inode_t *inode = NULL;
+
+    UPDATE_PROFILE_STATS(frame, WRITE);
+    if (frame->local) {
+        inode = frame->local;
+        frame->local = NULL;
+        ios_inode_ctx_get(inode, this, &iosstat);
+        if (iosstat) {
+            ios_bump_stats(this, iosstat, IOS_STATS_TYPE_WRITE);
+            BUMP_THROUGHPUT(iosstat, IOS_STATS_THRU_WRITE);
+            inode = NULL;
+            iosstat = NULL;
+        }
+    }
 
+    STACK_UNWIND_STRICT(writev, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    return 0;
+}
 
 int
-io_stats_readlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, const char *buf,
-                       struct iatt *sbuf)
+io_stats_copy_file_range_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                             int32_t op_ret, int32_t op_errno,
+                             struct iatt *stbuf, struct iatt *prebuf_dst,
+                             struct iatt *postbuf_dst, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, READLINK);
-        STACK_UNWIND_STRICT (readlink, frame, op_ret, op_errno, buf, sbuf);
-        return 0;
-}
+    UPDATE_PROFILE_STATS(frame, COPY_FILE_RANGE);
 
+    STACK_UNWIND_STRICT(copy_file_range, frame, op_ret, op_errno, stbuf,
+                        prebuf_dst, postbuf_dst, xdata);
+    return 0;
+}
 
 int
-io_stats_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno,
-                     inode_t *inode, struct iatt *buf,
-                     dict_t *xattr, struct iatt *postparent)
+io_stats_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, gf_dirent_t *buf,
+                      dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, LOOKUP);
-        STACK_UNWIND_STRICT (lookup, frame, op_ret, op_errno, inode, buf, xattr,
-                             postparent);
-        return 0;
-}
+    struct ios_stat *iosstat = NULL;
+    inode_t *inode = frame->local;
+
+    frame->local = NULL;
+
+    UPDATE_PROFILE_STATS(frame, READDIRP);
 
+    ios_inode_ctx_get(inode, this, &iosstat);
+
+    if (iosstat) {
+        ios_bump_stats(this, iosstat, IOS_STATS_TYPE_READDIRP);
+        iosstat = NULL;
+    }
+
+    STACK_UNWIND_STRICT(readdirp, frame, op_ret, op_errno, buf, xdata);
+    return 0;
+}
 
 int
-io_stats_symlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno,
-                      inode_t *inode, struct iatt *buf,
-                      struct iatt *preparent, struct iatt *postparent)
+io_stats_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, gf_dirent_t *buf,
+                     dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, SYMLINK);
-        STACK_UNWIND_STRICT (symlink, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, READDIR);
+    STACK_UNWIND_STRICT(readdir, frame, op_ret, op_errno, buf, xdata);
+    return 0;
 }
 
-
 int
-io_stats_mknod_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno,
-                    inode_t *inode, struct iatt *buf,
-                    struct iatt *preparent, struct iatt *postparent)
+io_stats_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                   struct iatt *postbuf, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, MKNOD);
-        STACK_UNWIND_STRICT (mknod, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, FSYNC);
+    STACK_UNWIND_STRICT(fsync, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+    return 0;
 }
 
+int
+io_stats_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *preop,
+                     struct iatt *postop, dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, SETATTR);
+    STACK_UNWIND_STRICT(setattr, frame, op_ret, op_errno, preop, postop, xdata);
+    return 0;
+}
 
 int
-io_stats_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno,
-                    inode_t *inode, struct iatt *buf,
-                    struct iatt *preparent, struct iatt *postparent)
+io_stats_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, MKDIR);
-        STACK_UNWIND_STRICT (mkdir, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, UNLINK);
+    STACK_UNWIND_STRICT(unlink, frame, op_ret, op_errno, preparent, postparent,
+                        xdata);
+    return 0;
 }
 
+int
+io_stats_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                    struct iatt *preoldparent, struct iatt *postoldparent,
+                    struct iatt *prenewparent, struct iatt *postnewparent,
+                    dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, RENAME);
+    STACK_UNWIND_STRICT(rename, frame, op_ret, op_errno, buf, preoldparent,
+                        postoldparent, prenewparent, postnewparent, xdata);
+    return 0;
+}
 
 int
-io_stats_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno,
-                   inode_t *inode, struct iatt *buf,
-                   struct iatt *preparent, struct iatt *postparent)
+io_stats_readlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, const char *buf,
+                      struct iatt *sbuf, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, LINK);
-        STACK_UNWIND_STRICT (link, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, READLINK);
+    STACK_UNWIND_STRICT(readlink, frame, op_ret, op_errno, buf, sbuf, xdata);
+    return 0;
 }
 
+int
+io_stats_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, inode_t *inode,
+                    struct iatt *buf, dict_t *xdata, struct iatt *postparent)
+{
+    UPDATE_PROFILE_STATS(frame, LOOKUP);
+    STACK_UNWIND_STRICT(lookup, frame, op_ret, op_errno, inode, buf, xdata,
+                        postparent);
+    return 0;
+}
 
 int
-io_stats_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno)
+io_stats_symlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, inode_t *inode,
+                     struct iatt *buf, struct iatt *preparent,
+                     struct iatt *postparent, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, FLUSH);
-        STACK_UNWIND_STRICT (flush, frame, op_ret, op_errno);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, SYMLINK);
+    STACK_UNWIND_STRICT(symlink, frame, op_ret, op_errno, inode, buf, preparent,
+                        postparent, xdata);
+    return 0;
 }
 
+int
+io_stats_mknod_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, inode_t *inode,
+                   struct iatt *buf, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, MKNOD);
+    STACK_UNWIND_STRICT(mknod, frame, op_ret, op_errno, inode, buf, preparent,
+                        postparent, xdata);
+    return 0;
+}
 
 int
-io_stats_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, fd_t *fd)
+io_stats_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, inode_t *inode,
+                   struct iatt *buf, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
 {
-        if (op_ret >= 0)
-                ios_fd_ctx_set (fd, this, 0);
+    char *path = frame->local;
 
-        END_FOP_LATENCY (frame, OPENDIR);
-        STACK_UNWIND_STRICT (opendir, frame, op_ret, op_errno, fd);
-        return 0;
-}
+    if (!path)
+        goto unwind;
+
+    UPDATE_PROFILE_STATS(frame, MKDIR);
+    if (op_ret < 0)
+        goto unwind;
+
+    /* allocate a struct ios_stat and set the inode ctx */
+    ios_init_iosstat(this, path, buf->ia_gfid, inode);
 
+unwind:
+    /* local is assigned with path */
+    GF_FREE(frame->local);
+    frame->local = NULL;
+    STACK_UNWIND_STRICT(mkdir, frame, op_ret, op_errno, inode, buf, preparent,
+                        postparent, xdata);
+    return 0;
+}
 
 int
-io_stats_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno,
-                    struct iatt *preparent, struct iatt *postparent)
+io_stats_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, inode_t *inode,
+                  struct iatt *buf, struct iatt *preparent,
+                  struct iatt *postparent, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, RMDIR);
-        STACK_UNWIND_STRICT (rmdir, frame, op_ret, op_errno,
-                             preparent, postparent);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, LINK);
+    STACK_UNWIND_STRICT(link, frame, op_ret, op_errno, inode, buf, preparent,
+                        postparent, xdata);
+    return 0;
 }
 
-
 int
-io_stats_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno,
-                       struct iatt *prebuf, struct iatt *postbuf)
+io_stats_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, TRUNCATE);
-        STACK_UNWIND_STRICT (truncate, frame, op_ret, op_errno,
-                             prebuf, postbuf);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, FLUSH);
+    STACK_UNWIND_STRICT(flush, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
-
 int
-io_stats_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, struct statvfs *buf)
+io_stats_opendir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, STATFS);
-        STACK_UNWIND_STRICT (statfs, frame, op_ret, op_errno, buf);
-        return 0;
+    struct ios_stat *iosstat = NULL;
+    int ret = -1;
+
+    UPDATE_PROFILE_STATS(frame, OPENDIR);
+    if (op_ret < 0)
+        goto unwind;
+
+    ios_fd_ctx_set(fd, this, 0);
+
+    ret = ios_inode_ctx_get(fd->inode, this, &iosstat);
+    if (!ret)
+        ios_bump_stats(this, iosstat, IOS_STATS_TYPE_OPENDIR);
+
+unwind:
+    STACK_UNWIND_STRICT(opendir, frame, op_ret, op_errno, fd, xdata);
+    return 0;
 }
 
+int
+io_stats_rmdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, RMDIR);
+
+    STACK_UNWIND_STRICT(rmdir, frame, op_ret, op_errno, preparent, postparent,
+                        xdata);
+    return 0;
+}
 
 int
-io_stats_setxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno)
+io_stats_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                      struct iatt *postbuf, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, SETXATTR);
-        STACK_UNWIND_STRICT (setxattr, frame, op_ret, op_errno);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, TRUNCATE);
+    STACK_UNWIND_STRICT(truncate, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    return 0;
 }
 
+int
+io_stats_statfs_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct statvfs *buf,
+                    dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, STATFS);
+    STACK_UNWIND_STRICT(statfs, frame, op_ret, op_errno, buf, xdata);
+    return 0;
+}
 
 int
-io_stats_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, dict_t *dict)
+io_stats_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, GETXATTR);
-        STACK_UNWIND_STRICT (getxattr, frame, op_ret, op_errno, dict);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, SETXATTR);
+    STACK_UNWIND_STRICT(setxattr, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
+int
+io_stats_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *dict,
+                      dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, GETXATTR);
+    STACK_UNWIND_STRICT(getxattr, frame, op_ret, op_errno, dict, xdata);
+    return 0;
+}
 
 int
-io_stats_removexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t op_ret, int32_t op_errno)
+io_stats_removexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, REMOVEXATTR);
-        STACK_UNWIND_STRICT (removexattr, frame, op_ret, op_errno);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, REMOVEXATTR);
+    STACK_UNWIND_STRICT(removexattr, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
+int
+io_stats_fsetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, FSETXATTR);
+    STACK_UNWIND_STRICT(fsetxattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
 int
-io_stats_fsyncdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno)
+io_stats_fgetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *dict,
+                       dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, FSYNCDIR);
-        STACK_UNWIND_STRICT (fsyncdir, frame, op_ret, op_errno);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, FGETXATTR);
+    STACK_UNWIND_STRICT(fgetxattr, frame, op_ret, op_errno, dict, xdata);
+    return 0;
 }
 
+int
+io_stats_fremovexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, FREMOVEXATTR);
+    STACK_UNWIND_STRICT(fremovexattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
 int
-io_stats_access_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno)
+io_stats_fsyncdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, ACCESS);
-        STACK_UNWIND_STRICT (access, frame, op_ret, op_errno);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, FSYNCDIR);
+    STACK_UNWIND_STRICT(fsyncdir, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
+int
+io_stats_access_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, ACCESS);
+    STACK_UNWIND_STRICT(access, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
 int
-io_stats_ftruncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno,
-                        struct iatt *prebuf, struct iatt *postbuf)
+io_stats_ftruncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                       struct iatt *postbuf, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, FTRUNCATE);
-        STACK_UNWIND_STRICT (ftruncate, frame, op_ret, op_errno,
-                             prebuf, postbuf);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, FTRUNCATE);
+    STACK_UNWIND_STRICT(ftruncate, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    return 0;
 }
 
+int
+io_stats_fstat_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                   dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, FSTAT);
+    STACK_UNWIND_STRICT(fstat, frame, op_ret, op_errno, buf, xdata);
+    return 0;
+}
 
 int
-io_stats_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, struct iatt *buf)
+io_stats_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                       struct iatt *postbuf, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, FSTAT);
-        STACK_UNWIND_STRICT (fstat, frame, op_ret, op_errno, buf);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, FALLOCATE);
+    STACK_UNWIND_STRICT(fallocate, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    return 0;
 }
 
+int
+io_stats_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, DISCARD);
+    STACK_UNWIND_STRICT(discard, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    return 0;
+}
 
 int
-io_stats_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno, struct gf_flock *lock)
+io_stats_zerofill_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                      struct iatt *postbuf, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, LK);
-        STACK_UNWIND_STRICT (lk, frame, op_ret, op_errno, lock);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, ZEROFILL);
+    STACK_UNWIND_STRICT(zerofill, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    return 0;
 }
 
+int32_t
+io_stats_ipc_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, IPC);
+    STACK_UNWIND_STRICT(ipc, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
 int
-io_stats_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno)
+io_stats_lk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct gf_flock *lock,
+                dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, ENTRYLK);
-        STACK_UNWIND_STRICT (entrylk, frame, op_ret, op_errno);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, LK);
+    STACK_UNWIND_STRICT(lk, frame, op_ret, op_errno, lock, xdata);
+    return 0;
 }
 
+int
+io_stats_entrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, ENTRYLK);
+    STACK_UNWIND_STRICT(entrylk, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
 int
-io_stats_xattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, dict_t *dict)
+io_stats_fentrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, XATTROP);
-        STACK_UNWIND_STRICT (xattrop, frame, op_ret, op_errno, dict);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, FENTRYLK);
+    STACK_UNWIND_STRICT(fentrylk, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
+int
+io_stats_rchecksum_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, uint32_t weak_checksum,
+                       uint8_t *strong_checksum, dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, RCHECKSUM);
+    STACK_UNWIND_STRICT(rchecksum, frame, op_ret, op_errno, weak_checksum,
+                        strong_checksum, xdata);
+    return 0;
+}
 
 int
-io_stats_fxattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, dict_t *dict)
+io_stats_seek_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, off_t offset, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, FXATTROP);
-        STACK_UNWIND_STRICT (fxattrop, frame, op_ret, op_errno, dict);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, SEEK);
+    STACK_UNWIND_STRICT(seek, frame, op_ret, op_errno, offset, xdata);
+    return 0;
 }
 
+int
+io_stats_lease_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct gf_lease *lease,
+                   dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, LEASE);
+    STACK_UNWIND_STRICT(lease, frame, op_ret, op_errno, lease, xdata);
+    return 0;
+}
 
 int
-io_stats_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno)
+io_stats_getactivelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno,
+                         lock_migration_info_t *locklist, dict_t *xdata)
 {
-        END_FOP_LATENCY (frame, INODELK);
-        STACK_UNWIND_STRICT (inodelk, frame, op_ret, op_errno);
-        return 0;
+    UPDATE_PROFILE_STATS(frame, GETACTIVELK);
+    STACK_UNWIND_STRICT(getactivelk, frame, op_ret, op_errno, locklist, xdata);
+    return 0;
 }
 
+int
+io_stats_setactivelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, SETACTIVELK);
+    STACK_UNWIND_STRICT(setactivelk, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
 int
-io_stats_entrylk (call_frame_t *frame, xlator_t *this,
-                  const char *volume, loc_t *loc, const char *basename,
-                  entrylk_cmd cmd, entrylk_type type)
+io_stats_compound_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, void *data,
+                      dict_t *xdata)
 {
-        BUMP_FOP (ENTRYLK);
+    UPDATE_PROFILE_STATS(frame, COMPOUND);
+    STACK_UNWIND_STRICT(compound, frame, op_ret, op_errno, data, xdata);
+    return 0;
+}
 
-        START_FOP_LATENCY (frame);
+int
+io_stats_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, dict_t *dict,
+                     dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, XATTROP);
+    STACK_UNWIND_STRICT(xattrop, frame, op_ret, op_errno, dict, xdata);
+    return 0;
+}
 
-        STACK_WIND (frame, io_stats_entrylk_cbk,
-                    FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->entrylk,
-                    volume, loc, basename, cmd, type);
-        return 0;
+int
+io_stats_fxattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *dict,
+                      dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, FXATTROP);
+    STACK_UNWIND_STRICT(fxattrop, frame, op_ret, op_errno, dict, xdata);
+    return 0;
 }
 
+int
+io_stats_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    UPDATE_PROFILE_STATS(frame, INODELK);
+    STACK_UNWIND_STRICT(inodelk, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
 int
-io_stats_inodelk (call_frame_t *frame, xlator_t *this,
-                  const char *volume, loc_t *loc, int32_t cmd, struct gf_flock *flock)
+io_stats_entrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+                 loc_t *loc, const char *basename, entrylk_cmd cmd,
+                 entrylk_type type, dict_t *xdata)
 {
+    START_FOP_LATENCY(frame);
 
-        BUMP_FOP (INODELK);
+    STACK_WIND(frame, io_stats_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, volume, loc, basename, cmd,
+               type, xdata);
+    return 0;
+}
 
-        START_FOP_LATENCY (frame);
+int
+io_stats_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+                  fd_t *fd, const char *basename, entrylk_cmd cmd,
+                  entrylk_type type, dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_inodelk_cbk,
-                    FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->inodelk,
-                    volume, loc, cmd, flock);
-        return 0;
+    STACK_WIND(frame, io_stats_fentrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fentrylk, volume, fd, basename, cmd,
+               type, xdata);
+    return 0;
 }
 
-
 int
-io_stats_finodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno)
+io_stats_inodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+                 loc_t *loc, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
 {
+    START_FOP_LATENCY(frame);
 
-        END_FOP_LATENCY (frame, FINODELK);
-        STACK_UNWIND_STRICT (finodelk, frame, op_ret, op_errno);
-        return 0;
+    STACK_WIND(frame, io_stats_inodelk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->inodelk, volume, loc, cmd, flock,
+               xdata);
+    return 0;
 }
 
-
 int
-io_stats_finodelk (call_frame_t *frame, xlator_t *this,
-                   const char *volume, fd_t *fd, int32_t cmd, struct gf_flock *flock)
+io_stats_finodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        BUMP_FOP (FINODELK);
+    UPDATE_PROFILE_STATS(frame, FINODELK);
+    STACK_UNWIND_STRICT(finodelk, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
-        START_FOP_LATENCY (frame);
+int
+io_stats_finodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+                  fd_t *fd, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_finodelk_cbk,
-                    FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->finodelk,
-                    volume, fd, cmd, flock);
-        return 0;
+    STACK_WIND(frame, io_stats_finodelk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->finodelk, volume, fd, cmd, flock,
+               xdata);
+    return 0;
 }
 
-
 int
-io_stats_xattrop (call_frame_t *frame, xlator_t *this,
-                  loc_t *loc, gf_xattrop_flags_t flags, dict_t *dict)
+io_stats_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                 gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
 {
-        BUMP_FOP (XATTROP);
+    START_FOP_LATENCY(frame);
 
-        START_FOP_LATENCY (frame);
+    STACK_WIND(frame, io_stats_xattrop_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->xattrop, loc, flags, dict, xdata);
+    return 0;
+}
 
-        STACK_WIND (frame, io_stats_xattrop_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->xattrop,
-                    loc, flags, dict);
+int
+io_stats_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                  gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
 
-        return 0;
+    STACK_WIND(frame, io_stats_fxattrop_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fxattrop, fd, flags, dict, xdata);
+    return 0;
 }
 
-
 int
-io_stats_fxattrop (call_frame_t *frame, xlator_t *this,
-                   fd_t *fd, gf_xattrop_flags_t flags, dict_t *dict)
+io_stats_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-        BUMP_FOP (FXATTROP);
+    START_FOP_LATENCY(frame);
 
-        START_FOP_LATENCY (frame);
+    STACK_WIND(frame, io_stats_lookup_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lookup, loc, xdata);
+    return 0;
+}
 
-        STACK_WIND (frame, io_stats_fxattrop_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fxattrop,
-                    fd, flags, dict);
+int
+io_stats_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
 
-        return 0;
+    STACK_WIND(frame, io_stats_stat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->stat, loc, xdata);
+    return 0;
 }
 
-
 int
-io_stats_lookup (call_frame_t *frame, xlator_t *this,
-                 loc_t *loc, dict_t *xattr_req)
+io_stats_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+                  dict_t *xdata)
 {
-        BUMP_FOP (LOOKUP);
+    START_FOP_LATENCY(frame);
 
-        START_FOP_LATENCY (frame);
+    STACK_WIND(frame, io_stats_readlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readlink, loc, size, xdata);
+    return 0;
+}
 
-        STACK_WIND (frame, io_stats_lookup_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->lookup,
-                    loc, xattr_req);
+int
+io_stats_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+               dev_t dev, mode_t umask, dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
 
-        return 0;
+    STACK_WIND(frame, io_stats_mknod_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mknod, loc, mode, dev, umask, xdata);
+    return 0;
 }
 
-
 int
-io_stats_stat (call_frame_t *frame, xlator_t *this, loc_t *loc)
+io_stats_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+               mode_t umask, dict_t *xdata)
 {
-        BUMP_FOP (STAT);
-
-        START_FOP_LATENCY (frame);
+    if (loc->path)
+        frame->local = gf_strdup(loc->path);
 
-        STACK_WIND (frame, io_stats_stat_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->stat,
-                    loc);
+    START_FOP_LATENCY(frame);
 
-        return 0;
+    STACK_WIND(frame, io_stats_mkdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mkdir, loc, mode, umask, xdata);
+    return 0;
 }
 
-
 int
-io_stats_readlink (call_frame_t *frame, xlator_t *this,
-                   loc_t *loc, size_t size)
+io_stats_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+                dict_t *xdata)
 {
-        BUMP_FOP (READLINK);
-
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_readlink_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->readlink,
-                    loc, size);
-
-        return 0;
+    STACK_WIND(frame, io_stats_unlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->unlink, loc, xflag, xdata);
+    return 0;
 }
 
-
 int
-io_stats_mknod (call_frame_t *frame, xlator_t *this,
-                loc_t *loc, mode_t mode, dev_t dev, dict_t *params)
+io_stats_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+               dict_t *xdata)
 {
-        BUMP_FOP (MKNOD);
+    START_FOP_LATENCY(frame);
 
-        START_FOP_LATENCY (frame);
+    STACK_WIND(frame, io_stats_rmdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rmdir, loc, flags, xdata);
+    return 0;
+}
 
-        STACK_WIND (frame, io_stats_mknod_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->mknod,
-                    loc, mode, dev, params);
+int
+io_stats_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+                 loc_t *loc, mode_t umask, dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
 
-        return 0;
+    STACK_WIND(frame, io_stats_symlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->symlink, linkpath, loc, umask, xdata);
+    return 0;
 }
 
-
 int
-io_stats_mkdir (call_frame_t *frame, xlator_t *this,
-                loc_t *loc, mode_t mode, dict_t *params)
+io_stats_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc,
+                loc_t *newloc, dict_t *xdata)
 {
-        BUMP_FOP (MKDIR);
+    START_FOP_LATENCY(frame);
 
-        START_FOP_LATENCY (frame);
-
-        STACK_WIND (frame, io_stats_mkdir_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->mkdir,
-                    loc, mode, params);
-        return 0;
+    STACK_WIND(frame, io_stats_rename_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
+    return 0;
 }
 
-
 int
-io_stats_unlink (call_frame_t *frame, xlator_t *this,
-                 loc_t *loc)
+io_stats_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+              dict_t *xdata)
 {
-        BUMP_FOP (UNLINK);
-
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_unlink_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->unlink,
-                    loc);
-        return 0;
+    STACK_WIND(frame, io_stats_link_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->link, oldloc, newloc, xdata);
+    return 0;
 }
 
-
 int
-io_stats_rmdir (call_frame_t *frame, xlator_t *this,
-                loc_t *loc, int flags)
+io_stats_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                 struct iatt *stbuf, int32_t valid, dict_t *xdata)
 {
-        BUMP_FOP (RMDIR);
+    START_FOP_LATENCY(frame);
 
-        START_FOP_LATENCY (frame);
+    STACK_WIND(frame, io_stats_setattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setattr, loc, stbuf, valid, xdata);
+    return 0;
+}
 
-        STACK_WIND (frame, io_stats_rmdir_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->rmdir,
-                    loc, flags);
+int
+io_stats_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+                  dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
 
-        return 0;
+    STACK_WIND(frame, io_stats_truncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+    return 0;
 }
 
-
 int
-io_stats_symlink (call_frame_t *frame, xlator_t *this,
-                  const char *linkpath, loc_t *loc, dict_t *params)
+io_stats_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+              fd_t *fd, dict_t *xdata)
 {
-        BUMP_FOP (SYMLINK);
+    if (loc->path)
+        frame->local = gf_strdup(loc->path);
 
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_symlink_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->symlink,
-                    linkpath, loc, params);
-
-        return 0;
+    STACK_WIND(frame, io_stats_open_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->open, loc, flags, fd, xdata);
+    return 0;
 }
 
-
 int
-io_stats_rename (call_frame_t *frame, xlator_t *this,
-                 loc_t *oldloc, loc_t *newloc)
+io_stats_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+                mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
 {
-        BUMP_FOP (RENAME);
-
-        START_FOP_LATENCY (frame);
+    if (loc->path)
+        frame->local = gf_strdup(loc->path);
 
-        STACK_WIND (frame, io_stats_rename_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->rename,
-                    oldloc, newloc);
+    START_FOP_LATENCY(frame);
 
-        return 0;
+    STACK_WIND(frame, io_stats_create_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->create, loc, flags, mode, umask, fd,
+               xdata);
+    return 0;
 }
 
-
 int
-io_stats_link (call_frame_t *frame, xlator_t *this,
-               loc_t *oldloc, loc_t *newloc)
+io_stats_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+               off_t offset, uint32_t flags, dict_t *xdata)
 {
-        BUMP_FOP (LINK);
+    frame->local = fd;
 
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_link_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->link,
-                    oldloc, newloc);
-        return 0;
+    STACK_WIND(frame, io_stats_readv_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readv, fd, size, offset, flags, xdata);
+    return 0;
 }
 
-
 int
-io_stats_setattr (call_frame_t *frame, xlator_t *this,
-                  loc_t *loc, struct iatt *stbuf, int32_t valid)
+io_stats_writev(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                struct iovec *vector, int32_t count, off_t offset,
+                uint32_t flags, struct iobref *iobref, dict_t *xdata)
 {
-        BUMP_FOP (SETATTR);
+    int len = 0;
 
-        START_FOP_LATENCY (frame);
+    if (fd->inode)
+        frame->local = fd->inode;
+    len = iov_length(vector, count);
 
-        STACK_WIND (frame, io_stats_setattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->setattr,
-                    loc, stbuf, valid);
+    ios_bump_write(this, fd, len);
+    START_FOP_LATENCY(frame);
 
-        return 0;
+    STACK_WIND(frame, io_stats_writev_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->writev, fd, vector, count, offset,
+               flags, iobref, xdata);
+    return 0;
 }
 
-
 int
-io_stats_truncate (call_frame_t *frame, xlator_t *this,
-                   loc_t *loc, off_t offset)
+io_stats_copy_file_range(call_frame_t *frame, xlator_t *this, fd_t *fd_in,
+                         off_t off_in, fd_t *fd_out, off_t off_out, size_t len,
+                         uint32_t flags, dict_t *xdata)
 {
-        BUMP_FOP (TRUNCATE);
+    START_FOP_LATENCY(frame);
 
-        START_FOP_LATENCY (frame);
+    STACK_WIND(frame, io_stats_copy_file_range_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->copy_file_range, fd_in, off_in, fd_out,
+               off_out, len, flags, xdata);
+    return 0;
+}
 
-        STACK_WIND (frame, io_stats_truncate_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->truncate,
-                    loc, offset);
+int
+io_stats_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
 
-        return 0;
+    STACK_WIND(frame, io_stats_statfs_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->statfs, loc, xdata);
+    return 0;
 }
 
-
 int
-io_stats_open (call_frame_t *frame, xlator_t *this,
-               loc_t *loc, int32_t flags, fd_t *fd, int32_t wbflags)
+io_stats_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
-        BUMP_FOP (OPEN);
+    START_FOP_LATENCY(frame);
 
-        frame->local = gf_strdup (loc->path);
+    STACK_WIND(frame, io_stats_flush_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->flush, fd, xdata);
+    return 0;
+}
 
-        START_FOP_LATENCY (frame);
+int
+io_stats_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+               dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_open_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->open,
-                    loc, flags, fd, wbflags);
-        return 0;
+    STACK_WIND(frame, io_stats_fsync_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsync, fd, flags, xdata);
+    return 0;
 }
 
+int
+conditional_dump(dict_t *dict, char *key, data_t *value, void *data)
+{
+    struct {
+        xlator_t *this;
+        inode_t *inode;
+        const char *path;
+    } * stub;
+    xlator_t *this = NULL;
+    char *filename = NULL;
+    FILE *logfp = NULL;
+    struct ios_dump_args args = {0};
+    int pid, namelen, dirlen;
+    char dump_key[100];
+    char *slash_ptr = NULL;
+    char *path_in_value = NULL;
+    char *identifier = NULL;
+    struct ios_conf *conf = NULL;
+
+    stub = data;
+    this = stub->this;
+    conf = this->private;
+
+    /* Don't do this on 'brick-side', only do this on client side */
+    /* Addresses CVE-2018-14659 */
+    if (this->ctx->process_mode != GF_CLIENT_PROCESS) {
+        gf_log(this->name, GF_LOG_DEBUG,
+               "taking io-stats dump using setxattr not permitted on brick."
+               " Use 'gluster profile' instead");
+        return -1;
+    }
+
+    /* Create a file name that is appended with the io-stats instance
+    name as well. This helps when there is more than a single io-stats
+    instance in the graph, or the client and server processes are running
+    on the same node */
+    /* For the sanity of where the file should be located, we should make
+       sure file is written only inside RUNDIR (ie, /var/run/gluster) */
+    /* TODO: provide an option to dump it to different directory of
+       choice, based on options */
+    /* name format: /var/run/gluster/<passed in path/filename>.<xlator name
+     * slashes to -> */
+
+    path_in_value = alloca0(value->len + 1);
+
+    /* We need a memcpy here because of the way dict_unserialize works */
+
+    memcpy(path_in_value, data_to_str(value), value->len);
+    path_in_value[value->len] = '\0';
+
+    if (strstr(path_in_value, "../")) {
+        gf_log(this->name, GF_LOG_ERROR, "%s: no \"../\" allowed in path",
+               path_in_value);
+        return -1;
+    }
+
+    if (path_in_value[0] == '/') {
+        path_in_value = path_in_value + 1;
+    }
+
+    dirlen = strlen(IOS_STATS_DUMP_DIR);
+    if (conf->unique_id) {
+        /* this->name will be the same for all bricks of the volume */
+        identifier = conf->unique_id;
+    } else {
+        identifier = this->name;
+    }
+
+    namelen = (dirlen + value->len + strlen(identifier) + 3);
+    /* +3 for '/', '.' and '\0' added in snprintf below*/
+
+    filename = alloca0(namelen);
+    snprintf(filename, namelen, "%s/%s.%s", IOS_STATS_DUMP_DIR, path_in_value,
+             identifier);
+
+    /* convert any slashes to '-' so that fopen works correctly */
+    slash_ptr = strchr(filename + dirlen + 1, '/');
+    while (slash_ptr) {
+        *slash_ptr = '-';
+        slash_ptr = strchr(slash_ptr, '/');
+    }
+
+    pid = getpid();
+
+    if (!strncmp(filename, "", 1)) {
+        gf_log(this->name, GF_LOG_ERROR, "No filename given");
+        return -1;
+    }
+    logfp = fopen(filename, "w+");
+    if (!logfp) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to open %s "
+               "for writing",
+               filename);
+        return -1;
+    }
+    sprintf(dump_key, "*io*stat*%d_json_dump", pid);
+    if (fnmatch(dump_key, key, 0) == 0) {
+        (void)ios_dump_args_init(&args, IOS_DUMP_TYPE_JSON_FILE, logfp);
+    } else {
+        (void)ios_dump_args_init(&args, IOS_DUMP_TYPE_FILE, logfp);
+    }
+    io_stats_dump(this, &args, GF_IOS_INFO_ALL, _gf_false);
+    fclose(logfp);
+    return 0;
+}
 
 int
-io_stats_create (call_frame_t *frame, xlator_t *this,
-                 loc_t *loc, int32_t flags, mode_t mode,
-                 fd_t *fd, dict_t *params)
+_ios_destroy_dump_thread(struct ios_conf *conf)
 {
-        BUMP_FOP (CREATE);
+    conf->dump_thread_should_die = _gf_true;
+    if (conf->dump_thread_running) {
+        (void)pthread_cancel(conf->dump_thread);
+        (void)pthread_join(conf->dump_thread, NULL);
+    }
+    return 0;
+}
 
-        frame->local = gf_strdup (loc->path);
+void *
+_ios_dump_thread(xlator_t *this)
+{
+    struct ios_conf *conf = NULL;
+    FILE *stats_logfp = NULL;
+    FILE *samples_logfp = NULL;
+    struct ios_dump_args args = {0};
+    int i;
+    int stats_bytes_written = 0;
+    int samples_bytes_written = 0;
+    char stats_filename[PATH_MAX];
+    char samples_filename[PATH_MAX];
+    char *xlator_name;
+    char *instance_name;
+    gf_boolean_t log_stats_fopen_failure = _gf_true;
+    gf_boolean_t log_samples_fopen_failure = _gf_true;
+    int old_cancel_type;
+
+    conf = this->private;
+    gf_log(this->name, GF_LOG_INFO,
+           "IO stats dump thread started, "
+           "polling IO stats every %d seconds",
+           conf->ios_dump_interval);
+    xlator_name = strdupa(conf->unique_id);
+    for (i = 0; i < strlen(xlator_name); i++) {
+        if (xlator_name[i] == '/')
+            xlator_name[i] = '_';
+    }
+    instance_name = this->instance_name;
+    if (this->name && strcmp(this->name, "glustershd") == 0) {
+        xlator_name = "shd";
+    } else if (this->prev && strcmp(this->prev->name, "nfs-server") == 0) {
+        xlator_name = "nfsd";
+        instance_name = this->prev->instance_name;
+    }
+    if (sys_mkdir(_IOS_DUMP_DIR, S_IRWXU | S_IRWXO | S_IRWXG) == (-1)) {
+        if (errno != EEXIST) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "could not create stats-dump directory %s", _IOS_DUMP_DIR);
+            goto out;
+        }
+    }
+    if (sys_mkdir(_IOS_SAMP_DIR, S_IRWXU | S_IRWXO | S_IRWXG) == (-1)) {
+        if (errno != EEXIST) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "could not create stats-sample directory %s", _IOS_SAMP_DIR);
+            goto out;
+        }
+    }
+    if (instance_name) {
+        stats_bytes_written = snprintf(stats_filename, PATH_MAX,
+                                       "%s/%s_%s_%s.dump", _IOS_DUMP_DIR,
+                                       __progname, xlator_name, instance_name);
+        samples_bytes_written = snprintf(
+            samples_filename, PATH_MAX, "%s/%s_%s_%s.samp", _IOS_SAMP_DIR,
+            __progname, xlator_name, instance_name);
+    } else {
+        stats_bytes_written = snprintf(stats_filename, PATH_MAX,
+                                       "%s/%s_%s.dump", _IOS_DUMP_DIR,
+                                       __progname, xlator_name);
+        samples_bytes_written = snprintf(samples_filename, PATH_MAX,
+                                         "%s/%s_%s.samp", _IOS_SAMP_DIR,
+                                         __progname, xlator_name);
+    }
+    if ((stats_bytes_written >= PATH_MAX) ||
+        (samples_bytes_written >= PATH_MAX)) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Invalid path for stats dump (%s) and/or latency "
+               "samples (%s)",
+               stats_filename, samples_filename);
+        goto out;
+    }
+    while (1) {
+        if (conf->dump_thread_should_die)
+            break;
+        (void)pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS,
+                                    &old_cancel_type);
+        sleep(conf->ios_dump_interval);
+        (void)pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, &old_cancel_type);
+        /*
+         * It's not clear whether we should reopen this each time, or
+         * just hold it open and rewind/truncate on each iteration.
+         * Leaving it alone for now.
+         */
+        stats_logfp = fopen(stats_filename, "w+");
+        if (stats_logfp) {
+            (void)ios_dump_args_init(&args, conf->dump_format, stats_logfp);
+            io_stats_dump(this, &args, GF_IOS_INFO_ALL, _gf_false);
+            fclose(stats_logfp);
+            log_stats_fopen_failure = _gf_true;
+        } else if (log_stats_fopen_failure) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "could not open stats-dump file %s (%s)", stats_filename,
+                   strerror(errno));
+            log_stats_fopen_failure = _gf_false;
+        }
+        samples_logfp = fopen(samples_filename, "w+");
+        if (samples_logfp) {
+            io_stats_dump_latency_samples_logfp(this, samples_logfp);
+            fclose(samples_logfp);
+            log_samples_fopen_failure = _gf_true;
+        } else if (log_samples_fopen_failure) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "could not open samples-dump file %s (%s)", samples_filename,
+                   strerror(errno));
+            log_samples_fopen_failure = _gf_false;
+        }
+    }
+out:
+    conf->dump_thread_running = _gf_false;
+    gf_log(this->name, GF_LOG_INFO, "IO stats dump thread terminated");
+    return NULL;
+}
 
-        START_FOP_LATENCY (frame);
+static gf_boolean_t
+match_special_xattr(dict_t *d, char *k, data_t *val, void *mdata)
+{
+    gf_boolean_t ret = _gf_false;
+    if (fnmatch("*io*stat*dump", k, 0) == 0) {
+        ret = _gf_true;
+    }
 
-        STACK_WIND (frame, io_stats_create_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->create,
-                    loc, flags, mode, fd, params);
-        return 0;
+    return ret;
 }
 
-
 int
-io_stats_readv (call_frame_t *frame, xlator_t *this,
-                fd_t *fd, size_t size, off_t offset)
+io_stats_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+                  int32_t flags, dict_t *xdata)
 {
-        BUMP_FOP (READ);
+    struct {
+        xlator_t *this;
+        inode_t *inode;
+        const char *path;
+    } stub;
 
-        frame->local = fd;
+    stub.this = this;
+    stub.inode = loc->inode;
+    stub.path = loc->path;
 
-        START_FOP_LATENCY (frame);
+    (void)dict_foreach_match(dict, match_special_xattr, NULL, conditional_dump,
+                             &stub);
 
-        STACK_WIND (frame, io_stats_readv_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->readv,
-                    fd, size, offset);
-        return 0;
-}
+    START_FOP_LATENCY(frame);
 
+    STACK_WIND(frame, io_stats_setxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setxattr, loc, dict, flags, xdata);
+    return 0;
+}
 
 int
-io_stats_writev (call_frame_t *frame, xlator_t *this,
-                 fd_t *fd, struct iovec *vector,
-                 int32_t count, off_t offset,
-                 struct iobref *iobref)
+io_stats_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                  const char *name, dict_t *xdata)
 {
-        int                 len = 0;
+    START_FOP_LATENCY(frame);
 
+    STACK_WIND(frame, io_stats_getxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->getxattr, loc, name, xdata);
+    return 0;
+}
 
-        len = iov_length (vector, count);
+int
+io_stats_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                     const char *name, dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
 
-        BUMP_FOP (WRITE);
-        BUMP_WRITE (fd, len);
+    STACK_WIND(frame, io_stats_removexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->removexattr, loc, name, xdata);
+    return 0;
+}
 
-        START_FOP_LATENCY (frame);
+int
+io_stats_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+                   int32_t flags, dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_writev_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->writev,
-                    fd, vector, count, offset, iobref);
-        return 0;
+    STACK_WIND(frame, io_stats_fsetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
+    return 0;
 }
 
-
 int
-io_stats_statfs (call_frame_t *frame, xlator_t *this,
-                 loc_t *loc)
+io_stats_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                   const char *name, dict_t *xdata)
 {
-        BUMP_FOP (STATFS);
-
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_statfs_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->statfs,
-                    loc);
-        return 0;
+    STACK_WIND(frame, io_stats_fgetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fgetxattr, fd, name, xdata);
+    return 0;
 }
 
-
 int
-io_stats_flush (call_frame_t *frame, xlator_t *this,
-                fd_t *fd)
+io_stats_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                      const char *name, dict_t *xdata)
 {
-        BUMP_FOP (FLUSH);
+    START_FOP_LATENCY(frame);
 
-        START_FOP_LATENCY (frame);
-
-        STACK_WIND (frame, io_stats_flush_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->flush,
-                    fd);
-        return 0;
+    STACK_WIND(frame, io_stats_fremovexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fremovexattr, fd, name, xdata);
+    return 0;
 }
 
-
 int
-io_stats_fsync (call_frame_t *frame, xlator_t *this,
-                fd_t *fd, int32_t flags)
+io_stats_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+                 dict_t *xdata)
 {
-        BUMP_FOP (FSYNC);
+    START_FOP_LATENCY(frame);
 
-        START_FOP_LATENCY (frame);
-
-        STACK_WIND (frame, io_stats_fsync_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fsync,
-                    fd, flags);
-        return 0;
+    STACK_WIND(frame, io_stats_opendir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->opendir, loc, fd, xdata);
+    return 0;
 }
 
+int
+io_stats_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                  off_t offset, dict_t *dict)
+{
+    frame->local = fd->inode;
+    START_FOP_LATENCY(frame);
 
-void
-conditional_dump (dict_t *dict, char *key, data_t *value, void *data)
-{
-        struct {
-                xlator_t       *this;
-                inode_t        *inode;
-                const char     *path;
-        } *stub;
-        xlator_t   *this = NULL;
-        inode_t    *inode = NULL;
-        const char *path = NULL;
-        char       *filename = NULL;
-
-        stub  = data;
-        this  = stub->this;
-        inode = stub->inode;
-        path  = stub->path;
-
-        filename = alloca (value->len + 1);
-        memset (filename, 0, value->len + 1);
-        memcpy (filename, data_to_str (value), value->len);
-
-        if (fnmatch ("*io*stat*dump", key, 0) == 0) {
-                io_stats_dump (this, filename, inode, path);
-        }
+    STACK_WIND(frame, io_stats_readdirp_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdirp, fd, size, offset, dict);
+    return 0;
 }
 
-
 int
-io_stats_setxattr (call_frame_t *frame, xlator_t *this,
-                   loc_t *loc, dict_t *dict,
-                   int32_t flags)
+io_stats_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                 off_t offset, dict_t *xdata)
 {
-        struct {
-                xlator_t     *this;
-                inode_t      *inode;
-                const char   *path;
-        } stub;
+    START_FOP_LATENCY(frame);
 
-        BUMP_FOP (SETXATTR);
+    STACK_WIND(frame, io_stats_readdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdir, fd, size, offset, xdata);
+    return 0;
+}
 
-        stub.this  = this;
-        stub.inode = loc->inode;
-        stub.path  = loc->path;
+int
+io_stats_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                  int32_t datasync, dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
 
-        dict_foreach (dict, conditional_dump, &stub);
+    STACK_WIND(frame, io_stats_fsyncdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsyncdir, fd, datasync, xdata);
+    return 0;
+}
 
-        START_FOP_LATENCY (frame);
+int
+io_stats_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+                dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_setxattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->setxattr,
-                    loc, dict, flags);
-        return 0;
+    STACK_WIND(frame, io_stats_access_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->access, loc, mask, xdata);
+    return 0;
 }
 
-
 int
-io_stats_getxattr (call_frame_t *frame, xlator_t *this,
-                   loc_t *loc, const char *name)
+io_stats_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                   dict_t *xdata)
 {
-        BUMP_FOP (GETXATTR);
-
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_getxattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->getxattr,
-                    loc, name);
-        return 0;
+    STACK_WIND(frame, io_stats_ftruncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+    return 0;
 }
 
-
 int
-io_stats_removexattr (call_frame_t *frame, xlator_t *this,
-                      loc_t *loc, const char *name)
+io_stats_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                  struct iatt *stbuf, int32_t valid, dict_t *xdata)
 {
-        BUMP_FOP (REMOVEXATTR);
+    START_FOP_LATENCY(frame);
 
-        START_FOP_LATENCY (frame);
+    STACK_WIND(frame, io_stats_setattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetattr, fd, stbuf, valid, xdata);
+    return 0;
+}
 
-        STACK_WIND (frame, io_stats_removexattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->removexattr,
-                    loc, name);
+int
+io_stats_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
 
-        return 0;
+    STACK_WIND(frame, io_stats_fstat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fstat, fd, xdata);
+    return 0;
 }
 
-
 int
-io_stats_opendir (call_frame_t *frame, xlator_t *this,
-                  loc_t *loc, fd_t *fd)
+io_stats_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+                   off_t offset, size_t len, dict_t *xdata)
 {
-        BUMP_FOP (OPENDIR);
+    START_FOP_LATENCY(frame);
 
-        START_FOP_LATENCY (frame);
+    STACK_WIND(frame, io_stats_fallocate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fallocate, fd, mode, offset, len,
+               xdata);
 
-        STACK_WIND (frame, io_stats_opendir_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->opendir,
-                    loc, fd);
-        return 0;
+    return 0;
 }
 
 int
-io_stats_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-                   off_t offset)
+io_stats_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                 size_t len, dict_t *xdata)
 {
-        BUMP_FOP (READDIRP);
-
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_readdirp_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->readdirp,
-                    fd, size, offset);
+    STACK_WIND(frame, io_stats_discard_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->discard, fd, offset, len, xdata);
 
-        return 0;
+    return 0;
 }
 
-
 int
-io_stats_readdir (call_frame_t *frame, xlator_t *this,
-                  fd_t *fd, size_t size, off_t offset)
+io_stats_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                  off_t len, dict_t *xdata)
 {
-        BUMP_FOP (READDIR);
+    START_FOP_LATENCY(frame);
 
-        START_FOP_LATENCY (frame);
+    STACK_WIND(frame, io_stats_zerofill_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->zerofill, fd, offset, len, xdata);
 
-        STACK_WIND (frame, io_stats_readdir_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->readdir,
-                    fd, size, offset);
-
-        return 0;
+    return 0;
 }
 
+int32_t
+io_stats_ipc(call_frame_t *frame, xlator_t *this, int32_t op, dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
+
+    STACK_WIND(frame, io_stats_ipc_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ipc, op, xdata);
+    return 0;
+}
 
 int
-io_stats_fsyncdir (call_frame_t *frame, xlator_t *this,
-                   fd_t *fd, int32_t datasync)
+io_stats_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+            struct gf_flock *lock, dict_t *xdata)
 {
-        BUMP_FOP (FSYNCDIR);
-
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_fsyncdir_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fsyncdir,
-                    fd, datasync);
-        return 0;
+    STACK_WIND(frame, io_stats_lk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lk, fd, cmd, lock, xdata);
+    return 0;
 }
 
-
 int
-io_stats_access (call_frame_t *frame, xlator_t *this,
-                 loc_t *loc, int32_t mask)
+io_stats_rchecksum(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                   int32_t len, dict_t *xdata)
 {
-        BUMP_FOP (ACCESS);
+    START_FOP_LATENCY(frame);
 
-        START_FOP_LATENCY (frame);
-
-        STACK_WIND (frame, io_stats_access_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->access,
-                    loc, mask);
-        return 0;
+    STACK_WIND(frame, io_stats_rchecksum_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rchecksum, fd, offset, len, xdata);
+    return 0;
 }
 
-
 int
-io_stats_ftruncate (call_frame_t *frame, xlator_t *this,
-                    fd_t *fd, off_t offset)
+io_stats_seek(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+              gf_seek_what_t what, dict_t *xdata)
 {
-        BUMP_FOP (FTRUNCATE);
+    START_FOP_LATENCY(frame);
 
-        START_FOP_LATENCY (frame);
+    STACK_WIND(frame, io_stats_seek_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->seek, fd, offset, what, xdata);
+    return 0;
+}
 
-        STACK_WIND (frame, io_stats_ftruncate_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->ftruncate,
-                    fd, offset);
+int
+io_stats_lease(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               struct gf_lease *lease, dict_t *xdata)
+{
+    START_FOP_LATENCY(frame);
 
-        return 0;
+    STACK_WIND(frame, io_stats_lease_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lease, loc, lease, xdata);
+    return 0;
 }
 
-
 int
-io_stats_fsetattr (call_frame_t *frame, xlator_t *this,
-                   fd_t *fd, struct iatt *stbuf, int32_t valid)
+io_stats_getactivelk(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                     dict_t *xdata)
 {
-        BUMP_FOP (FSETATTR);
+    START_FOP_LATENCY(frame);
 
-        START_FOP_LATENCY (frame);
-
-        STACK_WIND (frame, io_stats_setattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fsetattr,
-                    fd, stbuf, valid);
-        return 0;
+    STACK_WIND(frame, io_stats_getactivelk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->getactivelk, loc, xdata);
+    return 0;
 }
 
-
 int
-io_stats_fstat (call_frame_t *frame, xlator_t *this,
-                fd_t *fd)
+io_stats_setactivelk(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                     lock_migration_info_t *locklist, dict_t *xdata)
 {
-        BUMP_FOP (FSTAT);
-
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_fstat_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fstat,
-                    fd);
-        return 0;
+    STACK_WIND(frame, io_stats_setactivelk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setactivelk, loc, locklist, xdata);
+    return 0;
 }
 
-
 int
-io_stats_lk (call_frame_t *frame, xlator_t *this,
-             fd_t *fd, int32_t cmd, struct gf_flock *lock)
+io_stats_compound(call_frame_t *frame, xlator_t *this, void *args,
+                  dict_t *xdata)
 {
-        BUMP_FOP (LK);
-
-        START_FOP_LATENCY (frame);
+    START_FOP_LATENCY(frame);
 
-        STACK_WIND (frame, io_stats_lk_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->lk,
-                    fd, cmd, lock);
-        return 0;
+    STACK_WIND(frame, io_stats_compound_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->compound, args, xdata);
+    return 0;
 }
 
-
 int
-io_stats_release (xlator_t *this, fd_t *fd)
+io_stats_release(xlator_t *this, fd_t *fd)
 {
-        struct ios_fd  *iosfd = NULL;
+    struct ios_fd *iosfd = NULL;
+    struct ios_conf *conf = NULL;
 
-        BUMP_FOP (RELEASE);
+    BUMP_FOP(RELEASE);
 
-        ios_fd_ctx_get (fd, this, &iosfd);
-        if (iosfd) {
-                io_stats_dump_fd (this, iosfd);
-
-                if (iosfd->filename)
-                        GF_FREE (iosfd->filename);
-                GF_FREE (iosfd);
+    conf = this->private;
+    if (conf) {
+        LOCK(&conf->lock);
+        {
+            conf->cumulative.nr_opens--;
         }
+        UNLOCK(&conf->lock);
+    }
 
-        return 0;
-}
+    ios_fd_ctx_get(fd, this, &iosfd);
+    if (iosfd) {
+        io_stats_dump_fd(this, iosfd);
 
+        GF_FREE(iosfd->filename);
+        GF_FREE(iosfd);
+    }
+
+    return 0;
+}
 
 int
-io_stats_releasedir (xlator_t *this, fd_t *fd)
+io_stats_releasedir(xlator_t *this, fd_t *fd)
 {
-        BUMP_FOP (RELEASEDIR);
+    BUMP_FOP(RELEASEDIR);
 
-        return 0;
+    return 0;
 }
 
-
 int
-io_stats_forget (xlator_t *this, inode_t *inode)
+io_stats_forget(xlator_t *this, inode_t *inode)
 {
-        BUMP_FOP (FORGET);
-
-        return 0;
+    BUMP_FOP(FORGET);
+    ios_stats_cleanup(this, inode);
+    return 0;
 }
 
-int
-reconfigure (xlator_t *this, dict_t *options)
+static int
+ios_init_top_stats(struct ios_conf *conf)
 {
-        struct ios_conf    *conf = NULL;
-        char               *str = NULL;
-        int                 ret = 0;
-        char               *log_str = NULL;
-        glusterfs_ctx_t     *ctx = NULL;
+    int i = 0;
 
-        if (!this || !this->private)
-                return -1;
+    GF_ASSERT(conf);
 
-        conf = this->private;
+    for (i = 0; i < IOS_STATS_TYPE_MAX; i++) {
+        conf->list[i].iosstats = GF_CALLOC(1, sizeof(*conf->list[i].iosstats),
+                                           gf_io_stats_mt_ios_stat);
 
-        ret = dict_get_str (options, "dump-fd-stats", &str);
-        if (ret == 0) {
-                ret = gf_string2boolean (str, &conf->dump_fd_stats);
-                if (ret == -1) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "'dump-fd-stats' takes only boolean arguments");
-                        return -1;
-                }
+        if (!conf->list[i].iosstats)
+            return -1;
 
-                if (conf->dump_fd_stats) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"enabling dump-fd-stats");
-                }
-        }
+        INIT_LIST_HEAD(&conf->list[i].iosstats->list);
+        LOCK_INIT(&conf->list[i].lock);
+    }
 
-        ret = dict_get_str_boolean (options, "latency-measurement", 0);
-        if (ret != -1) {
-                if (conf->measure_latency != ret) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "changing latency measurement from %d to %d",
-                                conf->measure_latency, ret);
-                }
-                conf->measure_latency = ret;
-        }
-        ctx = glusterfs_ctx_get ();
-        if (!ctx)
-                return -1;
+    for (i = 0; i < IOS_STATS_THRU_MAX; i++) {
+        conf->thru_list[i].iosstats = GF_CALLOC(
+            1, sizeof(*conf->thru_list[i].iosstats), gf_io_stats_mt_ios_stat);
 
-        if (ctx->cmd_args.brick_name)
-                ret = dict_get_str (options, "log-level", &log_str);
-        else
-                ret = dict_get_str (options, "client-log-level", &log_str);
+        if (!conf->thru_list[i].iosstats)
+            return -1;
 
-        if (!ret) {
-                if (!is_gf_log_command(this, "trusted.glusterfs*set-log-level", log_str)) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                               "changing log-level to %s", log_str);
-                }
+        INIT_LIST_HEAD(&conf->thru_list[i].iosstats->list);
+        LOCK_INIT(&conf->thru_list[i].lock);
+    }
+
+    return 0;
+}
+
+static void
+ios_destroy_top_stats(struct ios_conf *conf)
+{
+    int i = 0;
+    struct ios_stat_head *list_head = NULL;
+    struct ios_stat_list *entry = NULL;
+    struct ios_stat_list *tmp = NULL;
+    struct ios_stat_list *list = NULL;
+    struct ios_stat *stat = NULL;
+
+    GF_ASSERT(conf);
+
+    LOCK(&conf->lock);
+
+    conf->cumulative.nr_opens = 0;
+    conf->cumulative.max_nr_opens = 0;
+    conf->cumulative.max_openfd_time.tv_sec = 0;
+    conf->cumulative.max_openfd_time.tv_usec = 0;
+
+    for (i = 0; i < IOS_STATS_TYPE_MAX; i++) {
+        list_head = &conf->list[i];
+        if (!list_head)
+            continue;
+        list_for_each_entry_safe(entry, tmp, &list_head->iosstats->list, list)
+        {
+            list = entry;
+            stat = list->iosstat;
+            ios_stat_unref(stat);
+            list_del(&list->list);
+            GF_FREE(list);
+            list_head->members--;
         }
-        return 0;
+        GF_FREE(list_head->iosstats);
+    }
+
+    for (i = 0; i < IOS_STATS_THRU_MAX; i++) {
+        list_head = &conf->thru_list[i];
+        if (!list_head)
+            continue;
+        list_for_each_entry_safe(entry, tmp, &list_head->iosstats->list, list)
+        {
+            list = entry;
+            stat = list->iosstat;
+            ios_stat_unref(stat);
+            list_del(&list->list);
+            GF_FREE(list);
+            list_head->members--;
+        }
+        GF_FREE(list_head->iosstats);
+    }
+
+    UNLOCK(&conf->lock);
+
+    return;
+}
+
+static void
+io_stats_clear(struct ios_conf *conf)
+{
+    time_t now = 0;
+
+    GF_ASSERT(conf);
+    now = gf_time();
+
+    LOCK(&conf->lock);
+    {
+        ios_global_stats_clear(&conf->cumulative, now);
+        ios_global_stats_clear(&conf->incremental, now);
+        conf->increment = 0;
+    }
+    UNLOCK(&conf->lock);
 }
 
 int32_t
-mem_acct_init (xlator_t *this)
+io_priv(xlator_t *this)
 {
-        int     ret = -1;
+    int i;
+    char key[GF_DUMP_MAX_BUF_LEN];
+    char key_prefix_cumulative[GF_DUMP_MAX_BUF_LEN];
+    char key_prefix_incremental[GF_DUMP_MAX_BUF_LEN];
+    double min, max, avg;
+    uint64_t count, total;
+    struct ios_conf *conf = NULL;
+
+    conf = this->private;
+    if (!conf)
+        return -1;
+
+    if (!conf->count_fop_hits || !conf->measure_latency)
+        return -1;
+
+    gf_proc_dump_write("cumulative.data_read", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(conf->cumulative.data_read));
+    gf_proc_dump_write("cumulative.data_written", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(conf->cumulative.data_written));
+
+    gf_proc_dump_write("incremental.data_read", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(conf->incremental.data_read));
+    gf_proc_dump_write("incremental.data_written", "%" GF_PRI_ATOMIC,
+                       GF_ATOMIC_GET(conf->incremental.data_written));
+
+    snprintf(key_prefix_cumulative, GF_DUMP_MAX_BUF_LEN, "%s.cumulative",
+             this->name);
+    snprintf(key_prefix_incremental, GF_DUMP_MAX_BUF_LEN, "%s.incremental",
+             this->name);
+
+    for (i = 0; i < GF_FOP_MAXVALUE; i++) {
+        count = GF_ATOMIC_GET(conf->cumulative.fop_hits[i]);
+        total = conf->cumulative.latency[i].total;
+        min = conf->cumulative.latency[i].min;
+        max = conf->cumulative.latency[i].max;
+        avg = conf->cumulative.latency[i].avg;
+
+        gf_proc_dump_build_key(key, key_prefix_cumulative, "%s",
+                               (char *)gf_fop_list[i]);
+
+        gf_proc_dump_write(key, "%" PRId64 ",%" PRId64 ",%.03f,%.03f,%.03f",
+                           count, total, min, max, avg);
+
+        count = GF_ATOMIC_GET(conf->incremental.fop_hits[i]);
+        total = conf->incremental.latency[i].total;
+        min = conf->incremental.latency[i].min;
+        max = conf->incremental.latency[i].max;
+        avg = conf->incremental.latency[i].avg;
+
+        gf_proc_dump_build_key(key, key_prefix_incremental, "%s",
+                               (char *)gf_fop_list[i]);
+
+        gf_proc_dump_write(key, "%" PRId64 ",%" PRId64 ",%.03f,%.03f,%.03f",
+                           count, total, min, max, avg);
+    }
+
+    return 0;
+}
 
-        if (!this)
-                return ret;
+static void
+ios_set_log_format_code(struct ios_conf *conf, char *dump_format_str)
+{
+    if (strcmp(dump_format_str, "json") == 0)
+        conf->dump_format = IOS_DUMP_TYPE_JSON_FILE;
+    else if (strcmp(dump_format_str, "text") == 0)
+        conf->dump_format = IOS_DUMP_TYPE_FILE;
+    else if (strcmp(dump_format_str, "dict") == 0)
+        conf->dump_format = IOS_DUMP_TYPE_DICT;
+    else if (strcmp(dump_format_str, "samples") == 0)
+        conf->dump_format = IOS_DUMP_TYPE_SAMPLES;
+}
 
-        ret = xlator_mem_acct_init (this, gf_io_stats_mt_end + 1);
-        
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
-                        " failed");
-                return ret;
-        }
+void
+xlator_set_loglevel(xlator_t *this, int log_level)
+{
+    glusterfs_ctx_t *ctx = NULL;
+    glusterfs_graph_t *active = NULL;
+    xlator_t *top = NULL;
+    xlator_t *trav = this;
 
-        return ret;
+    ctx = this->ctx;
+    GF_ASSERT(ctx);
+    active = ctx->active;
+    top = active->first;
+
+    if (log_level == -1)
+        return;
+
+    if (ctx->cmd_args.brick_mux) {
+        /* Set log-level for all brick xlators */
+        top->loglevel = log_level;
+
+        /* Set log-level for parent xlator */
+        if (this->parents)
+            this->parents->xlator->loglevel = log_level;
+
+        while (trav) {
+            trav->loglevel = log_level;
+            trav = trav->next;
+        }
+    } else {
+        gf_log_set_loglevel(this->ctx, log_level);
+    }
 }
 
 int
-init (xlator_t *this)
+reconfigure(xlator_t *this, dict_t *options)
 {
-        dict_t             *options = NULL;
-        struct ios_conf    *conf = NULL;
-        char               *str = NULL;
-        int                 ret = 0;
+    struct ios_conf *conf = NULL;
+    int ret = -1;
+    char *sys_log_str = NULL;
+    char *log_format_str = NULL;
+    char *logger_str = NULL;
+    char *dump_format_str = NULL;
+    int sys_log_level = -1;
+    char *log_str = NULL;
+    int log_level = -1;
+    int log_format = -1;
+    int logger = -1;
+    uint32_t log_buf_size = 0;
+    uint32_t log_flush_timeout = 0;
+    int32_t old_dump_interval;
+    int32_t threads;
+
+    if (!this || !this->private)
+        goto out;
+
+    conf = this->private;
+
+    GF_OPTION_RECONF("dump-fd-stats", conf->dump_fd_stats, options, bool, out);
+
+    GF_OPTION_RECONF("count-fop-hits", conf->count_fop_hits, options, bool,
+                     out);
+
+    GF_OPTION_RECONF("latency-measurement", conf->measure_latency, options,
+                     bool, out);
+
+    old_dump_interval = conf->ios_dump_interval;
+    GF_OPTION_RECONF("ios-dump-interval", conf->ios_dump_interval, options,
+                     int32, out);
+    if ((old_dump_interval <= 0) && (conf->ios_dump_interval > 0)) {
+        conf->dump_thread_running = _gf_true;
+        conf->dump_thread_should_die = _gf_false;
+        ret = gf_thread_create(&conf->dump_thread, NULL,
+                               (void *)&_ios_dump_thread, this, "iosdump");
+        if (ret) {
+            conf->dump_thread_running = _gf_false;
+            gf_log(this ? this->name : "io-stats", GF_LOG_ERROR,
+                   "Failed to start thread"
+                   "while reconfigure. Returning %d",
+                   ret);
+            goto out;
+        }
+    } else if ((old_dump_interval > 0) && (conf->ios_dump_interval == 0)) {
+        _ios_destroy_dump_thread(conf);
+    }
+
+    GF_OPTION_RECONF("ios-sample-interval", conf->ios_sample_interval, options,
+                     int32, out);
+    GF_OPTION_RECONF("ios-dump-format", dump_format_str, options, str, out);
+    ios_set_log_format_code(conf, dump_format_str);
+    GF_OPTION_RECONF("ios-sample-buf-size", conf->ios_sample_buf_size, options,
+                     int32, out);
+    GF_OPTION_RECONF("sys-log-level", sys_log_str, options, str, out);
+    if (sys_log_str) {
+        sys_log_level = glusterd_check_log_level(sys_log_str);
+        set_sys_log_level(sys_log_level);
+    }
+
+    GF_OPTION_RECONF("log-level", log_str, options, str, out);
+    if (log_str) {
+        log_level = glusterd_check_log_level(log_str);
+        /* Set loglevel for all children and server xlators */
+        xlator_set_loglevel(this, log_level);
+    }
+
+    GF_OPTION_RECONF("logger", logger_str, options, str, out);
+    if (logger_str) {
+        logger = gf_check_logger(logger_str);
+        gf_log_set_logger(logger);
+    }
+
+    GF_OPTION_RECONF("log-format", log_format_str, options, str, out);
+    if (log_format_str) {
+        log_format = gf_check_log_format(log_format_str);
+        gf_log_set_logformat(log_format);
+    }
+
+    GF_OPTION_RECONF("log-buf-size", log_buf_size, options, uint32, out);
+    gf_log_set_log_buf_size(log_buf_size);
+
+    GF_OPTION_RECONF("log-flush-timeout", log_flush_timeout, options, time,
+                     out);
+    gf_log_set_log_flush_timeout(log_flush_timeout);
+
+    GF_OPTION_RECONF("threads", threads, options, int32, out);
+    gf_async_adjust_threads(threads);
+
+    ret = 0;
+out:
+    gf_log(this ? this->name : "io-stats", GF_LOG_DEBUG,
+           "reconfigure returning %d", ret);
+    return ret;
+}
 
-        if (!this)
-                return -1;
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
 
-        if (!this->children || this->children->next) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "io_stats translator requires one subvolume");
-                return -1;
-        }
+    if (!this)
+        return ret;
 
-        if (!this->parents) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "dangling volume. check volfile ");
-        }
+    ret = xlator_mem_acct_init(this, gf_io_stats_mt_end + 1);
 
-        options = this->options;
+    if (ret != 0) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Memory accounting init"
+               " failed");
+        return ret;
+    }
 
-        conf = GF_CALLOC (1, sizeof(*conf), gf_io_stats_mt_ios_conf);
+    return ret;
+}
 
-        if (!conf) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory.");
-                return -1;
-        }
+void
+ios_conf_destroy(struct ios_conf *conf)
+{
+    if (!conf)
+        return;
 
-        LOCK_INIT (&conf->lock);
+    ios_destroy_top_stats(conf);
+    _ios_destroy_dump_thread(conf);
+    ios_destroy_sample_buf(conf->ios_sample_buf);
+    LOCK_DESTROY(&conf->lock);
+    gf_dnscache_deinit(conf->dnscache);
+    GF_FREE(conf);
+}
 
-        gettimeofday (&conf->cumulative.started_at, NULL);
-        gettimeofday (&conf->incremental.started_at, NULL);
+static void
+ios_init_stats(struct ios_global_stats *stats)
+{
+    int i = 0;
 
-        ret = dict_get_str (options, "dump-fd-stats", &str);
-        if (ret == 0) {
-                ret = gf_string2boolean (str, &conf->dump_fd_stats);
-                if (ret == -1) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "'dump-fd-stats' takes only boolean arguments");
-                        return -1;
-                }
+    GF_ATOMIC_INIT(stats->data_read, 0);
+    GF_ATOMIC_INIT(stats->data_written, 0);
 
-                if (conf->dump_fd_stats) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"enabling dump-fd-stats");
-                }
-        }
+    for (i = 0; i < IOS_BLOCK_COUNT_SIZE; i++) {
+        GF_ATOMIC_INIT(stats->block_count_write[i], 0);
+        GF_ATOMIC_INIT(stats->block_count_read[i], 0);
+    }
 
-        ret = dict_get_str_boolean (options, "latency-measurement", 0);
-        if (ret != -1) {
-                conf->measure_latency = ret;
-                if (conf->measure_latency)
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "enabling latency measurement");
-        }
+    for (i = 0; i < GF_FOP_MAXVALUE; i++)
+        GF_ATOMIC_INIT(stats->fop_hits[i], 0);
 
-        this->private = conf;
+    for (i = 0; i < GF_UPCALL_FLAGS_MAXVALUE; i++)
+        GF_ATOMIC_INIT(stats->upcall_hits[i], 0);
 
-        return 0;
+    stats->started_at = gf_time();
 }
 
+int
+init(xlator_t *this)
+{
+    struct ios_conf *conf = NULL;
+    char *volume_id = NULL;
+    char *sys_log_str = NULL;
+    char *logger_str = NULL;
+    char *log_format_str = NULL;
+    char *dump_format_str = NULL;
+    int logger = -1;
+    int log_format = -1;
+    int sys_log_level = -1;
+    char *log_str = NULL;
+    int log_level = -1;
+    int ret = -1;
+    uint32_t log_buf_size = 0;
+    uint32_t log_flush_timeout = 0;
+    int32_t threads;
+
+    if (!this)
+        return -1;
+
+    if (!this->children) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "io_stats translator requires at least one subvolume");
+        return -1;
+    }
+
+    if (!this->parents) {
+        /* This is very much valid as io-stats currently is loaded
+         * on top of volumes on both client and server, hence this is
+         * not an warning message */
+        gf_log(this->name, GF_LOG_DEBUG, "dangling volume. check volfile ");
+    }
+
+    conf = GF_CALLOC(1, sizeof(*conf), gf_io_stats_mt_ios_conf);
+
+    if (!conf)
+        goto out;
+
+    if (dict_get_str(this->options, "unique-id", &conf->unique_id) != 0) {
+        /* This is always set on servers, so we must be a client. */
+        conf->unique_id = this->name;
+    }
+
+    ret = dict_get_strn(this->options, "volume-id", SLEN("volume-id"),
+                        &volume_id);
+    if (!ret) {
+        strncpy(this->graph->volume_id, volume_id, GF_UUID_BUF_SIZE);
+    }
+    /*
+     * Init it just after calloc, so that we are sure the lock is inited
+     * in case of error paths.
+     */
+    LOCK_INIT(&conf->lock);
+    LOCK_INIT(&conf->ios_sampling_lock);
+
+    ios_init_stats(&conf->cumulative);
+    ios_init_stats(&conf->incremental);
+
+    ret = ios_init_top_stats(conf);
+    if (ret)
+        goto out;
+
+    GF_OPTION_INIT("dump-fd-stats", conf->dump_fd_stats, bool, out);
+
+    GF_OPTION_INIT("count-fop-hits", conf->count_fop_hits, bool, out);
+
+    GF_OPTION_INIT("latency-measurement", conf->measure_latency, bool, out);
+
+    GF_OPTION_INIT("ios-dump-interval", conf->ios_dump_interval, int32, out);
+
+    GF_OPTION_INIT("ios-sample-interval", conf->ios_sample_interval, int32,
+                   out);
+
+    GF_OPTION_INIT("ios-dump-format", dump_format_str, str, out);
+    ios_set_log_format_code(conf, dump_format_str);
+
+    GF_OPTION_INIT("ios-sample-buf-size", conf->ios_sample_buf_size, int32,
+                   out);
+
+    ret = ios_init_sample_buf(conf);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR, "Out of memory.");
+        goto out;
+    }
+
+    GF_OPTION_INIT("ios-dnscache-ttl-sec", conf->ios_dnscache_ttl_sec, int32,
+                   out);
+    conf->dnscache = gf_dnscache_init(conf->ios_dnscache_ttl_sec);
+    if (!conf->dnscache) {
+        ret = -1;
+        goto out;
+    }
+
+    GF_OPTION_INIT("sys-log-level", sys_log_str, str, out);
+    if (sys_log_str) {
+        sys_log_level = glusterd_check_log_level(sys_log_str);
+        set_sys_log_level(sys_log_level);
+    }
+
+    GF_OPTION_INIT("log-level", log_str, str, out);
+    if (log_str) {
+        log_level = glusterd_check_log_level(log_str);
+        if (DEFAULT_LOG_LEVEL != log_level)
+            gf_log_set_loglevel(this->ctx, log_level);
+    }
+
+    GF_OPTION_INIT("logger", logger_str, str, out);
+    if (logger_str) {
+        logger = gf_check_logger(logger_str);
+        gf_log_set_logger(logger);
+    }
+
+    GF_OPTION_INIT("log-format", log_format_str, str, out);
+    if (log_format_str) {
+        log_format = gf_check_log_format(log_format_str);
+        gf_log_set_logformat(log_format);
+    }
+
+    GF_OPTION_INIT("log-buf-size", log_buf_size, uint32, out);
+    gf_log_set_log_buf_size(log_buf_size);
+
+    GF_OPTION_INIT("log-flush-timeout", log_flush_timeout, time, out);
+    gf_log_set_log_flush_timeout(log_flush_timeout);
+
+    GF_OPTION_INIT("threads", threads, int32, out);
+    gf_async_adjust_threads(threads);
+
+    this->private = conf;
+    if (conf->ios_dump_interval > 0) {
+        conf->dump_thread_running = _gf_true;
+        conf->dump_thread_should_die = _gf_false;
+        ret = gf_thread_create(&conf->dump_thread, NULL,
+                               (void *)&_ios_dump_thread, this, "iosdump");
+        if (ret) {
+            conf->dump_thread_running = _gf_false;
+            gf_log(this ? this->name : "io-stats", GF_LOG_ERROR,
+                   "Failed to start thread"
+                   "in init. Returning %d",
+                   ret);
+            goto out;
+        }
+    }
+    return 0;
+out:
+    ios_conf_destroy(conf);
+    return ret;
+}
 
 void
-fini (xlator_t *this)
+fini(xlator_t *this)
 {
-        struct ios_conf *conf = NULL;
+    struct ios_conf *conf = NULL;
 
-        if (!this)
-                return;
+    if (!this)
+        return;
 
-        conf = this->private;
+    conf = this->private;
 
-        GF_FREE(conf);
+    ios_conf_destroy(conf);
+    this->private = NULL;
+    gf_log(this->name, GF_LOG_INFO, "io-stats translator unloaded");
+    return;
+}
 
-        gf_log (this->name, GF_LOG_NORMAL,
-                "io-stats translator unloaded");
-        return;
+int
+notify(xlator_t *this, int32_t event, void *data, ...)
+{
+    int ret = 0;
+    struct ios_dump_args args = {0};
+    dict_t *output = NULL;
+    dict_t *dict = NULL;
+    int32_t op = 0;
+    int32_t list_cnt = 0;
+    double throughput = 0;
+    double time = 0;
+    gf_boolean_t is_peek = _gf_false;
+    va_list ap;
+    struct gf_upcall *up_data = NULL;
+    struct gf_upcall_cache_invalidation *up_ci = NULL;
+
+    dict = data;
+    va_start(ap, data);
+    output = va_arg(ap, dict_t *);
+    va_end(ap);
+    switch (event) {
+        case GF_EVENT_TRANSLATOR_INFO:
+            ret = dict_get_str_boolean(dict, "clear-stats", _gf_false);
+            if (ret) {
+                ret = dict_set_int32(output, "top-op", op);
+                if (ret) {
+                    gf_log(this->name, GF_LOG_ERROR,
+                           "Failed to set top-op in dict");
+                    goto out;
+                }
+                ios_destroy_top_stats(this->private);
+                ret = ios_init_top_stats(this->private);
+                if (ret)
+                    gf_log(this->name, GF_LOG_ERROR,
+                           "Failed to reset top stats");
+                ret = dict_set_int32(output, "stats-cleared", ret ? 0 : 1);
+                if (ret)
+                    gf_log(this->name, GF_LOG_ERROR,
+                           "Failed to set stats-cleared"
+                           " in dict");
+                goto out;
+            }
+
+            ret = dict_get_int32(dict, "top-op", &op);
+            if (!ret) {
+                ret = dict_get_int32(dict, "list-cnt", &list_cnt);
+                if (op > IOS_STATS_TYPE_NONE && op < IOS_STATS_TYPE_MAX)
+                    ret = io_stats_dump_stats_to_dict(this, output, op,
+                                                      list_cnt);
+                if (op == IOS_STATS_TYPE_READ_THROUGHPUT ||
+                    op == IOS_STATS_TYPE_WRITE_THROUGHPUT) {
+                    ret = dict_get_double(dict, "throughput", &throughput);
+                    if (!ret) {
+                        ret = dict_get_double(dict, "time", &time);
+                        if (ret)
+                            goto out;
+                        ret = dict_set_double(output, "throughput", throughput);
+                        if (ret)
+                            goto out;
+                        ret = dict_set_double(output, "time", time);
+                        if (ret)
+                            goto out;
+                    }
+                    ret = 0;
+                }
+            } else {
+                ret = dict_get_int32(dict, "info-op", &op);
+                if (ret || op < GF_IOS_INFO_ALL || GF_IOS_INFO_CLEAR < op)
+                    op = GF_IOS_INFO_ALL;
+
+                ret = dict_set_int32(output, "info-op", op);
+                if (ret) {
+                    gf_log(this->name, GF_LOG_ERROR,
+                           "Failed to set info-op in dict");
+                    goto out;
+                }
+
+                if (GF_IOS_INFO_CLEAR == op) {
+                    io_stats_clear(this->private);
+
+                    ret = dict_set_int32(output, "stats-cleared", 1);
+                    if (ret)
+                        gf_log(this->name, GF_LOG_ERROR,
+                               "Failed to set stats-cleared"
+                               " in dict");
+                } else {
+                    ret = dict_get_str_boolean(dict, "peek", _gf_false);
+                    if (-1 != ret)
+                        is_peek = ret;
+
+                    (void)ios_dump_args_init(&args, IOS_DUMP_TYPE_DICT, output);
+                    ret = io_stats_dump(this, &args, op, is_peek);
+                }
+            }
+            break;
+        case GF_EVENT_UPCALL:
+            up_data = (struct gf_upcall *)data;
+            ios_bump_upcall(this, GF_UPCALL);
+
+            switch (up_data->event_type) {
+                case GF_UPCALL_RECALL_LEASE:
+                    ios_bump_upcall(this, GF_UPCALL_LEASE_RECALL);
+                    break;
+                case GF_UPCALL_CACHE_INVALIDATION:
+                    up_ci = (struct gf_upcall_cache_invalidation *)
+                                up_data->data;
+                    if (up_ci->flags & (UP_XATTR | UP_XATTR_RM))
+                        ios_bump_upcall(this, GF_UPCALL_CI_XATTR);
+                    if (up_ci->flags & IATT_UPDATE_FLAGS)
+                        ios_bump_upcall(this, GF_UPCALL_CI_STAT);
+                    if (up_ci->flags & UP_RENAME_FLAGS)
+                        ios_bump_upcall(this, GF_UPCALL_CI_RENAME);
+                    if (up_ci->flags & UP_FORGET)
+                        ios_bump_upcall(this, GF_UPCALL_CI_FORGET);
+                    if (up_ci->flags & UP_NLINK)
+                        ios_bump_upcall(this, GF_UPCALL_CI_NLINK);
+                    break;
+                default:
+                    gf_msg_debug(this->name, 0,
+                                 "Unknown upcall event "
+                                 "type :%d",
+                                 up_data->event_type);
+                    break;
+            }
+
+            default_notify(this, event, data);
+            break;
+        default:
+            default_notify(this, event, data);
+            break;
+    }
+out:
+    return ret;
 }
 
+struct xlator_dumpops dumpops = {.priv = io_priv};
 
 struct xlator_fops fops = {
-        .stat        = io_stats_stat,
-        .readlink    = io_stats_readlink,
-        .mknod       = io_stats_mknod,
-        .mkdir       = io_stats_mkdir,
-        .unlink      = io_stats_unlink,
-        .rmdir       = io_stats_rmdir,
-        .symlink     = io_stats_symlink,
-        .rename      = io_stats_rename,
-        .link        = io_stats_link,
-        .truncate    = io_stats_truncate,
-        .open        = io_stats_open,
-        .readv       = io_stats_readv,
-        .writev      = io_stats_writev,
-        .statfs      = io_stats_statfs,
-        .flush       = io_stats_flush,
-        .fsync       = io_stats_fsync,
-        .setxattr    = io_stats_setxattr,
-        .getxattr    = io_stats_getxattr,
-        .removexattr = io_stats_removexattr,
-        .opendir     = io_stats_opendir,
-        .readdir     = io_stats_readdir,
-        .readdirp    = io_stats_readdirp,
-        .fsyncdir    = io_stats_fsyncdir,
-        .access      = io_stats_access,
-        .ftruncate   = io_stats_ftruncate,
-        .fstat       = io_stats_fstat,
-        .create      = io_stats_create,
-        .lk          = io_stats_lk,
-        .inodelk     = io_stats_inodelk,
-        .finodelk    = io_stats_finodelk,
-        .entrylk     = io_stats_entrylk,
-        .lookup      = io_stats_lookup,
-        .xattrop     = io_stats_xattrop,
-        .fxattrop    = io_stats_fxattrop,
-        .setattr     = io_stats_setattr,
-        .fsetattr    = io_stats_fsetattr,
+    .stat = io_stats_stat,
+    .readlink = io_stats_readlink,
+    .mknod = io_stats_mknod,
+    .mkdir = io_stats_mkdir,
+    .unlink = io_stats_unlink,
+    .rmdir = io_stats_rmdir,
+    .symlink = io_stats_symlink,
+    .rename = io_stats_rename,
+    .link = io_stats_link,
+    .truncate = io_stats_truncate,
+    .open = io_stats_open,
+    .readv = io_stats_readv,
+    .writev = io_stats_writev,
+    .statfs = io_stats_statfs,
+    .flush = io_stats_flush,
+    .fsync = io_stats_fsync,
+    .setxattr = io_stats_setxattr,
+    .getxattr = io_stats_getxattr,
+    .removexattr = io_stats_removexattr,
+    .fsetxattr = io_stats_fsetxattr,
+    .fgetxattr = io_stats_fgetxattr,
+    .fremovexattr = io_stats_fremovexattr,
+    .opendir = io_stats_opendir,
+    .readdir = io_stats_readdir,
+    .readdirp = io_stats_readdirp,
+    .fsyncdir = io_stats_fsyncdir,
+    .access = io_stats_access,
+    .ftruncate = io_stats_ftruncate,
+    .fstat = io_stats_fstat,
+    .create = io_stats_create,
+    .lk = io_stats_lk,
+    .inodelk = io_stats_inodelk,
+    .finodelk = io_stats_finodelk,
+    .entrylk = io_stats_entrylk,
+    .fentrylk = io_stats_fentrylk,
+    .lookup = io_stats_lookup,
+    .xattrop = io_stats_xattrop,
+    .fxattrop = io_stats_fxattrop,
+    .setattr = io_stats_setattr,
+    .fsetattr = io_stats_fsetattr,
+    .fallocate = io_stats_fallocate,
+    .discard = io_stats_discard,
+    .zerofill = io_stats_zerofill,
+    .ipc = io_stats_ipc,
+    .rchecksum = io_stats_rchecksum,
+    .seek = io_stats_seek,
+    .lease = io_stats_lease,
+    .getactivelk = io_stats_getactivelk,
+    .setactivelk = io_stats_setactivelk,
+    .compound = io_stats_compound,
+    .copy_file_range = io_stats_copy_file_range,
 };
 
 struct xlator_cbks cbks = {
-        .release     = io_stats_release,
-        .releasedir  = io_stats_releasedir,
-        .forget      = io_stats_forget,
+    .release = io_stats_release,
+    .releasedir = io_stats_releasedir,
+    .forget = io_stats_forget,
 };
 
 struct volume_options options[] = {
-        { .key  = {"dump-fd-stats"},
-          .type = GF_OPTION_TYPE_BOOL,
-        },
-        { .key  = { "latency-measurement" },
-          .type = GF_OPTION_TYPE_BOOL,
-        },
-        { .key = {"log-level"},
-          .type = GF_OPTION_TYPE_STR,
-        },
-        { .key = {"client-log-level"},
-          .type = GF_OPTION_TYPE_STR,
-        },
-        { .key  = {NULL} },
+    {.key = {"dump-fd-stats"},
+     .op_version = {1},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = "If on stats related to file-operations would be "
+                    "tracked inside GlusterFS data-structures."},
+    {.key = {"ios-dump-interval"},
+     .type = GF_OPTION_TYPE_INT,
+     .op_version = {1},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .min = 0,
+     .max = 3600,
+     .default_value = "0",
+     .description = "Interval (in seconds) at which to auto-dump "
+                    "statistics. Zero disables automatic dumping."},
+    {.key = {"ios-sample-interval"},
+     .type = GF_OPTION_TYPE_INT,
+     .op_version = {1},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .min = 0,
+     .max = 65535,
+     .default_value = "0",
+     .description = "Interval in which we want to collect FOP latency "
+                    "samples.  2 means collect a sample every 2nd FOP."},
+    {.key = {"ios-dump-format"},
+     .type = GF_OPTION_TYPE_STR,
+     .op_version = {GD_OP_VERSION_3_12_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .default_value = "json",
+     .description = " The dump-format option specifies the format in which"
+                    " to dump the statistics. Select between \"text\", "
+                    "\"json\", \"dict\" and \"samples\". Default is "
+                    "\"json\".",
+     .value = {"text", "json", "dict", "samples"}},
+    {.key = {"ios-sample-buf-size"},
+     .type = GF_OPTION_TYPE_INT,
+     .op_version = {1},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .min = 1024,
+     .max = 1024 * 1024,
+     .default_value = "65535",
+     .description = "The maximum size of our FOP sampling ring buffer."},
+    {.key = {"ios-dnscache-ttl-sec"},
+     .type = GF_OPTION_TYPE_INT,
+     .op_version = {1},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .min = 1,
+     .max = 3600 * 72,
+     .default_value = "86400",
+     .description = "The interval after wish a cached DNS entry will be "
+                    "re-validated.  Default: 24 hrs"},
+    {.key = {"latency-measurement"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .op_version = {1},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .default_value = "off",
+     .description = "If on stats related to the latency of each operation "
+                    "would be tracked inside GlusterFS data-structures. "},
+    {
+        .key = {"count-fop-hits"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .op_version = {1},
+        .flags = OPT_FLAG_SETTABLE,
+        .tags = {"io-stats"},
+    },
+    {.key = {"log-level"},
+     .type = GF_OPTION_TYPE_STR,
+     .value = {"DEBUG", "WARNING", "ERROR", "INFO", "CRITICAL", "NONE",
+               "TRACE"}},
+
+    /* These are synthetic entries to assist validation of CLI's  *
+     *  volume set  command                                       */
+    {.key = {"client-log-level"},
+     .type = GF_OPTION_TYPE_STR,
+     .op_version = {1},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .default_value = "INFO",
+     .description = "Changes the log-level of the clients",
+     .value = {"DEBUG", "WARNING", "ERROR", "INFO", "CRITICAL", "NONE",
+               "TRACE"}},
+    {.key = {"sys-log-level"},
+     .type = GF_OPTION_TYPE_STR,
+     .default_value = "CRITICAL",
+     .description = "Gluster's syslog log-level",
+     .value = {"WARNING", "ERROR", "INFO", "CRITICAL"}},
+    {.key = {"brick-log-level"},
+     .type = GF_OPTION_TYPE_STR,
+     .op_version = {1},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .default_value = "INFO",
+     .description = "Changes the log-level of the bricks",
+     .value = {"DEBUG", "WARNING", "ERROR", "INFO", "CRITICAL", "NONE",
+               "TRACE"}},
+    {.key = {"logger"},
+     .type = GF_OPTION_TYPE_STR,
+     .value = {GF_LOGGER_GLUSTER_LOG, GF_LOGGER_SYSLOG}},
+    {.key = {"client-logger"},
+     .type = GF_OPTION_TYPE_STR,
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .default_value = GF_LOGGER_GLUSTER_LOG,
+     .description = "Changes the logging sub-system to log to, for the "
+                    "clients",
+     .value = {GF_LOGGER_GLUSTER_LOG, GF_LOGGER_SYSLOG}},
+    {.key = {"brick-logger"},
+     .type = GF_OPTION_TYPE_STR,
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .default_value = GF_LOGGER_GLUSTER_LOG,
+     .description = "Changes the logging sub-system to log to, for the "
+                    "bricks",
+     .value = {GF_LOGGER_GLUSTER_LOG, GF_LOGGER_SYSLOG}},
+    {.key = {"log-format"},
+     .type = GF_OPTION_TYPE_STR,
+     .value = {GF_LOG_FORMAT_NO_MSG_ID, GF_LOG_FORMAT_WITH_MSG_ID}},
+    {.key = {"client-log-format"},
+     .type = GF_OPTION_TYPE_STR,
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .default_value = GF_LOG_FORMAT_WITH_MSG_ID,
+     .description = "Changes log format for the clients",
+     .value = {GF_LOG_FORMAT_NO_MSG_ID, GF_LOG_FORMAT_WITH_MSG_ID}},
+    {.key = {"brick-log-format"},
+     .type = GF_OPTION_TYPE_STR,
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .default_value = GF_LOG_FORMAT_WITH_MSG_ID,
+     .description = "Changes the log format for the bricks",
+     .value = {GF_LOG_FORMAT_NO_MSG_ID, GF_LOG_FORMAT_WITH_MSG_ID}},
+    {
+        .key = {"log-buf-size"},
+        .type = GF_OPTION_TYPE_INT,
+        .min = GF_LOG_LRU_BUFSIZE_MIN,
+        .max = GF_LOG_LRU_BUFSIZE_MAX,
+        .default_value = "5",
+    },
+    {.key = {"client-log-buf-size"},
+     .type = GF_OPTION_TYPE_INT,
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .min = GF_LOG_LRU_BUFSIZE_MIN,
+     .max = GF_LOG_LRU_BUFSIZE_MAX,
+     .default_value = "5",
+     .description = "This option determines the maximum number of unique "
+                    "log messages that can be buffered for a time equal to"
+                    " the value of the option client-log-flush-timeout."},
+    {.key = {"brick-log-buf-size"},
+     .type = GF_OPTION_TYPE_INT,
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .min = GF_LOG_LRU_BUFSIZE_MIN,
+     .max = GF_LOG_LRU_BUFSIZE_MAX,
+     .default_value = "5",
+     .description = "This option determines the maximum number of unique "
+                    "log messages that can be buffered for a time equal to"
+                    " the value of the option brick-log-flush-timeout."},
+    {
+        .key = {"log-flush-timeout"},
+        .type = GF_OPTION_TYPE_TIME,
+        .min = GF_LOG_FLUSH_TIMEOUT_MIN,
+        .max = GF_LOG_FLUSH_TIMEOUT_MAX,
+        .default_value = "120",
+    },
+    {.key = {"client-log-flush-timeout"},
+     .type = GF_OPTION_TYPE_TIME,
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .min = GF_LOG_FLUSH_TIMEOUT_MIN,
+     .max = GF_LOG_FLUSH_TIMEOUT_MAX,
+     .default_value = "120",
+     .description = "This option determines the maximum number of unique "
+                    "log messages that can be buffered for a time equal to"
+                    " the value of the option client-log-flush-timeout."},
+    {.key = {"brick-log-flush-timeout"},
+     .type = GF_OPTION_TYPE_TIME,
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats"},
+     .min = GF_LOG_FLUSH_TIMEOUT_MIN,
+     .max = GF_LOG_FLUSH_TIMEOUT_MAX,
+     .default_value = "120",
+     .description = "This option determines the maximum number of unique "
+                    "log messages that can be buffered for a time equal to"
+                    " the value of the option brick-log-flush-timeout."},
+    {.key = {"unique-id"},
+     .type = GF_OPTION_TYPE_STR,
+     .default_value = "/no/such/path",
+     .description = "Unique ID for our files."},
+    {.key = {"global-threading"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .op_version = {GD_OP_VERSION_6_0},
+     .flags = OPT_FLAG_SETTABLE,
+     .tags = {"io-stats", "threading"},
+     .description = "This option enables the global threading support for "
+                    "bricks. If enabled, it's recommended to also enable "
+                    "'performance.iot-pass-through'"},
+    {.key = {"threads"}, .type = GF_OPTION_TYPE_INT},
+    {.key = {"brick-threads"},
+     .type = GF_OPTION_TYPE_INT,
+     .default_value = "16",
+     .min = 0,
+     .max = GF_ASYNC_MAX_THREADS,
+     .op_version = {GD_OP_VERSION_6_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"io-stats", "threading"},
+     .description = "When global threading is used, this value determines the "
+                    "maximum amount of threads that can be created on bricks"},
+    {.key = {"client-threads"},
+     .type = GF_OPTION_TYPE_INT,
+     .default_value = "16",
+     .min = 0,
+     .max = GF_ASYNC_MAX_THREADS,
+     .op_version = {GD_OP_VERSION_6_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC | OPT_FLAG_CLIENT_OPT,
+     .tags = {"io-stats", "threading"},
+     .description = "When global threading is used, this value determines the "
+                    "maximum amount of threads that can be created on clients"},
+    {.key = {"volume-id"},
+     .type = GF_OPTION_TYPE_STR,
+     .op_version = {GD_OP_VERSION_7_1},
+     .tags = {"global", "volume-id"},
+     .description =
+         "This option points to the 'unique' UUID particular to this "
+         "volume, which would be set in 'graph->volume_id'"},
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .dumpops = &dumpops,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "io-stats",
+    .category = GF_MAINTAINED,
 };
diff --git a/xlators/cluster/stripe/Makefile.am b/xlators/debug/sink/Makefile.am
index d471a3f9243..f2689244371 100644
--- a/xlators/cluster/stripe/Makefile.am
+++ b/xlators/debug/sink/Makefile.am
@@ -1,3 +1,2 @@
 SUBDIRS = src
 
-CLEANFILES = 
diff --git a/xlators/debug/sink/src/Makefile.am b/xlators/debug/sink/src/Makefile.am
new file mode 100644
index 00000000000..f952c2ce6bc
--- /dev/null
+++ b/xlators/debug/sink/src/Makefile.am
@@ -0,0 +1,14 @@
+xlator_LTLIBRARIES = sink.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/debug
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+        -I$(top_builddir)/rpc/xdr/src
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+sink_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+sink_la_SOURCES = sink.c
+sink_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+CLEANFILES =
+
diff --git a/xlators/debug/sink/src/sink.c b/xlators/debug/sink/src/sink.c
new file mode 100644
index 00000000000..9822bbb732e
--- /dev/null
+++ b/xlators/debug/sink/src/sink.c
@@ -0,0 +1,94 @@
+/*
+   Copyright (c) 2017 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+int32_t
+init(xlator_t *this)
+{
+    return 0;
+}
+
+void
+fini(xlator_t *this)
+{
+    return;
+}
+
+/*
+ * notify - when parent sends PARENT_UP, send CHILD_UP event from here
+ */
+int32_t
+notify(xlator_t *this, int32_t event, void *data, ...)
+{
+    switch (event) {
+        case GF_EVENT_PARENT_UP:
+            /* Tell the parent that this xlator is up */
+            default_notify(this, GF_EVENT_CHILD_UP, data);
+            break;
+        case GF_EVENT_PARENT_DOWN:
+            /* Tell the parent that this xlator is down */
+            default_notify(this, GF_EVENT_CHILD_DOWN, data);
+            break;
+        default:
+            break;
+    }
+
+    return 0;
+}
+
+/*
+ * A lookup on "/" is done while mounting or glfs_init() is performed. This
+ * needs to return a valid directory for the root of the mountpoint.
+ *
+ * In case this xlator is used for more advanced debugging, it will need to be
+ * extended to support different LOOKUPs too.
+ */
+static int32_t
+sink_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    struct iatt stbuf = {
+        0,
+    };
+    struct iatt postparent = {
+        0,
+    };
+
+    /* the root of the volume always need to be a directory */
+    stbuf.ia_type = IA_IFDIR;
+
+    STACK_UNWIND_STRICT(lookup, frame, 0, 0, loc ? loc->inode : NULL, &stbuf,
+                        xdata, &postparent);
+
+    return 0;
+}
+
+struct xlator_fops fops = {
+    .lookup = sink_lookup,
+};
+
+struct xlator_cbks cbks = {};
+
+struct volume_options options[] = {
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .op_version = {GD_OP_VERSION_3_12_0},
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "sink",
+    .category = GF_TECH_PREVIEW,
+};
diff --git a/xlators/debug/trace/src/Makefile.am b/xlators/debug/trace/src/Makefile.am
index 0f1679a049d..a37ea63af04 100644
--- a/xlators/debug/trace/src/Makefile.am
+++ b/xlators/debug/trace/src/Makefile.am
@@ -2,13 +2,16 @@
 xlator_LTLIBRARIES = trace.la
 xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/debug
 
-trace_la_LDFLAGS = -module -avoidversion
+trace_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
 
 trace_la_SOURCES = trace.c
 trace_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS)\
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
+noinst_HEADERS = trace.h trace-mem-types.h
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
 
 CLEANFILES = 
 
diff --git a/xlators/debug/trace/src/trace-mem-types.h b/xlators/debug/trace/src/trace-mem-types.h
new file mode 100644
index 00000000000..18a7e0414a6
--- /dev/null
+++ b/xlators/debug/trace/src/trace-mem-types.h
@@ -0,0 +1,20 @@
+/*
+   Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __TRACE_MEM_TYPES_H__
+#define __TRACE_MEM_TYPES_H__
+
+#include <glusterfs/mem-types.h>
+
+enum gf_trace_mem_types_ {
+    gf_trace_mt_trace_conf_t = gf_common_mt_end + 1,
+    gf_trace_mt_end
+};
+#endif
diff --git a/xlators/debug/trace/src/trace.c b/xlators/debug/trace/src/trace.c
index bc7a976dd78..6ed0ca00342 100644
--- a/xlators/debug/trace/src/trace.c
+++ b/xlators/debug/trace/src/trace.c
@@ -1,26 +1,15 @@
 /*
-  Copyright (c) 2006-2009 Gluster, Inc. <http://www.gluster.com>
+  Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
 */
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+#include "trace.h"
+#include "trace-mem-types.h"
 
 /**
  * xlators/debug/trace :
@@ -28,2203 +17,3518 @@
  *    their _cbk functions, which later passes the call to next layer.
  *    Very helpful translator for debugging.
  */
+#define TRACE_STAT_TO_STR(buf, str) trace_stat_to_str(buf, str, sizeof(str))
+
+static void
+trace_stat_to_str(struct iatt *buf, char *str, size_t len)
+{
+    char atime_buf[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    char mtime_buf[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    char ctime_buf[GF_TIMESTR_SIZE] = {
+        0,
+    };
+
+    if (!buf)
+        return;
 
-#include <time.h>
-#include <errno.h>
-#include "glusterfs.h"
-#include "xlator.h"
-#include "common-utils.h"
-
-#define ERR_EINVAL_NORETURN(cond)                               \
-        do                                                      \
-        {                                                       \
-                if ((cond))                                     \
-                {                                               \
-                        gf_log ("ERROR", GF_LOG_ERROR,          \
-                                "%s: %s: (%s) is true",         \
-                                __FILE__, __FUNCTION__, #cond); \
-                }                                               \
-        } while (0)
-
-
-typedef struct trace_private {
-        int32_t debug_flag;
-} trace_private_t;
-
-
-struct {
-        char *name;
-        int enabled;
-} trace_fop_names[GF_FOP_MAXVALUE];
-
-int trace_log_level = GF_LOG_NORMAL;
-
-static char *
-trace_stat_to_str (struct iatt *stbuf)
-{
-        char    *statstr           = NULL;
-        char     atime_buf[256]    = {0,};
-        char     mtime_buf[256]    = {0,};
-        char     ctime_buf[256]    = {0,};
-        int      asprint_ret_value = 0;
-        uint64_t ia_time           = 0;
-
-        ia_time = stbuf->ia_atime;
-        strftime (atime_buf, 256, "[%b %d %H:%M:%S]",
-                  localtime ((time_t *)&ia_time));
-
-        ia_time = stbuf->ia_mtime;
-        strftime (mtime_buf, 256, "[%b %d %H:%M:%S]",
-                  localtime ((time_t *)&ia_time));
-
-        ia_time = stbuf->ia_ctime;
-        strftime (ctime_buf, 256, "[%b %d %H:%M:%S]",
-                  localtime ((time_t *)&ia_time));
-
-        asprint_ret_value = gf_asprintf (&statstr,
-                                      "ia_ino=%"PRIu64
-                                      ", st_mode=%o, ia_nlink=%"GF_PRI_NLINK", "
-                                      "ia_uid=%d, ia_gid=%d, ia_size=%"PRId64", ia_blocks=%"PRId64
-                                      ", ia_atime=%s, ia_mtime=%s, ia_ctime=%s",
-                                      stbuf->ia_ino,
-                                      st_mode_from_ia (stbuf->ia_prot, stbuf->ia_type),
-                                      stbuf->ia_nlink, stbuf->ia_uid,
-                                      stbuf->ia_gid, stbuf->ia_size,
-                                      stbuf->ia_blocks, atime_buf,
-                                      mtime_buf, ctime_buf);
-
-        if (asprint_ret_value < 0)
-                statstr = NULL;
-
-        return statstr;
-}
-
-
-int
-trace_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, fd_t *fd,
-                  inode_t *inode, struct iatt *buf,
-                  struct iatt *preparent, struct iatt *postparent)
-{
-        char  *statstr = NULL;
-        char  *preparentstr = NULL;
-        char  *postparentstr = NULL;
-
-        if (trace_fop_names[GF_FOP_CREATE].enabled) {
-                if (op_ret >= 0) {
-                        statstr = trace_stat_to_str (buf);
-                        preparentstr = trace_stat_to_str (preparent);
-                        postparentstr = trace_stat_to_str (postparent);
-
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, fd=%p, ino=%"PRIu64" "
-                                "*stbuf {%s}, *preparent {%s}, *postparent = "
-                                "{%s})",
-                                frame->root->unique, op_ret, fd, inode->ino,
-                                statstr, preparentstr, postparentstr);
-
-                        if (statstr)
-                                GF_FREE (statstr);
-                        if (preparentstr)
-                                GF_FREE (preparentstr);
-                        if (postparentstr)
-                                GF_FREE (postparentstr);
-                } else {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                frame->root->unique, op_ret, op_errno);
-                }
-        }
+    gf_time_fmt(atime_buf, sizeof atime_buf, buf->ia_atime, gf_timefmt_dirent);
 
-        STACK_UNWIND_STRICT (create, frame, op_ret, op_errno, fd, inode, buf,
-                             preparent, postparent);
-        return 0;
-}
+    gf_time_fmt(mtime_buf, sizeof mtime_buf, buf->ia_mtime, gf_timefmt_dirent);
 
+    gf_time_fmt(ctime_buf, sizeof ctime_buf, buf->ia_ctime, gf_timefmt_dirent);
 
-int
-trace_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                int32_t op_ret, int32_t op_errno, fd_t *fd)
-{
+    snprintf(str, len,
+             "gfid=%s ino=%" PRIu64
+             ", mode=%o, "
+             "nlink=%" GF_PRI_NLINK ", uid=%u, gid=%u, size=%" PRIu64
+             ", "
+             "blocks=%" PRIu64
+             ", atime=%s mtime=%s ctime=%s "
+             "atime_sec=%" PRId64 ", atime_nsec=%" PRIu32
+             ","
+             " mtime_sec=%" PRId64 ", mtime_nsec=%" PRIu32
+             ", "
+             "ctime_sec=%" PRId64 ", ctime_nsec=%" PRIu32 "",
+             uuid_utoa(buf->ia_gfid), buf->ia_ino,
+             st_mode_from_ia(buf->ia_prot, buf->ia_type), buf->ia_nlink,
+             buf->ia_uid, buf->ia_gid, buf->ia_size, buf->ia_blocks, atime_buf,
+             mtime_buf, ctime_buf, buf->ia_atime, buf->ia_atime_nsec,
+             buf->ia_mtime, buf->ia_mtime_nsec, buf->ia_ctime,
+             buf->ia_ctime_nsec);
+}
+
+int
+dump_history_trace(circular_buffer_t *cb, void *data)
+{
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
 
-        if (trace_fop_names[GF_FOP_OPEN].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (op_ret=%d, op_errno=%d, *fd=%p)",
-                        frame->root->unique, op_ret, op_errno, fd);
-        }
-
-        STACK_UNWIND_STRICT (open, frame, op_ret, op_errno, fd);
-        return 0;
-}
-
-
-int
-trace_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        uint64_t ia_time = 0;
-        char     atime_buf[256];
-        char     mtime_buf[256];
-        char     ctime_buf[256];
-
-        if (trace_fop_names[GF_FOP_STAT].enabled) {
-                if (op_ret >= 0) {
-                        ia_time = buf->ia_atime;
-                        strftime (atime_buf, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        ia_time = buf->ia_mtime;
-                        strftime (mtime_buf, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        ia_time = buf->ia_ctime;
-                        strftime (ctime_buf, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, buf {"
-                                "ia_ino=%"PRIu64", st_mode=%o, ia_nlink=%"GF_PRI_NLINK", "
-                                "ia_uid=%d, ia_gid=%d, ia_rdev=%"PRIu64", ia_size=%"PRId64
-                                ", ia_blksize=%"GF_PRI_BLKSIZE", ia_blocks=%"PRId64", "
-                                "ia_atime=%s, ia_mtime=%s, ia_ctime=%s})",
-                                frame->root->unique, op_ret, buf->ia_ino,
-                                st_mode_from_ia (buf->ia_prot, buf->ia_type),
-                                buf->ia_nlink, buf->ia_uid, buf->ia_gid,
-                                buf->ia_rdev, buf->ia_size, buf->ia_blksize,
-                                buf->ia_blocks, atime_buf, mtime_buf, ctime_buf);
-                } else {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                frame->root->unique, op_ret, op_errno);
-                }
-        }
+    /* Since we are continuing with adding entries to the buffer even when
+       gettimeofday () fails, it's safe to check tm and then dump the time
+       at which the entry was added to the buffer */
 
-        STACK_UNWIND_STRICT (stat, frame, op_ret, op_errno, buf);
-        return 0;
-}
-
-
-int
-trace_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno, struct iovec *vector,
-                 int32_t count, struct iatt *buf, struct iobref *iobref)
-{
-        uint64_t ia_time = 0;
-        char     atime_buf[256];
-        char     mtime_buf[256];
-        char     ctime_buf[256];
-
-        if (trace_fop_names[GF_FOP_READ].enabled) {
-                if (op_ret >= 0) {
-                        ia_time = buf->ia_atime;
-                        strftime (atime_buf, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        ia_time = buf->ia_mtime;
-                        strftime (mtime_buf, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        ia_time = buf->ia_ctime;
-                        strftime (ctime_buf, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d, *buf {"
-                                "ia_ino=%"PRIu64", st_mode=%o, ia_nlink=%"GF_PRI_NLINK", "
-                                "ia_uid=%d, ia_gid=%d, ia_rdev=%"PRIu64", "
-                                "ia_size=%"PRId64", ia_blksize=%"GF_PRI_BLKSIZE", "
-                                "ia_blocks=%"PRId64", ia_atime=%s, ia_mtime=%s, ia_ctime=%s})",
-                                frame->root->unique, op_ret, op_errno, buf->ia_ino,
-                                st_mode_from_ia (buf->ia_prot, buf->ia_type),
-                                buf->ia_nlink, buf->ia_uid, buf->ia_gid,
-                                buf->ia_rdev, buf->ia_size, buf->ia_blksize, buf->ia_blocks,
-                                atime_buf, mtime_buf, ctime_buf);
-                } else {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                frame->root->unique, op_ret, op_errno);
-                }
-        }
+    gf_time_fmt_tv(timestr, sizeof timestr, &cb->tv, gf_timefmt_Ymd_T);
+    gf_proc_dump_write("TIME", "%s", timestr);
+
+    gf_proc_dump_write("FOP", "%s\n", (char *)cb->data);
 
-        STACK_UNWIND_STRICT (readv, frame, op_ret, op_errno, vector, count,
-                             buf, iobref);
-        return 0;
+    return 0;
 }
 
-
 int
-trace_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno,
-                  struct iatt *prebuf, struct iatt *postbuf)
-{
-        char  *preopstr = NULL;
-        char  *postopstr = NULL;
+trace_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, fd_t *fd, inode_t *inode,
+                 struct iatt *buf, struct iatt *preparent,
+                 struct iatt *postparent, dict_t *xdata)
+{
+    char statstr[1024] = {
+        0,
+    };
+    char preparentstr[1024] = {
+        0,
+    };
+    char postparentstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_CREATE].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret >= 0) {
+            TRACE_STAT_TO_STR(buf, statstr);
+            TRACE_STAT_TO_STR(preparent, preparentstr);
+            TRACE_STAT_TO_STR(postparent, postparentstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s (op_ret=%d, fd=%p"
+                     "*stbuf {%s}, *preparent {%s}, "
+                     "*postparent = {%s})",
+                     frame->root->unique, uuid_utoa(inode->gfid), op_ret, fd,
+                     statstr, preparentstr, postparentstr);
+
+            /* for 'release' log */
+            fd_ctx_set(fd, this, 0);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64 ": (op_ret=%d, op_errno=%d)",
+                     frame->root->unique, op_ret, op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(create, frame, op_ret, op_errno, fd, inode, buf,
+                       preparent, postparent, xdata);
+    return 0;
+}
+
+int
+trace_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_OPEN].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s op_ret=%d, op_errno=%d, "
+                 "*fd=%p",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret, op_errno,
+                 fd);
+
+        LOG_ELEMENT(conf, string);
+    }
+
+out:
+    /* for 'release' log */
+    if (op_ret >= 0)
+        fd_ctx_set(fd, this, 0);
+
+    TRACE_STACK_UNWIND(open, frame, op_ret, op_errno, fd, xdata);
+    return 0;
+}
+
+int
+trace_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, struct iatt *buf,
+               dict_t *xdata)
+{
+    char statstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_STAT].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(buf, statstr);
+            (void)snprintf(
+                string, sizeof(string), "%" PRId64 ": gfid=%s op_ret=%d buf=%s",
+                frame->root->unique, uuid_utoa(frame->local), op_ret, statstr);
+        } else {
+            (void)snprintf(string, sizeof(string),
+                           "%" PRId64
+                           ": gfid=%s op_ret=%d, "
+                           "op_errno=%d)",
+                           frame->root->unique, uuid_utoa(frame->local), op_ret,
+                           op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(stat, frame, op_ret, op_errno, buf, xdata);
+    return 0;
+}
+
+int
+trace_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iovec *vector,
+                int32_t count, struct iatt *buf, struct iobref *iobref,
+                dict_t *xdata)
+{
+    char statstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_READ].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret >= 0) {
+            TRACE_STAT_TO_STR(buf, statstr);
+            snprintf(
+                string, sizeof(string), "%" PRId64 ": gfid=%s op_ret=%d buf=%s",
+                frame->root->unique, uuid_utoa(frame->local), op_ret, statstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d)",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(readv, frame, op_ret, op_errno, vector, count, buf,
+                       iobref, xdata);
+    return 0;
+}
+
+int
+trace_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                 struct iatt *postbuf, dict_t *xdata)
+{
+    char preopstr[1024] = {
+        0,
+    };
+    char postopstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_WRITE].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret >= 0) {
+            TRACE_STAT_TO_STR(prebuf, preopstr);
+            TRACE_STAT_TO_STR(postbuf, postopstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": (op_ret=%d, "
+                     "*prebuf = {%s}, *postbuf = {%s})",
+                     frame->root->unique, op_ret, preopstr, postopstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(writev, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+    return 0;
+}
+
+int
+trace_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, gf_dirent_t *buf,
+                  dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_READDIR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 " : gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
+
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(readdir, frame, op_ret, op_errno, buf, xdata);
+
+    return 0;
+}
+
+int
+trace_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, gf_dirent_t *buf,
+                   dict_t *xdata)
+{
+    int count = 0;
+    char statstr[1024] = {
+        0,
+    };
+    char string[4096] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+    gf_dirent_t *entry = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_READDIRP].enabled) {
+        snprintf(string, sizeof(string),
+                 "%" PRId64 " : gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
+
+        LOG_ELEMENT(conf, string);
+    }
+    if (op_ret < 0)
+        goto out;
+
+    list_for_each_entry(entry, &buf->list, list)
+    {
+        count++;
+        TRACE_STAT_TO_STR(&entry->d_stat, statstr);
+        snprintf(string, sizeof(string),
+                 "entry no. %d, pargfid=%s, "
+                 "bname=%s *buf {%s}",
+                 count, uuid_utoa(frame->local), entry->d_name, statstr);
+        LOG_ELEMENT(conf, string);
+    }
+
+out:
+    TRACE_STACK_UNWIND(readdirp, frame, op_ret, op_errno, buf, xdata);
+    return 0;
+}
+
+int
+trace_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                struct iatt *postbuf, dict_t *xdata)
+{
+    char preopstr[1024] = {
+        0,
+    };
+    char postopstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FSYNC].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(prebuf, preopstr);
+            TRACE_STAT_TO_STR(postbuf, postopstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": (op_ret=%d, "
+                     "*prebuf = {%s}, *postbuf = {%s}",
+                     frame->root->unique, op_ret, preopstr, postopstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(fsync, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+
+    return 0;
+}
+
+int
+trace_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct iatt *statpre,
+                  struct iatt *statpost, dict_t *xdata)
+{
+    char preopstr[1024] = {
+        0,
+    };
+    char postopstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_SETATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(statpre, preopstr);
+            TRACE_STAT_TO_STR(statpost, postopstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": (op_ret=%d, "
+                     "*prebuf = {%s}, *postbuf = {%s})",
+                     frame->root->unique, op_ret, preopstr, postopstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d)",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(setattr, frame, op_ret, op_errno, statpre, statpost,
+                       xdata);
+    return 0;
+}
+
+int
+trace_fsetattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *statpre,
+                   struct iatt *statpost, dict_t *xdata)
+{
+    char preopstr[1024] = {
+        0,
+    };
+    char postopstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FSETATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(statpre, preopstr);
+            TRACE_STAT_TO_STR(statpost, postopstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": (op_ret=%d, "
+                     "*prebuf = {%s}, *postbuf = {%s})",
+                     frame->root->unique, op_ret, preopstr, postopstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d)",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(fsetattr, frame, op_ret, op_errno, statpre, statpost,
+                       xdata);
+    return 0;
+}
+
+int
+trace_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                 struct iatt *postparent, dict_t *xdata)
+{
+    char preparentstr[1024] = {
+        0,
+    };
+    char postparentstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_UNLINK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(preparent, preparentstr);
+            TRACE_STAT_TO_STR(postparent, postparentstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     " *preparent = {%s}, "
+                     "*postparent = {%s})",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     preparentstr, postparentstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d)",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(unlink, frame, op_ret, op_errno, preparent, postparent,
+                       xdata);
+    return 0;
+}
+
+int
+trace_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                 struct iatt *preoldparent, struct iatt *postoldparent,
+                 struct iatt *prenewparent, struct iatt *postnewparent,
+                 dict_t *xdata)
+{
+    char statstr[1024] = {
+        0,
+    };
+    char preoldparentstr[1024] = {
+        0,
+    };
+    char postoldparentstr[1024] = {
+        0,
+    };
+    char prenewparentstr[1024] = {
+        0,
+    };
+    char postnewparentstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_RENAME].enabled) {
+        char string[6044] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(buf, statstr);
+            TRACE_STAT_TO_STR(preoldparent, preoldparentstr);
+            TRACE_STAT_TO_STR(postoldparent, postoldparentstr);
+            TRACE_STAT_TO_STR(prenewparent, prenewparentstr);
+            TRACE_STAT_TO_STR(postnewparent, postnewparentstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": (op_ret=%d, "
+                     "*stbuf = {%s}, *preoldparent = {%s},"
+                     " *postoldparent = {%s}"
+                     " *prenewparent = {%s}, "
+                     "*postnewparent = {%s})",
+                     frame->root->unique, op_ret, statstr, preoldparentstr,
+                     postoldparentstr, prenewparentstr, postnewparentstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(rename, frame, op_ret, op_errno, buf, preoldparent,
+                       postoldparent, prenewparent, postnewparent, xdata);
+    return 0;
+}
+
+int
+trace_readlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, const char *buf,
+                   struct iatt *stbuf, dict_t *xdata)
+{
+    char statstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_READLINK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(stbuf, statstr);
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": (op_ret=%d, op_errno=%d,"
+                     "buf=%s, stbuf = { %s })",
+                     frame->root->unique, op_ret, op_errno, buf, statstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(readlink, frame, op_ret, op_errno, buf, stbuf, xdata);
+    return 0;
+}
+
+int
+trace_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, inode_t *inode,
+                 struct iatt *buf, dict_t *xdata, struct iatt *postparent)
+{
+    char statstr[1024] = {
+        0,
+    };
+    char postparentstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_LOOKUP].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(buf, statstr);
+            TRACE_STAT_TO_STR(postparent, postparentstr);
+            /* print buf->ia_gfid instead of inode->gfid,
+             * since if the inode is not yet linked to the
+             * inode table (fresh lookup) then null gfid
+             * will be printed.
+             */
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s (op_ret=%d "
+                     "*buf {%s}, *postparent {%s}",
+                     frame->root->unique, uuid_utoa(buf->ia_gfid), op_ret,
+                     statstr, postparentstr);
+
+            /* For 'forget' */
+            inode_ctx_put(inode, this, 0);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d)",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(lookup, frame, op_ret, op_errno, inode, buf, xdata,
+                       postparent);
+    return 0;
+}
+
+int
+trace_symlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, inode_t *inode,
+                  struct iatt *buf, struct iatt *preparent,
+                  struct iatt *postparent, dict_t *xdata)
+{
+    char statstr[1024] = {
+        0,
+    };
+    char preparentstr[1024] = {
+        0,
+    };
+    char postparentstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_SYMLINK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(buf, statstr);
+            TRACE_STAT_TO_STR(preparent, preparentstr);
+            TRACE_STAT_TO_STR(postparent, postparentstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s (op_ret=%d "
+                     "*stbuf = {%s}, *preparent = {%s}, "
+                     "*postparent = {%s})",
+                     frame->root->unique, uuid_utoa(inode->gfid), op_ret,
+                     statstr, preparentstr, postparentstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64 ": op_ret=%d, op_errno=%d", frame->root->unique,
+                     op_ret, op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(symlink, frame, op_ret, op_errno, inode, buf, preparent,
+                       postparent, xdata);
+    return 0;
+}
+
+int
+trace_mknod_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, inode_t *inode,
+                struct iatt *buf, struct iatt *preparent,
+                struct iatt *postparent, dict_t *xdata)
+{
+    char statstr[1024] = {
+        0,
+    };
+    char preparentstr[1024] = {
+        0,
+    };
+    char postparentstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    char string[4096] = {
+        0,
+    };
+    if (trace_fop_names[GF_FOP_MKNOD].enabled) {
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(buf, statstr);
+            TRACE_STAT_TO_STR(preparent, preparentstr);
+            TRACE_STAT_TO_STR(postparent, postparentstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s (op_ret=%d "
+                     "*stbuf = {%s}, *preparent = {%s}, "
+                     "*postparent = {%s})",
+                     frame->root->unique, uuid_utoa(inode->gfid), op_ret,
+                     statstr, preparentstr, postparentstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64 ": (op_ret=%d, op_errno=%d)",
+                     frame->root->unique, op_ret, op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(mknod, frame, op_ret, op_errno, inode, buf, preparent,
+                       postparent, xdata);
+    return 0;
+}
+
+int
+trace_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, inode_t *inode,
+                struct iatt *buf, struct iatt *preparent,
+                struct iatt *postparent, dict_t *xdata)
+{
+    char statstr[1024] = {
+        0,
+    };
+    char preparentstr[1024] = {
+        0,
+    };
+    char postparentstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_MKDIR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(buf, statstr);
+            TRACE_STAT_TO_STR(preparent, preparentstr);
+            TRACE_STAT_TO_STR(postparent, postparentstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s (op_ret=%d "
+                     ", *stbuf = {%s}, *prebuf = {%s}, "
+                     "*postbuf = {%s} )",
+                     frame->root->unique, uuid_utoa(inode->gfid), op_ret,
+                     statstr, preparentstr, postparentstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64 ": (op_ret=%d, op_errno=%d)",
+                     frame->root->unique, op_ret, op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(mkdir, frame, op_ret, op_errno, inode, buf, preparent,
+                       postparent, xdata);
+    return 0;
+}
+
+int
+trace_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, inode_t *inode,
+               struct iatt *buf, struct iatt *preparent,
+               struct iatt *postparent, dict_t *xdata)
+{
+    char statstr[1024] = {
+        0,
+    };
+    char preparentstr[1024] = {
+        0,
+    };
+    char postparentstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    char string[4096] = {
+        0,
+    };
+    if (trace_fop_names[GF_FOP_LINK].enabled) {
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(buf, statstr);
+            TRACE_STAT_TO_STR(preparent, preparentstr);
+            TRACE_STAT_TO_STR(postparent, postparentstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": (op_ret=%d, "
+                     "*stbuf = {%s},  *prebuf = {%s},"
+                     " *postbuf = {%s})",
+                     frame->root->unique, op_ret, statstr, preparentstr,
+                     postparentstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(link, frame, op_ret, op_errno, inode, buf, preparent,
+                       postparent, xdata);
+    return 0;
+}
+
+int
+trace_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    char string[4096] = {
+        0,
+    };
+    if (trace_fop_names[GF_FOP_FLUSH].enabled) {
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
+
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(flush, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+int
+trace_opendir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    char string[4096] = {
+        0,
+    };
+    if (trace_fop_names[GF_FOP_OPENDIR].enabled) {
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s op_ret=%d, op_errno=%d,"
+                 " fd=%p",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret, op_errno,
+                 fd);
+
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    /* for 'releasedir' log */
+    if (op_ret >= 0)
+        fd_ctx_set(fd, this, 0);
+
+    TRACE_STACK_UNWIND(opendir, frame, op_ret, op_errno, fd, xdata);
+    return 0;
+}
+
+int
+trace_rmdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                struct iatt *postparent, dict_t *xdata)
+{
+    char preparentstr[1024] = {
+        0,
+    };
+    char postparentstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_RMDIR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(preparent, preparentstr);
+            TRACE_STAT_TO_STR(postparent, postparentstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "*prebuf={%s},  *postbuf={%s}",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     preparentstr, postparentstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(rmdir, frame, op_ret, op_errno, preparent, postparent,
+                       xdata);
+    return 0;
+}
+
+int
+trace_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                   struct iatt *postbuf, dict_t *xdata)
+{
+    char preopstr[1024] = {
+        0,
+    };
+    char postopstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_TRUNCATE].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(prebuf, preopstr);
+            TRACE_STAT_TO_STR(postbuf, postopstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": (op_ret=%d, "
+                     "*prebuf = {%s}, *postbuf = {%s} )",
+                     frame->root->unique, op_ret, preopstr, postopstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(truncate, frame, op_ret, op_errno, prebuf, postbuf,
+                       xdata);
+    return 0;
+}
+
+int
+trace_statfs_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct statvfs *buf,
+                 dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_STATFS].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": ({f_bsize=%lu, "
+                     "f_frsize=%lu, "
+                     "f_blocks=%" GF_PRI_FSBLK ", f_bfree=%" GF_PRI_FSBLK
+                     ", "
+                     "f_bavail=%" GF_PRI_FSBLK
+                     ", "
+                     "f_files=%" GF_PRI_FSBLK
+                     ", "
+                     "f_ffree=%" GF_PRI_FSBLK
+                     ", "
+                     "f_favail=%" GF_PRI_FSBLK
+                     ", "
+                     "f_fsid=%lu, f_flag=%lu, "
+                     "f_namemax=%lu}) => ret=%d",
+                     frame->root->unique, buf->f_bsize, buf->f_frsize,
+                     buf->f_blocks, buf->f_bfree, buf->f_bavail, buf->f_files,
+                     buf->f_ffree, buf->f_favail, buf->f_fsid, buf->f_flag,
+                     buf->f_namemax, op_ret);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": (op_ret=%d, "
+                     "op_errno=%d)",
+                     frame->root->unique, op_ret, op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(statfs, frame, op_ret, op_errno, buf, xdata);
+    return 0;
+}
+
+int
+trace_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_SETXATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
+
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(setxattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+int
+trace_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *dict,
+                   dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_GETXATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s op_ret=%d, op_errno=%d,"
+                 " dict=%p",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret, op_errno,
+                 dict);
 
-        if (trace_fop_names[GF_FOP_WRITE].enabled) {
-                if (op_ret >= 0) {
-                        preopstr = trace_stat_to_str (prebuf);
-                        postopstr = trace_stat_to_str (postbuf);
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(getxattr, frame, op_ret, op_errno, dict, xdata);
+
+    return 0;
+}
+
+int
+trace_fsetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FSETXATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
 
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, ino = %"PRIu64
-                                ", *prebuf = {%s}, *postbuf = {%s})",
-                                frame->root->unique, op_ret, postbuf->ia_ino,
-                                preopstr, postopstr);
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(fsetxattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
-                        if (preopstr)
-                                GF_FREE (preopstr);
+int
+trace_fgetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *dict,
+                    dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
 
-                        if (postopstr)
-                                GF_FREE (postopstr);
-                } else {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                frame->root->unique, op_ret, op_errno);
-                }
-        }
+    conf = this->private;
 
-        STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, prebuf, postbuf);
-        return 0;
-}
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FGETXATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s op_ret=%d, op_errno=%d,"
+                 " dict=%p",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret, op_errno,
+                 dict);
 
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(fgetxattr, frame, op_ret, op_errno, dict, xdata);
 
+    return 0;
+}
 
 int
-trace_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, gf_dirent_t *buf)
+trace_removexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_READDIR].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64" :(op_ret=%d, op_errno=%d)",
-                        frame->root->unique, op_ret, op_errno);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_UNWIND_STRICT (readdir, frame, op_ret, op_errno, buf);
-
-        return 0;
-}
+    conf = this->private;
 
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_REMOVEXATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
+
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(removexattr, frame, op_ret, op_errno, xdata);
+
+    return 0;
+}
 
 int
-trace_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, gf_dirent_t *buf)
+trace_fsyncdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_READDIRP].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64" :(op_ret=%d, op_errno=%d)",
-                        frame->root->unique, op_ret, op_errno);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_UNWIND_STRICT (readdirp, frame, op_ret, op_errno, buf);
+    conf = this->private;
 
-        return 0;
-}
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FSYNCDIR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
+
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(fsyncdir, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+int
+trace_access_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_ACCESS].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s op_ret=%d, "
+                 "op_errno=%d)",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
+
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(access, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+int
+trace_ftruncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                    struct iatt *postbuf, dict_t *xdata)
+{
+    char prebufstr[1024] = {
+        0,
+    };
+    char postbufstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FTRUNCATE].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(prebuf, prebufstr);
+            TRACE_STAT_TO_STR(postbuf, postbufstr);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": op_ret=%d, "
+                     "*prebuf = {%s}, *postbuf = {%s} )",
+                     frame->root->unique, op_ret, prebufstr, postbufstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(ftruncate, frame, op_ret, op_errno, prebuf, postbuf,
+                       xdata);
+    return 0;
+}
+
+int
+trace_fstat_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                dict_t *xdata)
+{
+    char statstr[1024] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FSTAT].enabled) {
+        char string[4096] = {0.};
+        if (op_ret == 0) {
+            TRACE_STAT_TO_STR(buf, statstr);
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d "
+                     "buf=%s",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     statstr);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(fstat, frame, op_ret, op_errno, buf, xdata);
+    return 0;
+}
+
+int
+trace_lk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, struct gf_flock *lock, dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_LK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (op_ret == 0) {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "{l_type=%d, l_whence=%d, "
+                     "l_start=%" PRId64
+                     ", "
+                     "l_len=%" PRId64 ", l_pid=%u})",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     lock->l_type, lock->l_whence, lock->l_start, lock->l_len,
+                     lock->l_pid);
+        } else {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s op_ret=%d, "
+                     "op_errno=%d)",
+                     frame->root->unique, uuid_utoa(frame->local), op_ret,
+                     op_errno);
+        }
+
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(lk, frame, op_ret, op_errno, lock, xdata);
+    return 0;
+}
+
+int
+trace_entrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_ENTRYLK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
+
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(entrylk, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+int
+trace_fentrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FENTRYLK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
 
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(fentrylk, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
 int
-trace_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno,
-                 struct iatt *prebuf, struct iatt *postbuf)
+trace_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
 {
-        char  *preopstr = NULL;
-        char  *postopstr = NULL;
+    trace_conf_t *conf = NULL;
 
-        if (trace_fop_names[GF_FOP_FSYNC].enabled) {
-                if (op_ret >= 0) {
-                        preopstr = trace_stat_to_str (prebuf);
-                        postopstr = trace_stat_to_str (postbuf);
+    conf = this->private;
 
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, ino = %"PRIu64
-                                ", *prebuf = {%s}, *postbuf = {%s}",
-                                frame->root->unique, op_ret, postbuf->ia_ino,
-                                preopstr, postopstr);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_XATTROP].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
 
-                        if (preopstr)
-                                GF_FREE (preopstr);
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(xattrop, frame, op_ret, op_errno, dict, xdata);
+    return 0;
+}
 
-                        if (postopstr)
-                                GF_FREE (postopstr);
-                } else {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                frame->root->unique, op_ret, op_errno);
-                }
-        }
+int
+trace_fxattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *dict,
+                   dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
 
-        STACK_UNWIND_STRICT (fsync, frame, op_ret, op_errno, prebuf, postbuf);
-
-        return 0;
-}
-
-
-int
-trace_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno,
-                   struct iatt *statpre, struct iatt *statpost)
-{
-        uint64_t ia_time         = 0;
-        char     atime_pre[256]  = {0,};
-        char     mtime_pre[256]  = {0,};
-        char     ctime_pre[256]  = {0,};
-        char     atime_post[256] = {0,};
-        char     mtime_post[256] = {0,};
-        char     ctime_post[256] = {0,};
-
-        if (trace_fop_names[GF_FOP_SETATTR].enabled) {
-                if (op_ret >= 0) {
-                        ia_time = statpre->ia_atime;
-                        strftime (atime_pre, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        ia_time = statpre->ia_mtime;
-                        strftime (mtime_pre, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        ia_time = statpre->ia_ctime;
-                        strftime (ctime_pre, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        ia_time = statpost->ia_atime;
-                        strftime (atime_post, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        ia_time = statpost->ia_mtime;
-                        strftime (mtime_post, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        ia_time = statpost->ia_ctime;
-                        strftime (ctime_post, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, *statpre "
-                                "{ia_ino=%"PRIu64", st_mode=%o, ia_uid=%d, "
-                                "ia_gid=%d, ia_atime=%s, ia_mtime=%s, "
-                                "ia_ctime=%s}, *statpost {ia_ino=%"PRIu64", "
-                                "st_mode=%o, ia_uid=%d, ia_gid=%d, ia_atime=%s,"
-                                " ia_mtime=%s, ia_ctime=%s})",
-                                frame->root->unique, op_ret, statpre->ia_ino,
-                                st_mode_from_ia (statpre->ia_prot, statpre->ia_type),
-                                statpre->ia_uid,
-                                statpre->ia_gid, atime_pre, mtime_pre,
-                                ctime_pre, statpost->ia_ino,
-                                st_mode_from_ia (statpost->ia_prot, statpost->ia_type),
-                                statpost->ia_uid, statpost->ia_gid, atime_post,
-                                mtime_post, ctime_post);
-                } else {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                frame->root->unique, op_ret, op_errno);
-                }
-        }
+    conf = this->private;
 
-        STACK_UNWIND_STRICT (setattr, frame, op_ret, op_errno, statpre, statpost);
-        return 0;
-}
-
-
-int
-trace_fsetattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno,
-                    struct iatt *statpre, struct iatt *statpost)
-{
-        uint64_t ia_time         = 0;
-        char     atime_pre[256]  = {0,};
-        char     mtime_pre[256]  = {0,};
-        char     ctime_pre[256]  = {0,};
-        char     atime_post[256] = {0,};
-        char     mtime_post[256] = {0,};
-        char     ctime_post[256] = {0,};
-
-        if (trace_fop_names[GF_FOP_FSETATTR].enabled) {
-                if (op_ret >= 0) {
-                        ia_time = statpre->ia_atime;
-                        strftime (atime_pre, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        ia_time = statpre->ia_mtime; 
-                        strftime (mtime_pre, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        ia_time = statpre->ia_ctime;
-                        strftime (ctime_pre, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        ia_time = statpost->ia_atime;
-                        strftime (atime_post, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        ia_time = statpost->ia_mtime;
-                        strftime (mtime_post, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        ia_time = statpost->ia_ctime;
-                        strftime (ctime_post, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, *statpre "
-                                "{ia_ino=%"PRIu64", st_mode=%o, ia_uid=%d, "
-                                "ia_gid=%d, ia_atime=%s, ia_mtime=%s, "
-                                "ia_ctime=%s}, *statpost {ia_ino=%"PRIu64", "
-                                "st_mode=%o, ia_uid=%d, ia_gid=%d, ia_atime=%s,"
-                                " ia_mtime=%s, ia_ctime=%s})",
-                                frame->root->unique, op_ret, statpre->ia_ino,
-                                st_mode_from_ia (statpre->ia_prot, statpre->ia_type),
-                                statpre->ia_uid,
-                                statpre->ia_gid, atime_pre, mtime_pre,
-                                ctime_pre, statpost->ia_ino,
-                                st_mode_from_ia (statpost->ia_prot, statpost->ia_type),
-                                statpost->ia_uid, statpost->ia_gid, atime_post,
-                                mtime_post, ctime_post);
-                } else {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                frame->root->unique, op_ret, op_errno);
-                }
-        }
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FXATTROP].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
 
-        STACK_UNWIND_STRICT (fsetattr, frame, op_ret, op_errno,
-                             statpre, statpost);
-        return 0;
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(fxattrop, frame, op_ret, op_errno, dict, xdata);
+    return 0;
 }
 
-
 int
-trace_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno,
-                  struct iatt *preparent, struct iatt *postparent)
+trace_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        char *preparentstr = NULL;
-        char *postparentstr = NULL;
+    trace_conf_t *conf = NULL;
 
-        if (trace_fop_names[GF_FOP_UNLINK].enabled) {
-                if (op_ret >= 0) {
-                        preparentstr = trace_stat_to_str (preparent);
-                        postparentstr = trace_stat_to_str (postparent);
+    conf = this->private;
 
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, *preparent = {%s}, "
-                                "*postparent = {%s})",
-                                frame->root->unique, op_ret, preparentstr,
-                                postparentstr);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_INODELK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
 
-                        if (preparentstr)
-                                GF_FREE (preparentstr);
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(inodelk, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
-                        if (postparentstr)
-                                GF_FREE (postparentstr);
-                } else {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                frame->root->unique, op_ret, op_errno);
-                }
-        }
+int
+trace_finodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    trace_conf_t *conf = NULL;
 
-        STACK_UNWIND_STRICT (unlink, frame, op_ret, op_errno,
-                             preparent, postparent);
-        return 0;
-}
-
-
-int
-trace_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, struct iatt *buf,
-                  struct iatt *preoldparent, struct iatt *postoldparent,
-                  struct iatt *prenewparent, struct iatt *postnewparent)
-{
-        char  *statstr = NULL;
-        char  *preoldparentstr = NULL;
-        char  *postoldparentstr = NULL;
-        char  *prenewparentstr = NULL;
-        char  *postnewparentstr = NULL;
-
-        if (trace_fop_names[GF_FOP_RENAME].enabled) {
-                if (op_ret >= 0) {
-                        statstr = trace_stat_to_str (buf);
-                        preoldparentstr = trace_stat_to_str (preoldparent);
-                        postoldparentstr = trace_stat_to_str (postoldparent);
-
-                        prenewparentstr = trace_stat_to_str (prenewparent);
-                        postnewparentstr = trace_stat_to_str (postnewparent);
-
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, *stbuf = {%s}, "
-                                "*preoldparent = {%s}, *postoldparent = {%s}"
-                                " *prenewparent = {%s}, *postnewparent = {%s})",
-                                frame->root->unique, op_ret, statstr,
-                                preoldparentstr, postoldparentstr,
-                                prenewparentstr, postnewparentstr);
-
-                        if (preoldparentstr)
-                                GF_FREE (preoldparentstr);
-
-                        if (postoldparentstr)
-                                GF_FREE (postoldparentstr);
-
-                        if (prenewparentstr)
-                                GF_FREE (prenewparentstr);
-
-                        if (postnewparentstr)
-                                GF_FREE (postnewparentstr);
-                } else {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                frame->root->unique, op_ret, op_errno);
-                }
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (op_ret=%d, op_errno=%d, buf {ia_ino=%"PRIu64"})",
-                        frame->root->unique, op_ret, op_errno,
-                        (buf? buf->ia_ino : 0));
-        }
+    conf = this->private;
 
-        STACK_UNWIND_STRICT (rename, frame, op_ret, op_errno, buf,
-                             preoldparent, postoldparent,
-                             prenewparent, postnewparent);
-        return 0;
-}
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FINODELK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d, op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
 
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(finodelk, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
 int
-trace_readlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno,
-                    const char *buf, struct iatt *stbuf)
+trace_rchecksum_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, uint32_t weak_checksum,
+                    uint8_t *strong_checksum, dict_t *xdata)
 {
-        char *statstr = NULL;
+    trace_conf_t *conf = NULL;
 
-        if (trace_fop_names[GF_FOP_READLINK].enabled) {
+    conf = this->private;
 
-                if (op_ret == 0) {
-                        statstr = trace_stat_to_str (stbuf);
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d, buf=%s, "
-                                "stbuf = { %s })",
-                                frame->root->unique, op_ret, op_errno, buf,
-                                statstr);
-                } else
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d",
-                                frame->root->unique, op_ret, op_errno);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_RCHECKSUM].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s op_ret=%d op_errno=%d",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret,
+                 op_errno);
 
-                if (statstr)
-                        GF_FREE (statstr);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
-        STACK_UNWIND_STRICT (readlink, frame, op_ret, op_errno, buf, stbuf);
-        return 0;
+out:
+    TRACE_STACK_UNWIND(rchecksum, frame, op_ret, op_errno, weak_checksum,
+                       strong_checksum, xdata);
+
+    return 0;
 }
 
+/* *_cbk section over <----------> fop section start */
 
 int
-trace_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno,
-                  inode_t *inode, struct iatt *buf,
-                  dict_t *xattr, struct iatt *postparent)
+trace_entrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+              loc_t *loc, const char *basename, entrylk_cmd cmd,
+              entrylk_type type, dict_t *xdata)
 {
-        char  *statstr = NULL;
-        char  *postparentstr = NULL;
+    trace_conf_t *conf = NULL;
 
-        if (trace_fop_names[GF_FOP_LOOKUP].enabled) {
-                if (op_ret >= 0) {
-                        statstr = trace_stat_to_str (buf);
-                        postparentstr = trace_stat_to_str (postparent);
+    conf = this->private;
 
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, ino=%"PRIu64", "
-                                "*buf {%s}, *postparent {%s}",
-                                frame->root->unique, op_ret, inode->ino,
-                                statstr, postparentstr);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_ENTRYLK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s volume=%s, (path=%s "
+                 "basename=%s, cmd=%s, type=%s)",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), volume,
+                 loc->path, basename,
+                 ((cmd == ENTRYLK_LOCK) ? "ENTRYLK_LOCK" : "ENTRYLK_UNLOCK"),
+                 ((type == ENTRYLK_RDLCK) ? "ENTRYLK_RDLCK" : "ENTRYLK_WRLCK"));
 
-                        if (statstr)
-                                GF_FREE (statstr);
-                        if (postparentstr)
-                                GF_FREE (postparentstr);
-                } else {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                frame->root->unique, op_ret, op_errno);
-                }
-        }
+        frame->local = loc->inode->gfid;
 
-        STACK_UNWIND_STRICT (lookup, frame, op_ret, op_errno, inode, buf,
-                             xattr, postparent);
-        return 0;
-}
+        LOG_ELEMENT(conf, string);
+    }
 
+out:
+    STACK_WIND(frame, trace_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, volume, loc, basename, cmd,
+               type, xdata);
+    return 0;
+}
 
 int
-trace_symlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno,
-                   inode_t *inode, struct iatt *buf,
-                   struct iatt *preparent, struct iatt *postparent)
+trace_inodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+              loc_t *loc, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
 {
-        char  *statstr = NULL;
-        char  *preparentstr = NULL;
-        char  *postparentstr = NULL;
+    char *cmd_str = NULL;
+    char *type_str = NULL;
+    trace_conf_t *conf = NULL;
 
-        if (trace_fop_names[GF_FOP_SYMLINK].enabled) {
-                if (op_ret >= 0) {
-                        statstr = trace_stat_to_str (buf);
-                        preparentstr = trace_stat_to_str (preparent);
-                        postparentstr = trace_stat_to_str (postparent);
+    conf = this->private;
 
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, ino=%"PRIu64", "
-                                "*stbuf = {%s}, *preparent = {%s}, "
-                                "*postparent = {%s})",
-                                frame->root->unique, op_ret, inode->ino,
-                                statstr, preparentstr, postparentstr);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_INODELK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        switch (cmd) {
+#if F_GETLK != F_GETLK64
+            case F_GETLK64:
+#endif
+            case F_GETLK:
+                cmd_str = "GETLK";
+                break;
 
-                        if (statstr)
-                                GF_FREE (statstr);
+#if F_SETLK != F_SETLK64
+            case F_SETLK64:
+#endif
+            case F_SETLK:
+                cmd_str = "SETLK";
+                break;
 
-                        if (preparentstr)
-                                GF_FREE (preparentstr);
+#if F_SETLKW != F_SETLKW64
+            case F_SETLKW64:
+#endif
+            case F_SETLKW:
+                cmd_str = "SETLKW";
+                break;
+
+            default:
+                cmd_str = "UNKNOWN";
+                break;
+        }
+
+        switch (flock->l_type) {
+            case F_RDLCK:
+                type_str = "READ";
+                break;
+            case F_WRLCK:
+                type_str = "WRITE";
+                break;
+            case F_UNLCK:
+                type_str = "UNLOCK";
+                break;
+            default:
+                type_str = "UNKNOWN";
+                break;
+        }
+
+        snprintf(
+            string, sizeof(string),
+            "%" PRId64
+            ": gfid=%s volume=%s, (path=%s "
+            "cmd=%s, type=%s, start=%llu, len=%llu, "
+            "pid=%llu)",
+            frame->root->unique, uuid_utoa(loc->inode->gfid), volume, loc->path,
+            cmd_str, type_str, (unsigned long long)flock->l_start,
+            (unsigned long long)flock->l_len, (unsigned long long)flock->l_pid);
+
+        frame->local = loc->inode->gfid;
+
+        LOG_ELEMENT(conf, string);
+    }
+
+out:
+    STACK_WIND(frame, trace_inodelk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->inodelk, volume, loc, cmd, flock,
+               xdata);
+    return 0;
+}
+
+int
+trace_finodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+               fd_t *fd, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
+{
+    char *cmd_str = NULL;
+    char *type_str = NULL;
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FINODELK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        switch (cmd) {
+#if F_GETLK != F_GETLK64
+            case F_GETLK64:
+#endif
+            case F_GETLK:
+                cmd_str = "GETLK";
+                break;
 
-                        if (postparentstr)
-                                GF_FREE (postparentstr);
+#if F_SETLK != F_SETLK64
+            case F_SETLK64:
+#endif
+            case F_SETLK:
+                cmd_str = "SETLK";
+                break;
 
-                } else {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                frame->root->unique, op_ret, op_errno);
-                }
+#if F_SETLKW != F_SETLKW64
+            case F_SETLKW64:
+#endif
+            case F_SETLKW:
+                cmd_str = "SETLKW";
+                break;
+
+            default:
+                cmd_str = "UNKNOWN";
+                break;
         }
 
-        STACK_UNWIND_STRICT (symlink, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent);
-        return 0;
-}
+        switch (flock->l_type) {
+            case F_RDLCK:
+                type_str = "READ";
+                break;
+            case F_WRLCK:
+                type_str = "WRITE";
+                break;
+            case F_UNLCK:
+                type_str = "UNLOCK";
+                break;
+            default:
+                type_str = "UNKNOWN";
+                break;
+        }
+
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s volume=%s, (fd =%p "
+                 "cmd=%s, type=%s, start=%llu, len=%llu, "
+                 "pid=%llu)",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), volume, fd,
+                 cmd_str, type_str, (unsigned long long)flock->l_start,
+                 (unsigned long long)flock->l_len,
+                 (unsigned long long)flock->l_pid);
 
+        frame->local = fd->inode->gfid;
+
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    STACK_WIND(frame, trace_finodelk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->finodelk, volume, fd, cmd, flock,
+               xdata);
+    return 0;
+}
 
 int
-trace_mknod_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno,
-                 inode_t *inode, struct iatt *buf,
-                 struct iatt *preparent, struct iatt *postparent)
+trace_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+              gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
 {
-        char *statstr = NULL;
-        char *preparentstr = NULL;
-        char *postparentstr = NULL;
+    trace_conf_t *conf = NULL;
 
-        if (trace_fop_names[GF_FOP_MKNOD].enabled) {
-                if (op_ret >= 0) {
-                        statstr = trace_stat_to_str (buf);
-                        preparentstr = trace_stat_to_str (preparent);
-                        postparentstr = trace_stat_to_str (postparent);
+    conf = this->private;
 
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, ino=%"PRIu64", "
-                                "*stbuf = {%s}, *preparent = {%s}, "
-                                "*postparent = {%s})",
-                                frame->root->unique, op_ret, inode->ino,
-                                statstr, preparentstr, postparentstr);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_XATTROP].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s (path=%s flags=%d)", frame->root->unique,
+                 uuid_utoa(loc->inode->gfid), loc->path, flags);
 
-                        if (statstr)
-                                GF_FREE (statstr);
+        frame->local = loc->inode->gfid;
 
-                        if (preparentstr)
-                                GF_FREE (preparentstr);
+        LOG_ELEMENT(conf, string);
+    }
 
-                        if (postparentstr)
-                                GF_FREE (postparentstr);
-                } else {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                frame->root->unique, op_ret, op_errno);
-                }
-        }
+out:
+    STACK_WIND(frame, trace_xattrop_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->xattrop, loc, flags, dict, xdata);
 
-        STACK_UNWIND_STRICT (mknod, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent);
-        return 0;
+    return 0;
 }
 
-
 int
-trace_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno,
-                 inode_t *inode, struct iatt *buf,
-                 struct iatt *preparent, struct iatt *postparent)
+trace_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+               gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
 {
-        char  *statstr = NULL;
-        char  *preparentstr = NULL;
-        char  *postparentstr = NULL;
+    trace_conf_t *conf = NULL;
 
-        if (trace_fop_names[GF_FOP_MKDIR].enabled) {
-                if (op_ret >= 0) {
-                        statstr = trace_stat_to_str (buf);
-                        preparentstr = trace_stat_to_str (preparent);
-                        postparentstr = trace_stat_to_str (postparent);
+    conf = this->private;
 
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, ino = %"PRIu64
-                                ", *stbuf = {%s}, *prebuf = {%s}, "
-                                "*postbuf = {%s} )",
-                                frame->root->unique, op_ret, buf->ia_ino,
-                                statstr, preparentstr, postparentstr);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FXATTROP].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s fd=%p, flags=%d",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd, flags);
 
-                        if (statstr)
-                                GF_FREE (statstr);
+        frame->local = fd->inode->gfid;
 
-                        if (preparentstr)
-                                GF_FREE (preparentstr);
+        LOG_ELEMENT(conf, string);
+    }
 
-                        if (postparentstr)
-                                GF_FREE (postparentstr);
-                } else {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                frame->root->unique, op_ret, op_errno);
-                }
-        }
+out:
+    STACK_WIND(frame, trace_fxattrop_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fxattrop, fd, flags, dict, xdata);
 
-        STACK_UNWIND_STRICT (mkdir, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent);
-        return 0;
+    return 0;
 }
 
-
 int
-trace_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                int32_t op_ret, int32_t op_errno,
-                inode_t *inode, struct iatt *buf,
-                struct iatt *preparent, struct iatt *postparent)
+trace_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-        char  *statstr = NULL;
-        char  *preparentstr = NULL;
-        char  *postparentstr = NULL;
+    trace_conf_t *conf = NULL;
 
-        if (trace_fop_names[GF_FOP_LINK].enabled) {
-                if (op_ret >= 0) {
-                        statstr = trace_stat_to_str (buf);
-                        preparentstr = trace_stat_to_str (preparent);
-                        postparentstr = trace_stat_to_str (postparent);
+    conf = this->private;
 
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, ino = %"PRIu64
-                                ", *stbuf = {%s}, *prebuf = {%s}, "
-                                "*postbuf = {%s})",
-                                frame->root->unique, op_ret, buf->ia_ino,
-                                statstr, preparentstr, postparentstr);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_LOOKUP].enabled) {
+        char string[4096] = {
+            0,
+        };
+        /* TODO: print all the keys mentioned in xattr_req */
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s path=%s",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path);
 
-                        if (statstr)
-                                GF_FREE (statstr);
+        frame->local = loc->inode->gfid;
 
-                        if (preparentstr)
-                                GF_FREE (preparentstr);
+        LOG_ELEMENT(conf, string);
+    }
 
-                        if (postparentstr)
-                                GF_FREE (postparentstr);
-                } else {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                frame->root->unique, op_ret, op_errno);
-                }
-        }
+out:
+    STACK_WIND(frame, trace_lookup_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lookup, loc, xdata);
 
-        STACK_UNWIND_STRICT (link, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent);
-        return 0;
+    return 0;
 }
 
-
 int
-trace_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno)
+trace_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_FLUSH].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (op_ret=%d, op_errno=%d)",
-                        frame->root->unique, op_ret, op_errno);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_UNWIND_STRICT (flush, frame, op_ret, op_errno);
-        return 0;
-}
+    conf = this->private;
 
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_STAT].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s path=%s",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path);
 
-int
-trace_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, fd_t *fd)
-{
-        if (trace_fop_names[GF_FOP_OPENDIR].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (op_ret=%d, op_errno=%d, fd=%p)",
-                        frame->root->unique, op_ret, op_errno, fd);
-        }
+        frame->local = loc->inode->gfid;
 
-        STACK_UNWIND_STRICT (opendir, frame, op_ret, op_errno, fd);
-        return 0;
-}
+        LOG_ELEMENT(conf, string);
+    }
 
+out:
+    STACK_WIND(frame, trace_stat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->stat, loc, xdata);
+
+    return 0;
+}
 
 int
-trace_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno,
-                 struct iatt *preparent, struct iatt *postparent)
+trace_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+               dict_t *xdata)
 {
-        char  *preparentstr = NULL;
-        char  *postparentstr = NULL;
+    trace_conf_t *conf = NULL;
 
-        if (trace_fop_names[GF_FOP_RMDIR].enabled) {
-                if (op_ret >= 0) {
-                        preparentstr = trace_stat_to_str (preparent);
-                        postparentstr = trace_stat_to_str (postparent);
+    conf = this->private;
 
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, *prebuf = {%s}, "
-                                "*postbuf = {%s}",
-                                frame->root->unique, op_ret, preparentstr,
-                                postparentstr);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_READLINK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s path=%s, "
+                 "size=%" GF_PRI_SIZET ")",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                 size);
 
-                        if (preparentstr)
-                                GF_FREE (preparentstr);
+        frame->local = loc->inode->gfid;
 
-                        if (postparentstr)
-                                GF_FREE (postparentstr);
-                } else {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                frame->root->unique, op_ret, op_errno);
-                }
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
-        STACK_UNWIND_STRICT (rmdir, frame, op_ret, op_errno,
-                             preparent, postparent);
-        return 0;
-}
+out:
+    STACK_WIND(frame, trace_readlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readlink, loc, size, xdata);
 
+    return 0;
+}
 
 int
-trace_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno,
-                    struct iatt *prebuf, struct iatt *postbuf)
+trace_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+            dev_t dev, mode_t umask, dict_t *xdata)
 {
-        char  *preopstr = NULL;
-        char  *postopstr = NULL;
+    trace_conf_t *conf = NULL;
 
-        if (trace_fop_names[GF_FOP_TRUNCATE].enabled) {
-                if (op_ret >= 0) {
-                        preopstr = trace_stat_to_str (prebuf);
-                        postopstr = trace_stat_to_str (postbuf);
+    conf = this->private;
 
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, *prebuf = {%s}, "
-                                "*postbuf = {%s} )",
-                                frame->root->unique, op_ret, preopstr,
-                                postopstr);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_MKNOD].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s path=%s mode=%d "
+                 "umask=0%o, dev=%" GF_PRI_DEV ")",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                 mode, umask, dev);
 
-                        if (preopstr)
-                                GF_FREE (preopstr);
+        LOG_ELEMENT(conf, string);
+    }
 
-                        if (postopstr)
-                                GF_FREE (postopstr);
-                } else {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                frame->root->unique, op_ret, op_errno);
-                }
-        }
+out:
+    STACK_WIND(frame, trace_mknod_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mknod, loc, mode, dev, umask, xdata);
 
-        STACK_UNWIND_STRICT (truncate, frame, op_ret, op_errno, prebuf, postbuf);
-        return 0;
+    return 0;
 }
 
-
 int
-trace_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno, struct statvfs *buf)
+trace_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+            mode_t umask, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_STATFS].enabled) {
-                if (op_ret >= 0) {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": ({f_bsize=%lu, f_frsize=%lu, f_blocks=%"GF_PRI_FSBLK
-                                ", f_bfree=%"GF_PRI_FSBLK", f_bavail=%"GF_PRI_FSBLK", "
-                                "f_files=%"GF_PRI_FSBLK", f_ffree=%"GF_PRI_FSBLK", f_favail=%"
-                                GF_PRI_FSBLK", f_fsid=%lu, f_flag=%lu, f_namemax=%lu}) => ret=%d",
-                                frame->root->unique, buf->f_bsize, buf->f_frsize, buf->f_blocks,
-                                buf->f_bfree, buf->f_bavail, buf->f_files, buf->f_ffree,
-                                buf->f_favail, buf->f_fsid, buf->f_flag, buf->f_namemax, op_ret);
-                } else {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                frame->root->unique, op_ret, op_errno);
-                }
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_UNWIND_STRICT (statfs, frame, op_ret, op_errno, buf);
-        return 0;
-}
+    conf = this->private;
 
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_MKDIR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s path=%s mode=%d"
+                 " umask=0%o",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                 mode, umask);
 
-int
-trace_setxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno)
-{
-        if (trace_fop_names[GF_FOP_SETXATTR].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (op_ret=%d, op_errno=%d)",
-                        frame->root->unique, op_ret, op_errno);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
-        STACK_UNWIND_STRICT (setxattr, frame, op_ret, op_errno);
-        return 0;
+out:
+    STACK_WIND(frame, trace_mkdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mkdir, loc, mode, umask, xdata);
+    return 0;
 }
 
-
 int
-trace_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, dict_t *dict)
+trace_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+             dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_GETXATTR].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (op_ret=%d, op_errno=%d, dict=%p)",
-                        frame->root->unique, op_ret, op_errno, dict);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_UNWIND_STRICT (getxattr, frame, op_ret, op_errno, dict);
+    conf = this->private;
 
-        return 0;
-}
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_UNLINK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s path=%s flag=%d",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                 xflag);
 
+        frame->local = loc->inode->gfid;
+
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    STACK_WIND(frame, trace_unlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->unlink, loc, xflag, xdata);
+    return 0;
+}
 
 int
-trace_removexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno)
+trace_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+            dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_REMOVEXATTR].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (op_ret=%d, op_errno=%d)",
-                        frame->root->unique, op_ret, op_errno);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_UNWIND_STRICT (removexattr, frame, op_ret, op_errno);
+    conf = this->private;
 
-        return 0;
-}
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_RMDIR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s path=%s flags=%d", frame->root->unique,
+                 uuid_utoa(loc->inode->gfid), loc->path, flags);
 
+        frame->local = loc->inode->gfid;
 
-int
-trace_fsyncdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno)
-{
-        if (trace_fop_names[GF_FOP_FSYNCDIR].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (op_ret=%d, op_errno=%d)",
-                        frame->root->unique, op_ret, op_errno);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
-        STACK_UNWIND_STRICT (fsyncdir, frame, op_ret, op_errno);
-        return 0;
-}
+out:
+    STACK_WIND(frame, trace_rmdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rmdir, loc, flags, xdata);
 
+    return 0;
+}
 
 int
-trace_access_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                  int32_t op_ret, int32_t op_errno)
+trace_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+              loc_t *loc, mode_t umask, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_ACCESS].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (op_ret=%d, op_errno=%d)",
-                        frame->root->unique, op_ret, op_errno);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_UNWIND_STRICT (access, frame, op_ret, op_errno);
-        return 0;
-}
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_SYMLINK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s linkpath=%s, path=%s"
+                 " umask=0%o",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), linkpath,
+                 loc->path, umask);
 
+        LOG_ELEMENT(conf, string);
+    }
+
+out:
+    STACK_WIND(frame, trace_symlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->symlink, linkpath, loc, umask, xdata);
+
+    return 0;
+}
 
 int
-trace_ftruncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno,
-                     struct iatt *prebuf, struct iatt *postbuf)
+trace_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+             dict_t *xdata)
 {
-        char  *prebufstr = NULL;
-        char  *postbufstr = NULL;
+    char oldgfid[50] = {
+        0,
+    };
+    char newgfid[50] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
 
-        if (trace_fop_names[GF_FOP_FTRUNCATE].enabled) {
-                if (op_ret >= 0) {
-                        prebufstr = trace_stat_to_str (prebuf);
-                        postbufstr = trace_stat_to_str (postbuf);
+    conf = this->private;
 
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, *prebuf = {%s}, "
-                                "*postbuf = {%s} )",
-                                frame->root->unique, op_ret,
-                                prebufstr, postbufstr);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_RENAME].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (newloc->inode)
+            uuid_utoa_r(newloc->inode->gfid, newgfid);
+        else
+            strcpy(newgfid, "0");
 
-                        if (prebufstr)
-                                GF_FREE (prebufstr);
+        uuid_utoa_r(oldloc->inode->gfid, oldgfid);
 
-                        if (postbufstr)
-                                GF_FREE (postbufstr);
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": oldgfid=%s oldpath=%s --> "
+                 "newgfid=%s newpath=%s",
+                 frame->root->unique, oldgfid, oldloc->path, newgfid,
+                 newloc->path);
 
-                } else {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                frame->root->unique, op_ret, op_errno);
-                }
-        }
+        frame->local = oldloc->inode->gfid;
 
-        STACK_UNWIND_STRICT (ftruncate, frame, op_ret, op_errno, prebuf, postbuf);
-        return 0;
-}
-
-
-int
-trace_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        uint64_t ia_time        = 0;
-        char     atime_buf[256] = {0, };
-        char     mtime_buf[256] = {0, };
-        char     ctime_buf[256] = {0, };
-
-        if (trace_fop_names[GF_FOP_FSTAT].enabled) {
-                if (op_ret >= 0) {
-                        ia_time = buf->ia_atime;
-                        strftime (atime_buf, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        ia_time = buf->ia_mtime;
-                        strftime (mtime_buf, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        ia_time = buf->ia_ctime;
-                        strftime (ctime_buf, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, *buf {"
-                                "ia_ino=%"PRIu64", st_mode=%o, ia_nlink=%"GF_PRI_NLINK", "
-                                "ia_uid=%d, ia_gid=%d, ia_rdev=%"PRIu64", ia_size=%"PRId64", "
-                                "ia_blksize=%"GF_PRI_BLKSIZE", ia_blocks=%"PRId64", ia_atime=%s, "
-                                "ia_mtime=%s, ia_ctime=%s})",
-                                frame->root->unique, op_ret, buf->ia_ino,
-                                st_mode_from_ia (buf->ia_prot, buf->ia_type),
-                                buf->ia_nlink, buf->ia_uid, buf->ia_gid,
-                                buf->ia_rdev, buf->ia_size, buf->ia_blksize,
-                                buf->ia_blocks, atime_buf, mtime_buf, ctime_buf);
-                } else {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                frame->root->unique, op_ret, op_errno);
-                }
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
-        STACK_UNWIND_STRICT (fstat, frame, op_ret, op_errno, buf);
-        return 0;
-}
+out:
+    STACK_WIND(frame, trace_rename_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
 
+    return 0;
+}
 
 int
-trace_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-              int32_t op_ret, int32_t op_errno, struct gf_flock *lock)
+trace_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+           dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_LK].enabled) {
-                if (op_ret >= 0) {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, {l_type=%d, l_whence=%d, "
-                                "l_start=%"PRId64", l_len=%"PRId64", l_pid=%u})",
-                                frame->root->unique, op_ret, lock->l_type, lock->l_whence,
-                                lock->l_start, lock->l_len, lock->l_pid);
-                } else {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (op_ret=%d, op_errno=%d)",
-                                frame->root->unique, op_ret, op_errno);
-                }
-        }
+    char oldgfid[50] = {
+        0,
+    };
+    char newgfid[50] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
 
-        STACK_UNWIND_STRICT (lk, frame, op_ret, op_errno, lock);
-        return 0;
-}
+    conf = this->private;
 
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_LINK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (newloc->inode)
+            uuid_utoa_r(newloc->inode->gfid, newgfid);
+        else
+            strcpy(newgfid, "0");
 
+        uuid_utoa_r(oldloc->inode->gfid, oldgfid);
 
-int
-trace_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno)
-{
-        if (trace_fop_names[GF_FOP_ENTRYLK].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": op_ret=%d, op_errno=%d",
-                        frame->root->unique, op_ret, op_errno);
-        }
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": oldgfid=%s oldpath=%s --> "
+                 "newgfid=%s newpath=%s",
+                 frame->root->unique, oldgfid, oldloc->path, newgfid,
+                 newloc->path);
 
-        STACK_UNWIND_STRICT (entrylk, frame, op_ret, op_errno);
-        return 0;
-}
+        frame->local = oldloc->inode->gfid;
 
+        LOG_ELEMENT(conf, string);
+    }
+
+out:
+    STACK_WIND(frame, trace_link_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->link, oldloc, newloc, xdata);
+    return 0;
+}
 
 int
-trace_xattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, dict_t *dict)
+trace_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+              struct iatt *stbuf, int32_t valid, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_XATTROP].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (op_ret=%d, op_errno=%d)",
-                        frame->root->unique, op_ret, op_errno);
-        }
+    char actime_str[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    char modtime_str[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
 
-        STACK_UNWIND_STRICT (xattrop, frame, op_ret, op_errno, dict);
-        return 0;
-}
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_SETATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (valid & GF_SET_ATTR_MODE) {
+            snprintf(
+                string, sizeof(string), "%" PRId64 ": gfid=%s path=%s mode=%o)",
+                frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                st_mode_from_ia(stbuf->ia_prot, stbuf->ia_type));
 
+            LOG_ELEMENT(conf, string);
+            memset(string, 0, sizeof(string));
+        }
 
-int
-trace_fxattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, dict_t *dict)
-{
-        if (trace_fop_names[GF_FOP_FXATTROP].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (op_ret=%d, op_errno=%d)",
-                        frame->root->unique, op_ret, op_errno);
+        if (valid & (GF_SET_ATTR_UID | GF_SET_ATTR_GID)) {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s path=%s uid=%o,"
+                     " gid=%o",
+                     frame->root->unique, uuid_utoa(loc->inode->gfid),
+                     loc->path, stbuf->ia_uid, stbuf->ia_gid);
+
+            LOG_ELEMENT(conf, string);
+            memset(string, 0, sizeof(string));
         }
 
-        STACK_UNWIND_STRICT (fxattrop, frame, op_ret, op_errno, dict);
-        return 0;
-}
+        if (valid & (GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME)) {
+            gf_time_fmt(actime_str, sizeof actime_str, stbuf->ia_atime,
+                        gf_timefmt_bdT);
 
+            gf_time_fmt(modtime_str, sizeof modtime_str, stbuf->ia_mtime,
+                        gf_timefmt_bdT);
 
-int
-trace_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno)
-{
-        if (trace_fop_names[GF_FOP_INODELK].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": op_ret=%d, op_errno=%d",
-                        frame->root->unique, op_ret, op_errno);
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s path=%s "
+                     "ia_atime=%s, ia_mtime=%s",
+                     frame->root->unique, uuid_utoa(loc->inode->gfid),
+                     loc->path, actime_str, modtime_str);
+
+            LOG_ELEMENT(conf, string);
+            memset(string, 0, sizeof(string));
+        }
+        frame->local = loc->inode->gfid;
+    }
+
+out:
+    STACK_WIND(frame, trace_setattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setattr, loc, stbuf, valid, xdata);
+
+    return 0;
+}
+
+int
+trace_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+               struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    char actime_str[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    char modtime_str[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FSETATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        if (valid & GF_SET_ATTR_MODE) {
+            snprintf(string, sizeof(string),
+                     "%" PRId64 ": gfid=%s fd=%p, mode=%o", frame->root->unique,
+                     uuid_utoa(fd->inode->gfid), fd,
+                     st_mode_from_ia(stbuf->ia_prot, stbuf->ia_type));
+
+            LOG_ELEMENT(conf, string);
+            memset(string, 0, sizeof(string));
         }
 
-        STACK_UNWIND_STRICT (inodelk, frame, op_ret, op_errno);
-        return 0;
-}
+        if (valid & (GF_SET_ATTR_UID | GF_SET_ATTR_GID)) {
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s fd=%p, uid=%o, "
+                     "gid=%o",
+                     frame->root->unique, uuid_utoa(fd->inode->gfid), fd,
+                     stbuf->ia_uid, stbuf->ia_gid);
+
+            LOG_ELEMENT(conf, string);
+            memset(string, 0, sizeof(string));
+        }
+
+        if (valid & (GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME)) {
+            gf_time_fmt(actime_str, sizeof actime_str, stbuf->ia_atime,
+                        gf_timefmt_bdT);
+
+            gf_time_fmt(modtime_str, sizeof modtime_str, stbuf->ia_mtime,
+                        gf_timefmt_bdT);
+
+            snprintf(string, sizeof(string),
+                     "%" PRId64
+                     ": gfid=%s fd=%p "
+                     "ia_atime=%s, ia_mtime=%s",
+                     frame->root->unique, uuid_utoa(fd->inode->gfid), fd,
+                     actime_str, modtime_str);
 
+            LOG_ELEMENT(conf, string);
+            memset(string, 0, sizeof(string));
+        }
+        frame->local = fd->inode->gfid;
+    }
 
-int
-trace_entrylk (call_frame_t *frame, xlator_t *this,
-               const char *volume, loc_t *loc, const char *basename,
-               entrylk_cmd cmd, entrylk_type type)
+out:
+    STACK_WIND(frame, trace_fsetattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetattr, fd, stbuf, valid, xdata);
+
+    return 0;
+}
+
+static int
+trace_seek_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, off_t offset, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_ENTRYLK].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": volume=%s, (loc= {path=%s, ino=%"PRIu64"} basename=%s, cmd=%s, type=%s)",
-                        frame->root->unique, volume, loc->path, loc->inode->ino, basename,
-                        ((cmd == ENTRYLK_LOCK) ? "ENTRYLK_LOCK" : "ENTRYLK_UNLOCK"),
-                        ((type == ENTRYLK_RDLCK) ? "ENTRYLK_RDLCK" : "ENTRYLK_WRLCK"));
-        }
+    trace_conf_t *conf = this->private;
 
-        STACK_WIND (frame, trace_entrylk_cbk,
-                    FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->entrylk,
-                    volume, loc, basename, cmd, type);
-        return 0;
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_SEEK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s op_ret=%d op_errno=%d, "
+                 "offset=%" PRId64 "",
+                 frame->root->unique, uuid_utoa(frame->local), op_ret, op_errno,
+                 offset);
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    TRACE_STACK_UNWIND(seek, frame, op_ret, op_errno, offset, xdata);
+    return 0;
 }
 
+static int
+trace_seek(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+           gf_seek_what_t what, dict_t *xdata)
+{
+    trace_conf_t *conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_SEEK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s fd=%p "
+                 "offset=%" PRId64 " what=%d",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd, offset,
+                 what);
+
+        frame->local = fd->inode->gfid;
+        LOG_ELEMENT(conf, string);
+    }
+out:
+    STACK_WIND(frame, trace_seek_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->seek, fd, offset, what, xdata);
+
+    return 0;
+}
 
 int
-trace_inodelk (call_frame_t *frame, xlator_t *this, const char *volume,
-               loc_t *loc, int32_t cmd, struct gf_flock *flock)
+trace_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+               dict_t *xdata)
 {
-        char *cmd_str = NULL;
-        char *type_str = NULL;
+    trace_conf_t *conf = NULL;
 
-        if (trace_fop_names[GF_FOP_INODELK].enabled) {
-                switch (cmd) {
-#if F_GETLK != F_GETLK64
-                case F_GETLK64:
-#endif
-                case F_GETLK:
-                        cmd_str = "GETLK";
-                        break;
+    conf = this->private;
 
-#if F_SETLK != F_SETLK64
-                case F_SETLK64:
-#endif
-                case F_SETLK:
-                        cmd_str = "SETLK";
-                        break;
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_TRUNCATE].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s path=%s, "
+                 "offset=%" PRId64 "",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                 offset);
 
-#if F_SETLKW != F_SETLKW64
-                case F_SETLKW64:
-#endif
-                case F_SETLKW:
-                        cmd_str = "SETLKW";
-                        break;
-
-                default:
-                        cmd_str = "UNKNOWN";
-                        break;
-                }
-
-                switch (flock->l_type) {
-                case F_RDLCK:
-                        type_str = "READ";
-                        break;
-                case F_WRLCK:
-                        type_str = "WRITE";
-                        break;
-                case F_UNLCK:
-                        type_str = "UNLOCK";
-                        break;
-                default:
-                        type_str = "UNKNOWN";
-                        break;
-                }
-
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": volume=%s, (loc {path=%s, ino=%"PRIu64"}, "
-                        "cmd=%s, type=%s, start=%llu, len=%llu, pid=%llu)",
-                        frame->root->unique, volume, loc->path, loc->inode->ino,
-                        cmd_str, type_str, (unsigned long long) flock->l_start,
-                        (unsigned long long) flock->l_len,
-                        (unsigned long long) flock->l_pid);
-        }
+        frame->local = loc->inode->gfid;
 
-        STACK_WIND (frame, trace_inodelk_cbk,
-                    FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->inodelk,
-                    volume, loc, cmd, flock);
-        return 0;
-}
+        LOG_ELEMENT(conf, string);
+    }
 
+out:
+    STACK_WIND(frame, trace_truncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+
+    return 0;
+}
 
 int
-trace_finodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno)
+trace_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+           fd_t *fd, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_FINODELK].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": op_ret=%d, op_errno=%d",
-                        frame->root->unique, op_ret, op_errno);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_UNWIND_STRICT (finodelk, frame, op_ret, op_errno);
-        return 0;
-}
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_OPEN].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s path=%s flags=%d fd=%p",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                 flags, fd);
 
+        frame->local = loc->inode->gfid;
+
+        LOG_ELEMENT(conf, string);
+    }
+
+out:
+    STACK_WIND(frame, trace_open_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->open, loc, flags, fd, xdata);
+    return 0;
+}
 
 int
-trace_finodelk (call_frame_t *frame, xlator_t *this, const char *volume,
-                fd_t *fd, int32_t cmd, struct gf_flock *flock)
+trace_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+             mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
 {
-        char *cmd_str = NULL, *type_str = NULL;
+    trace_conf_t *conf = NULL;
 
-        if (trace_fop_names[GF_FOP_FINODELK].enabled) {
-                switch (cmd) {
-#if F_GETLK != F_GETLK64
-                case F_GETLK64:
-#endif
-                case F_GETLK:
-                        cmd_str = "GETLK";
-                        break;
+    conf = this->private;
 
-#if F_SETLK != F_SETLK64
-                case F_SETLK64:
-#endif
-                case F_SETLK:
-                        cmd_str = "SETLK";
-                        break;
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_CREATE].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s path=%s, fd=%p, "
+                 "flags=0%o mode=0%o umask=0%o",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                 fd, flags, mode, umask);
 
-#if F_SETLKW != F_SETLKW64
-                case F_SETLKW64:
-#endif
-                case F_SETLKW:
-                        cmd_str = "SETLKW";
-                        break;
-
-                default:
-                        cmd_str = "UNKNOWN";
-                        break;
-                }
-
-                switch (flock->l_type) {
-                case F_RDLCK:
-                        type_str = "READ";
-                        break;
-                case F_WRLCK:
-                        type_str = "WRITE";
-                        break;
-                case F_UNLCK:
-                        type_str = "UNLOCK";
-                        break;
-                default:
-                        type_str = "UNKNOWN";
-                        break;
-                }
-
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": volume=%s, (fd =%p, ino=%"PRIu64"}, "
-                        "cmd=%s, type=%s, start=%llu, len=%llu, pid=%llu)",
-                        frame->root->unique, volume, fd, fd->inode->ino,
-                        cmd_str, type_str, (unsigned long long) flock->l_start,
-                        (unsigned long long) flock->l_len,
-                        (unsigned long long) flock->l_pid);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
-        STACK_WIND (frame, trace_finodelk_cbk,
-                    FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->finodelk,
-                    volume, fd, cmd, flock);
-        return 0;
+out:
+    STACK_WIND(frame, trace_create_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->create, loc, flags, mode, umask, fd,
+               xdata);
+    return 0;
 }
 
-
 int
-trace_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc,
-               gf_xattrop_flags_t flags, dict_t *dict)
+trace_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t offset, uint32_t flags, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_XATTROP].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (path=%s, ino=%"PRIu64" flags=%d)",
-                        frame->root->unique, loc->path, loc->inode->ino, flags);
+    trace_conf_t *conf = NULL;
 
-        }
+    conf = this->private;
 
-        STACK_WIND (frame, trace_xattrop_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->xattrop,
-                    loc, flags, dict);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_READ].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s fd=%p, size=%" GF_PRI_SIZET
+                 "offset=%" PRId64 " flags=0%x)",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd, size,
+                 offset, flags);
 
-        return 0;
-}
+        frame->local = fd->inode->gfid;
 
+        LOG_ELEMENT(conf, string);
+    }
+
+out:
+    STACK_WIND(frame, trace_readv_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readv, fd, size, offset, flags, xdata);
+    return 0;
+}
 
 int
-trace_fxattrop (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                gf_xattrop_flags_t flags, dict_t *dict)
+trace_writev(call_frame_t *frame, xlator_t *this, fd_t *fd,
+             struct iovec *vector, int32_t count, off_t offset, uint32_t flags,
+             struct iobref *iobref, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_FXATTROP].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (fd=%p, flags=%d)",
-                        frame->root->unique, fd, flags);
+    trace_conf_t *conf = NULL;
+    int i = 0;
+    size_t total_size = 0;
 
-        }
+    conf = this->private;
 
-        STACK_WIND (frame, trace_fxattrop_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fxattrop,
-                    fd, flags, dict);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_WRITE].enabled) {
+        char string[4096] = {
+            0,
+        };
+        for (i = 0; i < count; i++)
+            total_size += vector[i].iov_len;
 
-        return 0;
-}
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s fd=%p, count=%d, "
+                 " offset=%" PRId64 " flags=0%x write_size=%zu",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd, count,
+                 offset, flags, total_size);
+
+        frame->local = fd->inode->gfid;
 
+        LOG_ELEMENT(conf, string);
+    }
+
+out:
+    STACK_WIND(frame, trace_writev_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->writev, fd, vector, count, offset,
+               flags, iobref, xdata);
+    return 0;
+}
 
 int
-trace_lookup (call_frame_t *frame, xlator_t *this,
-              loc_t *loc, dict_t *xattr_req)
+trace_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_LOOKUP].enabled) {
-                /* TODO: print all the keys mentioned in xattr_req */
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (loc {path=%s, ino=%"PRIu64"})",
-                        frame->root->unique, loc->path,
-                        loc->inode->ino);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_WIND (frame, trace_lookup_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->lookup,
-                    loc, xattr_req);
+    conf = this->private;
 
-        return 0;
-}
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_STATFS].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s path=%s",
+                 frame->root->unique,
+                 (loc->inode) ? uuid_utoa(loc->inode->gfid) : "0", loc->path);
+
+        LOG_ELEMENT(conf, string);
+    }
 
+out:
+    STACK_WIND(frame, trace_statfs_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->statfs, loc, xdata);
+    return 0;
+}
 
 int
-trace_stat (call_frame_t *frame, xlator_t *this, loc_t *loc)
+trace_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_STAT].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (loc {path=%s, ino=%"PRIu64"})",
-                        frame->root->unique, loc->path, loc->inode->ino);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_WIND (frame, trace_stat_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->stat,
-                    loc);
+    conf = this->private;
 
-        return 0;
-}
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FLUSH].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s fd=%p",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd);
+
+        frame->local = fd->inode->gfid;
 
+        LOG_ELEMENT(conf, string);
+    }
+
+out:
+    STACK_WIND(frame, trace_flush_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->flush, fd, xdata);
+    return 0;
+}
 
 int
-trace_readlink (call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size)
+trace_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+            dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_READLINK].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (loc {path=%s, ino=%"PRIu64"}, size=%"GF_PRI_SIZET")",
-                        frame->root->unique, loc->path, loc->inode->ino, size);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_WIND (frame, trace_readlink_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->readlink,
-                    loc, size);
+    conf = this->private;
 
-        return 0;
-}
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FSYNC].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s flags=%d fd=%p",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), flags, fd);
+
+        frame->local = fd->inode->gfid;
 
+        LOG_ELEMENT(conf, string);
+    }
+
+out:
+    STACK_WIND(frame, trace_fsync_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsync, fd, flags, xdata);
+    return 0;
+}
 
 int
-trace_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc,
-             mode_t mode, dev_t dev, dict_t *params)
+trace_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+               int32_t flags, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_MKNOD].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (loc {path=%s, ino=%"PRIu64"}, mode=%d, dev=%"GF_PRI_DEV")",
-                        frame->root->unique, loc->path, loc->inode->ino, mode, dev);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_WIND (frame, trace_mknod_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->mknod,
-                    loc, mode, dev, params);
+    conf = this->private;
 
-        return 0;
-}
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_SETXATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s path=%s flags=%d", frame->root->unique,
+                 uuid_utoa(loc->inode->gfid), loc->path, flags);
 
+        frame->local = loc->inode->gfid;
 
-int
-trace_mkdir (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
-             dict_t *params)
-{
-        if (trace_fop_names[GF_FOP_MKDIR].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (path=%s, ino=%"PRIu64", mode=%d)",
-                        frame->root->unique, loc->path,
-                        ((loc->inode)? loc->inode->ino : 0), mode);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
-        STACK_WIND (frame, trace_mkdir_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->mkdir,
-                    loc, mode, params);
-        return 0;
+out:
+    STACK_WIND(frame, trace_setxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setxattr, loc, dict, flags, xdata);
+    return 0;
 }
 
-
 int
-trace_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc)
+trace_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               const char *name, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_UNLINK].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (loc {path=%s, ino=%"PRIu64"})",
-                        frame->root->unique, loc->path, loc->inode->ino);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_WIND (frame, trace_unlink_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->unlink,
-                    loc);
-        return 0;
-}
+    conf = this->private;
 
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_GETXATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s path=%s name=%s",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                 name);
 
-int
-trace_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags)
-{
-        if (trace_fop_names[GF_FOP_RMDIR].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (loc {path=%s, ino=%"PRIu64"}, flags=%d)",
-                        frame->root->unique, loc->path, loc->inode->ino, flags);
-        }
+        frame->local = loc->inode->gfid;
 
-        STACK_WIND (frame, trace_rmdir_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->rmdir,
-                    loc, flags);
+        LOG_ELEMENT(conf, string);
+    }
 
-        return 0;
+out:
+    STACK_WIND(frame, trace_getxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->getxattr, loc, name, xdata);
+    return 0;
 }
 
-
 int
-trace_symlink (call_frame_t *frame, xlator_t *this, const char *linkpath,
-               loc_t *loc, dict_t *params)
+trace_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                  const char *name, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_SYMLINK].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (linkpath=%s, loc {path=%s, ino=%"PRIu64"})",
-                        frame->root->unique, linkpath, loc->path,
-                        ((loc->inode)? loc->inode->ino : 0));
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_WIND (frame, trace_symlink_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->symlink,
-                    linkpath, loc, params);
+    conf = this->private;
 
-        return 0;
-}
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_REMOVEXATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s path=%s name=%s",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                 name);
 
+        frame->local = loc->inode->gfid;
 
-int
-trace_rename (call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc)
-{
-        if (trace_fop_names[GF_FOP_RENAME].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (oldloc {path=%s, ino=%"PRIu64"}, "
-                        "newloc{path=%s, ino=%"PRIu64"})",
-                        frame->root->unique, oldloc->path, oldloc->ino,
-                        newloc->path, newloc->ino);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
-        STACK_WIND (frame, trace_rename_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->rename,
-                    oldloc, newloc);
+out:
+    STACK_WIND(frame, trace_removexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->removexattr, loc, name, xdata);
 
-        return 0;
+    return 0;
 }
 
-
 int
-trace_link (call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc)
+trace_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+              dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_LINK].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (oldloc {path=%s, ino=%"PRIu64"}, "
-                        "newloc {path=%s, ino=%"PRIu64"})",
-                        frame->root->unique, oldloc->path, oldloc->inode->ino,
-                        newloc->path, newloc->inode->ino);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_WIND (frame, trace_link_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->link,
-                    oldloc, newloc);
-        return 0;
-}
-
-
-int
-trace_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-               struct iatt *stbuf, int32_t valid)
-{
-        uint64_t ia_time          = 0;
-        char     actime_str[256]  = {0,};
-        char     modtime_str[256] = {0,};
-
-        if (trace_fop_names[GF_FOP_SETATTR].enabled) {
-                if (valid & GF_SET_ATTR_MODE) {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (loc {path=%s, ino=%"PRIu64"},"
-                                " mode=%o)", frame->root->unique, loc->path,
-                                loc->inode->ino,
-                                st_mode_from_ia (stbuf->ia_prot, stbuf->ia_type));
-                }
-
-                if (valid & (GF_SET_ATTR_UID | GF_SET_ATTR_GID)) {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (loc {path=%s, ino=%"PRIu64"},"
-                                " uid=%o, gid=%o)",
-                                frame->root->unique, loc->path, loc->inode->ino,
-                                stbuf->ia_uid, stbuf->ia_gid);
-                }
-
-                if (valid & (GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME)) {
-                        ia_time = stbuf->ia_atime;
-                        strftime (actime_str, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        ia_time = stbuf->ia_mtime;
-                        strftime (modtime_str, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
-
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (loc {path=%s, ino=%"PRIu64"}, "
-                                "*stbuf=%p {ia_atime=%s, ia_mtime=%s})",
-                                frame->root->unique, loc->path, loc->inode->ino,
-                                stbuf, actime_str, modtime_str);
-                }
-        }
+    conf = this->private;
 
-        STACK_WIND (frame, trace_setattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->setattr,
-                    loc, stbuf, valid);
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_OPENDIR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s path=%s fd=%p",
+                 frame->root->unique, uuid_utoa(loc->inode->gfid), loc->path,
+                 fd);
 
-        return 0;
-}
+        frame->local = loc->inode->gfid;
+
+        LOG_ELEMENT(conf, string);
+    }
 
+out:
+    STACK_WIND(frame, trace_opendir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->opendir, loc, fd, xdata);
+    return 0;
+}
 
 int
-trace_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                struct iatt *stbuf, int32_t valid)
+trace_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+               off_t offset, dict_t *dict)
 {
-        uint64_t ia_time          = 0;
-        char     actime_str[256]  = {0,};
-        char     modtime_str[256] = {0,};
-
-        if (trace_fop_names[GF_FOP_FSETATTR].enabled) {
-                if (valid & GF_SET_ATTR_MODE) {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (*fd=%p, mode=%o)",
-                                frame->root->unique, fd,
-                                st_mode_from_ia (stbuf->ia_prot, stbuf->ia_type));
-                }
+    trace_conf_t *conf = NULL;
 
-                if (valid & (GF_SET_ATTR_UID | GF_SET_ATTR_GID)) {
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (*fd=%p, uid=%o, gid=%o)",
-                                frame->root->unique, fd,
-                                stbuf->ia_uid, stbuf->ia_gid);
-                }
+    conf = this->private;
 
-                if (valid & (GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME)) {
-                        ia_time = stbuf->ia_atime;
-                        strftime (actime_str, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_READDIRP].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s fd=%p, size=%" GF_PRI_SIZET
+                 ", offset=%" PRId64 " dict=%p",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd, size,
+                 offset, dict);
 
-                        ia_time = stbuf->ia_mtime;
-                        strftime (modtime_str, 256, "[%b %d %H:%M:%S]",
-                                  localtime ((time_t *)&ia_time));
+        frame->local = fd->inode->gfid;
 
-                        gf_log (this->name, GF_LOG_NORMAL,
-                                "%"PRId64": (*fd=%p"
-                                "*stbuf=%p {ia_atime=%s, ia_mtime=%s})",
-                                frame->root->unique, fd, stbuf, actime_str,
-                                modtime_str);
-                }
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
-        STACK_WIND (frame, trace_fsetattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fsetattr,
-                    fd, stbuf, valid);
+out:
+    STACK_WIND(frame, trace_readdirp_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdirp, fd, size, offset, dict);
 
-        return 0;
+    return 0;
 }
 
-
 int
-trace_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                off_t offset)
+trace_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+              off_t offset, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_TRUNCATE].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (loc {path=%s, ino=%"PRIu64"}, offset=%"PRId64")",
-                        frame->root->unique, loc->path, loc->inode->ino, offset);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_WIND (frame, trace_truncate_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->truncate,
-                    loc, offset);
+    conf = this->private;
 
-        return 0;
-}
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_READDIR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s fd=%p, size=%" GF_PRI_SIZET
+                 ", offset=%" PRId64,
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd, size,
+                 offset);
 
+        frame->local = fd->inode->gfid;
 
-int
-trace_open (call_frame_t *frame, xlator_t *this, loc_t *loc,
-            int32_t flags, fd_t *fd, int32_t wbflags)
-{
-        if (trace_fop_names[GF_FOP_OPEN].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (loc {path=%s, ino=%"PRIu64"}, flags=%d, "
-                        "fd=%p, wbflags=%d)",
-                        frame->root->unique, loc->path, loc->inode->ino, flags,
-                        fd, wbflags);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
-        STACK_WIND (frame, trace_open_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->open,
-                    loc, flags, fd, wbflags);
-        return 0;
-}
+out:
+    STACK_WIND(frame, trace_readdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdir, fd, size, offset, xdata);
 
+    return 0;
+}
 
 int
-trace_create (call_frame_t *frame, xlator_t *this, loc_t *loc,
-              int32_t flags, mode_t mode, fd_t *fd, dict_t *params)
+trace_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+               dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_CREATE].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (loc {path=%s, ino=%"PRIu64"}, flags=0%o mode=0%o)",
-                        frame->root->unique, loc->path, loc->inode->ino, flags, mode);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_WIND (frame, trace_create_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->create,
-                    loc, flags, mode, fd, params);
-        return 0;
-}
+    conf = this->private;
 
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FSYNCDIR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s datasync=%d fd=%p", frame->root->unique,
+                 uuid_utoa(fd->inode->gfid), datasync, fd);
 
-int
-trace_readv (call_frame_t *frame, xlator_t *this, fd_t *fd,
-             size_t size, off_t offset)
-{
-        if (trace_fop_names[GF_FOP_READ].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (*fd=%p, size=%"GF_PRI_SIZET", offset=%"PRId64")",
-                        frame->root->unique, fd, size, offset);
-        }
+        frame->local = fd->inode->gfid;
 
-        STACK_WIND (frame, trace_readv_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->readv,
-                    fd, size, offset);
-        return 0;
-}
+        LOG_ELEMENT(conf, string);
+    }
 
+out:
+    STACK_WIND(frame, trace_fsyncdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsyncdir, fd, datasync, xdata);
+    return 0;
+}
 
 int
-trace_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
-              struct iovec *vector, int32_t count,
-              off_t offset, struct iobref *iobref)
+trace_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+             dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_WRITE].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (*fd=%p, *vector=%p, count=%d, offset=%"PRId64")",
-                        frame->root->unique, fd, vector, count, offset);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_WIND (frame, trace_writev_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->writev,
-                    fd, vector, count, offset, iobref);
-        return 0;
-}
+    conf = this->private;
 
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_ACCESS].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s path=%s mask=0%o", frame->root->unique,
+                 uuid_utoa(loc->inode->gfid), loc->path, mask);
 
-int
-trace_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc)
-{
-        if (trace_fop_names[GF_FOP_STATFS].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (loc {path=%s, ino=%"PRIu64"})",
-                        frame->root->unique, loc->path,
-                        ((loc->inode)? loc->inode->ino : 0));
-        }
+        frame->local = loc->inode->gfid;
 
-        STACK_WIND (frame, trace_statfs_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->statfs,
-                    loc);
-        return 0;
-}
+        LOG_ELEMENT(conf, string);
+    }
 
+out:
+    STACK_WIND(frame, trace_access_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->access, loc, mask, xdata);
+    return 0;
+}
 
-int
-trace_flush (call_frame_t *frame, xlator_t *this, fd_t *fd)
+int32_t
+trace_rchecksum(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                int32_t len, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_FLUSH].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (*fd=%p)",
-                        frame->root->unique, fd);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_WIND (frame, trace_flush_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->flush,
-                    fd);
-        return 0;
-}
+    conf = this->private;
 
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_RCHECKSUM].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s offset=%" PRId64 "len=%u fd=%p",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), offset, len,
+                 fd);
 
-int
-trace_fsync (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags)
-{
-        if (trace_fop_names[GF_FOP_FSYNC].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (flags=%d, *fd=%p)",
-                        frame->root->unique, flags, fd);
-        }
+        frame->local = fd->inode->gfid;
 
-        STACK_WIND (frame, trace_fsync_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fsync,
-                    fd, flags);
-        return 0;
-}
+        LOG_ELEMENT(conf, string);
+    }
 
+out:
+    STACK_WIND(frame, trace_rchecksum_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rchecksum, fd, offset, len, xdata);
 
-int
-trace_setxattr (call_frame_t *frame, xlator_t *this,
-                loc_t *loc, dict_t *dict, int32_t flags)
+    return 0;
+}
+
+int32_t
+trace_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+               fd_t *fd, const char *basename, entrylk_cmd cmd,
+               entrylk_type type, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_SETXATTR].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (loc {path=%s, ino=%"PRIu64"}, dict=%p, flags=%d)",
-                        frame->root->unique, loc->path,
-                        ((loc->inode)? loc->inode->ino : 0), dict, flags);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_WIND (frame, trace_setxattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->setxattr,
-                    loc, dict, flags);
-        return 0;
-}
+    conf = this->private;
 
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FENTRYLK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s volume=%s, (fd=%p "
+                 "basename=%s, cmd=%s, type=%s)",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), volume, fd,
+                 basename,
+                 ((cmd == ENTRYLK_LOCK) ? "ENTRYLK_LOCK" : "ENTRYLK_UNLOCK"),
+                 ((type == ENTRYLK_RDLCK) ? "ENTRYLK_RDLCK" : "ENTRYLK_WRLCK"));
 
-int
-trace_getxattr (call_frame_t *frame, xlator_t *this,
-                loc_t *loc, const char *name)
-{
-        if (trace_fop_names[GF_FOP_GETXATTR].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (loc {path=%s, ino=%"PRIu64"}), name=%s",
-                        frame->root->unique, loc->path,
-                        ((loc->inode)? loc->inode->ino : 0), name);
-        }
+        frame->local = fd->inode->gfid;
 
-        STACK_WIND (frame, trace_getxattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->getxattr,
-                    loc, name);
-        return 0;
-}
+        LOG_ELEMENT(conf, string);
+    }
 
+out:
+    STACK_WIND(frame, trace_fentrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fentrylk, volume, fd, basename, cmd,
+               type, xdata);
+    return 0;
+}
 
-int
-trace_removexattr (call_frame_t *frame, xlator_t *this,
-                   loc_t *loc, const char *name)
+int32_t
+trace_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+                dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_REMOVEXATTR].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (loc {path=%s, ino=%"PRIu64"}, name=%s)",
-                        frame->root->unique, loc->path,
-                        ((loc->inode)? loc->inode->ino : 0), name);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_WIND (frame, trace_removexattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->removexattr,
-                    loc, name);
+    conf = this->private;
 
-        return 0;
-}
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FGETXATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s fd=%p name=%s",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd, name);
 
+        frame->local = fd->inode->gfid;
 
-int
-trace_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd)
-{
-        if (trace_fop_names[GF_FOP_OPENDIR].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64":( loc {path=%s, ino=%"PRIu64"}, fd=%p)",
-                        frame->root->unique, loc->path, loc->inode->ino, fd);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
-        STACK_WIND (frame, trace_opendir_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->opendir,
-                    loc, fd);
-        return 0;
+out:
+    STACK_WIND(frame, trace_fgetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fgetxattr, fd, name, xdata);
+    return 0;
 }
 
-int
-trace_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-                off_t offset)
+int32_t
+trace_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+                int32_t flags, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_READDIRP].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (fd=%p, size=%"GF_PRI_SIZET", offset=%"PRId64")",
-                        frame->root->unique, fd, size, offset);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_WIND (frame, trace_readdirp_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->readdirp,
-                    fd, size, offset);
+    conf = this->private;
 
-        return 0;
-}
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FSETXATTR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s fd=%p flags=%d",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd, flags);
+
+        frame->local = fd->inode->gfid;
 
+        LOG_ELEMENT(conf, string);
+    }
+
+out:
+    STACK_WIND(frame, trace_fsetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
+    return 0;
+}
 
 int
-trace_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd,
-               size_t size, off_t offset)
+trace_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_READDIR].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (fd=%p, size=%"GF_PRI_SIZET", offset=%"PRId64")",
-                        frame->root->unique, fd, size, offset);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_WIND (frame, trace_readdir_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->readdir,
-                    fd, size, offset);
+    conf = this->private;
 
-        return 0;
-}
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FTRUNCATE].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64 ": gfid=%s offset=%" PRId64 " fd=%p",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), offset, fd);
 
+        frame->local = fd->inode->gfid;
 
-int
-trace_fsyncdir (call_frame_t *frame, xlator_t *this,
-                fd_t *fd, int32_t datasync)
-{
-        if (trace_fop_names[GF_FOP_FSYNCDIR].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (datasync=%d, *fd=%p)",
-                        frame->root->unique, datasync, fd);
-        }
+        LOG_ELEMENT(conf, string);
+    }
 
-        STACK_WIND (frame, trace_fsyncdir_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fsyncdir,
-                    fd, datasync);
-        return 0;
-}
+out:
+    STACK_WIND(frame, trace_ftruncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
 
+    return 0;
+}
 
 int
-trace_access (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask)
+trace_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_ACCESS].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (*loc {path=%s, ino=%"PRIu64"}, mask=0%o)",
-                        frame->root->unique, loc->path,
-                        ((loc->inode)? loc->inode->ino : 0), mask);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_WIND (frame, trace_access_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->access,
-                    loc, mask);
-        return 0;
-}
+    conf = this->private;
 
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_FSTAT].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "%" PRId64 ": gfid=%s fd=%p",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd);
 
-int
-trace_ftruncate (call_frame_t *frame, xlator_t *this,
-                 fd_t *fd, off_t offset)
-{
-        if (trace_fop_names[GF_FOP_FTRUNCATE].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (offset=%"PRId64", *fd=%p)",
-                        frame->root->unique, offset, fd);
-        }
+        frame->local = fd->inode->gfid;
 
-        STACK_WIND (frame, trace_ftruncate_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->ftruncate,
-                    fd, offset);
+        LOG_ELEMENT(conf, string);
+    }
 
-        return 0;
+out:
+    STACK_WIND(frame, trace_fstat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fstat, fd, xdata);
+    return 0;
 }
 
-
 int
-trace_fstat (call_frame_t *frame, xlator_t *this, fd_t *fd)
+trace_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+         struct gf_flock *lock, dict_t *xdata)
 {
-        if (trace_fop_names[GF_FOP_FSTAT].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (*fd=%p)",
-                        frame->root->unique, fd);
-        }
+    trace_conf_t *conf = NULL;
+
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_LK].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string),
+                 "%" PRId64
+                 ": gfid=%s fd=%p, cmd=%d, "
+                 "lock {l_type=%d, "
+                 "l_whence=%d, l_start=%" PRId64
+                 ", "
+                 "l_len=%" PRId64 ", l_pid=%u})",
+                 frame->root->unique, uuid_utoa(fd->inode->gfid), fd, cmd,
+                 lock->l_type, lock->l_whence, lock->l_start, lock->l_len,
+                 lock->l_pid);
+
+        frame->local = fd->inode->gfid;
 
-        STACK_WIND (frame, trace_fstat_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fstat,
-                    fd);
-        return 0;
+        LOG_ELEMENT(conf, string);
+    }
+
+out:
+    STACK_WIND(frame, trace_lk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lk, fd, cmd, lock, xdata);
+    return 0;
 }
 
+int32_t
+trace_forget(xlator_t *this, inode_t *inode)
+{
+    trace_conf_t *conf = NULL;
 
-int
-trace_lk (call_frame_t *frame, xlator_t *this, fd_t *fd,
-          int32_t cmd, struct gf_flock *lock)
+    conf = this->private;
+    /* If user want to understand when a lookup happens,
+       he should know about 'forget' too */
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_LOOKUP].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "gfid=%s", uuid_utoa(inode->gfid));
+
+        LOG_ELEMENT(conf, string);
+    }
+
+out:
+    return 0;
+}
+
+int32_t
+trace_releasedir(xlator_t *this, fd_t *fd)
 {
-        if (trace_fop_names[GF_FOP_LK].enabled) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "%"PRId64": (*fd=%p, cmd=%d, lock {l_type=%d, l_whence=%d, "
-                        "l_start=%"PRId64", l_len=%"PRId64", l_pid=%u})",
-                        frame->root->unique, fd, cmd, lock->l_type, lock->l_whence,
-                        lock->l_start, lock->l_len, lock->l_pid);
-        }
+    trace_conf_t *conf = NULL;
 
-        STACK_WIND (frame, trace_lk_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->lk,
-                    fd, cmd, lock);
-        return 0;
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_OPENDIR].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "gfid=%s fd=%p",
+                 uuid_utoa(fd->inode->gfid), fd);
+
+        LOG_ELEMENT(conf, string);
+    }
+
+out:
+    return 0;
 }
 
+int32_t
+trace_release(xlator_t *this, fd_t *fd)
+{
+    trace_conf_t *conf = NULL;
 
+    conf = this->private;
+
+    if (!conf->log_file && !conf->log_history)
+        goto out;
+    if (trace_fop_names[GF_FOP_OPEN].enabled ||
+        trace_fop_names[GF_FOP_CREATE].enabled) {
+        char string[4096] = {
+            0,
+        };
+        snprintf(string, sizeof(string), "gfid=%s fd=%p",
+                 uuid_utoa(fd->inode->gfid), fd);
+
+        LOG_ELEMENT(conf, string);
+    }
+
+out:
+    return 0;
+}
 
 void
-enable_all_calls (int enabled)
+enable_all_calls(int enabled)
 {
-        int i;
+    int i;
 
-        for (i = 0; i < GF_FOP_MAXVALUE; i++)
-                trace_fop_names[i].enabled = enabled;
+    for (i = 0; i < GF_FOP_MAXVALUE; i++)
+        trace_fop_names[i].enabled = enabled;
 }
 
-
 void
-enable_call (const char *name, int enabled)
+enable_call(const char *name, int enabled)
 {
-        int i;
-        for (i = 0; i < GF_FOP_MAXVALUE; i++)
-                if (!strcasecmp(trace_fop_names[i].name, name))
-                        trace_fop_names[i].enabled = enabled;
+    int i;
+    for (i = 0; i < GF_FOP_MAXVALUE; i++)
+        if (!strcasecmp(trace_fop_names[i].name, name))
+            trace_fop_names[i].enabled = enabled;
 }
 
-
 /*
   include = 1 for "include-ops"
   = 0 for "exclude-ops"
 */
 void
-process_call_list (const char *list, int include)
+process_call_list(const char *list, int include)
 {
-        enable_all_calls (include ? 0 : 1);
+    enable_all_calls(include ? 0 : 1);
 
-        char *call = strsep ((char **)&list, ",");
+    char *call = strsep((char **)&list, ",");
 
-        while (call) {
-                enable_call (call, include);
-                call = strsep ((char **)&list, ",");
-        }
+    while (call) {
+        enable_call(call, include);
+        call = strsep((char **)&list, ",");
+    }
 }
 
+int32_t
+trace_dump_history(xlator_t *this)
+{
+    int ret = -1;
+    char key_prefix[GF_DUMP_MAX_BUF_LEN] = {
+        0,
+    };
+    trace_conf_t *conf = NULL;
+
+    GF_VALIDATE_OR_GOTO("trace", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->history, out);
+
+    conf = this->private;
+    // Is it ok to return silently if log-history option his off?
+    if (conf && conf->log_history == _gf_true) {
+        gf_proc_dump_build_key(key_prefix, "xlator.debug.trace", "history");
+        gf_proc_dump_add_section("%s", key_prefix);
+        eh_dump(this->history, NULL, dump_history_trace);
+    }
+    ret = 0;
+
+out:
+    return ret;
+}
 
 int32_t
-init (xlator_t *this)
+mem_acct_init(xlator_t *this)
 {
-        dict_t *options = NULL;
-        char *includes = NULL, *excludes = NULL;
-        char *forced_loglevel = NULL;
+    int ret = -1;
 
-        if (!this)
-                return -1;
+    if (!this)
+        return ret;
 
-        if (!this->children || this->children->next) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "trace translator requires one subvolume");
-                return -1;
-        }
-        if (!this->parents) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "dangling volume. check volfile ");
-        }
+    ret = xlator_mem_acct_init(this, gf_trace_mt_end + 1);
 
+    if (ret != 0) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Memory accounting init"
+               " failed");
+        return ret;
+    }
 
-        options = this->options;
-        includes = data_to_str (dict_get (options, "include-ops"));
-        excludes = data_to_str (dict_get (options, "exclude-ops"));
+    return ret;
+}
 
-        {
-                int i;
-                for (i = 0; i < GF_FOP_MAXVALUE; i++) {
-                        trace_fop_names[i].name = (gf_fop_list[i] ?
-                                                   gf_fop_list[i] : ":O");
-                        trace_fop_names[i].enabled = 1;
-                }
-        }
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    int32_t ret = -1;
+    trace_conf_t *conf = NULL;
+    char *includes = NULL, *excludes = NULL;
 
-        if (includes && excludes) {
-                gf_log (this->name,
-                        GF_LOG_ERROR,
-                        "must specify only one of 'include-ops' and 'exclude-ops'");
-                return -1;
-        }
-        if (includes)
-                process_call_list (includes, 1);
-        if (excludes)
-                process_call_list (excludes, 0);
-
-        if (dict_get (options, "force-log-level")) {
-                forced_loglevel = data_to_str (dict_get (options,
-                                               "force-log-level"));
-                if (!forced_loglevel)
-                        goto setloglevel;
-
-                if (strcmp (forced_loglevel, "NORMAL") == 0)
-                        trace_log_level = GF_LOG_NORMAL;
-                else if (strcmp (forced_loglevel, "TRACE") == 0)
-                        trace_log_level = GF_LOG_TRACE;
-                else if (strcmp (forced_loglevel, "ERROR") == 0)
-                        trace_log_level = GF_LOG_ERROR;
-                else if (strcmp (forced_loglevel, "DEBUG") == 0)
-                        trace_log_level = GF_LOG_DEBUG;
-                else if (strcmp (forced_loglevel, "WARNING") == 0)
-                        trace_log_level = GF_LOG_WARNING;
-                else if (strcmp (forced_loglevel, "CRITICAL") == 0)
-                        trace_log_level = GF_LOG_CRITICAL;
-                else if (strcmp (forced_loglevel, "NONE") == 0)
-                        trace_log_level = GF_LOG_NONE;
+    GF_VALIDATE_OR_GOTO("quick-read", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+    GF_VALIDATE_OR_GOTO(this->name, options, out);
+
+    conf = this->private;
+
+    includes = data_to_str(dict_get(options, "include-ops"));
+    excludes = data_to_str(dict_get(options, "exclude-ops"));
+
+    {
+        int i;
+        for (i = 0; i < GF_FOP_MAXVALUE; i++) {
+            if (gf_fop_list[i])
+                strncpy(trace_fop_names[i].name, gf_fop_list[i],
+                        sizeof(trace_fop_names[i].name));
+            else
+                strncpy(trace_fop_names[i].name, ":0",
+                        sizeof(trace_fop_names[i].name));
+            trace_fop_names[i].enabled = 1;
+            trace_fop_names[i].name[sizeof(trace_fop_names[i].name) - 1] = 0;
         }
+    }
+
+    if (includes && excludes) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "must specify only one of 'include-ops' and "
+               "'exclude-ops'");
+        goto out;
+    }
+
+    if (includes)
+        process_call_list(includes, 1);
+    if (excludes)
+        process_call_list(excludes, 0);
+
+    /* Should resizing of the event-history be allowed in reconfigure?
+     * for which a new event_history might have to be allocated and the
+     * older history has to be freed.
+     */
+    GF_OPTION_RECONF("log-file", conf->log_file, options, bool, out);
+
+    GF_OPTION_RECONF("log-history", conf->log_history, options, bool, out);
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    dict_t *options = NULL;
+    char *includes = NULL, *excludes = NULL;
+    char *forced_loglevel = NULL;
+    eh_t *history = NULL;
+    int ret = -1;
+    uint64_t history_size = TRACE_DEFAULT_HISTORY_SIZE;
+    trace_conf_t *conf = NULL;
+
+    if (!this)
+        return -1;
+
+    if (!this->children || this->children->next) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "trace translator requires one subvolume");
+        return -1;
+    }
+    if (!this->parents) {
+        gf_log(this->name, GF_LOG_WARNING, "dangling volume. check volfile ");
+    }
+
+    conf = GF_CALLOC(1, sizeof(trace_conf_t), gf_trace_mt_trace_conf_t);
+    if (!conf) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "cannot allocate "
+               "xl->private");
+        return -1;
+    }
+
+    options = this->options;
+    includes = data_to_str(dict_get(options, "include-ops"));
+    excludes = data_to_str(dict_get(options, "exclude-ops"));
+
+    {
+        int i;
+        for (i = 0; i < GF_FOP_MAXVALUE; i++) {
+            if (gf_fop_list[i])
+                strncpy(trace_fop_names[i].name, gf_fop_list[i],
+                        sizeof(trace_fop_names[i].name));
+            else
+                strncpy(trace_fop_names[i].name, ":O",
+                        sizeof(trace_fop_names[i].name));
+            trace_fop_names[i].enabled = 1;
+            trace_fop_names[i].name[sizeof(trace_fop_names[i].name) - 1] = 0;
+        }
+    }
+
+    if (includes && excludes) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "must specify only one of 'include-ops' and "
+               "'exclude-ops'");
+        return -1;
+    }
+
+    if (includes)
+        process_call_list(includes, 1);
+    if (excludes)
+        process_call_list(excludes, 0);
+
+    GF_OPTION_INIT("history-size", history_size, size, out);
+    conf->history_size = history_size;
+
+    gf_log(this->name, GF_LOG_INFO, "history size %" PRIu64, history_size);
+
+    GF_OPTION_INIT("log-file", conf->log_file, bool, out);
+
+    gf_log(this->name, GF_LOG_INFO, "logging to file %s",
+           (conf->log_file == _gf_true) ? "enabled" : "disabled");
+
+    GF_OPTION_INIT("log-history", conf->log_history, bool, out);
+
+    gf_log(this->name, GF_LOG_DEBUG, "logging to history %s",
+           (conf->log_history == _gf_true) ? "enabled" : "disabled");
+
+    history = eh_new(history_size, _gf_false, NULL);
+    if (!history) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "event history cannot be "
+               "initialized");
+        return -1;
+    }
+
+    this->history = history;
+
+    conf->trace_log_level = GF_LOG_INFO;
+
+    if (dict_get(options, "force-log-level")) {
+        forced_loglevel = data_to_str(dict_get(options, "force-log-level"));
+        if (!forced_loglevel)
+            goto setloglevel;
+
+        if (strcmp(forced_loglevel, "INFO") == 0)
+            conf->trace_log_level = GF_LOG_INFO;
+        else if (strcmp(forced_loglevel, "TRACE") == 0)
+            conf->trace_log_level = GF_LOG_TRACE;
+        else if (strcmp(forced_loglevel, "ERROR") == 0)
+            conf->trace_log_level = GF_LOG_ERROR;
+        else if (strcmp(forced_loglevel, "DEBUG") == 0)
+            conf->trace_log_level = GF_LOG_DEBUG;
+        else if (strcmp(forced_loglevel, "WARNING") == 0)
+            conf->trace_log_level = GF_LOG_WARNING;
+        else if (strcmp(forced_loglevel, "CRITICAL") == 0)
+            conf->trace_log_level = GF_LOG_CRITICAL;
+        else if (strcmp(forced_loglevel, "NONE") == 0)
+            conf->trace_log_level = GF_LOG_NONE;
+    }
 
 setloglevel:
-        gf_log_set_loglevel (trace_log_level);
+    gf_log_set_loglevel(this->ctx, conf->trace_log_level);
+    this->private = conf;
+    ret = 0;
+out:
+    if (ret == -1) {
+        if (history)
+            GF_FREE(history);
+        if (conf)
+            GF_FREE(conf);
+    }
 
-        return 0;
+    return ret;
 }
 
 void
-fini (xlator_t *this)
+fini(xlator_t *this)
 {
-        if (!this)
-                return;
-
-        gf_log (this->name, GF_LOG_NORMAL,
-                "trace translator unloaded");
+    if (!this)
         return;
+
+    if (this->history)
+        eh_destroy(this->history);
+
+    gf_log(this->name, GF_LOG_INFO, "trace translator unloaded");
+    return;
 }
 
 struct xlator_fops fops = {
-        .stat        = trace_stat,
-        .readlink    = trace_readlink,
-        .mknod       = trace_mknod,
-        .mkdir       = trace_mkdir,
-        .unlink      = trace_unlink,
-        .rmdir       = trace_rmdir,
-        .symlink     = trace_symlink,
-        .rename      = trace_rename,
-        .link        = trace_link,
-        .truncate    = trace_truncate,
-        .open        = trace_open,
-        .readv       = trace_readv,
-        .writev      = trace_writev,
-        .statfs      = trace_statfs,
-        .flush       = trace_flush,
-        .fsync       = trace_fsync,
-        .setxattr    = trace_setxattr,
-        .getxattr    = trace_getxattr,
-        .removexattr = trace_removexattr,
-        .opendir     = trace_opendir,
-        .readdir     = trace_readdir,
-        .readdirp    = trace_readdirp,
-        .fsyncdir    = trace_fsyncdir,
-        .access      = trace_access,
-        .ftruncate   = trace_ftruncate,
-        .fstat       = trace_fstat,
-        .create      = trace_create,
-        .lk          = trace_lk,
-        .inodelk     = trace_inodelk,
-        .finodelk    = trace_finodelk,
-        .entrylk     = trace_entrylk,
-        .lookup      = trace_lookup,
-        .xattrop     = trace_xattrop,
-        .fxattrop    = trace_fxattrop,
-        .setattr     = trace_setattr,
-        .fsetattr    = trace_fsetattr,
+    .stat = trace_stat,
+    .readlink = trace_readlink,
+    .mknod = trace_mknod,
+    .mkdir = trace_mkdir,
+    .unlink = trace_unlink,
+    .rmdir = trace_rmdir,
+    .symlink = trace_symlink,
+    .rename = trace_rename,
+    .link = trace_link,
+    .truncate = trace_truncate,
+    .open = trace_open,
+    .readv = trace_readv,
+    .writev = trace_writev,
+    .statfs = trace_statfs,
+    .flush = trace_flush,
+    .fsync = trace_fsync,
+    .setxattr = trace_setxattr,
+    .getxattr = trace_getxattr,
+    .fsetxattr = trace_fsetxattr,
+    .fgetxattr = trace_fgetxattr,
+    .removexattr = trace_removexattr,
+    .opendir = trace_opendir,
+    .readdir = trace_readdir,
+    .readdirp = trace_readdirp,
+    .fsyncdir = trace_fsyncdir,
+    .access = trace_access,
+    .ftruncate = trace_ftruncate,
+    .fstat = trace_fstat,
+    .create = trace_create,
+    .lk = trace_lk,
+    .inodelk = trace_inodelk,
+    .finodelk = trace_finodelk,
+    .entrylk = trace_entrylk,
+    .fentrylk = trace_fentrylk,
+    .lookup = trace_lookup,
+    .rchecksum = trace_rchecksum,
+    .xattrop = trace_xattrop,
+    .fxattrop = trace_fxattrop,
+    .setattr = trace_setattr,
+    .fsetattr = trace_fsetattr,
+    .seek = trace_seek,
 };
 
-
 struct xlator_cbks cbks = {
+    .release = trace_release,
+    .releasedir = trace_releasedir,
+    .forget = trace_forget,
 };
 
 struct volume_options options[] = {
-        { .key  = {"include-ops", "include"},
-          .type = GF_OPTION_TYPE_STR,
-          /*.value = { ""} */
-        },
-        { .key  = {"exclude-ops", "exclude"},
-          .type = GF_OPTION_TYPE_STR
-          /*.value = { ""} */
-        },
-        { .key  = {NULL} },
+    {
+        .key = {"include-ops", "include"},
+        .type = GF_OPTION_TYPE_STR,
+        /*.value = { ""} */
+    },
+    {
+        .key = {"exclude-ops", "exclude"},
+        .type = GF_OPTION_TYPE_STR
+        /*.value = { ""} */
+    },
+    {
+        .key = {"history-size"},
+        .type = GF_OPTION_TYPE_SIZET,
+        .default_value = "1024",
+    },
+    {
+        .key = {"log-file"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "no",
+    },
+    {
+        .key = {"log-history"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "no",
+    },
+    {.key = {NULL}},
+};
+
+struct xlator_dumpops dumpops = {.history = trace_dump_history};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1},
+    .dumpops = &dumpops,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "trace",
+    .category = GF_TECH_PREVIEW,
 };
diff --git a/xlators/debug/trace/src/trace.h b/xlators/debug/trace/src/trace.h
new file mode 100644
index 00000000000..b16304799da
--- /dev/null
+++ b/xlators/debug/trace/src/trace.h
@@ -0,0 +1,55 @@
+/*
+   Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <time.h>
+#include <errno.h>
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/event-history.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/circ-buff.h>
+#include <glusterfs/statedump.h>
+#include <glusterfs/options.h>
+
+#define TRACE_DEFAULT_HISTORY_SIZE 1024
+
+typedef struct {
+    /* Since the longest fop name is fremovexattr i.e 12 characters, array size
+     * is kept 24, i.e double of the maximum.
+     */
+    char name[24];
+    int enabled;
+} trace_fop_name_t;
+
+trace_fop_name_t trace_fop_names[GF_FOP_MAXVALUE];
+
+typedef struct {
+    gf_boolean_t log_file;
+    gf_boolean_t log_history;
+    uint64_t history_size;
+    int trace_log_level;
+} trace_conf_t;
+
+#define TRACE_STACK_UNWIND(op, frame, params...)                               \
+    do {                                                                       \
+        frame->local = NULL;                                                   \
+        STACK_UNWIND_STRICT(op, frame, params);                                \
+    } while (0);
+
+#define LOG_ELEMENT(_conf, _string)                                            \
+    do {                                                                       \
+        if (_conf) {                                                           \
+            if ((_conf->log_history) == _gf_true)                              \
+                gf_log_eh("%s", _string);                                      \
+            if ((_conf->log_file) == _gf_true)                                 \
+                gf_log(THIS->name, _conf->trace_log_level, "%s", _string);     \
+        }                                                                      \
+    } while (0);
diff --git a/xlators/encryption/Makefile.am b/xlators/encryption/Makefile.am
deleted file mode 100644
index 2cbde680fac..00000000000
--- a/xlators/encryption/Makefile.am
+++ /dev/null
@@ -1,3 +0,0 @@
-SUBDIRS = rot-13
-
-CLEANFILES = 
diff --git a/xlators/encryption/rot-13/src/rot-13.c b/xlators/encryption/rot-13/src/rot-13.c
deleted file mode 100644
index 98bfd1b5aa4..00000000000
--- a/xlators/encryption/rot-13/src/rot-13.c
+++ /dev/null
@@ -1,205 +0,0 @@
-/*
-  Copyright (c) 2006-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-#include <ctype.h>
-#include <sys/uio.h>
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "xlator.h"
-#include "logging.h"
-
-#include "rot-13.h"
-
-/*
- * This is a rot13 ``encryption'' xlator. It rot13's data when 
- * writing to disk and rot13's it back when reading it. 
- * This xlator is meant as an example, NOT FOR PRODUCTION
- * USE ;) (hence no error-checking)
- */
-
-void 
-rot13 (char *buf, int len)
-{
-	int i;
-	for (i = 0; i < len; i++) {
-		if (buf[i] >= 'a' && buf[i] <= 'z')
-			buf[i] = 'a' + ((buf[i] - 'a' + 13) % 26);
-		else if (buf[i] >= 'A' && buf[i] <= 'Z')
-			buf[i] = 'A' + ((buf[i] - 'A' + 13) % 26);
-	}
-}
-
-void
-rot13_iovec (struct iovec *vector, int count)
-{
-	int i;
-	for (i = 0; i < count; i++) {
-		rot13 (vector[i].iov_base, vector[i].iov_len);
-	}
-}
-
-int32_t
-rot13_readv_cbk (call_frame_t *frame,
-                 void *cookie,
-                 xlator_t *this,
-                 int32_t op_ret,
-                 int32_t op_errno,
-                 struct iovec *vector,
-                 int32_t count,
-		 struct iatt *stbuf,
-                 struct iobref *iobref)
-{
-	rot_13_private_t *priv = (rot_13_private_t *)this->private;
-  
-	if (priv->decrypt_read)
-		rot13_iovec (vector, count);
-
-	STACK_UNWIND_STRICT (readv, frame, op_ret, op_errno, vector, count, stbuf, iobref);
-	return 0;
-}
-
-int32_t
-rot13_readv (call_frame_t *frame,
-             xlator_t *this,
-             fd_t *fd,
-             size_t size,
-             off_t offset)
-{
-	STACK_WIND (frame,
-		    rot13_readv_cbk,
-		    FIRST_CHILD (this),
-		    FIRST_CHILD (this)->fops->readv,
-		    fd, size, offset);
-	return 0;
-}
-
-int32_t
-rot13_writev_cbk (call_frame_t *frame,
-                  void *cookie,
-                  xlator_t *this,
-                  int32_t op_ret,
-                  int32_t op_errno,
-                  struct iatt *prebuf,
-		  struct iatt *postbuf)
-{
-	STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, prebuf, postbuf);
-	return 0;
-}
-
-int32_t
-rot13_writev (call_frame_t *frame,
-              xlator_t *this,
-              fd_t *fd,
-              struct iovec *vector,
-              int32_t count, 
-              off_t offset,
-              struct iobref *iobref)
-{
-	rot_13_private_t *priv = (rot_13_private_t *)this->private;
-	if (priv->encrypt_write)
-		rot13_iovec (vector, count);
-
-	STACK_WIND (frame, 
-		    rot13_writev_cbk,
-		    FIRST_CHILD (this),
-		    FIRST_CHILD (this)->fops->writev,
-		    fd, vector, count, offset,
-                    iobref);
-	return 0;
-}
-
-int32_t
-init (xlator_t *this)
-{
-	data_t *data = NULL;
-	rot_13_private_t *priv = NULL;
-
-	if (!this->children || this->children->next) {
-		gf_log ("rot13", GF_LOG_ERROR, 
-			"FATAL: rot13 should have exactly one child");
-		return -1;
-	}
-
-	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"dangling volume. check volfile ");
-	}
-
-	priv = GF_CALLOC (sizeof (rot_13_private_t), 1, 0);
-        if (!priv)
-                return -1;
-
-	priv->decrypt_read = 1;
-	priv->encrypt_write = 1;
-
-	data = dict_get (this->options, "encrypt-write");
-	if (data) {
-		if (gf_string2boolean (data->data, &priv->encrypt_write) == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"encrypt-write takes only boolean options");
-			return -1;
-		}
-	}
-
-	data = dict_get (this->options, "decrypt-read");
-	if (data) {
-		if (gf_string2boolean (data->data, &priv->decrypt_read) == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"decrypt-read takes only boolean options");
-			return -1;
-		}
-	}
-
-	this->private = priv;
-	gf_log ("rot13", GF_LOG_DEBUG, "rot13 xlator loaded");
-	return 0;
-}
-
-void
-fini (xlator_t *this)
-{
-	rot_13_private_t *priv = this->private;
-
-	GF_FREE (priv);
-
-	return;
-}
-
-struct xlator_fops fops = {
-	.readv        = rot13_readv,
-	.writev       = rot13_writev
-};
-
-struct xlator_cbks cbks = {
-};
-
-struct volume_options options[] = {
-	{ .key  = {"encrypt-write"}, 
-	  .type = GF_OPTION_TYPE_BOOL
-	},
-	{ .key  = {"decrypt-read"}, 
-	  .type = GF_OPTION_TYPE_BOOL 
-	},
-	{ .key  = {NULL} },
-};
diff --git a/xlators/encryption/rot-13/src/rot-13.h b/xlators/encryption/rot-13/src/rot-13.h
deleted file mode 100644
index 1fee3e44db2..00000000000
--- a/xlators/encryption/rot-13/src/rot-13.h
+++ /dev/null
@@ -1,33 +0,0 @@
-/*
-   Copyright (c) 2006-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef __ROT_13_H__
-#define __ROT_13_H__
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-typedef struct {
-	gf_boolean_t encrypt_write;
-	gf_boolean_t decrypt_read;
-} rot_13_private_t;
-
-#endif /* __ROT_13_H__ */
diff --git a/xlators/features/Makefile.am b/xlators/features/Makefile.am
index 8b23b939b0e..c57897f11ea 100644
--- a/xlators/features/Makefile.am
+++ b/xlators/features/Makefile.am
@@ -1,3 +1,14 @@
-SUBDIRS = locks trash quota read-only access-control mac-compat quiesce #path-converter # filter
+if BUILD_CLOUDSYNC
+  CLOUDSYNC_DIR = cloudsync
+endif
 
-CLEANFILES = 
+if BUILD_METADISP
+  METADISP_DIR = metadisp
+endif
+
+SUBDIRS = locks quota read-only quiesce marker index barrier arbiter upcall \
+	compress changelog gfid-access snapview-client snapview-server trash \
+	shard bit-rot leases selinux sdfs namespace $(CLOUDSYNC_DIR) thin-arbiter \
+	utime $(METADISP_DIR)
+
+CLEANFILES =
diff --git a/xlators/features/access-control/src/Makefile.am b/xlators/features/access-control/src/Makefile.am
deleted file mode 100644
index 6ab8cc4ec4e..00000000000
--- a/xlators/features/access-control/src/Makefile.am
+++ /dev/null
@@ -1,13 +0,0 @@
-xlator_LTLIBRARIES = access-control.la
-xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
-access_control_la_LDFLAGS = -module -avoidversion
-access_control_la_SOURCES = access-control.c
-access_control_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
-
-noinst_HEADERS = access-control.h
-
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS)\
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)\
-	-L$(xlatordir)/
-
-CLEANFILES =
diff --git a/xlators/features/access-control/src/access-control.c b/xlators/features/access-control/src/access-control.c
deleted file mode 100644
index 7fb0651215c..00000000000
--- a/xlators/features/access-control/src/access-control.c
+++ /dev/null
@@ -1,1862 +0,0 @@
-/*
-  Copyright (c) 2010 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "access-control.h"
-#include "xlator.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "iatt.h"
-
-/* Careful, this function erases the stub from frame->local. Dont call this if
- * a subsequent callback requires retaining access to the stub. This should be
- * called at the end of all access-control related operations, i.e. once the
- * frame will be handed off to the actual fop and the next callback that will
- * be called is the default callback. IOW, the function where call_resume is
- * called.
- * NOTE: this is required because FRAME_DESTROY tries to free frame->local if
- * it finds it to be non-NULL.
- */
-call_stub_t *
-__get_frame_stub (call_frame_t *fr)
-{
-        call_stub_t     *st = NULL;
-
-        if (!fr)
-                return NULL;
-
-        st = fr->local;
-        fr->local = NULL;
-
-        return st;
-}
-
-
-int
-ac_test_owner_access (struct iatt *ia, uid_t uid, int accesstest)
-{
-        int     ret = -1;
-
-        if (!ia)
-                return -1;
-
-        /* First test permissions using the uid. */
-        if (ia->ia_uid != uid) {
-                ret = -1;
-                goto out;
-        }
-
-        /* At this point we know, the uid matches that of the stat structure, so
-         * if the caller does not care, we should return success.
-         */
-        if (ac_test_dontcare (accesstest)) {
-                ret = 0;
-                goto out;
-        }
-
-        if (ac_test_read (accesstest))
-                ret = IA_PROT_RUSR (ia->ia_prot);
-
-        if (ac_test_write (accesstest))
-                ret = IA_PROT_WUSR (ia->ia_prot);
-
-        if (ac_test_exec (accesstest))
-                ret = IA_PROT_XUSR (ia->ia_prot);
-
-        /* For failed access test for owner, we need to return EACCES */
-        if (!ret)
-                ret = -1;
-        else
-                ret = 0;
-out:
-        return ret;
-}
-
-
-int
-ac_test_group_access (struct iatt *ia, gid_t gid, gid_t *auxgids, int auxcount,
-                      int accesstest)
-{
-        int     ret = -1;
-        int     testgid = -1;
-        int     x = 0;
-
-        if (!ia)
-                return -1;
-        /* First, determine which gid to test against. This will be determined
-         * by first checking which of the gids given to us match the gid in the
-         * stat. If none match, then we go to checking with others as the user.
-         */
-
-        /* If we are only given the primary gid. Dont depend on @auxgids
-         * being NULL since I know users of this function can pass statically
-         * allocated arrays which cant be NULL and yet contain no valid gids.
-         */
-
-        if ((ia->ia_gid != gid) && (auxcount == 0)) {
-                ret = -1;
-                goto out;
-        }
-
-        if (ia->ia_gid == gid)
-                testgid = gid;
-        else {
-                for (; x < auxcount; ++x) {
-                        if (ia->ia_gid == auxgids[x]) {
-                                testgid = ia->ia_gid;
-                                break;
-                        }
-                }
-        }
-
-        /* None of the gids match with the gid in the stat. */
-        if (testgid == -1) {
-                ret = -1;
-                goto out;
-        }
-
-        /* At this point, at least one gid matches that in the stat, now we must
-         * check whether the caller is interested in the access check at all.
-         */
-        if (ac_test_dontcare (accesstest)) {
-                ret = 0;
-                goto out;
-        }
-
-        if (ac_test_read (accesstest))
-                ret = IA_PROT_RGRP (ia->ia_prot);
-
-        if (ac_test_write (accesstest))
-                ret = IA_PROT_WGRP (ia->ia_prot);
-
-        if (ac_test_exec (accesstest))
-                ret = IA_PROT_XGRP (ia->ia_prot);
-
-        if (!ret)
-                ret = -1;
-        else
-                ret = 0;
-
-out:
-        return ret;
-}
-
-
-int
-ac_test_other_access (struct iatt *ia, int accesstest)
-{
-        int     ret = 0;
-
-        if (!ia)
-                return -1;
-
-        if (ac_test_read (accesstest))
-                ret = IA_PROT_ROTH (ia->ia_prot);
-
-        if (ac_test_write (accesstest))
-                ret = IA_PROT_WOTH (ia->ia_prot);
-
-        if (ac_test_exec (accesstest))
-                ret = IA_PROT_XOTH (ia->ia_prot);
-
-        if (!ret)
-                ret = -1;
-        else
-                ret = 0;
-
-        return ret;
-}
-
-
-/* Returns -1 on a failed access test with @operrno set to the relevant error
- * number.
- */
-int
-ac_test_access (struct iatt *ia, uid_t uid, gid_t gid, gid_t *auxgids,
-                int auxcount, int accesstest, int testwho, int *operrno)
-{
-        int             ret = -1;
-
-        if ((!ia) || (!operrno))
-                return -1;
-
-        if ((uid == 0) && (gid == 0)) {
-                gf_log (ACTRL, GF_LOG_TRACE, "Root has access");
-                return 0;
-        }
-
-        if (ac_test_owner (testwho)) {
-                gf_log (ACTRL, GF_LOG_TRACE, "Testing owner access");
-                ret = ac_test_owner_access (ia, uid, accesstest);
-        }
-
-        if (ret == 0) {
-                gf_log (ACTRL, GF_LOG_TRACE, "Owner has access");
-                goto out;
-        }
-
-        if (ac_test_group (testwho)) {
-                gf_log (ACTRL, GF_LOG_TRACE, "Testing group access");
-                ret = ac_test_group_access (ia, gid, auxgids, auxcount,
-                                            accesstest);
-        }
-
-        if (ret == 0) {
-                gf_log (ACTRL, GF_LOG_TRACE, "Group has access");
-                goto out;
-        }
-
-        if (ac_test_other (testwho)) {
-                gf_log (ACTRL, GF_LOG_TRACE, "Testing other access");
-                ret = ac_test_other_access (ia, accesstest);
-        }
-
-        if (ret == 0)
-                gf_log (ACTRL, GF_LOG_TRACE, "Other has access");
-out:
-        if (ret == -1) {
-                gf_log (ACTRL, GF_LOG_TRACE, "No access allowed");
-                *operrno = EPERM;
-        }
-
-        return ret;
-}
-
-
-int
-ac_loc_fill (loc_t *loc, inode_t *inode, inode_t *parent, char *path)
-{
-        int     ret = -EFAULT;
-
-        if (!loc)
-                return ret;
-
-        if (inode) {
-                loc->inode = inode_ref (inode);
-                loc->ino = inode->ino;
-        }
-
-        if (parent)
-                loc->parent = inode_ref (parent);
-
-        loc->path = gf_strdup (path);
-        if (!loc->path) {
-                gf_log (ACTRL, GF_LOG_ERROR, "strdup failed");
-                goto loc_wipe;
-        }
-
-        loc->name = strrchr (loc->path, '/');
-        if (loc->name)
-                loc->name++;
-        else
-                goto loc_wipe;
-
-        ret = 0;
-loc_wipe:
-        if (ret < 0)
-                loc_wipe (loc);
-
-        return ret;
-}
-
-
-int
-ac_inode_loc_fill (inode_t *inode, loc_t *loc)
-{
-        char            *resolvedpath = NULL;
-        inode_t         *parent = NULL;
-        int             ret = -EFAULT;
-
-        if ((!inode) || (!loc))
-                return ret;
-
-        if ((inode) && (inode->ino == 1))
-                goto ignore_parent;
-
-        parent = inode_parent (inode, 0, NULL);
-        if (!parent)
-                goto err;
-
-ignore_parent:
-        ret = inode_path (inode, NULL, &resolvedpath);
-        if (ret < 0)
-                goto err;
-
-        ret = ac_loc_fill (loc, inode, parent, resolvedpath);
-        if (ret < 0)
-                goto err;
-
-err:
-        if (parent)
-                inode_unref (parent);
-
-        if (resolvedpath)
-                GF_FREE (resolvedpath);
-
-        return ret;
-}
-
-
-int
-ac_parent_loc_fill (loc_t *parentloc, loc_t *childloc)
-{
-        if ((!parentloc) || (!childloc))
-                return -1;
-
-        return ac_inode_loc_fill (childloc->parent, parentloc);
-}
-
-
-int32_t
-ac_truncate_resume (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                    off_t offset)
-{
-        STACK_WIND (frame, default_truncate_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->truncate, loc, offset);
-        return 0;
-}
-
-
-int32_t
-ac_truncate_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        call_stub_t     *stub = NULL;
-
-        stub = __get_frame_stub (frame);
-        if (op_ret == -1)
-                goto out;
-
-        op_ret = ac_test_access (buf, frame->root->uid, frame->root->gid,
-                                 frame->root->groups, frame->root->ngrps,
-                                 ACCTEST_WRITE, ACCTEST_ANY, &op_errno);
-        if (op_ret == -1)
-                goto out;
-
-        call_resume (stub);
-out:
-        if (op_ret < 0) {
-                STACK_UNWIND_STRICT (truncate, frame, -1, op_errno, NULL, NULL);
-                if (stub)
-                        call_stub_destroy (stub);
-        }
-
-        return 0;
-}
-
-
-int32_t
-ac_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset)
-{
-        call_stub_t     *stub = NULL;
-        int             ret = -EFAULT;
-
-        stub = fop_truncate_stub (frame, ac_truncate_resume, loc, offset);
-	if (!stub) {
-		gf_log (this->name, GF_LOG_ERROR, "cannot create call stub: "
-                        "(out of memory)");
-                ret = -ENOMEM;
-                goto out;
-	}
-
-        frame->local = stub;
-        STACK_WIND (frame, ac_truncate_stat_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->stat, loc);
-
-        ret = 0;
-out:
-        if (ret < 0)
-                STACK_UNWIND_STRICT (truncate, frame, -1, -ret, NULL, NULL);
-
-        return 0;
-}
-
-
-int32_t
-ac_access_resume (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask)
-{
-        STACK_WIND (frame, default_access_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->access, loc, mask);
-        return 0;
-}
-
-
-int32_t
-ac_access_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        call_stub_t     *stub = NULL;
-        int32_t         mask = 0;
-        int             acctest = 0;
-
-        stub = __get_frame_stub (frame);
-        mask = stub->args.access.mask;
-
-        /* If mask requests test for file existence then do not
-         * return a failure with ENOENT, instead return a failed
-         * access test.
-         */
-        if (op_ret == -1) {
-                if (mask & F_OK)
-                        op_errno = EACCES;
-                else
-                        op_errno = errno;
-
-                goto out;
-        }
-
-        if (R_OK & mask)
-                acctest |= ACCTEST_READ;
-        else if (W_OK & mask)
-                acctest |= ACCTEST_WRITE;
-        else if (X_OK & mask)
-                acctest |= ACCTEST_EXEC;
-        else
-                acctest = 0;
-
-        op_ret = ac_test_access (buf, frame->root->uid, frame->root->gid,
-                                 frame->root->groups, frame->root->ngrps,
-                                 acctest, ACCTEST_ANY, &op_errno);
-        if (op_ret == -1)
-                goto out;
-
-        call_resume (stub);
-out:
-        if (op_ret < 0) {
-                STACK_UNWIND_STRICT (access, frame, -1, op_errno);
-                if (stub)
-                        call_stub_destroy (stub);
-        }
-
-        return 0;
-}
-
-
-int32_t
-ac_access (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask)
-{
-        call_stub_t     *stub = NULL;
-        int             ret = -EFAULT;
-
-        stub = fop_access_stub (frame, ac_access_resume, loc, mask);
-	if (!stub) {
-		gf_log (this->name, GF_LOG_ERROR, "cannot create call stub: "
-                        "(out of memory)");
-                ret = -ENOMEM;
-                goto out;
-	}
-
-        frame->local = stub;
-        STACK_WIND (frame, ac_access_stat_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->stat, loc);
-        ret = 0;
-
-out:
-        if (ret < 0)
-                STACK_UNWIND_STRICT (access, frame, -1, -ret);
-
-        return 0;
-}
-
-
-int32_t
-ac_readlink_resume (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                    size_t size)
-{
-        STACK_WIND (frame, default_readlink_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->readlink, loc, size);
-        return 0;
-}
-
-
-int32_t
-ac_readlink_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        call_stub_t     *stub = NULL;
-
-        stub = __get_frame_stub (frame);
-        if (op_ret == -1)
-                goto out;
-
-        op_ret = ac_test_access (buf, frame->root->uid, frame->root->gid,
-                                 frame->root->groups, frame->root->ngrps,
-                                 ACCTEST_READ, ACCTEST_ANY, &op_errno);
-        if (op_ret == -1)
-                goto out;
-
-        call_resume (stub);
-out:
-        if (op_ret < 0) {
-                STACK_UNWIND_STRICT (readlink, frame, -1, op_errno, NULL, NULL);
-                if (stub)
-                        call_stub_destroy (stub);
-        }
-
-        return 0;
-}
-
-
-int32_t
-ac_readlink (call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size)
-{
-        call_stub_t     *stub = NULL;
-        int             ret = -EFAULT;
-
-        stub = fop_readlink_stub (frame, ac_readlink_resume, loc, size);
-	if (!stub) {
-		gf_log (this->name, GF_LOG_ERROR, "cannot create call stub: "
-                        "(out of memory)");
-                ret = -ENOMEM;
-                goto out;
-	}
-
-        frame->local = stub;
-        STACK_WIND (frame, ac_readlink_stat_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->stat, loc);
-        ret = 0;
-
-out:
-        if (ret < 0)
-                STACK_UNWIND_STRICT (readlink, frame, -1, -ret, NULL, NULL);
-
-        return 0;
-}
-
-
-int
-ac_mknod_resume (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
-                 dev_t rdev, dict_t *params)
-{
-        STACK_WIND (frame, default_mknod_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->mknod, loc, mode, rdev, params);
-        return 0;
-}
-
-
-int32_t
-ac_mknod_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        call_stub_t     *stub = NULL;
-
-        stub = __get_frame_stub (frame);
-        if (op_ret == -1)
-                goto out;
-
-        op_ret = ac_test_access (buf, frame->root->uid, frame->root->gid,
-                                 frame->root->groups, frame->root->ngrps,
-                                 ACCTEST_WRITE, ACCTEST_ANY, &op_errno);
-        if (op_ret == -1) {
-                op_errno = EACCES;
-                goto out;
-        }
-
-        call_resume (stub);
-out:
-        if (op_ret < 0) {
-                STACK_UNWIND_STRICT (mknod, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL);
-                if (stub)
-                        call_stub_destroy (stub);
-        }
-
-        return 0;
-}
-
-
-int
-ac_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
-          dev_t rdev, dict_t *params)
-{
-        call_stub_t     *stub = NULL;
-        int             ret = -EFAULT;
-        loc_t           parentloc = {0, };
-
-        stub = fop_mknod_stub (frame, ac_mknod_resume, loc, mode, rdev, params);
-	if (!stub) {
-		gf_log (this->name, GF_LOG_ERROR, "cannot create call stub: "
-                        "(out of memory)");
-                ret = -ENOMEM;
-                goto out;
-	}
-
-        frame->local = stub;
-        ret = ac_parent_loc_fill (&parentloc, loc);
-        if (ret < 0)
-                goto out;
-
-        STACK_WIND (frame, ac_mknod_stat_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->stat, &parentloc);
-        loc_wipe (&parentloc);
-        ret = 0;
-
-out:
-        if (ret < 0) {
-                /* Erase any stored frame before unwinding. */
-                stub = __get_frame_stub (frame);
-                if (stub)
-                        call_stub_destroy (stub);
-                STACK_UNWIND_STRICT (mknod, frame, -1, -ret, NULL, NULL, NULL,
-                                     NULL);
-        }
-
-        return 0;
-}
-
-
-int
-ac_mkdir_resume (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
-                 dict_t *params)
-{
-        STACK_WIND (frame, default_mkdir_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->mkdir, loc, mode, params);
-        return 0;
-}
-
-
-int32_t
-ac_mkdir_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        call_stub_t     *stub = NULL;
-
-        stub = __get_frame_stub (frame);
-        if (op_ret == -1)
-                goto out;
-
-        op_ret = ac_test_access (buf, frame->root->uid, frame->root->gid,
-                                 frame->root->groups, frame->root->ngrps,
-                                 ACCTEST_WRITE, ACCTEST_ANY, &op_errno);
-        if (op_ret == -1) {
-                /* On a failed  write test on parent dir, we need to return
-                 * EACCES, not EPERM that is returned by default by
-                 * ac_test_access.
-                 */
-                op_errno = EACCES;
-                goto out;
-        }
-
-        call_resume (stub);
-out:
-        if (op_ret < 0) {
-                STACK_UNWIND_STRICT (mkdir, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL);
-                if (stub)
-                        call_stub_destroy (stub);
-        }
-
-        return 0;
-}
-
-
-int
-ac_mkdir (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
-          dict_t *params)
-{
-        call_stub_t     *stub = NULL;
-        int             ret = -EFAULT;
-        loc_t           parentloc = {0, };
-
-        stub = fop_mkdir_stub (frame, ac_mkdir_resume, loc, mode, params);
-	if (!stub) {
-		gf_log (this->name, GF_LOG_ERROR, "cannot create call stub: "
-                        "(out of memory)");
-                ret = -ENOMEM;
-                goto out;
-	}
-
-        frame->local = stub;
-        ret = ac_parent_loc_fill (&parentloc, loc);
-        if (ret < 0)
-                goto out;
-
-        STACK_WIND (frame, ac_mkdir_stat_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->stat, &parentloc);
-        loc_wipe (&parentloc);
-        ret = 0;
-
-out:
-        if (ret < 0) {
-                /* Erase the stored stub before unwinding. */
-                stub = __get_frame_stub (frame);
-                if (stub)
-                        call_stub_destroy (stub);
-                STACK_UNWIND_STRICT (mkdir, frame, -1, -ret, NULL, NULL, NULL,
-                                     NULL);
-        }
-
-        return 0;
-}
-
-
-int32_t
-ac_unlink_resume (call_frame_t *frame, xlator_t *this, loc_t *loc)
-{
-        STACK_WIND (frame, default_unlink_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->unlink, loc);
-        return 0;
-}
-
-
-int32_t
-ac_unlink_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        call_stub_t     *stub = NULL;
-
-        stub = __get_frame_stub (frame);
-        if (op_ret == -1)
-                goto out;
-
-        op_ret = ac_test_access (buf, frame->root->uid, frame->root->gid,
-                                 frame->root->groups, frame->root->ngrps,
-                                 ACCTEST_WRITE, ACCTEST_ANY, &op_errno);
-        if (op_ret == -1) {
-                op_errno = EACCES;
-                goto out;
-        }
-
-        call_resume (stub);
-out:
-        if (op_ret < 0) {
-                STACK_UNWIND_STRICT (unlink, frame, -1, op_errno, NULL, NULL);
-                if (stub)
-                        call_stub_destroy (stub);
-        }
-
-        return 0;
-}
-
-
-int32_t
-ac_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc)
-{
-        call_stub_t     *stub = NULL;
-        int             ret = -EFAULT;
-        loc_t           parentloc = {0, };
-
-        stub = fop_unlink_stub (frame, ac_unlink_resume, loc);
-	if (!stub) {
-		gf_log (this->name, GF_LOG_ERROR, "cannot create call stub: "
-                        "(out of memory)");
-                ret = -ENOMEM;
-                goto out;
-	}
-
-        frame->local = stub;
-        ret = ac_parent_loc_fill (&parentloc, loc);
-        if (ret < 0)
-                goto out;
-
-        STACK_WIND (frame, ac_unlink_stat_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->stat, &parentloc);
-        loc_wipe (&parentloc);
-        ret = 0;
-
-out:
-        if (ret < 0) {
-                /* Erase the stored stub before unwinding. */
-                stub = __get_frame_stub (frame);
-                if (stub)
-                        call_stub_destroy (stub);
-                STACK_UNWIND_STRICT (unlink, frame, -1, -ret, NULL, NULL);
-        }
-
-        return 0;
-}
-
-
-int
-ac_rmdir_resume (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags)
-{
-        STACK_WIND (frame, default_rmdir_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->rmdir, loc, flags);
-        return 0;
-}
-
-
-int32_t
-ac_rmdir_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        call_stub_t     *stub = NULL;
-
-        stub = __get_frame_stub (frame);
-        if (op_ret == -1)
-                goto out;
-
-        op_ret = ac_test_access (buf, frame->root->uid, frame->root->gid,
-                                 frame->root->groups, frame->root->ngrps,
-                                 ACCTEST_WRITE, ACCTEST_ANY, &op_errno);
-        if (op_ret == -1) {
-                op_errno = EACCES;
-                goto out;
-        }
-
-        call_resume (stub);
-out:
-        if (op_ret < 0) {
-                STACK_UNWIND_STRICT (rmdir, frame, -1, op_errno, NULL, NULL);
-                if (stub)
-                        call_stub_destroy (stub);
-        }
-
-        return 0;
-}
-
-
-int
-ac_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags)
-{
-        call_stub_t     *stub = NULL;
-        int             ret = -EFAULT;
-        loc_t           parentloc = {0, };
-
-        stub = fop_rmdir_stub (frame, ac_rmdir_resume, loc, flags);
-	if (!stub) {
-		gf_log (this->name, GF_LOG_ERROR, "cannot create call stub: "
-                        "(out of memory)");
-                ret = -ENOMEM;
-                goto out;
-	}
-
-        frame->local = stub;
-        ret = ac_parent_loc_fill (&parentloc, loc);
-        if (ret < 0)
-                goto out;
-
-        STACK_WIND (frame, ac_rmdir_stat_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->stat, &parentloc);
-        loc_wipe (&parentloc);
-        ret = 0;
-
-out:
-        if (ret < 0) {
-                /* Erase the stored stub before unwinding. */
-                stub = __get_frame_stub (frame);
-                if (stub)
-                        call_stub_destroy (stub);
-                STACK_UNWIND_STRICT (rmdir, frame, -1, -ret, NULL, NULL);
-        }
-
-        return 0;
-}
-
-
-int32_t
-ac_symlink_resume (call_frame_t *frame, xlator_t *this, const char *linkname,
-                   loc_t *loc, dict_t *params)
-{
-        STACK_WIND (frame, default_symlink_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->symlink, linkname, loc, params);
-        return 0;
-}
-
-
-int32_t
-ac_symlink_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        call_stub_t     *stub = NULL;
-
-        stub = __get_frame_stub (frame);
-        if (op_ret == -1)
-                goto out;
-
-        op_ret = ac_test_access (buf, frame->root->uid, frame->root->gid,
-                                 frame->root->groups, frame->root->ngrps,
-                                 ACCTEST_WRITE, ACCTEST_ANY, &op_errno);
-        if (op_ret == -1) {
-                op_errno = EACCES;
-                goto out;
-        }
-
-        call_resume (stub);
-out:
-        if (op_ret < 0) {
-                STACK_UNWIND_STRICT (symlink, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL);
-                if (stub)
-                        call_stub_destroy (stub);
-        }
-
-        return 0;
-}
-
-
-int
-ac_symlink (call_frame_t *frame, xlator_t *this, const char *linkname,
-            loc_t *loc, dict_t *params)
-{
-        call_stub_t     *stub = NULL;
-        int             ret = -EFAULT;
-        loc_t           parentloc = {0, };
-
-        stub = fop_symlink_stub (frame, ac_symlink_resume, linkname, loc,
-                                 params);
-	if (!stub) {
-		gf_log (this->name, GF_LOG_ERROR, "cannot create call stub: "
-                        "(out of memory)");
-                ret = -ENOMEM;
-                goto out;
-	}
-
-        frame->local = stub;
-        ret = ac_parent_loc_fill (&parentloc, loc);
-        if (ret < 0)
-                goto out;
-
-        STACK_WIND (frame, ac_symlink_stat_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->stat, &parentloc);
-        loc_wipe (&parentloc);
-        ret = 0;
-
-out:
-        if (ret < 0) {
-                /* Erase the stored stub before unwinding. */
-                stub = __get_frame_stub (frame);
-                if (stub)
-                        call_stub_destroy (stub);
-                STACK_UNWIND_STRICT (symlink, frame, -1, -ret, NULL, NULL, NULL,
-                                     NULL);
-        }
-
-        return 0;
-}
-
-
-int32_t
-ac_rename_resume (call_frame_t *frame, xlator_t *this, loc_t *oldloc,
-                  loc_t *newloc)
-{
-        STACK_WIND (frame, default_rename_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->rename, oldloc, newloc);
-        return 0;
-}
-
-
-int32_t
-ac_rename_dst_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        call_stub_t     *stub = NULL;
-
-        stub = __get_frame_stub (frame);
-        if (op_ret == -1)
-                goto out;
-
-        op_ret = ac_test_access (buf, frame->root->uid,
-                                 frame->root->gid, frame->root->groups,
-                                 frame->root->ngrps, ACCTEST_WRITE,
-                                 ACCTEST_ANY, &op_errno);
-        if (op_ret == -1) {
-                op_errno = EACCES;
-               goto out;
-        }
-
-        call_resume (stub);
-out:
-        if (op_ret < 0) {
-                STACK_UNWIND_STRICT (rename, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL, NULL);
-                if (stub)
-                        call_stub_destroy (stub);
-        }
-
-        return 0;
-}
-
-
-int32_t
-ac_rename_src_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        call_stub_t     *stub = NULL;
-        loc_t           parentloc = {0, };
-
-        stub = frame->local;
-        if (op_ret == -1)
-                goto out;
-
-        op_ret = ac_test_access (buf, frame->root->uid,
-                                 frame->root->gid, frame->root->groups,
-                                 frame->root->ngrps, ACCTEST_WRITE,
-                                 ACCTEST_ANY, &op_errno);
-        if (op_ret == -1) {
-                op_errno = EACCES;
-                goto out;
-        }
-
-        op_ret = ac_parent_loc_fill (&parentloc, &stub->args.rename.new);
-        if (op_ret < 0) {
-                op_errno = -EFAULT;
-                goto out;
-        }
-
-        STACK_WIND (frame, ac_rename_dst_stat_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->stat, &parentloc);
-        loc_wipe (&parentloc);
-
-out:
-        if (op_ret < 0) {
-                /* Erase the stored stub before unwinding. */
-                stub = __get_frame_stub (frame);
-                if (stub)
-                        call_stub_destroy (stub);
-                STACK_UNWIND_STRICT (rename, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL, NULL);
-        }
-
-        return 0;
-}
-
-
-int32_t
-ac_rename (call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc)
-{
-        call_stub_t     *stub = NULL;
-        int             ret = -EFAULT;
-        loc_t           parentloc = {0, };
-
-        stub = fop_rename_stub (frame, ac_rename_resume, oldloc, newloc);
-	if (!stub) {
-		gf_log (this->name, GF_LOG_ERROR, "cannot create call stub: "
-                        "(out of memory)");
-                ret = -ENOMEM;
-                goto out;
-	}
-
-        frame->local = stub;
-        ret = ac_parent_loc_fill (&parentloc, oldloc);
-        if (ret < 0)
-                goto out;
-
-        STACK_WIND (frame, ac_rename_src_stat_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->stat, &parentloc);
-        loc_wipe (&parentloc);
-        ret = 0;
-
-out:
-        if (ret < 0) {
-                /* Erase the stored stub before unwinding. */
-                stub = __get_frame_stub (frame);
-                if (stub)
-                        call_stub_destroy (stub);
-                STACK_UNWIND_STRICT (rename, frame, -1, -ret, NULL, NULL, NULL,
-                                     NULL, NULL);
-        }
-
-        return 0;
-}
-
-
-int32_t
-ac_link_resume (call_frame_t *frame, xlator_t *this, loc_t *oldloc,
-                loc_t *newloc)
-{
-        STACK_WIND (frame, default_link_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->link, oldloc, newloc);
-        return 0;
-}
-
-
-int32_t
-ac_link_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        call_stub_t     *stub = NULL;
-
-        stub = __get_frame_stub (frame);
-        if (op_ret == -1)
-                goto out;
-
-        op_ret = ac_test_access (buf, frame->root->uid, frame->root->gid,
-                                 frame->root->groups, frame->root->ngrps,
-                                 ACCTEST_WRITE, ACCTEST_ANY, &op_errno);
-        if (op_ret == -1) {
-                /* By default ac_test_access sets the op_errno to EPERM
-                 * but in the case of link, we need to return EACCES to meet
-                 * posix requirements when a write permission is not available
-                 * for the new directory.
-                 */
-                op_errno = EACCES;
-                goto out;
-        }
-
-        call_resume (stub);
-out:
-        if (op_ret < 0) {
-                STACK_UNWIND_STRICT (link, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL);
-                if (stub)
-                        call_stub_destroy (stub);
-        }
-
-        return 0;
-}
-
-
-int32_t
-ac_link (call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc)
-{
-        call_stub_t     *stub = NULL;
-        int             ret = -EFAULT;
-        loc_t           parentloc = {0, };
-
-        stub = fop_link_stub (frame, ac_link_resume, oldloc, newloc);
-	if (!stub) {
-		gf_log (this->name, GF_LOG_ERROR, "cannot create call stub: "
-                        "(out of memory)");
-                ret = -ENOMEM;
-                goto out;
-	}
-
-        frame->local = stub;
-        ret = ac_parent_loc_fill (&parentloc, newloc);
-        if (ret < 0)
-                goto out;
-
-        STACK_WIND (frame, ac_link_stat_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->stat, &parentloc);
-        loc_wipe (&parentloc);
-        ret = 0;
-
-out:
-        if (ret < 0) {
-                /* Erase the stored stub before unwinding. */
-                stub = __get_frame_stub (frame);
-                if (stub)
-                        call_stub_destroy (stub);
-                STACK_UNWIND_STRICT (link, frame, -1, -ret, NULL, NULL, NULL,
-                                     NULL);
-        }
-
-        return 0;
-}
-
-
-int32_t
-ac_create_resume (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                  int32_t flags, mode_t mode, fd_t *fd, dict_t *params)
-{
-        STACK_WIND (frame, default_create_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->create, loc, flags, mode,
-                    fd, params);
-        return 0;
-}
-
-
-int32_t
-ac_create_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        call_stub_t     *stub = NULL;
-
-        stub = __get_frame_stub (frame);
-        if (op_ret == -1)
-                goto out;
-
-        op_ret = ac_test_access (buf, frame->root->uid, frame->root->gid,
-                                 frame->root->groups, frame->root->ngrps,
-                                 ACCTEST_WRITE, ACCTEST_ANY, &op_errno);
-         if (op_ret == -1) {
-                op_errno = EACCES;
-                goto out;
-         }
-
-        call_resume (stub);
-out:
-        if (op_ret < 0) {
-                STACK_UNWIND_STRICT (create, frame, -1, op_errno, NULL, NULL,
-                                     NULL, NULL, NULL);
-                if (stub)
-                        call_stub_destroy (stub);
-        }
-
-        return 0;
-}
-
-
-int32_t
-ac_create (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
-           mode_t mode, fd_t *fd, dict_t *params)
-{
-        call_stub_t     *stub = NULL;
-        int             ret = -EFAULT;
-        loc_t           parentloc = {0, };
-
-        stub = fop_create_stub (frame, ac_create_resume, loc, flags, mode,
-                                fd, params);
-	if (!stub) {
-		gf_log (this->name, GF_LOG_ERROR, "cannot create call stub: "
-                        "(out of memory)");
-                ret = -ENOMEM;
-                goto out;
-	}
-
-        frame->local = stub;
-        ret = ac_parent_loc_fill (&parentloc, loc);
-        if (ret < 0)
-                goto out;
-
-        STACK_WIND (frame, ac_create_stat_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->stat, &parentloc);
-        loc_wipe (&parentloc);
-        ret = 0;
-
-out:
-        if (ret < 0) {
-                /* Erase the stored stub before unwinding. */
-                stub = __get_frame_stub (frame);
-                if (stub)
-                        call_stub_destroy (stub);
-                STACK_UNWIND_STRICT (create, frame, -1, -ret, NULL, NULL, NULL,
-                                     NULL, NULL);
-        }
-
-        return 0;
-}
-
-
-int32_t
-ac_open_resume (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
-                fd_t *fd, int32_t wbflags)
-{
-        STACK_WIND (frame, default_open_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->open, loc, flags, fd, wbflags);
-        return 0;
-}
-
-
-int32_t
-ac_open_create_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        call_stub_t     *stub = NULL;
-
-        stub = __get_frame_stub (frame);
-        if (op_ret == -1)
-                goto out;
-
-        op_ret = ac_test_access (buf, frame->root->uid, frame->root->gid,
-                                 frame->root->groups, frame->root->ngrps,
-                                 ACCTEST_WRITE, ACCTEST_ANY, &op_errno);
-         if (op_ret == -1) {
-                op_errno = EACCES;
-                goto out;
-         }
-
-        call_resume (stub);
-out:
-        if (op_ret < 0) {
-                STACK_UNWIND_STRICT (open, frame, -1, op_errno, NULL);
-                if (stub)
-                        call_stub_destroy (stub);
-        }
-
-        return 0;
-}
-
-
-int
-ac_open_create (call_stub_t *stub)
-{
-        int             ret = -EFAULT;
-        loc_t           parentloc = {0, };
-        xlator_t        *this = NULL;
-
-        if (!stub)
-                return ret;
-
-        ret = ac_parent_loc_fill (&parentloc, &stub->args.open.loc);
-        if (ret < 0)
-                goto out;
-
-        this = stub->frame->this;
-        STACK_WIND (stub->frame, ac_open_create_stat_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->stat, &parentloc);
-        loc_wipe (&parentloc);
-        ret = 0;
-
-out:
-        return ret;
-}
-
-
-int32_t
-ac_open_only_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        call_stub_t     *stub = NULL;
-        int             acctest = 0;
-        int32_t         flags = 0;
-
-        stub = __get_frame_stub (frame);
-        if (op_ret == -1)
-                goto out;
-
-        flags = stub->args.open.flags;
-        /* The permissions we test for depend on how the open needs to be
-         * performed. */
-        if ((flags & O_ACCMODE) == O_RDONLY)
-                acctest = ACCTEST_READ;
-        else if (((flags & O_ACCMODE) == O_RDWR) ||
-                 ((flags & O_ACCMODE) == O_WRONLY))
-                acctest = ACCTEST_WRITE;
-
-        op_ret = ac_test_access (buf, frame->root->uid, frame->root->gid,
-                                 frame->root->groups, frame->root->ngrps,
-                                 acctest, ACCTEST_ANY, &op_errno);
-        if (op_ret == -1)
-                goto out;
-
-        call_resume (stub);
-out:
-        if (op_ret < 0) {
-                STACK_UNWIND_STRICT (open, frame, -1, op_errno, NULL);
-                if (stub)
-                        call_stub_destroy (stub);
-        }
-
-        return 0;
-}
-
-
-int
-ac_open_only (call_stub_t *stub)
-{
-        int             ret = -EFAULT;
-        xlator_t        *this = NULL;
-
-        if (!stub)
-                return ret;
-
-        this = stub->frame->this;
-        STACK_WIND (stub->frame, ac_open_only_stat_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->stat, &stub->args.open.loc);
-        return 0;
-}
-
-int32_t
-ac_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
-         fd_t *fd, int32_t wbflags)
-{
-        call_stub_t     *stub = NULL;
-        int             ret = -EFAULT;
-
-        stub = fop_open_stub (frame, ac_open_resume, loc, flags, fd, wbflags);
-	if (!stub) {
-		gf_log (this->name, GF_LOG_ERROR, "cannot create call stub: "
-                        "(out of memory)");
-                ret = -ENOMEM;
-                goto out;
-	}
-
-        frame->local = stub;
-        /* If we are not supposed to create the file then there is no need to
-         * check the parent dir permissions. */
-        if (flags & O_CREAT)
-                ret = ac_open_create (stub);
-        else
-                ret = ac_open_only (stub);
-
-out:
-        if (ret < 0) {
-                /* Erase the stored stub before unwinding. */
-                stub = __get_frame_stub (frame);
-                if (stub)
-                        call_stub_destroy (stub);
-                STACK_UNWIND_STRICT (open, frame, -1, -ret, NULL);
-        }
-
-        return 0;
-}
-
-
-int32_t
-ac_readv_resume (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-                 off_t offset)
-{
-        STACK_WIND (frame, default_readv_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->readv, fd, size, offset);
-        return 0;
-}
-
-
-int32_t
-ac_readv_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                    int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        call_stub_t     *stub = NULL;
-
-        stub = __get_frame_stub (frame);
-        if (op_ret == -1)
-                goto out;
-
-        op_ret = ac_test_access (buf, frame->root->uid, frame->root->gid,
-                                 frame->root->groups, frame->root->ngrps,
-                                 ACCTEST_READ, ACCTEST_ANY, &op_errno);
-        if (op_ret == -1) {
-                op_errno = EACCES;
-                goto out;
-        }
-
-        call_resume (stub);
-out:
-        if (op_ret < 0) {
-                STACK_UNWIND_STRICT (readv, frame, -1, op_errno, NULL, 0, NULL,
-                                     NULL);
-                if (stub)
-                        call_stub_destroy (stub);
-        }
-
-        return 0;
-}
-
-
-int32_t
-ac_readv (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
-          off_t offset)
-{
-        call_stub_t     *stub = NULL;
-        int             ret = -EFAULT;
-
-        stub = fop_readv_stub (frame, ac_readv_resume, fd, size, offset);
-	if (!stub) {
-		gf_log (this->name, GF_LOG_ERROR, "cannot create call stub: "
-                        "(out of memory)");
-                ret = -ENOMEM;
-                goto out;
-	}
-
-        frame->local = stub;
-        STACK_WIND (frame, ac_readv_fstat_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->fstat, fd);
-        ret = 0;
-
-out:
-        if (ret < 0)
-                STACK_UNWIND_STRICT (readv, frame, -1, -ret, NULL, 0, NULL,
-                                     NULL);
-
-        return 0;
-}
-
-
-int32_t
-ac_writev_resume (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                  struct iovec *vector, int32_t count, off_t offset,
-                  struct iobref *iobref)
-{
-        STACK_WIND (frame, default_writev_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->writev, fd, vector, count, offset,
-                    iobref);
-        return 0;
-}
-
-
-int32_t
-ac_writev_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        call_stub_t     *stub = NULL;
-
-        stub = __get_frame_stub (frame);
-        if (op_ret == -1)
-                goto out;
-
-        op_ret = ac_test_access (buf, frame->root->uid, frame->root->gid,
-                                 frame->root->groups, frame->root->ngrps,
-                                 ACCTEST_WRITE, ACCTEST_ANY, &op_errno);
-        if (op_ret == -1) {
-                op_errno = EACCES;
-                goto out;
-        }
-
-        call_resume (stub);
-out:
-        if (op_ret < 0) {
-                STACK_UNWIND_STRICT (writev, frame, -1, op_errno, NULL, NULL);
-                if (stub)
-                        call_stub_destroy (stub);
-        }
-
-        return 0;
-}
-
-
-int32_t
-ac_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
-           int32_t count, off_t offset, struct iobref *iobref)
-{
-        call_stub_t     *stub = NULL;
-        int             ret = -EFAULT;
-
-        stub = fop_writev_stub (frame, ac_writev_resume, fd, vector, count,
-                                offset, iobref);
-	if (!stub) {
-		gf_log (this->name, GF_LOG_ERROR, "cannot create call stub: "
-                        "(out of memory)");
-                ret = -ENOMEM;
-                goto out;
-	}
-
-        frame->local = stub;
-        STACK_WIND (frame, ac_writev_fstat_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->fstat, fd);
-        ret = 0;
-
-out:
-        if (ret < 0)
-                STACK_UNWIND_STRICT (writev, frame, -1, -ret, NULL, NULL);
-
-        return 0;
-}
-
-
-int32_t
-ac_opendir_resume (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd)
-{
-        STACK_WIND (frame, default_opendir_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->opendir, loc, fd);
-        return 0;
-}
-
-
-int32_t
-ac_opendir_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        call_stub_t     *stub = NULL;
-
-        stub = __get_frame_stub (frame);
-        if (op_ret == -1)
-                goto out;
-
-        op_ret = ac_test_access (buf, frame->root->uid, frame->root->gid,
-                                 frame->root->groups, frame->root->ngrps,
-                                 ACCTEST_READ, ACCTEST_ANY, &op_errno);
-        if (op_ret == -1)
-                goto out;
-
-        call_resume (stub);
-out:
-        if (op_ret < 0) {
-                STACK_UNWIND_STRICT (opendir, frame, -1, op_errno, NULL);
-                if (stub)
-                        call_stub_destroy (stub);
-        }
-
-        return 0;
-}
-
-int32_t
-ac_opendir (call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd)
-{
-        call_stub_t     *stub = NULL;
-        int             ret = -EFAULT;
-
-        stub = fop_opendir_stub (frame, ac_opendir_resume, loc, fd);
-	if (!stub) {
-		gf_log (this->name, GF_LOG_ERROR, "cannot create call stub: "
-                        "(out of memory)");
-                ret = -ENOMEM;
-                goto out;
-	}
-
-        frame->local = stub;
-        STACK_WIND (frame, ac_opendir_stat_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->stat, loc);
-        ret = 0;
-
-out:
-        if (ret < 0)
-                STACK_UNWIND_STRICT (opendir, frame, -1, -ret, NULL);
-
-        return 0;
-}
-
-
-int32_t
-ac_setattr_resume (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                   struct iatt *buf, int32_t valid)
-{
-        STACK_WIND (frame, default_setattr_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->setattr, loc, buf, valid);
-        return 0;
-}
-
-
-int32_t
-ac_setattr_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                     int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        call_stub_t     *stub = NULL;
-        int32_t         valid = 0;
-        struct iatt     *setbuf = NULL;
-
-        stub = __get_frame_stub (frame);
-        if (op_ret == -1)
-                goto out;
-
-        op_ret = ac_test_access (buf, frame->root->uid, frame->root->gid,
-                                 frame->root->groups, frame->root->ngrps,
-                                 ACCTEST_DONTCARE, ACCTEST_OWNER,
-                                 &op_errno);
-        if (op_ret == -1)
-                goto out;
-
-        valid = stub->args.setattr.valid;
-        setbuf = &stub->args.setattr.stbuf;
-        if (gf_attr_uid_set (valid) || gf_attr_gid_set (valid)) {
-                /* chown returns EPERM if the operation would change the
-                 * ownership, but the effective user ID is not the
-                 * super-user and the process is not an owner of the file.
-                 * Ref: posix-testsuite/chown/07.t
-                 */
-                if ((frame->root->uid != 0) && (gf_attr_uid_set (valid))) {
-                        if (buf->ia_uid != setbuf->ia_uid) {
-                                op_ret = -1;
-                                op_errno = EPERM;
-                                goto out;
-                        }
-                }
-
-                /* non-super-user can modify file group if he is owner of a
-                 * file and gid he is setting is in his groups list.
-                 * Ref: posix-testsuite/chown/00.t
-                 */
-                if ((frame->root->uid != 0) && (gf_attr_gid_set (valid))) {
-                        if (frame->root->uid != buf->ia_uid) {
-                                op_ret = -1;
-                                op_errno = EPERM;
-                                goto out;
-                        }
-
-                        op_ret = ac_test_access (setbuf, 0, frame->root->gid,
-                                                 frame->root->groups,
-                                                 frame->root->ngrps,
-                                                 ACCTEST_DONTCARE,
-                                                 ACCTEST_GROUP, &op_errno);
-                        if (op_ret == -1)
-                                goto out;
-                }
-        }
-
-        call_resume (stub);
-out:
-        if (op_ret < 0) {
-                STACK_UNWIND_STRICT (setattr, frame, -1, op_errno, NULL, NULL);
-                if (stub)
-                        call_stub_destroy (stub);
-        }
-
-        return 0;
-}
-
-int32_t
-ac_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *buf,
-            int32_t valid)
-{
-        call_stub_t     *stub = NULL;
-        int             ret = -EFAULT;
-
-        stub = fop_setattr_stub (frame, ac_setattr_resume, loc, buf, valid);
-	if (!stub) {
-		gf_log (this->name, GF_LOG_ERROR, "cannot create call stub: "
-                        "(out of memory)");
-                ret = -ENOMEM;
-                goto out;
-	}
-
-        frame->local = stub;
-        STACK_WIND (frame, ac_setattr_stat_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->stat, loc);
-        ret = 0;
-
-out:
-        if (ret < 0)
-                STACK_UNWIND_STRICT (setattr, frame, -1, -ret, NULL, NULL);
-
-        return 0;
-}
-
-
-int32_t
-ac_fsetattr_resume (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                    struct iatt *buf, int32_t valid)
-{
-        STACK_WIND (frame, default_fsetattr_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fsetattr, fd, buf, valid);
-        return 0;
-}
-
-
-int32_t
-ac_fsetattr_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, struct iatt *buf)
-{
-        call_stub_t     *stub = NULL;
-        int32_t         valid = 0;
-        struct iatt     *setbuf = NULL;
-
-        stub = __get_frame_stub (frame);
-        if (op_ret == -1)
-                goto out;
-
-        op_ret = ac_test_access (buf, frame->root->uid, frame->root->gid,
-                                 frame->root->groups, frame->root->ngrps,
-                                 ACCTEST_DONTCARE, ACCTEST_OWNER,
-                                 &op_errno);
-        if (op_ret == -1)
-                goto out;
-
-        valid = stub->args.fsetattr.valid;
-        setbuf = &stub->args.fsetattr.stbuf;
-        if (gf_attr_uid_set (valid) && gf_attr_gid_set (valid)) {
-                /* chown returns EPERM if the operation would change the
-                 * ownership, but the effective user ID is not the
-                 * super-user and the process is not an owner of the file.
-                 * Ref: posix-testsuite/chown/07.t
-                 */
-                if ((frame->root->uid != 0) && (gf_attr_uid_set (valid))) {
-                        if (buf->ia_uid != setbuf->ia_uid) {
-                                op_ret = -1;
-                                op_errno = EPERM;
-                                goto out;
-                        }
-                }
-
-                /* non-super-user can modify file group if he is owner of a
-                 * file and gid he is setting is in his groups list.
-                 * Ref: posix-testsuite/chown/00.t
-                 */
-                if ((frame->root->uid != 0) && (gf_attr_gid_set (valid))) {
-                        if (frame->root->uid != buf->ia_uid) {
-                                op_ret = -1;
-                                op_errno = EPERM;
-                                goto out;
-                        }
-
-                        op_ret = ac_test_access (buf, 0, frame->root->gid,
-                                                 frame->root->groups,
-                                                 frame->root->ngrps,
-                                                 ACCTEST_DONTCARE,
-                                                 ACCTEST_GROUP, &op_errno);
-                        if (op_ret == -1)
-                                goto out;
-                }
-        }
-
-        call_resume (stub);
-out:
-        if (op_ret < 0) {
-                STACK_UNWIND_STRICT (fsetattr, frame, -1, op_errno, NULL, NULL);
-                if (stub)
-                        call_stub_destroy (stub);
-        }
-
-        return 0;
-}
-
-
-int32_t
-ac_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *buf,
-             int32_t valid)
-{
-        call_stub_t     *stub = NULL;
-        int             ret = -EFAULT;
-
-        stub = fop_fsetattr_stub (frame, ac_fsetattr_resume, fd, buf, valid);
-	if (!stub) {
-		gf_log (this->name, GF_LOG_ERROR, "cannot create call stub: "
-                        "(out of memory)");
-                ret = -ENOMEM;
-                goto out;
-	}
-
-        frame->local = stub;
-        STACK_WIND (frame, ac_fsetattr_fstat_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->fstat, fd);
-        ret = 0;
-
-out:
-        if (ret < 0)
-                STACK_UNWIND_STRICT (fsetattr, frame, -1, -ret, NULL, NULL);
-
-        return 0;
-}
-
-
-struct xlator_fops fops = {
-        .truncate       = ac_truncate,
-        .access         = ac_access,
-        .readlink       = ac_readlink,
-        .mknod          = ac_mknod,
-        .mkdir          = ac_mkdir,
-        .unlink         = ac_unlink,
-        .rmdir          = ac_rmdir,
-        .symlink        = ac_symlink,
-        .rename         = ac_rename,
-        .link           = ac_link,
-        .create         = ac_create,
-        .open           = ac_open,
-        .readv          = ac_readv,
-        .writev         = ac_writev,
-        .opendir        = ac_opendir,
-        .setattr        = ac_setattr,
-        .fsetattr       = ac_fsetattr,
-};
-
-int
-init (xlator_t *this)
-{
-        int ret = -1;
-
-	if (!this->children || this->children->next) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"FATAL: access-control not configured with "
-                        "exactly one  child");
-                goto out;
-	}
-
-	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"dangling volume. check volfile ");
-	}
-
-        ret = 0;
-out:
-        return ret;
-}
-
-void
-fini (xlator_t *this)
-{
-        return;
-}
-
-struct xlator_cbks cbks = {
-};
diff --git a/xlators/features/access-control/src/access-control.h b/xlators/features/access-control/src/access-control.h
deleted file mode 100644
index 8e2d6b6b966..00000000000
--- a/xlators/features/access-control/src/access-control.h
+++ /dev/null
@@ -1,55 +0,0 @@
-/*
-  Copyright (c) 2010 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-
-#ifndef __ACCESS_CONTROL_H_
-#define __ACCESS_CONTROL_H_
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#define ACTRL                   "access-control"
-#define ACCTEST_READ            0x1
-#define ACCTEST_WRITE           0x2
-#define ACCTEST_EXEC            0x4
-#define ACCTEST_DONTCARE        0x8
-
-/* Note if the caller is only interested in ownership test i.e. one of the below
-+ * in combination with GF_ACCTEST_DONTCARE, then only one type of user's owner
-+ * ship can be tested with one call to gf_test_access, i.e. we can only
-+ * check of either owner and group, if both need to be tested for a specific
-+ * (uid, gid) pair then two calls will be needed.
-+ */
-#define ACCTEST_OWNER           0x1
-#define ACCTEST_GROUP           0x2
-#define ACCTEST_OTHER           0x4
-
-/* Signifies any user, as long as we get access. */
-#define ACCTEST_ANY             (ACCTEST_OWNER | ACCTEST_GROUP | ACCTEST_OTHER)
-
-#define ac_test_owner(acc)      ((acc) & ACCTEST_OWNER)
-#define ac_test_group(acc)      ((acc) & ACCTEST_GROUP)
-#define ac_test_other(acc)      ((acc) & ACCTEST_OTHER)
-#define ac_test_dontcare(acc)   ((acc) & ACCTEST_DONTCARE)
-#define ac_test_read(acc)       ((acc) & ACCTEST_READ)
-#define ac_test_write(acc)      ((acc) & ACCTEST_WRITE)
-#define ac_test_exec(acc)       ((acc) & ACCTEST_EXEC)
-#endif
diff --git a/xlators/cluster/unify/Makefile.am b/xlators/features/arbiter/Makefile.am
index d471a3f9243..a985f42a877 100644
--- a/xlators/cluster/unify/Makefile.am
+++ b/xlators/features/arbiter/Makefile.am
@@ -1,3 +1,3 @@
 SUBDIRS = src
 
-CLEANFILES = 
+CLEANFILES =
diff --git a/xlators/features/arbiter/src/Makefile.am b/xlators/features/arbiter/src/Makefile.am
new file mode 100644
index 00000000000..badc42f37be
--- /dev/null
+++ b/xlators/features/arbiter/src/Makefile.am
@@ -0,0 +1,19 @@
+if WITH_SERVER
+xlator_LTLIBRARIES = arbiter.la
+endif
+
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+arbiter_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+arbiter_la_SOURCES = arbiter.c
+arbiter_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = arbiter.h arbiter-mem-types.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/arbiter/src/arbiter-mem-types.h b/xlators/features/arbiter/src/arbiter-mem-types.h
new file mode 100644
index 00000000000..05d18374c46
--- /dev/null
+++ b/xlators/features/arbiter/src/arbiter-mem-types.h
@@ -0,0 +1,18 @@
+/*
+  Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __ARBITER_MEM_TYPES_H__
+#define __ARBITER_MEM_TYPES_H__
+#include <glusterfs/mem-types.h>
+
+typedef enum gf_arbiter_mem_types_ {
+    gf_arbiter_mt_inode_ctx_t = gf_common_mt_end + 1,
+    gf_arbiter_mt_end
+} gf_arbiter_mem_types_t;
+#endif
diff --git a/xlators/features/arbiter/src/arbiter.c b/xlators/features/arbiter/src/arbiter.c
new file mode 100644
index 00000000000..83a97e3354b
--- /dev/null
+++ b/xlators/features/arbiter/src/arbiter.c
@@ -0,0 +1,380 @@
+/*
+  Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "arbiter.h"
+#include "arbiter-mem-types.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/logging.h>
+
+static arbiter_inode_ctx_t *
+__arbiter_inode_ctx_get(inode_t *inode, xlator_t *this)
+{
+    arbiter_inode_ctx_t *ctx = NULL;
+    int ret = 0;
+    uint64_t ctx_addr = 0;
+
+    ret = __inode_ctx_get(inode, this, &ctx_addr);
+    if (ret == 0) {
+        ctx = (arbiter_inode_ctx_t *)(long)ctx_addr;
+        goto out;
+    }
+
+    ctx = GF_CALLOC(1, sizeof(*ctx), gf_arbiter_mt_inode_ctx_t);
+    if (!ctx)
+        goto out;
+
+    ret = __inode_ctx_put(inode, this, (uint64_t)(uintptr_t)ctx);
+    if (ret) {
+        GF_FREE(ctx);
+        ctx = NULL;
+        gf_log_callingfn(this->name, GF_LOG_ERROR,
+                         "failed to "
+                         "set the inode ctx (%s)",
+                         uuid_utoa(inode->gfid));
+    }
+out:
+    return ctx;
+}
+
+static arbiter_inode_ctx_t *
+arbiter_inode_ctx_get(inode_t *inode, xlator_t *this)
+{
+    arbiter_inode_ctx_t *ctx = NULL;
+
+    LOCK(&inode->lock);
+    {
+        ctx = __arbiter_inode_ctx_get(inode, this);
+    }
+    UNLOCK(&inode->lock);
+    return ctx;
+}
+
+int32_t
+arbiter_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, inode_t *inode,
+                   struct iatt *buf, dict_t *xdata, struct iatt *postparent)
+{
+    arbiter_inode_ctx_t *ctx = NULL;
+
+    if (op_ret != 0)
+        goto unwind;
+    ctx = arbiter_inode_ctx_get(inode, this);
+    if (!ctx) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+    memcpy(&ctx->iattbuf, buf, sizeof(ctx->iattbuf));
+
+unwind:
+    STACK_UNWIND_STRICT(lookup, frame, op_ret, op_errno, inode, buf, xdata,
+                        postparent);
+    return 0;
+}
+
+int32_t
+arbiter_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    STACK_WIND(frame, arbiter_lookup_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lookup, loc, xdata);
+    return 0;
+}
+
+int32_t
+arbiter_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+                 dict_t *xdata)
+{
+    arbiter_inode_ctx_t *ctx = NULL;
+    struct iatt *buf = NULL;
+    int32_t op_ret = 0;
+    int32_t op_errno = 0;
+
+    ctx = arbiter_inode_ctx_get(loc->inode, this);
+    if (!ctx) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+    buf = &ctx->iattbuf;
+unwind:
+    STACK_UNWIND_STRICT(truncate, frame, op_ret, op_errno, buf, buf, NULL);
+    return 0;
+}
+
+int32_t
+arbiter_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                  dict_t *xdata)
+
+{
+    arbiter_inode_ctx_t *ctx = NULL;
+    struct iatt *buf = NULL;
+    int32_t op_ret = 0;
+    int32_t op_errno = 0;
+
+    ctx = arbiter_inode_ctx_get(fd->inode, this);
+    if (!ctx) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+    buf = &ctx->iattbuf;
+unwind:
+    STACK_UNWIND_STRICT(ftruncate, frame, op_ret, op_errno, buf, buf, NULL);
+    return 0;
+}
+
+dict_t *
+arbiter_fill_writev_xdata(fd_t *fd, dict_t *xdata, xlator_t *this)
+{
+    dict_t *rsp_xdata = NULL;
+    int32_t ret = 0;
+    int is_append = 1;
+
+    if (!fd || !fd->inode || gf_uuid_is_null(fd->inode->gfid)) {
+        goto out;
+    }
+
+    if (!xdata)
+        goto out;
+
+    rsp_xdata = dict_new();
+    if (!rsp_xdata)
+        goto out;
+
+    if (dict_get(xdata, GLUSTERFS_OPEN_FD_COUNT)) {
+        ret = dict_set_uint32(rsp_xdata, GLUSTERFS_OPEN_FD_COUNT,
+                              fd->inode->fd_count);
+        if (ret < 0) {
+            gf_msg_debug(this->name, 0,
+                         "Failed to set dict value"
+                         " for GLUSTERFS_OPEN_FD_COUNT");
+        }
+    }
+    if (dict_get(xdata, GLUSTERFS_WRITE_IS_APPEND)) {
+        ret = dict_set_uint32(rsp_xdata, GLUSTERFS_WRITE_IS_APPEND, is_append);
+        if (ret < 0) {
+            gf_msg_debug(this->name, 0,
+                         "Failed to set dict value"
+                         " for GLUSTERFS_WRITE_IS_APPEND");
+        }
+    }
+out:
+    return rsp_xdata;
+}
+
+int32_t
+arbiter_writev(call_frame_t *frame, xlator_t *this, fd_t *fd,
+               struct iovec *vector, int32_t count, off_t off, uint32_t flags,
+               struct iobref *iobref, dict_t *xdata)
+{
+    arbiter_inode_ctx_t *ctx = NULL;
+    struct iatt *buf = NULL;
+    dict_t *rsp_xdata = NULL;
+    int op_ret = 0;
+    int op_errno = 0;
+
+    ctx = arbiter_inode_ctx_get(fd->inode, this);
+    if (!ctx) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+    buf = &ctx->iattbuf;
+    op_ret = iov_length(vector, count);
+    rsp_xdata = arbiter_fill_writev_xdata(fd, xdata, this);
+unwind:
+    STACK_UNWIND_STRICT(writev, frame, op_ret, op_errno, buf, buf, rsp_xdata);
+    if (rsp_xdata)
+        dict_unref(rsp_xdata);
+    return 0;
+}
+
+int32_t
+arbiter_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                  int32_t keep_size, off_t offset, size_t len, dict_t *xdata)
+{
+    arbiter_inode_ctx_t *ctx = NULL;
+    struct iatt *buf = NULL;
+    int op_ret = 0;
+    int op_errno = 0;
+
+    ctx = arbiter_inode_ctx_get(fd->inode, this);
+    if (!ctx) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+    buf = &ctx->iattbuf;
+unwind:
+    STACK_UNWIND_STRICT(fallocate, frame, op_ret, op_errno, buf, buf, NULL);
+    return 0;
+}
+
+int32_t
+arbiter_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                size_t len, dict_t *xdata)
+{
+    arbiter_inode_ctx_t *ctx = NULL;
+    struct iatt *buf = NULL;
+    int op_ret = 0;
+    int op_errno = 0;
+
+    ctx = arbiter_inode_ctx_get(fd->inode, this);
+    if (!ctx) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+    buf = &ctx->iattbuf;
+unwind:
+    STACK_UNWIND_STRICT(discard, frame, op_ret, op_errno, buf, buf, NULL);
+    return 0;
+}
+
+int32_t
+arbiter_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                 off_t len, dict_t *xdata)
+{
+    arbiter_inode_ctx_t *ctx = NULL;
+    struct iatt *buf = NULL;
+    int op_ret = 0;
+    int op_errno = 0;
+
+    ctx = arbiter_inode_ctx_get(fd->inode, this);
+    if (!ctx) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+    buf = &ctx->iattbuf;
+unwind:
+    STACK_UNWIND_STRICT(zerofill, frame, op_ret, op_errno, buf, buf, NULL);
+    return 0;
+}
+
+static int32_t
+arbiter_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+              off_t offset, uint32_t flags, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(readv, frame, -1, ENOSYS, NULL, 0, NULL, NULL, NULL);
+    return 0;
+}
+
+static int32_t
+arbiter_seek(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             gf_seek_what_t what, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(seek, frame, -1, ENOSYS, 0, xdata);
+    return 0;
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    ret = xlator_mem_acct_init(this, gf_arbiter_mt_end + 1);
+    if (ret)
+        gf_log(this->name, GF_LOG_ERROR,
+               "Memory accounting "
+               "initialization failed.");
+    return ret;
+}
+
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    return 0;
+}
+
+int
+arbiter_forget(xlator_t *this, inode_t *inode)
+{
+    arbiter_inode_ctx_t *ctx = NULL;
+    uint64_t ctx_addr = 0;
+
+    inode_ctx_del(inode, this, &ctx_addr);
+    if (!ctx_addr)
+        return 0;
+    ctx = (arbiter_inode_ctx_t *)(long)ctx_addr;
+    GF_FREE(ctx);
+    return 0;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    if (!this->children || this->children->next) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "'arbiter' not configured with exactly one child");
+        return -1;
+    }
+
+    if (!this->parents)
+        gf_log(this->name, GF_LOG_ERROR, "dangling volume. check volfile ");
+
+    return 0;
+}
+
+void
+fini(xlator_t *this)
+{
+    return;
+}
+
+struct xlator_fops fops = {
+    .lookup = arbiter_lookup,
+
+    /* Return success for these inode write FOPS without winding it down to
+     * posix; this is needed for AFR write transaction logic to work.*/
+    .truncate = arbiter_truncate,
+    .writev = arbiter_writev,
+    .ftruncate = arbiter_ftruncate,
+    .fallocate = arbiter_fallocate,
+    .discard = arbiter_discard,
+    .zerofill = arbiter_zerofill,
+
+    /* AFR is not expected to wind these inode read FOPS initiated by the
+     * application to the arbiter brick. But in case a bug causes them
+     * to be called, we return ENOSYS. */
+    .readv = arbiter_readv,
+    .seek = arbiter_seek,
+
+    /* The following inode read FOPS initiated by the application are not
+     * wound by AFR either but internal logic like  shd, glfsheal and
+     * client side healing in AFR will send them for selfheal/ inode refresh
+     * operations etc.,so we need to wind them down to posix:
+     *
+     * (f)stat, readdir(p), readlink, (f)getxattr.*/
+
+    /* All other FOPs not listed here are safe to be wound down to posix.*/
+};
+
+struct xlator_cbks cbks = {
+    .forget = arbiter_forget,
+};
+
+struct volume_options options[] = {
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "arbiter",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/features/arbiter/src/arbiter.h b/xlators/features/arbiter/src/arbiter.h
new file mode 100644
index 00000000000..546db7b751a
--- /dev/null
+++ b/xlators/features/arbiter/src/arbiter.h
@@ -0,0 +1,21 @@
+/*
+  Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _ARBITER_H
+#define _ARBITER_H
+
+#include <glusterfs/locking.h>
+#include <glusterfs/common-utils.h>
+
+typedef struct arbiter_inode_ctx_ {
+    struct iatt iattbuf;
+} arbiter_inode_ctx_t;
+
+#endif /* _ARBITER_H */
diff --git a/xlators/features/filter/Makefile.am b/xlators/features/barrier/Makefile.am
index d471a3f9243..a985f42a877 100644
--- a/xlators/features/filter/Makefile.am
+++ b/xlators/features/barrier/Makefile.am
@@ -1,3 +1,3 @@
 SUBDIRS = src
 
-CLEANFILES = 
+CLEANFILES =
diff --git a/xlators/features/barrier/src/Makefile.am b/xlators/features/barrier/src/Makefile.am
new file mode 100644
index 00000000000..25099bc56e5
--- /dev/null
+++ b/xlators/features/barrier/src/Makefile.am
@@ -0,0 +1,17 @@
+xlator_LTLIBRARIES = barrier.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+barrier_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+barrier_la_SOURCES = barrier.c
+
+barrier_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = barrier.h barrier-mem-types.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/barrier/src/barrier-mem-types.h b/xlators/features/barrier/src/barrier-mem-types.h
new file mode 100644
index 00000000000..71ed7898d9c
--- /dev/null
+++ b/xlators/features/barrier/src/barrier-mem-types.h
@@ -0,0 +1,20 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __BARRIER_MEM_TYPES_H__
+#define __BARRIER_MEM_TYPES_H__
+
+#include <glusterfs/mem-types.h>
+
+enum gf_barrier_mem_types_ {
+    gf_barrier_mt_priv_t = gf_common_mt_end + 1,
+    gf_barrier_mt_end
+};
+#endif
diff --git a/xlators/features/barrier/src/barrier.c b/xlators/features/barrier/src/barrier.c
new file mode 100644
index 00000000000..852bbacb99d
--- /dev/null
+++ b/xlators/features/barrier/src/barrier.c
@@ -0,0 +1,809 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include "barrier.h"
+#include <glusterfs/defaults.h>
+#include <glusterfs/call-stub.h>
+
+#include <glusterfs/statedump.h>
+
+void
+barrier_local_set_gfid(call_frame_t *frame, uuid_t gfid, xlator_t *this)
+{
+    if (gfid) {
+        uuid_t *id = GF_MALLOC(sizeof(uuid_t), gf_common_mt_uuid_t);
+        if (!id) {
+            gf_log(this->name, GF_LOG_WARNING,
+                   "Could not set gfid"
+                   ". gfid will not be dumped in statedump file.");
+            return;
+        }
+        gf_uuid_copy(*id, gfid);
+        frame->local = id;
+    }
+}
+
+void
+barrier_local_free_gfid(call_frame_t *frame)
+{
+    if (frame->local) {
+        GF_FREE(frame->local);
+        frame->local = NULL;
+    }
+}
+
+int32_t
+barrier_truncate_cbk_resume(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int32_t op_ret, int32_t op_errno,
+                            struct iatt *prebuf, struct iatt *postbuf,
+                            dict_t *xdata)
+{
+    barrier_local_free_gfid(frame);
+    STACK_UNWIND_STRICT(truncate, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    return 0;
+}
+
+int32_t
+barrier_ftruncate_cbk_resume(call_frame_t *frame, void *cookie, xlator_t *this,
+                             int32_t op_ret, int32_t op_errno,
+                             struct iatt *prebuf, struct iatt *postbuf,
+                             dict_t *xdata)
+{
+    barrier_local_free_gfid(frame);
+    STACK_UNWIND_STRICT(ftruncate, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    return 0;
+}
+
+int32_t
+barrier_unlink_cbk_resume(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno,
+                          struct iatt *preparent, struct iatt *postparent,
+                          dict_t *xdata)
+{
+    barrier_local_free_gfid(frame);
+    STACK_UNWIND_STRICT(unlink, frame, op_ret, op_errno, preparent, postparent,
+                        xdata);
+    return 0;
+}
+
+int32_t
+barrier_rmdir_cbk_resume(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno,
+                         struct iatt *preparent, struct iatt *postparent,
+                         dict_t *xdata)
+{
+    barrier_local_free_gfid(frame);
+    STACK_UNWIND_STRICT(rmdir, frame, op_ret, op_errno, preparent, postparent,
+                        xdata);
+    return 0;
+}
+
+int32_t
+barrier_rename_cbk_resume(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                          struct iatt *preoldparent, struct iatt *postoldparent,
+                          struct iatt *prenewparent, struct iatt *postnewparent,
+                          dict_t *xdata)
+{
+    barrier_local_free_gfid(frame);
+    STACK_UNWIND_STRICT(rename, frame, op_ret, op_errno, buf, preoldparent,
+                        postoldparent, prenewparent, postnewparent, xdata);
+    return 0;
+}
+
+int32_t
+barrier_writev_cbk_resume(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                          struct iatt *postbuf, dict_t *xdata)
+{
+    barrier_local_free_gfid(frame);
+    STACK_UNWIND_STRICT(writev, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    return 0;
+}
+
+int32_t
+barrier_fsync_cbk_resume(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                         struct iatt *postbuf, dict_t *xdata)
+{
+    barrier_local_free_gfid(frame);
+    STACK_UNWIND_STRICT(fsync, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+    return 0;
+}
+
+int32_t
+barrier_removexattr_cbk_resume(call_frame_t *frame, void *cookie,
+                               xlator_t *this, int32_t op_ret, int32_t op_errno,
+                               dict_t *xdata)
+{
+    barrier_local_free_gfid(frame);
+    STACK_UNWIND_STRICT(removexattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+int32_t
+barrier_fremovexattr_cbk_resume(call_frame_t *frame, void *cookie,
+                                xlator_t *this, int32_t op_ret,
+                                int32_t op_errno, dict_t *xdata)
+{
+    barrier_local_free_gfid(frame);
+    STACK_UNWIND_STRICT(fremovexattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+int32_t
+barrier_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                   struct iatt *postbuf, dict_t *xdata)
+{
+    BARRIER_FOP_CBK(writev, out, frame, this, op_ret, op_errno, prebuf, postbuf,
+                    xdata);
+out:
+    return 0;
+}
+
+int32_t
+barrier_fremovexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    BARRIER_FOP_CBK(fremovexattr, out, frame, this, op_ret, op_errno, xdata);
+out:
+    return 0;
+}
+
+int32_t
+barrier_removexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    BARRIER_FOP_CBK(removexattr, out, frame, this, op_ret, op_errno, xdata);
+out:
+    return 0;
+}
+
+int32_t
+barrier_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+    BARRIER_FOP_CBK(truncate, out, frame, this, op_ret, op_errno, prebuf,
+                    postbuf, xdata);
+out:
+    return 0;
+}
+
+int32_t
+barrier_ftruncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                      struct iatt *postbuf, dict_t *xdata)
+{
+    BARRIER_FOP_CBK(ftruncate, out, frame, this, op_ret, op_errno, prebuf,
+                    postbuf, xdata);
+out:
+    return 0;
+}
+
+int32_t
+barrier_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                   struct iatt *preoldparent, struct iatt *postoldparent,
+                   struct iatt *prenewparent, struct iatt *postnewparent,
+                   dict_t *xdata)
+{
+    BARRIER_FOP_CBK(rename, out, frame, this, op_ret, op_errno, buf,
+                    preoldparent, postoldparent, prenewparent, postnewparent,
+                    xdata);
+out:
+    return 0;
+}
+
+int32_t
+barrier_rmdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                  struct iatt *postparent, dict_t *xdata)
+{
+    BARRIER_FOP_CBK(rmdir, out, frame, this, op_ret, op_errno, preparent,
+                    postparent, xdata);
+out:
+    return 0;
+}
+
+int32_t
+barrier_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
+{
+    BARRIER_FOP_CBK(unlink, out, frame, this, op_ret, op_errno, preparent,
+                    postparent, xdata);
+out:
+    return 0;
+}
+
+int32_t
+barrier_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                  struct iatt *postbuf, dict_t *xdata)
+{
+    BARRIER_FOP_CBK(fsync, out, frame, this, op_ret, op_errno, prebuf, postbuf,
+                    xdata);
+out:
+    return 0;
+}
+
+int32_t
+barrier_writev(call_frame_t *frame, xlator_t *this, fd_t *fd,
+               struct iovec *vector, int32_t count, off_t off, uint32_t flags,
+               struct iobref *iobref, dict_t *xdata)
+{
+    if (!((flags | fd->flags) & (O_SYNC | O_DSYNC))) {
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->writev, fd, vector, count, off,
+                        flags, iobref, xdata);
+
+        return 0;
+    }
+
+    barrier_local_set_gfid(frame, fd->inode->gfid, this);
+    STACK_WIND(frame, barrier_writev_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->writev, fd, vector, count, off, flags,
+               iobref, xdata);
+    return 0;
+}
+
+int32_t
+barrier_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                     const char *name, dict_t *xdata)
+{
+    barrier_local_set_gfid(frame, fd->inode->gfid, this);
+    STACK_WIND(frame, barrier_fremovexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fremovexattr, fd, name, xdata);
+    return 0;
+}
+
+int32_t
+barrier_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                    const char *name, dict_t *xdata)
+{
+    barrier_local_set_gfid(frame, loc->inode->gfid, this);
+    STACK_WIND(frame, barrier_removexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->removexattr, loc, name, xdata);
+    return 0;
+}
+
+int32_t
+barrier_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+                 dict_t *xdata)
+{
+    barrier_local_set_gfid(frame, loc->inode->gfid, this);
+    STACK_WIND(frame, barrier_truncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+    return 0;
+}
+
+int32_t
+barrier_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc,
+               loc_t *newloc, dict_t *xdata)
+{
+    barrier_local_set_gfid(frame, oldloc->inode->gfid, this);
+    STACK_WIND(frame, barrier_rename_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
+    return 0;
+}
+
+int
+barrier_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+              dict_t *xdata)
+{
+    barrier_local_set_gfid(frame, loc->inode->gfid, this);
+    STACK_WIND(frame, barrier_rmdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rmdir, loc, flags, xdata);
+    return 0;
+}
+
+int32_t
+barrier_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+               dict_t *xdata)
+{
+    barrier_local_set_gfid(frame, loc->inode->gfid, this);
+    STACK_WIND(frame, barrier_unlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->unlink, loc, xflag, xdata);
+    return 0;
+}
+
+int32_t
+barrier_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                  dict_t *xdata)
+{
+    barrier_local_set_gfid(frame, fd->inode->gfid, this);
+    STACK_WIND(frame, barrier_ftruncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+    return 0;
+}
+
+int32_t
+barrier_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+              dict_t *xdata)
+{
+    barrier_local_set_gfid(frame, fd->inode->gfid, this);
+    STACK_WIND(frame, barrier_fsync_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsync, fd, flags, xdata);
+    return 0;
+}
+
+call_stub_t *
+__barrier_dequeue(xlator_t *this, struct list_head *queue)
+{
+    call_stub_t *stub = NULL;
+    barrier_priv_t *priv = NULL;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    if (list_empty(queue))
+        goto out;
+
+    stub = list_entry(queue->next, call_stub_t, list);
+    list_del_init(&stub->list);
+
+out:
+    return stub;
+}
+
+void
+barrier_dequeue_all(xlator_t *this, struct list_head *queue)
+{
+    call_stub_t *stub = NULL;
+
+    gf_log(this->name, GF_LOG_INFO, "Dequeuing all the barriered fops");
+
+    /* TODO: Start the below task in a new thread */
+    while ((stub = __barrier_dequeue(this, queue)))
+        call_resume(stub);
+
+    gf_log(this->name, GF_LOG_INFO,
+           "Dequeuing the barriered fops is "
+           "finished");
+    return;
+}
+
+void
+barrier_timeout(void *data)
+{
+    xlator_t *this = NULL;
+    barrier_priv_t *priv = NULL;
+    struct list_head queue = {
+        0,
+    };
+
+    this = data;
+    THIS = this;
+    priv = this->private;
+
+    INIT_LIST_HEAD(&queue);
+
+    gf_log(this->name, GF_LOG_CRITICAL,
+           "Disabling barrier because of "
+           "the barrier timeout.");
+
+    LOCK(&priv->lock);
+    {
+        __barrier_disable(this, &queue);
+    }
+    UNLOCK(&priv->lock);
+
+    barrier_dequeue_all(this, &queue);
+
+    return;
+}
+
+void
+__barrier_enqueue(xlator_t *this, call_stub_t *stub)
+{
+    barrier_priv_t *priv = NULL;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    list_add_tail(&stub->list, &priv->queue);
+    priv->queue_size++;
+
+    return;
+}
+
+void
+__barrier_disable(xlator_t *this, struct list_head *queue)
+{
+    GF_UNUSED int ret = 0;
+    barrier_priv_t *priv = NULL;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    if (priv->timer) {
+        ret = gf_timer_call_cancel(this->ctx, priv->timer);
+        priv->timer = NULL;
+    }
+
+    list_splice_init(&priv->queue, queue);
+    priv->queue_size = 0;
+    priv->barrier_enabled = _gf_false;
+}
+
+int
+__barrier_enable(xlator_t *this, barrier_priv_t *priv)
+{
+    int ret = -1;
+
+    priv->timer = gf_timer_call_after(this->ctx, priv->timeout, barrier_timeout,
+                                      (void *)this);
+    if (!priv->timer) {
+        gf_log(this->name, GF_LOG_CRITICAL,
+               "Couldn't add barrier "
+               "timeout event.");
+        goto out;
+    }
+
+    priv->barrier_enabled = _gf_true;
+    ret = 0;
+out:
+    return ret;
+}
+
+int
+notify(xlator_t *this, int event, void *data, ...)
+{
+    barrier_priv_t *priv = this->private;
+    dict_t *dict = NULL;
+    int ret = -1;
+    int barrier_enabled = _gf_false;
+    struct list_head queue = {
+        0,
+    };
+
+    GF_ASSERT(priv);
+    INIT_LIST_HEAD(&queue);
+
+    switch (event) {
+        case GF_EVENT_TRANSLATOR_OP: {
+            dict = data;
+            barrier_enabled = dict_get_str_boolean(dict, "barrier", -1);
+
+            if (barrier_enabled == -1) {
+                gf_log(this->name, GF_LOG_ERROR,
+                       "Could not fetch "
+                       " barrier key from the dictionary.");
+                goto out;
+            }
+
+            LOCK(&priv->lock);
+            {
+                if (!priv->barrier_enabled) {
+                    if (barrier_enabled) {
+                        ret = __barrier_enable(this, priv);
+                    } else {
+                        UNLOCK(&priv->lock);
+                        gf_log(this->name, GF_LOG_ERROR, "Already disabled.");
+                        goto post_unlock;
+                    }
+                } else {
+                    if (!barrier_enabled) {
+                        __barrier_disable(this, &queue);
+                        ret = 0;
+                    } else {
+                        UNLOCK(&priv->lock);
+                        gf_log(this->name, GF_LOG_ERROR, "Already enabled");
+                        goto post_unlock;
+                    }
+                }
+            }
+            UNLOCK(&priv->lock);
+        post_unlock:
+            if (!list_empty(&queue))
+                barrier_dequeue_all(this, &queue);
+
+            break;
+        }
+        default: {
+            default_notify(this, event, data);
+            ret = 0;
+            goto out;
+        }
+    }
+out:
+    return ret;
+}
+
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    barrier_priv_t *priv = NULL;
+    int ret = -1;
+    gf_boolean_t barrier_enabled = _gf_false;
+    uint32_t timeout = {
+        0,
+    };
+    struct list_head queue = {
+        0,
+    };
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    GF_OPTION_RECONF("barrier", barrier_enabled, options, bool, out);
+    GF_OPTION_RECONF("barrier-timeout", timeout, options, time, out);
+
+    INIT_LIST_HEAD(&queue);
+
+    LOCK(&priv->lock);
+    {
+        if (!priv->barrier_enabled) {
+            if (barrier_enabled) {
+                ret = __barrier_enable(this, priv);
+                if (ret) {
+                    goto unlock;
+                }
+            }
+        } else {
+            if (!barrier_enabled) {
+                __barrier_disable(this, &queue);
+            }
+        }
+        priv->timeout.tv_sec = timeout;
+        ret = 0;
+    }
+unlock:
+    UNLOCK(&priv->lock);
+
+    if (!list_empty(&queue))
+        barrier_dequeue_all(this, &queue);
+
+out:
+    return ret;
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    ret = xlator_mem_acct_init(this, gf_barrier_mt_end + 1);
+    if (ret)
+        gf_log(this->name, GF_LOG_ERROR,
+               "Memory accounting "
+               "initialization failed.");
+
+    return ret;
+}
+
+int
+init(xlator_t *this)
+{
+    int ret = -1;
+    barrier_priv_t *priv = NULL;
+    uint32_t timeout = {
+        0,
+    };
+
+    if (!this->children || this->children->next) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "'barrier' not configured with exactly one child");
+        goto out;
+    }
+
+    if (!this->parents)
+        gf_log(this->name, GF_LOG_WARNING, "dangling volume. check volfile ");
+
+    priv = GF_CALLOC(1, sizeof(*priv), gf_barrier_mt_priv_t);
+    if (!priv)
+        goto out;
+
+    LOCK_INIT(&priv->lock);
+
+    GF_OPTION_INIT("barrier", priv->barrier_enabled, bool, out);
+    GF_OPTION_INIT("barrier-timeout", timeout, time, out);
+    priv->timeout.tv_sec = timeout;
+
+    INIT_LIST_HEAD(&priv->queue);
+
+    if (priv->barrier_enabled) {
+        ret = __barrier_enable(this, priv);
+        if (ret == -1)
+            goto out;
+    }
+
+    this->private = priv;
+    ret = 0;
+out:
+    if (ret && priv)
+        GF_FREE(priv);
+
+    return ret;
+}
+
+void
+fini(xlator_t *this)
+{
+    barrier_priv_t *priv = NULL;
+    struct list_head queue = {
+        0,
+    };
+
+    priv = this->private;
+    if (!priv)
+        goto out;
+
+    INIT_LIST_HEAD(&queue);
+
+    gf_log(this->name, GF_LOG_INFO,
+           "Disabling barriering and dequeuing "
+           "all the queued fops");
+    LOCK(&priv->lock);
+    {
+        __barrier_disable(this, &queue);
+    }
+    UNLOCK(&priv->lock);
+
+    if (!list_empty(&queue))
+        barrier_dequeue_all(this, &queue);
+
+    this->private = NULL;
+
+    LOCK_DESTROY(&priv->lock);
+    GF_FREE(priv);
+out:
+    return;
+}
+
+static void
+barrier_dump_stub(call_stub_t *stub, char *prefix)
+{
+    char key[GF_DUMP_MAX_BUF_LEN] = {
+        0,
+    };
+
+    gf_proc_dump_build_key(key, prefix, "fop");
+    gf_proc_dump_write(key, "%s", gf_fop_list[stub->fop]);
+
+    if (stub->frame->local) {
+        gf_proc_dump_build_key(key, prefix, "gfid");
+        gf_proc_dump_write(key, "%s",
+                           uuid_utoa(*(uuid_t *)(stub->frame->local)));
+    }
+    if (stub->args.loc.path) {
+        gf_proc_dump_build_key(key, prefix, "path");
+        gf_proc_dump_write(key, "%s", stub->args.loc.path);
+    }
+    if (stub->args.loc.name) {
+        gf_proc_dump_build_key(key, prefix, "name");
+        gf_proc_dump_write(key, "%s", stub->args.loc.name);
+    }
+
+    return;
+}
+
+static void
+__barrier_dump_queue(barrier_priv_t *priv)
+{
+    call_stub_t *stub = NULL;
+    char key[GF_DUMP_MAX_BUF_LEN] = {
+        0,
+    };
+    int i = 0;
+
+    GF_VALIDATE_OR_GOTO("barrier", priv, out);
+
+    list_for_each_entry(stub, &priv->queue, list)
+    {
+        snprintf(key, sizeof(key), "stub.%d", i++);
+        gf_proc_dump_add_section("%s", key);
+        barrier_dump_stub(stub, key);
+    }
+
+out:
+    return;
+}
+
+int
+barrier_dump_priv(xlator_t *this)
+{
+    int ret = -1;
+    char key[GF_DUMP_MAX_BUF_LEN] = {
+        0,
+    };
+    barrier_priv_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("barrier", this, out);
+
+    priv = this->private;
+    if (!priv)
+        return 0;
+
+    gf_proc_dump_build_key(key, "xlator.features.barrier", "priv");
+    gf_proc_dump_add_section("%s", key);
+    gf_proc_dump_build_key(key, "barrier", "enabled");
+
+    LOCK(&priv->lock);
+    {
+        gf_proc_dump_write(key, "%d", priv->barrier_enabled);
+        gf_proc_dump_build_key(key, "barrier", "timeout");
+        gf_proc_dump_write(key, "%ld", priv->timeout.tv_sec);
+        if (priv->barrier_enabled) {
+            gf_proc_dump_build_key(key, "barrier", "queue_size");
+            gf_proc_dump_write(key, "%d", priv->queue_size);
+            __barrier_dump_queue(priv);
+        }
+    }
+    UNLOCK(&priv->lock);
+
+out:
+    return ret;
+}
+
+struct xlator_fops fops = {
+
+    /* Barrier Class fops */
+    .rmdir = barrier_rmdir,
+    .unlink = barrier_unlink,
+    .rename = barrier_rename,
+    .removexattr = barrier_removexattr,
+    .fremovexattr = barrier_fremovexattr,
+    .truncate = barrier_truncate,
+    .ftruncate = barrier_ftruncate,
+    .fsync = barrier_fsync,
+
+    /* Writes with only O_SYNC flag */
+    .writev = barrier_writev,
+};
+
+struct xlator_dumpops dumpops = {
+    .priv = barrier_dump_priv,
+};
+
+struct xlator_cbks cbks;
+
+struct volume_options options[] = {
+    {.key = {"barrier"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "disable",
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .flags = OPT_FLAG_SETTABLE,
+     .description = "When \"enabled\", blocks acknowledgements to application "
+                    "for file operations such as rmdir, rename, unlink, "
+                    "removexattr, fremovexattr, truncate, ftruncate, "
+                    "write (with O_SYNC), fsync. It is turned \"off\" by "
+                    "default."},
+    {.key = {"barrier-timeout"},
+     .type = GF_OPTION_TYPE_TIME,
+     .default_value = BARRIER_TIMEOUT,
+     .op_version = {GD_OP_VERSION_3_6_0},
+     .flags = OPT_FLAG_SETTABLE,
+     .description = "After 'timeout' seconds since the time 'barrier' "
+                    "option was set to \"on\", acknowledgements to file "
+                    "operations are no longer blocked and previously "
+                    "blocked acknowledgements are sent to the application"},
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .dumpops = &dumpops,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "barrier",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/features/barrier/src/barrier.h b/xlators/features/barrier/src/barrier.h
new file mode 100644
index 00000000000..1337f311f7d
--- /dev/null
+++ b/xlators/features/barrier/src/barrier.h
@@ -0,0 +1,89 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __BARRIER_H__
+#define __BARRIER_H__
+
+#include "barrier-mem-types.h"
+#include <glusterfs/xlator.h>
+#include <glusterfs/timer.h>
+#include <glusterfs/call-stub.h>
+
+#define BARRIER_FOP_CBK(fop_name, label, frame, this, params...)               \
+    do {                                                                       \
+        barrier_priv_t *_priv = NULL;                                          \
+        call_stub_t *_stub = NULL;                                             \
+        gf_boolean_t _barrier_enabled = _gf_false;                             \
+        struct list_head queue = {                                             \
+            0,                                                                 \
+        };                                                                     \
+                                                                               \
+        INIT_LIST_HEAD(&queue);                                                \
+                                                                               \
+        _priv = this->private;                                                 \
+        GF_ASSERT(_priv);                                                      \
+                                                                               \
+        LOCK(&_priv->lock);                                                    \
+        {                                                                      \
+            if (_priv->barrier_enabled) {                                      \
+                _barrier_enabled = _priv->barrier_enabled;                     \
+                                                                               \
+                _stub = fop_##fop_name##_cbk_stub(                             \
+                    frame, barrier_##fop_name##_cbk_resume, params);           \
+                if (!_stub) {                                                  \
+                    __barrier_disable(this, &queue);                           \
+                    goto unlock;                                               \
+                }                                                              \
+                                                                               \
+                __barrier_enqueue(this, _stub);                                \
+            }                                                                  \
+        }                                                                      \
+    unlock:                                                                    \
+        UNLOCK(&_priv->lock);                                                  \
+                                                                               \
+        if (_stub)                                                             \
+            goto label;                                                        \
+                                                                               \
+        if (_barrier_enabled && !_stub) {                                      \
+            gf_log(this->name, GF_LOG_CRITICAL,                                \
+                   "Failed to barrier FOPs, disabling "                        \
+                   "barrier. FOP: %s, ERROR: %s",                              \
+                   #fop_name, strerror(ENOMEM));                               \
+            barrier_dequeue_all(this, &queue);                                 \
+        }                                                                      \
+        barrier_local_free_gfid(frame);                                        \
+        STACK_UNWIND_STRICT(fop_name, frame, params);                          \
+        goto label;                                                            \
+    } while (0)
+
+typedef struct {
+    gf_timer_t *timer;
+    gf_lock_t lock;
+    struct list_head queue;
+    struct timespec timeout;
+    uint32_t queue_size;
+    gf_boolean_t barrier_enabled;
+    char _pad[3]; /* manual padding */
+} barrier_priv_t;
+
+int
+__barrier_enable(xlator_t *this, barrier_priv_t *priv);
+void
+__barrier_enqueue(xlator_t *this, call_stub_t *stub);
+void
+__barrier_disable(xlator_t *this, struct list_head *queue);
+void
+barrier_timeout(void *data);
+void
+barrier_dequeue_all(xlator_t *this, struct list_head *queue);
+call_stub_t *
+__barrier_dequeue(xlator_t *this, struct list_head *queue);
+
+#endif
diff --git a/xlators/protocol/legacy/transport/ib-verbs/Makefile.am b/xlators/features/bit-rot/Makefile.am
index f963effea22..f963effea22 100644
--- a/xlators/protocol/legacy/transport/ib-verbs/Makefile.am
+++ b/xlators/features/bit-rot/Makefile.am
diff --git a/xlators/features/bit-rot/src/Makefile.am b/xlators/features/bit-rot/src/Makefile.am
new file mode 100644
index 00000000000..b5e4a7d62a0
--- /dev/null
+++ b/xlators/features/bit-rot/src/Makefile.am
@@ -0,0 +1 @@
+SUBDIRS = stub bitd
diff --git a/xlators/features/bit-rot/src/bitd/Makefile.am b/xlators/features/bit-rot/src/bitd/Makefile.am
new file mode 100644
index 00000000000..6db800e6565
--- /dev/null
+++ b/xlators/features/bit-rot/src/bitd/Makefile.am
@@ -0,0 +1,23 @@
+if WITH_SERVER
+xlator_LTLIBRARIES = bit-rot.la
+endif
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+bit_rot_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src/ -I$(top_builddir)/rpc/xdr/src/ \
+	-I$(top_srcdir)/rpc/rpc-lib/src -I$(CONTRIBDIR)/timer-wheel \
+	-I$(top_srcdir)/xlators/features/bit-rot/src/stub
+
+bit_rot_la_SOURCES = bit-rot.c bit-rot-scrub.c bit-rot-ssm.c \
+		     bit-rot-scrub-status.c
+bit_rot_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la \
+	$(top_builddir)/xlators/features/changelog/lib/src/libgfchangelog.la
+
+noinst_HEADERS = bit-rot.h bit-rot-scrub.h bit-rot-bitd-messages.h bit-rot-ssm.h \
+		 bit-rot-scrub-status.h
+
+AM_CFLAGS = -Wall -DBR_RATE_LIMIT_SIGNER $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-bitd-messages.h b/xlators/features/bit-rot/src/bitd/bit-rot-bitd-messages.h
new file mode 100644
index 00000000000..5bc5103a27c
--- /dev/null
+++ b/xlators/features/bit-rot/src/bitd/bit-rot-bitd-messages.h
@@ -0,0 +1,101 @@
+/*
+ Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+ This file is part of GlusterFS.
+
+ This file is licensed to you under your choice of the GNU Lesser
+ General Public License, version 3 or any later version (LGPLv3 or
+ later), or the GNU General Public License, version 2 (GPLv2), in all
+ cases as published by the Free Software Foundation.
+ */
+
+#ifndef _BITROT_BITD_MESSAGES_H_
+#define _BITROT_BITD_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(BITROT_BITD, BRB_MSG_FD_CREATE_FAILED, BRB_MSG_READV_FAILED,
+           BRB_MSG_BLOCK_READ_FAILED, BRB_MSG_CALC_CHECKSUM_FAILED,
+           BRB_MSG_NO_MEMORY, BRB_MSG_GET_SIGN_FAILED, BRB_MSG_SET_SIGN_FAILED,
+           BRB_MSG_OP_FAILED, BRB_MSG_READ_AND_SIGN_FAILED, BRB_MSG_SIGN_FAILED,
+           BRB_MSG_GET_SUBVOL_FAILED, BRB_MSG_SET_TIMER_FAILED,
+           BRB_MSG_GET_INFO_FAILED, BRB_MSG_PATH_FAILED, BRB_MSG_MARK_BAD_FILE,
+           BRB_MSG_TRIGGER_SIGN, BRB_MSG_REGISTER_FAILED,
+           BRB_MSG_CRAWLING_START, BRB_MSG_SPAWN_FAILED,
+           BRB_MSG_INVALID_SUBVOL_CHILD, BRB_MSG_SKIP_OBJECT, BRB_MSG_NO_CHILD,
+           BRB_MSG_CHECKSUM_MISMATCH, BRB_MSG_MARK_CORRUPTED,
+           BRB_MSG_CRAWLING_FINISH, BRB_MSG_CALC_ERROR, BRB_MSG_LOOKUP_FAILED,
+           BRB_MSG_PARTIAL_VERSION_PRESENCE, BRB_MSG_MEM_ACNT_FAILED,
+           BRB_MSG_TIMER_WHEEL_UNAVAILABLE, BRB_MSG_BITROT_LOADED,
+           BRB_MSG_SCALE_DOWN_FAILED, BRB_MSG_SCALE_UP_FAILED,
+           BRB_MSG_SCALE_DOWN_SCRUBBER, BRB_MSG_SCALING_UP_SCRUBBER,
+           BRB_MSG_UNKNOWN_THROTTLE, BRB_MSG_RATE_LIMIT_INFO,
+           BRB_MSG_SCRUB_INFO, BRB_MSG_CONNECTED_TO_BRICK, BRB_MSG_BRICK_INFO,
+           BRB_MSG_SUBVOL_CONNECT_FAILED, BRB_MSG_INVALID_SUBVOL,
+           BRB_MSG_RESCHEDULE_SCRUBBER_FAILED, BRB_MSG_SCRUB_START,
+           BRB_MSG_SCRUB_FINISH, BRB_MSG_SCRUB_RUNNING,
+           BRB_MSG_SCRUB_RESCHEDULED, BRB_MSG_SCRUB_TUNABLE,
+           BRB_MSG_SCRUB_THREAD_CLEANUP, BRB_MSG_SCRUBBER_CLEANED,
+           BRB_MSG_GENERIC_SSM_INFO, BRB_MSG_ZERO_TIMEOUT_BUG,
+           BRB_MSG_BAD_OBJ_READDIR_FAIL, BRB_MSG_SSM_FAILED,
+           BRB_MSG_SCRUB_WAIT_FAILED, BRB_MSG_TRIGGER_SIGN_FAILED,
+           BRB_MSG_EVENT_UNHANDLED, BRB_MSG_COULD_NOT_SCHEDULE_SCRUB,
+           BRB_MSG_THREAD_CREATION_FAILED, BRB_MSG_MEM_POOL_ALLOC,
+           BRB_MSG_SAVING_HASH_FAILED);
+
+#define BRB_MSG_FD_CREATE_FAILED_STR "failed to create fd for the inode"
+#define BRB_MSG_READV_FAILED_STR "readv failed"
+#define BRB_MSG_BLOCK_READ_FAILED_STR "reading block failed"
+#define BRB_MSG_NO_MEMORY_STR "failed to allocate memory"
+#define BRB_MSG_CALC_CHECKSUM_FAILED_STR "calculating checksum failed"
+#define BRB_MSG_GET_SIGN_FAILED_STR "failed to get the signature"
+#define BRB_MSG_SET_SIGN_FAILED_STR "signing failed"
+#define BRB_MSG_OP_FAILED_STR "failed on object"
+#define BRB_MSG_TRIGGER_SIGN_FAILED_STR "Could not trigger signing"
+#define BRB_MSG_READ_AND_SIGN_FAILED_STR "reading and signing of object failed"
+#define BRB_MSG_SET_TIMER_FAILED_STR "Failed to allocate object expiry timer"
+#define BRB_MSG_GET_SUBVOL_FAILED_STR                                          \
+    "failed to get the subvolume for the brick"
+#define BRB_MSG_PATH_FAILED_STR "path failed"
+#define BRB_MSG_SKIP_OBJECT_STR "Entry is marked corrupted. skipping"
+#define BRB_MSG_PARTIAL_VERSION_PRESENCE_STR                                   \
+    "PArtial version xattr presence detected, ignoring"
+#define BRB_MSG_TRIGGER_SIGN_STR "Triggering signing"
+#define BRB_MSG_CRAWLING_START_STR                                             \
+    "Crawling brick, scanning for unsigned objects"
+#define BRB_MSG_CRAWLING_FINISH_STR "Completed crawling brick"
+#define BRB_MSG_REGISTER_FAILED_STR "Register to changelog failed"
+#define BRB_MSG_SPAWN_FAILED_STR "failed to spawn"
+#define BRB_MSG_CONNECTED_TO_BRICK_STR "Connected to brick"
+#define BRB_MSG_LOOKUP_FAILED_STR "lookup on root failed"
+#define BRB_MSG_GET_INFO_FAILED_STR "failed to get stub info"
+#define BRB_MSG_SCRUB_THREAD_CLEANUP_STR "Error cleaning up scanner thread"
+#define BRB_MSG_SCRUBBER_CLEANED_STR "clened up scrubber for brick"
+#define BRB_MSG_SUBVOL_CONNECT_FAILED_STR                                      \
+    "callback handler for subvolume failed"
+#define BRB_MSG_MEM_ACNT_FAILED_STR "Memory accounting init failed"
+#define BRB_MSG_EVENT_UNHANDLED_STR "Event unhandled for child"
+#define BRB_MSG_INVALID_SUBVOL_STR "Got event from invalid subvolume"
+#define BRB_MSG_RESCHEDULE_SCRUBBER_FAILED_STR                                 \
+    "on demand scrub schedule failed. Scrubber is not in pending state."
+#define BRB_MSG_COULD_NOT_SCHEDULE_SCRUB_STR                                   \
+    "Could not schedule ondemand scrubbing. Scrubbing will continue "          \
+    "according to old frequency."
+#define BRB_MSG_THREAD_CREATION_FAILED_STR "thread creation failed"
+#define BRB_MSG_RATE_LIMIT_INFO_STR "Rate Limit Info"
+#define BRB_MSG_MEM_POOL_ALLOC_STR "failed to allocate mem-pool for timer"
+#define BRB_MSG_NO_CHILD_STR "FATAL: no children"
+#define BRB_MSG_TIMER_WHEEL_UNAVAILABLE_STR "global timer wheel unavailable"
+#define BRB_MSG_BITROT_LOADED_STR "bit-rot xlator loaded"
+#define BRB_MSG_SAVING_HASH_FAILED_STR                                         \
+    "failed to allocate memory for saving hash of the object"
+#endif /* !_BITROT_BITD_MESSAGES_H_ */
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.c b/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.c
new file mode 100644
index 00000000000..5cef2ffa5e5
--- /dev/null
+++ b/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.c
@@ -0,0 +1,78 @@
+/*
+  Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <string.h>
+#include <stdio.h>
+
+#include "bit-rot-scrub-status.h"
+
+void
+br_inc_unsigned_file_count(br_scrub_stats_t *scrub_stat)
+{
+    if (!scrub_stat)
+        return;
+
+    pthread_mutex_lock(&scrub_stat->lock);
+    {
+        scrub_stat->unsigned_files++;
+    }
+    pthread_mutex_unlock(&scrub_stat->lock);
+}
+
+void
+br_inc_scrubbed_file(br_scrub_stats_t *scrub_stat)
+{
+    if (!scrub_stat)
+        return;
+
+    pthread_mutex_lock(&scrub_stat->lock);
+    {
+        scrub_stat->scrubbed_files++;
+    }
+    pthread_mutex_unlock(&scrub_stat->lock);
+}
+
+void
+br_update_scrub_start_time(br_scrub_stats_t *scrub_stat, time_t time)
+{
+    if (!scrub_stat)
+        return;
+
+    pthread_mutex_lock(&scrub_stat->lock);
+    {
+        scrub_stat->scrub_start_time = time;
+    }
+    pthread_mutex_unlock(&scrub_stat->lock);
+}
+
+void
+br_update_scrub_finish_time(br_scrub_stats_t *scrub_stat, char *timestr,
+                            time_t time)
+{
+    int lst_size = 0;
+
+    if (!scrub_stat)
+        return;
+
+    lst_size = sizeof(scrub_stat->last_scrub_time);
+    if (strlen(timestr) >= lst_size)
+        return;
+
+    pthread_mutex_lock(&scrub_stat->lock);
+    {
+        scrub_stat->scrub_end_time = time;
+
+        scrub_stat->scrub_duration = scrub_stat->scrub_end_time -
+                                     scrub_stat->scrub_start_time;
+
+        snprintf(scrub_stat->last_scrub_time, lst_size, "%s", timestr);
+    }
+    pthread_mutex_unlock(&scrub_stat->lock);
+}
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.h b/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.h
new file mode 100644
index 00000000000..f022aa831eb
--- /dev/null
+++ b/xlators/features/bit-rot/src/bitd/bit-rot-scrub-status.h
@@ -0,0 +1,50 @@
+/*
+   Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __BIT_ROT_SCRUB_STATUS_H__
+#define __BIT_ROT_SCRUB_STATUS_H__
+
+#include <stdint.h>
+#include <sys/time.h>
+#include <pthread.h>
+
+#include <glusterfs/common-utils.h>
+
+struct br_scrub_stats {
+    uint64_t scrubbed_files; /* Total number of scrubbed files. */
+
+    uint64_t unsigned_files; /* Total number of unsigned files. */
+
+    uint64_t scrub_duration; /* Duration of last scrub. */
+
+    char last_scrub_time[GF_TIMESTR_SIZE]; /* Last scrub completion time. */
+
+    time_t scrub_start_time; /* Scrubbing starting time. */
+
+    time_t scrub_end_time; /* Scrubbing finishing time. */
+
+    int8_t scrub_running; /* Whether scrub running or not. */
+
+    pthread_mutex_t lock;
+};
+
+typedef struct br_scrub_stats br_scrub_stats_t;
+
+void
+br_inc_unsigned_file_count(br_scrub_stats_t *scrub_stat);
+void
+br_inc_scrubbed_file(br_scrub_stats_t *scrub_stat);
+void
+br_update_scrub_start_time(br_scrub_stats_t *scrub_stat, time_t time);
+void
+br_update_scrub_finish_time(br_scrub_stats_t *scrub_stat, char *timestr,
+                            time_t time);
+
+#endif /* __BIT_ROT_SCRUB_STATUS_H__ */
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-scrub.c b/xlators/features/bit-rot/src/bitd/bit-rot-scrub.c
new file mode 100644
index 00000000000..289dd53f610
--- /dev/null
+++ b/xlators/features/bit-rot/src/bitd/bit-rot-scrub.c
@@ -0,0 +1,2070 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <math.h>
+#include <ctype.h>
+#include <sys/uio.h>
+
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/common-utils.h>
+
+#include "bit-rot-scrub.h"
+#include <pthread.h>
+#include "bit-rot-bitd-messages.h"
+#include "bit-rot-scrub-status.h"
+#include <glusterfs/events.h>
+
+struct br_scrubbers {
+    pthread_t scrubthread;
+
+    struct list_head list;
+};
+
+struct br_fsscan_entry {
+    void *data;
+
+    loc_t parent;
+
+    gf_dirent_t *entry;
+
+    struct br_scanfs *fsscan; /* backpointer to subvolume scanner */
+
+    struct list_head list;
+};
+
+/**
+ * fetch signature extended attribute from an object's fd.
+ * NOTE: On success @xattr is not unref'd as @sign points
+ * to the dictionary value.
+ */
+static int32_t
+bitd_fetch_signature(xlator_t *this, br_child_t *child, fd_t *fd,
+                     dict_t **xattr, br_isignature_out_t **sign)
+{
+    int32_t ret = -1;
+
+    ret = syncop_fgetxattr(child->xl, fd, xattr, GLUSTERFS_GET_OBJECT_SIGNATURE,
+                           NULL, NULL);
+    if (ret < 0) {
+        br_log_object(this, "fgetxattr", fd->inode->gfid, -ret);
+        goto out;
+    }
+
+    ret = dict_get_ptr(*xattr, GLUSTERFS_GET_OBJECT_SIGNATURE, (void **)sign);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_GET_SIGN_FAILED,
+               "failed to extract signature info [GFID: %s]",
+               uuid_utoa(fd->inode->gfid));
+        goto unref_dict;
+    }
+
+    return 0;
+
+unref_dict:
+    dict_unref(*xattr);
+out:
+    return -1;
+}
+
+/**
+ * POST COMPUTE CHECK
+ *
+ * Checks to be performed before verifying calculated signature
+ * Object is skipped if:
+ *  - has stale signature
+ *  - mismatches versions caches in pre-compute check
+ */
+
+int32_t
+bitd_scrub_post_compute_check(xlator_t *this, br_child_t *child, fd_t *fd,
+                              unsigned long version,
+                              br_isignature_out_t **signature,
+                              br_scrub_stats_t *scrub_stat,
+                              gf_boolean_t skip_stat)
+{
+    int32_t ret = 0;
+    size_t signlen = 0;
+    dict_t *xattr = NULL;
+    br_isignature_out_t *signptr = NULL;
+
+    ret = bitd_fetch_signature(this, child, fd, &xattr, &signptr);
+    if (ret < 0) {
+        if (!skip_stat)
+            br_inc_unsigned_file_count(scrub_stat);
+        goto out;
+    }
+
+    /**
+     * Either the object got dirtied during the time the signature was
+     * calculated OR the version we saved during pre-compute check does
+     * not match now, implying that the object got dirtied and signed in
+     * between scrubs pre & post compute checks (checksum window).
+     *
+     * The log entry looks pretty ugly, but helps in debugging..
+     */
+    if (signptr->stale || (signptr->version != version)) {
+        if (!skip_stat)
+            br_inc_unsigned_file_count(scrub_stat);
+        gf_msg_debug(this->name, 0,
+                     "<STAGE: POST> Object [GFID: %s] "
+                     "either has a stale signature OR underwent "
+                     "signing during checksumming {Stale: %d | "
+                     "Version: %lu,%lu}",
+                     uuid_utoa(fd->inode->gfid), (signptr->stale) ? 1 : 0,
+                     version, signptr->version);
+        ret = -1;
+        goto unref_dict;
+    }
+
+    signlen = signptr->signaturelen;
+    *signature = GF_MALLOC(sizeof(br_isignature_out_t) + signlen,
+                           gf_common_mt_char);
+
+    (void)memcpy(*signature, signptr, sizeof(br_isignature_out_t) + signlen);
+
+    (*signature)->signaturelen = signlen;
+
+unref_dict:
+    dict_unref(xattr);
+out:
+    return ret;
+}
+
+static int32_t
+bitd_signature_staleness(xlator_t *this, br_child_t *child, fd_t *fd,
+                         int *stale, unsigned long *version,
+                         br_scrub_stats_t *scrub_stat, gf_boolean_t skip_stat)
+{
+    int32_t ret = -1;
+    dict_t *xattr = NULL;
+    br_isignature_out_t *signptr = NULL;
+
+    ret = bitd_fetch_signature(this, child, fd, &xattr, &signptr);
+    if (ret < 0) {
+        if (!skip_stat)
+            br_inc_unsigned_file_count(scrub_stat);
+        goto out;
+    }
+
+    /**
+     * save version for validation in post compute stage
+     * c.f. bitd_scrub_post_compute_check()
+     */
+    *stale = signptr->stale ? 1 : 0;
+    *version = signptr->version;
+
+    dict_unref(xattr);
+
+out:
+    return ret;
+}
+
+/**
+ * PRE COMPUTE CHECK
+ *
+ * Checks to be performed before initiating object signature calculation.
+ * An object is skipped if:
+ *  - it's already marked corrupted
+ *  - has stale signature
+ */
+int32_t
+bitd_scrub_pre_compute_check(xlator_t *this, br_child_t *child, fd_t *fd,
+                             unsigned long *version,
+                             br_scrub_stats_t *scrub_stat,
+                             gf_boolean_t skip_stat)
+{
+    int stale = 0;
+    int32_t ret = -1;
+
+    if (bitd_is_bad_file(this, child, NULL, fd)) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, BRB_MSG_SKIP_OBJECT,
+               "Object [GFID: %s] is marked corrupted, skipping..",
+               uuid_utoa(fd->inode->gfid));
+        goto out;
+    }
+
+    ret = bitd_signature_staleness(this, child, fd, &stale, version, scrub_stat,
+                                   skip_stat);
+    if (!ret && stale) {
+        if (!skip_stat)
+            br_inc_unsigned_file_count(scrub_stat);
+        gf_msg_debug(this->name, 0,
+                     "<STAGE: PRE> Object [GFID: %s] "
+                     "has stale signature",
+                     uuid_utoa(fd->inode->gfid));
+        ret = -1;
+    }
+
+out:
+    return ret;
+}
+
+/* static int */
+int
+bitd_compare_ckum(xlator_t *this, br_isignature_out_t *sign, unsigned char *md,
+                  inode_t *linked_inode, gf_dirent_t *entry, fd_t *fd,
+                  br_child_t *child, loc_t *loc)
+{
+    int ret = -1;
+    dict_t *xattr = NULL;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, sign, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+    GF_VALIDATE_OR_GOTO(this->name, child, out);
+    GF_VALIDATE_OR_GOTO(this->name, linked_inode, out);
+    GF_VALIDATE_OR_GOTO(this->name, md, out);
+    GF_VALIDATE_OR_GOTO(this->name, entry, out);
+
+    if (strncmp(sign->signature, (char *)md, sign->signaturelen) == 0) {
+        gf_msg_debug(this->name, 0,
+                     "%s [GFID: %s | Brick: %s] "
+                     "matches calculated checksum",
+                     loc->path, uuid_utoa(linked_inode->gfid),
+                     child->brick_path);
+        return 0;
+    }
+
+    gf_msg(this->name, GF_LOG_DEBUG, 0, BRB_MSG_CHECKSUM_MISMATCH,
+           "Object checksum mismatch: %s [GFID: %s | Brick: %s]", loc->path,
+           uuid_utoa(linked_inode->gfid), child->brick_path);
+    gf_msg(this->name, GF_LOG_ALERT, 0, BRB_MSG_CHECKSUM_MISMATCH,
+           "CORRUPTION DETECTED: Object %s {Brick: %s | GFID: %s}", loc->path,
+           child->brick_path, uuid_utoa(linked_inode->gfid));
+
+    /* Perform bad-file marking */
+    xattr = dict_new();
+    if (!xattr) {
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_set_int32(xattr, BITROT_OBJECT_BAD_KEY, _gf_true);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_MARK_BAD_FILE,
+               "Error setting bad-file marker for %s [GFID: %s | "
+               "Brick: %s]",
+               loc->path, uuid_utoa(linked_inode->gfid), child->brick_path);
+        goto dictfree;
+    }
+
+    gf_msg(this->name, GF_LOG_ALERT, 0, BRB_MSG_MARK_CORRUPTED,
+           "Marking"
+           " %s [GFID: %s | Brick: %s] as corrupted..",
+           loc->path, uuid_utoa(linked_inode->gfid), child->brick_path);
+    gf_event(EVENT_BITROT_BAD_FILE, "gfid=%s;path=%s;brick=%s",
+             uuid_utoa(linked_inode->gfid), loc->path, child->brick_path);
+    ret = syncop_fsetxattr(child->xl, fd, xattr, 0, NULL, NULL);
+    if (ret)
+        gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_MARK_BAD_FILE,
+               "Error marking object %s [GFID: %s] as corrupted", loc->path,
+               uuid_utoa(linked_inode->gfid));
+
+dictfree:
+    dict_unref(xattr);
+out:
+    return ret;
+}
+
+/**
+ * "The Scrubber"
+ *
+ * Perform signature validation for a given object with the assumption
+ * that the signature is SHA256 (because signer as of now _always_
+ * signs with SHA256).
+ */
+int
+br_scrubber_scrub_begin(xlator_t *this, struct br_fsscan_entry *fsentry)
+{
+    int32_t ret = -1;
+    fd_t *fd = NULL;
+    loc_t loc = {
+        0,
+    };
+    struct iatt iatt = {
+        0,
+    };
+    struct iatt parent_buf = {
+        0,
+    };
+    pid_t pid = 0;
+    br_child_t *child = NULL;
+    unsigned char *md = NULL;
+    inode_t *linked_inode = NULL;
+    br_isignature_out_t *sign = NULL;
+    unsigned long signedversion = 0;
+    gf_dirent_t *entry = NULL;
+    br_private_t *priv = NULL;
+    loc_t *parent = NULL;
+    gf_boolean_t skip_stat = _gf_false;
+    uuid_t shard_root_gfid = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("bit-rot", fsentry, out);
+
+    entry = fsentry->entry;
+    parent = &fsentry->parent;
+    child = fsentry->data;
+
+    priv = this->private;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", entry, out);
+    GF_VALIDATE_OR_GOTO("bit-rot", parent, out);
+    GF_VALIDATE_OR_GOTO("bit-rot", child, out);
+    GF_VALIDATE_OR_GOTO("bit-rot", priv, out);
+
+    pid = GF_CLIENT_PID_SCRUB;
+
+    ret = br_prepare_loc(this, child, parent, entry, &loc);
+    if (!ret)
+        goto out;
+
+    syncopctx_setfspid(&pid);
+
+    ret = syncop_lookup(child->xl, &loc, &iatt, &parent_buf, NULL, NULL);
+    if (ret) {
+        br_log_object_path(this, "lookup", loc.path, -ret);
+        goto out;
+    }
+
+    linked_inode = inode_link(loc.inode, parent->inode, loc.name, &iatt);
+    if (linked_inode)
+        inode_lookup(linked_inode);
+
+    gf_msg_debug(this->name, 0, "Scrubbing object %s [GFID: %s]", entry->d_name,
+                 uuid_utoa(linked_inode->gfid));
+
+    if (iatt.ia_type != IA_IFREG) {
+        gf_msg_debug(this->name, 0, "%s is not a regular file", entry->d_name);
+        ret = 0;
+        goto unref_inode;
+    }
+
+    if (IS_DHT_LINKFILE_MODE((&iatt))) {
+        gf_msg_debug(this->name, 0, "%s is a dht sticky bit file",
+                     entry->d_name);
+        ret = 0;
+        goto unref_inode;
+    }
+
+    /* skip updating scrub statistics for shard entries */
+    gf_uuid_parse(SHARD_ROOT_GFID, shard_root_gfid);
+    if (gf_uuid_compare(loc.pargfid, shard_root_gfid) == 0)
+        skip_stat = _gf_true;
+
+    /**
+     * open() an fd for subsequent operations
+     */
+    fd = fd_create(linked_inode, 0);
+    if (!fd) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_FD_CREATE_FAILED,
+               "failed to create fd for inode %s",
+               uuid_utoa(linked_inode->gfid));
+        goto unref_inode;
+    }
+
+    ret = syncop_open(child->xl, &loc, O_RDWR, fd, NULL, NULL);
+    if (ret) {
+        br_log_object(this, "open", linked_inode->gfid, -ret);
+        ret = -1;
+        goto unrefd;
+    }
+
+    fd_bind(fd);
+
+    /**
+     * perform pre compute checks before initiating checksum
+     * computation
+     *  - presence of bad object
+     *  - signature staleness
+     */
+    ret = bitd_scrub_pre_compute_check(this, child, fd, &signedversion,
+                                       &priv->scrub_stat, skip_stat);
+    if (ret)
+        goto unrefd; /* skip this object */
+
+    /* if all's good, proceed to calculate the hash */
+    md = GF_MALLOC(SHA256_DIGEST_LENGTH, gf_common_mt_char);
+    if (!md)
+        goto unrefd;
+
+    ret = br_calculate_obj_checksum(md, child, fd, &iatt);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_CALC_ERROR,
+               "error calculating hash for object [GFID: %s]",
+               uuid_utoa(fd->inode->gfid));
+        ret = -1;
+        goto free_md;
+    }
+
+    /**
+     * perform post compute checks as an object's signature may have
+     * become stale while scrubber calculated checksum.
+     */
+    ret = bitd_scrub_post_compute_check(this, child, fd, signedversion, &sign,
+                                        &priv->scrub_stat, skip_stat);
+    if (ret)
+        goto free_md;
+
+    ret = bitd_compare_ckum(this, sign, md, linked_inode, entry, fd, child,
+                            &loc);
+
+    if (!skip_stat)
+        br_inc_scrubbed_file(&priv->scrub_stat);
+
+    GF_FREE(sign); /* allocated on post-compute */
+
+    /** fd_unref() takes care of closing fd.. like syncop_close() */
+
+free_md:
+    GF_FREE(md);
+unrefd:
+    fd_unref(fd);
+unref_inode:
+    inode_unref(linked_inode);
+out:
+    loc_wipe(&loc);
+    return ret;
+}
+
+static void
+_br_lock_cleaner(void *arg)
+{
+    pthread_mutex_t *mutex = arg;
+
+    pthread_mutex_unlock(mutex);
+}
+
+static void
+wait_for_scrubbing(xlator_t *this, struct br_scanfs *fsscan)
+{
+    br_private_t *priv = NULL;
+    struct br_scrubber *fsscrub = NULL;
+
+    priv = this->private;
+    fsscrub = &priv->fsscrub;
+
+    pthread_cleanup_push(_br_lock_cleaner, &fsscan->waitlock);
+    pthread_mutex_lock(&fsscan->waitlock);
+    {
+        pthread_cleanup_push(_br_lock_cleaner, &fsscrub->mutex);
+        pthread_mutex_lock(&fsscrub->mutex);
+        {
+            list_replace_init(&fsscan->queued, &fsscan->ready);
+
+            /* wake up scrubbers */
+            pthread_cond_broadcast(&fsscrub->cond);
+        }
+        pthread_mutex_unlock(&fsscrub->mutex);
+        pthread_cleanup_pop(0);
+
+        while (fsscan->entries != 0)
+            pthread_cond_wait(&fsscan->waitcond, &fsscan->waitlock);
+    }
+    pthread_mutex_unlock(&fsscan->waitlock);
+    pthread_cleanup_pop(0);
+}
+
+static void
+_br_fsscan_inc_entry_count(struct br_scanfs *fsscan)
+{
+    fsscan->entries++;
+}
+
+static void
+_br_fsscan_dec_entry_count(struct br_scanfs *fsscan)
+{
+    if (--fsscan->entries == 0) {
+        pthread_mutex_lock(&fsscan->waitlock);
+        {
+            pthread_cond_signal(&fsscan->waitcond);
+        }
+        pthread_mutex_unlock(&fsscan->waitlock);
+    }
+}
+
+static void
+_br_fsscan_collect_entry(struct br_scanfs *fsscan,
+                         struct br_fsscan_entry *fsentry)
+{
+    list_add_tail(&fsentry->list, &fsscan->queued);
+    _br_fsscan_inc_entry_count(fsscan);
+}
+
+#define NR_ENTRIES (1 << 7) /* ..bulk scrubbing */
+
+int
+br_fsscanner_handle_entry(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent,
+                          void *data)
+{
+    int32_t ret = -1;
+    int scrub = 0;
+    br_child_t *child = NULL;
+    xlator_t *this = NULL;
+    struct br_scanfs *fsscan = NULL;
+    struct br_fsscan_entry *fsentry = NULL;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", subvol, error_return);
+    GF_VALIDATE_OR_GOTO("bit-rot", data, error_return);
+
+    child = data;
+    this = child->this;
+    fsscan = &child->fsscan;
+
+    _mask_cancellation();
+
+    fsentry = GF_CALLOC(1, sizeof(*fsentry), gf_br_mt_br_fsscan_entry_t);
+    if (!fsentry)
+        goto error_return;
+
+    {
+        fsentry->data = data;
+        fsentry->fsscan = &child->fsscan;
+
+        /* copy parent loc */
+        ret = loc_copy(&fsentry->parent, parent);
+        if (ret)
+            goto dealloc;
+
+        /* copy child entry */
+        fsentry->entry = entry_copy(entry);
+        if (!fsentry->entry)
+            goto locwipe;
+
+        INIT_LIST_HEAD(&fsentry->list);
+    }
+
+    LOCK(&fsscan->entrylock);
+    {
+        _br_fsscan_collect_entry(fsscan, fsentry);
+
+        /**
+         * need not be a equality check as entries may be pushed
+         * back onto the scanned queue when thread(s) are cleaned.
+         */
+        if (fsscan->entries >= NR_ENTRIES)
+            scrub = 1;
+    }
+    UNLOCK(&fsscan->entrylock);
+
+    _unmask_cancellation();
+
+    if (scrub)
+        wait_for_scrubbing(this, fsscan);
+
+    return 0;
+
+locwipe:
+    loc_wipe(&fsentry->parent);
+dealloc:
+    GF_FREE(fsentry);
+error_return:
+    return -1;
+}
+
+int32_t
+br_fsscan_deactivate(xlator_t *this)
+{
+    int ret = 0;
+    br_private_t *priv = NULL;
+    br_scrub_state_t nstate = 0;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    ret = gf_tw_del_timer(priv->timer_wheel, scrub_monitor->timer);
+    if (ret == 0) {
+        nstate = BR_SCRUB_STATE_STALLED;
+        gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO,
+               "Volume is under active scrubbing. Pausing scrub..");
+    } else {
+        nstate = BR_SCRUB_STATE_PAUSED;
+        gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO,
+               "Scrubber paused");
+    }
+
+    _br_monitor_set_scrub_state(scrub_monitor, nstate);
+
+    return 0;
+}
+
+static void
+br_scrubber_log_time(xlator_t *this, const char *sfx)
+{
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    br_private_t *priv = NULL;
+    time_t now = 0;
+
+    now = gf_time();
+    priv = this->private;
+
+    gf_time_fmt(timestr, sizeof(timestr), now, gf_timefmt_FT);
+
+    if (strcasecmp(sfx, "started") == 0) {
+        br_update_scrub_start_time(&priv->scrub_stat, now);
+        gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_START,
+               "Scrubbing %s at %s", sfx, timestr);
+    } else {
+        br_update_scrub_finish_time(&priv->scrub_stat, timestr, now);
+        gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_FINISH,
+               "Scrubbing %s at %s", sfx, timestr);
+    }
+}
+
+static void
+br_fsscanner_log_time(xlator_t *this, br_child_t *child, const char *sfx)
+{
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    time_t now = 0;
+
+    now = gf_time();
+    gf_time_fmt(timestr, sizeof(timestr), now, gf_timefmt_FT);
+
+    if (strcasecmp(sfx, "started") == 0) {
+        gf_msg_debug(this->name, 0, "Scrubbing \"%s\" %s at %s",
+                     child->brick_path, sfx, timestr);
+    } else {
+        gf_msg_debug(this->name, 0, "Scrubbing \"%s\" %s at %s",
+                     child->brick_path, sfx, timestr);
+    }
+}
+
+void
+br_child_set_scrub_state(br_child_t *child, gf_boolean_t state)
+{
+    child->active_scrubbing = state;
+}
+
+static void
+br_fsscanner_wait_until_kicked(xlator_t *this, br_child_t *child)
+{
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    pthread_cleanup_push(_br_lock_cleaner, &scrub_monitor->wakelock);
+    pthread_mutex_lock(&scrub_monitor->wakelock);
+    {
+        while (!scrub_monitor->kick)
+            pthread_cond_wait(&scrub_monitor->wakecond,
+                              &scrub_monitor->wakelock);
+
+        /* Child lock is to synchronize with disconnect events */
+        pthread_cleanup_push(_br_lock_cleaner, &child->lock);
+        pthread_mutex_lock(&child->lock);
+        {
+            scrub_monitor->active_child_count++;
+            br_child_set_scrub_state(child, _gf_true);
+        }
+        pthread_mutex_unlock(&child->lock);
+        pthread_cleanup_pop(0);
+    }
+    pthread_mutex_unlock(&scrub_monitor->wakelock);
+    pthread_cleanup_pop(0);
+}
+
+static void
+br_scrubber_entry_control(xlator_t *this)
+{
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    LOCK(&scrub_monitor->lock);
+    {
+        /* Move the state to BR_SCRUB_STATE_ACTIVE */
+        if (scrub_monitor->state == BR_SCRUB_STATE_PENDING)
+            scrub_monitor->state = BR_SCRUB_STATE_ACTIVE;
+        br_scrubber_log_time(this, "started");
+        priv->scrub_stat.scrub_running = 1;
+    }
+    UNLOCK(&scrub_monitor->lock);
+}
+
+static void
+br_scrubber_exit_control(xlator_t *this)
+{
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    LOCK(&scrub_monitor->lock);
+    {
+        br_scrubber_log_time(this, "finished");
+        priv->scrub_stat.scrub_running = 0;
+
+        if (scrub_monitor->state == BR_SCRUB_STATE_ACTIVE) {
+            (void)br_fsscan_activate(this);
+        } else {
+            UNLOCK(&scrub_monitor->lock);
+            gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO,
+                   "Volume waiting to get rescheduled..");
+            return;
+        }
+    }
+    UNLOCK(&scrub_monitor->lock);
+}
+
+static void
+br_fsscanner_entry_control(xlator_t *this, br_child_t *child)
+{
+    br_fsscanner_log_time(this, child, "started");
+}
+
+static void
+br_fsscanner_exit_control(xlator_t *this, br_child_t *child)
+{
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    if (!_br_is_child_connected(child)) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, BRB_MSG_SCRUB_INFO,
+               "Brick [%s] disconnected while scrubbing. Scrubbing "
+               "might be incomplete",
+               child->brick_path);
+    }
+
+    br_fsscanner_log_time(this, child, "finished");
+
+    pthread_cleanup_push(_br_lock_cleaner, &scrub_monitor->wakelock);
+    pthread_mutex_lock(&scrub_monitor->wakelock);
+    {
+        scrub_monitor->active_child_count--;
+        pthread_cleanup_push(_br_lock_cleaner, &child->lock);
+        pthread_mutex_lock(&child->lock);
+        {
+            br_child_set_scrub_state(child, _gf_false);
+        }
+        pthread_mutex_unlock(&child->lock);
+        pthread_cleanup_pop(0);
+
+        if (scrub_monitor->active_child_count == 0) {
+            /* The last child has finished scrubbing.
+             * Set the kick to false and  wake up other
+             * children who are waiting for the last
+             * child to complete scrubbing.
+             */
+            scrub_monitor->kick = _gf_false;
+            pthread_cond_broadcast(&scrub_monitor->wakecond);
+
+            /* Signal monitor thread waiting for the all
+             * the children to finish scrubbing.
+             */
+            pthread_cleanup_push(_br_lock_cleaner, &scrub_monitor->donelock);
+            pthread_mutex_lock(&scrub_monitor->donelock);
+            {
+                scrub_monitor->done = _gf_true;
+                pthread_cond_signal(&scrub_monitor->donecond);
+            }
+            pthread_mutex_unlock(&scrub_monitor->donelock);
+            pthread_cleanup_pop(0);
+        } else {
+            while (scrub_monitor->active_child_count)
+                pthread_cond_wait(&scrub_monitor->wakecond,
+                                  &scrub_monitor->wakelock);
+        }
+    }
+    pthread_mutex_unlock(&scrub_monitor->wakelock);
+    pthread_cleanup_pop(0);
+}
+
+void *
+br_fsscanner(void *arg)
+{
+    loc_t loc = {
+        0,
+    };
+    br_child_t *child = NULL;
+    xlator_t *this = NULL;
+    struct br_scanfs *fsscan = NULL;
+
+    child = arg;
+    this = child->this;
+    fsscan = &child->fsscan;
+
+    THIS = this;
+    loc.inode = child->table->root;
+
+    while (1) {
+        br_fsscanner_wait_until_kicked(this, child);
+        {
+            /* precursor for scrub */
+            br_fsscanner_entry_control(this, child);
+
+            /* scrub */
+            (void)syncop_ftw(child->xl, &loc, GF_CLIENT_PID_SCRUB, child,
+                             br_fsscanner_handle_entry);
+            if (!list_empty(&fsscan->queued))
+                wait_for_scrubbing(this, fsscan);
+
+            /* scrub exit criteria */
+            br_fsscanner_exit_control(this, child);
+        }
+    }
+
+    return NULL;
+}
+
+/**
+ * Keep this routine extremely simple and do not ever try to acquire
+ * child->lock here: it may lead to deadlock. Scrubber state is
+ * modified in br_fsscanner(). An intermediate state change to pause
+ * changes the scrub state to the _correct_ state by identifying a
+ * non-pending timer.
+ */
+void
+br_kickstart_scanner(struct gf_tw_timer_list *timer, void *data,
+                     unsigned long calltime)
+{
+    xlator_t *this = NULL;
+    struct br_monitor *scrub_monitor = data;
+    br_private_t *priv = NULL;
+
+    THIS = this = scrub_monitor->this;
+    priv = this->private;
+
+    /* Reset scrub statistics */
+    priv->scrub_stat.scrubbed_files = 0;
+    priv->scrub_stat.unsigned_files = 0;
+
+    /* Moves state from PENDING to ACTIVE */
+    (void)br_scrubber_entry_control(this);
+
+    /* kickstart scanning.. */
+    pthread_mutex_lock(&scrub_monitor->wakelock);
+    {
+        scrub_monitor->kick = _gf_true;
+        GF_ASSERT(scrub_monitor->active_child_count == 0);
+        pthread_cond_broadcast(&scrub_monitor->wakecond);
+    }
+    pthread_mutex_unlock(&scrub_monitor->wakelock);
+
+    return;
+}
+
+static uint32_t
+br_fsscan_calculate_delta(uint32_t times)
+{
+    return times;
+}
+
+#define BR_SCRUB_ONDEMAND (1)
+#define BR_SCRUB_MINUTE (60)
+#define BR_SCRUB_HOURLY (60 * 60)
+#define BR_SCRUB_DAILY (1 * 24 * 60 * 60)
+#define BR_SCRUB_WEEKLY (7 * 24 * 60 * 60)
+#define BR_SCRUB_BIWEEKLY (14 * 24 * 60 * 60)
+#define BR_SCRUB_MONTHLY (30 * 24 * 60 * 60)
+
+static unsigned int
+br_fsscan_calculate_timeout(scrub_freq_t freq)
+{
+    uint32_t timo = 0;
+
+    switch (freq) {
+        case BR_FSSCRUB_FREQ_MINUTE:
+            timo = br_fsscan_calculate_delta(BR_SCRUB_MINUTE);
+            break;
+        case BR_FSSCRUB_FREQ_HOURLY:
+            timo = br_fsscan_calculate_delta(BR_SCRUB_HOURLY);
+            break;
+        case BR_FSSCRUB_FREQ_DAILY:
+            timo = br_fsscan_calculate_delta(BR_SCRUB_DAILY);
+            break;
+        case BR_FSSCRUB_FREQ_WEEKLY:
+            timo = br_fsscan_calculate_delta(BR_SCRUB_WEEKLY);
+            break;
+        case BR_FSSCRUB_FREQ_BIWEEKLY:
+            timo = br_fsscan_calculate_delta(BR_SCRUB_BIWEEKLY);
+            break;
+        case BR_FSSCRUB_FREQ_MONTHLY:
+            timo = br_fsscan_calculate_delta(BR_SCRUB_MONTHLY);
+            break;
+        default:
+            timo = 0;
+    }
+
+    return timo;
+}
+
+int32_t
+br_fsscan_schedule(xlator_t *this)
+{
+    uint32_t timo = 0;
+    br_private_t *priv = NULL;
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    struct br_scrubber *fsscrub = NULL;
+    struct gf_tw_timer_list *timer = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    fsscrub = &priv->fsscrub;
+    scrub_monitor = &priv->scrub_monitor;
+
+    scrub_monitor->boot = gf_time();
+
+    timo = br_fsscan_calculate_timeout(fsscrub->frequency);
+    if (timo == 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_ZERO_TIMEOUT_BUG,
+               "BUG: Zero schedule timeout");
+        goto error_return;
+    }
+
+    scrub_monitor->timer = GF_CALLOC(1, sizeof(*scrub_monitor->timer),
+                                     gf_br_stub_mt_br_scanner_freq_t);
+    if (!scrub_monitor->timer)
+        goto error_return;
+
+    timer = scrub_monitor->timer;
+    INIT_LIST_HEAD(&timer->entry);
+
+    timer->data = scrub_monitor;
+    timer->expires = timo;
+    timer->function = br_kickstart_scanner;
+
+    gf_tw_add_timer(priv->timer_wheel, timer);
+    _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_PENDING);
+
+    gf_time_fmt(timestr, sizeof(timestr), (scrub_monitor->boot + timo),
+                gf_timefmt_FT);
+    gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO,
+           "Scrubbing is "
+           "scheduled to run at %s",
+           timestr);
+
+    return 0;
+
+error_return:
+    return -1;
+}
+
+int32_t
+br_fsscan_activate(xlator_t *this)
+{
+    uint32_t timo = 0;
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    time_t now = 0;
+    br_private_t *priv = NULL;
+    struct br_scrubber *fsscrub = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    fsscrub = &priv->fsscrub;
+    scrub_monitor = &priv->scrub_monitor;
+
+    now = gf_time();
+    timo = br_fsscan_calculate_timeout(fsscrub->frequency);
+    if (timo == 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_ZERO_TIMEOUT_BUG,
+               "BUG: Zero schedule timeout");
+        return -1;
+    }
+
+    pthread_mutex_lock(&scrub_monitor->donelock);
+    {
+        scrub_monitor->done = _gf_false;
+    }
+    pthread_mutex_unlock(&scrub_monitor->donelock);
+
+    gf_time_fmt(timestr, sizeof(timestr), now + timo, gf_timefmt_FT);
+    (void)gf_tw_mod_timer(priv->timer_wheel, scrub_monitor->timer, timo);
+
+    _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_PENDING);
+    gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO,
+           "Scrubbing is "
+           "rescheduled to run at %s",
+           timestr);
+
+    return 0;
+}
+
+int32_t
+br_fsscan_reschedule(xlator_t *this)
+{
+    int32_t ret = 0;
+    uint32_t timo = 0;
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    time_t now = 0;
+    br_private_t *priv = NULL;
+    struct br_scrubber *fsscrub = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    fsscrub = &priv->fsscrub;
+    scrub_monitor = &priv->scrub_monitor;
+
+    if (!fsscrub->frequency_reconf)
+        return 0;
+
+    now = gf_time();
+    timo = br_fsscan_calculate_timeout(fsscrub->frequency);
+    if (timo == 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_ZERO_TIMEOUT_BUG,
+               "BUG: Zero schedule timeout");
+        return -1;
+    }
+
+    gf_time_fmt(timestr, sizeof(timestr), now + timo, gf_timefmt_FT);
+
+    pthread_mutex_lock(&scrub_monitor->donelock);
+    {
+        scrub_monitor->done = _gf_false;
+    }
+    pthread_mutex_unlock(&scrub_monitor->donelock);
+
+    ret = gf_tw_mod_timer_pending(priv->timer_wheel, scrub_monitor->timer,
+                                  timo);
+    if (ret == 0)
+        gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO,
+               "Scrubber is currently running and would be "
+               "rescheduled after completion");
+    else {
+        _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_PENDING);
+        gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO,
+               "Scrubbing rescheduled to run at %s", timestr);
+    }
+
+    return 0;
+}
+
+int32_t
+br_fsscan_ondemand(xlator_t *this)
+{
+    int32_t ret = 0;
+    uint32_t timo = 0;
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    time_t now = 0;
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    now = gf_time();
+    timo = BR_SCRUB_ONDEMAND;
+    gf_time_fmt(timestr, sizeof(timestr), now + timo, gf_timefmt_FT);
+
+    pthread_mutex_lock(&scrub_monitor->donelock);
+    {
+        scrub_monitor->done = _gf_false;
+    }
+    pthread_mutex_unlock(&scrub_monitor->donelock);
+
+    ret = gf_tw_mod_timer_pending(priv->timer_wheel, scrub_monitor->timer,
+                                  timo);
+    if (ret == 0)
+        gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO,
+               "Scrubber is currently running and would be "
+               "rescheduled after completion");
+    else {
+        _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_PENDING);
+        gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO,
+               "Ondemand Scrubbing scheduled to run at %s", timestr);
+    }
+
+    return 0;
+}
+
+#define BR_SCRUB_THREAD_SCALE_LAZY 0
+#define BR_SCRUB_THREAD_SCALE_NORMAL 0.4
+#define BR_SCRUB_THREAD_SCALE_AGGRESSIVE 1.0
+
+#ifndef M_E
+#define M_E 2.718
+#endif
+
+/**
+ * This is just a simple exponential scale to a fixed value selected
+ * per throttle config. We probably need to be more smart and select
+ * the scale based on the number of processor cores too.
+ */
+static unsigned int
+br_scrubber_calc_scale(xlator_t *this, br_private_t *priv,
+                       scrub_throttle_t throttle)
+{
+    unsigned int scale = 0;
+
+    switch (throttle) {
+        case BR_SCRUB_THROTTLE_VOID:
+        case BR_SCRUB_THROTTLE_STALLED:
+            scale = 0;
+            break;
+        case BR_SCRUB_THROTTLE_LAZY:
+            scale = priv->child_count * pow(M_E, BR_SCRUB_THREAD_SCALE_LAZY);
+            break;
+        case BR_SCRUB_THROTTLE_NORMAL:
+            scale = priv->child_count * pow(M_E, BR_SCRUB_THREAD_SCALE_NORMAL);
+            break;
+        case BR_SCRUB_THROTTLE_AGGRESSIVE:
+            scale = priv->child_count *
+                    pow(M_E, BR_SCRUB_THREAD_SCALE_AGGRESSIVE);
+            break;
+        default:
+            gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_UNKNOWN_THROTTLE,
+                   "Unknown throttle %d", throttle);
+    }
+
+    return scale;
+}
+
+static br_child_t *
+_br_scrubber_get_next_child(struct br_scrubber *fsscrub)
+{
+    br_child_t *child = NULL;
+
+    child = list_first_entry(&fsscrub->scrublist, br_child_t, list);
+    list_rotate_left(&fsscrub->scrublist);
+
+    return child;
+}
+
+static void
+_br_scrubber_get_entry(br_child_t *child, struct br_fsscan_entry **fsentry)
+{
+    struct br_scanfs *fsscan = &child->fsscan;
+
+    if (list_empty(&fsscan->ready))
+        return;
+    *fsentry = list_first_entry(&fsscan->ready, struct br_fsscan_entry, list);
+    list_del_init(&(*fsentry)->list);
+}
+
+static void
+_br_scrubber_find_scrubbable_entry(struct br_scrubber *fsscrub,
+                                   struct br_fsscan_entry **fsentry)
+{
+    br_child_t *child = NULL;
+    br_child_t *firstchild = NULL;
+
+    while (1) {
+        while (list_empty(&fsscrub->scrublist))
+            pthread_cond_wait(&fsscrub->cond, &fsscrub->mutex);
+
+        firstchild = NULL;
+        for (child = _br_scrubber_get_next_child(fsscrub); child != firstchild;
+             child = _br_scrubber_get_next_child(fsscrub)) {
+            if (!firstchild)
+                firstchild = child;
+
+            _br_scrubber_get_entry(child, fsentry);
+            if (*fsentry)
+                break;
+        }
+
+        if (*fsentry)
+            break;
+
+        /* nothing to work on.. wait till available */
+        pthread_cond_wait(&fsscrub->cond, &fsscrub->mutex);
+    }
+}
+
+static void
+br_scrubber_pick_entry(struct br_scrubber *fsscrub,
+                       struct br_fsscan_entry **fsentry)
+{
+    pthread_cleanup_push(_br_lock_cleaner, &fsscrub->mutex);
+
+    pthread_mutex_lock(&fsscrub->mutex);
+    {
+        *fsentry = NULL;
+        _br_scrubber_find_scrubbable_entry(fsscrub, fsentry);
+    }
+    pthread_mutex_unlock(&fsscrub->mutex);
+
+    pthread_cleanup_pop(0);
+}
+
+struct br_scrub_entry {
+    gf_boolean_t scrubbed;
+    struct br_fsscan_entry *fsentry;
+};
+
+/**
+ * We need to be a bit careful here. These thread(s) are prone to cancellations
+ * when threads are scaled down (depending on the thottling value configured)
+ * and pausing scrub. A thread can get cancelled while it's waiting for entries
+ * in the ->pending queue or when an object is undergoing scrubbing.
+ */
+static void
+br_scrubber_entry_handle(void *arg)
+{
+    struct br_scanfs *fsscan = NULL;
+    struct br_scrub_entry *sentry = NULL;
+    struct br_fsscan_entry *fsentry = NULL;
+
+    sentry = arg;
+
+    fsentry = sentry->fsentry;
+    fsscan = fsentry->fsscan;
+
+    LOCK(&fsscan->entrylock);
+    {
+        if (sentry->scrubbed) {
+            _br_fsscan_dec_entry_count(fsscan);
+
+            /* cleanup ->entry */
+            fsentry->data = NULL;
+            fsentry->fsscan = NULL;
+            loc_wipe(&fsentry->parent);
+            gf_dirent_entry_free(fsentry->entry);
+
+            GF_FREE(sentry->fsentry);
+        } else {
+            /* (re)queue the entry again for scrub */
+            _br_fsscan_collect_entry(fsscan, sentry->fsentry);
+        }
+    }
+    UNLOCK(&fsscan->entrylock);
+}
+
+static void
+br_scrubber_scrub_entry(xlator_t *this, struct br_fsscan_entry *fsentry)
+{
+    struct br_scrub_entry sentry = {
+        0,
+    };
+
+    sentry.scrubbed = 0;
+    sentry.fsentry = fsentry;
+
+    pthread_cleanup_push(br_scrubber_entry_handle, &sentry);
+    {
+        (void)br_scrubber_scrub_begin(this, fsentry);
+        sentry.scrubbed = 1;
+    }
+    pthread_cleanup_pop(1);
+}
+
+void *
+br_scrubber_proc(void *arg)
+{
+    xlator_t *this = NULL;
+    struct br_scrubber *fsscrub = NULL;
+    struct br_fsscan_entry *fsentry = NULL;
+
+    fsscrub = arg;
+    THIS = this = fsscrub->this;
+
+    while (1) {
+        br_scrubber_pick_entry(fsscrub, &fsentry);
+        br_scrubber_scrub_entry(this, fsentry);
+        sleep(1);
+    }
+
+    return NULL;
+}
+
+static int32_t
+br_scrubber_scale_up(xlator_t *this, struct br_scrubber *fsscrub,
+                     unsigned int v1, unsigned int v2)
+{
+    int i = 0;
+    int32_t ret = -1;
+    int diff = 0;
+    struct br_scrubbers *scrub = NULL;
+
+    diff = (int)(v2 - v1);
+
+    gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCALING_UP_SCRUBBER,
+           "Scaling up scrubbers [%d => %d]", v1, v2);
+
+    for (i = 0; i < diff; i++) {
+        scrub = GF_CALLOC(diff, sizeof(*scrub), gf_br_mt_br_scrubber_t);
+        if (!scrub)
+            break;
+
+        INIT_LIST_HEAD(&scrub->list);
+        ret = gf_thread_create(&scrub->scrubthread, NULL, br_scrubber_proc,
+                               fsscrub, "brsproc");
+        if (ret)
+            break;
+
+        fsscrub->nr_scrubbers++;
+        list_add_tail(&scrub->list, &fsscrub->scrubbers);
+    }
+
+    if ((i != diff) && !scrub)
+        goto error_return;
+
+    if (i != diff) /* degraded scaling.. */
+        gf_msg(this->name, GF_LOG_WARNING, 0, BRB_MSG_SCALE_UP_FAILED,
+               "Could not fully scale up to %d scrubber(s). Spawned "
+               "%d/%d [total scrubber(s): %d]",
+               v2, i, diff, (v1 + i));
+
+    return 0;
+
+error_return:
+    return -1;
+}
+
+static int32_t
+br_scrubber_scale_down(xlator_t *this, struct br_scrubber *fsscrub,
+                       unsigned int v1, unsigned int v2)
+{
+    int i = 0;
+    int diff = 0;
+    int32_t ret = -1;
+    struct br_scrubbers *scrub = NULL;
+
+    diff = (int)(v1 - v2);
+
+    gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCALE_DOWN_SCRUBBER,
+           "Scaling down scrubbers [%d => %d]", v1, v2);
+
+    for (i = 0; i < diff; i++) {
+        scrub = list_first_entry(&fsscrub->scrubbers, struct br_scrubbers,
+                                 list);
+
+        list_del_init(&scrub->list);
+        ret = gf_thread_cleanup_xint(scrub->scrubthread);
+        if (ret)
+            break;
+        GF_FREE(scrub);
+
+        fsscrub->nr_scrubbers--;
+    }
+
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, BRB_MSG_SCALE_DOWN_FAILED,
+               "Could not fully scale down "
+               "to %d scrubber(s). Terminated %d/%d [total "
+               "scrubber(s): %d]",
+               v1, i, diff, (v2 - i));
+        ret = 0;
+    }
+
+    return ret;
+}
+
+static int32_t
+br_scrubber_configure(xlator_t *this, br_private_t *priv,
+                      struct br_scrubber *fsscrub, scrub_throttle_t nthrottle)
+{
+    int32_t ret = 0;
+    unsigned int v1 = 0;
+    unsigned int v2 = 0;
+
+    v1 = fsscrub->nr_scrubbers;
+    v2 = br_scrubber_calc_scale(this, priv, nthrottle);
+
+    if (v1 == v2)
+        return 0;
+
+    if (v1 > v2)
+        ret = br_scrubber_scale_down(this, fsscrub, v1, v2);
+    else
+        ret = br_scrubber_scale_up(this, fsscrub, v1, v2);
+
+    return ret;
+}
+
+static int32_t
+br_scrubber_fetch_option(xlator_t *this, char *opt, dict_t *options,
+                         char **value)
+{
+    if (options)
+        GF_OPTION_RECONF(opt, *value, options, str, error_return);
+    else
+        GF_OPTION_INIT(opt, *value, str, error_return);
+
+    return 0;
+
+error_return:
+    return -1;
+}
+
+/* internal "throttle" override */
+#define BR_SCRUB_STALLED "STALLED"
+
+/* TODO: token buket spec */
+static int32_t
+br_scrubber_handle_throttle(xlator_t *this, br_private_t *priv, dict_t *options,
+                            gf_boolean_t scrubstall)
+{
+    int32_t ret = 0;
+    char *tmp = NULL;
+    struct br_scrubber *fsscrub = NULL;
+    scrub_throttle_t nthrottle = BR_SCRUB_THROTTLE_VOID;
+
+    fsscrub = &priv->fsscrub;
+    fsscrub->throttle_reconf = _gf_false;
+
+    ret = br_scrubber_fetch_option(this, "scrub-throttle", options, &tmp);
+    if (ret)
+        goto error_return;
+
+    if (scrubstall)
+        tmp = BR_SCRUB_STALLED;
+
+    if (strcasecmp(tmp, "lazy") == 0)
+        nthrottle = BR_SCRUB_THROTTLE_LAZY;
+    else if (strcasecmp(tmp, "normal") == 0)
+        nthrottle = BR_SCRUB_THROTTLE_NORMAL;
+    else if (strcasecmp(tmp, "aggressive") == 0)
+        nthrottle = BR_SCRUB_THROTTLE_AGGRESSIVE;
+    else if (strcasecmp(tmp, BR_SCRUB_STALLED) == 0)
+        nthrottle = BR_SCRUB_THROTTLE_STALLED;
+    else
+        goto error_return;
+
+    /* on failure old throttling value is preserved */
+    ret = br_scrubber_configure(this, priv, fsscrub, nthrottle);
+    if (ret)
+        goto error_return;
+
+    if (fsscrub->throttle != nthrottle)
+        fsscrub->throttle_reconf = _gf_true;
+
+    fsscrub->throttle = nthrottle;
+    return 0;
+
+error_return:
+    return -1;
+}
+
+static int32_t
+br_scrubber_handle_stall(xlator_t *this, br_private_t *priv, dict_t *options,
+                         gf_boolean_t *scrubstall)
+{
+    int32_t ret = 0;
+    char *tmp = NULL;
+
+    ret = br_scrubber_fetch_option(this, "scrub-state", options, &tmp);
+    if (ret)
+        goto error_return;
+
+    if (strcasecmp(tmp, "pause") == 0) /* anything else is active */
+        *scrubstall = _gf_true;
+
+    return 0;
+
+error_return:
+    return -1;
+}
+
+static int32_t
+br_scrubber_handle_freq(xlator_t *this, br_private_t *priv, dict_t *options,
+                        gf_boolean_t scrubstall)
+{
+    int32_t ret = -1;
+    char *tmp = NULL;
+    scrub_freq_t frequency = BR_FSSCRUB_FREQ_HOURLY;
+    struct br_scrubber *fsscrub = NULL;
+
+    fsscrub = &priv->fsscrub;
+    fsscrub->frequency_reconf = _gf_true;
+
+    ret = br_scrubber_fetch_option(this, "scrub-freq", options, &tmp);
+    if (ret)
+        goto error_return;
+
+    if (scrubstall)
+        tmp = BR_SCRUB_STALLED;
+
+    if (strcasecmp(tmp, "hourly") == 0) {
+        frequency = BR_FSSCRUB_FREQ_HOURLY;
+    } else if (strcasecmp(tmp, "daily") == 0) {
+        frequency = BR_FSSCRUB_FREQ_DAILY;
+    } else if (strcasecmp(tmp, "weekly") == 0) {
+        frequency = BR_FSSCRUB_FREQ_WEEKLY;
+    } else if (strcasecmp(tmp, "biweekly") == 0) {
+        frequency = BR_FSSCRUB_FREQ_BIWEEKLY;
+    } else if (strcasecmp(tmp, "monthly") == 0) {
+        frequency = BR_FSSCRUB_FREQ_MONTHLY;
+    } else if (strcasecmp(tmp, "minute") == 0) {
+        frequency = BR_FSSCRUB_FREQ_MINUTE;
+    } else if (strcasecmp(tmp, BR_SCRUB_STALLED) == 0) {
+        frequency = BR_FSSCRUB_FREQ_STALLED;
+    } else
+        goto error_return;
+
+    if (fsscrub->frequency == frequency)
+        fsscrub->frequency_reconf = _gf_false;
+    else
+        fsscrub->frequency = frequency;
+
+    return 0;
+
+error_return:
+    return -1;
+}
+
+static void
+br_scrubber_log_option(xlator_t *this, br_private_t *priv,
+                       gf_boolean_t scrubstall)
+{
+    struct br_scrubber *fsscrub = &priv->fsscrub;
+    char *scrub_throttle_str[] = {
+        [BR_SCRUB_THROTTLE_LAZY] = "lazy",
+        [BR_SCRUB_THROTTLE_NORMAL] = "normal",
+        [BR_SCRUB_THROTTLE_AGGRESSIVE] = "aggressive",
+        [BR_SCRUB_THROTTLE_STALLED] = "stalled",
+    };
+
+    char *scrub_freq_str[] = {
+        [0] = "",
+        [BR_FSSCRUB_FREQ_HOURLY] = "hourly",
+        [BR_FSSCRUB_FREQ_DAILY] = "daily",
+        [BR_FSSCRUB_FREQ_WEEKLY] = "weekly",
+        [BR_FSSCRUB_FREQ_BIWEEKLY] = "biweekly",
+        [BR_FSSCRUB_FREQ_MONTHLY] = "monthly (30 days)",
+        [BR_FSSCRUB_FREQ_MINUTE] = "every minute",
+    };
+
+    if (scrubstall)
+        return; /* logged as pause */
+
+    if (fsscrub->frequency_reconf || fsscrub->throttle_reconf) {
+        if (fsscrub->throttle == BR_SCRUB_THROTTLE_VOID)
+            return;
+        gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_TUNABLE,
+               "SCRUB TUNABLES:: [Frequency: %s, Throttle: %s]",
+               scrub_freq_str[fsscrub->frequency],
+               scrub_throttle_str[fsscrub->throttle]);
+    }
+}
+
+int32_t
+br_scrubber_handle_options(xlator_t *this, br_private_t *priv, dict_t *options)
+{
+    int32_t ret = 0;
+    gf_boolean_t scrubstall = _gf_false; /* not as dangerous as it sounds */
+
+    ret = br_scrubber_handle_stall(this, priv, options, &scrubstall);
+    if (ret)
+        goto error_return;
+
+    ret = br_scrubber_handle_throttle(this, priv, options, scrubstall);
+    if (ret)
+        goto error_return;
+
+    ret = br_scrubber_handle_freq(this, priv, options, scrubstall);
+    if (ret)
+        goto error_return;
+
+    br_scrubber_log_option(this, priv, scrubstall);
+
+    return 0;
+
+error_return:
+    return -1;
+}
+
+inode_t *
+br_lookup_bad_obj_dir(xlator_t *this, br_child_t *child, uuid_t gfid)
+{
+    struct iatt statbuf = {
+        0,
+    };
+    inode_table_t *table = NULL;
+    int32_t ret = -1;
+    loc_t loc = {
+        0,
+    };
+    inode_t *linked_inode = NULL;
+    int32_t op_errno = 0;
+
+    GF_VALIDATE_OR_GOTO("bit-rot-scrubber", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+    GF_VALIDATE_OR_GOTO(this->name, child, out);
+
+    table = child->table;
+
+    loc.inode = inode_new(table);
+    if (!loc.inode) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_NO_MEMORY,
+               "failed to allocate a new inode for"
+               "bad object directory");
+        goto out;
+    }
+
+    gf_uuid_copy(loc.gfid, gfid);
+
+    ret = syncop_lookup(child->xl, &loc, &statbuf, NULL, NULL, NULL);
+    if (ret < 0) {
+        op_errno = -ret;
+        gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_LOOKUP_FAILED,
+               "failed to lookup the bad "
+               "objects directory (gfid: %s (%s))",
+               uuid_utoa(gfid), strerror(op_errno));
+        goto out;
+    }
+
+    linked_inode = inode_link(loc.inode, NULL, NULL, &statbuf);
+    if (linked_inode)
+        inode_lookup(linked_inode);
+
+out:
+    loc_wipe(&loc);
+    return linked_inode;
+}
+
+int32_t
+br_read_bad_object_dir(xlator_t *this, br_child_t *child, fd_t *fd,
+                       dict_t *dict)
+{
+    gf_dirent_t entries;
+    gf_dirent_t *entry = NULL;
+    int32_t ret = -1;
+    off_t offset = 0;
+    int32_t count = 0;
+    char key[32] = {
+        0,
+    };
+    dict_t *out_dict = NULL;
+
+    INIT_LIST_HEAD(&entries.list);
+
+    while ((ret = syncop_readdir(child->xl, fd, 131072, offset, &entries, NULL,
+                                 &out_dict))) {
+        if (ret < 0)
+            goto out;
+
+        list_for_each_entry(entry, &entries.list, list)
+        {
+            offset = entry->d_off;
+
+            snprintf(key, sizeof(key), "quarantine-%d", count);
+
+            /*
+             * ignore the dict_set errors for now. The intention is
+             * to get as many bad objects as possible instead of
+             * erroring out at the first failure.
+             */
+            ret = dict_set_dynstr_with_alloc(dict, key, entry->d_name);
+            if (!ret)
+                count++;
+
+            if (out_dict) {
+                dict_copy(out_dict, dict);
+                dict_unref(out_dict);
+                out_dict = NULL;
+            }
+        }
+
+        gf_dirent_free(&entries);
+    }
+
+    ret = count;
+    ret = dict_set_int32_sizen(dict, "count", count);
+
+out:
+    return ret;
+}
+
+int32_t
+br_get_bad_objects_from_child(xlator_t *this, dict_t *dict, br_child_t *child)
+{
+    inode_t *inode = NULL;
+    inode_table_t *table = NULL;
+    fd_t *fd = NULL;
+    int32_t ret = -1;
+    loc_t loc = {
+        0,
+    };
+    int32_t op_errno = 0;
+
+    GF_VALIDATE_OR_GOTO("bit-rot-scrubber", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+    GF_VALIDATE_OR_GOTO(this->name, child, out);
+    GF_VALIDATE_OR_GOTO(this->name, dict, out);
+
+    table = child->table;
+
+    inode = inode_find(table, BR_BAD_OBJ_CONTAINER);
+    if (!inode) {
+        inode = br_lookup_bad_obj_dir(this, child, BR_BAD_OBJ_CONTAINER);
+        if (!inode)
+            goto out;
+    }
+
+    fd = fd_create(inode, 0);
+    if (!fd) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_FD_CREATE_FAILED,
+               "fd creation for the bad "
+               "objects directory failed (gfid: %s)",
+               uuid_utoa(BR_BAD_OBJ_CONTAINER));
+        goto out;
+    }
+
+    loc.inode = inode;
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    ret = syncop_opendir(child->xl, &loc, fd, NULL, NULL);
+    if (ret < 0) {
+        op_errno = -ret;
+        fd_unref(fd);
+        fd = NULL;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_FD_CREATE_FAILED,
+               "failed to open the bad "
+               "objects directory %s",
+               uuid_utoa(BR_BAD_OBJ_CONTAINER));
+        goto out;
+    }
+
+    fd_bind(fd);
+
+    ret = br_read_bad_object_dir(this, child, fd, dict);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, BRB_MSG_BAD_OBJ_READDIR_FAIL,
+               "readdir of the bad "
+               "objects directory (%s) failed ",
+               uuid_utoa(BR_BAD_OBJ_CONTAINER));
+        goto out;
+    }
+
+    ret = 0;
+
+out:
+    loc_wipe(&loc);
+    if (fd)
+        fd_unref(fd);
+    return ret;
+}
+
+int32_t
+br_collect_bad_objects_of_child(xlator_t *this, br_child_t *child, dict_t *dict,
+                                dict_t *child_dict, int32_t total_count)
+{
+    int32_t ret = -1;
+    int32_t count = 0;
+    char key[32] = {
+        0,
+    };
+    char main_key[32] = {
+        0,
+    };
+    int32_t j = 0;
+    int32_t tmp_count = 0;
+    char *entry = NULL;
+    char tmp[PATH_MAX] = {
+        0,
+    };
+    char *path = NULL;
+    int32_t len = 0;
+
+    ret = dict_get_int32_sizen(child_dict, "count", &count);
+    if (ret)
+        goto out;
+
+    tmp_count = total_count;
+
+    for (j = 0; j < count; j++) {
+        len = snprintf(key, sizeof(key), "quarantine-%d", j);
+        ret = dict_get_strn(child_dict, key, len, &entry);
+        if (ret)
+            continue;
+
+        ret = dict_get_str(child_dict, entry, &path);
+        len = snprintf(tmp, PATH_MAX, "%s ==> BRICK: %s\n path: %s", entry,
+                       child->brick_path, path);
+        if ((len < 0) || (len >= PATH_MAX)) {
+            continue;
+        }
+        snprintf(main_key, sizeof(main_key), "quarantine-%d", tmp_count);
+
+        ret = dict_set_dynstr_with_alloc(dict, main_key, tmp);
+        if (!ret)
+            tmp_count++;
+        path = NULL;
+    }
+
+    ret = tmp_count;
+
+out:
+    return ret;
+}
+
+int32_t
+br_collect_bad_objects_from_children(xlator_t *this, dict_t *dict)
+{
+    int32_t ret = -1;
+    dict_t *child_dict = NULL;
+    int32_t i = 0;
+    int32_t total_count = 0;
+    br_child_t *child = NULL;
+    br_private_t *priv = NULL;
+    dict_t *tmp_dict = NULL;
+
+    priv = this->private;
+    tmp_dict = dict;
+
+    for (i = 0; i < priv->child_count; i++) {
+        child = &priv->children[i];
+        GF_ASSERT(child);
+        if (!_br_is_child_connected(child))
+            continue;
+
+        child_dict = dict_new();
+        if (!child_dict) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_NO_MEMORY,
+                   "failed to allocate dict");
+            continue;
+        }
+        ret = br_get_bad_objects_from_child(this, child_dict, child);
+        /*
+         * Continue asking the remaining children for the list of
+         * bad objects even though getting the list from one of them
+         * fails.
+         */
+        if (ret) {
+            dict_unref(child_dict);
+            continue;
+        }
+
+        ret = br_collect_bad_objects_of_child(this, child, tmp_dict, child_dict,
+                                              total_count);
+        if (ret < 0) {
+            dict_unref(child_dict);
+            continue;
+        }
+
+        total_count = ret;
+        dict_unref(child_dict);
+        child_dict = NULL;
+    }
+
+    ret = dict_set_int32(tmp_dict, "total-count", total_count);
+
+    return ret;
+}
+
+int32_t
+br_get_bad_objects_list(xlator_t *this, dict_t **dict)
+{
+    int32_t ret = -1;
+    dict_t *tmp_dict = NULL;
+
+    GF_VALIDATE_OR_GOTO("bir-rot-scrubber", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, dict, out);
+
+    tmp_dict = *dict;
+    if (!tmp_dict) {
+        tmp_dict = dict_new();
+        if (!tmp_dict) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_NO_MEMORY,
+                   "failed to allocate dict");
+            goto out;
+        }
+        *dict = tmp_dict;
+    }
+
+    ret = br_collect_bad_objects_from_children(this, tmp_dict);
+
+out:
+    return ret;
+}
+
+static int
+wait_for_scrub_to_finish(xlator_t *this)
+{
+    int ret = -1;
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", scrub_monitor, out);
+    GF_VALIDATE_OR_GOTO("bit-rot", this, out);
+
+    gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_INFO,
+           "Waiting for all children to start and finish scrub");
+
+    pthread_mutex_lock(&scrub_monitor->donelock);
+    {
+        while (!scrub_monitor->done)
+            pthread_cond_wait(&scrub_monitor->donecond,
+                              &scrub_monitor->donelock);
+    }
+    pthread_mutex_unlock(&scrub_monitor->donelock);
+    ret = 0;
+out:
+    return ret;
+}
+
+/**
+ * This function is executed in a separate thread. This is scrubber monitor
+ * thread that takes care of state machine.
+ */
+void *
+br_monitor_thread(void *arg)
+{
+    int32_t ret = 0;
+    xlator_t *this = NULL;
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    this = arg;
+    priv = this->private;
+
+    /*
+     * Since, this is the topmost xlator, THIS has to be set by bit-rot
+     * xlator itself (STACK_WIND won't help in this case). Also it has
+     * to be done for each thread that gets spawned. Otherwise, a new
+     * thread will get global_xlator's pointer when it does "THIS".
+     */
+    THIS = this;
+
+    scrub_monitor = &priv->scrub_monitor;
+
+    pthread_mutex_lock(&scrub_monitor->mutex);
+    {
+        while (!scrub_monitor->inited)
+            pthread_cond_wait(&scrub_monitor->cond, &scrub_monitor->mutex);
+    }
+    pthread_mutex_unlock(&scrub_monitor->mutex);
+
+    /* this needs to be serialized with reconfigure() */
+    pthread_mutex_lock(&priv->lock);
+    {
+        ret = br_scrub_state_machine(this, _gf_false);
+    }
+    pthread_mutex_unlock(&priv->lock);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, BRB_MSG_SSM_FAILED,
+               "Scrub state machine failed");
+        goto out;
+    }
+
+    while (1) {
+        /* Wait for all children to finish scrubbing */
+        ret = wait_for_scrub_to_finish(this);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, -ret, BRB_MSG_SCRUB_WAIT_FAILED,
+                   "Scrub wait failed");
+            goto out;
+        }
+
+        /* scrub exit criteria: Move the state to PENDING */
+        br_scrubber_exit_control(this);
+    }
+
+out:
+    return NULL;
+}
+
+static void
+br_set_scrub_state(struct br_monitor *scrub_monitor, br_scrub_state_t state)
+{
+    LOCK(&scrub_monitor->lock);
+    {
+        _br_monitor_set_scrub_state(scrub_monitor, state);
+    }
+    UNLOCK(&scrub_monitor->lock);
+}
+
+int32_t
+br_scrubber_monitor_init(xlator_t *this, br_private_t *priv)
+{
+    struct br_monitor *scrub_monitor = NULL;
+    int ret = 0;
+
+    scrub_monitor = &priv->scrub_monitor;
+
+    LOCK_INIT(&scrub_monitor->lock);
+    scrub_monitor->this = this;
+
+    scrub_monitor->inited = _gf_false;
+    pthread_mutex_init(&scrub_monitor->mutex, NULL);
+    pthread_cond_init(&scrub_monitor->cond, NULL);
+
+    scrub_monitor->kick = _gf_false;
+    scrub_monitor->active_child_count = 0;
+    pthread_mutex_init(&scrub_monitor->wakelock, NULL);
+    pthread_cond_init(&scrub_monitor->wakecond, NULL);
+
+    scrub_monitor->done = _gf_false;
+    pthread_mutex_init(&scrub_monitor->donelock, NULL);
+    pthread_cond_init(&scrub_monitor->donecond, NULL);
+
+    /* Set the state to INACTIVE */
+    br_set_scrub_state(&priv->scrub_monitor, BR_SCRUB_STATE_INACTIVE);
+
+    /* Start the monitor thread */
+    ret = gf_thread_create(&scrub_monitor->thread, NULL, br_monitor_thread,
+                           this, "brmon");
+    if (ret != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, -ret, BRB_MSG_SPAWN_FAILED,
+               "monitor thread creation failed");
+        ret = -1;
+        goto err;
+    }
+
+    return 0;
+err:
+    pthread_mutex_destroy(&scrub_monitor->mutex);
+    pthread_cond_destroy(&scrub_monitor->cond);
+
+    pthread_mutex_destroy(&scrub_monitor->wakelock);
+    pthread_cond_destroy(&scrub_monitor->wakecond);
+
+    pthread_mutex_destroy(&scrub_monitor->donelock);
+    pthread_cond_destroy(&scrub_monitor->donecond);
+
+    LOCK_DESTROY(&scrub_monitor->lock);
+
+    return ret;
+}
+
+int32_t
+br_scrubber_init(xlator_t *this, br_private_t *priv)
+{
+    struct br_scrubber *fsscrub = NULL;
+    int ret = 0;
+
+    priv->tbf = tbf_init(NULL, 0);
+    if (!priv->tbf)
+        return -1;
+
+    ret = br_scrubber_monitor_init(this, priv);
+    if (ret)
+        return -1;
+
+    fsscrub = &priv->fsscrub;
+
+    fsscrub->this = this;
+    fsscrub->throttle = BR_SCRUB_THROTTLE_VOID;
+
+    pthread_mutex_init(&fsscrub->mutex, NULL);
+    pthread_cond_init(&fsscrub->cond, NULL);
+
+    fsscrub->nr_scrubbers = 0;
+    INIT_LIST_HEAD(&fsscrub->scrubbers);
+    INIT_LIST_HEAD(&fsscrub->scrublist);
+
+    return 0;
+}
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-scrub.h b/xlators/features/bit-rot/src/bitd/bit-rot-scrub.h
new file mode 100644
index 00000000000..4e5f67bc021
--- /dev/null
+++ b/xlators/features/bit-rot/src/bitd/bit-rot-scrub.h
@@ -0,0 +1,46 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __BIT_ROT_SCRUB_H__
+#define __BIT_ROT_SCRUB_H__
+
+#include <glusterfs/xlator.h>
+#include "bit-rot.h"
+
+void *
+br_fsscanner(void *);
+
+int32_t
+br_fsscan_schedule(xlator_t *);
+int32_t
+br_fsscan_reschedule(xlator_t *);
+int32_t
+br_fsscan_activate(xlator_t *);
+int32_t
+br_fsscan_deactivate(xlator_t *);
+int32_t
+br_fsscan_ondemand(xlator_t *);
+
+int32_t
+br_scrubber_handle_options(xlator_t *, br_private_t *, dict_t *);
+
+int32_t
+br_scrubber_monitor_init(xlator_t *, br_private_t *);
+
+int32_t
+br_scrubber_init(xlator_t *, br_private_t *);
+
+int32_t
+br_collect_bad_objects_from_children(xlator_t *this, dict_t *dict);
+
+void
+br_child_set_scrub_state(br_child_t *, gf_boolean_t);
+
+#endif /* __BIT_ROT_SCRUB_H__ */
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-ssm.c b/xlators/features/bit-rot/src/bitd/bit-rot-ssm.c
new file mode 100644
index 00000000000..753e31a3b23
--- /dev/null
+++ b/xlators/features/bit-rot/src/bitd/bit-rot-ssm.c
@@ -0,0 +1,124 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include "bit-rot-ssm.h"
+#include "bit-rot-scrub.h"
+#include "bit-rot-bitd-messages.h"
+
+int
+br_scrub_ssm_noop(xlator_t *this)
+{
+    return 0;
+}
+
+int
+br_scrub_ssm_state_pause(xlator_t *this)
+{
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_GENERIC_SSM_INFO,
+           "Scrubber paused");
+    _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_PAUSED);
+    return 0;
+}
+
+int
+br_scrub_ssm_state_ipause(xlator_t *this)
+{
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_GENERIC_SSM_INFO,
+           "Scrubber paused");
+    _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_IPAUSED);
+    return 0;
+}
+
+int
+br_scrub_ssm_state_active(xlator_t *this)
+{
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    if (scrub_monitor->done) {
+        (void)br_fsscan_activate(this);
+    } else {
+        gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_GENERIC_SSM_INFO,
+               "Scrubbing resumed");
+        _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_ACTIVE);
+    }
+
+    return 0;
+}
+
+int
+br_scrub_ssm_state_stall(xlator_t *this)
+{
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_GENERIC_SSM_INFO,
+           "Volume is under active scrubbing. Pausing scrub..");
+    _br_monitor_set_scrub_state(scrub_monitor, BR_SCRUB_STATE_STALLED);
+    return 0;
+}
+
+static br_scrub_ssm_call *br_scrub_ssm[BR_SCRUB_MAXSTATES][BR_SCRUB_MAXEVENTS] =
+    {
+        /* INACTIVE */
+        {br_fsscan_schedule, br_scrub_ssm_state_ipause, br_scrub_ssm_noop},
+        /* PENDING  */
+        {br_fsscan_reschedule, br_fsscan_deactivate, br_fsscan_ondemand},
+        /* ACTIVE   */
+        {br_scrub_ssm_noop, br_scrub_ssm_state_stall, br_scrub_ssm_noop},
+        /* PAUSED   */
+        {br_fsscan_activate, br_scrub_ssm_noop, br_scrub_ssm_noop},
+        /* IPAUSED  */
+        {br_fsscan_schedule, br_scrub_ssm_noop, br_scrub_ssm_noop},
+        /* STALLED  */
+        {br_scrub_ssm_state_active, br_scrub_ssm_noop, br_scrub_ssm_noop},
+};
+
+int32_t
+br_scrub_state_machine(xlator_t *this, gf_boolean_t scrub_ondemand)
+{
+    br_private_t *priv = NULL;
+    br_scrub_ssm_call *call = NULL;
+    struct br_scrubber *fsscrub = NULL;
+    br_scrub_state_t currstate = 0;
+    br_scrub_event_t event = 0;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    fsscrub = &priv->fsscrub;
+    scrub_monitor = &priv->scrub_monitor;
+
+    currstate = scrub_monitor->state;
+    if (scrub_ondemand)
+        event = BR_SCRUB_EVENT_ONDEMAND;
+    else
+        event = _br_child_get_scrub_event(fsscrub);
+
+    call = br_scrub_ssm[currstate][event];
+    return call(this);
+}
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot-ssm.h b/xlators/features/bit-rot/src/bitd/bit-rot-ssm.h
new file mode 100644
index 00000000000..37b45a42eac
--- /dev/null
+++ b/xlators/features/bit-rot/src/bitd/bit-rot-ssm.h
@@ -0,0 +1,38 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __BIT_ROT_SSM_H__
+#define __BIT_ROT_SSM_H__
+
+#include <glusterfs/xlator.h>
+
+typedef enum br_scrub_state {
+    BR_SCRUB_STATE_INACTIVE = 0,
+    BR_SCRUB_STATE_PENDING,
+    BR_SCRUB_STATE_ACTIVE,
+    BR_SCRUB_STATE_PAUSED,
+    BR_SCRUB_STATE_IPAUSED,
+    BR_SCRUB_STATE_STALLED,
+    BR_SCRUB_MAXSTATES,
+} br_scrub_state_t;
+
+typedef enum br_scrub_event {
+    BR_SCRUB_EVENT_SCHEDULE = 0,
+    BR_SCRUB_EVENT_PAUSE,
+    BR_SCRUB_EVENT_ONDEMAND,
+    BR_SCRUB_MAXEVENTS,
+} br_scrub_event_t;
+
+struct br_monitor;
+
+int32_t
+br_scrub_state_machine(xlator_t *, gf_boolean_t);
+
+#endif /* __BIT_ROT_SSM_H__ */
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot.c b/xlators/features/bit-rot/src/bitd/bit-rot.c
new file mode 100644
index 00000000000..a2f1c343a1d
--- /dev/null
+++ b/xlators/features/bit-rot/src/bitd/bit-rot.c
@@ -0,0 +1,2232 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <ctype.h>
+
+#include <glusterfs/logging.h>
+#include <glusterfs/compat-errno.h>
+
+#include "bit-rot.h"
+#include "bit-rot-scrub.h"
+#include <pthread.h>
+#include "bit-rot-bitd-messages.h"
+
+#define BR_HASH_CALC_READ_SIZE (128 * 1024)
+
+typedef int32_t(br_child_handler)(xlator_t *, br_child_t *);
+
+struct br_child_event {
+    xlator_t *this;
+
+    br_child_t *child;
+
+    br_child_handler *call;
+
+    struct list_head list;
+};
+
+static int
+br_find_child_index(xlator_t *this, xlator_t *child)
+{
+    br_private_t *priv = NULL;
+    int i = -1;
+    int index = -1;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+    GF_VALIDATE_OR_GOTO(this->name, child, out);
+
+    priv = this->private;
+
+    for (i = 0; i < priv->child_count; i++) {
+        if (child == priv->children[i].xl) {
+            index = i;
+            break;
+        }
+    }
+
+out:
+    return index;
+}
+
+br_child_t *
+br_get_child_from_brick_path(xlator_t *this, char *brick_path)
+{
+    br_private_t *priv = NULL;
+    br_child_t *child = NULL;
+    br_child_t *tmp = NULL;
+    int i = 0;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+    GF_VALIDATE_OR_GOTO(this->name, brick_path, out);
+
+    priv = this->private;
+
+    pthread_mutex_lock(&priv->lock);
+    {
+        for (i = 0; i < priv->child_count; i++) {
+            tmp = &priv->children[i];
+            if (!strcmp(tmp->brick_path, brick_path)) {
+                child = tmp;
+                break;
+            }
+        }
+    }
+    pthread_mutex_unlock(&priv->lock);
+
+out:
+    return child;
+}
+
+/**
+ * probably we'll encapsulate brick inside our own structure when
+ * needed -- later.
+ */
+void *
+br_brick_init(void *xl, struct gf_brick_spec *brick)
+{
+    return brick;
+}
+
+/**
+ * and cleanup things here when allocated br_brick_init().
+ */
+void
+br_brick_fini(void *xl, char *brick, void *data)
+{
+    return;
+}
+
+/**
+ * TODO: Signature can contain null terminators which causes bitrot
+ * stub to store truncated hash as it depends on string length of
+ * the hash.
+ *
+ * FIX: Send the string length as part of the signature struct and
+ *      change stub to handle this change.
+ */
+static br_isignature_t *
+br_prepare_signature(const unsigned char *sign, unsigned long hashlen,
+                     int8_t hashtype, br_object_t *object)
+{
+    br_isignature_t *signature = NULL;
+
+    /* TODO: use mem-pool */
+    signature = GF_CALLOC(1, signature_size(hashlen + 1),
+                          gf_br_stub_mt_signature_t);
+    if (!signature)
+        return NULL;
+
+    /* object version */
+    signature->signedversion = object->signedversion;
+
+    /* signature length & type */
+    signature->signaturelen = hashlen;
+    signature->signaturetype = hashtype;
+
+    /* signature itself */
+    memcpy(signature->signature, (char *)sign, hashlen);
+    signature->signature[hashlen + 1] = '\0';
+
+    return signature;
+}
+
+gf_boolean_t
+bitd_is_bad_file(xlator_t *this, br_child_t *child, loc_t *loc, fd_t *fd)
+{
+    int32_t ret = -1;
+    dict_t *xattr = NULL;
+    inode_t *inode = NULL;
+    gf_boolean_t bad_file = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", this, out);
+
+    inode = (loc) ? loc->inode : fd->inode;
+
+    if (fd)
+        ret = syncop_fgetxattr(child->xl, fd, &xattr, BITROT_OBJECT_BAD_KEY,
+                               NULL, NULL);
+    else if (loc)
+        ret = syncop_getxattr(child->xl, loc, &xattr, BITROT_OBJECT_BAD_KEY,
+                              NULL, NULL);
+
+    if (!ret) {
+        gf_msg_debug(this->name, 0, "[GFID: %s] is marked corrupted",
+                     uuid_utoa(inode->gfid));
+        bad_file = _gf_true;
+    }
+
+    if (xattr)
+        dict_unref(xattr);
+
+out:
+    return bad_file;
+}
+
+/**
+ * Do a lookup on the gfid present within the object.
+ */
+static int32_t
+br_object_lookup(xlator_t *this, br_object_t *object, struct iatt *iatt,
+                 inode_t **linked_inode)
+{
+    int ret = -EINVAL;
+    loc_t loc = {
+        0,
+    };
+    inode_t *inode = NULL;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, object, out);
+
+    inode = inode_find(object->child->table, object->gfid);
+
+    if (inode)
+        loc.inode = inode;
+    else
+        loc.inode = inode_new(object->child->table);
+
+    if (!loc.inode) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    gf_uuid_copy(loc.gfid, object->gfid);
+
+    ret = syncop_lookup(object->child->xl, &loc, iatt, NULL, NULL, NULL);
+    if (ret < 0)
+        goto out;
+
+    /*
+     * The file might have been deleted by the application
+     * after getting the event, but before doing a lookup.
+     * So use linked_inode after inode_link is done.
+     */
+    *linked_inode = inode_link(loc.inode, NULL, NULL, iatt);
+    if (*linked_inode)
+        inode_lookup(*linked_inode);
+
+out:
+    loc_wipe(&loc);
+    return ret;
+}
+
+/**
+ * open the object with O_RDONLY flags and return the fd. How to let brick
+ * know that open is being done by bitd because syncop framework does not allow
+ * passing xdata -- may be use frame->root->pid itself.
+ */
+static int32_t
+br_object_open(xlator_t *this, br_object_t *object, inode_t *inode,
+               fd_t **openfd)
+{
+    int32_t ret = -1;
+    fd_t *fd = NULL;
+    loc_t loc = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("bit-rot", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, object, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+
+    ret = -EINVAL;
+    fd = fd_create(inode, 0);
+    if (!fd) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_FD_CREATE_FAILED,
+                "gfid=%s", uuid_utoa(inode->gfid), NULL);
+        goto out;
+    }
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    ret = syncop_open(object->child->xl, &loc, O_RDONLY, fd, NULL, NULL);
+    if (ret) {
+        br_log_object(this, "open", inode->gfid, -ret);
+        fd_unref(fd);
+        fd = NULL;
+    } else {
+        fd_bind(fd);
+        *openfd = fd;
+    }
+
+    loc_wipe(&loc);
+
+out:
+    return ret;
+}
+
+/**
+ * read 128k block from the object @object from the offset @offset
+ * and return the buffer.
+ */
+static int32_t
+br_object_read_block_and_sign(xlator_t *this, fd_t *fd, br_child_t *child,
+                              off_t offset, size_t size, SHA256_CTX *sha256)
+{
+    int32_t ret = -1;
+    tbf_t *tbf = NULL;
+    struct iovec *iovec = NULL;
+    struct iobref *iobref = NULL;
+    br_private_t *priv = NULL;
+    int count = 0;
+    int i = 0;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd->inode, out);
+    GF_VALIDATE_OR_GOTO(this->name, child, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    priv = this->private;
+
+    GF_VALIDATE_OR_GOTO(this->name, priv->tbf, out);
+    tbf = priv->tbf;
+
+    ret = syncop_readv(child->xl, fd, size, offset, 0, &iovec, &count, &iobref,
+                       NULL, NULL, NULL);
+
+    if (ret < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, BRB_MSG_READV_FAILED,
+                "gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        ret = -1;
+        goto out;
+    }
+
+    if (ret == 0)
+        goto out;
+
+    for (i = 0; i < count; i++) {
+        TBF_THROTTLE_BEGIN(tbf, TBF_OP_HASH, iovec[i].iov_len);
+        {
+            SHA256_Update(sha256, (const unsigned char *)(iovec[i].iov_base),
+                          iovec[i].iov_len);
+        }
+        TBF_THROTTLE_BEGIN(tbf, TBF_OP_HASH, iovec[i].iov_len);
+    }
+
+out:
+    if (iovec)
+        GF_FREE(iovec);
+
+    if (iobref)
+        iobref_unref(iobref);
+
+    return ret;
+}
+
+int32_t
+br_calculate_obj_checksum(unsigned char *md, br_child_t *child, fd_t *fd,
+                          struct iatt *iatt)
+{
+    int32_t ret = -1;
+    off_t offset = 0;
+    size_t block = BR_HASH_CALC_READ_SIZE;
+    xlator_t *this = NULL;
+
+    SHA256_CTX sha256;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", child, out);
+    GF_VALIDATE_OR_GOTO("bit-rot", iatt, out);
+    GF_VALIDATE_OR_GOTO("bit-rot", fd, out);
+
+    this = child->this;
+
+    SHA256_Init(&sha256);
+
+    while (1) {
+        ret = br_object_read_block_and_sign(this, fd, child, offset, block,
+                                            &sha256);
+        if (ret < 0) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_BLOCK_READ_FAILED,
+                    "offset=%" PRIu64, offset, "object-gfid=%s",
+                    uuid_utoa(fd->inode->gfid), NULL);
+            break;
+        }
+
+        if (ret == 0)
+            break;
+
+        offset += ret;
+    }
+
+    if (ret == 0)
+        SHA256_Final(md, &sha256);
+
+out:
+    return ret;
+}
+
+static int32_t
+br_object_checksum(unsigned char *md, br_object_t *object, fd_t *fd,
+                   struct iatt *iatt)
+{
+    return br_calculate_obj_checksum(md, object->child, fd, iatt);
+}
+
+static int32_t
+br_object_read_sign(inode_t *linked_inode, fd_t *fd, br_object_t *object,
+                    struct iatt *iatt)
+{
+    int32_t ret = -1;
+    xlator_t *this = NULL;
+    dict_t *xattr = NULL;
+    unsigned char *md = NULL;
+    br_isignature_t *sign = NULL;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", object, out);
+    GF_VALIDATE_OR_GOTO("bit-rot", linked_inode, out);
+    GF_VALIDATE_OR_GOTO("bit-rot", fd, out);
+
+    this = object->this;
+
+    md = GF_MALLOC(SHA256_DIGEST_LENGTH, gf_common_mt_char);
+    if (!md) {
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_SAVING_HASH_FAILED,
+                "object-gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        goto out;
+    }
+
+    ret = br_object_checksum(md, object, fd, iatt);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_CALC_CHECKSUM_FAILED,
+                "object-gfid=%s", uuid_utoa(linked_inode->gfid), NULL);
+        goto free_signature;
+    }
+
+    sign = br_prepare_signature(md, SHA256_DIGEST_LENGTH,
+                                BR_SIGNATURE_TYPE_SHA256, object);
+    if (!sign) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_GET_SIGN_FAILED,
+                "object-gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        goto free_signature;
+    }
+
+    xattr = dict_for_key_value(GLUSTERFS_SET_OBJECT_SIGNATURE, (void *)sign,
+                               signature_size(SHA256_DIGEST_LENGTH), _gf_true);
+
+    if (!xattr) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SET_SIGN_FAILED,
+                "dict-allocation object-gfid=%s", uuid_utoa(fd->inode->gfid),
+                NULL);
+        goto free_isign;
+    }
+
+    ret = syncop_fsetxattr(object->child->xl, fd, xattr, 0, NULL, NULL);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SET_SIGN_FAILED,
+                "fsetxattr object-gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        goto unref_dict;
+    }
+
+    ret = 0;
+
+unref_dict:
+    dict_unref(xattr);
+free_isign:
+    GF_FREE(sign);
+free_signature:
+    GF_FREE(md);
+out:
+    return ret;
+}
+
+static int
+br_object_sign_softerror(int32_t op_errno)
+{
+    return ((op_errno == ENOENT) || (op_errno == ESTALE) ||
+            (op_errno == ENODATA));
+}
+
+void
+br_log_object(xlator_t *this, char *op, uuid_t gfid, int32_t op_errno)
+{
+    int softerror = br_object_sign_softerror(op_errno);
+    if (softerror) {
+        gf_msg_debug(this->name, 0,
+                     "%s() failed on object %s "
+                     "[reason: %s]",
+                     op, uuid_utoa(gfid), strerror(op_errno));
+    } else {
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_OP_FAILED, "op=%s",
+                op, "gfid=%s", uuid_utoa(gfid), NULL);
+    }
+}
+
+void
+br_log_object_path(xlator_t *this, char *op, const char *path, int32_t op_errno)
+{
+    int softerror = br_object_sign_softerror(op_errno);
+    if (softerror) {
+        gf_msg_debug(this->name, 0,
+                     "%s() failed on object %s "
+                     "[reason: %s]",
+                     op, path, strerror(op_errno));
+    } else {
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_OP_FAILED, "op=%s",
+                op, "path=%s", path, NULL);
+    }
+}
+
+static void
+br_trigger_sign(xlator_t *this, br_child_t *child, inode_t *linked_inode,
+                loc_t *loc, gf_boolean_t need_reopen)
+{
+    fd_t *fd = NULL;
+    int32_t ret = -1;
+    uint32_t val = 0;
+    dict_t *dict = NULL;
+    pid_t pid = GF_CLIENT_PID_BITD;
+
+    syncopctx_setfspid(&pid);
+
+    val = (need_reopen == _gf_true) ? BR_OBJECT_REOPEN : BR_OBJECT_RESIGN;
+
+    dict = dict_new();
+    if (!dict)
+        goto out;
+
+    ret = dict_set_uint32(dict, BR_REOPEN_SIGN_HINT_KEY, val);
+    if (ret)
+        goto cleanup_dict;
+
+    ret = -1;
+    fd = fd_create(linked_inode, 0);
+    if (!fd) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_FD_CREATE_FAILED,
+                "gfid=%s", uuid_utoa(linked_inode->gfid), NULL);
+        goto cleanup_dict;
+    }
+
+    ret = syncop_open(child->xl, loc, O_RDWR, fd, NULL, NULL);
+    if (ret) {
+        br_log_object(this, "open", linked_inode->gfid, -ret);
+        goto unref_fd;
+    }
+
+    fd_bind(fd);
+
+    ret = syncop_fsetxattr(child->xl, fd, dict, 0, NULL, NULL);
+    if (ret)
+        br_log_object(this, "fsetxattr", linked_inode->gfid, -ret);
+
+    /* passthough: fd_unref() */
+
+unref_fd:
+    fd_unref(fd);
+cleanup_dict:
+    dict_unref(dict);
+out:
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, BRB_MSG_TRIGGER_SIGN_FAILED,
+                "gfid=%s", uuid_utoa(linked_inode->gfid), "reopen-hint-val=%d",
+                val, NULL);
+    }
+}
+
+static void
+br_object_resign(xlator_t *this, br_object_t *object, inode_t *linked_inode)
+{
+    loc_t loc = {
+        0,
+    };
+
+    loc.inode = inode_ref(linked_inode);
+    gf_uuid_copy(loc.gfid, linked_inode->gfid);
+
+    br_trigger_sign(this, object->child, linked_inode, &loc, _gf_false);
+
+    loc_wipe(&loc);
+}
+
+/**
+ * Sign a given object. This routine runs full throttle. There needs to be
+ * some form of priority scheduling and/or read burstness to avoid starving
+ * (or kicking) client I/O's.
+ */
+static int32_t
+br_sign_object(br_object_t *object)
+{
+    int32_t ret = -1;
+    inode_t *linked_inode = NULL;
+    xlator_t *this = NULL;
+    fd_t *fd = NULL;
+    struct iatt iatt = {
+        0,
+    };
+    pid_t pid = GF_CLIENT_PID_BITD;
+    br_sign_state_t sign_info = BR_SIGN_NORMAL;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", object, out);
+
+    this = object->this;
+
+    /**
+     * FIXME: This is required as signing an object is restricted to
+     * clients with special frame->root->pid. Change the way client
+     * pid is set.
+     */
+    syncopctx_setfspid(&pid);
+
+    ret = br_object_lookup(this, object, &iatt, &linked_inode);
+    if (ret) {
+        br_log_object(this, "lookup", object->gfid, -ret);
+        goto out;
+    }
+
+    /**
+     * For fd's that have notified for reopening, we send an explicit
+     * open() followed by a dummy write() call. This triggers the
+     * actual signing of the object.
+     */
+    sign_info = ntohl(object->sign_info);
+    if (sign_info == BR_SIGN_REOPEN_WAIT) {
+        br_object_resign(this, object, linked_inode);
+        goto unref_inode;
+    }
+
+    ret = br_object_open(this, object, linked_inode, &fd);
+    if (!fd) {
+        br_log_object(this, "open", object->gfid, -ret);
+        goto unref_inode;
+    }
+
+    /**
+     * we have an open file descriptor on the object. from here on,
+     * do not be generous to file operation errors.
+     */
+    gf_msg_debug(this->name, 0, "Signing object [%s]",
+                 uuid_utoa(linked_inode->gfid));
+
+    ret = br_object_read_sign(linked_inode, fd, object, &iatt);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_READ_AND_SIGN_FAILED,
+                "gfid=%s", uuid_utoa(linked_inode->gfid), NULL);
+        goto unref_fd;
+    }
+
+    ret = 0;
+
+unref_fd:
+    fd_unref(fd);
+unref_inode:
+    inode_unref(linked_inode);
+out:
+    return ret;
+}
+
+static br_object_t *
+__br_pick_object(br_private_t *priv)
+{
+    br_object_t *object = NULL;
+
+    while (list_empty(&priv->obj_queue->objects)) {
+        pthread_cond_wait(&priv->object_cond, &priv->lock);
+    }
+
+    object = list_first_entry(&priv->obj_queue->objects, br_object_t, list);
+    list_del_init(&object->list);
+
+    return object;
+}
+
+/**
+ * This is the place where the signing of the objects is triggered.
+ */
+void *
+br_process_object(void *arg)
+{
+    xlator_t *this = NULL;
+    br_object_t *object = NULL;
+    br_private_t *priv = NULL;
+    int32_t ret = -1;
+
+    this = arg;
+    priv = this->private;
+
+    THIS = this;
+
+    for (;;) {
+        pthread_mutex_lock(&priv->lock);
+        {
+            object = __br_pick_object(priv);
+        }
+        pthread_mutex_unlock(&priv->lock);
+
+        ret = br_sign_object(object);
+        if (ret && !br_object_sign_softerror(-ret))
+            gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SET_SIGN_FAILED,
+                    "gfid=%s", uuid_utoa(object->gfid), NULL);
+        GF_FREE(object);
+    }
+
+    return NULL;
+}
+
+/**
+ * This function gets kicked in once the object is expired from the
+ * timer wheel. This actually adds the object received via notification
+ * from the changelog to the queue from where the objects gets picked
+ * up for signing.
+ *
+ * This routine can be made lightweight by introducing an alternate
+ * timer-wheel API that dispatches _all_ expired objects in one-shot
+ * rather than an object at-a-time. This routine can then just simply
+ * be a call to list_splice_tail().
+ *
+ * NOTE: use call_time to instrument signing time in br_sign_object().
+ */
+void
+br_add_object_to_queue(struct gf_tw_timer_list *timer, void *data,
+                       unsigned long call_time)
+{
+    br_object_t *object = NULL;
+    xlator_t *this = NULL;
+    br_private_t *priv = NULL;
+
+    object = data;
+    this = object->this;
+    priv = this->private;
+
+    THIS = this;
+
+    pthread_mutex_lock(&priv->lock);
+    {
+        list_add_tail(&object->list, &priv->obj_queue->objects);
+        pthread_cond_broadcast(&priv->object_cond);
+    }
+    pthread_mutex_unlock(&priv->lock);
+
+    if (timer)
+        mem_put(timer);
+    return;
+}
+
+static br_object_t *
+br_initialize_object(xlator_t *this, br_child_t *child, changelog_event_t *ev)
+{
+    br_object_t *object = NULL;
+
+    object = GF_CALLOC(1, sizeof(*object), gf_br_mt_br_object_t);
+    if (!object)
+        goto out;
+    INIT_LIST_HEAD(&object->list);
+
+    object->this = this;
+    object->child = child;
+    gf_uuid_copy(object->gfid, ev->u.releasebr.gfid);
+
+    /* NOTE: it's BE, but no worry */
+    object->signedversion = ev->u.releasebr.version;
+    object->sign_info = ev->u.releasebr.sign_info;
+
+out:
+    return object;
+}
+
+static struct gf_tw_timer_list *
+br_initialize_timer(xlator_t *this, br_object_t *object, br_child_t *child,
+                    changelog_event_t *ev)
+{
+    br_private_t *priv = NULL;
+    struct gf_tw_timer_list *timer = NULL;
+
+    priv = this->private;
+
+    timer = mem_get0(child->timer_pool);
+    if (!timer)
+        goto out;
+    INIT_LIST_HEAD(&timer->entry);
+
+    timer->expires = priv->expiry_time;
+    if (!timer->expires)
+        timer->expires = 1;
+
+    timer->data = object;
+    timer->function = br_add_object_to_queue;
+    gf_tw_add_timer(priv->timer_wheel, timer);
+
+out:
+    return timer;
+}
+
+static int32_t
+br_schedule_object_reopen(xlator_t *this, br_object_t *object,
+                          br_child_t *child, changelog_event_t *ev)
+{
+    struct gf_tw_timer_list *timer = NULL;
+
+    timer = br_initialize_timer(this, object, child, ev);
+    if (!timer)
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SET_TIMER_FAILED,
+                "gfid=%s", uuid_utoa(object->gfid), NULL);
+    return timer ? 0 : -1;
+}
+
+static int32_t
+br_object_quicksign(xlator_t *this, br_object_t *object)
+{
+    br_add_object_to_queue(NULL, object, 0ULL);
+    return 0;
+}
+
+/**
+ * This callback function registered with the changelog is executed
+ * whenever a notification from the changelog is received. This should
+ * add the object (or the gfid) on which the notification has come to
+ * the timer-wheel with some expiry time.
+ *
+ * TODO: use mem-pool for allocations and maybe allocate timer and
+ * object as a single alloc and bifurcate their respective pointers.
+ */
+void
+br_brick_callback(void *xl, char *brick, void *data, changelog_event_t *ev)
+{
+    int32_t ret = 0;
+    uuid_t gfid = {
+        0,
+    };
+    xlator_t *this = NULL;
+    br_object_t *object = NULL;
+    br_child_t *child = NULL;
+    br_sign_state_t sign_info = BR_SIGN_INVALID;
+
+    this = xl;
+
+    GF_VALIDATE_OR_GOTO(this->name, ev, out);
+    GF_VALIDATE_OR_GOTO("bit-rot", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    GF_ASSERT(ev->ev_type == CHANGELOG_OP_TYPE_BR_RELEASE);
+    GF_ASSERT(!gf_uuid_is_null(ev->u.releasebr.gfid));
+
+    gf_uuid_copy(gfid, ev->u.releasebr.gfid);
+
+    gf_msg_debug(this->name, 0, "RELEASE EVENT [GFID %s]", uuid_utoa(gfid));
+
+    child = br_get_child_from_brick_path(this, brick);
+    if (!child) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_GET_SUBVOL_FAILED,
+                "brick=%s", brick, NULL);
+        goto out;
+    }
+
+    object = br_initialize_object(this, child, ev);
+    if (!object) {
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_NO_MEMORY,
+                "object-gfid=%s", uuid_utoa(gfid), NULL);
+        goto out;
+    }
+
+    /* sanity check */
+    sign_info = ntohl(object->sign_info);
+    GF_ASSERT(sign_info != BR_SIGN_NORMAL);
+
+    if (sign_info == BR_SIGN_REOPEN_WAIT)
+        ret = br_schedule_object_reopen(this, object, child, ev);
+    else
+        ret = br_object_quicksign(this, object);
+
+    if (ret)
+        goto free_object;
+
+    gf_msg_debug(this->name, 0, "->callback: brick [%s], type [%d]\n", brick,
+                 ev->ev_type);
+    return;
+
+free_object:
+    GF_FREE(object);
+out:
+    return;
+}
+
+void
+br_fill_brick_spec(struct gf_brick_spec *brick, char *path)
+{
+    brick->brick_path = gf_strdup(path);
+    brick->filter = CHANGELOG_OP_TYPE_BR_RELEASE;
+
+    brick->init = br_brick_init;
+    brick->fini = br_brick_fini;
+    brick->callback = br_brick_callback;
+    brick->connected = NULL;
+    brick->disconnected = NULL;
+}
+
+static gf_boolean_t
+br_check_object_need_sign(xlator_t *this, dict_t *xattr, br_child_t *child)
+{
+    int32_t ret = -1;
+    gf_boolean_t need_sign = _gf_false;
+    br_isignature_out_t *sign = NULL;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, xattr, out);
+    GF_VALIDATE_OR_GOTO(this->name, child, out);
+
+    ret = dict_get_ptr(xattr, GLUSTERFS_GET_OBJECT_SIGNATURE, (void **)&sign);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_GET_SIGN_FAILED,
+                "object-info", NULL);
+        goto out;
+    }
+
+    /* Object has been opened and hence dirty. Do not sign it */
+    if (sign->stale)
+        need_sign = _gf_true;
+
+out:
+    return need_sign;
+}
+
+int32_t
+br_prepare_loc(xlator_t *this, br_child_t *child, loc_t *parent,
+               gf_dirent_t *entry, loc_t *loc)
+{
+    int32_t ret = -1;
+    inode_t *inode = NULL;
+
+    inode = inode_grep(child->table, parent->inode, entry->d_name);
+    if (!inode)
+        loc->inode = inode_new(child->table);
+    else {
+        loc->inode = inode;
+        if (loc->inode->ia_type != IA_IFREG) {
+            gf_msg_debug(this->name, 0,
+                         "%s is not a regular "
+                         "file",
+                         entry->d_name);
+            ret = 0;
+            goto out;
+        }
+    }
+
+    loc->parent = inode_ref(parent->inode);
+    gf_uuid_copy(loc->pargfid, parent->inode->gfid);
+
+    ret = inode_path(parent->inode, entry->d_name, (char **)&loc->path);
+    if (ret < 0 || !loc->path) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_PATH_FAILED,
+                "inode_path=%s", entry->d_name, "parent-gfid=%s",
+                uuid_utoa(parent->inode->gfid), NULL);
+        goto out;
+    }
+
+    loc->name = strrchr(loc->path, '/');
+    if (loc->name)
+        loc->name++;
+
+    ret = 1;
+
+out:
+    return ret;
+}
+
+/**
+ * Oneshot crawler
+ * ---------------
+ * This is a catchup mechanism. Objects that remained unsigned from the
+ * last run for whatever reason (node crashes, reboots, etc..) become
+ * candidates for signing. This allows the signature to "catch up" with
+ * the current state of the object. Triggering signing is easy: perform
+ * an open() followed by a close() thereby resulting in call boomerang.
+ * (though not back to itself :))
+ */
+int
+bitd_oneshot_crawl(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent,
+                   void *data)
+{
+    int op_errno = 0;
+    br_child_t *child = NULL;
+    xlator_t *this = NULL;
+    loc_t loc = {
+        0,
+    };
+    struct iatt iatt = {
+        0,
+    };
+    struct iatt parent_buf = {
+        0,
+    };
+    dict_t *xattr = NULL;
+    int32_t ret = -1;
+    inode_t *linked_inode = NULL;
+    gf_boolean_t need_signing = _gf_false;
+    gf_boolean_t need_reopen = _gf_true;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", subvol, out);
+    GF_VALIDATE_OR_GOTO("bit-rot", data, out);
+
+    child = data;
+    this = child->this;
+
+    ret = br_prepare_loc(this, child, parent, entry, &loc);
+    if (!ret)
+        goto out;
+
+    ret = syncop_lookup(child->xl, &loc, &iatt, &parent_buf, NULL, NULL);
+    if (ret) {
+        br_log_object_path(this, "lookup", loc.path, -ret);
+        goto out;
+    }
+
+    linked_inode = inode_link(loc.inode, parent->inode, loc.name, &iatt);
+    if (linked_inode)
+        inode_lookup(linked_inode);
+
+    if (iatt.ia_type != IA_IFREG) {
+        gf_msg_debug(this->name, 0,
+                     "%s is not a regular file, "
+                     "skipping..",
+                     entry->d_name);
+        ret = 0;
+        goto unref_inode;
+    }
+
+    /**
+     * As of now, 2 cases  are possible and handled.
+     * 1) GlusterFS is upgraded from a previous version which does not
+     *    have any idea about bit-rot and have data in the filesystem.
+     *    In this case syncop_getxattr fails with ENODATA and the object
+     *    is signed. (In real, when crawler sends lookup, bit-rot-stub
+     *    creates the xattrs before returning lookup reply)
+     * 2) Bit-rot was not enabled or BitD was does for some reasons, during
+     *    which some files were created, but since BitD was down, were not
+     *    signed.
+     * If the file was just created and was being written some data when
+     * the down BitD came up, then bit-rot stub should be intelligent to
+     * identify this case (by comparing the ongoing version or by checking
+     * if there are any fds present for that inode) and handle properly.
+     */
+
+    if (bitd_is_bad_file(this, child, &loc, NULL)) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, BRB_MSG_SKIP_OBJECT, "path=%s",
+                loc.path, NULL);
+        goto unref_inode;
+    }
+
+    ret = syncop_getxattr(child->xl, &loc, &xattr,
+                          GLUSTERFS_GET_OBJECT_SIGNATURE, NULL, NULL);
+    if (ret < 0) {
+        op_errno = -ret;
+        br_log_object(this, "getxattr", linked_inode->gfid, op_errno);
+
+        /**
+         * No need to sign the zero byte objects as the signing
+         * happens upon first modification of the object.
+         */
+        if (op_errno == ENODATA && (iatt.ia_size != 0))
+            need_signing = _gf_true;
+        if (op_errno == EINVAL)
+            gf_smsg(this->name, GF_LOG_WARNING, 0,
+                    BRB_MSG_PARTIAL_VERSION_PRESENCE, "gfid=%s",
+                    uuid_utoa(linked_inode->gfid), NULL);
+    } else {
+        need_signing = br_check_object_need_sign(this, xattr, child);
+
+        /*
+         * If we are here means, bitrot daemon has started. Is it just
+         * a simple restart of the daemon or is it started because the
+         * feature is enabled is something hard to determine. Hence,
+         * if need_signing is false (because bit-rot version and signature
+         * are present), then still go ahead and sign it.
+         */
+        if (!need_signing) {
+            need_signing = _gf_true;
+            need_reopen = _gf_true;
+        }
+    }
+
+    if (!need_signing)
+        goto unref_dict;
+
+    gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_TRIGGER_SIGN, "path=%s",
+            loc.path, "gfid=%s", uuid_utoa(linked_inode->gfid), "Brick-path=%s",
+            child->brick_path, NULL);
+    br_trigger_sign(this, child, linked_inode, &loc, need_reopen);
+
+    ret = 0;
+
+unref_dict:
+    if (xattr)
+        dict_unref(xattr);
+unref_inode:
+    inode_unref(linked_inode);
+out:
+    loc_wipe(&loc);
+
+    return ret;
+}
+
+#define BR_CRAWL_THROTTLE_COUNT 50
+#define BR_CRAWL_THROTTLE_ZZZ 5
+
+void *
+br_oneshot_signer(void *arg)
+{
+    loc_t loc = {
+        0,
+    };
+    xlator_t *this = NULL;
+    br_child_t *child = NULL;
+
+    child = arg;
+    this = child->this;
+
+    THIS = this;
+
+    gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_CRAWLING_START, "brick-path=%s",
+            child->brick_path, NULL);
+
+    loc.inode = child->table->root;
+    (void)syncop_ftw_throttle(child->xl, &loc, GF_CLIENT_PID_BITD, child,
+                              bitd_oneshot_crawl, BR_CRAWL_THROTTLE_COUNT,
+                              BR_CRAWL_THROTTLE_ZZZ);
+
+    gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_CRAWLING_FINISH,
+            "brick-path=%s", child->brick_path, NULL);
+
+    return NULL;
+}
+
+static void
+br_set_child_state(br_child_t *child, br_child_state_t state)
+{
+    pthread_mutex_lock(&child->lock);
+    {
+        _br_set_child_state(child, state);
+    }
+    pthread_mutex_unlock(&child->lock);
+}
+
+/**
+ * At this point a thread is spawned to crawl the filesystem (in
+ * tortoise pace) to sign objects that were not signed in previous run(s).
+ * Such objects are identified by examining it's dirtyness and timestamp.
+ *
+ *    pick object:
+ *       signature_is_stale() && (object_timestamp() <= stub_init_time())
+ *
+ * Also, we register to the changelog library to subscribe for event
+ * notifications.
+ */
+static int32_t
+br_enact_signer(xlator_t *this, br_child_t *child, br_stub_init_t *stub)
+{
+    int32_t ret = 0;
+    br_private_t *priv = NULL;
+    struct gf_brick_spec *brick = NULL;
+
+    priv = this->private;
+
+    brick = GF_CALLOC(1, sizeof(struct gf_brick_spec),
+                      gf_common_mt_gf_brick_spec_t);
+    if (!brick)
+        goto error_return;
+
+    br_fill_brick_spec(brick, stub->export);
+    ret = gf_changelog_register_generic(brick, 1, 1,
+                                        this->ctx->cmd_args.log_file, -1, this);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, BRB_MSG_REGISTER_FAILED, NULL);
+        goto dealloc;
+    }
+
+    child->threadrunning = 0;
+    ret = gf_thread_create(&child->thread, NULL, br_oneshot_signer, child,
+                           "brosign");
+    if (ret)
+        gf_smsg(this->name, GF_LOG_WARNING, 0, BRB_MSG_SPAWN_FAILED,
+                "FS-crawler-thread", NULL);
+    else
+        child->threadrunning = 1;
+
+    /* it's OK to continue, "old" objects would be signed when modified */
+    list_add_tail(&child->list, &priv->signing);
+    return 0;
+
+dealloc:
+    GF_FREE(brick);
+error_return:
+    return -1;
+}
+
+static int32_t
+br_launch_scrubber(xlator_t *this, br_child_t *child, struct br_scanfs *fsscan,
+                   struct br_scrubber *fsscrub)
+{
+    int32_t ret = -1;
+    br_private_t *priv = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+
+    scrub_monitor = &priv->scrub_monitor;
+    ret = gf_thread_create(&child->thread, NULL, br_fsscanner, child,
+                           "brfsscan");
+    if (ret != 0) {
+        gf_smsg(this->name, GF_LOG_ALERT, 0, BRB_MSG_SPAWN_FAILED,
+                "bitrot-scrubber-daemon Brick-path=%s", child->brick_path,
+                NULL);
+        goto error_return;
+    }
+
+    /* Signal monitor to kick off state machine*/
+    pthread_mutex_lock(&scrub_monitor->mutex);
+    {
+        if (!scrub_monitor->inited)
+            pthread_cond_signal(&scrub_monitor->cond);
+        scrub_monitor->inited = _gf_true;
+    }
+    pthread_mutex_unlock(&scrub_monitor->mutex);
+
+    /**
+     * Everything has been setup.. add this subvolume to scrubbers
+     * list.
+     */
+    pthread_mutex_lock(&fsscrub->mutex);
+    {
+        list_add_tail(&child->list, &fsscrub->scrublist);
+        pthread_cond_broadcast(&fsscrub->cond);
+    }
+    pthread_mutex_unlock(&fsscrub->mutex);
+
+    return 0;
+
+error_return:
+    return -1;
+}
+
+static int32_t
+br_enact_scrubber(xlator_t *this, br_child_t *child)
+{
+    int32_t ret = 0;
+    br_private_t *priv = NULL;
+    struct br_scanfs *fsscan = NULL;
+    struct br_scrubber *fsscrub = NULL;
+
+    priv = this->private;
+
+    fsscan = &child->fsscan;
+    fsscrub = &priv->fsscrub;
+
+    /**
+     * if this child already witnesses a successful connection earlier
+     * there's no need to initialize mutexes, condvars, etc..
+     */
+    if (_br_child_witnessed_connection(child))
+        return br_launch_scrubber(this, child, fsscan, fsscrub);
+
+    LOCK_INIT(&fsscan->entrylock);
+    pthread_mutex_init(&fsscan->waitlock, NULL);
+    pthread_cond_init(&fsscan->waitcond, NULL);
+
+    fsscan->entries = 0;
+    INIT_LIST_HEAD(&fsscan->queued);
+    INIT_LIST_HEAD(&fsscan->ready);
+
+    ret = br_launch_scrubber(this, child, fsscan, fsscrub);
+    if (ret)
+        goto error_return;
+
+    return 0;
+
+error_return:
+    LOCK_DESTROY(&fsscan->entrylock);
+    pthread_mutex_destroy(&fsscan->waitlock);
+    pthread_cond_destroy(&fsscan->waitcond);
+
+    return -1;
+}
+
+static int32_t
+br_child_enaction(xlator_t *this, br_child_t *child, br_stub_init_t *stub)
+{
+    int32_t ret = -1;
+    br_private_t *priv = this->private;
+
+    pthread_mutex_lock(&child->lock);
+    {
+        if (priv->iamscrubber)
+            ret = br_enact_scrubber(this, child);
+        else
+            ret = br_enact_signer(this, child, stub);
+
+        if (!ret) {
+            child->witnessed = 1;
+            _br_set_child_state(child, BR_CHILD_STATE_CONNECTED);
+            gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_CONNECTED_TO_BRICK,
+                    "brick-path=%s", child->brick_path, NULL);
+        }
+    }
+    pthread_mutex_unlock(&child->lock);
+
+    return ret;
+}
+
+/**
+ * This routine fetches various attributes associated with a child which
+ * is basically a subvolume. Attributes include brick path and the stub
+ * birth time. This is done by performing a lookup on the root followed
+ * by getxattr() on a virtual key. Depending on the configuration, the
+ * process either acts as a signer or a scrubber.
+ */
+int32_t
+br_brick_connect(xlator_t *this, br_child_t *child)
+{
+    int32_t ret = -1;
+    loc_t loc = {
+        0,
+    };
+    struct iatt buf = {
+        0,
+    };
+    struct iatt parent = {
+        0,
+    };
+    br_stub_init_t *stub = NULL;
+    dict_t *xattr = NULL;
+    int op_errno = 0;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, child, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    br_child_set_scrub_state(child, _gf_false);
+    br_set_child_state(child, BR_CHILD_STATE_INITIALIZING);
+
+    loc.inode = inode_ref(child->table->root);
+    gf_uuid_copy(loc.gfid, loc.inode->gfid);
+    loc.path = gf_strdup("/");
+
+    ret = syncop_lookup(child->xl, &loc, &buf, &parent, NULL, NULL);
+    if (ret) {
+        op_errno = -ret;
+        ret = -1;
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_LOOKUP_FAILED,
+                NULL);
+        goto wipeloc;
+    }
+
+    ret = syncop_getxattr(child->xl, &loc, &xattr,
+                          GLUSTERFS_GET_BR_STUB_INIT_TIME, NULL, NULL);
+    if (ret) {
+        op_errno = -ret;
+        ret = -1;
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno, BRB_MSG_GET_INFO_FAILED,
+                NULL);
+        goto wipeloc;
+    }
+
+    ret = dict_get_ptr(xattr, GLUSTERFS_GET_BR_STUB_INIT_TIME, (void **)&stub);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_GET_INFO_FAILED, NULL);
+        goto free_dict;
+    }
+
+    memcpy(child->brick_path, stub->export, strlen(stub->export) + 1);
+    child->tv.tv_sec = ntohl(stub->timebuf[0]);
+    child->tv.tv_usec = ntohl(stub->timebuf[1]);
+
+    ret = br_child_enaction(this, child, stub);
+
+free_dict:
+    dict_unref(xattr);
+wipeloc:
+    loc_wipe(&loc);
+out:
+    if (ret)
+        br_set_child_state(child, BR_CHILD_STATE_CONNFAILED);
+    return ret;
+}
+
+/* TODO: cleanup signer */
+static int32_t
+br_cleanup_signer(xlator_t *this, br_child_t *child)
+{
+    return 0;
+}
+
+static int32_t
+br_cleanup_scrubber(xlator_t *this, br_child_t *child)
+{
+    int32_t ret = 0;
+    br_private_t *priv = NULL;
+    struct br_scrubber *fsscrub = NULL;
+    struct br_monitor *scrub_monitor = NULL;
+
+    priv = this->private;
+    fsscrub = &priv->fsscrub;
+    scrub_monitor = &priv->scrub_monitor;
+
+    if (_br_is_child_scrub_active(child)) {
+        scrub_monitor->active_child_count--;
+        br_child_set_scrub_state(child, _gf_false);
+    }
+
+    /**
+     * 0x0: child (brick) goes out of rotation
+     *
+     * This is fully safe w.r.t. entries for this child being actively
+     * scrubbed. Each of the scrubber thread(s) would finish scrubbing
+     * the entry (probably failing due to disconnection) and either
+     * putting the entry back into the queue or continuing further.
+     * Either way, pending entries for this child's queue need not be
+     * drained; entries just sit there in the queued/ready list to be
+     * consumed later upon re-connection.
+     */
+    pthread_mutex_lock(&fsscrub->mutex);
+    {
+        list_del_init(&child->list);
+    }
+    pthread_mutex_unlock(&fsscrub->mutex);
+
+    /**
+     * 0x1: cleanup scanner thread
+     *
+     * The pending timer needs to be removed _after_ cleaning up the
+     * filesystem scanner (scheduling the next scrub time is not a
+     * cancellation point).
+     */
+    ret = gf_thread_cleanup_xint(child->thread);
+    if (ret)
+        gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUB_THREAD_CLEANUP, NULL);
+
+    gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_SCRUBBER_CLEANED,
+            "brick-path=%s", child->brick_path, NULL);
+
+    return 0;
+}
+
+/**
+ * OK.. this child has made it's mind to go down the drain. So,
+ * let's clean up what it touched. (NOTE: there's no need to clean
+ * the inode table, it's just reused taking care of stale inodes)
+ */
+int32_t
+br_brick_disconnect(xlator_t *this, br_child_t *child)
+{
+    int32_t ret = 0;
+    struct br_monitor *scrub_monitor = NULL;
+    br_private_t *priv = this->private;
+
+    scrub_monitor = &priv->scrub_monitor;
+
+    /* Lock order should be wakelock and then child lock to
+     * dead locks.
+     */
+    pthread_mutex_lock(&scrub_monitor->wakelock);
+    {
+        pthread_mutex_lock(&child->lock);
+        {
+            if (!_br_is_child_connected(child))
+                goto unblock;
+
+            /* child is on death row.. */
+            _br_set_child_state(child, BR_CHILD_STATE_DISCONNECTED);
+
+            if (priv->iamscrubber)
+                ret = br_cleanup_scrubber(this, child);
+            else
+                ret = br_cleanup_signer(this, child);
+        }
+    unblock:
+        pthread_mutex_unlock(&child->lock);
+    }
+    pthread_mutex_unlock(&scrub_monitor->wakelock);
+
+    return ret;
+}
+
+/**
+ * This function is executed in a separate thread. The thread gets the
+ * brick from where CHILD_UP has received from the queue and gets the
+ * information regarding that brick (such as brick path).
+ */
+void *
+br_handle_events(void *arg)
+{
+    int32_t ret = 0;
+    xlator_t *this = NULL;
+    br_private_t *priv = NULL;
+    br_child_t *child = NULL;
+    struct br_child_event *childev = NULL;
+
+    this = arg;
+    priv = this->private;
+
+    /*
+     * Since, this is the topmost xlator, THIS has to be set by bit-rot
+     * xlator itself (STACK_WIND won't help in this case). Also it has
+     * to be done for each thread that gets spawned. Otherwise, a new
+     * thread will get global_xlator's pointer when it does "THIS".
+     */
+    THIS = this;
+
+    while (1) {
+        pthread_mutex_lock(&priv->lock);
+        {
+            while (list_empty(&priv->bricks))
+                pthread_cond_wait(&priv->cond, &priv->lock);
+
+            childev = list_first_entry(&priv->bricks, struct br_child_event,
+                                       list);
+            list_del_init(&childev->list);
+        }
+        pthread_mutex_unlock(&priv->lock);
+
+        child = childev->child;
+        ret = childev->call(this, child);
+        if (ret)
+            gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_SUBVOL_CONNECT_FAILED,
+                    "name=%s", child->xl->name, NULL);
+        GF_FREE(childev);
+    }
+
+    return NULL;
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int32_t ret = -1;
+
+    if (!this)
+        return ret;
+
+    ret = xlator_mem_acct_init(this, gf_br_stub_mt_end + 1);
+
+    if (ret != 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, BRB_MSG_MEM_ACNT_FAILED, NULL);
+        return ret;
+    }
+
+    return ret;
+}
+
+static void
+_br_qchild_event(xlator_t *this, br_child_t *child, br_child_handler *call)
+{
+    br_private_t *priv = NULL;
+    struct br_child_event *childev = NULL;
+
+    priv = this->private;
+
+    childev = GF_CALLOC(1, sizeof(*childev), gf_br_mt_br_child_event_t);
+    if (!childev) {
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_EVENT_UNHANDLED,
+                "Brick-name=%s", child->xl->name, NULL);
+        return;
+    }
+
+    INIT_LIST_HEAD(&childev->list);
+    childev->this = this;
+    childev->child = child;
+    childev->call = call;
+
+    list_add_tail(&childev->list, &priv->bricks);
+}
+
+int
+br_scrubber_status_get(xlator_t *this, dict_t **dict)
+{
+    int ret = -1;
+    br_private_t *priv = NULL;
+    struct br_scrub_stats *scrub_stats = NULL;
+
+    priv = this->private;
+
+    GF_VALIDATE_OR_GOTO("bit-rot", priv, out);
+
+    scrub_stats = &priv->scrub_stat;
+
+    ret = br_get_bad_objects_list(this, dict);
+    if (ret) {
+        gf_msg_debug(this->name, 0,
+                     "Failed to collect corrupt "
+                     "files");
+    }
+
+    ret = dict_set_int8(*dict, "scrub-running", scrub_stats->scrub_running);
+    if (ret) {
+        gf_msg_debug(this->name, 0,
+                     "Failed setting scrub_running "
+                     "entry to the dictionary");
+    }
+
+    ret = dict_set_uint64(*dict, "scrubbed-files", scrub_stats->scrubbed_files);
+    if (ret) {
+        gf_msg_debug(this->name, 0,
+                     "Failed to setting scrubbed file "
+                     "entry to the dictionary");
+    }
+
+    ret = dict_set_uint64(*dict, "unsigned-files", scrub_stats->unsigned_files);
+    if (ret) {
+        gf_msg_debug(this->name, 0,
+                     "Failed to set unsigned file count"
+                     " entry to the dictionary");
+    }
+
+    ret = dict_set_uint64(*dict, "scrub-duration", scrub_stats->scrub_duration);
+    if (ret) {
+        gf_msg_debug(this->name, 0,
+                     "Failed to set scrub duration"
+                     " entry to the dictionary");
+    }
+
+    ret = dict_set_dynstr_with_alloc(*dict, "last-scrub-time",
+                                     scrub_stats->last_scrub_time);
+    if (ret) {
+        gf_msg_debug(this->name, 0,
+                     "Failed to set "
+                     "last scrub time value");
+    }
+
+out:
+    return ret;
+}
+
+int
+notify(xlator_t *this, int32_t event, void *data, ...)
+{
+    int idx = -1;
+    int ret = -1;
+    xlator_t *subvol = NULL;
+    br_child_t *child = NULL;
+    br_private_t *priv = NULL;
+    dict_t *output = NULL;
+    va_list ap;
+    struct br_monitor *scrub_monitor = NULL;
+
+    subvol = (xlator_t *)data;
+    priv = this->private;
+    scrub_monitor = &priv->scrub_monitor;
+
+    gf_msg_trace(this->name, 0, "Notification received: %d", event);
+
+    idx = br_find_child_index(this, subvol);
+
+    switch (event) {
+        case GF_EVENT_CHILD_UP:
+            if (idx < 0) {
+                gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_INVALID_SUBVOL,
+                        "event=%d", event, NULL);
+                goto out;
+            }
+
+            pthread_mutex_lock(&priv->lock);
+            {
+                child = &priv->children[idx];
+                if (child->child_up == 1)
+                    goto unblock_0;
+                priv->up_children++;
+
+                child->child_up = 1;
+                child->xl = subvol;
+                if (!child->table)
+                    child->table = inode_table_new(4096, subvol);
+
+                _br_qchild_event(this, child, br_brick_connect);
+                pthread_cond_signal(&priv->cond);
+            }
+        unblock_0:
+            pthread_mutex_unlock(&priv->lock);
+
+            if (priv->up_children == priv->child_count)
+                default_notify(this, event, data);
+            break;
+
+        case GF_EVENT_CHILD_DOWN:
+            if (idx < 0) {
+                gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_INVALID_SUBVOL,
+                        "event=%d", event, NULL);
+                goto out;
+            }
+
+            pthread_mutex_lock(&priv->lock);
+            {
+                child = &priv->children[idx];
+                if (child->child_up == 0)
+                    goto unblock_1;
+
+                child->child_up = 0;
+                priv->up_children--;
+
+                _br_qchild_event(this, child, br_brick_disconnect);
+                pthread_cond_signal(&priv->cond);
+            }
+        unblock_1:
+            pthread_mutex_unlock(&priv->lock);
+
+            if (priv->up_children == 0)
+                default_notify(this, event, data);
+            break;
+
+        case GF_EVENT_SCRUB_STATUS:
+            gf_msg_debug(this->name, GF_LOG_INFO,
+                         "BitRot scrub status "
+                         "called");
+            va_start(ap, data);
+            output = va_arg(ap, dict_t *);
+            va_end(ap);
+
+            ret = br_scrubber_status_get(this, &output);
+            gf_msg_debug(this->name, 0, "returning %d", ret);
+            break;
+
+        case GF_EVENT_SCRUB_ONDEMAND:
+            gf_log(this->name, GF_LOG_INFO,
+                   "BitRot scrub ondemand "
+                   "called");
+
+            if (scrub_monitor->state != BR_SCRUB_STATE_PENDING) {
+                gf_smsg(this->name, GF_LOG_ERROR, 0,
+                        BRB_MSG_RESCHEDULE_SCRUBBER_FAILED, "Current-state=%d",
+                        scrub_monitor->state, NULL);
+                return -2;
+            }
+
+            /* Needs synchronization with reconfigure thread */
+            pthread_mutex_lock(&priv->lock);
+            {
+                ret = br_scrub_state_machine(this, _gf_true);
+            }
+            pthread_mutex_unlock(&priv->lock);
+
+            if (ret) {
+                gf_smsg(this->name, GF_LOG_ERROR, 0,
+                        BRB_MSG_COULD_NOT_SCHEDULE_SCRUB, NULL);
+            }
+            gf_msg_debug(this->name, 0, "returning %d", ret);
+            break;
+        default:
+            default_notify(this, event, data);
+    }
+
+out:
+    return 0;
+}
+
+static void
+br_fini_signer(xlator_t *this, br_private_t *priv)
+{
+    int i = 0;
+
+    if (priv == NULL)
+        return;
+
+    for (; i < priv->signer_th_count; i++) {
+        (void)gf_thread_cleanup_xint(priv->obj_queue->workers[i]);
+    }
+    GF_FREE(priv->obj_queue->workers);
+
+    pthread_cond_destroy(&priv->object_cond);
+}
+
+/**
+ * Initialize signer specific structures, spawn worker threads.
+ */
+
+static int32_t
+br_init_signer(xlator_t *this, br_private_t *priv)
+{
+    int i = 0;
+    int32_t ret = -1;
+
+    /* initialize gfchangelog xlator context */
+    ret = gf_changelog_init(this);
+    if (ret)
+        goto out;
+
+    pthread_cond_init(&priv->object_cond, NULL);
+
+    priv->obj_queue = GF_CALLOC(1, sizeof(*priv->obj_queue),
+                                gf_br_mt_br_ob_n_wk_t);
+    if (!priv->obj_queue)
+        goto cleanup_cond;
+    INIT_LIST_HEAD(&priv->obj_queue->objects);
+
+    priv->obj_queue->workers = GF_CALLOC(
+        priv->signer_th_count, sizeof(pthread_t), gf_br_mt_br_worker_t);
+    if (!priv->obj_queue->workers)
+        goto cleanup_obj_queue;
+
+    for (i = 0; i < priv->signer_th_count; i++) {
+        ret = gf_thread_create(&priv->obj_queue->workers[i], NULL,
+                               br_process_object, this, "brpobj");
+        if (ret != 0) {
+            gf_smsg(this->name, GF_LOG_ERROR, -ret,
+                    BRB_MSG_THREAD_CREATION_FAILED, NULL);
+            ret = -1;
+            goto cleanup_threads;
+        }
+    }
+
+    return 0;
+
+cleanup_threads:
+    for (i--; i >= 0; i--) {
+        (void)gf_thread_cleanup_xint(priv->obj_queue->workers[i]);
+    }
+    GF_FREE(priv->obj_queue->workers);
+
+cleanup_obj_queue:
+    GF_FREE(priv->obj_queue);
+
+cleanup_cond:
+    /* that's explicit */
+    pthread_cond_destroy(&priv->object_cond);
+out:
+    return -1;
+}
+
+/**
+ * For signer, only rate limit CPU usage (during hash calculation) when
+ * compiled with -DBR_RATE_LIMIT_SIGNER cflags, else let it run full
+ * throttle.
+ */
+static int32_t
+br_rate_limit_signer(xlator_t *this, int child_count, int numbricks)
+{
+    br_private_t *priv = NULL;
+    tbf_opspec_t spec = {
+        0,
+    };
+
+    priv = this->private;
+
+    spec.op = TBF_OP_HASH;
+    spec.rate = 0;
+    spec.maxlimit = 0;
+
+    /**
+     * OK. Most implementations of TBF I've come across generate tokens
+     * every second (UML, etc..) and some chose sub-second granularity
+     * (blk-iothrottle cgroups). TBF algorithm itself does not enforce
+     * any logic for choosing generation interval and it seems pretty
+     * logical as one could jack up token count per interval w.r.t.
+     * generation rate.
+     *
+     * Value used here is chosen based on a series of test(s) performed
+     * to balance object signing time and not maxing out on all available
+     * CPU cores. It's obvious to have seconds granularity and jack up
+     * token count per interval, thereby achieving close to similar
+     * results. Let's stick to this as it seems to be working fine for
+     * the set of ops that are throttled.
+     **/
+    spec.token_gen_interval = 600000; /* In usec */
+
+#ifdef BR_RATE_LIMIT_SIGNER
+
+    double contribution = 0;
+    contribution = ((double)1 - ((double)child_count / (double)numbricks));
+    if (contribution == 0)
+        contribution = 1;
+    spec.rate = BR_HASH_CALC_READ_SIZE * contribution;
+    spec.maxlimit = priv->signer_th_count * BR_HASH_CALC_READ_SIZE;
+
+#endif
+
+    if (!spec.rate)
+        gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_RATE_LIMIT_INFO,
+                "FULL THROTTLE", NULL);
+    else
+        gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_RATE_LIMIT_INFO,
+                "tokens/sec-rate=%lu", spec.rate, "maxlimit=%lu", spec.maxlimit,
+                NULL);
+
+    priv->tbf = tbf_init(&spec, 1);
+    return priv->tbf ? 0 : -1;
+}
+
+static int32_t
+br_signer_handle_options(xlator_t *this, br_private_t *priv, dict_t *options)
+{
+    if (options) {
+        GF_OPTION_RECONF("expiry-time", priv->expiry_time, options, uint32,
+                         error_return);
+        GF_OPTION_RECONF("signer-threads", priv->signer_th_count, options,
+                         uint32, error_return);
+    } else {
+        GF_OPTION_INIT("expiry-time", priv->expiry_time, uint32, error_return);
+        GF_OPTION_INIT("signer-threads", priv->signer_th_count, uint32,
+                       error_return);
+    }
+
+    return 0;
+
+error_return:
+    return -1;
+}
+
+static int32_t
+br_signer_init(xlator_t *this, br_private_t *priv)
+{
+    int32_t ret = 0;
+    int numbricks = 0;
+
+    GF_OPTION_INIT("expiry-time", priv->expiry_time, uint32, error_return);
+    GF_OPTION_INIT("brick-count", numbricks, int32, error_return);
+    GF_OPTION_INIT("signer-threads", priv->signer_th_count, uint32,
+                   error_return);
+
+    ret = br_rate_limit_signer(this, priv->child_count, numbricks);
+    if (ret)
+        goto error_return;
+
+    ret = br_init_signer(this, priv);
+    if (ret)
+        goto cleanup_tbf;
+
+    return 0;
+
+cleanup_tbf:
+    /* cleanup TBF */
+error_return:
+    return -1;
+}
+
+static void
+br_free_scrubber_monitor(xlator_t *this, br_private_t *priv)
+{
+    struct br_monitor *scrub_monitor = &priv->scrub_monitor;
+
+    if (scrub_monitor->timer) {
+        (void)gf_tw_del_timer(priv->timer_wheel, scrub_monitor->timer);
+
+        GF_FREE(scrub_monitor->timer);
+        scrub_monitor->timer = NULL;
+    }
+
+    (void)gf_thread_cleanup_xint(scrub_monitor->thread);
+
+    /* Clean up cond and mutex variables */
+    pthread_mutex_destroy(&scrub_monitor->mutex);
+    pthread_cond_destroy(&scrub_monitor->cond);
+
+    pthread_mutex_destroy(&scrub_monitor->wakelock);
+    pthread_cond_destroy(&scrub_monitor->wakecond);
+
+    pthread_mutex_destroy(&scrub_monitor->donelock);
+    pthread_cond_destroy(&scrub_monitor->donecond);
+
+    LOCK_DESTROY(&scrub_monitor->lock);
+}
+
+static void
+br_free_children(xlator_t *this, br_private_t *priv, int count)
+{
+    br_child_t *child = NULL;
+
+    for (--count; count >= 0; count--) {
+        child = &priv->children[count];
+        mem_pool_destroy(child->timer_pool);
+        pthread_mutex_destroy(&child->lock);
+    }
+
+    GF_FREE(priv->children);
+    priv->children = NULL;
+}
+
+static int
+br_init_children(xlator_t *this, br_private_t *priv)
+{
+    int i = 0;
+    br_child_t *child = NULL;
+    xlator_list_t *trav = NULL;
+
+    priv->child_count = xlator_subvolume_count(this);
+    priv->children = GF_CALLOC(priv->child_count, sizeof(*priv->children),
+                               gf_br_mt_br_child_t);
+    if (!priv->children)
+        goto err;
+
+    trav = this->children;
+    while (trav) {
+        child = &priv->children[i];
+
+        pthread_mutex_init(&child->lock, NULL);
+        child->witnessed = 0;
+
+        br_set_child_state(child, BR_CHILD_STATE_DISCONNECTED);
+
+        child->this = this;
+        child->xl = trav->xlator;
+
+        child->timer_pool = mem_pool_new(struct gf_tw_timer_list, 4096);
+        if (!child->timer_pool) {
+            gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_MEM_POOL_ALLOC,
+                    NULL);
+            errno = ENOMEM;
+            goto freechild;
+        }
+
+        INIT_LIST_HEAD(&child->list);
+
+        i++;
+        trav = trav->next;
+    }
+
+    return 0;
+
+freechild:
+    br_free_children(this, priv, i);
+err:
+    return -1;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    int32_t ret = -1;
+    br_private_t *priv = NULL;
+
+    if (!this->children) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_NO_CHILD, NULL);
+        goto out;
+    }
+
+    priv = GF_CALLOC(1, sizeof(*priv), gf_br_mt_br_private_t);
+    if (!priv) {
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, BRB_MSG_NO_MEMORY, NULL);
+        goto out;
+    }
+
+    GF_OPTION_INIT("scrubber", priv->iamscrubber, bool, free_priv);
+
+    ret = br_init_children(this, priv);
+    if (ret)
+        goto free_priv;
+
+    pthread_mutex_init(&priv->lock, NULL);
+    pthread_cond_init(&priv->cond, NULL);
+
+    INIT_LIST_HEAD(&priv->bricks);
+    INIT_LIST_HEAD(&priv->signing);
+
+    priv->timer_wheel = glusterfs_ctx_tw_get(this->ctx);
+    if (!priv->timer_wheel) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_TIMER_WHEEL_UNAVAILABLE,
+                NULL);
+        goto cleanup;
+    }
+
+    this->private = priv;
+
+    if (!priv->iamscrubber) {
+        ret = br_signer_init(this, priv);
+        if (!ret)
+            ret = br_signer_handle_options(this, priv, NULL);
+    } else {
+        ret = br_scrubber_init(this, priv);
+        if (!ret)
+            ret = br_scrubber_handle_options(this, priv, NULL);
+    }
+
+    if (ret)
+        goto cleanup;
+
+    ret = gf_thread_create(&priv->thread, NULL, br_handle_events, this,
+                           "brhevent");
+    if (ret != 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, -ret, BRB_MSG_THREAD_CREATION_FAILED,
+                NULL);
+        ret = -1;
+    }
+
+    if (!ret) {
+        gf_smsg(this->name, GF_LOG_INFO, 0, BRB_MSG_BITROT_LOADED, "mode=%s",
+                (priv->iamscrubber) ? "SCRUBBER" : "SIGNER", NULL);
+        return 0;
+    }
+
+cleanup:
+    (void)pthread_cond_destroy(&priv->cond);
+    (void)pthread_mutex_destroy(&priv->lock);
+
+    br_free_children(this, priv, priv->child_count);
+
+free_priv:
+    GF_FREE(priv);
+out:
+    this->private = NULL;
+    return -1;
+}
+
+void
+fini(xlator_t *this)
+{
+    br_private_t *priv = this->private;
+
+    if (!priv)
+        return;
+
+    if (!priv->iamscrubber)
+        br_fini_signer(this, priv);
+    else
+        (void)br_free_scrubber_monitor(this, priv);
+
+    br_free_children(this, priv, priv->child_count);
+
+    this->private = NULL;
+    GF_FREE(priv);
+
+    glusterfs_ctx_tw_put(this->ctx);
+
+    return;
+}
+
+static void
+br_reconfigure_monitor(xlator_t *this)
+{
+    int32_t ret = 0;
+
+    ret = br_scrub_state_machine(this, _gf_false);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRB_MSG_COULD_NOT_SCHEDULE_SCRUB,
+                NULL);
+    }
+}
+
+static int
+br_reconfigure_scrubber(xlator_t *this, dict_t *options)
+{
+    int32_t ret = -1;
+    br_private_t *priv = NULL;
+
+    priv = this->private;
+
+    pthread_mutex_lock(&priv->lock);
+    {
+        ret = br_scrubber_handle_options(this, priv, options);
+    }
+    pthread_mutex_unlock(&priv->lock);
+
+    if (ret)
+        goto err;
+
+    /* change state for all _up_ subvolume(s) */
+    pthread_mutex_lock(&priv->lock);
+    {
+        br_reconfigure_monitor(this);
+    }
+    pthread_mutex_unlock(&priv->lock);
+
+err:
+    return ret;
+}
+
+static int
+br_reconfigure_signer(xlator_t *this, dict_t *options)
+{
+    br_private_t *priv = this->private;
+
+    return br_signer_handle_options(this, priv, options);
+}
+
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    int ret = 0;
+    br_private_t *priv = NULL;
+
+    priv = this->private;
+
+    if (priv->iamscrubber)
+        ret = br_reconfigure_scrubber(this, options);
+    else
+        ret = br_reconfigure_signer(this, options);
+
+    return ret;
+}
+
+struct xlator_fops fops;
+
+struct xlator_cbks cbks;
+
+struct volume_options options[] = {
+    {
+        .key = {"expiry-time"},
+        .type = GF_OPTION_TYPE_INT,
+        .default_value = SIGNING_TIMEOUT,
+        .op_version = {GD_OP_VERSION_3_7_0},
+        .flags = OPT_FLAG_SETTABLE,
+        .description = "Waiting time for an object on which it waits "
+                       "before it is signed",
+    },
+    {
+        .key = {"brick-count"},
+        .type = GF_OPTION_TYPE_STR,
+        .description = "Total number of bricks for the current node for "
+                       "all volumes in the trusted storage pool.",
+    },
+    {
+        .key = {"scrubber", "scrub"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "false",
+        .op_version = {GD_OP_VERSION_3_7_0},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_FORCE,
+        .description = "option to run as a scrubber",
+    },
+    {
+        .key = {"scrub-throttle"},
+        .type = GF_OPTION_TYPE_STR,
+        .default_value = "lazy",
+        .op_version = {GD_OP_VERSION_3_7_0},
+        .flags = OPT_FLAG_SETTABLE,
+        .description = "Scrub-throttle value is a measure of how fast "
+                       "or slow the scrubber scrubs the filesystem for "
+                       "volume <VOLNAME>",
+    },
+    {
+        .key = {"scrub-freq"},
+        .type = GF_OPTION_TYPE_STR,
+        .default_value = "biweekly",
+        .op_version = {GD_OP_VERSION_3_7_0},
+        .flags = OPT_FLAG_SETTABLE,
+        .description = "Scrub frequency for volume <VOLNAME>",
+    },
+    {
+        .key = {"scrub-state"},
+        .type = GF_OPTION_TYPE_STR,
+        .default_value = "active",
+        .op_version = {GD_OP_VERSION_4_0_0},
+        .flags = OPT_FLAG_SETTABLE,
+        .description = "Pause/Resume scrub. Upon resume, scrubber "
+                       "continues from where it left off.",
+    },
+    {
+        .key = {"signer-threads"},
+        .type = GF_OPTION_TYPE_INT,
+        .default_value = BR_WORKERS,
+        .op_version = {GD_OP_VERSION_8_0},
+        .flags = OPT_FLAG_SETTABLE,
+        .description = "Number of signing process threads. As a best "
+                       "practice, set this to the number of processor cores",
+    },
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "bit-rot",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot.h b/xlators/features/bit-rot/src/bitd/bit-rot.h
new file mode 100644
index 00000000000..8ac7dcdac3d
--- /dev/null
+++ b/xlators/features/bit-rot/src/bitd/bit-rot.h
@@ -0,0 +1,302 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __BIT_ROT_H__
+#define __BIT_ROT_H__
+
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/dict.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/syncop.h>
+#include <glusterfs/syncop-utils.h>
+#include "changelog.h"
+#include "timer-wheel.h"
+
+#include <glusterfs/throttle-tbf.h>
+#include "bit-rot-ssm.h"
+
+#include "bit-rot-common.h"
+#include "bit-rot-stub-mem-types.h"
+#include "bit-rot-scrub-status.h"
+
+#include <openssl/sha.h>
+
+typedef enum scrub_throttle {
+    BR_SCRUB_THROTTLE_VOID = -1,
+    BR_SCRUB_THROTTLE_LAZY = 0,
+    BR_SCRUB_THROTTLE_NORMAL = 1,
+    BR_SCRUB_THROTTLE_AGGRESSIVE = 2,
+    BR_SCRUB_THROTTLE_STALLED = 3,
+} scrub_throttle_t;
+
+typedef enum scrub_freq {
+    BR_FSSCRUB_FREQ_HOURLY = 1,
+    BR_FSSCRUB_FREQ_DAILY,
+    BR_FSSCRUB_FREQ_WEEKLY,
+    BR_FSSCRUB_FREQ_BIWEEKLY,
+    BR_FSSCRUB_FREQ_MONTHLY,
+    BR_FSSCRUB_FREQ_MINUTE,
+    BR_FSSCRUB_FREQ_STALLED,
+} scrub_freq_t;
+
+#define signature_size(hl) (sizeof(br_isignature_t) + hl + 1)
+
+struct br_scanfs {
+    gf_lock_t entrylock;
+
+    pthread_mutex_t waitlock;
+    pthread_cond_t waitcond;
+
+    unsigned int entries;
+    struct list_head queued;
+    struct list_head ready;
+};
+
+/* just need three states to track child status */
+typedef enum br_child_state {
+    BR_CHILD_STATE_CONNECTED = 1,
+    BR_CHILD_STATE_INITIALIZING,
+    BR_CHILD_STATE_CONNFAILED,
+    BR_CHILD_STATE_DISCONNECTED,
+} br_child_state_t;
+
+struct br_child {
+    pthread_mutex_t lock;     /* protects child state */
+    char witnessed;           /* witnessed at least one successful
+                                 connection */
+    br_child_state_t c_state; /* current state of this child */
+
+    char child_up;             /* Indicates whether this child is
+                                  up or not */
+    xlator_t *xl;              /* client xlator corresponding to
+                                  this child */
+    inode_table_t *table;      /* inode table for this child */
+    char brick_path[PATH_MAX]; /* brick export directory of this
+                                  child */
+    struct list_head list;     /* hook to attach to the list of
+                                  UP children */
+    xlator_t *this;            /* Bit rot xlator */
+
+    pthread_t thread;  /* initial crawler for unsigned
+                          object(s) or scrub crawler */
+    int threadrunning; /* active thread */
+
+    struct mem_pool *timer_pool; /* timer-wheel's timer mem-pool */
+
+    struct timeval tv;
+
+    struct br_scanfs fsscan; /* per subvolume FS scanner */
+
+    gf_boolean_t active_scrubbing; /* Actively scrubbing or not */
+};
+
+typedef struct br_child br_child_t;
+
+struct br_obj_n_workers {
+    struct list_head objects; /* queue of objects expired from the
+                                 timer wheel and ready to be picked
+                                 up for signing */
+    pthread_t *workers;       /* Threads which pick up the objects
+                                 from the above queue and start
+                                 signing each object */
+};
+
+struct br_scrubber {
+    xlator_t *this;
+
+    scrub_throttle_t throttle;
+
+    /**
+     * frequency of scanning for this subvolume. this should
+     * normally be per-child, but since all children follow the
+     * same frequency for a volume, this option ends up here
+     * instead of br_child_t.
+     */
+    scrub_freq_t frequency;
+
+    gf_boolean_t frequency_reconf;
+    gf_boolean_t throttle_reconf;
+
+    pthread_mutex_t mutex;
+    pthread_cond_t cond;
+
+    unsigned int nr_scrubbers;
+    struct list_head scrubbers;
+
+    /**
+     * list of "rotatable" subvolume(s) undergoing scrubbing
+     */
+    struct list_head scrublist;
+};
+
+struct br_monitor {
+    gf_lock_t lock;
+    pthread_t thread; /* Monitor thread */
+
+    gf_boolean_t inited;
+    pthread_mutex_t mutex;
+    pthread_cond_t cond; /* Thread starts and will be waiting on cond.
+                            First child which is up wakes this up */
+
+    xlator_t *this;
+    /* scheduler */
+    uint32_t boot;
+
+    int32_t active_child_count; /* Number of children currently scrubbing */
+    gf_boolean_t kick;          /* This variable tracks the scrubber is
+                                 * kicked or not. Both 'kick' and
+                                 * 'active_child_count' uses the same pair
+                                 * of mutex-cond variable, i.e, wakelock and
+                                 * wakecond. */
+
+    pthread_mutex_t wakelock;
+    pthread_cond_t wakecond;
+
+    gf_boolean_t done;
+    pthread_mutex_t donelock;
+    pthread_cond_t donecond;
+
+    struct gf_tw_timer_list *timer;
+    br_scrub_state_t state; /* current scrub state */
+};
+
+typedef struct br_obj_n_workers br_obj_n_workers_t;
+
+typedef struct br_private br_private_t;
+
+typedef void (*br_scrubbed_file_update)(br_private_t *priv);
+
+struct br_private {
+    pthread_mutex_t lock;
+
+    struct list_head bricks; /* list of bricks from which enents
+                                have been received */
+
+    struct list_head signing;
+
+    pthread_cond_t object_cond; /* handling signing of objects */
+    int child_count;
+    br_child_t *children; /* list of subvolumes */
+    int up_children;
+
+    pthread_cond_t cond; /* handling CHILD_UP notifications */
+    pthread_t thread;    /* thread for connecting each UP
+                            child with changelog */
+
+    struct tvec_base *timer_wheel; /* timer wheel where the objects which
+                                      changelog has sent sits and waits
+                                      for expiry */
+    br_obj_n_workers_t *obj_queue; /* place holder for all the objects
+                                      that are expired from timer wheel
+                                      and ready to be picked up for
+                                      signing and the workers which sign
+                                      the objects */
+
+    uint32_t expiry_time; /* objects "wait" time */
+
+    uint32_t signer_th_count; /* Number of signing process threads */
+
+    tbf_t *tbf; /* token bucket filter */
+
+    gf_boolean_t iamscrubber; /* function as a fs scrubber */
+
+    struct br_scrub_stats scrub_stat; /* statistics of scrub*/
+
+    struct br_scrubber fsscrub; /* scrubbers for this subvolume */
+
+    struct br_monitor scrub_monitor; /* scrubber monitor */
+};
+
+struct br_object {
+    xlator_t *this;
+
+    uuid_t gfid;
+
+    unsigned long signedversion; /* version against which this object will
+                                    be signed */
+    br_child_t *child;           /* object's subvolume */
+
+    int sign_info;
+
+    struct list_head list; /* hook to add to the queue once the
+                              object is expired from timer wheel */
+    void *data;
+};
+
+typedef struct br_object br_object_t;
+typedef int32_t(br_scrub_ssm_call)(xlator_t *);
+
+void
+br_log_object(xlator_t *, char *, uuid_t, int32_t);
+
+void
+br_log_object_path(xlator_t *, char *, const char *, int32_t);
+
+int32_t
+br_calculate_obj_checksum(unsigned char *, br_child_t *, fd_t *, struct iatt *);
+
+int32_t
+br_prepare_loc(xlator_t *, br_child_t *, loc_t *, gf_dirent_t *, loc_t *);
+
+gf_boolean_t
+bitd_is_bad_file(xlator_t *, br_child_t *, loc_t *, fd_t *);
+
+static inline void
+_br_set_child_state(br_child_t *child, br_child_state_t state)
+{
+    child->c_state = state;
+}
+
+static inline int
+_br_is_child_connected(br_child_t *child)
+{
+    return (child->c_state == BR_CHILD_STATE_CONNECTED);
+}
+
+static inline int
+_br_is_child_scrub_active(br_child_t *child)
+{
+    return child->active_scrubbing;
+}
+
+static inline int
+_br_child_failed_conn(br_child_t *child)
+{
+    return (child->c_state == BR_CHILD_STATE_CONNFAILED);
+}
+
+static inline int
+_br_child_witnessed_connection(br_child_t *child)
+{
+    return (child->witnessed == 1);
+}
+
+/* scrub state */
+static inline void
+_br_monitor_set_scrub_state(struct br_monitor *scrub_monitor,
+                            br_scrub_state_t state)
+{
+    scrub_monitor->state = state;
+}
+
+static inline br_scrub_event_t
+_br_child_get_scrub_event(struct br_scrubber *fsscrub)
+{
+    return (fsscrub->frequency == BR_FSSCRUB_FREQ_STALLED)
+               ? BR_SCRUB_EVENT_PAUSE
+               : BR_SCRUB_EVENT_SCHEDULE;
+}
+
+int32_t
+br_get_bad_objects_list(xlator_t *this, dict_t **dict);
+
+#endif /* __BIT_ROT_H__ */
diff --git a/xlators/features/bit-rot/src/stub/Makefile.am b/xlators/features/bit-rot/src/stub/Makefile.am
new file mode 100644
index 00000000000..f13de7145fc
--- /dev/null
+++ b/xlators/features/bit-rot/src/stub/Makefile.am
@@ -0,0 +1,20 @@
+if WITH_SERVER
+xlator_LTLIBRARIES = bitrot-stub.la
+endif
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+bitrot_stub_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+bitrot_stub_la_SOURCES = bit-rot-stub-helpers.c bit-rot-stub.c
+bitrot_stub_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = bit-rot-stub.h bit-rot-common.h bit-rot-stub-mem-types.h \
+	bit-rot-object-version.h bit-rot-stub-messages.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src \
+	-I$(top_srcdir)/rpc/rpc-lib/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/bit-rot/src/stub/bit-rot-common.h b/xlators/features/bit-rot/src/stub/bit-rot-common.h
new file mode 100644
index 00000000000..20561aa7764
--- /dev/null
+++ b/xlators/features/bit-rot/src/stub/bit-rot-common.h
@@ -0,0 +1,178 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __BIT_ROT_COMMON_H__
+#define __BIT_ROT_COMMON_H__
+
+#include <glusterfs/glusterfs.h>
+#include "bit-rot-object-version.h"
+
+#define BR_VXATTR_VERSION (1 << 0)
+#define BR_VXATTR_SIGNATURE (1 << 1)
+
+#define BR_VXATTR_SIGN_MISSING (BR_VXATTR_SIGNATURE)
+#define BR_VXATTR_ALL_MISSING (BR_VXATTR_VERSION | BR_VXATTR_SIGNATURE)
+
+#define BR_BAD_OBJ_CONTAINER                                                   \
+    (uuid_t) { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 8 }
+
+typedef enum br_vxattr_state {
+    BR_VXATTR_STATUS_FULL = 0,
+    BR_VXATTR_STATUS_MISSING = 1,
+    BR_VXATTR_STATUS_UNSIGNED = 2,
+    BR_VXATTR_STATUS_INVALID = 3,
+} br_vxattr_status_t;
+
+typedef enum br_sign_state {
+    BR_SIGN_INVALID = -1,
+    BR_SIGN_NORMAL = 0,
+    BR_SIGN_REOPEN_WAIT = 1,
+    BR_SIGN_QUICK = 2,
+} br_sign_state_t;
+
+static inline br_vxattr_status_t
+br_version_xattr_state(dict_t *xattr, br_version_t **obuf,
+                       br_signature_t **sbuf, gf_boolean_t *objbad)
+{
+    int32_t ret = 0;
+    int32_t vxattr = 0;
+    br_vxattr_status_t status;
+    void *data = NULL;
+
+    /**
+     * The key being present in the dict indicates the xattr was set on
+     * disk. The presence of xattr itself as of now is suffecient to say
+     * the the object is bad.
+     */
+    *objbad = _gf_false;
+    ret = dict_get_bin(xattr, BITROT_OBJECT_BAD_KEY, (void **)&data);
+    if (!ret)
+        *objbad = _gf_true;
+
+    ret = dict_get_bin(xattr, BITROT_CURRENT_VERSION_KEY, (void **)obuf);
+    if (ret)
+        vxattr |= BR_VXATTR_VERSION;
+
+    ret = dict_get_bin(xattr, BITROT_SIGNING_VERSION_KEY, (void **)sbuf);
+    if (ret)
+        vxattr |= BR_VXATTR_SIGNATURE;
+
+    switch (vxattr) {
+        case 0:
+            status = BR_VXATTR_STATUS_FULL;
+            break;
+        case BR_VXATTR_SIGN_MISSING:
+            status = BR_VXATTR_STATUS_UNSIGNED;
+            break;
+        case BR_VXATTR_ALL_MISSING:
+            status = BR_VXATTR_STATUS_MISSING;
+            break;
+        default:
+            status = BR_VXATTR_STATUS_INVALID;
+    }
+
+    return status;
+}
+
+/**
+ * in-memory representation of signature used by signer for object
+ * signing.
+ */
+typedef struct br_isignature_in {
+    int8_t signaturetype; /* signature type            */
+
+    unsigned long signedversion; /* version against which the
+                                    object was signed         */
+
+    size_t signaturelen; /* signature length          */
+    char signature[0];   /* object signature          */
+} br_isignature_t;
+
+/**
+ * in-memory representation of signature used by scrubber for object
+ * verification.
+ */
+typedef struct br_isignature_out {
+    char stale; /* stale signature?          */
+
+    unsigned long version; /* current signed version    */
+
+    uint32_t time[2]; /* time when the object
+                         got dirtied               */
+
+    int8_t signaturetype; /* hash type                 */
+    size_t signaturelen;  /* signature length          */
+    char signature[0];    /* signature (hash)          */
+} br_isignature_out_t;
+
+typedef struct br_stub_init {
+    uint32_t timebuf[2];
+    char export[PATH_MAX];
+} br_stub_init_t;
+
+typedef enum {
+    BR_SIGNATURE_TYPE_VOID = -1,  /* object is not signed       */
+    BR_SIGNATURE_TYPE_ZERO = 0,   /* min boundary               */
+    BR_SIGNATURE_TYPE_SHA256 = 1, /* signed with SHA256         */
+    BR_SIGNATURE_TYPE_MAX = 2,    /* max boundary               */
+} br_signature_type;
+
+/* BitRot stub start time (virtual xattr) */
+#define GLUSTERFS_GET_BR_STUB_INIT_TIME "trusted.glusterfs.bit-rot.stub-init"
+
+/* signing/reopen hint */
+#define BR_OBJECT_RESIGN 0
+#define BR_OBJECT_REOPEN 1
+#define BR_REOPEN_SIGN_HINT_KEY "trusted.glusterfs.bit-rot.reopen-hint"
+
+static inline int
+br_is_signature_type_valid(int8_t signaturetype)
+{
+    return ((signaturetype > BR_SIGNATURE_TYPE_ZERO) &&
+            (signaturetype < BR_SIGNATURE_TYPE_MAX));
+}
+
+static inline void
+br_set_default_ongoingversion(br_version_t *buf, uint32_t *tv)
+{
+    buf->ongoingversion = BITROT_DEFAULT_CURRENT_VERSION;
+    buf->timebuf[0] = tv[0];
+    buf->timebuf[1] = tv[1];
+}
+
+static inline void
+br_set_default_signature(br_signature_t *buf, size_t *size)
+{
+    buf->signaturetype = (int8_t)BR_SIGNATURE_TYPE_VOID;
+    buf->signedversion = BITROT_DEFAULT_SIGNING_VERSION;
+
+    *size = sizeof(br_signature_t); /* no signature */
+}
+
+static inline void
+br_set_ongoingversion(br_version_t *buf, unsigned long version, uint32_t *tv)
+{
+    buf->ongoingversion = version;
+    buf->timebuf[0] = tv[0];
+    buf->timebuf[1] = tv[1];
+}
+
+static inline void
+br_set_signature(br_signature_t *buf, br_isignature_t *sign,
+                 size_t signaturelen, size_t *size)
+{
+    buf->signaturetype = sign->signaturetype;
+    buf->signedversion = ntohl(sign->signedversion);
+
+    memcpy(buf->signature, sign->signature, signaturelen);
+    *size = sizeof(br_signature_t) + signaturelen;
+}
+
+#endif /* __BIT_ROT_COMMON_H__ */
diff --git a/xlators/features/bit-rot/src/stub/bit-rot-object-version.h b/xlators/features/bit-rot/src/stub/bit-rot-object-version.h
new file mode 100644
index 00000000000..7ae6a5200df
--- /dev/null
+++ b/xlators/features/bit-rot/src/stub/bit-rot-object-version.h
@@ -0,0 +1,30 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __BIT_ROT_OBJECT_VERSION_H
+#define __BIT_ROT_OBJECT_VERSION_H
+
+/**
+ * on-disk formats for ongoing version and object signature.
+ */
+typedef struct br_version {
+    unsigned long ongoingversion;
+    uint32_t timebuf[2];
+} br_version_t;
+
+typedef struct __attribute__((__packed__)) br_signature {
+    int8_t signaturetype;
+
+    unsigned long signedversion;
+
+    char signature[0];
+} br_signature_t;
+
+#endif
diff --git a/xlators/features/bit-rot/src/stub/bit-rot-stub-helpers.c b/xlators/features/bit-rot/src/stub/bit-rot-stub-helpers.c
new file mode 100644
index 00000000000..8ac13a09941
--- /dev/null
+++ b/xlators/features/bit-rot/src/stub/bit-rot-stub-helpers.c
@@ -0,0 +1,796 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include "bit-rot-stub.h"
+
+br_stub_fd_t *
+br_stub_fd_new(void)
+{
+    br_stub_fd_t *br_stub_fd = NULL;
+
+    br_stub_fd = GF_CALLOC(1, sizeof(*br_stub_fd), gf_br_stub_mt_br_stub_fd_t);
+
+    return br_stub_fd;
+}
+
+int
+__br_stub_fd_ctx_set(xlator_t *this, fd_t *fd, br_stub_fd_t *br_stub_fd)
+{
+    uint64_t value = 0;
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("bit-rot-stub", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+    GF_VALIDATE_OR_GOTO(this->name, br_stub_fd, out);
+
+    value = (uint64_t)(long)br_stub_fd;
+
+    ret = __fd_ctx_set(fd, this, value);
+
+out:
+    return ret;
+}
+
+br_stub_fd_t *
+__br_stub_fd_ctx_get(xlator_t *this, fd_t *fd)
+{
+    br_stub_fd_t *br_stub_fd = NULL;
+    uint64_t value = 0;
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("bit-rot-stub", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    ret = __fd_ctx_get(fd, this, &value);
+    if (ret)
+        return NULL;
+
+    br_stub_fd = (br_stub_fd_t *)((long)value);
+
+out:
+    return br_stub_fd;
+}
+
+br_stub_fd_t *
+br_stub_fd_ctx_get(xlator_t *this, fd_t *fd)
+{
+    br_stub_fd_t *br_stub_fd = NULL;
+
+    GF_VALIDATE_OR_GOTO("bit-rot-stub", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    LOCK(&fd->lock);
+    {
+        br_stub_fd = __br_stub_fd_ctx_get(this, fd);
+    }
+    UNLOCK(&fd->lock);
+
+out:
+    return br_stub_fd;
+}
+
+int32_t
+br_stub_fd_ctx_set(xlator_t *this, fd_t *fd, br_stub_fd_t *br_stub_fd)
+{
+    int32_t ret = -1;
+
+    GF_VALIDATE_OR_GOTO("bit-rot-stub", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+    GF_VALIDATE_OR_GOTO(this->name, br_stub_fd, out);
+
+    LOCK(&fd->lock);
+    {
+        ret = __br_stub_fd_ctx_set(this, fd, br_stub_fd);
+    }
+    UNLOCK(&fd->lock);
+
+out:
+    return ret;
+}
+
+/**
+ * Adds an entry to the bad objects directory.
+ * @gfid: gfid of the bad object being added to the bad objects directory
+ */
+int
+br_stub_add(xlator_t *this, uuid_t gfid)
+{
+    char gfid_path[BR_PATH_MAX_PLUS] = {0};
+    char bad_gfid_path[BR_PATH_MAX_PLUS] = {0};
+    int ret = 0;
+    br_stub_private_t *priv = NULL;
+    struct stat st = {0};
+
+    priv = this->private;
+    GF_ASSERT_AND_GOTO_WITH_ERROR(this->name, !gf_uuid_is_null(gfid), out,
+                                  errno, EINVAL);
+
+    snprintf(gfid_path, sizeof(gfid_path), "%s/%s", priv->stub_basepath,
+             uuid_utoa(gfid));
+
+    ret = sys_stat(gfid_path, &st);
+    if (!ret)
+        goto out;
+    snprintf(bad_gfid_path, sizeof(bad_gfid_path), "%s/stub-%s",
+             priv->stub_basepath, uuid_utoa(priv->bad_object_dir_gfid));
+
+    ret = sys_link(bad_gfid_path, gfid_path);
+    if (ret) {
+        if ((errno != ENOENT) && (errno != EMLINK) && (errno != EEXIST))
+            goto out;
+
+        /*
+         * Continue with success. At least we'll have half of the
+         * functionality, in the sense, object is marked bad and
+         * would be inaccessible. It's only scrub status that would
+         * show up less number of objects. That's fine as we'll have
+         * the log files that will have the missing information.
+         */
+        gf_smsg(this->name, GF_LOG_WARNING, errno, BRS_MSG_LINK_FAIL, "gfid=%s",
+                uuid_utoa(gfid), NULL);
+    }
+
+    return 0;
+out:
+    return -1;
+}
+
+int
+br_stub_del(xlator_t *this, uuid_t gfid)
+{
+    int32_t op_errno __attribute__((unused)) = 0;
+    br_stub_private_t *priv = NULL;
+    int ret = 0;
+    char gfid_path[BR_PATH_MAX_PLUS] = {0};
+
+    priv = this->private;
+    GF_ASSERT_AND_GOTO_WITH_ERROR(this->name, !gf_uuid_is_null(gfid), out,
+                                  op_errno, EINVAL);
+    snprintf(gfid_path, sizeof(gfid_path), "%s/%s", priv->stub_basepath,
+             uuid_utoa(gfid));
+    ret = sys_unlink(gfid_path);
+    if (ret && (errno != ENOENT)) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, BRS_MSG_BAD_OBJ_UNLINK_FAIL,
+                "path=%s", gfid_path, NULL);
+        ret = -errno;
+        goto out;
+    }
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+static int
+br_stub_check_stub_directory(xlator_t *this, char *fullpath)
+{
+    int ret = 0;
+    struct stat st = {
+        0,
+    };
+    char oldpath[BR_PATH_MAX_PLUS] = {0};
+    br_stub_private_t *priv = NULL;
+
+    priv = this->private;
+
+    snprintf(oldpath, sizeof(oldpath), "%s/%s", priv->export,
+             OLD_BR_STUB_QUARANTINE_DIR);
+
+    ret = sys_stat(fullpath, &st);
+    if (!ret && !S_ISDIR(st.st_mode))
+        goto error_return;
+    if (ret) {
+        if (errno != ENOENT)
+            goto error_return;
+        ret = sys_stat(oldpath, &st);
+        if (ret)
+            ret = mkdir_p(fullpath, 0600, _gf_true);
+        else
+            ret = sys_rename(oldpath, fullpath);
+    }
+
+    if (ret)
+        gf_smsg(this->name, GF_LOG_ERROR, errno, BRS_MSG_BAD_OBJECT_DIR_FAIL,
+                "create-path=%s", fullpath, NULL);
+    return ret;
+
+error_return:
+    gf_smsg(this->name, GF_LOG_ERROR, errno, BRS_MSG_BAD_OBJECT_DIR_FAIL,
+            "verify-path=%s", fullpath, NULL);
+    return -1;
+}
+
+/**
+ * Function to create the container for the bad objects within the bad objects
+ * directory.
+ */
+static int
+br_stub_check_stub_file(xlator_t *this, char *path)
+{
+    int ret = 0;
+    int fd = -1;
+    struct stat st = {
+        0,
+    };
+
+    ret = sys_stat(path, &st);
+    if (!ret && !S_ISREG(st.st_mode))
+        goto error_return;
+    if (ret) {
+        if (errno != ENOENT)
+            goto error_return;
+        fd = sys_creat(path, 0);
+        if (fd < 0)
+            gf_smsg(this->name, GF_LOG_ERROR, errno,
+                    BRS_MSG_BAD_OBJECT_DIR_FAIL, "create-path=%s", path, NULL);
+    }
+
+    if (fd >= 0) {
+        sys_close(fd);
+        ret = 0;
+    }
+
+    return ret;
+
+error_return:
+    gf_smsg(this->name, GF_LOG_ERROR, errno, BRS_MSG_BAD_OBJECT_DIR_FAIL,
+            "verify-path=%s", path, NULL);
+    return -1;
+}
+
+int
+br_stub_dir_create(xlator_t *this, br_stub_private_t *priv)
+{
+    int ret = -1;
+    char fullpath[BR_PATH_MAX_PLUS] = {
+        0,
+    };
+    char stub_gfid_path[BR_PATH_MAX_PLUS] = {
+        0,
+    };
+
+    gf_uuid_copy(priv->bad_object_dir_gfid, BR_BAD_OBJ_CONTAINER);
+
+    if (snprintf(fullpath, sizeof(fullpath), "%s", priv->stub_basepath) >=
+        sizeof(fullpath))
+        goto out;
+
+    if (snprintf(stub_gfid_path, sizeof(stub_gfid_path), "%s/stub-%s",
+                 priv->stub_basepath, uuid_utoa(priv->bad_object_dir_gfid)) >=
+        sizeof(stub_gfid_path))
+        goto out;
+
+    ret = br_stub_check_stub_directory(this, fullpath);
+    if (ret)
+        goto out;
+    ret = br_stub_check_stub_file(this, stub_gfid_path);
+    if (ret)
+        goto out;
+
+    return 0;
+
+out:
+    return -1;
+}
+
+call_stub_t *
+__br_stub_dequeue(struct list_head *callstubs)
+{
+    call_stub_t *stub = NULL;
+
+    if (!list_empty(callstubs)) {
+        stub = list_entry(callstubs->next, call_stub_t, list);
+        list_del_init(&stub->list);
+    }
+
+    return stub;
+}
+
+void
+__br_stub_enqueue(struct list_head *callstubs, call_stub_t *stub)
+{
+    list_add_tail(&stub->list, callstubs);
+}
+
+void
+br_stub_worker_enqueue(xlator_t *this, call_stub_t *stub)
+{
+    br_stub_private_t *priv = NULL;
+
+    priv = this->private;
+    pthread_mutex_lock(&priv->container.bad_lock);
+    {
+        __br_stub_enqueue(&priv->container.bad_queue, stub);
+        pthread_cond_signal(&priv->container.bad_cond);
+    }
+    pthread_mutex_unlock(&priv->container.bad_lock);
+}
+
+void *
+br_stub_worker(void *data)
+{
+    br_stub_private_t *priv = NULL;
+    xlator_t *this = NULL;
+    call_stub_t *stub = NULL;
+
+    THIS = data;
+    this = data;
+    priv = this->private;
+
+    for (;;) {
+        pthread_mutex_lock(&priv->container.bad_lock);
+        {
+            while (list_empty(&priv->container.bad_queue)) {
+                (void)pthread_cond_wait(&priv->container.bad_cond,
+                                        &priv->container.bad_lock);
+            }
+
+            stub = __br_stub_dequeue(&priv->container.bad_queue);
+        }
+        pthread_mutex_unlock(&priv->container.bad_lock);
+
+        if (stub) /* guard against spurious wakeups */
+            call_resume(stub);
+    }
+
+    return NULL;
+}
+
+int32_t
+br_stub_lookup_wrapper(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                       dict_t *xattr_req)
+{
+    br_stub_private_t *priv = NULL;
+    struct stat lstatbuf = {0};
+    int ret = 0;
+    int32_t op_errno = EINVAL;
+    int32_t op_ret = -1;
+    struct iatt stbuf = {
+        0,
+    };
+    struct iatt postparent = {
+        0,
+    };
+    dict_t *xattr = NULL;
+    gf_boolean_t ver_enabled = _gf_false;
+
+    BR_STUB_VER_ENABLED_IN_CALLPATH(frame, ver_enabled);
+    priv = this->private;
+    BR_STUB_VER_COND_GOTO(priv, (!ver_enabled), done);
+
+    VALIDATE_OR_GOTO(loc, done);
+    if (gf_uuid_compare(loc->gfid, priv->bad_object_dir_gfid))
+        goto done;
+
+    ret = sys_lstat(priv->stub_basepath, &lstatbuf);
+    if (ret) {
+        gf_msg_debug(this->name, errno,
+                     "Stat failed on stub bad "
+                     "object dir");
+        op_errno = errno;
+        goto done;
+    } else if (!S_ISDIR(lstatbuf.st_mode)) {
+        gf_msg_debug(this->name, errno,
+                     "bad object container is not "
+                     "a directory");
+        op_errno = ENOTDIR;
+        goto done;
+    }
+
+    iatt_from_stat(&stbuf, &lstatbuf);
+    gf_uuid_copy(stbuf.ia_gfid, priv->bad_object_dir_gfid);
+
+    op_ret = op_errno = 0;
+    xattr = dict_new();
+    if (!xattr) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+    }
+
+done:
+    STACK_UNWIND_STRICT(lookup, frame, op_ret, op_errno, loc->inode, &stbuf,
+                        xattr, &postparent);
+    if (xattr)
+        dict_unref(xattr);
+    return 0;
+}
+
+static int
+is_bad_gfid_file_current(char *filename, uuid_t gfid)
+{
+    char current_stub_gfid[GF_UUID_BUF_SIZE + 16] = {
+        0,
+    };
+
+    snprintf(current_stub_gfid, sizeof current_stub_gfid, "stub-%s",
+             uuid_utoa(gfid));
+    return (!strcmp(filename, current_stub_gfid));
+}
+
+static void
+check_delete_stale_bad_file(xlator_t *this, char *filename)
+{
+    int ret = 0;
+    struct stat st = {0};
+    char filepath[BR_PATH_MAX_PLUS] = {0};
+    br_stub_private_t *priv = NULL;
+
+    priv = this->private;
+
+    if (is_bad_gfid_file_current(filename, priv->bad_object_dir_gfid))
+        return;
+
+    snprintf(filepath, sizeof(filepath), "%s/%s", priv->stub_basepath,
+             filename);
+
+    ret = sys_stat(filepath, &st);
+    if (!ret && st.st_nlink == 1)
+        sys_unlink(filepath);
+}
+
+static int
+br_stub_fill_readdir(fd_t *fd, br_stub_fd_t *fctx, DIR *dir, off_t off,
+                     size_t size, gf_dirent_t *entries)
+{
+    off_t in_case = -1;
+    off_t last_off = 0;
+    size_t filled = 0;
+    int count = 0;
+    int32_t this_size = -1;
+    gf_dirent_t *this_entry = NULL;
+    xlator_t *this = NULL;
+    struct dirent *entry = NULL;
+    struct dirent scratch[2] = {
+        {
+            0,
+        },
+    };
+
+    this = THIS;
+    if (!off) {
+        rewinddir(dir);
+    } else {
+        seekdir(dir, off);
+#ifndef GF_LINUX_HOST_OS
+        if ((u_long)telldir(dir) != off && off != fctx->bad_object.dir_eof) {
+            gf_smsg(THIS->name, GF_LOG_ERROR, 0,
+                    BRS_MSG_BAD_OBJECT_DIR_SEEK_FAIL, "off=(0x%llx)", off,
+                    "dir=%p", dir, NULL);
+            errno = EINVAL;
+            count = -1;
+            goto out;
+        }
+#endif /* GF_LINUX_HOST_OS */
+    }
+
+    while (filled <= size) {
+        in_case = (u_long)telldir(dir);
+
+        if (in_case == -1) {
+            gf_smsg(THIS->name, GF_LOG_ERROR, 0,
+                    BRS_MSG_BAD_OBJECT_DIR_TELL_FAIL, "dir=%p", dir, "err=%s",
+                    strerror(errno), NULL);
+            goto out;
+        }
+
+        errno = 0;
+        entry = sys_readdir(dir, scratch);
+        if (!entry || errno != 0) {
+            if (errno == EBADF) {
+                gf_smsg(THIS->name, GF_LOG_WARNING, 0,
+                        BRS_MSG_BAD_OBJECT_DIR_READ_FAIL, "dir=%p", dir,
+                        "err=%s", strerror(errno), NULL);
+                goto out;
+            }
+            break;
+        }
+
+        if (!strcmp(entry->d_name, ".") || !strcmp(entry->d_name, ".."))
+            continue;
+
+        if (!strncmp(entry->d_name, "stub-", strlen("stub-"))) {
+            check_delete_stale_bad_file(this, entry->d_name);
+            continue;
+        }
+
+        this_size = max(sizeof(gf_dirent_t), sizeof(gfs3_dirplist)) +
+                    strlen(entry->d_name) + 1;
+
+        if (this_size + filled > size) {
+            seekdir(dir, in_case);
+#ifndef GF_LINUX_HOST_OS
+            if ((u_long)telldir(dir) != in_case &&
+                in_case != fctx->bad_object.dir_eof) {
+                gf_smsg(THIS->name, GF_LOG_ERROR, 0,
+                        BRS_MSG_BAD_OBJECT_DIR_SEEK_FAIL, "in_case=(0x%llx)",
+                        in_case, "dir=%p", dir, NULL);
+                errno = EINVAL;
+                count = -1;
+                goto out;
+            }
+#endif /* GF_LINUX_HOST_OS */
+            break;
+        }
+
+        this_entry = gf_dirent_for_name(entry->d_name);
+
+        if (!this_entry) {
+            gf_smsg(THIS->name, GF_LOG_ERROR, 0,
+                    BRS_MSG_CREATE_GF_DIRENT_FAILED, "entry-name=%s",
+                    entry->d_name, "err=%s", strerror(errno), NULL);
+            goto out;
+        }
+        /*
+         * we store the offset of next entry here, which is
+         * probably not intended, but code using syncop_readdir()
+         * (glfs-heal.c, afr-self-heald.c, pump.c) rely on it
+         * for directory read resumption.
+         */
+        last_off = (u_long)telldir(dir);
+        this_entry->d_off = last_off;
+        this_entry->d_ino = entry->d_ino;
+
+        list_add_tail(&this_entry->list, &entries->list);
+
+        filled += this_size;
+        count++;
+    }
+
+    if ((!sys_readdir(dir, scratch) && (errno == 0))) {
+        /* Indicate EOF */
+        errno = ENOENT;
+        /* Remember EOF offset for later detection */
+        fctx->bad_object.dir_eof = last_off;
+    }
+out:
+    return count;
+}
+
+int32_t
+br_stub_readdir_wrapper(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                        size_t size, off_t off, dict_t *xdata)
+{
+    br_stub_fd_t *fctx = NULL;
+    DIR *dir = NULL;
+    int ret = -1;
+    int32_t op_ret = -1;
+    int32_t op_errno = 0;
+    int count = 0;
+    gf_dirent_t entries;
+    gf_boolean_t xdata_unref = _gf_false;
+    dict_t *dict = NULL;
+
+    INIT_LIST_HEAD(&entries.list);
+
+    fctx = br_stub_fd_ctx_get(this, fd);
+    if (!fctx) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, BRS_MSG_GET_FD_CONTEXT_FAILED,
+                "fd=%p", fd, NULL);
+        op_errno = -ret;
+        goto done;
+    }
+
+    dir = fctx->bad_object.dir;
+
+    if (!dir) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, BRS_MSG_BAD_HANDLE_DIR_NULL,
+                "fd=%p", fd, NULL);
+        op_errno = EINVAL;
+        goto done;
+    }
+
+    count = br_stub_fill_readdir(fd, fctx, dir, off, size, &entries);
+
+    /* pick ENOENT to indicate EOF */
+    op_errno = errno;
+    op_ret = count;
+
+    dict = xdata;
+    (void)br_stub_bad_objects_path(this, fd, &entries, &dict);
+    if (!xdata && dict) {
+        xdata = dict;
+        xdata_unref = _gf_true;
+    }
+
+done:
+    STACK_UNWIND_STRICT(readdir, frame, op_ret, op_errno, &entries, xdata);
+    gf_dirent_free(&entries);
+    if (xdata_unref)
+        dict_unref(xdata);
+    return 0;
+}
+
+/**
+ * This function is called to mainly obtain the paths of the corrupt
+ * objects (files as of now). Currently scrub status prints only the
+ * gfid of the corrupted files. Reason is, bitrot-stub maintains the
+ * list of the corrupted objects as entries inside the quarantine
+ * directory (<brick export>/.glusterfs/quarantine)
+ *
+ * And the name of each entry in the qurantine directory is the gfid
+ * of the corrupted object. So scrub status will just show that info.
+ * But it helps the users a lot if the actual path to the object is
+ * also reported. Hence the below function to get that information.
+ * The function allocates a new dict to be returned (if it does not
+ * get one from the caller of readdir i.e. scrubber as of now), and
+ * stores the paths of each corrupted gfid there. The gfid is used as
+ * the key and path is used as the value.
+ *
+ * NOTE: The path will be there in following situations
+ * 1) gfid2path option has been enabled (posix xlator option)
+ *    and the corrupted file contains the path as an extended
+ *    attribute.
+ * 2) If the gfid2path option is not enabled, OR if the xattr
+ *    is absent, then the inode table should have it.
+ *    The path will be there if a name based lookup has happened
+ *    on the file which has been corrupted. With lookup a inode and
+ *    dentry would be created in the inode table. And the path is
+ *    constructed using the in memory inode and dentry. If a lookup
+ *    has not happened OR the inode corresponding to the corrupted
+ *    file does not exist in the inode table (because it got purged
+ *    as lru limit of the inodes exceeded) OR a nameless lookup had
+ *    happened to populate the inode in the inode table, then the
+ *    path will not be printed in scrub and only the gfid will be there.
+ **/
+int
+br_stub_bad_objects_path(xlator_t *this, fd_t *fd, gf_dirent_t *entries,
+                         dict_t **dict)
+{
+    gf_dirent_t *entry = NULL;
+    inode_t *inode = NULL;
+    char *hpath = NULL;
+    uuid_t gfid = {0};
+    int ret = -1;
+    dict_t *tmp_dict = NULL;
+    char str_gfid[64] = {0};
+
+    if (list_empty(&entries->list))
+        return 0;
+
+    tmp_dict = *dict;
+
+    if (!tmp_dict) {
+        tmp_dict = dict_new();
+        /*
+         * If the allocation of dict fails then no need treat it
+         * it as a error. This path (or function) is executed when
+         * "gluster volume bitrot <volume name> scrub status" is
+         * executed, to get the list of the corrupted objects.
+         * And the motive of this function is to get the paths of
+         * the corrupted objects. If the dict allocation fails, then
+         * the scrub status will only show the gfids of those corrupted
+         * objects (which is the behavior as of the time of this patch
+         * being worked upon). So just return and only the gfids will
+         * be shown.
+         */
+        if (!tmp_dict) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_ALLOC_FAILED, NULL);
+            goto out;
+        }
+    }
+
+    list_for_each_entry(entry, &entries->list, list)
+    {
+        gf_uuid_clear(gfid);
+        gf_uuid_parse(entry->d_name, gfid);
+
+        inode = inode_find(fd->inode->table, gfid);
+
+        /* No need to check the return value here.
+         * Because @hpath is examined.
+         */
+        (void)br_stub_get_path_of_gfid(this, fd->inode, inode, gfid, &hpath);
+
+        if (hpath) {
+            gf_msg_debug(this->name, 0,
+                         "path of the corrupted "
+                         "object (gfid: %s) is %s",
+                         uuid_utoa(gfid), hpath);
+            br_stub_entry_xattr_fill(this, hpath, entry, tmp_dict);
+        } else
+            gf_smsg(this->name, GF_LOG_WARNING, 0, BRS_MSG_PATH_GET_FAILED,
+                    "gfid=%s", uuid_utoa_r(gfid, str_gfid), NULL);
+
+        inode = NULL;
+        hpath = NULL;
+    }
+
+    ret = 0;
+    *dict = tmp_dict;
+
+out:
+    return ret;
+}
+
+int
+br_stub_get_path_of_gfid(xlator_t *this, inode_t *parent, inode_t *inode,
+                         uuid_t gfid, char **path)
+{
+    int32_t ret = -1;
+    char gfid_str[64] = {0};
+
+    GF_VALIDATE_OR_GOTO("bitrot-stub", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, parent, out);
+    GF_VALIDATE_OR_GOTO(this->name, path, out);
+
+    /* Above, No need to validate the @inode for hard resolution. Because
+     * inode can be NULL and if it is NULL, then syncop_gfid_to_path_hard
+     * will allocate a new inode and proceed. So no need to bother about
+     * @inode. Because we need it only to send a syncop_getxattr call
+     * from inside syncop_gfid_to_path_hard. And getxattr fetches the
+     * path from the backend.
+     */
+
+    ret = syncop_gfid_to_path_hard(parent->table, FIRST_CHILD(this), gfid,
+                                   inode, path, _gf_true);
+    if (ret < 0)
+        gf_smsg(this->name, GF_LOG_WARNING, 0, BRS_MSG_PATH_GET_FAILED,
+                "gfid=%s", uuid_utoa_r(gfid, gfid_str), NULL);
+
+    /*
+     * Try with soft resolution of path if hard resolve fails. Because
+     * checking the xattr on disk to get the path of a inode (or gfid)
+     * is dependent on whether that option is enabled in the posix
+     * xlator or not. If it is not enabled, then hard resolution by
+     * checking the on disk xattr fails.
+     *
+     * Thus in such situations fall back to the soft resolution which
+     * mainly depends on the inode_path() function. And for using
+     * inode_path, @inode has to be linked i.e. a successful lookup should
+     * have happened on the gfid (or the path) to link the inode to the
+     * inode table. And if @inode is NULL, means, the inode has not been
+     * found in the inode table and better not to do inode_path() on the
+     * inode which has not been linked.
+     */
+    if (ret < 0 && inode) {
+        ret = syncop_gfid_to_path_hard(parent->table, FIRST_CHILD(this), gfid,
+                                       inode, path, _gf_false);
+        if (ret < 0)
+            gf_smsg(this->name, GF_LOG_WARNING, 0, BRS_MSG_PATH_GET_FAILED,
+                    "from-memory  gfid=%s", uuid_utoa_r(gfid, gfid_str), NULL);
+    }
+
+out:
+    return ret;
+}
+
+/**
+ * NOTE: If the file has multiple hardlinks (in gluster volume
+ * namespace), the path would be one of the hardlinks. Its up to
+ * the user to find the remaining hardlinks (using find -samefile)
+ * and remove them.
+ **/
+void
+br_stub_entry_xattr_fill(xlator_t *this, char *hpath, gf_dirent_t *entry,
+                         dict_t *dict)
+{
+    int32_t ret = -1;
+
+    GF_VALIDATE_OR_GOTO("bit-rot-stub", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, hpath, out);
+
+    /*
+     * Use the entry->d_name (which is nothing but the gfid of the
+     * corrupted object) as the key. And the value will be the actual
+     * path of that object (or file).
+     *
+     * ALso ignore the dict_set errors. scrubber will get the gfid of
+     * the corrupted object for sure. So, for now lets just log the
+     * dict_set_dynstr failure and move on.
+     */
+
+    ret = dict_set_dynstr(dict, entry->d_name, hpath);
+    if (ret)
+        gf_smsg(this->name, GF_LOG_WARNING, 0, BRS_MSG_DICT_SET_FAILED,
+                "path=%s", hpath, "object-name=%s", entry->d_name, NULL);
+out:
+    return;
+}
diff --git a/xlators/features/bit-rot/src/stub/bit-rot-stub-mem-types.h b/xlators/features/bit-rot/src/stub/bit-rot-stub-mem-types.h
new file mode 100644
index 00000000000..9d93caf069f
--- /dev/null
+++ b/xlators/features/bit-rot/src/stub/bit-rot-stub-mem-types.h
@@ -0,0 +1,36 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef _BR_MEM_TYPES_H
+#define _BR_MEM_TYPES_H
+
+#include <glusterfs/mem-types.h>
+
+enum br_mem_types {
+    gf_br_stub_mt_private_t = gf_common_mt_end + 1,
+    gf_br_stub_mt_version_t,
+    gf_br_stub_mt_inode_ctx_t,
+    gf_br_stub_mt_signature_t,
+    gf_br_mt_br_private_t,
+    gf_br_mt_br_child_t,
+    gf_br_mt_br_object_t,
+    gf_br_mt_br_ob_n_wk_t,
+    gf_br_mt_br_scrubber_t,
+    gf_br_mt_br_fsscan_entry_t,
+    gf_br_stub_mt_br_stub_fd_t,
+    gf_br_stub_mt_br_scanner_freq_t,
+    gf_br_stub_mt_sigstub_t,
+    gf_br_mt_br_child_event_t,
+    gf_br_stub_mt_misc,
+    gf_br_mt_br_worker_t,
+    gf_br_stub_mt_end,
+};
+
+#endif
diff --git a/xlators/features/bit-rot/src/stub/bit-rot-stub-messages.h b/xlators/features/bit-rot/src/stub/bit-rot-stub-messages.h
new file mode 100644
index 00000000000..6c15a166f18
--- /dev/null
+++ b/xlators/features/bit-rot/src/stub/bit-rot-stub-messages.h
@@ -0,0 +1,117 @@
+/*
+ Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+ This file is part of GlusterFS.
+
+ This file is licensed to you under your choice of the GNU Lesser
+ General Public License, version 3 or any later version (LGPLv3 or
+ later), or the GNU General Public License, version 2 (GPLv2), in all
+ cases as published by the Free Software Foundation.
+ */
+
+#ifndef _BITROT_STUB_MESSAGES_H_
+#define _BITROT_STUB_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(BITROT_STUB, BRS_MSG_NO_MEMORY, BRS_MSG_SET_EVENT_FAILED,
+           BRS_MSG_MEM_ACNT_FAILED, BRS_MSG_CREATE_FRAME_FAILED,
+           BRS_MSG_SET_CONTEXT_FAILED, BRS_MSG_CHANGE_VERSION_FAILED,
+           BRS_MSG_ADD_FD_TO_LIST_FAILED, BRS_MSG_SET_FD_CONTEXT_FAILED,
+           BRS_MSG_CREATE_ANONYMOUS_FD_FAILED, BRS_MSG_NO_CHILD,
+           BRS_MSG_STUB_ALLOC_FAILED, BRS_MSG_GET_INODE_CONTEXT_FAILED,
+           BRS_MSG_CANCEL_SIGN_THREAD_FAILED, BRS_MSG_ADD_FD_TO_INODE,
+           BRS_MSG_SIGN_VERSION_ERROR, BRS_MSG_BAD_OBJ_MARK_FAIL,
+           BRS_MSG_NON_SCRUB_BAD_OBJ_MARK, BRS_MSG_REMOVE_INTERNAL_XATTR,
+           BRS_MSG_SET_INTERNAL_XATTR, BRS_MSG_BAD_OBJECT_ACCESS,
+           BRS_MSG_BAD_CONTAINER_FAIL, BRS_MSG_BAD_OBJECT_DIR_FAIL,
+           BRS_MSG_BAD_OBJECT_DIR_SEEK_FAIL, BRS_MSG_BAD_OBJECT_DIR_TELL_FAIL,
+           BRS_MSG_BAD_OBJECT_DIR_READ_FAIL, BRS_MSG_GET_FD_CONTEXT_FAILED,
+           BRS_MSG_BAD_HANDLE_DIR_NULL, BRS_MSG_BAD_OBJ_THREAD_FAIL,
+           BRS_MSG_BAD_OBJ_DIR_CLOSE_FAIL, BRS_MSG_LINK_FAIL,
+           BRS_MSG_BAD_OBJ_UNLINK_FAIL, BRS_MSG_DICT_SET_FAILED,
+           BRS_MSG_PATH_GET_FAILED, BRS_MSG_NULL_LOCAL,
+           BRS_MSG_SPAWN_SIGN_THRD_FAILED, BRS_MSG_KILL_SIGN_THREAD,
+           BRS_MSG_NON_BITD_PID, BRS_MSG_SIGN_PREPARE_FAIL,
+           BRS_MSG_USING_DEFAULT_THREAD_SIZE, BRS_MSG_ALLOC_MEM_FAILED,
+           BRS_MSG_DICT_ALLOC_FAILED, BRS_MSG_CREATE_GF_DIRENT_FAILED,
+           BRS_MSG_ALLOC_FAILED, BRS_MSG_PATH_XATTR_GET_FAILED,
+           BRS_MSG_VERSION_PREPARE_FAIL);
+
+#define BRS_MSG_MEM_ACNT_FAILED_STR "Memory accounting init failed"
+#define BRS_MSG_BAD_OBJ_THREAD_FAIL_STR "pthread_init failed"
+#define BRS_MSG_USING_DEFAULT_THREAD_SIZE_STR "Using default thread stack size"
+#define BRS_MSG_NO_CHILD_STR "FATAL: no children"
+#define BRS_MSG_SPAWN_SIGN_THRD_FAILED_STR                                     \
+    "failed to create the new thread for signer"
+#define BRS_MSG_BAD_CONTAINER_FAIL_STR                                         \
+    "failed to launch the thread for storing bad gfids"
+#define BRS_MSG_CANCEL_SIGN_THREAD_FAILED_STR                                  \
+    "Could not cancel sign serializer thread"
+#define BRS_MSG_KILL_SIGN_THREAD_STR "killed the signer thread"
+#define BRS_MSG_GET_INODE_CONTEXT_FAILED_STR                                   \
+    "failed to init the inode context for the inode"
+#define BRS_MSG_ADD_FD_TO_INODE_STR "failed to add fd to the inode"
+#define BRS_MSG_NO_MEMORY_STR "local allocation failed"
+#define BRS_MSG_BAD_OBJECT_ACCESS_STR "bad object accessed. Returning"
+#define BRS_MSG_SIGN_VERSION_ERROR_STR "Signing version exceeds current version"
+#define BRS_MSG_NON_BITD_PID_STR                                               \
+    "PID from where signature request came, does not belong to bit-rot "       \
+    "daemon. Unwinding the fop"
+#define BRS_MSG_SIGN_PREPARE_FAIL_STR                                          \
+    "failed to prepare the signature. Unwinding the fop"
+#define BRS_MSG_VERSION_PREPARE_FAIL_STR                                       \
+    "failed to prepare the version. Unwinding the fop"
+#define BRS_MSG_STUB_ALLOC_FAILED_STR "failed to allocate stub fop, Unwinding"
+#define BRS_MSG_BAD_OBJ_MARK_FAIL_STR "failed to mark object as bad"
+#define BRS_MSG_NON_SCRUB_BAD_OBJ_MARK_STR                                     \
+    "bad object marking is not from the scrubber"
+#define BRS_MSG_ALLOC_MEM_FAILED_STR "failed to allocate memory"
+#define BRS_MSG_SET_INTERNAL_XATTR_STR "called on the internal xattr"
+#define BRS_MSG_REMOVE_INTERNAL_XATTR_STR "removexattr called on internal xattr"
+#define BRS_MSG_CREATE_ANONYMOUS_FD_FAILED_STR                                 \
+    "failed to create anonymous fd for the inode"
+#define BRS_MSG_ADD_FD_TO_LIST_FAILED_STR "failed add fd to the list"
+#define BRS_MSG_SET_FD_CONTEXT_FAILED_STR                                      \
+    "failed to set the fd context for the file"
+#define BRS_MSG_NULL_LOCAL_STR "local is NULL"
+#define BRS_MSG_DICT_ALLOC_FAILED_STR                                          \
+    "dict allocation failed: cannot send IPC FOP to changelog"
+#define BRS_MSG_SET_EVENT_FAILED_STR "cannot set release event in dict"
+#define BRS_MSG_CREATE_FRAME_FAILED_STR "create_frame() failure"
+#define BRS_MSG_BAD_OBJ_DIR_CLOSE_FAIL_STR "closedir error"
+#define BRS_MSG_LINK_FAIL_STR "failed to record gfid"
+#define BRS_MSG_BAD_OBJ_UNLINK_FAIL_STR                                        \
+    "failed to delete bad object link from quaratine directory"
+#define BRS_MSG_BAD_OBJECT_DIR_FAIL_STR "failed stub directory"
+#define BRS_MSG_BAD_OBJECT_DIR_SEEK_FAIL_STR                                   \
+    "seekdir failed. Invalid argument (offset reused from another DIR * "      \
+    "structure)"
+#define BRS_MSG_BAD_OBJECT_DIR_TELL_FAIL_STR "telldir failed on dir"
+#define BRS_MSG_BAD_OBJECT_DIR_READ_FAIL_STR "readdir failed on dir"
+#define BRS_MSG_CREATE_GF_DIRENT_FAILED_STR "could not create gf_dirent"
+#define BRS_MSG_GET_FD_CONTEXT_FAILED_STR "pfd is NULL"
+#define BRS_MSG_BAD_HANDLE_DIR_NULL_STR "dir if NULL"
+#define BRS_MSG_ALLOC_FAILED_STR                                               \
+    "failed to allocate new dict for saving the paths of the corrupted "       \
+    "objects. Scrub status will only display the gfid"
+#define BRS_MSG_PATH_GET_FAILED_STR "failed to get the path"
+#define BRS_MSG_PATH_XATTR_GET_FAILED_STR                                      \
+    "failed to get the path xattr from disk for the gfid. Trying to get path " \
+    "from the memory"
+#define BRS_MSG_DICT_SET_FAILED_STR                                            \
+    "failed to set the actual path as the value in the dict for the "          \
+    "corrupted object"
+#define BRS_MSG_SET_CONTEXT_FAILED_STR                                         \
+    "could not set fd context for release callback"
+#define BRS_MSG_CHANGE_VERSION_FAILED_STR "change version failed"
+#endif /* !_BITROT_STUB_MESSAGES_H_ */
diff --git a/xlators/features/bit-rot/src/stub/bit-rot-stub.c b/xlators/features/bit-rot/src/stub/bit-rot-stub.c
new file mode 100644
index 00000000000..447dd47ff41
--- /dev/null
+++ b/xlators/features/bit-rot/src/stub/bit-rot-stub.c
@@ -0,0 +1,3590 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <ctype.h>
+#include <sys/uio.h>
+#include <signal.h>
+
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/logging.h>
+#include "changelog.h"
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/call-stub.h>
+
+#include "bit-rot-stub.h"
+#include "bit-rot-stub-mem-types.h"
+#include "bit-rot-stub-messages.h"
+#include "bit-rot-common.h"
+
+#define BR_STUB_REQUEST_COOKIE 0x1
+
+void
+br_stub_lock_cleaner(void *arg)
+{
+    pthread_mutex_t *clean_mutex = arg;
+
+    pthread_mutex_unlock(clean_mutex);
+    return;
+}
+
+void *
+br_stub_signth(void *);
+
+struct br_stub_signentry {
+    unsigned long v;
+
+    call_stub_t *stub;
+
+    struct list_head list;
+};
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int32_t ret = -1;
+
+    if (!this)
+        return ret;
+
+    ret = xlator_mem_acct_init(this, gf_br_stub_mt_end + 1);
+
+    if (ret != 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, BRS_MSG_MEM_ACNT_FAILED, NULL);
+        return ret;
+    }
+
+    return ret;
+}
+
+int
+br_stub_bad_object_container_init(xlator_t *this, br_stub_private_t *priv)
+{
+    pthread_attr_t w_attr;
+    int ret = -1;
+
+    ret = pthread_cond_init(&priv->container.bad_cond, NULL);
+    if (ret != 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_BAD_OBJ_THREAD_FAIL,
+                "cond_init ret=%d", ret, NULL);
+        goto out;
+    }
+
+    ret = pthread_mutex_init(&priv->container.bad_lock, NULL);
+    if (ret != 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_BAD_OBJ_THREAD_FAIL,
+                "mutex_init ret=%d", ret, NULL);
+        goto cleanup_cond;
+    }
+
+    ret = pthread_attr_init(&w_attr);
+    if (ret != 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_BAD_OBJ_THREAD_FAIL,
+                "attr_init ret=%d", ret, NULL);
+        goto cleanup_lock;
+    }
+
+    ret = pthread_attr_setstacksize(&w_attr, BAD_OBJECT_THREAD_STACK_SIZE);
+    if (ret == EINVAL) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0,
+                BRS_MSG_USING_DEFAULT_THREAD_SIZE, NULL);
+    }
+
+    INIT_LIST_HEAD(&priv->container.bad_queue);
+    ret = br_stub_dir_create(this, priv);
+    if (ret < 0)
+        goto cleanup_lock;
+
+    ret = gf_thread_create(&priv->container.thread, &w_attr, br_stub_worker,
+                           this, "brswrker");
+    if (ret)
+        goto cleanup_attr;
+
+    return 0;
+
+cleanup_attr:
+    pthread_attr_destroy(&w_attr);
+cleanup_lock:
+    pthread_mutex_destroy(&priv->container.bad_lock);
+cleanup_cond:
+    pthread_cond_destroy(&priv->container.bad_cond);
+out:
+    return -1;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    int ret = 0;
+    char *tmp = NULL;
+    struct timeval tv = {
+        0,
+    };
+    br_stub_private_t *priv = NULL;
+
+    if (!this->children) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_NO_CHILD, NULL);
+        goto error_return;
+    }
+
+    priv = GF_CALLOC(1, sizeof(*priv), gf_br_stub_mt_private_t);
+    if (!priv)
+        goto error_return;
+
+    priv->local_pool = mem_pool_new(br_stub_local_t, 512);
+    if (!priv->local_pool)
+        goto free_priv;
+
+    GF_OPTION_INIT("bitrot", priv->do_versioning, bool, free_mempool);
+
+    GF_OPTION_INIT("export", tmp, str, free_mempool);
+
+    if (snprintf(priv->export, PATH_MAX, "%s", tmp) >= PATH_MAX)
+        goto free_mempool;
+
+    if (snprintf(priv->stub_basepath, sizeof(priv->stub_basepath), "%s/%s",
+                 priv->export,
+                 BR_STUB_QUARANTINE_DIR) >= sizeof(priv->stub_basepath))
+        goto free_mempool;
+
+    (void)gettimeofday(&tv, NULL);
+
+    /* boot time is in network endian format */
+    priv->boot[0] = htonl(tv.tv_sec);
+    priv->boot[1] = htonl(tv.tv_usec);
+
+    pthread_mutex_init(&priv->lock, NULL);
+    pthread_cond_init(&priv->cond, NULL);
+    INIT_LIST_HEAD(&priv->squeue);
+
+    /* Thread creations need 'this' to be passed so that THIS can be
+     * assigned inside the thread. So setting this->private here.
+     */
+    this->private = priv;
+    if (!priv->do_versioning)
+        return 0;
+
+    ret = gf_thread_create(&priv->signth, NULL, br_stub_signth, this,
+                           "brssign");
+    if (ret != 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, BRS_MSG_SPAWN_SIGN_THRD_FAILED,
+                NULL);
+        goto cleanup_lock;
+    }
+
+    ret = br_stub_bad_object_container_init(this, priv);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_BAD_CONTAINER_FAIL, NULL);
+        goto cleanup_lock;
+    }
+
+    gf_msg_debug(this->name, 0, "bit-rot stub loaded");
+
+    return 0;
+
+cleanup_lock:
+    pthread_cond_destroy(&priv->cond);
+    pthread_mutex_destroy(&priv->lock);
+free_mempool:
+    mem_pool_destroy(priv->local_pool);
+    priv->local_pool = NULL;
+free_priv:
+    GF_FREE(priv);
+    this->private = NULL;
+error_return:
+    return -1;
+}
+
+/* TODO:
+ * As of now enabling bitrot option does 2 things.
+ * 1) Start the Bitrot Daemon which signs the objects (currently files only)
+ *    upon getting notified by the stub.
+ * 2) Enable versioning of the objects. Object versions (again files only) are
+ *    incremented upon modification.
+ * So object versioning is tied to bitrot daemon's signing. In future, object
+ * versioning might be necessary for other things as well apart from bit-rot
+ * detection (well that's the objective of bringing in object-versioning :)).
+ * In that case, better to make versioning a new option and letting it to be
+ * enabled despite bit-rot detection is not needed.
+ * Ex: ICAP.
+ */
+int32_t
+reconfigure(xlator_t *this, dict_t *options)
+{
+    int32_t ret = -1;
+    br_stub_private_t *priv = NULL;
+
+    priv = this->private;
+
+    GF_OPTION_RECONF("bitrot", priv->do_versioning, options, bool, err);
+    if (priv->do_versioning && !priv->signth) {
+        ret = gf_thread_create(&priv->signth, NULL, br_stub_signth, this,
+                               "brssign");
+        if (ret != 0) {
+            gf_smsg(this->name, GF_LOG_WARNING, 0,
+                    BRS_MSG_SPAWN_SIGN_THRD_FAILED, NULL);
+            goto err;
+        }
+
+        ret = br_stub_bad_object_container_init(this, priv);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_BAD_CONTAINER_FAIL,
+                    NULL);
+            goto err;
+        }
+    } else {
+        if (priv->signth) {
+            if (gf_thread_cleanup_xint(priv->signth)) {
+                gf_smsg(this->name, GF_LOG_ERROR, 0,
+                        BRS_MSG_CANCEL_SIGN_THREAD_FAILED, NULL);
+            } else {
+                gf_smsg(this->name, GF_LOG_INFO, 0, BRS_MSG_KILL_SIGN_THREAD,
+                        NULL);
+                priv->signth = 0;
+            }
+        }
+
+        if (priv->container.thread) {
+            if (gf_thread_cleanup_xint(priv->container.thread)) {
+                gf_smsg(this->name, GF_LOG_ERROR, 0,
+                        BRS_MSG_CANCEL_SIGN_THREAD_FAILED, NULL);
+            }
+            priv->container.thread = 0;
+        }
+    }
+
+    ret = 0;
+    return ret;
+err:
+    if (priv->signth) {
+        if (gf_thread_cleanup_xint(priv->signth)) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0,
+                    BRS_MSG_CANCEL_SIGN_THREAD_FAILED, NULL);
+        }
+        priv->signth = 0;
+    }
+
+    if (priv->container.thread) {
+        if (gf_thread_cleanup_xint(priv->container.thread)) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0,
+                    BRS_MSG_CANCEL_SIGN_THREAD_FAILED, NULL);
+        }
+        priv->container.thread = 0;
+    }
+    ret = -1;
+    return ret;
+}
+
+int
+notify(xlator_t *this, int event, void *data, ...)
+{
+    br_stub_private_t *priv = NULL;
+
+    if (!this)
+        return 0;
+
+    priv = this->private;
+    if (!priv)
+        return 0;
+
+    default_notify(this, event, data);
+    return 0;
+}
+
+void
+fini(xlator_t *this)
+{
+    int32_t ret = 0;
+    br_stub_private_t *priv = this->private;
+    struct br_stub_signentry *sigstub = NULL;
+    call_stub_t *stub = NULL;
+
+    if (!priv)
+        return;
+
+    if (!priv->do_versioning)
+        goto cleanup;
+
+    ret = gf_thread_cleanup_xint(priv->signth);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_CANCEL_SIGN_THREAD_FAILED,
+                NULL);
+        goto out;
+    }
+    priv->signth = 0;
+
+    while (!list_empty(&priv->squeue)) {
+        sigstub = list_first_entry(&priv->squeue, struct br_stub_signentry,
+                                   list);
+        list_del_init(&sigstub->list);
+
+        call_stub_destroy(sigstub->stub);
+        GF_FREE(sigstub);
+    }
+
+    ret = gf_thread_cleanup_xint(priv->container.thread);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_CANCEL_SIGN_THREAD_FAILED,
+                NULL);
+        goto out;
+    }
+
+    priv->container.thread = 0;
+
+    while (!list_empty(&priv->container.bad_queue)) {
+        stub = list_first_entry(&priv->container.bad_queue, call_stub_t, list);
+        list_del_init(&stub->list);
+        call_stub_destroy(stub);
+    }
+
+    pthread_mutex_destroy(&priv->container.bad_lock);
+    pthread_cond_destroy(&priv->container.bad_cond);
+
+cleanup:
+    pthread_mutex_destroy(&priv->lock);
+    pthread_cond_destroy(&priv->cond);
+
+    if (priv->local_pool) {
+        mem_pool_destroy(priv->local_pool);
+        priv->local_pool = NULL;
+    }
+
+    this->private = NULL;
+    GF_FREE(priv);
+
+out:
+    return;
+}
+
+static int
+br_stub_alloc_versions(br_version_t **obuf, br_signature_t **sbuf,
+                       size_t signaturelen)
+{
+    void *mem = NULL;
+    size_t size = 0;
+
+    if (obuf)
+        size += sizeof(br_version_t);
+    if (sbuf)
+        size += sizeof(br_signature_t) + signaturelen;
+
+    mem = GF_CALLOC(1, size, gf_br_stub_mt_version_t);
+    if (!mem)
+        goto error_return;
+
+    if (obuf) {
+        *obuf = (br_version_t *)mem;
+        mem = ((char *)mem + sizeof(br_version_t));
+    }
+    if (sbuf) {
+        *sbuf = (br_signature_t *)mem;
+    }
+
+    return 0;
+
+error_return:
+    return -1;
+}
+
+static void
+br_stub_dealloc_versions(void *mem)
+{
+    GF_FREE(mem);
+}
+
+static br_stub_local_t *
+br_stub_alloc_local(xlator_t *this)
+{
+    br_stub_private_t *priv = this->private;
+
+    return mem_get0(priv->local_pool);
+}
+
+static void
+br_stub_dealloc_local(br_stub_local_t *ptr)
+{
+    if (!ptr)
+        return;
+
+    mem_put(ptr);
+}
+
+static int
+br_stub_prepare_version_request(xlator_t *this, dict_t *dict,
+                                br_version_t *obuf, unsigned long oversion)
+{
+    br_stub_private_t *priv = NULL;
+
+    priv = this->private;
+    br_set_ongoingversion(obuf, oversion, priv->boot);
+
+    return dict_set_bin(dict, BITROT_CURRENT_VERSION_KEY, (void *)obuf,
+                        sizeof(br_version_t));
+}
+
+static int
+br_stub_prepare_signing_request(dict_t *dict, br_signature_t *sbuf,
+                                br_isignature_t *sign, size_t signaturelen)
+{
+    size_t size = 0;
+
+    br_set_signature(sbuf, sign, signaturelen, &size);
+
+    return dict_set_bin(dict, BITROT_SIGNING_VERSION_KEY, (void *)sbuf, size);
+}
+
+/**
+ * initialize an inode context starting with a given ongoing version.
+ * a fresh lookup() or a first creat() call initializes the inode
+ * context, hence the inode is marked dirty. this routine also
+ * initializes the transient inode version.
+ */
+static int
+br_stub_init_inode_versions(xlator_t *this, fd_t *fd, inode_t *inode,
+                            unsigned long version, gf_boolean_t markdirty,
+                            gf_boolean_t bad_object, uint64_t *ctx_addr)
+{
+    int32_t ret = 0;
+    br_stub_inode_ctx_t *ctx = NULL;
+
+    ctx = GF_CALLOC(1, sizeof(br_stub_inode_ctx_t), gf_br_stub_mt_inode_ctx_t);
+    if (!ctx)
+        goto error_return;
+
+    INIT_LIST_HEAD(&ctx->fd_list);
+    (markdirty) ? __br_stub_mark_inode_dirty(ctx)
+                : __br_stub_mark_inode_synced(ctx);
+    __br_stub_set_ongoing_version(ctx, version);
+
+    if (bad_object)
+        __br_stub_mark_object_bad(ctx);
+
+    if (fd) {
+        ret = br_stub_add_fd_to_inode(this, fd, ctx);
+        if (ret)
+            goto free_ctx;
+    }
+
+    ret = br_stub_set_inode_ctx(this, inode, ctx);
+    if (ret)
+        goto free_ctx;
+
+    if (ctx_addr)
+        *ctx_addr = (uint64_t)(uintptr_t)ctx;
+    return 0;
+
+free_ctx:
+    GF_FREE(ctx);
+error_return:
+    return -1;
+}
+
+/**
+ * modify the ongoing version of an inode.
+ */
+static int
+br_stub_mod_inode_versions(xlator_t *this, fd_t *fd, inode_t *inode,
+                           unsigned long version)
+{
+    int32_t ret = -1;
+    br_stub_inode_ctx_t *ctx = 0;
+
+    LOCK(&inode->lock);
+    {
+        ctx = __br_stub_get_ongoing_version_ctx(this, inode, NULL);
+        if (ctx == NULL)
+            goto unblock;
+        if (__br_stub_is_inode_dirty(ctx)) {
+            __br_stub_set_ongoing_version(ctx, version);
+            __br_stub_mark_inode_synced(ctx);
+        }
+
+        ret = 0;
+    }
+unblock:
+    UNLOCK(&inode->lock);
+
+    return ret;
+}
+
+static void
+br_stub_fill_local(br_stub_local_t *local, call_stub_t *stub, fd_t *fd,
+                   inode_t *inode, uuid_t gfid, int versioningtype,
+                   unsigned long memversion)
+{
+    local->fopstub = stub;
+    local->versioningtype = versioningtype;
+    local->u.context.version = memversion;
+    if (fd)
+        local->u.context.fd = fd_ref(fd);
+    if (inode)
+        local->u.context.inode = inode_ref(inode);
+    gf_uuid_copy(local->u.context.gfid, gfid);
+}
+
+static void
+br_stub_cleanup_local(br_stub_local_t *local)
+{
+    if (!local)
+        return;
+
+    local->fopstub = NULL;
+    local->versioningtype = 0;
+    local->u.context.version = 0;
+    if (local->u.context.fd) {
+        fd_unref(local->u.context.fd);
+        local->u.context.fd = NULL;
+    }
+    if (local->u.context.inode) {
+        inode_unref(local->u.context.inode);
+        local->u.context.inode = NULL;
+    }
+    memset(local->u.context.gfid, '\0', sizeof(uuid_t));
+}
+
+static int
+br_stub_need_versioning(xlator_t *this, fd_t *fd, gf_boolean_t *versioning,
+                        gf_boolean_t *modified, br_stub_inode_ctx_t **ctx)
+{
+    int32_t ret = -1;
+    uint64_t ctx_addr = 0;
+    br_stub_inode_ctx_t *c = NULL;
+    unsigned long version = BITROT_DEFAULT_CURRENT_VERSION;
+
+    *versioning = _gf_false;
+    *modified = _gf_false;
+
+    /* Bitrot stub inode context was initialized only in lookup, create
+     * and mknod cbk path. Object versioning was enabled by default
+     * irrespective of bitrot enabled or not. But it's made optional now.
+     * As a consequence there could be cases where getting inode ctx would
+     * fail because it's not set yet.
+     * e.g., If versioning (with bitrot enable) is enabled while I/O is
+     * happening, it could directly get other fops like writev without
+     * lookup, where getting inode ctx would fail. Hence initialize the
+     * inode ctx on failure to get ctx. This is done in all places where
+     * applicable.
+     */
+    ret = br_stub_get_inode_ctx(this, fd->inode, &ctx_addr);
+    if (ret < 0) {
+        ret = br_stub_init_inode_versions(this, fd, fd->inode, version,
+                                          _gf_true, _gf_false, &ctx_addr);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0,
+                    BRS_MSG_GET_INODE_CONTEXT_FAILED, "gfid=%s",
+                    uuid_utoa(fd->inode->gfid), NULL);
+            goto error_return;
+        }
+    }
+
+    c = (br_stub_inode_ctx_t *)(long)ctx_addr;
+
+    LOCK(&fd->inode->lock);
+    {
+        if (__br_stub_is_inode_dirty(c))
+            *versioning = _gf_true;
+        if (__br_stub_is_inode_modified(c))
+            *modified = _gf_true;
+    }
+    UNLOCK(&fd->inode->lock);
+
+    if (ctx)
+        *ctx = c;
+    return 0;
+
+error_return:
+    return -1;
+}
+
+static int32_t
+br_stub_anon_fd_ctx(xlator_t *this, fd_t *fd, br_stub_inode_ctx_t *ctx)
+{
+    int32_t ret = -1;
+    br_stub_fd_t *br_stub_fd = NULL;
+
+    br_stub_fd = br_stub_fd_ctx_get(this, fd);
+    if (!br_stub_fd) {
+        ret = br_stub_add_fd_to_inode(this, fd, ctx);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_ADD_FD_TO_INODE,
+                    "gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+            goto out;
+        }
+    }
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+static int
+br_stub_versioning_prep(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                        br_stub_inode_ctx_t *ctx)
+{
+    int32_t ret = -1;
+    br_stub_local_t *local = NULL;
+
+    local = br_stub_alloc_local(this);
+    if (!local) {
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, BRS_MSG_NO_MEMORY, "gfid=%s",
+                uuid_utoa(fd->inode->gfid), NULL);
+        goto error_return;
+    }
+
+    if (fd_is_anonymous(fd)) {
+        ret = br_stub_anon_fd_ctx(this, fd, ctx);
+        if (ret)
+            goto free_local;
+    }
+
+    frame->local = local;
+
+    return 0;
+
+free_local:
+    br_stub_dealloc_local(local);
+error_return:
+    return -1;
+}
+
+static int
+br_stub_mark_inode_modified(xlator_t *this, br_stub_local_t *local)
+{
+    fd_t *fd = NULL;
+    int32_t ret = 0;
+    uint64_t ctx_addr = 0;
+    br_stub_inode_ctx_t *ctx = NULL;
+    unsigned long version = BITROT_DEFAULT_CURRENT_VERSION;
+
+    fd = local->u.context.fd;
+
+    ret = br_stub_get_inode_ctx(this, fd->inode, &ctx_addr);
+    if (ret < 0) {
+        ret = br_stub_init_inode_versions(this, fd, fd->inode, version,
+                                          _gf_true, _gf_false, &ctx_addr);
+        if (ret)
+            goto error_return;
+    }
+
+    ctx = (br_stub_inode_ctx_t *)(long)ctx_addr;
+
+    LOCK(&fd->inode->lock);
+    {
+        __br_stub_set_inode_modified(ctx);
+    }
+    UNLOCK(&fd->inode->lock);
+
+    return 0;
+
+error_return:
+    return -1;
+}
+
+/**
+ * The possible return values from br_stub_is_bad_object () are:
+ * 1) 0  => as per the inode context object is not bad
+ * 2) -1 => Failed to get the inode context itself
+ * 3) -2 => As per the inode context object is bad
+ * Both -ve values means the fop which called this function is failed
+ * and error is returned upwards.
+ */
+static int
+br_stub_check_bad_object(xlator_t *this, inode_t *inode, int32_t *op_ret,
+                         int32_t *op_errno)
+{
+    int ret = -1;
+    unsigned long version = BITROT_DEFAULT_CURRENT_VERSION;
+
+    ret = br_stub_is_bad_object(this, inode);
+    if (ret == -2) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_BAD_OBJECT_ACCESS,
+                "gfid=%s", uuid_utoa(inode->gfid), NULL);
+        *op_ret = -1;
+        *op_errno = EIO;
+    }
+
+    if (ret == -1) {
+        ret = br_stub_init_inode_versions(this, NULL, inode, version, _gf_true,
+                                          _gf_false, NULL);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0,
+                    BRS_MSG_GET_INODE_CONTEXT_FAILED, "gfid=%s",
+                    uuid_utoa(inode->gfid), NULL);
+            *op_ret = -1;
+            *op_errno = EINVAL;
+        }
+    }
+
+    return ret;
+}
+
+/**
+ * callback for inode/fd versioning
+ */
+int
+br_stub_fd_incversioning_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                             int op_ret, int op_errno, dict_t *xdata)
+{
+    fd_t *fd = NULL;
+    inode_t *inode = NULL;
+    unsigned long version = 0;
+    br_stub_local_t *local = NULL;
+
+    local = (br_stub_local_t *)frame->local;
+    if (op_ret < 0)
+        goto done;
+    fd = local->u.context.fd;
+    inode = local->u.context.inode;
+    version = local->u.context.version;
+
+    op_ret = br_stub_mod_inode_versions(this, fd, inode, version);
+    if (op_ret < 0)
+        op_errno = EINVAL;
+
+done:
+    if (op_ret < 0) {
+        frame->local = NULL;
+        call_unwind_error(local->fopstub, -1, op_errno);
+        br_stub_cleanup_local(local);
+        br_stub_dealloc_local(local);
+    } else {
+        call_resume(local->fopstub);
+    }
+    return 0;
+}
+
+/**
+ * Initial object versioning
+ *
+ * Version persists two (2) extended attributes as explained below:
+ *   1. Current (ongoing) version: This is incremented on an writev ()
+ *      or truncate () and is the running version for an object.
+ *   2. Signing version: This is the version against which an object
+ *      was signed (checksummed).
+ *
+ * During initial versioning, both ongoing and signing versions are
+ * set of one and zero respectively. A write() call increments the
+ * ongoing version as an indication of modification to the object.
+ * Additionally this needs to be persisted on disk and needs to be
+ * durable: fsync().. :-/
+ * As an optimization only the first write() synchronizes the ongoing
+ * version to disk, subsequent write()s before the *last* release()
+ * are no-op's.
+ *
+ * create(), just like lookup() initializes the object versions to
+ * the default. As an optimization this is not a durable operation:
+ * in case of a crash, hard reboot etc.. absence of versioning xattrs
+ * is ignored in scrubber along with the one time crawler explicitly
+ * triggering signing for such objects.
+ *
+ * c.f. br_stub_writev() / br_stub_truncate()
+ */
+
+/**
+ * perform full or incremental versioning on an inode pointd by an
+ * fd. incremental versioning is done when an inode is dirty and a
+ * writeback is triggered.
+ */
+
+int
+br_stub_fd_versioning(xlator_t *this, call_frame_t *frame, call_stub_t *stub,
+                      dict_t *dict, fd_t *fd, br_stub_version_cbk *callback,
+                      unsigned long memversion, int versioningtype, int durable)
+{
+    int32_t ret = -1;
+    int flags = 0;
+    dict_t *xdata = NULL;
+    br_stub_local_t *local = NULL;
+
+    xdata = dict_new();
+    if (!xdata)
+        goto done;
+
+    ret = dict_set_int32(xdata, GLUSTERFS_INTERNAL_FOP_KEY, 1);
+    if (ret)
+        goto dealloc_xdata;
+
+    if (durable) {
+        ret = dict_set_int32(xdata, GLUSTERFS_DURABLE_OP, 0);
+        if (ret)
+            goto dealloc_xdata;
+    }
+
+    local = frame->local;
+
+    br_stub_fill_local(local, stub, fd, fd->inode, fd->inode->gfid,
+                       versioningtype, memversion);
+
+    STACK_WIND(frame, callback, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
+
+    ret = 0;
+
+dealloc_xdata:
+    dict_unref(xdata);
+done:
+    return ret;
+}
+
+static int
+br_stub_perform_incversioning(xlator_t *this, call_frame_t *frame,
+                              call_stub_t *stub, fd_t *fd,
+                              br_stub_inode_ctx_t *ctx)
+{
+    int32_t ret = -1;
+    dict_t *dict = NULL;
+    br_version_t *obuf = NULL;
+    unsigned long writeback_version = 0;
+    int op_errno = 0;
+    br_stub_local_t *local = NULL;
+
+    op_errno = EINVAL;
+    local = frame->local;
+
+    writeback_version = __br_stub_writeback_version(ctx);
+
+    op_errno = ENOMEM;
+    dict = dict_new();
+    if (!dict)
+        goto out;
+    ret = br_stub_alloc_versions(&obuf, NULL, 0);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_ALLOC_MEM_FAILED,
+                "gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        goto out;
+    }
+    ret = br_stub_prepare_version_request(this, dict, obuf, writeback_version);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_VERSION_PREPARE_FAIL,
+                "gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        br_stub_dealloc_versions(obuf);
+        goto out;
+    }
+
+    ret = br_stub_fd_versioning(
+        this, frame, stub, dict, fd, br_stub_fd_incversioning_cbk,
+        writeback_version, BR_STUB_INCREMENTAL_VERSIONING, !WRITEBACK_DURABLE);
+out:
+    if (dict)
+        dict_unref(dict);
+    if (ret) {
+        if (local)
+            frame->local = NULL;
+        call_unwind_error(stub, -1, op_errno);
+        if (local) {
+            br_stub_cleanup_local(local);
+            br_stub_dealloc_local(local);
+        }
+    }
+
+    return ret;
+}
+
+/** {{{ */
+
+/* fsetxattr() */
+
+int32_t
+br_stub_perform_objsign(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                        dict_t *dict, int flags, dict_t *xdata)
+{
+    STACK_WIND(frame, default_fsetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
+
+    dict_unref(xdata);
+    return 0;
+}
+
+void *
+br_stub_signth(void *arg)
+{
+    xlator_t *this = arg;
+    br_stub_private_t *priv = this->private;
+    struct br_stub_signentry *sigstub = NULL;
+
+    THIS = this;
+    while (1) {
+        /*
+         * Disabling bit-rot feature leads to this particular thread
+         * getting cleaned up by reconfigure via a call to the function
+         * gf_thread_cleanup_xint (which in turn calls pthread_cancel
+         * and pthread_join). But, if this thread had held the mutex
+         * &priv->lock at the time of cancellation, then it leads to
+         * deadlock in future when bit-rot feature is enabled (which
+         * again spawns this thread which cant hold the lock as the
+         * mutex is still held by the previous instance of the thread
+         * which got killed). Also, the br_stub_handle_object_signature
+         * function which is called whenever file has to be signed
+         * also gets blocked as it too attempts to acquire &priv->lock.
+         *
+         * So, arrange for the lock to be unlocked as part of the
+         * cleanup of this thread using pthread_cleanup_push and
+         * pthread_cleanup_pop.
+         */
+        pthread_cleanup_push(br_stub_lock_cleaner, &priv->lock);
+        pthread_mutex_lock(&priv->lock);
+        {
+            while (list_empty(&priv->squeue))
+                pthread_cond_wait(&priv->cond, &priv->lock);
+
+            sigstub = list_first_entry(&priv->squeue, struct br_stub_signentry,
+                                       list);
+            list_del_init(&sigstub->list);
+        }
+        pthread_mutex_unlock(&priv->lock);
+        pthread_cleanup_pop(0);
+
+        call_resume(sigstub->stub);
+
+        GF_FREE(sigstub);
+    }
+
+    return NULL;
+}
+
+static gf_boolean_t
+br_stub_internal_xattr(dict_t *dict)
+{
+    if (dict_get(dict, GLUSTERFS_SET_OBJECT_SIGNATURE) ||
+        dict_get(dict, GLUSTERFS_GET_OBJECT_SIGNATURE) ||
+        dict_get(dict, BR_REOPEN_SIGN_HINT_KEY) ||
+        dict_get(dict, BITROT_OBJECT_BAD_KEY) ||
+        dict_get(dict, BITROT_SIGNING_VERSION_KEY) ||
+        dict_get(dict, BITROT_CURRENT_VERSION_KEY))
+        return _gf_true;
+
+    return _gf_false;
+}
+
+int
+orderq(struct list_head *elem1, struct list_head *elem2)
+{
+    struct br_stub_signentry *s1 = NULL;
+    struct br_stub_signentry *s2 = NULL;
+
+    s1 = list_entry(elem1, struct br_stub_signentry, list);
+    s2 = list_entry(elem2, struct br_stub_signentry, list);
+
+    return (s1->v > s2->v);
+}
+
+static int
+br_stub_compare_sign_version(xlator_t *this, inode_t *inode,
+                             br_signature_t *sbuf, dict_t *dict,
+                             int *fakesuccess)
+{
+    int32_t ret = -1;
+    uint64_t tmp_ctx = 0;
+    gf_boolean_t invalid = _gf_false;
+    br_stub_inode_ctx_t *ctx = NULL;
+
+    GF_VALIDATE_OR_GOTO("bit-rot-stub", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+    GF_VALIDATE_OR_GOTO(this->name, sbuf, out);
+    GF_VALIDATE_OR_GOTO(this->name, dict, out);
+
+    ret = br_stub_get_inode_ctx(this, inode, &tmp_ctx);
+    if (ret) {
+        dict_del(dict, BITROT_SIGNING_VERSION_KEY);
+        goto out;
+    }
+
+    ctx = (br_stub_inode_ctx_t *)(long)tmp_ctx;
+
+    LOCK(&inode->lock);
+    {
+        if (ctx->currentversion < sbuf->signedversion) {
+            invalid = _gf_true;
+        } else if (ctx->currentversion > sbuf->signedversion) {
+            gf_msg_debug(this->name, 0,
+                         "\"Signing version\" "
+                         "(%lu) lower than \"Current version \" "
+                         "(%lu)",
+                         ctx->currentversion, sbuf->signedversion);
+            *fakesuccess = 1;
+        }
+    }
+    UNLOCK(&inode->lock);
+
+    if (invalid) {
+        ret = -1;
+        gf_smsg(this->name, GF_LOG_WARNING, 0, BRS_MSG_SIGN_VERSION_ERROR,
+                "Signing-ver=%lu", sbuf->signedversion, "current-ver=%lu",
+                ctx->currentversion, NULL);
+    }
+
+out:
+    return ret;
+}
+
+static int
+br_stub_prepare_signature(xlator_t *this, dict_t *dict, inode_t *inode,
+                          br_isignature_t *sign, int *fakesuccess)
+{
+    int32_t ret = -1;
+    size_t signaturelen = 0;
+    br_signature_t *sbuf = NULL;
+
+    if (!br_is_signature_type_valid(sign->signaturetype))
+        goto out;
+
+    signaturelen = sign->signaturelen;
+    ret = br_stub_alloc_versions(NULL, &sbuf, signaturelen);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_ALLOC_MEM_FAILED,
+                "gfid=%s", uuid_utoa(inode->gfid), NULL);
+        ret = -1;
+        goto out;
+    }
+    ret = br_stub_prepare_signing_request(dict, sbuf, sign, signaturelen);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_SIGN_PREPARE_FAIL,
+                "gfid=%s", uuid_utoa(inode->gfid), NULL);
+        ret = -1;
+        br_stub_dealloc_versions(sbuf);
+        goto out;
+    }
+
+    /* At this point sbuf has been added to dict, so the memory will be freed
+     * when the data from the dict is destroyed
+     */
+    ret = br_stub_compare_sign_version(this, inode, sbuf, dict, fakesuccess);
+out:
+    return ret;
+}
+
+static void
+br_stub_handle_object_signature(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                                dict_t *dict, br_isignature_t *sign,
+                                dict_t *xdata)
+{
+    int32_t ret = -1;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    int fakesuccess = 0;
+    br_stub_private_t *priv = NULL;
+    struct br_stub_signentry *sigstub = NULL;
+
+    priv = this->private;
+
+    if (frame->root->pid != GF_CLIENT_PID_BITD) {
+        gf_smsg(this->name, GF_LOG_WARNING, op_errno, BRS_MSG_NON_BITD_PID,
+                "PID=%d", frame->root->pid, NULL);
+        goto dofop;
+    }
+
+    ret = br_stub_prepare_signature(this, dict, fd->inode, sign, &fakesuccess);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, BRS_MSG_SIGN_PREPARE_FAIL,
+                "gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        goto dofop;
+    }
+    if (fakesuccess) {
+        op_ret = op_errno = 0;
+        goto dofop;
+    }
+
+    dict_del(dict, GLUSTERFS_SET_OBJECT_SIGNATURE);
+
+    ret = -1;
+    if (!xdata) {
+        xdata = dict_new();
+        if (!xdata)
+            goto dofop;
+    } else {
+        dict_ref(xdata);
+    }
+
+    ret = dict_set_int32(xdata, GLUSTERFS_DURABLE_OP, 0);
+    if (ret)
+        goto unref_dict;
+
+    /* prepare dispatch stub to order object signing */
+    sigstub = GF_CALLOC(1, sizeof(*sigstub), gf_br_stub_mt_sigstub_t);
+    if (!sigstub)
+        goto unref_dict;
+
+    INIT_LIST_HEAD(&sigstub->list);
+    sigstub->v = ntohl(sign->signedversion);
+    sigstub->stub = fop_fsetxattr_stub(frame, br_stub_perform_objsign, fd, dict,
+                                       0, xdata);
+    if (!sigstub->stub)
+        goto cleanup_stub;
+
+    pthread_mutex_lock(&priv->lock);
+    {
+        list_add_order(&sigstub->list, &priv->squeue, orderq);
+        pthread_cond_signal(&priv->cond);
+    }
+    pthread_mutex_unlock(&priv->lock);
+
+    return;
+
+cleanup_stub:
+    GF_FREE(sigstub);
+unref_dict:
+    dict_unref(xdata);
+dofop:
+    STACK_UNWIND_STRICT(fsetxattr, frame, op_ret, op_errno, NULL);
+}
+
+int32_t
+br_stub_fsetxattr_resume(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    int32_t ret = -1;
+    br_stub_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+
+    ret = br_stub_mark_inode_modified(this, local);
+    if (ret) {
+        op_ret = -1;
+        op_errno = EINVAL;
+    }
+
+    STACK_UNWIND_STRICT(fsetxattr, frame, op_ret, op_errno, xdata);
+
+    br_stub_cleanup_local(local);
+    br_stub_dealloc_local(local);
+
+    return 0;
+}
+
+/**
+ * Handles object reopens. Object reopens can be of 3 types. 2 are from
+ * oneshot crawler and 1 from the regular signer.
+ * ONESHOT CRAWLER:
+ * For those objects which were created before bitrot was enabled. oneshow
+ * crawler crawls the namespace and signs all the objects. It has to do
+ * the versioning before making bit-rot-stub send a sign notification.
+ * So it sends fsetxattr with BR_OBJECT_REOPEN as the value. And bit-rot-stub
+ * upon getting BR_OBJECT_REOPEN value checks if the version has to be
+ * increased or not. By default the version will be increased. But if the
+ * object is modified before BR_OBJECT_REOPEN from oneshot crawler, then
+ * versioning need not be done. In that case simply a success is returned.
+ * SIGNER:
+ * Signer wait for 2 minutes upon getting the notification from bit-rot-stub
+ * and then it sends a dummy write (in reality a fsetxattr) call, to change
+ * the state of the inode from REOPEN_WAIT to SIGN_QUICK. The funny part here
+ * is though the inode's state is REOPEN_WAIT, the call sent by signer is
+ * BR_OBJECT_RESIGN. Once the state is changed to SIGN_QUICK, then yet another
+ * notification is sent upon release (RESIGN would have happened via fsetxattr,
+ * so a fd is needed) and the object is signed truly this time.
+ * There is a challenge in the above RESIGN method by signer. After sending
+ * the 1st notification, the inode could be forgotten before RESIGN request
+ * is received. In that case, the inode's context (the newly looked up inode)
+ * would not indicate the inode as being modified (it would be in the default
+ * state) and because of this, a SIGN_QUICK notification to truly sign the
+ * object would not be sent. So, this is how its handled.
+ * if (request == RESIGN) {
+ *    if (inode->sign_info == NORMAL) {
+ *        mark_inode_non_dirty;
+ *        mark_inode_modified;
+ *    }
+ *    GOBACK (means unwind without doing versioning)
+ * }
+ */
+static void
+br_stub_handle_object_reopen(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                             uint32_t val)
+{
+    int32_t ret = -1;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    call_stub_t *stub = NULL;
+    gf_boolean_t inc_version = _gf_false;
+    gf_boolean_t modified = _gf_false;
+    br_stub_inode_ctx_t *ctx = NULL;
+    br_stub_local_t *local = NULL;
+    gf_boolean_t goback = _gf_true;
+
+    ret = br_stub_need_versioning(this, fd, &inc_version, &modified, &ctx);
+    if (ret)
+        goto unwind;
+
+    LOCK(&fd->inode->lock);
+    {
+        if ((val == BR_OBJECT_REOPEN) && inc_version)
+            goback = _gf_false;
+        if (val == BR_OBJECT_RESIGN && ctx->info_sign == BR_SIGN_NORMAL) {
+            __br_stub_mark_inode_synced(ctx);
+            __br_stub_set_inode_modified(ctx);
+        }
+        (void)__br_stub_inode_sign_state(ctx, GF_FOP_FSETXATTR, fd);
+    }
+    UNLOCK(&fd->inode->lock);
+
+    if (goback) {
+        op_ret = op_errno = 0;
+        goto unwind;
+    }
+
+    ret = br_stub_versioning_prep(frame, this, fd, ctx);
+    if (ret)
+        goto unwind;
+    local = frame->local;
+
+    stub = fop_fsetxattr_cbk_stub(frame, br_stub_fsetxattr_resume, 0, 0, NULL);
+    if (!stub) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_STUB_ALLOC_FAILED,
+                "fsetxattr gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        goto cleanup_local;
+    }
+
+    (void)br_stub_perform_incversioning(this, frame, stub, fd, ctx);
+    return;
+
+cleanup_local:
+    br_stub_cleanup_local(local);
+    br_stub_dealloc_local(local);
+
+unwind:
+    frame->local = NULL;
+    STACK_UNWIND_STRICT(fsetxattr, frame, op_ret, op_errno, NULL);
+}
+
+/**
+ * This function only handles bad file identification. Instead of checking in
+ * fops like open, readv, writev whether the object is bad or not by doing
+ * getxattr calls, better to catch them when scrubber marks it as bad.
+ * So this callback is called only when the fsetxattr is sent by the scrubber
+ * to mark the object as bad.
+ */
+int
+br_stub_fsetxattr_bad_object_cbk(call_frame_t *frame, void *cookie,
+                                 xlator_t *this, int32_t op_ret,
+                                 int32_t op_errno, dict_t *xdata)
+{
+    br_stub_local_t *local = NULL;
+    int32_t ret = -1;
+
+    local = frame->local;
+    frame->local = NULL;
+
+    if (op_ret < 0)
+        goto unwind;
+
+    /*
+     * What to do if marking the object as bad fails? (i.e. in memory
+     * marking within the inode context. If we are here means fsetxattr
+     * fop has succeeded on disk and the bad object xattr has been set).
+     * We can return failure to scruber, but there is nothing the scrubber
+     * can do with it (it might assume that the on disk setxattr itself has
+     * failed). The main purpose of this operation is to help identify the
+     * bad object by checking the inode context itself (thus avoiding the
+     * necessity of doing a getxattr fop on the disk).
+     *
+     * So as of now, success itself is being returned even though inode
+     * context set operation fails.
+     * In future if there is any change in the policy which can handle this,
+     * then appropriate response should be sent (i.e. success or error).
+     */
+    ret = br_stub_mark_object_bad(this, local->u.context.inode);
+    if (ret)
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_BAD_OBJ_MARK_FAIL,
+                "gfid=%s", uuid_utoa(local->u.context.inode->gfid), NULL);
+
+    ret = br_stub_add(this, local->u.context.inode->gfid);
+
+unwind:
+    STACK_UNWIND_STRICT(fsetxattr, frame, op_ret, op_errno, xdata);
+    br_stub_cleanup_local(local);
+    br_stub_dealloc_local(local);
+    return 0;
+}
+
+static int32_t
+br_stub_handle_bad_object_key(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                              dict_t *dict, int flags, dict_t *xdata)
+{
+    br_stub_local_t *local = NULL;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+
+    if (frame->root->pid != GF_CLIENT_PID_SCRUB) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_NON_SCRUB_BAD_OBJ_MARK,
+                "gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        goto unwind;
+    }
+
+    local = br_stub_alloc_local(this);
+    if (!local) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_ALLOC_MEM_FAILED,
+                "fsetxattr gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+
+    br_stub_fill_local(local, NULL, fd, fd->inode, fd->inode->gfid,
+                       BR_STUB_NO_VERSIONING, 0);
+    frame->local = local;
+
+    STACK_WIND(frame, br_stub_fsetxattr_bad_object_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
+    return 0;
+unwind:
+    STACK_UNWIND_STRICT(fsetxattr, frame, op_ret, op_errno, NULL);
+    return 0;
+}
+
+/**
+ * As of now, versioning is done by the stub (though as a setxattr
+ * operation) as part of inode modification operations such as writev,
+ * truncate, ftruncate. And signing is done by BitD by a fsetxattr call.
+ * So any kind of setxattr coming on the versioning and the signing xattr is
+ * not allowed (i.e. BITROT_CURRENT_VERSION_KEY and BITROT_SIGNING_VERSION_KEY).
+ * In future if BitD/scrubber are allowed to change the versioning
+ * xattrs (though I cannot see a reason for it as of now), then the below
+ * function can be modified to block setxattr on version for only applications.
+ *
+ * NOTE: BitD sends sign request on GLUSTERFS_SET_OBJECT_SIGNATURE key.
+ *       BITROT_SIGNING_VERSION_KEY is the xattr used to save the signature.
+ *
+ */
+static int32_t
+br_stub_handle_internal_xattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                              char *key)
+{
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+
+    gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_SET_INTERNAL_XATTR,
+            "setxattr key=%s", key, "inode-gfid=%s", uuid_utoa(fd->inode->gfid),
+            NULL);
+
+    STACK_UNWIND_STRICT(fsetxattr, frame, op_ret, op_errno, NULL);
+    return 0;
+}
+
+static void
+br_stub_dump_xattr(xlator_t *this, dict_t *dict, int *op_errno)
+{
+    char *format = "(%s:%s)";
+    char *dump = NULL;
+
+    dump = GF_CALLOC(1, BR_STUB_DUMP_STR_SIZE, gf_br_stub_mt_misc);
+    if (!dump) {
+        *op_errno = ENOMEM;
+        goto out;
+    }
+    dict_dump_to_str(dict, dump, BR_STUB_DUMP_STR_SIZE, format);
+    gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_SET_INTERNAL_XATTR,
+            "fsetxattr dump=%s", dump, NULL);
+out:
+    if (dump) {
+        GF_FREE(dump);
+    }
+    return;
+}
+
+int
+br_stub_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+                  int flags, dict_t *xdata)
+{
+    int32_t ret = 0;
+    uint32_t val = 0;
+    br_isignature_t *sign = NULL;
+    br_stub_private_t *priv = NULL;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+
+    priv = this->private;
+
+    if ((frame->root->pid != GF_CLIENT_PID_BITD &&
+         frame->root->pid != GF_CLIENT_PID_SCRUB) &&
+        br_stub_internal_xattr(dict)) {
+        br_stub_dump_xattr(this, dict, &op_errno);
+        goto unwind;
+    }
+
+    if (!priv->do_versioning)
+        goto wind;
+
+    if (!IA_ISREG(fd->inode->ia_type))
+        goto wind;
+
+    /* object signature request */
+    ret = dict_get_bin(dict, GLUSTERFS_SET_OBJECT_SIGNATURE, (void **)&sign);
+    if (!ret) {
+        gf_msg_debug(this->name, 0, "got SIGNATURE request on %s",
+                     uuid_utoa(fd->inode->gfid));
+        br_stub_handle_object_signature(frame, this, fd, dict, sign, xdata);
+        goto done;
+    }
+
+    /* signing xattr */
+    if (dict_get(dict, BITROT_SIGNING_VERSION_KEY)) {
+        br_stub_handle_internal_xattr(frame, this, fd,
+                                      BITROT_SIGNING_VERSION_KEY);
+        goto done;
+    }
+
+    /* version xattr */
+    if (dict_get(dict, BITROT_CURRENT_VERSION_KEY)) {
+        br_stub_handle_internal_xattr(frame, this, fd,
+                                      BITROT_CURRENT_VERSION_KEY);
+        goto done;
+    }
+
+    if (dict_get(dict, GLUSTERFS_GET_OBJECT_SIGNATURE)) {
+        br_stub_handle_internal_xattr(frame, this, fd,
+                                      GLUSTERFS_GET_OBJECT_SIGNATURE);
+        goto done;
+    }
+
+    /* object reopen request */
+    ret = dict_get_uint32(dict, BR_REOPEN_SIGN_HINT_KEY, &val);
+    if (!ret) {
+        br_stub_handle_object_reopen(frame, this, fd, val);
+        goto done;
+    }
+
+    /* handle bad object */
+    if (dict_get(dict, BITROT_OBJECT_BAD_KEY)) {
+        br_stub_handle_bad_object_key(frame, this, fd, dict, flags, xdata);
+        goto done;
+    }
+
+wind:
+    STACK_WIND(frame, default_fsetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
+    return 0;
+
+unwind:
+    STACK_UNWIND_STRICT(fsetxattr, frame, op_ret, op_errno, NULL);
+
+done:
+    return 0;
+}
+
+/**
+ * Currently BitD and scrubber are doing fsetxattr to either sign the object
+ * or to mark it as bad. Hence setxattr on any of those keys is denied directly
+ * without checking from where the fop is coming.
+ * Later, if BitD or Scrubber does setxattr of those keys, then appropriate
+ * check has to be added below.
+ */
+int
+br_stub_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+                 int flags, dict_t *xdata)
+{
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+
+    if (br_stub_internal_xattr(dict)) {
+        br_stub_dump_xattr(this, dict, &op_errno);
+        goto unwind;
+    }
+
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->setxattr,
+                    loc, dict, flags, xdata);
+    return 0;
+unwind:
+    STACK_UNWIND_STRICT(setxattr, frame, op_ret, op_errno, NULL);
+    return 0;
+}
+
+/** }}} */
+
+/** {{{ */
+
+/* {f}removexattr() */
+
+int32_t
+br_stub_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                    const char *name, dict_t *xdata)
+{
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+
+    if (!strcmp(BITROT_OBJECT_BAD_KEY, name) ||
+        !strcmp(BITROT_SIGNING_VERSION_KEY, name) ||
+        !strcmp(BITROT_CURRENT_VERSION_KEY, name)) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, BRS_MSG_REMOVE_INTERNAL_XATTR,
+                "name=%s", name, "file-path=%s", loc->path, NULL);
+        goto unwind;
+    }
+
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->removexattr, loc, name, xdata);
+    return 0;
+unwind:
+    STACK_UNWIND_STRICT(removexattr, frame, op_ret, op_errno, NULL);
+    return 0;
+}
+
+int32_t
+br_stub_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                     const char *name, dict_t *xdata)
+{
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+
+    if (!strcmp(BITROT_OBJECT_BAD_KEY, name) ||
+        !strcmp(BITROT_SIGNING_VERSION_KEY, name) ||
+        !strcmp(BITROT_CURRENT_VERSION_KEY, name)) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, BRS_MSG_REMOVE_INTERNAL_XATTR,
+                "name=%s", name, "inode-gfid=%s", uuid_utoa(fd->inode->gfid),
+                NULL);
+        goto unwind;
+    }
+
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->fremovexattr, fd, name, xdata);
+    return 0;
+unwind:
+    STACK_UNWIND_STRICT(fremovexattr, frame, op_ret, op_errno, NULL);
+    return 0;
+}
+
+/** }}} */
+
+/** {{{ */
+
+/* {f}getxattr() */
+
+int
+br_stub_listxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
+{
+    if (op_ret < 0)
+        goto unwind;
+
+    br_stub_remove_vxattrs(xattr, _gf_true);
+
+unwind:
+    STACK_UNWIND_STRICT(getxattr, frame, op_ret, op_errno, xattr, xdata);
+    return 0;
+}
+
+/**
+ * ONE SHOT CRAWLER from BitD signs the objects that it encounters while
+ * crawling, if the object is identified as stale by the stub. Stub follows
+ * the below logic to mark an object as stale or not.
+ * If the ongoing version and the signed_version match, then the object is not
+ * stale. Just return. Otherwise if they does not match, then it means one
+ * of the below things.
+ * 1) If the inode does not need write back of the version and the sign state is
+ *    is NORMAL, then some active i/o is going on the object. So skip it.
+ *    A notification will be sent to trigger the sign once the release is
+ *    received on the object.
+ * 2) If inode does not need writeback of the version and the sign state is
+ *    either reopen wait or quick sign, then it means:
+ *    A) BitD restarted and it is not sure whether the object it encountered
+ *       while crawling is in its timer wheel or not. Since there is no way to
+ *       scan the timer wheel as of now, ONE SHOT CRAWLER just goes ahead and
+ *       signs the object. Since the inode does not need writeback, version will
+ *       not be incremented and directly the object will be signed.
+ * 3) If the inode needs writeback, then it means the inode was forgotten after
+ *    the versioning and it has to be signed now.
+ *
+ * This is the algorithm followed:
+ * if (ongoing_version == signed_version); then
+ *     object_is_not_stale;
+ *     return;
+ * else; then
+ *      if (!inode_needs_writeback && inode_sign_state != NORMAL); then
+ *            object_is_stale;
+ *      if (inode_needs_writeback); then
+ *            object_is_stale;
+ *
+ * For SCRUBBER, no need to check for the sign state and inode writeback.
+ * If the ondisk ongoingversion and the ondisk signed version does not match,
+ * then treat the object as stale.
+ */
+char
+br_stub_is_object_stale(xlator_t *this, call_frame_t *frame, inode_t *inode,
+                        br_version_t *obuf, br_signature_t *sbuf)
+{
+    uint64_t ctx_addr = 0;
+    br_stub_inode_ctx_t *ctx = NULL;
+    int32_t ret = -1;
+    char stale = 0;
+
+    if (obuf->ongoingversion == sbuf->signedversion)
+        goto out;
+
+    if (frame->root->pid == GF_CLIENT_PID_SCRUB) {
+        stale = 1;
+        goto out;
+    }
+
+    ret = br_stub_get_inode_ctx(this, inode, &ctx_addr);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_GET_INODE_CONTEXT_FAILED,
+                "gfid=%s", uuid_utoa(inode->gfid), NULL);
+        goto out;
+    }
+
+    ctx = (br_stub_inode_ctx_t *)(long)ctx_addr;
+
+    LOCK(&inode->lock);
+    {
+        if ((!__br_stub_is_inode_dirty(ctx) &&
+             ctx->info_sign != BR_SIGN_NORMAL) ||
+            __br_stub_is_inode_dirty(ctx))
+            stale = 1;
+    }
+    UNLOCK(&inode->lock);
+
+out:
+    return stale;
+}
+
+int
+br_stub_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
+{
+    int32_t ret = 0;
+    size_t totallen = 0;
+    size_t signaturelen = 0;
+    br_stub_private_t *priv = NULL;
+    br_version_t *obuf = NULL;
+    br_signature_t *sbuf = NULL;
+    br_isignature_out_t *sign = NULL;
+    br_vxattr_status_t status;
+    br_stub_local_t *local = NULL;
+    inode_t *inode = NULL;
+    gf_boolean_t bad_object = _gf_false;
+    gf_boolean_t ver_enabled = _gf_false;
+
+    BR_STUB_VER_ENABLED_IN_CALLPATH(frame, ver_enabled);
+    priv = this->private;
+
+    if (op_ret < 0)
+        goto unwind;
+    BR_STUB_VER_COND_GOTO(priv, (!ver_enabled), delkeys);
+
+    if (cookie != (void *)BR_STUB_REQUEST_COOKIE)
+        goto unwind;
+
+    local = frame->local;
+    frame->local = NULL;
+    if (!local) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto unwind;
+    }
+    inode = local->u.context.inode;
+
+    op_ret = -1;
+    status = br_version_xattr_state(xattr, &obuf, &sbuf, &bad_object);
+
+    op_errno = EIO;
+    if (bad_object)
+        goto delkeys;
+
+    op_errno = EINVAL;
+    if (status == BR_VXATTR_STATUS_INVALID)
+        goto delkeys;
+
+    op_errno = ENODATA;
+    if ((status == BR_VXATTR_STATUS_MISSING) ||
+        (status == BR_VXATTR_STATUS_UNSIGNED))
+        goto delkeys;
+
+    /**
+     * okay.. we have enough information to satisfy the request,
+     * namely: version and signing extended attribute. what's
+     * pending is the signature length -- that's figured out
+     * indirectly via the size of the _whole_ xattr and the
+     * on-disk signing xattr header size.
+     */
+    op_errno = EINVAL;
+    ret = dict_get_uint32(xattr, BITROT_SIGNING_XATTR_SIZE_KEY,
+                          (uint32_t *)&signaturelen);
+    if (ret)
+        goto delkeys;
+
+    signaturelen -= sizeof(br_signature_t);
+    totallen = sizeof(br_isignature_out_t) + signaturelen;
+
+    op_errno = ENOMEM;
+    sign = GF_CALLOC(1, totallen, gf_br_stub_mt_signature_t);
+    if (!sign)
+        goto delkeys;
+
+    sign->time[0] = obuf->timebuf[0];
+    sign->time[1] = obuf->timebuf[1];
+
+    /* Object's dirty state & current signed version */
+    sign->version = sbuf->signedversion;
+    sign->stale = br_stub_is_object_stale(this, frame, inode, obuf, sbuf);
+
+    /* Object's signature */
+    sign->signaturelen = signaturelen;
+    sign->signaturetype = sbuf->signaturetype;
+    (void)memcpy(sign->signature, sbuf->signature, signaturelen);
+
+    op_errno = EINVAL;
+    ret = dict_set_bin(xattr, GLUSTERFS_GET_OBJECT_SIGNATURE, (void *)sign,
+                       totallen);
+    if (ret < 0) {
+        GF_FREE(sign);
+        goto delkeys;
+    }
+    op_errno = 0;
+    op_ret = totallen;
+
+delkeys:
+    br_stub_remove_vxattrs(xattr, _gf_true);
+
+unwind:
+    STACK_UNWIND_STRICT(getxattr, frame, op_ret, op_errno, xattr, xdata);
+    br_stub_cleanup_local(local);
+    br_stub_dealloc_local(local);
+    return 0;
+}
+
+static void
+br_stub_send_stub_init_time(call_frame_t *frame, xlator_t *this)
+{
+    int op_ret = 0;
+    int op_errno = 0;
+    dict_t *xattr = NULL;
+    br_stub_init_t stub = {
+        {
+            0,
+        },
+    };
+    br_stub_private_t *priv = NULL;
+
+    priv = this->private;
+
+    xattr = dict_new();
+    if (!xattr) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+
+    stub.timebuf[0] = priv->boot[0];
+    stub.timebuf[1] = priv->boot[1];
+    memcpy(stub.export, priv->export, strlen(priv->export) + 1);
+
+    op_ret = dict_set_static_bin(xattr, GLUSTERFS_GET_BR_STUB_INIT_TIME,
+                                 (void *)&stub, sizeof(br_stub_init_t));
+    if (op_ret < 0) {
+        op_errno = EINVAL;
+        goto unwind;
+    }
+
+    op_ret = sizeof(br_stub_init_t);
+
+unwind:
+    STACK_UNWIND_STRICT(getxattr, frame, op_ret, op_errno, xattr, NULL);
+
+    if (xattr)
+        dict_unref(xattr);
+}
+
+int
+br_stub_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                 const char *name, dict_t *xdata)
+{
+    void *cookie = NULL;
+    static uuid_t rootgfid = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1};
+    fop_getxattr_cbk_t cbk = br_stub_getxattr_cbk;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    br_stub_local_t *local = NULL;
+    br_stub_private_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("bit-rot-stub", this, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, loc, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, unwind);
+
+    if (!name) {
+        cbk = br_stub_listxattr_cbk;
+        goto wind;
+    }
+
+    if (br_stub_is_internal_xattr(name))
+        goto unwind;
+
+    priv = this->private;
+    BR_STUB_VER_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+
+    /**
+     * If xattr is node-uuid and the inode is marked bad, return EIO.
+     * Returning EIO would result in AFR to choose correct node-uuid
+     * corresponding to the subvolume * where the good copy of the
+     * file resides.
+     */
+    if (IA_ISREG(loc->inode->ia_type) && XATTR_IS_NODE_UUID(name) &&
+        br_stub_check_bad_object(this, loc->inode, &op_ret, &op_errno)) {
+        goto unwind;
+    }
+
+    /**
+     * this special extended attribute is allowed only on root
+     */
+    if (name &&
+        (strncmp(name, GLUSTERFS_GET_BR_STUB_INIT_TIME,
+                 sizeof(GLUSTERFS_GET_BR_STUB_INIT_TIME) - 1) == 0) &&
+        ((gf_uuid_compare(loc->gfid, rootgfid) == 0) ||
+         (gf_uuid_compare(loc->inode->gfid, rootgfid) == 0))) {
+        BR_STUB_RESET_LOCAL_NULL(frame);
+        br_stub_send_stub_init_time(frame, this);
+        return 0;
+    }
+
+    if (!IA_ISREG(loc->inode->ia_type))
+        goto wind;
+
+    if (name && (strncmp(name, GLUSTERFS_GET_OBJECT_SIGNATURE,
+                         sizeof(GLUSTERFS_GET_OBJECT_SIGNATURE) - 1) == 0)) {
+        cookie = (void *)BR_STUB_REQUEST_COOKIE;
+
+        local = br_stub_alloc_local(this);
+        if (!local) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto unwind;
+        }
+
+        br_stub_fill_local(local, NULL, NULL, loc->inode, loc->inode->gfid,
+                           BR_STUB_NO_VERSIONING, 0);
+        frame->local = local;
+    }
+
+wind:
+    STACK_WIND_COOKIE(frame, cbk, cookie, FIRST_CHILD(this),
+                      FIRST_CHILD(this)->fops->getxattr, loc, name, xdata);
+    return 0;
+unwind:
+    BR_STUB_RESET_LOCAL_NULL(frame);
+    STACK_UNWIND_STRICT(getxattr, frame, op_ret, op_errno, NULL, NULL);
+    return 0;
+}
+
+int
+br_stub_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                  const char *name, dict_t *xdata)
+{
+    void *cookie = NULL;
+    static uuid_t rootgfid = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1};
+    fop_fgetxattr_cbk_t cbk = br_stub_getxattr_cbk;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    br_stub_local_t *local = NULL;
+    br_stub_private_t *priv = NULL;
+
+    priv = this->private;
+
+    if (!name) {
+        cbk = br_stub_listxattr_cbk;
+        goto wind;
+    }
+
+    if (br_stub_is_internal_xattr(name))
+        goto unwind;
+
+    BR_STUB_VER_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+
+    /**
+     * If xattr is node-uuid and the inode is marked bad, return EIO.
+     * Returning EIO would result in AFR to choose correct node-uuid
+     * corresponding to the subvolume * where the good copy of the
+     * file resides.
+     */
+    if (IA_ISREG(fd->inode->ia_type) && XATTR_IS_NODE_UUID(name) &&
+        br_stub_check_bad_object(this, fd->inode, &op_ret, &op_errno)) {
+        goto unwind;
+    }
+
+    /**
+     * this special extended attribute is allowed only on root
+     */
+    if (name &&
+        (strncmp(name, GLUSTERFS_GET_BR_STUB_INIT_TIME,
+                 sizeof(GLUSTERFS_GET_BR_STUB_INIT_TIME) - 1) == 0) &&
+        (gf_uuid_compare(fd->inode->gfid, rootgfid) == 0)) {
+        BR_STUB_RESET_LOCAL_NULL(frame);
+        br_stub_send_stub_init_time(frame, this);
+        return 0;
+    }
+
+    if (!IA_ISREG(fd->inode->ia_type))
+        goto wind;
+
+    if (name && (strncmp(name, GLUSTERFS_GET_OBJECT_SIGNATURE,
+                         sizeof(GLUSTERFS_GET_OBJECT_SIGNATURE) - 1) == 0)) {
+        cookie = (void *)BR_STUB_REQUEST_COOKIE;
+
+        local = br_stub_alloc_local(this);
+        if (!local) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto unwind;
+        }
+
+        br_stub_fill_local(local, NULL, fd, fd->inode, fd->inode->gfid,
+                           BR_STUB_NO_VERSIONING, 0);
+        frame->local = local;
+    }
+
+wind:
+    STACK_WIND_COOKIE(frame, cbk, cookie, FIRST_CHILD(this),
+                      FIRST_CHILD(this)->fops->fgetxattr, fd, name, xdata);
+    return 0;
+unwind:
+    BR_STUB_RESET_LOCAL_NULL(frame);
+    STACK_UNWIND_STRICT(fgetxattr, frame, op_ret, op_errno, NULL, NULL);
+    return 0;
+}
+
+int32_t
+br_stub_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+              off_t offset, uint32_t flags, dict_t *xdata)
+{
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    int32_t ret = -1;
+    br_stub_private_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("bit-rot-stub", this, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, frame, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, fd, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, fd->inode, unwind);
+
+    priv = this->private;
+    if (!priv->do_versioning)
+        goto wind;
+
+    ret = br_stub_check_bad_object(this, fd->inode, &op_ret, &op_errno);
+    if (ret)
+        goto unwind;
+
+wind:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->readv,
+                    fd, size, offset, flags, xdata);
+    return 0;
+
+unwind:
+    STACK_UNWIND_STRICT(readv, frame, op_ret, op_errno, NULL, 0, NULL, NULL,
+                        NULL);
+    return 0;
+}
+
+/**
+ * The first write response on the first fd in the list of fds will set
+ * the flag to indicate that the inode is modified. The subsequent write
+ * respnses coming on either the first fd or some other fd will not change
+ * the fd. The inode-modified flag is unset only upon release of all the
+ * fds.
+ */
+int32_t
+br_stub_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                   struct iatt *postbuf, dict_t *xdata)
+{
+    int32_t ret = 0;
+    br_stub_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+
+    if (op_ret < 0)
+        goto unwind;
+
+    ret = br_stub_mark_inode_modified(this, local);
+    if (ret) {
+        op_ret = -1;
+        op_errno = EINVAL;
+    }
+
+unwind:
+    STACK_UNWIND_STRICT(writev, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+
+    br_stub_cleanup_local(local);
+    br_stub_dealloc_local(local);
+
+    return 0;
+}
+
+int32_t
+br_stub_writev_resume(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                      struct iovec *vector, int32_t count, off_t offset,
+                      uint32_t flags, struct iobref *iobref, dict_t *xdata)
+{
+    STACK_WIND(frame, br_stub_writev_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->writev, fd, vector, count, offset,
+               flags, iobref, xdata);
+    return 0;
+}
+
+/**
+ * This is probably the most crucial part about the whole versioning thing.
+ * There's absolutely no differentiation as such between an anonymous fd
+ * and a regular fd except the fd context initialization. Object versioning
+ * is performed when the inode is dirty. Parallel write operations are no
+ * special with each write performing object versioning followed by marking
+ * the inode as non-dirty (synced). This is followed by the actual operation
+ * (writev() in this case) which on a success marks the inode as modified.
+ * This prevents signing of objects that have not been modified.
+ */
+int32_t
+br_stub_writev(call_frame_t *frame, xlator_t *this, fd_t *fd,
+               struct iovec *vector, int32_t count, off_t offset,
+               uint32_t flags, struct iobref *iobref, dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    gf_boolean_t inc_version = _gf_false;
+    gf_boolean_t modified = _gf_false;
+    br_stub_inode_ctx_t *ctx = NULL;
+    int32_t ret = -1;
+    fop_writev_cbk_t cbk = default_writev_cbk;
+    br_stub_local_t *local = NULL;
+    br_stub_private_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("bit-rot-stub", this, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, frame, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, fd, unwind);
+
+    priv = this->private;
+    if (!priv->do_versioning)
+        goto wind;
+
+    ret = br_stub_need_versioning(this, fd, &inc_version, &modified, &ctx);
+    if (ret)
+        goto unwind;
+
+    ret = br_stub_check_bad_object(this, fd->inode, &op_ret, &op_errno);
+    if (ret)
+        goto unwind;
+
+    /**
+     * The inode is not dirty and also witnessed at least one successful
+     * modification operation. Therefore, subsequent operations need not
+     * perform any special tracking.
+     */
+    if (!inc_version && modified)
+        goto wind;
+
+    /**
+     * okay.. so, either the inode needs versioning or the modification
+     * needs to be tracked. ->cbk is set to the appropriate callback
+     * routine for this.
+     * NOTE: ->local needs to be deallocated on failures from here on.
+     */
+    ret = br_stub_versioning_prep(frame, this, fd, ctx);
+    if (ret)
+        goto unwind;
+
+    local = frame->local;
+    if (!inc_version) {
+        br_stub_fill_local(local, NULL, fd, fd->inode, fd->inode->gfid,
+                           BR_STUB_NO_VERSIONING, 0);
+        cbk = br_stub_writev_cbk;
+        goto wind;
+    }
+
+    stub = fop_writev_stub(frame, br_stub_writev_resume, fd, vector, count,
+                           offset, flags, iobref, xdata);
+
+    if (!stub) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_STUB_ALLOC_FAILED,
+                "write  gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        goto cleanup_local;
+    }
+
+    /* Perform Versioning */
+    return br_stub_perform_incversioning(this, frame, stub, fd, ctx);
+
+wind:
+    STACK_WIND(frame, cbk, FIRST_CHILD(this), FIRST_CHILD(this)->fops->writev,
+               fd, vector, count, offset, flags, iobref, xdata);
+    return 0;
+
+cleanup_local:
+    br_stub_cleanup_local(local);
+    br_stub_dealloc_local(local);
+
+unwind:
+    frame->local = NULL;
+    STACK_UNWIND_STRICT(writev, frame, op_ret, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int32_t
+br_stub_ftruncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                      struct iatt *postbuf, dict_t *xdata)
+{
+    int32_t ret = -1;
+    br_stub_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+
+    if (op_ret < 0)
+        goto unwind;
+
+    ret = br_stub_mark_inode_modified(this, local);
+    if (ret) {
+        op_ret = -1;
+        op_errno = EINVAL;
+    }
+
+unwind:
+    STACK_UNWIND_STRICT(ftruncate, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+
+    br_stub_cleanup_local(local);
+    br_stub_dealloc_local(local);
+
+    return 0;
+}
+
+int32_t
+br_stub_ftruncate_resume(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                         off_t offset, dict_t *xdata)
+{
+    STACK_WIND(frame, br_stub_ftruncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+    return 0;
+}
+
+/* c.f. br_stub_writev() for explanation */
+int32_t
+br_stub_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                  dict_t *xdata)
+{
+    br_stub_local_t *local = NULL;
+    call_stub_t *stub = NULL;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    gf_boolean_t inc_version = _gf_false;
+    gf_boolean_t modified = _gf_false;
+    br_stub_inode_ctx_t *ctx = NULL;
+    int32_t ret = -1;
+    fop_ftruncate_cbk_t cbk = default_ftruncate_cbk;
+    br_stub_private_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("bit-rot-stub", this, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, frame, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, fd, unwind);
+
+    priv = this->private;
+    if (!priv->do_versioning)
+        goto wind;
+
+    ret = br_stub_need_versioning(this, fd, &inc_version, &modified, &ctx);
+    if (ret)
+        goto unwind;
+
+    ret = br_stub_check_bad_object(this, fd->inode, &op_ret, &op_errno);
+    if (ret)
+        goto unwind;
+
+    if (!inc_version && modified)
+        goto wind;
+
+    ret = br_stub_versioning_prep(frame, this, fd, ctx);
+    if (ret)
+        goto unwind;
+
+    local = frame->local;
+    if (!inc_version) {
+        br_stub_fill_local(local, NULL, fd, fd->inode, fd->inode->gfid,
+                           BR_STUB_NO_VERSIONING, 0);
+        cbk = br_stub_ftruncate_cbk;
+        goto wind;
+    }
+
+    stub = fop_ftruncate_stub(frame, br_stub_ftruncate_resume, fd, offset,
+                              xdata);
+    if (!stub) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_STUB_ALLOC_FAILED,
+                "ftruncate gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        goto cleanup_local;
+    }
+
+    return br_stub_perform_incversioning(this, frame, stub, fd, ctx);
+
+wind:
+    STACK_WIND(frame, cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+    return 0;
+
+cleanup_local:
+    br_stub_cleanup_local(local);
+    br_stub_dealloc_local(local);
+
+unwind:
+    frame->local = NULL;
+    STACK_UNWIND_STRICT(ftruncate, frame, op_ret, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int32_t
+br_stub_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+    int32_t ret = 0;
+    br_stub_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+
+    if (op_ret < 0)
+        goto unwind;
+
+    ret = br_stub_mark_inode_modified(this, local);
+    if (ret) {
+        op_ret = -1;
+        op_errno = EINVAL;
+    }
+
+unwind:
+    STACK_UNWIND_STRICT(truncate, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    br_stub_cleanup_local(local);
+    br_stub_dealloc_local(local);
+    return 0;
+}
+
+int32_t
+br_stub_truncate_resume(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                        off_t offset, dict_t *xdata)
+{
+    br_stub_local_t *local = frame->local;
+
+    fd_unref(local->u.context.fd);
+    STACK_WIND(frame, br_stub_ftruncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+    return 0;
+}
+
+/**
+ * Bit-rot-stub depends heavily on the fd based operations to for doing
+ * versioning and sending notification. It starts tracking the operation
+ * upon getting first fd based modify operation by doing versioning and
+ * sends notification when last fd using which the inode was modified is
+ * released.
+ * But for truncate there is no fd and hence it becomes difficult to do
+ * the versioning and send notification. It is handled by doing versioning
+ * on an anonymous fd. The fd will be valid till the completion of the
+ * truncate call. It guarantees that release on this anonymous fd will happen
+ * after the truncate call and notification is sent after the truncate call.
+ *
+ * c.f. br_writev_cbk() for explanation
+ */
+int32_t
+br_stub_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+                 dict_t *xdata)
+{
+    br_stub_local_t *local = NULL;
+    call_stub_t *stub = NULL;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    gf_boolean_t inc_version = _gf_false;
+    gf_boolean_t modified = _gf_false;
+    br_stub_inode_ctx_t *ctx = NULL;
+    int32_t ret = -1;
+    fd_t *fd = NULL;
+    fop_truncate_cbk_t cbk = default_truncate_cbk;
+    br_stub_private_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("bit-rot-stub", this, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, frame, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, loc, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, unwind);
+
+    priv = this->private;
+    if (!priv->do_versioning)
+        goto wind;
+
+    fd = fd_anonymous(loc->inode);
+    if (!fd) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_CREATE_ANONYMOUS_FD_FAILED,
+                "inode-gfid=%s", uuid_utoa(loc->inode->gfid), NULL);
+        goto unwind;
+    }
+
+    ret = br_stub_need_versioning(this, fd, &inc_version, &modified, &ctx);
+    if (ret)
+        goto cleanup_fd;
+
+    ret = br_stub_check_bad_object(this, fd->inode, &op_ret, &op_errno);
+    if (ret)
+        goto unwind;
+
+    if (!inc_version && modified)
+        goto wind;
+
+    ret = br_stub_versioning_prep(frame, this, fd, ctx);
+    if (ret)
+        goto cleanup_fd;
+
+    local = frame->local;
+    if (!inc_version) {
+        br_stub_fill_local(local, NULL, fd, fd->inode, fd->inode->gfid,
+                           BR_STUB_NO_VERSIONING, 0);
+        cbk = br_stub_truncate_cbk;
+        goto wind;
+    }
+
+    stub = fop_truncate_stub(frame, br_stub_truncate_resume, loc, offset,
+                             xdata);
+    if (!stub) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_STUB_ALLOC_FAILED,
+                "truncate gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        goto cleanup_local;
+    }
+
+    return br_stub_perform_incversioning(this, frame, stub, fd, ctx);
+
+wind:
+    STACK_WIND(frame, cbk, FIRST_CHILD(this), FIRST_CHILD(this)->fops->truncate,
+               loc, offset, xdata);
+    if (fd)
+        fd_unref(fd);
+    return 0;
+
+cleanup_local:
+    br_stub_cleanup_local(local);
+    br_stub_dealloc_local(local);
+cleanup_fd:
+    fd_unref(fd);
+unwind:
+    frame->local = NULL;
+    STACK_UNWIND_STRICT(truncate, frame, op_ret, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+/** }}} */
+
+/** {{{ */
+
+/* open() */
+
+/**
+ * It's probably worth mentioning a bit about why some of the housekeeping
+ * work is done in open() call path, rather than the callback path.
+ * Two (or more) open()'s in parallel can race and lead to a situation
+ * where a release() gets triggered (possibly after a series of write()
+ * calls) when *other* open()'s have still not reached callback path
+ * thereby having an active fd on an inode that is in process of getting
+ * signed with the current version.
+ *
+ * Maintaining fd list in the call path ensures that a release() would
+ * not be triggered if an open() call races ahead (followed by a close())
+ * threby finding non-empty fd list.
+ */
+
+int
+br_stub_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+             fd_t *fd, dict_t *xdata)
+{
+    int32_t ret = -1;
+    br_stub_inode_ctx_t *ctx = NULL;
+    uint64_t ctx_addr = 0;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    br_stub_private_t *priv = NULL;
+    unsigned long version = BITROT_DEFAULT_CURRENT_VERSION;
+
+    GF_VALIDATE_OR_GOTO("bit-rot-stub", this, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, loc, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, fd, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, fd->inode, unwind);
+
+    priv = this->private;
+
+    if (!priv->do_versioning)
+        goto wind;
+
+    ret = br_stub_get_inode_ctx(this, fd->inode, &ctx_addr);
+    if (ret) {
+        ret = br_stub_init_inode_versions(this, fd, fd->inode, version,
+                                          _gf_true, _gf_false, &ctx_addr);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0,
+                    BRS_MSG_GET_INODE_CONTEXT_FAILED, "path=%s", loc->path,
+                    "gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+            goto unwind;
+        }
+    }
+
+    ctx = (br_stub_inode_ctx_t *)(long)ctx_addr;
+
+    ret = br_stub_check_bad_object(this, fd->inode, &op_ret, &op_errno);
+    if (ret)
+        goto unwind;
+
+    if (frame->root->pid == GF_CLIENT_PID_SCRUB)
+        goto wind;
+
+    if (flags == O_RDONLY)
+        goto wind;
+
+    ret = br_stub_add_fd_to_inode(this, fd, ctx);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_ADD_FD_TO_LIST_FAILED,
+                "gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        goto unwind;
+    }
+
+wind:
+    STACK_WIND(frame, default_open_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->open, loc, flags, fd, xdata);
+    return 0;
+unwind:
+    STACK_UNWIND_STRICT(open, frame, op_ret, op_errno, NULL, NULL);
+    return 0;
+}
+
+/** }}} */
+
+/** {{{ */
+
+/* creat() */
+
+/**
+ * This routine registers a release callback for the given fd and adds the
+ * fd to the inode context fd tracking list.
+ */
+int32_t
+br_stub_add_fd_to_inode(xlator_t *this, fd_t *fd, br_stub_inode_ctx_t *ctx)
+{
+    int32_t ret = -1;
+    br_stub_fd_t *br_stub_fd = NULL;
+
+    ret = br_stub_require_release_call(this, fd, &br_stub_fd);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_SET_FD_CONTEXT_FAILED,
+                "gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        goto out;
+    }
+
+    LOCK(&fd->inode->lock);
+    {
+        list_add_tail(&ctx->fd_list, &br_stub_fd->list);
+    }
+    UNLOCK(&fd->inode->lock);
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+int
+br_stub_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int op_ret, int op_errno, fd_t *fd, inode_t *inode,
+                   struct iatt *stbuf, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
+{
+    int32_t ret = 0;
+    uint64_t ctx_addr = 0;
+    br_stub_inode_ctx_t *ctx = NULL;
+    unsigned long version = BITROT_DEFAULT_CURRENT_VERSION;
+    br_stub_private_t *priv = NULL;
+
+    priv = this->private;
+
+    if (op_ret < 0)
+        goto unwind;
+
+    if (!priv->do_versioning)
+        goto unwind;
+
+    ret = br_stub_get_inode_ctx(this, fd->inode, &ctx_addr);
+    if (ret < 0) {
+        ret = br_stub_init_inode_versions(this, fd, inode, version, _gf_true,
+                                          _gf_false, &ctx_addr);
+        if (ret) {
+            op_ret = -1;
+            op_errno = EINVAL;
+        }
+    } else {
+        ctx = (br_stub_inode_ctx_t *)(long)ctx_addr;
+        ret = br_stub_add_fd_to_inode(this, fd, ctx);
+    }
+
+unwind:
+    STACK_UNWIND_STRICT(create, frame, op_ret, op_errno, fd, inode, stbuf,
+                        preparent, postparent, xdata);
+    return 0;
+}
+
+int
+br_stub_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+               mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+    GF_VALIDATE_OR_GOTO("bit-rot-stub", this, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, loc, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, fd, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, fd->inode, unwind);
+
+    STACK_WIND(frame, br_stub_create_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->create, loc, flags, mode, umask, fd,
+               xdata);
+    return 0;
+unwind:
+    STACK_UNWIND_STRICT(create, frame, -1, EINVAL, NULL, NULL, NULL, NULL, NULL,
+                        NULL);
+    return 0;
+}
+
+int
+br_stub_mknod_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                  int op_errno, inode_t *inode, struct iatt *stbuf,
+                  struct iatt *preparent, struct iatt *postparent,
+                  dict_t *xdata)
+{
+    int32_t ret = -1;
+    unsigned long version = BITROT_DEFAULT_CURRENT_VERSION;
+    br_stub_private_t *priv = NULL;
+
+    priv = this->private;
+
+    if (op_ret < 0)
+        goto unwind;
+
+    if (!priv->do_versioning)
+        goto unwind;
+
+    ret = br_stub_init_inode_versions(this, NULL, inode, version, _gf_true,
+                                      _gf_false, NULL);
+    /**
+     * Like lookup, if init_inode_versions fail, return EINVAL
+     */
+    if (ret) {
+        op_ret = -1;
+        op_errno = EINVAL;
+    }
+
+unwind:
+    STACK_UNWIND_STRICT(mknod, frame, op_ret, op_errno, inode, stbuf, preparent,
+                        postparent, xdata);
+    return 0;
+}
+
+int
+br_stub_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+              dev_t dev, mode_t umask, dict_t *xdata)
+{
+    GF_VALIDATE_OR_GOTO("bit-rot-stub", this, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, loc, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, unwind);
+
+    STACK_WIND(frame, br_stub_mknod_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mknod, loc, mode, dev, umask, xdata);
+    return 0;
+unwind:
+    STACK_UNWIND_STRICT(mknod, frame, -1, EINVAL, NULL, NULL, NULL, NULL, NULL);
+    return 0;
+}
+
+/** }}} */
+
+/**
+ * As of now, only lookup searches for bad object xattr and marks the
+ * object as bad in its inode context if the xattr is present. But there
+ * is a possibility that, at the time of the lookup the object was not
+ * marked bad (i.e. bad object xattr was not set), and later its marked
+ * as bad. In this case, object is not bad, so when a fop such as open or
+ * readv or writev comes on the object, the fop will be sent downward instead
+ * of sending as error upwards.
+ * The solution for this is to do a getxattr for the below list of fops.
+ * lookup, readdirp, open, readv, writev.
+ * But doing getxattr for each of the above fops might be costly.
+ * So another method followed is to catch the bad file marking by the scrubber
+ * and set that info within the object's inode context. In this way getxattr
+ * calls can be avoided and bad objects can be caught instantly. Fetching the
+ * xattr is needed only in lookups when there is a brick restart or inode
+ * forget.
+ *
+ * If the dict (@xattr) is NULL, then how should that be handled? Fail the
+ * lookup operation? Or let it continue with version being initialized to
+ * BITROT_DEFAULT_CURRENT_VERSION. But what if the version was different
+ * on disk (and also a right signature was there), but posix failed to
+ * successfully allocate the dict? Posix does not treat call back xdata
+ * creattion failure as the lookup failure.
+ */
+static int32_t
+br_stub_lookup_version(xlator_t *this, uuid_t gfid, inode_t *inode,
+                       dict_t *xattr)
+{
+    unsigned long version = 0;
+    br_version_t *obuf = NULL;
+    br_signature_t *sbuf = NULL;
+    br_vxattr_status_t status;
+    gf_boolean_t bad_object = _gf_false;
+
+    /**
+     * versioning xattrs were requested from POSIX. if available, figure
+     * out the correct version to use in the inode context (start with
+     * the default version if unavailable). As of now versions are not
+     * persisted on-disk. The inode is marked dirty, so that the first
+     * operation (such as write(), etc..) triggers synchronization to
+     * disk.
+     */
+    status = br_version_xattr_state(xattr, &obuf, &sbuf, &bad_object);
+    version = ((status == BR_VXATTR_STATUS_FULL) ||
+               (status == BR_VXATTR_STATUS_UNSIGNED))
+                  ? obuf->ongoingversion
+                  : BITROT_DEFAULT_CURRENT_VERSION;
+
+    /**
+     * If signature is there, but version is not there then that status is
+     * is treated as INVALID. So in that case, we should not initialize the
+     * inode context with wrong version names etc.
+     */
+    if (status == BR_VXATTR_STATUS_INVALID)
+        return -1;
+
+    return br_stub_init_inode_versions(this, NULL, inode, version, _gf_true,
+                                       bad_object, NULL);
+}
+
+/** {{{ */
+
+int32_t
+br_stub_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+                dict_t *xdata)
+{
+    br_stub_private_t *priv = NULL;
+    br_stub_fd_t *fd_ctx = NULL;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+
+    priv = this->private;
+    if (gf_uuid_compare(fd->inode->gfid, priv->bad_object_dir_gfid))
+        goto normal;
+
+    fd_ctx = br_stub_fd_new();
+    if (!fd_ctx) {
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+
+    fd_ctx->bad_object.dir_eof = -1;
+    fd_ctx->bad_object.dir = sys_opendir(priv->stub_basepath);
+    if (!fd_ctx->bad_object.dir) {
+        op_errno = errno;
+        goto err_freectx;
+    }
+
+    op_ret = br_stub_fd_ctx_set(this, fd, fd_ctx);
+    if (!op_ret)
+        goto unwind;
+
+    sys_closedir(fd_ctx->bad_object.dir);
+
+err_freectx:
+    GF_FREE(fd_ctx);
+unwind:
+    STACK_UNWIND_STRICT(opendir, frame, op_ret, op_errno, fd, NULL);
+    return 0;
+
+normal:
+    STACK_WIND(frame, default_opendir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->opendir, loc, fd, xdata);
+    return 0;
+}
+
+int32_t
+br_stub_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                off_t off, dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    br_stub_private_t *priv = NULL;
+
+    priv = this->private;
+    if (!priv->do_versioning)
+        goto out;
+
+    if (gf_uuid_compare(fd->inode->gfid, priv->bad_object_dir_gfid))
+        goto out;
+    stub = fop_readdir_stub(frame, br_stub_readdir_wrapper, fd, size, off,
+                            xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(readdir, frame, -1, ENOMEM, NULL, NULL);
+        return 0;
+    }
+    br_stub_worker_enqueue(this, stub);
+    return 0;
+out:
+    STACK_WIND(frame, default_readdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdir, fd, size, off, xdata);
+    return 0;
+}
+
+int
+br_stub_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, gf_dirent_t *entries,
+                     dict_t *dict)
+{
+    int32_t ret = 0;
+    uint64_t ctxaddr = 0;
+    gf_dirent_t *entry = NULL;
+    br_stub_private_t *priv = NULL;
+    gf_boolean_t ver_enabled = _gf_false;
+
+    BR_STUB_VER_ENABLED_IN_CALLPATH(frame, ver_enabled);
+    priv = this->private;
+    BR_STUB_VER_COND_GOTO(priv, (!ver_enabled), unwind);
+
+    if (op_ret < 0)
+        goto unwind;
+
+    list_for_each_entry(entry, &entries->list, list)
+    {
+        if ((strcmp(entry->d_name, ".") == 0) ||
+            (strcmp(entry->d_name, "..") == 0))
+            continue;
+
+        if (!IA_ISREG(entry->d_stat.ia_type))
+            continue;
+
+        /*
+         * Readdirp for most part is a bulk lookup for all the entries
+         * present in the directory being read. Ideally, for each
+         * entry, the handling should be similar to that of a lookup
+         * callback. But for now, just keeping this as it has been
+         * until now (which means, this comment has been added much
+         * later as part of a change that wanted to send the flag
+         * of true/false to br_stub_remove_vxattrs to indicate whether
+         * the bad-object xattr should be removed from the entry->dict
+         * or not). Until this change, the function br_stub_remove_vxattrs
+         * was just removing all the xattrs associated with bit-rot-stub
+         * (like version, bad-object, signature etc). But, there are
+         * scenarios where we only want to send bad-object xattr and not
+         * others. So this comment is part of that change which also
+         * mentions about another possible change that might be needed
+         * in future.
+         * But for now, adding _gf_true means functionally its same as
+         * what this function was doing before. Just remove all the stub
+         * related xattrs.
+         */
+        ret = br_stub_get_inode_ctx(this, entry->inode, &ctxaddr);
+        if (ret < 0)
+            ctxaddr = 0;
+        if (ctxaddr) { /* already has the context */
+            br_stub_remove_vxattrs(entry->dict, _gf_true);
+            continue;
+        }
+
+        ret = br_stub_lookup_version(this, entry->inode->gfid, entry->inode,
+                                     entry->dict);
+        br_stub_remove_vxattrs(entry->dict, _gf_true);
+        if (ret) {
+            /**
+             * there's no per-file granularity support in case of
+             * failure. let's fail the entire request for now..
+             */
+            break;
+        }
+    }
+
+    if (ret) {
+        op_ret = -1;
+        op_errno = EINVAL;
+    }
+
+unwind:
+    STACK_UNWIND_STRICT(readdirp, frame, op_ret, op_errno, entries, dict);
+
+    return 0;
+}
+
+int
+br_stub_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                 off_t offset, dict_t *dict)
+{
+    int32_t ret = -1;
+    int op_errno = 0;
+    gf_boolean_t xref = _gf_false;
+    br_stub_private_t *priv = NULL;
+
+    priv = this->private;
+    BR_STUB_VER_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+
+    op_errno = ENOMEM;
+    if (!dict) {
+        dict = dict_new();
+        if (!dict)
+            goto unwind;
+    } else {
+        dict = dict_ref(dict);
+    }
+
+    xref = _gf_true;
+
+    op_errno = EINVAL;
+    ret = dict_set_uint32(dict, BITROT_CURRENT_VERSION_KEY, 0);
+    if (ret)
+        goto unwind;
+    ret = dict_set_uint32(dict, BITROT_SIGNING_VERSION_KEY, 0);
+    if (ret)
+        goto unwind;
+    ret = dict_set_uint32(dict, BITROT_OBJECT_BAD_KEY, 0);
+    if (ret)
+        goto unwind;
+
+wind:
+    STACK_WIND(frame, br_stub_readdirp_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdirp, fd, size, offset, dict);
+    goto unref_dict;
+
+unwind:
+    if (frame->local == (void *)0x1)
+        frame->local = NULL;
+    STACK_UNWIND_STRICT(readdirp, frame, -1, op_errno, NULL, NULL);
+    return 0;
+
+unref_dict:
+    if (xref)
+        dict_unref(dict);
+    return 0;
+}
+
+/** }}} */
+
+/** {{{ */
+
+/* lookup() */
+
+/**
+ * This function mainly handles the ENOENT error for the bad objects. Though
+ * br_stub_forget () handles removal of the link for the bad object from the
+ * quarantine directory, its better to handle it in lookup as well, where
+ * a failed lookup on a bad object with ENOENT, will trigger deletion of the
+ * link for the bad object from quarantine directory. So whoever comes first
+ * either forget () or lookup () will take care of removing the link.
+ */
+void
+br_stub_handle_lookup_error(xlator_t *this, inode_t *inode, int32_t op_errno)
+{
+    int32_t ret = -1;
+    uint64_t ctx_addr = 0;
+    br_stub_inode_ctx_t *ctx = NULL;
+
+    if (op_errno != ENOENT)
+        goto out;
+
+    if (!inode_is_linked(inode))
+        goto out;
+
+    ret = br_stub_get_inode_ctx(this, inode, &ctx_addr);
+    if (ret)
+        goto out;
+
+    ctx = (br_stub_inode_ctx_t *)(long)ctx_addr;
+
+    LOCK(&inode->lock);
+    {
+        if (__br_stub_is_bad_object(ctx))
+            (void)br_stub_del(this, inode->gfid);
+    }
+    UNLOCK(&inode->lock);
+
+    if (__br_stub_is_bad_object(ctx)) {
+        /* File is not present, might be deleted for recovery,
+         * del the bitrot inode context
+         */
+        ctx_addr = 0;
+        inode_ctx_del(inode, this, &ctx_addr);
+        if (ctx_addr) {
+            ctx = (br_stub_inode_ctx_t *)(long)ctx_addr;
+            GF_FREE(ctx);
+        }
+    }
+
+out:
+    return;
+}
+
+int
+br_stub_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int op_ret, int op_errno, inode_t *inode, struct iatt *stbuf,
+                   dict_t *xattr, struct iatt *postparent)
+{
+    int32_t ret = 0;
+    br_stub_private_t *priv = NULL;
+    gf_boolean_t ver_enabled = _gf_false;
+    gf_boolean_t remove_bad_file_marker = _gf_true;
+
+    BR_STUB_VER_ENABLED_IN_CALLPATH(frame, ver_enabled);
+    priv = this->private;
+
+    if (op_ret < 0) {
+        (void)br_stub_handle_lookup_error(this, inode, op_errno);
+
+        /*
+         * If the lookup error is not ENOENT, then it is better
+         * to send the bad file marker to the higher layer (if
+         * it has been set)
+         */
+        if (op_errno != ENOENT)
+            remove_bad_file_marker = _gf_false;
+        goto delkey;
+    }
+
+    BR_STUB_VER_COND_GOTO(priv, (!ver_enabled), delkey);
+
+    if (!IA_ISREG(stbuf->ia_type))
+        goto unwind;
+
+    /**
+     * If the object is bad, then "bad inode" marker has to be sent back
+     * in resoinse, for revalidated lookups as well. Some xlators such as
+     * quick-read might cache the data in revalidated lookup as fresh
+     * lookup would anyway have sent "bad inode" marker.
+     * In general send bad inode marker for every lookup operation on the
+     * bad object.
+     */
+    if (cookie != (void *)BR_STUB_REQUEST_COOKIE) {
+        ret = br_stub_mark_xdata_bad_object(this, inode, xattr);
+        if (ret) {
+            op_ret = -1;
+            op_errno = EIO;
+            /*
+             * This flag ensures that in the label @delkey below,
+             * bad file marker is not removed from the dictinary,
+             * but other virtual xattrs (such as version, signature)
+             * are removed.
+             */
+            remove_bad_file_marker = _gf_false;
+        }
+        goto delkey;
+    }
+
+    ret = br_stub_lookup_version(this, stbuf->ia_gfid, inode, xattr);
+    if (ret < 0) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto delkey;
+    }
+
+    /**
+     * If the object is bad, send "bad inode" marker back in response
+     * for xlator(s) to act accordingly (such as quick-read, etc..)
+     */
+    ret = br_stub_mark_xdata_bad_object(this, inode, xattr);
+    if (ret) {
+        /**
+         * aaha! bad object, but sorry we would not
+         * satisfy the request on allocation failures.
+         */
+        op_ret = -1;
+        op_errno = EIO;
+        goto delkey;
+    }
+
+delkey:
+    br_stub_remove_vxattrs(xattr, remove_bad_file_marker);
+unwind:
+    STACK_UNWIND_STRICT(lookup, frame, op_ret, op_errno, inode, stbuf, xattr,
+                        postparent);
+
+    return 0;
+}
+
+int
+br_stub_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    int32_t ret = 0;
+    int op_errno = 0;
+    void *cookie = NULL;
+    uint64_t ctx_addr = 0;
+    gf_boolean_t xref = _gf_false;
+    br_stub_private_t *priv = NULL;
+    call_stub_t *stub = NULL;
+
+    GF_VALIDATE_OR_GOTO("bit-rot-stub", this, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, loc, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, unwind);
+
+    priv = this->private;
+
+    BR_STUB_VER_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+
+    if (!gf_uuid_compare(loc->gfid, priv->bad_object_dir_gfid) ||
+        !gf_uuid_compare(loc->pargfid, priv->bad_object_dir_gfid)) {
+        stub = fop_lookup_stub(frame, br_stub_lookup_wrapper, loc, xdata);
+        if (!stub) {
+            op_errno = ENOMEM;
+            goto unwind;
+        }
+        br_stub_worker_enqueue(this, stub);
+        return 0;
+    }
+
+    ret = br_stub_get_inode_ctx(this, loc->inode, &ctx_addr);
+    if (ret < 0)
+        ctx_addr = 0;
+    if (ctx_addr != 0)
+        goto wind;
+
+    /**
+     * fresh lookup: request version keys from POSIX
+     */
+    op_errno = ENOMEM;
+    if (!xdata) {
+        xdata = dict_new();
+        if (!xdata)
+            goto unwind;
+    } else {
+        xdata = dict_ref(xdata);
+    }
+
+    xref = _gf_true;
+
+    /**
+     * Requesting both xattrs provides a way of sanity checking the
+     * object. Anomaly checking is done in cbk by examining absence
+     * of either or both xattrs.
+     */
+    op_errno = EINVAL;
+    ret = dict_set_uint32(xdata, BITROT_CURRENT_VERSION_KEY, 0);
+    if (ret)
+        goto unwind;
+    ret = dict_set_uint32(xdata, BITROT_SIGNING_VERSION_KEY, 0);
+    if (ret)
+        goto unwind;
+    ret = dict_set_uint32(xdata, BITROT_OBJECT_BAD_KEY, 0);
+    if (ret)
+        goto unwind;
+    cookie = (void *)BR_STUB_REQUEST_COOKIE;
+
+wind:
+    STACK_WIND_COOKIE(frame, br_stub_lookup_cbk, cookie, FIRST_CHILD(this),
+                      FIRST_CHILD(this)->fops->lookup, loc, xdata);
+    goto dealloc_dict;
+
+unwind:
+    if (frame->local == (void *)0x1)
+        frame->local = NULL;
+    STACK_UNWIND_STRICT(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+dealloc_dict:
+    if (xref)
+        dict_unref(xdata);
+    return 0;
+}
+
+/** }}} */
+
+/** {{{ */
+
+/* stat */
+int
+br_stub_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    int32_t ret = 0;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    br_stub_private_t *priv = NULL;
+
+    priv = this->private;
+
+    if (!priv->do_versioning)
+        goto wind;
+
+    if (!IA_ISREG(loc->inode->ia_type))
+        goto wind;
+
+    ret = br_stub_check_bad_object(this, loc->inode, &op_ret, &op_errno);
+    if (ret)
+        goto unwind;
+
+wind:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->stat,
+                    loc, xdata);
+    return 0;
+
+unwind:
+    STACK_UNWIND_STRICT(stat, frame, op_ret, op_errno, NULL, NULL);
+    return 0;
+}
+
+/* fstat */
+int
+br_stub_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    int32_t ret = 0;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    br_stub_private_t *priv = NULL;
+
+    priv = this->private;
+
+    if (!priv->do_versioning)
+        goto wind;
+
+    if (!IA_ISREG(fd->inode->ia_type))
+        goto wind;
+
+    ret = br_stub_check_bad_object(this, fd->inode, &op_ret, &op_errno);
+    if (ret)
+        goto unwind;
+
+wind:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->fstat,
+                    fd, xdata);
+    return 0;
+
+unwind:
+    STACK_UNWIND_STRICT(fstat, frame, op_ret, op_errno, NULL, NULL);
+    return 0;
+}
+
+/** }}} */
+
+/** {{{ */
+
+/* unlink() */
+
+int
+br_stub_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
+{
+    br_stub_local_t *local = NULL;
+    inode_t *inode = NULL;
+    uint64_t ctx_addr = 0;
+    br_stub_inode_ctx_t *ctx = NULL;
+    int32_t ret = -1;
+    br_stub_private_t *priv = NULL;
+    gf_boolean_t ver_enabled = _gf_false;
+
+    BR_STUB_VER_ENABLED_IN_CALLPATH(frame, ver_enabled);
+    priv = this->private;
+    BR_STUB_VER_COND_GOTO(priv, (!ver_enabled), unwind);
+
+    local = frame->local;
+    frame->local = NULL;
+
+    if (op_ret < 0)
+        goto unwind;
+
+    if (!local) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, BRS_MSG_NULL_LOCAL, NULL);
+        goto unwind;
+    }
+    inode = local->u.context.inode;
+    if (!IA_ISREG(inode->ia_type))
+        goto unwind;
+
+    ret = br_stub_get_inode_ctx(this, inode, &ctx_addr);
+    if (ret) {
+        /**
+         * If the inode is bad AND context is not there, then there
+         * is a possibility of the gfid of the object being listed
+         * in the quarantine directory and will be shown in the
+         * bad objects list. So continuing with the fop with a
+         * warning log. The entry from the quarantine directory
+         * has to be removed manually. Its not a good idea to fail
+         * the fop, as the object has already been deleted.
+         */
+        gf_smsg(this->name, GF_LOG_WARNING, 0, BRS_MSG_GET_INODE_CONTEXT_FAILED,
+                "inode-gfid=%s", uuid_utoa(inode->gfid), NULL);
+        goto unwind;
+    }
+
+    ctx = (br_stub_inode_ctx_t *)(long)ctx_addr;
+
+    LOCK(&inode->lock);
+    {
+        /**
+         * Ignoring the return value of br_stub_del ().
+         * There is not much that can be done if unlinking
+         * of the entry in the quarantine directory fails.
+         * The failure is logged.
+         */
+        if (__br_stub_is_bad_object(ctx))
+            (void)br_stub_del(this, inode->gfid);
+    }
+    UNLOCK(&inode->lock);
+
+unwind:
+    STACK_UNWIND_STRICT(unlink, frame, op_ret, op_errno, preparent, postparent,
+                        xdata);
+    br_stub_cleanup_local(local);
+    br_stub_dealloc_local(local);
+    return 0;
+}
+
+int
+br_stub_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int flag,
+               dict_t *xdata)
+{
+    br_stub_local_t *local = NULL;
+    int32_t op_ret = -1;
+    int32_t op_errno = 0;
+    br_stub_private_t *priv = NULL;
+
+    priv = this->private;
+    BR_STUB_VER_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+
+    local = br_stub_alloc_local(this);
+    if (!local) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, BRS_MSG_ALLOC_MEM_FAILED,
+                "local path=%s", loc->path, "gfid=%s",
+                uuid_utoa(loc->inode->gfid), NULL);
+        goto unwind;
+    }
+
+    br_stub_fill_local(local, NULL, NULL, loc->inode, loc->inode->gfid,
+                       BR_STUB_NO_VERSIONING, 0);
+
+    frame->local = local;
+
+wind:
+    STACK_WIND(frame, br_stub_unlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->unlink, loc, flag, xdata);
+    return 0;
+
+unwind:
+    if (frame->local == (void *)0x1)
+        frame->local = NULL;
+    STACK_UNWIND_STRICT(unlink, frame, op_ret, op_errno, NULL, NULL, NULL);
+    return 0;
+}
+
+/** }}} */
+
+/** {{{ */
+
+/* forget() */
+
+int
+br_stub_forget(xlator_t *this, inode_t *inode)
+{
+    uint64_t ctx_addr = 0;
+    br_stub_inode_ctx_t *ctx = NULL;
+
+    inode_ctx_del(inode, this, &ctx_addr);
+    if (!ctx_addr)
+        return 0;
+
+    ctx = (br_stub_inode_ctx_t *)(long)ctx_addr;
+
+    GF_FREE(ctx);
+
+    return 0;
+}
+
+/** }}} */
+
+/** {{{ */
+
+int32_t
+br_stub_noop(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, dict_t *xdata)
+{
+    STACK_DESTROY(frame->root);
+    return 0;
+}
+
+static void
+br_stub_send_ipc_fop(xlator_t *this, fd_t *fd, unsigned long releaseversion,
+                     int sign_info)
+{
+    int32_t op = 0;
+    int32_t ret = 0;
+    dict_t *xdata = NULL;
+    call_frame_t *frame = NULL;
+    changelog_event_t ev = {
+        0,
+    };
+
+    ev.ev_type = CHANGELOG_OP_TYPE_BR_RELEASE;
+    ev.u.releasebr.version = releaseversion;
+    ev.u.releasebr.sign_info = sign_info;
+    gf_uuid_copy(ev.u.releasebr.gfid, fd->inode->gfid);
+
+    xdata = dict_new();
+    if (!xdata) {
+        gf_smsg(this->name, GF_LOG_WARNING, ENOMEM, BRS_MSG_DICT_ALLOC_FAILED,
+                NULL);
+        goto out;
+    }
+
+    ret = dict_set_static_bin(xdata, "RELEASE-EVENT", &ev, CHANGELOG_EV_SIZE);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, BRS_MSG_SET_EVENT_FAILED, NULL);
+        goto dealloc_dict;
+    }
+
+    frame = create_frame(this, this->ctx->pool);
+    if (!frame) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, BRS_MSG_CREATE_FRAME_FAILED,
+                NULL);
+        goto dealloc_dict;
+    }
+
+    op = GF_IPC_TARGET_CHANGELOG;
+    STACK_WIND(frame, br_stub_noop, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ipc, op, xdata);
+
+dealloc_dict:
+    dict_unref(xdata);
+out:
+    return;
+}
+
+/**
+ * This is how the state machine of sign info works:
+ * 3 states:
+ * 1) BR_SIGN_NORMAL => The default State of the inode
+ * 2) BR_SIGN_REOPEN_WAIT => A release has been sent and is waiting for reopen
+ * 3) BR_SIGN_QUICK => reopen has happened and this release should trigger sign
+ * 2 events:
+ * 1) GF_FOP_RELEASE
+ * 2) GF_FOP_WRITE (actually a dummy write for BitD)
+ *
+ * This is how states are changed based on events:
+ * EVENT: GF_FOP_RELEASE:
+ * if (state == BR_SIGN_NORMAL) ; then
+ *     set state = BR_SIGN_REOPEN_WAIT;
+ * if (state == BR_SIGN_QUICK); then
+ *     set state = BR_SIGN_NORMAL;
+ * EVENT: GF_FOP_WRITE:
+ *  if (state == BR_SIGN_REOPEN_WAIT); then
+ *     set state = BR_SIGN_QUICK;
+ */
+br_sign_state_t
+__br_stub_inode_sign_state(br_stub_inode_ctx_t *ctx, glusterfs_fop_t fop,
+                           fd_t *fd)
+{
+    br_sign_state_t sign_info = BR_SIGN_INVALID;
+
+    switch (fop) {
+        case GF_FOP_FSETXATTR:
+            sign_info = ctx->info_sign = BR_SIGN_QUICK;
+            break;
+
+        case GF_FOP_RELEASE:
+            GF_ASSERT(ctx->info_sign != BR_SIGN_REOPEN_WAIT);
+
+            if (ctx->info_sign == BR_SIGN_NORMAL) {
+                sign_info = ctx->info_sign = BR_SIGN_REOPEN_WAIT;
+            } else {
+                sign_info = ctx->info_sign;
+                ctx->info_sign = BR_SIGN_NORMAL;
+            }
+
+            break;
+        default:
+            break;
+    }
+
+    return sign_info;
+}
+
+int32_t
+br_stub_release(xlator_t *this, fd_t *fd)
+{
+    int32_t ret = 0;
+    int32_t flags = 0;
+    inode_t *inode = NULL;
+    unsigned long releaseversion = 0;
+    br_stub_inode_ctx_t *ctx = NULL;
+    uint64_t tmp = 0;
+    br_stub_fd_t *br_stub_fd = NULL;
+    int32_t signinfo = 0;
+
+    inode = fd->inode;
+
+    LOCK(&inode->lock);
+    {
+        ctx = __br_stub_get_ongoing_version_ctx(this, inode, NULL);
+        if (ctx == NULL)
+            goto unblock;
+        br_stub_fd = br_stub_fd_ctx_get(this, fd);
+        if (br_stub_fd) {
+            list_del_init(&br_stub_fd->list);
+        }
+
+        ret = __br_stub_can_trigger_release(inode, ctx, &releaseversion);
+        if (!ret)
+            goto unblock;
+
+        signinfo = __br_stub_inode_sign_state(ctx, GF_FOP_RELEASE, fd);
+        signinfo = htonl(signinfo);
+
+        /* inode back to initital state: mark dirty */
+        if (ctx->info_sign == BR_SIGN_NORMAL) {
+            __br_stub_mark_inode_dirty(ctx);
+            __br_stub_unset_inode_modified(ctx);
+        }
+    }
+unblock:
+    UNLOCK(&inode->lock);
+
+    if (ret) {
+        gf_msg_debug(this->name, 0,
+                     "releaseversion: %lu | flags: %d "
+                     "| signinfo: %d",
+                     (unsigned long)ntohl(releaseversion), flags,
+                     ntohl(signinfo));
+        br_stub_send_ipc_fop(this, fd, releaseversion, signinfo);
+    }
+
+    ret = fd_ctx_del(fd, this, &tmp);
+    br_stub_fd = (br_stub_fd_t *)(long)tmp;
+
+    GF_FREE(br_stub_fd);
+
+    return 0;
+}
+
+int32_t
+br_stub_releasedir(xlator_t *this, fd_t *fd)
+{
+    br_stub_fd_t *fctx = NULL;
+    uint64_t ctx = 0;
+    int ret = 0;
+
+    ret = fd_ctx_del(fd, this, &ctx);
+    if (ret < 0)
+        goto out;
+
+    fctx = (br_stub_fd_t *)(long)ctx;
+    if (fctx->bad_object.dir) {
+        ret = sys_closedir(fctx->bad_object.dir);
+        if (ret)
+            gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_BAD_OBJ_DIR_CLOSE_FAIL,
+                    "error=%s", strerror(errno), NULL);
+    }
+
+    GF_FREE(fctx);
+out:
+    return 0;
+}
+
+/** }}} */
+
+/** {{{ */
+
+/* ictxmerge */
+
+void
+br_stub_ictxmerge(xlator_t *this, fd_t *fd, inode_t *inode,
+                  inode_t *linked_inode)
+{
+    int32_t ret = 0;
+    uint64_t ctxaddr = 0;
+    uint64_t lctxaddr = 0;
+    br_stub_inode_ctx_t *ctx = NULL;
+    br_stub_inode_ctx_t *lctx = NULL;
+    br_stub_fd_t *br_stub_fd = NULL;
+
+    ret = br_stub_get_inode_ctx(this, inode, &ctxaddr);
+    if (ret < 0)
+        goto done;
+    ctx = (br_stub_inode_ctx_t *)(uintptr_t)ctxaddr;
+
+    LOCK(&linked_inode->lock);
+    {
+        ret = __br_stub_get_inode_ctx(this, linked_inode, &lctxaddr);
+        if (ret < 0)
+            goto unblock;
+        lctx = (br_stub_inode_ctx_t *)(uintptr_t)lctxaddr;
+
+        GF_ASSERT(list_is_singular(&ctx->fd_list));
+        br_stub_fd = list_first_entry(&ctx->fd_list, br_stub_fd_t, list);
+        if (br_stub_fd) {
+            GF_ASSERT(br_stub_fd->fd == fd);
+            list_move_tail(&br_stub_fd->list, &lctx->fd_list);
+        }
+    }
+unblock:
+    UNLOCK(&linked_inode->lock);
+
+done:
+    return;
+}
+
+/** }}} */
+
+struct xlator_fops fops = {
+    .lookup = br_stub_lookup,
+    .stat = br_stub_stat,
+    .fstat = br_stub_fstat,
+    .open = br_stub_open,
+    .create = br_stub_create,
+    .readdirp = br_stub_readdirp,
+    .getxattr = br_stub_getxattr,
+    .fgetxattr = br_stub_fgetxattr,
+    .fsetxattr = br_stub_fsetxattr,
+    .writev = br_stub_writev,
+    .truncate = br_stub_truncate,
+    .ftruncate = br_stub_ftruncate,
+    .mknod = br_stub_mknod,
+    .readv = br_stub_readv,
+    .removexattr = br_stub_removexattr,
+    .fremovexattr = br_stub_fremovexattr,
+    .setxattr = br_stub_setxattr,
+    .opendir = br_stub_opendir,
+    .readdir = br_stub_readdir,
+    .unlink = br_stub_unlink,
+};
+
+struct xlator_cbks cbks = {
+    .forget = br_stub_forget,
+    .release = br_stub_release,
+    .ictxmerge = br_stub_ictxmerge,
+};
+
+struct volume_options options[] = {
+    {.key = {"bitrot"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .op_version = {GD_OP_VERSION_3_7_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_FORCE,
+     .tags = {"bitrot"},
+     .description = "enable/disable bitrot stub"},
+    {.key = {"export"},
+     .type = GF_OPTION_TYPE_PATH,
+     .op_version = {GD_OP_VERSION_3_7_0},
+     .tags = {"bitrot"},
+     .description = "brick path for versioning",
+     .default_value = "{{ brick.path }}"},
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "bitrot-stub",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/features/bit-rot/src/stub/bit-rot-stub.h b/xlators/features/bit-rot/src/stub/bit-rot-stub.h
new file mode 100644
index 00000000000..edd79a77e4f
--- /dev/null
+++ b/xlators/features/bit-rot/src/stub/bit-rot-stub.h
@@ -0,0 +1,515 @@
+/*
+  Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+#ifndef __BIT_ROT_STUB_H__
+#define __BIT_ROT_STUB_H__
+
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/dict.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/call-stub.h>
+#include "bit-rot-stub-mem-types.h"
+#include <glusterfs/syscall.h>
+#include <glusterfs/common-utils.h>
+#include "bit-rot-common.h"
+#include "bit-rot-stub-messages.h"
+#include "glusterfs3-xdr.h"
+#include <glusterfs/syncop.h>
+#include <glusterfs/syncop-utils.h>
+
+#define BAD_OBJECT_THREAD_STACK_SIZE ((size_t)(1024 * 1024))
+#define BR_STUB_DUMP_STR_SIZE 65536
+
+#define BR_PATH_MAX_EXTRA (PATH_MAX + 1024)
+#define BR_PATH_MAX_PLUS (PATH_MAX + 2048)
+
+/*
+ * Oops. Spelling mistake. Correcting it
+ */
+#define OLD_BR_STUB_QUARANTINE_DIR GF_HIDDEN_PATH "/quanrantine"
+#define BR_STUB_QUARANTINE_DIR GF_HIDDEN_PATH "/quarantine"
+
+/* do not reference frame->local in cbk unless initialized.
+ * Assigned 0x1 marks verisoning flag between call path and
+ * cbk path.
+ */
+#define BR_STUB_VER_NOT_ACTIVE_THEN_GOTO(frame, priv, label)                   \
+    do {                                                                       \
+        if (priv->do_versioning)                                               \
+            frame->local = (void *)0x1;                                        \
+        else                                                                   \
+            goto label;                                                        \
+    } while (0)
+
+#define BR_STUB_VER_COND_GOTO(priv, cond, label)                               \
+    do {                                                                       \
+        if (!priv->do_versioning || cond)                                      \
+            goto label;                                                        \
+    } while (0)
+
+#define BR_STUB_VER_ENABLED_IN_CALLPATH(frame, flag)                           \
+    do {                                                                       \
+        if (frame->local)                                                      \
+            flag = _gf_true;                                                   \
+        if (frame->local == (void *)0x1)                                       \
+            frame->local = NULL;                                               \
+    } while (0)
+
+#define BR_STUB_RESET_LOCAL_NULL(frame)                                        \
+    do {                                                                       \
+        if (frame->local == (void *)0x1)                                       \
+            frame->local = NULL;                                               \
+    } while (0)
+
+typedef int(br_stub_version_cbk)(call_frame_t *, void *, xlator_t *, int32_t,
+                                 int32_t, dict_t *);
+
+typedef struct br_stub_inode_ctx {
+    int need_writeback;           /* does the inode need
+                                        a writeback to disk? */
+    unsigned long currentversion; /* ongoing version */
+
+    int info_sign;
+    struct list_head fd_list; /* list of open fds or fds participating in
+                                 write operations */
+    gf_boolean_t bad_object;
+} br_stub_inode_ctx_t;
+
+typedef struct br_stub_fd {
+    fd_t *fd;
+    struct list_head list;
+    struct bad_object_dir {
+        DIR *dir;
+        off_t dir_eof;
+    } bad_object;
+} br_stub_fd_t;
+
+#define I_DIRTY (1 << 0) /* inode needs writeback */
+#define I_MODIFIED (1 << 1)
+#define WRITEBACK_DURABLE 1 /* writeback is durable */
+
+/**
+ * This could just have been a plain struct without unions and all,
+ * but we may need additional things in the future.
+ */
+typedef struct br_stub_local {
+    call_stub_t *fopstub; /* stub for original fop */
+
+    int versioningtype; /* not much used atm */
+
+    union {
+        struct br_stub_ctx {
+            fd_t *fd;
+            uuid_t gfid;
+            inode_t *inode;
+            unsigned long version;
+        } context;
+    } u;
+} br_stub_local_t;
+
+#define BR_STUB_NO_VERSIONING (1 << 0)
+#define BR_STUB_INCREMENTAL_VERSIONING (1 << 1)
+
+typedef struct br_stub_private {
+    gf_boolean_t do_versioning;
+
+    uint32_t boot[2];
+    char export[PATH_MAX];
+
+    pthread_mutex_t lock;
+    pthread_cond_t cond;
+
+    struct list_head squeue; /* ordered signing queue */
+    pthread_t signth;
+    struct bad_objects_container {
+        pthread_t thread;
+        pthread_mutex_t bad_lock;
+        pthread_cond_t bad_cond;
+        struct list_head bad_queue;
+    } container;
+    struct mem_pool *local_pool;
+
+    char stub_basepath[BR_PATH_MAX_EXTRA];
+
+    uuid_t bad_object_dir_gfid;
+} br_stub_private_t;
+
+br_stub_fd_t *
+br_stub_fd_new(void);
+
+int
+__br_stub_fd_ctx_set(xlator_t *this, fd_t *fd, br_stub_fd_t *br_stub_fd);
+
+br_stub_fd_t *
+__br_stub_fd_ctx_get(xlator_t *this, fd_t *fd);
+
+br_stub_fd_t *
+br_stub_fd_ctx_get(xlator_t *this, fd_t *fd);
+
+int32_t
+br_stub_fd_ctx_set(xlator_t *this, fd_t *fd, br_stub_fd_t *br_stub_fd);
+
+static inline gf_boolean_t
+__br_stub_is_bad_object(br_stub_inode_ctx_t *ctx)
+{
+    return ctx->bad_object;
+}
+
+static inline void
+__br_stub_mark_object_bad(br_stub_inode_ctx_t *ctx)
+{
+    ctx->bad_object = _gf_true;
+}
+
+/* inode writeback helpers */
+static inline void
+__br_stub_mark_inode_dirty(br_stub_inode_ctx_t *ctx)
+{
+    ctx->need_writeback |= I_DIRTY;
+}
+
+static inline void
+__br_stub_mark_inode_synced(br_stub_inode_ctx_t *ctx)
+{
+    ctx->need_writeback &= ~I_DIRTY;
+}
+
+static inline int
+__br_stub_is_inode_dirty(br_stub_inode_ctx_t *ctx)
+{
+    return (ctx->need_writeback & I_DIRTY);
+}
+
+/* inode mofification markers */
+static inline void
+__br_stub_set_inode_modified(br_stub_inode_ctx_t *ctx)
+{
+    ctx->need_writeback |= I_MODIFIED;
+}
+
+static inline void
+__br_stub_unset_inode_modified(br_stub_inode_ctx_t *ctx)
+{
+    ctx->need_writeback &= ~I_MODIFIED;
+}
+
+static inline int
+__br_stub_is_inode_modified(br_stub_inode_ctx_t *ctx)
+{
+    return (ctx->need_writeback & I_MODIFIED);
+}
+
+static inline int
+br_stub_require_release_call(xlator_t *this, fd_t *fd, br_stub_fd_t **fd_ctx)
+{
+    int32_t ret = 0;
+    br_stub_fd_t *br_stub_fd = NULL;
+
+    br_stub_fd = br_stub_fd_new();
+    if (!br_stub_fd)
+        return -1;
+
+    br_stub_fd->fd = fd;
+    INIT_LIST_HEAD(&br_stub_fd->list);
+
+    ret = br_stub_fd_ctx_set(this, fd, br_stub_fd);
+    if (ret)
+        gf_smsg(this->name, GF_LOG_WARNING, 0, BRS_MSG_SET_CONTEXT_FAILED,
+                NULL);
+    else
+        *fd_ctx = br_stub_fd;
+
+    return ret;
+}
+
+/* get/set inode context helpers */
+
+static inline int
+__br_stub_get_inode_ctx(xlator_t *this, inode_t *inode, uint64_t *ctx)
+{
+    return __inode_ctx_get(inode, this, ctx);
+}
+
+static inline int
+br_stub_get_inode_ctx(xlator_t *this, inode_t *inode, uint64_t *ctx)
+{
+    int ret = -1;
+
+    LOCK(&inode->lock);
+    {
+        ret = __br_stub_get_inode_ctx(this, inode, ctx);
+    }
+    UNLOCK(&inode->lock);
+
+    return ret;
+}
+
+static inline int
+br_stub_set_inode_ctx(xlator_t *this, inode_t *inode, br_stub_inode_ctx_t *ctx)
+{
+    uint64_t ctx_addr = (uint64_t)(uintptr_t)ctx;
+    return inode_ctx_set(inode, this, &ctx_addr);
+}
+
+/* version get/set helpers */
+
+static inline unsigned long
+__br_stub_writeback_version(br_stub_inode_ctx_t *ctx)
+{
+    return (ctx->currentversion + 1);
+}
+
+static inline void
+__br_stub_set_ongoing_version(br_stub_inode_ctx_t *ctx, unsigned long version)
+{
+    if (ctx->currentversion < version)
+        ctx->currentversion = version;
+    else
+        gf_smsg("bit-rot-stub", GF_LOG_WARNING, 0,
+                BRS_MSG_CHANGE_VERSION_FAILED, "current version=%lu",
+                ctx->currentversion, "new version=%lu", version, NULL);
+}
+
+static inline int
+__br_stub_can_trigger_release(inode_t *inode, br_stub_inode_ctx_t *ctx,
+                              unsigned long *version)
+{
+    /**
+     * If the inode is modified, then it has to be dirty. An inode is
+     * marked dirty once version is increased. Its marked as modified
+     * when the modification call (write/truncate) which triggered
+     * the versioning is successful.
+     */
+    if (__br_stub_is_inode_modified(ctx) && list_empty(&ctx->fd_list) &&
+        (ctx->info_sign != BR_SIGN_REOPEN_WAIT)) {
+        GF_ASSERT(__br_stub_is_inode_dirty(ctx) == 0);
+
+        if (version)
+            *version = htonl(ctx->currentversion);
+        return 1;
+    }
+
+    return 0;
+}
+
+static inline int32_t
+br_stub_get_ongoing_version(xlator_t *this, inode_t *inode,
+                            unsigned long *version)
+{
+    int32_t ret = 0;
+    uint64_t ctx_addr = 0;
+    br_stub_inode_ctx_t *ctx = NULL;
+
+    LOCK(&inode->lock);
+    {
+        ret = __inode_ctx_get(inode, this, &ctx_addr);
+        if (ret < 0)
+            goto unblock;
+        ctx = (br_stub_inode_ctx_t *)(long)ctx_addr;
+        *version = ctx->currentversion;
+    }
+unblock:
+    UNLOCK(&inode->lock);
+
+    return ret;
+}
+
+/**
+ * fetch the current version from inode and return the context.
+ * inode->lock should be held before invoking this as context
+ * *needs* to be valid in the caller.
+ */
+static inline br_stub_inode_ctx_t *
+__br_stub_get_ongoing_version_ctx(xlator_t *this, inode_t *inode,
+                                  unsigned long *version)
+{
+    int32_t ret = 0;
+    uint64_t ctx_addr = 0;
+    br_stub_inode_ctx_t *ctx = NULL;
+
+    ret = __inode_ctx_get(inode, this, &ctx_addr);
+    if (ret < 0)
+        return NULL;
+    ctx = (br_stub_inode_ctx_t *)(long)ctx_addr;
+    if (version)
+        *version = ctx->currentversion;
+
+    return ctx;
+}
+
+/* filter for xattr fetch */
+static inline int
+br_stub_is_internal_xattr(const char *name)
+{
+    if (name && ((strncmp(name, BITROT_CURRENT_VERSION_KEY,
+                          SLEN(BITROT_CURRENT_VERSION_KEY)) == 0) ||
+                 (strncmp(name, BITROT_SIGNING_VERSION_KEY,
+                          SLEN(BITROT_SIGNING_VERSION_KEY)) == 0)))
+        return 1;
+    return 0;
+}
+
+static inline void
+br_stub_remove_vxattrs(dict_t *xattr, gf_boolean_t remove_bad_marker)
+{
+    if (xattr) {
+        /*
+         * When a file is corrupted, bad-object should be
+         * set in the dict. But, other info such as version,
+         * signature etc should not be set. Hence the flag
+         * remove_bad_marker. The consumer should know whether
+         * to send the bad-object info in the dict or not.
+         */
+        if (remove_bad_marker)
+            dict_del(xattr, BITROT_OBJECT_BAD_KEY);
+        dict_del(xattr, BITROT_CURRENT_VERSION_KEY);
+        dict_del(xattr, BITROT_SIGNING_VERSION_KEY);
+        dict_del(xattr, BITROT_SIGNING_XATTR_SIZE_KEY);
+    }
+}
+
+/**
+ * This function returns the below values for different situations
+ * 0  => as per the inode context object is not bad
+ * -1 => Failed to get the inode context itself
+ * -2 => As per the inode context object is bad
+ * Both -ve values means the fop which called this function is failed
+ * and error is returned upwards.
+ * In future if needed or more errors have to be handled, then those
+ * errors can be made into enums.
+ */
+static inline int
+br_stub_is_bad_object(xlator_t *this, inode_t *inode)
+{
+    int bad_object = 0;
+    gf_boolean_t tmp = _gf_false;
+    uint64_t ctx_addr = 0;
+    br_stub_inode_ctx_t *ctx = NULL;
+    int32_t ret = -1;
+
+    ret = br_stub_get_inode_ctx(this, inode, &ctx_addr);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_GET_INODE_CONTEXT_FAILED,
+                "inode-gfid=%s", uuid_utoa(inode->gfid), NULL);
+        bad_object = -1;
+        goto out;
+    }
+
+    ctx = (br_stub_inode_ctx_t *)(long)ctx_addr;
+
+    LOCK(&inode->lock);
+    {
+        tmp = __br_stub_is_bad_object(ctx);
+        if (tmp)
+            bad_object = -2;
+    }
+    UNLOCK(&inode->lock);
+
+out:
+    return bad_object;
+}
+
+static inline int32_t
+br_stub_mark_object_bad(xlator_t *this, inode_t *inode)
+{
+    int32_t ret = -1;
+    uint64_t ctx_addr = 0;
+    br_stub_inode_ctx_t *ctx = NULL;
+
+    ret = br_stub_get_inode_ctx(this, inode, &ctx_addr);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, BRS_MSG_GET_INODE_CONTEXT_FAILED,
+                "inode-gfid=%s", uuid_utoa(inode->gfid), NULL);
+        goto out;
+    }
+
+    ctx = (br_stub_inode_ctx_t *)(long)ctx_addr;
+
+    LOCK(&inode->lock);
+    {
+        __br_stub_mark_object_bad(ctx);
+    }
+    UNLOCK(&inode->lock);
+
+out:
+    return ret;
+}
+
+/**
+ * There is a possibility that dict_set might fail. The o/p of dict_set is
+ * given to the caller and the caller has to decide what to do.
+ */
+static inline int32_t
+br_stub_mark_xdata_bad_object(xlator_t *this, inode_t *inode, dict_t *xdata)
+{
+    int32_t ret = 0;
+
+    if (br_stub_is_bad_object(this, inode) == -2)
+        ret = dict_set_int32(xdata, GLUSTERFS_BAD_INODE, 1);
+
+    return ret;
+}
+
+int32_t
+br_stub_add_fd_to_inode(xlator_t *this, fd_t *fd, br_stub_inode_ctx_t *ctx);
+
+br_sign_state_t
+__br_stub_inode_sign_state(br_stub_inode_ctx_t *ctx, glusterfs_fop_t fop,
+                           fd_t *fd);
+
+int
+br_stub_dir_create(xlator_t *this, br_stub_private_t *priv);
+
+int
+br_stub_add(xlator_t *this, uuid_t gfid);
+
+int32_t
+br_stub_create_stub_gfid(xlator_t *this, char *stub_gfid_path, uuid_t gfid);
+
+int
+br_stub_dir_create(xlator_t *this, br_stub_private_t *priv);
+
+call_stub_t *
+__br_stub_dequeue(struct list_head *callstubs);
+
+void
+__br_stub_enqueue(struct list_head *callstubs, call_stub_t *stub);
+
+void
+br_stub_worker_enqueue(xlator_t *this, call_stub_t *stub);
+
+void *
+br_stub_worker(void *data);
+
+int32_t
+br_stub_lookup_wrapper(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                       dict_t *xattr_req);
+
+int32_t
+br_stub_readdir_wrapper(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                        size_t size, off_t off, dict_t *xdata);
+
+int
+br_stub_del(xlator_t *this, uuid_t gfid);
+
+int
+br_stub_bad_objects_path(xlator_t *this, fd_t *fd, gf_dirent_t *entries,
+                         dict_t **dict);
+
+void
+br_stub_entry_xattr_fill(xlator_t *this, char *hpath, gf_dirent_t *entry,
+                         dict_t *dict);
+
+int
+br_stub_get_path_of_gfid(xlator_t *this, inode_t *parent, inode_t *inode,
+                         uuid_t gfid, char **path);
+
+#endif /* __BIT_ROT_STUB_H__ */
diff --git a/xlators/features/changelog/Makefile.am b/xlators/features/changelog/Makefile.am
new file mode 100644
index 00000000000..153bb685076
--- /dev/null
+++ b/xlators/features/changelog/Makefile.am
@@ -0,0 +1,3 @@
+SUBDIRS = src lib
+
+CLEANFILES =
diff --git a/xlators/encryption/rot-13/Makefile.am b/xlators/features/changelog/lib/Makefile.am
index d471a3f9243..a985f42a877 100644
--- a/xlators/encryption/rot-13/Makefile.am
+++ b/xlators/features/changelog/lib/Makefile.am
@@ -1,3 +1,3 @@
 SUBDIRS = src
 
-CLEANFILES = 
+CLEANFILES =
diff --git a/xlators/features/changelog/lib/examples/c/get-changes-multi.c b/xlators/features/changelog/lib/examples/c/get-changes-multi.c
new file mode 100644
index 00000000000..5ea5bbb6630
--- /dev/null
+++ b/xlators/features/changelog/lib/examples/c/get-changes-multi.c
@@ -0,0 +1,90 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+/**
+ * Compile it using:
+ *  gcc -o getchanges-multi `pkg-config --cflags libgfchangelog` \
+ *  get-changes-multi.c `pkg-config --libs libgfchangelog`
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <sys/un.h>
+#include <limits.h>
+#include <sys/socket.h>
+#include <sys/types.h>
+#include <errno.h>
+
+#include "changelog.h"
+
+void *
+brick_init(void *xl, struct gf_brick_spec *brick)
+{
+    return brick;
+}
+
+void
+brick_fini(void *xl, char *brick, void *data)
+{
+    return;
+}
+
+void
+brick_callback(void *xl, char *brick, void *data, changelog_event_t *ev)
+{
+    printf("->callback: (brick,type) [%s:%d]\n", brick, ev->ev_type);
+}
+
+void
+fill_brick_spec(struct gf_brick_spec *brick, char *path)
+{
+    brick->brick_path = strdup(path);
+    brick->filter = CHANGELOG_OP_TYPE_BR_RELEASE;
+
+    brick->init = brick_init;
+    brick->fini = brick_fini;
+    brick->callback = brick_callback;
+    brick->connected = NULL;
+    brick->disconnected = NULL;
+}
+
+int
+main(int argc, char **argv)
+{
+    int ret = 0;
+    void *bricks = NULL;
+    struct gf_brick_spec *brick = NULL;
+
+    bricks = calloc(2, sizeof(struct gf_brick_spec));
+    if (!bricks)
+        goto error_return;
+
+    brick = (struct gf_brick_spec *)bricks;
+    fill_brick_spec(brick, "/export/z1/zwoop");
+
+    brick++;
+    fill_brick_spec(brick, "/export/z2/zwoop");
+
+    ret = gf_changelog_init(NULL);
+    if (ret)
+        goto error_return;
+
+    ret = gf_changelog_register_generic((struct gf_brick_spec *)bricks, 2, 0,
+                                        "/tmp/multi-changes.log", 9, NULL);
+    if (ret)
+        goto error_return;
+
+    /* let callbacks do the job */
+    select(0, NULL, NULL, NULL, NULL);
+
+error_return:
+    return -1;
+}
diff --git a/xlators/features/changelog/lib/examples/c/get-changes.c b/xlators/features/changelog/lib/examples/c/get-changes.c
new file mode 100644
index 00000000000..8bc651c24a4
--- /dev/null
+++ b/xlators/features/changelog/lib/examples/c/get-changes.c
@@ -0,0 +1,93 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+/**
+ * get set of new changes every 10 seconds (just print the file names)
+ *
+ * Compile it using:
+ *  gcc -o getchanges `pkg-config --cflags libgfchangelog` get-changes.c \
+ *  `pkg-config --libs libgfchangelog`
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <sys/un.h>
+#include <limits.h>
+#include <sys/socket.h>
+#include <sys/types.h>
+#include <errno.h>
+
+#include "changelog.h"
+
+#define handle_error(fn) printf("%s (reason: %s)\n", fn, strerror(errno))
+
+int
+main(int argc, char **argv)
+{
+    int i = 0;
+    int ret = 0;
+    ssize_t nr_changes = 0;
+    ssize_t changes = 0;
+    char fbuf[PATH_MAX] = {
+        0,
+    };
+
+    ret = gf_changelog_init(NULL);
+    if (ret) {
+        handle_error("Init failed");
+        goto out;
+    }
+
+    /* get changes for brick "/home/vshankar/export/yow/yow-1" */
+    ret = gf_changelog_register("/export/z1/zwoop", "/tmp/scratch",
+                                "/tmp/change.log", 9, 5);
+    if (ret) {
+        handle_error("register failed");
+        goto out;
+    }
+
+    while (1) {
+        i = 0;
+        nr_changes = gf_changelog_scan();
+        if (nr_changes < 0) {
+            handle_error("scan(): ");
+            break;
+        }
+
+        if (nr_changes == 0)
+            goto next;
+
+        printf("Got %ld changelog files\n", nr_changes);
+
+        while ((changes = gf_changelog_next_change(fbuf, PATH_MAX)) > 0) {
+            printf("changelog file [%d]: %s\n", ++i, fbuf);
+
+            /* process changelog */
+            /* ... */
+            /* ... */
+            /* ... */
+            /* done processing */
+
+            ret = gf_changelog_done(fbuf);
+            if (ret)
+                handle_error("gf_changelog_done");
+        }
+
+        if (changes == -1)
+            handle_error("gf_changelog_next_change");
+
+    next:
+        sleep(10);
+    }
+
+out:
+    return ret;
+}
diff --git a/xlators/features/changelog/lib/examples/c/get-history.c b/xlators/features/changelog/lib/examples/c/get-history.c
new file mode 100644
index 00000000000..3e888d75ca6
--- /dev/null
+++ b/xlators/features/changelog/lib/examples/c/get-history.c
@@ -0,0 +1,116 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+/**
+ * get set of new changes every 10 seconds (just print the file names)
+ *
+ * Compile it using:
+ *  gcc -o gethistory `pkg-config --cflags libgfchangelog` get-history.c \
+ *  `pkg-config --libs libgfchangelog`
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <sys/un.h>
+#include <limits.h>
+#include <sys/socket.h>
+#include <sys/types.h>
+#include <errno.h>
+
+#include "changelog.h"
+
+#define handle_error(fn) printf("%s (reason: %s)\n", fn, strerror(errno))
+
+int
+main(int argc, char **argv)
+{
+    int i = 0;
+    int ret = 0;
+    ssize_t nr_changes = 0;
+    ssize_t changes = 0;
+    char fbuf[PATH_MAX] = {
+        0,
+    };
+    unsigned long end_ts = 0;
+
+    ret = gf_changelog_init(NULL);
+    if (ret) {
+        handle_error("init failed");
+        goto out;
+    }
+
+    ret = gf_changelog_register("/export/z1/zwoop", "/tmp/scratch_v1",
+                                "/tmp/changes.log", 9, 5);
+    if (ret) {
+        handle_error("register failed");
+        goto out;
+    }
+
+    int a, b;
+    printf("give the two numbers start and end\t");
+    scanf("%d%d", &a, &b);
+    ret = gf_history_changelog("/export/z1/zwoop/.glusterfs/changelogs", a, b,
+                               3, &end_ts);
+    if (ret == -1) {
+        printf("history failed");
+        goto out;
+    }
+
+    printf("end time till when changelog available : %d , ret(%d) \t", end_ts,
+           ret);
+    fflush(stdout);
+
+    while (1) {
+        nr_changes = gf_history_changelog_scan();
+        printf("scanned, nr_changes : %d\n", nr_changes);
+        if (nr_changes < 0) {
+            handle_error("scan(): ");
+            break;
+        }
+
+        if (nr_changes == 0) {
+            printf("done scanning \n");
+            goto out;
+        }
+
+        printf("Got %ld changelog files\n", nr_changes);
+
+        while ((changes = gf_history_changelog_next_change(fbuf, PATH_MAX)) >
+               0) {
+            printf("changelog file [%d]: %s\n", ++i, fbuf);
+
+            /* process changelog */
+            /* ... */
+            /* ... */
+            /* ... */
+            /* done processing */
+
+            ret = gf_history_changelog_done(fbuf);
+            if (ret)
+                handle_error("gf_changelog_done");
+        }
+        /*
+        if (changes == -1)
+                handle_error ("gf_changelog_next_change");
+        if (nr_changes ==1){
+                printf("continue scanning\n");
+        }
+
+        if(nr_changes == 0){
+                printf("done scanning \n");
+                goto out;
+        }
+        */
+    }
+
+out:
+    return ret;
+}
diff --git a/xlators/features/changelog/lib/examples/python/changes.py b/xlators/features/changelog/lib/examples/python/changes.py
new file mode 100755
index 00000000000..c410d3b000d
--- /dev/null
+++ b/xlators/features/changelog/lib/examples/python/changes.py
@@ -0,0 +1,34 @@
+#!/usr/bin/python3
+
+from __future__ import print_function
+import os
+import sys
+import time
+import libgfchangelog
+
+cl = libgfchangelog.Changes()
+
+def get_changes(brick, scratch_dir, log_file, log_level, interval):
+    change_list = []
+    try:
+        cl.cl_init()
+        cl.cl_register(brick, scratch_dir, log_file, log_level)
+        while True:
+            cl.cl_scan()
+            change_list = cl.cl_getchanges()
+            if change_list:
+                print(change_list)
+            for change in change_list:
+                print(('done with %s' % (change)))
+                cl.cl_done(change)
+            time.sleep(interval)
+    except OSError:
+        ex = sys.exc_info()[1]
+        print(ex)
+
+if __name__ == '__main__':
+    if len(sys.argv) != 6:
+        print(("usage: %s <brick> <scratch-dir> <log-file> <fetch-interval>"
+              % (sys.argv[0])))
+        sys.exit(1)
+    get_changes(sys.argv[1], sys.argv[2], sys.argv[3], 9, int(sys.argv[4]))
diff --git a/xlators/features/changelog/lib/examples/python/libgfchangelog.py b/xlators/features/changelog/lib/examples/python/libgfchangelog.py
new file mode 100644
index 00000000000..2da9f2d2a8c
--- /dev/null
+++ b/xlators/features/changelog/lib/examples/python/libgfchangelog.py
@@ -0,0 +1,71 @@
+import os
+from ctypes import *
+from ctypes.util import find_library
+
+class Changes(object):
+    libgfc = CDLL(find_library("gfchangelog"), mode=RTLD_GLOBAL,
+                  use_errno=True)
+
+    @classmethod
+    def geterrno(cls):
+        return get_errno()
+
+    @classmethod
+    def raise_oserr(cls):
+        errn = cls.geterrno()
+        raise OSError(errn, os.strerror(errn))
+
+    @classmethod
+    def _get_api(cls, call):
+        return getattr(cls.libgfc, call)
+
+    @classmethod
+    def cl_init(cls):
+        ret = cls._get_api('gf_changelog_init')(None)
+        if ret == -1:
+            cls.raise_changelog_err()
+
+    @classmethod
+    def cl_register(cls, brick, path, log_file, log_level, retries = 0):
+        ret = cls._get_api('gf_changelog_register')(brick, path,
+                                                    log_file, log_level, retries)
+        if ret == -1:
+            cls.raise_oserr()
+
+    @classmethod
+    def cl_scan(cls):
+        ret = cls._get_api('gf_changelog_scan')()
+        if ret == -1:
+            cls.raise_oserr()
+
+    @classmethod
+    def cl_startfresh(cls):
+        ret = cls._get_api('gf_changelog_start_fresh')()
+        if ret == -1:
+            cls.raise_oserr()
+
+    @classmethod
+    def cl_getchanges(cls):
+        """ remove hardcoding for path name length """
+        def clsort(f):
+            return f.split('.')[-1]
+        changes = []
+        buf = create_string_buffer('\0', 4096)
+        call = cls._get_api('gf_changelog_next_change')
+
+        while True:
+            ret = call(buf, 4096)
+            if ret in (0, -1):
+                break;
+            changes.append(buf.raw[:ret-1])
+        if ret == -1:
+            cls.raise_oserr()
+        # cleanup tracker
+        cls.cl_startfresh()
+        return sorted(changes, key=clsort)
+
+    @classmethod
+    def cl_done(cls, clfile):
+        ret = cls._get_api('gf_changelog_done')(clfile)
+        if ret == -1:
+            cls.raise_oserr()
diff --git a/xlators/features/changelog/lib/src/Makefile.am b/xlators/features/changelog/lib/src/Makefile.am
new file mode 100644
index 00000000000..c933ec53ed2
--- /dev/null
+++ b/xlators/features/changelog/lib/src/Makefile.am
@@ -0,0 +1,35 @@
+libgfchangelog_la_CFLAGS = -Wall $(GF_CFLAGS) $(GF_DARWIN_LIBGLUSTERFS_CFLAGS) \
+	-DDATADIR=\"$(localstatedir)\"
+
+libgfchangelog_la_CPPFLAGS = $(GF_CPPFLAGS) -D__USE_FILE_OFFSET64 -D__USE_LARGEFILE64 -fpic \
+	-I../../../src/ -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/xlators/features/changelog/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src \
+	-I$(top_srcdir)/rpc/rpc-lib/src \
+	-I$(top_srcdir)/rpc/rpc-transport/socket/src \
+	-DDATADIR=\"$(localstatedir)\"
+
+libgfchangelog_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la \
+	$(top_builddir)/rpc/xdr/src/libgfxdr.la \
+	$(top_builddir)/rpc/rpc-lib/src/libgfrpc.la
+
+libgfchangelog_la_LDFLAGS = $(GF_LDFLAGS) \
+        -version-info $(LIBGFCHANGELOG_LT_VERSION) \
+        $(GF_NO_UNDEFINED)
+
+lib_LTLIBRARIES = libgfchangelog.la
+
+CONTRIB_BUILDDIR = $(top_builddir)/contrib
+
+libgfchangelog_la_SOURCES = gf-changelog.c gf-changelog-journal-handler.c \
+	gf-changelog-helpers.c gf-changelog-api.c gf-history-changelog.c \
+	gf-changelog-rpc.c gf-changelog-reborp.c \
+	$(top_srcdir)/xlators/features/changelog/src/changelog-rpc-common.c
+
+noinst_HEADERS = gf-changelog-helpers.h gf-changelog-rpc.h \
+	gf-changelog-journal.h changelog-lib-messages.h
+
+CLEANFILES =
+
+$(top_builddir)/libglusterfs/src/libglusterfs.la:
+	$(MAKE) -C $(top_builddir)/libglusterfs/src/ all
diff --git a/xlators/features/changelog/lib/src/changelog-lib-messages.h b/xlators/features/changelog/lib/src/changelog-lib-messages.h
new file mode 100644
index 00000000000..d7fe7274353
--- /dev/null
+++ b/xlators/features/changelog/lib/src/changelog-lib-messages.h
@@ -0,0 +1,74 @@
+/*
+ Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+ This file is part of GlusterFS.
+
+ This file is licensed to you under your choice of the GNU Lesser
+ General Public License, version 3 or any later version (LGPLv3 or
+ later), or the GNU General Public License, version 2 (GPLv2), in all
+ cases as published by the Free Software Foundation.
+ */
+
+#ifndef _CHANGELOG_LIB_MESSAGES_H_
+#define _CHANGELOG_LIB_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(
+    CHANGELOG_LIB, CHANGELOG_LIB_MSG_OPEN_FAILED,
+    CHANGELOG_LIB_MSG_FAILED_TO_RMDIR,
+    CHANGELOG_LIB_MSG_SCRATCH_DIR_ENTRIES_CREATION_ERROR,
+    CHANGELOG_LIB_MSG_THREAD_CREATION_FAILED, CHANGELOG_LIB_MSG_OPENDIR_ERROR,
+    CHANGELOG_LIB_MSG_RENAME_FAILED, CHANGELOG_LIB_MSG_READ_ERROR,
+    CHANGELOG_LIB_MSG_HTIME_ERROR, CHANGELOG_LIB_MSG_GET_TIME_ERROR,
+    CHANGELOG_LIB_MSG_WRITE_FAILED, CHANGELOG_LIB_MSG_PTHREAD_ERROR,
+    CHANGELOG_LIB_MSG_MMAP_FAILED, CHANGELOG_LIB_MSG_MUNMAP_FAILED,
+    CHANGELOG_LIB_MSG_ASCII_ERROR, CHANGELOG_LIB_MSG_STAT_FAILED,
+    CHANGELOG_LIB_MSG_GET_XATTR_FAILED, CHANGELOG_LIB_MSG_PUBLISH_ERROR,
+    CHANGELOG_LIB_MSG_PARSE_ERROR, CHANGELOG_LIB_MSG_MIN_MAX_INFO,
+    CHANGELOG_LIB_MSG_CLEANUP_ERROR, CHANGELOG_LIB_MSG_UNLINK_FAILED,
+    CHANGELOG_LIB_MSG_NOTIFY_REGISTER_FAILED,
+    CHANGELOG_LIB_MSG_INVOKE_RPC_FAILED, CHANGELOG_LIB_MSG_DRAINING_EVENT_INFO,
+    CHANGELOG_LIB_MSG_CLEANING_BRICK_ENTRY_INFO,
+    CHANGELOG_LIB_MSG_FREEING_ENTRY_INFO, CHANGELOG_LIB_MSG_XDR_DECODING_FAILED,
+    CHANGELOG_LIB_MSG_NOTIFY_REGISTER_INFO,
+    CHANGELOG_LIB_MSG_THREAD_CLEANUP_WARNING,
+    CHANGELOG_LIB_MSG_COPY_FROM_BUFFER_FAILED,
+    CHANGELOG_LIB_MSG_PTHREAD_JOIN_FAILED, CHANGELOG_LIB_MSG_HIST_FAILED,
+    CHANGELOG_LIB_MSG_DRAINED_EVENT_INFO, CHANGELOG_LIB_MSG_PARSE_ERROR_CEASED,
+    CHANGELOG_LIB_MSG_REQUESTING_INFO, CHANGELOG_LIB_MSG_FINAL_INFO);
+
+#define CHANGELOG_LIB_MSG_NOTIFY_REGISTER_INFO_STR "Registering brick"
+#define CHANGELOG_LIB_MSG_RENAME_FAILED_STR "error moving changelog file"
+#define CHANGELOG_LIB_MSG_OPEN_FAILED_STR "cannot open changelog file"
+#define CHANGELOG_LIB_MSG_UNLINK_FAILED_STR "failed to unlink"
+#define CHANGELOG_LIB_MSG_FAILED_TO_RMDIR_STR "failed to rmdir"
+#define CHANGELOG_LIB_MSG_STAT_FAILED_STR "stat failed on changelog file"
+#define CHANGELOG_LIB_MSG_PARSE_ERROR_STR "could not parse changelog"
+#define CHANGELOG_LIB_MSG_PARSE_ERROR_CEASED_STR                               \
+    "parsing error, ceased publishing..."
+#define CHANGELOG_LIB_MSG_HTIME_ERROR_STR "fop failed on htime file"
+#define CHANGELOG_LIB_MSG_GET_XATTR_FAILED_STR                                 \
+    "error extracting max timstamp from htime file"
+#define CHANGELOG_LIB_MSG_MIN_MAX_INFO_STR "changelogs min max"
+#define CHANGELOG_LIB_MSG_REQUESTING_INFO_STR "Requesting historical changelogs"
+#define CHANGELOG_LIB_MSG_FINAL_INFO_STR "FINAL"
+#define CHANGELOG_LIB_MSG_HIST_FAILED_STR                                      \
+    "Requested changelog range is not available"
+#define CHANGELOG_LIB_MSG_GET_TIME_ERROR_STR "wrong result"
+#define CHANGELOG_LIB_MSG_CLEANING_BRICK_ENTRY_INFO_STR                        \
+    "Cleaning brick entry for brick"
+#define CHANGELOG_LIB_MSG_DRAINING_EVENT_INFO_STR "Draining event"
+#define CHANGELOG_LIB_MSG_DRAINED_EVENT_INFO_STR "Drained event"
+#define CHANGELOG_LIB_MSG_FREEING_ENTRY_INFO_STR "freeing entry"
+
+#endif /* !_CHANGELOG_MESSAGES_H_ */
diff --git a/xlators/features/changelog/lib/src/gf-changelog-api.c b/xlators/features/changelog/lib/src/gf-changelog-api.c
new file mode 100644
index 00000000000..81a5cbfec10
--- /dev/null
+++ b/xlators/features/changelog/lib/src/gf-changelog-api.c
@@ -0,0 +1,224 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/compat-uuid.h>
+#include <glusterfs/globals.h>
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/syscall.h>
+
+#include "gf-changelog-helpers.h"
+#include "gf-changelog-journal.h"
+#include "changelog-mem-types.h"
+#include "changelog-lib-messages.h"
+
+int
+gf_changelog_done(char *file)
+{
+    int ret = -1;
+    char *buffer = NULL;
+    xlator_t *this = NULL;
+    gf_changelog_journal_t *jnl = NULL;
+    char to_path[PATH_MAX] = {
+        0,
+    };
+
+    errno = EINVAL;
+
+    this = THIS;
+    if (!this)
+        goto out;
+
+    jnl = (gf_changelog_journal_t *)GF_CHANGELOG_GET_API_PTR(this);
+    if (!jnl)
+        goto out;
+
+    if (!file || !strlen(file))
+        goto out;
+
+    /* make sure 'file' is inside ->jnl_working_dir */
+    buffer = realpath(file, NULL);
+    if (!buffer)
+        goto out;
+
+    if (strncmp(jnl->jnl_working_dir, buffer, strlen(jnl->jnl_working_dir)))
+        goto out;
+
+    (void)snprintf(to_path, PATH_MAX, "%s%s", jnl->jnl_processed_dir,
+                   basename(buffer));
+    gf_msg_debug(this->name, 0, "moving %s to processed directory", file);
+    ret = sys_rename(buffer, to_path);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                CHANGELOG_LIB_MSG_RENAME_FAILED, "from=%s", file, "to=%s",
+                to_path, NULL);
+        goto out;
+    }
+
+    ret = 0;
+
+out:
+    if (buffer)
+        free(buffer); /* allocated by realpath() */
+    return ret;
+}
+
+/**
+ * @API
+ *  for a set of changelogs, start from the beginning
+ */
+int
+gf_changelog_start_fresh()
+{
+    xlator_t *this = NULL;
+    gf_changelog_journal_t *jnl = NULL;
+
+    this = THIS;
+    if (!this)
+        goto out;
+
+    errno = EINVAL;
+
+    jnl = (gf_changelog_journal_t *)GF_CHANGELOG_GET_API_PTR(this);
+    if (!jnl)
+        goto out;
+
+    if (gf_ftruncate(jnl->jnl_fd, 0))
+        goto out;
+
+    return 0;
+
+out:
+    return -1;
+}
+
+/**
+ * @API
+ * return the next changelog file entry. zero means all chanelogs
+ * consumed.
+ */
+ssize_t
+gf_changelog_next_change(char *bufptr, size_t maxlen)
+{
+    ssize_t size = -1;
+    int tracker_fd = 0;
+    xlator_t *this = NULL;
+    gf_changelog_journal_t *jnl = NULL;
+    char buffer[PATH_MAX] = {
+        0,
+    };
+
+    errno = EINVAL;
+
+    this = THIS;
+    if (!this)
+        goto out;
+
+    jnl = (gf_changelog_journal_t *)GF_CHANGELOG_GET_API_PTR(this);
+    if (!jnl)
+        goto out;
+
+    tracker_fd = jnl->jnl_fd;
+
+    size = gf_readline(tracker_fd, buffer, maxlen);
+    if (size < 0) {
+        size = -1;
+        goto out;
+    }
+
+    if (size == 0)
+        goto out;
+
+    memcpy(bufptr, buffer, size - 1);
+    bufptr[size - 1] = '\0';
+
+out:
+    return size;
+}
+
+/**
+ * @API
+ *  gf_changelog_scan() - scan and generate a list of change entries
+ *
+ * calling this api multiple times (without calling gf_changlog_done())
+ * would result new changelogs(s) being refreshed in the tracker file.
+ * This call also acts as a cancellation point for the consumer.
+ */
+ssize_t
+gf_changelog_scan()
+{
+    int tracker_fd = 0;
+    size_t off = 0;
+    xlator_t *this = NULL;
+    size_t nr_entries = 0;
+    gf_changelog_journal_t *jnl = NULL;
+    struct dirent *entry = NULL;
+    struct dirent scratch[2] = {
+        {
+            0,
+        },
+    };
+    char buffer[PATH_MAX] = {
+        0,
+    };
+
+    this = THIS;
+    if (!this)
+        goto out;
+
+    jnl = (gf_changelog_journal_t *)GF_CHANGELOG_GET_API_PTR(this);
+    if (!jnl)
+        goto out;
+    if (JNL_IS_API_DISCONNECTED(jnl)) {
+        errno = ENOTCONN;
+        goto out;
+    }
+
+    errno = EINVAL;
+
+    tracker_fd = jnl->jnl_fd;
+    if (gf_ftruncate(tracker_fd, 0))
+        goto out;
+
+    rewinddir(jnl->jnl_dir);
+
+    for (;;) {
+        errno = 0;
+        entry = sys_readdir(jnl->jnl_dir, scratch);
+        if (!entry || errno != 0)
+            break;
+
+        if (!strcmp(basename(entry->d_name), ".") ||
+            !strcmp(basename(entry->d_name), ".."))
+            continue;
+
+        nr_entries++;
+
+        GF_CHANGELOG_FILL_BUFFER(jnl->jnl_processing_dir, buffer, off,
+                                 strlen(jnl->jnl_processing_dir));
+        GF_CHANGELOG_FILL_BUFFER(entry->d_name, buffer, off,
+                                 strlen(entry->d_name));
+        GF_CHANGELOG_FILL_BUFFER("\n", buffer, off, 1);
+
+        if (gf_changelog_write(tracker_fd, buffer, off) != off) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, CHANGELOG_LIB_MSG_WRITE_FAILED,
+                   "error writing changelog filename"
+                   " to tracker file");
+            break;
+        }
+        off = 0;
+    }
+
+    if (!entry) {
+        if (gf_lseek(tracker_fd, 0, SEEK_SET) != -1)
+            return nr_entries;
+    }
+out:
+    return -1;
+}
diff --git a/xlators/features/changelog/lib/src/gf-changelog-helpers.c b/xlators/features/changelog/lib/src/gf-changelog-helpers.c
new file mode 100644
index 00000000000..75f8a6dfc08
--- /dev/null
+++ b/xlators/features/changelog/lib/src/gf-changelog-helpers.c
@@ -0,0 +1,170 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include "changelog-mem-types.h"
+#include "gf-changelog-helpers.h"
+#include "changelog-lib-messages.h"
+#include <glusterfs/syscall.h>
+
+size_t
+gf_changelog_write(int fd, char *buffer, size_t len)
+{
+    ssize_t size = 0;
+    size_t written = 0;
+
+    while (written < len) {
+        size = sys_write(fd, buffer + written, len - written);
+        if (size <= 0)
+            break;
+
+        written += size;
+    }
+
+    return written;
+}
+
+void
+gf_rfc3986_encode_space_newline(unsigned char *s, char *enc, char *estr)
+{
+    for (; *s; s++) {
+        if (estr[*s])
+            sprintf(enc, "%c", estr[*s]);
+        else
+            sprintf(enc, "%%%02X", *s);
+        while (*++enc)
+            ;
+    }
+}
+
+/**
+ * thread safe version of readline with buffering
+ * (taken from Unix Network Programming Volume I, W.R. Stevens)
+ *
+ * This is favoured over fgets() as we'd need to ftruncate()
+ * (see gf_changelog_scan() API) to record new changelog files.
+ * stream open functions does have a truncate like api (although
+ * that can be done via @fflush(fp), @ftruncate(fd) and @fseek(fp),
+ * but this involves mixing POSIX file descriptors and stream FILE *).
+ *
+ * NOTE: This implementation still does work with more than one fd's
+ *       used to perform gf_readline(). For this very reason it's not
+ *       made a part of libglusterfs.
+ */
+
+static __thread read_line_t thread_tsd = {};
+
+static ssize_t
+my_read(read_line_t *tsd, int fd, char *ptr)
+{
+    if (tsd->rl_cnt <= 0) {
+        tsd->rl_cnt = sys_read(fd, tsd->rl_buf, MAXLINE);
+
+        if (tsd->rl_cnt < 0)
+            return -1;
+        else if (tsd->rl_cnt == 0)
+            return 0;
+        tsd->rl_bufptr = tsd->rl_buf;
+    }
+
+    tsd->rl_cnt--;
+    *ptr = *tsd->rl_bufptr++;
+    return 1;
+}
+
+ssize_t
+gf_readline(int fd, void *vptr, size_t maxlen)
+{
+    size_t n = 0;
+    size_t rc = 0;
+    char c = ' ';
+    char *ptr = NULL;
+    read_line_t *tsd = &thread_tsd;
+
+    ptr = vptr;
+    for (n = 1; n < maxlen; n++) {
+        if ((rc = my_read(tsd, fd, &c)) == 1) {
+            *ptr++ = c;
+            if (c == '\n')
+                break;
+        } else if (rc == 0) {
+            *ptr = '\0';
+            return (n - 1);
+        } else
+            return -1;
+    }
+
+    *ptr = '\0';
+    return n;
+}
+
+off_t
+gf_lseek(int fd, off_t offset, int whence)
+{
+    off_t off = 0;
+    read_line_t *tsd = &thread_tsd;
+
+    off = sys_lseek(fd, offset, whence);
+    if (off == -1)
+        return -1;
+
+    tsd->rl_cnt = 0;
+    tsd->rl_bufptr = tsd->rl_buf;
+
+    return off;
+}
+
+int
+gf_ftruncate(int fd, off_t length)
+{
+    read_line_t *tsd = &thread_tsd;
+
+    if (sys_ftruncate(fd, 0))
+        return -1;
+
+    tsd->rl_cnt = 0;
+    tsd->rl_bufptr = tsd->rl_buf;
+
+    return 0;
+}
+
+int
+gf_thread_cleanup(xlator_t *this, pthread_t thread)
+{
+    int ret = 0;
+    void *res = NULL;
+
+    ret = pthread_cancel(thread);
+    if (ret != 0) {
+        gf_msg(this->name, GF_LOG_WARNING, 0,
+               CHANGELOG_LIB_MSG_THREAD_CLEANUP_WARNING,
+               "Failed to send cancellation to thread");
+        goto error_return;
+    }
+
+    ret = pthread_join(thread, &res);
+    if (ret != 0) {
+        gf_msg(this->name, GF_LOG_WARNING, 0,
+               CHANGELOG_LIB_MSG_THREAD_CLEANUP_WARNING,
+               "failed to join thread");
+        goto error_return;
+    }
+
+    if (res != PTHREAD_CANCELED) {
+        gf_msg(this->name, GF_LOG_WARNING, 0,
+               CHANGELOG_LIB_MSG_THREAD_CLEANUP_WARNING,
+               "Thread could not be cleaned up");
+        goto error_return;
+    }
+
+    return 0;
+
+error_return:
+    return -1;
+}
diff --git a/xlators/features/changelog/lib/src/gf-changelog-helpers.h b/xlators/features/changelog/lib/src/gf-changelog-helpers.h
new file mode 100644
index 00000000000..9c609d33172
--- /dev/null
+++ b/xlators/features/changelog/lib/src/gf-changelog-helpers.h
@@ -0,0 +1,255 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef _GF_CHANGELOG_HELPERS_H
+#define _GF_CHANGELOG_HELPERS_H
+
+#include <unistd.h>
+#include <dirent.h>
+#include <limits.h>
+#include <glusterfs/locking.h>
+
+#include <glusterfs/xlator.h>
+
+#include "changelog.h"
+
+#include "changelog-rpc-common.h"
+#include "gf-changelog-journal.h"
+
+#define GF_CHANGELOG_TRACKER "tracker"
+
+#define GF_CHANGELOG_CURRENT_DIR ".current"
+#define GF_CHANGELOG_PROCESSED_DIR ".processed"
+#define GF_CHANGELOG_PROCESSING_DIR ".processing"
+#define GF_CHANGELOG_HISTORY_DIR ".history"
+#define TIMESTAMP_LENGTH 10
+
+#ifndef MAXLINE
+#define MAXLINE 4096
+#endif
+
+#define GF_CHANGELOG_FILL_BUFFER(ptr, ascii, off, len)                         \
+    do {                                                                       \
+        memcpy(ascii + off, ptr, len);                                         \
+        off += len;                                                            \
+    } while (0)
+
+typedef struct read_line {
+    int rl_cnt;
+    char *rl_bufptr;
+    char rl_buf[MAXLINE];
+} read_line_t;
+
+struct gf_changelog;
+struct gf_event;
+
+/**
+ * Event list for ordered event notification
+ *
+ * ->next_seq holds the next _expected_ sequence number.
+ */
+struct gf_event_list {
+    pthread_mutex_t lock; /* protects this structure */
+    pthread_cond_t cond;
+
+    pthread_t invoker;
+
+    unsigned long next_seq; /* next sequence number expected:
+                               zero during bootstrap */
+
+    struct gf_changelog *entry; /* backpointer to it's brick
+                                   encapsulator (entry) */
+    struct list_head events;    /* list of events */
+};
+
+/**
+ * include a refcount if it's of use by additional layers
+ */
+struct gf_event {
+    int count;
+
+    unsigned long seq;
+
+    struct list_head list;
+
+    struct iovec iov[0];
+};
+#define GF_EVENT_CALLOC_SIZE(cnt, len)                                         \
+    (sizeof(struct gf_event) + (cnt * sizeof(struct iovec)) + len)
+
+/**
+ * assign the base address of the IO vector to the correct memory
+o * area and set it's addressable length.
+ */
+#define GF_EVENT_ASSIGN_IOVEC(vec, event, len, pos)                            \
+    do {                                                                       \
+        vec->iov_base = ((char *)event) + sizeof(struct gf_event) +            \
+                        (event->count * sizeof(struct iovec)) + pos;           \
+        vec->iov_len = len;                                                    \
+        pos += len;                                                            \
+    } while (0)
+
+typedef enum gf_changelog_conn_state {
+    GF_CHANGELOG_CONN_STATE_PENDING = 0,
+    GF_CHANGELOG_CONN_STATE_ACCEPTED,
+    GF_CHANGELOG_CONN_STATE_DISCONNECTED,
+} gf_changelog_conn_state_t;
+
+/**
+ * An instance of this structure is allocated for each brick for which
+ * notifications are streamed.
+ */
+typedef struct gf_changelog {
+    gf_lock_t statelock;
+    gf_changelog_conn_state_t connstate;
+
+    xlator_t *this;
+
+    struct list_head list; /* list of instances */
+
+    char brick[PATH_MAX]; /* brick path for this end-point */
+
+    changelog_rpc_t grpc; /* rpc{-clnt,svc} for this brick */
+#define RPC_PROBER(ent) ent->grpc.rpc
+#define RPC_REBORP(ent) ent->grpc.svc
+#define RPC_SOCK(ent) ent->grpc.sock
+
+    unsigned int notify; /* notification flag(s) */
+
+    FINI *fini;               /* destructor callback */
+    CALLBACK *callback;       /* event callback dispatcher */
+    CONNECT *connected;       /* connect callback */
+    DISCONNECT *disconnected; /* disconnection callback */
+
+    void *ptr;           /* owner specific private data */
+    xlator_t *invokerxl; /* consumers _this_, if valid,
+                            assigned to THIS before cbk is
+                            invoked */
+
+    gf_boolean_t ordered;
+
+    void (*queueevent)(struct gf_event_list *, struct gf_event *);
+    void (*pickevent)(struct gf_event_list *, struct gf_event **);
+
+    struct gf_event_list event;
+} gf_changelog_t;
+
+static inline int
+gf_changelog_filter_check(gf_changelog_t *entry, changelog_event_t *event)
+{
+    if (event->ev_type & entry->notify)
+        return 1;
+    return 0;
+}
+
+#define GF_NEED_ORDERED_EVENTS(ent) (ent->ordered == _gf_true)
+
+/** private structure */
+typedef struct gf_private {
+    pthread_mutex_t lock; /* protects ->connections, cleanups */
+    pthread_cond_t cond;
+
+    void *api; /* pointer for API access */
+
+    pthread_t poller;            /* event poller thread */
+    pthread_t connectionjanitor; /* connection cleaner */
+
+    struct list_head connections; /* list of connections */
+    struct list_head cleanups;    /* list of connection to be
+                                     cleaned up */
+} gf_private_t;
+
+#define GF_CHANGELOG_GET_API_PTR(this) (((gf_private_t *)this->private)->api)
+
+/**
+ * upcall: invoke callback with _correct_ THIS
+ */
+#define GF_CHANGELOG_INVOKE_CBK(this, cbk, brick, args...)                     \
+    do {                                                                       \
+        xlator_t *old_this = NULL;                                             \
+        xlator_t *invokerxl = NULL;                                            \
+                                                                               \
+        invokerxl = entry->invokerxl;                                          \
+        old_this = this;                                                       \
+                                                                               \
+        if (invokerxl) {                                                       \
+            THIS = invokerxl;                                                  \
+        }                                                                      \
+                                                                               \
+        cbk(invokerxl, brick, args);                                           \
+        THIS = old_this;                                                       \
+                                                                               \
+    } while (0)
+
+#define SAVE_THIS(xl)                                                          \
+    do {                                                                       \
+        old_this = xl;                                                         \
+        THIS = master;                                                         \
+    } while (0)
+
+#define RESTORE_THIS()                                                         \
+    do {                                                                       \
+        if (old_this)                                                          \
+            THIS = old_this;                                                   \
+    } while (0)
+
+/** APIs and the rest */
+
+void *
+gf_changelog_process(void *data);
+
+void
+gf_rfc3986_encode_space_newline(unsigned char *s, char *enc, char *estr);
+
+size_t
+gf_changelog_write(int fd, char *buffer, size_t len);
+
+ssize_t
+gf_readline(int fd, void *vptr, size_t maxlen);
+
+int
+gf_ftruncate(int fd, off_t length);
+
+off_t
+gf_lseek(int fd, off_t offset, int whence);
+
+int
+gf_changelog_consume(xlator_t *this, gf_changelog_journal_t *jnl,
+                     char *from_path, gf_boolean_t no_publish);
+int
+gf_changelog_publish(xlator_t *this, gf_changelog_journal_t *jnl,
+                     char *from_path);
+int
+gf_thread_cleanup(xlator_t *this, pthread_t thread);
+void *
+gf_changelog_callback_invoker(void *arg);
+
+int
+gf_cleanup_event(xlator_t *, struct gf_event_list *);
+
+/* (un)ordered event queueing */
+void
+queue_ordered_event(struct gf_event_list *, struct gf_event *);
+
+void
+queue_unordered_event(struct gf_event_list *, struct gf_event *);
+
+/* (un)ordered event picking */
+void
+pick_event_ordered(struct gf_event_list *, struct gf_event **);
+
+void
+pick_event_unordered(struct gf_event_list *, struct gf_event **);
+
+/* connection janitor thread */
+void *
+gf_changelog_connection_janitor(void *);
+
+#endif
diff --git a/xlators/features/changelog/lib/src/gf-changelog-journal-handler.c b/xlators/features/changelog/lib/src/gf-changelog-journal-handler.c
new file mode 100644
index 00000000000..7f6e2329e71
--- /dev/null
+++ b/xlators/features/changelog/lib/src/gf-changelog-journal-handler.c
@@ -0,0 +1,1029 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/compat-uuid.h>
+#include <glusterfs/globals.h>
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/syscall.h>
+#include <glusterfs/compat-errno.h>
+
+#include "gf-changelog-helpers.h"
+
+/* from the changelog translator */
+#include "changelog-misc.h"
+#include "changelog-mem-types.h"
+
+#include "gf-changelog-journal.h"
+#include "changelog-lib-messages.h"
+
+extern int byebye;
+
+enum changelog_versions { VERSION_1_1 = 0, VERSION_1_2 = 1 };
+
+/**
+ * number of gfid records after fop number
+ */
+int nr_gfids[2][GF_FOP_MAXVALUE] = {{
+                                        [GF_FOP_MKNOD] = 1,
+                                        [GF_FOP_MKDIR] = 1,
+                                        [GF_FOP_UNLINK] = 1,
+                                        [GF_FOP_RMDIR] = 1,
+                                        [GF_FOP_SYMLINK] = 1,
+                                        [GF_FOP_RENAME] = 2,
+                                        [GF_FOP_LINK] = 1,
+                                        [GF_FOP_CREATE] = 1,
+                                    },
+                                    {
+                                        [GF_FOP_MKNOD] = 1,
+                                        [GF_FOP_MKDIR] = 1,
+                                        [GF_FOP_UNLINK] = 2,
+                                        [GF_FOP_RMDIR] = 2,
+                                        [GF_FOP_SYMLINK] = 1,
+                                        [GF_FOP_RENAME] = 2,
+                                        [GF_FOP_LINK] = 1,
+                                        [GF_FOP_CREATE] = 1,
+                                    }};
+
+int nr_extra_recs[2][GF_FOP_MAXVALUE] = {{
+                                             [GF_FOP_MKNOD] = 3,
+                                             [GF_FOP_MKDIR] = 3,
+                                             [GF_FOP_UNLINK] = 0,
+                                             [GF_FOP_RMDIR] = 0,
+                                             [GF_FOP_SYMLINK] = 0,
+                                             [GF_FOP_RENAME] = 0,
+                                             [GF_FOP_LINK] = 0,
+                                             [GF_FOP_CREATE] = 3,
+                                         },
+                                         {
+                                             [GF_FOP_MKNOD] = 3,
+                                             [GF_FOP_MKDIR] = 3,
+                                             [GF_FOP_UNLINK] = 0,
+                                             [GF_FOP_RMDIR] = 0,
+                                             [GF_FOP_SYMLINK] = 0,
+                                             [GF_FOP_RENAME] = 0,
+                                             [GF_FOP_LINK] = 0,
+                                             [GF_FOP_CREATE] = 3,
+                                         }};
+
+static char *
+binary_to_ascii(uuid_t uuid)
+{
+    return uuid_utoa(uuid);
+}
+
+static char *
+conv_noop(char *ptr)
+{
+    return ptr;
+}
+
+#define VERIFY_SEPARATOR(ptr, plen, perr)                                      \
+    {                                                                          \
+        if (*(ptr + plen) != '\0') {                                           \
+            perr = 1;                                                          \
+            break;                                                             \
+        }                                                                      \
+    }
+
+#define MOVER_MOVE(mover, nleft, bytes)                                        \
+    {                                                                          \
+        mover += bytes;                                                        \
+        nleft -= bytes;                                                        \
+    }
+
+#define PARSE_GFID(mov, ptr, le, fn, perr)                                     \
+    {                                                                          \
+        VERIFY_SEPARATOR(mov, le, perr);                                       \
+        ptr = fn(mov);                                                         \
+        if (!ptr) {                                                            \
+            perr = 1;                                                          \
+            break;                                                             \
+        }                                                                      \
+    }
+
+#define FILL_AND_MOVE(pt, buf, of, mo, nl, le)                                 \
+    {                                                                          \
+        GF_CHANGELOG_FILL_BUFFER(pt, buf, of, strlen(pt));                     \
+        MOVER_MOVE(mo, nl, le);                                                \
+    }
+
+#define PARSE_GFID_MOVE(ptr, uuid, mover, nleft, perr)                         \
+    {                                                                          \
+        memcpy(uuid, mover, sizeof(uuid_t));                                   \
+        ptr = binary_to_ascii(uuid);                                           \
+        if (!ptr) {                                                            \
+            perr = 1;                                                          \
+            break;                                                             \
+        }                                                                      \
+        MOVER_MOVE(mover, nleft, sizeof(uuid_t));                              \
+    }
+
+#define LINE_BUFSIZE (3 * PATH_MAX) /* enough buffer for extra chars too */
+
+/**
+ * using mmap() makes parsing easy. fgets() cannot be used here as
+ * the binary gfid could contain a line-feed (0x0A), in that case fgets()
+ * would read an incomplete line and parsing would fail. using POSIX fds
+ * would result is additional code to maintain state in case of partial
+ * reads of data (where multiple entries do not fit extirely in the buffer).
+ *
+ * mmap() gives the flexibility of pointing to an offset in the file
+ * without us worrying about reading it in memory (VM does that for us for
+ * free).
+ */
+
+static int
+gf_changelog_parse_binary(xlator_t *this, gf_changelog_journal_t *jnl,
+                          int from_fd, int to_fd, size_t start_offset,
+                          struct stat *stbuf, int version_idx)
+
+{
+    int ret = -1;
+    off_t off = 0;
+    off_t nleft = 0;
+    uuid_t uuid = {
+        0,
+    };
+    char *ptr = NULL;
+    char *bname_start = NULL;
+    char *bname_end = NULL;
+    char *mover = NULL;
+    void *start = NULL;
+    char current_mover = ' ';
+    size_t blen = 0;
+    int parse_err = 0;
+    char *ascii = NULL;
+
+    ascii = GF_CALLOC(LINE_BUFSIZE, sizeof(char), gf_common_mt_char);
+
+    nleft = stbuf->st_size;
+
+    start = mmap(NULL, nleft, PROT_READ, MAP_PRIVATE, from_fd, 0);
+    if (start == MAP_FAILED) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, CHANGELOG_LIB_MSG_MMAP_FAILED,
+               "mmap() error");
+        goto out;
+    }
+
+    mover = start;
+
+    MOVER_MOVE(mover, nleft, start_offset);
+
+    while (nleft > 0) {
+        off = blen = 0;
+        ptr = bname_start = bname_end = NULL;
+
+        current_mover = *mover;
+
+        switch (current_mover) {
+            case 'D':
+            case 'M':
+                MOVER_MOVE(mover, nleft, 1);
+                PARSE_GFID_MOVE(ptr, uuid, mover, nleft, parse_err);
+
+                break;
+
+            case 'E':
+                MOVER_MOVE(mover, nleft, 1);
+                PARSE_GFID_MOVE(ptr, uuid, mover, nleft, parse_err);
+
+                bname_start = mover;
+                bname_end = strchr(mover, '\n');
+                if (bname_end == NULL) {
+                    parse_err = 1;
+                    break;
+                }
+
+                blen = bname_end - bname_start;
+                MOVER_MOVE(mover, nleft, blen);
+
+                break;
+
+            default:
+                parse_err = 1;
+        }
+
+        if (parse_err)
+            break;
+
+        GF_CHANGELOG_FILL_BUFFER(&current_mover, ascii, off, 1);
+        GF_CHANGELOG_FILL_BUFFER(" ", ascii, off, 1);
+        GF_CHANGELOG_FILL_BUFFER(ptr, ascii, off, strlen(ptr));
+        if (blen)
+            GF_CHANGELOG_FILL_BUFFER(bname_start, ascii, off, blen);
+        GF_CHANGELOG_FILL_BUFFER("\n", ascii, off, 1);
+
+        if (gf_changelog_write(to_fd, ascii, off) != off) {
+            gf_msg(this->name, GF_LOG_ERROR, errno,
+                   CHANGELOG_LIB_MSG_ASCII_ERROR,
+                   "processing binary changelog failed due to "
+                   " error in writing ascii change");
+            break;
+        }
+
+        MOVER_MOVE(mover, nleft, 1);
+    }
+
+    if ((nleft == 0) && (!parse_err))
+        ret = 0;
+
+    if (munmap(start, stbuf->st_size))
+        gf_msg(this->name, GF_LOG_ERROR, errno, CHANGELOG_LIB_MSG_MUNMAP_FAILED,
+               "munmap() error");
+out:
+    if (ascii)
+        GF_FREE(ascii);
+    return ret;
+}
+
+/**
+ * ascii decoder:
+ *  - separate out one entry from another
+ *  - use fop name rather than fop number
+ */
+static int
+gf_changelog_parse_ascii(xlator_t *this, gf_changelog_journal_t *jnl,
+                         int from_fd, int to_fd, size_t start_offset,
+                         struct stat *stbuf, int version_idx)
+{
+    int ng = 0;
+    int ret = -1;
+    int fop = 0;
+    int len = 0;
+    off_t off = 0;
+    off_t nleft = 0;
+    char *ptr = NULL;
+    char *eptr = NULL;
+    void *start = NULL;
+    char *mover = NULL;
+    int parse_err = 0;
+    char current_mover = ' ';
+    char *ascii = NULL;
+    const char *fopname = NULL;
+
+    ascii = GF_CALLOC(LINE_BUFSIZE, sizeof(char), gf_common_mt_char);
+
+    nleft = stbuf->st_size;
+
+    start = mmap(NULL, nleft, PROT_READ, MAP_PRIVATE, from_fd, 0);
+    if (start == MAP_FAILED) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, CHANGELOG_LIB_MSG_MMAP_FAILED,
+               "mmap() error");
+        goto out;
+    }
+
+    mover = start;
+
+    MOVER_MOVE(mover, nleft, start_offset);
+
+    while (nleft > 0) {
+        off = 0;
+        current_mover = *mover;
+
+        GF_CHANGELOG_FILL_BUFFER(&current_mover, ascii, off, 1);
+        GF_CHANGELOG_FILL_BUFFER(" ", ascii, off, 1);
+
+        switch (current_mover) {
+            case 'D':
+                MOVER_MOVE(mover, nleft, 1);
+
+                /* target gfid */
+                PARSE_GFID(mover, ptr, UUID_CANONICAL_FORM_LEN, conv_noop,
+                           parse_err);
+                FILL_AND_MOVE(ptr, ascii, off, mover, nleft,
+                              UUID_CANONICAL_FORM_LEN);
+                break;
+            case 'M':
+                MOVER_MOVE(mover, nleft, 1);
+
+                /* target gfid */
+                PARSE_GFID(mover, ptr, UUID_CANONICAL_FORM_LEN, conv_noop,
+                           parse_err);
+                FILL_AND_MOVE(ptr, ascii, off, mover, nleft,
+                              UUID_CANONICAL_FORM_LEN);
+                FILL_AND_MOVE(" ", ascii, off, mover, nleft, 1);
+
+                /* fop */
+                len = strlen(mover);
+                VERIFY_SEPARATOR(mover, len, parse_err);
+
+                fop = atoi(mover);
+                fopname = gf_fop_list[fop];
+                if (fopname == NULL) {
+                    parse_err = 1;
+                    break;
+                }
+
+                MOVER_MOVE(mover, nleft, len);
+
+                len = strlen(fopname);
+                GF_CHANGELOG_FILL_BUFFER(fopname, ascii, off, len);
+
+                break;
+
+            case 'E':
+                MOVER_MOVE(mover, nleft, 1);
+
+                /* target gfid */
+                PARSE_GFID(mover, ptr, UUID_CANONICAL_FORM_LEN, conv_noop,
+                           parse_err);
+                FILL_AND_MOVE(ptr, ascii, off, mover, nleft,
+                              UUID_CANONICAL_FORM_LEN);
+                FILL_AND_MOVE(" ", ascii, off, mover, nleft, 1);
+
+                /* fop */
+                len = strlen(mover);
+                VERIFY_SEPARATOR(mover, len, parse_err);
+
+                fop = atoi(mover);
+                fopname = gf_fop_list[fop];
+                if (fopname == NULL) {
+                    parse_err = 1;
+                    break;
+                }
+
+                MOVER_MOVE(mover, nleft, len);
+
+                len = strlen(fopname);
+                GF_CHANGELOG_FILL_BUFFER(fopname, ascii, off, len);
+
+                ng = nr_extra_recs[version_idx][fop];
+                for (; ng > 0; ng--) {
+                    MOVER_MOVE(mover, nleft, 1);
+                    len = strlen(mover);
+                    VERIFY_SEPARATOR(mover, len, parse_err);
+
+                    GF_CHANGELOG_FILL_BUFFER(" ", ascii, off, 1);
+                    FILL_AND_MOVE(mover, ascii, off, mover, nleft, len);
+                }
+
+                /* pargfid + bname */
+                ng = nr_gfids[version_idx][fop];
+                while (ng-- > 0) {
+                    MOVER_MOVE(mover, nleft, 1);
+                    len = strlen(mover);
+                    if (!len) {
+                        MOVER_MOVE(mover, nleft, 1);
+                        continue;
+                    }
+
+                    GF_CHANGELOG_FILL_BUFFER(" ", ascii, off, 1);
+
+                    PARSE_GFID(mover, ptr, len, conv_noop, parse_err);
+                    eptr = calloc(3, strlen(ptr));
+                    if (!eptr) {
+                        parse_err = 1;
+                        break;
+                    }
+
+                    gf_rfc3986_encode_space_newline((unsigned char *)ptr, eptr,
+                                                    jnl->rfc3986_space_newline);
+                    FILL_AND_MOVE(eptr, ascii, off, mover, nleft, len);
+                    free(eptr);
+                }
+
+                break;
+            default:
+                parse_err = 1;
+        }
+
+        if (parse_err)
+            break;
+
+        GF_CHANGELOG_FILL_BUFFER("\n", ascii, off, 1);
+
+        if (gf_changelog_write(to_fd, ascii, off) != off) {
+            gf_msg(this->name, GF_LOG_ERROR, errno,
+                   CHANGELOG_LIB_MSG_ASCII_ERROR,
+                   "processing ascii changelog failed due to "
+                   " error in writing change");
+            break;
+        }
+
+        MOVER_MOVE(mover, nleft, 1);
+    }
+
+    if ((nleft == 0) && (!parse_err))
+        ret = 0;
+
+    if (munmap(start, stbuf->st_size))
+        gf_msg(this->name, GF_LOG_ERROR, errno, CHANGELOG_LIB_MSG_MUNMAP_FAILED,
+               "munmap() error");
+
+out:
+    if (ascii)
+        GF_FREE(ascii);
+
+    return ret;
+}
+
+static int
+gf_changelog_decode(xlator_t *this, gf_changelog_journal_t *jnl, int from_fd,
+                    int to_fd, struct stat *stbuf, int *zerob)
+{
+    int ret = -1;
+    int encoding = -1;
+    int major_version = -1;
+    int minor_version = -1;
+    int version_idx = -1;
+    size_t elen = 0;
+    char buffer[1024] = {
+        0,
+    };
+
+    CHANGELOG_GET_HEADER_INFO(from_fd, buffer, sizeof(buffer), encoding,
+                              major_version, minor_version, elen);
+    if (encoding == -1) /* unknown encoding */
+        goto out;
+
+    if (major_version == -1) /* unknown major version */
+        goto out;
+
+    if (minor_version == -1) /* unknown minor version */
+        goto out;
+
+    if (!CHANGELOG_VALID_ENCODING(encoding))
+        goto out;
+
+    if (elen == stbuf->st_size) {
+        *zerob = 1;
+        goto out;
+    }
+
+    if (major_version == 1 && minor_version == 1) {
+        version_idx = VERSION_1_1;
+    } else if (major_version == 1 && minor_version == 2) {
+        version_idx = VERSION_1_2;
+    }
+
+    if (version_idx == -1) /* unknown version number */
+        goto out;
+
+    /**
+     * start processing after the header
+     */
+    if (sys_lseek(from_fd, elen, SEEK_SET) < 0) {
+        goto out;
+    }
+    switch (encoding) {
+        case CHANGELOG_ENCODE_BINARY:
+            /**
+             * this ideally should have been a part of changelog-encoders.c
+             * (ie. part of the changelog translator).
+             */
+            ret = gf_changelog_parse_binary(this, jnl, from_fd, to_fd, elen,
+                                            stbuf, version_idx);
+            break;
+
+        case CHANGELOG_ENCODE_ASCII:
+            ret = gf_changelog_parse_ascii(this, jnl, from_fd, to_fd, elen,
+                                           stbuf, version_idx);
+            break;
+    }
+
+out:
+    return ret;
+}
+
+int
+gf_changelog_publish(xlator_t *this, gf_changelog_journal_t *jnl,
+                     char *from_path)
+{
+    int ret = 0;
+    char dest[PATH_MAX] = {
+        0,
+    };
+    char to_path[PATH_MAX] = {
+        0,
+    };
+    struct stat stbuf = {
+        0,
+    };
+
+    if (snprintf(to_path, PATH_MAX, "%s%s", jnl->jnl_current_dir,
+                 basename(from_path)) >= PATH_MAX)
+        return -1;
+
+    /* handle zerob file that won't exist in current */
+    ret = sys_stat(to_path, &stbuf);
+    if (ret) {
+        if (errno == ENOENT)
+            ret = 0;
+        goto out;
+    }
+
+    if (snprintf(dest, PATH_MAX, "%s%s", jnl->jnl_processing_dir,
+                 basename(from_path)) >= PATH_MAX)
+        return -1;
+
+    ret = sys_rename(to_path, dest);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                CHANGELOG_LIB_MSG_RENAME_FAILED, "from=%s", to_path, "to=%s",
+                dest, NULL);
+    }
+
+out:
+    return ret;
+}
+
+int
+gf_changelog_consume(xlator_t *this, gf_changelog_journal_t *jnl,
+                     char *from_path, gf_boolean_t no_publish)
+{
+    int ret = -1;
+    int fd1 = 0;
+    int fd2 = 0;
+    int zerob = 0;
+    struct stat stbuf = {
+        0,
+    };
+    char dest[PATH_MAX] = {
+        0,
+    };
+    char to_path[PATH_MAX] = {
+        0,
+    };
+
+    if (snprintf(to_path, PATH_MAX, "%s%s", jnl->jnl_current_dir,
+                 basename(from_path)) >= PATH_MAX)
+        goto out;
+    if (snprintf(dest, PATH_MAX, "%s%s", jnl->jnl_processing_dir,
+                 basename(from_path)) >= PATH_MAX)
+        goto out;
+
+    ret = sys_stat(from_path, &stbuf);
+    if (ret || !S_ISREG(stbuf.st_mode)) {
+        ret = -1;
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_LIB_MSG_STAT_FAILED,
+                "path=%s", from_path, NULL);
+        goto out;
+    }
+
+    fd1 = open(from_path, O_RDONLY);
+    if (fd1 < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_LIB_MSG_OPEN_FAILED,
+                "path=%s", from_path, NULL);
+        goto out;
+    }
+
+    fd2 = open(to_path, O_CREAT | O_TRUNC | O_RDWR,
+               S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
+    if (fd2 < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_LIB_MSG_OPEN_FAILED,
+                "path=%s", to_path, NULL);
+        goto close_fd;
+    } else {
+        ret = gf_changelog_decode(this, jnl, fd1, fd2, &stbuf, &zerob);
+
+        sys_close(fd2);
+
+        if (!ret) {
+            /* move it to processing on a successful
+               decode */
+            if (no_publish == _gf_true)
+                goto close_fd;
+            ret = sys_rename(to_path, dest);
+            if (ret)
+                gf_smsg(this->name, GF_LOG_ERROR, errno,
+                        CHANGELOG_LIB_MSG_RENAME_FAILED, "from=%s", to_path,
+                        "to=%s", dest, NULL);
+        }
+
+        /* remove it from .current if it's an empty file */
+        if (zerob) {
+            /* zerob changelogs must be unlinked */
+            ret = sys_unlink(to_path);
+            if (ret)
+                gf_smsg(this->name, GF_LOG_ERROR, errno,
+                        CHANGELOG_LIB_MSG_UNLINK_FAILED, "name=empty changelog",
+                        "path=%s", to_path, NULL);
+        }
+    }
+
+close_fd:
+    sys_close(fd1);
+
+out:
+    return ret;
+}
+
+void *
+gf_changelog_process(void *data)
+{
+    xlator_t *this = NULL;
+    gf_changelog_journal_t *jnl = NULL;
+    gf_changelog_entry_t *entry = NULL;
+    gf_changelog_processor_t *jnl_proc = NULL;
+
+    jnl = data;
+    jnl_proc = jnl->jnl_proc;
+    THIS = jnl->this;
+    this = jnl->this;
+
+    while (1) {
+        pthread_mutex_lock(&jnl_proc->lock);
+        {
+            while (list_empty(&jnl_proc->entries)) {
+                jnl_proc->waiting = _gf_true;
+                pthread_cond_wait(&jnl_proc->cond, &jnl_proc->lock);
+            }
+
+            entry = list_first_entry(&jnl_proc->entries, gf_changelog_entry_t,
+                                     list);
+            if (entry)
+                list_del(&entry->list);
+
+            jnl_proc->waiting = _gf_false;
+        }
+        pthread_mutex_unlock(&jnl_proc->lock);
+
+        if (entry) {
+            (void)gf_changelog_consume(this, jnl, entry->path, _gf_false);
+            GF_FREE(entry);
+        }
+    }
+
+    return NULL;
+}
+
+void
+gf_changelog_queue_journal(gf_changelog_processor_t *jnl_proc,
+                           changelog_event_t *event)
+{
+    size_t len = 0;
+    gf_changelog_entry_t *entry = NULL;
+
+    entry = GF_CALLOC(1, sizeof(gf_changelog_entry_t),
+                      gf_changelog_mt_libgfchangelog_entry_t);
+    if (!entry)
+        return;
+    INIT_LIST_HEAD(&entry->list);
+
+    len = strlen(event->u.journal.path);
+    (void)memcpy(entry->path, event->u.journal.path, len + 1);
+    entry->path[len] = '\0';
+
+    pthread_mutex_lock(&jnl_proc->lock);
+    {
+        list_add_tail(&entry->list, &jnl_proc->entries);
+        if (jnl_proc->waiting)
+            pthread_cond_signal(&jnl_proc->cond);
+    }
+    pthread_mutex_unlock(&jnl_proc->lock);
+
+    return;
+}
+
+void
+gf_changelog_handle_journal(void *xl, char *brick, void *cbkdata,
+                            changelog_event_t *event)
+{
+    gf_changelog_journal_t *jnl = NULL;
+    gf_changelog_processor_t *jnl_proc = NULL;
+
+    jnl = cbkdata;
+    jnl_proc = jnl->jnl_proc;
+
+    gf_changelog_queue_journal(jnl_proc, event);
+}
+
+void
+gf_changelog_journal_disconnect(void *xl, char *brick, void *data)
+{
+    gf_changelog_journal_t *jnl = NULL;
+
+    jnl = data;
+
+    pthread_spin_lock(&jnl->lock);
+    {
+        JNL_SET_API_STATE(jnl, JNL_API_DISCONNECTED);
+    };
+    pthread_spin_unlock(&jnl->lock);
+}
+
+void
+gf_changelog_journal_connect(void *xl, char *brick, void *data)
+{
+    gf_changelog_journal_t *jnl = NULL;
+
+    jnl = data;
+
+    pthread_spin_lock(&jnl->lock);
+    {
+        JNL_SET_API_STATE(jnl, JNL_API_CONNECTED);
+    };
+    pthread_spin_unlock(&jnl->lock);
+
+    return;
+}
+
+void
+gf_changelog_cleanup_processor(gf_changelog_journal_t *jnl)
+{
+    int ret = 0;
+    xlator_t *this = NULL;
+    gf_changelog_processor_t *jnl_proc = NULL;
+
+    this = THIS;
+    if (!this || !jnl || !jnl->jnl_proc)
+        goto error_return;
+
+    jnl_proc = jnl->jnl_proc;
+
+    ret = gf_thread_cleanup(this, jnl_proc->processor);
+    if (ret != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, CHANGELOG_LIB_MSG_CLEANUP_ERROR,
+               "failed to cleanup processor thread");
+        goto error_return;
+    }
+
+    (void)pthread_mutex_destroy(&jnl_proc->lock);
+    (void)pthread_cond_destroy(&jnl_proc->cond);
+
+    GF_FREE(jnl_proc);
+
+error_return:
+    return;
+}
+
+int
+gf_changelog_init_processor(gf_changelog_journal_t *jnl)
+{
+    int ret = -1;
+    gf_changelog_processor_t *jnl_proc = NULL;
+
+    jnl_proc = GF_CALLOC(1, sizeof(gf_changelog_processor_t),
+                         gf_changelog_mt_libgfchangelog_t);
+    if (!jnl_proc)
+        goto error_return;
+
+    ret = pthread_mutex_init(&jnl_proc->lock, NULL);
+    if (ret != 0)
+        goto free_jnl_proc;
+    ret = pthread_cond_init(&jnl_proc->cond, NULL);
+    if (ret != 0)
+        goto cleanup_mutex;
+
+    INIT_LIST_HEAD(&jnl_proc->entries);
+    jnl_proc->waiting = _gf_false;
+    jnl->jnl_proc = jnl_proc;
+
+    ret = gf_thread_create(&jnl_proc->processor, NULL, gf_changelog_process,
+                           jnl, "clogproc");
+    if (ret != 0) {
+        jnl->jnl_proc = NULL;
+        goto cleanup_cond;
+    }
+
+    return 0;
+
+cleanup_cond:
+    (void)pthread_cond_destroy(&jnl_proc->cond);
+cleanup_mutex:
+    (void)pthread_mutex_destroy(&jnl_proc->lock);
+free_jnl_proc:
+    GF_FREE(jnl_proc);
+error_return:
+    return -1;
+}
+
+static void
+gf_changelog_cleanup_fds(gf_changelog_journal_t *jnl)
+{
+    /* tracker fd */
+    if (jnl->jnl_fd != -1)
+        sys_close(jnl->jnl_fd);
+    /* processing dir */
+    if (jnl->jnl_dir)
+        sys_closedir(jnl->jnl_dir);
+
+    if (jnl->jnl_working_dir)
+        free(jnl->jnl_working_dir); /* allocated by realpath */
+}
+
+static int
+gf_changelog_open_dirs(xlator_t *this, gf_changelog_journal_t *jnl)
+{
+    int ret = -1;
+    DIR *dir = NULL;
+    int tracker_fd = 0;
+    char tracker_path[PATH_MAX] = {
+        0,
+    };
+
+    /* .current */
+    (void)snprintf(jnl->jnl_current_dir, PATH_MAX,
+                   "%s/" GF_CHANGELOG_CURRENT_DIR "/", jnl->jnl_working_dir);
+    ret = recursive_rmdir(jnl->jnl_current_dir);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                CHANGELOG_LIB_MSG_FAILED_TO_RMDIR, "path=%s",
+                jnl->jnl_current_dir, NULL);
+        goto out;
+    }
+    ret = mkdir_p(jnl->jnl_current_dir, 0600, _gf_false);
+    if (ret)
+        goto out;
+
+    /* .processed */
+    (void)snprintf(jnl->jnl_processed_dir, PATH_MAX,
+                   "%s/" GF_CHANGELOG_PROCESSED_DIR "/", jnl->jnl_working_dir);
+    ret = mkdir_p(jnl->jnl_processed_dir, 0600, _gf_false);
+    if (ret)
+        goto out;
+
+    /* .processing */
+    (void)snprintf(jnl->jnl_processing_dir, PATH_MAX,
+                   "%s/" GF_CHANGELOG_PROCESSING_DIR "/", jnl->jnl_working_dir);
+    ret = recursive_rmdir(jnl->jnl_processing_dir);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                CHANGELOG_LIB_MSG_FAILED_TO_RMDIR, "path=%s",
+                jnl->jnl_processing_dir, NULL);
+        goto out;
+    }
+
+    ret = mkdir_p(jnl->jnl_processing_dir, 0600, _gf_false);
+    if (ret)
+        goto out;
+
+    dir = sys_opendir(jnl->jnl_processing_dir);
+    if (!dir) {
+        gf_msg("", GF_LOG_ERROR, errno, CHANGELOG_LIB_MSG_OPENDIR_ERROR,
+               "opendir() error");
+        goto out;
+    }
+
+    jnl->jnl_dir = dir;
+
+    (void)snprintf(tracker_path, PATH_MAX, "%s/" GF_CHANGELOG_TRACKER,
+                   jnl->jnl_working_dir);
+
+    tracker_fd = open(tracker_path, O_CREAT | O_APPEND | O_RDWR,
+                      S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
+    if (tracker_fd < 0) {
+        sys_closedir(jnl->jnl_dir);
+        ret = -1;
+        goto out;
+    }
+
+    jnl->jnl_fd = tracker_fd;
+    ret = 0;
+out:
+    return ret;
+}
+
+int
+gf_changelog_init_history(xlator_t *this, gf_changelog_journal_t *jnl,
+                          char *brick_path)
+{
+    int i = 0;
+    int ret = 0;
+    char hist_scratch_dir[PATH_MAX] = {
+        0,
+    };
+
+    jnl->hist_jnl = GF_CALLOC(1, sizeof(*jnl),
+                              gf_changelog_mt_libgfchangelog_t);
+    if (!jnl->hist_jnl)
+        goto error_return;
+
+    jnl->hist_jnl->jnl_dir = NULL;
+    jnl->hist_jnl->jnl_fd = -1;
+
+    (void)snprintf(hist_scratch_dir, PATH_MAX,
+                   "%s/" GF_CHANGELOG_HISTORY_DIR "/", jnl->jnl_working_dir);
+
+    ret = mkdir_p(hist_scratch_dir, 0600, _gf_false);
+    if (ret)
+        goto dealloc_hist;
+
+    jnl->hist_jnl->jnl_working_dir = realpath(hist_scratch_dir, NULL);
+    if (!jnl->hist_jnl->jnl_working_dir)
+        goto dealloc_hist;
+
+    ret = gf_changelog_open_dirs(this, jnl->hist_jnl);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, CHANGELOG_LIB_MSG_OPENDIR_ERROR,
+               "could not create entries in history scratch dir");
+        goto dealloc_hist;
+    }
+
+    if (snprintf(jnl->hist_jnl->jnl_brickpath, PATH_MAX, "%s", brick_path) >=
+        PATH_MAX)
+        goto dealloc_hist;
+
+    for (i = 0; i < 256; i++) {
+        jnl->hist_jnl->rfc3986_space_newline[i] = (i == ' ' || i == '\n' ||
+                                                   i == '%')
+                                                      ? 0
+                                                      : i;
+    }
+
+    return 0;
+
+dealloc_hist:
+    GF_FREE(jnl->hist_jnl);
+    jnl->hist_jnl = NULL;
+error_return:
+    return -1;
+}
+
+void
+gf_changelog_journal_fini(void *xl, char *brick, void *data)
+{
+    gf_changelog_journal_t *jnl = NULL;
+
+    jnl = data;
+
+    gf_changelog_cleanup_processor(jnl);
+
+    gf_changelog_cleanup_fds(jnl);
+    if (jnl->hist_jnl)
+        gf_changelog_cleanup_fds(jnl->hist_jnl);
+
+    GF_FREE(jnl);
+}
+
+void *
+gf_changelog_journal_init(void *xl, struct gf_brick_spec *brick)
+{
+    int i = 0;
+    int ret = 0;
+    xlator_t *this = NULL;
+    struct stat buf = {
+        0,
+    };
+    char *scratch_dir = NULL;
+    gf_changelog_journal_t *jnl = NULL;
+
+    this = xl;
+    scratch_dir = (char *)brick->ptr;
+
+    jnl = GF_CALLOC(1, sizeof(gf_changelog_journal_t),
+                    gf_changelog_mt_libgfchangelog_t);
+    if (!jnl)
+        goto error_return;
+
+    if (snprintf(jnl->jnl_brickpath, PATH_MAX, "%s", brick->brick_path) >=
+        PATH_MAX)
+        goto dealloc_private;
+
+    if (sys_stat(scratch_dir, &buf) && errno == ENOENT) {
+        ret = mkdir_p(scratch_dir, 0600, _gf_true);
+        if (ret)
+            goto dealloc_private;
+    }
+
+    jnl->jnl_working_dir = realpath(scratch_dir, NULL);
+    if (!jnl->jnl_working_dir)
+        goto dealloc_private;
+
+    ret = gf_changelog_open_dirs(this, jnl);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, CHANGELOG_LIB_MSG_OPENDIR_ERROR,
+               "could not create entries in scratch dir");
+        goto dealloc_private;
+    }
+
+    /* RFC 3986 {de,en}coding */
+    for (i = 0; i < 256; i++) {
+        jnl->rfc3986_space_newline[i] = (i == ' ' || i == '\n' || i == '%') ? 0
+                                                                            : i;
+    }
+
+    ret = gf_changelog_init_history(this, jnl, brick->brick_path);
+    if (ret)
+        goto cleanup_fds;
+
+    /* initialize journal processor */
+    jnl->this = this;
+    ret = gf_changelog_init_processor(jnl);
+    if (ret)
+        goto cleanup_fds;
+
+    JNL_SET_API_STATE(jnl, JNL_API_CONN_INPROGESS);
+    ret = pthread_spin_init(&jnl->lock, 0);
+    if (ret != 0)
+        goto cleanup_processor;
+    return jnl;
+
+cleanup_processor:
+    gf_changelog_cleanup_processor(jnl);
+cleanup_fds:
+    gf_changelog_cleanup_fds(jnl);
+    if (jnl->hist_jnl)
+        gf_changelog_cleanup_fds(jnl->hist_jnl);
+dealloc_private:
+    GF_FREE(jnl);
+error_return:
+    return NULL;
+}
diff --git a/xlators/features/changelog/lib/src/gf-changelog-journal.h b/xlators/features/changelog/lib/src/gf-changelog-journal.h
new file mode 100644
index 00000000000..ba5b9bf827e
--- /dev/null
+++ b/xlators/features/changelog/lib/src/gf-changelog-journal.h
@@ -0,0 +1,116 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __GF_CHANGELOG_JOURNAL_H
+#define __GF_CHANGELOG_JOURNAL_H
+
+#include <unistd.h>
+#include <pthread.h>
+
+#include "changelog.h"
+
+enum api_conn {
+    JNL_API_CONNECTED,
+    JNL_API_CONN_INPROGESS,
+    JNL_API_DISCONNECTED,
+};
+
+typedef struct gf_changelog_entry {
+    char path[PATH_MAX];
+
+    struct list_head list;
+} gf_changelog_entry_t;
+
+typedef struct gf_changelog_processor {
+    pthread_mutex_t lock; /* protects ->entries */
+    pthread_cond_t cond;  /* waiter during empty list */
+    gf_boolean_t waiting;
+
+    pthread_t processor; /* thread-id of journal processing thread */
+
+    struct list_head entries;
+} gf_changelog_processor_t;
+
+typedef struct gf_changelog_journal {
+    DIR *jnl_dir; /* 'processing' directory stream */
+
+    int jnl_fd; /* fd to the tracker file */
+
+    char jnl_brickpath[PATH_MAX]; /* brick path for this end-point */
+
+    gf_changelog_processor_t *jnl_proc;
+
+    char *jnl_working_dir; /* scratch directory */
+
+    char jnl_current_dir[PATH_MAX];
+    char jnl_processed_dir[PATH_MAX];
+    char jnl_processing_dir[PATH_MAX];
+
+    char rfc3986_space_newline[256]; /* RFC 3986 string encoding */
+
+    struct gf_changelog_journal *hist_jnl;
+    int hist_done; /* holds 0 done scanning,
+                      1 keep scanning and -1 error */
+
+    pthread_spinlock_t lock;
+    int connected;
+    xlator_t *this;
+} gf_changelog_journal_t;
+
+#define JNL_SET_API_STATE(jnl, state) (jnl->connected = state)
+#define JNL_IS_API_DISCONNECTED(jnl) (jnl->connected == JNL_API_DISCONNECTED)
+
+/* History API */
+typedef struct gf_changelog_history_data {
+    int len;
+
+    int htime_fd;
+
+    /* parallelism count */
+    int n_parallel;
+
+    /* history from, to indexes */
+    unsigned long from;
+    unsigned long to;
+    xlator_t *this;
+} gf_changelog_history_data_t;
+
+typedef struct gf_changelog_consume_data {
+    /** set of inputs */
+
+    /* fd to read from */
+    int fd;
+
+    /* from @offset */
+    off_t offset;
+
+    xlator_t *this;
+
+    gf_changelog_journal_t *jnl;
+
+    /** set of outputs */
+
+    /* return value */
+    int retval;
+
+    /* journal processed */
+    char changelog[PATH_MAX];
+} gf_changelog_consume_data_t;
+
+/* event handler */
+CALLBACK gf_changelog_handle_journal;
+
+/* init, connect & disconnect handler */
+INIT gf_changelog_journal_init;
+FINI gf_changelog_journal_fini;
+CONNECT gf_changelog_journal_connect;
+DISCONNECT gf_changelog_journal_disconnect;
+
+#endif
diff --git a/xlators/features/changelog/lib/src/gf-changelog-reborp.c b/xlators/features/changelog/lib/src/gf-changelog-reborp.c
new file mode 100644
index 00000000000..56b11cbb705
--- /dev/null
+++ b/xlators/features/changelog/lib/src/gf-changelog-reborp.c
@@ -0,0 +1,413 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include "changelog-misc.h"
+#include "changelog-mem-types.h"
+
+#include "gf-changelog-helpers.h"
+#include "changelog-rpc-common.h"
+#include "changelog-lib-messages.h"
+
+#include <glusterfs/syscall.h>
+
+/**
+ * Reverse socket: actual data transfer handler. Connection
+ * initiator is PROBER, data transfer is REBORP.
+ */
+
+static struct rpcsvc_program *gf_changelog_reborp_programs[];
+
+void *
+gf_changelog_connection_janitor(void *arg)
+{
+    int32_t ret = 0;
+    xlator_t *this = NULL;
+    gf_private_t *priv = NULL;
+    gf_changelog_t *entry = NULL;
+    struct gf_event *event = NULL;
+    struct gf_event_list *ev = NULL;
+    unsigned long drained = 0;
+
+    this = arg;
+    THIS = this;
+
+    priv = this->private;
+
+    while (1) {
+        pthread_mutex_lock(&priv->lock);
+        {
+            while (list_empty(&priv->cleanups))
+                pthread_cond_wait(&priv->cond, &priv->lock);
+
+            entry = list_first_entry(&priv->cleanups, gf_changelog_t, list);
+            list_del_init(&entry->list);
+        }
+        pthread_mutex_unlock(&priv->lock);
+
+        drained = 0;
+        ev = &entry->event;
+
+        gf_smsg(this->name, GF_LOG_INFO, 0,
+                CHANGELOG_LIB_MSG_CLEANING_BRICK_ENTRY_INFO, "brick=%s",
+                entry->brick, NULL);
+
+        /* 0x0: disable rpc-clnt */
+        rpc_clnt_disable(RPC_PROBER(entry));
+
+        /* 0x1: cleanup callback invoker thread */
+        ret = gf_cleanup_event(this, ev);
+        if (ret)
+            continue;
+
+        /* 0x2: drain pending events */
+        while (!list_empty(&ev->events)) {
+            event = list_first_entry(&ev->events, struct gf_event, list);
+            gf_smsg(this->name, GF_LOG_INFO, 0,
+                    CHANGELOG_LIB_MSG_DRAINING_EVENT_INFO, "seq=%lu",
+                    event->seq, "payload=%d", event->count, NULL);
+
+            GF_FREE(event);
+            drained++;
+        }
+
+        gf_smsg(this->name, GF_LOG_INFO, 0,
+                CHANGELOG_LIB_MSG_DRAINED_EVENT_INFO, "num=%lu", drained, NULL);
+
+        /* 0x3: freeup brick entry */
+        gf_smsg(this->name, GF_LOG_INFO, 0,
+                CHANGELOG_LIB_MSG_FREEING_ENTRY_INFO, "entry=%p", entry, NULL);
+        LOCK_DESTROY(&entry->statelock);
+        GF_FREE(entry);
+    }
+
+    return NULL;
+}
+
+int
+gf_changelog_reborp_rpcsvc_notify(rpcsvc_t *rpc, void *mydata,
+                                  rpcsvc_event_t event, void *data)
+{
+    int ret = 0;
+    xlator_t *this = NULL;
+    gf_changelog_t *entry = NULL;
+
+    if (!(event == RPCSVC_EVENT_ACCEPT || event == RPCSVC_EVENT_DISCONNECT))
+        return 0;
+
+    entry = mydata;
+    this = entry->this;
+
+    switch (event) {
+        case RPCSVC_EVENT_ACCEPT:
+            ret = sys_unlink(RPC_SOCK(entry));
+            if (ret != 0)
+                gf_smsg(this->name, GF_LOG_WARNING, errno,
+                        CHANGELOG_LIB_MSG_UNLINK_FAILED, "name=reverse socket",
+                        "path=%s", RPC_SOCK(entry), NULL);
+            if (entry->connected)
+                GF_CHANGELOG_INVOKE_CBK(this, entry->connected, entry->brick,
+                                        entry->ptr);
+            break;
+        case RPCSVC_EVENT_DISCONNECT:
+            if (entry->disconnected)
+                GF_CHANGELOG_INVOKE_CBK(this, entry->disconnected, entry->brick,
+                                        entry->ptr);
+            /* passthrough */
+        default:
+            break;
+    }
+
+    return 0;
+}
+
+rpcsvc_t *
+gf_changelog_reborp_init_rpc_listner(xlator_t *this, char *path, char *sock,
+                                     void *cbkdata)
+{
+    CHANGELOG_MAKE_TMP_SOCKET_PATH(path, sock, UNIX_PATH_MAX);
+    return changelog_rpc_server_init(this, sock, cbkdata,
+                                     gf_changelog_reborp_rpcsvc_notify,
+                                     gf_changelog_reborp_programs);
+}
+
+/**
+ * This is dirty and painful as of now until there is event filtering in the
+ * server. The entire event buffer is scanned and interested events are picked,
+ * whereas we _should_ be notified with the events we were interested in
+ * (selected at the time of probe). As of now this is complete BS and needs
+ * fixture ASAP. I just made it work, it needs to be better.
+ *
+ * @FIXME: cleanup this bugger once server filters events.
+ */
+void
+gf_changelog_invoke_callback(gf_changelog_t *entry, struct iovec **vec,
+                             int payloadcnt)
+{
+    int i = 0;
+    int evsize = 0;
+    xlator_t *this = NULL;
+    changelog_event_t *event = NULL;
+
+    this = entry->this;
+
+    for (; i < payloadcnt; i++) {
+        event = (changelog_event_t *)vec[i]->iov_base;
+        evsize = vec[i]->iov_len / CHANGELOG_EV_SIZE;
+
+        for (; evsize > 0; evsize--, event++) {
+            if (gf_changelog_filter_check(entry, event)) {
+                GF_CHANGELOG_INVOKE_CBK(this, entry->callback, entry->brick,
+                                        entry->ptr, event);
+            }
+        }
+    }
+}
+
+/**
+ * Ordered event handler is self-adaptive.. if the event sequence number
+ * is what's expected (->next_seq) there is no ordering list that's
+ * maintained. On out-of-order event notifications, event buffers are
+ * dynamically allocated and ordered.
+ */
+
+int
+__is_expected_sequence(struct gf_event_list *ev, struct gf_event *event)
+{
+    return (ev->next_seq == event->seq);
+}
+
+int
+__can_process_event(struct gf_event_list *ev, struct gf_event **event)
+{
+    *event = list_first_entry(&ev->events, struct gf_event, list);
+
+    if (__is_expected_sequence(ev, *event)) {
+        list_del(&(*event)->list);
+        ev->next_seq++;
+        return 1;
+    }
+
+    return 0;
+}
+
+void
+pick_event_ordered(struct gf_event_list *ev, struct gf_event **event)
+{
+    pthread_mutex_lock(&ev->lock);
+    {
+        while (list_empty(&ev->events) || !__can_process_event(ev, event))
+            pthread_cond_wait(&ev->cond, &ev->lock);
+    }
+    pthread_mutex_unlock(&ev->lock);
+}
+
+void
+pick_event_unordered(struct gf_event_list *ev, struct gf_event **event)
+{
+    pthread_mutex_lock(&ev->lock);
+    {
+        while (list_empty(&ev->events))
+            pthread_cond_wait(&ev->cond, &ev->lock);
+        *event = list_first_entry(&ev->events, struct gf_event, list);
+        list_del(&(*event)->list);
+    }
+    pthread_mutex_unlock(&ev->lock);
+}
+
+void *
+gf_changelog_callback_invoker(void *arg)
+{
+    xlator_t *this = NULL;
+    gf_changelog_t *entry = NULL;
+    struct iovec *vec = NULL;
+    struct gf_event *event = NULL;
+    struct gf_event_list *ev = NULL;
+
+    ev = arg;
+    entry = ev->entry;
+    THIS = this = entry->this;
+
+    while (1) {
+        entry->pickevent(ev, &event);
+
+        vec = (struct iovec *)&event->iov;
+        gf_changelog_invoke_callback(entry, &vec, event->count);
+
+        GF_FREE(event);
+    }
+
+    return NULL;
+}
+
+static int
+orderfn(struct list_head *pos1, struct list_head *pos2)
+{
+    struct gf_event *event1 = NULL;
+    struct gf_event *event2 = NULL;
+
+    event1 = list_entry(pos1, struct gf_event, list);
+    event2 = list_entry(pos2, struct gf_event, list);
+
+    if (event1->seq > event2->seq)
+        return 1;
+    return -1;
+}
+
+void
+queue_ordered_event(struct gf_event_list *ev, struct gf_event *event)
+{
+    /* add event to the ordered event list and wake up listener(s) */
+    pthread_mutex_lock(&ev->lock);
+    {
+        list_add_order(&event->list, &ev->events, orderfn);
+        if (!ev->next_seq)
+            ev->next_seq = event->seq;
+        if (ev->next_seq == event->seq)
+            pthread_cond_signal(&ev->cond);
+    }
+    pthread_mutex_unlock(&ev->lock);
+}
+
+void
+queue_unordered_event(struct gf_event_list *ev, struct gf_event *event)
+{
+    /* add event to the tail of the queue and wake up listener(s) */
+    pthread_mutex_lock(&ev->lock);
+    {
+        list_add_tail(&event->list, &ev->events);
+        pthread_cond_signal(&ev->cond);
+    }
+    pthread_mutex_unlock(&ev->lock);
+}
+
+int
+gf_changelog_event_handler(rpcsvc_request_t *req, xlator_t *this,
+                           gf_changelog_t *entry)
+{
+    int i = 0;
+    size_t payloadlen = 0;
+    ssize_t len = 0;
+    int payloadcnt = 0;
+    changelog_event_req rpc_req = {
+        0,
+    };
+    changelog_event_rsp rpc_rsp = {
+        0,
+    };
+    struct iovec *vec = NULL;
+    struct gf_event *event = NULL;
+    struct gf_event_list *ev = NULL;
+
+    ev = &entry->event;
+
+    len = xdr_to_generic(req->msg[0], &rpc_req,
+                         (xdrproc_t)xdr_changelog_event_req);
+    if (len < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0,
+               CHANGELOG_LIB_MSG_XDR_DECODING_FAILED, "xdr decoding failed");
+        req->rpc_err = GARBAGE_ARGS;
+        goto handle_xdr_error;
+    }
+
+    if (len < req->msg[0].iov_len) {
+        payloadcnt = 1;
+        payloadlen = (req->msg[0].iov_len - len);
+    }
+    for (i = 1; i < req->count; i++) {
+        payloadcnt++;
+        payloadlen += req->msg[i].iov_len;
+    }
+
+    event = GF_CALLOC(1, GF_EVENT_CALLOC_SIZE(payloadcnt, payloadlen),
+                      gf_changelog_mt_libgfchangelog_event_t);
+    if (!event)
+        goto handle_xdr_error;
+    INIT_LIST_HEAD(&event->list);
+
+    payloadlen = 0;
+    event->seq = rpc_req.seq;
+    event->count = payloadcnt;
+
+    /* deep copy IO vectors */
+    vec = &event->iov[0];
+    GF_EVENT_ASSIGN_IOVEC(vec, event, (req->msg[0].iov_len - len), payloadlen);
+    (void)memcpy(vec->iov_base, req->msg[0].iov_base + len, vec->iov_len);
+
+    for (i = 1; i < req->count; i++) {
+        vec = &event->iov[i];
+        GF_EVENT_ASSIGN_IOVEC(vec, event, req->msg[i].iov_len, payloadlen);
+        (void)memcpy(event->iov[i].iov_base, req->msg[i].iov_base,
+                     req->msg[i].iov_len);
+    }
+
+    gf_msg_debug(this->name, 0,
+                 "seq: %" PRIu64 " [%s] (time: %" PRIu64 ".%" PRIu64
+                 "), "
+                 "(vec: %d, len: %zd)",
+                 rpc_req.seq, entry->brick, rpc_req.tv_sec, rpc_req.tv_usec,
+                 payloadcnt, payloadlen);
+
+    /* dispatch event */
+    entry->queueevent(ev, event);
+
+    /* ack sequence number */
+    rpc_rsp.op_ret = 0;
+    rpc_rsp.seq = rpc_req.seq;
+
+    goto submit_rpc;
+
+handle_xdr_error:
+    rpc_rsp.op_ret = -1;
+    rpc_rsp.seq = 0; /* invalid */
+submit_rpc:
+    return changelog_rpc_sumbit_reply(req, &rpc_rsp, NULL, 0, NULL,
+                                      (xdrproc_t)xdr_changelog_event_rsp);
+}
+
+int
+gf_changelog_reborp_handle_event(rpcsvc_request_t *req)
+{
+    xlator_t *this = NULL;
+    rpcsvc_t *svc = NULL;
+    gf_changelog_t *entry = NULL;
+
+    svc = rpcsvc_request_service(req);
+    entry = svc->mydata;
+
+    this = THIS = entry->this;
+
+    return gf_changelog_event_handler(req, this, entry);
+}
+
+static rpcsvc_actor_t gf_changelog_reborp_actors[CHANGELOG_REV_PROC_MAX] = {
+    [CHANGELOG_REV_PROC_EVENT] = {"CHANGELOG EVENT HANDLER",
+                                  gf_changelog_reborp_handle_event, NULL,
+                                  CHANGELOG_REV_PROC_EVENT, DRC_NA, 0},
+};
+
+/**
+ * Do not use synctask as the RPC layer dereferences ->mydata as THIS.
+ * In gf_changelog_setup_rpc(), @cbkdata is of type @gf_changelog_t,
+ * and that's required to invoke the callback with the appropriate
+ * brick path and it's private data.
+ */
+static struct rpcsvc_program gf_changelog_reborp_prog = {
+    .progname = "LIBGFCHANGELOG REBORP",
+    .prognum = CHANGELOG_REV_RPC_PROCNUM,
+    .progver = CHANGELOG_REV_RPC_PROCVER,
+    .numactors = CHANGELOG_REV_PROC_MAX,
+    .actors = gf_changelog_reborp_actors,
+    .synctask = _gf_false,
+};
+
+static struct rpcsvc_program *gf_changelog_reborp_programs[] = {
+    &gf_changelog_reborp_prog,
+    NULL,
+};
diff --git a/xlators/features/changelog/lib/src/gf-changelog-rpc.c b/xlators/features/changelog/lib/src/gf-changelog-rpc.c
new file mode 100644
index 00000000000..8ec6ffbcebc
--- /dev/null
+++ b/xlators/features/changelog/lib/src/gf-changelog-rpc.c
@@ -0,0 +1,98 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include "gf-changelog-rpc.h"
+#include "changelog-misc.h"
+#include "changelog-mem-types.h"
+
+struct rpc_clnt_program gf_changelog_clnt;
+
+/* TODO: piggyback reconnect to called (upcall) */
+int
+gf_changelog_rpc_notify(struct rpc_clnt *rpc, void *mydata,
+                        rpc_clnt_event_t event, void *data)
+{
+    switch (event) {
+        case RPC_CLNT_CONNECT:
+            break;
+        case RPC_CLNT_DISCONNECT:
+        case RPC_CLNT_MSG:
+        case RPC_CLNT_DESTROY:
+        case RPC_CLNT_PING:
+            break;
+    }
+
+    return 0;
+}
+
+struct rpc_clnt *
+gf_changelog_rpc_init(xlator_t *this, gf_changelog_t *entry)
+{
+    char sockfile[UNIX_PATH_MAX] = {
+        0,
+    };
+
+    CHANGELOG_MAKE_SOCKET_PATH(entry->brick, sockfile, UNIX_PATH_MAX);
+    return changelog_rpc_client_init(this, entry, sockfile,
+                                     gf_changelog_rpc_notify);
+}
+
+/**
+ * remote procedure calls declarations.
+ */
+
+int
+gf_probe_changelog_cbk(struct rpc_req *req, struct iovec *iovec, int count,
+                       void *myframe)
+{
+    return 0;
+}
+
+int
+gf_probe_changelog_filter(call_frame_t *frame, xlator_t *this, void *data)
+{
+    char *sock = NULL;
+    gf_changelog_t *entry = NULL;
+    changelog_probe_req req = {
+        0,
+    };
+
+    entry = data;
+    sock = RPC_SOCK(entry);
+
+    (void)memcpy(&req.sock, sock, strlen(sock));
+    req.filter = entry->notify;
+
+    /* invoke RPC */
+    return changelog_rpc_sumbit_req(
+        RPC_PROBER(entry), (void *)&req, frame, &gf_changelog_clnt,
+        CHANGELOG_RPC_PROBE_FILTER, NULL, 0, NULL, this, gf_probe_changelog_cbk,
+        (xdrproc_t)xdr_changelog_probe_req);
+}
+
+int
+gf_changelog_invoke_rpc(xlator_t *this, gf_changelog_t *entry, int procidx)
+{
+    return changelog_invoke_rpc(this, RPC_PROBER(entry), &gf_changelog_clnt,
+                                procidx, entry);
+}
+
+struct rpc_clnt_procedure gf_changelog_procs[CHANGELOG_RPC_PROC_MAX] = {
+    [CHANGELOG_RPC_PROC_NULL] = {"NULL", NULL},
+    [CHANGELOG_RPC_PROBE_FILTER] = {"PROBE FILTER", gf_probe_changelog_filter},
+};
+
+struct rpc_clnt_program gf_changelog_clnt = {
+    .progname = "LIBGFCHANGELOG",
+    .prognum = CHANGELOG_RPC_PROGNUM,
+    .progver = CHANGELOG_RPC_PROGVER,
+    .numproc = CHANGELOG_RPC_PROC_MAX,
+    .proctable = gf_changelog_procs,
+};
diff --git a/xlators/features/changelog/lib/src/gf-changelog-rpc.h b/xlators/features/changelog/lib/src/gf-changelog-rpc.h
new file mode 100644
index 00000000000..5c82d6f1c08
--- /dev/null
+++ b/xlators/features/changelog/lib/src/gf-changelog-rpc.h
@@ -0,0 +1,28 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __GF_CHANGELOG_RPC_H
+#define __GF_CHANGELOG_RPC_H
+
+#include <glusterfs/xlator.h>
+
+#include "gf-changelog-helpers.h"
+#include "changelog-rpc-common.h"
+
+struct rpc_clnt *
+gf_changelog_rpc_init(xlator_t *, gf_changelog_t *);
+
+int
+gf_changelog_invoke_rpc(xlator_t *, gf_changelog_t *, int);
+
+rpcsvc_t *
+gf_changelog_reborp_init_rpc_listner(xlator_t *, char *, char *, void *);
+
+#endif
diff --git a/xlators/features/changelog/lib/src/gf-changelog.c b/xlators/features/changelog/lib/src/gf-changelog.c
new file mode 100644
index 00000000000..57c3d39ef76
--- /dev/null
+++ b/xlators/features/changelog/lib/src/gf-changelog.c
@@ -0,0 +1,652 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <errno.h>
+#include <dirent.h>
+#include <stddef.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <sys/un.h>
+#include <sys/time.h>
+#include <sys/resource.h>
+
+#ifndef _GNU_SOURCE
+#define _GNU_SOURCE
+#endif
+#include <string.h>
+
+#include <glusterfs/globals.h>
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/syncop.h>
+
+#include "gf-changelog-rpc.h"
+#include "gf-changelog-helpers.h"
+
+/* from the changelog translator */
+#include "changelog-misc.h"
+#include "changelog-mem-types.h"
+#include "changelog-lib-messages.h"
+
+/**
+ * Global singleton xlator pointer for the library, initialized
+ * during library load. This should probably be hidden inside
+ * an initialized object which is an handle for the consumer.
+ *
+ * TODO: do away with the global..
+ */
+xlator_t *master = NULL;
+
+static inline gf_private_t *
+gf_changelog_alloc_priv()
+{
+    int ret = 0;
+    gf_private_t *priv = NULL;
+
+    priv = GF_CALLOC(1, sizeof(*priv), gf_changelog_mt_priv_t);
+    if (!priv)
+        goto error_return;
+    INIT_LIST_HEAD(&priv->connections);
+    INIT_LIST_HEAD(&priv->cleanups);
+
+    ret = pthread_mutex_init(&priv->lock, NULL);
+    if (ret != 0)
+        goto free_priv;
+    ret = pthread_cond_init(&priv->cond, NULL);
+    if (ret != 0)
+        goto cleanup_mutex;
+
+    priv->api = NULL;
+    return priv;
+
+cleanup_mutex:
+    (void)pthread_mutex_destroy(&priv->lock);
+free_priv:
+    GF_FREE(priv);
+error_return:
+    return NULL;
+}
+
+#define GF_CHANGELOG_EVENT_POOL_SIZE 16384
+#define GF_CHANGELOG_EVENT_THREAD_COUNT 4
+
+static int
+gf_changelog_ctx_defaults_init(glusterfs_ctx_t *ctx)
+{
+    cmd_args_t *cmd_args = NULL;
+    struct rlimit lim = {
+        0,
+    };
+    call_pool_t *pool = NULL;
+    int ret = -1;
+
+    ret = xlator_mem_acct_init(THIS, gf_changelog_mt_end);
+    if (ret != 0)
+        return -1;
+
+    ctx->process_uuid = generate_glusterfs_ctx_id();
+    if (!ctx->process_uuid)
+        return -1;
+
+    ctx->page_size = 128 * GF_UNIT_KB;
+
+    ctx->iobuf_pool = iobuf_pool_new();
+    if (!ctx->iobuf_pool)
+        goto free_pool;
+
+    ctx->event_pool = gf_event_pool_new(GF_CHANGELOG_EVENT_POOL_SIZE,
+                                        GF_CHANGELOG_EVENT_THREAD_COUNT);
+    if (!ctx->event_pool)
+        goto free_pool;
+
+    pool = GF_CALLOC(1, sizeof(call_pool_t),
+                     gf_changelog_mt_libgfchangelog_call_pool_t);
+    if (!pool)
+        goto free_pool;
+
+    /* frame_mem_pool size 112 * 64 */
+    pool->frame_mem_pool = mem_pool_new(call_frame_t, 32);
+    if (!pool->frame_mem_pool)
+        goto free_pool;
+
+    /* stack_mem_pool size 256 * 128 */
+    pool->stack_mem_pool = mem_pool_new(call_stack_t, 16);
+
+    if (!pool->stack_mem_pool)
+        goto free_pool;
+
+    ctx->stub_mem_pool = mem_pool_new(call_stub_t, 16);
+    if (!ctx->stub_mem_pool)
+        goto free_pool;
+
+    ctx->dict_pool = mem_pool_new(dict_t, 32);
+    if (!ctx->dict_pool)
+        goto free_pool;
+
+    ctx->dict_pair_pool = mem_pool_new(data_pair_t, 512);
+    if (!ctx->dict_pair_pool)
+        goto free_pool;
+
+    ctx->dict_data_pool = mem_pool_new(data_t, 512);
+    if (!ctx->dict_data_pool)
+        goto free_pool;
+
+    ctx->logbuf_pool = mem_pool_new(log_buf_t, 256);
+    if (!ctx->logbuf_pool)
+        goto free_pool;
+
+    INIT_LIST_HEAD(&pool->all_frames);
+    LOCK_INIT(&pool->lock);
+    ctx->pool = pool;
+
+    LOCK_INIT(&ctx->lock);
+
+    cmd_args = &ctx->cmd_args;
+
+    INIT_LIST_HEAD(&cmd_args->xlator_options);
+
+    lim.rlim_cur = RLIM_INFINITY;
+    lim.rlim_max = RLIM_INFINITY;
+    setrlimit(RLIMIT_CORE, &lim);
+
+    return 0;
+
+free_pool:
+    if (pool) {
+        GF_FREE(pool->frame_mem_pool);
+
+        GF_FREE(pool->stack_mem_pool);
+
+        GF_FREE(pool);
+    }
+
+    GF_FREE(ctx->stub_mem_pool);
+
+    GF_FREE(ctx->dict_pool);
+
+    GF_FREE(ctx->dict_pair_pool);
+
+    GF_FREE(ctx->dict_data_pool);
+
+    GF_FREE(ctx->logbuf_pool);
+
+    GF_FREE(ctx->iobuf_pool);
+
+    GF_FREE(ctx->event_pool);
+
+    return -1;
+}
+
+/* TODO: cleanup ctx defaults */
+void
+gf_changelog_cleanup_this(xlator_t *this)
+{
+    glusterfs_ctx_t *ctx = NULL;
+
+    if (!this)
+        return;
+
+    ctx = this->ctx;
+    syncenv_destroy(ctx->env);
+    free(ctx);
+
+    this->private = NULL;
+    this->ctx = NULL;
+
+    mem_pools_fini();
+}
+
+static int
+gf_changelog_init_context()
+{
+    glusterfs_ctx_t *ctx = NULL;
+
+    ctx = glusterfs_ctx_new();
+    if (!ctx)
+        goto error_return;
+
+    if (glusterfs_globals_init(ctx))
+        goto free_ctx;
+
+    THIS->ctx = ctx;
+    if (gf_changelog_ctx_defaults_init(ctx))
+        goto free_ctx;
+
+    ctx->env = syncenv_new(0, 0, 0);
+    if (!ctx->env)
+        goto free_ctx;
+    return 0;
+
+free_ctx:
+    free(ctx);
+    THIS->ctx = NULL;
+error_return:
+    return -1;
+}
+
+static int
+gf_changelog_init_master()
+{
+    int ret = 0;
+
+    ret = gf_changelog_init_context();
+    mem_pools_init();
+
+    return ret;
+}
+
+/* TODO: cleanup clnt/svc on failure */
+int
+gf_changelog_setup_rpc(xlator_t *this, gf_changelog_t *entry, int proc)
+{
+    int ret = 0;
+    rpcsvc_t *svc = NULL;
+    struct rpc_clnt *rpc = NULL;
+
+    /**
+     * Initialize a connect back socket. A probe() RPC call to the server
+     * triggers a reverse connect.
+     */
+    svc = gf_changelog_reborp_init_rpc_listner(this, entry->brick,
+                                               RPC_SOCK(entry), entry);
+    if (!svc)
+        goto error_return;
+    RPC_REBORP(entry) = svc;
+
+    /* Initialize an RPC client */
+    rpc = gf_changelog_rpc_init(this, entry);
+    if (!rpc)
+        goto error_return;
+    RPC_PROBER(entry) = rpc;
+
+    /**
+     * @FIXME
+     * till we have connection state machine, let's delay the RPC call
+     * for now..
+     */
+    sleep(2);
+
+    /**
+     * Probe changelog translator for reverse connection. After a successful
+     * call, there's less use of the client and can be disconnected, but
+     * let's leave the connection active for any future RPC calls.
+     */
+    ret = gf_changelog_invoke_rpc(this, entry, proc);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, CHANGELOG_LIB_MSG_INVOKE_RPC_FAILED,
+               "Could not initiate probe RPC, bailing out!!!");
+        goto error_return;
+    }
+
+    return 0;
+
+error_return:
+    return -1;
+}
+
+int
+gf_cleanup_event(xlator_t *this, struct gf_event_list *ev)
+{
+    int ret = 0;
+
+    ret = gf_thread_cleanup(this, ev->invoker);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret,
+               CHANGELOG_LIB_MSG_CLEANUP_ERROR,
+               "cannot cleanup callback invoker thread."
+               " Not freeing resources");
+        return -1;
+    }
+
+    ev->entry = NULL;
+
+    return 0;
+}
+
+static int
+gf_init_event(gf_changelog_t *entry)
+{
+    int ret = 0;
+    struct gf_event_list *ev = NULL;
+
+    ev = &entry->event;
+    ev->entry = entry;
+
+    ret = pthread_mutex_init(&ev->lock, NULL);
+    if (ret != 0)
+        goto error_return;
+    ret = pthread_cond_init(&ev->cond, NULL);
+    if (ret != 0)
+        goto cleanup_mutex;
+    INIT_LIST_HEAD(&ev->events);
+
+    ev->next_seq = 0; /* bootstrap sequencing */
+
+    if (GF_NEED_ORDERED_EVENTS(entry)) {
+        entry->pickevent = pick_event_ordered;
+        entry->queueevent = queue_ordered_event;
+    } else {
+        entry->pickevent = pick_event_unordered;
+        entry->queueevent = queue_unordered_event;
+    }
+
+    ret = gf_thread_create(&ev->invoker, NULL, gf_changelog_callback_invoker,
+                           ev, "clogcbki");
+    if (ret != 0) {
+        entry->pickevent = NULL;
+        entry->queueevent = NULL;
+        goto cleanup_cond;
+    }
+
+    return 0;
+
+cleanup_cond:
+    (void)pthread_cond_destroy(&ev->cond);
+cleanup_mutex:
+    (void)pthread_mutex_destroy(&ev->lock);
+error_return:
+    return -1;
+}
+
+/**
+ * TODO:
+ *  - cleanup invoker thread
+ *  - cleanup event list
+ *  - destroy rpc{-clnt, svc}
+ */
+int
+gf_cleanup_brick_connection(xlator_t *this, gf_changelog_t *entry)
+{
+    return 0;
+}
+
+int
+gf_cleanup_connections(xlator_t *this)
+{
+    return 0;
+}
+
+static int
+gf_setup_brick_connection(xlator_t *this, struct gf_brick_spec *brick,
+                          gf_boolean_t ordered, void *xl)
+{
+    int ret = 0;
+    gf_private_t *priv = NULL;
+    gf_changelog_t *entry = NULL;
+
+    priv = this->private;
+
+    if (!brick->callback || !brick->init || !brick->fini)
+        goto error_return;
+
+    entry = GF_CALLOC(1, sizeof(*entry), gf_changelog_mt_libgfchangelog_t);
+    if (!entry)
+        goto error_return;
+    INIT_LIST_HEAD(&entry->list);
+
+    LOCK_INIT(&entry->statelock);
+    entry->connstate = GF_CHANGELOG_CONN_STATE_PENDING;
+
+    entry->notify = brick->filter;
+    if (snprintf(entry->brick, PATH_MAX, "%s", brick->brick_path) >= PATH_MAX)
+        goto free_entry;
+
+    entry->this = this;
+    entry->invokerxl = xl;
+
+    entry->ordered = ordered;
+    ret = gf_init_event(entry);
+    if (ret)
+        goto free_entry;
+
+    entry->fini = brick->fini;
+    entry->callback = brick->callback;
+    entry->connected = brick->connected;
+    entry->disconnected = brick->disconnected;
+
+    entry->ptr = brick->init(this, brick);
+    if (!entry->ptr)
+        goto cleanup_event;
+    priv->api = entry->ptr; /* pointer to API, if required */
+
+    pthread_mutex_lock(&priv->lock);
+    {
+        list_add_tail(&entry->list, &priv->connections);
+    }
+    pthread_mutex_unlock(&priv->lock);
+
+    ret = gf_changelog_setup_rpc(this, entry, CHANGELOG_RPC_PROBE_FILTER);
+    if (ret)
+        goto cleanup_event;
+    return 0;
+
+cleanup_event:
+    (void)gf_cleanup_event(this, &entry->event);
+free_entry:
+    gf_msg_debug(this->name, 0, "freeing entry %p", entry);
+    list_del(&entry->list); /* FIXME: kludge for now */
+    GF_FREE(entry);
+error_return:
+    return -1;
+}
+
+int
+gf_changelog_register_brick(xlator_t *this, struct gf_brick_spec *brick,
+                            gf_boolean_t ordered, void *xl)
+{
+    return gf_setup_brick_connection(this, brick, ordered, xl);
+}
+
+static int
+gf_changelog_setup_logging(xlator_t *this, char *logfile, int loglevel)
+{
+    /* passing ident as NULL means to use default ident for syslog */
+    if (gf_log_init(this->ctx, logfile, NULL))
+        return -1;
+
+    gf_log_set_loglevel(this->ctx, (loglevel == -1) ? GF_LOG_INFO : loglevel);
+    return 0;
+}
+
+static int
+gf_changelog_set_master(xlator_t *master, void *xl)
+{
+    int32_t ret = 0;
+    xlator_t *this = NULL;
+    xlator_t *old_this = NULL;
+    gf_private_t *priv = NULL;
+
+    this = xl;
+    if (!this || !this->ctx) {
+        ret = gf_changelog_init_master();
+        if (ret)
+            return -1;
+        this = THIS;
+    }
+
+    master->ctx = this->ctx;
+
+    INIT_LIST_HEAD(&master->volume_options);
+    SAVE_THIS(THIS);
+
+    ret = xlator_mem_acct_init(THIS, gf_changelog_mt_end);
+    if (ret != 0)
+        goto restore_this;
+
+    priv = gf_changelog_alloc_priv();
+    if (!priv) {
+        ret = -1;
+        goto restore_this;
+    }
+
+    if (!xl) {
+        /* poller thread */
+        ret = gf_thread_create(&priv->poller, NULL, changelog_rpc_poller, THIS,
+                               "clogpoll");
+        if (ret != 0) {
+            GF_FREE(priv);
+            gf_msg(master->name, GF_LOG_ERROR, 0,
+                   CHANGELOG_LIB_MSG_THREAD_CREATION_FAILED,
+                   "failed to spawn poller thread");
+            goto restore_this;
+        }
+    }
+
+    master->private = priv;
+
+restore_this:
+    RESTORE_THIS();
+
+    return ret;
+}
+
+int
+gf_changelog_init(void *xl)
+{
+    int ret = 0;
+    gf_private_t *priv = NULL;
+
+    if (master)
+        return 0;
+
+    master = calloc(1, sizeof(*master));
+    if (!master)
+        goto error_return;
+
+    master->name = strdup("gfchangelog");
+    if (!master->name)
+        goto dealloc_master;
+
+    ret = gf_changelog_set_master(master, xl);
+    if (ret)
+        goto dealloc_name;
+
+    priv = master->private;
+    ret = gf_thread_create(&priv->connectionjanitor, NULL,
+                           gf_changelog_connection_janitor, master, "clogjan");
+    if (ret != 0) {
+        /* TODO: cleanup priv, mutex (poller thread for !xl) */
+        goto dealloc_name;
+    }
+
+    return 0;
+
+dealloc_name:
+    free(master->name);
+dealloc_master:
+    free(master);
+    master = NULL;
+error_return:
+    return -1;
+}
+
+int
+gf_changelog_register_generic(struct gf_brick_spec *bricks, int count,
+                              int ordered, char *logfile, int lvl, void *xl)
+{
+    int ret = 0;
+    xlator_t *this = NULL;
+    xlator_t *old_this = NULL;
+    struct gf_brick_spec *brick = NULL;
+    gf_boolean_t need_order = _gf_false;
+
+    SAVE_THIS(xl);
+
+    this = THIS;
+    if (!this)
+        goto error_return;
+
+    ret = gf_changelog_setup_logging(this, logfile, lvl);
+    if (ret)
+        goto error_return;
+
+    need_order = (ordered) ? _gf_true : _gf_false;
+
+    brick = bricks;
+    while (count--) {
+        gf_smsg(this->name, GF_LOG_INFO, 0,
+                CHANGELOG_LIB_MSG_NOTIFY_REGISTER_INFO, "brick=%s",
+                brick->brick_path, "notify_filter=%d", brick->filter, NULL);
+
+        ret = gf_changelog_register_brick(this, brick, need_order, xl);
+        if (ret != 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   CHANGELOG_LIB_MSG_NOTIFY_REGISTER_FAILED,
+                   "Error registering with changelog xlator");
+            break;
+        }
+
+        brick++;
+    }
+
+    if (ret != 0)
+        goto cleanup_inited_bricks;
+
+    RESTORE_THIS();
+    return 0;
+
+cleanup_inited_bricks:
+    gf_cleanup_connections(this);
+error_return:
+    RESTORE_THIS();
+    return -1;
+}
+
+/**
+ * @API
+ *  gf_changelog_register()
+ *
+ * This is _NOT_ a generic register API. It's a special API to handle
+ * updates at a journal granulality. This is used by consumers wanting
+ * to process persistent journal such as geo-replication via a set of
+ * APIs. All of this is required to maintain backward compatibility.
+ * Owner specific private data is stored in ->api (in gf_private_t),
+ * which is used by APIs to access it's private data. This limits
+ * the library access to a single brick, but that's how it used to
+ * be anyway. Furthermore, this API solely _owns_ "this", therefore
+ * callers already having a notion of "this" are expected to use the
+ * newer API.
+ *
+ * Newer applications wanting to use this library need not face this
+ * limitation and reply of the much more feature rich generic register
+ * API, which is purely callback based.
+ *
+ * NOTE: @max_reconnects is not used but required for backward compat.
+ *
+ * For generic API, refer gf_changelog_register_generic().
+ */
+int
+gf_changelog_register(char *brick_path, char *scratch_dir, char *log_file,
+                      int log_level, int max_reconnects)
+{
+    struct gf_brick_spec brick = {
+        0,
+    };
+
+    if (master)
+        THIS = master;
+    else
+        return -1;
+
+    brick.brick_path = brick_path;
+    brick.filter = CHANGELOG_OP_TYPE_JOURNAL;
+
+    brick.init = gf_changelog_journal_init;
+    brick.fini = gf_changelog_journal_fini;
+    brick.callback = gf_changelog_handle_journal;
+    brick.connected = gf_changelog_journal_connect;
+    brick.disconnected = gf_changelog_journal_disconnect;
+
+    brick.ptr = scratch_dir;
+
+    return gf_changelog_register_generic(&brick, 1, 1, log_file, log_level,
+                                         NULL);
+}
diff --git a/xlators/features/changelog/lib/src/gf-history-changelog.c b/xlators/features/changelog/lib/src/gf-history-changelog.c
new file mode 100644
index 00000000000..a16219f3664
--- /dev/null
+++ b/xlators/features/changelog/lib/src/gf-history-changelog.c
@@ -0,0 +1,1020 @@
+#include <errno.h>
+#include <dirent.h>
+#include <stddef.h>
+#include <sys/types.h>
+
+#ifndef _GNU_SOURCE
+#define _GNU_SOURCE
+#endif
+#include <string.h>
+
+#include <glusterfs/globals.h>
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/syscall.h>
+
+#include "gf-changelog-helpers.h"
+#include "gf-changelog-journal.h"
+
+/* from the changelog translator */
+#include "changelog-misc.h"
+#include "changelog-lib-messages.h"
+#include "changelog-mem-types.h"
+
+/**
+ * @API
+ * gf_history_changelog_done:
+ *    Move processed history changelog file from .processing
+ *    to .processed
+ *
+ * ARGUMENTS:
+ *    file(IN): path to processed history changelog file in
+ *    .processing directory.
+ *
+ * RETURN VALUE:
+ *    0: On success.
+ *   -1: On error.
+ */
+int
+gf_history_changelog_done(char *file)
+{
+    int ret = -1;
+    char *buffer = NULL;
+    xlator_t *this = NULL;
+    gf_changelog_journal_t *jnl = NULL;
+    gf_changelog_journal_t *hist_jnl = NULL;
+    char to_path[PATH_MAX] = {
+        0,
+    };
+
+    errno = EINVAL;
+
+    this = THIS;
+    if (!this)
+        goto out;
+
+    jnl = (gf_changelog_journal_t *)GF_CHANGELOG_GET_API_PTR(this);
+    if (!jnl)
+        goto out;
+
+    hist_jnl = jnl->hist_jnl;
+    if (!hist_jnl)
+        goto out;
+
+    if (!file || !strlen(file))
+        goto out;
+
+    /* make sure 'file' is inside ->jnl_working_dir */
+    buffer = realpath(file, NULL);
+    if (!buffer)
+        goto out;
+
+    if (strncmp(hist_jnl->jnl_working_dir, buffer,
+                strlen(hist_jnl->jnl_working_dir)))
+        goto out;
+
+    (void)snprintf(to_path, PATH_MAX, "%s%s", hist_jnl->jnl_processed_dir,
+                   basename(buffer));
+    gf_msg_debug(this->name, 0, "moving %s to processed directory", file);
+    ret = sys_rename(buffer, to_path);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                CHANGELOG_LIB_MSG_RENAME_FAILED, "from=%s", file, "to=%s",
+                to_path, NULL);
+        goto out;
+    }
+
+    ret = 0;
+
+out:
+    if (buffer)
+        free(buffer); /* allocated by realpath() */
+    return ret;
+}
+
+/**
+ * @API
+ *  gf_history_changelog_start_fresh:
+ *     For a set of changelogs, start from the beginning.
+ *     It will truncates the history tracker fd.
+ *
+ *  RETURN VALUES:
+ *     0: On success.
+ *    -1: On error.
+ */
+int
+gf_history_changelog_start_fresh()
+{
+    xlator_t *this = NULL;
+    gf_changelog_journal_t *jnl = NULL;
+    gf_changelog_journal_t *hist_jnl = NULL;
+
+    this = THIS;
+    if (!this)
+        goto out;
+
+    errno = EINVAL;
+
+    jnl = (gf_changelog_journal_t *)GF_CHANGELOG_GET_API_PTR(this);
+    if (!jnl)
+        goto out;
+
+    hist_jnl = jnl->hist_jnl;
+    if (!hist_jnl)
+        goto out;
+
+    if (gf_ftruncate(hist_jnl->jnl_fd, 0))
+        goto out;
+
+    return 0;
+
+out:
+    return -1;
+}
+
+/**
+ * @API
+ *  gf_history_changelog_next_change:
+ *     Return the next history changelog file entry. Zero means all
+ *     history chanelogs are consumed.
+ *
+ *  ARGUMENTS:
+ *     bufptr(OUT): Path to unprocessed history changelog file
+ *                  from tracker file.
+ *     maxlen(IN): Usually PATH_MAX.
+ *
+ *  RETURN VALUES:
+ *     size: On success.
+ *     -1  : On error.
+ */
+ssize_t
+gf_history_changelog_next_change(char *bufptr, size_t maxlen)
+{
+    ssize_t size = -1;
+    int tracker_fd = 0;
+    xlator_t *this = NULL;
+    gf_changelog_journal_t *jnl = NULL;
+    gf_changelog_journal_t *hist_jnl = NULL;
+    char buffer[PATH_MAX] = {
+        0,
+    };
+
+    if (maxlen > PATH_MAX) {
+        errno = ENAMETOOLONG;
+        goto out;
+    }
+
+    errno = EINVAL;
+
+    this = THIS;
+    if (!this)
+        goto out;
+
+    jnl = (gf_changelog_journal_t *)GF_CHANGELOG_GET_API_PTR(this);
+    if (!jnl)
+        goto out;
+
+    hist_jnl = jnl->hist_jnl;
+    if (!hist_jnl)
+        goto out;
+
+    tracker_fd = hist_jnl->jnl_fd;
+
+    size = gf_readline(tracker_fd, buffer, maxlen);
+    if (size < 0) {
+        size = -1;
+        goto out;
+    }
+
+    if (size == 0)
+        goto out;
+
+    memcpy(bufptr, buffer, size - 1);
+    bufptr[size - 1] = '\0';
+
+out:
+    return size;
+}
+
+/**
+ * @API
+ *  gf_history_changelog_scan:
+ *     Scan and generate a list of change entries.
+ *     Calling this api multiple times (without calling gf_changlog_done())
+ *     would result new changelogs(s) being refreshed in the tracker file.
+ *     This call also acts as a cancellation point for the consumer.
+ *
+ *  RETURN VALUES:
+ *      +ve integer : success and keep scanning.(count of changelogs)
+ *      0           : success and done scanning.
+ *     -1           : error.
+ *
+ *  NOTE: After first 0 return call_get_next change for once more time
+ *        to empty the tracker
+ *
+ */
+ssize_t
+gf_history_changelog_scan()
+{
+    int tracker_fd = 0;
+    size_t off = 0;
+    xlator_t *this = NULL;
+    size_t nr_entries = 0;
+    gf_changelog_journal_t *jnl = NULL;
+    gf_changelog_journal_t *hist_jnl = NULL;
+    struct dirent *entry = NULL;
+    struct dirent scratch[2] = {
+        {
+            0,
+        },
+    };
+    char buffer[PATH_MAX] = {
+        0,
+    };
+    static int is_last_scan;
+
+    this = THIS;
+    if (!this)
+        goto out;
+
+    jnl = (gf_changelog_journal_t *)GF_CHANGELOG_GET_API_PTR(this);
+    if (!jnl)
+        goto out;
+    if (JNL_IS_API_DISCONNECTED(jnl)) {
+        errno = ENOTCONN;
+        goto out;
+    }
+
+    hist_jnl = jnl->hist_jnl;
+    if (!hist_jnl)
+        goto out;
+
+retry:
+    if (is_last_scan == 1)
+        return 0;
+    if (hist_jnl->hist_done == 0)
+        is_last_scan = 1;
+
+    errno = EINVAL;
+    if (hist_jnl->hist_done == -1)
+        goto out;
+
+    tracker_fd = hist_jnl->jnl_fd;
+
+    if (gf_ftruncate(tracker_fd, 0))
+        goto out;
+
+    rewinddir(hist_jnl->jnl_dir);
+
+    for (;;) {
+        errno = 0;
+        entry = sys_readdir(hist_jnl->jnl_dir, scratch);
+        if (!entry || errno != 0)
+            break;
+
+        if (strcmp(basename(entry->d_name), ".") == 0 ||
+            strcmp(basename(entry->d_name), "..") == 0)
+            continue;
+
+        nr_entries++;
+
+        GF_CHANGELOG_FILL_BUFFER(hist_jnl->jnl_processing_dir, buffer, off,
+                                 strlen(hist_jnl->jnl_processing_dir));
+        GF_CHANGELOG_FILL_BUFFER(entry->d_name, buffer, off,
+                                 strlen(entry->d_name));
+        GF_CHANGELOG_FILL_BUFFER("\n", buffer, off, 1);
+
+        if (gf_changelog_write(tracker_fd, buffer, off) != off) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, CHANGELOG_LIB_MSG_WRITE_FAILED,
+                   "error writing changelog filename"
+                   " to tracker file");
+            break;
+        }
+        off = 0;
+    }
+
+    gf_msg_debug(this->name, 0, "hist_done %d, is_last_scan: %d",
+                 hist_jnl->hist_done, is_last_scan);
+
+    if (!entry) {
+        if (gf_lseek(tracker_fd, 0, SEEK_SET) != -1) {
+            if (nr_entries > 0)
+                return nr_entries;
+            else {
+                sleep(1);
+                goto retry;
+            }
+        }
+    }
+out:
+    return -1;
+}
+
+/*
+ * Gets timestamp value at the changelog path at index.
+ * Returns 0 on success(updates given time-stamp), -1 on failure.
+ */
+int
+gf_history_get_timestamp(int fd, int index, int len, unsigned long *ts)
+{
+    xlator_t *this = NULL;
+    int n_read = -1;
+    char path_buf[PATH_MAX] = {
+        0,
+    };
+    char *iter = path_buf;
+    size_t offset = index * (len + 1);
+    unsigned long value = 0;
+    int ret = 0;
+
+    this = THIS;
+    if (!this) {
+        return -1;
+    }
+
+    n_read = sys_pread(fd, path_buf, len, offset);
+    if (n_read < 0) {
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, errno, CHANGELOG_LIB_MSG_READ_ERROR,
+               "could not read from htime file");
+        goto out;
+    }
+    iter += len - TIMESTAMP_LENGTH;
+    sscanf(iter, "%lu", &value);
+out:
+    if (ret == 0)
+        *ts = value;
+    return ret;
+}
+
+/*
+ * Function to ensure correctness of search
+ * Checks whether @value is there next to @target_index or not
+ */
+int
+gf_history_check(int fd, int target_index, unsigned long value, int len)
+{
+    int ret = 0;
+    unsigned long ts1 = 0;
+    unsigned long ts2 = 0;
+
+    if (target_index == 0) {
+        ret = gf_history_get_timestamp(fd, target_index, len, &ts1);
+        if (ret == -1)
+            goto out;
+        if (value <= ts1)
+            goto out;
+        else {
+            ret = -1;
+            goto out;
+        }
+    }
+
+    ret = gf_history_get_timestamp(fd, target_index, len, &ts1);
+    if (ret == -1)
+        goto out;
+    ret = gf_history_get_timestamp(fd, target_index - 1, len, &ts2);
+    if (ret == -1)
+        goto out;
+
+    if ((value <= ts1) && (value > ts2)) {
+        goto out;
+    } else
+        ret = -1;
+out:
+    return ret;
+}
+
+/*
+ * This is a "binary search" based search function which checks neighbours
+ * for in-range availability of the value to be searched and provides the
+ * index at which the changelog file nearest to the requested timestamp(value)
+ * can be read from.
+ *
+ * Actual offset can be calculated as (index* (len+1) ).
+ * "1" is because the changelog paths are null terminated.
+ *
+ * @path        : Htime file to search in
+ * @value       : time stamp to search
+ * @from        : start index to search
+ * @to          : end index to search
+ * @len         : length of fixes length strings separated by null
+ */
+
+int
+gf_history_b_search(int fd, unsigned long value, unsigned long from,
+                    unsigned long to, int len)
+{
+    int m_index = -1;
+    unsigned long cur_value = 0;
+    unsigned long ts1 = 0;
+    int ret = 0;
+
+    m_index = (from + to) / 2;
+
+    if ((to - from) <= 1) {
+        /* either one or 2 changelogs left */
+        if (to != from) {
+            /* check if value is less or greater than to
+             * return accordingly
+             */
+            ret = gf_history_get_timestamp(fd, from, len, &ts1);
+            if (ret == -1)
+                goto out;
+            if (ts1 >= value) {
+                /* actually compatision should be
+                 * exactly == but considering
+                 *
+                 * case of only 2 changelogs in htime file
+                 */
+                return from;
+            } else
+                return to;
+        } else
+            return to;
+    }
+
+    ret = gf_history_get_timestamp(fd, m_index, len, &cur_value);
+    if (ret == -1)
+        goto out;
+    if (cur_value == value) {
+        return m_index;
+    } else if (value > cur_value) {
+        ret = gf_history_get_timestamp(fd, m_index + 1, len, &cur_value);
+        if (ret == -1)
+            goto out;
+        if (value < cur_value)
+            return m_index + 1;
+        else
+            return gf_history_b_search(fd, value, m_index + 1, to, len);
+    } else {
+        if (m_index == 0) {
+            /*  we are sure that values exists
+             *  in this htime file
+             */
+            return 0;
+        } else {
+            ret = gf_history_get_timestamp(fd, m_index - 1, len, &cur_value);
+            if (ret == -1)
+                goto out;
+            if (value > cur_value) {
+                return m_index;
+            } else
+                return gf_history_b_search(fd, value, from, m_index - 1, len);
+        }
+    }
+out:
+    return -1;
+}
+
+/*
+ * Description: Checks if the changelog path is usable or not,
+ *              which is differentiated by checking for "changelog"
+ *              in the path and not "CHANGELOG".
+ *
+ * Returns:
+ * 1 : Yes, usable ( contains "CHANGELOG" )
+ * 0 : No, Not usable ( contains, "changelog")
+ */
+int
+gf_is_changelog_usable(char *cl_path)
+{
+    int ret = -1;
+    const char low_c[] = "changelog";
+    char *str_ret = NULL;
+    char *bname = NULL;
+
+    bname = basename(cl_path);
+
+    str_ret = strstr(bname, low_c);
+
+    if (str_ret != NULL)
+        ret = 0;
+    else
+        ret = 1;
+
+    return ret;
+}
+
+void *
+gf_changelog_consume_wrap(void *data)
+{
+    int ret = -1;
+    ssize_t nread = 0;
+    xlator_t *this = NULL;
+    gf_changelog_consume_data_t *ccd = NULL;
+
+    ccd = (gf_changelog_consume_data_t *)data;
+    this = ccd->this;
+
+    ccd->retval = -1;
+
+    nread = sys_pread(ccd->fd, ccd->changelog, PATH_MAX - 1, ccd->offset);
+    if (nread < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, CHANGELOG_LIB_MSG_READ_ERROR,
+               "cannot read from history metadata file");
+        goto out;
+    }
+
+    /* TODO: handle short reads and EOF. */
+    if (gf_is_changelog_usable(ccd->changelog) == 1) {
+        ret = gf_changelog_consume(ccd->this, ccd->jnl, ccd->changelog,
+                                   _gf_true);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_LIB_MSG_PARSE_ERROR,
+                    "name=%s", ccd->changelog, NULL);
+            goto out;
+        }
+    }
+    ccd->retval = 0;
+
+out:
+    return NULL;
+}
+
+/**
+ * "gf_history_consume" is a worker function for history.
+ * parses and moves changelogs files from index "from"
+ * to index "to" in open htime file whose fd is "fd".
+ */
+
+#define MAX_PARALLELS 10
+
+void *
+gf_history_consume(void *data)
+{
+    xlator_t *this = NULL;
+    gf_changelog_journal_t *jnl = NULL;
+    gf_changelog_journal_t *hist_jnl = NULL;
+    int ret = 0;
+    int iter = 0;
+    int fd = -1;
+    int from = -1;
+    int to = -1;
+    int len = -1;
+    int n_parallel = 0;
+    int n_envoked = 0;
+    gf_boolean_t publish = _gf_true;
+    pthread_t th_id[MAX_PARALLELS] = {
+        0,
+    };
+    gf_changelog_history_data_t *hist_data = NULL;
+    gf_changelog_consume_data_t ccd[MAX_PARALLELS] = {
+        {0},
+    };
+    gf_changelog_consume_data_t *curr = NULL;
+
+    hist_data = (gf_changelog_history_data_t *)data;
+    if (hist_data == NULL) {
+        ret = -1;
+        goto out;
+    }
+
+    fd = hist_data->htime_fd;
+    from = hist_data->from;
+    to = hist_data->to;
+    len = hist_data->len;
+    n_parallel = hist_data->n_parallel;
+
+    THIS = hist_data->this;
+    this = hist_data->this;
+    if (!this) {
+        ret = -1;
+        goto out;
+    }
+
+    jnl = (gf_changelog_journal_t *)GF_CHANGELOG_GET_API_PTR(this);
+    if (!jnl) {
+        ret = -1;
+        goto out;
+    }
+
+    hist_jnl = jnl->hist_jnl;
+    if (!hist_jnl) {
+        ret = -1;
+        goto out;
+    }
+
+    while (from <= to) {
+        n_envoked = 0;
+
+        for (iter = 0; (iter < n_parallel) && (from <= to); iter++) {
+            curr = &ccd[iter];
+
+            curr->this = this;
+            curr->jnl = hist_jnl;
+            curr->fd = fd;
+            curr->offset = from * (len + 1);
+
+            curr->retval = 0;
+            memset(curr->changelog, '\0', PATH_MAX);
+
+            ret = gf_thread_create(&th_id[iter], NULL,
+                                   gf_changelog_consume_wrap, curr,
+                                   "clogc%03hx", (iter + 1) & 0x3ff);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, ret,
+                       CHANGELOG_LIB_MSG_THREAD_CREATION_FAILED,
+                       "could not create consume-thread");
+                goto sync;
+            } else
+                n_envoked++;
+
+            from++;
+        }
+
+    sync:
+        for (iter = 0; iter < n_envoked; iter++) {
+            ret = pthread_join(th_id[iter], NULL);
+            if (ret) {
+                publish = _gf_false;
+                gf_msg(this->name, GF_LOG_ERROR, ret,
+                       CHANGELOG_LIB_MSG_PTHREAD_JOIN_FAILED,
+                       "pthread_join() error");
+                /* try to join the rest */
+                continue;
+            }
+
+            if (publish == _gf_false)
+                continue;
+
+            curr = &ccd[iter];
+            if (ccd->retval) {
+                publish = _gf_false;
+                gf_smsg(this->name, GF_LOG_ERROR, 0,
+                        CHANGELOG_LIB_MSG_PARSE_ERROR_CEASED, NULL);
+                continue;
+            }
+
+            ret = gf_changelog_publish(curr->this, curr->jnl, curr->changelog);
+            if (ret) {
+                publish = _gf_false;
+                gf_msg(this->name, GF_LOG_ERROR, 0,
+                       CHANGELOG_LIB_MSG_PUBLISH_ERROR,
+                       "publish error, ceased publishing...");
+            }
+        }
+    }
+
+    /* informing "parsing done". */
+    hist_jnl->hist_done = (publish == _gf_true) ? 0 : -1;
+
+out:
+    if (fd != -1)
+        (void)sys_close(fd);
+    GF_FREE(hist_data);
+    return NULL;
+}
+
+/**
+ * @API
+ * gf_history_changelog() : Get/parse historical changelogs and get them ready
+ * for consumption.
+ *
+ * Arguments:
+ * @changelog_dir : Directory location from where history changelogs are
+ * supposed to be consumed.
+ * @start: Unix timestamp FROM where changelogs should be consumed.
+ * @end: Unix timestamp TO where changelogsshould be consumed.
+ * @n_parallel : degree of parallelism while changelog parsing.
+ * @actual_end : the end time till where changelogs are available.
+ *
+ * Return:
+ * Returns <timestamp> on success, the last time till where changelogs are
+ *      available.
+ * Returns -1 on failure(error).
+ */
+
+/**
+ * Extract timestamp range from a historical metadata file
+ * Returns:
+ *    0 : Success ({min,max}_ts with the appropriate values)
+ *   -1 : Failure
+ *   -2 : Ignore this metadata file and process next
+ */
+int
+gf_changelog_extract_min_max(const char *dname, const char *htime_dir, int *fd,
+                             unsigned long *total, unsigned long *min_ts,
+                             unsigned long *max_ts)
+{
+    int ret = -1;
+    xlator_t *this = NULL;
+    char htime_file[PATH_MAX] = {
+        0,
+    };
+    struct stat stbuf = {
+        0,
+    };
+    char *iter = NULL;
+    char x_value[30] = {
+        0,
+    };
+
+    this = THIS;
+
+    snprintf(htime_file, PATH_MAX, "%s/%s", htime_dir, dname);
+
+    iter = (htime_file + strlen(htime_file) - TIMESTAMP_LENGTH);
+    sscanf(iter, "%lu", min_ts);
+
+    ret = sys_stat(htime_file, &stbuf);
+    if (ret) {
+        ret = -1;
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_LIB_MSG_HTIME_ERROR,
+                "op=stat", "path=%s", htime_file, NULL);
+        goto out;
+    }
+
+    /* ignore everything except regular files */
+    if (!S_ISREG(stbuf.st_mode)) {
+        ret = -2;
+        goto out;
+    }
+
+    *fd = open(htime_file, O_RDONLY);
+    if (*fd < 0) {
+        ret = -1;
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_LIB_MSG_HTIME_ERROR,
+                "op=open", "path=%s", htime_file, NULL);
+        goto out;
+    }
+
+    /* Looks good, extract max timestamp */
+    ret = sys_fgetxattr(*fd, HTIME_KEY, x_value, sizeof(x_value));
+    if (ret < 0) {
+        ret = -1;
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                CHANGELOG_LIB_MSG_GET_XATTR_FAILED, "path=%s", htime_file,
+                NULL);
+        goto out;
+    }
+
+    sscanf(x_value, "%lu:%lu", max_ts, total);
+    gf_smsg(this->name, GF_LOG_INFO, 0, CHANGELOG_LIB_MSG_MIN_MAX_INFO,
+            "min=%lu", *min_ts, "max=%lu", *max_ts, "total_changelogs=%lu",
+            *total, NULL);
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+/* gf_history_changelog returns actual_end and spawns threads to
+ * parse historical changelogs. The return values are as follows.
+ *     0 : On success
+ *     1 : Successful, but partial historical changelogs available,
+ *         end time falls into different htime file or future time
+ *    -2 : Error, requested historical changelog not available, not
+ *         even partial
+ *    -1 : On any error
+ */
+int
+gf_history_changelog(char *changelog_dir, unsigned long start,
+                     unsigned long end, int n_parallel,
+                     unsigned long *actual_end)
+{
+    int ret = 0;
+    int len = -1;
+    int fd = -1;
+    int n_read = -1;
+    unsigned long min_ts = 0;
+    unsigned long max_ts = 0;
+    unsigned long end2 = 0;
+    unsigned long ts1 = 0;
+    unsigned long ts2 = 0;
+    unsigned long to = 0;
+    unsigned long from = 0;
+    unsigned long total_changelog = 0;
+    xlator_t *this = NULL;
+    gf_changelog_journal_t *jnl = NULL;
+    gf_changelog_journal_t *hist_jnl = NULL;
+    gf_changelog_history_data_t *hist_data = NULL;
+    DIR *dirp = NULL;
+    struct dirent *entry = NULL;
+    struct dirent scratch[2] = {
+        {
+            0,
+        },
+    };
+    pthread_t consume_th = 0;
+    char htime_dir[PATH_MAX] = {
+        0,
+    };
+    char buffer[PATH_MAX] = {
+        0,
+    };
+    gf_boolean_t partial_history = _gf_false;
+
+    pthread_attr_t attr;
+
+    this = THIS;
+    if (!this) {
+        ret = -1;
+        goto out;
+    }
+
+    ret = pthread_attr_init(&attr);
+    if (ret != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, CHANGELOG_LIB_MSG_PTHREAD_ERROR,
+               "Pthread init failed");
+        return -1;
+    }
+
+    jnl = (gf_changelog_journal_t *)GF_CHANGELOG_GET_API_PTR(this);
+    if (!jnl) {
+        ret = -1;
+        goto out;
+    }
+
+    hist_jnl = (gf_changelog_journal_t *)jnl->hist_jnl;
+    if (!hist_jnl) {
+        ret = -1;
+        goto out;
+    }
+
+    gf_smsg(this->name, GF_LOG_INFO, 0, CHANGELOG_LIB_MSG_REQUESTING_INFO,
+            "start=%lu", start, "end=%lu", end, NULL);
+
+    /* basic sanity check */
+    if (start > end || n_parallel <= 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_LIB_MSG_HIST_FAILED,
+                "start=%lu", start, "end=%lu", end, "thread_count=%d",
+                n_parallel, NULL);
+        ret = -1;
+        goto out;
+    }
+
+    /* cap parallelism count */
+    if (n_parallel > MAX_PARALLELS)
+        n_parallel = MAX_PARALLELS;
+
+    CHANGELOG_FILL_HTIME_DIR(changelog_dir, htime_dir);
+
+    dirp = sys_opendir(htime_dir);
+    if (dirp == NULL) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_LIB_MSG_HTIME_ERROR,
+                "op=opendir", "path=%s", htime_dir, NULL);
+        ret = -1;
+        goto out;
+    }
+
+    for (;;) {
+        errno = 0;
+
+        entry = sys_readdir(dirp, scratch);
+
+        if (!entry || errno != 0) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno,
+                    CHANGELOG_LIB_MSG_HIST_FAILED, "start=%lu", start,
+                    "end=%lu", end, NULL);
+            ret = -2;
+            break;
+        }
+
+        ret = gf_changelog_extract_min_max(entry->d_name, htime_dir, &fd,
+                                           &total_changelog, &min_ts, &max_ts);
+        if (ret) {
+            if (-2 == ret)
+                continue;
+            goto out;
+        }
+
+        if (start >= min_ts && start < max_ts) {
+            /**
+             * TODO: handle short reads later...
+             */
+            n_read = sys_read(fd, buffer, PATH_MAX);
+            if (n_read < 0) {
+                ret = -1;
+                gf_msg(this->name, GF_LOG_ERROR, errno,
+                       CHANGELOG_LIB_MSG_READ_ERROR,
+                       "unable to read htime file");
+                goto out;
+            }
+
+            len = strlen(buffer);
+
+            /**
+             * search @start in the htime file returning it's index
+             * (@from)
+             */
+            from = gf_history_b_search(fd, start, 0, total_changelog - 1, len);
+
+            /* ensuring correctness of gf_b_search */
+            if (gf_history_check(fd, from, start, len) != 0) {
+                ret = -1;
+                gf_smsg(this->name, GF_LOG_ERROR, 0,
+                        CHANGELOG_LIB_MSG_GET_TIME_ERROR, "for=start",
+                        "start=%lu", start, "idx=%lu", from, NULL);
+                goto out;
+            }
+
+            end2 = (end <= max_ts) ? end : max_ts;
+
+            /* Check if end falls out of same HTIME file. The end
+             * falling to a different htime file or changelog
+             * disable-enable is detected only after 20 seconds.
+             * This is required because, applications generally
+             * asks historical changelogs till current time and
+             * it is possible changelog is not rolled over yet.
+             * So, buffer time of default rollover time plus 5
+             * seconds is subtracted.  If the application requests
+             * the end time with in half a minute of changelog
+             * disable, it's not detected as changelog disable and
+             * it's application's responsibility to retry after
+             * 20 seconds before confirming it as partial history.
+             */
+            if ((end - 20) > max_ts) {
+                partial_history = _gf_true;
+            }
+
+            /**
+             * search @end2 in htime file returning it's index (@to)
+             */
+            to = gf_history_b_search(fd, end2, 0, total_changelog - 1, len);
+
+            if (gf_history_check(fd, to, end2, len) != 0) {
+                ret = -1;
+                gf_smsg(this->name, GF_LOG_ERROR, 0,
+                        CHANGELOG_LIB_MSG_GET_TIME_ERROR, "for=end",
+                        "start=%lu", end2, "idx=%lu", to, NULL);
+                goto out;
+            }
+
+            ret = gf_history_get_timestamp(fd, from, len, &ts1);
+            if (ret == -1)
+                goto out;
+
+            ret = gf_history_get_timestamp(fd, to, len, &ts2);
+            if (ret == -1)
+                goto out;
+
+            gf_smsg(this->name, GF_LOG_INFO, 0, CHANGELOG_LIB_MSG_FINAL_INFO,
+                    "from=%lu", ts1, "to=%lu", ts2, "changes=%lu",
+                    (to - from + 1), NULL);
+
+            hist_data = GF_CALLOC(1, sizeof(gf_changelog_history_data_t),
+                                  gf_changelog_mt_history_data_t);
+
+            hist_data->htime_fd = fd;
+            hist_data->from = from;
+            hist_data->to = to;
+            hist_data->len = len;
+            hist_data->n_parallel = n_parallel;
+            hist_data->this = this;
+
+            ret = pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
+            if (ret != 0) {
+                gf_msg(this->name, GF_LOG_ERROR, ret,
+                       CHANGELOG_LIB_MSG_PTHREAD_ERROR,
+                       "unable to sets the detach"
+                       " state attribute");
+                ret = -1;
+                goto out;
+            }
+
+            /* spawn a thread for background parsing & publishing */
+            ret = gf_thread_create(&consume_th, &attr, gf_history_consume,
+                                   hist_data, "cloghcon");
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, ret,
+                       CHANGELOG_LIB_MSG_THREAD_CREATION_FAILED,
+                       "creation of consume parent-thread"
+                       " failed.");
+                ret = -1;
+                goto out;
+            }
+
+            goto out;
+
+        } else { /* end of range check */
+            gf_smsg(this->name, GF_LOG_ERROR, errno,
+                    CHANGELOG_LIB_MSG_HIST_FAILED, "start=%lu", start,
+                    "end=%lu", end, "chlog_min=%lu", min_ts, "chlog_max=%lu",
+                    max_ts, NULL);
+        }
+    } /* end of readdir() */
+
+out:
+    if (dirp != NULL)
+        (void)sys_closedir(dirp);
+
+    if (ret < 0) {
+        if (fd != -1)
+            (void)sys_close(fd);
+        GF_FREE(hist_data);
+        (void)pthread_attr_destroy(&attr);
+
+        return ret;
+    }
+
+    hist_jnl->hist_done = 1;
+    *actual_end = ts2;
+
+    if (partial_history) {
+        ret = 1;
+    }
+
+    return ret;
+}
diff --git a/xlators/features/changelog/src/Makefile.am b/xlators/features/changelog/src/Makefile.am
new file mode 100644
index 00000000000..eee7dfa238d
--- /dev/null
+++ b/xlators/features/changelog/src/Makefile.am
@@ -0,0 +1,29 @@
+xlator_LTLIBRARIES = changelog.la
+
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+noinst_HEADERS = changelog-helpers.h changelog-mem-types.h changelog-rt.h \
+	changelog-rpc-common.h changelog-misc.h changelog-encoders.h \
+	changelog-rpc-common.h changelog-rpc.h changelog-ev-handle.h \
+	changelog-messages.h
+
+changelog_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+changelog_la_SOURCES = changelog.c changelog-rt.c changelog-helpers.c \
+	changelog-encoders.c changelog-rpc.c changelog-barrier.c \
+	changelog-rpc-common.c changelog-ev-handle.c
+changelog_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la \
+	$(top_builddir)/rpc/xdr/src/libgfxdr.la \
+	$(top_builddir)/rpc/rpc-lib/src/libgfrpc.la
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src \
+	-I$(top_srcdir)/rpc/rpc-lib/src \
+	-I$(top_srcdir)/rpc/rpc-transport/socket/src \
+	-I$(top_srcdir)/xlators/features/changelog/lib/src/ \
+	-fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -D$(GF_HOST_OS) \
+	-DDATADIR=\"$(localstatedir)\"
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/changelog/src/changelog-barrier.c b/xlators/features/changelog/src/changelog-barrier.c
new file mode 100644
index 00000000000..0fb89ddb127
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-barrier.c
@@ -0,0 +1,131 @@
+/*
+     Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+     This file is part of GlusterFS.
+
+     This file is licensed to you under your choice of the GNU Lesser
+     General Public License, version 3 or any later version (LGPLv3 or
+     later), or the GNU General Public License, version 2 (GPLv2), in all
+     cases as published by the Free Software Foundation.
+*/
+
+#include "changelog-helpers.h"
+#include "changelog-messages.h"
+#include <glusterfs/call-stub.h>
+
+/* Enqueue a stub*/
+void
+__chlog_barrier_enqueue(xlator_t *this, call_stub_t *stub)
+{
+    changelog_priv_t *priv = NULL;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    list_add_tail(&stub->list, &priv->queue);
+    priv->queue_size++;
+
+    return;
+}
+
+/* Dequeue a stub */
+call_stub_t *
+__chlog_barrier_dequeue(xlator_t *this, struct list_head *queue)
+{
+    call_stub_t *stub = NULL;
+    changelog_priv_t *priv = NULL;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    if (list_empty(queue))
+        goto out;
+
+    stub = list_entry(queue->next, call_stub_t, list);
+    list_del_init(&stub->list);
+
+out:
+    return stub;
+}
+
+/* Dequeue all the stubs and call corresponding resume functions */
+void
+chlog_barrier_dequeue_all(xlator_t *this, struct list_head *queue)
+{
+    call_stub_t *stub = NULL;
+
+    gf_smsg(this->name, GF_LOG_INFO, 0, CHANGELOG_MSG_DEQUEUING_BARRIER_FOPS,
+            NULL);
+
+    while ((stub = __chlog_barrier_dequeue(this, queue)))
+        call_resume(stub);
+
+    gf_smsg(this->name, GF_LOG_INFO, 0,
+            CHANGELOG_MSG_DEQUEUING_BARRIER_FOPS_FINISHED, NULL);
+    return;
+}
+
+/* Function called on changelog barrier timeout */
+void
+chlog_barrier_timeout(void *data)
+{
+    xlator_t *this = NULL;
+    changelog_priv_t *priv = NULL;
+    struct list_head queue = {
+        0,
+    };
+
+    this = data;
+    THIS = this;
+    priv = this->private;
+
+    INIT_LIST_HEAD(&queue);
+
+    gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_MSG_BARRIER_TIMEOUT, NULL);
+
+    LOCK(&priv->lock);
+    {
+        __chlog_barrier_disable(this, &queue);
+    }
+    UNLOCK(&priv->lock);
+
+    chlog_barrier_dequeue_all(this, &queue);
+
+    return;
+}
+
+/* Disable changelog barrier enable flag */
+void
+__chlog_barrier_disable(xlator_t *this, struct list_head *queue)
+{
+    changelog_priv_t *priv = this->private;
+    GF_ASSERT(priv);
+
+    if (priv->timer) {
+        gf_timer_call_cancel(this->ctx, priv->timer);
+        priv->timer = NULL;
+    }
+
+    list_splice_init(&priv->queue, queue);
+    priv->queue_size = 0;
+    priv->barrier_enabled = _gf_false;
+}
+
+/* Enable chagelog barrier enable with timer */
+int
+__chlog_barrier_enable(xlator_t *this, changelog_priv_t *priv)
+{
+    int ret = -1;
+
+    priv->timer = gf_timer_call_after(this->ctx, priv->timeout,
+                                      chlog_barrier_timeout, (void *)this);
+    if (!priv->timer) {
+        gf_smsg(this->name, GF_LOG_CRITICAL, 0,
+                CHANGELOG_MSG_TIMEOUT_ADD_FAILED, NULL);
+        goto out;
+    }
+
+    priv->barrier_enabled = _gf_true;
+    ret = 0;
+out:
+    return ret;
+}
diff --git a/xlators/features/changelog/src/changelog-encoders.c b/xlators/features/changelog/src/changelog-encoders.c
new file mode 100644
index 00000000000..63754516c2e
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-encoders.c
@@ -0,0 +1,232 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include "changelog-encoders.h"
+
+size_t
+entry_fn(void *data, char *buffer, gf_boolean_t encode)
+{
+    char *tmpbuf = NULL;
+    size_t bufsz = 0;
+    struct changelog_entry_fields *ce = NULL;
+
+    ce = (struct changelog_entry_fields *)data;
+
+    if (encode) {
+        tmpbuf = uuid_utoa(ce->cef_uuid);
+        CHANGELOG_FILL_BUFFER(buffer, bufsz, tmpbuf, strlen(tmpbuf));
+    } else {
+        CHANGELOG_FILL_BUFFER(buffer, bufsz, ce->cef_uuid, sizeof(uuid_t));
+    }
+
+    CHANGELOG_FILL_BUFFER(buffer, bufsz, "/", 1);
+    CHANGELOG_FILL_BUFFER(buffer, bufsz, ce->cef_bname, strlen(ce->cef_bname));
+    return bufsz;
+}
+
+size_t
+del_entry_fn(void *data, char *buffer, gf_boolean_t encode)
+{
+    char *tmpbuf = NULL;
+    size_t bufsz = 0;
+    struct changelog_entry_fields *ce = NULL;
+
+    ce = (struct changelog_entry_fields *)data;
+
+    if (encode) {
+        tmpbuf = uuid_utoa(ce->cef_uuid);
+        CHANGELOG_FILL_BUFFER(buffer, bufsz, tmpbuf, strlen(tmpbuf));
+    } else {
+        CHANGELOG_FILL_BUFFER(buffer, bufsz, ce->cef_uuid, sizeof(uuid_t));
+    }
+
+    CHANGELOG_FILL_BUFFER(buffer, bufsz, "/", 1);
+    CHANGELOG_FILL_BUFFER(buffer, bufsz, ce->cef_bname, strlen(ce->cef_bname));
+    CHANGELOG_FILL_BUFFER(buffer, bufsz, "\0", 1);
+
+    if (ce->cef_path[0] == '\0') {
+        CHANGELOG_FILL_BUFFER(buffer, bufsz, "\0", 1);
+    } else {
+        CHANGELOG_FILL_BUFFER(buffer, bufsz, ce->cef_path,
+                              strlen(ce->cef_path));
+    }
+
+    return bufsz;
+}
+
+size_t
+fop_fn(void *data, char *buffer, gf_boolean_t encode)
+{
+    char buf[10] = {
+        0,
+    };
+    size_t bufsz = 0;
+    glusterfs_fop_t fop = 0;
+
+    fop = *(glusterfs_fop_t *)data;
+
+    if (encode) {
+        (void)snprintf(buf, sizeof(buf), "%d", fop);
+        CHANGELOG_FILL_BUFFER(buffer, bufsz, buf, strlen(buf));
+    } else
+        CHANGELOG_FILL_BUFFER(buffer, bufsz, &fop, sizeof(fop));
+
+    return bufsz;
+}
+
+size_t
+number_fn(void *data, char *buffer, gf_boolean_t encode)
+{
+    size_t bufsz = 0;
+    unsigned int nr = 0;
+    char buf[20] = {
+        0,
+    };
+
+    nr = *(unsigned int *)data;
+
+    if (encode) {
+        (void)snprintf(buf, sizeof(buf), "%u", nr);
+        CHANGELOG_FILL_BUFFER(buffer, bufsz, buf, strlen(buf));
+    } else
+        CHANGELOG_FILL_BUFFER(buffer, bufsz, &nr, sizeof(unsigned int));
+
+    return bufsz;
+}
+
+void
+entry_free_fn(void *data)
+{
+    changelog_opt_t *co = data;
+
+    if (!co)
+        return;
+
+    GF_FREE(co->co_entry.cef_bname);
+}
+
+void
+del_entry_free_fn(void *data)
+{
+    changelog_opt_t *co = data;
+
+    if (!co)
+        return;
+
+    GF_FREE(co->co_entry.cef_bname);
+    GF_FREE(co->co_entry.cef_path);
+}
+
+/**
+ * try to write all data in one shot
+ */
+
+static void
+changelog_encode_write_xtra(changelog_log_data_t *cld, char *buffer,
+                            size_t *off, gf_boolean_t encode)
+{
+    int i = 0;
+    size_t offset = 0;
+    void *data = NULL;
+    changelog_opt_t *co = NULL;
+
+    offset = *off;
+
+    co = (changelog_opt_t *)cld->cld_ptr;
+
+    for (; i < cld->cld_xtra_records; i++, co++) {
+        CHANGELOG_FILL_BUFFER(buffer, offset, "\0", 1);
+
+        switch (co->co_type) {
+            case CHANGELOG_OPT_REC_FOP:
+                data = &co->co_fop;
+                break;
+            case CHANGELOG_OPT_REC_ENTRY:
+                data = &co->co_entry;
+                break;
+            case CHANGELOG_OPT_REC_UINT32:
+                data = &co->co_uint32;
+                break;
+        }
+
+        if (co->co_convert)
+            offset += co->co_convert(data, buffer + offset, encode);
+        else /* no coversion: write it out as it is */
+            CHANGELOG_FILL_BUFFER(buffer, offset, data, co->co_len);
+    }
+
+    *off = offset;
+}
+
+int
+changelog_encode_ascii(xlator_t *this, changelog_log_data_t *cld)
+{
+    size_t off = 0;
+    size_t gfid_len = 0;
+    char *gfid_str = NULL;
+    char *buffer = NULL;
+    changelog_priv_t *priv = NULL;
+
+    priv = this->private;
+
+    gfid_str = uuid_utoa(cld->cld_gfid);
+    gfid_len = strlen(gfid_str);
+
+    /* extra bytes for decorations */
+    buffer = alloca(gfid_len + cld->cld_ptr_len + 10);
+    CHANGELOG_STORE_ASCII(priv, buffer, off, gfid_str, gfid_len, cld);
+
+    if (cld->cld_xtra_records)
+        changelog_encode_write_xtra(cld, buffer, &off, _gf_true);
+
+    CHANGELOG_FILL_BUFFER(buffer, off, "\0", 1);
+
+    return changelog_write_change(priv, buffer, off);
+}
+
+int
+changelog_encode_binary(xlator_t *this, changelog_log_data_t *cld)
+{
+    size_t off = 0;
+    char *buffer = NULL;
+    changelog_priv_t *priv = NULL;
+
+    priv = this->private;
+
+    /* extra bytes for decorations */
+    buffer = alloca(sizeof(uuid_t) + cld->cld_ptr_len + 10);
+    CHANGELOG_STORE_BINARY(priv, buffer, off, cld->cld_gfid, cld);
+
+    if (cld->cld_xtra_records)
+        changelog_encode_write_xtra(cld, buffer, &off, _gf_false);
+
+    CHANGELOG_FILL_BUFFER(buffer, off, "\0", 1);
+
+    return changelog_write_change(priv, buffer, off);
+}
+
+static struct changelog_encoder cb_encoder[] = {
+    [CHANGELOG_ENCODE_BINARY] =
+        {
+            .encoder = CHANGELOG_ENCODE_BINARY,
+            .encode = changelog_encode_binary,
+        },
+    [CHANGELOG_ENCODE_ASCII] =
+        {
+            .encoder = CHANGELOG_ENCODE_ASCII,
+            .encode = changelog_encode_ascii,
+        },
+};
+
+void
+changelog_encode_change(changelog_priv_t *priv)
+{
+    priv->ce = &cb_encoder[priv->encode_mode];
+}
diff --git a/xlators/features/changelog/src/changelog-encoders.h b/xlators/features/changelog/src/changelog-encoders.h
new file mode 100644
index 00000000000..26252696d56
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-encoders.h
@@ -0,0 +1,50 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef _CHANGELOG_ENCODERS_H
+#define _CHANGELOG_ENCODERS_H
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "changelog-helpers.h"
+
+#define CHANGELOG_STORE_ASCII(priv, buf, off, gfid, gfid_len, cld)             \
+    do {                                                                       \
+        CHANGELOG_FILL_BUFFER(buffer, off, priv->maps[cld->cld_type], 1);      \
+        CHANGELOG_FILL_BUFFER(buffer, off, gfid, gfid_len);                    \
+    } while (0)
+
+#define CHANGELOG_STORE_BINARY(priv, buf, off, gfid, cld)                      \
+    do {                                                                       \
+        CHANGELOG_FILL_BUFFER(buffer, off, priv->maps[cld->cld_type], 1);      \
+        CHANGELOG_FILL_BUFFER(buffer, off, gfid, sizeof(uuid_t));              \
+    } while (0)
+
+size_t
+entry_fn(void *data, char *buffer, gf_boolean_t encode);
+size_t
+del_entry_fn(void *data, char *buffer, gf_boolean_t encode);
+size_t
+fop_fn(void *data, char *buffer, gf_boolean_t encode);
+size_t
+number_fn(void *data, char *buffer, gf_boolean_t encode);
+void
+entry_free_fn(void *data);
+void
+del_entry_free_fn(void *data);
+int
+changelog_encode_binary(xlator_t *, changelog_log_data_t *);
+int
+changelog_encode_ascii(xlator_t *, changelog_log_data_t *);
+void
+changelog_encode_change(changelog_priv_t *);
+
+#endif /* _CHANGELOG_ENCODERS_H */
diff --git a/xlators/features/changelog/src/changelog-ev-handle.c b/xlators/features/changelog/src/changelog-ev-handle.c
new file mode 100644
index 00000000000..aa94459de5a
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-ev-handle.c
@@ -0,0 +1,412 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include "changelog-ev-handle.h"
+#include "changelog-rpc-common.h"
+#include "changelog-helpers.h"
+
+struct rpc_clnt_program changelog_ev_program;
+
+#define NR_IOVEC (MAX_IOVEC - 3)
+struct ev_rpc_vec {
+    int count;
+    struct iovec vector[NR_IOVEC];
+
+    /* sequence number */
+    unsigned long seq;
+};
+
+struct ev_rpc {
+    rbuf_list_t *rlist;
+    struct rpc_clnt *rpc;
+    struct ev_rpc_vec vec;
+};
+
+/**
+ * As of now this just does the minimal (retval logging). Going further
+ * un-acknowledges sequence numbers can be retransmitted and other
+ * intelligence can be built into the server.
+ */
+int
+changelog_event_dispatch_cbk(struct rpc_req *req, struct iovec *iov, int count,
+                             void *myframe)
+{
+    return 0;
+}
+
+/* dispatcher RPC */
+int
+changelog_dispatch_vec(call_frame_t *frame, xlator_t *this,
+                       struct rpc_clnt *rpc, struct ev_rpc_vec *vec)
+{
+    struct timeval tv = {
+        0,
+    };
+    changelog_event_req req = {
+        0,
+    };
+
+    (void)gettimeofday(&tv, NULL);
+
+    /**
+     * Event dispatch RPC header contains a sequence number for each
+     * dispatch. This allows the receiver to order the request before
+     * processing.
+     */
+    req.seq = vec->seq;
+    req.tv_sec = tv.tv_sec;
+    req.tv_usec = tv.tv_usec;
+
+    return changelog_rpc_sumbit_req(
+        rpc, (void *)&req, frame, &changelog_ev_program,
+        CHANGELOG_REV_PROC_EVENT, vec->vector, vec->count, NULL, this,
+        changelog_event_dispatch_cbk, (xdrproc_t)xdr_changelog_event_req);
+}
+
+int
+changelog_event_dispatch_rpc(call_frame_t *frame, xlator_t *this, void *data)
+{
+    int idx = 0;
+    int count = 0;
+    int ret = 0;
+    unsigned long sequence = 0;
+    rbuf_iovec_t *rvec = NULL;
+    struct ev_rpc *erpc = NULL;
+    struct rlist_iter riter = {
+        {
+            0,
+        },
+    };
+
+    /* dispatch NR_IOVEC IO vectors at a time. */
+
+    erpc = data;
+    sequence = erpc->rlist->seq[0];
+
+    rlist_iter_init(&riter, erpc->rlist);
+
+    rvec_for_each_entry(rvec, &riter)
+    {
+        idx = count % NR_IOVEC;
+        if (++count == NR_IOVEC) {
+            erpc->vec.vector[idx] = rvec->iov;
+            erpc->vec.seq = sequence++;
+            erpc->vec.count = NR_IOVEC;
+
+            ret = changelog_dispatch_vec(frame, this, erpc->rpc, &erpc->vec);
+            if (ret)
+                break;
+            count = 0;
+            continue;
+        }
+
+        erpc->vec.vector[idx] = rvec->iov;
+    }
+
+    if (ret)
+        goto error_return;
+
+    idx = count % NR_IOVEC;
+    if (idx) {
+        erpc->vec.seq = sequence;
+        erpc->vec.count = idx;
+
+        ret = changelog_dispatch_vec(frame, this, erpc->rpc, &erpc->vec);
+    }
+
+error_return:
+    return ret;
+}
+
+int
+changelog_rpc_notify(struct rpc_clnt *rpc, void *mydata, rpc_clnt_event_t event,
+                     void *data)
+{
+    xlator_t *this = NULL;
+    changelog_rpc_clnt_t *crpc = NULL;
+    changelog_clnt_t *c_clnt = NULL;
+    changelog_priv_t *priv = NULL;
+    changelog_ev_selector_t *selection = NULL;
+    uint64_t clntcnt = 0;
+    uint64_t xprtcnt = 0;
+
+    crpc = mydata;
+    this = crpc->this;
+    c_clnt = crpc->c_clnt;
+
+    priv = this->private;
+
+    switch (event) {
+        case RPC_CLNT_CONNECT:
+            selection = &priv->ev_selection;
+            GF_ATOMIC_INC(priv->clntcnt);
+
+            LOCK(&c_clnt->wait_lock);
+            {
+                LOCK(&c_clnt->active_lock);
+                {
+                    changelog_select_event(this, selection, crpc->filter);
+                    list_move_tail(&crpc->list, &c_clnt->active);
+                }
+                UNLOCK(&c_clnt->active_lock);
+            }
+            UNLOCK(&c_clnt->wait_lock);
+
+            break;
+        case RPC_CLNT_DISCONNECT:
+            rpc_clnt_disable(crpc->rpc);
+
+            /* rpc_clnt_disable doesn't unref the rpc. It just marks
+             * the rpc as disabled and cancels reconnection timer.
+             * Hence unref the rpc object to free it.
+             */
+            rpc_clnt_unref(crpc->rpc);
+
+            if (priv)
+                selection = &priv->ev_selection;
+
+            LOCK(&crpc->lock);
+            {
+                if (selection)
+                    changelog_deselect_event(this, selection, crpc->filter);
+                changelog_set_disconnect_flag(crpc, _gf_true);
+            }
+            UNLOCK(&crpc->lock);
+            LOCK(&c_clnt->active_lock);
+            {
+                list_del_init(&crpc->list);
+            }
+            UNLOCK(&c_clnt->active_lock);
+
+            break;
+        case RPC_CLNT_MSG:
+        case RPC_CLNT_DESTROY:
+            /* Free up mydata */
+            changelog_rpc_clnt_unref(crpc);
+            clntcnt = GF_ATOMIC_DEC(priv->clntcnt);
+            xprtcnt = GF_ATOMIC_GET(priv->xprtcnt);
+            if (this->cleanup_starting) {
+                if (!clntcnt && !xprtcnt)
+                    changelog_process_cleanup_event(this);
+            }
+            break;
+        case RPC_CLNT_PING:
+            break;
+    }
+
+    return 0;
+}
+
+void *
+changelog_ev_connector(void *data)
+{
+    xlator_t *this = NULL;
+    changelog_clnt_t *c_clnt = NULL;
+    changelog_rpc_clnt_t *crpc = NULL;
+
+    c_clnt = data;
+    this = c_clnt->this;
+
+    while (1) {
+        pthread_mutex_lock(&c_clnt->pending_lock);
+        {
+            while (list_empty(&c_clnt->pending))
+                pthread_cond_wait(&c_clnt->pending_cond, &c_clnt->pending_lock);
+            crpc = list_first_entry(&c_clnt->pending, changelog_rpc_clnt_t,
+                                    list);
+            crpc->rpc = changelog_rpc_client_init(this, crpc, crpc->sock,
+                                                  changelog_rpc_notify);
+            if (!crpc->rpc) {
+                gf_smsg(this->name, GF_LOG_ERROR, 0,
+                        CHANGELOG_MSG_RPC_CONNECT_ERROR, "path=%s", crpc->sock,
+                        NULL);
+                crpc->cleanup(crpc);
+                goto mutex_unlock;
+            }
+
+            LOCK(&c_clnt->wait_lock);
+            {
+                list_move_tail(&crpc->list, &c_clnt->waitq);
+            }
+            UNLOCK(&c_clnt->wait_lock);
+        }
+    mutex_unlock:
+        pthread_mutex_unlock(&c_clnt->pending_lock);
+    }
+
+    return NULL;
+}
+
+void
+changelog_ev_cleanup_connections(xlator_t *this, changelog_clnt_t *c_clnt)
+{
+    changelog_rpc_clnt_t *crpc = NULL;
+
+    /* cleanup active connections */
+    LOCK(&c_clnt->active_lock);
+    {
+        list_for_each_entry(crpc, &c_clnt->active, list)
+        {
+            rpc_clnt_disable(crpc->rpc);
+        }
+    }
+    UNLOCK(&c_clnt->active_lock);
+}
+
+/**
+ * TODO: granularize lock
+ *
+ * If we have multiple threads dispatching events, doing it this way is
+ * a performance bottleneck.
+ */
+
+static changelog_rpc_clnt_t *
+get_client(changelog_clnt_t *c_clnt, struct list_head **next)
+{
+    changelog_rpc_clnt_t *crpc = NULL;
+
+    LOCK(&c_clnt->active_lock);
+    {
+        if (*next == &c_clnt->active)
+            goto unblock;
+        crpc = list_entry(*next, changelog_rpc_clnt_t, list);
+        /* ref rpc as DISCONNECT might unref the rpc asynchronously */
+        changelog_rpc_clnt_ref(crpc);
+        rpc_clnt_ref(crpc->rpc);
+        *next = (*next)->next;
+    }
+unblock:
+    UNLOCK(&c_clnt->active_lock);
+
+    return crpc;
+}
+
+static void
+put_client(changelog_clnt_t *c_clnt, changelog_rpc_clnt_t *crpc)
+{
+    LOCK(&c_clnt->active_lock);
+    {
+        rpc_clnt_unref(crpc->rpc);
+        changelog_rpc_clnt_unref(crpc);
+    }
+    UNLOCK(&c_clnt->active_lock);
+}
+
+void
+_dispatcher(rbuf_list_t *rlist, void *arg)
+{
+    xlator_t *this = NULL;
+    changelog_clnt_t *c_clnt = NULL;
+    changelog_rpc_clnt_t *crpc = NULL;
+    struct ev_rpc erpc = {
+        0,
+    };
+    struct list_head *next = NULL;
+
+    c_clnt = arg;
+    this = c_clnt->this;
+
+    erpc.rlist = rlist;
+    next = c_clnt->active.next;
+
+    while (1) {
+        crpc = get_client(c_clnt, &next);
+        if (!crpc)
+            break;
+        erpc.rpc = crpc->rpc;
+        (void)changelog_invoke_rpc(this, crpc->rpc, &changelog_ev_program,
+                                   CHANGELOG_REV_PROC_EVENT, &erpc);
+        put_client(c_clnt, crpc);
+    }
+}
+
+/** this is called under rotbuff's lock */
+void
+sequencer(rbuf_list_t *rlist, void *mydata)
+{
+    unsigned long range = 0;
+    changelog_clnt_t *c_clnt = 0;
+
+    c_clnt = mydata;
+
+    range = (RLIST_ENTRY_COUNT(rlist)) / NR_IOVEC;
+    if ((RLIST_ENTRY_COUNT(rlist)) % NR_IOVEC)
+        range++;
+    RLIST_STORE_SEQ(rlist, c_clnt->sequence, range);
+
+    c_clnt->sequence += range;
+}
+
+void *
+changelog_ev_dispatch(void *data)
+{
+    int ret = 0;
+    void *opaque = NULL;
+    xlator_t *this = NULL;
+    changelog_clnt_t *c_clnt = NULL;
+    struct timeval tv = {
+        0,
+    };
+
+    c_clnt = data;
+    this = c_clnt->this;
+
+    while (1) {
+        /* TODO: change this to be pthread cond based.. later */
+
+        tv.tv_sec = 1;
+        tv.tv_usec = 0;
+        select(0, NULL, NULL, NULL, &tv);
+
+        ret = rbuf_get_buffer(c_clnt->rbuf, &opaque, sequencer, c_clnt);
+        if (ret != RBUF_CONSUMABLE) {
+            if (ret != RBUF_EMPTY)
+                gf_smsg(this->name, GF_LOG_WARNING, 0,
+                        CHANGELOG_MSG_BUFFER_STARVATION_ERROR,
+                        "Failed to get buffer for RPC dispatch",
+                        "rbuf_retval=%d", ret, NULL);
+            continue;
+        }
+
+        ret = rbuf_wait_for_completion(c_clnt->rbuf, opaque, _dispatcher,
+                                       c_clnt);
+        if (ret)
+            gf_smsg(this->name, GF_LOG_WARNING, 0,
+                    CHANGELOG_MSG_PUT_BUFFER_FAILED, NULL);
+    }
+
+    return NULL;
+}
+
+void
+changelog_ev_queue_connection(changelog_clnt_t *c_clnt,
+                              changelog_rpc_clnt_t *crpc)
+{
+    pthread_mutex_lock(&c_clnt->pending_lock);
+    {
+        list_add_tail(&crpc->list, &c_clnt->pending);
+        pthread_cond_signal(&c_clnt->pending_cond);
+    }
+    pthread_mutex_unlock(&c_clnt->pending_lock);
+}
+
+struct rpc_clnt_procedure changelog_ev_procs[CHANGELOG_REV_PROC_MAX] = {
+    [CHANGELOG_REV_PROC_NULL] = {"NULL", NULL},
+    [CHANGELOG_REV_PROC_EVENT] = {"EVENT DISPATCH",
+                                  changelog_event_dispatch_rpc},
+};
+
+struct rpc_clnt_program changelog_ev_program = {
+    .progname = "CHANGELOG EVENT DISPATCHER",
+    .prognum = CHANGELOG_REV_RPC_PROCNUM,
+    .progver = CHANGELOG_REV_RPC_PROCVER,
+    .numproc = CHANGELOG_REV_PROC_MAX,
+    .proctable = changelog_ev_procs,
+};
diff --git a/xlators/features/changelog/src/changelog-ev-handle.h b/xlators/features/changelog/src/changelog-ev-handle.h
new file mode 100644
index 00000000000..cc1af58a276
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-ev-handle.h
@@ -0,0 +1,136 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __CHANGELOG_EV_HANDLE_H
+#define __CHANGELOG_EV_HANDLE_H
+
+#include <glusterfs/list.h>
+#include <glusterfs/xlator.h>
+#include "rpc-clnt.h"
+
+#include <glusterfs/rot-buffs.h>
+
+struct changelog_clnt;
+
+typedef struct changelog_rpc_clnt {
+    xlator_t *this;
+
+    gf_lock_t lock;
+
+    gf_atomic_t ref;
+    gf_boolean_t disconnected;
+
+    unsigned int filter;
+    char sock[UNIX_PATH_MAX];
+
+    struct changelog_clnt *c_clnt; /* back pointer to list holder */
+
+    struct rpc_clnt *rpc; /* RPC client endpoint */
+
+    struct list_head list; /* ->pending, ->waitq, ->active */
+
+    void (*cleanup)(struct changelog_rpc_clnt *); /* cleanup handler */
+} changelog_rpc_clnt_t;
+
+static inline void
+changelog_rpc_clnt_ref(changelog_rpc_clnt_t *crpc)
+{
+    GF_ATOMIC_INC(crpc->ref);
+}
+
+static inline void
+changelog_set_disconnect_flag(changelog_rpc_clnt_t *crpc, gf_boolean_t flag)
+{
+    crpc->disconnected = flag;
+}
+
+static inline int
+changelog_rpc_clnt_is_disconnected(changelog_rpc_clnt_t *crpc)
+{
+    return (crpc->disconnected == _gf_true);
+}
+
+static inline void
+changelog_rpc_clnt_unref(changelog_rpc_clnt_t *crpc)
+{
+    gf_boolean_t gone = _gf_false;
+    uint64_t ref = 0;
+
+    ref = GF_ATOMIC_DEC(crpc->ref);
+
+    if (!ref && changelog_rpc_clnt_is_disconnected(crpc)) {
+        list_del(&crpc->list);
+        gone = _gf_true;
+    }
+
+    if (gone)
+        crpc->cleanup(crpc);
+}
+
+/**
+ * This structure holds pending and active clients. On probe RPC all
+ * an instance of the above structure (@changelog_rpc_clnt) is placed
+ * in ->pending and gets moved to ->active on a successful connect.
+ *
+ * locking rules:
+ *
+ * Manipulating ->pending
+ * ->pending_lock
+ *    ->pending
+ *
+ * Manipulating ->active
+ * ->active_lock
+ *    ->active
+ *
+ * Moving object from ->pending to ->active
+ * ->pending_lock
+ *   ->active_lock
+ *
+ * Objects are _never_ moved from ->active to ->pending, i.e., during
+ * disconnection, the object is destroyed. Well, we could have tried
+ * to reconnect, but that's pure waste.. let the other end reconnect.
+ */
+
+typedef struct changelog_clnt {
+    xlator_t *this;
+
+    /* pending connections */
+    pthread_mutex_t pending_lock;
+    pthread_cond_t pending_cond;
+    struct list_head pending;
+
+    /* current active connections */
+    gf_lock_t active_lock;
+    struct list_head active;
+
+    gf_lock_t wait_lock;
+    struct list_head waitq;
+
+    /* consumer part of rot-buffs */
+    rbuf_t *rbuf;
+    unsigned long sequence;
+} changelog_clnt_t;
+
+void *
+changelog_ev_connector(void *);
+
+void *
+changelog_ev_dispatch(void *);
+
+/* APIs */
+void
+changelog_ev_queue_connection(changelog_clnt_t *, changelog_rpc_clnt_t *);
+
+void
+changelog_ev_cleanup_connections(xlator_t *, changelog_clnt_t *);
+
+void
+changelog_process_cleanup_event(xlator_t *);
+#endif
diff --git a/xlators/features/changelog/src/changelog-helpers.c b/xlators/features/changelog/src/changelog-helpers.c
new file mode 100644
index 00000000000..e561997d858
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-helpers.c
@@ -0,0 +1,1977 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/iobuf.h>
+#include <glusterfs/syscall.h>
+
+#include "changelog-helpers.h"
+#include "changelog-encoders.h"
+#include "changelog-mem-types.h"
+#include "changelog-messages.h"
+
+#include "changelog-encoders.h"
+#include "changelog-rpc-common.h"
+#include <pthread.h>
+#include <time.h>
+
+static void
+changelog_cleanup_free_mutex(void *arg_mutex)
+{
+    pthread_mutex_t *p_mutex = (pthread_mutex_t *)arg_mutex;
+
+    if (p_mutex)
+        pthread_mutex_unlock(p_mutex);
+}
+
+int
+changelog_thread_cleanup(xlator_t *this, pthread_t thr_id)
+{
+    int ret = 0;
+    void *retval = NULL;
+
+    /* send a cancel request to the thread */
+    ret = pthread_cancel(thr_id);
+    if (ret != 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                CHANGELOG_MSG_PTHREAD_CANCEL_FAILED, NULL);
+        goto out;
+    }
+
+    ret = pthread_join(thr_id, &retval);
+    if ((ret != 0) || (retval != PTHREAD_CANCELED)) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                CHANGELOG_MSG_PTHREAD_CANCEL_FAILED, NULL);
+    }
+
+out:
+    return ret;
+}
+
+void *
+changelog_get_usable_buffer(changelog_local_t *local)
+{
+    changelog_log_data_t *cld = NULL;
+
+    if (!local)
+        return NULL;
+
+    cld = &local->cld;
+    if (!cld->cld_iobuf)
+        return NULL;
+
+    return cld->cld_iobuf->ptr;
+}
+
+static int
+changelog_selector_index(unsigned int selector)
+{
+    return (ffs(selector) - 1);
+}
+
+int
+changelog_ev_selected(xlator_t *this, changelog_ev_selector_t *selection,
+                      unsigned int selector)
+{
+    int idx = 0;
+
+    idx = changelog_selector_index(selector);
+    gf_msg_debug(this->name, 0, "selector ref count for %d (idx: %d): %d",
+                 selector, idx, selection->ref[idx]);
+    /* this can be lockless */
+    return (idx < CHANGELOG_EV_SELECTION_RANGE && (selection->ref[idx] > 0));
+}
+
+void
+changelog_select_event(xlator_t *this, changelog_ev_selector_t *selection,
+                       unsigned int selector)
+{
+    int idx = 0;
+
+    LOCK(&selection->reflock);
+    {
+        while (selector) {
+            idx = changelog_selector_index(selector);
+            if (idx < CHANGELOG_EV_SELECTION_RANGE) {
+                selection->ref[idx]++;
+                gf_msg_debug(this->name, 0, "selecting event %d", idx);
+            }
+            selector &= ~(1 << idx);
+        }
+    }
+    UNLOCK(&selection->reflock);
+}
+
+void
+changelog_deselect_event(xlator_t *this, changelog_ev_selector_t *selection,
+                         unsigned int selector)
+{
+    int idx = 0;
+
+    LOCK(&selection->reflock);
+    {
+        while (selector) {
+            idx = changelog_selector_index(selector);
+            if (idx < CHANGELOG_EV_SELECTION_RANGE) {
+                selection->ref[idx]--;
+                gf_msg_debug(this->name, 0, "de-selecting event %d", idx);
+            }
+            selector &= ~(1 << idx);
+        }
+    }
+    UNLOCK(&selection->reflock);
+}
+
+int
+changelog_init_event_selection(xlator_t *this,
+                               changelog_ev_selector_t *selection)
+{
+    int ret = 0;
+    int j = CHANGELOG_EV_SELECTION_RANGE;
+
+    ret = LOCK_INIT(&selection->reflock);
+    if (ret != 0)
+        return -1;
+
+    LOCK(&selection->reflock);
+    {
+        while (j--) {
+            selection->ref[j] = 0;
+        }
+    }
+    UNLOCK(&selection->reflock);
+
+    return 0;
+}
+
+static void
+changelog_perform_dispatch(xlator_t *this, changelog_priv_t *priv, void *mem,
+                           size_t size)
+{
+    char *buf = NULL;
+    void *opaque = NULL;
+
+    buf = rbuf_reserve_write_area(priv->rbuf, size, &opaque);
+    if (!buf) {
+        gf_msg_callingfn(this->name, GF_LOG_WARNING, 0,
+                         CHANGELOG_MSG_DISPATCH_EVENT_FAILED,
+                         "failed to dispatch event");
+        return;
+    }
+
+    memcpy(buf, mem, size);
+    rbuf_write_complete(opaque);
+}
+
+void
+changelog_dispatch_event(xlator_t *this, changelog_priv_t *priv,
+                         changelog_event_t *ev)
+{
+    changelog_ev_selector_t *selection = NULL;
+
+    selection = &priv->ev_selection;
+    if (changelog_ev_selected(this, selection, ev->ev_type)) {
+        changelog_perform_dispatch(this, priv, ev, CHANGELOG_EV_SIZE);
+    }
+}
+
+void
+changelog_set_usable_record_and_length(changelog_local_t *local, size_t len,
+                                       int xr)
+{
+    changelog_log_data_t *cld = NULL;
+
+    cld = &local->cld;
+
+    cld->cld_ptr_len = len;
+    cld->cld_xtra_records = xr;
+}
+
+void
+changelog_local_cleanup(xlator_t *xl, changelog_local_t *local)
+{
+    int i = 0;
+    changelog_opt_t *co = NULL;
+    changelog_log_data_t *cld = NULL;
+
+    if (!local)
+        return;
+
+    cld = &local->cld;
+
+    /* cleanup dynamic allocation for extra records */
+    if (cld->cld_xtra_records) {
+        co = (changelog_opt_t *)cld->cld_ptr;
+        for (; i < cld->cld_xtra_records; i++, co++)
+            if (co->co_free)
+                co->co_free(co);
+    }
+
+    CHANGELOG_IOBUF_UNREF(cld->cld_iobuf);
+
+    if (local->inode)
+        inode_unref(local->inode);
+
+    mem_put(local);
+}
+
+int
+changelog_write(int fd, char *buffer, size_t len)
+{
+    ssize_t size = 0;
+    size_t written = 0;
+
+    while (written < len) {
+        size = sys_write(fd, buffer + written, len - written);
+        if (size <= 0)
+            break;
+
+        written += size;
+    }
+
+    return (written != len);
+}
+
+int
+htime_update(xlator_t *this, changelog_priv_t *priv, time_t ts, char *buffer)
+{
+    char changelog_path[PATH_MAX + 1] = {
+        0,
+    };
+    int len = -1;
+    char x_value[25] = {
+        0,
+    };
+    /* time stamp(10) + : (1) + rolltime (12 ) + buffer (2) */
+    int ret = 0;
+
+    if (priv->htime_fd == -1) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_MSG_HTIME_ERROR,
+                "reason=fd not available", NULL);
+        ret = -1;
+        goto out;
+    }
+    len = snprintf(changelog_path, PATH_MAX, "%s", buffer);
+    if (len >= PATH_MAX) {
+        ret = -1;
+        goto out;
+    }
+    if (changelog_write(priv->htime_fd, (void *)changelog_path, len + 1) < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_MSG_HTIME_ERROR,
+                "reason=write failed", NULL);
+        ret = -1;
+        goto out;
+    }
+
+    len = snprintf(x_value, sizeof(x_value), "%ld:%d", ts,
+                   priv->rollover_count);
+    if (len >= sizeof(x_value)) {
+        ret = -1;
+        goto out;
+    }
+
+    if (sys_fsetxattr(priv->htime_fd, HTIME_KEY, x_value, len, XATTR_REPLACE)) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_HTIME_ERROR,
+                "reason=xattr updation failed", "XATTR_REPLACE=true",
+                "changelog=%s", changelog_path, NULL);
+
+        if (sys_fsetxattr(priv->htime_fd, HTIME_KEY, x_value, len, 0)) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_HTIME_ERROR,
+                    "reason=xattr updation failed", "changelog=%s",
+                    changelog_path, NULL);
+            ret = -1;
+            goto out;
+        }
+    }
+
+    priv->rollover_count += 1;
+
+out:
+    return ret;
+}
+
+/*
+ * Description: Check if the changelog to rollover is empty or not.
+ * It is assumed that fd passed is already verified.
+ *
+ * Returns:
+ * 1 : If found empty, changed path from "CHANGELOG.<TS>" to "changelog.<TS>"
+ * 0 : If NOT empty, proceed usual.
+ */
+int
+cl_is_empty(xlator_t *this, int fd)
+{
+    int ret = -1;
+    size_t elen = 0;
+    int encoding = -1;
+    char buffer[1024] = {
+        0,
+    };
+    struct stat stbuf = {
+        0,
+    };
+    int major_version = -1;
+    int minor_version = -1;
+
+    ret = sys_fstat(fd, &stbuf);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_FSTAT_OP_FAILED,
+                NULL);
+        goto out;
+    }
+
+    ret = sys_lseek(fd, 0, SEEK_SET);
+    if (ret == -1) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_LSEEK_OP_FAILED,
+                NULL);
+        goto out;
+    }
+
+    CHANGELOG_GET_HEADER_INFO(fd, buffer, sizeof(buffer), encoding,
+                              major_version, minor_version, elen);
+
+    if (elen == stbuf.st_size) {
+        ret = 1;
+    } else {
+        ret = 0;
+    }
+
+out:
+    return ret;
+}
+
+/*
+ * Description: Updates "CHANGELOG" to "changelog" for writing changelog path
+ * to htime file.
+ *
+ * Returns:
+ * 0  : Success
+ * -1 : Error
+ */
+int
+update_path(xlator_t *this, char *cl_path)
+{
+    const char low_cl[] = "changelog";
+    const char up_cl[] = "CHANGELOG";
+    char *found = NULL;
+    int ret = -1;
+
+    found = strstr(cl_path, up_cl);
+
+    if (found == NULL) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_PATH_NOT_FOUND,
+                NULL);
+        goto out;
+    } else {
+        memcpy(found, low_cl, sizeof(low_cl) - 1);
+    }
+
+    ret = 0;
+out:
+    return ret;
+}
+
+static int
+changelog_rollover_changelog(xlator_t *this, changelog_priv_t *priv, time_t ts)
+{
+    int ret = -1;
+    int notify = 0;
+    int cl_empty_flag = 0;
+    struct tm *gmt;
+    char yyyymmdd[40];
+    char ofile[PATH_MAX] = {
+        0,
+    };
+    char nfile[PATH_MAX] = {
+        0,
+    };
+    char nfile_dir[PATH_MAX] = {
+        0,
+    };
+    changelog_event_t ev = {
+        0,
+    };
+
+    if (priv->changelog_fd != -1) {
+        ret = sys_fsync(priv->changelog_fd);
+        if (ret < 0) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno,
+                    CHANGELOG_MSG_FSYNC_OP_FAILED, NULL);
+        }
+        ret = cl_is_empty(this, priv->changelog_fd);
+        if (ret == 1) {
+            cl_empty_flag = 1;
+        } else if (ret == -1) {
+            /* Log error but proceed as usual */
+            gf_smsg(this->name, GF_LOG_WARNING, 0,
+                    CHANGELOG_MSG_DETECT_EMPTY_CHANGELOG_FAILED, NULL);
+        }
+        sys_close(priv->changelog_fd);
+        priv->changelog_fd = -1;
+    }
+
+    /* Get GMT time. */
+    gmt = gmtime(&ts);
+
+    strftime(yyyymmdd, sizeof(yyyymmdd), "%Y/%m/%d", gmt);
+
+    (void)snprintf(ofile, PATH_MAX, "%s/" CHANGELOG_FILE_NAME,
+                   priv->changelog_dir);
+    (void)snprintf(nfile, PATH_MAX, "%s/%s/" CHANGELOG_FILE_NAME ".%ld",
+                   priv->changelog_dir, yyyymmdd, ts);
+    (void)snprintf(nfile_dir, PATH_MAX, "%s/%s", priv->changelog_dir, yyyymmdd);
+
+    if (cl_empty_flag == 1) {
+        ret = sys_unlink(ofile);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno,
+                    CHANGELOG_MSG_UNLINK_OP_FAILED, "path=%s", ofile, NULL);
+            ret = 0; /* Error in unlinking empty changelog should
+                        not break further changelog operation, so
+                        reset return value to 0*/
+        }
+    } else {
+        ret = sys_rename(ofile, nfile);
+
+        /* Changelog file rename gets ENOENT when parent dir doesn't exist */
+        if (errno == ENOENT) {
+            ret = mkdir_p(nfile_dir, 0600, _gf_true);
+
+            if ((ret == -1) && (EEXIST != errno)) {
+                gf_smsg(this->name, GF_LOG_ERROR, errno,
+                        CHANGELOG_MSG_MKDIR_ERROR, "%s", nfile_dir, NULL);
+                goto out;
+            }
+
+            ret = sys_rename(ofile, nfile);
+        }
+
+        if (ret && (errno == ENOENT)) {
+            ret = 0;
+            goto out;
+        }
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_RENAME_ERROR,
+                    "from=%s", ofile, "to=%s", nfile, NULL);
+        }
+    }
+
+    if (!ret && (cl_empty_flag == 0)) {
+        notify = 1;
+    }
+
+    if (!ret) {
+        if (cl_empty_flag) {
+            update_path(this, nfile);
+        }
+        ret = htime_update(this, priv, ts, nfile);
+        if (ret == -1) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_MSG_HTIME_ERROR,
+                    NULL);
+            goto out;
+        }
+    }
+
+    if (notify) {
+        ev.ev_type = CHANGELOG_OP_TYPE_JOURNAL;
+        memcpy(ev.u.journal.path, nfile, strlen(nfile) + 1);
+        changelog_dispatch_event(this, priv, &ev);
+    }
+out:
+    /* If this is explicit rollover initiated by snapshot,
+     * wakeup reconfigure thread waiting for changelog to
+     * rollover. This should happen even in failure cases as
+     * well otherwise snapshot will timeout and fail. Hence
+     * moved under out.
+     */
+    if (priv->explicit_rollover) {
+        priv->explicit_rollover = _gf_false;
+
+        pthread_mutex_lock(&priv->bn.bnotify_mutex);
+        {
+            if (ret) {
+                priv->bn.bnotify_error = _gf_true;
+                gf_smsg(this->name, GF_LOG_ERROR, 0,
+                        CHANGELOG_MSG_EXPLICIT_ROLLOVER_FAILED, NULL);
+            } else {
+                gf_smsg(this->name, GF_LOG_INFO, 0, CHANGELOG_MSG_BNOTIFY_INFO,
+                        "changelog=%s", nfile, NULL);
+            }
+            priv->bn.bnotify = _gf_false;
+            pthread_cond_signal(&priv->bn.bnotify_cond);
+        }
+        pthread_mutex_unlock(&priv->bn.bnotify_mutex);
+    }
+    return ret;
+}
+
+int
+filter_cur_par_dirs(const struct dirent *entry)
+{
+    if (entry == NULL)
+        return 0;
+
+    if ((strcmp(entry->d_name, ".") == 0) || (strcmp(entry->d_name, "..") == 0))
+        return 0;
+    else
+        return 1;
+}
+
+/*
+ * find_current_htime:
+ *       It finds the latest htime file and sets the HTIME_CURRENT
+ *       xattr.
+ *       RETURN VALUE:
+ *           -1 : Error
+ *           ret: Number of directory entries;
+ */
+
+int
+find_current_htime(int ht_dir_fd, const char *ht_dir_path, char *ht_file_bname)
+{
+    struct dirent **namelist = NULL;
+    int ret = 0;
+    int cnt = 0;
+    int i = 0;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+    GF_ASSERT(ht_dir_path);
+
+    cnt = scandir(ht_dir_path, &namelist, filter_cur_par_dirs, alphasort);
+    if (cnt < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_SCAN_DIR_FAILED,
+                NULL);
+    } else if (cnt > 0) {
+        if (snprintf(ht_file_bname, NAME_MAX, "%s",
+                     namelist[cnt - 1]->d_name) >= NAME_MAX) {
+            ret = -1;
+            goto out;
+        }
+        if (sys_fsetxattr(ht_dir_fd, HTIME_CURRENT, ht_file_bname,
+                          strlen(ht_file_bname), 0)) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno,
+                    CHANGELOG_MSG_FSETXATTR_FAILED, "HTIME_CURRENT", NULL);
+            ret = -1;
+            goto out;
+        }
+
+        if (sys_fsync(ht_dir_fd) < 0) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno,
+                    CHANGELOG_MSG_FSYNC_OP_FAILED, NULL);
+            ret = -1;
+            goto out;
+        }
+    }
+
+out:
+    for (i = 0; i < cnt; i++)
+        free(namelist[i]);
+    free(namelist);
+
+    if (ret)
+        cnt = ret;
+
+    return cnt;
+}
+
+/* Returns 0 on successful open of htime file
+ * returns -1 on failure or error
+ */
+int
+htime_open(xlator_t *this, changelog_priv_t *priv, time_t ts)
+{
+    int ht_file_fd = -1;
+    int ht_dir_fd = -1;
+    int ret = 0;
+    int cnt = 0;
+    char ht_dir_path[PATH_MAX] = {
+        0,
+    };
+    char ht_file_path[PATH_MAX] = {
+        0,
+    };
+    char ht_file_bname[NAME_MAX] = {
+        0,
+    };
+    char x_value[NAME_MAX] = {
+        0,
+    };
+    int flags = 0;
+    unsigned long min_ts = 0;
+    unsigned long max_ts = 0;
+    unsigned long total = 0;
+    unsigned long total1 = 0;
+    ssize_t size = 0;
+    struct stat stat_buf = {
+        0,
+    };
+    unsigned long record_len = 0;
+    int32_t len = 0;
+
+    CHANGELOG_FILL_HTIME_DIR(priv->changelog_dir, ht_dir_path);
+
+    /* Open htime directory to get HTIME_CURRENT */
+    ht_dir_fd = open(ht_dir_path, O_RDONLY);
+    if (ht_dir_fd == -1) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_OPEN_FAILED,
+                "path=%s", ht_dir_path, NULL);
+        ret = -1;
+        goto out;
+    }
+
+    size = sys_fgetxattr(ht_dir_fd, HTIME_CURRENT, ht_file_bname,
+                         sizeof(ht_file_bname));
+    if (size < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_FGETXATTR_FAILED,
+                "name=HTIME_CURRENT", NULL);
+
+        /* If upgrade scenario, find the latest HTIME.TSTAMP file
+         * and use the same. If error, create a new HTIME.TSTAMP
+         * file.
+         */
+        cnt = find_current_htime(ht_dir_fd, ht_dir_path, ht_file_bname);
+        if (cnt <= 0) {
+            gf_smsg(this->name, GF_LOG_INFO, errno,
+                    CHANGELOG_MSG_NO_HTIME_CURRENT, NULL);
+            sys_close(ht_dir_fd);
+            return htime_create(this, priv, ts);
+        }
+
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                CHANGELOG_MSG_HTIME_CURRENT_ERROR, NULL);
+    }
+
+    gf_smsg(this->name, GF_LOG_INFO, 0, CHANGELOG_MSG_HTIME_CURRENT, "path=%s",
+            ht_file_bname, NULL);
+    len = snprintf(ht_file_path, PATH_MAX, "%s/%s", ht_dir_path, ht_file_bname);
+    if ((len < 0) || (len >= PATH_MAX)) {
+        ret = -1;
+        goto out;
+    }
+
+    /* Open in append mode as existing htime file is used */
+    flags |= (O_RDWR | O_SYNC | O_APPEND);
+    ht_file_fd = open(ht_file_path, flags,
+                      S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
+    if (ht_file_fd < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_OPEN_FAILED,
+                "path=%s", ht_file_path, NULL);
+        ret = -1;
+        goto out;
+    }
+
+    /* save this htime_fd in priv->htime_fd */
+    priv->htime_fd = ht_file_fd;
+
+    ret = sys_fstat(ht_file_fd, &stat_buf);
+    if (ret < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_HTIME_STAT_ERROR,
+                "path=%s", ht_file_path, NULL);
+        ret = -1;
+        goto out;
+    }
+
+    /* Initialize rollover-number in priv to current number */
+    size = sys_fgetxattr(ht_file_fd, HTIME_KEY, x_value, sizeof(x_value));
+    if (size < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_FGETXATTR_FAILED,
+                "name=%s", HTIME_KEY, "path=%s", ht_file_path, NULL);
+        ret = -1;
+        goto out;
+    }
+
+    sscanf(x_value, "%lu:%lu", &max_ts, &total);
+
+    /* 22 = 1(/) + 20(CHANGELOG.TIMESTAMP) + 1(\x00) */
+    record_len = strlen(priv->changelog_dir) + 22;
+    total1 = stat_buf.st_size / record_len;
+    if (total != total1) {
+        gf_smsg(this->name, GF_LOG_INFO, 0, CHANGELOG_MSG_TOTAL_LOG_INFO,
+                "xattr_total=%lu", total, "size_total=%lu", total1, NULL);
+    }
+
+    gf_smsg(this->name, GF_LOG_INFO, 0, CHANGELOG_MSG_TOTAL_LOG_INFO, "min=%lu",
+            min_ts, "max=%lu", max_ts, "total_changelogs=%lu", total, NULL);
+
+    if (total < total1)
+        priv->rollover_count = total1 + 1;
+    else
+        priv->rollover_count = total + 1;
+
+out:
+    if (ht_dir_fd != -1)
+        sys_close(ht_dir_fd);
+    return ret;
+}
+
+/* Returns 0 on successful creation of htime file
+ * returns -1 on failure or error
+ */
+int
+htime_create(xlator_t *this, changelog_priv_t *priv, time_t ts)
+{
+    int ht_file_fd = -1;
+    int ht_dir_fd = -1;
+    int ret = 0;
+    char ht_dir_path[PATH_MAX] = {
+        0,
+    };
+    char ht_file_path[PATH_MAX] = {
+        0,
+    };
+    char ht_file_bname[NAME_MAX + 1] = {
+        0,
+    };
+    int flags = 0;
+    int32_t len = 0;
+
+    gf_smsg(this->name, GF_LOG_INFO, 0, CHANGELOG_MSG_NEW_HTIME_FILE,
+            "name=%ld", ts, NULL);
+
+    CHANGELOG_FILL_HTIME_DIR(priv->changelog_dir, ht_dir_path);
+
+    /* get the htime file name in ht_file_path */
+    len = snprintf(ht_file_path, PATH_MAX, "%s/%s.%ld", ht_dir_path,
+                   HTIME_FILE_NAME, ts);
+    if ((len < 0) || (len >= PATH_MAX)) {
+        ret = -1;
+        goto out;
+    }
+
+    flags |= (O_CREAT | O_RDWR | O_SYNC);
+    ht_file_fd = open(ht_file_path, flags,
+                      S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
+    if (ht_file_fd < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_OPEN_FAILED,
+                "path=%s", ht_file_path, NULL);
+        ret = -1;
+        goto out;
+    }
+
+    if (sys_fsetxattr(ht_file_fd, HTIME_KEY, HTIME_INITIAL_VALUE,
+                      sizeof(HTIME_INITIAL_VALUE) - 1, 0)) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                CHANGELOG_MSG_XATTR_INIT_FAILED, NULL);
+        ret = -1;
+        goto out;
+    }
+
+    ret = sys_fsync(ht_file_fd);
+    if (ret < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_FSYNC_OP_FAILED,
+                NULL);
+        goto out;
+    }
+
+    /* save this htime_fd in priv->htime_fd */
+    priv->htime_fd = ht_file_fd;
+
+    ht_file_fd = -1;
+
+    /* Set xattr HTIME_CURRENT on htime directory to htime filename */
+    ht_dir_fd = open(ht_dir_path, O_RDONLY);
+    if (ht_dir_fd == -1) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_OPEN_FAILED,
+                "path=%s", ht_dir_path, NULL);
+        ret = -1;
+        goto out;
+    }
+
+    (void)snprintf(ht_file_bname, sizeof(ht_file_bname), "%s.%ld",
+                   HTIME_FILE_NAME, ts);
+    if (sys_fsetxattr(ht_dir_fd, HTIME_CURRENT, ht_file_bname,
+                      strlen(ht_file_bname), 0)) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_FSETXATTR_FAILED,
+                " HTIME_CURRENT", NULL);
+        ret = -1;
+        goto out;
+    }
+
+    ret = sys_fsync(ht_dir_fd);
+    if (ret < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_FSYNC_OP_FAILED,
+                NULL);
+        goto out;
+    }
+
+    /* initialize rollover-number in priv to 1 */
+    priv->rollover_count = 1;
+
+out:
+    if (ht_dir_fd != -1)
+        sys_close(ht_dir_fd);
+    if (ht_file_fd != -1)
+        sys_close(ht_file_fd);
+    return ret;
+}
+
+/* Description:
+ *      Opens the snap changelog to log call path fops in it.
+ *      This changelos name is "CHANGELOG.SNAP", stored in
+ *      path ".glusterfs/changelogs/csnap".
+ * Returns:
+ *       0  : On success.
+ *      -1  : On failure.
+ */
+int
+changelog_snap_open(xlator_t *this, changelog_priv_t *priv)
+{
+    int fd = -1;
+    int ret = 0;
+    int flags = 0;
+    char buffer[1024] = {
+        0,
+    };
+    char c_snap_path[PATH_MAX] = {
+        0,
+    };
+    char csnap_dir_path[PATH_MAX] = {
+        0,
+    };
+    int32_t len = 0;
+
+    CHANGELOG_FILL_CSNAP_DIR(priv->changelog_dir, csnap_dir_path);
+
+    len = snprintf(c_snap_path, PATH_MAX, "%s/" CSNAP_FILE_NAME,
+                   csnap_dir_path);
+    if ((len < 0) || (len >= PATH_MAX)) {
+        ret = -1;
+        goto out;
+    }
+
+    flags |= (O_CREAT | O_RDWR | O_TRUNC);
+
+    fd = open(c_snap_path, flags, S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
+    if (fd < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_OPEN_FAILED,
+                "path=%s", c_snap_path, NULL);
+        ret = -1;
+        goto out;
+    }
+    priv->c_snap_fd = fd;
+
+    (void)snprintf(buffer, 1024, CHANGELOG_HEADER, CHANGELOG_VERSION_MAJOR,
+                   CHANGELOG_VERSION_MINOR, priv->ce->encoder);
+    ret = changelog_snap_write_change(priv, buffer, strlen(buffer));
+    if (ret < 0) {
+        sys_close(priv->c_snap_fd);
+        priv->c_snap_fd = -1;
+        goto out;
+    }
+
+out:
+    return ret;
+}
+
+/*
+ * Description:
+ *      Starts logging fop details in CSNAP journal.
+ * Returns:
+ *       0 : On success.
+ *      -1 : On Failure.
+ */
+int
+changelog_snap_logging_start(xlator_t *this, changelog_priv_t *priv)
+{
+    int ret = 0;
+
+    ret = changelog_snap_open(this, priv);
+    gf_smsg(this->name, GF_LOG_INFO, 0, CHANGELOG_MSG_SNAP_INFO, "starting",
+            NULL);
+
+    return ret;
+}
+
+/*
+ * Description:
+ *      Stops logging fop details in CSNAP journal.
+ * Returns:
+ *       0 : On success.
+ *      -1 : On Failure.
+ */
+int
+changelog_snap_logging_stop(xlator_t *this, changelog_priv_t *priv)
+{
+    int ret = 0;
+
+    sys_close(priv->c_snap_fd);
+    priv->c_snap_fd = -1;
+
+    gf_smsg(this->name, GF_LOG_INFO, 0, CHANGELOG_MSG_SNAP_INFO, "Stopped",
+            NULL);
+
+    return ret;
+}
+
+int
+changelog_open_journal(xlator_t *this, changelog_priv_t *priv)
+{
+    int fd = 0;
+    int ret = -1;
+    int flags = 0;
+    char buffer[1024] = {
+        0,
+    };
+    char changelog_path[PATH_MAX] = {
+        0,
+    };
+
+    (void)snprintf(changelog_path, PATH_MAX, "%s/" CHANGELOG_FILE_NAME,
+                   priv->changelog_dir);
+
+    flags |= (O_CREAT | O_RDWR);
+    if (priv->fsync_interval == 0)
+        flags |= O_SYNC;
+
+    fd = open(changelog_path, flags, S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
+    if (fd < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_OPEN_FAILED,
+                "path=%s", changelog_path, NULL);
+        goto out;
+    }
+
+    priv->changelog_fd = fd;
+
+    (void)snprintf(buffer, 1024, CHANGELOG_HEADER, CHANGELOG_VERSION_MAJOR,
+                   CHANGELOG_VERSION_MINOR, priv->ce->encoder);
+    ret = changelog_write_change(priv, buffer, strlen(buffer));
+    if (ret) {
+        sys_close(priv->changelog_fd);
+        priv->changelog_fd = -1;
+        goto out;
+    }
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+int
+changelog_start_next_change(xlator_t *this, changelog_priv_t *priv, time_t ts,
+                            gf_boolean_t finale)
+{
+    int ret = -1;
+
+    ret = changelog_rollover_changelog(this, priv, ts);
+
+    if (!ret && !finale)
+        ret = changelog_open_journal(this, priv);
+
+    return ret;
+}
+
+/**
+ * return the length of entry
+ */
+size_t
+changelog_entry_length()
+{
+    return sizeof(changelog_log_data_t);
+}
+
+void
+changelog_fill_rollover_data(changelog_log_data_t *cld, gf_boolean_t is_last)
+{
+    cld->cld_type = CHANGELOG_TYPE_ROLLOVER;
+    cld->cld_roll_time = gf_time();
+    cld->cld_finale = is_last;
+}
+
+int
+changelog_snap_write_change(changelog_priv_t *priv, char *buffer, size_t len)
+{
+    return changelog_write(priv->c_snap_fd, buffer, len);
+}
+
+int
+changelog_write_change(changelog_priv_t *priv, char *buffer, size_t len)
+{
+    return changelog_write(priv->changelog_fd, buffer, len);
+}
+
+/*
+ * Descriptions:
+ *      Writes fop details in ascii format to CSNAP.
+ * Issues:
+ *      Not Encoding agnostic.
+ * Returns:
+ *      0 : On Success.
+ *     -1 : On Failure.
+ */
+int
+changelog_snap_handle_ascii_change(xlator_t *this, changelog_log_data_t *cld)
+{
+    size_t off = 0;
+    size_t gfid_len = 0;
+    char *gfid_str = NULL;
+    char *buffer = NULL;
+    changelog_priv_t *priv = NULL;
+    int ret = 0;
+
+    if (this == NULL) {
+        ret = -1;
+        goto out;
+    }
+
+    priv = this->private;
+
+    if (priv == NULL) {
+        ret = -1;
+        goto out;
+    }
+
+    gfid_str = uuid_utoa(cld->cld_gfid);
+    gfid_len = strlen(gfid_str);
+
+    /*  extra bytes for decorations */
+    buffer = alloca(gfid_len + cld->cld_ptr_len + 10);
+    CHANGELOG_STORE_ASCII(priv, buffer, off, gfid_str, gfid_len, cld);
+
+    CHANGELOG_FILL_BUFFER(buffer, off, "\0", 1);
+
+    ret = changelog_snap_write_change(priv, buffer, off);
+
+    if (ret < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_MSG_WRITE_FAILED,
+                "csnap", NULL);
+    }
+    gf_smsg(this->name, GF_LOG_INFO, 0, CHANGELOG_MSG_WROTE_TO_CSNAP, NULL);
+    ret = 0;
+out:
+    return ret;
+}
+
+int
+changelog_handle_change(xlator_t *this, changelog_priv_t *priv,
+                        changelog_log_data_t *cld)
+{
+    int ret = 0;
+
+    if (CHANGELOG_TYPE_IS_ROLLOVER(cld->cld_type)) {
+        changelog_encode_change(priv);
+        ret = changelog_start_next_change(this, priv, cld->cld_roll_time,
+                                          cld->cld_finale);
+        if (ret)
+            gf_smsg(this->name, GF_LOG_ERROR, 0,
+                    CHANGELOG_MSG_GET_TIME_OP_FAILED, NULL);
+        goto out;
+    }
+
+    /**
+     * case when there is reconfigure done (disabling changelog) and there
+     * are still fops that have updates in prgress.
+     */
+    if (priv->changelog_fd == -1)
+        return 0;
+
+    if (CHANGELOG_TYPE_IS_FSYNC(cld->cld_type)) {
+        ret = sys_fsync(priv->changelog_fd);
+        if (ret < 0) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno,
+                    CHANGELOG_MSG_FSYNC_OP_FAILED, NULL);
+        }
+        goto out;
+    }
+
+    ret = priv->ce->encode(this, cld);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_MSG_WRITE_FAILED,
+                "changelog", NULL);
+    }
+
+out:
+    return ret;
+}
+
+changelog_local_t *
+changelog_local_init(xlator_t *this, inode_t *inode, uuid_t gfid,
+                     int xtra_records, gf_boolean_t update_flag)
+{
+    changelog_local_t *local = NULL;
+    struct iobuf *iobuf = NULL;
+
+    /**
+     * We relax the presence of inode if @update_flag is true.
+     * The caller (implementation of the fop) needs to be careful to
+     * not blindly use local->inode.
+     */
+    if (!update_flag && !inode) {
+        gf_msg_callingfn(this->name, GF_LOG_WARNING, 0,
+                         CHANGELOG_MSG_INODE_NOT_FOUND,
+                         "inode needed for version checking !!!");
+
+        goto out;
+    }
+
+    if (xtra_records) {
+        iobuf = iobuf_get2(this->ctx->iobuf_pool,
+                           xtra_records * CHANGELOG_OPT_RECORD_LEN);
+        if (!iobuf)
+            goto out;
+    }
+
+    local = mem_get0(this->local_pool);
+    if (!local) {
+        CHANGELOG_IOBUF_UNREF(iobuf);
+        goto out;
+    }
+
+    local->update_no_check = update_flag;
+
+    gf_uuid_copy(local->cld.cld_gfid, gfid);
+
+    local->cld.cld_iobuf = iobuf;
+    local->cld.cld_xtra_records = 0; /* set by the caller */
+
+    if (inode)
+        local->inode = inode_ref(inode);
+
+out:
+    return local;
+}
+
+int
+changelog_forget(xlator_t *this, inode_t *inode)
+{
+    uint64_t ctx_addr = 0;
+    changelog_inode_ctx_t *ctx = NULL;
+
+    inode_ctx_del(inode, this, &ctx_addr);
+    if (!ctx_addr)
+        return 0;
+
+    ctx = (changelog_inode_ctx_t *)(long)ctx_addr;
+    GF_FREE(ctx);
+
+    return 0;
+}
+
+int
+changelog_inject_single_event(xlator_t *this, changelog_priv_t *priv,
+                              changelog_log_data_t *cld)
+{
+    return priv->cd.dispatchfn(this, priv, priv->cd.cd_data, cld, NULL);
+}
+
+/* Wait till all the black fops are drained */
+void
+changelog_drain_black_fops(xlator_t *this, changelog_priv_t *priv)
+{
+    int ret = 0;
+
+    /* clean up framework of pthread_mutex is required here as
+     * 'reconfigure' terminates the changelog_rollover thread
+     * on graph change.
+     */
+    pthread_cleanup_push(changelog_cleanup_free_mutex,
+                         &priv->dm.drain_black_mutex);
+    ret = pthread_mutex_lock(&priv->dm.drain_black_mutex);
+    if (ret)
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_PTHREAD_ERROR,
+                "error=%d", ret, NULL);
+    while (priv->dm.black_fop_cnt > 0) {
+        gf_msg_debug(this->name, 0, "Conditional wait on black fops: %ld",
+                     priv->dm.black_fop_cnt);
+        priv->dm.drain_wait_black = _gf_true;
+        ret = pthread_cond_wait(&priv->dm.drain_black_cond,
+                                &priv->dm.drain_black_mutex);
+        if (ret)
+            gf_smsg(this->name, GF_LOG_ERROR, errno,
+                    CHANGELOG_MSG_PTHREAD_COND_WAIT_FAILED, "error=%d", ret,
+                    NULL);
+    }
+    priv->dm.drain_wait_black = _gf_false;
+    ret = pthread_mutex_unlock(&priv->dm.drain_black_mutex);
+    if (ret)
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_PTHREAD_ERROR,
+                "error=%d", ret, NULL);
+    pthread_cleanup_pop(0);
+    gf_msg_debug(this->name, 0, "Woke up: Conditional wait on black fops");
+}
+
+/* Wait till all the white  fops are drained */
+void
+changelog_drain_white_fops(xlator_t *this, changelog_priv_t *priv)
+{
+    int ret = 0;
+
+    /* clean up framework of pthread_mutex is required here as
+     * 'reconfigure' terminates the changelog_rollover thread
+     * on graph change.
+     */
+    pthread_cleanup_push(changelog_cleanup_free_mutex,
+                         &priv->dm.drain_white_mutex);
+    ret = pthread_mutex_lock(&priv->dm.drain_white_mutex);
+    if (ret)
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_PTHREAD_ERROR,
+                "error=%d", ret, NULL);
+    while (priv->dm.white_fop_cnt > 0) {
+        gf_msg_debug(this->name, 0, "Conditional wait on white fops : %ld",
+                     priv->dm.white_fop_cnt);
+        priv->dm.drain_wait_white = _gf_true;
+        ret = pthread_cond_wait(&priv->dm.drain_white_cond,
+                                &priv->dm.drain_white_mutex);
+        if (ret)
+            gf_smsg(this->name, GF_LOG_ERROR, errno,
+                    CHANGELOG_MSG_PTHREAD_COND_WAIT_FAILED, "error=%d", ret,
+                    NULL);
+    }
+    priv->dm.drain_wait_white = _gf_false;
+    ret = pthread_mutex_unlock(&priv->dm.drain_white_mutex);
+    if (ret)
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_PTHREAD_ERROR,
+                "error=%d", ret, NULL);
+    pthread_cleanup_pop(0);
+    gf_msg_debug(this->name, 0, "Woke up: Conditional wait on white fops");
+}
+
+/**
+ * TODO: these threads have many thing in common (wake up after
+ * a certain time etc..). move them into separate routine.
+ */
+void *
+changelog_rollover(void *data)
+{
+    int ret = 0;
+    xlator_t *this = NULL;
+    struct timespec tv = {
+        0,
+    };
+    changelog_log_data_t cld = {
+        0,
+    };
+    changelog_time_slice_t *slice = NULL;
+    changelog_priv_t *priv = data;
+
+    this = priv->cr.this;
+    slice = &priv->slice;
+
+    while (1) {
+        (void)pthread_testcancel();
+
+        tv.tv_sec = gf_time() + priv->rollover_time;
+        tv.tv_nsec = 0;
+        ret = 0; /* Reset ret to zero */
+
+        /* The race between actual rollover and explicit rollover is
+         * handled. If actual rollover is being done and the
+         * explicit rollover event comes, the event is not missed.
+         * Since explicit rollover sets 'cr.notify' to true, this
+         * thread doesn't wait on 'pthread_cond_timedwait'.
+         */
+        pthread_cleanup_push(changelog_cleanup_free_mutex, &priv->cr.lock);
+        pthread_mutex_lock(&priv->cr.lock);
+        {
+            while (ret == 0 && !priv->cr.notify)
+                ret = pthread_cond_timedwait(&priv->cr.cond, &priv->cr.lock,
+                                             &tv);
+            if (ret == 0)
+                priv->cr.notify = _gf_false;
+        }
+        pthread_mutex_unlock(&priv->cr.lock);
+        pthread_cleanup_pop(0);
+
+        if (ret == 0) {
+            gf_smsg(this->name, GF_LOG_INFO, 0, CHANGELOG_MSG_BARRIER_INFO,
+                    NULL);
+            priv->explicit_rollover = _gf_true;
+        } else if (ret && ret != ETIMEDOUT) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno,
+                    CHANGELOG_MSG_SELECT_FAILED, NULL);
+            continue;
+        } else if (ret && ret == ETIMEDOUT) {
+            gf_msg_debug(this->name, 0, "Wokeup on timeout");
+        }
+
+        /* Reading curent_color without lock is fine here
+         * as it is only modified here and is next to reading.
+         */
+        if (priv->current_color == FOP_COLOR_BLACK) {
+            LOCK(&priv->lock);
+            priv->current_color = FOP_COLOR_WHITE;
+            UNLOCK(&priv->lock);
+            gf_msg_debug(this->name, 0,
+                         "Black fops"
+                         " to be drained:%ld",
+                         priv->dm.black_fop_cnt);
+            changelog_drain_black_fops(this, priv);
+        } else {
+            LOCK(&priv->lock);
+            priv->current_color = FOP_COLOR_BLACK;
+            UNLOCK(&priv->lock);
+            gf_msg_debug(this->name, 0,
+                         "White fops"
+                         " to be drained:%ld",
+                         priv->dm.white_fop_cnt);
+            changelog_drain_white_fops(this, priv);
+        }
+
+        /* Adding delay of 1 second only during explicit rollover:
+         *
+         * Changelog rollover can happen either due to actual
+         * or the explicit rollover during snapshot. Actual
+         * rollover is controlled by tuneable called 'rollover-time'.
+         * The minimum granularity for rollover-time is 1 second.
+         * Explicit rollover is asynchronous in nature and happens
+         * during snapshot.
+         *
+         * Basically, rollover renames the current CHANGELOG file
+         * to CHANGELOG.TIMESTAMP. Let's assume, at time 't1',
+         * actual and explicit rollover raced against  each
+         * other and actual rollover won the race renaming the
+         * CHANGELOG file to CHANGELOG.t1 and opens a new
+         * CHANGELOG file. There is high chance that, an immediate
+         * explicit rollover at time 't1' can happen with in the same
+         * second to rename CHANGELOG file to CHANGELOG.t1 resulting in
+         * purging the earlier CHANGELOG.t1 file created by actual
+         * rollover. So adding a delay of 1 second guarantees unique
+         * CHANGELOG.TIMESTAMP during  explicit rollover.
+         */
+        if (priv->explicit_rollover == _gf_true)
+            sleep(1);
+
+        changelog_fill_rollover_data(&cld, _gf_false);
+
+        _mask_cancellation();
+
+        LOCK(&priv->lock);
+        {
+            ret = changelog_inject_single_event(this, priv, &cld);
+            if (!ret)
+                SLICE_VERSION_UPDATE(slice);
+        }
+        UNLOCK(&priv->lock);
+
+        _unmask_cancellation();
+    }
+
+    return NULL;
+}
+
+void *
+changelog_fsync_thread(void *data)
+{
+    int ret = 0;
+    xlator_t *this = NULL;
+    struct timeval tv = {
+        0,
+    };
+    changelog_log_data_t cld = {
+        0,
+    };
+    changelog_priv_t *priv = data;
+
+    this = priv->cf.this;
+    cld.cld_type = CHANGELOG_TYPE_FSYNC;
+
+    while (1) {
+        (void)pthread_testcancel();
+
+        tv.tv_sec = priv->fsync_interval;
+        tv.tv_usec = 0;
+
+        ret = select(0, NULL, NULL, NULL, &tv);
+        if (ret)
+            continue;
+
+        _mask_cancellation();
+
+        ret = changelog_inject_single_event(this, priv, &cld);
+        if (ret)
+            gf_smsg(this->name, GF_LOG_ERROR, 0,
+                    CHANGELOG_MSG_INJECT_FSYNC_FAILED, NULL);
+
+        _unmask_cancellation();
+    }
+
+    return NULL;
+}
+
+/* macros for inode/changelog version checks */
+
+#define INODE_VERSION_UPDATE(priv, inode, iver, slice, type)                   \
+    do {                                                                       \
+        LOCK(&inode->lock);                                                    \
+        {                                                                      \
+            LOCK(&priv->lock);                                                 \
+            {                                                                  \
+                *iver = slice->changelog_version[type];                        \
+            }                                                                  \
+            UNLOCK(&priv->lock);                                               \
+        }                                                                      \
+        UNLOCK(&inode->lock);                                                  \
+    } while (0)
+
+#define INODE_VERSION_EQUALS_SLICE(priv, ver, slice, type, upd)                \
+    do {                                                                       \
+        LOCK(&priv->lock);                                                     \
+        {                                                                      \
+            upd = (ver == slice->changelog_version[type]) ? _gf_false          \
+                                                          : _gf_true;          \
+        }                                                                      \
+        UNLOCK(&priv->lock);                                                   \
+    } while (0)
+
+static int
+__changelog_inode_ctx_set(xlator_t *this, inode_t *inode,
+                          changelog_inode_ctx_t *ctx)
+{
+    uint64_t ctx_addr = (uint64_t)(uintptr_t)ctx;
+    return __inode_ctx_set(inode, this, &ctx_addr);
+}
+
+/**
+ * one shot routine to get the address and the value of a inode version
+ * for a particular type.
+ */
+changelog_inode_ctx_t *
+__changelog_inode_ctx_get(xlator_t *this, inode_t *inode, unsigned long **iver,
+                          unsigned long *version, changelog_log_type type)
+{
+    int ret = 0;
+    uint64_t ctx_addr = 0;
+    changelog_inode_ctx_t *ctx = NULL;
+
+    ret = __inode_ctx_get(inode, this, &ctx_addr);
+    if (ret < 0)
+        ctx_addr = 0;
+    if (ctx_addr != 0) {
+        ctx = (changelog_inode_ctx_t *)(long)ctx_addr;
+        goto out;
+    }
+
+    ctx = GF_CALLOC(1, sizeof(*ctx), gf_changelog_mt_inode_ctx_t);
+    if (!ctx)
+        goto out;
+
+    ret = __changelog_inode_ctx_set(this, inode, ctx);
+    if (ret) {
+        GF_FREE(ctx);
+        ctx = NULL;
+    }
+
+out:
+    if (ctx && iver && version) {
+        *iver = CHANGELOG_INODE_VERSION_TYPE(ctx, type);
+        *version = **iver;
+    }
+
+    return ctx;
+}
+
+static changelog_inode_ctx_t *
+changelog_inode_ctx_get(xlator_t *this, inode_t *inode, unsigned long **iver,
+                        unsigned long *version, changelog_log_type type)
+{
+    changelog_inode_ctx_t *ctx = NULL;
+
+    LOCK(&inode->lock);
+    {
+        ctx = __changelog_inode_ctx_get(this, inode, iver, version, type);
+    }
+    UNLOCK(&inode->lock);
+
+    return ctx;
+}
+
+/**
+ * This is the main update routine. Locking has been made granular so as to
+ * maximize parallelism of fops - I'll try to explain it below using execution
+ * timelines.
+ *
+ * Basically, the contention is between multiple execution threads of this
+ * routine and the roll-over thread. So, instead of having a big lock, we hold
+ * granular locks: inode->lock and priv->lock. Now I'll explain what happens
+ * when there is an update and a roll-over at just about the same time.
+ * NOTE:
+ *  - the dispatcher itself synchronizes updates via it's own lock
+ *  - the slice version in incremented by the roll-over thread
+ *
+ * Case 1: When the rollover thread wins before the inode version can be
+ * compared with the slice version.
+ *
+ *          [updater]                 |             [rollover]
+ *                                    |
+ *                                    |           <SLICE: 1, 1, 1>
+ * <changelog_update>                 |
+ *   <changelog_inode_ctx_get>        |
+ *      <CTX: 1, 1, 1>                |
+ *                                    |         <dispatch-rollover-event>
+ *                                    |         LOCK (&priv->lock)
+ *                                    |            <SLICE_VERSION_UPDATE>
+ *                                    |              <SLICE: 2, 2, 2>
+ *                                    |         UNLOCK (&priv->lock)
+ *                                    |
+ * LOCK (&priv->lock)                 |
+ *   <INODE_VERSION_EQUALS_SLICE>     |
+ *    I: 1 <-> S: 2                   |
+ *    update: true                    |
+ * UNLOCK (&priv->lock)               |
+ *                                    |
+ * <if update == true>                |
+ *  <dispath-update-event>            |
+ *  <INODE_VERSION_UPDATE>            |
+ *   LOCK (&inode->lock)              |
+ *    LOCK (&priv->lock)              |
+ *     <CTX: 2, 1, 1>                 |
+ *    UNLOCK (&priv->lock)            |
+ *   UNLOCK (&inode->lock)            |
+ *
+ * Therefore, the change gets recorded in the next change (no lost change). If
+ * the slice version was ahead of the inode version (say I:1, S: 2), then
+ * anyway the comparison would result in a update (I: 1, S: 3).
+ *
+ * If the rollover time is too less, then there is another contention when the
+ * updater tries to bring up inode version to the slice version (this is also
+ * the case when the roll-over thread wakes up during INODE_VERSION_UPDATE.
+ *
+ *   <CTX: 1, 1, 1>                   |       <SLICE: 2, 2, 2>
+ *                                    |
+ *                                    |
+ * <dispath-update-event>             |
+ * <INODE_VERSION_UPDATE>             |
+ *  LOCK (&inode->lock)               |
+ *   LOCK (&priv->lock)               |
+ *    <CTX: 2, 1, 1>                  |
+ *   UNLOCK (&priv->lock)             |
+ *  UNLOCK (&inode->lock)             |
+ *                                    |         <dispatch-rollover-event>
+ *                                    |         LOCK (&priv->lock)
+ *                                    |            <SLICE_VERSION_UPDATE>
+ *                                    |              <SLICE: 3, 3, 3>
+ *                                    |         UNLOCK (&priv->lock)
+ *
+ *
+ * Case 2: When the fop thread wins
+ *
+ *          [updater]                 |             [rollover]
+ *                                    |
+ *                                    |           <SLICE: 1, 1, 1>
+ * <changelog_update>                 |
+ *   <changelog_inode_ctx_get>        |
+ *      <CTX: 0, 0, 0>                |
+ *                                    |
+ * LOCK (&priv->lock)                 |
+ *   <INODE_VERSION_EQUALS_SLICE>     |
+ *    I: 0 <-> S: 1                   |
+ *    update: true                    |
+ * UNLOCK (&priv->lock)               |
+ *                                    |         <dispatch-rollover-event>
+ *                                    |         LOCK (&priv->lock)
+ *                                    |            <SLICE_VERSION_UPDATE>
+ *                                    |              <SLICE: 2, 2, 2>
+ *                                    |         UNLOCK (&priv->lock)
+ * <if update == true>                |
+ *  <dispath-update-event>            |
+ *  <INODE_VERSION_UPDATE>            |
+ *   LOCK (&inode->lock)              |
+ *    LOCK (&priv->lock)              |
+ *     <CTX: 2, 0, 0>                 |
+ *    UNLOCK (&priv->lock)            |
+ *   UNLOCK (&inode->lock)            |
+ *
+ * Here again, if the inode version was equal to the slice version (I: 1, S: 1)
+ * then there is no need to record an update (as the equality of the two version
+ * signifies an update was recorded in the current time slice).
+ */
+void
+changelog_update(xlator_t *this, changelog_priv_t *priv,
+                 changelog_local_t *local, changelog_log_type type)
+{
+    int ret = 0;
+    unsigned long *iver = NULL;
+    unsigned long version = 0;
+    inode_t *inode = NULL;
+    changelog_time_slice_t *slice = NULL;
+    changelog_inode_ctx_t *ctx = NULL;
+    changelog_log_data_t *cld_0 = NULL;
+    changelog_log_data_t *cld_1 = NULL;
+    changelog_local_t *next_local = NULL;
+    gf_boolean_t need_upd = _gf_true;
+
+    slice = &priv->slice;
+
+    /**
+     * for fops that do not require inode version checking
+     */
+    if (local->update_no_check)
+        goto update;
+
+    inode = local->inode;
+
+    ctx = changelog_inode_ctx_get(this, inode, &iver, &version, type);
+    if (!ctx)
+        goto update;
+
+    INODE_VERSION_EQUALS_SLICE(priv, version, slice, type, need_upd);
+
+update:
+    if (need_upd) {
+        cld_0 = &local->cld;
+        cld_0->cld_type = type;
+
+        if ((next_local = local->prev_entry) != NULL) {
+            cld_1 = &next_local->cld;
+            cld_1->cld_type = type;
+        }
+
+        ret = priv->cd.dispatchfn(this, priv, priv->cd.cd_data, cld_0, cld_1);
+
+        /**
+         * update after the dispatcher has successfully done
+         * it's job.
+         */
+        if (!local->update_no_check && iver && !ret)
+            INODE_VERSION_UPDATE(priv, inode, iver, slice, type);
+    }
+
+    return;
+}
+
+/* Begin: Geo-rep snapshot dependency changes */
+
+/* changelog_color_fop_and_inc_cnt: Assign color and inc fop cnt.
+ *
+ * Assigning color and increment of corresponding fop count should happen
+ * in a lock (i.e., there should be no window between them). If it does not,
+ * we might miss draining those fops which are colored but not yet incremented
+ * the count. Let's assume black fops are draining. If the black fop count
+ * reaches zero, we say draining is completed but we miss black fops which are
+ * not incremented fop count but color is assigned black.
+ */
+
+void
+changelog_color_fop_and_inc_cnt(xlator_t *this, changelog_priv_t *priv,
+                                changelog_local_t *local)
+{
+    if (!priv || !local)
+        return;
+
+    LOCK(&priv->lock);
+    {
+        local->color = priv->current_color;
+        changelog_inc_fop_cnt(this, priv, local);
+    }
+    UNLOCK(&priv->lock);
+}
+
+/* Increments the respective fop counter based on the fop color */
+void
+changelog_inc_fop_cnt(xlator_t *this, changelog_priv_t *priv,
+                      changelog_local_t *local)
+{
+    int ret = 0;
+
+    if (local) {
+        if (local->color == FOP_COLOR_BLACK) {
+            ret = pthread_mutex_lock(&priv->dm.drain_black_mutex);
+            CHANGELOG_PTHREAD_ERROR_HANDLE_0(ret, out);
+            {
+                priv->dm.black_fop_cnt++;
+            }
+            ret = pthread_mutex_unlock(&priv->dm.drain_black_mutex);
+            CHANGELOG_PTHREAD_ERROR_HANDLE_0(ret, out);
+        } else {
+            ret = pthread_mutex_lock(&priv->dm.drain_white_mutex);
+            CHANGELOG_PTHREAD_ERROR_HANDLE_0(ret, out);
+            {
+                priv->dm.white_fop_cnt++;
+            }
+            ret = pthread_mutex_unlock(&priv->dm.drain_white_mutex);
+            CHANGELOG_PTHREAD_ERROR_HANDLE_0(ret, out);
+        }
+    }
+out:
+    return;
+}
+
+/* Decrements the respective fop counter based on the fop color */
+void
+changelog_dec_fop_cnt(xlator_t *this, changelog_priv_t *priv,
+                      changelog_local_t *local)
+{
+    int ret = 0;
+
+    if (local) {
+        if (local->color == FOP_COLOR_BLACK) {
+            ret = pthread_mutex_lock(&priv->dm.drain_black_mutex);
+            CHANGELOG_PTHREAD_ERROR_HANDLE_0(ret, out);
+            {
+                priv->dm.black_fop_cnt--;
+                if (priv->dm.black_fop_cnt == 0 &&
+                    priv->dm.drain_wait_black == _gf_true) {
+                    ret = pthread_cond_signal(&priv->dm.drain_black_cond);
+                    CHANGELOG_PTHREAD_ERROR_HANDLE_2(
+                        ret, out, priv->dm.drain_black_mutex);
+                    gf_msg_debug(this->name, 0,
+                                 "Signalled "
+                                 "draining of black");
+                }
+            }
+            ret = pthread_mutex_unlock(&priv->dm.drain_black_mutex);
+            CHANGELOG_PTHREAD_ERROR_HANDLE_0(ret, out);
+        } else {
+            ret = pthread_mutex_lock(&priv->dm.drain_white_mutex);
+            CHANGELOG_PTHREAD_ERROR_HANDLE_0(ret, out);
+            {
+                priv->dm.white_fop_cnt--;
+                if (priv->dm.white_fop_cnt == 0 &&
+                    priv->dm.drain_wait_white == _gf_true) {
+                    ret = pthread_cond_signal(&priv->dm.drain_white_cond);
+                    CHANGELOG_PTHREAD_ERROR_HANDLE_2(
+                        ret, out, priv->dm.drain_white_mutex);
+                    gf_msg_debug(this->name, 0,
+                                 "Signalled "
+                                 "draining of white");
+                }
+            }
+            ret = pthread_mutex_unlock(&priv->dm.drain_white_mutex);
+            CHANGELOG_PTHREAD_ERROR_HANDLE_0(ret, out);
+        }
+    }
+out:
+    return;
+}
+
+/* Write to a pipe setup between changelog main thread and changelog
+ * rollover thread to initiate explicit rollover of changelog journal.
+ */
+int
+changelog_barrier_notify(changelog_priv_t *priv, char *buf)
+{
+    int ret = 0;
+
+    pthread_mutex_lock(&priv->cr.lock);
+    {
+        ret = pthread_cond_signal(&priv->cr.cond);
+        priv->cr.notify = _gf_true;
+    }
+    pthread_mutex_unlock(&priv->cr.lock);
+    return ret;
+}
+
+/* Clean up flags set on barrier notification */
+void
+changelog_barrier_cleanup(xlator_t *this, changelog_priv_t *priv,
+                          struct list_head *queue)
+{
+    int ret = 0;
+
+    LOCK(&priv->bflags.lock);
+    priv->bflags.barrier_ext = _gf_false;
+    UNLOCK(&priv->bflags.lock);
+
+    ret = pthread_mutex_lock(&priv->bn.bnotify_mutex);
+    CHANGELOG_PTHREAD_ERROR_HANDLE_0(ret, out);
+    {
+        priv->bn.bnotify = _gf_false;
+    }
+    ret = pthread_mutex_unlock(&priv->bn.bnotify_mutex);
+    CHANGELOG_PTHREAD_ERROR_HANDLE_0(ret, out);
+
+    /* Disable changelog barrier and dequeue fops */
+    LOCK(&priv->lock);
+    {
+        if (priv->barrier_enabled == _gf_true)
+            __chlog_barrier_disable(this, queue);
+        else
+            ret = -1;
+    }
+    UNLOCK(&priv->lock);
+    if (ret == 0)
+        chlog_barrier_dequeue_all(this, queue);
+
+out:
+    return;
+}
+/* End: Geo-Rep snapshot dependency changes */
+
+int32_t
+changelog_fill_entry_buf(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                         changelog_local_t **local)
+{
+    changelog_opt_t *co = NULL;
+    size_t xtra_len = 0;
+    char *dup_path = NULL;
+    char *bname = NULL;
+    inode_t *parent = NULL;
+
+    GF_ASSERT(this);
+
+    parent = inode_parent(loc->inode, 0, 0);
+    if (!parent) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, CHANGELOG_MSG_INODE_NOT_FOUND,
+                "type=parent", "gfid=%s", uuid_utoa(loc->inode->gfid), NULL);
+        goto err;
+    }
+
+    CHANGELOG_INIT_NOCHECK(this, *local, loc->inode, loc->inode->gfid, 5);
+    if (!(*local)) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_MSG_LOCAL_INIT_FAILED,
+                NULL);
+        goto err;
+    }
+
+    co = changelog_get_usable_buffer(*local);
+    if (!co) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_MSG_GET_BUFFER_FAILED,
+                NULL);
+        goto err;
+    }
+
+    if (loc->inode->ia_type == IA_IFDIR) {
+        CHANGLOG_FILL_FOP_NUMBER(co, GF_FOP_MKDIR, fop_fn, xtra_len);
+        co++;
+        CHANGELOG_FILL_UINT32(co, S_IFDIR | 0755, number_fn, xtra_len);
+        co++;
+    } else {
+        CHANGLOG_FILL_FOP_NUMBER(co, GF_FOP_CREATE, fop_fn, xtra_len);
+        co++;
+        CHANGELOG_FILL_UINT32(co, S_IFREG | 0644, number_fn, xtra_len);
+        co++;
+    }
+
+    CHANGELOG_FILL_UINT32(co, frame->root->uid, number_fn, xtra_len);
+    co++;
+
+    CHANGELOG_FILL_UINT32(co, frame->root->gid, number_fn, xtra_len);
+    co++;
+
+    dup_path = gf_strdup(loc->path);
+    bname = basename(dup_path);
+
+    CHANGELOG_FILL_ENTRY(co, parent->gfid, bname, entry_fn, entry_free_fn,
+                         xtra_len, err);
+    changelog_set_usable_record_and_length(*local, xtra_len, 5);
+
+    if (dup_path)
+        GF_FREE(dup_path);
+    if (parent)
+        inode_unref(parent);
+    return 0;
+
+err:
+    if (dup_path)
+        GF_FREE(dup_path);
+    if (parent)
+        inode_unref(parent);
+    return -1;
+}
+
+/*
+ * resolve_pargfid_to_path:
+ *      It converts given pargfid to path by doing recursive readlinks at the
+ * backend. If bname is given, it suffixes bname to pargfid to form the
+ * complete path else it doesn't. It allocates memory for the path and is
+ * caller's responsibility to free the same. If bname is NULL and pargfid
+ * is ROOT, then it returns "."
+ */
+
+int
+resolve_pargfid_to_path(xlator_t *this, const uuid_t pgfid, char **path,
+                        char *bname)
+{
+    char *linkname = NULL;
+    char *dir_handle = NULL;
+    char *pgfidstr = NULL;
+    char *saveptr = NULL;
+    ssize_t len = 0;
+    int ret = 0;
+    uuid_t tmp_gfid = {
+        0,
+    };
+    uuid_t pargfid = {
+        0,
+    };
+    changelog_priv_t *priv = NULL;
+    char gpath[PATH_MAX] = {
+        0,
+    };
+    char result[PATH_MAX] = {
+        0,
+    };
+    char *dir_name = NULL;
+    char pre_dir_name[PATH_MAX] = {
+        0,
+    };
+
+    GF_ASSERT(this);
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    gf_uuid_copy(pargfid, pgfid);
+    if (!path || gf_uuid_is_null(pargfid)) {
+        ret = -1;
+        goto out;
+    }
+
+    if (__is_root_gfid(pargfid)) {
+        if (bname)
+            *path = gf_strdup(bname);
+        else
+            *path = gf_strdup(".");
+        return ret;
+    }
+
+    dir_handle = alloca(PATH_MAX);
+    linkname = alloca(PATH_MAX);
+    (void)snprintf(gpath, PATH_MAX, "%s/.glusterfs/", priv->changelog_brick);
+
+    while (!(__is_root_gfid(pargfid))) {
+        len = snprintf(dir_handle, PATH_MAX, "%s/%02x/%02x/%s", gpath,
+                       pargfid[0], pargfid[1], uuid_utoa(pargfid));
+        if ((len < 0) || (len >= PATH_MAX)) {
+            ret = -1;
+            goto out;
+        }
+
+        len = sys_readlink(dir_handle, linkname, PATH_MAX);
+        if (len < 0) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno,
+                    CHANGELOG_MSG_READLINK_OP_FAILED,
+                    "could not read the "
+                    "link from the gfid handle",
+                    "handle=%s", dir_handle, NULL);
+            ret = -1;
+            goto out;
+        }
+
+        linkname[len] = '\0';
+
+        pgfidstr = strtok_r(linkname + strlen("../../00/00/"), "/", &saveptr);
+        dir_name = strtok_r(NULL, "/", &saveptr);
+
+        len = snprintf(result, PATH_MAX, "%s/%s", dir_name, pre_dir_name);
+        if ((len < 0) || (len >= PATH_MAX)) {
+            ret = -1;
+            goto out;
+        }
+        if (snprintf(pre_dir_name, len + 1, "%s", result) >= len + 1) {
+            ret = -1;
+            goto out;
+        }
+
+        gf_uuid_parse(pgfidstr, tmp_gfid);
+        gf_uuid_copy(pargfid, tmp_gfid);
+    }
+
+    if (bname)
+        strncat(result, bname, strlen(bname) + 1);
+
+    *path = gf_strdup(result);
+
+out:
+    return ret;
+}
diff --git a/xlators/features/changelog/src/changelog-helpers.h b/xlators/features/changelog/src/changelog-helpers.h
new file mode 100644
index 00000000000..38fa7590c32
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-helpers.h
@@ -0,0 +1,716 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef _CHANGELOG_HELPERS_H
+#define _CHANGELOG_HELPERS_H
+
+#include <glusterfs/locking.h>
+#include <glusterfs/timer.h>
+#include "pthread.h"
+#include <glusterfs/iobuf.h>
+#include <glusterfs/rot-buffs.h>
+
+#include "changelog-misc.h"
+#include <glusterfs/call-stub.h>
+
+#include "rpcsvc.h"
+#include "changelog-ev-handle.h"
+
+#include "changelog.h"
+#include "changelog-messages.h"
+
+/**
+ * the changelog entry
+ */
+typedef struct changelog_log_data {
+    /* rollover related */
+    time_t cld_roll_time;
+
+    /* reopen changelog? */
+    gf_boolean_t cld_finale;
+
+    changelog_log_type cld_type;
+
+    /**
+     * sincd gfid is _always_ a necessity, it's not a part
+     * of the iobuf. by doing this we do not add any overhead
+     * for data and metadata related fops.
+     */
+    uuid_t cld_gfid;
+
+    /**
+     * iobufs are used for optionals records: pargfid, path,
+     * write offsets etc.. It's the fop implementers job
+     * to allocate (iobuf_get() in the fop) and get unref'ed
+     * in the callback (CHANGELOG_STACK_UNWIND).
+     */
+    struct iobuf *cld_iobuf;
+
+#define cld_ptr cld_iobuf->ptr
+
+    /**
+     * after allocation you can point this to the length of
+     * usable data, but make sure it does not exceed the
+     * the size of the requested iobuf.
+     */
+    size_t cld_iobuf_len;
+
+#define cld_ptr_len cld_iobuf_len
+
+    /**
+     * number of optional records
+     */
+    int cld_xtra_records;
+} changelog_log_data_t;
+
+/**
+ * holder for dispatch function and private data
+ */
+
+typedef struct changelog_priv changelog_priv_t;
+
+typedef struct changelog_dispatcher {
+    void *cd_data;
+    int (*dispatchfn)(xlator_t *, changelog_priv_t *, void *,
+                      changelog_log_data_t *, changelog_log_data_t *);
+} changelog_dispatcher_t;
+
+struct changelog_bootstrap {
+    changelog_mode_t mode;
+    int (*ctor)(xlator_t *, changelog_dispatcher_t *);
+    int (*dtor)(xlator_t *, changelog_dispatcher_t *);
+};
+
+struct changelog_encoder {
+    changelog_encoder_t encoder;
+    int (*encode)(xlator_t *, changelog_log_data_t *);
+};
+
+/* xlator private */
+
+typedef struct changelog_time_slice {
+    /**
+     * version of changelog file, incremented each time changes
+     * rollover.
+     */
+    unsigned long changelog_version[CHANGELOG_MAX_TYPE];
+} changelog_time_slice_t;
+
+typedef struct changelog_rollover {
+    /* rollover thread */
+    pthread_t rollover_th;
+
+    xlator_t *this;
+
+    pthread_mutex_t lock;
+    pthread_cond_t cond;
+    gf_boolean_t notify;
+} changelog_rollover_t;
+
+typedef struct changelog_fsync {
+    /* fsync() thread */
+    pthread_t fsync_th;
+
+    xlator_t *this;
+} changelog_fsync_t;
+
+/* Draining during changelog rollover (for geo-rep snapshot dependency):
+ * --------------------------------------------------------------------
+ * The introduction of draining of in-transit fops during changelog rollover
+ * (both explicit/timeout triggered) requires coloring of fops. Basically the
+ * implementation requires two counters, one counter which keeps the count of
+ * current intransit fops which should end up in current changelog and the other
+ * counter to keep track of incoming fops which should be drained as part of
+ * next changelog rollover event. The fops are colored w.r.t these counters.
+ * The fops that are to be drained as part of current changelog rollover is
+ * given one color and the fops which keep incoming during this and not
+ * necessarily should end up in current changelog and should be drained as part
+ * of next changelog rollover are given other color. The color switching
+ * continues with each changelog rollover. Two colors(black and white) are
+ * chosen here and initially black is chosen is default.
+ */
+
+typedef enum chlog_fop_color {
+    FOP_COLOR_BLACK,
+    FOP_COLOR_WHITE
+} chlog_fop_color_t;
+
+/* Barrier notify variable */
+typedef struct barrier_notify {
+    pthread_mutex_t bnotify_mutex;
+    pthread_cond_t bnotify_cond;
+    gf_boolean_t bnotify;
+    gf_boolean_t bnotify_error;
+} barrier_notify_t;
+
+/* Two separate mutex and conditional variable set is used
+ * to drain white and black fops. */
+
+typedef struct drain_mgmt {
+    pthread_mutex_t drain_black_mutex;
+    pthread_cond_t drain_black_cond;
+    pthread_mutex_t drain_white_mutex;
+    pthread_cond_t drain_white_cond;
+    /* Represents black fops count in-transit */
+    unsigned long black_fop_cnt;
+    /* Represents white fops count in-transit */
+    unsigned long white_fop_cnt;
+    gf_boolean_t drain_wait_black;
+    gf_boolean_t drain_wait_white;
+} drain_mgmt_t;
+
+/* External barrier as a result of snap on/off indicating flag*/
+typedef struct barrier_flags {
+    gf_lock_t lock;
+    gf_boolean_t barrier_ext;
+} barrier_flags_t;
+
+/* Event selection */
+typedef struct changelog_ev_selector {
+    gf_lock_t reflock;
+
+    /**
+     * Array of references for each selection bit.
+     */
+    unsigned int ref[CHANGELOG_EV_SELECTION_RANGE];
+} changelog_ev_selector_t;
+
+/* changelog's private structure */
+struct changelog_priv {
+    /* changelog journalling */
+    gf_boolean_t active;
+
+    /* changelog live notifications */
+    gf_boolean_t rpc_active;
+
+    /* to generate unique socket file per brick */
+    char *changelog_brick;
+
+    /* logging directory */
+    char *changelog_dir;
+
+    /* htime directory */
+    char *htime_dir;
+
+    /* one file for all changelog types */
+    int changelog_fd;
+
+    /* htime fd for current changelog session */
+    int htime_fd;
+
+    /*  c_snap_fd is fd for call-path changelog */
+    int c_snap_fd;
+
+    /* rollover_count used by htime */
+    int rollover_count;
+
+    gf_lock_t lock;
+
+    /*  lock to synchronize CSNAP updation */
+    gf_lock_t c_snap_lock;
+
+    /* written end of the pipe */
+    int wfd;
+
+    /* rollover time */
+    int32_t rollover_time;
+
+    /* fsync() interval */
+    int32_t fsync_interval;
+
+    /* changelog type maps */
+    const char *maps[CHANGELOG_MAX_TYPE];
+
+    /* time slicer */
+    changelog_time_slice_t slice;
+
+    /* context of the updater */
+    changelog_dispatcher_t cd;
+
+    /* context of the rollover thread */
+    changelog_rollover_t cr;
+
+    /* context of fsync thread */
+    changelog_fsync_t cf;
+
+    /* operation mode */
+    changelog_mode_t op_mode;
+
+    /* bootstrap routine for 'current' logger */
+    struct changelog_bootstrap *cb;
+
+    /* encoder mode */
+    changelog_encoder_t encode_mode;
+
+    /* encoder */
+    struct changelog_encoder *ce;
+
+    /**
+     * snapshot dependency changes
+     */
+
+    /* Draining of fops*/
+    drain_mgmt_t dm;
+
+    /* Represents the active color. Initially by default black */
+    chlog_fop_color_t current_color;
+
+    /* flag to determine explicit rollover is triggered */
+    gf_boolean_t explicit_rollover;
+
+    /* barrier notification variable protected by mutex */
+    barrier_notify_t bn;
+
+    /* barrier on/off indicating flags */
+    barrier_flags_t bflags;
+
+    /* changelog barrier on/off indicating flag */
+    gf_boolean_t barrier_enabled;
+    struct list_head queue;
+    uint32_t queue_size;
+    gf_timer_t *timer;
+    struct timespec timeout;
+
+    /**
+     * buffers, RPC, event selection, notifications and other
+     * beasts.
+     */
+
+    /* epoll pthread */
+    pthread_t poller;
+
+    /* rotational buffer */
+    rbuf_t *rbuf;
+
+    /* changelog RPC server */
+    rpcsvc_t *rpc;
+
+    /* event selection */
+    changelog_ev_selector_t ev_selection;
+
+    /* client handling (reverse connection) */
+    pthread_t connector;
+
+    int nr_dispatchers;
+    pthread_t *ev_dispatcher;
+
+    changelog_clnt_t connections;
+
+    /* glusterfind dependency to capture paths on deleted entries*/
+    gf_boolean_t capture_del_path;
+
+    /* Save total no. of listners */
+    gf_atomic_t listnercnt;
+
+    /* Save total no. of xprt are associated with listner */
+    gf_atomic_t xprtcnt;
+
+    /* Save xprt list */
+    struct list_head xprt_list;
+
+    /* Save total no. of client connection */
+    gf_atomic_t clntcnt;
+
+    /* Save cleanup brick in victim */
+    xlator_t *victim;
+
+    /* Status to save cleanup notify status */
+    gf_boolean_t notify_down;
+};
+
+struct changelog_local {
+    inode_t *inode;
+    gf_boolean_t update_no_check;
+
+    changelog_log_data_t cld;
+
+    /**
+     * ->prev_entry is used in cases when there needs to be
+     * additional changelog entry for the parent (eg. rename)
+     * It's analogous to ->next in single linked list world,
+     * but we call it as ->prev_entry... ha ha ha
+     */
+    struct changelog_local *prev_entry;
+
+    /* snap dependency changes */
+    chlog_fop_color_t color;
+};
+
+typedef struct changelog_local changelog_local_t;
+
+/* inode version is stored in inode ctx */
+typedef struct changelog_inode_ctx {
+    unsigned long iversion[CHANGELOG_MAX_TYPE];
+} changelog_inode_ctx_t;
+
+#define CHANGELOG_INODE_VERSION_TYPE(ctx, type) &(ctx->iversion[type])
+
+/**
+ * Optional Records:
+ *  fops that need to save additional information request a array of
+ *  @changelog_opt_t struct. The array is allocated via @iobufs.
+ */
+typedef enum {
+    CHANGELOG_OPT_REC_FOP,
+    CHANGELOG_OPT_REC_ENTRY,
+    CHANGELOG_OPT_REC_UINT32,
+} changelog_optional_rec_type_t;
+
+struct changelog_entry_fields {
+    uuid_t cef_uuid;
+    char *cef_bname;
+    char *cef_path;
+};
+
+typedef struct {
+    /**
+     * @co_covert can be used to do post-processing of the record before
+     * it's persisted to the CHANGELOG. If this is NULL, then the record
+     * is persisted as per it's in memory format.
+     */
+    size_t (*co_convert)(void *data, char *buffer, gf_boolean_t encode);
+
+    /* release routines */
+    void (*co_free)(void *data);
+
+    /* type of the field */
+    changelog_optional_rec_type_t co_type;
+
+    /**
+     * sizeof of the 'valid' field in the union. This field is not used if
+     * @co_convert is specified.
+     */
+    size_t co_len;
+
+    union {
+        unsigned int co_uint32;
+        glusterfs_fop_t co_fop;
+        struct changelog_entry_fields co_entry;
+    };
+} changelog_opt_t;
+
+#define CHANGELOG_OPT_RECORD_LEN sizeof(changelog_opt_t)
+
+/**
+ * helpers routines
+ */
+
+int
+changelog_thread_cleanup(xlator_t *this, pthread_t thr_id);
+
+void *
+changelog_get_usable_buffer(changelog_local_t *local);
+
+void
+changelog_set_usable_record_and_length(changelog_local_t *local, size_t len,
+                                       int xr);
+void
+changelog_local_cleanup(xlator_t *xl, changelog_local_t *local);
+changelog_local_t *
+changelog_local_init(xlator_t *this, inode_t *inode, uuid_t gfid,
+                     int xtra_records, gf_boolean_t update_flag);
+int
+changelog_start_next_change(xlator_t *this, changelog_priv_t *priv, time_t ts,
+                            gf_boolean_t finale);
+int
+changelog_open_journal(xlator_t *this, changelog_priv_t *priv);
+void
+changelog_fill_rollover_data(changelog_log_data_t *cld, gf_boolean_t is_last);
+int
+changelog_inject_single_event(xlator_t *this, changelog_priv_t *priv,
+                              changelog_log_data_t *cld);
+size_t
+changelog_entry_length();
+int
+changelog_write(int fd, char *buffer, size_t len);
+int
+changelog_write_change(changelog_priv_t *priv, char *buffer, size_t len);
+int
+changelog_handle_change(xlator_t *this, changelog_priv_t *priv,
+                        changelog_log_data_t *cld);
+void
+changelog_update(xlator_t *this, changelog_priv_t *priv,
+                 changelog_local_t *local, changelog_log_type type);
+void *
+changelog_rollover(void *data);
+void *
+changelog_fsync_thread(void *data);
+int
+changelog_forget(xlator_t *this, inode_t *inode);
+int
+htime_update(xlator_t *this, changelog_priv_t *priv, time_t ts, char *buffer);
+int
+htime_open(xlator_t *this, changelog_priv_t *priv, time_t ts);
+int
+htime_create(xlator_t *this, changelog_priv_t *priv, time_t ts);
+
+/* Geo-Rep snapshot dependency changes */
+void
+changelog_color_fop_and_inc_cnt(xlator_t *this, changelog_priv_t *priv,
+                                changelog_local_t *local);
+void
+changelog_inc_fop_cnt(xlator_t *this, changelog_priv_t *priv,
+                      changelog_local_t *local);
+void
+changelog_dec_fop_cnt(xlator_t *this, changelog_priv_t *priv,
+                      changelog_local_t *local);
+int
+changelog_barrier_notify(changelog_priv_t *priv, char *buf);
+void
+changelog_barrier_cleanup(xlator_t *this, changelog_priv_t *priv,
+                          struct list_head *queue);
+void
+changelog_drain_white_fops(xlator_t *this, changelog_priv_t *priv);
+void
+changelog_drain_black_fops(xlator_t *this, changelog_priv_t *priv);
+
+/* Crash consistency of changelog wrt snapshot */
+int
+changelog_snap_logging_stop(xlator_t *this, changelog_priv_t *priv);
+int
+changelog_snap_logging_start(xlator_t *this, changelog_priv_t *priv);
+int
+changelog_snap_open(xlator_t *this, changelog_priv_t *priv);
+int
+changelog_snap_handle_ascii_change(xlator_t *this, changelog_log_data_t *cld);
+int
+changelog_snap_write_change(changelog_priv_t *priv, char *buffer, size_t len);
+
+/* Changelog barrier routines */
+void
+__chlog_barrier_enqueue(xlator_t *this, call_stub_t *stub);
+void
+__chlog_barrier_disable(xlator_t *this, struct list_head *queue);
+void
+chlog_barrier_dequeue_all(xlator_t *this, struct list_head *queue);
+call_stub_t *
+__chlog_barrier_dequeue(xlator_t *this, struct list_head *queue);
+int
+__chlog_barrier_enable(xlator_t *this, changelog_priv_t *priv);
+
+int32_t
+changelog_fill_entry_buf(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                         changelog_local_t **local);
+
+/* event selection routines */
+void
+changelog_select_event(xlator_t *, changelog_ev_selector_t *, unsigned int);
+void
+changelog_deselect_event(xlator_t *, changelog_ev_selector_t *, unsigned int);
+int
+changelog_init_event_selection(xlator_t *, changelog_ev_selector_t *);
+int
+changelog_ev_selected(xlator_t *, changelog_ev_selector_t *, unsigned int);
+void
+changelog_dispatch_event(xlator_t *, changelog_priv_t *, changelog_event_t *);
+
+changelog_inode_ctx_t *
+__changelog_inode_ctx_get(xlator_t *, inode_t *, unsigned long **,
+                          unsigned long *, changelog_log_type);
+int
+resolve_pargfid_to_path(xlator_t *this, const uuid_t gfid, char **path,
+                        char *bname);
+
+/* macros */
+
+#define CHANGELOG_STACK_UNWIND(fop, frame, params...)                          \
+    do {                                                                       \
+        changelog_local_t *__local = NULL;                                     \
+        xlator_t *__xl = NULL;                                                 \
+        if (frame) {                                                           \
+            __local = frame->local;                                            \
+            __xl = frame->this;                                                \
+            frame->local = NULL;                                               \
+        }                                                                      \
+        STACK_UNWIND_STRICT(fop, frame, params);                               \
+        if (__local && __local->prev_entry)                                    \
+            changelog_local_cleanup(__xl, __local->prev_entry);                \
+        changelog_local_cleanup(__xl, __local);                                \
+    } while (0)
+
+#define CHANGELOG_IOBUF_REF(iobuf)                                             \
+    do {                                                                       \
+        if (iobuf)                                                             \
+            iobuf_ref(iobuf);                                                  \
+    } while (0)
+
+#define CHANGELOG_IOBUF_UNREF(iobuf)                                           \
+    do {                                                                       \
+        if (iobuf)                                                             \
+            iobuf_unref(iobuf);                                                \
+    } while (0)
+
+#define CHANGELOG_FILL_BUFFER(buffer, off, val, len)                           \
+    do {                                                                       \
+        memcpy(buffer + off, val, len);                                        \
+        off += len;                                                            \
+    } while (0)
+
+#define SLICE_VERSION_UPDATE(slice)                                            \
+    do {                                                                       \
+        int i = 0;                                                             \
+        for (; i < CHANGELOG_MAX_TYPE; i++) {                                  \
+            slice->changelog_version[i]++;                                     \
+        }                                                                      \
+    } while (0)
+
+#define CHANGELOG_FILL_UINT32(co, number, converter, xlen)                     \
+    do {                                                                       \
+        co->co_convert = converter;                                            \
+        co->co_free = NULL;                                                    \
+        co->co_type = CHANGELOG_OPT_REC_UINT32;                                \
+        co->co_uint32 = number;                                                \
+        xlen += sizeof(unsigned int);                                          \
+    } while (0)
+
+#define CHANGLOG_FILL_FOP_NUMBER(co, fop, converter, xlen)                     \
+    do {                                                                       \
+        co->co_convert = converter;                                            \
+        co->co_free = NULL;                                                    \
+        co->co_type = CHANGELOG_OPT_REC_FOP;                                   \
+        co->co_fop = fop;                                                      \
+        xlen += sizeof(fop);                                                   \
+    } while (0)
+
+#define CHANGELOG_FILL_ENTRY(co, pargfid, bname, converter, freefn, xlen,      \
+                             label)                                            \
+    do {                                                                       \
+        co->co_convert = converter;                                            \
+        co->co_free = freefn;                                                  \
+        co->co_type = CHANGELOG_OPT_REC_ENTRY;                                 \
+        gf_uuid_copy(co->co_entry.cef_uuid, pargfid);                          \
+        co->co_entry.cef_bname = gf_strdup(bname);                             \
+        if (!co->co_entry.cef_bname)                                           \
+            goto label;                                                        \
+        xlen += (UUID_CANONICAL_FORM_LEN + strlen(bname));                     \
+    } while (0)
+
+#define CHANGELOG_FILL_ENTRY_DIR_PATH(co, pargfid, bname, converter,           \
+                                      del_freefn, xlen, label, capture_del)    \
+    do {                                                                       \
+        co->co_convert = converter;                                            \
+        co->co_free = del_freefn;                                              \
+        co->co_type = CHANGELOG_OPT_REC_ENTRY;                                 \
+        gf_uuid_copy(co->co_entry.cef_uuid, pargfid);                          \
+        co->co_entry.cef_bname = gf_strdup(bname);                             \
+        if (!co->co_entry.cef_bname)                                           \
+            goto label;                                                        \
+        xlen += (UUID_CANONICAL_FORM_LEN + strlen(bname));                     \
+        if (!capture_del ||                                                    \
+            resolve_pargfid_to_path(this, pargfid, &(co->co_entry.cef_path),   \
+                                    co->co_entry.cef_bname)) {                 \
+            co->co_entry.cef_path = gf_strdup("\0");                           \
+            xlen += 1;                                                         \
+        } else {                                                               \
+            xlen += (strlen(co->co_entry.cef_path));                           \
+        }                                                                      \
+    } while (0)
+
+#define CHANGELOG_INIT(this, local, inode, gfid, xrec)                         \
+    local = changelog_local_init(this, inode, gfid, xrec, _gf_false)
+
+#define CHANGELOG_INIT_NOCHECK(this, local, inode, gfid, xrec)                 \
+    local = changelog_local_init(this, inode, gfid, xrec, _gf_true)
+
+#define CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, label)                     \
+    do {                                                                       \
+        if (!priv->active)                                                     \
+            goto label;                                                        \
+        /* ignore rebalance process's activity. */                             \
+        if ((frame->root->pid == GF_CLIENT_PID_DEFRAG) ||                      \
+            (frame->root->pid == GF_CLIENT_PID_TIER_DEFRAG))                   \
+            goto label;                                                        \
+    } while (0)
+
+/* If it is a METADATA entry and fop num being GF_FOP_NULL, don't
+ * log in the changelog as it is of no use. And also if it is
+ * logged, since slicing version checking is done for metadata
+ * entries, the subsequent entries with valid fop num which falls
+ * to same changelog will be missed. Hence check for boundary
+ * condition.
+ */
+#define CHANGELOG_OP_BOUNDARY_CHECK(frame, label)                              \
+    do {                                                                       \
+        if (frame->root->op <= GF_FOP_NULL ||                                  \
+            frame->root->op >= GF_FOP_MAXVALUE)                                \
+            goto label;                                                        \
+    } while (0)
+
+/**
+ * ignore internal fops for all clients except AFR self-heal daemon
+ */
+#define CHANGELOG_IF_INTERNAL_FOP_THEN_GOTO(frame, dict, label)                \
+    do {                                                                       \
+        if ((frame->root->pid != GF_CLIENT_PID_SELF_HEALD) && dict &&          \
+            dict_get(dict, GLUSTERFS_INTERNAL_FOP_KEY))                        \
+            goto label;                                                        \
+    } while (0)
+
+#define CHANGELOG_COND_GOTO(priv, cond, label)                                 \
+    do {                                                                       \
+        if (!priv->active || cond)                                             \
+            goto label;                                                        \
+    } while (0)
+
+/* Begin: Geo-Rep snapshot dependency changes */
+
+#define DICT_ERROR -1
+#define BARRIER_OFF 0
+#define BARRIER_ON 1
+#define DICT_DEFAULT 2
+
+#define CHANGELOG_NOT_ON_THEN_GOTO(priv, ret, label)                           \
+    do {                                                                       \
+        if (!priv->active) {                                                   \
+            gf_smsg(this->name, GF_LOG_WARNING, 0,                             \
+                    CHANGELOG_MSG_CHANGELOG_NOT_ACTIVE, NULL);                 \
+            ret = 0;                                                           \
+            goto label;                                                        \
+        }                                                                      \
+    } while (0)
+
+/* Log pthread error and goto label */
+#define CHANGELOG_PTHREAD_ERROR_HANDLE_0(ret, label)                           \
+    do {                                                                       \
+        if (ret) {                                                             \
+            gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_MSG_PTHREAD_ERROR,  \
+                    "error=%d", ret, NULL);                                    \
+            ret = -1;                                                          \
+            goto label;                                                        \
+        }                                                                      \
+    } while (0);
+
+/* Log pthread error, set flag and goto label */
+#define CHANGELOG_PTHREAD_ERROR_HANDLE_1(ret, label, flag)                     \
+    do {                                                                       \
+        if (ret) {                                                             \
+            gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_MSG_PTHREAD_ERROR,  \
+                    "error=%d", ret, NULL);                                    \
+            ret = -1;                                                          \
+            flag = _gf_true;                                                   \
+            goto label;                                                        \
+        }                                                                      \
+    } while (0)
+
+/* Log pthread error, unlock mutex and goto label */
+#define CHANGELOG_PTHREAD_ERROR_HANDLE_2(ret, label, mutex)                    \
+    do {                                                                       \
+        if (ret) {                                                             \
+            gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_MSG_PTHREAD_ERROR,  \
+                    "error=%d", ret, NULL);                                    \
+            ret = -1;                                                          \
+            pthread_mutex_unlock(&mutex);                                      \
+            goto label;                                                        \
+        }                                                                      \
+    } while (0)
+
+/* End: Geo-Rep snapshot dependency changes */
+
+#endif /* _CHANGELOG_HELPERS_H */
diff --git a/xlators/features/changelog/src/changelog-mem-types.h b/xlators/features/changelog/src/changelog-mem-types.h
new file mode 100644
index 00000000000..a2d8a9cbe93
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-mem-types.h
@@ -0,0 +1,34 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef _CHANGELOG_MEM_TYPES_H
+#define _CHANGELOG_MEM_TYPES_H
+
+#include <glusterfs/mem-types.h>
+
+enum gf_changelog_mem_types {
+    gf_changelog_mt_priv_t = gf_common_mt_end + 1,
+    gf_changelog_mt_str_t = gf_common_mt_end + 2,
+    gf_changelog_mt_batch_t = gf_common_mt_end + 3,
+    gf_changelog_mt_rt_t = gf_common_mt_end + 4,
+    gf_changelog_mt_inode_ctx_t = gf_common_mt_end + 5,
+    gf_changelog_mt_rpc_clnt_t = gf_common_mt_end + 6,
+    gf_changelog_mt_libgfchangelog_t = gf_common_mt_end + 7,
+    gf_changelog_mt_libgfchangelog_entry_t = gf_common_mt_end + 8,
+    gf_changelog_mt_libgfchangelog_rl_t = gf_common_mt_end + 9,
+    gf_changelog_mt_changelog_buffer_t = gf_common_mt_end + 10,
+    gf_changelog_mt_history_data_t = gf_common_mt_end + 11,
+    gf_changelog_mt_libgfchangelog_call_pool_t = gf_common_mt_end + 12,
+    gf_changelog_mt_libgfchangelog_event_t = gf_common_mt_end + 13,
+    gf_changelog_mt_ev_dispatcher_t = gf_common_mt_end + 14,
+    gf_changelog_mt_end
+};
+
+#endif
diff --git a/xlators/features/changelog/src/changelog-messages.h b/xlators/features/changelog/src/changelog-messages.h
new file mode 100644
index 00000000000..cb0e16c85d8
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-messages.h
@@ -0,0 +1,172 @@
+/*
+ Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+ This file is part of GlusterFS.
+
+ This file is licensed to you under your choice of the GNU Lesser
+ General Public License, version 3 or any later version (LGPLv3 or
+ later), or the GNU General Public License, version 2 (GPLv2), in all
+ cases as published by the Free Software Foundation.
+ */
+
+#ifndef _CHANGELOG_MESSAGES_H_
+#define _CHANGELOG_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(
+    CHANGELOG, CHANGELOG_MSG_OPEN_FAILED, CHANGELOG_MSG_BARRIER_FOP_FAILED,
+    CHANGELOG_MSG_VOL_MISCONFIGURED, CHANGELOG_MSG_RENAME_ERROR,
+    CHANGELOG_MSG_READ_ERROR, CHANGELOG_MSG_HTIME_ERROR,
+    CHANGELOG_MSG_PTHREAD_MUTEX_INIT_FAILED,
+    CHANGELOG_MSG_PTHREAD_COND_INIT_FAILED, CHANGELOG_MSG_CHILD_MISCONFIGURED,
+    CHANGELOG_MSG_DIR_OPTIONS_NOT_SET, CHANGELOG_MSG_CLOSE_ERROR,
+    CHANGELOG_MSG_PIPE_CREATION_ERROR, CHANGELOG_MSG_DICT_GET_FAILED,
+    CHANGELOG_MSG_BARRIER_INFO, CHANGELOG_MSG_BARRIER_ERROR,
+    CHANGELOG_MSG_GET_TIME_OP_FAILED, CHANGELOG_MSG_WRITE_FAILED,
+    CHANGELOG_MSG_PTHREAD_ERROR, CHANGELOG_MSG_INODE_NOT_FOUND,
+    CHANGELOG_MSG_FSYNC_OP_FAILED, CHANGELOG_MSG_TOTAL_LOG_INFO,
+    CHANGELOG_MSG_SNAP_INFO, CHANGELOG_MSG_SELECT_FAILED,
+    CHANGELOG_MSG_FCNTL_FAILED, CHANGELOG_MSG_BNOTIFY_INFO,
+    CHANGELOG_MSG_ENTRY_BUF_INFO, CHANGELOG_MSG_CHANGELOG_NOT_ACTIVE,
+    CHANGELOG_MSG_LOCAL_INIT_FAILED, CHANGELOG_MSG_NOTIFY_REGISTER_FAILED,
+    CHANGELOG_MSG_PROGRAM_NAME_REG_FAILED, CHANGELOG_MSG_HANDLE_PROBE_ERROR,
+    CHANGELOG_MSG_SET_FD_CONTEXT, CHANGELOG_MSG_FREEUP_FAILED,
+    CHANGELOG_MSG_RECONFIGURE, CHANGELOG_MSG_RPC_SUBMIT_REPLY_FAILED,
+    CHANGELOG_MSG_RPC_BUILD_ERROR, CHANGELOG_MSG_RPC_CONNECT_ERROR,
+    CHANGELOG_MSG_RPC_START_ERROR, CHANGELOG_MSG_BUFFER_STARVATION_ERROR,
+    CHANGELOG_MSG_SCAN_DIR_FAILED, CHANGELOG_MSG_FSETXATTR_FAILED,
+    CHANGELOG_MSG_FGETXATTR_FAILED, CHANGELOG_MSG_CLEANUP_ON_ACTIVE_REF,
+    CHANGELOG_MSG_DISPATCH_EVENT_FAILED, CHANGELOG_MSG_PUT_BUFFER_FAILED,
+    CHANGELOG_MSG_PTHREAD_COND_WAIT_FAILED, CHANGELOG_MSG_PTHREAD_CANCEL_FAILED,
+    CHANGELOG_MSG_INJECT_FSYNC_FAILED, CHANGELOG_MSG_CREATE_FRAME_FAILED,
+    CHANGELOG_MSG_FSTAT_OP_FAILED, CHANGELOG_MSG_LSEEK_OP_FAILED,
+    CHANGELOG_MSG_STRSTR_OP_FAILED, CHANGELOG_MSG_UNLINK_OP_FAILED,
+    CHANGELOG_MSG_DETECT_EMPTY_CHANGELOG_FAILED,
+    CHANGELOG_MSG_READLINK_OP_FAILED, CHANGELOG_MSG_EXPLICIT_ROLLOVER_FAILED,
+    CHANGELOG_MSG_RPCSVC_NOTIFY_FAILED, CHANGELOG_MSG_MEMORY_INIT_FAILED,
+    CHANGELOG_MSG_NO_MEMORY, CHANGELOG_MSG_HTIME_STAT_ERROR,
+    CHANGELOG_MSG_HTIME_CURRENT_ERROR, CHANGELOG_MSG_BNOTIFY_COND_INFO,
+    CHANGELOG_MSG_NO_HTIME_CURRENT, CHANGELOG_MSG_HTIME_CURRENT,
+    CHANGELOG_MSG_NEW_HTIME_FILE, CHANGELOG_MSG_MKDIR_ERROR,
+    CHANGELOG_MSG_PATH_NOT_FOUND, CHANGELOG_MSG_XATTR_INIT_FAILED,
+    CHANGELOG_MSG_WROTE_TO_CSNAP, CHANGELOG_MSG_UNUSED_0,
+    CHANGELOG_MSG_GET_BUFFER_FAILED, CHANGELOG_MSG_BARRIER_STATE_NOTIFY,
+    CHANGELOG_MSG_BARRIER_DISABLED, CHANGELOG_MSG_BARRIER_ALREADY_DISABLED,
+    CHANGELOG_MSG_BARRIER_ON_ERROR, CHANGELOG_MSG_BARRIER_ENABLE,
+    CHANGELOG_MSG_BARRIER_KEY_NOT_FOUND, CHANGELOG_MSG_ERROR_IN_DICT_GET,
+    CHANGELOG_MSG_UNUSED_1, CHANGELOG_MSG_UNUSED_2,
+    CHANGELOG_MSG_DEQUEUING_BARRIER_FOPS,
+    CHANGELOG_MSG_DEQUEUING_BARRIER_FOPS_FINISHED,
+    CHANGELOG_MSG_BARRIER_TIMEOUT, CHANGELOG_MSG_TIMEOUT_ADD_FAILED,
+    CHANGELOG_MSG_CLEANUP_ALREADY_SET);
+
+#define CHANGELOG_MSG_BARRIER_FOP_FAILED_STR                                   \
+    "failed to barrier FOPs, disabling changelog barrier"
+#define CHANGELOG_MSG_MEMORY_INIT_FAILED_STR "memory accounting init failed"
+#define CHANGELOG_MSG_NO_MEMORY_STR "failed to create local memory pool"
+#define CHANGELOG_MSG_ENTRY_BUF_INFO_STR                                       \
+    "Entry cannot be captured for gfid, Capturing DATA entry."
+#define CHANGELOG_MSG_PTHREAD_ERROR_STR "pthread error"
+#define CHANGELOG_MSG_PTHREAD_MUTEX_INIT_FAILED_STR "pthread_mutex_init failed"
+#define CHANGELOG_MSG_PTHREAD_COND_INIT_FAILED_STR "pthread_cond_init failed"
+#define CHANGELOG_MSG_HTIME_ERROR_STR "failed to update HTIME file"
+#define CHANGELOG_MSG_HTIME_STAT_ERROR_STR "unable to stat htime file"
+#define CHANGELOG_MSG_HTIME_CURRENT_ERROR_STR "Error extracting HTIME_CURRENT."
+#define CHANGELOG_MSG_UNLINK_OP_FAILED_STR "error unlinking empty changelog"
+#define CHANGELOG_MSG_RENAME_ERROR_STR "error renaming"
+#define CHANGELOG_MSG_MKDIR_ERROR_STR "unable to create directory"
+#define CHANGELOG_MSG_BNOTIFY_INFO_STR                                         \
+    "Explicit rollover changelog signaling bnotify"
+#define CHANGELOG_MSG_BNOTIFY_COND_INFO_STR "Woke up: bnotify conditional wait"
+#define CHANGELOG_MSG_RECONFIGURE_STR "Reconfigure: Changelog Enable"
+#define CHANGELOG_MSG_NO_HTIME_CURRENT_STR                                     \
+    "HTIME_CURRENT not found. Changelog enabled before init"
+#define CHANGELOG_MSG_HTIME_CURRENT_STR "HTIME_CURRENT"
+#define CHANGELOG_MSG_NEW_HTIME_FILE_STR                                       \
+    "Changelog enable: Creating new HTIME file"
+#define CHANGELOG_MSG_FGETXATTR_FAILED_STR "fgetxattr failed"
+#define CHANGELOG_MSG_TOTAL_LOG_INFO_STR "changelog info"
+#define CHANGELOG_MSG_PTHREAD_COND_WAIT_FAILED_STR "pthread cond wait failed"
+#define CHANGELOG_MSG_INODE_NOT_FOUND_STR "inode not found"
+#define CHANGELOG_MSG_READLINK_OP_FAILED_STR                                   \
+    "could not read the link from the gfid handle"
+#define CHANGELOG_MSG_OPEN_FAILED_STR "unable to open file"
+#define CHANGELOG_MSG_RPC_CONNECT_ERROR_STR "failed to connect back"
+#define CHANGELOG_MSG_BUFFER_STARVATION_ERROR_STR                              \
+    "Failed to get buffer for RPC dispatch"
+#define CHANGELOG_MSG_PTHREAD_CANCEL_FAILED_STR "could not cancel thread"
+#define CHANGELOG_MSG_FSTAT_OP_FAILED_STR "Could not stat (CHANGELOG)"
+#define CHANGELOG_MSG_LSEEK_OP_FAILED_STR "Could not lseek (changelog)"
+#define CHANGELOG_MSG_PATH_NOT_FOUND_STR                                       \
+    "Could not find CHANGELOG in changelog path"
+#define CHANGELOG_MSG_FSYNC_OP_FAILED_STR "fsync failed"
+#define CHANGELOG_MSG_DETECT_EMPTY_CHANGELOG_FAILED_STR                        \
+    "Error detecting empty changelog"
+#define CHANGELOG_MSG_EXPLICIT_ROLLOVER_FAILED_STR                             \
+    "Fail snapshot because of previous errors"
+#define CHANGELOG_MSG_SCAN_DIR_FAILED_STR "scandir failed"
+#define CHANGELOG_MSG_FSETXATTR_FAILED_STR "fsetxattr failed"
+#define CHANGELOG_MSG_XATTR_INIT_FAILED_STR "Htime xattr initialization failed"
+#define CHANGELOG_MSG_SNAP_INFO_STR "log in call path"
+#define CHANGELOG_MSG_WRITE_FAILED_STR "error writing to disk"
+#define CHANGELOG_MSG_WROTE_TO_CSNAP_STR "Successfully wrote to csnap"
+#define CHANGELOG_MSG_GET_TIME_OP_FAILED_STR "Problem rolling over changelog(s)"
+#define CHANGELOG_MSG_BARRIER_INFO_STR "Explicit wakeup on barrier notify"
+#define CHANGELOG_MSG_SELECT_FAILED_STR "pthread_cond_timedwait failed"
+#define CHANGELOG_MSG_INJECT_FSYNC_FAILED_STR "failed to inject fsync event"
+#define CHANGELOG_MSG_LOCAL_INIT_FAILED_STR                                    \
+    "changelog local initialization failed"
+#define CHANGELOG_MSG_GET_BUFFER_FAILED_STR "Failed to get buffer"
+#define CHANGELOG_MSG_SET_FD_CONTEXT_STR                                       \
+    "could not set fd context(for release cbk)"
+#define CHANGELOG_MSG_DICT_GET_FAILED_STR "Barrier failed"
+#define CHANGELOG_MSG_BARRIER_STATE_NOTIFY_STR "Barrier notification"
+#define CHANGELOG_MSG_BARRIER_ERROR_STR                                        \
+    "Received another barrier off notification while already off"
+#define CHANGELOG_MSG_BARRIER_DISABLED_STR "disabled changelog barrier"
+#define CHANGELOG_MSG_BARRIER_ALREADY_DISABLED_STR                             \
+    "Changelog barrier already disabled"
+#define CHANGELOG_MSG_BARRIER_ON_ERROR_STR                                     \
+    "Received another barrier on notification when last one is not served yet"
+#define CHANGELOG_MSG_BARRIER_ENABLE_STR "Enabled changelog barrier"
+#define CHANGELOG_MSG_BARRIER_KEY_NOT_FOUND_STR "barrier key not found"
+#define CHANGELOG_MSG_ERROR_IN_DICT_GET_STR                                    \
+    "Something went wrong in dict_get_str_boolean"
+#define CHANGELOG_MSG_DIR_OPTIONS_NOT_SET_STR "changelog-dir option is not set"
+#define CHANGELOG_MSG_FREEUP_FAILED_STR "could not cleanup bootstrapper"
+#define CHANGELOG_MSG_CHILD_MISCONFIGURED_STR                                  \
+    "translator needs a single subvolume"
+#define CHANGELOG_MSG_VOL_MISCONFIGURED_STR                                    \
+    "dangling volume. please check volfile"
+#define CHANGELOG_MSG_DEQUEUING_BARRIER_FOPS_STR                               \
+    "Dequeuing all the changelog barriered fops"
+#define CHANGELOG_MSG_DEQUEUING_BARRIER_FOPS_FINISHED_STR                      \
+    "Dequeuing changelog barriered fops is finished"
+#define CHANGELOG_MSG_BARRIER_TIMEOUT_STR                                      \
+    "Disabling changelog barrier because of the timeout"
+#define CHANGELOG_MSG_TIMEOUT_ADD_FAILED_STR                                   \
+    "Couldn't add changelog barrier timeout event"
+#define CHANGELOG_MSG_RPC_BUILD_ERROR_STR "failed to build rpc options"
+#define CHANGELOG_MSG_NOTIFY_REGISTER_FAILED_STR "failed to register notify"
+#define CHANGELOG_MSG_RPC_START_ERROR_STR "failed to start rpc"
+#define CHANGELOG_MSG_CREATE_FRAME_FAILED_STR "failed to create frame"
+#define CHANGELOG_MSG_RPC_SUBMIT_REPLY_FAILED_STR "failed to serialize reply"
+#define CHANGELOG_MSG_PROGRAM_NAME_REG_FAILED_STR "cannot register program"
+#define CHANGELOG_MSG_CHANGELOG_NOT_ACTIVE_STR                                 \
+    "Changelog is not active, return success"
+#define CHANGELOG_MSG_PUT_BUFFER_FAILED_STR                                    \
+    "failed to put buffer after consumption"
+#define CHANGELOG_MSG_CLEANUP_ALREADY_SET_STR                                  \
+    "cleanup_starting flag is already set for xl"
+#define CHANGELOG_MSG_HANDLE_PROBE_ERROR_STR "xdr decoding error"
+#endif /* !_CHANGELOG_MESSAGES_H_ */
diff --git a/xlators/features/changelog/src/changelog-misc.h b/xlators/features/changelog/src/changelog-misc.h
new file mode 100644
index 00000000000..e2addc09414
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-misc.h
@@ -0,0 +1,131 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef _CHANGELOG_MISC_H
+#define _CHANGELOG_MISC_H
+
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/common-utils.h>
+
+#define CHANGELOG_MAX_TYPE 4
+#define CHANGELOG_FILE_NAME "CHANGELOG"
+#define HTIME_FILE_NAME "HTIME"
+#define CSNAP_FILE_NAME "CHANGELOG.SNAP"
+#define HTIME_KEY "trusted.glusterfs.htime"
+#define HTIME_CURRENT "trusted.glusterfs.current_htime"
+#define HTIME_INITIAL_VALUE "0:0"
+
+#define CHANGELOG_VERSION_MAJOR 1
+#define CHANGELOG_VERSION_MINOR 2
+
+#define CHANGELOG_UNIX_SOCK DEFAULT_VAR_RUN_DIRECTORY "/changelog-%s.sock"
+#define CHANGELOG_TMP_UNIX_SOCK DEFAULT_VAR_RUN_DIRECTORY "/.%s%lu.sock"
+
+/**
+ * header starts with the version and the format of the changelog.
+ * 'version' not much of a use now.
+ */
+#define CHANGELOG_HEADER                                                       \
+    "GlusterFS Changelog | version: v%d.%d | encoding : %d\n"
+
+#define CHANGELOG_MAKE_SOCKET_PATH(brick_path, sockpath, len)                  \
+    do {                                                                       \
+        char xxh64[GF_XXH64_DIGEST_LENGTH * 2 + 1] = {                         \
+            0,                                                                 \
+        };                                                                     \
+        gf_xxh64_wrapper((unsigned char *)brick_path, strlen(brick_path),      \
+                         GF_XXHSUM64_DEFAULT_SEED, xxh64);                     \
+        (void)snprintf(sockpath, len, CHANGELOG_UNIX_SOCK, xxh64);             \
+    } while (0)
+
+#define CHANGELOG_MAKE_TMP_SOCKET_PATH(brick_path, sockpath, len)              \
+    do {                                                                       \
+        unsigned long pid = 0;                                                 \
+        char xxh64[GF_XXH64_DIGEST_LENGTH * 2 + 1] = {                         \
+            0,                                                                 \
+        };                                                                     \
+        pid = (unsigned long)getpid();                                         \
+        gf_xxh64_wrapper((unsigned char *)brick_path, strlen(brick_path),      \
+                         GF_XXHSUM64_DEFAULT_SEED, xxh64);                     \
+        (void)snprintf(sockpath, len, CHANGELOG_TMP_UNIX_SOCK, xxh64, pid);    \
+    } while (0)
+
+/**
+ * ... used by libgfchangelog.
+ */
+#define CHANGELOG_GET_HEADER_INFO(fd, buffer, len, enc, maj, min, elen)        \
+    do {                                                                       \
+        FILE *fp;                                                              \
+        int fd_dup;                                                            \
+                                                                               \
+        enc = -1;                                                              \
+        maj = -1;                                                              \
+        min = -1;                                                              \
+        fd_dup = dup(fd);                                                      \
+                                                                               \
+        if (fd_dup != -1) {                                                    \
+            fp = fdopen(fd_dup, "r");                                          \
+            if (fp) {                                                          \
+                if (fgets(buffer, len, fp)) {                                  \
+                    elen = strlen(buffer);                                     \
+                    sscanf(buffer, CHANGELOG_HEADER, &maj, &min, &enc);        \
+                }                                                              \
+                fclose(fp);                                                    \
+            } else {                                                           \
+                sys_close(fd_dup);                                             \
+            }                                                                  \
+        }                                                                      \
+    } while (0)
+
+#define CHANGELOG_FILL_HTIME_DIR(changelog_dir, path)                          \
+    do {                                                                       \
+        snprintf(path, sizeof(path), "%s/htime", changelog_dir);               \
+    } while (0)
+
+#define CHANGELOG_FILL_CSNAP_DIR(changelog_dir, path)                          \
+    do {                                                                       \
+        snprintf(path, sizeof(path), "%s/csnap", changelog_dir);               \
+    } while (0)
+/**
+ * everything after 'CHANGELOG_TYPE_METADATA_XATTR' are internal types
+ * (ie. none of the fops trigger this type of event), hence
+ * CHANGELOG_MAX_TYPE = 4
+ */
+typedef enum {
+    CHANGELOG_TYPE_DATA = 0,
+    CHANGELOG_TYPE_METADATA,
+    CHANGELOG_TYPE_ENTRY,
+    CHANGELOG_TYPE_METADATA_XATTR,
+    CHANGELOG_TYPE_ROLLOVER,
+    CHANGELOG_TYPE_FSYNC,
+} changelog_log_type;
+
+/* operation modes - RT for now */
+typedef enum {
+    CHANGELOG_MODE_RT = 0,
+} changelog_mode_t;
+
+/* encoder types */
+
+typedef enum {
+    CHANGELOG_ENCODE_MIN = 0,
+    CHANGELOG_ENCODE_BINARY,
+    CHANGELOG_ENCODE_ASCII,
+    CHANGELOG_ENCODE_MAX,
+} changelog_encoder_t;
+
+#define CHANGELOG_VALID_ENCODING(enc)                                          \
+    (enc > CHANGELOG_ENCODE_MIN && enc < CHANGELOG_ENCODE_MAX)
+
+#define CHANGELOG_TYPE_IS_ENTRY(type) (type == CHANGELOG_TYPE_ENTRY)
+#define CHANGELOG_TYPE_IS_ROLLOVER(type) (type == CHANGELOG_TYPE_ROLLOVER)
+#define CHANGELOG_TYPE_IS_FSYNC(type) (type == CHANGELOG_TYPE_FSYNC)
+
+#endif /* _CHANGELOG_MISC_H */
diff --git a/xlators/features/changelog/src/changelog-rpc-common.c b/xlators/features/changelog/src/changelog-rpc-common.c
new file mode 100644
index 00000000000..125246a17e1
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-rpc-common.c
@@ -0,0 +1,359 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include "changelog-rpc-common.h"
+#include "changelog-messages.h"
+
+#include <glusterfs/syscall.h>
+/**
+*****************************************************
+                  Client Interface
+*****************************************************
+*/
+
+/**
+ * Initialize and return an RPC client object for a given unix
+ * domain socket.
+ */
+
+void *
+changelog_rpc_poller(void *arg)
+{
+    xlator_t *this = arg;
+
+    (void)gf_event_dispatch(this->ctx->event_pool);
+    return NULL;
+}
+
+struct rpc_clnt *
+changelog_rpc_client_init(xlator_t *this, void *cbkdata, char *sockfile,
+                          rpc_clnt_notify_t fn)
+{
+    int ret = 0;
+    struct rpc_clnt *rpc = NULL;
+    dict_t *options = NULL;
+
+    if (!cbkdata)
+        cbkdata = this;
+
+    options = dict_new();
+    if (!options)
+        goto error_return;
+
+    ret = rpc_transport_unix_options_build(options, sockfile, 0);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_MSG_RPC_BUILD_ERROR,
+                NULL);
+        goto dealloc_dict;
+    }
+
+    rpc = rpc_clnt_new(options, this, this->name, 16);
+    if (!rpc)
+        goto dealloc_dict;
+
+    ret = rpc_clnt_register_notify(rpc, fn, cbkdata);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0,
+                CHANGELOG_MSG_NOTIFY_REGISTER_FAILED, NULL);
+        goto dealloc_rpc_clnt;
+    }
+
+    ret = rpc_clnt_start(rpc);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_MSG_RPC_START_ERROR,
+                NULL);
+        goto dealloc_rpc_clnt;
+    }
+
+    dict_unref(options);
+    return rpc;
+
+dealloc_rpc_clnt:
+    rpc_clnt_unref(rpc);
+dealloc_dict:
+    dict_unref(options);
+error_return:
+    return NULL;
+}
+
+/**
+ * Generic RPC client routine to dispatch a request to an
+ * RPC server.
+ */
+int
+changelog_rpc_sumbit_req(struct rpc_clnt *rpc, void *req, call_frame_t *frame,
+                         rpc_clnt_prog_t *prog, int procnum,
+                         struct iovec *payload, int payloadcnt,
+                         struct iobref *iobref, xlator_t *this,
+                         fop_cbk_fn_t cbkfn, xdrproc_t xdrproc)
+{
+    int ret = 0;
+    int count = 0;
+    struct iovec iov = {
+        0,
+    };
+    struct iobuf *iobuf = NULL;
+    char new_iobref = 0;
+    ssize_t xdr_size = 0;
+
+    GF_ASSERT(this);
+
+    if (req) {
+        xdr_size = xdr_sizeof(xdrproc, req);
+
+        iobuf = iobuf_get2(this->ctx->iobuf_pool, xdr_size);
+        if (!iobuf) {
+            goto out;
+        };
+
+        if (!iobref) {
+            iobref = iobref_new();
+            if (!iobref) {
+                goto out;
+            }
+
+            new_iobref = 1;
+        }
+
+        iobref_add(iobref, iobuf);
+
+        iov.iov_base = iobuf->ptr;
+        iov.iov_len = iobuf_size(iobuf);
+
+        /* Create the xdr payload */
+        ret = xdr_serialize_generic(iov, req, xdrproc);
+        if (ret == -1) {
+            goto out;
+        }
+
+        iov.iov_len = ret;
+        count = 1;
+    }
+
+    ret = rpc_clnt_submit(rpc, prog, procnum, cbkfn, &iov, count, payload,
+                          payloadcnt, iobref, frame, NULL, 0, NULL, 0, NULL);
+
+out:
+    if (new_iobref)
+        iobref_unref(iobref);
+    if (iobuf)
+        iobuf_unref(iobuf);
+    return ret;
+}
+
+/**
+ * Entry point to perform a remote procedure call
+ */
+int
+changelog_invoke_rpc(xlator_t *this, struct rpc_clnt *rpc,
+                     rpc_clnt_prog_t *prog, int procidx, void *arg)
+{
+    int ret = 0;
+    call_frame_t *frame = NULL;
+    rpc_clnt_procedure_t *proc = NULL;
+
+    if (!this || !prog)
+        goto error_return;
+
+    frame = create_frame(this, this->ctx->pool);
+    if (!frame) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_MSG_CREATE_FRAME_FAILED,
+                NULL);
+        goto error_return;
+    }
+
+    proc = &prog->proctable[procidx];
+    if (proc->fn)
+        ret = proc->fn(frame, this, arg);
+
+    STACK_DESTROY(frame->root);
+    return ret;
+
+error_return:
+    return -1;
+}
+
+/**
+*****************************************************
+                  Server Interface
+*****************************************************
+*/
+
+struct iobuf *
+__changelog_rpc_serialize_reply(rpcsvc_request_t *req, void *arg,
+                                struct iovec *outmsg, xdrproc_t xdrproc)
+{
+    struct iobuf *iob = NULL;
+    ssize_t retlen = 0;
+    ssize_t rsp_size = 0;
+
+    rsp_size = xdr_sizeof(xdrproc, arg);
+    iob = iobuf_get2(req->svc->ctx->iobuf_pool, rsp_size);
+    if (!iob)
+        goto error_return;
+
+    iobuf_to_iovec(iob, outmsg);
+
+    retlen = xdr_serialize_generic(*outmsg, arg, xdrproc);
+    if (retlen == -1)
+        goto unref_iob;
+
+    outmsg->iov_len = retlen;
+    return iob;
+
+unref_iob:
+    iobuf_unref(iob);
+error_return:
+    return NULL;
+}
+
+int
+changelog_rpc_sumbit_reply(rpcsvc_request_t *req, void *arg,
+                           struct iovec *payload, int payloadcount,
+                           struct iobref *iobref, xdrproc_t xdrproc)
+{
+    int ret = -1;
+    struct iobuf *iob = NULL;
+    struct iovec iov = {
+        0,
+    };
+    char new_iobref = 0;
+
+    if (!req)
+        goto return_ret;
+
+    if (!iobref) {
+        iobref = iobref_new();
+        if (!iobref)
+            goto return_ret;
+        new_iobref = 1;
+    }
+
+    iob = __changelog_rpc_serialize_reply(req, arg, &iov, xdrproc);
+    if (!iob)
+        gf_smsg("", GF_LOG_ERROR, 0, CHANGELOG_MSG_RPC_SUBMIT_REPLY_FAILED,
+                NULL);
+    else
+        iobref_add(iobref, iob);
+
+    ret = rpcsvc_submit_generic(req, &iov, 1, payload, payloadcount, iobref);
+
+    if (new_iobref)
+        iobref_unref(iobref);
+    if (iob)
+        iobuf_unref(iob);
+return_ret:
+    return ret;
+}
+
+void
+changelog_rpc_server_destroy(xlator_t *this, rpcsvc_t *rpc, char *sockfile,
+                             rpcsvc_notify_t fn, struct rpcsvc_program **progs)
+{
+    rpcsvc_listener_t *listener = NULL;
+    rpcsvc_listener_t *next = NULL;
+    struct rpcsvc_program *prog = NULL;
+    rpc_transport_t *trans = NULL;
+
+    if (!rpc)
+        return;
+
+    while (*progs) {
+        prog = *progs;
+        (void)rpcsvc_program_unregister(rpc, prog);
+        progs++;
+    }
+
+    list_for_each_entry_safe(listener, next, &rpc->listeners, list)
+    {
+        if (listener->trans) {
+            trans = listener->trans;
+            rpc_transport_disconnect(trans, _gf_false);
+        }
+    }
+
+    (void)rpcsvc_unregister_notify(rpc, fn, this);
+
+    /* TODO Avoid freeing rpc object in case of brick multiplex
+       after freeing rpc object svc->rpclock corrupted and it takes
+       more time to detach a brick
+    */
+    if (!this->cleanup_starting) {
+        if (rpc->rxpool) {
+            mem_pool_destroy(rpc->rxpool);
+            rpc->rxpool = NULL;
+        }
+        GF_FREE(rpc);
+    }
+}
+
+rpcsvc_t *
+changelog_rpc_server_init(xlator_t *this, char *sockfile, void *cbkdata,
+                          rpcsvc_notify_t fn, struct rpcsvc_program **progs)
+{
+    int ret = 0;
+    rpcsvc_t *rpc = NULL;
+    dict_t *options = NULL;
+    struct rpcsvc_program *prog = NULL;
+
+    if (!cbkdata)
+        cbkdata = this;
+
+    options = dict_new();
+    if (!options)
+        return NULL;
+
+    ret = rpcsvc_transport_unix_options_build(options, sockfile);
+    if (ret)
+        goto dealloc_dict;
+
+    rpc = rpcsvc_init(this, this->ctx, options, 8);
+    if (rpc == NULL) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_MSG_RPC_START_ERROR,
+                NULL);
+        goto dealloc_dict;
+    }
+
+    ret = rpcsvc_register_notify(rpc, fn, cbkdata);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0,
+                CHANGELOG_MSG_NOTIFY_REGISTER_FAILED, NULL);
+        goto dealloc_rpc;
+    }
+
+    ret = rpcsvc_create_listeners(rpc, options, this->name);
+    if (ret != 1) {
+        gf_msg_debug(this->name, 0, "failed to create listeners");
+        goto dealloc_rpc;
+    }
+
+    while (*progs) {
+        prog = *progs;
+        ret = rpcsvc_program_register(rpc, prog, _gf_false);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0,
+                    CHANGELOG_MSG_PROGRAM_NAME_REG_FAILED, "name%s",
+                    prog->progname, "prognum=%d", prog->prognum, "pogver=%d",
+                    prog->progver, NULL);
+            goto dealloc_rpc;
+        }
+
+        progs++;
+    }
+
+    dict_unref(options);
+    return rpc;
+
+dealloc_rpc:
+    GF_FREE(rpc);
+dealloc_dict:
+    dict_unref(options);
+    return NULL;
+}
diff --git a/xlators/features/changelog/src/changelog-rpc-common.h b/xlators/features/changelog/src/changelog-rpc-common.h
new file mode 100644
index 00000000000..4d9aa2c694b
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-rpc-common.h
@@ -0,0 +1,85 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __CHANGELOG_RPC_COMMON_H
+#define __CHANGELOG_RPC_COMMON_H
+
+#include "rpcsvc.h"
+#include "rpc-clnt.h"
+#include <glusterfs/gf-event.h>
+#include <glusterfs/call-stub.h>
+
+#include "changelog-xdr.h"
+#include "xdr-generic.h"
+
+#include "changelog.h"
+
+/**
+ * Let's keep this non-configurable for now.
+ */
+#define NR_ROTT_BUFFS 4
+#define NR_DISPATCHERS (NR_ROTT_BUFFS - 1)
+
+enum changelog_rpc_procnum {
+    CHANGELOG_RPC_PROC_NULL = 0,
+    CHANGELOG_RPC_PROBE_FILTER = 1,
+    CHANGELOG_RPC_PROC_MAX = 2,
+};
+
+#define CHANGELOG_RPC_PROGNUM 1885957735
+#define CHANGELOG_RPC_PROGVER 1
+
+/**
+ * reverse connection: data xfer path
+ */
+enum changelog_reverse_rpc_procnum {
+    CHANGELOG_REV_PROC_NULL = 0,
+    CHANGELOG_REV_PROC_EVENT = 1,
+    CHANGELOG_REV_PROC_MAX = 2,
+};
+
+#define CHANGELOG_REV_RPC_PROCNUM 1886350951
+#define CHANGELOG_REV_RPC_PROCVER 1
+
+typedef struct changelog_rpc {
+    rpcsvc_t *svc;
+    struct rpc_clnt *rpc;
+    char sock[UNIX_PATH_MAX]; /* tied to server */
+} changelog_rpc_t;
+
+/* event poller */
+void *
+changelog_rpc_poller(void *);
+
+/* CLIENT API */
+struct rpc_clnt *
+changelog_rpc_client_init(xlator_t *, void *, char *, rpc_clnt_notify_t);
+
+int
+changelog_rpc_sumbit_req(struct rpc_clnt *, void *, call_frame_t *,
+                         rpc_clnt_prog_t *, int, struct iovec *, int,
+                         struct iobref *, xlator_t *, fop_cbk_fn_t, xdrproc_t);
+
+int
+changelog_invoke_rpc(xlator_t *, struct rpc_clnt *, rpc_clnt_prog_t *, int,
+                     void *);
+
+/* SERVER API */
+int
+changelog_rpc_sumbit_reply(rpcsvc_request_t *, void *, struct iovec *, int,
+                           struct iobref *, xdrproc_t);
+rpcsvc_t *
+changelog_rpc_server_init(xlator_t *, char *, void *, rpcsvc_notify_t,
+                          struct rpcsvc_program **);
+void
+changelog_rpc_server_destroy(xlator_t *, rpcsvc_t *, char *, rpcsvc_notify_t,
+                             struct rpcsvc_program **);
+
+#endif
diff --git a/xlators/features/changelog/src/changelog-rpc.c b/xlators/features/changelog/src/changelog-rpc.c
new file mode 100644
index 00000000000..440b88091a6
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-rpc.c
@@ -0,0 +1,440 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/syscall.h>
+#include "changelog-rpc.h"
+#include "changelog-mem-types.h"
+#include "changelog-ev-handle.h"
+
+static struct rpcsvc_program *changelog_programs[];
+
+static void
+changelog_cleanup_dispatchers(xlator_t *this, changelog_priv_t *priv, int count)
+{
+    for (count--; count >= 0; count--) {
+        (void)changelog_thread_cleanup(this, priv->ev_dispatcher[count]);
+        priv->ev_dispatcher[count] = 0;
+    }
+}
+
+int
+changelog_cleanup_rpc_threads(xlator_t *this, changelog_priv_t *priv)
+{
+    int ret = 0;
+    changelog_clnt_t *conn = NULL;
+
+    conn = &priv->connections;
+    if (!conn)
+        return 0;
+
+    /** terminate RPC thread(s) */
+    ret = changelog_thread_cleanup(this, priv->connector);
+    if (ret != 0)
+        goto error_return;
+    priv->connector = 0;
+
+    /** terminate dispatcher thread(s) */
+    changelog_cleanup_dispatchers(this, priv, priv->nr_dispatchers);
+
+    /* destroy locks */
+    ret = pthread_mutex_destroy(&conn->pending_lock);
+    if (ret != 0)
+        goto error_return;
+    ret = pthread_cond_destroy(&conn->pending_cond);
+    if (ret != 0)
+        goto error_return;
+    ret = LOCK_DESTROY(&conn->active_lock);
+    if (ret != 0)
+        goto error_return;
+    ret = LOCK_DESTROY(&conn->wait_lock);
+    if (ret != 0)
+        goto error_return;
+    return 0;
+
+error_return:
+    return -1;
+}
+
+static int
+changelog_init_rpc_threads(xlator_t *this, changelog_priv_t *priv, rbuf_t *rbuf,
+                           int nr_dispatchers)
+{
+    int j = 0;
+    int ret = 0;
+    changelog_clnt_t *conn = NULL;
+
+    conn = &priv->connections;
+
+    conn->this = this;
+    conn->rbuf = rbuf;
+    conn->sequence = 1; /* start with sequence number one */
+
+    INIT_LIST_HEAD(&conn->pending);
+    INIT_LIST_HEAD(&conn->active);
+    INIT_LIST_HEAD(&conn->waitq);
+
+    ret = pthread_mutex_init(&conn->pending_lock, NULL);
+    if (ret)
+        goto error_return;
+    ret = pthread_cond_init(&conn->pending_cond, NULL);
+    if (ret)
+        goto cleanup_pending_lock;
+
+    ret = LOCK_INIT(&conn->active_lock);
+    if (ret)
+        goto cleanup_pending_cond;
+    ret = LOCK_INIT(&conn->wait_lock);
+    if (ret)
+        goto cleanup_active_lock;
+
+    /* spawn reverse connection thread */
+    ret = gf_thread_create(&priv->connector, NULL, changelog_ev_connector, conn,
+                           "clogecon");
+    if (ret != 0)
+        goto cleanup_wait_lock;
+
+    /* spawn dispatcher thread(s) */
+    priv->ev_dispatcher = GF_CALLOC(nr_dispatchers, sizeof(pthread_t),
+                                    gf_changelog_mt_ev_dispatcher_t);
+    if (!priv->ev_dispatcher)
+        goto cleanup_connector;
+
+    /* spawn dispatcher threads */
+    for (; j < nr_dispatchers; j++) {
+        ret = gf_thread_create(&priv->ev_dispatcher[j], NULL,
+                               changelog_ev_dispatch, conn, "clogd%03hx",
+                               j & 0x3ff);
+        if (ret != 0) {
+            changelog_cleanup_dispatchers(this, priv, j);
+            break;
+        }
+    }
+
+    if (ret != 0)
+        goto cleanup_connector;
+
+    priv->nr_dispatchers = nr_dispatchers;
+    return 0;
+
+cleanup_connector:
+    (void)pthread_cancel(priv->connector);
+cleanup_wait_lock:
+    LOCK_DESTROY(&conn->wait_lock);
+cleanup_active_lock:
+    LOCK_DESTROY(&conn->active_lock);
+cleanup_pending_cond:
+    (void)pthread_cond_destroy(&conn->pending_cond);
+cleanup_pending_lock:
+    (void)pthread_mutex_destroy(&conn->pending_lock);
+error_return:
+    return -1;
+}
+
+int
+changelog_rpcsvc_notify(rpcsvc_t *rpc, void *xl, rpcsvc_event_t event,
+                        void *data)
+{
+    xlator_t *this = NULL;
+    rpc_transport_t *trans = NULL;
+    rpc_transport_t *xprt = NULL;
+    rpc_transport_t *xp_next = NULL;
+    changelog_priv_t *priv = NULL;
+    uint64_t listnercnt = 0;
+    uint64_t xprtcnt = 0;
+    uint64_t clntcnt = 0;
+    rpcsvc_listener_t *listener = NULL;
+    rpcsvc_listener_t *next = NULL;
+    gf_boolean_t listner_found = _gf_false;
+    socket_private_t *sockpriv = NULL;
+
+    if (!xl || !data || !rpc) {
+        gf_msg_callingfn("changelog", GF_LOG_WARNING, 0,
+                         CHANGELOG_MSG_RPCSVC_NOTIFY_FAILED,
+                         "Calling rpc_notify without initializing");
+        goto out;
+    }
+
+    this = xl;
+    trans = data;
+    priv = this->private;
+
+    if (!priv) {
+        gf_msg_callingfn("changelog", GF_LOG_WARNING, 0,
+                         CHANGELOG_MSG_RPCSVC_NOTIFY_FAILED,
+                         "Calling rpc_notify without priv initializing");
+        goto out;
+    }
+
+    if (event == RPCSVC_EVENT_ACCEPT) {
+        GF_ATOMIC_INC(priv->xprtcnt);
+        LOCK(&priv->lock);
+        {
+            list_add_tail(&trans->list, &priv->xprt_list);
+        }
+        UNLOCK(&priv->lock);
+        goto out;
+    }
+
+    if (event == RPCSVC_EVENT_DISCONNECT) {
+        list_for_each_entry_safe(listener, next, &rpc->listeners, list)
+        {
+            if (listener && listener->trans) {
+                if (listener->trans == trans) {
+                    listnercnt = GF_ATOMIC_DEC(priv->listnercnt);
+                    listner_found = _gf_true;
+                    rpcsvc_listener_destroy(listener);
+                }
+            }
+        }
+
+        if (listnercnt > 0) {
+            goto out;
+        }
+        if (listner_found) {
+            LOCK(&priv->lock);
+            list_for_each_entry_safe(xprt, xp_next, &priv->xprt_list, list)
+            {
+                sockpriv = (socket_private_t *)(xprt->private);
+                gf_log("changelog", GF_LOG_INFO,
+                       "Send disconnect"
+                       " on socket %d",
+                       sockpriv->sock);
+                rpc_transport_disconnect(xprt, _gf_false);
+            }
+            UNLOCK(&priv->lock);
+            goto out;
+        }
+        LOCK(&priv->lock);
+        {
+            list_del_init(&trans->list);
+        }
+        UNLOCK(&priv->lock);
+
+        xprtcnt = GF_ATOMIC_DEC(priv->xprtcnt);
+        clntcnt = GF_ATOMIC_GET(priv->clntcnt);
+        if (!xprtcnt && !clntcnt) {
+            changelog_process_cleanup_event(this);
+        }
+    }
+
+out:
+    return 0;
+}
+
+void
+changelog_process_cleanup_event(xlator_t *this)
+{
+    gf_boolean_t cleanup_notify = _gf_false;
+    changelog_priv_t *priv = NULL;
+    char sockfile[UNIX_PATH_MAX] = {
+        0,
+    };
+
+    if (!this)
+        return;
+    priv = this->private;
+    if (!priv)
+        return;
+
+    LOCK(&priv->lock);
+    {
+        cleanup_notify = priv->notify_down;
+        priv->notify_down = _gf_true;
+    }
+    UNLOCK(&priv->lock);
+
+    if (priv->victim && !cleanup_notify) {
+        default_notify(this, GF_EVENT_PARENT_DOWN, priv->victim);
+
+        if (priv->rpc) {
+            /* sockfile path could have been saved to avoid this */
+            CHANGELOG_MAKE_SOCKET_PATH(priv->changelog_brick, sockfile,
+                                       UNIX_PATH_MAX);
+            sys_unlink(sockfile);
+            (void)rpcsvc_unregister_notify(priv->rpc, changelog_rpcsvc_notify,
+                                           this);
+            if (priv->rpc->rxpool) {
+                mem_pool_destroy(priv->rpc->rxpool);
+                priv->rpc->rxpool = NULL;
+            }
+            GF_FREE(priv->rpc);
+            priv->rpc = NULL;
+        }
+    }
+}
+
+void
+changelog_destroy_rpc_listner(xlator_t *this, changelog_priv_t *priv)
+{
+    char sockfile[UNIX_PATH_MAX] = {
+        0,
+    };
+
+    /* sockfile path could have been saved to avoid this */
+    CHANGELOG_MAKE_SOCKET_PATH(priv->changelog_brick, sockfile, UNIX_PATH_MAX);
+    changelog_rpc_server_destroy(this, priv->rpc, sockfile,
+                                 changelog_rpcsvc_notify, changelog_programs);
+}
+
+rpcsvc_t *
+changelog_init_rpc_listener(xlator_t *this, changelog_priv_t *priv,
+                            rbuf_t *rbuf, int nr_dispatchers)
+{
+    int ret = 0;
+    char sockfile[UNIX_PATH_MAX] = {
+        0,
+    };
+    rpcsvc_t *svcp;
+
+    ret = changelog_init_rpc_threads(this, priv, rbuf, nr_dispatchers);
+    if (ret)
+        return NULL;
+
+    CHANGELOG_MAKE_SOCKET_PATH(priv->changelog_brick, sockfile, UNIX_PATH_MAX);
+    (void)sys_unlink(sockfile);
+    svcp = changelog_rpc_server_init(
+        this, sockfile, NULL, changelog_rpcsvc_notify, changelog_programs);
+    return svcp;
+}
+
+void
+changelog_rpc_clnt_cleanup(changelog_rpc_clnt_t *crpc)
+{
+    if (!crpc)
+        return;
+    crpc->c_clnt = NULL;
+    LOCK_DESTROY(&crpc->lock);
+    GF_FREE(crpc);
+}
+
+static changelog_rpc_clnt_t *
+changelog_rpc_clnt_init(xlator_t *this, changelog_probe_req *rpc_req,
+                        changelog_clnt_t *c_clnt)
+{
+    int ret = 0;
+    changelog_rpc_clnt_t *crpc = NULL;
+
+    crpc = GF_CALLOC(1, sizeof(*crpc), gf_changelog_mt_rpc_clnt_t);
+    if (!crpc)
+        goto error_return;
+    INIT_LIST_HEAD(&crpc->list);
+
+    /* Take a ref, the last unref will be on RPC_CLNT_DESTROY
+     * which comes as a result of last rpc_clnt_unref.
+     */
+    GF_ATOMIC_INIT(crpc->ref, 1);
+    changelog_set_disconnect_flag(crpc, _gf_false);
+
+    crpc->filter = rpc_req->filter;
+    (void)memcpy(crpc->sock, rpc_req->sock, strlen(rpc_req->sock));
+
+    crpc->this = this;
+    crpc->c_clnt = c_clnt;
+    crpc->cleanup = changelog_rpc_clnt_cleanup;
+
+    ret = LOCK_INIT(&crpc->lock);
+    if (ret != 0)
+        goto dealloc_crpc;
+    return crpc;
+
+dealloc_crpc:
+    GF_FREE(crpc);
+error_return:
+    return NULL;
+}
+
+/**
+ * Actor declarations
+ */
+
+/**
+ * @probe_handler
+ * A probe RPC call spawns a connect back to the caller. Caller also
+ * passes an hint which acts as a filter for selecting updates.
+ */
+
+int
+changelog_handle_probe(rpcsvc_request_t *req)
+{
+    int ret = 0;
+    xlator_t *this = NULL;
+    rpcsvc_t *svc = NULL;
+    changelog_priv_t *priv = NULL;
+    changelog_clnt_t *c_clnt = NULL;
+    changelog_rpc_clnt_t *crpc = NULL;
+
+    changelog_probe_req rpc_req = {
+        0,
+    };
+    changelog_probe_rsp rpc_rsp = {
+        0,
+    };
+
+    this = req->trans->xl;
+    if (this->cleanup_starting) {
+        gf_smsg(this->name, GF_LOG_DEBUG, 0, CHANGELOG_MSG_CLEANUP_ALREADY_SET,
+                NULL);
+        return 0;
+    }
+
+    ret = xdr_to_generic(req->msg[0], &rpc_req,
+                         (xdrproc_t)xdr_changelog_probe_req);
+    if (ret < 0) {
+        gf_smsg("", GF_LOG_ERROR, 0, CHANGELOG_MSG_HANDLE_PROBE_ERROR, NULL);
+        req->rpc_err = GARBAGE_ARGS;
+        goto handle_xdr_error;
+    }
+
+    /* ->xl hidden in rpcsvc */
+    svc = rpcsvc_request_service(req);
+    this = svc->xl;
+    priv = this->private;
+    c_clnt = &priv->connections;
+
+    crpc = changelog_rpc_clnt_init(this, &rpc_req, c_clnt);
+    if (!crpc)
+        goto handle_xdr_error;
+
+    changelog_ev_queue_connection(c_clnt, crpc);
+    rpc_rsp.op_ret = 0;
+
+    goto submit_rpc;
+
+handle_xdr_error:
+    rpc_rsp.op_ret = -1;
+submit_rpc:
+    (void)changelog_rpc_sumbit_reply(req, &rpc_rsp, NULL, 0, NULL,
+                                     (xdrproc_t)xdr_changelog_probe_rsp);
+    return 0;
+}
+
+/**
+ * RPC declarations
+ */
+
+static rpcsvc_actor_t changelog_svc_actors[CHANGELOG_RPC_PROC_MAX] = {
+    [CHANGELOG_RPC_PROBE_FILTER] = {"CHANGELOG PROBE FILTER",
+                                    changelog_handle_probe, NULL,
+                                    CHANGELOG_RPC_PROBE_FILTER, DRC_NA, 0},
+};
+
+static struct rpcsvc_program changelog_svc_prog = {
+    .progname = CHANGELOG_RPC_PROGNAME,
+    .prognum = CHANGELOG_RPC_PROGNUM,
+    .progver = CHANGELOG_RPC_PROGVER,
+    .numactors = CHANGELOG_RPC_PROC_MAX,
+    .actors = changelog_svc_actors,
+    .synctask = _gf_true,
+};
+
+static struct rpcsvc_program *changelog_programs[] = {
+    &changelog_svc_prog,
+    NULL,
+};
diff --git a/xlators/features/changelog/src/changelog-rpc.h b/xlators/features/changelog/src/changelog-rpc.h
new file mode 100644
index 00000000000..b1707565249
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-rpc.h
@@ -0,0 +1,31 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __CHANGELOG_RPC_H
+#define __CHANGELOG_RPC_H
+
+#include <glusterfs/xlator.h>
+#include "changelog-helpers.h"
+
+/* one time */
+#include "socket.h"
+#include "changelog-rpc-common.h"
+
+#define CHANGELOG_RPC_PROGNAME "GlusterFS Changelog"
+
+rpcsvc_t *
+changelog_init_rpc_listener(xlator_t *, changelog_priv_t *, rbuf_t *, int);
+
+void
+changelog_destroy_rpc_listner(xlator_t *, changelog_priv_t *);
+
+int
+changelog_cleanup_rpc_threads(xlator_t *this, changelog_priv_t *priv);
+#endif
diff --git a/xlators/features/changelog/src/changelog-rt.c b/xlators/features/changelog/src/changelog-rt.c
new file mode 100644
index 00000000000..841545ae359
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-rt.c
@@ -0,0 +1,66 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/logging.h>
+
+#include "changelog-rt.h"
+#include "changelog-mem-types.h"
+
+int
+changelog_rt_init(xlator_t *this, changelog_dispatcher_t *cd)
+{
+    changelog_rt_t *crt = NULL;
+
+    crt = GF_CALLOC(1, sizeof(*crt), gf_changelog_mt_rt_t);
+    if (!crt)
+        return -1;
+
+    LOCK_INIT(&crt->lock);
+
+    cd->cd_data = crt;
+    cd->dispatchfn = &changelog_rt_enqueue;
+
+    return 0;
+}
+
+int
+changelog_rt_fini(xlator_t *this, changelog_dispatcher_t *cd)
+{
+    changelog_rt_t *crt = NULL;
+
+    crt = cd->cd_data;
+
+    LOCK_DESTROY(&crt->lock);
+    GF_FREE(crt);
+
+    return 0;
+}
+
+int
+changelog_rt_enqueue(xlator_t *this, changelog_priv_t *priv, void *cbatch,
+                     changelog_log_data_t *cld_0, changelog_log_data_t *cld_1)
+{
+    int ret = 0;
+    changelog_rt_t *crt = NULL;
+
+    crt = (changelog_rt_t *)cbatch;
+
+    LOCK(&crt->lock);
+    {
+        ret = changelog_handle_change(this, priv, cld_0);
+        if (!ret && cld_1)
+            ret = changelog_handle_change(this, priv, cld_1);
+    }
+    UNLOCK(&crt->lock);
+
+    return ret;
+}
diff --git a/xlators/features/changelog/src/changelog-rt.h b/xlators/features/changelog/src/changelog-rt.h
new file mode 100644
index 00000000000..28b9827d85b
--- /dev/null
+++ b/xlators/features/changelog/src/changelog-rt.h
@@ -0,0 +1,33 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef _CHANGELOG_RT_H
+#define _CHANGELOG_RT_H
+
+#include <glusterfs/locking.h>
+#include <glusterfs/timer.h>
+#include "pthread.h"
+
+#include "changelog-helpers.h"
+
+/* unused as of now - may be you would need it later */
+typedef struct changelog_rt {
+    gf_lock_t lock;
+} changelog_rt_t;
+
+int
+changelog_rt_init(xlator_t *this, changelog_dispatcher_t *cd);
+int
+changelog_rt_fini(xlator_t *this, changelog_dispatcher_t *cd);
+int
+changelog_rt_enqueue(xlator_t *this, changelog_priv_t *priv, void *cbatch,
+                     changelog_log_data_t *cld_0, changelog_log_data_t *cld_1);
+
+#endif /* _CHANGELOG_RT_H */
diff --git a/xlators/features/changelog/src/changelog.c b/xlators/features/changelog/src/changelog.c
new file mode 100644
index 00000000000..6a6e5af859e
--- /dev/null
+++ b/xlators/features/changelog/src/changelog.c
@@ -0,0 +1,2989 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/syscall.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/iobuf.h>
+
+#include "changelog-rt.h"
+
+#include "changelog-encoders.h"
+#include "changelog-mem-types.h"
+#include "changelog-messages.h"
+
+#include <pthread.h>
+#include <signal.h>
+
+#include "changelog-rpc.h"
+#include "errno.h"
+
+static struct changelog_bootstrap cb_bootstrap[] = {
+    {
+        .mode = CHANGELOG_MODE_RT,
+        .ctor = changelog_rt_init,
+        .dtor = changelog_rt_fini,
+    },
+};
+
+static int
+changelog_init_rpc(xlator_t *this, changelog_priv_t *priv);
+
+static int
+changelog_init(xlator_t *this, changelog_priv_t *priv);
+
+/* Entry operations - TYPE III */
+
+/**
+ * entry operations do not undergo inode version checking.
+ */
+
+/* {{{ */
+
+/* rmdir */
+
+int32_t
+changelog_rmdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    CHANGELOG_COND_GOTO(priv, ((op_ret < 0) || !local), unwind);
+
+    changelog_update(this, priv, local, CHANGELOG_TYPE_ENTRY);
+
+unwind:
+    changelog_dec_fop_cnt(this, priv, local);
+    CHANGELOG_STACK_UNWIND(rmdir, frame, op_ret, op_errno, preparent,
+                           postparent, xdata);
+    return 0;
+}
+
+int32_t
+changelog_rmdir_resume(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                       int xflags, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+
+    priv = this->private;
+
+    gf_msg_debug(this->name, 0, "Dequeue rmdir");
+    changelog_color_fop_and_inc_cnt(this, priv, frame->local);
+    STACK_WIND(frame, changelog_rmdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rmdir, loc, xflags, xdata);
+    return 0;
+}
+
+int32_t
+changelog_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflags,
+                dict_t *xdata)
+{
+    size_t xtra_len = 0;
+    changelog_priv_t *priv = NULL;
+    changelog_opt_t *co = NULL;
+    call_stub_t *stub = NULL;
+    struct list_head queue = {
+        0,
+    };
+    gf_boolean_t barrier_enabled = _gf_false;
+
+    INIT_LIST_HEAD(&queue);
+
+    priv = this->private;
+    CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+
+    CHANGELOG_INIT_NOCHECK(this, frame->local, NULL, loc->inode->gfid, 2);
+
+    co = changelog_get_usable_buffer(frame->local);
+    if (!co)
+        goto wind;
+
+    CHANGLOG_FILL_FOP_NUMBER(co, frame->root->op, fop_fn, xtra_len);
+
+    co++;
+    if (priv->capture_del_path) {
+        CHANGELOG_FILL_ENTRY_DIR_PATH(co, loc->pargfid, loc->name, del_entry_fn,
+                                      del_entry_free_fn, xtra_len, wind,
+                                      _gf_true);
+    } else {
+        CHANGELOG_FILL_ENTRY_DIR_PATH(co, loc->pargfid, loc->name, del_entry_fn,
+                                      del_entry_free_fn, xtra_len, wind,
+                                      _gf_false);
+    }
+
+    changelog_set_usable_record_and_length(frame->local, xtra_len, 2);
+
+    /* changelog barrier */
+    /* Color assignment and increment of fop_cnt for rmdir/unlink/rename
+     * should be made with in priv lock if changelog barrier is not enabled.
+     * Because if counter is not incremented yet, draining wakes up and
+     * publishes the changelog but later these fops might hit the disk and
+     * present in snapped volume but where as the intention is these fops
+     * should not be present in snapped volume.
+     */
+    LOCK(&priv->lock);
+    {
+        if ((barrier_enabled = priv->barrier_enabled)) {
+            stub = fop_rmdir_stub(frame, changelog_rmdir_resume, loc, xflags,
+                                  xdata);
+            if (!stub)
+                __chlog_barrier_disable(this, &queue);
+            else
+                __chlog_barrier_enqueue(this, stub);
+        } else {
+            ((changelog_local_t *)frame->local)->color = priv->current_color;
+            changelog_inc_fop_cnt(this, priv, frame->local);
+        }
+    }
+    UNLOCK(&priv->lock);
+
+    if (barrier_enabled && stub) {
+        gf_msg_debug(this->name, 0, "Enqueue rmdir");
+        goto out;
+    }
+    if (barrier_enabled && !stub) {
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM,
+                CHANGELOG_MSG_BARRIER_FOP_FAILED, "fop=rmdir", NULL);
+        chlog_barrier_dequeue_all(this, &queue);
+    }
+
+    /* changelog barrier */
+
+wind:
+    STACK_WIND(frame, changelog_rmdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rmdir, loc, xflags, xdata);
+out:
+    return 0;
+}
+
+/* unlink */
+
+int32_t
+changelog_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                     struct iatt *postparent, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    CHANGELOG_COND_GOTO(priv, ((op_ret < 0) || !local), unwind);
+
+    changelog_update(this, priv, local, CHANGELOG_TYPE_ENTRY);
+
+unwind:
+    changelog_dec_fop_cnt(this, priv, local);
+    CHANGELOG_STACK_UNWIND(unlink, frame, op_ret, op_errno, preparent,
+                           postparent, xdata);
+    return 0;
+}
+
+int32_t
+changelog_unlink_resume(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                        int xflags, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+
+    priv = this->private;
+
+    gf_msg_debug(this->name, 0, "Dequeue unlink");
+    changelog_color_fop_and_inc_cnt(this, priv, frame->local);
+    STACK_WIND(frame, changelog_unlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->unlink, loc, xflags, xdata);
+    return 0;
+}
+
+int32_t
+changelog_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflags,
+                 dict_t *xdata)
+{
+    size_t xtra_len = 0;
+    changelog_priv_t *priv = NULL;
+    changelog_opt_t *co = NULL;
+    call_stub_t *stub = NULL;
+    struct list_head queue = {
+        0,
+    };
+    gf_boolean_t barrier_enabled = _gf_false;
+    dht_changelog_rename_info_t *info = NULL;
+    int ret = 0;
+    char *old_name = NULL;
+    char *new_name = NULL;
+    char *nname = NULL;
+
+    INIT_LIST_HEAD(&queue);
+    priv = this->private;
+
+    CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+
+    ret = dict_get_bin(xdata, DHT_CHANGELOG_RENAME_OP_KEY, (void **)&info);
+    if (!ret) { /* special case: unlink considered as rename */
+        /* 3 == fop + oldloc + newloc */
+        old_name = alloca(info->oldname_len);
+        new_name = alloca(info->newname_len);
+        CHANGELOG_INIT_NOCHECK(this, frame->local, NULL, loc->inode->gfid, 3);
+
+        co = changelog_get_usable_buffer(frame->local);
+        if (!co)
+            goto wind;
+
+        CHANGLOG_FILL_FOP_NUMBER(co, GF_FOP_RENAME, fop_fn, xtra_len);
+
+        co++;
+        strncpy(old_name, info->buffer, info->oldname_len);
+        CHANGELOG_FILL_ENTRY(co, info->old_pargfid, old_name, entry_fn,
+                             entry_free_fn, xtra_len, wind);
+
+        co++;
+        /* new name resides just after old name */
+        nname = info->buffer + info->oldname_len;
+        strncpy(new_name, nname, info->newname_len);
+        CHANGELOG_FILL_ENTRY(co, info->new_pargfid, new_name, entry_fn,
+                             entry_free_fn, xtra_len, wind);
+
+        changelog_set_usable_record_and_length(frame->local, xtra_len, 3);
+    } else { /* default unlink */
+        CHANGELOG_IF_INTERNAL_FOP_THEN_GOTO(frame, xdata, wind);
+        CHANGELOG_INIT_NOCHECK(this, frame->local, NULL, loc->inode->gfid, 2);
+
+        co = changelog_get_usable_buffer(frame->local);
+        if (!co)
+            goto wind;
+
+        CHANGLOG_FILL_FOP_NUMBER(co, frame->root->op, fop_fn, xtra_len);
+
+        co++;
+        if (priv->capture_del_path) {
+            CHANGELOG_FILL_ENTRY_DIR_PATH(co, loc->pargfid, loc->name,
+                                          del_entry_fn, del_entry_free_fn,
+                                          xtra_len, wind, _gf_true);
+        } else {
+            CHANGELOG_FILL_ENTRY_DIR_PATH(co, loc->pargfid, loc->name,
+                                          del_entry_fn, del_entry_free_fn,
+                                          xtra_len, wind, _gf_false);
+        }
+
+        changelog_set_usable_record_and_length(frame->local, xtra_len, 2);
+    }
+
+    /* changelog barrier */
+    LOCK(&priv->lock);
+    {
+        if ((barrier_enabled = priv->barrier_enabled)) {
+            stub = fop_unlink_stub(frame, changelog_unlink_resume, loc, xflags,
+                                   xdata);
+            if (!stub)
+                __chlog_barrier_disable(this, &queue);
+            else
+                __chlog_barrier_enqueue(this, stub);
+        } else {
+            ((changelog_local_t *)frame->local)->color = priv->current_color;
+            changelog_inc_fop_cnt(this, priv, frame->local);
+        }
+    }
+    UNLOCK(&priv->lock);
+
+    if (barrier_enabled && stub) {
+        gf_msg_debug(this->name, 0, "Enqueue unlink");
+        goto out;
+    }
+    if (barrier_enabled && !stub) {
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM,
+                CHANGELOG_MSG_BARRIER_FOP_FAILED, "fop=unlink", NULL);
+        chlog_barrier_dequeue_all(this, &queue);
+    }
+
+    /* changelog barrier */
+
+wind:
+    STACK_WIND(frame, changelog_unlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->unlink, loc, xflags, xdata);
+out:
+    return 0;
+}
+
+/* rename */
+
+int32_t
+changelog_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                     struct iatt *preoldparent, struct iatt *postoldparent,
+                     struct iatt *prenewparent, struct iatt *postnewparent,
+                     dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+    CHANGELOG_COND_GOTO(priv, ((op_ret < 0) || !local), unwind);
+    changelog_update(this, priv, local, CHANGELOG_TYPE_ENTRY);
+unwind:
+    changelog_dec_fop_cnt(this, priv, local);
+    CHANGELOG_STACK_UNWIND(rename, frame, op_ret, op_errno, buf, preoldparent,
+                           postoldparent, prenewparent, postnewparent, xdata);
+    return 0;
+}
+
+int32_t
+changelog_rename_resume(call_frame_t *frame, xlator_t *this, loc_t *oldloc,
+                        loc_t *newloc, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+
+    priv = this->private;
+
+    gf_msg_debug(this->name, 0, "Dequeue rename");
+    changelog_color_fop_and_inc_cnt(this, priv, frame->local);
+    STACK_WIND(frame, changelog_rename_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
+    return 0;
+}
+
+int32_t
+changelog_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc,
+                 loc_t *newloc, dict_t *xdata)
+{
+    size_t xtra_len = 0;
+    changelog_priv_t *priv = NULL;
+    changelog_opt_t *co = NULL;
+    call_stub_t *stub = NULL;
+    struct list_head queue = {
+        0,
+    };
+    gf_boolean_t barrier_enabled = _gf_false;
+    dht_changelog_rename_info_t *info = NULL;
+    int ret = 0;
+
+    INIT_LIST_HEAD(&queue);
+
+    priv = this->private;
+    CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+
+    ret = dict_get_bin(xdata, DHT_CHANGELOG_RENAME_OP_KEY, (void **)&info);
+    if (ret && oldloc->inode->ia_type != IA_IFDIR) {
+        /* xdata "NOT" set for a non-directory,
+         * Special rename => avoid logging */
+        goto wind;
+    }
+
+    /* 3 == fop + oldloc + newloc */
+    CHANGELOG_INIT_NOCHECK(this, frame->local, NULL, oldloc->inode->gfid, 3);
+
+    co = changelog_get_usable_buffer(frame->local);
+    if (!co)
+        goto wind;
+
+    CHANGLOG_FILL_FOP_NUMBER(co, frame->root->op, fop_fn, xtra_len);
+
+    co++;
+    CHANGELOG_FILL_ENTRY(co, oldloc->pargfid, oldloc->name, entry_fn,
+                         entry_free_fn, xtra_len, wind);
+
+    co++;
+    CHANGELOG_FILL_ENTRY(co, newloc->pargfid, newloc->name, entry_fn,
+                         entry_free_fn, xtra_len, wind);
+
+    changelog_set_usable_record_and_length(frame->local, xtra_len, 3);
+    /* changelog barrier */
+    LOCK(&priv->lock);
+    {
+        if ((barrier_enabled = priv->barrier_enabled)) {
+            stub = fop_rename_stub(frame, changelog_rename_resume, oldloc,
+                                   newloc, xdata);
+            if (!stub)
+                __chlog_barrier_disable(this, &queue);
+            else
+                __chlog_barrier_enqueue(this, stub);
+        } else {
+            ((changelog_local_t *)frame->local)->color = priv->current_color;
+            changelog_inc_fop_cnt(this, priv, frame->local);
+        }
+    }
+    UNLOCK(&priv->lock);
+
+    if (barrier_enabled && stub) {
+        gf_msg_debug(this->name, 0, "Enqueue rename");
+        goto out;
+    }
+    if (barrier_enabled && !stub) {
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM,
+                CHANGELOG_MSG_BARRIER_FOP_FAILED, "fop=rename", NULL);
+        chlog_barrier_dequeue_all(this, &queue);
+    }
+    /* changelog barrier */
+
+wind:
+    STACK_WIND(frame, changelog_rename_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
+out:
+    return 0;
+}
+
+/* link */
+
+int32_t
+changelog_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, inode_t *inode,
+                   struct iatt *buf, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    CHANGELOG_COND_GOTO(priv, ((op_ret < 0) || !local), unwind);
+
+    changelog_update(this, priv, local, CHANGELOG_TYPE_ENTRY);
+
+unwind:
+    changelog_dec_fop_cnt(this, priv, local);
+    CHANGELOG_STACK_UNWIND(link, frame, op_ret, op_errno, inode, buf, preparent,
+                           postparent, xdata);
+    return 0;
+}
+
+int32_t
+changelog_link_resume(call_frame_t *frame, xlator_t *this, loc_t *oldloc,
+                      loc_t *newloc, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("changelog", this, out);
+    GF_VALIDATE_OR_GOTO("changelog", this->fops, out);
+    GF_VALIDATE_OR_GOTO("changelog", frame, out);
+
+    priv = this->private;
+
+    gf_msg_debug(this->name, 0, "Dequeuing link");
+    changelog_color_fop_and_inc_cnt(this, priv, frame->local);
+    STACK_WIND(frame, changelog_link_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->link, oldloc, newloc, xdata);
+    return 0;
+out:
+    return -1;
+}
+int32_t
+changelog_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc,
+               loc_t *newloc, dict_t *xdata)
+{
+    size_t xtra_len = 0;
+    changelog_priv_t *priv = NULL;
+    changelog_opt_t *co = NULL;
+    call_stub_t *stub = NULL;
+    struct list_head queue = {
+        0,
+    };
+    gf_boolean_t barrier_enabled = _gf_false;
+
+    priv = this->private;
+
+    CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+    CHANGELOG_IF_INTERNAL_FOP_THEN_GOTO(frame, xdata, wind);
+
+    CHANGELOG_INIT_NOCHECK(this, frame->local, NULL, oldloc->gfid, 2);
+
+    co = changelog_get_usable_buffer(frame->local);
+    if (!co)
+        goto wind;
+
+    CHANGLOG_FILL_FOP_NUMBER(co, frame->root->op, fop_fn, xtra_len);
+
+    co++;
+    CHANGELOG_FILL_ENTRY(co, newloc->pargfid, newloc->name, entry_fn,
+                         entry_free_fn, xtra_len, wind);
+
+    changelog_set_usable_record_and_length(frame->local, xtra_len, 2);
+
+    LOCK(&priv->lock);
+    {
+        if ((barrier_enabled = priv->barrier_enabled)) {
+            stub = fop_link_stub(frame, changelog_link_resume, oldloc, newloc,
+                                 xdata);
+            if (!stub)
+                __chlog_barrier_disable(this, &queue);
+            else
+                __chlog_barrier_enqueue(this, stub);
+        } else {
+            ((changelog_local_t *)frame->local)->color = priv->current_color;
+            changelog_inc_fop_cnt(this, priv, frame->local);
+        }
+    }
+    UNLOCK(&priv->lock);
+
+    if (barrier_enabled && stub) {
+        gf_msg_debug(this->name, 0, "Enqueued link");
+        goto out;
+    }
+
+    if (barrier_enabled && !stub) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_MSG_BARRIER_FOP_FAILED,
+                "fop=link", NULL);
+        chlog_barrier_dequeue_all(this, &queue);
+    }
+wind:
+    STACK_WIND(frame, changelog_link_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->link, oldloc, newloc, xdata);
+out:
+    return 0;
+}
+
+/* mkdir */
+
+int32_t
+changelog_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, inode_t *inode,
+                    struct iatt *buf, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    CHANGELOG_COND_GOTO(priv, ((op_ret < 0) || !local), unwind);
+
+    changelog_update(this, priv, local, CHANGELOG_TYPE_ENTRY);
+
+unwind:
+    changelog_dec_fop_cnt(this, priv, local);
+    CHANGELOG_STACK_UNWIND(mkdir, frame, op_ret, op_errno, inode, buf,
+                           preparent, postparent, xdata);
+    return 0;
+}
+
+int32_t
+changelog_mkdir_resume(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                       mode_t mode, mode_t umask, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("changelog", this, out);
+    GF_VALIDATE_OR_GOTO("changelog", this->fops, out);
+    GF_VALIDATE_OR_GOTO("changelog", frame, out);
+
+    priv = this->private;
+
+    gf_msg_debug(this->name, 0, "Dequeuing mkdir");
+    changelog_color_fop_and_inc_cnt(this, priv, frame->local);
+    STACK_WIND(frame, changelog_mkdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mkdir, loc, mode, umask, xdata);
+    return 0;
+out:
+    return -1;
+}
+
+int32_t
+changelog_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+                mode_t umask, dict_t *xdata)
+{
+    int ret = -1;
+    uuid_t gfid = {
+        0,
+    };
+    size_t xtra_len = 0;
+    changelog_priv_t *priv = NULL;
+    changelog_opt_t *co = NULL;
+    call_stub_t *stub = NULL;
+    struct list_head queue = {
+        0,
+    };
+    gf_boolean_t barrier_enabled = _gf_false;
+
+    priv = this->private;
+    CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+
+    ret = dict_get_gfuuid(xdata, "gfid-req", &gfid);
+    if (ret) {
+        gf_msg_debug(this->name, 0, "failed to get gfid from dict");
+        goto wind;
+    }
+
+    CHANGELOG_INIT_NOCHECK(this, frame->local, NULL, gfid, 5);
+
+    co = changelog_get_usable_buffer(frame->local);
+    if (!co)
+        goto wind;
+
+    CHANGLOG_FILL_FOP_NUMBER(co, frame->root->op, fop_fn, xtra_len);
+    co++;
+
+    CHANGELOG_FILL_UINT32(co, S_IFDIR | mode, number_fn, xtra_len);
+    co++;
+
+    CHANGELOG_FILL_UINT32(co, frame->root->uid, number_fn, xtra_len);
+    co++;
+
+    CHANGELOG_FILL_UINT32(co, frame->root->gid, number_fn, xtra_len);
+    co++;
+
+    CHANGELOG_FILL_ENTRY(co, loc->pargfid, loc->name, entry_fn, entry_free_fn,
+                         xtra_len, wind);
+
+    changelog_set_usable_record_and_length(frame->local, xtra_len, 5);
+
+    LOCK(&priv->lock);
+    {
+        if ((barrier_enabled = priv->barrier_enabled)) {
+            stub = fop_mkdir_stub(frame, changelog_mkdir_resume, loc, mode,
+                                  umask, xdata);
+            if (!stub)
+                __chlog_barrier_disable(this, &queue);
+            else
+                __chlog_barrier_enqueue(this, stub);
+        } else {
+            ((changelog_local_t *)frame->local)->color = priv->current_color;
+            changelog_inc_fop_cnt(this, priv, frame->local);
+        }
+    }
+    UNLOCK(&priv->lock);
+
+    if (barrier_enabled && stub) {
+        gf_msg_debug(this->name, 0, "Enqueued mkdir");
+        goto out;
+    }
+
+    if (barrier_enabled && !stub) {
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM,
+                CHANGELOG_MSG_BARRIER_FOP_FAILED, "fop=mkdir", NULL);
+        chlog_barrier_dequeue_all(this, &queue);
+    }
+
+wind:
+    STACK_WIND(frame, changelog_mkdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mkdir, loc, mode, umask, xdata);
+out:
+    return 0;
+}
+
+/* symlink */
+
+int32_t
+changelog_symlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, inode_t *inode,
+                      struct iatt *buf, struct iatt *preparent,
+                      struct iatt *postparent, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    CHANGELOG_COND_GOTO(priv, ((op_ret < 0) || !local), unwind);
+
+    changelog_update(this, priv, local, CHANGELOG_TYPE_ENTRY);
+
+unwind:
+    changelog_dec_fop_cnt(this, priv, local);
+    CHANGELOG_STACK_UNWIND(symlink, frame, op_ret, op_errno, inode, buf,
+                           preparent, postparent, xdata);
+    return 0;
+}
+
+int32_t
+changelog_symlink_resume(call_frame_t *frame, xlator_t *this,
+                         const char *linkname, loc_t *loc, mode_t umask,
+                         dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("changelog", this, out);
+    GF_VALIDATE_OR_GOTO("changelog", this->fops, out);
+    GF_VALIDATE_OR_GOTO("changelog", frame, out);
+
+    priv = this->private;
+
+    gf_msg_debug(this->name, 0, "Dequeuing symlink");
+    changelog_color_fop_and_inc_cnt(this, priv, frame->local);
+    STACK_WIND(frame, changelog_symlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->symlink, linkname, loc, umask, xdata);
+    return 0;
+out:
+    return -1;
+}
+
+int32_t
+changelog_symlink(call_frame_t *frame, xlator_t *this, const char *linkname,
+                  loc_t *loc, mode_t umask, dict_t *xdata)
+{
+    int ret = -1;
+    size_t xtra_len = 0;
+    uuid_t gfid = {
+        0,
+    };
+    changelog_priv_t *priv = NULL;
+    changelog_opt_t *co = NULL;
+    call_stub_t *stub = NULL;
+    struct list_head queue = {
+        0,
+    };
+    gf_boolean_t barrier_enabled = _gf_false;
+
+    priv = this->private;
+    CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+
+    ret = dict_get_gfuuid(xdata, "gfid-req", &gfid);
+    if (ret) {
+        gf_msg_debug(this->name, 0, "failed to get gfid from dict");
+        goto wind;
+    }
+
+    CHANGELOG_INIT_NOCHECK(this, frame->local, NULL, gfid, 2);
+
+    co = changelog_get_usable_buffer(frame->local);
+    if (!co)
+        goto wind;
+
+    CHANGLOG_FILL_FOP_NUMBER(co, frame->root->op, fop_fn, xtra_len);
+    co++;
+
+    CHANGELOG_FILL_ENTRY(co, loc->pargfid, loc->name, entry_fn, entry_free_fn,
+                         xtra_len, wind);
+
+    changelog_set_usable_record_and_length(frame->local, xtra_len, 2);
+
+    LOCK(&priv->lock);
+    {
+        if ((barrier_enabled = priv->barrier_enabled)) {
+            stub = fop_symlink_stub(frame, changelog_symlink_resume, linkname,
+                                    loc, umask, xdata);
+            if (!stub)
+                __chlog_barrier_disable(this, &queue);
+            else
+                __chlog_barrier_enqueue(this, stub);
+        } else {
+            ((changelog_local_t *)frame->local)->color = priv->current_color;
+            changelog_inc_fop_cnt(this, priv, frame->local);
+        }
+    }
+    UNLOCK(&priv->lock);
+
+    if (barrier_enabled && stub) {
+        gf_msg_debug(this->name, 0, "Enqueued symlink");
+        goto out;
+    }
+
+    if (barrier_enabled && !stub) {
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM,
+                CHANGELOG_MSG_BARRIER_FOP_FAILED, "fop=symlink", NULL);
+        chlog_barrier_dequeue_all(this, &queue);
+    }
+
+wind:
+    STACK_WIND(frame, changelog_symlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->symlink, linkname, loc, umask, xdata);
+out:
+    return 0;
+}
+
+/* mknod */
+
+int32_t
+changelog_mknod_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, inode_t *inode,
+                    struct iatt *buf, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    CHANGELOG_COND_GOTO(priv, ((op_ret < 0) || !local), unwind);
+
+    changelog_update(this, priv, local, CHANGELOG_TYPE_ENTRY);
+
+unwind:
+    changelog_dec_fop_cnt(this, priv, local);
+    CHANGELOG_STACK_UNWIND(mknod, frame, op_ret, op_errno, inode, buf,
+                           preparent, postparent, xdata);
+    return 0;
+}
+
+int32_t
+changelog_mknod_resume(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                       mode_t mode, dev_t rdev, mode_t umask, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("changelog", this, out);
+    GF_VALIDATE_OR_GOTO("changelog", this->fops, out);
+    GF_VALIDATE_OR_GOTO("changelog", frame, out);
+
+    priv = this->private;
+
+    gf_msg_debug(this->name, 0, "Dequeuing mknod");
+    changelog_color_fop_and_inc_cnt(this, priv, frame->local);
+    STACK_WIND(frame, changelog_mknod_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mknod, loc, mode, rdev, umask, xdata);
+    return 0;
+out:
+    return -1;
+}
+
+int32_t
+changelog_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+                dev_t dev, mode_t umask, dict_t *xdata)
+{
+    int ret = -1;
+    uuid_t gfid = {
+        0,
+    };
+    size_t xtra_len = 0;
+    changelog_priv_t *priv = NULL;
+    changelog_opt_t *co = NULL;
+    call_stub_t *stub = NULL;
+    struct list_head queue = {
+        0,
+    };
+    gf_boolean_t barrier_enabled = _gf_false;
+
+    priv = this->private;
+
+    /* Check whether changelog active */
+    if (!(priv->active))
+        goto wind;
+
+    /* Check whether rebalance activity */
+    if (frame->root->pid == GF_CLIENT_PID_DEFRAG)
+        goto wind;
+
+    /* If tier-dht linkto is SET, ignore about verifiying :
+     * 1. Whether internal fop AND
+     * 2. Whether tier rebalance process activity (this will help in
+     * recording mknod if tier rebalance process calls this mknod) */
+    if (!(dict_get(xdata, "trusted.tier.tier-dht.linkto"))) {
+        CHANGELOG_IF_INTERNAL_FOP_THEN_GOTO(frame, xdata, wind);
+        if (frame->root->pid == GF_CLIENT_PID_TIER_DEFRAG)
+            goto wind;
+    }
+
+    ret = dict_get_gfuuid(xdata, "gfid-req", &gfid);
+    if (ret) {
+        gf_msg_debug(this->name, 0, "failed to get gfid from dict");
+        goto wind;
+    }
+
+    CHANGELOG_INIT_NOCHECK(this, frame->local, NULL, gfid, 5);
+
+    co = changelog_get_usable_buffer(frame->local);
+    if (!co)
+        goto wind;
+
+    CHANGLOG_FILL_FOP_NUMBER(co, frame->root->op, fop_fn, xtra_len);
+    co++;
+
+    CHANGELOG_FILL_UINT32(co, mode, number_fn, xtra_len);
+    co++;
+
+    CHANGELOG_FILL_UINT32(co, frame->root->uid, number_fn, xtra_len);
+    co++;
+
+    CHANGELOG_FILL_UINT32(co, frame->root->gid, number_fn, xtra_len);
+    co++;
+
+    CHANGELOG_FILL_ENTRY(co, loc->pargfid, loc->name, entry_fn, entry_free_fn,
+                         xtra_len, wind);
+
+    changelog_set_usable_record_and_length(frame->local, xtra_len, 5);
+
+    LOCK(&priv->lock);
+    {
+        if ((barrier_enabled = priv->barrier_enabled)) {
+            stub = fop_mknod_stub(frame, changelog_mknod_resume, loc, mode, dev,
+                                  umask, xdata);
+            if (!stub)
+                __chlog_barrier_disable(this, &queue);
+            else
+                __chlog_barrier_enqueue(this, stub);
+        } else {
+            ((changelog_local_t *)frame->local)->color = priv->current_color;
+            changelog_inc_fop_cnt(this, priv, frame->local);
+        }
+    }
+    UNLOCK(&priv->lock);
+
+    if (barrier_enabled && stub) {
+        gf_msg_debug(this->name, 0, "Enqueued mknod");
+        goto out;
+    }
+
+    if (barrier_enabled && !stub) {
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM,
+                CHANGELOG_MSG_BARRIER_FOP_FAILED, "fop=mknod", NULL);
+        chlog_barrier_dequeue_all(this, &queue);
+    }
+
+wind:
+    STACK_WIND(frame, changelog_mknod_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mknod, loc, mode, dev, umask, xdata);
+out:
+    return 0;
+}
+
+/* create */
+
+int32_t
+changelog_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, fd_t *fd, inode_t *inode,
+                     struct iatt *buf, struct iatt *preparent,
+                     struct iatt *postparent, dict_t *xdata)
+{
+    int32_t ret = 0;
+    changelog_priv_t *priv = NULL;
+    changelog_local_t *local = NULL;
+    changelog_event_t ev = {
+        0,
+    };
+
+    priv = this->private;
+    local = frame->local;
+
+    CHANGELOG_COND_GOTO(priv, ((op_ret < 0) || !local), unwind);
+
+    /* fill the event structure.. similar to open() */
+    ev.ev_type = CHANGELOG_OP_TYPE_CREATE;
+    gf_uuid_copy(ev.u.create.gfid, buf->ia_gfid);
+    ev.u.create.flags = fd->flags;
+    changelog_dispatch_event(this, priv, &ev);
+
+    if (changelog_ev_selected(this, &priv->ev_selection,
+                              CHANGELOG_OP_TYPE_RELEASE)) {
+        ret = fd_ctx_set(fd, this, (uint64_t)(long)0x1);
+        if (ret)
+            gf_smsg(this->name, GF_LOG_WARNING, 0, CHANGELOG_MSG_SET_FD_CONTEXT,
+                    NULL);
+    }
+
+    changelog_update(this, priv, local, CHANGELOG_TYPE_ENTRY);
+
+unwind:
+    changelog_dec_fop_cnt(this, priv, local);
+    CHANGELOG_STACK_UNWIND(create, frame, op_ret, op_errno, fd, inode, buf,
+                           preparent, postparent, xdata);
+    return 0;
+}
+
+int32_t
+changelog_create_resume(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                        int32_t flags, mode_t mode, mode_t umask, fd_t *fd,
+                        dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("changelog", this, out);
+    GF_VALIDATE_OR_GOTO("changelog", this->fops, out);
+    GF_VALIDATE_OR_GOTO("changelog", frame, out);
+
+    priv = this->private;
+
+    gf_msg_debug(this->name, 0, "Dequeuing create");
+    changelog_color_fop_and_inc_cnt(this, priv, frame->local);
+    STACK_WIND(frame, changelog_create_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->create, loc, flags, mode, umask, fd,
+               xdata);
+    return 0;
+
+out:
+    return -1;
+}
+
+int32_t
+changelog_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+                 mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+    int ret = -1;
+    uuid_t gfid = {
+        0,
+    };
+    changelog_opt_t *co = NULL;
+    changelog_priv_t *priv = NULL;
+    size_t xtra_len = 0;
+    call_stub_t *stub = NULL;
+    struct list_head queue = {
+        0,
+    };
+    gf_boolean_t barrier_enabled = _gf_false;
+
+    priv = this->private;
+    CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+
+    ret = dict_get_gfuuid(xdata, "gfid-req", &gfid);
+    if (ret) {
+        gf_msg_debug(this->name, 0, "failed to get gfid from dict");
+        goto wind;
+    }
+
+    /* init with two extra records */
+    CHANGELOG_INIT_NOCHECK(this, frame->local, NULL, gfid, 5);
+    if (!frame->local)
+        goto wind;
+
+    co = changelog_get_usable_buffer(frame->local);
+    if (!co)
+        goto wind;
+
+    CHANGLOG_FILL_FOP_NUMBER(co, frame->root->op, fop_fn, xtra_len);
+    co++;
+
+    CHANGELOG_FILL_UINT32(co, mode, number_fn, xtra_len);
+    co++;
+
+    CHANGELOG_FILL_UINT32(co, frame->root->uid, number_fn, xtra_len);
+    co++;
+
+    CHANGELOG_FILL_UINT32(co, frame->root->gid, number_fn, xtra_len);
+    co++;
+
+    CHANGELOG_FILL_ENTRY(co, loc->pargfid, loc->name, entry_fn, entry_free_fn,
+                         xtra_len, wind);
+
+    changelog_set_usable_record_and_length(frame->local, xtra_len, 5);
+
+    LOCK(&priv->lock);
+    {
+        if ((barrier_enabled = priv->barrier_enabled)) {
+            stub = fop_create_stub(frame, changelog_create_resume, loc, flags,
+                                   mode, umask, fd, xdata);
+            if (!stub)
+                __chlog_barrier_disable(this, &queue);
+            else
+                __chlog_barrier_enqueue(this, stub);
+        } else {
+            ((changelog_local_t *)frame->local)->color = priv->current_color;
+            changelog_inc_fop_cnt(this, priv, frame->local);
+        }
+    }
+    UNLOCK(&priv->lock);
+
+    if (barrier_enabled && stub) {
+        gf_msg_debug(this->name, 0, "Enqueued create");
+        goto out;
+    }
+
+    if (barrier_enabled && !stub) {
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM,
+                CHANGELOG_MSG_BARRIER_FOP_FAILED, "fop=create", NULL);
+        chlog_barrier_dequeue_all(this, &queue);
+    }
+
+wind:
+    STACK_WIND(frame, changelog_create_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->create, loc, flags, mode, umask, fd,
+               xdata);
+out:
+    return 0;
+}
+
+/* }}} */
+
+/* Metadata modification fops - TYPE II */
+
+/* {{{ */
+
+/* {f}setattr */
+
+int32_t
+changelog_fsetattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno,
+                       struct iatt *preop_stbuf, struct iatt *postop_stbuf,
+                       dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    CHANGELOG_COND_GOTO(priv, ((op_ret < 0) || !local), unwind);
+
+    changelog_update(this, priv, local, CHANGELOG_TYPE_METADATA);
+
+unwind:
+    changelog_dec_fop_cnt(this, priv, local);
+    CHANGELOG_STACK_UNWIND(fsetattr, frame, op_ret, op_errno, preop_stbuf,
+                           postop_stbuf, xdata);
+
+    return 0;
+}
+
+int32_t
+changelog_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                   struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_opt_t *co = NULL;
+    size_t xtra_len = 0;
+
+    priv = this->private;
+    CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+
+    CHANGELOG_OP_BOUNDARY_CHECK(frame, wind);
+
+    CHANGELOG_INIT(this, frame->local, fd->inode, fd->inode->gfid, 1);
+    if (!frame->local)
+        goto wind;
+
+    co = changelog_get_usable_buffer(frame->local);
+    if (!co)
+        goto wind;
+
+    CHANGLOG_FILL_FOP_NUMBER(co, frame->root->op, fop_fn, xtra_len);
+
+    changelog_set_usable_record_and_length(frame->local, xtra_len, 1);
+
+wind:
+    changelog_color_fop_and_inc_cnt(this, priv, frame->local);
+    STACK_WIND(frame, changelog_fsetattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetattr, fd, stbuf, valid, xdata);
+    return 0;
+}
+
+int32_t
+changelog_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno,
+                      struct iatt *preop_stbuf, struct iatt *postop_stbuf,
+                      dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    CHANGELOG_COND_GOTO(priv, ((op_ret < 0) || !local), unwind);
+
+    changelog_update(this, priv, local, CHANGELOG_TYPE_METADATA);
+
+unwind:
+    changelog_dec_fop_cnt(this, priv, local);
+    CHANGELOG_STACK_UNWIND(setattr, frame, op_ret, op_errno, preop_stbuf,
+                           postop_stbuf, xdata);
+
+    return 0;
+}
+
+int32_t
+changelog_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                  struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_opt_t *co = NULL;
+    size_t xtra_len = 0;
+    uuid_t shard_root_gfid = {
+        0,
+    };
+
+    priv = this->private;
+    CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+
+    CHANGELOG_IF_INTERNAL_FOP_THEN_GOTO(frame, xdata, wind);
+
+    /* Do not record META on .shard */
+    gf_uuid_parse(SHARD_ROOT_GFID, shard_root_gfid);
+    if (gf_uuid_compare(loc->gfid, shard_root_gfid) == 0) {
+        goto wind;
+    }
+
+    CHANGELOG_OP_BOUNDARY_CHECK(frame, wind);
+
+    CHANGELOG_INIT(this, frame->local, loc->inode, loc->inode->gfid, 1);
+    if (!frame->local)
+        goto wind;
+
+    co = changelog_get_usable_buffer(frame->local);
+    if (!co)
+        goto wind;
+
+    CHANGLOG_FILL_FOP_NUMBER(co, frame->root->op, fop_fn, xtra_len);
+
+    changelog_set_usable_record_and_length(frame->local, xtra_len, 1);
+
+wind:
+    changelog_color_fop_and_inc_cnt(this, priv, frame->local);
+    STACK_WIND(frame, changelog_setattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setattr, loc, stbuf, valid, xdata);
+    return 0;
+}
+
+/* {f}removexattr */
+
+int32_t
+changelog_fremovexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    CHANGELOG_COND_GOTO(priv, ((op_ret < 0) || !local), unwind);
+
+    changelog_update(this, priv, local, CHANGELOG_TYPE_METADATA_XATTR);
+
+unwind:
+    changelog_dec_fop_cnt(this, priv, local);
+    CHANGELOG_STACK_UNWIND(fremovexattr, frame, op_ret, op_errno, xdata);
+
+    return 0;
+}
+
+int32_t
+changelog_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                       const char *name, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_opt_t *co = NULL;
+    size_t xtra_len = 0;
+
+    priv = this->private;
+    CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+
+    CHANGELOG_OP_BOUNDARY_CHECK(frame, wind);
+
+    CHANGELOG_INIT(this, frame->local, fd->inode, fd->inode->gfid, 1);
+
+    co = changelog_get_usable_buffer(frame->local);
+    if (!co)
+        goto wind;
+
+    CHANGLOG_FILL_FOP_NUMBER(co, frame->root->op, fop_fn, xtra_len);
+
+    changelog_set_usable_record_and_length(frame->local, xtra_len, 1);
+
+wind:
+    changelog_color_fop_and_inc_cnt(this, priv, frame->local);
+    STACK_WIND(frame, changelog_fremovexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fremovexattr, fd, name, xdata);
+    return 0;
+}
+
+int32_t
+changelog_removexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    CHANGELOG_COND_GOTO(priv, ((op_ret < 0) || !local), unwind);
+
+    changelog_update(this, priv, local, CHANGELOG_TYPE_METADATA_XATTR);
+
+unwind:
+    changelog_dec_fop_cnt(this, priv, local);
+    CHANGELOG_STACK_UNWIND(removexattr, frame, op_ret, op_errno, xdata);
+
+    return 0;
+}
+
+int32_t
+changelog_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                      const char *name, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_opt_t *co = NULL;
+    size_t xtra_len = 0;
+
+    priv = this->private;
+    CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+
+    CHANGELOG_OP_BOUNDARY_CHECK(frame, wind);
+
+    CHANGELOG_INIT(this, frame->local, loc->inode, loc->inode->gfid, 1);
+
+    co = changelog_get_usable_buffer(frame->local);
+    if (!co)
+        goto wind;
+
+    CHANGLOG_FILL_FOP_NUMBER(co, frame->root->op, fop_fn, xtra_len);
+
+    changelog_set_usable_record_and_length(frame->local, xtra_len, 1);
+
+wind:
+    changelog_color_fop_and_inc_cnt(this, priv, frame->local);
+    STACK_WIND(frame, changelog_removexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->removexattr, loc, name, xdata);
+    return 0;
+}
+
+/* {f}setxattr */
+
+int32_t
+changelog_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    CHANGELOG_COND_GOTO(priv, ((op_ret < 0) || !local), unwind);
+
+    changelog_update(this, priv, local, CHANGELOG_TYPE_METADATA_XATTR);
+
+unwind:
+    changelog_dec_fop_cnt(this, priv, local);
+    CHANGELOG_STACK_UNWIND(setxattr, frame, op_ret, op_errno, xdata);
+
+    return 0;
+}
+
+/* changelog_handle_virtual_xattr:
+ *         Handles virtual setxattr 'glusterfs.geo-rep.trigger-sync' on files.
+ *         Following is the behaviour based on the value of xattr.
+ *                         1: Captures only DATA entry in changelog.
+ *                         2: Tries to captures both ENTRY and DATA entry in
+ *                            changelog. If failed to get pargfid, only DATA
+ *                            entry is captured.
+ *           any other value: ENOTSUP is returned.
+ */
+static void
+changelog_handle_virtual_xattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                               dict_t *dict)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_local_t *local = NULL;
+    int32_t value = 0;
+    int ret = 0;
+    int dict_ret = 0;
+    gf_boolean_t valid = _gf_false;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    dict_ret = dict_get_int32(dict, GF_XATTR_TRIGGER_SYNC, &value);
+
+    if ((dict_ret == 0 && value == 1) && ((loc->inode->ia_type == IA_IFDIR) ||
+                                          (loc->inode->ia_type == IA_IFREG)))
+        valid = _gf_true;
+
+    if (valid) {
+        ret = changelog_fill_entry_buf(frame, this, loc, &local);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_INFO, 0, CHANGELOG_MSG_ENTRY_BUF_INFO,
+                    "gfid=%s", uuid_utoa(loc->inode->gfid), NULL);
+            goto unwind;
+        }
+        changelog_update(this, priv, local, CHANGELOG_TYPE_ENTRY);
+
+    unwind:
+        /* Capture DATA only if it's a file. */
+        if (loc->inode->ia_type != IA_IFDIR)
+            changelog_update(this, priv, frame->local, CHANGELOG_TYPE_DATA);
+        /* Assign local to prev_entry, so unwind will take
+         * care of cleanup. */
+        ((changelog_local_t *)(frame->local))->prev_entry = local;
+        CHANGELOG_STACK_UNWIND(setxattr, frame, 0, 0, NULL);
+        return;
+    } else {
+        CHANGELOG_STACK_UNWIND(setxattr, frame, -1, ENOTSUP, NULL);
+        return;
+    }
+}
+
+int32_t
+changelog_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   dict_t *dict, int32_t flags, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_opt_t *co = NULL;
+    size_t xtra_len = 0;
+
+    priv = this->private;
+    CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+
+    CHANGELOG_OP_BOUNDARY_CHECK(frame, wind);
+
+    CHANGELOG_INIT(this, frame->local, loc->inode, loc->inode->gfid, 1);
+
+    /* On setting this virtual xattr on a file, an explicit data
+     * sync is triggered from geo-rep as CREATE|DATA entry is
+     * recorded in changelog based on xattr value.
+     */
+    if (dict_get(dict, GF_XATTR_TRIGGER_SYNC)) {
+        changelog_handle_virtual_xattr(frame, this, loc, dict);
+        return 0;
+    }
+
+    co = changelog_get_usable_buffer(frame->local);
+    if (!co)
+        goto wind;
+
+    CHANGLOG_FILL_FOP_NUMBER(co, frame->root->op, fop_fn, xtra_len);
+
+    changelog_set_usable_record_and_length(frame->local, xtra_len, 1);
+
+wind:
+    changelog_color_fop_and_inc_cnt(this, priv, frame->local);
+    STACK_WIND(frame, changelog_setxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setxattr, loc, dict, flags, xdata);
+    return 0;
+}
+
+int32_t
+changelog_fsetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    CHANGELOG_COND_GOTO(priv, ((op_ret < 0) || !local), unwind);
+
+    changelog_update(this, priv, local, CHANGELOG_TYPE_METADATA_XATTR);
+
+unwind:
+    changelog_dec_fop_cnt(this, priv, local);
+    CHANGELOG_STACK_UNWIND(fsetxattr, frame, op_ret, op_errno, xdata);
+
+    return 0;
+}
+
+int32_t
+changelog_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+                    int32_t flags, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_opt_t *co = NULL;
+    size_t xtra_len = 0;
+
+    priv = this->private;
+    CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+    CHANGELOG_IF_INTERNAL_FOP_THEN_GOTO(frame, xdata, wind);
+
+    CHANGELOG_OP_BOUNDARY_CHECK(frame, wind);
+
+    CHANGELOG_INIT(this, frame->local, fd->inode, fd->inode->gfid, 1);
+
+    co = changelog_get_usable_buffer(frame->local);
+    if (!co)
+        goto wind;
+
+    CHANGLOG_FILL_FOP_NUMBER(co, frame->root->op, fop_fn, xtra_len);
+
+    changelog_set_usable_record_and_length(frame->local, xtra_len, 1);
+
+wind:
+    changelog_color_fop_and_inc_cnt(this, priv, frame->local);
+    STACK_WIND(frame, changelog_fsetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
+    return 0;
+}
+
+int32_t
+changelog_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xattr,
+                      dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    CHANGELOG_COND_GOTO(priv, ((op_ret < 0) || !local), unwind);
+
+    changelog_update(this, priv, local, CHANGELOG_TYPE_METADATA);
+
+unwind:
+    changelog_dec_fop_cnt(this, priv, local);
+    CHANGELOG_STACK_UNWIND(xattrop, frame, op_ret, op_errno, xattr, xdata);
+
+    return 0;
+}
+
+int32_t
+changelog_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                  gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_opt_t *co = NULL;
+    size_t xtra_len = 0;
+    int ret = 0;
+    void *size_attr = NULL;
+
+    priv = this->private;
+    CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+    ret = dict_get_ptr(xattr, GF_XATTR_SHARD_FILE_SIZE, &size_attr);
+    if (ret)
+        goto wind;
+
+    CHANGELOG_OP_BOUNDARY_CHECK(frame, wind);
+
+    CHANGELOG_INIT(this, frame->local, loc->inode, loc->inode->gfid, 1);
+
+    co = changelog_get_usable_buffer(frame->local);
+    if (!co)
+        goto wind;
+
+    CHANGLOG_FILL_FOP_NUMBER(co, frame->root->op, fop_fn, xtra_len);
+
+    changelog_set_usable_record_and_length(frame->local, xtra_len, 1);
+
+wind:
+    changelog_color_fop_and_inc_cnt(this, priv, frame->local);
+    STACK_WIND(frame, changelog_xattrop_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->xattrop, loc, optype, xattr, xdata);
+    return 0;
+}
+
+int32_t
+changelog_fxattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *xattr,
+                       dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    CHANGELOG_COND_GOTO(priv, ((op_ret < 0) || !local), unwind);
+
+    changelog_update(this, priv, local, CHANGELOG_TYPE_METADATA_XATTR);
+
+unwind:
+    changelog_dec_fop_cnt(this, priv, local);
+    CHANGELOG_STACK_UNWIND(fxattrop, frame, op_ret, op_errno, xattr, xdata);
+
+    return 0;
+}
+
+int32_t
+changelog_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                   gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_opt_t *co = NULL;
+    size_t xtra_len = 0;
+    void *size_attr = NULL;
+    int ret = 0;
+
+    priv = this->private;
+    CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+    ret = dict_get_ptr(xattr, GF_XATTR_SHARD_FILE_SIZE, &size_attr);
+    if (ret)
+        goto wind;
+
+    CHANGELOG_OP_BOUNDARY_CHECK(frame, wind);
+
+    CHANGELOG_INIT(this, frame->local, fd->inode, fd->inode->gfid, 1);
+
+    co = changelog_get_usable_buffer(frame->local);
+    if (!co)
+        goto wind;
+
+    CHANGLOG_FILL_FOP_NUMBER(co, frame->root->op, fop_fn, xtra_len);
+
+    changelog_set_usable_record_and_length(frame->local, xtra_len, 1);
+
+wind:
+    changelog_color_fop_and_inc_cnt(this, priv, frame->local);
+    STACK_WIND(frame, changelog_fxattrop_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fxattrop, fd, optype, xattr, xdata);
+    return 0;
+}
+/* }}} */
+
+/* Data modification fops - TYPE I */
+
+/* {{{ */
+
+/* {f}truncate() */
+
+int32_t
+changelog_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                       struct iatt *postbuf, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    CHANGELOG_COND_GOTO(priv, ((op_ret < 0) || !local), unwind);
+
+    changelog_update(this, priv, local, CHANGELOG_TYPE_DATA);
+
+unwind:
+    changelog_dec_fop_cnt(this, priv, local);
+    CHANGELOG_STACK_UNWIND(truncate, frame, op_ret, op_errno, prebuf, postbuf,
+                           xdata);
+    return 0;
+}
+
+int32_t
+changelog_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   off_t offset, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+
+    priv = this->private;
+    CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+
+    CHANGELOG_INIT(this, frame->local, loc->inode, loc->inode->gfid, 0);
+    LOCK(&priv->c_snap_lock);
+    {
+        if (priv->c_snap_fd != -1 && priv->barrier_enabled == _gf_true) {
+            changelog_snap_handle_ascii_change(
+                this, &(((changelog_local_t *)(frame->local))->cld));
+        }
+    }
+    UNLOCK(&priv->c_snap_lock);
+
+wind:
+    changelog_color_fop_and_inc_cnt(this, priv, frame->local);
+    STACK_WIND(frame, changelog_truncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+    return 0;
+}
+
+int32_t
+changelog_ftruncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                        struct iatt *postbuf, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    CHANGELOG_COND_GOTO(priv, ((op_ret < 0) || !local), unwind);
+
+    changelog_update(this, priv, local, CHANGELOG_TYPE_DATA);
+
+unwind:
+    changelog_dec_fop_cnt(this, priv, local);
+    CHANGELOG_STACK_UNWIND(ftruncate, frame, op_ret, op_errno, prebuf, postbuf,
+                           xdata);
+    return 0;
+}
+
+int32_t
+changelog_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                    dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+
+    priv = this->private;
+    CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+
+    CHANGELOG_INIT(this, frame->local, fd->inode, fd->inode->gfid, 0);
+    LOCK(&priv->c_snap_lock);
+    {
+        if (priv->c_snap_fd != -1 && priv->barrier_enabled == _gf_true) {
+            changelog_snap_handle_ascii_change(
+                this, &(((changelog_local_t *)(frame->local))->cld));
+        }
+    }
+    UNLOCK(&priv->c_snap_lock);
+
+wind:
+    changelog_color_fop_and_inc_cnt(this, priv, frame->local);
+    STACK_WIND(frame, changelog_ftruncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+    return 0;
+}
+
+/* writev() */
+
+int32_t
+changelog_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+    changelog_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    CHANGELOG_COND_GOTO(priv, ((op_ret <= 0) || !local), unwind);
+
+    changelog_update(this, priv, local, CHANGELOG_TYPE_DATA);
+
+unwind:
+    changelog_dec_fop_cnt(this, priv, local);
+    CHANGELOG_STACK_UNWIND(writev, frame, op_ret, op_errno, prebuf, postbuf,
+                           xdata);
+    return 0;
+}
+
+int32_t
+changelog_writev(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                 struct iovec *vector, int32_t count, off_t offset,
+                 uint32_t flags, struct iobref *iobref, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+
+    priv = this->private;
+    CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+
+    CHANGELOG_INIT(this, frame->local, fd->inode, fd->inode->gfid, 0);
+    LOCK(&priv->c_snap_lock);
+    {
+        if (priv->c_snap_fd != -1 && priv->barrier_enabled == _gf_true) {
+            changelog_snap_handle_ascii_change(
+                this, &(((changelog_local_t *)(frame->local))->cld));
+        }
+    }
+    UNLOCK(&priv->c_snap_lock);
+
+wind:
+    changelog_color_fop_and_inc_cnt(this, priv, frame->local);
+    STACK_WIND(frame, changelog_writev_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->writev, fd, vector, count, offset,
+               flags, iobref, xdata);
+    return 0;
+}
+
+/* }}} */
+
+/* open, release and other beasts */
+
+/* {{{ */
+
+int
+changelog_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int op_ret, int op_errno, fd_t *fd, dict_t *xdata)
+{
+    int ret = 0;
+    changelog_priv_t *priv = NULL;
+    changelog_event_t ev = {
+        0,
+    };
+    gf_boolean_t logopen = _gf_false;
+
+    priv = this->private;
+    if (frame->local) {
+        frame->local = NULL;
+        logopen = _gf_true;
+    }
+
+    CHANGELOG_COND_GOTO(priv, ((op_ret < 0) || !logopen), unwind);
+
+    /* fill the event structure */
+    ev.ev_type = CHANGELOG_OP_TYPE_OPEN;
+    gf_uuid_copy(ev.u.open.gfid, fd->inode->gfid);
+    ev.u.open.flags = fd->flags;
+    changelog_dispatch_event(this, priv, &ev);
+
+    if (changelog_ev_selected(this, &priv->ev_selection,
+                              CHANGELOG_OP_TYPE_RELEASE)) {
+        ret = fd_ctx_set(fd, this, (uint64_t)(long)0x1);
+        if (ret)
+            gf_smsg(this->name, GF_LOG_WARNING, 0, CHANGELOG_MSG_SET_FD_CONTEXT,
+                    NULL);
+    }
+
+unwind:
+    CHANGELOG_STACK_UNWIND(open, frame, op_ret, op_errno, fd, xdata);
+    return 0;
+}
+
+int
+changelog_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+               fd_t *fd, dict_t *xdata)
+{
+    changelog_priv_t *priv = NULL;
+
+    priv = this->private;
+    CHANGELOG_NOT_ACTIVE_THEN_GOTO(frame, priv, wind);
+
+    frame->local = (void *)0x1; /* do not dereference in ->cbk */
+
+wind:
+    STACK_WIND(frame, changelog_open_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->open, loc, flags, fd, xdata);
+    return 0;
+}
+
+/* }}} */
+
+/* {{{ */
+
+/* }}} */
+
+int32_t
+_changelog_generic_dispatcher(dict_t *dict, char *key, data_t *value,
+                              void *data)
+{
+    xlator_t *this = NULL;
+    changelog_priv_t *priv = NULL;
+
+    this = data;
+    priv = this->private;
+
+    changelog_dispatch_event(this, priv, (changelog_event_t *)value->data);
+    return 0;
+}
+
+/**
+ * changelog ipc dispatches events, pointers of which are passed in
+ * @xdata. Dispatching is orderless (whatever order dict_foreach()
+ * traverses the dictionary).
+ */
+int32_t
+changelog_ipc(call_frame_t *frame, xlator_t *this, int32_t op, dict_t *xdata)
+{
+    if (op != GF_IPC_TARGET_CHANGELOG)
+        goto wind;
+
+    /* it's for us, do the job */
+    if (xdata)
+        (void)dict_foreach(xdata, _changelog_generic_dispatcher, this);
+
+    STACK_UNWIND_STRICT(ipc, frame, 0, 0, NULL);
+    return 0;
+
+wind:
+    STACK_WIND(frame, default_ipc_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ipc, op, xdata);
+    return 0;
+}
+
+/* {{{ */
+
+int32_t
+changelog_release(xlator_t *this, fd_t *fd)
+{
+    changelog_event_t ev = {
+        0,
+    };
+    changelog_priv_t *priv = NULL;
+
+    priv = this->private;
+
+    ev.ev_type = CHANGELOG_OP_TYPE_RELEASE;
+    gf_uuid_copy(ev.u.release.gfid, fd->inode->gfid);
+    changelog_dispatch_event(this, priv, &ev);
+
+    (void)fd_ctx_del(fd, this, NULL);
+
+    return 0;
+}
+
+/* }}} */
+
+/**
+ * The
+ *   - @init ()
+ *   - @fini ()
+ *   - @reconfigure ()
+ *   ... and helper routines
+ */
+
+/**
+ * needed if there are more operation modes in the future.
+ */
+static void
+changelog_assign_opmode(changelog_priv_t *priv, char *mode)
+{
+    if (strncmp(mode, "realtime", 8) == 0) {
+        priv->op_mode = CHANGELOG_MODE_RT;
+    }
+}
+
+static void
+changelog_assign_encoding(changelog_priv_t *priv, char *enc)
+{
+    if (strncmp(enc, "binary", 6) == 0) {
+        priv->encode_mode = CHANGELOG_ENCODE_BINARY;
+    } else if (strncmp(enc, "ascii", 5) == 0) {
+        priv->encode_mode = CHANGELOG_ENCODE_ASCII;
+    }
+}
+
+static void
+changelog_assign_barrier_timeout(changelog_priv_t *priv, uint32_t timeout)
+{
+    LOCK(&priv->lock);
+    {
+        priv->timeout.tv_sec = timeout;
+    }
+    UNLOCK(&priv->lock);
+}
+
+/* cleanup any helper threads that are running */
+static void
+changelog_cleanup_helper_threads(xlator_t *this, changelog_priv_t *priv)
+{
+    if (priv->cr.rollover_th) {
+        (void)changelog_thread_cleanup(this, priv->cr.rollover_th);
+        priv->cr.rollover_th = 0;
+    }
+
+    if (priv->cf.fsync_th) {
+        (void)changelog_thread_cleanup(this, priv->cf.fsync_th);
+        priv->cf.fsync_th = 0;
+    }
+}
+
+/* spawn helper thread; cleaning up in case of errors */
+static int
+changelog_spawn_helper_threads(xlator_t *this, changelog_priv_t *priv)
+{
+    int ret = 0;
+
+    /* Geo-Rep snapshot dependency:
+     *
+     * To implement explicit rollover of changlog journal on barrier
+     * notification, a pipe is created to communicate between
+     * 'changelog_rollover' thread and changelog main thread. The select
+     * call used to wait till roll-over time in changelog_rollover thread
+     * is modified to wait on read end of the pipe. When barrier
+     * notification comes (i.e, in 'reconfigure'), select in
+     * changelog_rollover thread is woken up explicitly by writing into
+     * the write end of the pipe in 'reconfigure'.
+     */
+
+    priv->cr.notify = _gf_false;
+    priv->cr.this = this;
+    ret = gf_thread_create(&priv->cr.rollover_th, NULL, changelog_rollover,
+                           priv, "clogro");
+    if (ret)
+        goto out;
+
+    if (priv->fsync_interval) {
+        priv->cf.this = this;
+        ret = gf_thread_create(&priv->cf.fsync_th, NULL, changelog_fsync_thread,
+                               priv, "clogfsyn");
+    }
+
+    if (ret)
+        changelog_cleanup_helper_threads(this, priv);
+
+out:
+    return ret;
+}
+
+int
+notify(xlator_t *this, int event, void *data, ...)
+{
+    changelog_priv_t *priv = NULL;
+    dict_t *dict = NULL;
+    char buf[1] = {1};
+    int barrier = DICT_DEFAULT;
+    gf_boolean_t bclean_req = _gf_false;
+    int ret = 0;
+    int ret1 = 0;
+    struct list_head queue = {
+        0,
+    };
+    uint64_t xprtcnt = 0;
+    uint64_t clntcnt = 0;
+    changelog_clnt_t *conn = NULL;
+    gf_boolean_t cleanup_notify = _gf_false;
+    char sockfile[UNIX_PATH_MAX] = {
+        0,
+    };
+    rpcsvc_listener_t *listener = NULL;
+    rpcsvc_listener_t *next = NULL;
+
+    INIT_LIST_HEAD(&queue);
+
+    priv = this->private;
+    if (!priv)
+        goto out;
+
+    if (event == GF_EVENT_PARENT_DOWN) {
+        priv->victim = data;
+        gf_log(this->name, GF_LOG_INFO,
+               "cleanup changelog rpc connection of brick %s",
+               priv->victim->name);
+
+        if (priv->rpc_active) {
+            this->cleanup_starting = 1;
+            changelog_destroy_rpc_listner(this, priv);
+            conn = &priv->connections;
+            if (conn)
+                changelog_ev_cleanup_connections(this, conn);
+            xprtcnt = GF_ATOMIC_GET(priv->xprtcnt);
+            clntcnt = GF_ATOMIC_GET(priv->clntcnt);
+            if (!xprtcnt && !clntcnt) {
+                LOCK(&priv->lock);
+                {
+                    cleanup_notify = priv->notify_down;
+                    priv->notify_down = _gf_true;
+                }
+                UNLOCK(&priv->lock);
+                if (priv->rpc) {
+                    list_for_each_entry_safe(listener, next,
+                                             &priv->rpc->listeners, list)
+                    {
+                        if (listener->trans) {
+                            rpc_transport_unref(listener->trans);
+                        }
+                    }
+                    rpcsvc_destroy(priv->rpc);
+                    priv->rpc = NULL;
+                }
+                CHANGELOG_MAKE_SOCKET_PATH(priv->changelog_brick, sockfile,
+                                           UNIX_PATH_MAX);
+                sys_unlink(sockfile);
+                if (!cleanup_notify)
+                    default_notify(this, GF_EVENT_PARENT_DOWN, data);
+            }
+        } else {
+            default_notify(this, GF_EVENT_PARENT_DOWN, data);
+        }
+        goto out;
+    }
+
+    if (event == GF_EVENT_TRANSLATOR_OP) {
+        dict = data;
+
+        barrier = dict_get_str_boolean(dict, "barrier", DICT_DEFAULT);
+
+        switch (barrier) {
+            case DICT_ERROR:
+                gf_smsg(this->name, GF_LOG_ERROR, 0,
+                        CHANGELOG_MSG_DICT_GET_FAILED, "dict_get_str_boolean",
+                        NULL);
+                ret = -1;
+                goto out;
+
+            case BARRIER_OFF:
+                gf_smsg(this->name, GF_LOG_INFO, 0,
+                        CHANGELOG_MSG_BARRIER_STATE_NOTIFY, "off", NULL);
+
+                CHANGELOG_NOT_ON_THEN_GOTO(priv, ret, out);
+                LOCK(&priv->c_snap_lock);
+                {
+                    changelog_snap_logging_stop(this, priv);
+                }
+                UNLOCK(&priv->c_snap_lock);
+
+                LOCK(&priv->bflags.lock);
+                {
+                    if (priv->bflags.barrier_ext == _gf_false)
+                        ret = -1;
+                }
+                UNLOCK(&priv->bflags.lock);
+
+                if (ret == -1) {
+                    gf_smsg(this->name, GF_LOG_ERROR, 0,
+                            CHANGELOG_MSG_BARRIER_ERROR, NULL);
+                    goto out;
+                }
+
+                /* Stop changelog barrier and dequeue all fops */
+                LOCK(&priv->lock);
+                {
+                    if (priv->barrier_enabled == _gf_true)
+                        __chlog_barrier_disable(this, &queue);
+                    else
+                        ret = -1;
+                }
+                UNLOCK(&priv->lock);
+                /* If ret = -1, then changelog barrier is already
+                 * disabled because of error or timeout.
+                 */
+                if (ret == 0) {
+                    chlog_barrier_dequeue_all(this, &queue);
+                    gf_smsg(this->name, GF_LOG_INFO, 0,
+                            CHANGELOG_MSG_BARRIER_DISABLED, NULL);
+                } else {
+                    gf_smsg(this->name, GF_LOG_ERROR, 0,
+                            CHANGELOG_MSG_BARRIER_ALREADY_DISABLED, NULL);
+                }
+
+                LOCK(&priv->bflags.lock);
+                {
+                    priv->bflags.barrier_ext = _gf_false;
+                }
+                UNLOCK(&priv->bflags.lock);
+
+                goto out;
+
+            case BARRIER_ON:
+                gf_smsg(this->name, GF_LOG_INFO, 0,
+                        CHANGELOG_MSG_BARRIER_STATE_NOTIFY, "on", NULL);
+
+                CHANGELOG_NOT_ON_THEN_GOTO(priv, ret, out);
+                LOCK(&priv->c_snap_lock);
+                {
+                    changelog_snap_logging_start(this, priv);
+                }
+                UNLOCK(&priv->c_snap_lock);
+
+                LOCK(&priv->bflags.lock);
+                {
+                    if (priv->bflags.barrier_ext == _gf_true)
+                        ret = -1;
+                    else
+                        priv->bflags.barrier_ext = _gf_true;
+                }
+                UNLOCK(&priv->bflags.lock);
+
+                if (ret == -1) {
+                    gf_smsg(this->name, GF_LOG_ERROR, 0,
+                            CHANGELOG_MSG_BARRIER_ON_ERROR, NULL);
+                    goto out;
+                }
+
+                ret = pthread_mutex_lock(&priv->bn.bnotify_mutex);
+                CHANGELOG_PTHREAD_ERROR_HANDLE_1(ret, out, bclean_req);
+                {
+                    priv->bn.bnotify = _gf_true;
+                }
+                ret = pthread_mutex_unlock(&priv->bn.bnotify_mutex);
+                CHANGELOG_PTHREAD_ERROR_HANDLE_1(ret, out, bclean_req);
+
+                /* Start changelog barrier */
+                LOCK(&priv->lock);
+                {
+                    ret = __chlog_barrier_enable(this, priv);
+                }
+                UNLOCK(&priv->lock);
+                if (ret == -1) {
+                    changelog_barrier_cleanup(this, priv, &queue);
+                    goto out;
+                }
+
+                gf_smsg(this->name, GF_LOG_INFO, 0,
+                        CHANGELOG_MSG_BARRIER_ENABLE, NULL);
+
+                ret = changelog_barrier_notify(priv, buf);
+                if (ret) {
+                    gf_smsg(this->name, GF_LOG_ERROR, 0,
+                            CHANGELOG_MSG_WRITE_FAILED, "Explicit roll over",
+                            NULL);
+                    changelog_barrier_cleanup(this, priv, &queue);
+                    ret = -1;
+                    goto out;
+                }
+
+                ret = pthread_mutex_lock(&priv->bn.bnotify_mutex);
+                CHANGELOG_PTHREAD_ERROR_HANDLE_1(ret, out, bclean_req);
+                {
+                    /* The while condition check is required here to
+                     * handle spurious wakeup of cond wait that can
+                     * happen with pthreads. See man page */
+                    while (priv->bn.bnotify == _gf_true) {
+                        ret = pthread_cond_wait(&priv->bn.bnotify_cond,
+                                                &priv->bn.bnotify_mutex);
+                        CHANGELOG_PTHREAD_ERROR_HANDLE_1(ret, out, bclean_req);
+                    }
+                    if (priv->bn.bnotify_error == _gf_true) {
+                        ret = -1;
+                        priv->bn.bnotify_error = _gf_false;
+                    }
+                }
+                ret1 = pthread_mutex_unlock(&priv->bn.bnotify_mutex);
+                CHANGELOG_PTHREAD_ERROR_HANDLE_1(ret1, out, bclean_req);
+                gf_smsg(this->name, GF_LOG_INFO, 0,
+                        CHANGELOG_MSG_BNOTIFY_COND_INFO, NULL);
+
+                goto out;
+
+            case DICT_DEFAULT:
+                gf_smsg(this->name, GF_LOG_ERROR, 0,
+                        CHANGELOG_MSG_BARRIER_KEY_NOT_FOUND, NULL);
+                ret = -1;
+                goto out;
+
+            default:
+                gf_smsg(this->name, GF_LOG_ERROR, EINVAL,
+                        CHANGELOG_MSG_ERROR_IN_DICT_GET, NULL);
+                ret = -1;
+                goto out;
+        }
+    } else {
+        ret = default_notify(this, event, data);
+    }
+
+out:
+    if (bclean_req)
+        changelog_barrier_cleanup(this, priv, &queue);
+
+    return ret;
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    if (!this)
+        return ret;
+
+    ret = xlator_mem_acct_init(this, gf_changelog_mt_end + 1);
+
+    if (ret != 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, ENOMEM,
+                CHANGELOG_MSG_MEMORY_INIT_FAILED, NULL);
+        return ret;
+    }
+
+    return ret;
+}
+
+static int
+changelog_init(xlator_t *this, changelog_priv_t *priv)
+{
+    int i = 0;
+    int ret = 0;
+    changelog_log_data_t cld = {
+        0,
+    };
+
+    priv->maps[CHANGELOG_TYPE_DATA] = "D ";
+    priv->maps[CHANGELOG_TYPE_METADATA] = "M ";
+    priv->maps[CHANGELOG_TYPE_METADATA_XATTR] = "M ";
+    priv->maps[CHANGELOG_TYPE_ENTRY] = "E ";
+
+    for (; i < CHANGELOG_MAX_TYPE; i++) {
+        /* start with version 1 */
+        priv->slice.changelog_version[i] = 1;
+    }
+
+    if (!priv->active)
+        return ret;
+
+    /**
+     * start with a fresh changelog file every time. this is done
+     * in case there was an encoding change. so... things are kept
+     * simple here.
+     */
+    changelog_fill_rollover_data(&cld, _gf_false);
+
+    ret = htime_open(this, priv, cld.cld_roll_time);
+    /* call htime open with cld's rollover_time */
+    if (ret)
+        goto out;
+
+    LOCK(&priv->lock);
+    {
+        ret = changelog_inject_single_event(this, priv, &cld);
+    }
+    UNLOCK(&priv->lock);
+
+    /* ... and finally spawn the helpers threads */
+    ret = changelog_spawn_helper_threads(this, priv);
+
+out:
+    return ret;
+}
+
+/**
+ * Init barrier related condition variables and locks
+ */
+static int
+changelog_barrier_pthread_init(xlator_t *this, changelog_priv_t *priv)
+{
+    gf_boolean_t bn_mutex_init = _gf_false;
+    gf_boolean_t bn_cond_init = _gf_false;
+    gf_boolean_t dm_mutex_black_init = _gf_false;
+    gf_boolean_t dm_cond_black_init = _gf_false;
+    gf_boolean_t dm_mutex_white_init = _gf_false;
+    gf_boolean_t dm_cond_white_init = _gf_false;
+    gf_boolean_t cr_mutex_init = _gf_false;
+    gf_boolean_t cr_cond_init = _gf_false;
+    int ret = 0;
+
+    if ((ret = pthread_mutex_init(&priv->bn.bnotify_mutex, NULL)) != 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                CHANGELOG_MSG_PTHREAD_MUTEX_INIT_FAILED, "name=bnotify",
+                "ret=%d", ret, NULL);
+        ret = -1;
+        goto out;
+    }
+    bn_mutex_init = _gf_true;
+
+    if ((ret = pthread_cond_init(&priv->bn.bnotify_cond, NULL)) != 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                CHANGELOG_MSG_PTHREAD_COND_INIT_FAILED, "name=bnotify",
+                "ret=%d", ret, NULL);
+        ret = -1;
+        goto out;
+    }
+    bn_cond_init = _gf_true;
+
+    if ((ret = pthread_mutex_init(&priv->dm.drain_black_mutex, NULL)) != 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                CHANGELOG_MSG_PTHREAD_MUTEX_INIT_FAILED, "name=drain_black",
+                "ret=%d", ret, NULL);
+        ret = -1;
+        goto out;
+    }
+    dm_mutex_black_init = _gf_true;
+
+    if ((ret = pthread_cond_init(&priv->dm.drain_black_cond, NULL)) != 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                CHANGELOG_MSG_PTHREAD_COND_INIT_FAILED, "name=drain_black",
+                "ret=%d", ret, NULL);
+        ret = -1;
+        goto out;
+    }
+    dm_cond_black_init = _gf_true;
+
+    if ((ret = pthread_mutex_init(&priv->dm.drain_white_mutex, NULL)) != 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                CHANGELOG_MSG_PTHREAD_MUTEX_INIT_FAILED, "name=drain_white",
+                "ret=%d", ret, NULL);
+        ret = -1;
+        goto out;
+    }
+    dm_mutex_white_init = _gf_true;
+
+    if ((ret = pthread_cond_init(&priv->dm.drain_white_cond, NULL)) != 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                CHANGELOG_MSG_PTHREAD_COND_INIT_FAILED, "name=drain_white",
+                "ret=%d", ret, NULL);
+        ret = -1;
+        goto out;
+    }
+    dm_cond_white_init = _gf_true;
+
+    if ((pthread_mutex_init(&priv->cr.lock, NULL)) != 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                CHANGELOG_MSG_PTHREAD_MUTEX_INIT_FAILED,
+                "name=changelog_rollover", "ret=%d", ret, NULL);
+        ret = -1;
+        goto out;
+    }
+    cr_mutex_init = _gf_true;
+
+    if ((pthread_cond_init(&priv->cr.cond, NULL)) != 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                CHANGELOG_MSG_PTHREAD_COND_INIT_FAILED,
+                "changelog_rollover cond init failed", "ret=%d", ret, NULL);
+        ret = -1;
+        goto out;
+    }
+    cr_cond_init = _gf_true;
+out:
+    if (ret) {
+        if (bn_mutex_init)
+            pthread_mutex_destroy(&priv->bn.bnotify_mutex);
+        if (bn_cond_init)
+            pthread_cond_destroy(&priv->bn.bnotify_cond);
+        if (dm_mutex_black_init)
+            pthread_mutex_destroy(&priv->dm.drain_black_mutex);
+        if (dm_cond_black_init)
+            pthread_cond_destroy(&priv->dm.drain_black_cond);
+        if (dm_mutex_white_init)
+            pthread_mutex_destroy(&priv->dm.drain_white_mutex);
+        if (dm_cond_white_init)
+            pthread_cond_destroy(&priv->dm.drain_white_cond);
+        if (cr_mutex_init)
+            pthread_mutex_destroy(&priv->cr.lock);
+        if (cr_cond_init)
+            pthread_cond_destroy(&priv->cr.cond);
+    }
+    return ret;
+}
+
+/* Destroy barrier related condition variables and locks */
+static void
+changelog_barrier_pthread_destroy(changelog_priv_t *priv)
+{
+    pthread_mutex_destroy(&priv->bn.bnotify_mutex);
+    pthread_cond_destroy(&priv->bn.bnotify_cond);
+    pthread_mutex_destroy(&priv->dm.drain_black_mutex);
+    pthread_cond_destroy(&priv->dm.drain_black_cond);
+    pthread_mutex_destroy(&priv->dm.drain_white_mutex);
+    pthread_cond_destroy(&priv->dm.drain_white_cond);
+    pthread_mutex_destroy(&priv->cr.lock);
+    pthread_cond_destroy(&priv->cr.cond);
+    LOCK_DESTROY(&priv->bflags.lock);
+}
+
+static void
+changelog_cleanup_rpc(xlator_t *this, changelog_priv_t *priv)
+{
+    /* terminate rpc server */
+    if (!this->cleanup_starting)
+        changelog_destroy_rpc_listner(this, priv);
+
+    (void)changelog_cleanup_rpc_threads(this, priv);
+    /* cleanup rot buffs */
+    rbuf_dtor(priv->rbuf);
+
+    /* cleanup poller thread */
+    if (priv->poller)
+        (void)changelog_thread_cleanup(this, priv->poller);
+}
+
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    int ret = 0;
+    char *tmp = NULL;
+    changelog_priv_t *priv = NULL;
+    gf_boolean_t active_earlier = _gf_true;
+    gf_boolean_t active_now = _gf_true;
+    gf_boolean_t rpc_active_earlier = _gf_true;
+    gf_boolean_t rpc_active_now = _gf_true;
+    gf_boolean_t iniate_rpc = _gf_false;
+    changelog_time_slice_t *slice = NULL;
+    changelog_log_data_t cld = {
+        0,
+    };
+    char htime_dir[PATH_MAX] = {
+        0,
+    };
+    char csnap_dir[PATH_MAX] = {
+        0,
+    };
+    uint32_t timeout = 0;
+
+    priv = this->private;
+    if (!priv)
+        goto out;
+
+    ret = -1;
+    active_earlier = priv->active;
+    rpc_active_earlier = priv->rpc_active;
+
+    /* first stop the rollover and the fsync thread */
+    changelog_cleanup_helper_threads(this, priv);
+
+    GF_OPTION_RECONF("changelog-dir", tmp, options, str, out);
+    if (!tmp) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_MSG_DIR_OPTIONS_NOT_SET,
+                NULL);
+        goto out;
+    }
+
+    GF_FREE(priv->changelog_dir);
+    priv->changelog_dir = gf_strdup(tmp);
+    if (!priv->changelog_dir)
+        goto out;
+
+    ret = mkdir_p(priv->changelog_dir, 0600, _gf_true);
+
+    if (ret)
+        goto out;
+    CHANGELOG_FILL_HTIME_DIR(priv->changelog_dir, htime_dir);
+    ret = mkdir_p(htime_dir, 0600, _gf_true);
+
+    if (ret)
+        goto out;
+
+    CHANGELOG_FILL_CSNAP_DIR(priv->changelog_dir, csnap_dir);
+    ret = mkdir_p(csnap_dir, 0600, _gf_true);
+
+    if (ret)
+        goto out;
+
+    GF_OPTION_RECONF("changelog", active_now, options, bool, out);
+    GF_OPTION_RECONF("changelog-notification", rpc_active_now, options, bool,
+                     out);
+
+    /* If journalling is enabled, enable rpc notifications */
+    if (active_now && !active_earlier) {
+        if (!rpc_active_earlier)
+            iniate_rpc = _gf_true;
+    }
+
+    if (rpc_active_now && !rpc_active_earlier) {
+        iniate_rpc = _gf_true;
+    }
+
+    /* TODO: Disable of changelog-notifications is not supported for now
+     * as there is no clean way of cleaning up of rpc resources
+     */
+
+    if (iniate_rpc) {
+        ret = changelog_init_rpc(this, priv);
+        if (ret)
+            goto out;
+        priv->rpc_active = _gf_true;
+    }
+
+    /**
+     * changelog_handle_change() handles changes that could possibly
+     * have been submit changes before changelog deactivation.
+     */
+    if (!active_now)
+        priv->active = _gf_false;
+
+    GF_OPTION_RECONF("op-mode", tmp, options, str, out);
+    changelog_assign_opmode(priv, tmp);
+
+    tmp = NULL;
+
+    GF_OPTION_RECONF("encoding", tmp, options, str, out);
+    changelog_assign_encoding(priv, tmp);
+
+    GF_OPTION_RECONF("rollover-time", priv->rollover_time, options, int32, out);
+    GF_OPTION_RECONF("fsync-interval", priv->fsync_interval, options, int32,
+                     out);
+    GF_OPTION_RECONF("changelog-barrier-timeout", timeout, options, time, out);
+    changelog_assign_barrier_timeout(priv, timeout);
+
+    GF_OPTION_RECONF("capture-del-path", priv->capture_del_path, options, bool,
+                     out);
+
+    if (active_now || active_earlier) {
+        changelog_fill_rollover_data(&cld, !active_now);
+
+        slice = &priv->slice;
+
+        LOCK(&priv->lock);
+        {
+            ret = changelog_inject_single_event(this, priv, &cld);
+            if (!ret && active_now)
+                SLICE_VERSION_UPDATE(slice);
+        }
+        UNLOCK(&priv->lock);
+
+        if (ret)
+            goto out;
+
+        if (active_now) {
+            if (!active_earlier) {
+                gf_smsg(this->name, GF_LOG_INFO, 0, CHANGELOG_MSG_RECONFIGURE,
+                        NULL);
+                htime_create(this, priv, gf_time());
+            }
+            ret = changelog_spawn_helper_threads(this, priv);
+        }
+    }
+
+out:
+    if (ret) {
+        /* TODO */
+    } else {
+        gf_msg_debug(this->name, 0, "changelog reconfigured");
+        if (active_now && priv)
+            priv->active = _gf_true;
+    }
+
+    return ret;
+}
+
+static void
+changelog_freeup_options(xlator_t *this, changelog_priv_t *priv)
+{
+    int ret = 0;
+
+    ret = priv->cb->dtor(this, &priv->cd);
+    if (ret)
+        gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_MSG_FREEUP_FAILED, NULL);
+    GF_FREE(priv->changelog_brick);
+    GF_FREE(priv->changelog_dir);
+}
+
+static int
+changelog_init_options(xlator_t *this, changelog_priv_t *priv)
+{
+    int ret = 0;
+    char *tmp = NULL;
+    uint32_t timeout = 0;
+    char htime_dir[PATH_MAX] = {
+        0,
+    };
+    char csnap_dir[PATH_MAX] = {
+        0,
+    };
+
+    GF_OPTION_INIT("changelog-brick", tmp, str, error_return);
+    priv->changelog_brick = gf_strdup(tmp);
+    if (!priv->changelog_brick)
+        goto error_return;
+
+    tmp = NULL;
+
+    GF_OPTION_INIT("changelog-dir", tmp, str, dealloc_1);
+    priv->changelog_dir = gf_strdup(tmp);
+    if (!priv->changelog_dir)
+        goto dealloc_1;
+
+    tmp = NULL;
+
+    /**
+     * create the directory even if change-logging would be inactive
+     * so that consumers can _look_ into it (finding nothing...)
+     */
+    ret = mkdir_p(priv->changelog_dir, 0600, _gf_true);
+
+    if (ret)
+        goto dealloc_2;
+
+    CHANGELOG_FILL_HTIME_DIR(priv->changelog_dir, htime_dir);
+    ret = mkdir_p(htime_dir, 0600, _gf_true);
+    if (ret)
+        goto dealloc_2;
+
+    CHANGELOG_FILL_CSNAP_DIR(priv->changelog_dir, csnap_dir);
+    ret = mkdir_p(csnap_dir, 0600, _gf_true);
+    if (ret)
+        goto dealloc_2;
+
+    GF_OPTION_INIT("changelog", priv->active, bool, dealloc_2);
+    GF_OPTION_INIT("changelog-notification", priv->rpc_active, bool, dealloc_2);
+    GF_OPTION_INIT("capture-del-path", priv->capture_del_path, bool, dealloc_2);
+
+    GF_OPTION_INIT("op-mode", tmp, str, dealloc_2);
+    changelog_assign_opmode(priv, tmp);
+
+    tmp = NULL;
+
+    GF_OPTION_INIT("encoding", tmp, str, dealloc_2);
+    changelog_assign_encoding(priv, tmp);
+    changelog_encode_change(priv);
+
+    GF_OPTION_INIT("rollover-time", priv->rollover_time, int32, dealloc_2);
+
+    GF_OPTION_INIT("fsync-interval", priv->fsync_interval, int32, dealloc_2);
+
+    GF_OPTION_INIT("changelog-barrier-timeout", timeout, time, dealloc_2);
+    changelog_assign_barrier_timeout(priv, timeout);
+
+    GF_ASSERT(cb_bootstrap[priv->op_mode].mode == priv->op_mode);
+    priv->cb = &cb_bootstrap[priv->op_mode];
+
+    /* ... now bootstrap the logger */
+    ret = priv->cb->ctor(this, &priv->cd);
+    if (ret)
+        goto dealloc_2;
+
+    priv->changelog_fd = -1;
+
+    return 0;
+
+dealloc_2:
+    GF_FREE(priv->changelog_dir);
+dealloc_1:
+    GF_FREE(priv->changelog_brick);
+error_return:
+    return -1;
+}
+
+static int
+changelog_init_rpc(xlator_t *this, changelog_priv_t *priv)
+{
+    rpcsvc_t *rpc = NULL;
+    changelog_ev_selector_t *selection = NULL;
+
+    selection = &priv->ev_selection;
+
+    /* initialize event selection */
+    changelog_init_event_selection(this, selection);
+
+    priv->rbuf = rbuf_init(NR_ROTT_BUFFS);
+    if (!priv->rbuf)
+        goto cleanup_thread;
+
+    rpc = changelog_init_rpc_listener(this, priv, priv->rbuf, NR_DISPATCHERS);
+    if (!rpc)
+        goto cleanup_rbuf;
+    priv->rpc = rpc;
+
+    return 0;
+
+cleanup_rbuf:
+    rbuf_dtor(priv->rbuf);
+cleanup_thread:
+    if (priv->poller)
+        (void)changelog_thread_cleanup(this, priv->poller);
+
+    return -1;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    int ret = -1;
+    changelog_priv_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("changelog", this, error_return);
+
+    if (!this->children || this->children->next) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_MSG_CHILD_MISCONFIGURED,
+                NULL);
+        goto error_return;
+    }
+
+    if (!this->parents) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, CHANGELOG_MSG_VOL_MISCONFIGURED,
+                NULL);
+        goto error_return;
+    }
+
+    priv = GF_CALLOC(1, sizeof(*priv), gf_changelog_mt_priv_t);
+    if (!priv)
+        goto error_return;
+
+    this->local_pool = mem_pool_new(changelog_local_t, 64);
+    if (!this->local_pool) {
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, CHANGELOG_MSG_NO_MEMORY,
+                NULL);
+        goto cleanup_priv;
+    }
+
+    LOCK_INIT(&priv->lock);
+    LOCK_INIT(&priv->c_snap_lock);
+    GF_ATOMIC_INIT(priv->listnercnt, 0);
+    GF_ATOMIC_INIT(priv->clntcnt, 0);
+    GF_ATOMIC_INIT(priv->xprtcnt, 0);
+    INIT_LIST_HEAD(&priv->xprt_list);
+    priv->htime_fd = -1;
+
+    ret = changelog_init_options(this, priv);
+    if (ret)
+        goto cleanup_mempool;
+
+    /* snap dependency changes */
+    priv->dm.black_fop_cnt = 0;
+    priv->dm.white_fop_cnt = 0;
+    priv->dm.drain_wait_black = _gf_false;
+    priv->dm.drain_wait_white = _gf_false;
+    priv->current_color = FOP_COLOR_BLACK;
+    priv->explicit_rollover = _gf_false;
+
+    priv->cr.notify = _gf_false;
+    /* Mutex is not needed as threads are not spawned yet */
+    priv->bn.bnotify = _gf_false;
+    priv->bn.bnotify_error = _gf_false;
+    ret = changelog_barrier_pthread_init(this, priv);
+    if (ret)
+        goto cleanup_options;
+    LOCK_INIT(&priv->bflags.lock);
+    priv->bflags.barrier_ext = _gf_false;
+
+    /* Changelog barrier init */
+    INIT_LIST_HEAD(&priv->queue);
+    priv->barrier_enabled = _gf_false;
+
+    if (priv->rpc_active || priv->active) {
+        /* RPC ball rolling.. */
+        ret = changelog_init_rpc(this, priv);
+        if (ret)
+            goto cleanup_barrier;
+        priv->rpc_active = _gf_true;
+    }
+
+    ret = changelog_init(this, priv);
+    if (ret)
+        goto cleanup_rpc;
+
+    gf_msg_debug(this->name, 0, "changelog translator loaded");
+
+    this->private = priv;
+    return 0;
+
+cleanup_rpc:
+    if (priv->rpc_active) {
+        changelog_cleanup_rpc(this, priv);
+    }
+cleanup_barrier:
+    changelog_barrier_pthread_destroy(priv);
+cleanup_options:
+    changelog_freeup_options(this, priv);
+cleanup_mempool:
+    mem_pool_destroy(this->local_pool);
+    this->local_pool = NULL;
+cleanup_priv:
+    GF_FREE(priv);
+error_return:
+    this->private = NULL;
+    return -1;
+}
+
+void
+fini(xlator_t *this)
+{
+    changelog_priv_t *priv = NULL;
+    struct list_head queue = {
+        0,
+    };
+
+    priv = this->private;
+
+    if (priv) {
+        if (priv->active || priv->rpc_active) {
+            /* terminate RPC server/threads */
+            changelog_cleanup_rpc(this, priv);
+            GF_FREE(priv->ev_dispatcher);
+        }
+        /* call barrier_disable to cancel timer */
+        if (priv->barrier_enabled)
+            __chlog_barrier_disable(this, &queue);
+
+        /* cleanup barrier related objects */
+        changelog_barrier_pthread_destroy(priv);
+
+        /* cleanup helper threads */
+        changelog_cleanup_helper_threads(this, priv);
+
+        /* cleanup allocated options */
+        changelog_freeup_options(this, priv);
+
+        /* deallocate mempool */
+        mem_pool_destroy(this->local_pool);
+
+        if (priv->htime_fd != -1) {
+            sys_close(priv->htime_fd);
+        }
+
+        /* finally, dealloac private variable */
+        GF_FREE(priv);
+    }
+
+    this->private = NULL;
+    this->local_pool = NULL;
+
+    return;
+}
+
+struct xlator_fops fops = {
+    .open = changelog_open,
+    .mknod = changelog_mknod,
+    .mkdir = changelog_mkdir,
+    .create = changelog_create,
+    .symlink = changelog_symlink,
+    .writev = changelog_writev,
+    .truncate = changelog_truncate,
+    .ftruncate = changelog_ftruncate,
+    .link = changelog_link,
+    .rename = changelog_rename,
+    .unlink = changelog_unlink,
+    .rmdir = changelog_rmdir,
+    .setattr = changelog_setattr,
+    .fsetattr = changelog_fsetattr,
+    .setxattr = changelog_setxattr,
+    .fsetxattr = changelog_fsetxattr,
+    .removexattr = changelog_removexattr,
+    .fremovexattr = changelog_fremovexattr,
+    .ipc = changelog_ipc,
+    .xattrop = changelog_xattrop,
+    .fxattrop = changelog_fxattrop,
+};
+
+struct xlator_cbks cbks = {
+    .forget = changelog_forget,
+    .release = changelog_release,
+};
+
+struct volume_options options[] = {
+    {.key = {"changelog"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = "enable/disable change-logging",
+     .op_version = {3},
+     .flags = OPT_FLAG_SETTABLE,
+     .level = OPT_STATUS_BASIC,
+     .tags = {"journal", "georep", "glusterfind"}},
+    {.key = {"changelog-notification"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = "enable/disable changelog live notification",
+     .op_version = {3},
+     .level = OPT_STATUS_BASIC,
+     .tags = {"bitrot", "georep"}},
+    {.key = {"changelog-brick"},
+     .type = GF_OPTION_TYPE_PATH,
+     .description = "brick path to generate unique socket file name."
+                    " should be the export directory of the volume strictly.",
+     .default_value = "{{ brick.path }}",
+     .op_version = {3},
+     .tags = {"journal"}},
+    {.key = {"changelog-dir"},
+     .type = GF_OPTION_TYPE_PATH,
+     .description = "directory for the changelog files",
+     .default_value = "{{ brick.path }}/.glusterfs/changelogs",
+     .op_version = {3},
+     .flags = OPT_FLAG_SETTABLE,
+     .level = OPT_STATUS_ADVANCED,
+     .tags = {"journal", "georep", "glusterfind"}},
+    {.key = {"op-mode"},
+     .type = GF_OPTION_TYPE_STR,
+     .default_value = "realtime",
+     .value = {"realtime"},
+     .description = "operation mode - futuristic operation modes",
+     .op_version = {3},
+     .tags = {"journal"}},
+    {.key = {"encoding"},
+     .type = GF_OPTION_TYPE_STR,
+     .default_value = "ascii",
+     .value = {"binary", "ascii"},
+     .description = "encoding type for changelogs",
+     .op_version = {3},
+     .flags = OPT_FLAG_SETTABLE,
+     .level = OPT_STATUS_ADVANCED,
+     .tags = {"journal"}},
+    {.key = {"rollover-time"},
+     .default_value = "15",
+     .type = GF_OPTION_TYPE_TIME,
+     .description = "time to switch to a new changelog file (in seconds)",
+     .op_version = {3},
+     .flags = OPT_FLAG_SETTABLE,
+     .level = OPT_STATUS_ADVANCED,
+     .tags = {"journal", "georep", "glusterfind"}},
+    {.key = {"fsync-interval"},
+     .type = GF_OPTION_TYPE_TIME,
+     .default_value = "5",
+     .description = "do not open CHANGELOG file with O_SYNC mode."
+                    " instead perform fsync() at specified intervals",
+     .op_version = {3},
+     .flags = OPT_FLAG_SETTABLE,
+     .level = OPT_STATUS_ADVANCED,
+     .tags = {"journal"}},
+    {.key = {"changelog-barrier-timeout"},
+     .type = GF_OPTION_TYPE_TIME,
+     .default_value = BARRIER_TIMEOUT,
+     .description = "After 'timeout' seconds since the time 'barrier' "
+                    "option was set to \"on\", unlink/rmdir/rename  "
+                    "operations are no longer blocked and previously "
+                    "blocked fops are allowed to go through",
+     .op_version = {3},
+     .flags = OPT_FLAG_SETTABLE,
+     .level = OPT_STATUS_ADVANCED,
+     .tags = {"journal"}},
+    {.key = {"capture-del-path"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .description = "enable/disable capturing paths of deleted entries",
+     .op_version = {3},
+     .flags = OPT_FLAG_SETTABLE,
+     .level = OPT_STATUS_BASIC,
+     .tags = {"journal", "glusterfind"}},
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "changelog",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/features/cloudsync/Makefile.am b/xlators/features/cloudsync/Makefile.am
new file mode 100644
index 00000000000..a985f42a877
--- /dev/null
+++ b/xlators/features/cloudsync/Makefile.am
@@ -0,0 +1,3 @@
+SUBDIRS = src
+
+CLEANFILES =
diff --git a/xlators/features/cloudsync/src/Makefile.am b/xlators/features/cloudsync/src/Makefile.am
new file mode 100644
index 00000000000..e2a277e372b
--- /dev/null
+++ b/xlators/features/cloudsync/src/Makefile.am
@@ -0,0 +1,46 @@
+SUBDIRS = cloudsync-plugins
+
+xlator_LTLIBRARIES = cloudsync.la
+
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+cloudsync_sources = cloudsync.c
+
+CLOUDSYNC_SRC = $(top_srcdir)/xlators/features/cloudsync/src
+CLOUDSYNC_BLD = $(top_builddir)/xlators/features/cloudsync/src
+
+cloudsynccommon_sources = $(CLOUDSYNC_SRC)/cloudsync-common.c
+
+noinst_HEADERS = $(CLOUDSYNC_BLD)/cloudsync.h \
+		 $(CLOUDSYNC_BLD)/cloudsync-mem-types.h \
+		 $(CLOUDSYNC_BLD)/cloudsync-messages.h \
+		 $(CLOUDSYNC_BLD)/cloudsync-common.h
+
+cloudsync_la_SOURCES = $(cloudsync_sources) $(cloudsynccommon_sources)
+
+nodist_cloudsync_la_SOURCES = cloudsync-autogen-fops.c cloudsync-autogen-fops.h
+BUILT_SOURCES = cloudsync-autogen-fops.h
+
+cloudsync_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+cloudsync_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la $(LIB_DL)
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src -I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src \
+	      -DCS_PLUGINDIR=\"$(libdir)/glusterfs/$(PACKAGE_VERSION)/cloudsync-plugins\"
+AM_CFLAGS = -Wall -fno-strict-aliasing $(GF_CFLAGS)
+
+noinst_PYTHON = cloudsync-fops-c.py cloudsync-fops-h.py
+EXTRA_DIST = cloudsync-autogen-fops-tmpl.c cloudsync-autogen-fops-tmpl.h
+
+cloudsync-autogen-fops.c: cloudsync-fops-c.py cloudsync-autogen-fops-tmpl.c
+	$(PYTHON) $(CLOUDSYNC_SRC)/cloudsync-fops-c.py \
+	$(CLOUDSYNC_SRC)/cloudsync-autogen-fops-tmpl.c > $@
+
+cloudsync-autogen-fops.h: cloudsync-fops-h.py cloudsync-autogen-fops-tmpl.h
+	$(PYTHON) $(CLOUDSYNC_SRC)/cloudsync-fops-h.py \
+	$(CLOUDSYNC_SRC)/cloudsync-autogen-fops-tmpl.h > $@
+
+CLEANFILES = $(nodist_cloudsync_la_SOURCES)
+
+uninstall-local:
+	rm -f $(DESTDIR)$(xlatordir)/cloudsync.so
diff --git a/xlators/features/cloudsync/src/cloudsync-autogen-fops-tmpl.c b/xlators/features/cloudsync/src/cloudsync-autogen-fops-tmpl.c
new file mode 100644
index 00000000000..ee63f983980
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-autogen-fops-tmpl.c
@@ -0,0 +1,30 @@
+/*
+  Copyright (c) 2008-2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+/* File: cloudsync-autogen-fops-tmpl.c
+ * This file contains the CLOUDSYNC autogenerated FOPs. This is run through
+ * the code generator, generator.py to generate the required FOPs.
+ */
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include <dlfcn.h>
+
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+#include "cloudsync.h"
+#include "cloudsync-common.h"
+#include <glusterfs/call-stub.h>
+
+#pragma generate
diff --git a/xlators/features/cloudsync/src/cloudsync-autogen-fops-tmpl.h b/xlators/features/cloudsync/src/cloudsync-autogen-fops-tmpl.h
new file mode 100644
index 00000000000..d922c77d8aa
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-autogen-fops-tmpl.h
@@ -0,0 +1,24 @@
+/*
+  Copyright (c) 2008-2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+/* File: clousync-autogen-fops-tmpl.h
+ * This file contains the cloudsync autogenerated FOPs declarations.
+ */
+
+#ifndef _CLOUDSYNC_AUTOGEN_FOPS_H
+#define _CLOUDSYNC_AUTOGEN_FOPS_H
+
+#include <glusterfs/xlator.h>
+#include "cloudsync.h"
+#include "cloudsync-common.h"
+
+#pragma generate
+
+#endif /* _CLOUDSYNC_AUTOGEN_FOPS_H */
diff --git a/xlators/features/cloudsync/src/cloudsync-common.c b/xlators/features/cloudsync/src/cloudsync-common.c
new file mode 100644
index 00000000000..445a31b90e7
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-common.c
@@ -0,0 +1,60 @@
+/*
+  Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "cloudsync-common.h"
+
+void
+cs_xattrinfo_wipe(cs_local_t *local)
+{
+    if (local->xattrinfo.lxattr) {
+        if (local->xattrinfo.lxattr->file_path)
+            GF_FREE(local->xattrinfo.lxattr->file_path);
+
+        if (local->xattrinfo.lxattr->volname)
+            GF_FREE(local->xattrinfo.lxattr->volname);
+
+        GF_FREE(local->xattrinfo.lxattr);
+    }
+}
+
+void
+cs_local_wipe(xlator_t *this, cs_local_t *local)
+{
+    if (!local)
+        return;
+
+    loc_wipe(&local->loc);
+
+    if (local->fd) {
+        fd_unref(local->fd);
+        local->fd = NULL;
+    }
+
+    if (local->stub) {
+        call_stub_destroy(local->stub);
+        local->stub = NULL;
+    }
+
+    if (local->xattr_req)
+        dict_unref(local->xattr_req);
+
+    if (local->xattr_rsp)
+        dict_unref(local->xattr_rsp);
+
+    if (local->dlfd)
+        fd_unref(local->dlfd);
+
+    if (local->remotepath)
+        GF_FREE(local->remotepath);
+
+    cs_xattrinfo_wipe(local);
+
+    mem_put(local);
+}
diff --git a/xlators/features/cloudsync/src/cloudsync-common.h b/xlators/features/cloudsync/src/cloudsync-common.h
new file mode 100644
index 00000000000..11d233460a4
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-common.h
@@ -0,0 +1,134 @@
+/*
+  Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+#ifndef _CLOUDSYNC_COMMON_H
+#define _CLOUDSYNC_COMMON_H
+
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/call-stub.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/syncop.h>
+#include <glusterfs/compat-errno.h>
+#include "cloudsync-mem-types.h"
+#include "cloudsync-messages.h"
+
+typedef struct cs_loc_xattr {
+    char *file_path;
+    uuid_t uuid;
+    uuid_t gfid;
+    char *volname;
+} cs_loc_xattr_t;
+
+typedef struct cs_size_xattr {
+    uint64_t size;
+    uint64_t blksize;
+    uint64_t blocks;
+} cs_size_xattr_t;
+
+typedef struct cs_local {
+    loc_t loc;
+    fd_t *fd;
+    call_stub_t *stub;
+    call_frame_t *main_frame;
+    int op_errno;
+    int op_ret;
+    fd_t *dlfd;
+    off_t dloffset;
+    struct iatt stbuf;
+    dict_t *xattr_rsp;
+    dict_t *xattr_req;
+    glusterfs_fop_t fop;
+    gf_boolean_t locked;
+    int call_cnt;
+    inode_t *inode;
+    char *remotepath;
+
+    struct {
+        /* offset, flags and size are the information needed
+         * by read fop for remote read operation. These will be
+         * populated in cloudsync read fop, before being passed
+         * on to the plugin performing remote read.
+         */
+        off_t offset;
+        uint32_t flags;
+        size_t size;
+        cs_loc_xattr_t *lxattr;
+    } xattrinfo;
+
+} cs_local_t;
+
+typedef int (*fop_download_t)(call_frame_t *frame, void *config);
+
+typedef int (*fop_remote_read_t)(call_frame_t *, void *);
+
+typedef void *(*store_init)(xlator_t *this);
+
+typedef int (*store_reconfigure)(xlator_t *this, dict_t *options);
+
+typedef void (*store_fini)(void *config);
+
+struct cs_remote_stores {
+    char *name;                    /* store name */
+    void *config;                  /* store related information */
+    fop_download_t dlfop;          /* store specific download function */
+    fop_remote_read_t rdfop;       /* store specific read function */
+    store_init init;               /* store init to initialize store config */
+    store_reconfigure reconfigure; /* reconfigure store config */
+    store_fini fini;
+    void *handle; /* shared library handle*/
+};
+
+typedef struct cs_private {
+    xlator_t *this;
+    struct cs_remote_stores *stores;
+    gf_boolean_t abortdl;
+    pthread_spinlock_t lock;
+    gf_boolean_t remote_read;
+} cs_private_t;
+
+void
+cs_local_wipe(xlator_t *this, cs_local_t *local);
+
+void
+cs_xattrinfo_wipe(cs_local_t *local);
+
+#define CS_STACK_UNWIND(fop, frame, params...)                                 \
+    do {                                                                       \
+        cs_local_t *__local = NULL;                                            \
+        xlator_t *__xl = NULL;                                                 \
+        if (frame) {                                                           \
+            __xl = frame->this;                                                \
+            __local = frame->local;                                            \
+            frame->local = NULL;                                               \
+        }                                                                      \
+        STACK_UNWIND_STRICT(fop, frame, params);                               \
+        cs_local_wipe(__xl, __local);                                          \
+    } while (0)
+
+#define CS_STACK_DESTROY(frame)                                                \
+    do {                                                                       \
+        cs_local_t *__local = NULL;                                            \
+        xlator_t *__xl = NULL;                                                 \
+        __xl = frame->this;                                                    \
+        __local = frame->local;                                                \
+        frame->local = NULL;                                                   \
+        STACK_DESTROY(frame->root);                                            \
+        cs_local_wipe(__xl, __local);                                          \
+    } while (0)
+
+typedef struct store_methods {
+    int (*fop_download)(call_frame_t *frame, void *config);
+    int (*fop_remote_read)(call_frame_t *, void *);
+    /* return type should be the store config */
+    void *(*fop_init)(xlator_t *this);
+    int (*fop_reconfigure)(xlator_t *this, dict_t *options);
+    void (*fop_fini)(void *config);
+} store_methods_t;
+
+#endif /* _CLOUDSYNC_COMMON_H */
diff --git a/xlators/features/cloudsync/src/cloudsync-fops-c.py b/xlators/features/cloudsync/src/cloudsync-fops-c.py
new file mode 100755
index 00000000000..c27df97ae58
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-fops-c.py
@@ -0,0 +1,324 @@
+#!/usr/bin/python3
+
+from __future__ import print_function
+import os
+import sys
+
+curdir = os.path.dirname(sys.argv[0])
+gendir = os.path.join(curdir, '../../../../libglusterfs/src')
+sys.path.append(gendir)
+from generator import ops, fop_subs, cbk_subs, generate
+
+FD_DATA_MODIFYING_OP_FOP_TEMPLATE = """
+int32_t
+cs_@NAME@ (call_frame_t *frame, xlator_t *this,
+           @LONG_ARGS@)
+{
+        int                         op_errno        = EINVAL ;
+        cs_local_t                 *local           = NULL;
+        int                         ret             = 0;
+        cs_inode_ctx_t             *ctx             = NULL;
+        gf_cs_obj_state             state           = -1;
+
+        VALIDATE_OR_GOTO (frame, err);
+        VALIDATE_OR_GOTO (this, err);
+        VALIDATE_OR_GOTO (fd, err);
+
+        local = cs_local_init (this, frame, NULL, fd, GF_FOP_@UPNAME@);
+        if (!local) {
+
+                gf_msg (this->name, GF_LOG_ERROR, 0, 0, "local init failed");
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        __cs_inode_ctx_get (this, fd->inode, &ctx);
+
+        if (ctx)
+                state = __cs_get_file_state (fd->inode, ctx);
+        else
+                state = GF_CS_LOCAL;
+
+        xdata = xdata ? dict_ref (xdata) : dict_new ();
+
+        if (!xdata) {
+                gf_msg (this->name, GF_LOG_ERROR, 0, 0, "insufficient memory");
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        local->xattr_req = xdata;
+
+        ret = dict_set_uint32 (local->xattr_req, GF_CS_OBJECT_STATUS, 1);
+        if (ret) {
+                gf_msg (this->name, GF_LOG_ERROR, 0, 0, "dict_set failed key:"
+                        " %s", GF_CS_OBJECT_STATUS);
+                goto err;
+        }
+
+        local->stub = fop_@NAME@_stub (frame, cs_resume_@NAME@,
+                                       @SHORT_ARGS@);
+        if (!local->stub) {
+                gf_msg (this->name, GF_LOG_ERROR, 0, 0, "insufficient memory");
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+
+        if (state == GF_CS_LOCAL) {
+                STACK_WIND (frame, cs_@NAME@_cbk,
+                            FIRST_CHILD(this), FIRST_CHILD(this)->fops->@NAME@,
+                            @SHORT_ARGS@);
+        } else {
+                local->call_cnt++;
+                ret = locate_and_execute (frame);
+                if (ret) {
+                        op_errno = ENOMEM;
+                        goto err;
+                }
+        }
+
+        return 0;
+
+err:
+        CS_STACK_UNWIND (@NAME@, frame, -1, op_errno, @CBK_ERROR_ARGS@);
+
+        return 0;
+}
+"""
+
+FD_DATA_MODIFYING_RESUME_OP_FOP_TEMPLATE = """
+int32_t
+cs_resume_@NAME@ (call_frame_t *frame, xlator_t *this,
+                  @LONG_ARGS@)
+{
+        int              ret    = 0;
+
+        ret = cs_resume_postprocess (this, frame, fd->inode);
+        if (ret) {
+                goto unwind;
+        }
+
+        cs_inodelk_unlock (frame);
+
+        STACK_WIND (frame, cs_@NAME@_cbk,
+                    FIRST_CHILD(this), FIRST_CHILD(this)->fops->@NAME@,
+                    @SHORT_ARGS@);
+
+        return 0;
+
+unwind:
+
+        cs_inodelk_unlock (frame);
+
+        cs_common_cbk (frame);
+
+        return 0;
+}
+"""
+FD_DATA_MODIFYING_OP_FOP_CBK_TEMPLATE = """
+int32_t
+cs_@NAME@_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno,
+               @LONG_ARGS@)
+{
+        cs_local_t      *local = NULL;
+        int              ret = 0;
+        uint64_t         val = 0;
+        fd_t            *fd = NULL;
+
+        local = frame->local;
+        fd = local->fd;
+
+        /* Do we need lock here? */
+        local->call_cnt++;
+
+        if (op_ret == -1) {
+                ret = dict_get_uint64 (xdata, GF_CS_OBJECT_STATUS, &val);
+                if (ret == 0) {
+                        if (val == GF_CS_ERROR) {
+                                gf_msg (this->name, GF_LOG_ERROR, 0, 0,
+                                        "could not get file state, unwinding");
+                                op_ret = -1;
+                                op_errno = EIO;
+                                goto unwind;
+                        } else {
+                                __cs_inode_ctx_update (this, fd->inode, val);
+                                gf_msg (this->name, GF_LOG_INFO, 0, 0,
+                                        " state = %" PRIu64, val);
+
+                                if (local->call_cnt == 1 &&
+                                    (val == GF_CS_REMOTE ||
+                                     val == GF_CS_DOWNLOADING))  {
+                                        gf_msg (this->name, GF_LOG_INFO, 0,
+                                                0, " will repair and download "
+                                                "the file, current state : %"
+                                                PRIu64, val);
+                                        goto repair;
+                                } else {
+                                        gf_msg (this->name, GF_LOG_ERROR, 0, 0,
+                                                "second @NAME@, Unwinding");
+                                        goto unwind;
+                                }
+                        }
+                } else {
+                        gf_msg (this->name, GF_LOG_ERROR, 0, 0, "file state "
+                                "could not be figured, unwinding");
+                        goto unwind;
+                }
+        } else {
+                /* successful @NAME@ => file is local */
+                __cs_inode_ctx_update (this, fd->inode, GF_CS_LOCAL);
+                gf_msg (this->name, GF_LOG_INFO, 0, 0, "state : GF_CS_LOCAL"
+                        ", @NAME@ successful");
+
+                goto unwind;
+        }
+
+repair:
+        ret = locate_and_execute (frame);
+        if (ret) {
+                goto unwind;
+        }
+
+        return 0;
+
+unwind:
+        CS_STACK_UNWIND (@NAME@, frame, op_ret, op_errno, @SHORT_ARGS@);
+
+        return 0;
+}
+"""
+
+LOC_STAT_OP_FOP_TEMPLATE = """
+int32_t
+cs_@NAME@ (call_frame_t *frame, xlator_t *this,
+           @LONG_ARGS@)
+{
+        int              op_errno = EINVAL;
+        cs_local_t      *local = NULL;
+        int              ret   = 0;
+
+        local = cs_local_init (this, frame, loc, NULL, GF_FOP_@UPNAME@);
+        if (!local) {
+                gf_msg (this->name, GF_LOG_ERROR, 0, 0, "local is NULL");
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        if (loc->inode->ia_type == IA_IFDIR)
+                goto wind;
+
+        xdata = xdata ? dict_ref (xdata) : dict_new ();
+
+        if (!xdata) {
+                gf_msg (this->name, GF_LOG_ERROR, 0, 0, "insufficient memory");
+                op_errno = ENOMEM;
+                goto err;
+        }
+
+        local->xattr_req = xdata;
+
+        ret = dict_set_uint32 (local->xattr_req, GF_CS_OBJECT_STATUS, 1);
+        if (ret) {
+                gf_msg (this->name, GF_LOG_ERROR, 0, 0, "dict_set failed key:"
+                        " %s", GF_CS_OBJECT_STATUS);
+                goto err;
+        }
+
+wind:
+        STACK_WIND (frame, cs_@NAME@_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->@NAME@,
+                    @SHORT_ARGS@);
+
+        return 0;
+err:
+        CS_STACK_UNWIND (@NAME@, frame, -1, op_errno, @CBK_ERROR_ARGS@);
+
+        return 0;
+}
+"""
+
+LOC_STAT_OP_FOP_CBK_TEMPLATE = """
+int32_t
+cs_@NAME@_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno,
+               @LONG_ARGS@)
+{
+        int              ret = 0;
+        uint64_t         val = 0;
+        loc_t           *loc = NULL;
+        cs_local_t      *local = NULL;
+
+        local = frame->local;
+
+        loc = &local->loc;
+
+        if (op_ret == 0) {
+                ret = dict_get_uint64 (xdata, GF_CS_OBJECT_STATUS, &val);
+                if (!ret) {
+                        ret = __cs_inode_ctx_update (this, loc->inode, val);
+                        if (ret) {
+                                gf_msg (this->name, GF_LOG_ERROR, 0, 0,
+                                        "ctx update failed");
+                        }
+                }
+        } else {
+                cs_inode_ctx_reset (this, loc->inode);
+        }
+
+        CS_STACK_UNWIND (@NAME@, frame, op_ret, op_errno, @SHORT_ARGS@);
+
+        return 0;
+}
+"""
+
+# All xlator FOPs are covered in the following section just to create a clarity
+# The lists themselves are not used.
+entry_ops = ['mknod', 'mkdir', 'unlink', 'rmdir', 'symlink', 'rename', 'link',
+             'create']
+special_ops = ['statfs', 'lookup', 'ipc', 'compound', 'icreate', 'namelink']
+ignored_ops = ['getspec']
+inode_ops = ['stat', 'readlink', 'truncate', 'open', 'setxattr', 'getxattr',
+             'removexattr', 'opendir', 'access', 'inodelk', 'entrylk',
+             'xattrop', 'setattr', 'lease', 'getactivelk', 'setactivelk',
+             'discover']
+fd_ops = ['readv', 'writev', 'flush', 'fsync', 'fsyncdir', 'ftruncate',
+          'fstat', 'lk', 'readdir', 'finodelk', 'fentrylk', 'fxattrop',
+          'fsetxattr', 'fgetxattr', 'rchecksum', 'fsetattr', 'readdirp',
+          'fremovexattr', 'fallocate', 'discard', 'zerofill', 'seek']
+
+
+# These are the current actual lists used to generate the code
+
+# The following list contains fops which are fd based that modifies data
+fd_data_modify_op_fop_template = ['writev', 'flush', 'fsync',
+                                  'ftruncate', 'rchecksum', 'fallocate',
+                                  'discard', 'zerofill', 'seek']
+
+# The following list contains fops which are entry based that does not change
+# data
+loc_stat_op_fop_template = ['lookup', 'stat', 'discover', 'access', 'setattr',
+                            'getattr']
+
+# These fops need a separate implementation
+special_fops = ['statfs', 'setxattr', 'unlink', 'getxattr',
+                'truncate', 'fstat', 'readv', 'readdirp']
+
+def gen_defaults():
+    for name in ops:
+        if name in fd_data_modify_op_fop_template:
+            print(generate(FD_DATA_MODIFYING_OP_FOP_CBK_TEMPLATE, name, cbk_subs))
+            print(generate(FD_DATA_MODIFYING_RESUME_OP_FOP_TEMPLATE, name, fop_subs))
+            print(generate(FD_DATA_MODIFYING_OP_FOP_TEMPLATE, name, fop_subs))
+        elif name in loc_stat_op_fop_template:
+            print(generate(LOC_STAT_OP_FOP_CBK_TEMPLATE, name, cbk_subs))
+            print(generate(LOC_STAT_OP_FOP_TEMPLATE, name, fop_subs))
+
+for l in open(sys.argv[1], 'r').readlines():
+    if l.find('#pragma generate') != -1:
+        print("/* BEGIN GENERATED CODE - DO NOT MODIFY */")
+        gen_defaults()
+        print("/* END GENERATED CODE */")
+    else:
+        print(l[:-1])
diff --git a/xlators/features/cloudsync/src/cloudsync-fops-h.py b/xlators/features/cloudsync/src/cloudsync-fops-h.py
new file mode 100755
index 00000000000..faa2de651a7
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-fops-h.py
@@ -0,0 +1,31 @@
+#!/usr/bin/python3
+
+from __future__ import print_function
+import os
+import sys
+
+curdir = os.path.dirname(sys.argv[0])
+gendir = os.path.join(curdir, '../../../../libglusterfs/src')
+sys.path.append(gendir)
+from generator import ops, fop_subs, cbk_subs, generate
+
+OP_FOP_TEMPLATE = """
+int32_t
+cs_@NAME@ (call_frame_t *frame, xlator_t *this,
+                   @LONG_ARGS@);
+"""
+
+def gen_defaults():
+    for name, value in ops.items():
+        if name == 'getspec':
+            continue
+        print(generate(OP_FOP_TEMPLATE, name, fop_subs))
+
+
+for l in open(sys.argv[1], 'r').readlines():
+    if l.find('#pragma generate') != -1:
+        print("/* BEGIN GENERATED CODE - DO NOT MODIFY */")
+        gen_defaults()
+        print("/* END GENERATED CODE */")
+    else:
+        print(l[:-1])
diff --git a/xlators/features/cloudsync/src/cloudsync-mem-types.h b/xlators/features/cloudsync/src/cloudsync-mem-types.h
new file mode 100644
index 00000000000..220346405d0
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-mem-types.h
@@ -0,0 +1,22 @@
+/*
+ *   Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com>
+ *   This file is part of GlusterFS.
+ *
+ *   This file is licensed to you under your choice of the GNU Lesser
+ *   General Public License, version 3 or any later version (LGPLv3 or
+ *   later), or the GNU General Public License, version 2 (GPLv2), in all
+ *   cases as published by the Free Software Foundation.
+ */
+
+#ifndef __CLOUDSYNC_MEM_TYPES_H__
+#define __CLOUDSYNC_MEM_TYPES_H__
+
+#include <glusterfs/mem-types.h>
+enum cs_mem_types_ {
+    gf_cs_mt_cs_private_t = gf_common_mt_end + 1,
+    gf_cs_mt_cs_remote_stores_t,
+    gf_cs_mt_cs_inode_ctx_t,
+    gf_cs_mt_cs_lxattr_t,
+    gf_cs_mt_end
+};
+#endif /* __CLOUDSYNC_MEM_TYPES_H__ */
diff --git a/xlators/features/cloudsync/src/cloudsync-messages.h b/xlators/features/cloudsync/src/cloudsync-messages.h
new file mode 100644
index 00000000000..fb08f72de7f
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-messages.h
@@ -0,0 +1,16 @@
+/*
+ *   Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com>
+ *   This file is part of GlusterFS.
+ *
+ *   This file is licensed to you under your choice of the GNU Lesser
+ *   General Public License, version 3 or any later version (LGPLv3 or
+ *   later), or the GNU General Public License, version 2 (GPLv2), in all
+ *   cases as published by the Free Software Foundation.
+ */
+
+#ifndef __CLOUDSYNC_MESSAGES_H__
+#define __CLOUDSYNC_MESSAGES_H__
+
+/*TODO: define relevant message ids */
+
+#endif /* __CLOUDSYNC_MESSAGES_H__ */
diff --git a/xlators/features/cloudsync/src/cloudsync-plugins/Makefile.am b/xlators/features/cloudsync/src/cloudsync-plugins/Makefile.am
new file mode 100644
index 00000000000..a985f42a877
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-plugins/Makefile.am
@@ -0,0 +1,3 @@
+SUBDIRS = src
+
+CLEANFILES =
diff --git a/xlators/features/cloudsync/src/cloudsync-plugins/src/Makefile.am b/xlators/features/cloudsync/src/cloudsync-plugins/src/Makefile.am
new file mode 100644
index 00000000000..fb6b0580c6d
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-plugins/src/Makefile.am
@@ -0,0 +1,11 @@
+if BUILD_AMAZONS3_PLUGIN
+  AMAZONS3_DIR = cloudsyncs3
+endif
+
+if BUILD_CVLT_PLUGIN
+  CVLT_DIR = cvlt
+endif
+
+SUBDIRS = ${AMAZONS3_DIR} ${CVLT_DIR}
+
+CLEANFILES =
diff --git a/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/Makefile.am b/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/Makefile.am
new file mode 100644
index 00000000000..a985f42a877
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/Makefile.am
@@ -0,0 +1,3 @@
+SUBDIRS = src
+
+CLEANFILES =
diff --git a/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/src/Makefile.am b/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/src/Makefile.am
new file mode 100644
index 00000000000..6509426ef87
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/src/Makefile.am
@@ -0,0 +1,12 @@
+csp_LTLIBRARIES = cloudsyncs3.la
+cspdir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/cloudsync-plugins
+
+cloudsyncs3_la_SOURCES = libcloudsyncs3.c  $(top_srcdir)/xlators/features/cloudsync/src/cloudsync-common.c
+cloudsyncs3_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+cloudsyncs3_la_LDFLAGS = -module -export-symbols $(top_srcdir)/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/src/libcloudsyncs3.sym $(GF_XLATOR_LDFLAGS)
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src   -I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src -lcurlpp -lcryptopp
+noinst_HEADERS = libcloudsyncs3.h libcloudsyncs3-mem-types.h
+AM_CFLAGS = -Wall -fno-strict-aliasing $(GF_CFLAGS) -lcurl -lcrypto -I$(top_srcdir)/xlators/features/cloudsync/src
+CLEANFILES =
+
+EXTRA_DIST = libcloudsyncs3.sym
diff --git a/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/src/libcloudsyncs3-mem-types.h b/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/src/libcloudsyncs3-mem-types.h
new file mode 100644
index 00000000000..7ccfcc9f4b6
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/src/libcloudsyncs3-mem-types.h
@@ -0,0 +1,19 @@
+/*
+ *   Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com>
+ *   This file is part of GlusterFS.
+ *
+ *   This file is licensed to you under your choice of the GNU Lesser
+ *   General Public License, version 3 or any later version (LGPLv3 or
+ *   later), or the GNU General Public License, version 2 (GPLv2), in all
+ *   cases as published by the Free Software Foundation.
+ */
+
+#ifndef __LIBAWS_MEM_TYPES_H__
+#define __LIBAWS_MEM_TYPES_H__
+
+#include <glusterfs/mem-types.h>
+enum libaws_mem_types_ {
+    gf_libaws_mt_aws_private_t = gf_common_mt_end + 1,
+    gf_libaws_mt_end
+};
+#endif /* __CLOUDSYNC_MEM_TYPES_H__ */
diff --git a/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/src/libcloudsyncs3.c b/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/src/libcloudsyncs3.c
new file mode 100644
index 00000000000..23c3599825a
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/src/libcloudsyncs3.c
@@ -0,0 +1,584 @@
+/*
+  Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <stdlib.h>
+#include <openssl/hmac.h>
+#include <openssl/evp.h>
+#include <openssl/bio.h>
+#include <openssl/buffer.h>
+#include <openssl/crypto.h>
+#include <curl/curl.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/glusterfs.h>
+#include "libcloudsyncs3.h"
+#include "cloudsync-common.h"
+
+#define RESOURCE_SIZE 4096
+
+store_methods_t store_ops = {
+    .fop_download = aws_download_s3,
+    .fop_init = aws_init,
+    .fop_reconfigure = aws_reconfigure,
+    .fop_fini = aws_fini,
+};
+
+typedef struct aws_private {
+    char *hostname;
+    char *bucketid;
+    char *awssekey;
+    char *awskeyid;
+    gf_boolean_t abortdl;
+    pthread_spinlock_t lock;
+} aws_private_t;
+
+void *
+aws_init(xlator_t *this)
+{
+    aws_private_t *priv = NULL;
+    char *temp_str = NULL;
+    int ret = 0;
+
+    priv = GF_CALLOC(1, sizeof(aws_private_t), gf_libaws_mt_aws_private_t);
+    if (!priv) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "insufficient memory");
+        return NULL;
+    }
+
+    priv->abortdl = _gf_false;
+
+    pthread_spin_init(&priv->lock, PTHREAD_PROCESS_PRIVATE);
+
+    pthread_spin_lock(&(priv->lock));
+    {
+        if (dict_get_str(this->options, "s3plugin-seckey", &temp_str) == 0) {
+            priv->awssekey = gf_strdup(temp_str);
+            if (!priv->awssekey) {
+                gf_msg(this->name, GF_LOG_ERROR, ENOMEM, 0,
+                       "initializing aws secret key failed");
+                ret = -1;
+                goto unlock;
+            }
+        }
+
+        if (dict_get_str(this->options, "s3plugin-keyid", &temp_str) == 0) {
+            priv->awskeyid = gf_strdup(temp_str);
+            if (!priv->awskeyid) {
+                gf_msg(this->name, GF_LOG_ERROR, ENOMEM, 0,
+                       "initializing aws key ID failed");
+                ret = -1;
+                goto unlock;
+            }
+        }
+
+        if (dict_get_str(this->options, "s3plugin-bucketid", &temp_str) == 0) {
+            priv->bucketid = gf_strdup(temp_str);
+            if (!priv->bucketid) {
+                gf_msg(this->name, GF_LOG_ERROR, ENOMEM, 0,
+                       "initializing aws bucketid failed");
+
+                ret = -1;
+                goto unlock;
+            }
+        }
+
+        if (dict_get_str(this->options, "s3plugin-hostname", &temp_str) == 0) {
+            priv->hostname = gf_strdup(temp_str);
+            if (!priv->hostname) {
+                gf_msg(this->name, GF_LOG_ERROR, ENOMEM, 0,
+                       "initializing aws hostname failed");
+
+                ret = -1;
+                goto unlock;
+            }
+        }
+
+        gf_msg_debug(this->name, 0,
+                     "stored key: %s id: %s "
+                     "bucketid %s hostname: %s",
+                     priv->awssekey, priv->awskeyid, priv->bucketid,
+                     priv->hostname);
+    }
+unlock:
+    pthread_spin_unlock(&(priv->lock));
+
+    if (ret == -1) {
+        GF_FREE(priv->awskeyid);
+        GF_FREE(priv->awssekey);
+        GF_FREE(priv->bucketid);
+        GF_FREE(priv->hostname);
+        GF_FREE(priv);
+        priv = NULL;
+    }
+
+    return (void *)priv;
+}
+
+int
+aws_reconfigure(xlator_t *this, dict_t *options)
+{
+    aws_private_t *priv = NULL;
+    char *temp_str = NULL;
+    int ret = 0;
+    cs_private_t *cspriv = NULL;
+
+    cspriv = this->private;
+
+    priv = cspriv->stores->config;
+
+    if (!priv) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "null priv");
+        return -1;
+    }
+
+    pthread_spin_lock(&(priv->lock));
+    {
+        if (dict_get_str(options, "s3plugin-seckey", &temp_str) == 0) {
+            priv->awssekey = gf_strdup(temp_str);
+            if (!priv->awssekey) {
+                gf_msg(this->name, GF_LOG_ERROR, ENOMEM, 0,
+                       "initializing aws secret key failed");
+                ret = -1;
+                goto out;
+            }
+        }
+
+        if (dict_get_str(options, "s3plugin-keyid", &temp_str) == 0) {
+            priv->awskeyid = gf_strdup(temp_str);
+            if (!priv->awskeyid) {
+                gf_msg(this->name, GF_LOG_ERROR, ENOMEM, 0,
+                       "initializing aws key ID failed");
+                ret = -1;
+                goto out;
+            }
+        }
+
+        if (dict_get_str(options, "s3plugin-bucketid", &temp_str) == 0) {
+            priv->bucketid = gf_strdup(temp_str);
+            if (!priv->bucketid) {
+                gf_msg(this->name, GF_LOG_ERROR, ENOMEM, 0,
+                       "initializing aws bucketid failed");
+                ret = -1;
+                goto out;
+            }
+        }
+
+        if (dict_get_str(options, "s3plugin-hostname", &temp_str) == 0) {
+            priv->hostname = gf_strdup(temp_str);
+            if (!priv->hostname) {
+                gf_msg(this->name, GF_LOG_ERROR, ENOMEM, 0,
+                       "initializing aws hostname failed");
+                ret = -1;
+                goto out;
+            }
+        }
+    }
+out:
+    pthread_spin_unlock(&(priv->lock));
+
+    gf_msg_debug(this->name, 0,
+                 "stored key: %s id: %s "
+                 "bucketid %s hostname: %s",
+                 priv->awssekey, priv->awskeyid, priv->bucketid,
+                 priv->hostname);
+
+    return ret;
+}
+
+void
+aws_fini(void *config)
+{
+    aws_private_t *priv = NULL;
+
+    priv = (aws_private_t *)priv;
+
+    if (priv) {
+        GF_FREE(priv->hostname);
+        GF_FREE(priv->bucketid);
+        GF_FREE(priv->awssekey);
+        GF_FREE(priv->awskeyid);
+
+        pthread_spin_destroy(&priv->lock);
+        GF_FREE(priv);
+    }
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+
+    ret = xlator_mem_acct_init(this, gf_libaws_mt_end + 1);
+
+    if (ret != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "Memory accounting init failed");
+        return ret;
+    }
+out:
+    return ret;
+}
+char *
+aws_form_request(char *resource, char **date, char *reqtype, char *bucketid,
+                 char *filepath)
+{
+    char httpdate[256];
+    time_t ctime;
+    struct tm *gtime = NULL;
+    char *sign_req = NULL;
+    int signreq_len = -1;
+    int date_len = -1;
+    int res_len = -1;
+
+    ctime = gf_time();
+    gtime = gmtime(&ctime);
+
+    date_len = strftime(httpdate, sizeof(httpdate),
+                        "%a, %d %b %Y %H:%M:%S +0000", gtime);
+
+    *date = gf_strndup(httpdate, date_len);
+    if (*date == NULL) {
+        gf_msg("CS", GF_LOG_ERROR, ENOMEM, 0,
+               "memory allocation "
+               "failure for date");
+        goto out;
+    }
+
+    res_len = snprintf(resource, RESOURCE_SIZE, "%s/%s", bucketid, filepath);
+
+    gf_msg_debug("CS", 0, "resource %s", resource);
+
+    /* 6 accounts for the 4 new line chars, one forward slash and
+     * one null char */
+    signreq_len = res_len + date_len + strlen(reqtype) + 6;
+
+    sign_req = GF_MALLOC(signreq_len, gf_common_mt_char);
+    if (sign_req == NULL) {
+        gf_msg("CS", GF_LOG_ERROR, ENOMEM, 0,
+               "memory allocation "
+               "failure for sign_req");
+        goto out;
+    }
+
+    snprintf(sign_req, signreq_len, "%s\n\n%s\n%s\n/%s", reqtype, "", *date,
+             resource);
+
+out:
+    return sign_req;
+}
+
+char *
+aws_b64_encode(const unsigned char *input, int length)
+{
+    BIO *bio, *b64;
+    BUF_MEM *bptr;
+    char *buff = NULL;
+
+    b64 = BIO_new(BIO_f_base64());
+    bio = BIO_new(BIO_s_mem());
+    b64 = BIO_push(b64, bio);
+    BIO_write(b64, input, length);
+    BIO_flush(b64);
+    BIO_get_mem_ptr(b64, &bptr);
+
+    buff = GF_MALLOC(bptr->length, gf_common_mt_char);
+    memcpy(buff, bptr->data, bptr->length - 1);
+    buff[bptr->length - 1] = 0;
+
+    BIO_free_all(b64);
+
+    return buff;
+}
+
+char *
+aws_sign_request(char *const str, char *awssekey)
+{
+#if (OPENSSL_VERSION_NUMBER < 0x1010002f)
+    HMAC_CTX ctx;
+#endif
+    HMAC_CTX *pctx = NULL;
+    ;
+
+    unsigned char md[256];
+    unsigned len;
+    char *base64 = NULL;
+
+#if (OPENSSL_VERSION_NUMBER < 0x1010002f)
+    HMAC_CTX_init(&ctx);
+    pctx = &ctx;
+#else
+    pctx = HMAC_CTX_new();
+#endif
+    HMAC_Init_ex(pctx, awssekey, strlen(awssekey), EVP_sha1(), NULL);
+    HMAC_Update(pctx, (unsigned char *)str, strlen(str));
+    HMAC_Final(pctx, (unsigned char *)md, &len);
+
+#if (OPENSSL_VERSION_NUMBER < 0x1010002f)
+    HMAC_CTX_cleanup(pctx);
+#else
+    HMAC_CTX_free(pctx);
+#endif
+    base64 = aws_b64_encode(md, len);
+
+    return base64;
+}
+
+int
+aws_dlwritev_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+                 dict_t *xdata)
+{
+    aws_private_t *priv = NULL;
+
+    if (op_ret == -1) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, op_errno,
+               "write failed "
+               ". Aborting Download");
+
+        priv = this->private;
+        pthread_spin_lock(&(priv->lock));
+        {
+            priv->abortdl = _gf_true;
+        }
+        pthread_spin_unlock(&(priv->lock));
+    }
+
+    CS_STACK_DESTROY(frame);
+
+    return op_ret;
+}
+
+size_t
+aws_write_callback(void *dlbuf, size_t size, size_t nitems, void *mainframe)
+{
+    call_frame_t *frame = NULL;
+    fd_t *dlfd = NULL;
+    int ret = 0;
+    cs_local_t *local = NULL;
+    struct iovec iov = {
+        0,
+    };
+    struct iobref *iobref = NULL;
+    struct iobuf *iobuf = NULL;
+    struct iovec dliov = {
+        0,
+    };
+    size_t tsize = 0;
+    xlator_t *this = NULL;
+    cs_private_t *xl_priv = NULL;
+    aws_private_t *priv = NULL;
+    call_frame_t *dlframe = NULL;
+
+    frame = (call_frame_t *)mainframe;
+    this = frame->this;
+    xl_priv = this->private;
+    priv = xl_priv->stores->config;
+
+    pthread_spin_lock(&(priv->lock));
+    {
+        /* returning size other than the size passed from curl will
+         * abort further download*/
+        if (priv->abortdl) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, 0, "aborting download");
+            pthread_spin_unlock(&(priv->lock));
+            return 0;
+        }
+    }
+    pthread_spin_unlock(&(priv->lock));
+
+    local = frame->local;
+    dlfd = local->dlfd;
+    tsize = size * nitems;
+
+    dliov.iov_base = (void *)dlbuf;
+    dliov.iov_len = tsize;
+
+    ret = iobuf_copy(this->ctx->iobuf_pool, &dliov, 1, &iobref, &iobuf, &iov);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "iobuf_copy failed");
+        goto out;
+    }
+
+    /* copy frame */
+    dlframe = copy_frame(frame);
+    if (!dlframe) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "copy_frame failed");
+        tsize = 0;
+        goto out;
+    }
+
+    STACK_WIND(dlframe, aws_dlwritev_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->writev, dlfd, &iov, 1, local->dloffset,
+               0, iobref, NULL);
+
+    local->dloffset += tsize;
+
+out:
+    if (iobuf)
+        iobuf_unref(iobuf);
+    if (iobref)
+        iobref_unref(iobref);
+
+    return tsize;
+}
+
+int
+aws_download_s3(call_frame_t *frame, void *config)
+{
+    char *buf;
+    int bufsize = -1;
+    CURL *handle = NULL;
+    struct curl_slist *slist = NULL;
+    struct curl_slist *tmp = NULL;
+    xlator_t *this = NULL;
+    int ret = 0;
+    int debug = 1;
+    CURLcode res;
+    char errbuf[CURL_ERROR_SIZE];
+    size_t len = 0;
+    long responsecode;
+    char *sign_req = NULL;
+    char *date = NULL;
+    char *const reqtype = "GET";
+    char *signature = NULL;
+    cs_local_t *local = NULL;
+    char resource[RESOURCE_SIZE] = {
+        0,
+    };
+    aws_private_t *priv = NULL;
+
+    this = frame->this;
+
+    local = frame->local;
+
+    priv = (aws_private_t *)config;
+
+    if (!priv->bucketid || !priv->hostname || !priv->awssekey ||
+        !priv->awskeyid) {
+        ret = -1;
+        goto out;
+    }
+
+    sign_req = aws_form_request(resource, &date, reqtype, priv->bucketid,
+                                local->remotepath);
+    if (!sign_req) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "null sign_req, "
+               "aborting download");
+        ret = -1;
+        goto out;
+    }
+
+    gf_msg_debug("CS", 0, "sign_req %s date %s", sign_req, date);
+
+    signature = aws_sign_request(sign_req, priv->awssekey);
+    if (!signature) {
+        gf_msg("CS", GF_LOG_ERROR, 0, 0,
+               "null signature, "
+               "aborting download");
+        ret = -1;
+        goto out;
+    }
+
+    handle = curl_easy_init();
+    this = frame->this;
+
+    /* special numbers 6, 20, 10 accounts for static characters in the
+     * below snprintf string format arguments*/
+    bufsize = strlen(date) + 6 + strlen(priv->awskeyid) + strlen(signature) +
+              20 + strlen(priv->hostname) + 10;
+
+    buf = (char *)alloca(bufsize);
+    if (!buf) {
+        gf_msg("CS", GF_LOG_ERROR, ENOMEM, 0,
+               "mem allocation "
+               "failed for buf");
+        ret = -1;
+        goto out;
+    }
+
+    snprintf(buf, bufsize, "Date: %s", date);
+    slist = curl_slist_append(slist, buf);
+    snprintf(buf, bufsize, "Authorization: AWS %s:%s", priv->awskeyid,
+             signature);
+    slist = curl_slist_append(slist, buf);
+    snprintf(buf, bufsize, "https://%s/%s", priv->hostname, resource);
+
+    if (gf_log_get_loglevel() >= GF_LOG_DEBUG) {
+        tmp = slist;
+        while (tmp) {
+            gf_msg_debug(this->name, 0, "slist for curl - %s", tmp->data);
+            tmp = tmp->next;
+        }
+    }
+
+    curl_easy_setopt(handle, CURLOPT_HTTPHEADER, slist);
+    curl_easy_setopt(handle, CURLOPT_URL, buf);
+    curl_easy_setopt(handle, CURLOPT_WRITEFUNCTION, aws_write_callback);
+    curl_easy_setopt(handle, CURLOPT_WRITEDATA, frame);
+    curl_easy_setopt(handle, CURLOPT_VERBOSE, debug);
+    curl_easy_setopt(handle, CURLOPT_ERRORBUFFER, errbuf);
+
+    res = curl_easy_perform(handle);
+    if (res != CURLE_OK) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "download failed. err: %s\n",
+               curl_easy_strerror(res));
+        ret = -1;
+        len = strlen(errbuf);
+        if (len) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, 0, "curl failure %s", errbuf);
+        } else {
+            gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+                   "curl error "
+                   "%s\n",
+                   curl_easy_strerror(res));
+        }
+    }
+
+    if (res == CURLE_OK) {
+        curl_easy_getinfo(handle, CURLINFO_RESPONSE_CODE, &responsecode);
+        gf_msg_debug(this->name, 0, "response code %ld", responsecode);
+        if (responsecode != 200) {
+            ret = -1;
+            gf_msg(this->name, GF_LOG_ERROR, 0, 0, "curl download failed");
+        }
+    }
+
+    curl_slist_free_all(slist);
+    curl_easy_cleanup(handle);
+
+out:
+    if (sign_req)
+        GF_FREE(sign_req);
+    if (date)
+        GF_FREE(date);
+    if (signature)
+        GF_FREE(signature);
+
+    return ret;
+}
+
+struct volume_options cs_options[] = {
+    {.key = {"s3plugin-seckey"},
+     .type = GF_OPTION_TYPE_STR,
+     .description = "aws secret key"},
+    {.key = {"s3plugin-keyid"},
+     .type = GF_OPTION_TYPE_STR,
+     .description = "aws key ID"
+
+    },
+    {.key = {"s3plugin-bucketid"},
+     .type = GF_OPTION_TYPE_STR,
+     .description = "aws bucketid"},
+    {.key = {"s3plugin-hostname"},
+     .type = GF_OPTION_TYPE_STR,
+     .description = "aws hostname e.g. s3.amazonaws.com"},
+    {.key = {NULL}},
+};
diff --git a/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/src/libcloudsyncs3.h b/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/src/libcloudsyncs3.h
new file mode 100644
index 00000000000..85ae669486b
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/src/libcloudsyncs3.h
@@ -0,0 +1,50 @@
+/*
+  Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+#ifndef _LIBAWS_H
+#define _LIBAWS_H
+
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/call-stub.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/syncop.h>
+#include <curl/curl.h>
+#include "cloudsync-common.h"
+#include "libcloudsyncs3-mem-types.h"
+
+char *
+aws_b64_encode(const unsigned char *input, int length);
+
+size_t
+aws_write_callback(void *dlbuf, size_t size, size_t nitems, void *mainframe);
+
+int
+aws_download_s3(call_frame_t *frame, void *config);
+
+int
+aws_dlwritev_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+                 dict_t *xdata);
+
+void *
+aws_init(xlator_t *this);
+
+int
+aws_reconfigure(xlator_t *this, dict_t *options);
+
+char *
+aws_form_request(char *resource, char **date, char *reqtype, char *bucketid,
+                 char *filepath);
+char *
+aws_sign_request(char *const str, char *awssekey);
+
+void
+aws_fini(void *config);
+
+#endif
diff --git a/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/src/libcloudsyncs3.sym b/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/src/libcloudsyncs3.sym
new file mode 100644
index 00000000000..0bc273670d5
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-plugins/src/cloudsyncs3/src/libcloudsyncs3.sym
@@ -0,0 +1 @@
+store_ops
diff --git a/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/Makefile.am b/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/Makefile.am
new file mode 100644
index 00000000000..a985f42a877
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/Makefile.am
@@ -0,0 +1,3 @@
+SUBDIRS = src
+
+CLEANFILES =
diff --git a/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/Makefile.am b/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/Makefile.am
new file mode 100644
index 00000000000..b512464f157
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/Makefile.am
@@ -0,0 +1,12 @@
+csp_LTLIBRARIES = cloudsynccvlt.la
+cspdir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/cloudsync-plugins
+
+cloudsynccvlt_la_SOURCES = libcvlt.c  $(top_srcdir)/xlators/features/cloudsync/src/cloudsync-common.c
+cloudsynccvlt_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+cloudsynccvlt_la_LDFLAGS = -module -avoid-version -export-symbols $(top_srcdir)/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/libcloudsynccvlt.sym
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src   -I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+noinst_HEADERS = archivestore.h libcvlt.h libcvlt-mem-types.h cvlt-messages.h
+AM_CFLAGS = -Wall -fno-strict-aliasing $(GF_CFLAGS) -I$(top_srcdir)/xlators/features/cloudsync/src
+CLEANFILES =
+
+EXTRA_DIST = libcloudsynccvlt.sym
diff --git a/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/archivestore.h b/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/archivestore.h
new file mode 100644
index 00000000000..7230ef77337
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/archivestore.h
@@ -0,0 +1,203 @@
+/*
+  Copyright (c) 2018 Commvault Systems, Inc. <http://www.commvault.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __ARCHIVESTORE_H__
+#define __ARCHIVESTORE_H__
+
+#include <stdlib.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <dlfcn.h>
+#include <uuid/uuid.h>
+
+#define CS_XATTR_ARCHIVE_UUID "trusted.cloudsync.uuid"
+#define CS_XATTR_PRODUCT_ID "trusted.cloudsync.product-id"
+#define CS_XATTR_STORE_ID "trusted.cloudsync.store-id"
+
+struct _archstore_methods;
+typedef struct _archstore_methods archstore_methods_t;
+
+struct _archstore_desc {
+    void *priv; /* Private field for store mgmt.   */
+                /* To be used only by archive store*/
+};
+typedef struct _archstore_desc archstore_desc_t;
+
+struct _archstore_info {
+    char *id;         /* Identifier for the archivestore */
+    uint32_t idlen;   /* Length of identifier string     */
+    char *prod;       /* Name of the data mgmt. product  */
+    uint32_t prodlen; /* Length of the product string    */
+};
+typedef struct _archstore_info archstore_info_t;
+
+struct _archstore_fileinfo {
+    uuid_t uuid;         /* uuid of the file                */
+    char *path;          /* file path                       */
+    uint32_t pathlength; /* length of file path             */
+};
+typedef struct _archstore_fileinfo archstore_fileinfo_t;
+
+struct _app_callback_info {
+    archstore_info_t *src_archstore;
+    archstore_fileinfo_t *src_archfile;
+    archstore_info_t *dest_archstore;
+    archstore_fileinfo_t *dest_archfile;
+};
+typedef struct _app_callback_info app_callback_info_t;
+
+typedef void (*app_callback_t)(archstore_desc_t *, app_callback_info_t *,
+                               void *, int64_t, int32_t);
+
+enum _archstore_scan_type { FULL = 1, INCREMENTAL = 2 };
+typedef enum _archstore_scan_type archstore_scan_type_t;
+
+typedef int32_t archstore_errno_t;
+
+/*
+ * Initialize archive store.
+ * arg1  pointer to structure containing archive store information
+ * arg2  error number if any generated during the initialization
+ * arg3  name of the log file
+ */
+typedef int32_t (*init_archstore_t)(archstore_desc_t *, archstore_errno_t *,
+                                    const char *);
+
+/*
+ * Clean up archive store.
+ * arg1  pointer to structure containing archive store information
+ * arg2  error number if any generated during the cleanup
+ */
+typedef int32_t (*term_archstore_t)(archstore_desc_t *, archstore_errno_t *);
+
+/*
+ * Read the contents of the file from archive store
+ * arg1  pointer to structure containing archive store description
+ * arg2  pointer to structure containing archive store information
+ * arg3  pointer to structure containing information about file to be read
+ * arg4  offset in the file from which data should be read
+ * arg5  buffer where the data should be read
+ * arg6  number of bytes of data to be read
+ * arg7  error number if any generated during the read from file
+ * arg8  callback handler to be invoked after the data is read
+ * arg9  cookie to be passed when callback is invoked
+ */
+typedef int32_t (*read_archstore_t)(archstore_desc_t *, archstore_info_t *,
+                                    archstore_fileinfo_t *, off_t, char *,
+                                    size_t, archstore_errno_t *, app_callback_t,
+                                    void *);
+
+/*
+ * Restore the contents of the file from archive store
+ * This is basically in-place restore
+ * arg1  pointer to structure containing archive store description
+ * arg2  pointer to structure containing archive store information
+ * arg3  pointer to structure containing information about file to be restored
+ * arg4  error number if any generated during the file restore
+ * arg5  callback to be invoked after the file is restored
+ * arg6  cookie to be passed when callback is invoked
+ */
+typedef int32_t (*recall_archstore_t)(archstore_desc_t *, archstore_info_t *,
+                                      archstore_fileinfo_t *,
+                                      archstore_errno_t *, app_callback_t,
+                                      void *);
+
+/*
+ * Restore the contents of the file from archive store to a different store
+ * This is basically out-of-place restore
+ * arg1  pointer to structure containing archive store description
+ * arg2  pointer to structure containing source archive store information
+ * arg3  pointer to structure containing information about file to be restored
+ * arg4  pointer to structure containing destination archive store information
+ * arg5  pointer to structure containing information about the location to
+         which the file will be restored
+ * arg6  error number if any generated during the file restore
+ * arg7  callback to be invoked after the file is restored
+ * arg8  cookie to be passed when callback is invoked
+ */
+typedef int32_t (*restore_archstore_t)(archstore_desc_t *, archstore_info_t *,
+                                       archstore_fileinfo_t *,
+                                       archstore_info_t *,
+                                       archstore_fileinfo_t *,
+                                       archstore_errno_t *, app_callback_t,
+                                       void *);
+
+/*
+ * Archive the contents of the file to archive store
+ * arg1  pointer to structure containing archive store description
+ * arg2  pointer to structure containing source archive store information
+ * arg3  pointer to structure containing information about files to be archived
+ * arg4  pointer to structure containing destination archive store information
+ * arg5  pointer to structure containing information about files that failed
+ *       to be archived
+ * arg6  error number if any generated during the file archival
+ * arg7  callback to be invoked after the file is archived
+ * arg8  cookie to be passed when callback is invoked
+ */
+typedef int32_t (*archive_archstore_t)(archstore_desc_t *, archstore_info_t *,
+                                       archstore_fileinfo_t *,
+                                       archstore_info_t *,
+                                       archstore_fileinfo_t *,
+                                       archstore_errno_t *, app_callback_t,
+                                       void *);
+
+/*
+ * Backup list of files provided in the input file
+ * arg1  pointer to structure containing archive store description
+ * arg2  pointer to structure containing source archive store information
+ * arg3  pointer to structure containing information about files to be backed up
+ * arg4  pointer to structure containing destination archive store information
+ * arg5  pointer to structure containing information about files that failed
+ *       to be backed up
+ * arg6  error number if any generated during the file archival
+ * arg7  callback to be invoked after the file is archived
+ * arg8  cookie to be passed when callback is invoked
+ */
+typedef int32_t (*backup_archstore_t)(archstore_desc_t *, archstore_info_t *,
+                                      archstore_fileinfo_t *,
+                                      archstore_info_t *,
+                                      archstore_fileinfo_t *,
+                                      archstore_errno_t *, app_callback_t,
+                                      void *);
+
+/*
+ * Scan the contents of a store and determine the files which need to be
+ * backed up.
+ * arg1  pointer to structure containing archive store description
+ * arg2  pointer to structure containing archive store information
+ * arg3  type of scan whether full or incremental
+ * arg4  path to file that contains list of files to be backed up
+ * arg5  error number if any generated during scan operation
+ */
+typedef int32_t (*scan_archstore_t)(archstore_desc_t *, archstore_info_t *,
+                                    archstore_scan_type_t, char *,
+                                    archstore_errno_t *);
+
+struct _archstore_methods {
+    init_archstore_t init;
+    term_archstore_t fini;
+    backup_archstore_t backup;
+    archive_archstore_t archive;
+    scan_archstore_t scan;
+    restore_archstore_t restore;
+    recall_archstore_t recall;
+    read_archstore_t read;
+};
+
+typedef int (*get_archstore_methods_t)(archstore_methods_t *);
+
+/*
+ * Single function that will be invoked by applications for extracting
+ * the function pointers to all data management functions.
+ */
+int32_t
+get_archstore_methods(archstore_methods_t *);
+
+#endif /* End of __ARCHIVESTORE_H__ */
diff --git a/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/cvlt-messages.h b/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/cvlt-messages.h
new file mode 100644
index 00000000000..57c9aa77da0
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/cvlt-messages.h
@@ -0,0 +1,30 @@
+/*
+ Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+ This file is part of GlusterFS.
+
+ This file is licensed to you under your choice of the GNU Lesser
+ General Public License, version 3 or any later version (LGPLv3 or
+ later), or the GNU General Public License, version 2 (GPLv2), in all
+ cases as published by the Free Software Foundation.
+ */
+
+#ifndef _CVLT_MESSAGES_H_
+#define _CVLT_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(CVLT, CVLT_EXTRACTION_FAILED, CVLT_FREE,
+           CVLT_RESOURCE_ALLOCATION_FAILED, CVLT_RESTORE_FAILED,
+           CVLT_READ_FAILED, CVLT_NO_MEMORY, CVLT_DLOPEN_FAILED);
+
+#endif /* !_CVLT_MESSAGES_H_ */
diff --git a/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/libcloudsynccvlt.sym b/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/libcloudsynccvlt.sym
new file mode 100644
index 00000000000..0bc273670d5
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/libcloudsynccvlt.sym
@@ -0,0 +1 @@
+store_ops
diff --git a/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/libcvlt-mem-types.h b/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/libcvlt-mem-types.h
new file mode 100644
index 00000000000..c24fab8bfe7
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/libcvlt-mem-types.h
@@ -0,0 +1,19 @@
+/*
+ *   Copyright (c) 2018 Commvault Systems, Inc. <http://www.commvault.com>
+ *   This file is part of GlusterFS.
+ *
+ *   This file is licensed to you under your choice of the GNU Lesser
+ *   General Public License, version 3 or any later version (LGPLv3 or
+ *   later), or the GNU General Public License, version 2 (GPLv2), in all
+ *   cases as published by the Free Software Foundation.
+ */
+
+#ifndef __LIBCVLT_MEM_TYPES_H__
+#define __LIBCVLT_MEM_TYPES_H__
+
+#include <glusterfs/mem-types.h>
+enum libcvlt_mem_types_ {
+    gf_libcvlt_mt_cvlt_private_t = gf_common_mt_end + 1,
+    gf_libcvlt_mt_end
+};
+#endif /* __LIBCVLT_MEM_TYPES_H__ */
diff --git a/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/libcvlt.c b/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/libcvlt.c
new file mode 100644
index 00000000000..5b7272bb448
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/libcvlt.c
@@ -0,0 +1,842 @@
+#include <stdlib.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/glusterfs.h>
+#include "libcvlt.h"
+#include "cloudsync-common.h"
+#include "cvlt-messages.h"
+
+#define LIBARCHIVE_SO "libopenarchive.so"
+#define ALIGN_SIZE 4096
+#define CVLT_TRAILER "cvltv1"
+
+store_methods_t store_ops = {
+    .fop_download = cvlt_download,
+    .fop_init = cvlt_init,
+    .fop_reconfigure = cvlt_reconfigure,
+    .fop_fini = cvlt_fini,
+    .fop_remote_read = cvlt_read,
+};
+
+static const int32_t num_req = 32;
+static const int32_t num_iatt = 32;
+static char *plugin = "cvlt_cloudSync";
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    if (!this)
+        return ret;
+
+    ret = xlator_mem_acct_init(this, gf_libcvlt_mt_end + 1);
+
+    if (ret != 0) {
+        return ret;
+    }
+
+    return ret;
+}
+
+static void
+cvlt_free_resources(archive_t *arch)
+{
+    /*
+     * We will release all the resources that were allocated by the xlator.
+     * Check whether there are any buffers which have not been released
+     * back to a mempool.
+     */
+
+    if (arch->handle) {
+        dlclose(arch->handle);
+    }
+
+    if (arch->iobuf_pool) {
+        iobuf_pool_destroy(arch->iobuf_pool);
+    }
+
+    if (arch->req_pool) {
+        mem_pool_destroy(arch->req_pool);
+        arch->req_pool = NULL;
+    }
+
+    return;
+}
+
+static int32_t
+cvlt_extract_store_fops(xlator_t *this, archive_t *arch)
+{
+    int32_t op_ret = -1;
+    get_archstore_methods_t get_archstore_methods;
+
+    /*
+     * libopenarchive.so defines methods for performing data management
+     * operations. We will extract the methods from library and these
+     * methods will be invoked for moving data between glusterfs volume
+     * and the data management product.
+     */
+
+    VALIDATE_OR_GOTO(arch, err);
+
+    arch->handle = dlopen(LIBARCHIVE_SO, RTLD_NOW);
+    if (!arch->handle) {
+        gf_msg(plugin, GF_LOG_ERROR, 0, CVLT_DLOPEN_FAILED,
+               " failed to open %s ", LIBARCHIVE_SO);
+        return op_ret;
+    }
+
+    dlerror(); /* Clear any existing error */
+
+    get_archstore_methods = dlsym(arch->handle, "get_archstore_methods");
+    if (!get_archstore_methods) {
+        gf_msg(plugin, GF_LOG_ERROR, 0, CVLT_EXTRACTION_FAILED,
+               " Error extracting get_archstore_methods()");
+        dlclose(arch->handle);
+        arch->handle = NULL;
+        return op_ret;
+    }
+
+    op_ret = get_archstore_methods(&(arch->fops));
+    if (op_ret) {
+        gf_msg(plugin, GF_LOG_ERROR, 0, CVLT_EXTRACTION_FAILED,
+               " Failed to extract methods in get_archstore_methods");
+        dlclose(arch->handle);
+        arch->handle = NULL;
+        return op_ret;
+    }
+
+err:
+    return op_ret;
+}
+
+static int32_t
+cvlt_alloc_resources(xlator_t *this, archive_t *arch, int num_req, int num_iatt)
+{
+    /*
+     * Initialize information about all the memory pools that will be
+     * used by this xlator.
+     */
+    arch->nreqs = 0;
+
+    arch->req_pool = NULL;
+
+    arch->handle = NULL;
+    arch->xl = this;
+
+    arch->req_pool = mem_pool_new(cvlt_request_t, num_req);
+    if (!arch->req_pool) {
+        goto err;
+    }
+
+    arch->iobuf_pool = iobuf_pool_new();
+    if (!arch->iobuf_pool) {
+        goto err;
+    }
+
+    if (cvlt_extract_store_fops(this, arch)) {
+        goto err;
+    }
+
+    return 0;
+
+err:
+
+    return -1;
+}
+
+static void
+cvlt_req_init(cvlt_request_t *req)
+{
+    sem_init(&(req->sem), 0, 0);
+
+    return;
+}
+
+static void
+cvlt_req_destroy(cvlt_request_t *req)
+{
+    if (req->iobuf) {
+        iobuf_unref(req->iobuf);
+    }
+
+    if (req->iobref) {
+        iobref_unref(req->iobref);
+    }
+
+    sem_destroy(&(req->sem));
+
+    return;
+}
+
+static cvlt_request_t *
+cvlt_alloc_req(archive_t *arch)
+{
+    cvlt_request_t *reqptr = NULL;
+
+    if (!arch) {
+        goto err;
+    }
+
+    if (arch->req_pool) {
+        reqptr = mem_get0(arch->req_pool);
+        if (reqptr) {
+            cvlt_req_init(reqptr);
+        }
+    }
+
+    if (reqptr) {
+        LOCK(&(arch->lock));
+        arch->nreqs++;
+        UNLOCK(&(arch->lock));
+    }
+
+err:
+    return reqptr;
+}
+
+static int32_t
+cvlt_free_req(archive_t *arch, cvlt_request_t *reqptr)
+{
+    if (!reqptr) {
+        goto err;
+    }
+
+    if (!arch) {
+        goto err;
+    }
+
+    if (arch->req_pool) {
+        /*
+         * Free the request resources if they exist.
+         */
+
+        cvlt_req_destroy(reqptr);
+        mem_put(reqptr);
+
+        LOCK(&(arch->lock));
+        arch->nreqs--;
+        UNLOCK(&(arch->lock));
+    }
+
+    return 0;
+
+err:
+    return -1;
+}
+
+static int32_t
+cvlt_init_xlator(xlator_t *this, archive_t *arch, int num_req, int num_iatt)
+{
+    int32_t ret = -1;
+    int32_t errnum = -1;
+    int32_t locked = 0;
+
+    /*
+     * Perform all the initializations needed for brining up the xlator.
+     */
+    if (!arch) {
+        goto err;
+    }
+
+    LOCK_INIT(&(arch->lock));
+    LOCK(&(arch->lock));
+
+    locked = 1;
+
+    ret = cvlt_alloc_resources(this, arch, num_req, num_iatt);
+
+    if (ret) {
+        goto err;
+    }
+
+    /*
+     * Now that the fops have been extracted initialize the store
+     */
+    ret = arch->fops.init(&(arch->descinfo), &errnum, plugin);
+    if (ret) {
+        goto err;
+    }
+
+    UNLOCK(&(arch->lock));
+    locked = 0;
+    ret = 0;
+
+    return ret;
+
+err:
+    if (arch) {
+        cvlt_free_resources(arch);
+
+        if (locked) {
+            UNLOCK(&(arch->lock));
+        }
+    }
+
+    return ret;
+}
+
+static int32_t
+cvlt_term_xlator(archive_t *arch)
+{
+    int32_t errnum = -1;
+
+    if (!arch) {
+        goto err;
+    }
+
+    LOCK(&(arch->lock));
+
+    /*
+     * Release the resources that have been allocated inside store
+     */
+    arch->fops.fini(&(arch->descinfo), &errnum);
+
+    cvlt_free_resources(arch);
+
+    UNLOCK(&(arch->lock));
+
+    GF_FREE(arch);
+
+    return 0;
+
+err:
+    return -1;
+}
+
+static int32_t
+cvlt_init_store_info(archive_t *priv, archstore_info_t *store_info)
+{
+    if (!store_info) {
+        return -1;
+    }
+
+    store_info->prod = priv->product_id;
+    store_info->prodlen = strlen(priv->product_id);
+
+    store_info->id = priv->store_id;
+    store_info->idlen = strlen(priv->store_id);
+
+    return 0;
+}
+
+static int32_t
+cvlt_init_file_info(cs_loc_xattr_t *xattr, archstore_fileinfo_t *file_info)
+{
+    if (!xattr || !file_info) {
+        return -1;
+    }
+
+    gf_uuid_copy(file_info->uuid, xattr->uuid);
+    file_info->path = xattr->file_path;
+    file_info->pathlength = strlen(xattr->file_path);
+
+    return 0;
+}
+
+static int32_t
+cvlt_init_gluster_store_info(cs_loc_xattr_t *xattr,
+                             archstore_info_t *store_info)
+{
+    static char *product = "glusterfs";
+
+    if (!xattr || !store_info) {
+        return -1;
+    }
+
+    store_info->prod = product;
+    store_info->prodlen = strlen(product);
+
+    store_info->id = xattr->volname;
+    store_info->idlen = strlen(xattr->volname);
+
+    return 0;
+}
+
+static int32_t
+cvlt_init_gluster_file_info(cs_loc_xattr_t *xattr,
+                            archstore_fileinfo_t *file_info)
+{
+    if (!xattr || !file_info) {
+        return -1;
+    }
+
+    gf_uuid_copy(file_info->uuid, xattr->gfid);
+    file_info->path = xattr->file_path;
+    file_info->pathlength = strlen(xattr->file_path);
+
+    return 0;
+}
+
+static void
+cvlt_copy_stat_info(struct iatt *buf, cs_size_xattr_t *xattrs)
+{
+    /*
+     * If the file was archived then the reported size will not be a
+     * correct one. We need to fix this.
+     */
+    if (buf && xattrs) {
+        buf->ia_size = xattrs->size;
+        buf->ia_blksize = xattrs->blksize;
+        buf->ia_blocks = xattrs->blocks;
+    }
+
+    return;
+}
+
+static void
+cvlt_readv_complete(archstore_desc_t *desc, app_callback_info_t *cbkinfo,
+                    void *cookie, int64_t op_ret, int32_t op_errno)
+{
+    struct iovec iov;
+    xlator_t *this = NULL;
+    struct iatt postbuf = {
+        0,
+    };
+    call_frame_t *frame = NULL;
+    cvlt_request_t *req = (cvlt_request_t *)cookie;
+    cs_local_t *local = NULL;
+    cs_private_t *cspriv = NULL;
+    archive_t *priv = NULL;
+
+    frame = req->frame;
+    this = frame->this;
+    local = frame->local;
+
+    cspriv = this->private;
+    priv = (archive_t *)cspriv->stores->config;
+
+    if (strcmp(priv->trailer, CVLT_TRAILER)) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    gf_msg_debug(plugin, 0,
+                 " Read callback invoked offset:%" PRIu64 "bytes: %" PRIu64
+                 " op : %d ret : %" PRId64 " errno : %d",
+                 req->offset, req->bytes, req->op_type, op_ret, op_errno);
+
+    if (op_ret < 0) {
+        goto out;
+    }
+
+    req->iobref = iobref_new();
+    if (!req->iobref) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    iobref_add(req->iobref, req->iobuf);
+    iov.iov_base = iobuf_ptr(req->iobuf);
+    iov.iov_len = op_ret;
+
+    cvlt_copy_stat_info(&postbuf, &(req->szxattr));
+
+    /*
+     * Hack to notify higher layers of EOF.
+     */
+    if (!postbuf.ia_size || (req->offset + iov.iov_len >= postbuf.ia_size)) {
+        gf_msg_debug(plugin, 0, " signalling end-of-file for uuid=%s",
+                     uuid_utoa(req->file_info.uuid));
+        op_errno = ENOENT;
+    }
+
+out:
+
+    STACK_UNWIND_STRICT(readv, frame, op_ret, op_errno, &iov, 1, &postbuf,
+                        req->iobref, local->xattr_rsp);
+
+    cvlt_free_req(priv, req);
+
+    return;
+}
+
+static void
+cvlt_download_complete(archstore_desc_t *store, app_callback_info_t *cbk_info,
+                       void *cookie, int64_t ret, int errcode)
+{
+    cvlt_request_t *req = (cvlt_request_t *)cookie;
+
+    gf_msg_debug(plugin, 0,
+                 " Download callback invoked  ret : %" PRId64 " errno : %d",
+                 ret, errcode);
+
+    req->op_ret = ret;
+    req->op_errno = errcode;
+    sem_post(&(req->sem));
+
+    return;
+}
+
+void *
+cvlt_init(xlator_t *this)
+{
+    int ret = 0;
+    archive_t *priv = NULL;
+
+    if (!this->children || this->children->next) {
+        gf_msg(plugin, GF_LOG_ERROR, ENOMEM, 0,
+               "should have exactly one child");
+        ret = -1;
+        goto out;
+    }
+
+    if (!this->parents) {
+        gf_msg(plugin, GF_LOG_ERROR, ENOMEM, 0,
+               "dangling volume. check volfile");
+        ret = -1;
+        goto out;
+    }
+
+    priv = GF_CALLOC(1, sizeof(archive_t), gf_libcvlt_mt_cvlt_private_t);
+    if (!priv) {
+        ret = -1;
+        goto out;
+    }
+
+    priv->trailer = CVLT_TRAILER;
+    if (cvlt_init_xlator(this, priv, num_req, num_iatt)) {
+        gf_msg(plugin, GF_LOG_ERROR, ENOMEM, 0, "xlator init failed");
+        ret = -1;
+        goto out;
+    }
+
+    GF_OPTION_INIT("cloudsync-store-id", priv->store_id, str, out);
+    GF_OPTION_INIT("cloudsync-product-id", priv->product_id, str, out);
+
+    gf_msg(plugin, GF_LOG_INFO, 0, 0,
+           "store id is : %s "
+           "product id is : %s.",
+           priv->store_id, priv->product_id);
+out:
+    if (ret == -1) {
+        cvlt_term_xlator(priv);
+        return (NULL);
+    }
+    return priv;
+}
+
+int
+cvlt_reconfigure(xlator_t *this, dict_t *options)
+{
+    cs_private_t *cspriv = NULL;
+    archive_t *priv = NULL;
+
+    cspriv = this->private;
+    priv = (archive_t *)cspriv->stores->config;
+
+    if (strcmp(priv->trailer, CVLT_TRAILER))
+        goto out;
+
+    GF_OPTION_RECONF("cloudsync-store-id", priv->store_id, options, str, out);
+
+    GF_OPTION_RECONF("cloudsync-product-id", priv->product_id, options, str,
+                     out);
+    gf_msg_debug(plugin, 0,
+                 "store id is : %s "
+                 "product id is : %s.",
+                 priv->store_id, priv->product_id);
+    return 0;
+out:
+    return -1;
+}
+
+void
+cvlt_fini(void *config)
+{
+    archive_t *priv = NULL;
+
+    priv = (archive_t *)config;
+
+    if (strcmp(priv->trailer, CVLT_TRAILER))
+        return;
+
+    cvlt_term_xlator(priv);
+    gf_msg(plugin, GF_LOG_INFO, 0, CVLT_FREE, " released xlator resources");
+    return;
+}
+
+int
+cvlt_download(call_frame_t *frame, void *config)
+{
+    archive_t *parch = NULL;
+    cs_local_t *local = frame->local;
+    cs_loc_xattr_t *locxattr = local->xattrinfo.lxattr;
+    cvlt_request_t *req = NULL;
+    archstore_info_t dest_storeinfo;
+    archstore_fileinfo_t dest_fileinfo;
+    int32_t op_ret, op_errno;
+
+    parch = (archive_t *)config;
+
+    if (strcmp(parch->trailer, CVLT_TRAILER)) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    gf_msg_debug(plugin, 0, " download invoked for uuid = %s  gfid=%s ",
+                 locxattr->uuid, uuid_utoa(locxattr->gfid));
+
+    if (!(parch->fops.restore)) {
+        op_errno = ELIBBAD;
+        goto err;
+    }
+
+    /*
+     * Download needs to be processed. Allocate a request.
+     */
+    req = cvlt_alloc_req(parch);
+
+    if (!req) {
+        gf_msg(plugin, GF_LOG_ERROR, ENOMEM, CVLT_RESOURCE_ALLOCATION_FAILED,
+               " failed to allocated request for gfid=%s",
+               uuid_utoa(locxattr->gfid));
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    /*
+     * Initialize the request object.
+     */
+    req->op_type = CVLT_RESTORE_OP;
+    req->frame = frame;
+
+    /*
+     * The file is currently residing inside a data management store.
+     * To restore the file contents we need to provide the information
+     * about data management store.
+     */
+    op_ret = cvlt_init_store_info(parch, &(req->store_info));
+    if (op_ret < 0) {
+        gf_msg(plugin, GF_LOG_ERROR, 0, CVLT_EXTRACTION_FAILED,
+               " failed to extract store info for gfid=%s",
+               uuid_utoa(locxattr->gfid));
+        goto err;
+    }
+
+    op_ret = cvlt_init_file_info(locxattr, &(req->file_info));
+    if (op_ret < 0) {
+        gf_msg(plugin, GF_LOG_ERROR, 0, CVLT_EXTRACTION_FAILED,
+               " failed to extract file info for gfid=%s",
+               uuid_utoa(locxattr->gfid));
+        goto err;
+    }
+
+    /*
+     * We need to perform in-place restore of the file from data management
+     * store to gusterfs volume.
+     */
+    op_ret = cvlt_init_gluster_store_info(locxattr, &dest_storeinfo);
+    if (op_ret < 0) {
+        gf_msg(plugin, GF_LOG_ERROR, 0, CVLT_EXTRACTION_FAILED,
+               " failed to extract destination store info for gfid=%s",
+               uuid_utoa(locxattr->gfid));
+        goto err;
+    }
+
+    op_ret = cvlt_init_gluster_file_info(locxattr, &dest_fileinfo);
+    if (op_ret < 0) {
+        gf_msg(plugin, GF_LOG_ERROR, 0, CVLT_EXTRACTION_FAILED,
+               " failed to extract file info for gfid=%s",
+               uuid_utoa(locxattr->gfid));
+        goto err;
+    }
+
+    /*
+     * Submit the restore request.
+     */
+    op_ret = parch->fops.restore(&(parch->descinfo), &(req->store_info),
+                                 &(req->file_info), &dest_storeinfo,
+                                 &dest_fileinfo, &op_errno,
+                                 cvlt_download_complete, req);
+    if (op_ret < 0) {
+        gf_msg(plugin, GF_LOG_ERROR, 0, CVLT_RESTORE_FAILED,
+               " failed to restore file gfid=%s from data management store",
+               uuid_utoa(locxattr->gfid));
+        goto err;
+    }
+
+    /*
+     * Wait for the restore to complete.
+     */
+    sem_wait(&(req->sem));
+
+    if (req->op_ret < 0) {
+        gf_msg(plugin, GF_LOG_ERROR, 0, CVLT_RESTORE_FAILED,
+               " restored failed for gfid=%s", uuid_utoa(locxattr->gfid));
+        goto err;
+    }
+
+    if (req) {
+        cvlt_free_req(parch, req);
+    }
+
+    return 0;
+
+err:
+
+    if (req) {
+        cvlt_free_req(parch, req);
+    }
+
+    return -1;
+}
+
+int
+cvlt_read(call_frame_t *frame, void *config)
+{
+    int32_t op_ret = -1;
+    int32_t op_errno = 0;
+    archive_t *parch = NULL;
+    cvlt_request_t *req = NULL;
+    struct iovec iov = {
+        0,
+    };
+    struct iobref *iobref;
+    size_t size = 0;
+    off_t off = 0;
+
+    cs_local_t *local = frame->local;
+    cs_loc_xattr_t *locxattr = local->xattrinfo.lxattr;
+
+    size = local->xattrinfo.size;
+    off = local->xattrinfo.offset;
+
+    parch = (archive_t *)config;
+
+    if (strcmp(parch->trailer, CVLT_TRAILER)) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    gf_msg_debug(plugin, 0,
+                 " read invoked for gfid = %s offset = %" PRIu64
+                 " file_size = %" PRIu64,
+                 uuid_utoa(locxattr->gfid), off, local->stbuf.ia_size);
+
+    if (off >= local->stbuf.ia_size) {
+        /*
+         * Hack to notify higher layers of EOF.
+         */
+
+        op_errno = ENOENT;
+        op_ret = 0;
+
+        gf_msg(plugin, GF_LOG_ERROR, 0, CVLT_READ_FAILED,
+               " reporting end-of-file for gfid=%s", uuid_utoa(locxattr->gfid));
+
+        goto err;
+    }
+
+    if (!size) {
+        op_errno = EINVAL;
+
+        gf_msg(plugin, GF_LOG_ERROR, 0, CVLT_READ_FAILED,
+               " zero size read attempted on gfid=%s",
+               uuid_utoa(locxattr->gfid));
+        goto err;
+    }
+
+    if (!(parch->fops.read)) {
+        op_errno = ELIBBAD;
+        goto err;
+    }
+
+    /*
+     * The read request need to be processed. Allocate a request.
+     */
+    req = cvlt_alloc_req(parch);
+
+    if (!req) {
+        gf_msg(plugin, GF_LOG_ERROR, ENOMEM, CVLT_NO_MEMORY,
+               " failed to allocated request for gfid=%s",
+               uuid_utoa(locxattr->gfid));
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    req->iobuf = iobuf_get_page_aligned(parch->iobuf_pool, size, ALIGN_SIZE);
+    if (!req->iobuf) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    /*
+     * Initialize the request object.
+     */
+    req->op_type = CVLT_READ_OP;
+    req->offset = off;
+    req->bytes = size;
+    req->frame = frame;
+    req->szxattr.size = local->stbuf.ia_size;
+    req->szxattr.blocks = local->stbuf.ia_blocks;
+    req->szxattr.blksize = local->stbuf.ia_blksize;
+
+    /*
+     * The file is currently residing inside a data management store.
+     * To read the file contents we need to provide the information
+     * about data management store.
+     */
+    op_ret = cvlt_init_store_info(parch, &(req->store_info));
+    if (op_ret < 0) {
+        gf_msg(plugin, GF_LOG_ERROR, 0, CVLT_EXTRACTION_FAILED,
+               " failed to extract store info for gfid=%s"
+               " offset=%" PRIu64 " size=%" GF_PRI_SIZET
+               ", "
+               " buf=%p",
+               uuid_utoa(locxattr->gfid), off, size, req->iobuf->ptr);
+        goto err;
+    }
+
+    op_ret = cvlt_init_file_info(locxattr, &(req->file_info));
+    if (op_ret < 0) {
+        gf_msg(plugin, GF_LOG_ERROR, 0, CVLT_EXTRACTION_FAILED,
+               " failed to extract file info for gfid=%s"
+               " offset=%" PRIu64 " size=%" GF_PRI_SIZET
+               ", "
+               " buf=%p",
+               uuid_utoa(locxattr->gfid), off, size, req->iobuf->ptr);
+        goto err;
+    }
+
+    /*
+     * Submit the read request.
+     */
+    op_ret = parch->fops.read(&(parch->descinfo), &(req->store_info),
+                              &(req->file_info), off, req->iobuf->ptr, size,
+                              &op_errno, cvlt_readv_complete, req);
+
+    if (op_ret < 0) {
+        gf_msg(plugin, GF_LOG_ERROR, 0, CVLT_EXTRACTION_FAILED,
+               " read failed on gfid=%s"
+               " offset=%" PRIu64 " size=%" GF_PRI_SIZET
+               ", "
+               " buf=%p",
+               uuid_utoa(locxattr->gfid), off, size, req->iobuf->ptr);
+        goto err;
+    }
+
+    return 0;
+
+err:
+
+    iobref = iobref_new();
+    gf_msg_debug(plugin, 0, " read unwinding stack op_ret = %d, op_errno = %d",
+                 op_ret, op_errno);
+
+    STACK_UNWIND_STRICT(readv, frame, op_ret, op_errno, &iov, 1,
+                        &(local->stbuf), iobref, local->xattr_rsp);
+
+    if (iobref) {
+        iobref_unref(iobref);
+    }
+
+    if (req) {
+        cvlt_free_req(parch, req);
+    }
+
+    return 0;
+}
diff --git a/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/libcvlt.h b/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/libcvlt.h
new file mode 100644
index 00000000000..c45ac948f6c
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync-plugins/src/cvlt/src/libcvlt.h
@@ -0,0 +1,84 @@
+/*
+  Copyright (c) 2018 Commvault Systems, Inc. <http://www.commvault.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+#ifndef _LIBCVLT_H
+#define _LIBCVLT_H
+
+#include <semaphore.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/call-stub.h>
+#include <glusterfs/syncop.h>
+#include <glusterfs/compat-errno.h>
+#include "cloudsync-common.h"
+#include "libcvlt-mem-types.h"
+#include "archivestore.h"
+
+enum _cvlt_op {
+    CVLT_READ_OP = 1,
+    CVLT_WRITE_OP = 2,
+    CVLT_RESTORE_OP = 3,
+    CVLT_ARCHIVE_OP = 4,
+    CVLT_LOOKUP_OP = 5,
+    CVLT_XATTR_OP = 6,
+    CVLT_STAT_OP = 7,
+    CVLT_FSTAT_op = 8,
+    CVLT_UNDEF_OP = 127
+};
+typedef enum _cvlt_op cvlt_op_t;
+
+struct _archive;
+struct _cvlt_request {
+    uint64_t offset;
+    uint64_t bytes;
+    struct iobuf *iobuf;
+    struct iobref *iobref;
+    call_frame_t *frame;
+    cvlt_op_t op_type;
+    int32_t op_ret;
+    int32_t op_errno;
+    xlator_t *this;
+    sem_t sem;
+    archstore_info_t store_info;
+    archstore_fileinfo_t file_info;
+    cs_size_xattr_t szxattr;
+};
+typedef struct _cvlt_request cvlt_request_t;
+
+struct _archive {
+    gf_lock_t lock;                /* lock for controlling access   */
+    xlator_t *xl;                  /* xlator                        */
+    void *handle;                  /* handle returned from dlopen   */
+    int32_t nreqs;                 /* num requests active           */
+    struct mem_pool *req_pool;     /* pool for requests             */
+    struct iobuf_pool *iobuf_pool; /* iobuff pool                   */
+    archstore_desc_t descinfo;     /* Archive store descriptor info */
+    archstore_methods_t fops;      /* function pointers             */
+    char *product_id;
+    char *store_id;
+    char *trailer;
+};
+typedef struct _archive archive_t;
+
+void *
+cvlt_init(xlator_t *);
+
+int
+cvlt_reconfigure(xlator_t *, dict_t *);
+
+void
+cvlt_fini(void *);
+
+int
+cvlt_download(call_frame_t *, void *);
+
+int
+cvlt_read(call_frame_t *, void *);
+
+#endif
diff --git a/xlators/features/cloudsync/src/cloudsync.c b/xlators/features/cloudsync/src/cloudsync.c
new file mode 100644
index 00000000000..7f0b9e563b8
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync.c
@@ -0,0 +1,2076 @@
+/*
+ *   Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com>
+ *   This file is part of GlusterFS.
+ *
+ *   This file is licensed to you under your choice of the GNU Lesser
+ *   General Public License, version 3 or any later version (LGPLv3 or
+ *   later), or the GNU General Public License, version 2 (GPLv2), in all
+ *   cases as published by the Free Software Foundation.
+ */
+
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+#include "cloudsync.h"
+#include "cloudsync-common.h"
+#include <glusterfs/call-stub.h>
+#include "cloudsync-autogen-fops.h"
+
+#include <string.h>
+#include <dlfcn.h>
+
+static void
+cs_cleanup_private(cs_private_t *priv)
+{
+    if (priv) {
+        if (priv->stores) {
+            priv->stores->fini(priv->stores->config);
+            GF_FREE(priv->stores);
+        }
+
+        pthread_spin_destroy(&priv->lock);
+        GF_FREE(priv);
+    }
+
+    return;
+}
+
+static struct cs_plugin plugins[] = {
+    {.name = "cloudsyncs3",
+     .library = "cloudsyncs3.so",
+     .description = "cloudsync s3 store."},
+#if defined(__linux__)
+    {.name = "cvlt",
+     .library = "cloudsynccvlt.so",
+     .description = "Commvault content store."},
+#endif
+    {.name = NULL},
+};
+
+int
+cs_init(xlator_t *this)
+{
+    cs_private_t *priv = NULL;
+    gf_boolean_t per_vol = _gf_false;
+    int ret = 0;
+    char *libpath = NULL;
+    store_methods_t *store_methods = NULL;
+    void *handle = NULL;
+    char *temp_str = NULL;
+    int index = 0;
+    char *libname = NULL;
+
+    priv = GF_CALLOC(1, sizeof(*priv), gf_cs_mt_cs_private_t);
+    if (!priv) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "insufficient memory");
+        goto out;
+    }
+
+    priv->this = this;
+
+    this->local_pool = mem_pool_new(cs_local_t, 512);
+    if (!this->local_pool) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, ENOMEM, "initialisation failed.");
+        ret = -1;
+        goto out;
+    }
+
+    this->private = priv;
+
+    GF_OPTION_INIT("cloudsync-remote-read", priv->remote_read, bool, out);
+
+    /* temp workaround. Should be configurable through glusterd*/
+    per_vol = _gf_true;
+
+    if (per_vol) {
+        if (dict_get_str_sizen(this->options, "cloudsync-storetype",
+                               &temp_str) == 0) {
+            for (index = 0; plugins[index].name; index++) {
+                if (!strcmp(temp_str, plugins[index].name)) {
+                    libname = plugins[index].library;
+                    break;
+                }
+            }
+        } else {
+            ret = 0;
+        }
+
+        if (!libname) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, 0, "no plugin enabled");
+            ret = 0;
+            goto out;
+        }
+
+        ret = gf_asprintf(&libpath, "%s/%s", CS_PLUGINDIR, libname);
+        if (ret == -1) {
+            goto out;
+        }
+
+        handle = dlopen(libpath, RTLD_NOW);
+        if (!handle) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, 0,
+                   "could not "
+                   "load the required library. %s",
+                   dlerror());
+            ret = 0;
+            goto out;
+        } else {
+            gf_msg(this->name, GF_LOG_INFO, 0, 0,
+                   "loading library:%s successful", libname);
+        }
+
+        priv->stores = GF_CALLOC(1, sizeof(struct cs_remote_stores),
+                                 gf_cs_mt_cs_remote_stores_t);
+        if (!priv->stores) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+                   "Could not "
+                   "allocate memory for priv->stores");
+            ret = -1;
+            goto out;
+        }
+
+        (void)dlerror(); /* clear out previous error string */
+
+        /* load library methods */
+        store_methods = (store_methods_t *)dlsym(handle, "store_ops");
+        if (!store_methods) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, 0, "null store_methods %s",
+                   dlerror());
+            ret = -1;
+            goto out;
+        }
+
+        (void)dlerror();
+
+        if (priv->remote_read) {
+            priv->stores->rdfop = store_methods->fop_remote_read;
+            if (!priv->stores->rdfop) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+                       "failed to get"
+                       " read fop %s",
+                       dlerror());
+                ret = -1;
+                goto out;
+            }
+        }
+
+        priv->stores->dlfop = store_methods->fop_download;
+        if (!priv->stores->dlfop) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+                   "failed to get"
+                   " download fop %s",
+                   dlerror());
+            ret = -1;
+            goto out;
+        }
+
+        (void)dlerror();
+        priv->stores->init = store_methods->fop_init;
+        if (!priv->stores->init) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+                   "failed to get"
+                   " init fop %s",
+                   dlerror());
+            ret = -1;
+            goto out;
+        }
+
+        (void)dlerror();
+        priv->stores->reconfigure = store_methods->fop_reconfigure;
+        if (!priv->stores->reconfigure) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+                   "failed to get"
+                   " reconfigure fop %s",
+                   dlerror());
+            ret = -1;
+            goto out;
+        }
+
+        priv->stores->handle = handle;
+
+        priv->stores->config = (void *)((priv->stores->init)(this));
+        if (!priv->stores->config) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, 0, "null config");
+            ret = -1;
+            goto out;
+        }
+    }
+
+    ret = 0;
+
+out:
+    if (ret == -1) {
+        if (this->local_pool) {
+            mem_pool_destroy(this->local_pool);
+            this->local_pool = NULL;
+        }
+
+        cs_cleanup_private(priv);
+
+        if (handle) {
+            dlclose(handle);
+        }
+    }
+
+    GF_FREE(libpath);
+
+    return ret;
+}
+
+int
+cs_forget(xlator_t *this, inode_t *inode)
+{
+    uint64_t ctx_int = 0;
+    cs_inode_ctx_t *ctx = NULL;
+
+    inode_ctx_del(inode, this, &ctx_int);
+    if (!ctx_int)
+        return 0;
+
+    ctx = (cs_inode_ctx_t *)(uintptr_t)ctx_int;
+
+    GF_FREE(ctx);
+    return 0;
+}
+
+void
+cs_fini(xlator_t *this)
+{
+    cs_private_t *priv = NULL;
+    priv = this->private;
+
+    cs_cleanup_private(priv);
+}
+
+int
+cs_reconfigure(xlator_t *this, dict_t *options)
+{
+    cs_private_t *priv = NULL;
+    int ret = 0;
+
+    priv = this->private;
+    if (!priv) {
+        ret = -1;
+        goto out;
+    }
+
+    GF_OPTION_RECONF("cloudsync-remote-read", priv->remote_read, options, bool,
+                     out);
+
+    /* needed only for per volume configuration*/
+    ret = priv->stores->reconfigure(this, options);
+
+out:
+    return ret;
+}
+
+int32_t
+cs_mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("cloudsync", this, out);
+
+    ret = xlator_mem_acct_init(this, gf_cs_mt_end + 1);
+
+    if (ret != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "Memory accounting init failed");
+        return ret;
+    }
+out:
+    return ret;
+}
+
+int32_t
+cs_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t off, dict_t *xdata)
+{
+    int ret = 0;
+    int op_errno = ENOMEM;
+
+    if (!xdata) {
+        xdata = dict_new();
+        if (!xdata) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, ENOMEM,
+                   "failed to create "
+                   "dict");
+            goto err;
+        }
+    }
+
+    ret = dict_set_uint32(xdata, GF_CS_OBJECT_STATUS, 1);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "dict_set failed key:"
+               " %s",
+               GF_CS_OBJECT_STATUS);
+        goto err;
+    }
+
+    STACK_WIND(frame, default_readdirp_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdirp, fd, size, off, xdata);
+    return 0;
+err:
+    STACK_UNWIND_STRICT(readdirp, frame, -1, op_errno, NULL, NULL);
+    return 0;
+}
+
+int32_t
+cs_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                struct iatt *postbuf, dict_t *xdata)
+{
+    cs_local_t *local = NULL;
+    int ret = 0;
+    uint64_t val = 0;
+
+    local = frame->local;
+
+    local->call_cnt++;
+
+    if (op_ret == -1) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "truncate failed");
+        ret = dict_get_uint64(xdata, GF_CS_OBJECT_STATUS, &val);
+        if (ret == 0) {
+            if (val == GF_CS_ERROR) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+                       "could not get file state, unwinding");
+                op_ret = -1;
+                op_errno = EIO;
+                goto unwind;
+            } else {
+                __cs_inode_ctx_update(this, local->loc.inode, val);
+                gf_msg(this->name, GF_LOG_INFO, 0, 0, " state = %" PRIu64, val);
+
+                if (local->call_cnt == 1 &&
+                    (val == GF_CS_REMOTE || val == GF_CS_DOWNLOADING)) {
+                    gf_msg(this->name, GF_LOG_WARNING, 0, 0,
+                           "will repair and download "
+                           "the file, current state : %" PRIu64,
+                           val);
+                    goto repair;
+                } else {
+                    gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+                           "second truncate, Unwinding");
+                    goto unwind;
+                }
+            }
+        } else {
+            gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+                   "file state "
+                   "could not be figured, unwinding");
+            goto unwind;
+        }
+    } else {
+        /* successful write => file is local */
+        __cs_inode_ctx_update(this, local->loc.inode, GF_CS_LOCAL);
+        gf_msg(this->name, GF_LOG_INFO, 0, 0,
+               "state : GF_CS_LOCAL"
+               ", truncate successful");
+
+        goto unwind;
+    }
+
+repair:
+    ret = locate_and_execute(frame);
+    if (ret) {
+        goto unwind;
+    }
+
+    return 0;
+
+unwind:
+    CS_STACK_UNWIND(truncate, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+    return 0;
+}
+
+int32_t
+cs_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+            dict_t *xdata)
+{
+    cs_local_t *local = NULL;
+    int ret = 0;
+    cs_inode_ctx_t *ctx = NULL;
+    gf_cs_obj_state state = -1;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+
+    local = cs_local_init(this, frame, loc, NULL, GF_FOP_TRUNCATE);
+    if (!local) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "local init failed");
+        goto err;
+    }
+
+    __cs_inode_ctx_get(this, loc->inode, &ctx);
+
+    if (ctx)
+        state = __cs_get_file_state(loc->inode, ctx);
+    else
+        state = GF_CS_LOCAL;
+
+    local->xattr_req = xdata ? dict_ref(xdata) : (xdata = dict_new());
+
+    ret = dict_set_uint32(local->xattr_req, GF_CS_OBJECT_STATUS, 1);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "dict_set failed key:"
+               " %s",
+               GF_CS_OBJECT_STATUS);
+        goto err;
+    }
+
+    local->stub = fop_truncate_stub(frame, cs_resume_truncate, loc, offset,
+                                    xdata);
+    if (!local->stub) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "insufficient memory");
+        goto err;
+    }
+
+    if (state == GF_CS_LOCAL) {
+        STACK_WIND(frame, cs_truncate_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+
+    } else {
+        local->call_cnt++;
+        ret = locate_and_execute(frame);
+        if (ret) {
+            goto err;
+        }
+    }
+
+    return 0;
+err:
+    CS_STACK_UNWIND(truncate, frame, -1, ENOMEM, NULL, NULL, NULL);
+    return 0;
+}
+
+int32_t
+cs_statfs_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct statvfs *buf, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(statfs, frame, op_ret, op_errno, buf, xdata);
+    return 0;
+}
+
+int32_t
+cs_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    STACK_WIND(frame, cs_statfs_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->statfs, loc, xdata);
+    return 0;
+}
+
+int32_t
+cs_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(getxattr, frame, op_ret, op_errno, dict, xdata);
+    return 0;
+}
+
+int32_t
+cs_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, const char *name,
+            dict_t *xattr_req)
+{
+    STACK_WIND(frame, cs_getxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->getxattr, loc, name, xattr_req);
+    return 0;
+}
+
+int32_t
+cs_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    cs_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->locked)
+        cs_inodelk_unlock(frame);
+
+    CS_STACK_UNWIND(setxattr, frame, op_ret, op_errno, xdata);
+
+    return 0;
+}
+
+int32_t
+cs_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+            int32_t flags, dict_t *xdata)
+{
+    data_t *tmp = NULL;
+    cs_local_t *local = NULL;
+    int ret = 0;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+
+    local = cs_local_init(this, frame, loc, NULL, GF_FOP_SETXATTR);
+    if (!local) {
+        ret = -1;
+        goto err;
+    }
+
+    local->xattr_req = xdata ? dict_ref(xdata) : (xdata = dict_new());
+
+    tmp = dict_get_sizen(dict, GF_CS_OBJECT_UPLOAD_COMPLETE);
+    if (tmp) {
+        /* Value of key should be the atime */
+        local->stub = fop_setxattr_stub(frame, cs_resume_setxattr, loc, dict,
+                                        flags, xdata);
+
+        if (!local->stub)
+            goto err;
+
+        ret = locate_and_execute(frame);
+        if (ret) {
+            goto err;
+        }
+
+        return 0;
+    }
+
+    STACK_WIND(frame, cs_setxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setxattr, loc, dict, flags, xdata);
+    return 0;
+err:
+    CS_STACK_UNWIND(setxattr, frame, -1, errno, NULL);
+    return 0;
+}
+
+int32_t
+cs_fgetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(fgetxattr, frame, op_ret, op_errno, dict, xdata);
+    return 0;
+}
+
+int32_t
+cs_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+             dict_t *xdata)
+{
+    STACK_WIND(frame, cs_fgetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fgetxattr, fd, name, xdata);
+    return 0;
+}
+
+int32_t
+cs_fsetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(fsetxattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+int32_t
+cs_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+             int32_t flags, dict_t *xdata)
+{
+    STACK_WIND(frame, cs_fsetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
+    return 0;
+}
+
+int32_t
+cs_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct iatt *preparent, struct iatt *postparent,
+              dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(unlink, frame, op_ret, op_errno, preparent, postparent,
+                        xdata);
+    return 0;
+}
+
+int32_t
+cs_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+          dict_t *xattr_req)
+{
+    cs_local_t *local = NULL;
+    int ret = 0;
+
+    local = cs_local_init(this, frame, loc, NULL, GF_FOP_UNLINK);
+    if (!local)
+        goto err;
+
+    local->xattr_req = xattr_req ? dict_ref(xattr_req) : dict_new();
+
+    ret = dict_set_uint32(local->xattr_req, GF_CS_OBJECT_STATUS, 1);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "dict_set failed key:"
+               " %s",
+               GF_CS_OBJECT_STATUS);
+        goto err;
+    }
+    STACK_WIND(frame, cs_unlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->unlink, loc, flags, local->xattr_req);
+    return 0;
+err:
+    CS_STACK_UNWIND(unlink, frame, -1, errno, NULL, NULL, NULL);
+    return 0;
+}
+
+int32_t
+cs_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, fd_t *fd, dict_t *xdata)
+{
+    int ret = 0;
+    uint64_t val = 0;
+
+    if (op_ret == 0) {
+        ret = dict_get_uint64(xdata, GF_CS_OBJECT_STATUS, &val);
+        if (!ret) {
+            ret = __cs_inode_ctx_update(this, fd->inode, val);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, 0, "ctx update failed");
+            }
+        }
+    } else {
+        cs_inode_ctx_reset(this, fd->inode);
+    }
+
+    CS_STACK_UNWIND(open, frame, op_ret, op_errno, fd, xdata);
+    return 0;
+}
+
+int32_t
+cs_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+        fd_t *fd, dict_t *xattr_req)
+{
+    cs_local_t *local = NULL;
+    int ret = 0;
+
+    local = cs_local_init(this, frame, NULL, fd, GF_FOP_OPEN);
+    if (!local)
+        goto err;
+
+    local->xattr_req = xattr_req ? dict_ref(xattr_req) : dict_new();
+
+    ret = dict_set_uint32(local->xattr_req, GF_CS_OBJECT_STATUS, 1);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "dict_set failed key:"
+               " %s",
+               GF_CS_OBJECT_STATUS);
+        goto err;
+    }
+
+    STACK_WIND(frame, cs_open_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->open, loc, flags, fd, local->xattr_req);
+    return 0;
+err:
+    CS_STACK_UNWIND(open, frame, -1, errno, NULL, NULL);
+    return 0;
+}
+
+int32_t
+cs_fstat_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, struct iatt *buf, dict_t *xdata)
+{
+    int ret = 0;
+    uint64_t val = 0;
+    fd_t *fd = NULL;
+    cs_local_t *local = NULL;
+
+    local = frame->local;
+
+    fd = local->fd;
+
+    if (op_ret == 0) {
+        ret = dict_get_uint64(xdata, GF_CS_OBJECT_STATUS, &val);
+        if (!ret) {
+            gf_msg_debug(this->name, 0, "state %" PRIu64, val);
+            ret = __cs_inode_ctx_update(this, fd->inode, val);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, 0, "ctx update failed");
+            }
+        }
+    } else {
+        cs_inode_ctx_reset(this, fd->inode);
+    }
+
+    CS_STACK_UNWIND(fstat, frame, op_ret, op_errno, buf, xdata);
+
+    return 0;
+}
+
+int32_t
+cs_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xattr_req)
+{
+    cs_local_t *local = NULL;
+    int ret = 0;
+
+    local = cs_local_init(this, frame, NULL, fd, GF_FOP_FSTAT);
+    if (!local)
+        goto err;
+
+    if (fd->inode->ia_type == IA_IFDIR)
+        goto wind;
+
+    local->xattr_req = xattr_req ? dict_ref(xattr_req) : dict_new();
+
+    ret = dict_set_uint32(local->xattr_req, GF_CS_OBJECT_STATUS, 1);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "dict_set failed key:"
+               " %s",
+               GF_CS_OBJECT_STATUS);
+        goto err;
+    }
+
+wind:
+    STACK_WIND(frame, cs_fstat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fstat, fd, local->xattr_req);
+    return 0;
+err:
+    CS_STACK_UNWIND(fstat, frame, -1, errno, NULL, NULL);
+    return 0;
+}
+
+cs_local_t *
+cs_local_init(xlator_t *this, call_frame_t *frame, loc_t *loc, fd_t *fd,
+              glusterfs_fop_t fop)
+{
+    cs_local_t *local = NULL;
+    int ret = 0;
+
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto out;
+
+    if (loc) {
+        ret = loc_copy(&local->loc, loc);
+        if (ret)
+            goto out;
+    }
+
+    if (fd) {
+        local->fd = fd_ref(fd);
+    }
+
+    local->op_ret = -1;
+    local->op_errno = EUCLEAN;
+    local->fop = fop;
+    local->dloffset = 0;
+    frame->local = local;
+    local->locked = _gf_false;
+    local->call_cnt = 0;
+out:
+    if (ret) {
+        if (local)
+            mem_put(local);
+        local = NULL;
+    }
+
+    return local;
+}
+
+call_frame_t *
+cs_lock_frame(call_frame_t *parent_frame)
+{
+    call_frame_t *lock_frame = NULL;
+
+    lock_frame = copy_frame(parent_frame);
+
+    if (lock_frame == NULL)
+        goto out;
+
+    set_lk_owner_from_ptr(&lock_frame->root->lk_owner, parent_frame->root);
+
+out:
+    return lock_frame;
+}
+
+void
+cs_lock_wipe(call_frame_t *lock_frame)
+{
+    CS_STACK_DESTROY(lock_frame);
+}
+
+int32_t
+cs_inodelk_unlock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    cs_lock_wipe(frame);
+
+    return 0;
+}
+
+int
+cs_inodelk_unlock(call_frame_t *main_frame)
+{
+    xlator_t *this = NULL;
+    struct gf_flock flock = {
+        0,
+    };
+    call_frame_t *lock_frame = NULL;
+    cs_local_t *lock_local = NULL;
+    cs_local_t *main_local = NULL;
+    int ret = 0;
+
+    this = main_frame->this;
+    main_local = main_frame->local;
+
+    lock_frame = cs_lock_frame(main_frame);
+    if (!lock_frame)
+        goto out;
+
+    lock_local = cs_local_init(this, lock_frame, NULL, NULL, 0);
+    if (!lock_local)
+        goto out;
+
+    ret = cs_build_loc(&lock_local->loc, main_frame);
+    if (ret) {
+        goto out;
+    }
+
+    flock.l_type = F_UNLCK;
+
+    main_local->locked = _gf_false;
+
+    STACK_WIND(lock_frame, cs_inodelk_unlock_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->inodelk, CS_LOCK_DOMAIN,
+               &lock_local->loc, F_SETLKW, &flock, NULL);
+
+    return 0;
+
+out:
+    gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+           "Stale lock would be found on"
+           " server");
+
+    if (lock_frame)
+        cs_lock_wipe(lock_frame);
+
+    return 0;
+}
+
+int
+cs_download_task(void *arg)
+{
+    call_frame_t *frame = NULL;
+    xlator_t *this = NULL;
+    cs_private_t *priv = NULL;
+    int ret = -1;
+    char *sign_req = NULL;
+    fd_t *fd = NULL;
+    cs_local_t *local = NULL;
+    dict_t *dict = NULL;
+
+    frame = (call_frame_t *)arg;
+
+    this = frame->this;
+
+    priv = this->private;
+
+    if (!priv->stores) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "No remote store "
+               "plugins found");
+        ret = -1;
+        goto out;
+    }
+
+    local = frame->local;
+
+    if (local->fd)
+        fd = fd_anonymous(local->fd->inode);
+    else
+        fd = fd_anonymous(local->loc.inode);
+
+    if (!fd) {
+        gf_msg("CS", GF_LOG_ERROR, 0, 0, "fd creation failed");
+        ret = -1;
+        goto out;
+    }
+
+    local->dlfd = fd;
+    local->dloffset = 0;
+
+    dict = dict_new();
+    if (!dict) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, ENOMEM,
+               "failed to create "
+               "dict");
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_set_uint32(dict, GF_CS_OBJECT_DOWNLOADING, 1);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "dict_set failed");
+        ret = -1;
+        goto out;
+    }
+
+    ret = syncop_fsetxattr(this, local->fd, dict, 0, NULL, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "fsetxattr failed "
+               "key %s",
+               GF_CS_OBJECT_DOWNLOADING);
+        ret = -1;
+        goto out;
+    }
+    /*this calling method is for per volume setting */
+    ret = priv->stores->dlfop(frame, priv->stores->config);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "download failed"
+               ", remotepath: %s",
+               local->remotepath);
+
+        /*using dlfd as it is anonymous and have RDWR flag*/
+        ret = syncop_ftruncate(FIRST_CHILD(this), local->dlfd, 0, NULL, NULL,
+                               NULL, NULL);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, -ret, "ftruncate failed");
+        } else {
+            gf_msg_debug(this->name, 0, "ftruncate succeed");
+        }
+
+        ret = -1;
+        goto out;
+    } else {
+        gf_msg(this->name, GF_LOG_INFO, 0, 0,
+               "download success, path"
+               " : %s",
+               local->remotepath);
+
+        ret = syncop_fremovexattr(this, local->fd, GF_CS_OBJECT_REMOTE, NULL,
+                                  NULL);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, -ret,
+                   "removexattr failed, remotexattr");
+            ret = -1;
+            goto out;
+        } else {
+            gf_msg_debug(this->name, 0,
+                         "fremovexattr success, "
+                         "path : %s",
+                         local->remotepath);
+        }
+
+        ret = syncop_fremovexattr(this, local->fd, GF_CS_OBJECT_DOWNLOADING,
+                                  NULL, NULL);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, -ret,
+                   "removexattr failed, downloading xattr, path %s",
+                   local->remotepath);
+            ret = -1;
+            goto out;
+        } else {
+            gf_msg_debug(this->name, 0,
+                         "fremovexattr success"
+                         " path  %s",
+                         local->remotepath);
+        }
+    }
+
+out:
+    GF_FREE(sign_req);
+
+    if (dict)
+        dict_unref(dict);
+
+    if (fd) {
+        fd_unref(fd);
+        local->dlfd = NULL;
+    }
+
+    return ret;
+}
+
+int
+cs_download(call_frame_t *frame)
+{
+    int ret = 0;
+    cs_local_t *local = NULL;
+    xlator_t *this = NULL;
+
+    local = frame->local;
+    this = frame->this;
+
+    if (!local->remotepath) {
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "remote path not"
+               " available. Check posix logs to resolve");
+        goto out;
+    }
+
+    ret = cs_download_task((void *)frame);
+out:
+    return ret;
+}
+
+int
+cs_set_xattr_req(call_frame_t *frame)
+{
+    cs_local_t *local = NULL;
+    GF_UNUSED int ret = 0;
+
+    local = frame->local;
+
+    /* When remote reads are performed (i.e. reads on remote store),
+     * there needs to be a way to associate a file on gluster volume
+     * with its correspnding file on the remote store. In order to do
+     * that, a unique key can be maintained as an xattr
+     * (GF_CS_XATTR_ARCHIVE_UUID)on the stub file on gluster bricks.
+     * This xattr should be provided to the plugin to
+     * perform the read fop on the correct file. This assumes that the file
+     * hierarchy and name need not be the same on remote store as that of
+     * the gluster volume.
+     */
+    ret = dict_set_sizen_str_sizen(local->xattr_req, GF_CS_XATTR_ARCHIVE_UUID,
+                                   "1");
+
+    return 0;
+}
+
+int
+cs_update_xattrs(call_frame_t *frame, dict_t *xdata)
+{
+    cs_local_t *local = NULL;
+    xlator_t *this = NULL;
+    int size = -1;
+    GF_UNUSED int ret = 0;
+
+    local = frame->local;
+    this = frame->this;
+
+    local->xattrinfo.lxattr = GF_CALLOC(1, sizeof(cs_loc_xattr_t),
+                                        gf_cs_mt_cs_lxattr_t);
+    if (!local->xattrinfo.lxattr) {
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        goto err;
+    }
+
+    gf_uuid_copy(local->xattrinfo.lxattr->gfid, local->loc.gfid);
+
+    if (local->remotepath) {
+        local->xattrinfo.lxattr->file_path = gf_strdup(local->remotepath);
+        if (!local->xattrinfo.lxattr->file_path) {
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            goto err;
+        }
+    }
+
+    ret = dict_get_gfuuid(xdata, GF_CS_XATTR_ARCHIVE_UUID,
+                          &(local->xattrinfo.lxattr->uuid));
+
+    if (ret) {
+        gf_uuid_clear(local->xattrinfo.lxattr->uuid);
+    }
+    size = strlen(this->name) - strlen("-cloudsync") + 1;
+    local->xattrinfo.lxattr->volname = GF_CALLOC(1, size, gf_common_mt_char);
+    if (!local->xattrinfo.lxattr->volname) {
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        goto err;
+    }
+    strncpy(local->xattrinfo.lxattr->volname, this->name, size - 1);
+    local->xattrinfo.lxattr->volname[size - 1] = '\0';
+
+    return 0;
+err:
+    cs_xattrinfo_wipe(local);
+    return -1;
+}
+
+int
+cs_serve_readv(call_frame_t *frame, off_t offset, size_t size, uint32_t flags)
+{
+    xlator_t *this = NULL;
+    cs_private_t *priv = NULL;
+    int ret = -1;
+    fd_t *fd = NULL;
+    cs_local_t *local = NULL;
+
+    local = frame->local;
+    this = frame->this;
+    priv = this->private;
+
+    if (!local->remotepath) {
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "remote path not"
+               " available. Check posix logs to resolve");
+        goto out;
+    }
+
+    if (!priv->stores) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "No remote store "
+               "plugins found");
+        ret = -1;
+        goto out;
+    }
+
+    if (local->fd) {
+        fd = fd_anonymous(local->fd->inode);
+    } else {
+        fd = fd_anonymous(local->loc.inode);
+    }
+
+    local->xattrinfo.size = size;
+    local->xattrinfo.offset = offset;
+    local->xattrinfo.flags = flags;
+
+    if (!fd) {
+        gf_msg("CS", GF_LOG_ERROR, 0, 0, "fd creation failed");
+        ret = -1;
+        goto out;
+    }
+
+    local->dlfd = fd;
+    local->dloffset = offset;
+
+    /*this calling method is for per volume setting */
+    ret = priv->stores->rdfop(frame, priv->stores->config);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "read failed"
+               ", remotepath: %s",
+               local->remotepath);
+        ret = -1;
+        goto out;
+    } else {
+        gf_msg(this->name, GF_LOG_INFO, 0, 0,
+               "read success, path"
+               " : %s",
+               local->remotepath);
+    }
+
+out:
+    if (fd) {
+        fd_unref(fd);
+        local->dlfd = NULL;
+    }
+    return ret;
+}
+
+int32_t
+cs_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, struct iovec *vector, int32_t count,
+             struct iatt *stbuf, struct iobref *iobref, dict_t *xdata)
+{
+    cs_local_t *local = NULL;
+    int ret = 0;
+    uint64_t val = 0;
+    fd_t *fd = NULL;
+
+    local = frame->local;
+    fd = local->fd;
+
+    local->call_cnt++;
+
+    if (op_ret == -1) {
+        ret = dict_get_uint64(xdata, GF_CS_OBJECT_STATUS, &val);
+        if (ret == 0) {
+            if (val == GF_CS_ERROR) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+                       "could not get file state, unwinding");
+                op_ret = -1;
+                op_errno = EIO;
+                goto unwind;
+            } else {
+                __cs_inode_ctx_update(this, fd->inode, val);
+                gf_msg(this->name, GF_LOG_INFO, 0, 0, " state = %" PRIu64, val);
+
+                if (local->call_cnt == 1 &&
+                    (val == GF_CS_REMOTE || val == GF_CS_DOWNLOADING)) {
+                    gf_msg(this->name, GF_LOG_INFO, 0, 0,
+                           " will read from remote : %" PRIu64, val);
+                    goto repair;
+                } else {
+                    gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+                           "second readv, Unwinding");
+                    goto unwind;
+                }
+            }
+        } else {
+            gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+                   "file state "
+                   "could not be figured, unwinding");
+            goto unwind;
+        }
+    } else {
+        /* successful readv => file is local */
+        __cs_inode_ctx_update(this, fd->inode, GF_CS_LOCAL);
+        gf_msg(this->name, GF_LOG_INFO, 0, 0,
+               "state : GF_CS_LOCAL"
+               ", readv successful");
+
+        goto unwind;
+    }
+
+repair:
+    ret = locate_and_execute(frame);
+    if (ret) {
+        goto unwind;
+    }
+
+    return 0;
+
+unwind:
+    CS_STACK_UNWIND(readv, frame, op_ret, op_errno, vector, count, stbuf,
+                    iobref, xdata);
+
+    return 0;
+}
+
+int32_t
+cs_resume_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                off_t offset, uint32_t flags, dict_t *xdata)
+{
+    int ret = 0;
+
+    ret = cs_resume_postprocess(this, frame, fd->inode);
+    if (ret) {
+        goto unwind;
+    }
+
+    cs_inodelk_unlock(frame);
+
+    STACK_WIND(frame, cs_readv_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readv, fd, size, offset, flags, xdata);
+
+    return 0;
+
+unwind:
+    cs_inodelk_unlock(frame);
+
+    cs_common_cbk(frame);
+
+    return 0;
+}
+
+int32_t
+cs_resume_remote_readv(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                       size_t size, off_t offset, uint32_t flags, dict_t *xdata)
+{
+    int ret = 0;
+    cs_local_t *local = NULL;
+    gf_cs_obj_state state = -1;
+    cs_inode_ctx_t *ctx = NULL;
+
+    cs_inodelk_unlock(frame);
+
+    local = frame->local;
+    if (!local) {
+        ret = -1;
+        goto unwind;
+    }
+
+    __cs_inode_ctx_get(this, fd->inode, &ctx);
+
+    state = __cs_get_file_state(fd->inode, ctx);
+    if (state == GF_CS_ERROR) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "status is GF_CS_ERROR."
+               " Aborting readv");
+        local->op_ret = -1;
+        local->op_errno = EREMOTE;
+        ret = -1;
+        goto unwind;
+    }
+
+    /* Serve readv from remote store only if it is remote. */
+    gf_msg_debug(this->name, 0, "status of file %s is %d",
+                 local->remotepath ? local->remotepath : "", state);
+
+    /* We will reach this condition if local inode ctx had REMOTE
+     * state when the control was in cs_readv but after stat
+     * we got an updated state saying that the file is LOCAL.
+     */
+    if (state == GF_CS_LOCAL) {
+        STACK_WIND(frame, cs_readv_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->readv, fd, size, offset, flags,
+                   xdata);
+    } else if (state == GF_CS_REMOTE) {
+        ret = cs_resume_remote_readv_postprocess(this, frame, fd->inode, offset,
+                                                 size, flags);
+        /* Failed to submit the remote readv fop to plugin */
+        if (ret) {
+            local->op_ret = -1;
+            local->op_errno = EREMOTE;
+            goto unwind;
+        }
+        /* When the file is in any other intermediate state,
+         * we should not perform remote reads.
+         */
+    } else {
+        local->op_ret = -1;
+        local->op_errno = EINVAL;
+        goto unwind;
+    }
+
+    return 0;
+
+unwind:
+    cs_common_cbk(frame);
+
+    return 0;
+}
+
+int32_t
+cs_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+         off_t offset, uint32_t flags, dict_t *xdata)
+{
+    int op_errno = ENOMEM;
+    cs_local_t *local = NULL;
+    int ret = 0;
+    cs_inode_ctx_t *ctx = NULL;
+    gf_cs_obj_state state = -1;
+    cs_private_t *priv = NULL;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    priv = this->private;
+
+    local = cs_local_init(this, frame, NULL, fd, GF_FOP_READ);
+    if (!local) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "local init failed");
+        goto err;
+    }
+
+    __cs_inode_ctx_get(this, fd->inode, &ctx);
+
+    if (ctx)
+        state = __cs_get_file_state(fd->inode, ctx);
+    else
+        state = GF_CS_LOCAL;
+
+    local->xattr_req = xdata ? dict_ref(xdata) : (xdata = dict_new());
+
+    ret = dict_set_uint32(local->xattr_req, GF_CS_OBJECT_STATUS, 1);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "dict_set failed key:"
+               " %s",
+               GF_CS_OBJECT_STATUS);
+        goto err;
+    }
+
+    if (priv->remote_read) {
+        local->stub = fop_readv_stub(frame, cs_resume_remote_readv, fd, size,
+                                     offset, flags, xdata);
+    } else {
+        local->stub = fop_readv_stub(frame, cs_resume_readv, fd, size, offset,
+                                     flags, xdata);
+    }
+    if (!local->stub) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "insufficient memory");
+        goto err;
+    }
+
+    if (state == GF_CS_LOCAL) {
+        STACK_WIND(frame, cs_readv_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->readv, fd, size, offset, flags,
+                   xdata);
+    } else {
+        local->call_cnt++;
+        ret = locate_and_execute(frame);
+        if (ret) {
+            goto err;
+        }
+    }
+
+    return 0;
+
+err:
+    CS_STACK_UNWIND(readv, frame, -1, op_errno, NULL, -1, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int
+cs_resume_remote_readv_postprocess(xlator_t *this, call_frame_t *frame,
+                                   inode_t *inode, off_t offset, size_t size,
+                                   uint32_t flags)
+{
+    int ret = 0;
+
+    ret = cs_serve_readv(frame, offset, size, flags);
+
+    return ret;
+}
+
+int
+cs_stat_check_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                  int op_errno, struct iatt *stbuf, dict_t *xdata)
+{
+    cs_local_t *local = NULL;
+    call_stub_t *stub = NULL;
+    char *filepath = NULL;
+    int ret = 0;
+    inode_t *inode = NULL;
+    uint64_t val = 0;
+
+    local = frame->local;
+
+    if (op_ret == -1) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        gf_msg(this->name, GF_LOG_ERROR, 0, op_errno, "stat check failed");
+        goto err;
+    } else {
+        if (local->fd)
+            inode = local->fd->inode;
+        else
+            inode = local->loc.inode;
+
+        if (!inode) {
+            local->op_ret = -1;
+            local->op_errno = EINVAL;
+            gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+                   "null inode "
+                   "returned");
+            goto err;
+        }
+
+        ret = dict_get_uint64(xdata, GF_CS_OBJECT_STATUS, &val);
+        if (ret == 0) {
+            if (val == GF_CS_ERROR) {
+                cs_inode_ctx_reset(this, inode);
+                local->op_ret = -1;
+                local->op_errno = EIO;
+                gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+                       "status = GF_CS_ERROR. failed to get "
+                       " file state");
+                goto err;
+            } else {
+                ret = __cs_inode_ctx_update(this, inode, val);
+                gf_msg_debug(this->name, 0, "status : %" PRIu64, val);
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0, 0, "ctx update failed");
+                    local->op_ret = -1;
+                    local->op_errno = ENOMEM;
+                    goto err;
+                }
+            }
+        } else {
+            gf_msg_debug(this->name, 0, "status not found in dict");
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            goto err;
+        }
+
+        ret = dict_get_str_sizen(xdata, GF_CS_OBJECT_REMOTE, &filepath);
+        if (filepath) {
+            gf_msg_debug(this->name, 0, "filepath returned %s", filepath);
+            local->remotepath = gf_strdup(filepath);
+            if (!local->remotepath) {
+                local->op_ret = -1;
+                local->op_errno = ENOMEM;
+                goto err;
+            }
+        } else {
+            gf_msg_debug(this->name, 0, "NULL filepath");
+        }
+
+        ret = cs_update_xattrs(frame, xdata);
+        if (ret)
+            goto err;
+
+        local->op_ret = 0;
+        local->xattr_rsp = dict_ref(xdata);
+        memcpy(&local->stbuf, stbuf, sizeof(struct iatt));
+    }
+
+    stub = local->stub;
+    local->stub = NULL;
+    call_resume(stub);
+
+    return 0;
+err:
+    cs_inodelk_unlock(frame);
+
+    cs_common_cbk(frame);
+
+    return 0;
+}
+
+int
+cs_do_stat_check(call_frame_t *main_frame)
+{
+    cs_local_t *local = NULL;
+    xlator_t *this = NULL;
+    int ret = 0;
+
+    local = main_frame->local;
+    this = main_frame->this;
+
+    ret = dict_set_uint32(local->xattr_req, GF_CS_OBJECT_REPAIR, 256);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "dict_set failed");
+        goto err;
+    }
+
+    cs_set_xattr_req(main_frame);
+
+    if (local->fd) {
+        STACK_WIND(main_frame, cs_stat_check_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fstat, local->fd, local->xattr_req);
+    } else {
+        STACK_WIND(main_frame, cs_stat_check_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->stat, &local->loc,
+                   local->xattr_req);
+    }
+
+    return 0;
+
+err:
+    cs_inodelk_unlock(main_frame);
+
+    cs_common_cbk(main_frame);
+
+    return 0;
+}
+
+void
+cs_common_cbk(call_frame_t *frame)
+{
+    glusterfs_fop_t fop = -1;
+    cs_local_t *local = NULL;
+
+    local = frame->local;
+
+    fop = local->fop;
+
+    /*Note: Only the failure case needs to be handled here. Since for
+     * successful stat check the fop will resume anyway. The unwind can
+     * happen from the fop_cbk and each cbk can unlock the inodelk in case
+     * a lock was taken before. The lock status can be stored in frame */
+
+    /* for failure case  */
+
+    /*TODO: add other fops*/
+    switch (fop) {
+        case GF_FOP_WRITE:
+            CS_STACK_UNWIND(writev, frame, local->op_ret, local->op_errno, NULL,
+                            NULL, NULL);
+            break;
+
+        case GF_FOP_SETXATTR:
+            CS_STACK_UNWIND(setxattr, frame, local->op_ret, local->op_errno,
+                            NULL);
+            break;
+        case GF_FOP_READ:
+            CS_STACK_UNWIND(readv, frame, local->op_ret, local->op_errno, NULL,
+                            0, NULL, NULL, NULL);
+            break;
+        case GF_FOP_FTRUNCATE:
+            CS_STACK_UNWIND(ftruncate, frame, local->op_ret, local->op_errno,
+                            NULL, NULL, NULL);
+            break;
+
+        case GF_FOP_TRUNCATE:
+            CS_STACK_UNWIND(truncate, frame, local->op_ret, local->op_errno,
+                            NULL, NULL, NULL);
+            break;
+        default:
+            break;
+    }
+
+    return;
+}
+
+int
+cs_blocking_inodelk_cbk(call_frame_t *lock_frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    cs_local_t *main_local = NULL;
+    call_frame_t *main_frame = NULL;
+    cs_local_t *lock_local = NULL;
+
+    lock_local = lock_frame->local;
+
+    main_frame = lock_local->main_frame;
+    main_local = main_frame->local;
+
+    if (op_ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "inodelk failed");
+        main_local->op_errno = op_errno;
+        main_local->op_ret = op_ret;
+        goto err;
+    }
+
+    main_local->locked = _gf_true;
+
+    cs_lock_wipe(lock_frame);
+
+    cs_do_stat_check(main_frame);
+
+    return 0;
+err:
+    cs_common_cbk(main_frame);
+
+    cs_lock_wipe(lock_frame);
+
+    return 0;
+}
+
+int
+cs_build_loc(loc_t *loc, call_frame_t *frame)
+{
+    cs_local_t *local = NULL;
+    int ret = -1;
+
+    local = frame->local;
+
+    if (local->fd) {
+        loc->inode = inode_ref(local->fd->inode);
+        if (loc->inode) {
+            gf_uuid_copy(loc->gfid, loc->inode->gfid);
+            ret = 0;
+            goto out;
+        } else {
+            ret = -1;
+            goto out;
+        }
+    } else {
+        loc->inode = inode_ref(local->loc.inode);
+        if (loc->inode) {
+            gf_uuid_copy(loc->gfid, loc->inode->gfid);
+            ret = 0;
+            goto out;
+        } else {
+            ret = -1;
+            goto out;
+        }
+    }
+out:
+    return ret;
+}
+
+int
+cs_blocking_inodelk(call_frame_t *parent_frame)
+{
+    call_frame_t *lock_frame = NULL;
+    cs_local_t *lock_local = NULL;
+    xlator_t *this = NULL;
+    struct gf_flock flock = {
+        0,
+    };
+    int ret = 0;
+
+    this = parent_frame->this;
+
+    lock_frame = cs_lock_frame(parent_frame);
+    if (!lock_frame) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "insuffcient memory");
+        goto err;
+    }
+
+    lock_local = cs_local_init(this, lock_frame, NULL, NULL, 0);
+    if (!lock_local) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "local init failed");
+        goto err;
+    }
+
+    lock_local->main_frame = parent_frame;
+
+    flock.l_type = F_WRLCK;
+
+    ret = cs_build_loc(&lock_local->loc, parent_frame);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "build_loc failed");
+        goto err;
+    }
+
+    STACK_WIND(lock_frame, cs_blocking_inodelk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->inodelk, CS_LOCK_DOMAIN,
+               &lock_local->loc, F_SETLKW, &flock, NULL);
+
+    return 0;
+err:
+    if (lock_frame)
+        cs_lock_wipe(lock_frame);
+
+    return -1;
+}
+
+int
+locate_and_execute(call_frame_t *frame)
+{
+    int ret = 0;
+
+    ret = cs_blocking_inodelk(frame);
+
+    if (ret)
+        return -1;
+    else
+        return 0;
+}
+
+int32_t
+cs_resume_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   off_t offset, dict_t *xattr_req)
+{
+    cs_local_t *local = NULL;
+    int ret = 0;
+
+    local = frame->local;
+
+    ret = cs_resume_postprocess(this, frame, loc->inode);
+    if (ret) {
+        goto unwind;
+    }
+
+    cs_inodelk_unlock(frame);
+
+    STACK_WIND(frame, cs_truncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->truncate, loc, offset,
+               local->xattr_req);
+
+    return 0;
+
+unwind:
+    cs_inodelk_unlock(frame);
+
+    cs_common_cbk(frame);
+
+    return 0;
+}
+
+int32_t
+cs_resume_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   dict_t *dict, int32_t flags, dict_t *xdata)
+{
+    cs_local_t *local = NULL;
+    cs_inode_ctx_t *ctx = NULL;
+    gf_cs_obj_state state = GF_CS_ERROR;
+
+    local = frame->local;
+
+    __cs_inode_ctx_get(this, loc->inode, &ctx);
+
+    state = __cs_get_file_state(loc->inode, ctx);
+
+    if (state == GF_CS_ERROR) {
+        /* file is already remote */
+        local->op_ret = -1;
+        local->op_errno = EINVAL;
+        gf_msg(this->name, GF_LOG_WARNING, 0, 0,
+               "file %s , could not figure file state", loc->path);
+        goto unwind;
+    }
+
+    if (state == GF_CS_REMOTE) {
+        /* file is already remote */
+        local->op_ret = -1;
+        local->op_errno = EINVAL;
+        gf_msg(this->name, GF_LOG_WARNING, 0, EINVAL,
+               "file %s is already remote", loc->path);
+        goto unwind;
+    }
+
+    if (state == GF_CS_DOWNLOADING) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, 0,
+               " file is in downloading state.");
+        local->op_ret = -1;
+        local->op_errno = EINVAL;
+        goto unwind;
+    }
+
+    STACK_WIND(frame, cs_setxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setxattr, loc, dict, flags,
+               local->xattr_req);
+
+    return 0;
+unwind:
+    cs_inodelk_unlock(frame);
+
+    cs_common_cbk(frame);
+
+    return 0;
+}
+
+gf_cs_obj_state
+__cs_get_file_state(inode_t *inode, cs_inode_ctx_t *ctx)
+{
+    gf_cs_obj_state state = -1;
+
+    if (!ctx)
+        return GF_CS_ERROR;
+
+    LOCK(&inode->lock);
+    {
+        state = ctx->state;
+    }
+    UNLOCK(&inode->lock);
+
+    return state;
+}
+
+void
+__cs_inode_ctx_get(xlator_t *this, inode_t *inode, cs_inode_ctx_t **ctx)
+{
+    uint64_t ctxint = 0;
+    int ret = 0;
+
+    LOCK(&inode->lock);
+    {
+        ret = __inode_ctx_get(inode, this, &ctxint);
+    }
+    UNLOCK(&inode->lock);
+
+    if (ret)
+        *ctx = NULL;
+    else
+        *ctx = (cs_inode_ctx_t *)(uintptr_t)ctxint;
+
+    return;
+}
+
+int
+__cs_inode_ctx_update(xlator_t *this, inode_t *inode, uint64_t val)
+{
+    cs_inode_ctx_t *ctx = NULL;
+    uint64_t ctxint = 0;
+    int ret = 0;
+
+    LOCK(&inode->lock);
+    {
+        ret = __inode_ctx_get(inode, this, &ctxint);
+        if (ret) {
+            ctx = GF_CALLOC(1, sizeof(*ctx), gf_cs_mt_cs_inode_ctx_t);
+            if (!ctx) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, 0, "ctx allocation failed");
+                ret = -1;
+                goto out;
+            }
+
+            ctx->state = val;
+
+            ctxint = (uint64_t)(uintptr_t)ctx;
+
+            ret = __inode_ctx_set(inode, this, &ctxint);
+            if (ret) {
+                GF_FREE(ctx);
+                goto out;
+            }
+        } else {
+            ctx = (cs_inode_ctx_t *)(uintptr_t)ctxint;
+
+            ctx->state = val;
+        }
+    }
+
+out:
+    UNLOCK(&inode->lock);
+
+    return ret;
+}
+
+int
+cs_inode_ctx_reset(xlator_t *this, inode_t *inode)
+{
+    cs_inode_ctx_t *ctx = NULL;
+    uint64_t ctxint = 0;
+
+    inode_ctx_del(inode, this, &ctxint);
+    if (!ctxint) {
+        return 0;
+    }
+
+    ctx = (cs_inode_ctx_t *)(uintptr_t)ctxint;
+
+    GF_FREE(ctx);
+    return 0;
+}
+
+int
+cs_resume_postprocess(xlator_t *this, call_frame_t *frame, inode_t *inode)
+{
+    cs_local_t *local = NULL;
+    gf_cs_obj_state state = -1;
+    cs_inode_ctx_t *ctx = NULL;
+    int ret = 0;
+
+    local = frame->local;
+    if (!local) {
+        ret = -1;
+        goto out;
+    }
+
+    __cs_inode_ctx_get(this, inode, &ctx);
+
+    state = __cs_get_file_state(inode, ctx);
+    if (state == GF_CS_ERROR) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "status is GF_CS_ERROR."
+               " Aborting write");
+        local->op_ret = -1;
+        local->op_errno = EREMOTE;
+        ret = -1;
+        goto out;
+    }
+
+    if (state == GF_CS_REMOTE || state == GF_CS_DOWNLOADING) {
+        gf_msg_debug(this->name, 0, "status is %d", state);
+        ret = cs_download(frame);
+        if (ret == 0) {
+            gf_msg_debug(this->name, 0, "Winding for Final Write");
+        } else {
+            gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+                   " download failed, unwinding writev");
+            local->op_ret = -1;
+            local->op_errno = EREMOTE;
+            ret = -1;
+        }
+    }
+out:
+    return ret;
+}
+
+int32_t
+cs_fdctx_to_dict(xlator_t *this, fd_t *fd, dict_t *dict)
+{
+    return 0;
+}
+
+int32_t
+cs_inode(xlator_t *this)
+{
+    return 0;
+}
+
+int32_t
+cs_inode_to_dict(xlator_t *this, dict_t *dict)
+{
+    return 0;
+}
+
+int32_t
+cs_history(xlator_t *this)
+{
+    return 0;
+}
+
+int32_t
+cs_fd(xlator_t *this)
+{
+    return 0;
+}
+
+int32_t
+cs_fd_to_dict(xlator_t *this, dict_t *dict)
+{
+    return 0;
+}
+
+int32_t
+cs_fdctx(xlator_t *this, fd_t *fd)
+{
+    return 0;
+}
+
+int32_t
+cs_inodectx(xlator_t *this, inode_t *ino)
+{
+    return 0;
+}
+
+int32_t
+cs_inodectx_to_dict(xlator_t *this, inode_t *ino, dict_t *dict)
+{
+    return 0;
+}
+
+int32_t
+cs_priv_to_dict(xlator_t *this, dict_t *dict, char *brickname)
+{
+    return 0;
+}
+
+int32_t
+cs_priv(xlator_t *this)
+{
+    return 0;
+}
+
+int
+cs_notify(xlator_t *this, int event, void *data, ...)
+{
+    return default_notify(this, event, data);
+}
+
+struct xlator_fops cs_fops = {
+    .stat = cs_stat,
+    .readdirp = cs_readdirp,
+    .truncate = cs_truncate,
+    .seek = cs_seek,
+    .statfs = cs_statfs,
+    .fallocate = cs_fallocate,
+    .discard = cs_discard,
+    .getxattr = cs_getxattr,
+    .writev = cs_writev,
+    .setxattr = cs_setxattr,
+    .fgetxattr = cs_fgetxattr,
+    .lookup = cs_lookup,
+    .fsetxattr = cs_fsetxattr,
+    .readv = cs_readv,
+    .ftruncate = cs_ftruncate,
+    .rchecksum = cs_rchecksum,
+    .unlink = cs_unlink,
+    .open = cs_open,
+    .fstat = cs_fstat,
+    .zerofill = cs_zerofill,
+};
+
+struct xlator_cbks cs_cbks = {
+    .forget = cs_forget,
+};
+
+struct xlator_dumpops cs_dumpops = {
+    .fdctx_to_dict = cs_fdctx_to_dict,
+    .inode = cs_inode,
+    .inode_to_dict = cs_inode_to_dict,
+    .history = cs_history,
+    .fd = cs_fd,
+    .fd_to_dict = cs_fd_to_dict,
+    .fdctx = cs_fdctx,
+    .inodectx = cs_inodectx,
+    .inodectx_to_dict = cs_inodectx_to_dict,
+    .priv_to_dict = cs_priv_to_dict,
+    .priv = cs_priv,
+};
+
+struct volume_options cs_options[] = {
+    {.key = {"cloudsync-storetype"},
+     .type = GF_OPTION_TYPE_STR,
+     .description = "Defines which remote store is enabled"},
+    {.key = {"cloudsync-remote-read"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .description = "Defines a remote read fop when on"},
+    {.key = {"cloudsync-store-id"},
+     .type = GF_OPTION_TYPE_STR,
+     .description = "Defines a volume wide store id"},
+    {.key = {"cloudsync-product-id"},
+     .type = GF_OPTION_TYPE_STR,
+     .description = "Defines a volume wide product id"},
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = cs_init,
+    .fini = cs_fini,
+    .notify = cs_notify,
+    .reconfigure = cs_reconfigure,
+    .mem_acct_init = cs_mem_acct_init,
+    .dumpops = &cs_dumpops,
+    .fops = &cs_fops,
+    .cbks = &cs_cbks,
+    .options = cs_options,
+    .identifier = "cloudsync",
+    .category = GF_TECH_PREVIEW,
+};
diff --git a/xlators/features/cloudsync/src/cloudsync.h b/xlators/features/cloudsync/src/cloudsync.h
new file mode 100644
index 00000000000..d24141978d6
--- /dev/null
+++ b/xlators/features/cloudsync/src/cloudsync.h
@@ -0,0 +1,123 @@
+/*
+ *   Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com>
+ *   This file is part of GlusterFS.
+ *
+ *   This file is licensed to you under your choice of the GNU Lesser
+ *   General Public License, version 3 or any later version (LGPLv3 or
+ *   later), or the GNU General Public License, version 2 (GPLv2), in all
+ *   cases as published by the Free Software Foundation.
+ */
+
+#ifndef __CLOUDSYNC_H__
+#define __CLOUDSYNC_H__
+
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/syncop.h>
+#include <glusterfs/call-stub.h>
+#include "cloudsync-common.h"
+#include "cloudsync-autogen-fops.h"
+
+#define ALIGN_SIZE 4096
+#define CS_LOCK_DOMAIN "cs.protect.file.stat"
+typedef struct cs_dlstore {
+    off_t off;
+    struct iovec *vector;
+    int32_t count;
+    struct iobref *iobref;
+    uint32_t flags;
+} cs_dlstore;
+
+typedef struct cs_inode_ctx {
+    cs_loc_xattr_t locxattr;
+    gf_cs_obj_state state;
+} cs_inode_ctx_t;
+
+struct cs_plugin {
+    char *name;        /* store name */
+    char *library;     /* library to load for the given store */
+    char *description; /* description about the store */
+};
+
+cs_local_t *
+cs_local_init(xlator_t *this, call_frame_t *frame, loc_t *loc, fd_t *fd,
+              glusterfs_fop_t fop);
+
+int
+locate_and_execute(call_frame_t *frame);
+
+int32_t
+cs_resume_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   dict_t *dict, int32_t flags, dict_t *xdata);
+
+int32_t
+cs_inodelk_unlock_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata);
+
+size_t
+cs_write_callback(void *lcurlbuf, size_t size, size_t nitems, void *frame);
+
+void
+cs_common_cbk(call_frame_t *frame);
+
+gf_boolean_t
+cs_is_file_remote(struct iatt *stbuf, dict_t *xattr);
+
+int32_t
+cs_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata);
+int
+cs_build_loc(loc_t *loc, call_frame_t *frame);
+
+int
+cs_blocking_inodelk_cbk(call_frame_t *lock_frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata);
+
+int
+cs_read_authinfo(xlator_t *this);
+
+int
+__cs_inode_ctx_update(xlator_t *this, inode_t *inode, uint64_t val);
+
+int
+cs_inode_ctx_reset(xlator_t *this, inode_t *inode);
+
+void
+__cs_inode_ctx_get(xlator_t *this, inode_t *inode, cs_inode_ctx_t **ctx);
+
+gf_cs_obj_state
+__cs_get_file_state(inode_t *inode, cs_inode_ctx_t *ctx);
+
+int
+cs_inodelk_unlock(call_frame_t *main_frame);
+
+int
+cs_resume_postprocess(xlator_t *this, call_frame_t *frame, inode_t *inode);
+
+int32_t
+cs_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                struct iatt *postbuf, dict_t *xdata);
+int32_t
+cs_resume_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   off_t offset, dict_t *xattr_req);
+
+int32_t
+cs_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, struct iovec *vector, int32_t count,
+             struct iatt *stbuf, struct iobref *iobref, dict_t *xdata);
+int32_t
+cs_resume_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                off_t offset, uint32_t flags, dict_t *xdata);
+int32_t
+cs_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+         off_t offset, uint32_t flags, dict_t *xdata);
+
+int
+cs_resume_remote_readv_postprocess(xlator_t *this, call_frame_t *frame,
+                                   inode_t *inode, off_t offset, size_t size,
+                                   uint32_t flags);
+int
+cs_serve_readv(call_frame_t *frame, off_t offset, size_t size, uint32_t flags);
+#endif /* __CLOUDSYNC_H__ */
diff --git a/xlators/features/compress/Makefile.am b/xlators/features/compress/Makefile.am
new file mode 100644
index 00000000000..a985f42a877
--- /dev/null
+++ b/xlators/features/compress/Makefile.am
@@ -0,0 +1,3 @@
+SUBDIRS = src
+
+CLEANFILES =
diff --git a/xlators/features/compress/src/Makefile.am b/xlators/features/compress/src/Makefile.am
new file mode 100644
index 00000000000..98271a9f3fc
--- /dev/null
+++ b/xlators/features/compress/src/Makefile.am
@@ -0,0 +1,19 @@
+xlator_LTLIBRARIES = cdc.la
+
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+noinst_HEADERS = cdc.h cdc-mem-types.h
+
+cdc_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+cdc_la_SOURCES = cdc.c cdc-helper.c
+cdc_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la $(ZLIB_LIBS)
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src \
+	-fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -D$(GF_HOST_OS) \
+	$(LIBZ_CFLAGS)
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/compress/src/cdc-helper.c b/xlators/features/compress/src/cdc-helper.c
new file mode 100644
index 00000000000..f973ff56cf5
--- /dev/null
+++ b/xlators/features/compress/src/cdc-helper.c
@@ -0,0 +1,527 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/syscall.h>
+
+#include "cdc.h"
+#include "cdc-mem-types.h"
+
+#ifdef HAVE_LIB_Z
+#include "zlib.h"
+#endif
+
+#ifdef HAVE_LIB_Z
+/* gzip header looks something like this
+ * (RFC 1950)
+ *
+ * +---+---+---+---+---+---+---+---+---+---+
+ * |ID1|ID2|CM |FLG|     MTIME     |XFL|OS |
+ * +---+---+---+---+---+---+---+---+---+---+
+ *
+ * Data is usually sent without this header i.e
+ * Data sent = <compressed-data> + trailer(8)
+ * The trailer contains the checksum.
+ *
+ * gzip_header is added only during debugging.
+ * Refer to the function cdc_dump_iovec_to_disk
+ */
+static const char gzip_header[10] = {'\037', '\213', Z_DEFLATED,  0, 0, 0, 0,
+                                     0,      0,      GF_CDC_OS_ID};
+
+static int32_t
+cdc_next_iovec(xlator_t *this, cdc_info_t *ci)
+{
+    int ret = -1;
+
+    ci->ncount++;
+    /* check for iovec overflow -- should not happen */
+    if (ci->ncount == MAX_IOVEC) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Zlib output buffer overflow"
+               " ->ncount (%d) | ->MAX_IOVEC (%d)",
+               ci->ncount, MAX_IOVEC);
+        goto out;
+    }
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+static void
+cdc_put_long(unsigned char *string, unsigned long x)
+{
+    string[0] = (unsigned char)(x & 0xff);
+    string[1] = (unsigned char)((x & 0xff00) >> 8);
+    string[2] = (unsigned char)((x & 0xff0000) >> 16);
+    string[3] = (unsigned char)((x & 0xff000000) >> 24);
+}
+
+static unsigned long
+cdc_get_long(unsigned char *buf)
+{
+    return ((unsigned long)buf[0]) | (((unsigned long)buf[1]) << 8) |
+           (((unsigned long)buf[2]) << 16) | (((unsigned long)buf[3]) << 24);
+}
+
+static int32_t
+cdc_init_gzip_trailer(xlator_t *this, cdc_priv_t *priv, cdc_info_t *ci)
+{
+    int ret = -1;
+    char *buf = NULL;
+
+    ret = cdc_next_iovec(this, ci);
+    if (ret)
+        goto out;
+
+    buf = CURR_VEC(ci).iov_base = (char *)GF_CALLOC(1, GF_CDC_VALIDATION_SIZE,
+                                                    gf_cdc_mt_gzip_trailer_t);
+
+    if (!CURR_VEC(ci).iov_base)
+        goto out;
+
+    CURR_VEC(ci).iov_len = GF_CDC_VALIDATION_SIZE;
+
+    cdc_put_long((unsigned char *)&buf[0], ci->crc);
+    cdc_put_long((unsigned char *)&buf[4], ci->stream.total_in);
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+static int32_t
+cdc_alloc_iobuf_and_init_vec(xlator_t *this, cdc_priv_t *priv, cdc_info_t *ci,
+                             int size)
+{
+    int ret = -1;
+    int alloc_len = 0;
+    struct iobuf *iobuf = NULL;
+
+    ret = cdc_next_iovec(this, ci);
+    if (ret)
+        goto out;
+
+    alloc_len = size ? size : ci->buffer_size;
+
+    iobuf = iobuf_get2(this->ctx->iobuf_pool, alloc_len);
+    if (!iobuf)
+        goto out;
+
+    ret = iobref_add(ci->iobref, iobuf);
+    if (ret)
+        goto out;
+
+    /* Initialize this iovec */
+    CURR_VEC(ci).iov_base = iobuf->ptr;
+    CURR_VEC(ci).iov_len = alloc_len;
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+static void
+cdc_init_zlib_output_stream(cdc_priv_t *priv, cdc_info_t *ci, int size)
+{
+    ci->stream.next_out = (unsigned char *)CURR_VEC(ci).iov_base;
+    ci->stream.avail_out = size ? size : ci->buffer_size;
+}
+
+/* This routine is for testing and debugging only.
+ * Data written = header(10) + <compressed-data> + trailer(8)
+ * So each gzip dump file is at least 18 bytes in size.
+ */
+void
+cdc_dump_iovec_to_disk(xlator_t *this, cdc_info_t *ci, const char *file)
+{
+    int i = 0;
+    int fd = 0;
+    size_t written = 0;
+    size_t total_written = 0;
+
+    fd = open(file, O_WRONLY | O_CREAT | O_TRUNC, 0777);
+    if (fd < 0) {
+        gf_log(this->name, GF_LOG_ERROR, "Cannot open file: %s", file);
+        return;
+    }
+
+    written = sys_write(fd, (char *)gzip_header, 10);
+    total_written += written;
+    for (i = 0; i < ci->ncount; i++) {
+        written = sys_write(fd, (char *)ci->vec[i].iov_base,
+                            ci->vec[i].iov_len);
+        total_written += written;
+    }
+
+    gf_log(this->name, GF_LOG_DEBUG, "dump'd %zu bytes to %s", total_written,
+           GF_CDC_DEBUG_DUMP_FILE);
+
+    sys_close(fd);
+}
+
+static int32_t
+cdc_flush_libz_buffer(cdc_priv_t *priv, xlator_t *this, cdc_info_t *ci,
+                      int (*libz_func)(z_streamp, int), int flush)
+{
+    int32_t ret = Z_OK;
+    int done = 0;
+    unsigned int deflate_len = 0;
+
+    for (;;) {
+        deflate_len = ci->buffer_size - ci->stream.avail_out;
+
+        if (deflate_len != 0) {
+            CURR_VEC(ci).iov_len = deflate_len;
+
+            ret = cdc_alloc_iobuf_and_init_vec(this, priv, ci, 0);
+            if (ret) {
+                ret = Z_MEM_ERROR;
+                break;
+            }
+
+            /* Re-position Zlib output buffer */
+            cdc_init_zlib_output_stream(priv, ci, 0);
+        }
+
+        if (done) {
+            ci->ncount--;
+            break;
+        }
+
+        ret = libz_func(&ci->stream, flush);
+
+        if (ret == Z_BUF_ERROR) {
+            ret = Z_OK;
+            ci->ncount--;
+            break;
+        }
+
+        done = (ci->stream.avail_out != 0 || ret == Z_STREAM_END);
+
+        if (ret != Z_OK && ret != Z_STREAM_END)
+            break;
+    }
+
+    return ret;
+}
+
+static int32_t
+do_cdc_compress(struct iovec *vec, xlator_t *this, cdc_priv_t *priv,
+                cdc_info_t *ci)
+{
+    int ret = -1;
+
+    /* Initialize defalte */
+    ret = deflateInit2(&ci->stream, priv->cdc_level, Z_DEFLATED,
+                       priv->window_size, priv->mem_level, Z_DEFAULT_STRATEGY);
+
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR, "unable to init Zlib (retval: %d)",
+               ret);
+        goto out;
+    }
+
+    ret = cdc_alloc_iobuf_and_init_vec(this, priv, ci, 0);
+    if (ret)
+        goto out;
+
+    /* setup output buffer */
+    cdc_init_zlib_output_stream(priv, ci, 0);
+
+    /* setup input buffer */
+    ci->stream.next_in = (unsigned char *)vec->iov_base;
+    ci->stream.avail_in = vec->iov_len;
+
+    ci->crc = crc32(ci->crc, (const Bytef *)vec->iov_base, vec->iov_len);
+
+    gf_log(this->name, GF_LOG_DEBUG, "crc=%lu len=%d buffer_size=%d", ci->crc,
+           ci->stream.avail_in, ci->buffer_size);
+
+    /* compress !! */
+    while (ci->stream.avail_in != 0) {
+        if (ci->stream.avail_out == 0) {
+            CURR_VEC(ci).iov_len = ci->buffer_size;
+
+            ret = cdc_alloc_iobuf_and_init_vec(this, priv, ci, 0);
+            if (ret)
+                break;
+
+            /* Re-position Zlib output buffer */
+            cdc_init_zlib_output_stream(priv, ci, 0);
+        }
+
+        ret = deflate(&ci->stream, Z_NO_FLUSH);
+        if (ret != Z_OK)
+            break;
+    }
+
+out:
+    return ret;
+}
+
+int32_t
+cdc_compress(xlator_t *this, cdc_priv_t *priv, cdc_info_t *ci, dict_t **xdata)
+{
+    int ret = -1;
+    int i = 0;
+
+    ci->iobref = iobref_new();
+    if (!ci->iobref)
+        goto out;
+
+    if (!*xdata) {
+        *xdata = dict_new();
+        if (!*xdata) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "Cannot allocate xdata"
+                   " dict");
+            goto out;
+        }
+    }
+
+    /* data */
+    for (i = 0; i < ci->count; i++) {
+        ret = do_cdc_compress(&ci->vector[i], this, priv, ci);
+        if (ret != Z_OK)
+            goto deflate_cleanup_out;
+    }
+
+    /* flush zlib buffer */
+    ret = cdc_flush_libz_buffer(priv, this, ci, deflate, Z_FINISH);
+    if (!(ret == Z_OK || ret == Z_STREAM_END)) {
+        gf_log(this->name, GF_LOG_ERROR, "Compression Error: ret (%d)", ret);
+        ret = -1;
+        goto deflate_cleanup_out;
+    }
+
+    /* trailer */
+    ret = cdc_init_gzip_trailer(this, priv, ci);
+    if (ret)
+        goto deflate_cleanup_out;
+
+    gf_log(this->name, GF_LOG_DEBUG, "Compressed %ld to %ld bytes",
+           ci->stream.total_in, ci->stream.total_out);
+
+    ci->nbytes = ci->stream.total_out + GF_CDC_VALIDATION_SIZE;
+
+    /* set deflated canary value for identification */
+    ret = dict_set_int32(*xdata, GF_CDC_DEFLATE_CANARY_VAL, 1);
+    if (ret) {
+        /* Send uncompressed data if we can't _tell_ the client
+         * that deflated data is on it's way. So, we just log
+         * the failure and continue as usual.
+         */
+        gf_log(this->name, GF_LOG_ERROR,
+               "Data deflated, but could not set canary"
+               " value in dict for identification");
+    }
+
+    /* This is to be used in testing */
+    if (priv->debug) {
+        cdc_dump_iovec_to_disk(this, ci, GF_CDC_DEBUG_DUMP_FILE);
+    }
+
+deflate_cleanup_out:
+    (void)deflateEnd(&ci->stream);
+
+out:
+    return ret;
+}
+
+/* deflate content is checked by the presence of a canary
+ * value in the dict as the key
+ */
+static int32_t
+cdc_check_content_for_deflate(dict_t *xdata)
+{
+    return dict_get(xdata, GF_CDC_DEFLATE_CANARY_VAL) ? -1 : 0;
+}
+
+static unsigned long
+cdc_extract_crc(char *trailer)
+{
+    return cdc_get_long((unsigned char *)&trailer[0]);
+}
+
+static unsigned long
+cdc_extract_size(char *trailer)
+{
+    return cdc_get_long((unsigned char *)&trailer[4]);
+}
+
+static int32_t
+cdc_validate_inflate(cdc_info_t *ci, unsigned long crc, unsigned long len)
+{
+    return !((crc == ci->crc)
+             /* inflated length is hidden inside
+              * Zlib stream struct */
+             && (len == ci->stream.total_out));
+}
+
+static int32_t
+do_cdc_decompress(xlator_t *this, cdc_priv_t *priv, cdc_info_t *ci)
+{
+    int ret = -1;
+    int i = 0;
+    int len = 0;
+    char *inflte = NULL;
+    char *trailer = NULL;
+    struct iovec vec = {
+        0,
+    };
+    unsigned long computed_crc = 0;
+    unsigned long computed_len = 0;
+
+    ret = inflateInit2(&ci->stream, priv->window_size);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR, "Zlib: Unable to initialize inflate");
+        goto out;
+    }
+
+    vec = THIS_VEC(ci, 0);
+
+    trailer = (char *)(((char *)vec.iov_base) + vec.iov_len -
+                       GF_CDC_VALIDATION_SIZE);
+
+    /* CRC of uncompressed data */
+    computed_crc = cdc_extract_crc(trailer);
+
+    /* size of uncomrpessed data */
+    computed_len = cdc_extract_size(trailer);
+
+    gf_log(this->name, GF_LOG_DEBUG, "crc=%lu len=%lu buffer_size=%d",
+           computed_crc, computed_len, ci->buffer_size);
+
+    inflte = vec.iov_base;
+    len = vec.iov_len - GF_CDC_VALIDATION_SIZE;
+
+    /* allocate buffer of the original length of the data */
+    ret = cdc_alloc_iobuf_and_init_vec(this, priv, ci, 0);
+    if (ret)
+        goto out;
+
+    /* setup output buffer */
+    cdc_init_zlib_output_stream(priv, ci, 0);
+
+    /* setup input buffer */
+    ci->stream.next_in = (unsigned char *)inflte;
+    ci->stream.avail_in = len;
+
+    while (ci->stream.avail_in != 0) {
+        if (ci->stream.avail_out == 0) {
+            CURR_VEC(ci).iov_len = ci->buffer_size;
+
+            ret = cdc_alloc_iobuf_and_init_vec(this, priv, ci, 0);
+            if (ret)
+                break;
+
+            /* Re-position Zlib output buffer */
+            cdc_init_zlib_output_stream(priv, ci, 0);
+        }
+
+        ret = inflate(&ci->stream, Z_NO_FLUSH);
+        if (ret == Z_STREAM_ERROR)
+            break;
+    }
+
+    /* flush zlib buffer */
+    ret = cdc_flush_libz_buffer(priv, this, ci, inflate, Z_SYNC_FLUSH);
+    if (!(ret == Z_OK || ret == Z_STREAM_END)) {
+        gf_log(this->name, GF_LOG_ERROR, "Decompression Error: ret (%d)", ret);
+        ret = -1;
+        goto out;
+    }
+
+    /* compute CRC of the uncompresses data to check for
+     * correctness */
+
+    for (i = 0; i < ci->ncount; i++) {
+        ci->crc = crc32(ci->crc, (const Bytef *)ci->vec[i].iov_base,
+                        ci->vec[i].iov_len);
+    }
+
+    /* validate inflated data */
+    ret = cdc_validate_inflate(ci, computed_crc, computed_len);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Checksum or length mismatched in inflated data");
+    }
+
+out:
+    return ret;
+}
+
+int32_t
+cdc_decompress(xlator_t *this, cdc_priv_t *priv, cdc_info_t *ci, dict_t *xdata)
+{
+    int32_t ret = -1;
+
+    /* check for deflate content */
+    if (!cdc_check_content_for_deflate(xdata)) {
+        gf_log(this->name, GF_LOG_DEBUG,
+               "Content not deflated, passing through ...");
+        goto passthrough_out;
+    }
+
+    ci->iobref = iobref_new();
+    if (!ci->iobref)
+        goto passthrough_out;
+
+    /* do we need to do this? can we assume that one iovec
+     * will hold per request data every time?
+     *
+     * server/client protocol seems to deal with a single
+     * iovec even if op_ret > 1M. So, it looks ok to
+     * assume that a single iovec will contain all the
+     * data (This saves us a lot from finding the trailer
+     * and the data since it could have been split-up onto
+     * two adjacent iovec's.
+     *
+     * But, in case this translator is loaded above quick-read
+     * for some reason, then it's entirely possible that we get
+     * multiple iovec's...
+     *
+     * This case (handled below) is not tested. (by loading the
+     * xlator below quick-read)
+     */
+
+    /* @@ I_HOPE_THIS_IS_NEVER_HIT */
+    if (ci->count > 1) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "unable to handle"
+               " multiple iovecs (%d in number)",
+               ci->count);
+        goto inflate_cleanup_out;
+        /* TODO: coallate all iovecs in one */
+    }
+
+    ret = do_cdc_decompress(this, priv, ci);
+    if (ret)
+        goto inflate_cleanup_out;
+
+    ci->nbytes = ci->stream.total_out;
+
+    gf_log(this->name, GF_LOG_DEBUG, "Inflated %ld to %ld bytes",
+           ci->stream.total_in, ci->stream.total_out);
+
+inflate_cleanup_out:
+    (void)inflateEnd(&ci->stream);
+
+passthrough_out:
+    return ret;
+}
+
+#endif
diff --git a/xlators/features/compress/src/cdc-mem-types.h b/xlators/features/compress/src/cdc-mem-types.h
new file mode 100644
index 00000000000..928afdd2efe
--- /dev/null
+++ b/xlators/features/compress/src/cdc-mem-types.h
@@ -0,0 +1,23 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __CDC_MEM_TYPES_H
+#define __CDC_MEM_TYPES_H
+
+#include <glusterfs/mem-types.h>
+
+enum gf_cdc_mem_types {
+    gf_cdc_mt_priv_t = gf_common_mt_end + 1,
+    gf_cdc_mt_vec_t = gf_common_mt_end + 2,
+    gf_cdc_mt_gzip_trailer_t = gf_common_mt_end + 3,
+    gf_cdc_mt_end = gf_common_mt_end + 4,
+};
+
+#endif
diff --git a/xlators/features/compress/src/cdc.c b/xlators/features/compress/src/cdc.c
new file mode 100644
index 00000000000..b0b51e914ed
--- /dev/null
+++ b/xlators/features/compress/src/cdc.c
@@ -0,0 +1,348 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <sys/uio.h>
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/logging.h>
+
+#include "cdc.h"
+#include "cdc-mem-types.h"
+
+static void
+cdc_cleanup_iobref(cdc_info_t *ci)
+{
+    assert(ci->iobref != NULL);
+    iobref_clear(ci->iobref);
+}
+
+int32_t
+cdc_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct iovec *vector, int32_t count,
+              struct iatt *stbuf, struct iobref *iobref, dict_t *xdata)
+{
+    int ret = -1;
+    cdc_priv_t *priv = NULL;
+    cdc_info_t ci = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("cdc", this, default_out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, default_out);
+
+    priv = this->private;
+
+    if (op_ret <= 0)
+        goto default_out;
+
+    if ((priv->min_file_size != 0) && (op_ret < priv->min_file_size))
+        goto default_out;
+
+    ci.count = count;
+    ci.ibytes = op_ret;
+    ci.vector = vector;
+    ci.buf = NULL;
+    ci.iobref = NULL;
+    ci.ncount = 0;
+    ci.crc = 0;
+    ci.buffer_size = GF_CDC_DEF_BUFFERSIZE;
+
+    /* A readv compresses on the server side and decompresses on the client side
+     */
+    if (priv->op_mode == GF_CDC_MODE_SERVER) {
+        ret = cdc_compress(this, priv, &ci, &xdata);
+    } else if (priv->op_mode == GF_CDC_MODE_CLIENT) {
+        ret = cdc_decompress(this, priv, &ci, xdata);
+    } else {
+        gf_log(this->name, GF_LOG_ERROR, "Invalid operation mode (%d)",
+               priv->op_mode);
+    }
+
+    if (ret)
+        goto default_out;
+
+    STACK_UNWIND_STRICT(readv, frame, ci.nbytes, op_errno, ci.vec, ci.ncount,
+                        stbuf, iobref, xdata);
+    cdc_cleanup_iobref(&ci);
+    return 0;
+
+default_out:
+    STACK_UNWIND_STRICT(readv, frame, op_ret, op_errno, vector, count, stbuf,
+                        iobref, xdata);
+    return 0;
+}
+
+int32_t
+cdc_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+          off_t offset, uint32_t flags, dict_t *xdata)
+{
+    fop_readv_cbk_t cbk = NULL;
+
+#ifdef HAVE_LIB_Z
+    cbk = cdc_readv_cbk;
+#else
+    cbk = default_readv_cbk;
+#endif
+    STACK_WIND(frame, cbk, FIRST_CHILD(this), FIRST_CHILD(this)->fops->readv,
+               fd, size, offset, flags, xdata);
+    return 0;
+}
+
+int32_t
+cdc_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+               struct iatt *postbuf, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(writev, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    return 0;
+}
+
+int32_t
+cdc_writev(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
+           int32_t count, off_t offset, uint32_t flags, struct iobref *iobref,
+           dict_t *xdata)
+{
+    int ret = -1;
+    cdc_priv_t *priv = NULL;
+    cdc_info_t ci = {
+        0,
+    };
+    size_t isize = 0;
+
+    GF_VALIDATE_OR_GOTO("cdc", this, err);
+    GF_VALIDATE_OR_GOTO(this->name, frame, err);
+
+    priv = this->private;
+
+    isize = iov_length(vector, count);
+
+    if (isize <= 0)
+        goto default_out;
+
+    if ((priv->min_file_size != 0) && (isize < priv->min_file_size))
+        goto default_out;
+
+    ci.count = count;
+    ci.ibytes = isize;
+    ci.vector = vector;
+    ci.buf = NULL;
+    ci.iobref = NULL;
+    ci.ncount = 0;
+    ci.crc = 0;
+    ci.buffer_size = GF_CDC_DEF_BUFFERSIZE;
+
+    /* A writev compresses on the client side and decompresses on the server
+     * side
+     */
+    if (priv->op_mode == GF_CDC_MODE_CLIENT) {
+        ret = cdc_compress(this, priv, &ci, &xdata);
+    } else if (priv->op_mode == GF_CDC_MODE_SERVER) {
+        ret = cdc_decompress(this, priv, &ci, xdata);
+    } else {
+        gf_log(this->name, GF_LOG_ERROR, "Invalid operation mode (%d) ",
+               priv->op_mode);
+    }
+
+    if (ret)
+        goto default_out;
+
+    STACK_WIND(frame, cdc_writev_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->writev, fd, ci.vec, ci.ncount, offset,
+               flags, iobref, xdata);
+
+    cdc_cleanup_iobref(&ci);
+    return 0;
+
+default_out:
+    STACK_WIND(frame, cdc_writev_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->writev, fd, vector, count, offset,
+               flags, iobref, xdata);
+    return 0;
+err:
+    STACK_UNWIND_STRICT(writev, frame, -1, EINVAL, NULL, NULL, NULL);
+    return 0;
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    if (!this)
+        return ret;
+
+    ret = xlator_mem_acct_init(this, gf_cdc_mt_end);
+
+    if (ret != 0) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Memory accounting init"
+               "failed");
+        return ret;
+    }
+
+    return ret;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    int ret = -1;
+    char *temp_str = NULL;
+    cdc_priv_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("cdc", this, err);
+
+    if (!this->children || this->children->next) {
+        gf_log(this->name, GF_LOG_ERROR, "Need subvolume == 1");
+        goto err;
+    }
+
+    if (!this->parents) {
+        gf_log(this->name, GF_LOG_WARNING, "Dangling volume. Check volfile");
+    }
+
+    priv = GF_CALLOC(1, sizeof(*priv), gf_cdc_mt_priv_t);
+    if (!priv) {
+        goto err;
+    }
+
+    /* Check if debug mode is turned on */
+    GF_OPTION_INIT("debug", priv->debug, bool, err);
+    if (priv->debug) {
+        gf_log(this->name, GF_LOG_DEBUG, "CDC debug option turned on");
+    }
+
+    /* Set Gzip Window Size */
+    GF_OPTION_INIT("window-size", priv->window_size, int32, err);
+    if ((priv->window_size > GF_CDC_MAX_WINDOWSIZE) ||
+        (priv->window_size < GF_CDC_DEF_WINDOWSIZE)) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "Invalid gzip window size (%d), using default",
+               priv->window_size);
+        priv->window_size = GF_CDC_DEF_WINDOWSIZE;
+    }
+
+    /* Set Gzip (De)Compression Level */
+    GF_OPTION_INIT("compression-level", priv->cdc_level, int32, err);
+    if (((priv->cdc_level < 1) || (priv->cdc_level > 9)) &&
+        (priv->cdc_level != GF_CDC_DEF_COMPRESSION)) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "Invalid gzip (de)compression level (%d),"
+               " using default",
+               priv->cdc_level);
+        priv->cdc_level = GF_CDC_DEF_COMPRESSION;
+    }
+
+    /* Set Gzip Memory Level */
+    GF_OPTION_INIT("mem-level", priv->mem_level, int32, err);
+    if ((priv->mem_level < 1) || (priv->mem_level > 9)) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "Invalid gzip memory level, using the default");
+        priv->mem_level = GF_CDC_DEF_MEMLEVEL;
+    }
+
+    /* Set min file size to enable compression */
+    GF_OPTION_INIT("min-size", priv->min_file_size, int32, err);
+
+    /* Mode of operation - Server/Client */
+    ret = dict_get_str(this->options, "mode", &temp_str);
+    if (ret) {
+        gf_log(this->name, GF_LOG_CRITICAL, "Operation mode not specified !!");
+        goto err;
+    }
+
+    if (GF_CDC_MODE_IS_CLIENT(temp_str)) {
+        priv->op_mode = GF_CDC_MODE_CLIENT;
+    } else if (GF_CDC_MODE_IS_SERVER(temp_str)) {
+        priv->op_mode = GF_CDC_MODE_SERVER;
+    } else {
+        gf_log(this->name, GF_LOG_CRITICAL,
+               "Bogus operation mode (%s) specified", temp_str);
+        goto err;
+    }
+
+    this->private = priv;
+    gf_log(this->name, GF_LOG_DEBUG, "CDC xlator loaded in (%s) mode",
+           temp_str);
+    return 0;
+
+err:
+    if (priv)
+        GF_FREE(priv);
+
+    return -1;
+}
+
+void
+fini(xlator_t *this)
+{
+    cdc_priv_t *priv = this->private;
+
+    if (priv)
+        GF_FREE(priv);
+    this->private = NULL;
+    return;
+}
+
+struct xlator_fops fops = {
+    .readv = cdc_readv,
+    .writev = cdc_writev,
+};
+
+struct xlator_cbks cbks = {};
+
+struct volume_options options[] = {
+    {.key = {"window-size"},
+     .default_value = "-15",
+     .type = GF_OPTION_TYPE_INT,
+     .description = "Size of the zlib history buffer."},
+    {.key = {"mem-level"},
+     .default_value = "8",
+     .type = GF_OPTION_TYPE_INT,
+     .description = "Memory allocated for internal compression state. "
+                    "1 uses minimum memory but is slow and reduces "
+                    "compression ratio; memLevel=9 uses maximum memory "
+                    "for optimal speed. The default value is 8."},
+    {.key = {"compression-level"},
+     .default_value = "-1",
+     .type = GF_OPTION_TYPE_INT,
+     .description = "Compression levels \n"
+                    "0 : no compression, 1 : best speed, \n"
+                    "9 : best compression, -1 : default compression "},
+    {.key = {"min-size"},
+     .default_value = "0",
+     .type = GF_OPTION_TYPE_INT,
+     .description = "Data is compressed only when its size exceeds this."},
+    {.key = {"mode"},
+     .value = {"server", "client"},
+     .type = GF_OPTION_TYPE_STR,
+     .description = "Set on the basis of where the xlator is loaded. "
+                    "This option should NOT be configured by user."},
+    {.key = {"debug"},
+     .default_value = "false",
+     .type = GF_OPTION_TYPE_BOOL,
+     .description = "This is used in testing. Will dump compressed data "
+                    "to disk as a gzip file."},
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {GD_OP_VERSION_3_9_0},
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "cdc",
+    .category = GF_TECH_PREVIEW,
+};
diff --git a/xlators/features/compress/src/cdc.h b/xlators/features/compress/src/cdc.h
new file mode 100644
index 00000000000..cb87b06a989
--- /dev/null
+++ b/xlators/features/compress/src/cdc.h
@@ -0,0 +1,99 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __CDC_H
+#define __CDC_H
+
+#ifdef HAVE_LIB_Z
+#include "zlib.h"
+#endif
+
+#include <glusterfs/xlator.h>
+
+#ifndef MAX_IOVEC
+#define MAX_IOVEC 16
+#endif
+
+typedef struct cdc_priv {
+    int window_size;
+    int mem_level;
+    int cdc_level;
+    int min_file_size;
+    int op_mode;
+    gf_boolean_t debug;
+    gf_lock_t lock;
+} cdc_priv_t;
+
+typedef struct cdc_info {
+    /* input bits */
+    int count;
+    int32_t ibytes;
+    struct iovec *vector;
+    struct iatt *buf;
+
+    /* output bits */
+    int ncount;
+    int nbytes;
+    int buffer_size;
+    struct iovec vec[MAX_IOVEC];
+    struct iobref *iobref;
+
+    /* zlib bits */
+#ifdef HAVE_LIB_Z
+    z_stream stream;
+#endif
+    unsigned long crc;
+} cdc_info_t;
+
+#define NVEC(ci) (ci->ncount - 1)
+#define CURR_VEC(ci) ci->vec[ci->ncount - 1]
+#define THIS_VEC(ci, i) ci->vector[i]
+
+/* Gzip defaults */
+#define GF_CDC_DEF_WINDOWSIZE -15 /* default value */
+#define GF_CDC_MAX_WINDOWSIZE -8  /* max value     */
+
+#ifdef HAVE_LIB_Z
+#define GF_CDC_DEF_COMPRESSION Z_DEFAULT_COMPRESSION
+#else
+#define GF_CDC_DEF_COMPRESSION -1
+#endif
+
+#define GF_CDC_DEF_MEMLEVEL 8
+#define GF_CDC_DEF_BUFFERSIZE 262144  // 256K - default compression buffer size
+
+/* Operation mode
+ * If xlator is loaded on client, readv decompresses and writev compresses
+ * If xlator is loaded on server, readv compresses and writev decompresses
+ */
+#define GF_CDC_MODE_CLIENT 0
+#define GF_CDC_MODE_SERVER 1
+
+/* min size of data to do cmpression
+ * 0 == compress even 1byte
+ */
+#define GF_CDC_MIN_CHUNK_SIZE 0
+
+#define GF_CDC_VALIDATION_SIZE 8
+
+#define GF_CDC_OS_ID 0xFF
+#define GF_CDC_DEFLATE_CANARY_VAL "deflate"
+#define GF_CDC_DEBUG_DUMP_FILE "/tmp/cdcdump.gz"
+
+#define GF_CDC_MODE_IS_CLIENT(m) (strcmp(m, "client") == 0)
+
+#define GF_CDC_MODE_IS_SERVER(m) (strcmp(m, "server") == 0)
+
+int32_t
+cdc_compress(xlator_t *this, cdc_priv_t *priv, cdc_info_t *ci, dict_t **xdata);
+int32_t
+cdc_decompress(xlator_t *this, cdc_priv_t *priv, cdc_info_t *ci, dict_t *xdata);
+
+#endif
diff --git a/xlators/features/filter/src/Makefile.am b/xlators/features/filter/src/Makefile.am
deleted file mode 100644
index d473b9ea16d..00000000000
--- a/xlators/features/filter/src/Makefile.am
+++ /dev/null
@@ -1,15 +0,0 @@
-xlator_LTLIBRARIES = filter.la
-xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/testing/features
-
-filter_la_LDFLAGS = -module -avoidversion
-
-filter_la_SOURCES = filter.c
-filter_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la 
-
-noinst_HEADERS = filter-mem-types.h
-
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS) \
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
-
-CLEANFILES = 
-
diff --git a/xlators/features/filter/src/filter-mem-types.h b/xlators/features/filter/src/filter-mem-types.h
deleted file mode 100644
index db41a88776e..00000000000
--- a/xlators/features/filter/src/filter-mem-types.h
+++ /dev/null
@@ -1,30 +0,0 @@
-/*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef __FILTER_MEM_TYPES_H__
-#define __FILTER_MEM_TYPES_H__
-
-#include "mem-types.h"
-
-enum gf_filter_mem_types_ {
-        gf_filter_mt_gf_filter = gf_common_mt_end + 1,
-        gf_filter_mt_end
-};
-#endif
-
diff --git a/xlators/features/filter/src/filter.c b/xlators/features/filter/src/filter.c
deleted file mode 100644
index a813e0c631b..00000000000
--- a/xlators/features/filter/src/filter.c
+++ /dev/null
@@ -1,1744 +0,0 @@
-/*
-  Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <unistd.h>
-
-#include "glusterfs.h"
-#include "logging.h"
-#include "dict.h"
-#include "xlator.h"
-#include "filter-mem-types.h"
-
-#define GF_FILTER_NOBODY_UID         65534
-#define GF_FILTER_NOBODY_GID         65534
-#define GF_FILTER_ROOT_UID           0
-#define GF_FILTER_ROOT_GID           0
-
-#define GF_MAXIMUM_FILTERING_ALLOWED 32
-
-/*
-  option root-filtering on (off by default)
-  option translate-uid <uid-range=newuid,uid=newuid>
-  option translate-gid <gid-range=newgid,gid=newgid>
-  option read-only <yes|true>
-  option fixed-uid <uid>
-  option fixed-gid <gid>
-  option filter-uid <uid-range,uid>
-  option filter-gid <gid-range,gid> // not supported yet
-
-*/
-
-struct gf_filter {
-	/* Flags */
-	gf_boolean_t complete_read_only;
-	char fixed_uid_set;
-	char fixed_gid_set;
-	char partial_filter;
-
-	/* Options */
-	/* Mapping/Filtering/Translate whatever you want to call */
-	int translate_num_uid_entries;
-	int translate_num_gid_entries;
-	int translate_input_uid[GF_MAXIMUM_FILTERING_ALLOWED][2];
-	int translate_output_uid[GF_MAXIMUM_FILTERING_ALLOWED];
-	int translate_input_gid[GF_MAXIMUM_FILTERING_ALLOWED][2];
-	int translate_output_gid[GF_MAXIMUM_FILTERING_ALLOWED];
-
-	/* Fixed uid/gid */
-	int fixed_uid;
-	int fixed_gid;
-
-	/* Filter */
-	int filter_num_uid_entries;
-	int filter_num_gid_entries;
-	int filter_input_uid[GF_MAXIMUM_FILTERING_ALLOWED][2];
-	int filter_input_gid[GF_MAXIMUM_FILTERING_ALLOWED][2];
-	
-};
-
-/* update_frame: The main logic of the whole translator.
-   Return values:
-   0: no change
-   // TRANSLATE
-   1: only uid changed 
-   2: only gid changed
-   3: both uid/gid changed
-   // FILTER
-   4: uid in filter range
-   5: gid in filter range  // not supported yet
-   6: complete fs is readonly
-*/
-
-#define GF_FILTER_NO_CHANGE    0
-#define	GF_FILTER_MAP_UID      1
-#define GF_FILTER_MAP_GID      2
-#define GF_FILTER_MAP_BOTH     3
-#define GF_FILTER_FILTER_UID   4
-#define GF_FILTER_FILTER_GID   5
-#define GF_FILTER_RO_FS        6
-
-static int32_t
-update_frame (call_frame_t *frame,
-	      inode_t *inode,
-	      struct gf_filter *filter)
-{
-	uid_t    uid = 0;
-	int32_t  idx = 0;
-	int32_t  ret = 0;
-	int32_t  dictret = 0;
-	uint64_t tmp_uid = 0;
-	
-	for (idx = 0; idx < filter->translate_num_uid_entries; idx++) {
-		if ((frame->root->uid >=filter->translate_input_uid[idx][0]) &&
-		    (frame->root->uid <=filter->translate_input_uid[idx][1])) {
-			dictret = inode_ctx_get (inode, frame->this, &tmp_uid);
-			uid = (uid_t)tmp_uid;
-			if (dictret == 0) {
-				if (frame->root->uid != uid)
-					ret = GF_FILTER_MAP_UID;
-			} else {
-				ret = GF_FILTER_MAP_UID;
-			}
-			break;
-		}
-	}
-	
-	for (idx = 0; idx < filter->translate_num_gid_entries; idx++) {
-		if ((frame->root->gid >=filter->translate_input_gid[idx][0]) &&
-		    (frame->root->gid <=filter->translate_input_gid[idx][1])) {
-			if (ret == GF_FILTER_NO_CHANGE) 
-				ret = GF_FILTER_MAP_GID;
-			else 
-				ret = GF_FILTER_MAP_BOTH;
-			break;
-		}
-	}
-
-
-	if (filter->complete_read_only)
-		return GF_FILTER_RO_FS;
-	
-	if (filter->partial_filter) {
-		dictret = inode_ctx_get (inode, frame->this, &tmp_uid);
-		uid = (uid_t)tmp_uid;
-		if (dictret != -1) {
-			for (idx = 0; idx < filter->filter_num_uid_entries; 
-			     idx++) {
-				if ((uid >=filter->filter_input_uid[idx][0]) &&
-				    (uid <=filter->filter_input_uid[idx][1])) {
-					return GF_FILTER_FILTER_UID;
-				}
-			}
-		}
-	}
-
-	return ret;
-}
-
-/* if 'root' don't change the uid/gid */
-static int32_t
-update_stat (struct iatt *stbuf,
-	     struct gf_filter *filter)
-{
-	int32_t idx = 0;
-	for (idx = 0; idx < filter->translate_num_uid_entries; idx++) {
-		if (stbuf->ia_uid == GF_FILTER_ROOT_UID)
-			continue;
-		if ((stbuf->ia_uid >= filter->translate_input_uid[idx][0]) &&
-		    (stbuf->ia_uid <= filter->translate_input_uid[idx][1])) {
-			stbuf->ia_uid = filter->translate_output_uid[idx];
-			break;
-		}
-	}
-	
-	for (idx = 0; idx < filter->translate_num_gid_entries; idx++) {
-		if (stbuf->ia_gid == GF_FILTER_ROOT_GID)
-			continue;
-		if ((stbuf->ia_gid >= filter->translate_input_gid[idx][0]) &&
-		    (stbuf->ia_gid <= filter->translate_input_gid[idx][1])) {
-			stbuf->ia_gid = filter->translate_output_gid[idx];
-			break;
-		}
-	}
-
-	if (filter->fixed_uid_set) {
-		stbuf->ia_uid = filter->fixed_uid;
-	}
-
-	if (filter->fixed_gid_set) {
-		stbuf->ia_gid = filter->fixed_gid;
-	}
-	
-	return 0;
-}
-
-static int32_t 
-filter_lookup_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-		   inode_t *inode,
-		   struct iatt *buf,
-		   dict_t *dict,
-                   struct iatt *postparent)
-{
-	int ret = 0;
-	if (op_ret >= 0) {
-		update_stat (buf, this->private);
-		ret = inode_ctx_put (inode, this, (uint64_t)(long)buf->ia_uid);
-		if (ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"couldn't set context");
-		}
-
-		update_stat (postparent, this->private);
-	}
-	STACK_UNWIND (frame, op_ret, op_errno, inode, buf, dict, postparent);
-	return 0;
-}
-
-int32_t 
-filter_lookup (call_frame_t *frame,
-	       xlator_t *this,
-	       loc_t *loc,
-	       dict_t *xattr_req)
-{
-	STACK_WIND (frame,
-		    filter_lookup_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->lookup,
-		    loc,
-		    xattr_req);
-	return 0;
-}
-
-
-static int32_t
-filter_stat_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 struct iatt *buf)
-{
-	if (op_ret >= 0) {
-		update_stat (buf, this->private);
-	}
-	STACK_UNWIND (frame, op_ret, op_errno, buf);
-	return 0;
-}
-
-int32_t
-filter_stat (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc)
-{
-	STACK_WIND (frame,
-		    filter_stat_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->stat,
-		    loc);
-	return 0;
-}
-
-static int32_t
-filter_setattr_cbk (call_frame_t *frame,
-                    void *cookie,
-                    xlator_t *this,
-                    int32_t op_ret,
-                    int32_t op_errno,
-                    struct iatt *preop,
-                    struct iatt *postop)
-{
-	if (op_ret >= 0) {
-		update_stat (preop, this->private);
-		update_stat (postop, this->private);
-	}
-	STACK_UNWIND (frame, op_ret, op_errno, preop, postop);
-	return 0;
-}
-
-int32_t
-filter_setattr (call_frame_t *frame,
-                xlator_t *this,
-                loc_t *loc,
-                struct iatt *stbuf,
-                int32_t valid)
-{
-	int32_t ret = 0;
-	ret = update_frame (frame, loc->inode, this->private);
-	switch (ret) {
-	case GF_FILTER_MAP_UID:
-		if (loc->inode->st_mode & S_IWGRP)
-			break;
-	case GF_FILTER_MAP_BOTH:
-		if (loc->inode->st_mode & S_IWOTH)
-			break;
-		gf_log (this->name, GF_LOG_DEBUG,
-                        "%s: returning permission denied", loc->path);
-		STACK_UNWIND (frame, -1, EPERM, NULL, NULL, NULL);
-		return 0;
-
-	case GF_FILTER_FILTER_UID:
-	case GF_FILTER_FILTER_GID:
-	case GF_FILTER_RO_FS:
-		STACK_UNWIND (frame, -1, EROFS, NULL, NULL);
-		return 0;
-	default:
-		break;
-	}
-
-	STACK_WIND (frame,
-		    filter_setattr_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->setattr,
-		    loc,
-		    stbuf, valid);
-	return 0;
-}
-
-static int32_t
-filter_fsetattr_cbk (call_frame_t *frame,
-                     void *cookie,
-                     xlator_t *this,
-                     int32_t op_ret,
-                     int32_t op_errno,
-                     struct iatt *preop,
-                     struct iatt *postop)
-{
-	if (op_ret >= 0) {
-                update_stat (preop, this->private);
-		update_stat (postop, this->private);
-	}
-	STACK_UNWIND (frame,
-		      op_ret,
-		      op_errno,
-		      preop, postop);
-	return 0;
-}
-
-int32_t
-filter_fsetattr (call_frame_t *frame,
-                 xlator_t *this,
-                 fd_t *fd,
-                 struct iatt *stbuf,
-                 int32_t valid)
-{
-	STACK_WIND (frame,
-		    filter_fsetattr_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->fsetattr,
-		    fd,
-		    stbuf, valid);
-	return 0;
-}
-
-
-static int32_t
-filter_truncate_cbk (call_frame_t *frame,
-		     void *cookie,
-		     xlator_t *this,
-		     int32_t op_ret,
-		     int32_t op_errno,
-		     struct iatt *prebuf,
-                     struct iatt *postbuf)
-{
-	if (op_ret >= 0) {
-		update_stat (prebuf, this->private);
-		update_stat (postbuf, this->private);
-	}
-	STACK_UNWIND (frame, op_ret, op_errno, prebuf, postbuf);
-	return 0;
-}
-
-int32_t
-filter_truncate (call_frame_t *frame,
-		 xlator_t *this,
-		 loc_t *loc,
-		 off_t offset)
-{
-	int32_t ret = 0;
-	ret = update_frame (frame, loc->inode, this->private);
-	switch (ret) {
-	case GF_FILTER_MAP_UID:
-		if (loc->inode->st_mode & S_IWGRP)
-			break;
-	case GF_FILTER_MAP_BOTH:
-		if (loc->inode->st_mode & S_IWOTH)
-			break;
-		gf_log (this->name, GF_LOG_DEBUG, "%s: returning permission denied", loc->path);
-		STACK_UNWIND (frame, -1, EPERM, NULL, NULL);
-		return 0;
-		
-	case GF_FILTER_FILTER_UID:
-	case GF_FILTER_FILTER_GID:
-	case GF_FILTER_RO_FS:
-		STACK_UNWIND (frame, -1, EROFS, NULL, NULL);
-		return 0;
-	}			
-
-	STACK_WIND (frame,
-		    filter_truncate_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->truncate,
-		    loc,
-		    offset);
-	return 0;
-}
-
-static int32_t
-filter_ftruncate_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno,
-		      struct iatt *prebuf,
-                      struct iatt *postbuf)
-{
-	if (op_ret >= 0) {
-		update_stat (prebuf, this->private);
-		update_stat (postbuf, this->private);
-	}
-	STACK_UNWIND (frame, op_ret, op_errno, prebuf, postbuf);
-	return 0;
-}
-
-int32_t
-filter_ftruncate (call_frame_t *frame,
-		  xlator_t *this,
-		  fd_t *fd,
-		  off_t offset)
-{
-	STACK_WIND (frame,
-		    filter_ftruncate_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->ftruncate,
-		    fd,
-		    offset);
-	return 0;
-}
-
-
-static int32_t
-filter_readlink_cbk (call_frame_t *frame,
-		     void *cookie,
-		     xlator_t *this,
-		     int32_t op_ret,
-		     int32_t op_errno,
-		     const char *path,
-                     struct iatt *sbuf)
-{
-        if (op_ret >= 0)
-                update_stat (sbuf, this->private);
-
-	STACK_UNWIND (frame, op_ret, op_errno, path, sbuf);
-	return 0;
-}
-
-int32_t
-filter_readlink (call_frame_t *frame,
-		 xlator_t *this,
-		 loc_t *loc,
-		 size_t size)
-{
-	int32_t ret = 0;
-	ret = update_frame (frame, loc->inode, this->private);
-	switch (ret) {
-	case GF_FILTER_MAP_UID:
-		if (loc->inode->st_mode & S_IRGRP)
-			break;
-	case GF_FILTER_MAP_BOTH:
-		if (loc->inode->st_mode & S_IROTH)
-			break;
-		gf_log (this->name, GF_LOG_DEBUG, "%s: returning permission denied", loc->path);
-		STACK_UNWIND (frame, -1, EPERM, NULL);
-		return 0;
-	}			
-	STACK_WIND (frame,
-		    filter_readlink_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->readlink,
-		    loc,
-		    size);
-	return 0;
-}
-
-
-static int32_t
-filter_mknod_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  inode_t *inode,
-                  struct iatt *buf,
-                  struct iatt *preparent,
-                  struct iatt *postparent)
-{
-	int ret = 0;
-
-	if (op_ret >= 0) {
-		update_stat (buf, this->private);
-		ret = inode_ctx_put (inode, this, (uint64_t)(long)buf->ia_uid);
-		if (ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"couldn't set context");
-		}
-
-		update_stat (preparent, this->private);
-		update_stat (postparent, this->private);
-	}
-	STACK_UNWIND (frame, op_ret, op_errno, inode, buf,
-                      preparent, postparent);
-	return 0;
-}
-
-int32_t
-filter_mknod (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      mode_t mode,
-	      dev_t rdev)
-{
-	int ret = 0;
-	inode_t *parent = loc->parent;
-	ret = update_frame (frame, loc->inode, this->private);
-	switch (ret) {		
-	case GF_FILTER_MAP_UID:
-		if (parent->st_mode & S_IWGRP)
-			break;
-	case GF_FILTER_MAP_BOTH:
-		if (parent->st_mode & S_IWOTH)
-			break;
-		gf_log (this->name, GF_LOG_DEBUG, "%s: returning permission denied", loc->path);
-		STACK_UNWIND (frame, -1, EPERM, NULL, NULL,
-                              NULL, NULL);
-		return 0;
-
-	case GF_FILTER_FILTER_UID:
-	case GF_FILTER_FILTER_GID:
-	case GF_FILTER_RO_FS:
-		STACK_UNWIND (frame, -1, EROFS, NULL, NULL,
-                              NULL, NULL);
-		return 0;
-	}
-	STACK_WIND (frame,
-		    filter_mknod_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->mknod,
-		    loc, mode, rdev);
-	return 0;
-}
-
-static int32_t
-filter_mkdir_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  inode_t *inode,
-                  struct iatt *buf,
-                  struct iatt *preparent,
-                  struct iatt *postparent)
-{
-	int ret = 0;
-	if (op_ret >= 0) {
-		update_stat (buf, this->private);
-		ret = inode_ctx_put (inode, this, (uint64_t)(long)buf->ia_uid);
-		if (ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"couldn't set context");
-		}
-
-		update_stat (preparent, this->private);
-		update_stat (postparent, this->private);
-	}
-	STACK_UNWIND (frame, op_ret, op_errno, inode, buf,
-                      preparent, postparent);
-	return 0;
-}
-
-int32_t
-filter_mkdir (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      mode_t mode)
-{
-	int ret = 0;
-	inode_t *parent = loc->parent;
-	ret = update_frame (frame, loc->inode, this->private);
-	switch (ret) {		
-	case GF_FILTER_MAP_UID:
-		if (parent->st_mode & S_IWGRP)
-			break;
-	case GF_FILTER_MAP_BOTH:
-		if (parent->st_mode & S_IWOTH)
-			break;
-		gf_log (this->name, GF_LOG_DEBUG, "%s: returning permission denied", loc->path);
-		STACK_UNWIND (frame, -1, EPERM, NULL, NULL,
-                              NULL, NULL);
-		return 0;
-
-	case GF_FILTER_FILTER_UID:
-	case GF_FILTER_FILTER_GID:
-	case GF_FILTER_RO_FS:
-		STACK_UNWIND (frame, -1, EROFS, NULL, NULL,
-                              NULL, NULL);
-		return 0;
-	}
-	STACK_WIND (frame,
-		    filter_mkdir_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->mkdir,
-		    loc, mode);
-	return 0;
-}
-
-static int32_t
-filter_unlink_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-                   struct iatt *preparent,
-                   struct iatt *postparent)
-{
-        if (op_ret >= 0) {
-		update_stat (preparent, this->private);
-		update_stat (postparent, this->private);
-        }
-
-	STACK_UNWIND (frame, op_ret, op_errno, preparent, postparent);
-	return 0;
-}
-
-int32_t
-filter_unlink (call_frame_t *frame,
-	       xlator_t *this,
-	       loc_t *loc)
-{
-	int32_t ret = 0;
-	inode_t *parent = loc->parent;
-	if (!parent)
-		parent = inode_parent (loc->inode, 0, NULL);
-	ret = update_frame (frame, loc->inode, this->private);
-	switch (ret) {
-	case GF_FILTER_MAP_UID:
-		if (parent->st_mode & S_IWGRP)
-			break;
-		if (loc->inode->st_mode & S_IWGRP)
-			break;
-	case GF_FILTER_MAP_BOTH:
-		if (parent->st_mode & S_IWOTH)
-			break;
-		if (loc->inode->st_mode & S_IWOTH)
-			break;
-		gf_log (this->name, GF_LOG_DEBUG, "%s: returning permission denied", loc->path);
-		STACK_UNWIND (frame, -1, EPERM, NULL, NULL);
-		return 0;
-	case GF_FILTER_FILTER_UID:
-	case GF_FILTER_FILTER_GID:
-	case GF_FILTER_RO_FS:
-		STACK_UNWIND (frame, -1, EROFS, NULL, NULL);
-		return 0;
-	}
-	STACK_WIND (frame,
-		    filter_unlink_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->unlink,
-		    loc);
-	return 0;
-}
-
-static int32_t
-filter_rmdir_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-                  struct iatt *preparent,
-                  struct iatt *postparent)
-{
-        if (op_ret >= 0) {
-		update_stat (preparent, this->private);
-		update_stat (postparent, this->private);
-        }
-
-	STACK_UNWIND (frame, op_ret, op_errno, preparent, postparent);
-	return 0;
-}
-
-int32_t
-filter_rmdir (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc)
-{
-	int32_t ret = 0;
-	inode_t *parent = loc->parent;
-	if (!parent)
-		parent = inode_parent (loc->inode, 0, NULL);
-	ret = update_frame (frame, loc->inode, this->private);
-	switch (ret) {
-	case GF_FILTER_MAP_UID:
-		if (parent->st_mode & S_IWGRP)
-			break;
-		if (loc->inode->st_mode & S_IWGRP)
-			break;
-	case GF_FILTER_MAP_BOTH:
-		if (parent->st_mode & S_IWOTH)
-			break;
-		if (loc->inode->st_mode & S_IWOTH)
-			break;
-		gf_log (this->name, GF_LOG_DEBUG, "%s: returning permission denied", loc->path);
-		STACK_UNWIND (frame, -1, EPERM, NULL, NULL);
-		return 0;
-	case GF_FILTER_FILTER_UID:
-	case GF_FILTER_FILTER_GID:
-	case GF_FILTER_RO_FS:
-                STACK_UNWIND (frame, -1, EROFS, NULL, NULL);
-                return 0;
-	}			
-	STACK_WIND (frame,
-		    filter_rmdir_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->rmdir,
-		    loc);
-	return 0;
-}
-
-static int32_t
-filter_symlink_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    inode_t *inode,
-                    struct iatt *buf,
-                    struct iatt *preparent,
-                    struct iatt *postparent)
-{
-	int ret = 0;
-	if (op_ret >= 0) {
-		update_stat (buf, this->private);
-		ret = inode_ctx_put (inode, this, (uint64_t)(long)buf->ia_uid);
-		if (ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"couldn't set context");
-		}
-
-		update_stat (preparent, this->private);
-		update_stat (postparent, this->private);
-	}
-	STACK_UNWIND (frame, op_ret, op_errno, inode, buf,
-                      preparent, postparent);
-	return 0;
-}
-
-int32_t
-filter_symlink (call_frame_t *frame,
-		xlator_t *this,
-		const char *linkpath,
-		loc_t *loc)
-{
-	int ret = 0;
-	inode_t *parent = loc->parent;
-	ret = update_frame (frame, loc->inode, this->private);
-	switch (ret) {		
-	case GF_FILTER_MAP_UID:
-		if (parent->st_mode & S_IWGRP)
-			break;
-	case GF_FILTER_MAP_BOTH:
-		if (parent->st_mode & S_IWOTH)
-			break;
-		gf_log (this->name, GF_LOG_DEBUG, "%s: returning permission denied", loc->path);
-		STACK_UNWIND (frame, -1, EPERM, NULL, NULL,
-                              NULL, NULL);
-		return 0;
-
-	case GF_FILTER_FILTER_UID:
-	case GF_FILTER_FILTER_GID:
-	case GF_FILTER_RO_FS:
-		STACK_UNWIND (frame, -1, EROFS, NULL, NULL,
-                              NULL, NULL);
-		return 0;
-	}
-	STACK_WIND (frame,
-		    filter_symlink_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->symlink,
-		    linkpath, loc);
-	return 0;
-}
-
-
-static int32_t
-filter_rename_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-		   struct iatt *buf,
-                   struct iatt *preoldparent,
-                   struct iatt *postoldparent,
-                   struct iatt *prenewparent,
-                   struct iatt *postnewparent)
-{
-	if (op_ret >= 0) {
-		update_stat (buf, this->private);
-
-		update_stat (preoldparent, this->private);
-		update_stat (postoldparent, this->private);
-
-		update_stat (prenewparent, this->private);
-		update_stat (postnewparent, this->private);
-	}
-
-	STACK_UNWIND (frame, op_ret, op_errno, buf,
-                      preoldparent, postoldparent,
-                      prenewparent, postnewparent);
-	return 0;
-}
-
-int32_t
-filter_rename (call_frame_t *frame,
-	       xlator_t *this,
-	       loc_t *oldloc,
-	       loc_t *newloc)
-{
-	int32_t ret = 0;
-	inode_t *parent = oldloc->parent;
-	if (!parent)
-		parent = inode_parent (oldloc->inode, 0, NULL);
-	ret = update_frame (frame, oldloc->inode, this->private);
-	switch (ret) {		
-	case GF_FILTER_MAP_UID:
-		if (parent->st_mode & S_IWGRP)
-			break;
-		if (oldloc->inode->st_mode & S_IWGRP)
-			break;
-	case GF_FILTER_MAP_BOTH:
-		if (parent->st_mode & S_IWOTH)
-			break;
-		if (oldloc->inode->st_mode & S_IWOTH)
-			break;
-		gf_log (this->name, GF_LOG_DEBUG, 
-			"%s -> %s: returning permission denied", oldloc->path, newloc->path);
-		STACK_UNWIND (frame, -1, EPERM, NULL,
-                              NULL, NULL,
-                              NULL, NULL);
-		return 0;
-
-	case GF_FILTER_FILTER_UID:
-	case GF_FILTER_FILTER_GID:
-	case GF_FILTER_RO_FS:
-		STACK_UNWIND (frame, -1, EROFS, NULL,
-                              NULL, NULL,
-                              NULL, NULL);
-		return 0;
-	}
-	STACK_WIND (frame,
-		    filter_rename_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->rename,
-		    oldloc, newloc);
-	return 0;
-}
-
-
-static int32_t
-filter_link_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 inode_t *inode,
-                 struct iatt *buf,
-                 struct iatt *preparent,
-                 struct iatt *postparent)
-{
-	int ret = 0;
-	if (op_ret >= 0) {
-		update_stat (buf, this->private);
-		ret = inode_ctx_put (inode, this, (uint64_t)(long)buf->ia_uid);
-		if (ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"couldn't set context");
-		}
-
-		update_stat (preparent, this->private);
-		update_stat (postparent, this->private);
-	}
-	STACK_UNWIND (frame, op_ret, op_errno, inode, buf,
-                      preparent, postparent);
-	return 0;
-}
-
-int32_t
-filter_link (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *oldloc,
-	     loc_t *newloc)
-{
-	int ret = 0;
-	ret = update_frame (frame, oldloc->inode, this->private);
-	switch (ret) {
-	case GF_FILTER_FILTER_UID:
-	case GF_FILTER_FILTER_GID:
-	case GF_FILTER_RO_FS:
-		STACK_UNWIND (frame, -1, EROFS, NULL, NULL,
-                              NULL, NULL);
-		return 0;
-	}
-	STACK_WIND (frame,
-		    filter_link_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->link,
-		    oldloc, newloc);
-	return 0;
-}
-
-
-static int32_t
-filter_create_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-		   fd_t *fd,
-		   inode_t *inode,
-		   struct iatt *buf,
-                   struct iatt *preparent,
-                   struct iatt *postparent)
-{
-	int ret = 0;
-	if (op_ret >= 0) {
-		update_stat (buf, this->private);
-		ret = inode_ctx_put (inode, this, (uint64_t)(long)buf->ia_uid);
-		if (ret == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"couldn't set context");
-		}
-		update_stat (preparent, this->private);
-		update_stat (postparent, this->private);
-	}
-	STACK_UNWIND (frame, op_ret, op_errno, fd, inode, buf,
-                      preparent, postparent);
-	return 0;
-}
-
-int32_t
-filter_create (call_frame_t *frame,
-	       xlator_t *this,
-	       loc_t *loc,
-	       int32_t flags,
-	       mode_t mode, fd_t *fd)
-{
-	int ret = 0;
-	inode_t *parent = loc->parent;
-	ret = update_frame (frame, loc->inode, this->private);
-	switch (ret) {		
-	case GF_FILTER_MAP_UID:
-		if (parent->st_mode & S_IWGRP)
-			break;
-	case GF_FILTER_MAP_BOTH:
-		if (parent->st_mode & S_IWOTH)
-			break;
-		gf_log (this->name, GF_LOG_DEBUG, "%s: returning permission denied", loc->path);
-		STACK_UNWIND (frame, -1, EPERM, NULL, NULL, NULL,
-                              NULL, NULL);
-		return 0;
-
-	case GF_FILTER_FILTER_UID:
-	case GF_FILTER_FILTER_GID:
-	case GF_FILTER_RO_FS:
-		STACK_UNWIND (frame, -1, EROFS, NULL, NULL, NULL,
-                              NULL, NULL);
-		return 0;
-	}
-	STACK_WIND (frame, filter_create_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->create,
-		    loc, flags, mode, fd);
-	return 0;
-}
-
-static int32_t
-filter_open_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 fd_t *fd)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, fd);
-	return 0;
-}
-
-int32_t
-filter_open (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     int32_t flags, 
-	     fd_t *fd,
-             int32_t wbflags)
-{
-	int32_t ret = 0;
-	ret = update_frame (frame, loc->inode, this->private);
-	switch (ret) {
-	case GF_FILTER_MAP_UID:
-		if (loc->inode->st_mode & S_IWGRP)
-			break;
-		if (!(((flags & O_ACCMODE) == O_WRONLY)
-                      || ((flags & O_ACCMODE) == O_RDWR))
-		    && (loc->inode->st_mode & S_IRGRP))
-			break;
-	case GF_FILTER_MAP_BOTH:
-		if (loc->inode->st_mode & S_IWOTH)
-				break;
-		if (!(((flags & O_ACCMODE) == O_WRONLY)
-                      || ((flags & O_ACCMODE) == O_RDWR))
-		    && (loc->inode->st_mode & S_IROTH))
-			break;
-		gf_log (this->name, GF_LOG_DEBUG, 
-			"%s: returning permission denied (mode: 0%o, flag=0%o)", 
-			loc->path, loc->inode->st_mode, flags);
-		STACK_UNWIND (frame, -1, EPERM, fd);
-		return 0;
-	case GF_FILTER_FILTER_UID:
-	case GF_FILTER_FILTER_GID:
-	case GF_FILTER_RO_FS:
-		if (!(((flags & O_ACCMODE) == O_WRONLY)
-                      || ((flags & O_ACCMODE) == O_RDWR)))
-			break;
-		STACK_UNWIND (frame, -1, EROFS, NULL);
-		return 0;
-		
-	}
-	STACK_WIND (frame,
-		    filter_open_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->open,
-		    loc, flags, fd, wbflags);
-	return 0;
-}
-
-static int32_t
-filter_readv_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  struct iovec *vector,
-		  int32_t count,
-		  struct iatt *stbuf,
-                  struct iobref *iobref)
-{
-	if (op_ret >= 0) {
-		update_stat (stbuf, this->private);
-	}
-	STACK_UNWIND (frame,
-		      op_ret,
-		      op_errno,
-		      vector,
-		      count,
-		      stbuf,
-                      iobref);
-	return 0;
-}
-
-int32_t
-filter_readv (call_frame_t *frame,
-	      xlator_t *this,
-	      fd_t *fd,
-	      size_t size,
-	      off_t offset)
-{
-	STACK_WIND (frame,
-		    filter_readv_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->readv,
-		    fd,
-		    size,
-		    offset);
-	return 0;
-}
-
-
-static int32_t
-filter_writev_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-                   struct iatt *prebuf,
-		   struct iatt *postbuf)
-{
-	if (op_ret >= 0) {
-		update_stat (prebuf, this->private);
-		update_stat (postbuf, this->private);
-	}
-	STACK_UNWIND (frame,
-		      op_ret,
-		      op_errno,
-		      prebuf,
-                      postbuf);
-	return 0;
-}
-
-int32_t
-filter_writev (call_frame_t *frame,
-	       xlator_t *this,
-	       fd_t *fd,
-	       struct iovec *vector,
-	       int32_t count,
-	       off_t off,
-               struct iobref *iobref)
-{
-	int32_t ret = 0;
-	ret = update_frame (frame, fd->inode, this->private);
-	switch (ret) {
-	case GF_FILTER_FILTER_UID:
-	case GF_FILTER_FILTER_GID:
-	case GF_FILTER_RO_FS:
-		STACK_UNWIND (frame, -1, EROFS, NULL, NULL);
-		return 0;
-	}
-
-	STACK_WIND (frame,
-		    filter_writev_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->writev,
-		    fd,
-		    vector,
-		    count,
-		    off,
-                    iobref);
-	return 0;
-}
-
-static int32_t
-filter_fstat_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  struct iatt *buf)
-{
-	if (op_ret >= 0) {
-		update_stat (buf, this->private);
-	}
-	STACK_UNWIND (frame,
-		      op_ret,
-		      op_errno,
-		      buf);
-	return 0;
-}
-
-int32_t
-filter_fstat (call_frame_t *frame,
-	      xlator_t *this,
-	      fd_t *fd)
-{
-	STACK_WIND (frame,
-		    filter_fstat_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->fstat,
-		    fd);
-	return 0;
-}
-
-static int32_t
-filter_opendir_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    fd_t *fd)
-{
-	STACK_UNWIND (frame,
-		      op_ret,
-		      op_errno,
-		      fd);
-	return 0;
-}
-
-int32_t
-filter_opendir (call_frame_t *frame,
-		xlator_t *this,
-		loc_t *loc, fd_t *fd)
-{
-	int32_t ret = 0;
-	ret = update_frame (frame, loc->inode, this->private);
-	switch (ret) {
-	case GF_FILTER_MAP_UID:
-		if (loc->inode->st_mode & S_IWGRP)
-			break;
-		if (loc->inode->st_mode & S_IRGRP)
-			break;
-	case GF_FILTER_MAP_BOTH:
-		if (loc->inode->st_mode & S_IWOTH)
-			break;
-		if (loc->inode->st_mode & S_IROTH)
-			break;
-		gf_log (this->name, GF_LOG_DEBUG, "%s: returning permission denied", loc->path);
-		STACK_UNWIND (frame, -1, EPERM, fd);
-		return 0;
-	}			
-	STACK_WIND (frame,
-		    filter_opendir_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->opendir,
-		    loc, fd);
-	return 0;
-}
-
-
-static int32_t
-filter_setxattr_cbk (call_frame_t *frame,
-		     void *cookie,
-		     xlator_t *this,
-		     int32_t op_ret,
-		     int32_t op_errno)
-{
-	STACK_UNWIND (frame,
-		      op_ret,
-		      op_errno);
-	return 0;
-}
-
-int32_t
-filter_setxattr (call_frame_t *frame,
-		 xlator_t *this,
-		 loc_t *loc,
-		 dict_t *dict,
-		 int32_t flags)
-{
-
-	int32_t ret = 0;
-	ret = update_frame (frame, loc->inode, this->private);
-	switch (ret) {
-	case GF_FILTER_MAP_UID:
-		if (loc->inode->st_mode & S_IWGRP)
-			break;
-	case GF_FILTER_MAP_BOTH:
-		if (loc->inode->st_mode & S_IWOTH)
-			break;
-		gf_log (this->name, GF_LOG_DEBUG, "%s: returning permission denied", loc->path);
-		STACK_UNWIND (frame, -1, EPERM);
-		return 0;
-	case GF_FILTER_FILTER_UID:
-	case GF_FILTER_FILTER_GID:
-	case GF_FILTER_RO_FS:
-		STACK_UNWIND (frame, -1, EROFS);
-		return 0;
-	}			
-
-	STACK_WIND (frame,
-		    filter_setxattr_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->setxattr,
-		    loc,
-		    dict,
-		    flags);
-	return 0;
-}
-
-static int32_t
-filter_getxattr_cbk (call_frame_t *frame,
-		     void *cookie,
-		     xlator_t *this,
-		     int32_t op_ret,
-		     int32_t op_errno,
-		     dict_t *dict)
-{
-	STACK_UNWIND (frame,
-		      op_ret,
-		      op_errno,
-		      dict);
-	return 0;
-}
-
-int32_t
-filter_getxattr (call_frame_t *frame,
-		 xlator_t *this,
-		 loc_t *loc,
-		 const char *name)
-{
-	int32_t ret = 0;
-	ret = update_frame (frame, loc->inode, this->private);
-	switch (ret) {
-	case GF_FILTER_MAP_UID:
-		if (loc->inode->st_mode & S_IRGRP)
-			break;
-	case GF_FILTER_MAP_BOTH:
-		if (loc->inode->st_mode & S_IROTH)
-			break;
-		gf_log (this->name, GF_LOG_DEBUG, "%s: returning permission denied", loc->path);
-		STACK_UNWIND (frame, -1, EPERM, NULL);
-		return 0;
-	}			
-
-	STACK_WIND (frame,
-		    filter_getxattr_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->getxattr,
-		    loc,
-		    name);
-	return 0;
-}
-
-static int32_t
-filter_removexattr_cbk (call_frame_t *frame,
-			void *cookie,
-			xlator_t *this,
-			int32_t op_ret,
-			int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-int32_t
-filter_removexattr (call_frame_t *frame,
-		    xlator_t *this,
-		    loc_t *loc,
-		    const char *name)
-{
-	int32_t ret = 0;
-	ret = update_frame (frame, loc->inode, this->private);
-	switch (ret) {
-	case GF_FILTER_MAP_UID:
-		if (loc->inode->st_mode & S_IWGRP)
-			break;
-	case GF_FILTER_MAP_BOTH:
-		if (loc->inode->st_mode & S_IWOTH)
-			break;
-		gf_log (this->name, GF_LOG_DEBUG, "%s: returning permission denied", loc->path);
-		STACK_UNWIND (frame, -1, EPERM);
-		return 0;
-	case GF_FILTER_FILTER_UID:
-	case GF_FILTER_FILTER_GID:
-	case GF_FILTER_RO_FS:
-		STACK_UNWIND (frame, -1, EROFS);
-		return 0;
-	}			
-
-	STACK_WIND (frame,
-		    filter_removexattr_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->removexattr,
-		    loc,
-		    name);
-	return 0;
-}
-
-int32_t
-mem_acct_init (xlator_t *this)
-{
-        int     ret = -1;
-
-        if (!this)
-                return ret;
-
-        ret = xlator_mem_acct_init (this, gf_filter_mt_end + 1);
-        
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
-                        "failed");
-                return ret;
-        }
-
-        return ret;
-}
-
-int32_t 
-init (xlator_t *this)
-{
-	char *value = NULL;
-	char *tmp_str = NULL;
-	char *tmp_str1 = NULL;
-	char *tmp_str2 = NULL;
-	char *dup_str = NULL;
-	char *input_value_str1 = NULL;
-	char *input_value_str2 = NULL;
-	char *output_value_str = NULL;
-	int32_t input_value = 0;
-	int32_t output_value = 0;
-	data_t *option_data = NULL;
-	struct gf_filter *filter = NULL;
-	gf_boolean_t tmp_bool = 0;
-
-	if (!this->children || this->children->next) {
-		gf_log (this->name,
-			GF_LOG_ERROR,
-			"translator not configured with exactly one child");
-		return -1;
-	}
-	
-	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"dangling volume. check volfile ");
-	}
-
-	filter = GF_CALLOC (sizeof (*filter), 1, gf_filter_mt_gf_filter);
-	ERR_ABORT (filter);
-	
-	if (dict_get (this->options, "read-only")) {
-		value = data_to_str (dict_get (this->options, "read-only"));
-		if (gf_string2boolean (value, &filter->complete_read_only) == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"wrong value provided for 'read-only'");
-			return -1;
-		}
-	}
-
-	if (dict_get (this->options, "root-squashing")) {
-		value = data_to_str (dict_get (this->options, "root-squashing"));
-		if (gf_string2boolean (value, &tmp_bool) == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"wrong value provided for 'root-squashing'");
-			return -1;
-		}
-		if (tmp_bool) {
-			filter->translate_num_uid_entries = 1;
-			filter->translate_num_gid_entries = 1;
-			filter->translate_input_uid[0][0] = GF_FILTER_ROOT_UID; /* root */
-			filter->translate_input_uid[0][1] = GF_FILTER_ROOT_UID; /* root */
-			filter->translate_input_gid[0][0] = GF_FILTER_ROOT_GID; /* root */
-			filter->translate_input_gid[0][1] = GF_FILTER_ROOT_GID; /* root */
-			filter->translate_output_uid[0] = GF_FILTER_NOBODY_UID;
-			filter->translate_output_gid[0] = GF_FILTER_NOBODY_GID;
-		}
-	}
-
-	if (dict_get (this->options, "translate-uid")) {
-		option_data = dict_get (this->options, "translate-uid");
-		value = strtok_r (option_data->data, ",", &tmp_str);
-		while (value) {
-			dup_str = gf_strdup (value);
-			input_value_str1 = strtok_r (dup_str, "=", &tmp_str1);
-			if (input_value_str1) {
-				/* Check for n-m */
-				char *temp_string = gf_strdup (input_value_str1);
-				input_value_str2 = strtok_r (temp_string, "-", &tmp_str2);
-				if (gf_string2int (input_value_str2, &input_value) != 0) {
-					gf_log (this->name, GF_LOG_ERROR, 
-						"invalid number format \"%s\"", 
-						input_value_str2);
-					return -1;
-				}
-				filter->translate_input_uid[filter->translate_num_uid_entries][0] = input_value;
-				input_value_str2 = strtok_r (NULL, "-", &tmp_str2);
-				if (input_value_str2) {
-					if (gf_string2int (input_value_str2, &input_value) != 0) {
-						gf_log (this->name, GF_LOG_ERROR, 
-							"invalid number format \"%s\"", 
-							input_value_str2);
-						return -1;
-					}
-				}
-				filter->translate_input_uid[filter->translate_num_uid_entries][1] = input_value;
-				GF_FREE (temp_string);
-				output_value_str = strtok_r (NULL, "=", &tmp_str1);
-				if (output_value_str) {
-					if (gf_string2int (output_value_str, &output_value) != 0) {
-						gf_log (this->name, GF_LOG_ERROR, 
-							"invalid number format \"%s\"", 
-							output_value_str);
-						return -1;
-					}
-				} else {
-					gf_log (this->name, GF_LOG_ERROR, 
-						"mapping string not valid");
-					return -1;
-				}
-			} else {
-				gf_log (this->name, GF_LOG_ERROR, 
-					"mapping string not valid");
-				return -1;
-			}
-			filter->translate_output_uid[filter->translate_num_uid_entries]   = output_value;
-			gf_log (this->name, 
-				GF_LOG_DEBUG, 
-				"pair %d: input uid '%d' will be changed to uid '%d'", 
-				filter->translate_num_uid_entries, input_value, output_value);
-
-			filter->translate_num_uid_entries++;
-			if (filter->translate_num_uid_entries == GF_MAXIMUM_FILTERING_ALLOWED)
-				break;
-			value = strtok_r (NULL, ",", &tmp_str);
-			GF_FREE (dup_str);
-		}
-	}
-
-	tmp_str1 = NULL;
-	tmp_str2 = NULL;
-	tmp_str  = NULL;
-
-	if (dict_get (this->options, "translate-gid")) {
-		option_data = dict_get (this->options, "translate-gid");
-		value = strtok_r (option_data->data, ",", &tmp_str);
-		while (value) {
-			dup_str = gf_strdup (value);
-			input_value_str1 = strtok_r (dup_str, "=", &tmp_str1);
-			if (input_value_str1) {
-				/* Check for n-m */
-				char *temp_string = gf_strdup (input_value_str1);
-				input_value_str2 = strtok_r (temp_string, "-", &tmp_str2);
-				if (gf_string2int (input_value_str2, &input_value) != 0) {
-					gf_log (this->name, GF_LOG_ERROR, 
-						"invalid number format \"%s\"", 
-						input_value_str2);
-					return -1;
-				}
-				filter->translate_input_gid[filter->translate_num_gid_entries][0] = input_value;
-				input_value_str2 = strtok_r (NULL, "-", &tmp_str2);
-				if (input_value_str2) {
-					if (gf_string2int (input_value_str2, &input_value) != 0) {
-						gf_log (this->name, GF_LOG_ERROR, 
-							"invalid number format \"%s\"", 
-							input_value_str2);
-						return -1;
-					}
-				}
-				filter->translate_input_gid[filter->translate_num_gid_entries][1] = input_value;
-				GF_FREE (temp_string);
-				output_value_str = strtok_r (NULL, "=", &tmp_str1);
-				if (output_value_str) {
-					if (gf_string2int (output_value_str, &output_value) != 0) {
-						gf_log (this->name, GF_LOG_ERROR, 
-							"invalid number format \"%s\"", 
-							output_value_str);
-						return -1;
-					}
-				} else {
-					gf_log (this->name, GF_LOG_ERROR, 
-						"translate-gid value not valid");
-					return -1;
-				}
-			} else {
-				gf_log (this->name, GF_LOG_ERROR, 
-					"translate-gid value not valid");
-				return -1;
-			}
-			
-			filter->translate_output_gid[filter->translate_num_gid_entries] = output_value;
-			
-			gf_log (this->name, GF_LOG_DEBUG, 
-				"pair %d: input gid '%d' will be changed to gid '%d'", 
-				filter->translate_num_gid_entries, input_value, output_value);
-			
-			filter->translate_num_gid_entries++;
-			if (filter->translate_num_gid_entries == GF_MAXIMUM_FILTERING_ALLOWED)
-				break;
-			value = strtok_r (NULL, ",", &tmp_str);
-			GF_FREE (dup_str);
-		}
-	}
-
-	tmp_str  = NULL;
-	tmp_str1 = NULL;
-
-	if (dict_get (this->options, "filter-uid")) {
-		option_data = dict_get (this->options, "filter-uid");
-		value = strtok_r (option_data->data, ",", &tmp_str);
-		while (value) {
-			dup_str = gf_strdup (value);
-			/* Check for n-m */
-			input_value_str1 = strtok_r (dup_str, "-", &tmp_str1);
-			if (gf_string2int (input_value_str1, &input_value) != 0) {
-				gf_log (this->name, GF_LOG_ERROR, 
-					"invalid number format \"%s\"", 
-					input_value_str1);
-				return -1;
-			}
-			filter->filter_input_uid[filter->filter_num_uid_entries][0] = input_value;
-			input_value_str1 = strtok_r (NULL, "-", &tmp_str1);
-			if (input_value_str1) {
-				if (gf_string2int (input_value_str1, &input_value) != 0) {
-					gf_log (this->name, GF_LOG_ERROR, 
-						"invalid number format \"%s\"", 
-						input_value_str1);
-					return -1;
-				}
-			}
-			filter->filter_input_uid[filter->filter_num_uid_entries][1] = input_value;
-
-			gf_log (this->name, 
-				GF_LOG_DEBUG, 
-				"filter [%d]: input uid(s) '%s' will be filtered", 
-				filter->filter_num_uid_entries, dup_str);
-			
-			filter->filter_num_uid_entries++;
-			if (filter->filter_num_uid_entries == GF_MAXIMUM_FILTERING_ALLOWED)
-				break;
-			value = strtok_r (NULL, ",", &tmp_str);
-			GF_FREE (dup_str);
-		}
-		filter->partial_filter = 1;
-	}
-
-	tmp_str  = NULL;
-	tmp_str1 = NULL;
-
-	if (dict_get (this->options, "filter-gid")) {
-		option_data = dict_get (this->options, "filter-gid");
-		value = strtok_r (option_data->data, ",", &tmp_str);
-		while (value) {
-			dup_str = gf_strdup (value);
-			/* Check for n-m */
-			input_value_str1 = strtok_r (dup_str, "-", &tmp_str1);
-			if (gf_string2int (input_value_str1, &input_value) != 0) {
-				gf_log (this->name, GF_LOG_ERROR, 
-					"invalid number format \"%s\"", 
-					input_value_str1);
-				return -1;
-			}
-			filter->filter_input_gid[filter->filter_num_gid_entries][0] = input_value;
-			input_value_str1 = strtok_r (NULL, "-", &tmp_str1);
-			if (input_value_str1) {
-				if (gf_string2int (input_value_str1, &input_value) != 0) {
-					gf_log (this->name, GF_LOG_ERROR, 
-						"invalid number format \"%s\"", 
-						input_value_str1);
-					return -1;
-				}
-			}
-			filter->filter_input_gid[filter->filter_num_gid_entries][1] = input_value;
-
-			gf_log (this->name, 
-				GF_LOG_DEBUG, 
-				"filter [%d]: input gid(s) '%s' will be filtered", 
-				filter->filter_num_gid_entries, dup_str);
-			
-			filter->filter_num_gid_entries++;
-			if (filter->filter_num_gid_entries == GF_MAXIMUM_FILTERING_ALLOWED)
-				break;
-			value = strtok_r (NULL, ",", &tmp_str);
-			GF_FREE (dup_str);
-		}
-		gf_log (this->name, GF_LOG_ERROR, "this option is not supported currently.. exiting");
-		return -1;
-		filter->partial_filter = 1;
-	}
-
-	if (dict_get (this->options, "fixed-uid")) {
-		option_data = dict_get (this->options, "fixed-uid");
-		if (gf_string2int (option_data->data, &input_value) != 0) {
-			gf_log (this->name, GF_LOG_ERROR, 
-				"invalid number format \"%s\"", 
-				option_data->data);
-			return -1;
-		}
-		filter->fixed_uid = input_value;
-		filter->fixed_uid_set = 1;
-	}
-
-	if (dict_get (this->options, "fixed-gid")) {
-		option_data = dict_get (this->options, "fixed-gid");
-		if (gf_string2int (option_data->data, &input_value) != 0) {
-			gf_log (this->name, GF_LOG_ERROR, 
-				"invalid number format \"%s\"", 
-				option_data->data);
-			return -1;
-		}
-		filter->fixed_gid = input_value;
-		filter->fixed_gid_set = 1;
-	}
-
-	this->private = filter;
-	return 0;
-}
-
-
-void
-fini (xlator_t *this)
-{
-	struct gf_filter *filter = this->private;
-
-	GF_FREE (filter);
-
-	return;
-}
-
-
-struct xlator_fops fops = {
-	.lookup      = filter_lookup,
-	.stat        = filter_stat,
-	.fstat       = filter_fstat,
-	.readlink    = filter_readlink,
-	.mknod       = filter_mknod,
-	.mkdir       = filter_mkdir,
-	.unlink      = filter_unlink,
-	.rmdir       = filter_rmdir,
-	.symlink     = filter_symlink,
-	.rename      = filter_rename,
-	.link        = filter_link,
-	.truncate    = filter_truncate,
-	.ftruncate   = filter_ftruncate,
-	.create      = filter_create,
-	.open        = filter_open,
-	.readv       = filter_readv,
-	.writev      = filter_writev,
-	.setxattr    = filter_setxattr,
-	.getxattr    = filter_getxattr,
-	.removexattr = filter_removexattr,
-	.opendir     = filter_opendir,
-        .setattr     = filter_setattr,
-        .fsetattr    = filter_fsetattr,
-};
-
-struct xlator_cbks cbks = {
-};
-
-struct volume_options options[] = {
-	{ .key  = { "root-squashing" }, 
-	  .type = GF_OPTION_TYPE_BOOL 
-	},
-	{ .key  = { "read-only" }, 
-	  .type = GF_OPTION_TYPE_BOOL
-	},
-	{ .key  = { "fixed-uid" },  
-	  .type = GF_OPTION_TYPE_INT
-	},
-	{ .key  = { "fixed-gid" },  
-	  .type = GF_OPTION_TYPE_INT
-	},
-	{ .key  = { "translate-uid" },  
-	  .type = GF_OPTION_TYPE_ANY 
-	},
-	{ .key  = { "translate-gid" },  
-	  .type = GF_OPTION_TYPE_ANY
-	},
-	{ .key  = { "filter-uid" },  
-	  .type = GF_OPTION_TYPE_ANY 
-	},
-	{ .key  = { "filter-gid" },  
-	  .type = GF_OPTION_TYPE_ANY 
-	},
-	{ .key = {NULL} },
-};
diff --git a/xlators/bindings/python/Makefile.am b/xlators/features/gfid-access/Makefile.am
index af437a64d6d..af437a64d6d 100644
--- a/xlators/bindings/python/Makefile.am
+++ b/xlators/features/gfid-access/Makefile.am
diff --git a/xlators/features/gfid-access/src/Makefile.am b/xlators/features/gfid-access/src/Makefile.am
new file mode 100644
index 00000000000..ff95604c4de
--- /dev/null
+++ b/xlators/features/gfid-access/src/Makefile.am
@@ -0,0 +1,16 @@
+xlator_LTLIBRARIES = gfid-access.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+gfid_access_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+gfid_access_la_SOURCES = gfid-access.c
+gfid_access_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = gfid-access.h gfid-access-mem-types.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/gfid-access/src/gfid-access-mem-types.h b/xlators/features/gfid-access/src/gfid-access-mem-types.h
new file mode 100644
index 00000000000..1c4d0b93de2
--- /dev/null
+++ b/xlators/features/gfid-access/src/gfid-access-mem-types.h
@@ -0,0 +1,22 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef _GFID_ACCESS_MEM_TYPES_H
+#define _GFID_ACCESS_MEM_TYPES_H
+
+#include <glusterfs/mem-types.h>
+
+enum gf_changelog_mem_types {
+    gf_gfid_access_mt_priv_t = gf_common_mt_end + 1,
+    gf_gfid_access_mt_gfid_t,
+    gf_gfid_access_mt_end
+};
+
+#endif
diff --git a/xlators/features/gfid-access/src/gfid-access.c b/xlators/features/gfid-access/src/gfid-access.c
new file mode 100644
index 00000000000..3fea5672a21
--- /dev/null
+++ b/xlators/features/gfid-access/src/gfid-access.c
@@ -0,0 +1,1420 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include "gfid-access.h"
+#include <glusterfs/inode.h>
+#include <glusterfs/byte-order.h>
+#include <glusterfs/statedump.h>
+
+int
+ga_valid_inode_loc_copy(loc_t *dst, loc_t *src, xlator_t *this)
+{
+    int ret = 0;
+    uint64_t value = 0;
+
+    /* if its an entry operation, on the virtual */
+    /* directory inode as parent, we need to handle */
+    /* it properly */
+    ret = loc_copy(dst, src);
+    if (ret < 0)
+        goto out;
+
+    /*
+     * Change ALL virtual inodes with real-inodes in loc
+     */
+    if (dst->parent) {
+        ret = inode_ctx_get(dst->parent, this, &value);
+        if (ret < 0) {
+            ret = 0;  // real-inode
+            goto out;
+        }
+        inode_unref(dst->parent);
+        dst->parent = inode_ref((inode_t *)(uintptr_t)value);
+        gf_uuid_copy(dst->pargfid, dst->parent->gfid);
+    }
+
+    if (dst->inode) {
+        ret = inode_ctx_get(dst->inode, this, &value);
+        if (ret < 0) {
+            ret = 0;  // real-inode
+            goto out;
+        }
+        inode_unref(dst->inode);
+        dst->inode = inode_ref((inode_t *)(uintptr_t)value);
+        gf_uuid_copy(dst->gfid, dst->inode->gfid);
+    }
+out:
+
+    return ret;
+}
+
+void
+ga_newfile_args_free(ga_newfile_args_t *args)
+{
+    if (!args)
+        goto out;
+
+    GF_FREE(args->bname);
+
+    if (S_ISLNK(args->st_mode) && args->args.symlink.linkpath) {
+        GF_FREE(args->args.symlink.linkpath);
+        args->args.symlink.linkpath = NULL;
+    }
+
+    mem_put(args);
+out:
+    return;
+}
+
+void
+ga_heal_args_free(ga_heal_args_t *args)
+{
+    if (!args)
+        goto out;
+
+    GF_FREE(args->bname);
+
+    mem_put(args);
+out:
+    return;
+}
+
+ga_newfile_args_t *
+ga_newfile_parse_args(xlator_t *this, data_t *data)
+{
+    ga_newfile_args_t *args = NULL;
+    ga_private_t *priv = NULL;
+    int len = 0;
+    int blob_len = 0;
+    int min_len = 0;
+    void *blob = NULL;
+
+    priv = this->private;
+
+    blob = data->data;
+    blob_len = data->len;
+
+    min_len = sizeof(args->uid) + sizeof(args->gid) + sizeof(args->gfid) +
+              sizeof(args->st_mode) + 2 + 2;
+    if (blob_len < min_len) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Invalid length: Total length is less "
+               "than minimum length.");
+        goto err;
+    }
+
+    args = mem_get0(priv->newfile_args_pool);
+    if (args == NULL)
+        goto err;
+
+    args->uid = ntoh32(*(uint32_t *)blob);
+    blob += sizeof(uint32_t);
+    blob_len -= sizeof(uint32_t);
+
+    args->gid = ntoh32(*(uint32_t *)blob);
+    blob += sizeof(uint32_t);
+    blob_len -= sizeof(uint32_t);
+
+    memcpy(args->gfid, blob, sizeof(args->gfid));
+    blob += sizeof(args->gfid);
+    blob_len -= sizeof(args->gfid);
+
+    args->st_mode = ntoh32(*(uint32_t *)blob);
+    blob += sizeof(uint32_t);
+    blob_len -= sizeof(uint32_t);
+
+    len = strnlen(blob, blob_len);
+    if (len == blob_len) {
+        gf_log(this->name, GF_LOG_ERROR, "gfid: %s. No null byte present.",
+               args->gfid);
+        goto err;
+    }
+
+    args->bname = GF_MALLOC(len + 1, gf_common_mt_char);
+    if (args->bname == NULL)
+        goto err;
+
+    memcpy(args->bname, blob, (len + 1));
+    blob += (len + 1);
+    blob_len -= (len + 1);
+
+    if (S_ISDIR(args->st_mode)) {
+        if (blob_len < sizeof(uint32_t)) {
+            gf_log(this->name, GF_LOG_ERROR, "gfid: %s. Invalid length",
+                   args->gfid);
+            goto err;
+        }
+        args->args.mkdir.mode = ntoh32(*(uint32_t *)blob);
+        blob += sizeof(uint32_t);
+        blob_len -= sizeof(uint32_t);
+
+        if (blob_len < sizeof(uint32_t)) {
+            gf_log(this->name, GF_LOG_ERROR, "gfid: %s. Invalid length",
+                   args->gfid);
+            goto err;
+        }
+        args->args.mkdir.umask = ntoh32(*(uint32_t *)blob);
+        blob_len -= sizeof(uint32_t);
+        if (blob_len < 0) {
+            gf_log(this->name, GF_LOG_ERROR, "gfid: %s. Invalid length",
+                   args->gfid);
+            goto err;
+        }
+    } else if (S_ISLNK(args->st_mode)) {
+        len = strnlen(blob, blob_len);
+        if (len == blob_len) {
+            gf_log(this->name, GF_LOG_ERROR, "gfid: %s. Invalid length",
+                   args->gfid);
+            goto err;
+        }
+        args->args.symlink.linkpath = GF_MALLOC(len + 1, gf_common_mt_char);
+        if (args->args.symlink.linkpath == NULL)
+            goto err;
+
+        memcpy(args->args.symlink.linkpath, blob, (len + 1));
+        blob_len -= (len + 1);
+    } else {
+        if (blob_len < sizeof(uint32_t)) {
+            gf_log(this->name, GF_LOG_ERROR, "gfid: %s. Invalid length",
+                   args->gfid);
+            goto err;
+        }
+        args->args.mknod.mode = ntoh32(*(uint32_t *)blob);
+        blob += sizeof(uint32_t);
+        blob_len -= sizeof(uint32_t);
+
+        if (blob_len < sizeof(uint32_t)) {
+            gf_log(this->name, GF_LOG_ERROR, "gfid: %s. Invalid length",
+                   args->gfid);
+            goto err;
+        }
+        args->args.mknod.rdev = ntoh32(*(uint32_t *)blob);
+        blob += sizeof(uint32_t);
+        blob_len -= sizeof(uint32_t);
+
+        if (blob_len < sizeof(uint32_t)) {
+            gf_log(this->name, GF_LOG_ERROR, "gfid: %s. Invalid length",
+                   args->gfid);
+            goto err;
+        }
+        args->args.mknod.umask = ntoh32(*(uint32_t *)blob);
+        blob_len -= sizeof(uint32_t);
+    }
+
+    if (blob_len) {
+        gf_log(this->name, GF_LOG_ERROR, "gfid: %s. Invalid length",
+               args->gfid);
+        goto err;
+    }
+
+    return args;
+
+err:
+    if (args)
+        ga_newfile_args_free(args);
+
+    return NULL;
+}
+
+ga_heal_args_t *
+ga_heal_parse_args(xlator_t *this, data_t *data)
+{
+    ga_heal_args_t *args = NULL;
+    ga_private_t *priv = NULL;
+    void *blob = NULL;
+    int len = 0;
+    int blob_len = 0;
+
+    blob = data->data;
+    blob_len = data->len;
+
+    priv = this->private;
+
+    /* bname should at least contain a character */
+    if (blob_len < (sizeof(args->gfid) + 2))
+        goto err;
+
+    args = mem_get0(priv->heal_args_pool);
+    if (!args)
+        goto err;
+
+    memcpy(args->gfid, blob, sizeof(args->gfid));
+    blob += sizeof(args->gfid);
+    blob_len -= sizeof(args->gfid);
+
+    len = strnlen(blob, blob_len);
+    if (len == blob_len)
+        goto err;
+
+    args->bname = GF_MALLOC(len + 1, gf_common_mt_char);
+    if (!args->bname)
+        goto err;
+
+    memcpy(args->bname, blob, len);
+    args->bname[len] = '\0';
+    blob_len -= (len + 1);
+
+    if (blob_len)
+        goto err;
+
+    return args;
+
+err:
+    if (args)
+        ga_heal_args_free(args);
+
+    return NULL;
+}
+
+static int32_t
+ga_fill_tmp_loc(loc_t *loc, xlator_t *this, uuid_t gfid, char *bname,
+                dict_t *xdata, loc_t *new_loc)
+{
+    int ret = -1;
+    uint64_t value = 0;
+    inode_t *parent = NULL;
+    unsigned char *gfid_ptr = NULL;
+
+    parent = loc->inode;
+    ret = inode_ctx_get(loc->inode, this, &value);
+    if (!ret) {
+        parent = (void *)(uintptr_t)value;
+        if (gf_uuid_is_null(parent->gfid))
+            parent = loc->inode;
+    }
+
+    /* parent itself should be looked up */
+    gf_uuid_copy(new_loc->pargfid, parent->gfid);
+    new_loc->parent = inode_ref(parent);
+
+    new_loc->inode = inode_grep(parent->table, parent, bname);
+    if (!new_loc->inode) {
+        new_loc->inode = inode_new(parent->table);
+        gf_uuid_copy(new_loc->inode->gfid, gfid);
+    }
+
+    loc_path(new_loc, bname);
+    if (new_loc->path) {
+        new_loc->name = strrchr(new_loc->path, '/');
+        if (new_loc->name)
+            new_loc->name++;
+    }
+
+    gfid_ptr = GF_MALLOC(sizeof(uuid_t), gf_common_mt_uuid_t);
+    if (!gfid_ptr) {
+        ret = -1;
+        goto out;
+    }
+    gf_uuid_copy(gfid_ptr, gfid);
+    ret = dict_set_gfuuid(xdata, "gfid-req", gfid_ptr, false);
+    if (ret < 0)
+        goto out;
+
+    ret = 0;
+
+out:
+    if (ret && gfid_ptr)
+        GF_FREE(gfid_ptr);
+    return ret;
+}
+
+static gf_boolean_t
+__is_gfid_access_dir(uuid_t gfid)
+{
+    static uuid_t aux_gfid = {0, 0, 0, 0, 0, 0, 0, 0,
+                              0, 0, 0, 0, 0, 0, 0, GF_AUX_GFID};
+
+    if (gf_uuid_compare(gfid, aux_gfid) == 0)
+        return _gf_true;
+
+    return _gf_false;
+}
+
+int32_t
+ga_forget(xlator_t *this, inode_t *inode)
+{
+    int ret = -1;
+    uint64_t value = 0;
+    inode_t *tmp_inode = NULL;
+
+    ret = inode_ctx_del(inode, this, &value);
+    if (ret)
+        goto out;
+
+    tmp_inode = (void *)(uintptr_t)value;
+    inode_unref(tmp_inode);
+
+out:
+    return 0;
+}
+
+static int
+ga_heal_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, inode_t *inode, struct iatt *stat, dict_t *dict,
+            struct iatt *postparent)
+{
+    call_frame_t *orig_frame = NULL;
+
+    orig_frame = frame->local;
+    frame->local = NULL;
+
+    /* don't worry about inode linking and other stuff. They'll happen on
+     * the next lookup.
+     */
+    STACK_DESTROY(frame->root);
+
+    STACK_UNWIND_STRICT(setxattr, orig_frame, op_ret, op_errno, dict);
+
+    return 0;
+}
+
+static int
+ga_newentry_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, inode_t *inode,
+                struct iatt *buf, struct iatt *preparent,
+                struct iatt *postparent, dict_t *xdata)
+{
+    ga_local_t *local = NULL;
+
+    local = frame->local;
+
+    /* don't worry about inode linking and other stuff. They'll happen on
+     * the next lookup.
+     */
+    frame->local = NULL;
+    STACK_DESTROY(frame->root);
+
+    STACK_UNWIND_STRICT(setxattr, local->orig_frame, op_ret, op_errno, xdata);
+
+    if (local->xdata)
+        dict_unref(local->xdata);
+    loc_wipe(&local->loc);
+    mem_put(local);
+
+    return 0;
+}
+
+static int
+ga_newentry_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, inode_t *inode,
+                       struct iatt *stat, dict_t *xdata,
+                       struct iatt *postparent)
+
+{
+    ga_local_t *local = NULL;
+
+    local = frame->local;
+
+    if ((op_ret < 0) && ((op_errno != ENOENT) && (op_errno != ESTALE)))
+        goto err;
+
+    STACK_WIND(frame, ga_newentry_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mknod, &local->loc, local->mode,
+               local->rdev, local->umask, local->xdata);
+    return 0;
+
+err:
+    frame->local = NULL;
+    STACK_DESTROY(frame->root);
+    STACK_UNWIND_STRICT(setxattr, local->orig_frame, op_ret, op_errno, xdata);
+    if (local->xdata)
+        dict_unref(local->xdata);
+    loc_wipe(&local->loc);
+    mem_put(local);
+
+    return 0;
+}
+
+int32_t
+ga_new_entry(call_frame_t *frame, xlator_t *this, loc_t *loc, data_t *data,
+             dict_t *xdata)
+{
+    int ret = -1;
+    ga_newfile_args_t *args = NULL;
+    loc_t tmp_loc = {
+        0,
+    };
+    call_frame_t *new_frame = NULL;
+    ga_local_t *local = NULL;
+    uuid_t gfid = {
+        0,
+    };
+
+    if (!xdata) {
+        xdata = dict_new();
+    } else {
+        xdata = dict_ref(xdata);
+    }
+
+    if (!xdata) {
+        ret = -1;
+        goto out;
+    }
+
+    args = ga_newfile_parse_args(this, data);
+    if (!args)
+        goto out;
+
+    ret = gf_uuid_parse(args->gfid, gfid);
+    if (ret)
+        goto out;
+
+    ret = ga_fill_tmp_loc(loc, this, gfid, args->bname, xdata, &tmp_loc);
+    if (ret)
+        goto out;
+
+    new_frame = copy_frame(frame);
+    if (!new_frame)
+        goto out;
+
+    local = mem_get0(this->local_pool);
+    local->orig_frame = frame;
+
+    loc_copy(&local->loc, &tmp_loc);
+
+    new_frame->local = local;
+    new_frame->root->uid = args->uid;
+    new_frame->root->gid = args->gid;
+
+    if (S_ISDIR(args->st_mode)) {
+        STACK_WIND(new_frame, ga_newentry_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->mkdir, &tmp_loc,
+                   args->args.mkdir.mode, args->args.mkdir.umask, xdata);
+    } else if (S_ISLNK(args->st_mode)) {
+        STACK_WIND(new_frame, ga_newentry_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->symlink,
+                   args->args.symlink.linkpath, &tmp_loc, 0, xdata);
+    } else {
+        /* use 07777 (4 7s) for considering the Sticky bits etc) */
+        ((ga_local_t *)new_frame->local)->mode = (S_IFMT & args->st_mode) |
+                                                 (07777 &
+                                                  args->args.mknod.mode);
+
+        ((ga_local_t *)new_frame->local)->umask = args->args.mknod.umask;
+        ((ga_local_t *)new_frame->local)->rdev = args->args.mknod.rdev;
+        ((ga_local_t *)new_frame->local)->xdata = dict_ref(xdata);
+
+        /* send a named lookup, so that dht can cleanup up stale linkto
+         * files etc.
+         */
+        STACK_WIND(new_frame, ga_newentry_lookup_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->lookup, &tmp_loc, NULL);
+    }
+
+    ret = 0;
+out:
+    ga_newfile_args_free(args);
+
+    if (xdata)
+        dict_unref(xdata);
+
+    loc_wipe(&tmp_loc);
+
+    return ret;
+}
+
+int32_t
+ga_heal_entry(call_frame_t *frame, xlator_t *this, loc_t *loc, data_t *data,
+              dict_t *xdata)
+{
+    int ret = -1;
+    ga_heal_args_t *args = NULL;
+    loc_t tmp_loc = {
+        0,
+    };
+    call_frame_t *new_frame = NULL;
+    uuid_t gfid = {
+        0,
+    };
+
+    args = ga_heal_parse_args(this, data);
+    if (!args)
+        goto out;
+
+    ret = gf_uuid_parse(args->gfid, gfid);
+    if (ret)
+        goto out;
+
+    if (!xdata)
+        xdata = dict_new();
+    else
+        xdata = dict_ref(xdata);
+
+    if (!xdata) {
+        ret = -1;
+        goto out;
+    }
+
+    ret = ga_fill_tmp_loc(loc, this, gfid, args->bname, xdata, &tmp_loc);
+    if (ret)
+        goto out;
+
+    new_frame = copy_frame(frame);
+    if (!new_frame)
+        goto out;
+
+    new_frame->local = (void *)frame;
+
+    STACK_WIND(new_frame, ga_heal_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lookup, &tmp_loc, xdata);
+
+    ret = 0;
+out:
+    if (args)
+        ga_heal_args_free(args);
+
+    loc_wipe(&tmp_loc);
+
+    if (xdata)
+        dict_unref(xdata);
+
+    return ret;
+}
+
+int32_t
+ga_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(setxattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+int32_t
+ga_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+            int32_t flags, dict_t *xdata)
+{
+    data_t *data = NULL;
+    int op_errno = ENOMEM;
+    int ret = 0;
+    loc_t ga_loc = {
+        0,
+    };
+
+    GFID_ACCESS_INODE_OP_CHECK(loc, op_errno, err);
+
+    data = dict_get(dict, GF_FUSE_AUX_GFID_NEWFILE);
+    if (data) {
+        ret = ga_new_entry(frame, this, loc, data, xdata);
+        if (ret)
+            goto err;
+        return 0;
+    }
+
+    data = dict_get(dict, GF_FUSE_AUX_GFID_HEAL);
+    if (data) {
+        ret = ga_heal_entry(frame, this, loc, data, xdata);
+        if (ret)
+            goto err;
+        return 0;
+    }
+
+    // If the inode is a virtual inode change the inode otherwise perform
+    // the operation on same inode
+    ret = ga_valid_inode_loc_copy(&ga_loc, loc, this);
+    if (ret < 0)
+        goto err;
+
+    STACK_WIND(frame, ga_setxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setxattr, &ga_loc, dict, flags, xdata);
+
+    loc_wipe(&ga_loc);
+    return 0;
+err:
+    STACK_UNWIND_STRICT(setxattr, frame, -1, op_errno, xdata);
+    return 0;
+}
+
+int32_t
+ga_virtual_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, inode_t *inode,
+                      struct iatt *buf, dict_t *xdata, struct iatt *postparent)
+{
+    int ret = 0;
+    inode_t *cbk_inode = NULL;
+    inode_t *true_inode = NULL;
+    uuid_t random_gfid = {
+        0,
+    };
+    inode_t *linked_inode = NULL;
+
+    if (frame->local)
+        cbk_inode = frame->local;
+    else
+        cbk_inode = inode_ref(inode);
+
+    frame->local = NULL;
+    if (op_ret)
+        goto unwind;
+
+    if (!IA_ISDIR(buf->ia_type))
+        goto unwind;
+
+    /* need to send back a different inode for linking in itable */
+    if (cbk_inode == inode) {
+        /* check if the inode is in the 'itable' or
+           if its just previously discover()'d inode */
+        true_inode = inode_find(inode->table, buf->ia_gfid);
+        if (!true_inode) {
+            /* This unref is for 'inode_ref()' done in beginning.
+               This is needed as cbk_inode is allocated new inode
+               whose unref is taken at the end*/
+            inode_unref(cbk_inode);
+            cbk_inode = inode_new(inode->table);
+
+            if (!cbk_inode) {
+                op_ret = -1;
+                op_errno = ENOMEM;
+                goto unwind;
+            }
+            /* the inode is not present in itable, ie, the actual
+               path is not yet looked up. Use the current inode
+               itself for now */
+
+            linked_inode = inode_link(inode, NULL, NULL, buf);
+            inode = linked_inode;
+        } else {
+            /* 'inode_ref()' has been done in inode_find() */
+            inode = true_inode;
+        }
+
+        ret = inode_ctx_put(cbk_inode, this, (uint64_t)(uintptr_t)inode);
+        if (ret) {
+            gf_log(this->name, GF_LOG_WARNING,
+                   "failed to set the inode ctx with"
+                   "the actual inode");
+            if (inode)
+                inode_unref(inode);
+        }
+        inode = NULL;
+    }
+
+    if (!gf_uuid_is_null(cbk_inode->gfid)) {
+        /* if the previous linked inode is used, use the
+           same gfid */
+        gf_uuid_copy(random_gfid, cbk_inode->gfid);
+    } else {
+        /* replace the buf->ia_gfid to a random gfid
+           for directory, for files, what we received is fine */
+        gf_uuid_generate(random_gfid);
+    }
+
+    gf_uuid_copy(buf->ia_gfid, random_gfid);
+
+    buf->ia_ino = gfid_to_ino(buf->ia_gfid);
+
+unwind:
+    /* Lookup on non-existing gfid returns ESTALE.
+       Convert into ENOENT for virtual lookup*/
+    if (op_errno == ESTALE)
+        op_errno = ENOENT;
+
+    STACK_UNWIND_STRICT(lookup, frame, op_ret, op_errno, cbk_inode, buf, xdata,
+                        postparent);
+
+    /* Also handles inode_unref of frame->local if done in ga_lookup */
+    if (cbk_inode)
+        inode_unref(cbk_inode);
+
+    return 0;
+}
+
+int32_t
+ga_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, inode_t *inode, struct iatt *buf, dict_t *xdata,
+              struct iatt *postparent)
+{
+    ga_private_t *priv = NULL;
+
+    /* if the entry in question is not 'root',
+       then follow the normal path */
+    if (op_ret || !__is_root_gfid(buf->ia_gfid))
+        goto unwind;
+
+    priv = this->private;
+
+    /* do we need to copy root stbuf every time? */
+    /* mostly yes, as we want to have the 'stat' info show latest
+       in every _cbk() */
+
+    /* keep the reference for root stat buf */
+    priv->root_stbuf = *buf;
+    priv->gfiddir_stbuf = priv->root_stbuf;
+    priv->gfiddir_stbuf.ia_gfid[15] = GF_AUX_GFID;
+    priv->gfiddir_stbuf.ia_ino = GF_AUX_GFID;
+
+unwind:
+    STACK_UNWIND_STRICT(lookup, frame, op_ret, op_errno, inode, buf, xdata,
+                        postparent);
+    return 0;
+}
+
+int32_t
+ga_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    ga_private_t *priv = NULL;
+    int ret = -1;
+    uuid_t tmp_gfid = {
+        0,
+    };
+    loc_t tmp_loc = {
+        0,
+    };
+    uint64_t value = 0;
+    inode_t *inode = NULL;
+    inode_t *true_inode = NULL;
+    int32_t op_errno = ENOENT;
+
+    priv = this->private;
+
+    /* Handle nameless lookup on ".gfid" */
+    if (!loc->parent && __is_gfid_access_dir(loc->gfid)) {
+        STACK_UNWIND_STRICT(lookup, frame, 0, 0, loc->inode,
+                            &priv->gfiddir_stbuf, xdata, &priv->root_stbuf);
+        return 0;
+    }
+
+    /* if its discover(), no need for any action here */
+    if (!loc->name)
+        goto wind;
+
+    /* if its revalidate, and inode is not of type directory,
+       proceed with 'wind' */
+    if (loc->inode && loc->inode->ia_type && !IA_ISDIR(loc->inode->ia_type)) {
+        /* a revalidate on ".gfid/<dentry>" is possible, check for it */
+        if (((loc->parent && __is_gfid_access_dir(loc->parent->gfid)) ||
+             __is_gfid_access_dir(loc->pargfid))) {
+            /* here, just send 'loc->gfid' and 'loc->inode' */
+            tmp_loc.inode = inode_ref(loc->inode);
+            gf_uuid_copy(tmp_loc.gfid, loc->inode->gfid);
+
+            STACK_WIND(frame, default_lookup_cbk, FIRST_CHILD(this),
+                       FIRST_CHILD(this)->fops->lookup, &tmp_loc, xdata);
+
+            inode_unref(tmp_loc.inode);
+
+            return 0;
+        }
+
+        /* not something to bother, continue the flow */
+        goto wind;
+    }
+
+    /* need to check if the lookup is on virtual dir */
+    if ((loc->name && !strcmp(GF_GFID_DIR, loc->name)) &&
+        ((loc->parent && __is_root_gfid(loc->parent->gfid)) ||
+         __is_root_gfid(loc->pargfid))) {
+        /* this means, the query is on '/.gfid', return the fake stat,
+           and say success */
+
+        STACK_UNWIND_STRICT(lookup, frame, 0, 0, loc->inode,
+                            &priv->gfiddir_stbuf, xdata, &priv->root_stbuf);
+        return 0;
+    }
+
+    /* now, check if the lookup() is on an existing entry,
+       but on gfid-path */
+    if (!((loc->parent && __is_gfid_access_dir(loc->parent->gfid)) ||
+          __is_gfid_access_dir(loc->pargfid))) {
+        if (!loc->parent)
+            goto wind;
+
+        ret = inode_ctx_get(loc->parent, this, &value);
+        if (ret)
+            goto wind;
+
+        inode = (inode_t *)(uintptr_t)value;
+
+        ret = loc_copy_overload_parent(&tmp_loc, loc, inode);
+        if (ret)
+            goto err;
+
+        STACK_WIND(frame, ga_lookup_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->lookup, &tmp_loc, xdata);
+
+        loc_wipe(&tmp_loc);
+        return 0;
+    }
+
+    /* make sure the 'basename' is actually a 'canonical-gfid',
+       otherwise, return error */
+    ret = gf_uuid_parse(loc->name, tmp_gfid);
+    if (ret)
+        goto err;
+
+    /* if its fresh lookup, go ahead and send it down, if not,
+       for directory, we need indirection to actual dir inode */
+    if (!(loc->inode && loc->inode->ia_type))
+        goto discover;
+
+    /* revalidate on directory */
+    ret = inode_ctx_get(loc->inode, this, &value);
+    if (ret)
+        goto err;
+
+    inode = (void *)(uintptr_t)value;
+
+    /* valid inode, already looked up, work on that */
+    if (inode->ia_type)
+        goto discover;
+
+    /* check if the inode is in the 'itable' or
+       if its just previously discover()'d inode */
+    true_inode = inode_find(loc->inode->table, tmp_gfid);
+    if (true_inode) {
+        /* time do another lookup and update the context
+           with proper inode */
+        op_errno = ESTALE;
+        /* 'inode_ref()' done in inode_find */
+        inode_unref(true_inode);
+        goto err;
+    }
+
+discover:
+    /* for the virtual entries, we don't need to send 'gfid-req' key, as
+       for these entries, we don't want to 'set' a new gfid */
+    if (xdata)
+        dict_del(xdata, "gfid-req");
+
+    gf_uuid_copy(tmp_loc.gfid, tmp_gfid);
+
+    /* if revalidate, then we need to have the proper reference */
+    if (inode) {
+        tmp_loc.inode = inode_ref(inode);
+        frame->local = inode_ref(loc->inode);
+    } else {
+        tmp_loc.inode = inode_ref(loc->inode);
+    }
+
+    STACK_WIND(frame, ga_virtual_lookup_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lookup, &tmp_loc, xdata);
+
+    inode_unref(tmp_loc.inode);
+
+    return 0;
+
+wind:
+    /* used for all the normal lookup path */
+    STACK_WIND(frame, ga_lookup_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lookup, loc, xdata);
+
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(lookup, frame, -1, op_errno, loc->inode,
+                        &priv->gfiddir_stbuf, xdata, &priv->root_stbuf);
+    return 0;
+}
+
+int
+ga_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+         mode_t umask, dict_t *xdata)
+{
+    int op_errno = ENOMEM;
+
+    GFID_ACCESS_ENTRY_OP_CHECK(loc, op_errno, err);
+
+    STACK_WIND(frame, default_mkdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mkdir, loc, mode, umask, xdata);
+
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(mkdir, frame, -1, op_errno, loc->inode, NULL, NULL,
+                        NULL, xdata);
+    return 0;
+}
+
+int
+ga_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+          mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+    int op_errno = ENOMEM;
+
+    GFID_ACCESS_ENTRY_OP_CHECK(loc, op_errno, err);
+
+    STACK_WIND(frame, default_create_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->create, loc, flags, mode, umask, fd,
+               xdata);
+    return 0;
+err:
+    STACK_UNWIND_STRICT(create, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                        NULL, xdata);
+
+    return 0;
+}
+
+int
+ga_symlink(call_frame_t *frame, xlator_t *this, const char *linkname,
+           loc_t *loc, mode_t umask, dict_t *xdata)
+{
+    int op_errno = ENOMEM;
+
+    GFID_ACCESS_ENTRY_OP_CHECK(loc, op_errno, err);
+
+    STACK_WIND(frame, default_symlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->symlink, linkname, loc, umask, xdata);
+    return 0;
+err:
+    STACK_UNWIND_STRICT(symlink, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                        xdata);
+
+    return 0;
+}
+
+int
+ga_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+         dev_t rdev, mode_t umask, dict_t *xdata)
+{
+    int op_errno = ENOMEM;
+
+    GFID_ACCESS_ENTRY_OP_CHECK(loc, op_errno, err);
+
+    STACK_WIND(frame, default_mknod_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mknod, loc, mode, rdev, umask, xdata);
+
+    return 0;
+err:
+    STACK_UNWIND_STRICT(mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                        xdata);
+
+    return 0;
+}
+
+int
+ga_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flag,
+         dict_t *xdata)
+{
+    int op_errno = ENOMEM;
+    int ret = -1;
+    loc_t ga_loc = {
+        0,
+    };
+
+    GFID_ACCESS_ENTRY_OP_CHECK(loc, op_errno, err);
+
+    ret = ga_valid_inode_loc_copy(&ga_loc, loc, this);
+    if (ret < 0)
+        goto err;
+
+    STACK_WIND(frame, default_rmdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rmdir, &ga_loc, flag, xdata);
+
+    loc_wipe(&ga_loc);
+    return 0;
+err:
+    STACK_UNWIND_STRICT(rmdir, frame, -1, op_errno, NULL, NULL, xdata);
+
+    return 0;
+}
+
+int
+ga_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t xflag,
+          dict_t *xdata)
+{
+    int op_errno = ENOMEM;
+    int ret = -1;
+    loc_t ga_loc = {
+        0,
+    };
+
+    GFID_ACCESS_ENTRY_OP_CHECK(loc, op_errno, err);
+
+    ret = ga_valid_inode_loc_copy(&ga_loc, loc, this);
+    if (ret < 0)
+        goto err;
+
+    STACK_WIND(frame, default_unlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->unlink, &ga_loc, xflag, xdata);
+
+    loc_wipe(&ga_loc);
+    return 0;
+err:
+    STACK_UNWIND_STRICT(unlink, frame, -1, op_errno, NULL, NULL, xdata);
+
+    return 0;
+}
+
+int
+ga_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+          dict_t *xdata)
+{
+    int op_errno = ENOMEM;
+    int ret = 0;
+    loc_t ga_oldloc = {
+        0,
+    };
+    loc_t ga_newloc = {
+        0,
+    };
+
+    GFID_ACCESS_ENTRY_OP_CHECK(oldloc, op_errno, err);
+    GFID_ACCESS_ENTRY_OP_CHECK(newloc, op_errno, err);
+
+    ret = ga_valid_inode_loc_copy(&ga_oldloc, oldloc, this);
+    if (ret < 0)
+        goto err;
+
+    ret = ga_valid_inode_loc_copy(&ga_newloc, newloc, this);
+    if (ret < 0) {
+        loc_wipe(&ga_oldloc);
+        goto err;
+    }
+
+    STACK_WIND(frame, default_rename_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rename, &ga_oldloc, &ga_newloc, xdata);
+
+    loc_wipe(&ga_newloc);
+    loc_wipe(&ga_oldloc);
+    return 0;
+err:
+    STACK_UNWIND_STRICT(rename, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                        NULL, xdata);
+
+    return 0;
+}
+
+int
+ga_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+        dict_t *xdata)
+{
+    int op_errno = ENOMEM;
+    int ret = 0;
+    loc_t ga_oldloc = {
+        0,
+    };
+    loc_t ga_newloc = {
+        0,
+    };
+
+    GFID_ACCESS_ENTRY_OP_CHECK(oldloc, op_errno, err);
+    GFID_ACCESS_ENTRY_OP_CHECK(newloc, op_errno, err);
+
+    ret = ga_valid_inode_loc_copy(&ga_oldloc, oldloc, this);
+    if (ret < 0)
+        goto err;
+
+    ret = ga_valid_inode_loc_copy(&ga_newloc, newloc, this);
+    if (ret < 0) {
+        loc_wipe(&ga_oldloc);
+        goto err;
+    }
+
+    STACK_WIND(frame, default_link_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->link, &ga_oldloc, &ga_newloc, xdata);
+
+    loc_wipe(&ga_newloc);
+    loc_wipe(&ga_oldloc);
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(link, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                        xdata);
+
+    return 0;
+}
+
+int32_t
+ga_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+           dict_t *xdata)
+{
+    int op_errno = ENOMEM;
+
+    GFID_ACCESS_INODE_OP_CHECK(loc, op_errno, err);
+
+    /* also check if the loc->inode itself is virtual
+       inode, if yes, return with failure, mainly because we
+       can't handle all the readdirp and other things on it. */
+    if (inode_ctx_get(loc->inode, this, NULL) == 0) {
+        op_errno = ENOTSUP;
+        goto err;
+    }
+
+    STACK_WIND(frame, default_opendir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->opendir, loc, fd, xdata);
+    return 0;
+err:
+    STACK_UNWIND_STRICT(opendir, frame, -1, op_errno, NULL, xdata);
+
+    return 0;
+}
+
+int32_t
+ga_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, const char *name,
+            dict_t *xdata)
+{
+    int op_errno = ENOMEM;
+    int ret = -1;
+    loc_t ga_loc = {
+        0,
+    };
+
+    GFID_ACCESS_INODE_OP_CHECK(loc, op_errno, err);
+    ret = ga_valid_inode_loc_copy(&ga_loc, loc, this);
+    if (ret < 0)
+        goto err;
+
+    STACK_WIND(frame, default_getxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->getxattr, &ga_loc, name, xdata);
+
+    loc_wipe(&ga_loc);
+
+    return 0;
+err:
+    STACK_UNWIND_STRICT(getxattr, frame, -1, op_errno, NULL, xdata);
+
+    return 0;
+}
+
+int32_t
+ga_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    int op_errno = ENOMEM;
+    int ret = -1;
+    loc_t ga_loc = {
+        0,
+    };
+    ga_private_t *priv = NULL;
+
+    priv = this->private;
+    /* If stat is on ".gfid" itself, do not wind further,
+     * return fake stat and return success.
+     */
+    if (__is_gfid_access_dir(loc->gfid))
+        goto out;
+
+    ret = ga_valid_inode_loc_copy(&ga_loc, loc, this);
+    if (ret < 0)
+        goto err;
+
+    STACK_WIND(frame, default_stat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->stat, &ga_loc, xdata);
+
+    loc_wipe(&ga_loc);
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(stat, frame, -1, op_errno, NULL, xdata);
+
+    return 0;
+
+out:
+    STACK_UNWIND_STRICT(stat, frame, 0, 0, &priv->gfiddir_stbuf, xdata);
+    return 0;
+}
+
+int32_t
+ga_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *stbuf,
+           int32_t valid, dict_t *xdata)
+{
+    int op_errno = ENOMEM;
+    int ret = -1;
+    loc_t ga_loc = {
+        0,
+    };
+
+    GFID_ACCESS_INODE_OP_CHECK(loc, op_errno, err);
+    ret = ga_valid_inode_loc_copy(&ga_loc, loc, this);
+    if (ret < 0)
+        goto err;
+
+    STACK_WIND(frame, default_setattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setattr, &ga_loc, stbuf, valid, xdata);
+
+    loc_wipe(&ga_loc);
+    return 0;
+err:
+    STACK_UNWIND_STRICT(setattr, frame, -1, op_errno, NULL, NULL, xdata);
+
+    return 0;
+}
+
+int32_t
+ga_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               const char *name, dict_t *xdata)
+{
+    int op_errno = ENOMEM;
+    int ret = -1;
+    loc_t ga_loc = {
+        0,
+    };
+
+    GFID_ACCESS_INODE_OP_CHECK(loc, op_errno, err);
+    ret = ga_valid_inode_loc_copy(&ga_loc, loc, this);
+    if (ret < 0)
+        goto err;
+
+    STACK_WIND(frame, default_removexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->removexattr, &ga_loc, name, xdata);
+
+    loc_wipe(&ga_loc);
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(removexattr, frame, -1, op_errno, xdata);
+
+    return 0;
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    if (!this)
+        return ret;
+
+    ret = xlator_mem_acct_init(this, gf_gfid_access_mt_end + 1);
+
+    if (ret != 0) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "Memory accounting"
+               " init failed");
+        return ret;
+    }
+
+    return ret;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    ga_private_t *priv = NULL;
+    int ret = -1;
+
+    if (!this->children || this->children->next) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "not configured with exactly one child. exiting");
+        goto out;
+    }
+
+    /* This can be the top of graph in certain cases */
+    if (!this->parents) {
+        gf_log(this->name, GF_LOG_DEBUG, "dangling volume. check volfile ");
+    }
+
+    /* TODO: define a mem-type structure */
+    priv = GF_CALLOC(1, sizeof(*priv), gf_gfid_access_mt_priv_t);
+    if (!priv)
+        goto out;
+
+    priv->newfile_args_pool = mem_pool_new(ga_newfile_args_t, 512);
+    if (!priv->newfile_args_pool)
+        goto out;
+
+    priv->heal_args_pool = mem_pool_new(ga_heal_args_t, 512);
+    if (!priv->heal_args_pool)
+        goto out;
+
+    this->local_pool = mem_pool_new(ga_local_t, 16);
+    if (!this->local_pool) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to create local_t's memory pool");
+        goto out;
+    }
+
+    this->private = priv;
+
+    ret = 0;
+out:
+    if (ret && priv) {
+        if (priv->newfile_args_pool)
+            mem_pool_destroy(priv->newfile_args_pool);
+        GF_FREE(priv);
+    }
+
+    return ret;
+}
+
+void
+fini(xlator_t *this)
+{
+    ga_private_t *priv = NULL;
+    priv = this->private;
+    this->private = NULL;
+
+    if (priv) {
+        if (priv->newfile_args_pool)
+            mem_pool_destroy(priv->newfile_args_pool);
+        if (priv->heal_args_pool)
+            mem_pool_destroy(priv->heal_args_pool);
+        GF_FREE(priv);
+    }
+
+    return;
+}
+
+int32_t
+ga_dump_inodectx(xlator_t *this, inode_t *inode)
+{
+    int ret = -1;
+    uint64_t value = 0;
+    inode_t *tmp_inode = NULL;
+    char key_prefix[GF_DUMP_MAX_BUF_LEN] = {
+        0,
+    };
+
+    ret = inode_ctx_get(inode, this, &value);
+    if (ret == 0) {
+        tmp_inode = (void *)(uintptr_t)value;
+        gf_proc_dump_build_key(key_prefix, this->name, "inode");
+        gf_proc_dump_add_section("%s", key_prefix);
+        gf_proc_dump_write("real-gfid", "%s", uuid_utoa(tmp_inode->gfid));
+    }
+
+    return 0;
+}
+
+struct xlator_fops fops = {
+    .lookup = ga_lookup,
+
+    /* entry fops */
+    .mkdir = ga_mkdir,
+    .mknod = ga_mknod,
+    .create = ga_create,
+    .symlink = ga_symlink,
+    .link = ga_link,
+    .unlink = ga_unlink,
+    .rmdir = ga_rmdir,
+    .rename = ga_rename,
+
+    /* handle any other directory operations here */
+    .opendir = ga_opendir,
+    .stat = ga_stat,
+    .setattr = ga_setattr,
+    .getxattr = ga_getxattr,
+    .removexattr = ga_removexattr,
+
+    /* special fop to handle more entry creations */
+    .setxattr = ga_setxattr,
+};
+
+struct xlator_cbks cbks = {
+    .forget = ga_forget,
+};
+
+struct xlator_dumpops dumpops = {
+    .inodectx = ga_dump_inodectx,
+};
+
+struct volume_options options[] = {
+    /* This translator doesn't take any options, or provide any options */
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1},
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "gfid-access",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/features/gfid-access/src/gfid-access.h b/xlators/features/gfid-access/src/gfid-access.h
new file mode 100644
index 00000000000..b1e255e56c0
--- /dev/null
+++ b/xlators/features/gfid-access/src/gfid-access.h
@@ -0,0 +1,107 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#ifndef __GFID_ACCESS_H__
+#define __GFID_ACCESS_H__
+
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/dict.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+#include "gfid-access-mem-types.h"
+
+#define UUID_CANONICAL_FORM_LEN 36
+
+#define GF_FUSE_AUX_GFID_NEWFILE "glusterfs.gfid.newfile"
+#define GF_FUSE_AUX_GFID_HEAL "glusterfs.gfid.heal"
+
+#define GF_GFID_KEY "GLUSTERFS_GFID"
+#define GF_GFID_DIR ".gfid"
+#define GF_AUX_GFID 0xd
+
+#define GFID_ACCESS_ENTRY_OP_CHECK(loc, err, lbl)                              \
+    do {                                                                       \
+        /* need to check if the lookup is on virtual dir */                    \
+        if ((loc->name && !strcmp(GF_GFID_DIR, loc->name)) &&                  \
+            ((loc->parent && __is_root_gfid(loc->parent->gfid)) ||             \
+             __is_root_gfid(loc->pargfid))) {                                  \
+            err = ENOTSUP;                                                     \
+            goto lbl;                                                          \
+        }                                                                      \
+                                                                               \
+        /* now, check if the lookup() is on an existing */                     \
+        /* entry, but on gfid-path */                                          \
+        if ((loc->parent && __is_gfid_access_dir(loc->parent->gfid)) ||        \
+            __is_gfid_access_dir(loc->pargfid)) {                              \
+            err = EPERM;                                                       \
+            goto lbl;                                                          \
+        }                                                                      \
+    } while (0)
+
+#define GFID_ACCESS_INODE_OP_CHECK(loc, err, lbl)                              \
+    do {                                                                       \
+        /*Check if it is on .gfid*/                                            \
+        if (__is_gfid_access_dir(loc->gfid)) {                                 \
+            err = ENOTSUP;                                                     \
+            goto lbl;                                                          \
+        }                                                                      \
+    } while (0)
+typedef struct {
+    unsigned int uid;
+    unsigned int gid;
+    char gfid[UUID_CANONICAL_FORM_LEN + 1];
+    unsigned int st_mode;
+    char *bname;
+
+    union {
+        struct _symlink_in {
+            char *linkpath;
+        } __attribute__((__packed__)) symlink;
+
+        struct _mknod_in {
+            unsigned int mode;
+            unsigned int rdev;
+            unsigned int umask;
+        } __attribute__((__packed__)) mknod;
+
+        struct _mkdir_in {
+            unsigned int mode;
+            unsigned int umask;
+        } __attribute__((__packed__)) mkdir;
+    } __attribute__((__packed__)) args;
+} __attribute__((__packed__)) ga_newfile_args_t;
+
+typedef struct {
+    char gfid[UUID_CANONICAL_FORM_LEN + 1];
+    char *bname; /* a null terminated basename */
+} __attribute__((__packed__)) ga_heal_args_t;
+
+struct ga_private {
+    /* root inode's stbuf */
+    struct iatt root_stbuf;
+    struct iatt gfiddir_stbuf;
+    struct mem_pool *newfile_args_pool;
+    struct mem_pool *heal_args_pool;
+};
+typedef struct ga_private ga_private_t;
+
+struct __ga_local {
+    call_frame_t *orig_frame;
+    unsigned int uid;
+    unsigned int gid;
+    loc_t loc;
+    mode_t mode;
+    dev_t rdev;
+    mode_t umask;
+    dict_t *xdata;
+};
+typedef struct __ga_local ga_local_t;
+
+#endif /* __GFID_ACCESS_H__ */
diff --git a/xlators/features/index/Makefile.am b/xlators/features/index/Makefile.am
new file mode 100644
index 00000000000..a985f42a877
--- /dev/null
+++ b/xlators/features/index/Makefile.am
@@ -0,0 +1,3 @@
+SUBDIRS = src
+
+CLEANFILES =
diff --git a/xlators/features/index/src/Makefile.am b/xlators/features/index/src/Makefile.am
new file mode 100644
index 00000000000..c71c238c163
--- /dev/null
+++ b/xlators/features/index/src/Makefile.am
@@ -0,0 +1,19 @@
+if WITH_SERVER
+xlator_LTLIBRARIES = index.la
+endif
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+index_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+index_la_SOURCES = index.c
+index_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = index.h index-mem-types.h index-messages.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src \
+	-I$(top_srcdir)/rpc/rpc-lib/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/index/src/index-mem-types.h b/xlators/features/index/src/index-mem-types.h
new file mode 100644
index 00000000000..58833d0ec9b
--- /dev/null
+++ b/xlators/features/index/src/index-mem-types.h
@@ -0,0 +1,23 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __INDEX_MEM_TYPES_H__
+#define __INDEX_MEM_TYPES_H__
+
+#include <glusterfs/mem-types.h>
+
+enum gf_index_mem_types_ {
+    gf_index_mt_priv_t = gf_common_mt_end + 1,
+    gf_index_inode_ctx_t,
+    gf_index_fd_ctx_t,
+    gf_index_mt_local_t,
+    gf_index_mt_end
+};
+#endif
diff --git a/xlators/features/index/src/index-messages.h b/xlators/features/index/src/index-messages.h
new file mode 100644
index 00000000000..364f17cd34e
--- /dev/null
+++ b/xlators/features/index/src/index-messages.h
@@ -0,0 +1,33 @@
+/*
+ Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com>
+ This file is part of GlusterFS.
+
+ This file is licensed to you under your choice of the GNU Lesser
+ General Public License, version 3 or any later version (LGPLv3 or
+ later), or the GNU General Public License, version 2 (GPLv2), in all
+ cases as published by the Free Software Foundation.
+ */
+
+#ifndef _INDEX_MESSAGES_H_
+#define _INDEX_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(INDEX, INDEX_MSG_INDEX_DIR_CREATE_FAILED,
+           INDEX_MSG_INDEX_READDIR_FAILED, INDEX_MSG_INDEX_ADD_FAILED,
+           INDEX_MSG_INDEX_DEL_FAILED, INDEX_MSG_DICT_SET_FAILED,
+           INDEX_MSG_INODE_CTX_GET_SET_FAILED, INDEX_MSG_INVALID_ARGS,
+           INDEX_MSG_FD_OP_FAILED, INDEX_MSG_WORKER_THREAD_CREATE_FAILED,
+           INDEX_MSG_INVALID_GRAPH);
+
+#endif /* !_INDEX_MESSAGES_H_ */
diff --git a/xlators/features/index/src/index.c b/xlators/features/index/src/index.c
new file mode 100644
index 00000000000..4abb2c73ce5
--- /dev/null
+++ b/xlators/features/index/src/index.c
@@ -0,0 +1,2682 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include "index.h"
+#include <glusterfs/options.h>
+#include "glusterfs3-xdr.h"
+#include <glusterfs/syscall.h>
+#include <glusterfs/syncop.h>
+#include <glusterfs/common-utils.h>
+#include "index-messages.h"
+#include <ftw.h>
+#include <libgen.h> /* for dirname() */
+#include <signal.h>
+
+#define XATTROP_SUBDIR "xattrop"
+#define DIRTY_SUBDIR "dirty"
+#define ENTRY_CHANGES_SUBDIR "entry-changes"
+
+struct index_syncop_args {
+    inode_t *parent;
+    gf_dirent_t *entries;
+    char *path;
+};
+
+static char *index_vgfid_xattrs[XATTROP_TYPE_END] = {
+    [XATTROP] = GF_XATTROP_INDEX_GFID,
+    [DIRTY] = GF_XATTROP_DIRTY_GFID,
+    [ENTRY_CHANGES] = GF_XATTROP_ENTRY_CHANGES_GFID};
+
+static char *index_subdirs[XATTROP_TYPE_END] = {
+    [XATTROP] = XATTROP_SUBDIR,
+    [DIRTY] = DIRTY_SUBDIR,
+    [ENTRY_CHANGES] = ENTRY_CHANGES_SUBDIR};
+
+int
+index_get_type_from_vgfid(index_priv_t *priv, uuid_t vgfid)
+{
+    int i = 0;
+
+    for (i = 0; i < XATTROP_TYPE_END; i++) {
+        if (gf_uuid_compare(priv->internal_vgfid[i], vgfid) == 0)
+            return i;
+    }
+    return -1;
+}
+
+gf_boolean_t
+index_is_virtual_gfid(index_priv_t *priv, uuid_t vgfid)
+{
+    if (index_get_type_from_vgfid(priv, vgfid) < 0)
+        return _gf_false;
+    return _gf_true;
+}
+
+static int
+__index_inode_ctx_get(inode_t *inode, xlator_t *this, index_inode_ctx_t **ctx)
+{
+    int ret = 0;
+    index_inode_ctx_t *ictx = NULL;
+    uint64_t tmpctx = 0;
+
+    ret = __inode_ctx_get(inode, this, &tmpctx);
+    if (!ret) {
+        ictx = (index_inode_ctx_t *)(long)tmpctx;
+        goto out;
+    }
+    ictx = GF_CALLOC(1, sizeof(*ictx), gf_index_inode_ctx_t);
+    if (!ictx) {
+        ret = -1;
+        goto out;
+    }
+
+    INIT_LIST_HEAD(&ictx->callstubs);
+    ret = __inode_ctx_put(inode, this, (uint64_t)(uintptr_t)ictx);
+    if (ret) {
+        GF_FREE(ictx);
+        ictx = NULL;
+        goto out;
+    }
+out:
+    if (ictx)
+        *ctx = ictx;
+    return ret;
+}
+
+static int
+index_inode_ctx_get(inode_t *inode, xlator_t *this, index_inode_ctx_t **ctx)
+{
+    int ret = 0;
+
+    LOCK(&inode->lock);
+    {
+        ret = __index_inode_ctx_get(inode, this, ctx);
+    }
+    UNLOCK(&inode->lock);
+
+    return ret;
+}
+
+static gf_boolean_t
+index_is_subdir_of_entry_changes(xlator_t *this, inode_t *inode)
+{
+    index_inode_ctx_t *ctx = NULL;
+    int ret = 0;
+
+    if (!inode)
+        return _gf_false;
+
+    ret = index_inode_ctx_get(inode, this, &ctx);
+    if ((ret == 0) && !gf_uuid_is_null(ctx->virtual_pargfid))
+        return _gf_true;
+    return _gf_false;
+}
+
+static int
+index_get_type_from_vgfid_xattr(const char *name)
+{
+    int i = 0;
+
+    for (i = 0; i < XATTROP_TYPE_END; i++) {
+        if (strcmp(name, index_vgfid_xattrs[i]) == 0)
+            return i;
+    }
+    return -1;
+}
+
+gf_boolean_t
+index_is_fop_on_internal_inode(xlator_t *this, inode_t *inode, uuid_t gfid)
+{
+    index_priv_t *priv = this->private;
+    uuid_t vgfid = {0};
+
+    if (!inode)
+        return _gf_false;
+
+    if (gfid && !gf_uuid_is_null(gfid))
+        gf_uuid_copy(vgfid, gfid);
+    else
+        gf_uuid_copy(vgfid, inode->gfid);
+
+    if (index_is_virtual_gfid(priv, vgfid))
+        return _gf_true;
+    if (index_is_subdir_of_entry_changes(this, inode))
+        return _gf_true;
+    return _gf_false;
+}
+
+static gf_boolean_t
+index_is_vgfid_xattr(const char *name)
+{
+    if (index_get_type_from_vgfid_xattr(name) < 0)
+        return _gf_false;
+    return _gf_true;
+}
+
+call_stub_t *
+__index_dequeue(struct list_head *callstubs)
+{
+    call_stub_t *stub = NULL;
+
+    if (!list_empty(callstubs)) {
+        stub = list_entry(callstubs->next, call_stub_t, list);
+        list_del_init(&stub->list);
+    }
+
+    return stub;
+}
+
+static void
+__index_enqueue(struct list_head *callstubs, call_stub_t *stub)
+{
+    list_add_tail(&stub->list, callstubs);
+}
+
+static void
+worker_enqueue(xlator_t *this, call_stub_t *stub)
+{
+    index_priv_t *priv = NULL;
+
+    priv = this->private;
+    pthread_mutex_lock(&priv->mutex);
+    {
+        __index_enqueue(&priv->callstubs, stub);
+        GF_ATOMIC_INC(priv->stub_cnt);
+        pthread_cond_signal(&priv->cond);
+    }
+    pthread_mutex_unlock(&priv->mutex);
+}
+
+void *
+index_worker(void *data)
+{
+    index_priv_t *priv = NULL;
+    xlator_t *this = NULL;
+    call_stub_t *stub = NULL;
+    gf_boolean_t bye = _gf_false;
+
+    THIS = data;
+    this = data;
+    priv = this->private;
+
+    for (;;) {
+        pthread_mutex_lock(&priv->mutex);
+        {
+            while (list_empty(&priv->callstubs)) {
+                if (priv->down) {
+                    bye = _gf_true; /*Avoid wait*/
+                    break;
+                }
+                (void)pthread_cond_wait(&priv->cond, &priv->mutex);
+                if (priv->down) {
+                    bye = _gf_true;
+                    break;
+                }
+            }
+            if (!bye)
+                stub = __index_dequeue(&priv->callstubs);
+            if (bye) {
+                priv->curr_count--;
+                if (priv->curr_count == 0)
+                    pthread_cond_broadcast(&priv->cond);
+            }
+        }
+        pthread_mutex_unlock(&priv->mutex);
+
+        if (stub) { /* guard against spurious wakeups */
+            call_resume(stub);
+            GF_ATOMIC_DEC(priv->stub_cnt);
+        }
+        stub = NULL;
+        if (bye)
+            break;
+    }
+
+    return NULL;
+}
+
+static void
+make_index_dir_path(char *base, const char *subdir, char *index_dir, size_t len)
+{
+    snprintf(index_dir, len, "%s/%s", base, subdir);
+}
+
+int
+index_dir_create(xlator_t *this, const char *subdir)
+{
+    int ret = 0;
+    struct stat st = {0};
+    char fullpath[PATH_MAX] = {0};
+    char path[PATH_MAX] = {0};
+    char *dir = NULL;
+    index_priv_t *priv = NULL;
+    size_t len = 0;
+    size_t pathlen = 0;
+
+    priv = this->private;
+    make_index_dir_path(priv->index_basepath, subdir, fullpath,
+                        sizeof(fullpath));
+    ret = sys_stat(fullpath, &st);
+    if (!ret) {
+        if (!S_ISDIR(st.st_mode))
+            ret = -2;
+        goto out;
+    }
+
+    pathlen = strlen(fullpath);
+    if ((pathlen > 1) && fullpath[pathlen - 1] == '/')
+        fullpath[pathlen - 1] = '\0';
+    dir = strchr(fullpath, '/');
+    while (dir) {
+        dir = strchr(dir + 1, '/');
+        if (dir)
+            len = pathlen - strlen(dir);
+        else
+            len = pathlen;
+        strncpy(path, fullpath, len);
+        path[len] = '\0';
+        ret = sys_mkdir(path, 0600);
+        if (ret && (errno != EEXIST))
+            goto out;
+    }
+    ret = 0;
+out:
+    if (ret == -1) {
+        gf_msg(this->name, GF_LOG_ERROR, errno,
+               INDEX_MSG_INDEX_DIR_CREATE_FAILED,
+               "%s/%s: Failed to "
+               "create",
+               priv->index_basepath, subdir);
+    } else if (ret == -2) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOTDIR,
+               INDEX_MSG_INDEX_DIR_CREATE_FAILED,
+               "%s/%s: Failed to "
+               "create, path exists, not a directory ",
+               priv->index_basepath, subdir);
+    }
+    return ret;
+}
+
+void
+index_get_index(index_priv_t *priv, uuid_t index)
+{
+    LOCK(&priv->lock);
+    {
+        gf_uuid_copy(index, priv->index);
+    }
+    UNLOCK(&priv->lock);
+}
+
+void
+index_generate_index(index_priv_t *priv, uuid_t index)
+{
+    LOCK(&priv->lock);
+    {
+        // To prevent duplicate generates.
+        // This method fails if number of contending threads is greater
+        // than MAX_LINK count of the fs
+        if (!gf_uuid_compare(priv->index, index))
+            gf_uuid_generate(priv->index);
+        gf_uuid_copy(index, priv->index);
+    }
+    UNLOCK(&priv->lock);
+}
+
+static void
+make_index_path(char *base, const char *subdir, uuid_t index, char *index_path,
+                size_t len)
+{
+    make_index_dir_path(base, subdir, index_path, len);
+    snprintf(index_path + strlen(index_path), len - strlen(index_path),
+             "/%s-%s", subdir, uuid_utoa(index));
+}
+
+static void
+make_gfid_path(char *base, const char *subdir, uuid_t gfid, char *gfid_path,
+               size_t len)
+{
+    make_index_dir_path(base, subdir, gfid_path, len);
+    snprintf(gfid_path + strlen(gfid_path), len - strlen(gfid_path), "/%s",
+             uuid_utoa(gfid));
+}
+
+static void
+make_file_path(char *base, const char *subdir, const char *filename,
+               char *file_path, size_t len)
+{
+    make_index_dir_path(base, subdir, file_path, len);
+    snprintf(file_path + strlen(file_path), len - strlen(file_path), "/%s",
+             filename);
+}
+
+static int
+is_index_file_current(char *filename, uuid_t priv_index, char *subdir)
+{
+    char current_index[GF_UUID_BUF_SIZE + 16] = {
+        0,
+    };
+
+    snprintf(current_index, sizeof current_index, "%s-%s", subdir,
+             uuid_utoa(priv_index));
+    return (!strcmp(filename, current_index));
+}
+
+static void
+check_delete_stale_index_file(xlator_t *this, char *filename, char *subdir)
+{
+    int ret = 0;
+    struct stat st = {0};
+    char filepath[PATH_MAX] = {0};
+    index_priv_t *priv = NULL;
+
+    priv = this->private;
+
+    if (is_index_file_current(filename, priv->index, subdir))
+        return;
+
+    make_file_path(priv->index_basepath, subdir, filename, filepath,
+                   sizeof(filepath));
+    ret = sys_stat(filepath, &st);
+    if (!ret && st.st_nlink == 1)
+        sys_unlink(filepath);
+}
+
+static void
+index_set_link_count(index_priv_t *priv, int64_t count,
+                     index_xattrop_type_t type)
+{
+    switch (type) {
+        case XATTROP:
+            LOCK(&priv->lock);
+            {
+                priv->pending_count = count;
+            }
+            UNLOCK(&priv->lock);
+            break;
+        default:
+            break;
+    }
+}
+
+static void
+index_get_link_count(index_priv_t *priv, int64_t *count,
+                     index_xattrop_type_t type)
+{
+    switch (type) {
+        case XATTROP:
+            LOCK(&priv->lock);
+            {
+                *count = priv->pending_count;
+            }
+            UNLOCK(&priv->lock);
+            break;
+        default:
+            break;
+    }
+}
+
+static void
+index_dec_link_count(index_priv_t *priv, index_xattrop_type_t type)
+{
+    switch (type) {
+        case XATTROP:
+            LOCK(&priv->lock);
+            {
+                priv->pending_count--;
+                if (priv->pending_count == 0)
+                    priv->pending_count--;
+            }
+            UNLOCK(&priv->lock);
+            break;
+        default:
+            break;
+    }
+}
+
+char *
+index_get_subdir_from_type(index_xattrop_type_t type)
+{
+    if (type < XATTROP || type >= XATTROP_TYPE_END)
+        return NULL;
+    return index_subdirs[type];
+}
+
+char *
+index_get_subdir_from_vgfid(index_priv_t *priv, uuid_t vgfid)
+{
+    return index_get_subdir_from_type(index_get_type_from_vgfid(priv, vgfid));
+}
+
+static int
+index_fill_readdir(fd_t *fd, index_fd_ctx_t *fctx, DIR *dir, off_t off,
+                   size_t size, gf_dirent_t *entries)
+{
+    off_t in_case = -1;
+    off_t last_off = 0;
+    size_t filled = 0;
+    int count = 0;
+    struct dirent *entry = NULL;
+    struct dirent scratch[2] = {
+        {
+            0,
+        },
+    };
+    int32_t this_size = -1;
+    gf_dirent_t *this_entry = NULL;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    if (!off) {
+        rewinddir(dir);
+    } else {
+        seekdir(dir, off);
+#ifndef GF_LINUX_HOST_OS
+        if ((u_long)telldir(dir) != off && off != fctx->dir_eof) {
+            gf_msg(THIS->name, GF_LOG_ERROR, EINVAL,
+                   INDEX_MSG_INDEX_READDIR_FAILED,
+                   "seekdir(0x%llx) failed on dir=%p: "
+                   "Invalid argument (offset reused from "
+                   "another DIR * structure?)",
+                   off, dir);
+            errno = EINVAL;
+            count = -1;
+            goto out;
+        }
+#endif /* GF_LINUX_HOST_OS */
+    }
+
+    while (filled <= size) {
+        in_case = (u_long)telldir(dir);
+
+        if (in_case == -1) {
+            gf_msg(THIS->name, GF_LOG_ERROR, errno,
+                   INDEX_MSG_INDEX_READDIR_FAILED, "telldir failed on dir=%p",
+                   dir);
+            goto out;
+        }
+
+        errno = 0;
+        entry = sys_readdir(dir, scratch);
+        if (!entry || errno != 0) {
+            if (errno == EBADF) {
+                gf_msg(THIS->name, GF_LOG_WARNING, errno,
+                       INDEX_MSG_INDEX_READDIR_FAILED,
+                       "readdir failed on dir=%p", dir);
+                goto out;
+            }
+            break;
+        }
+
+        if (!strncmp(entry->d_name, XATTROP_SUBDIR "-",
+                     strlen(XATTROP_SUBDIR "-"))) {
+            check_delete_stale_index_file(this, entry->d_name, XATTROP_SUBDIR);
+            continue;
+        } else if (!strncmp(entry->d_name, DIRTY_SUBDIR "-",
+                            strlen(DIRTY_SUBDIR "-"))) {
+            check_delete_stale_index_file(this, entry->d_name, DIRTY_SUBDIR);
+            continue;
+        }
+
+        this_size = max(sizeof(gf_dirent_t), sizeof(gfs3_dirplist)) +
+                    strlen(entry->d_name) + 1;
+
+        if (this_size + filled > size) {
+            seekdir(dir, in_case);
+#ifndef GF_LINUX_HOST_OS
+            if ((u_long)telldir(dir) != in_case && in_case != fctx->dir_eof) {
+                gf_msg(THIS->name, GF_LOG_ERROR, EINVAL,
+                       INDEX_MSG_INDEX_READDIR_FAILED,
+                       "seekdir(0x%llx) failed on dir=%p: "
+                       "Invalid argument (offset reused from "
+                       "another DIR * structure?)",
+                       in_case, dir);
+                errno = EINVAL;
+                count = -1;
+                goto out;
+            }
+#endif /* GF_LINUX_HOST_OS */
+            break;
+        }
+
+        this_entry = gf_dirent_for_name(entry->d_name);
+
+        if (!this_entry) {
+            gf_msg(THIS->name, GF_LOG_ERROR, errno,
+                   INDEX_MSG_INDEX_READDIR_FAILED,
+                   "could not create gf_dirent for entry %s", entry->d_name);
+            goto out;
+        }
+        /*
+         * we store the offset of next entry here, which is
+         * probably not intended, but code using syncop_readdir()
+         * (glfs-heal.c, afr-self-heald.c, pump.c) rely on it
+         * for directory read resumption.
+         */
+        last_off = (u_long)telldir(dir);
+        this_entry->d_off = last_off;
+        this_entry->d_ino = entry->d_ino;
+
+        list_add_tail(&this_entry->list, &entries->list);
+
+        filled += this_size;
+        count++;
+    }
+
+    errno = 0;
+
+    if ((!sys_readdir(dir, scratch) && (errno == 0))) {
+        /* Indicate EOF */
+        errno = ENOENT;
+        /* Remember EOF offset for later detection */
+        fctx->dir_eof = last_off;
+    }
+out:
+    return count;
+}
+
+int
+index_link_to_base(xlator_t *this, char *fpath, const char *subdir)
+{
+    int ret = 0;
+    int fd = 0;
+    int op_errno = 0;
+    uuid_t index = {0};
+    index_priv_t *priv = this->private;
+    char base[PATH_MAX] = {0};
+
+    index_get_index(priv, index);
+    make_index_path(priv->index_basepath, subdir, index, base, sizeof(base));
+
+    ret = sys_link(base, fpath);
+    if (!ret || (errno == EEXIST)) {
+        ret = 0;
+        goto out;
+    }
+
+    op_errno = errno;
+    if (op_errno == ENOENT) {
+        ret = index_dir_create(this, subdir);
+        if (ret) {
+            op_errno = errno;
+            goto out;
+        }
+    } else if (op_errno == EMLINK) {
+        index_generate_index(priv, index);
+        make_index_path(priv->index_basepath, subdir, index, base,
+                        sizeof(base));
+    } else {
+        goto out;
+    }
+
+    op_errno = 0;
+    fd = sys_creat(base, 0);
+    if ((fd < 0) && (errno != EEXIST)) {
+        op_errno = errno;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, INDEX_MSG_INDEX_ADD_FAILED,
+               "%s: Not able to "
+               "create index",
+               fpath);
+        goto out;
+    }
+
+    if (fd >= 0)
+        sys_close(fd);
+
+    ret = sys_link(base, fpath);
+    if (ret && (errno != EEXIST)) {
+        op_errno = errno;
+        gf_msg(this->name, GF_LOG_ERROR, errno, INDEX_MSG_INDEX_ADD_FAILED,
+               "%s: Not able to "
+               "add to index",
+               fpath);
+        goto out;
+    }
+out:
+    return -op_errno;
+}
+
+int
+index_add(xlator_t *this, uuid_t gfid, const char *subdir,
+          index_xattrop_type_t type)
+{
+    char gfid_path[PATH_MAX] = {0};
+    int ret = -1;
+    index_priv_t *priv = NULL;
+    struct stat st = {0};
+
+    priv = this->private;
+
+    if (gf_uuid_is_null(gfid)) {
+        GF_ASSERT(0);
+        goto out;
+    }
+
+    make_gfid_path(priv->index_basepath, subdir, gfid, gfid_path,
+                   sizeof(gfid_path));
+
+    ret = sys_stat(gfid_path, &st);
+    if (!ret)
+        goto out;
+    ret = index_link_to_base(this, gfid_path, subdir);
+out:
+    return ret;
+}
+
+int
+index_del(xlator_t *this, uuid_t gfid, const char *subdir, int type)
+{
+    int32_t op_errno __attribute__((unused)) = 0;
+    index_priv_t *priv = NULL;
+    int ret = 0;
+    char gfid_path[PATH_MAX] = {0};
+    char rename_dst[PATH_MAX] = {
+        0,
+    };
+    uuid_t uuid;
+
+    priv = this->private;
+    GF_ASSERT_AND_GOTO_WITH_ERROR(this->name, !gf_uuid_is_null(gfid), out,
+                                  op_errno, EINVAL);
+    make_gfid_path(priv->index_basepath, subdir, gfid, gfid_path,
+                   sizeof(gfid_path));
+
+    if ((strcmp(subdir, ENTRY_CHANGES_SUBDIR)) == 0) {
+        ret = sys_rmdir(gfid_path);
+        /* rmdir above could fail with ENOTEMPTY if the indices under
+         * it were created when granular-entry-heal was enabled, whereas
+         * the actual heal that happened was non-granular (or full) in
+         * nature, resulting in name indices getting left out. To
+         * clean up this directory without it affecting the IO path perf,
+         * the directory is renamed to a unique name under
+         * indices/entry-changes. Self-heal will pick up this entry
+         * during crawl and on lookup into the file system figure that
+         * the index is stale and subsequently wipe it out using rmdir().
+         */
+        if ((ret) && (errno == ENOTEMPTY)) {
+            gf_uuid_generate(uuid);
+            make_gfid_path(priv->index_basepath, subdir, uuid, rename_dst,
+                           sizeof(rename_dst));
+            ret = sys_rename(gfid_path, rename_dst);
+        }
+    } else {
+        ret = sys_unlink(gfid_path);
+    }
+
+    if (ret && (errno != ENOENT)) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, INDEX_MSG_INDEX_DEL_FAILED,
+               "%s: failed to delete"
+               " from index",
+               gfid_path);
+        ret = -errno;
+        goto out;
+    }
+
+    index_dec_link_count(priv, type);
+    ret = 0;
+out:
+    return ret;
+}
+
+static gf_boolean_t
+_is_xattr_in_watchlist(dict_t *d, char *k, data_t *v, void *tmp)
+{
+    if (!strncmp(k, tmp, strlen(k)))
+        return _gf_true;
+
+    return _gf_false;
+}
+
+static gf_boolean_t
+is_xattr_in_watchlist(dict_t *this, char *key, data_t *value, void *matchdata)
+{
+    int ret = -1;
+
+    // matchdata is a list of xattrs
+    // key is strncmp'ed with each xattr in matchdata.
+    // ret will be 0 if key pattern is not present in the matchdata
+    // else ret will be count number of xattrs the key pattern-matches with.
+    ret = dict_foreach_match(matchdata, _is_xattr_in_watchlist, key,
+                             dict_null_foreach_fn, NULL);
+
+    if (ret > 0)
+        return _gf_true;
+    return _gf_false;
+}
+
+static int
+index_find_xattr_type(dict_t *d, char *k, data_t *v)
+{
+    int idx = -1;
+    index_priv_t *priv = THIS->private;
+
+    if (priv->dirty_watchlist &&
+        is_xattr_in_watchlist(d, k, v, priv->dirty_watchlist))
+        idx = DIRTY;
+    else if (priv->pending_watchlist &&
+             is_xattr_in_watchlist(d, k, v, priv->pending_watchlist))
+        idx = XATTROP;
+
+    return idx;
+}
+
+int
+index_fill_zero_array(dict_t *d, char *k, data_t *v, void *adata)
+{
+    int idx = -1;
+    int *zfilled = adata;
+    // zfilled array contains `state` for all types xattrs.
+    // state : whether the gfid file of this file exists in
+    // corresponding xattr directory or not.
+
+    idx = index_find_xattr_type(d, k, v);
+    if (idx == -1)
+        return 0;
+    zfilled[idx] = 0;
+    return 0;
+}
+
+static int
+_check_key_is_zero_filled(dict_t *d, char *k, data_t *v, void *tmp)
+{
+    int *zfilled = tmp;
+    int idx = -1;
+
+    idx = index_find_xattr_type(d, k, v);
+    if (idx == -1)
+        return 0;
+
+    /* Along with checking that the value of a key is zero filled
+     * the key's corresponding index should be assigned
+     * appropriate value.
+     * zfilled[idx] will be 0(false) if value not zero.
+     *              will be 1(true) if value is zero.
+     */
+    if (mem_0filled((const char *)v->data, v->len)) {
+        zfilled[idx] = 0;
+        return 0;
+    }
+
+    /* If zfilled[idx] was previously 0, it means at least
+     * one xattr of its "kind" is non-zero. Keep its value
+     * the same.
+     */
+    if (zfilled[idx])
+        zfilled[idx] = 1;
+    return 0;
+}
+
+int
+index_entry_create(xlator_t *this, inode_t *inode, char *filename)
+{
+    int ret = -1;
+    int op_errno = 0;
+    char pgfid_path[PATH_MAX] = {0};
+    char entry_path[PATH_MAX] = {0};
+    index_priv_t *priv = NULL;
+    index_inode_ctx_t *ctx = NULL;
+    int32_t len = 0;
+
+    priv = this->private;
+
+    GF_ASSERT_AND_GOTO_WITH_ERROR(this->name, !gf_uuid_is_null(inode->gfid),
+                                  out, op_errno, EINVAL);
+    GF_ASSERT_AND_GOTO_WITH_ERROR(this->name, filename, out, op_errno, EINVAL);
+
+    ret = index_inode_ctx_get(inode, this, &ctx);
+    if (ret) {
+        op_errno = EINVAL;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno,
+               INDEX_MSG_INODE_CTX_GET_SET_FAILED,
+               "Not able to get inode ctx for %s", uuid_utoa(inode->gfid));
+        goto out;
+    }
+
+    make_gfid_path(priv->index_basepath, ENTRY_CHANGES_SUBDIR, inode->gfid,
+                   pgfid_path, sizeof(pgfid_path));
+
+    if (ctx->state[ENTRY_CHANGES] != IN) {
+        ret = sys_mkdir(pgfid_path, 0600);
+        if (ret != 0 && errno != EEXIST) {
+            op_errno = errno;
+            goto out;
+        }
+        ctx->state[ENTRY_CHANGES] = IN;
+    }
+
+    if (strchr(filename, '/')) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, INDEX_MSG_INDEX_ADD_FAILED,
+               "Got invalid entry (%s) for pargfid path (%s)", filename,
+               pgfid_path);
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    len = snprintf(entry_path, sizeof(entry_path), "%s/%s", pgfid_path,
+                   filename);
+    if ((len < 0) || (len >= sizeof(entry_path))) {
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    op_errno = 0;
+
+    ret = index_link_to_base(this, entry_path, ENTRY_CHANGES_SUBDIR);
+out:
+    if (op_errno)
+        ret = -op_errno;
+    return ret;
+}
+
+int
+index_entry_delete(xlator_t *this, uuid_t pgfid, char *filename)
+{
+    int ret = 0;
+    int op_errno = 0;
+    char pgfid_path[PATH_MAX] = {0};
+    char entry_path[PATH_MAX] = {0};
+    index_priv_t *priv = NULL;
+    int32_t len = 0;
+
+    priv = this->private;
+
+    GF_ASSERT_AND_GOTO_WITH_ERROR(this->name, !gf_uuid_is_null(pgfid), out,
+                                  op_errno, EINVAL);
+    GF_ASSERT_AND_GOTO_WITH_ERROR(this->name, filename, out, op_errno, EINVAL);
+
+    make_gfid_path(priv->index_basepath, ENTRY_CHANGES_SUBDIR, pgfid,
+                   pgfid_path, sizeof(pgfid_path));
+
+    if (strchr(filename, '/')) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, INDEX_MSG_INDEX_DEL_FAILED,
+               "Got invalid entry (%s) for pargfid path (%s)", filename,
+               pgfid_path);
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    len = snprintf(entry_path, sizeof(entry_path), "%s/%s", pgfid_path,
+                   filename);
+    if ((len < 0) || (len >= sizeof(entry_path))) {
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    ret = sys_unlink(entry_path);
+    if (ret && (errno != ENOENT)) {
+        op_errno = errno;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, INDEX_MSG_INDEX_DEL_FAILED,
+               "%s: failed to delete from index/entry-changes", entry_path);
+    }
+
+out:
+    return -op_errno;
+}
+
+int
+index_entry_action(xlator_t *this, inode_t *inode, dict_t *xdata, char *key)
+{
+    int ret = 0;
+    char *filename = NULL;
+
+    ret = dict_get_str(xdata, key, &filename);
+    if (ret != 0) {
+        ret = 0;
+        goto out;
+    }
+
+    if (strcmp(key, GF_XATTROP_ENTRY_IN_KEY) == 0)
+        ret = index_entry_create(this, inode, filename);
+    else if (strcmp(key, GF_XATTROP_ENTRY_OUT_KEY) == 0)
+        ret = index_entry_delete(this, inode->gfid, filename);
+
+out:
+    return ret;
+}
+
+void
+_index_action(xlator_t *this, inode_t *inode, int *zfilled)
+{
+    int ret = 0;
+    int i = 0;
+    index_inode_ctx_t *ctx = NULL;
+    char *subdir = NULL;
+
+    ret = index_inode_ctx_get(inode, this, &ctx);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL,
+               INDEX_MSG_INODE_CTX_GET_SET_FAILED,
+               "Not able to get"
+               " inode context for %s.",
+               uuid_utoa(inode->gfid));
+        goto out;
+    }
+
+    for (i = 0; i < XATTROP_TYPE_END; i++) {
+        subdir = index_get_subdir_from_type(i);
+        if (zfilled[i] == 1) {
+            if (ctx->state[i] == NOTIN)
+                continue;
+            ret = index_del(this, inode->gfid, subdir, i);
+            if (!ret)
+                ctx->state[i] = NOTIN;
+        } else if (zfilled[i] == 0) {
+            if (ctx->state[i] == IN)
+                continue;
+            ret = index_add(this, inode->gfid, subdir, i);
+            if (!ret)
+                ctx->state[i] = IN;
+        }
+    }
+out:
+    return;
+}
+
+static void
+index_init_state(xlator_t *this, inode_t *inode, index_inode_ctx_t *ctx,
+                 char *subdir)
+{
+    int ret = -1;
+    char pgfid_path[PATH_MAX] = {0};
+    struct stat st = {0};
+    index_priv_t *priv = NULL;
+
+    priv = this->private;
+
+    make_gfid_path(priv->index_basepath, subdir, inode->gfid, pgfid_path,
+                   sizeof(pgfid_path));
+
+    ret = sys_stat(pgfid_path, &st);
+    if (ret == 0)
+        ctx->state[ENTRY_CHANGES] = IN;
+    else if (ret != 0 && errno == ENOENT)
+        ctx->state[ENTRY_CHANGES] = NOTIN;
+
+    return;
+}
+
+void
+xattrop_index_action(xlator_t *this, index_local_t *local, dict_t *xattr,
+                     dict_match_t match, void *match_data)
+{
+    int ret = 0;
+    int zfilled[XATTROP_TYPE_END] = {
+        0,
+    };
+    int8_t value = 0;
+    char *subdir = NULL;
+    dict_t *req_xdata = NULL;
+    inode_t *inode = NULL;
+    index_inode_ctx_t *ctx = NULL;
+
+    inode = local->inode;
+    req_xdata = local->xdata;
+
+    memset(zfilled, -1, sizeof(zfilled));
+    ret = dict_foreach_match(xattr, match, match_data,
+                             _check_key_is_zero_filled, zfilled);
+    _index_action(this, inode, zfilled);
+
+    if (req_xdata) {
+        ret = index_entry_action(this, inode, req_xdata,
+                                 GF_XATTROP_ENTRY_OUT_KEY);
+
+        ret = dict_get_int8(req_xdata, GF_XATTROP_PURGE_INDEX, &value);
+        if ((ret) || (value == 0))
+            goto out;
+    }
+
+    if (zfilled[XATTROP] != 1)
+        goto out;
+
+    if (inode->ia_type != IA_IFDIR)
+        goto out;
+
+    subdir = index_get_subdir_from_type(ENTRY_CHANGES);
+    ret = index_inode_ctx_get(inode, this, &ctx);
+    if (ctx->state[ENTRY_CHANGES] == UNKNOWN)
+        index_init_state(this, inode, ctx, subdir);
+    if (ctx->state[ENTRY_CHANGES] == IN) {
+        ret = index_del(this, inode->gfid, subdir, ENTRY_CHANGES);
+        ctx->state[ENTRY_CHANGES] = NOTIN;
+    }
+
+out:
+    return;
+}
+
+static gf_boolean_t
+index_xattrop_track(xlator_t *this, gf_xattrop_flags_t flags, dict_t *dict)
+{
+    index_priv_t *priv = this->private;
+
+    if (flags == GF_XATTROP_ADD_ARRAY)
+        return _gf_true;
+
+    if (flags != GF_XATTROP_ADD_ARRAY64)
+        return _gf_false;
+
+    if (!priv->pending_watchlist)
+        return _gf_false;
+
+    if (dict_foreach_match(dict, is_xattr_in_watchlist, priv->pending_watchlist,
+                           dict_null_foreach_fn, NULL) > 0)
+        return _gf_true;
+
+    return _gf_false;
+}
+
+int
+index_inode_path(xlator_t *this, inode_t *inode, char *dirpath, size_t len)
+{
+    char *subdir = NULL;
+    int ret = 0;
+    index_priv_t *priv = NULL;
+    index_inode_ctx_t *ictx = NULL;
+
+    priv = this->private;
+    if (!index_is_fop_on_internal_inode(this, inode, NULL)) {
+        ret = -EINVAL;
+        goto out;
+    }
+
+    subdir = index_get_subdir_from_vgfid(priv, inode->gfid);
+    if (subdir) {
+        if (len <= strlen(priv->index_basepath) + 1 /*'/'*/ + strlen(subdir)) {
+            ret = -EINVAL;
+            goto out;
+        }
+        make_index_dir_path(priv->index_basepath, subdir, dirpath, len);
+    } else {
+        ret = index_inode_ctx_get(inode, this, &ictx);
+        if (ret)
+            goto out;
+        if (gf_uuid_is_null(ictx->virtual_pargfid)) {
+            ret = -EINVAL;
+            goto out;
+        }
+        make_index_dir_path(priv->index_basepath, ENTRY_CHANGES_SUBDIR, dirpath,
+                            len);
+        if (len <= strlen(dirpath) + 1 /*'/'*/ + SLEN(UUID0_STR)) {
+            ret = -EINVAL;
+            goto out;
+        }
+        strcat(dirpath, "/");
+        strcat(dirpath, uuid_utoa(ictx->virtual_pargfid));
+    }
+out:
+    return ret;
+}
+
+int
+__index_fd_ctx_get(fd_t *fd, xlator_t *this, index_fd_ctx_t **ctx)
+{
+    int ret = 0;
+    index_fd_ctx_t *fctx = NULL;
+    uint64_t tmpctx = 0;
+    char dirpath[PATH_MAX] = {0};
+
+    ret = __fd_ctx_get(fd, this, &tmpctx);
+    if (!ret) {
+        fctx = (index_fd_ctx_t *)(long)tmpctx;
+        *ctx = fctx;
+        goto out;
+    }
+
+    ret = index_inode_path(this, fd->inode, dirpath, sizeof(dirpath));
+    if (ret)
+        goto out;
+
+    fctx = GF_CALLOC(1, sizeof(*fctx), gf_index_fd_ctx_t);
+    if (!fctx) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    fctx->dir = sys_opendir(dirpath);
+    if (!fctx->dir) {
+        ret = -errno;
+        GF_FREE(fctx);
+        fctx = NULL;
+        goto out;
+    }
+    fctx->dir_eof = -1;
+
+    ret = __fd_ctx_set(fd, this, (uint64_t)(long)fctx);
+    if (ret) {
+        (void)sys_closedir(fctx->dir);
+        GF_FREE(fctx);
+        fctx = NULL;
+        ret = -EINVAL;
+        goto out;
+    }
+    *ctx = fctx;
+out:
+    return ret;
+}
+
+int
+index_fd_ctx_get(fd_t *fd, xlator_t *this, index_fd_ctx_t **ctx)
+{
+    int ret = 0;
+    LOCK(&fd->lock);
+    {
+        ret = __index_fd_ctx_get(fd, this, ctx);
+    }
+    UNLOCK(&fd->lock);
+    return ret;
+}
+
+// new - Not NULL means start a fop
+// new - NULL means done processing the fop
+void
+index_queue_process(xlator_t *this, inode_t *inode, call_stub_t *new)
+{
+    call_stub_t *stub = NULL;
+    index_inode_ctx_t *ctx = NULL;
+    int ret = 0;
+    call_frame_t *frame = NULL;
+
+    LOCK(&inode->lock);
+    {
+        ret = __index_inode_ctx_get(inode, this, &ctx);
+        if (ret)
+            goto unlock;
+
+        if (new) {
+            __index_enqueue(&ctx->callstubs, new);
+            new = NULL;
+        } else {
+            ctx->processing = _gf_false;
+        }
+
+        if (!ctx->processing) {
+            stub = __index_dequeue(&ctx->callstubs);
+            if (stub)
+                ctx->processing = _gf_true;
+            else
+                ctx->processing = _gf_false;
+        }
+    }
+unlock:
+    UNLOCK(&inode->lock);
+
+    if (ret && new) {
+        frame = new->frame;
+        if (new->fop == GF_FOP_XATTROP) {
+            INDEX_STACK_UNWIND(xattrop, frame, -1, ENOMEM, NULL, NULL);
+        } else if (new->fop == GF_FOP_FXATTROP) {
+            INDEX_STACK_UNWIND(fxattrop, frame, -1, ENOMEM, NULL, NULL);
+        }
+        call_stub_destroy(new);
+    } else if (stub) {
+        call_resume(stub);
+    }
+    return;
+}
+
+static int
+xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, dict_t *xattr, dict_t *xdata, dict_match_t match,
+            dict_t *matchdata)
+{
+    inode_t *inode = NULL;
+    index_local_t *local = NULL;
+
+    local = frame->local;
+    inode = inode_ref(local->inode);
+
+    if (op_ret < 0)
+        goto out;
+
+    xattrop_index_action(this, local, xattr, match, matchdata);
+out:
+    INDEX_STACK_UNWIND(xattrop, frame, op_ret, op_errno, xattr, xdata);
+    index_queue_process(this, inode, NULL);
+    inode_unref(inode);
+
+    return 0;
+}
+
+int32_t
+index_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, dict_t *xattr,
+                  dict_t *xdata)
+{
+    index_priv_t *priv = this->private;
+
+    xattrop_cbk(frame, cookie, this, op_ret, op_errno, xattr, xdata,
+                is_xattr_in_watchlist, priv->complete_watchlist);
+    return 0;
+}
+
+int32_t
+index_xattrop64_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *xattr,
+                    dict_t *xdata)
+{
+    index_priv_t *priv = this->private;
+
+    return xattrop_cbk(frame, cookie, this, op_ret, op_errno, xattr, xdata,
+                       is_xattr_in_watchlist, priv->pending_watchlist);
+}
+
+void
+index_xattrop_do(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+                 gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+    int ret = -1;
+    int zfilled[XATTROP_TYPE_END] = {
+        0,
+    };
+    index_local_t *local = NULL;
+    fop_xattrop_cbk_t x_cbk = NULL;
+
+    local = frame->local;
+
+    if (optype == GF_XATTROP_ADD_ARRAY)
+        x_cbk = index_xattrop_cbk;
+    else
+        x_cbk = index_xattrop64_cbk;
+
+    // In wind phase bring the gfid into index. This way if the brick crashes
+    // just after posix performs xattrop before _cbk reaches index xlator
+    // we will still have the gfid in index.
+    memset(zfilled, -1, sizeof(zfilled));
+
+    /* Foreach xattr, set corresponding index of zfilled to 1
+     * zfilled[index] = 1 implies the xattr's value is zero filled
+     * and should be added in its corresponding subdir.
+     *
+     * zfilled should be set to 1 only for those index that
+     * exist in xattr variable. This is to distinguish
+     * between different types of volumes.
+     * For e.g., if the check is not made,
+     * zfilled[DIRTY] is set to 1 for EC volumes,
+     * index file will be tried to create in indices/dirty dir
+     * which doesn't exist for an EC volume.
+     */
+    ret = dict_foreach(xattr, index_fill_zero_array, zfilled);
+
+    _index_action(this, local->inode, zfilled);
+    if (xdata)
+        ret = index_entry_action(this, local->inode, xdata,
+                                 GF_XATTROP_ENTRY_IN_KEY);
+    if (ret < 0) {
+        x_cbk(frame, NULL, this, -1, -ret, NULL, NULL);
+        return;
+    }
+
+    if (loc)
+        STACK_WIND(frame, x_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->xattrop, loc, optype, xattr, xdata);
+    else
+        STACK_WIND(frame, x_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fxattrop, fd, optype, xattr, xdata);
+}
+
+int
+index_xattrop_wrapper(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                      gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+    index_xattrop_do(frame, this, loc, NULL, optype, xattr, xdata);
+    return 0;
+}
+
+int
+index_fxattrop_wrapper(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                       gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+    index_xattrop_do(frame, this, NULL, fd, optype, xattr, xdata);
+    return 0;
+}
+
+int32_t
+index_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+              gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    index_local_t *local = NULL;
+
+    if (!index_xattrop_track(this, flags, dict))
+        goto out;
+
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto err;
+
+    frame->local = local;
+    local->inode = inode_ref(loc->inode);
+    if (xdata)
+        local->xdata = dict_ref(xdata);
+    stub = fop_xattrop_stub(frame, index_xattrop_wrapper, loc, flags, dict,
+                            xdata);
+
+err:
+    if ((!local) || (!stub)) {
+        INDEX_STACK_UNWIND(xattrop, frame, -1, ENOMEM, NULL, NULL);
+        return 0;
+    }
+
+    index_queue_process(this, loc->inode, stub);
+    return 0;
+out:
+    STACK_WIND(frame, default_xattrop_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->xattrop, loc, flags, dict, xdata);
+    return 0;
+}
+
+int32_t
+index_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+               gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    index_local_t *local = NULL;
+
+    if (!index_xattrop_track(this, flags, dict))
+        goto out;
+
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto err;
+
+    frame->local = local;
+    local->inode = inode_ref(fd->inode);
+    if (xdata)
+        local->xdata = dict_ref(xdata);
+    stub = fop_fxattrop_stub(frame, index_fxattrop_wrapper, fd, flags, dict,
+                             xdata);
+
+err:
+    if ((!local) || (!stub)) {
+        INDEX_STACK_UNWIND(fxattrop, frame, -1, ENOMEM, NULL, xdata);
+        return 0;
+    }
+
+    index_queue_process(this, fd->inode, stub);
+    return 0;
+out:
+    STACK_WIND(frame, default_fxattrop_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fxattrop, fd, flags, dict, xdata);
+    return 0;
+}
+
+uint64_t
+index_entry_count(xlator_t *this, char *subdir)
+{
+    uint64_t count = 0;
+    index_priv_t *priv = NULL;
+    DIR *dirp = NULL;
+    struct dirent *entry = NULL;
+    struct dirent scratch[2] = {
+        {
+            0,
+        },
+    };
+    char index_dir[PATH_MAX] = {
+        0,
+    };
+
+    priv = this->private;
+
+    make_index_dir_path(priv->index_basepath, subdir, index_dir,
+                        sizeof(index_dir));
+
+    dirp = sys_opendir(index_dir);
+    if (!dirp)
+        return 0;
+
+    for (;;) {
+        errno = 0;
+        entry = sys_readdir(dirp, scratch);
+        if (!entry || errno != 0)
+            break;
+
+        if (strcmp(entry->d_name, ".") == 0 || strcmp(entry->d_name, "..") == 0)
+            continue;
+
+        if (!strncmp(entry->d_name, subdir, strlen(subdir)))
+            continue;
+
+        count++;
+    }
+
+    (void)sys_closedir(dirp);
+
+    return count;
+}
+
+int32_t
+index_getxattr_wrapper(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                       const char *name, dict_t *xdata)
+{
+    index_priv_t *priv = NULL;
+    dict_t *xattr = NULL;
+    int ret = 0;
+    int vgfid_type = 0;
+    uint64_t count = 0;
+
+    priv = this->private;
+
+    xattr = dict_new();
+    if (!xattr) {
+        ret = -ENOMEM;
+        goto done;
+    }
+
+    vgfid_type = index_get_type_from_vgfid_xattr(name);
+    if (vgfid_type >= 0) {
+        ret = dict_set_static_bin(xattr, (char *)name,
+                                  priv->internal_vgfid[vgfid_type],
+                                  sizeof(priv->internal_vgfid[vgfid_type]));
+        if (ret) {
+            ret = -EINVAL;
+            gf_msg(this->name, GF_LOG_ERROR, -ret, INDEX_MSG_DICT_SET_FAILED,
+                   "xattrop index "
+                   "gfid set failed");
+            goto done;
+        }
+    }
+
+    /* TODO: Need to check what kind of link-counts are needed for
+     * ENTRY-CHANGES before refactor of this block with array*/
+    if (strcmp(name, GF_XATTROP_INDEX_COUNT) == 0) {
+        count = index_entry_count(this, XATTROP_SUBDIR);
+
+        ret = dict_set_uint64(xattr, (char *)name, count);
+        if (ret) {
+            ret = -EINVAL;
+            gf_msg(this->name, GF_LOG_ERROR, -ret, INDEX_MSG_DICT_SET_FAILED,
+                   "xattrop index "
+                   "count set failed");
+            goto done;
+        }
+    } else if (strcmp(name, GF_XATTROP_DIRTY_COUNT) == 0) {
+        count = index_entry_count(this, DIRTY_SUBDIR);
+
+        ret = dict_set_uint64(xattr, (char *)name, count);
+        if (ret) {
+            ret = -EINVAL;
+            gf_msg(this->name, GF_LOG_ERROR, -ret, INDEX_MSG_DICT_SET_FAILED,
+                   "dirty index "
+                   "count set failed");
+            goto done;
+        }
+    }
+done:
+    if (ret)
+        STACK_UNWIND_STRICT(getxattr, frame, -1, -ret, xattr, NULL);
+    else
+        STACK_UNWIND_STRICT(getxattr, frame, 0, 0, xattr, NULL);
+
+    if (xattr)
+        dict_unref(xattr);
+
+    return 0;
+}
+
+static int
+index_save_pargfid_for_entry_changes(xlator_t *this, loc_t *loc, char *path)
+{
+    index_priv_t *priv = NULL;
+    index_inode_ctx_t *ctx = NULL;
+    int ret = 0;
+
+    priv = this->private;
+    if (!loc)
+        return -1;
+    if (gf_uuid_compare(loc->pargfid, priv->internal_vgfid[ENTRY_CHANGES]))
+        return 0;
+
+    ret = index_inode_ctx_get(loc->inode, this, &ctx);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL,
+               INDEX_MSG_INODE_CTX_GET_SET_FAILED,
+               "Unable to get inode context for %s", path);
+        return -EINVAL;
+    }
+    ret = gf_uuid_parse(loc->name, ctx->virtual_pargfid);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL,
+               INDEX_MSG_INODE_CTX_GET_SET_FAILED,
+               "Unable to store "
+               "virtual gfid in inode context for %s",
+               path);
+        return -EINVAL;
+    }
+    return 0;
+}
+
+int32_t
+index_lookup_wrapper(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                     dict_t *xattr_req)
+{
+    index_priv_t *priv = NULL;
+    struct stat lstatbuf = {0};
+    int ret = 0;
+    int32_t op_errno = EINVAL;
+    int32_t op_ret = -1;
+    uint64_t val = IA_INVAL;
+    char path[PATH_MAX] = {0};
+    struct iatt stbuf = {
+        0,
+    };
+    struct iatt postparent = {
+        0,
+    };
+    dict_t *xattr = NULL;
+    gf_boolean_t is_dir = _gf_false;
+    char *subdir = NULL;
+    loc_t iloc = {0};
+
+    priv = this->private;
+    loc_copy(&iloc, loc);
+
+    VALIDATE_OR_GOTO(loc, done);
+    if (index_is_fop_on_internal_inode(this, loc->parent, loc->pargfid)) {
+        subdir = index_get_subdir_from_vgfid(priv, loc->pargfid);
+        ret = index_inode_path(this, loc->parent, path, sizeof(path));
+        if (ret < 0) {
+            op_errno = -ret;
+            goto done;
+        }
+        ret = snprintf(path + strlen(path), PATH_MAX - strlen(path), "/%s",
+                       loc->name);
+
+        if ((ret < 0) || (ret > (PATH_MAX - strlen(path)))) {
+            op_errno = EINVAL;
+            op_ret = -1;
+            goto done;
+        }
+
+    } else if (index_is_virtual_gfid(priv, loc->gfid)) {
+        subdir = index_get_subdir_from_vgfid(priv, loc->gfid);
+        make_index_dir_path(priv->index_basepath, subdir, path, sizeof(path));
+        is_dir = _gf_true;
+
+        if ((xattr_req) && (dict_get(xattr_req, GF_INDEX_IA_TYPE_GET_REQ))) {
+            if (0 == strcmp(subdir, index_get_subdir_from_type(ENTRY_CHANGES)))
+                val = IA_IFDIR;
+            else
+                val = IA_IFREG;
+        }
+    } else {
+        if (!inode_is_linked(loc->inode)) {
+            inode_unref(iloc.inode);
+            iloc.inode = inode_find(loc->inode->table, loc->gfid);
+        }
+        ret = index_inode_path(this, iloc.inode, path, sizeof(path));
+        if (ret < 0) {
+            op_errno = -ret;
+            goto done;
+        }
+    }
+    ret = sys_lstat(path, &lstatbuf);
+    if (ret) {
+        gf_msg_debug(this->name, errno, "Stat failed on %s dir ", path);
+        op_errno = errno;
+        goto done;
+    } else if (!S_ISDIR(lstatbuf.st_mode) && is_dir) {
+        op_errno = ENOTDIR;
+        gf_msg_debug(this->name, op_errno,
+                     "Stat failed on %s dir, "
+                     "not a directory",
+                     path);
+        goto done;
+    }
+    xattr = dict_new();
+    if (!xattr) {
+        op_errno = ENOMEM;
+        goto done;
+    }
+
+    if (val != IA_INVAL) {
+        ret = dict_set_uint64(xattr, GF_INDEX_IA_TYPE_GET_RSP, val);
+        if (ret) {
+            op_ret = -1;
+            op_errno = -ret;
+            goto done;
+        }
+    }
+
+    iatt_from_stat(&stbuf, &lstatbuf);
+    if (is_dir || inode_is_linked(iloc.inode))
+        loc_gfid(&iloc, stbuf.ia_gfid);
+    else
+        gf_uuid_generate(stbuf.ia_gfid);
+
+    ret = index_save_pargfid_for_entry_changes(this, &iloc, path);
+    if (ret) {
+        op_ret = -1;
+        op_errno = -ret;
+        goto done;
+    }
+
+    stbuf.ia_ino = -1;
+    op_ret = 0;
+done:
+    STACK_UNWIND_STRICT(lookup, frame, op_ret, op_errno,
+                        loc ? loc->inode : NULL, &stbuf, xattr, &postparent);
+    if (xattr)
+        dict_unref(xattr);
+    loc_wipe(&iloc);
+    return 0;
+}
+
+int
+index_get_gfid_type(void *opaque)
+{
+    gf_dirent_t *entry = NULL;
+    xlator_t *this = THIS;
+    struct index_syncop_args *args = opaque;
+    loc_t loc = {0};
+    struct iatt iatt = {0};
+    int ret = 0;
+
+    list_for_each_entry(entry, &args->entries->list, list)
+    {
+        if (strcmp(entry->d_name, ".") == 0 || strcmp(entry->d_name, "..") == 0)
+            continue;
+
+        loc_wipe(&loc);
+
+        entry->d_type = gf_d_type_from_ia_type(IA_INVAL);
+        entry->d_stat.ia_type = IA_INVAL;
+        if (gf_uuid_parse(entry->d_name, loc.gfid))
+            continue;
+
+        loc.inode = inode_find(args->parent->table, loc.gfid);
+        if (loc.inode) {
+            entry->d_stat.ia_type = loc.inode->ia_type;
+            entry->d_type = gf_d_type_from_ia_type(loc.inode->ia_type);
+            continue;
+        }
+        loc.inode = inode_new(args->parent->table);
+        if (!loc.inode)
+            continue;
+        ret = syncop_lookup(FIRST_CHILD(this), &loc, &iatt, 0, 0, 0);
+        if (ret == 0) {
+            entry->d_type = gf_d_type_from_ia_type(iatt.ia_type);
+            entry->d_stat = iatt;
+        }
+    }
+    loc_wipe(&loc);
+
+    return 0;
+}
+
+int32_t
+index_readdir_wrapper(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                      size_t size, off_t off, dict_t *xdata)
+{
+    index_fd_ctx_t *fctx = NULL;
+    index_priv_t *priv = NULL;
+    DIR *dir = NULL;
+    int ret = -1;
+    int32_t op_ret = -1;
+    int32_t op_errno = 0;
+    int count = 0;
+    gf_dirent_t entries;
+    struct index_syncop_args args = {0};
+
+    priv = this->private;
+    INIT_LIST_HEAD(&entries.list);
+
+    ret = index_fd_ctx_get(fd, this, &fctx);
+    if (ret < 0) {
+        op_errno = -ret;
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, INDEX_MSG_FD_OP_FAILED,
+               "pfd is NULL, fd=%p", fd);
+        goto done;
+    }
+
+    dir = fctx->dir;
+    if (!dir) {
+        op_errno = EINVAL;
+        gf_msg(this->name, GF_LOG_WARNING, op_errno,
+               INDEX_MSG_INDEX_READDIR_FAILED, "dir is NULL for fd=%p", fd);
+        goto done;
+    }
+
+    count = index_fill_readdir(fd, fctx, dir, off, size, &entries);
+
+    /* pick ENOENT to indicate EOF */
+    op_errno = errno;
+    op_ret = count;
+    if (index_is_virtual_gfid(priv, fd->inode->gfid) && xdata &&
+        dict_get(xdata, "get-gfid-type")) {
+        args.parent = fd->inode;
+        args.entries = &entries;
+        ret = synctask_new(this->ctx->env, index_get_gfid_type, NULL, NULL,
+                           &args);
+    }
+done:
+    STACK_UNWIND_STRICT(readdir, frame, op_ret, op_errno, &entries, NULL);
+    gf_dirent_free(&entries);
+    return 0;
+}
+
+int
+deletion_handler(const char *fpath, const struct stat *sb, int typeflag,
+                 struct FTW *ftwbuf)
+{
+    ia_type_t type = IA_INVAL;
+
+    switch (sb->st_mode & S_IFMT) {
+        case S_IFREG:
+            sys_unlink(fpath);
+            break;
+
+        case S_IFDIR:
+            sys_rmdir(fpath);
+            break;
+        default:
+            type = ia_type_from_st_mode(sb->st_mode);
+            gf_msg(THIS->name, GF_LOG_WARNING, EINVAL, INDEX_MSG_INVALID_ARGS,
+                   "%s neither a regular file nor a directory - type:%s", fpath,
+                   gf_inode_type_to_str(type));
+            break;
+    }
+    return 0;
+}
+
+static int
+index_wipe_index_subdir(void *opaque)
+{
+    struct index_syncop_args *args = opaque;
+
+    nftw(args->path, deletion_handler, 1, FTW_DEPTH | FTW_PHYS);
+    return 0;
+}
+
+static void
+index_get_parent_iatt(struct iatt *parent, char *path, loc_t *loc,
+                      int32_t *op_ret, int32_t *op_errno)
+{
+    int ret = -1;
+    struct stat lstatbuf = {
+        0,
+    };
+
+    ret = sys_lstat(path, &lstatbuf);
+    if (ret < 0) {
+        *op_ret = -1;
+        *op_errno = errno;
+        return;
+    }
+
+    iatt_from_stat(parent, &lstatbuf);
+    gf_uuid_copy(parent->ia_gfid, loc->pargfid);
+    parent->ia_ino = -1;
+
+    return;
+}
+
+int
+index_rmdir_wrapper(call_frame_t *frame, xlator_t *this, loc_t *loc, int flag,
+                    dict_t *xdata)
+{
+    int ret = 0;
+    int32_t op_ret = 0;
+    int32_t op_errno = 0;
+    char *subdir = NULL;
+    char index_dir[PATH_MAX] = {0};
+    char index_subdir[PATH_MAX] = {0};
+    uuid_t gfid = {0};
+    struct iatt preparent = {0};
+    struct iatt postparent = {0};
+    index_priv_t *priv = NULL;
+    index_xattrop_type_t type = XATTROP_TYPE_UNSET;
+    struct index_syncop_args args = {
+        0,
+    };
+
+    priv = this->private;
+
+    type = index_get_type_from_vgfid(priv, loc->pargfid);
+    subdir = index_get_subdir_from_vgfid(priv, loc->pargfid);
+    make_index_dir_path(priv->index_basepath, subdir, index_dir,
+                        sizeof(index_dir));
+
+    index_get_parent_iatt(&preparent, index_dir, loc, &op_ret, &op_errno);
+    if (op_ret < 0)
+        goto done;
+
+    gf_uuid_parse(loc->name, gfid);
+    make_gfid_path(priv->index_basepath, subdir, gfid, index_subdir,
+                   sizeof(index_subdir));
+
+    if (flag == 0) {
+        ret = index_del(this, gfid, subdir, type);
+        if (ret < 0) {
+            op_ret = -1;
+            op_errno = -ret;
+            goto done;
+        }
+    } else {
+        args.path = index_subdir;
+        ret = synctask_new(this->ctx->env, index_wipe_index_subdir, NULL, NULL,
+                           &args);
+    }
+
+    index_get_parent_iatt(&postparent, index_dir, loc, &op_ret, &op_errno);
+    if (op_ret < 0)
+        goto done;
+
+done:
+    INDEX_STACK_UNWIND(rmdir, frame, op_ret, op_errno, &preparent, &postparent,
+                       xdata);
+    return 0;
+}
+
+int
+index_unlink_wrapper(call_frame_t *frame, xlator_t *this, loc_t *loc, int flag,
+                     dict_t *xdata)
+{
+    index_priv_t *priv = NULL;
+    index_inode_ctx_t *ictx = NULL;
+    int32_t op_ret = 0;
+    int32_t op_errno = 0;
+    int ret = 0;
+    index_xattrop_type_t type = XATTROP_TYPE_UNSET;
+    struct iatt preparent = {0};
+    struct iatt postparent = {0};
+    char index_dir[PATH_MAX] = {0};
+    char filepath[PATH_MAX] = {0};
+    uuid_t gfid = {0};
+    char *subdir = NULL;
+
+    priv = this->private;
+    type = index_get_type_from_vgfid(priv, loc->pargfid);
+    ret = index_inode_path(this, loc->parent, index_dir, sizeof(index_dir));
+    if (ret < 0) {
+        op_ret = -1;
+        op_errno = -ret;
+        goto done;
+    }
+
+    index_get_parent_iatt(&preparent, index_dir, loc, &op_ret, &op_errno);
+    if (op_ret < 0)
+        goto done;
+
+    if (type <= XATTROP_TYPE_UNSET) {
+        ret = index_inode_ctx_get(loc->parent, this, &ictx);
+        if ((ret == 0) && gf_uuid_is_null(ictx->virtual_pargfid)) {
+            ret = -EINVAL;
+        }
+        if (ret == 0) {
+            ret = index_entry_delete(this, ictx->virtual_pargfid,
+                                     (char *)loc->name);
+        }
+    } else if (type == ENTRY_CHANGES) {
+        make_file_path(priv->index_basepath, ENTRY_CHANGES_SUBDIR,
+                       (char *)loc->name, filepath, sizeof(filepath));
+        ret = sys_unlink(filepath);
+    } else {
+        subdir = index_get_subdir_from_type(type);
+        gf_uuid_parse(loc->name, gfid);
+        ret = index_del(this, gfid, subdir, type);
+    }
+    if (ret < 0) {
+        op_ret = -1;
+        op_errno = -ret;
+        goto done;
+    }
+
+    index_get_parent_iatt(&postparent, index_dir, loc, &op_ret, &op_errno);
+    if (op_ret < 0)
+        goto done;
+done:
+    INDEX_STACK_UNWIND(unlink, frame, op_ret, op_errno, &preparent, &postparent,
+                       xdata);
+    return 0;
+}
+
+int32_t
+index_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               const char *name, dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+
+    if (!name ||
+        (!index_is_vgfid_xattr(name) && strcmp(GF_XATTROP_INDEX_COUNT, name) &&
+         strcmp(GF_XATTROP_DIRTY_COUNT, name)))
+        goto out;
+
+    stub = fop_getxattr_stub(frame, index_getxattr_wrapper, loc, name, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(getxattr, frame, -1, ENOMEM, NULL, NULL);
+        return 0;
+    }
+    worker_enqueue(this, stub);
+    return 0;
+out:
+    STACK_WIND(frame, default_getxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->getxattr, loc, name, xdata);
+    return 0;
+}
+
+int64_t
+index_fetch_link_count(xlator_t *this, index_xattrop_type_t type)
+{
+    index_priv_t *priv = this->private;
+    char *subdir = NULL;
+    struct stat lstatbuf = {
+        0,
+    };
+    int ret = -1;
+    int64_t count = -1;
+    DIR *dirp = NULL;
+    struct dirent *entry = NULL;
+    struct dirent scratch[2] = {
+        {
+            0,
+        },
+    };
+    char index_dir[PATH_MAX] = {
+        0,
+    };
+    char index_path[PATH_MAX] = {
+        0,
+    };
+
+    subdir = index_get_subdir_from_type(type);
+    make_index_dir_path(priv->index_basepath, subdir, index_dir,
+                        sizeof(index_dir));
+
+    dirp = sys_opendir(index_dir);
+    if (!dirp)
+        goto out;
+
+    for (;;) {
+        errno = 0;
+        entry = sys_readdir(dirp, scratch);
+        if (!entry || errno != 0) {
+            if (count == -1)
+                count = 0;
+            goto out;
+        }
+
+        if (strcmp(entry->d_name, ".") == 0 || strcmp(entry->d_name, "..") == 0)
+            continue;
+
+        make_file_path(priv->index_basepath, subdir, entry->d_name, index_path,
+                       sizeof(index_path));
+
+        ret = sys_lstat(index_path, &lstatbuf);
+        if (ret < 0) {
+            count = -2;
+            continue;
+        } else {
+            count = lstatbuf.st_nlink - 1;
+            if (count == 0)
+                continue;
+            else
+                break;
+        }
+    }
+out:
+    if (dirp)
+        (void)sys_closedir(dirp);
+    return count;
+}
+
+dict_t *
+index_fill_link_count(xlator_t *this, dict_t *xdata)
+{
+    int ret = -1;
+    index_priv_t *priv = NULL;
+    int64_t count = -1;
+
+    priv = this->private;
+    xdata = (xdata) ? dict_ref(xdata) : dict_new();
+    if (!xdata)
+        goto out;
+
+    index_get_link_count(priv, &count, XATTROP);
+    if (count < 0) {
+        count = index_fetch_link_count(this, XATTROP);
+        index_set_link_count(priv, count, XATTROP);
+    }
+
+    if (count == 0) {
+        ret = dict_set_int8(xdata, "link-count", 0);
+        if (ret < 0)
+            gf_msg(this->name, GF_LOG_ERROR, EINVAL, INDEX_MSG_DICT_SET_FAILED,
+                   "Unable to set link-count");
+    } else {
+        ret = dict_set_int8(xdata, "link-count", 1);
+        if (ret < 0)
+            gf_msg(this->name, GF_LOG_ERROR, EINVAL, INDEX_MSG_DICT_SET_FAILED,
+                   "Unable to set link-count");
+    }
+
+out:
+    return xdata;
+}
+
+int32_t
+index_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, inode_t *inode,
+                 struct iatt *buf, dict_t *xdata, struct iatt *postparent)
+{
+    xdata = index_fill_link_count(this, xdata);
+    STACK_UNWIND_STRICT(lookup, frame, op_ret, op_errno, inode, buf, xdata,
+                        postparent);
+    if (xdata)
+        dict_unref(xdata);
+    return 0;
+}
+
+int32_t
+index_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
+{
+    inode_t *inode = NULL;
+    call_stub_t *stub = NULL;
+    char *flag = NULL;
+    int ret = -1;
+
+    if (!index_is_fop_on_internal_inode(this, loc->parent, loc->pargfid) &&
+        !index_is_fop_on_internal_inode(this, loc->inode, loc->gfid)) {
+        if (!inode_is_linked(loc->inode)) {
+            inode = inode_find(loc->inode->table, loc->gfid);
+            if (!index_is_fop_on_internal_inode(this, inode, loc->gfid)) {
+                inode_unref(inode);
+                goto normal;
+            }
+            inode_unref(inode);
+        } else {
+            goto normal;
+        }
+    }
+
+    stub = fop_lookup_stub(frame, index_lookup_wrapper, loc, xattr_req);
+    if (!stub) {
+        STACK_UNWIND_STRICT(lookup, frame, -1, ENOMEM, loc->inode, NULL, NULL,
+                            NULL);
+        return 0;
+    }
+    worker_enqueue(this, stub);
+    return 0;
+normal:
+    ret = dict_get_str_sizen(xattr_req, "link-count", &flag);
+    if ((ret == 0) && (strcmp(flag, GF_XATTROP_INDEX_COUNT) == 0)) {
+        STACK_WIND(frame, index_lookup_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->lookup, loc, xattr_req);
+    } else {
+        STACK_WIND(frame, default_lookup_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->lookup, loc, xattr_req);
+    }
+
+    return 0;
+}
+
+int32_t
+index_fstat_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                dict_t *xdata)
+{
+    xdata = index_fill_link_count(this, xdata);
+    STACK_UNWIND_STRICT(fstat, frame, op_ret, op_errno, buf, xdata);
+    if (xdata)
+        dict_unref(xdata);
+    return 0;
+}
+
+int32_t
+index_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    int ret = -1;
+    char *flag = NULL;
+
+    ret = dict_get_str(xdata, "link-count", &flag);
+    if ((ret == 0) && (strcmp(flag, GF_XATTROP_INDEX_COUNT) == 0)) {
+        STACK_WIND(frame, index_fstat_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fstat, fd, xdata);
+    } else {
+        STACK_WIND(frame, default_fstat_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fstat, fd, xdata);
+    }
+
+    return 0;
+}
+
+int32_t
+index_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+              dict_t *xdata)
+{
+    if (!index_is_fop_on_internal_inode(this, fd->inode, NULL))
+        goto normal;
+
+    frame->local = NULL;
+    STACK_UNWIND_STRICT(opendir, frame, 0, 0, fd, NULL);
+    return 0;
+
+normal:
+    STACK_WIND(frame, default_opendir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->opendir, loc, fd, xdata);
+    return 0;
+}
+
+int32_t
+index_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+              off_t off, dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+
+    if (!index_is_fop_on_internal_inode(this, fd->inode, NULL))
+        goto out;
+
+    stub = fop_readdir_stub(frame, index_readdir_wrapper, fd, size, off, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(readdir, frame, -1, ENOMEM, NULL, NULL);
+        return 0;
+    }
+    worker_enqueue(this, stub);
+    return 0;
+out:
+    STACK_WIND(frame, default_readdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdir, fd, size, off, xdata);
+    return 0;
+}
+
+int
+index_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+             dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+
+    if (!index_is_fop_on_internal_inode(this, loc->parent, NULL))
+        goto out;
+
+    stub = fop_unlink_stub(frame, index_unlink_wrapper, loc, xflag, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(unlink, frame, -1, ENOMEM, NULL, NULL, NULL);
+        return 0;
+    }
+    worker_enqueue(this, stub);
+    return 0;
+out:
+    STACK_WIND(frame, default_unlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->unlink, loc, xflag, xdata);
+    return 0;
+}
+
+int
+index_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+            dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+
+    if (!index_is_fop_on_internal_inode(this, loc->parent, NULL))
+        goto out;
+
+    stub = fop_rmdir_stub(frame, index_rmdir_wrapper, loc, flags, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(rmdir, frame, -1, ENOMEM, NULL, NULL, NULL);
+        return 0;
+    }
+    worker_enqueue(this, stub);
+    return 0;
+out:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->rmdir,
+                    loc, flags, xdata);
+    return 0;
+}
+
+int
+index_make_xattrop_watchlist(xlator_t *this, index_priv_t *priv,
+                             char *watchlist, index_xattrop_type_t type)
+{
+    char *delim = NULL;
+    char *dup_watchlist = NULL;
+    char *key = NULL;
+    char *saveptr = NULL;
+    dict_t *xattrs = NULL;
+    data_t *dummy = NULL;
+    int ret = 0;
+
+    if (!watchlist)
+        return 0;
+
+    dup_watchlist = gf_strdup(watchlist);
+    if (!dup_watchlist)
+        return -1;
+
+    xattrs = dict_new();
+    if (!xattrs) {
+        ret = -1;
+        goto out;
+    }
+
+    dummy = int_to_data(1);
+    if (!dummy) {
+        ret = -1;
+        goto out;
+    }
+
+    data_ref(dummy);
+
+    delim = ",";
+    key = strtok_r(dup_watchlist, delim, &saveptr);
+    while (key) {
+        if (strlen(key) == 0) {
+            ret = -1;
+            goto out;
+        }
+
+        ret = dict_set(xattrs, key, dummy);
+        if (ret)
+            goto out;
+
+        key = strtok_r(NULL, delim, &saveptr);
+    }
+
+    switch (type) {
+        case DIRTY:
+            priv->dirty_watchlist = dict_copy_with_ref(xattrs,
+                                                       priv->dirty_watchlist);
+            if (!priv->dirty_watchlist) {
+                ret = -1;
+                goto out;
+            }
+            break;
+        case XATTROP:
+            priv->pending_watchlist = dict_copy_with_ref(
+                xattrs, priv->pending_watchlist);
+            if (!priv->pending_watchlist) {
+                ret = -1;
+                goto out;
+            }
+            break;
+        default:
+            break;
+    }
+
+    ret = 0;
+out:
+    if (xattrs)
+        dict_unref(xattrs);
+
+    GF_FREE(dup_watchlist);
+
+    if (dummy)
+        data_unref(dummy);
+
+    return ret;
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    ret = xlator_mem_acct_init(this, gf_index_mt_end + 1);
+
+    return ret;
+}
+
+int
+init(xlator_t *this)
+{
+    int i = 0;
+    int ret = -1;
+    int64_t count = -1;
+    index_priv_t *priv = NULL;
+    pthread_attr_t w_attr;
+    gf_boolean_t mutex_inited = _gf_false;
+    gf_boolean_t cond_inited = _gf_false;
+    gf_boolean_t attr_inited = _gf_false;
+    char *watchlist = NULL;
+    char *dirtylist = NULL;
+    char *pendinglist = NULL;
+    char *index_base_parent = NULL;
+    char *tmp = NULL;
+
+    if (!this->children || this->children->next) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, INDEX_MSG_INVALID_GRAPH,
+               "'index' not configured with exactly one child");
+        goto out;
+    }
+
+    if (!this->parents) {
+        gf_msg(this->name, GF_LOG_WARNING, EINVAL, INDEX_MSG_INVALID_GRAPH,
+               "dangling volume. check volfile ");
+    }
+
+    priv = GF_CALLOC(1, sizeof(*priv), gf_index_mt_priv_t);
+    if (!priv)
+        goto out;
+
+    LOCK_INIT(&priv->lock);
+    if ((ret = pthread_cond_init(&priv->cond, NULL)) != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, ret, INDEX_MSG_INVALID_ARGS,
+               "pthread_cond_init failed");
+        goto out;
+    }
+    cond_inited = _gf_true;
+
+    if ((ret = pthread_mutex_init(&priv->mutex, NULL)) != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, ret, INDEX_MSG_INVALID_ARGS,
+               "pthread_mutex_init failed");
+        goto out;
+    }
+    mutex_inited = _gf_true;
+
+    if ((ret = pthread_attr_init(&w_attr)) != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, ret, INDEX_MSG_INVALID_ARGS,
+               "pthread_attr_init failed");
+        goto out;
+    }
+    attr_inited = _gf_true;
+
+    ret = pthread_attr_setstacksize(&w_attr, INDEX_THREAD_STACK_SIZE);
+    if (ret == EINVAL) {
+        gf_msg(this->name, GF_LOG_WARNING, ret, INDEX_MSG_INVALID_ARGS,
+               "Using default thread stack size");
+    }
+
+    GF_OPTION_INIT("index-base", priv->index_basepath, path, out);
+    tmp = gf_strdup(priv->index_basepath);
+    index_base_parent = dirname(tmp);
+    if (gf_lstat_dir(index_base_parent, NULL) != 0) {
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, errno,
+               INDEX_MSG_INDEX_DIR_CREATE_FAILED,
+               "Failed to find parent dir (%s) of index basepath %s.",
+               index_base_parent, priv->index_basepath);
+        goto out;
+    }
+
+    GF_OPTION_INIT("xattrop64-watchlist", watchlist, str, out);
+    ret = index_make_xattrop_watchlist(this, priv, watchlist, XATTROP);
+    if (ret)
+        goto out;
+
+    GF_OPTION_INIT("xattrop-dirty-watchlist", dirtylist, str, out);
+    ret = index_make_xattrop_watchlist(this, priv, dirtylist, DIRTY);
+    if (ret)
+        goto out;
+
+    GF_OPTION_INIT("xattrop-pending-watchlist", pendinglist, str, out);
+    ret = index_make_xattrop_watchlist(this, priv, pendinglist, XATTROP);
+    if (ret)
+        goto out;
+
+    if (priv->dirty_watchlist)
+        priv->complete_watchlist = dict_copy_with_ref(priv->dirty_watchlist,
+                                                      priv->complete_watchlist);
+    if (priv->pending_watchlist)
+        priv->complete_watchlist = dict_copy_with_ref(priv->pending_watchlist,
+                                                      priv->complete_watchlist);
+
+    gf_uuid_generate(priv->index);
+    for (i = 0; i < XATTROP_TYPE_END; i++)
+        gf_uuid_generate(priv->internal_vgfid[i]);
+
+    INIT_LIST_HEAD(&priv->callstubs);
+    GF_ATOMIC_INIT(priv->stub_cnt, 0);
+
+    this->local_pool = mem_pool_new(index_local_t, 64);
+    if (!this->local_pool) {
+        ret = -1;
+        goto out;
+    }
+
+    this->private = priv;
+
+    ret = index_dir_create(this, XATTROP_SUBDIR);
+    if (ret < 0)
+        goto out;
+
+    if (priv->dirty_watchlist) {
+        ret = index_dir_create(this, DIRTY_SUBDIR);
+        if (ret < 0)
+            goto out;
+    }
+
+    ret = index_dir_create(this, ENTRY_CHANGES_SUBDIR);
+    if (ret < 0)
+        goto out;
+
+    /*init indices files counts*/
+    count = index_fetch_link_count(this, XATTROP);
+    index_set_link_count(priv, count, XATTROP);
+    priv->down = _gf_false;
+
+    priv->curr_count = 0;
+    ret = gf_thread_create(&priv->thread, &w_attr, index_worker, this,
+                           "idxwrker");
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, ret,
+               INDEX_MSG_WORKER_THREAD_CREATE_FAILED,
+               "Failed to create worker thread, aborting");
+        goto out;
+    }
+    priv->curr_count++;
+    ret = 0;
+out:
+    GF_FREE(tmp);
+
+    if (ret) {
+        if (cond_inited)
+            pthread_cond_destroy(&priv->cond);
+        if (mutex_inited)
+            pthread_mutex_destroy(&priv->mutex);
+        if (priv && priv->dirty_watchlist)
+            dict_unref(priv->dirty_watchlist);
+        if (priv && priv->pending_watchlist)
+            dict_unref(priv->pending_watchlist);
+        if (priv && priv->complete_watchlist)
+            dict_unref(priv->complete_watchlist);
+        if (priv)
+            GF_FREE(priv);
+        this->private = NULL;
+        mem_pool_destroy(this->local_pool);
+        this->local_pool = NULL;
+    }
+
+    if (attr_inited)
+        pthread_attr_destroy(&w_attr);
+    return ret;
+}
+
+void
+fini(xlator_t *this)
+{
+    index_priv_t *priv = NULL;
+
+    priv = this->private;
+    if (!priv)
+        goto out;
+
+    priv->down = _gf_true;
+    pthread_cond_broadcast(&priv->cond);
+    if (priv->thread) {
+        gf_thread_cleanup_xint(priv->thread);
+        priv->thread = 0;
+    }
+    this->private = NULL;
+    LOCK_DESTROY(&priv->lock);
+    pthread_cond_destroy(&priv->cond);
+    pthread_mutex_destroy(&priv->mutex);
+    if (priv->dirty_watchlist)
+        dict_unref(priv->dirty_watchlist);
+    if (priv->pending_watchlist)
+        dict_unref(priv->pending_watchlist);
+    if (priv->complete_watchlist)
+        dict_unref(priv->complete_watchlist);
+    GF_FREE(priv);
+
+    if (this->local_pool) {
+        mem_pool_destroy(this->local_pool);
+        this->local_pool = NULL;
+    }
+out:
+    return;
+}
+
+int
+index_forget(xlator_t *this, inode_t *inode)
+{
+    uint64_t tmp_cache = 0;
+    if (!inode_ctx_del(inode, this, &tmp_cache))
+        GF_FREE((index_inode_ctx_t *)(long)tmp_cache);
+
+    return 0;
+}
+
+int32_t
+index_releasedir(xlator_t *this, fd_t *fd)
+{
+    index_fd_ctx_t *fctx = NULL;
+    uint64_t ctx = 0;
+    int ret = 0;
+
+    ret = fd_ctx_del(fd, this, &ctx);
+    if (ret < 0)
+        goto out;
+
+    fctx = (index_fd_ctx_t *)(long)ctx;
+    if (fctx->dir) {
+        ret = sys_closedir(fctx->dir);
+        if (ret)
+            gf_msg(this->name, GF_LOG_ERROR, errno, INDEX_MSG_FD_OP_FAILED,
+                   "closedir error");
+    }
+
+    GF_FREE(fctx);
+out:
+    return 0;
+}
+
+int32_t
+index_release(xlator_t *this, fd_t *fd)
+{
+    index_fd_ctx_t *fctx = NULL;
+    uint64_t ctx = 0;
+    int ret = 0;
+
+    ret = fd_ctx_del(fd, this, &ctx);
+    if (ret < 0)
+        goto out;
+
+    fctx = (index_fd_ctx_t *)(long)ctx;
+    GF_FREE(fctx);
+out:
+    return 0;
+}
+
+int
+notify(xlator_t *this, int event, void *data, ...)
+{
+    int ret = 0;
+    index_priv_t *priv = NULL;
+    uint64_t stub_cnt = 0;
+    xlator_t *victim = data;
+    struct timespec sleep_till = {
+        0,
+    };
+
+    if (!this)
+        return 0;
+
+    priv = this->private;
+    if (!priv)
+        return 0;
+
+    if ((event == GF_EVENT_PARENT_DOWN) && victim->cleanup_starting) {
+        stub_cnt = GF_ATOMIC_GET(priv->stub_cnt);
+        timespec_now_realtime(&sleep_till);
+        sleep_till.tv_sec += 1;
+
+        /* Wait for draining stub from queue before notify PARENT_DOWN */
+        pthread_mutex_lock(&priv->mutex);
+        {
+            while (stub_cnt) {
+                (void)pthread_cond_timedwait(&priv->cond, &priv->mutex,
+                                             &sleep_till);
+                stub_cnt = GF_ATOMIC_GET(priv->stub_cnt);
+            }
+        }
+        pthread_mutex_unlock(&priv->mutex);
+        gf_log(this->name, GF_LOG_INFO,
+               "Notify GF_EVENT_PARENT_DOWN for brick %s", victim->name);
+    }
+
+    if ((event == GF_EVENT_CHILD_DOWN) && victim->cleanup_starting) {
+        pthread_mutex_lock(&priv->mutex);
+        {
+            priv->down = _gf_true;
+            pthread_cond_broadcast(&priv->cond);
+            while (priv->curr_count)
+                pthread_cond_wait(&priv->cond, &priv->mutex);
+        }
+        pthread_mutex_unlock(&priv->mutex);
+
+        gf_log(this->name, GF_LOG_INFO,
+               "Notify GF_EVENT_CHILD_DOWN for brick %s", victim->name);
+    }
+
+    ret = default_notify(this, event, data);
+    return ret;
+}
+
+struct xlator_fops fops = {
+    .xattrop = index_xattrop,
+    .fxattrop = index_fxattrop,
+
+    // interface functions follow
+    .getxattr = index_getxattr,
+    .lookup = index_lookup,
+    .opendir = index_opendir,
+    .readdir = index_readdir,
+    .unlink = index_unlink,
+    .rmdir = index_rmdir,
+    .fstat = index_fstat,
+};
+
+struct xlator_dumpops dumpops;
+
+struct xlator_cbks cbks = {.forget = index_forget,
+                           .release = index_release,
+                           .releasedir = index_releasedir};
+
+struct volume_options options[] = {
+    {.key = {"index-base"},
+     .type = GF_OPTION_TYPE_PATH,
+     .description = "path where the index files need to be stored",
+     .default_value = "{{ brick.path }}/.glusterfs/indices"},
+    {.key = {"xattrop64-watchlist"},
+     .type = GF_OPTION_TYPE_STR,
+     .description = "Comma separated list of xattrs that are watched",
+     .default_value = "trusted.ec.dirty"},
+    {.key = {"xattrop-dirty-watchlist"},
+     .type = GF_OPTION_TYPE_STR,
+     .description = "Comma separated list of xattrs that are watched",
+     .default_value = "trusted.afr.dirty"},
+    {.key = {"xattrop-pending-watchlist"},
+     .type = GF_OPTION_TYPE_STR,
+     .description = "Comma separated list of xattrs that are watched",
+     .default_value = "trusted.afr.{{ volume.name }}"},
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .dumpops = &dumpops,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "index",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/features/index/src/index.h b/xlators/features/index/src/index.h
new file mode 100644
index 00000000000..a2b6e6e2570
--- /dev/null
+++ b/xlators/features/index/src/index.h
@@ -0,0 +1,86 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __INDEX_H__
+#define __INDEX_H__
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/call-stub.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/byte-order.h>
+#include <glusterfs/common-utils.h>
+#include "index-mem-types.h"
+
+#define INDEX_THREAD_STACK_SIZE ((size_t)(1024 * 1024))
+
+typedef enum { UNKNOWN, IN, NOTIN } index_state_t;
+
+typedef enum {
+    XATTROP_TYPE_UNSET = -1,
+    XATTROP,
+    DIRTY,
+    ENTRY_CHANGES,
+    XATTROP_TYPE_END
+} index_xattrop_type_t;
+
+typedef struct index_inode_ctx {
+    gf_boolean_t processing;
+    struct list_head callstubs;
+    int state[XATTROP_TYPE_END];
+    uuid_t virtual_pargfid; /* virtual gfid of dir under
+                              .glusterfs/indices/entry-changes. */
+} index_inode_ctx_t;
+
+typedef struct index_fd_ctx {
+    DIR *dir;
+    off_t dir_eof;
+} index_fd_ctx_t;
+
+typedef struct index_priv {
+    char *index_basepath;
+    char *dirty_basepath;
+    uuid_t index;
+    gf_lock_t lock;
+    uuid_t internal_vgfid[XATTROP_TYPE_END];
+    struct list_head callstubs;
+    pthread_mutex_t mutex;
+    pthread_cond_t cond;
+    dict_t *dirty_watchlist;
+    dict_t *pending_watchlist;
+    dict_t *complete_watchlist;
+    int64_t pending_count;
+    pthread_t thread;
+    gf_boolean_t down;
+    gf_atomic_t stub_cnt;
+    int32_t curr_count;
+} index_priv_t;
+
+typedef struct index_local {
+    inode_t *inode;
+    dict_t *xdata;
+} index_local_t;
+
+#define INDEX_STACK_UNWIND(fop, frame, params...)                              \
+    do {                                                                       \
+        index_local_t *__local = NULL;                                         \
+        if (frame) {                                                           \
+            __local = frame->local;                                            \
+            frame->local = NULL;                                               \
+        }                                                                      \
+        STACK_UNWIND_STRICT(fop, frame, params);                               \
+        if (__local) {                                                         \
+            inode_unref(__local->inode);                                       \
+            if (__local->xdata)                                                \
+                dict_unref(__local->xdata);                                    \
+            mem_put(__local);                                                  \
+        }                                                                      \
+    } while (0)
+
+#endif
diff --git a/xlators/features/leases/Makefile.am b/xlators/features/leases/Makefile.am
new file mode 100644
index 00000000000..a985f42a877
--- /dev/null
+++ b/xlators/features/leases/Makefile.am
@@ -0,0 +1,3 @@
+SUBDIRS = src
+
+CLEANFILES =
diff --git a/xlators/features/leases/src/Makefile.am b/xlators/features/leases/src/Makefile.am
new file mode 100644
index 00000000000..a1aef10e299
--- /dev/null
+++ b/xlators/features/leases/src/Makefile.am
@@ -0,0 +1,20 @@
+if WITH_SERVER
+xlator_LTLIBRARIES = leases.la
+endif
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+leases_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+leases_la_SOURCES = leases.c leases-internal.c
+
+leases_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = leases.h leases-mem-types.h leases-messages.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src \
+	-I$(CONTRIBDIR)/timer-wheel
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/leases/src/leases-internal.c b/xlators/features/leases/src/leases-internal.c
new file mode 100644
index 00000000000..56dee244281
--- /dev/null
+++ b/xlators/features/leases/src/leases-internal.c
@@ -0,0 +1,1412 @@
+/*
+   Copyright (c) 2015-2016 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "leases.h"
+
+/* Mutex locks used in this xlator and their order of acquisition:
+ * Check lease conflict:
+ *         lease_ctx lock
+ *                 add_timer => internal timer locks
+ *         lease_ctx unlock
+ *
+ * Add/remove lease:
+ *         lease_ctx lock
+ *                 add_timer => internal timer locks
+ *                 OR
+ *                 priv lock => Adding/removing to/from the cleanup client list
+ *                 priv unlock
+ *         lease_ctx unlock
+ *
+ * Timer thread:
+ *         Timer internal lock
+ *                 priv lock => By timer handler
+ *                 priv unlock
+ *         Timer internal unlock
+ *
+ * Expired recall cleanup thread:
+ *         priv lock
+ *                 priv condwait
+ *         priv unlock
+ *         lease_ctx lock
+ *                 priv lock
+ *                 priv unlock
+ *         lease_ctx unlock
+ */
+
+/*
+ * Check if lease_lk is enabled
+ * Return Value:
+ * _gf_true  - lease lock option enabled
+ * _gf_false - lease lock option disabled
+ */
+gf_boolean_t
+is_leases_enabled(xlator_t *this)
+{
+    leases_private_t *priv = NULL;
+    gf_boolean_t is_enabled = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("leases", this, out);
+
+    if (this->private) {
+        priv = (leases_private_t *)this->private;
+        is_enabled = priv->leases_enabled;
+    }
+out:
+    return is_enabled;
+}
+
+/*
+ * Get the recall_leaselk_timeout
+ * Return Value:
+ * timeout value(in seconds) set as an option to this xlator.
+ * -1 error case
+ */
+static int32_t
+get_recall_lease_timeout(xlator_t *this)
+{
+    leases_private_t *priv = NULL;
+    int32_t timeout = -1;
+
+    GF_VALIDATE_OR_GOTO("leases", this, out);
+
+    if (this->private) {
+        priv = (leases_private_t *)this->private;
+        timeout = priv->recall_lease_timeout;
+    }
+out:
+    return timeout;
+}
+
+static void
+__dump_leases_info(xlator_t *this, lease_inode_ctx_t *lease_ctx)
+{
+    lease_id_entry_t *lease_entry = NULL;
+    lease_id_entry_t *tmp = NULL;
+
+    GF_VALIDATE_OR_GOTO("leases", this, out);
+    GF_VALIDATE_OR_GOTO("leases", lease_ctx, out);
+
+    gf_msg_debug(this->name, 0,
+                 "Lease held on this inode, lease_type: %d,"
+                 " lease_cnt:%" PRIu64
+                 ", RD lease:%d, RW lease:%d, "
+                 "openfd cnt:%" PRIu64,
+                 lease_ctx->lease_type, lease_ctx->lease_cnt,
+                 lease_ctx->lease_type_cnt[GF_RD_LEASE],
+                 lease_ctx->lease_type_cnt[GF_RW_LEASE], lease_ctx->openfd_cnt);
+
+    list_for_each_entry_safe(lease_entry, tmp, &lease_ctx->lease_id_list,
+                             lease_id_list)
+    {
+        gf_msg_debug(this->name, 0,
+                     "Leases held by client: %s, lease "
+                     "ID:%s, RD lease:%d, RW lease:%d, lease_type: %d, "
+                     "lease_cnt:%" PRIu64,
+                     lease_entry->client_uid, lease_entry->lease_id,
+                     lease_entry->lease_type_cnt[GF_RD_LEASE],
+                     lease_entry->lease_type_cnt[GF_RW_LEASE],
+                     lease_entry->lease_type, lease_entry->lease_cnt);
+    }
+out:
+    return;
+}
+
+static int
+__lease_ctx_set(inode_t *inode, xlator_t *this)
+{
+    lease_inode_ctx_t *inode_ctx = NULL;
+    int ret = -1;
+    uint64_t ctx = 0;
+
+    GF_VALIDATE_OR_GOTO("leases", inode, out);
+    GF_VALIDATE_OR_GOTO("leases", this, out);
+
+    ret = __inode_ctx_get(inode, this, &ctx);
+    if (!ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, LEASE_MSG_INVAL_INODE_CTX,
+               "inode_ctx_get failed");
+        goto out;
+    }
+
+    inode_ctx = GF_CALLOC(1, sizeof(*inode_ctx),
+                          gf_leases_mt_lease_inode_ctx_t);
+    GF_CHECK_ALLOC(inode_ctx, ret, out);
+
+    pthread_mutex_init(&inode_ctx->lock, NULL);
+    INIT_LIST_HEAD(&inode_ctx->lease_id_list);
+    INIT_LIST_HEAD(&inode_ctx->blocked_list);
+
+    inode_ctx->lease_cnt = 0;
+
+    ret = __inode_ctx_set(inode, this, (uint64_t *)inode_ctx);
+    if (ret) {
+        GF_FREE(inode_ctx);
+        gf_msg(this->name, GF_LOG_INFO, 0, LEASE_MSG_INVAL_INODE_CTX,
+               "failed to set inode ctx (%p)", inode);
+    }
+out:
+    return ret;
+}
+
+static lease_inode_ctx_t *
+__lease_ctx_get(inode_t *inode, xlator_t *this)
+{
+    lease_inode_ctx_t *inode_ctx = NULL;
+    uint64_t ctx = 0;
+    int ret = 0;
+
+    GF_VALIDATE_OR_GOTO("leases", inode, out);
+    GF_VALIDATE_OR_GOTO("leases", this, out);
+
+    ret = __inode_ctx_get(inode, this, &ctx);
+    if (ret < 0) {
+        ret = __lease_ctx_set(inode, this);
+        if (ret < 0)
+            goto out;
+
+        ret = __inode_ctx_get(inode, this, &ctx);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, LEASE_MSG_INVAL_INODE_CTX,
+                   "failed to get inode ctx (%p)", inode);
+            goto out;
+        }
+    }
+
+    inode_ctx = (lease_inode_ctx_t *)(long)ctx;
+out:
+    return inode_ctx;
+}
+
+lease_inode_ctx_t *
+lease_ctx_get(inode_t *inode, xlator_t *this)
+{
+    lease_inode_ctx_t *inode_ctx = NULL;
+
+    GF_VALIDATE_OR_GOTO("leases", inode, out);
+    GF_VALIDATE_OR_GOTO("leases", this, out);
+
+    LOCK(&inode->lock);
+    {
+        inode_ctx = __lease_ctx_get(inode, this);
+    }
+    UNLOCK(&inode->lock);
+out:
+    return inode_ctx;
+}
+
+static lease_id_entry_t *
+new_lease_id_entry(call_frame_t *frame, const char *lease_id)
+{
+    lease_id_entry_t *lease_entry = NULL;
+
+    GF_VALIDATE_OR_GOTO("leases", frame, out);
+    GF_VALIDATE_OR_GOTO("leases", lease_id, out);
+
+    lease_entry = GF_CALLOC(1, sizeof(*lease_entry),
+                            gf_leases_mt_lease_id_entry_t);
+    if (!lease_entry) {
+        gf_msg(frame->this->name, GF_LOG_ERROR, ENOMEM, LEASE_MSG_NO_MEM,
+               "Memory allocation for lease_entry failed");
+        return NULL;
+    }
+
+    INIT_LIST_HEAD(&lease_entry->lease_id_list);
+    lease_entry->lease_type = NONE;
+    lease_entry->lease_cnt = 0;
+    lease_entry->recall_time = get_recall_lease_timeout(frame->this);
+    lease_entry->client_uid = gf_strdup(frame->root->client->client_uid);
+    if (!lease_entry->client_uid) {
+        gf_msg(frame->this->name, GF_LOG_ERROR, ENOMEM, LEASE_MSG_NO_MEM,
+               "Memory allocation for client_uid failed");
+        GF_FREE(lease_entry);
+        lease_entry = NULL;
+        goto out;
+    }
+
+    memcpy(lease_entry->lease_id, lease_id, LEASE_ID_SIZE);
+out:
+    return lease_entry;
+}
+
+static void
+__destroy_lease_id_entry(lease_id_entry_t *lease_entry)
+{
+    GF_VALIDATE_OR_GOTO("leases", lease_entry, out);
+
+    list_del_init(&lease_entry->lease_id_list);
+    GF_FREE(lease_entry->client_uid);
+    GF_FREE(lease_entry);
+out:
+    return;
+}
+
+static inline gf_boolean_t
+__is_same_lease_id(const char *k1, const char *k2)
+{
+    if (memcmp(k1, k2, strlen(k1)) == 0)
+        return _gf_true;
+
+    return _gf_false;
+}
+
+/* Checks if there are any leases, other than the leases taken
+ * by the given lease_id
+ */
+static gf_boolean_t
+__another_lease_found(lease_inode_ctx_t *lease_ctx, const char *lease_id)
+{
+    lease_id_entry_t *lease_entry = NULL;
+    lease_id_entry_t *tmp = NULL;
+    gf_boolean_t found_lease = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("leases", lease_id, out);
+    GF_VALIDATE_OR_GOTO("leases", lease_ctx, out);
+
+    list_for_each_entry_safe(lease_entry, tmp, &lease_ctx->lease_id_list,
+                             lease_id_list)
+    {
+        if (!__is_same_lease_id(lease_id, lease_entry->lease_id)) {
+            if (lease_entry->lease_cnt > 0) {
+                found_lease = _gf_true;
+                break;
+            }
+        }
+    }
+out:
+    return found_lease;
+}
+
+/* Returns the lease_id_entry for a given lease_id and a given inode.
+ * Return values:
+ * NULL - If no client entry found
+ * lease_id_entry_t* - a pointer to the client entry if found
+ */
+static lease_id_entry_t *
+__get_lease_id_entry(lease_inode_ctx_t *lease_ctx, const char *lease_id)
+{
+    lease_id_entry_t *lease_entry = NULL;
+    lease_id_entry_t *tmp = NULL;
+    lease_id_entry_t *found = NULL;
+
+    GF_VALIDATE_OR_GOTO("leases", lease_id, out);
+    GF_VALIDATE_OR_GOTO("leases", lease_ctx, out);
+
+    list_for_each_entry_safe(lease_entry, tmp, &lease_ctx->lease_id_list,
+                             lease_id_list)
+    {
+        if (__is_same_lease_id(lease_id, lease_entry->lease_id)) {
+            found = lease_entry;
+            gf_msg_debug("leases", 0,
+                         "lease ID entry found "
+                         "Client UID:%s, lease id:%s",
+                         lease_entry->client_uid,
+                         leaseid_utoa(lease_entry->lease_id));
+            break;
+        }
+    }
+out:
+    return found;
+}
+
+/* Returns the lease_id_entry for a given lease_id and a given inode,
+ * if none found creates one.
+ * Return values:
+ * lease_id_entry_t* - a pointer to the client entry
+ */
+static lease_id_entry_t *
+__get_or_new_lease_entry(call_frame_t *frame, const char *lease_id,
+                         lease_inode_ctx_t *lease_ctx)
+{
+    lease_id_entry_t *lease_entry = NULL;
+
+    GF_VALIDATE_OR_GOTO("leases", frame, out);
+    GF_VALIDATE_OR_GOTO("leases", lease_id, out);
+    GF_VALIDATE_OR_GOTO("leases", lease_ctx, out);
+
+    lease_entry = __get_lease_id_entry(lease_ctx, lease_id);
+    if (!lease_entry) { /* create one */
+        lease_entry = new_lease_id_entry(frame, lease_id);
+        if (!lease_entry)
+            goto out;
+
+        list_add_tail(&lease_entry->lease_id_list, &lease_ctx->lease_id_list);
+
+        gf_msg_debug(frame->this->name, 0,
+                     "lease ID entry added,"
+                     " Client UID:%s, lease id:%s",
+                     lease_entry->client_uid,
+                     leaseid_utoa(lease_entry->lease_id));
+    }
+out:
+    return lease_entry;
+}
+
+static lease_inode_t *
+new_lease_inode(inode_t *inode)
+{
+    lease_inode_t *l_inode = GF_MALLOC(sizeof(*l_inode),
+                                       gf_leases_mt_lease_inode_t);
+    if (!l_inode)
+        goto out;
+
+    INIT_LIST_HEAD(&l_inode->list);
+    l_inode->inode = inode_ref(inode);
+out:
+    return l_inode;
+}
+
+static void
+__destroy_lease_inode(lease_inode_t *l_inode)
+{
+    list_del_init(&l_inode->list);
+    inode_unref(l_inode->inode);
+    GF_FREE(l_inode);
+}
+
+static lease_client_t *
+new_lease_client(const char *client_uid)
+{
+    lease_client_t *clnt = GF_MALLOC(sizeof(*clnt),
+                                     gf_leases_mt_lease_client_t);
+    if (!clnt)
+        goto out;
+
+    INIT_LIST_HEAD(&clnt->client_list);
+    INIT_LIST_HEAD(&clnt->inode_list);
+    clnt->client_uid = gf_strdup(client_uid);
+out:
+    return clnt;
+}
+
+static void
+__destroy_lease_client(lease_client_t *clnt)
+{
+    list_del_init(&clnt->inode_list);
+    list_del_init(&clnt->client_list);
+    GF_FREE(clnt);
+
+    return;
+}
+
+static lease_client_t *
+__get_lease_client(xlator_t *this, leases_private_t *priv,
+                   const char *client_uid)
+{
+    lease_client_t *clnt = NULL;
+    lease_client_t *tmp = NULL;
+    lease_client_t *found = NULL;
+
+    list_for_each_entry_safe(clnt, tmp, &priv->client_list, client_list)
+    {
+        if ((strcmp(clnt->client_uid, client_uid) == 0)) {
+            found = clnt;
+            gf_msg_debug(this->name, 0,
+                         "Client:%s already found "
+                         "in the cleanup list",
+                         client_uid);
+            break;
+        }
+    }
+    return found;
+}
+
+static lease_client_t *
+__get_or_new_lease_client(xlator_t *this, leases_private_t *priv,
+                          const char *client_uid)
+{
+    lease_client_t *found = NULL;
+
+    found = __get_lease_client(this, priv, client_uid);
+    if (!found) {
+        found = new_lease_client(client_uid);
+        if (!found)
+            goto out;
+        list_add_tail(&found->client_list, &priv->client_list);
+        gf_msg_debug(this->name, 0,
+                     "Adding a new client:%s entry "
+                     "to the cleanup list",
+                     client_uid);
+    }
+out:
+    return found;
+}
+
+static int
+add_inode_to_client_list(xlator_t *this, inode_t *inode, const char *client_uid)
+{
+    leases_private_t *priv = this->private;
+    lease_client_t *clnt = NULL;
+
+    lease_inode_t *lease_inode = new_lease_inode(inode);
+    if (!lease_inode)
+        return -ENOMEM;
+
+    pthread_mutex_lock(&priv->mutex);
+    {
+        clnt = __get_or_new_lease_client(this, priv, client_uid);
+        if (!clnt) {
+            pthread_mutex_unlock(&priv->mutex);
+            __destroy_lease_inode(lease_inode);
+            return -ENOMEM;
+        }
+        list_add_tail(&clnt->inode_list, &lease_inode->list);
+    }
+    pthread_mutex_unlock(&priv->mutex);
+    gf_msg_debug(this->name, 0,
+                 "Added a new inode:%p to the client(%s) "
+                 "cleanup list, gfid(%s)",
+                 inode, client_uid, uuid_utoa(inode->gfid));
+    return 0;
+}
+
+/* Add lease entry to the corresponding client entry.
+ * Return values:
+ * 0 Success
+ * -1 Failure
+ */
+static int
+__add_lease(call_frame_t *frame, inode_t *inode, lease_inode_ctx_t *lease_ctx,
+            const char *client_uid, struct gf_lease *lease)
+{
+    lease_id_entry_t *lease_entry = NULL;
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("leases", frame, out);
+    GF_VALIDATE_OR_GOTO("leases", client_uid, out);
+    GF_VALIDATE_OR_GOTO("leases", lease_ctx, out);
+    GF_VALIDATE_OR_GOTO("leases", inode, out);
+    GF_VALIDATE_OR_GOTO("leases", lease, out);
+
+    gf_msg_trace(frame->this->name, 0,
+                 "Granting lease lock to client %s with lease id %s"
+                 " on gfid(%s)",
+                 client_uid, leaseid_utoa(lease->lease_id),
+                 uuid_utoa(inode->gfid));
+
+    lease_entry = __get_or_new_lease_entry(frame, lease->lease_id, lease_ctx);
+    if (!lease_entry) {
+        errno = ENOMEM;
+        goto out;
+    }
+
+    lease_entry->lease_type_cnt[lease->lease_type]++;
+    lease_entry->lease_cnt++;
+    lease_entry->lease_type |= lease->lease_type;
+    /* If this is the first lease taken by the client on the file, then
+     * add this inode/file to the client disconnect cleanup list
+     */
+    if (lease_entry->lease_cnt == 1) {
+        add_inode_to_client_list(frame->this, inode, client_uid);
+    }
+
+    lease_ctx->lease_cnt++;
+    lease_ctx->lease_type_cnt[lease->lease_type]++;
+    lease_ctx->lease_type |= lease->lease_type;
+
+    /* Take a ref for the first lock taken on this inode. Corresponding
+     * unref when all the leases are unlocked or during DISCONNECT
+     * Ref is required because the inode on which lease is acquired should
+     * not be deleted when lru cleanup kicks in*/
+    if (lease_ctx->lease_cnt == 1) {
+        lease_ctx->inode = inode_ref(inode);
+    }
+
+    ret = 0;
+out:
+    return ret;
+}
+
+static gf_boolean_t
+__is_clnt_lease_none(const char *client_uid, lease_inode_ctx_t *lease_ctx)
+{
+    gf_boolean_t lease_none = _gf_true;
+    lease_id_entry_t *lease_entry = NULL;
+    lease_id_entry_t *tmp = NULL;
+
+    list_for_each_entry_safe(lease_entry, tmp, &lease_ctx->lease_id_list,
+                             lease_id_list)
+    {
+        if ((strcmp(client_uid, lease_entry->client_uid) == 0) &&
+            (lease_entry->lease_cnt != 0)) {
+            lease_none = _gf_false;
+            break;
+        }
+    }
+
+    return lease_none;
+}
+
+static int
+__remove_inode_from_clnt_list(xlator_t *this, lease_client_t *clnt,
+                              inode_t *inode)
+{
+    int ret = -1;
+    lease_inode_t *l_inode = NULL;
+    lease_inode_t *tmp1 = NULL;
+
+    list_for_each_entry_safe(l_inode, tmp1, &clnt->inode_list, list)
+    {
+        if (l_inode->inode == inode) {
+            __destroy_lease_inode(l_inode);
+            gf_msg_debug(this->name, 0,
+                         "Removed the inode from the client cleanup list");
+            ret = 0;
+        }
+    }
+    /* TODO: Remove the client entry from the cleanup list */
+
+    return ret;
+}
+
+static int
+remove_from_clnt_list(xlator_t *this, const char *client_uid, inode_t *inode)
+{
+    leases_private_t *priv = NULL;
+    int ret = -1;
+    lease_client_t *clnt = NULL;
+
+    priv = this->private;
+    if (!priv)
+        goto out;
+
+    pthread_mutex_lock(&priv->mutex);
+    {
+        clnt = __get_lease_client(this, priv, client_uid);
+        if (!clnt) {
+            pthread_mutex_unlock(&priv->mutex);
+            gf_msg(this->name, GF_LOG_ERROR, 0, LEASE_MSG_CLNT_NOTFOUND,
+                   "There is no client entry found in the cleanup list");
+            goto out;
+        }
+        ret = __remove_inode_from_clnt_list(this, clnt, inode);
+        if (ret) {
+            pthread_mutex_unlock(&priv->mutex);
+            gf_msg(this->name, GF_LOG_ERROR, 0, LEASE_MSG_INODE_NOTFOUND,
+                   "There is no inode entry found in the cleanup list");
+            goto out;
+        }
+    }
+    pthread_mutex_unlock(&priv->mutex);
+out:
+    return ret;
+}
+
+/* Remove lease entry in the corresponding client entry.
+ */
+static int
+__remove_lease(xlator_t *this, inode_t *inode, lease_inode_ctx_t *lease_ctx,
+               const char *client_uid, struct gf_lease *lease)
+{
+    lease_id_entry_t *lease_entry = NULL;
+    int ret = 0;
+    int32_t lease_type = 0;
+    leases_private_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("leases", lease_ctx, out);
+    GF_VALIDATE_OR_GOTO("leases", lease, out);
+
+    priv = this->private;
+
+    gf_msg_trace(this->name, 0,
+                 "Removing lease entry for client: %s, "
+                 "lease type:%d, lease id:%s",
+                 client_uid, lease->lease_type, leaseid_utoa(lease->lease_id));
+
+    /* There could be a race where in server recalled the lease and by the time
+     * client sends lease_unlock request, server may have revoked it. To handle
+     * such cases, if lease doesnt exist treat it as noop and return success.
+     */
+    lease_entry = __get_lease_id_entry(lease_ctx, lease->lease_id);
+    if (!lease_entry) {
+        gf_msg(this->name, GF_LOG_INFO, 0, LEASE_MSG_INVAL_UNLK_LEASE,
+               "Got unlock lease request from client:%s, but has no "
+               "corresponding lock",
+               client_uid);
+        ret = 0;
+        goto out;
+    }
+
+    if (!(lease_entry->lease_type & lease->lease_type)) {
+        gf_msg(this->name, GF_LOG_INFO, 0, LEASE_MSG_INVAL_UNLK_LEASE,
+               "Got unlock lease request from client:%s for an invalid "
+               "lease_type",
+               client_uid);
+        ret = -EINVAL;
+        errno = EINVAL;
+        goto out;
+    }
+    lease_type = lease->lease_type;
+    lease_entry->lease_type_cnt[lease_type]--;
+    lease_entry->lease_cnt--;
+
+    lease_ctx->lease_type_cnt[lease_type]--;
+    lease_ctx->lease_cnt--;
+
+    if (lease_entry->lease_type_cnt[lease_type] == 0)
+        lease_entry->lease_type = lease_entry->lease_type & (~lease_type);
+
+    if (lease_ctx->lease_type_cnt[lease_type] == 0)
+        lease_ctx->lease_type = lease_ctx->lease_type & (~lease_type);
+
+    if (lease_entry->lease_cnt == 0) {
+        if (__is_clnt_lease_none(client_uid, lease_ctx)) {
+            gf_msg_trace(this->name, 0,
+                         "Client(%s) has no leases"
+                         " on gfid (%s), hence removing the inode"
+                         " from the client cleanup list",
+                         client_uid, uuid_utoa(inode->gfid));
+            remove_from_clnt_list(this, client_uid, lease_ctx->inode);
+        }
+        __destroy_lease_id_entry(lease_entry);
+        lease_ctx->blocked_fops_resuming = _gf_true;
+    }
+
+    if (lease_ctx->lease_cnt == 0 && lease_ctx->timer) {
+        ret = gf_tw_del_timer(priv->timer_wheel, lease_ctx->timer);
+        lease_ctx->recall_in_progress = _gf_false;
+        lease_ctx->timer = NULL;
+    }
+out:
+    return ret;
+}
+
+static gf_boolean_t
+__is_lease_grantable(xlator_t *this, lease_inode_ctx_t *lease_ctx,
+                     struct gf_lease *lease, inode_t *inode)
+{
+    uint32_t fd_count = 0;
+    int32_t flags = 0;
+    fd_t *iter_fd = NULL;
+    gf_boolean_t grant = _gf_false;
+    int ret = 0;
+    lease_fd_ctx_t *fd_ctx = NULL;
+    uint64_t ctx = 0;
+
+    GF_VALIDATE_OR_GOTO("leases", lease_ctx, out);
+    GF_VALIDATE_OR_GOTO("leases", lease, out);
+    GF_VALIDATE_OR_GOTO("leases", inode, out);
+
+    if (lease_ctx->recall_in_progress) {
+        gf_msg_debug(this->name, 0,
+                     "Recall in progress, hence "
+                     "failing the lease request");
+        grant = _gf_false;
+        goto out;
+    }
+
+    if (lease_ctx->blocked_fops_resuming) {
+        gf_msg_debug(this->name, 0,
+                     "Previously blocked fops resuming, hence "
+                     "failing the lease request");
+        grant = _gf_false;
+        goto out;
+    }
+
+    LOCK(&inode->lock);
+    {
+        list_for_each_entry(iter_fd, &inode->fd_list, inode_list)
+        {
+            ret = fd_ctx_get(iter_fd, this, &ctx);
+            if (ret < 0) {
+                grant = _gf_false;
+                UNLOCK(&inode->lock);
+                gf_msg(this->name, GF_LOG_ERROR, 0, LEASE_MSG_INVAL_FD_CTX,
+                       "Unable to get fd ctx");
+                goto out;
+            }
+            fd_ctx = (lease_fd_ctx_t *)(long)ctx;
+
+            /* Check for open fd conflict, note that open fds from
+             * the same lease id is not checked for conflict, as it is
+             * lease id based lease.
+             */
+            if (fd_ctx->client_uid != NULL &&
+                !__is_same_lease_id(fd_ctx->lease_id, lease->lease_id)) {
+                fd_count++;
+                flags |= iter_fd->flags;
+            }
+        }
+    }
+    UNLOCK(&inode->lock);
+
+    gf_msg_debug(this->name, 0, "open fd count:%d flags:%d", fd_count, flags);
+
+    __dump_leases_info(this, lease_ctx);
+
+    switch (lease->lease_type) {
+        case GF_RD_LEASE:
+            /* check open fd conflict */
+            if ((fd_count > 0) && ((flags & O_WRONLY) || (flags & O_RDWR))) {
+                grant = _gf_false;
+                break;
+            }
+
+            /* check for conflict with existing leases */
+            if (lease_ctx->lease_type == NONE ||
+                lease_ctx->lease_type == GF_RD_LEASE ||
+                !(__another_lease_found(lease_ctx, lease->lease_id)))
+                grant = _gf_true;
+            else
+                grant = _gf_false;
+            break;
+
+        case GF_RW_LEASE:
+            /* check open fd conflict; conflict if there are any fds open
+             * other than the client on which the lease is requested. */
+            if (fd_count > 0) {
+                grant = _gf_false;
+                break;
+            }
+
+            /* check existing lease conflict */
+            if (lease_ctx->lease_type == NONE ||
+                !(__another_lease_found(lease_ctx, lease->lease_id)))
+                grant = _gf_true;
+            else
+                grant = _gf_false;
+            break;
+
+        default:
+            gf_msg(this->name, GF_LOG_ERROR, EINVAL, LEASE_MSG_INVAL_LEASE_TYPE,
+                   "Invalid lease type specified");
+            break;
+    }
+out:
+    return grant;
+}
+
+static void
+do_blocked_fops(xlator_t *this, lease_inode_ctx_t *lease_ctx)
+{
+    struct list_head wind_list;
+    fop_stub_t *blk_fop = NULL;
+    fop_stub_t *tmp = NULL;
+
+    INIT_LIST_HEAD(&wind_list);
+
+    pthread_mutex_lock(&lease_ctx->lock);
+    {
+        if (!lease_ctx->blocked_fops_resuming) {
+            /* lease_ctx->blocked_fops_resuming will be set
+             * only when the last lease is released. That
+             * is when we need to resume blocked fops and unref
+             * the inode taken in __add_lease (when lease_cnt == 1).
+             * Return otherwise.
+             */
+            pthread_mutex_unlock(&lease_ctx->lock);
+            return;
+        }
+
+        list_for_each_entry_safe(blk_fop, tmp, &lease_ctx->blocked_list, list)
+        {
+            list_del_init(&blk_fop->list);
+            list_add_tail(&blk_fop->list, &wind_list);
+        }
+    }
+    pthread_mutex_unlock(&lease_ctx->lock);
+
+    gf_msg_trace(this->name, 0, "Executing the blocked stubs on gfid(%s)",
+                 uuid_utoa(lease_ctx->inode->gfid));
+    list_for_each_entry_safe(blk_fop, tmp, &wind_list, list)
+    {
+        list_del_init(&blk_fop->list);
+        gf_msg_trace(this->name, 0, "Executing fop:%d", blk_fop->stub->fop);
+        call_resume(blk_fop->stub);
+        GF_FREE(blk_fop);
+    }
+
+    pthread_mutex_lock(&lease_ctx->lock);
+    {
+        lease_ctx->lease_type = NONE;
+        /* unref the inode taken in __add_lease
+         * (when lease_cnt == 1) */
+        lease_ctx->blocked_fops_resuming = _gf_false;
+        inode_unref(lease_ctx->inode);
+        lease_ctx->inode = NULL;
+    }
+    pthread_mutex_unlock(&lease_ctx->lock);
+
+    return;
+}
+
+void
+recall_lease_timer_handler(struct gf_tw_timer_list *timer, void *data,
+                           unsigned long calltime)
+{
+    inode_t *inode = NULL;
+    lease_inode_t *lease_inode = NULL;
+    leases_private_t *priv = NULL;
+    lease_timer_data_t *timer_data = NULL;
+
+    timer_data = data;
+
+    priv = timer_data->this->private;
+    inode = timer_data->inode;
+    lease_inode = new_lease_inode(inode);
+    if (!lease_inode) {
+        errno = ENOMEM;
+        goto out;
+    }
+    pthread_mutex_lock(&priv->mutex);
+    {
+        list_add_tail(&lease_inode->list, &priv->recall_list);
+        pthread_cond_broadcast(&priv->cond);
+    }
+    pthread_mutex_unlock(&priv->mutex);
+out:
+    /* unref the inode_ref taken by timer_data in __recall_lease */
+    inode_unref(timer_data->inode);
+
+    GF_FREE(timer);
+}
+
+static void
+__recall_lease(xlator_t *this, lease_inode_ctx_t *lease_ctx)
+{
+    lease_id_entry_t *lease_entry = NULL;
+    lease_id_entry_t *tmp = NULL;
+    struct gf_upcall up_req = {
+        0,
+    };
+    struct gf_upcall_recall_lease recall_req = {
+        0,
+    };
+    int notify_ret = -1;
+    struct gf_tw_timer_list *timer = NULL;
+    leases_private_t *priv = NULL;
+    lease_timer_data_t *timer_data = NULL;
+    time_t recall_time;
+
+    if (lease_ctx->recall_in_progress) {
+        gf_msg_debug(this->name, 0,
+                     "Lease recall is already in "
+                     "progress, hence not sending another recall");
+        goto out;
+    }
+
+    priv = this->private;
+    recall_time = gf_time();
+    list_for_each_entry_safe(lease_entry, tmp, &lease_ctx->lease_id_list,
+                             lease_id_list)
+    {
+        gf_uuid_copy(up_req.gfid, lease_ctx->inode->gfid);
+        up_req.client_uid = lease_entry->client_uid;
+        up_req.event_type = GF_UPCALL_RECALL_LEASE;
+        up_req.data = &recall_req;
+
+        notify_ret = this->notify(this, GF_EVENT_UPCALL, &up_req);
+        if (notify_ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, LEASE_MSG_RECALL_FAIL,
+                   "Recall notification to client: %s failed",
+                   lease_entry->client_uid);
+            /* Do not return from here, continue registering the timer,
+               this is required mostly o keep replicas in sync*/
+        } else {
+            gf_msg_debug(this->name, 0,
+                         "Recall lease (all)"
+                         "notification sent to client %s",
+                         lease_entry->client_uid);
+        }
+
+        lease_ctx->recall_in_progress = _gf_true;
+        lease_entry->recall_time = recall_time;
+    }
+    timer = GF_MALLOC(sizeof(*timer), gf_common_mt_tw_timer_list);
+    if (!timer) {
+        goto out;
+    }
+    timer_data = GF_MALLOC(sizeof(lease_timer_data_t),
+                           gf_leases_mt_timer_data_t);
+    if (!timer_data) {
+        GF_FREE(timer);
+        goto out;
+    }
+
+    timer_data->inode = inode_ref(lease_ctx->inode);
+    timer_data->this = this;
+    timer->data = timer_data;
+
+    INIT_LIST_HEAD(&timer->entry);
+    timer->expires = get_recall_lease_timeout(this);
+    timer->function = recall_lease_timer_handler;
+    lease_ctx->timer = timer;
+    gf_tw_add_timer(priv->timer_wheel, timer);
+    gf_msg_trace(this->name, 0,
+                 "Registering timer "
+                 "%p, after "
+                 "sending recall",
+                 timer);
+out:
+    return;
+}
+
+/* ret = 0; STACK_UNWIND Success
+ * ret = -1; STACK_UNWIND failure
+ */
+int
+process_lease_req(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                  struct gf_lease *lease)
+{
+    int ret = 0;
+    char *client_uid = NULL;
+    lease_inode_ctx_t *lease_ctx = NULL;
+
+    GF_VALIDATE_OR_GOTO("leases", frame, out);
+    GF_VALIDATE_OR_GOTO("leases", this, out);
+    GF_VALIDATE_OR_GOTO("leases", inode, out);
+    GF_VALIDATE_OR_GOTO("leases", lease, out);
+
+    client_uid = frame->root->client->client_uid;
+
+    if (!is_valid_lease_id(lease->lease_id)) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, LEASE_MSG_INVAL_LEASE_ID,
+               "Invalid lease id, from"
+               "client:%s",
+               client_uid);
+        ret = -EINVAL;
+        errno = EINVAL;
+        goto out;
+    }
+
+    lease_ctx = lease_ctx_get(inode, this);
+    if (!lease_ctx) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, LEASE_MSG_NO_MEM,
+               "Unable to create/get inode ctx, "
+               "inode:%p",
+               inode);
+        ret = -ENOMEM;
+        errno = ENOMEM;
+        goto out;
+    }
+
+    gf_msg_debug(this->name, 0,
+                 "Lease request from client: %s, "
+                 "lease type:%d, lease cmd:%d, lease ID:%s, gfid:%s",
+                 client_uid, lease->lease_type, lease->cmd,
+                 leaseid_utoa(lease->lease_id), uuid_utoa(inode->gfid));
+
+    pthread_mutex_lock(&lease_ctx->lock);
+    {
+        switch (lease->cmd) {
+            case GF_GET_LEASE:
+                lease->lease_type = lease_ctx->lease_type;
+                gf_msg_debug(this->name, 0,
+                             "Get lease, existing lease"
+                             "type: %d",
+                             lease_ctx->lease_type);
+                /*TODO:Should it consider lease id or client_uid?*/
+                break;
+
+            case GF_SET_LEASE:
+                if (__is_lease_grantable(this, lease_ctx, lease, inode)) {
+                    __add_lease(frame, inode, lease_ctx, client_uid, lease);
+                    ret = 0;
+                } else {
+                    gf_msg_debug(this->name, GF_LOG_DEBUG,
+                                 "Not granting the conflicting lease"
+                                 " request from %s on gfid(%s)",
+                                 client_uid, uuid_utoa(inode->gfid));
+                    __recall_lease(this, lease_ctx);
+                    ret = -1;
+                }
+                break;
+            case GF_UNLK_LEASE:
+                ret = __remove_lease(this, inode, lease_ctx, client_uid, lease);
+                if ((ret >= 0) && (lease_ctx->lease_cnt == 0)) {
+                    pthread_mutex_unlock(&lease_ctx->lock);
+                    goto unblock;
+                }
+                break;
+            default:
+                ret = -EINVAL;
+                break;
+        }
+    }
+    pthread_mutex_unlock(&lease_ctx->lock);
+
+    return ret;
+
+unblock:
+    do_blocked_fops(this, lease_ctx);
+out:
+    return ret;
+}
+
+/* ret = 1 conflict
+ * ret = 0 no conflict
+ */
+gf_boolean_t
+__check_lease_conflict(call_frame_t *frame, lease_inode_ctx_t *lease_ctx,
+                       const char *lease_id, gf_boolean_t is_write)
+{
+    gf_lease_types_t lease_type = {
+        0,
+    };
+    gf_boolean_t conflicts = _gf_false;
+    lease_id_entry_t *lease_entry = NULL;
+
+    GF_VALIDATE_OR_GOTO("leases", frame, out);
+    GF_VALIDATE_OR_GOTO("leases", lease_ctx, out);
+
+    lease_type = lease_ctx->lease_type;
+
+    /* If the fop is rename or unlink conflict the lease even if its
+     * from the same client??
+     */
+    if ((frame->root->op == GF_FOP_RENAME) ||
+        (frame->root->op == GF_FOP_UNLINK)) {
+        conflicts = _gf_true;
+        goto recall;
+    }
+
+    /* As internal fops are used to maintain data integrity but do not
+     * make modififications to the client data, no need to conflict with
+     * them.
+     *
+     * @todo: like for locks, even lease state has to be handled by
+     * rebalance or self-heal daemon process. */
+    if (frame->root->pid < 0) {
+        conflicts = _gf_false;
+        goto recall;
+    }
+
+    /* If lease_id is not sent, set conflicts = true if there is
+     * an existing lease */
+    if (!lease_id && (lease_ctx->lease_cnt > 0)) {
+        conflicts = _gf_true;
+        goto recall;
+    }
+
+    switch (lease_type) {
+        case (GF_RW_LEASE | GF_RD_LEASE):
+        case GF_RW_LEASE:
+            lease_entry = __get_lease_id_entry(lease_ctx, lease_id);
+            if (lease_entry && (lease_entry->lease_type & GF_RW_LEASE))
+                conflicts = _gf_false;
+            else
+                conflicts = _gf_true;
+            break;
+        case GF_RD_LEASE:
+            if (is_write && __another_lease_found(lease_ctx, lease_id))
+                conflicts = _gf_true;
+            else
+                conflicts = _gf_false;
+            break;
+        default:
+            break;
+    }
+
+recall:
+    /* If there is a conflict found and recall is not already sent to all
+     * the clients, then send recall to each of the client holding lease.
+     */
+    if (conflicts)
+        __recall_lease(frame->this, lease_ctx);
+out:
+    return conflicts;
+}
+
+/* Return values:
+ * -1 : error, unwind the fop
+ * WIND_FOP: No conflict, wind the fop
+ * BLOCK_FOP: Found a conflicting lease, block the fop
+ */
+int
+check_lease_conflict(call_frame_t *frame, inode_t *inode, const char *lease_id,
+                     uint32_t fop_flags)
+{
+    lease_inode_ctx_t *lease_ctx = NULL;
+    gf_boolean_t is_blocking_fop = _gf_false;
+    gf_boolean_t is_write_fop = _gf_false;
+    gf_boolean_t conflicts = _gf_false;
+    int ret = WIND_FOP;
+
+    lease_ctx = lease_ctx_get(inode, frame->this);
+    if (!lease_ctx) {
+        gf_msg(frame->this->name, GF_LOG_WARNING, ENOMEM, LEASE_MSG_NO_MEM,
+               "Unable to create/get inode ctx");
+        ret = -1;
+        errno = ENOMEM;
+        goto out;
+    }
+
+    is_blocking_fop = ((fop_flags & BLOCKING_FOP) != 0);
+    is_write_fop = ((fop_flags & DATA_MODIFY_FOP) != 0);
+
+    pthread_mutex_lock(&lease_ctx->lock);
+    {
+        if (lease_ctx->lease_type == NONE) {
+            pthread_mutex_unlock(&lease_ctx->lock);
+            gf_msg_debug(frame->this->name, 0,
+                         "No leases found continuing with the"
+                         " fop:%s",
+                         gf_fop_list[frame->root->op]);
+            ret = WIND_FOP;
+            goto out;
+        }
+        conflicts = __check_lease_conflict(frame, lease_ctx, lease_id,
+                                           is_write_fop);
+        if (conflicts) {
+            if (is_blocking_fop) {
+                gf_msg_debug(frame->this->name, 0,
+                             "Fop: %s "
+                             "conflicting existing "
+                             "lease: %d, blocking the"
+                             "fop",
+                             gf_fop_list[frame->root->op],
+                             lease_ctx->lease_type);
+                ret = BLOCK_FOP;
+            } else {
+                gf_msg_debug(frame->this->name, 0,
+                             "Fop: %s "
+                             "conflicting existing "
+                             "lease: %d, sending "
+                             "EAGAIN",
+                             gf_fop_list[frame->root->op],
+                             lease_ctx->lease_type);
+                errno = EAGAIN;
+                ret = -1;
+            }
+        }
+    }
+    pthread_mutex_unlock(&lease_ctx->lock);
+out:
+    return ret;
+}
+
+static int
+remove_clnt_leases(const char *client_uid, inode_t *inode, xlator_t *this)
+{
+    lease_inode_ctx_t *lease_ctx = NULL;
+    lease_id_entry_t *lease_entry = NULL;
+    lease_id_entry_t *tmp = NULL;
+    int ret = 0;
+    int i = 0;
+
+    lease_ctx = lease_ctx_get(inode, this);
+    if (!lease_ctx) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, LEASE_MSG_INVAL_INODE_CTX,
+               "Unable to create/get inode ctx");
+        ret = -1;
+        errno = ENOMEM;
+        goto out;
+    }
+
+    pthread_mutex_lock(&lease_ctx->lock);
+    {
+        list_for_each_entry_safe(lease_entry, tmp, &lease_ctx->lease_id_list,
+                                 lease_id_list)
+        {
+            if (strcmp(client_uid, lease_entry->client_uid) == 0) {
+                for (i = 0; i < GF_LEASE_MAX_TYPE; i++) {
+                    lease_ctx->lease_type_cnt[i] -= lease_entry
+                                                        ->lease_type_cnt[i];
+                }
+                lease_ctx->lease_cnt -= lease_entry->lease_cnt;
+                __destroy_lease_id_entry(lease_entry);
+                if (lease_ctx->lease_cnt == 0) {
+                    lease_ctx->blocked_fops_resuming = _gf_true;
+                    pthread_mutex_unlock(&lease_ctx->lock);
+                    goto unblock;
+                }
+            }
+        }
+    }
+    pthread_mutex_unlock(&lease_ctx->lock);
+out:
+    return ret;
+
+unblock:
+    do_blocked_fops(this, lease_ctx);
+    return ret;
+}
+
+int
+cleanup_client_leases(xlator_t *this, const char *client_uid)
+{
+    lease_client_t *clnt = NULL;
+    lease_client_t *tmp = NULL;
+    struct list_head cleanup_list = {
+        0,
+    };
+    lease_inode_t *l_inode = NULL;
+    lease_inode_t *tmp1 = NULL;
+    leases_private_t *priv = NULL;
+    int ret = 0;
+
+    priv = this->private;
+    if (!priv) {
+        ret = -1;
+        errno = EINVAL;
+        goto out;
+    }
+
+    INIT_LIST_HEAD(&cleanup_list);
+    pthread_mutex_lock(&priv->mutex);
+    {
+        list_for_each_entry_safe(clnt, tmp, &priv->client_list, client_list)
+        {
+            if ((strcmp(clnt->client_uid, client_uid) == 0)) {
+                list_for_each_entry_safe(l_inode, tmp1, &clnt->inode_list, list)
+                {
+                    list_del_init(&l_inode->list);
+                    list_add_tail(&l_inode->list, &cleanup_list);
+                }
+                __destroy_lease_client(clnt);
+                break;
+            }
+        }
+    }
+    pthread_mutex_unlock(&priv->mutex);
+
+    l_inode = tmp1 = NULL;
+    list_for_each_entry_safe(l_inode, tmp1, &cleanup_list, list)
+    {
+        remove_clnt_leases(client_uid, l_inode->inode, this);
+        __destroy_lease_inode(l_inode);
+    }
+out:
+    return ret;
+}
+
+static void
+__remove_all_leases(xlator_t *this, lease_inode_ctx_t *lease_ctx)
+{
+    int i = 0;
+    lease_id_entry_t *lease_entry = NULL;
+    lease_id_entry_t *tmp = NULL;
+
+    if (lease_ctx->lease_cnt == 0) {
+        /* No leases to remove. Return */
+        return;
+    }
+    __dump_leases_info(this, lease_ctx);
+
+    list_for_each_entry_safe(lease_entry, tmp, &lease_ctx->lease_id_list,
+                             lease_id_list)
+    {
+        lease_entry->lease_cnt = 0;
+        remove_from_clnt_list(this, lease_entry->client_uid, lease_ctx->inode);
+        __destroy_lease_id_entry(lease_entry);
+    }
+    INIT_LIST_HEAD(&lease_ctx->lease_id_list);
+    for (i = 0; i <= GF_LEASE_MAX_TYPE; i++)
+        lease_ctx->lease_type_cnt[i] = 0;
+    lease_ctx->lease_type = 0;
+    lease_ctx->lease_cnt = 0;
+    lease_ctx->recall_in_progress = _gf_false;
+    lease_ctx->timer = NULL;
+    lease_ctx->blocked_fops_resuming = _gf_true;
+
+    /* TODO:
+     * - Mark the corresponding fd bad. Could be done on client side
+     * as a result of recall
+     * - Free the lease_ctx
+     */
+    return;
+}
+
+static int
+remove_all_leases(xlator_t *this, inode_t *inode)
+{
+    lease_inode_ctx_t *lease_ctx = NULL;
+    int ret = 0;
+
+    GF_VALIDATE_OR_GOTO("leases", inode, out);
+
+    lease_ctx = lease_ctx_get(inode, this);
+    if (!lease_ctx) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, LEASE_MSG_INVAL_INODE_CTX,
+               "Unable to create/get inode ctx");
+        ret = -1;
+        errno = ENOMEM;
+        goto out;
+    }
+
+    pthread_mutex_lock(&lease_ctx->lock);
+    {
+        __remove_all_leases(this, lease_ctx);
+    }
+    pthread_mutex_unlock(&lease_ctx->lock);
+
+    do_blocked_fops(this, lease_ctx);
+out:
+    return ret;
+}
+
+void *
+expired_recall_cleanup(void *data)
+{
+    struct timespec sleep_till = {
+        0,
+    };
+    struct list_head recall_cleanup_list;
+    lease_inode_t *recall_entry = NULL;
+    lease_inode_t *tmp = NULL;
+    leases_private_t *priv = NULL;
+    xlator_t *this = NULL;
+    time_t time_now;
+
+    GF_VALIDATE_OR_GOTO("leases", data, out);
+
+    this = data;
+    priv = this->private;
+
+    gf_msg_debug(this->name, 0, "Started the expired_recall_cleanup thread");
+
+    while (1) {
+        time_now = gf_time();
+        pthread_mutex_lock(&priv->mutex);
+        {
+            if (priv->fini) {
+                pthread_mutex_unlock(&priv->mutex);
+                goto out;
+            }
+            INIT_LIST_HEAD(&recall_cleanup_list);
+            if (list_empty(&priv->recall_list)) {
+                sleep_till.tv_sec = time_now + 600;
+                pthread_cond_timedwait(&priv->cond, &priv->mutex, &sleep_till);
+            }
+            if (!list_empty(&priv->recall_list)) {
+                gf_msg_debug(this->name, 0, "Found expired recalls");
+                list_for_each_entry_safe(recall_entry, tmp, &priv->recall_list,
+                                         list)
+                {
+                    list_del_init(&recall_entry->list);
+                    list_add_tail(&recall_entry->list, &recall_cleanup_list);
+                }
+            }
+        }
+        pthread_mutex_unlock(&priv->mutex);
+
+        recall_entry = tmp = NULL;
+        list_for_each_entry_safe(recall_entry, tmp, &recall_cleanup_list, list)
+        {
+            gf_msg_debug(this->name, 0,
+                         "Recall lease was sent on"
+                         " inode:%p, recall timer has expired"
+                         " and clients haven't unlocked the lease"
+                         " hence cleaning up leases on the inode",
+                         recall_entry->inode);
+            remove_all_leases(this, recall_entry->inode);
+            /* no need to take priv->mutex lock as this entry
+             * reference is removed from global recall list. */
+            __destroy_lease_inode(recall_entry);
+        }
+    }
+
+out:
+    return NULL;
+}
diff --git a/xlators/features/leases/src/leases-mem-types.h b/xlators/features/leases/src/leases-mem-types.h
new file mode 100644
index 00000000000..25664b44156
--- /dev/null
+++ b/xlators/features/leases/src/leases-mem-types.h
@@ -0,0 +1,27 @@
+/*
+   Copyright (c) 2015-2016 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __LEASES_MEM_TYPES_H__
+#define __LEASES_MEM_TYPES_H__
+
+#include <glusterfs/mem-types.h>
+
+enum gf_leases_mem_types_ {
+    gf_leases_mt_private_t = gf_common_mt_end + 1,
+    gf_leases_mt_lease_client_t,
+    gf_leases_mt_lease_inode_t,
+    gf_leases_mt_fd_ctx_t,
+    gf_leases_mt_lease_inode_ctx_t,
+    gf_leases_mt_lease_id_entry_t,
+    gf_leases_mt_fop_stub_t,
+    gf_leases_mt_timer_data_t,
+    gf_leases_mt_end
+};
+#endif
diff --git a/xlators/features/leases/src/leases-messages.h b/xlators/features/leases/src/leases-messages.h
new file mode 100644
index 00000000000..da696b832de
--- /dev/null
+++ b/xlators/features/leases/src/leases-messages.h
@@ -0,0 +1,33 @@
+/*
+ Copyright (c) 2015-2016 Red Hat, Inc. <http://www.redhat.com>
+ This file is part of GlusterFS.
+
+ This file is licensed to you under your choice of the GNU Lesser
+ General Public License, version 3 or any later version (LGPLv3 or
+ later), or the GNU General Public License, version 2 (GPLv2), in all
+ cases as published by the Free Software Foundation.
+ */
+
+#ifndef _LEASES_MESSAGES_H_
+#define _LEASES_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(LEASES, LEASE_MSG_NO_MEM, LEASE_MSG_RECALL_FAIL,
+           LEASE_MSG_INVAL_LEASE_ID, LEASE_MSG_INVAL_UNLK_LEASE,
+           LEASE_MSG_INVAL_INODE_CTX, LEASE_MSG_NOT_ENABLED,
+           LEASE_MSG_NO_TIMER_WHEEL, LEASE_MSG_CLNT_NOTFOUND,
+           LEASE_MSG_INODE_NOTFOUND, LEASE_MSG_INVAL_FD_CTX,
+           LEASE_MSG_INVAL_LEASE_TYPE);
+
+#endif /* !_LEASES_MESSAGES_H_ */
diff --git a/xlators/features/leases/src/leases.c b/xlators/features/leases/src/leases.c
new file mode 100644
index 00000000000..04bee50ba3f
--- /dev/null
+++ b/xlators/features/leases/src/leases.c
@@ -0,0 +1,1168 @@
+/*
+   Copyright (c) 2015-2016 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "leases.h"
+
+int32_t
+leases_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(open, frame, op_ret, op_errno, fd, xdata);
+
+    return 0;
+}
+
+int32_t
+leases_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+            fd_t *fd, dict_t *xdata)
+{
+    uint32_t fop_flags = 0;
+    int32_t op_errno = EINVAL;
+    int ret = 0;
+    lease_fd_ctx_t *fd_ctx = NULL;
+    char *lease_id = NULL;
+
+    EXIT_IF_LEASES_OFF(this, out);
+    EXIT_IF_INTERNAL_FOP(frame, xdata, out);
+
+    fd_ctx = GF_CALLOC(1, sizeof(*fd_ctx), gf_leases_mt_fd_ctx_t);
+    if (!fd_ctx) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    fd_ctx->client_uid = gf_strdup(frame->root->client->client_uid);
+    if (!fd_ctx->client_uid) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    GET_FLAGS(frame->root->op, flags);
+    GET_LEASE_ID(xdata, lease_id, frame->root->client->client_uid);
+    if (lease_id != NULL)
+        memcpy(fd_ctx->lease_id, lease_id, LEASE_ID_SIZE);
+    else
+        memset(fd_ctx->lease_id, 0, LEASE_ID_SIZE);
+
+    ret = fd_ctx_set(fd, this, (uint64_t)(uintptr_t)fd_ctx);
+    if (ret) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    ret = check_lease_conflict(frame, fd->inode, lease_id, fop_flags);
+    if (ret < 0)
+        goto err;
+    else if (ret == BLOCK_FOP)
+        goto block;
+    else if (ret == WIND_FOP)
+        goto out;
+
+block:
+    LEASE_BLOCK_FOP(fd->inode, open, frame, this, loc, flags, fd, xdata);
+    return 0;
+
+out:
+    STACK_WIND(frame, leases_open_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->open, loc, flags, fd, xdata);
+    return 0;
+
+err:
+    if (fd_ctx) {
+        GF_FREE(fd_ctx->client_uid);
+        GF_FREE(fd_ctx);
+    }
+
+    STACK_UNWIND_STRICT(open, frame, -1, op_errno, NULL, NULL);
+    return 0;
+}
+
+int32_t
+leases_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                  int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+                  dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(writev, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+
+    return 0;
+}
+
+int32_t
+leases_writev(call_frame_t *frame, xlator_t *this, fd_t *fd,
+              struct iovec *vector, int count, off_t off, uint32_t flags,
+              struct iobref *iobref, dict_t *xdata)
+{
+    uint32_t fop_flags = 0;
+    char *lease_id = NULL;
+    int ret = 0;
+
+    EXIT_IF_LEASES_OFF(this, out);
+    EXIT_IF_INTERNAL_FOP(frame, xdata, out);
+
+    GET_LEASE_ID(xdata, lease_id, frame->root->client->client_uid);
+    GET_FLAGS(frame->root->op, fd->flags);
+
+    ret = check_lease_conflict(frame, fd->inode, lease_id, fop_flags);
+    if (ret < 0)
+        goto err;
+    else if (ret == BLOCK_FOP)
+        goto block;
+    else if (ret == WIND_FOP)
+        goto out;
+
+block:
+    LEASE_BLOCK_FOP(fd->inode, writev, frame, this, fd, vector, count, off,
+                    flags, iobref, xdata);
+    return 0;
+
+out:
+    STACK_WIND(frame, leases_writev_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->writev, fd, vector, count, off, flags,
+               iobref, xdata);
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(writev, frame, -1, errno, NULL, NULL, NULL);
+    return 0;
+}
+
+int32_t
+leases_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, struct iovec *vector, int count,
+                 struct iatt *stbuf, struct iobref *iobref, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(readv, frame, op_ret, op_errno, vector, count, stbuf,
+                        iobref, xdata);
+
+    return 0;
+}
+
+int32_t
+leases_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+             off_t offset, uint32_t flags, dict_t *xdata)
+{
+    uint32_t fop_flags = 0;
+    char *lease_id = NULL;
+    int ret = 0;
+
+    EXIT_IF_LEASES_OFF(this, out);
+    EXIT_IF_INTERNAL_FOP(frame, xdata, out);
+
+    GET_LEASE_ID(xdata, lease_id, frame->root->client->client_uid);
+    GET_FLAGS(frame->root->op, fd->flags);
+
+    ret = check_lease_conflict(frame, fd->inode, lease_id, fop_flags);
+    if (ret < 0)
+        goto err;
+    else if (ret == BLOCK_FOP)
+        goto block;
+    else if (ret == WIND_FOP)
+        goto out;
+
+block:
+    LEASE_BLOCK_FOP(fd->inode, readv, frame, this, fd, size, offset, flags,
+                    xdata);
+    return 0;
+
+out:
+    STACK_WIND(frame, leases_readv_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readv, fd, size, offset, flags, xdata);
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(readv, frame, -1, errno, NULL, 0, NULL, NULL, NULL);
+    return 0;
+}
+
+int32_t
+leases_lk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct gf_flock *lock, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(lk, frame, op_ret, op_errno, lock, xdata);
+
+    return 0;
+}
+
+int32_t
+leases_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+          struct gf_flock *flock, dict_t *xdata)
+{
+    uint32_t fop_flags = 0;
+    char *lease_id = NULL;
+    int ret = 0;
+
+    EXIT_IF_LEASES_OFF(this, out);
+    EXIT_IF_INTERNAL_FOP(frame, xdata, out);
+
+    GET_LEASE_ID(xdata, lease_id, frame->root->client->client_uid);
+    GET_FLAGS_LK(cmd, flock->l_type, fd->flags);
+
+    ret = check_lease_conflict(frame, fd->inode, lease_id, fop_flags);
+    if (ret < 0)
+        goto err;
+    else if (ret == BLOCK_FOP)
+        goto block;
+    else if (ret == WIND_FOP)
+        goto out;
+
+block:
+    LEASE_BLOCK_FOP(fd->inode, lk, frame, this, fd, cmd, flock, xdata);
+    return 0;
+
+out:
+    STACK_WIND(frame, leases_lk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lk, fd, cmd, flock, xdata);
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(lk, frame, -1, errno, NULL, NULL);
+    return 0;
+}
+
+int32_t
+leases_lease(call_frame_t *frame, xlator_t *this, loc_t *loc,
+             struct gf_lease *lease, dict_t *xdata)
+{
+    int32_t op_errno = 0;
+    int ret = 0;
+    struct gf_lease nullease = {
+        0,
+    };
+    int32_t op_ret = 0;
+
+    EXIT_IF_LEASES_OFF(this, out);
+    EXIT_IF_INTERNAL_FOP(frame, xdata, out);
+
+    ret = process_lease_req(frame, this, loc->inode, lease);
+    if (ret < 0) {
+        op_errno = -ret;
+        op_ret = -1;
+    }
+    goto unwind;
+
+out:
+    gf_msg(this->name, GF_LOG_ERROR, EINVAL, LEASE_MSG_NOT_ENABLED,
+           "\"features/leases\" translator is not enabled. "
+           "You need to enable it for proper functioning of your "
+           "application");
+    op_errno = ENOSYS;
+    op_ret = -1;
+
+unwind:
+    STACK_UNWIND_STRICT(lease, frame, op_ret, op_errno,
+                        (op_errno == ENOSYS) ? &nullease : lease, xdata);
+    return 0;
+}
+
+int32_t
+leases_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int op_ret, int op_errno, struct iatt *prebuf,
+                    struct iatt *postbuf, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(truncate, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+
+    return 0;
+}
+
+int32_t
+leases_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+                dict_t *xdata)
+{
+    uint32_t fop_flags = 0;
+    char *lease_id = NULL;
+    int ret = 0;
+
+    EXIT_IF_LEASES_OFF(this, out);
+    EXIT_IF_INTERNAL_FOP(frame, xdata, out);
+
+    GET_LEASE_ID(xdata, lease_id, frame->root->client->client_uid);
+    GET_FLAGS(frame->root->op, 0);
+
+    ret = check_lease_conflict(frame, loc->inode, lease_id, fop_flags);
+    if (ret < 0)
+        goto err;
+    else if (ret == BLOCK_FOP)
+        goto block;
+    else if (ret == WIND_FOP)
+        goto out;
+
+block:
+    LEASE_BLOCK_FOP(loc->inode, truncate, frame, this, loc, offset, xdata);
+    return 0;
+
+out:
+    STACK_WIND(frame, leases_truncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(truncate, frame, -1, errno, NULL, NULL, NULL);
+    return 0;
+}
+
+int32_t
+leases_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int op_ret, int op_errno, struct iatt *statpre,
+                   struct iatt *statpost, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(setattr, frame, op_ret, op_errno, statpre, statpost,
+                        xdata);
+
+    return 0;
+}
+
+int32_t
+leases_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    uint32_t fop_flags = 0;
+    char *lease_id = NULL;
+    int ret = 0;
+
+    EXIT_IF_LEASES_OFF(this, out);
+    EXIT_IF_INTERNAL_FOP(frame, xdata, out);
+
+    GET_LEASE_ID(xdata, lease_id, frame->root->client->client_uid);
+    GET_FLAGS(frame->root->op, 0);
+
+    ret = check_lease_conflict(frame, loc->inode, lease_id, fop_flags);
+    if (ret < 0)
+        goto err;
+    else if (ret == BLOCK_FOP)
+        goto block;
+    else if (ret == WIND_FOP)
+        goto out;
+
+block:
+    LEASE_BLOCK_FOP(loc->inode, setattr, frame, this, loc, stbuf, valid, xdata);
+    return 0;
+
+out:
+    STACK_WIND(frame, leases_setattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setattr, loc, stbuf, valid, xdata);
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(setattr, frame, -1, errno, NULL, NULL, NULL);
+    return 0;
+}
+
+int32_t
+leases_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct iatt *stbuf,
+                  struct iatt *preoldparent, struct iatt *postoldparent,
+                  struct iatt *prenewparent, struct iatt *postnewparent,
+                  dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(rename, frame, op_ret, op_errno, stbuf, preoldparent,
+                        postoldparent, prenewparent, postnewparent, xdata);
+
+    return 0;
+}
+
+int32_t
+leases_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+              dict_t *xdata)
+{
+    uint32_t fop_flags = 0;
+    char *lease_id = NULL;
+    int ret = 0;
+
+    EXIT_IF_LEASES_OFF(this, out);
+    EXIT_IF_INTERNAL_FOP(frame, xdata, out);
+
+    /* should the lease be also checked for newloc */
+    GET_LEASE_ID(xdata, lease_id, frame->root->client->client_uid);
+    GET_FLAGS(frame->root->op, 0);
+
+    ret = check_lease_conflict(frame, oldloc->inode, lease_id, fop_flags);
+    if (ret < 0)
+        goto err;
+    else if (ret == BLOCK_FOP)
+        goto block;
+    else if (ret == WIND_FOP)
+        goto out;
+
+block:
+    LEASE_BLOCK_FOP(oldloc->inode, rename, frame, this, oldloc, newloc, xdata);
+    return 0;
+
+out:
+    STACK_WIND(frame, leases_rename_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(rename, frame, -1, errno, NULL, NULL, NULL, NULL, NULL,
+                        NULL);
+    return 0;
+}
+
+int32_t
+leases_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                  int op_errno, struct iatt *preparent, struct iatt *postparent,
+                  dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(unlink, frame, op_ret, op_errno, preparent, postparent,
+                        xdata);
+
+    return 0;
+}
+
+int32_t
+leases_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+              dict_t *xdata)
+{
+    uint32_t fop_flags = 0;
+    char *lease_id = NULL;
+    int ret = 0;
+
+    EXIT_IF_LEASES_OFF(this, out);
+    EXIT_IF_INTERNAL_FOP(frame, xdata, out);
+
+    GET_LEASE_ID(xdata, lease_id, frame->root->client->client_uid);
+    GET_FLAGS(frame->root->op, 0);
+
+    ret = check_lease_conflict(frame, loc->inode, lease_id, fop_flags);
+    if (ret < 0)
+        goto err;
+    else if (ret == BLOCK_FOP)
+        goto block;
+    else if (ret == WIND_FOP)
+        goto out;
+
+block:
+    LEASE_BLOCK_FOP(loc->inode, unlink, frame, this, loc, xflag, xdata);
+    return 0;
+
+out:
+    STACK_WIND(frame, leases_unlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->unlink, loc, xflag, xdata);
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(unlink, frame, -1, errno, NULL, NULL, NULL);
+    return 0;
+}
+
+int32_t
+leases_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                int op_errno, inode_t *inode, struct iatt *stbuf,
+                struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(link, frame, op_ret, op_errno, inode, stbuf, preparent,
+                        postparent, xdata);
+
+    return 0;
+}
+
+int32_t
+leases_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+            dict_t *xdata)
+{
+    uint32_t fop_flags = 0;
+    char *lease_id = NULL;
+    int ret = 0;
+
+    EXIT_IF_LEASES_OFF(this, out);
+    EXIT_IF_INTERNAL_FOP(frame, xdata, out);
+
+    GET_LEASE_ID(xdata, lease_id, frame->root->client->client_uid);
+    GET_FLAGS(frame->root->op, 0);
+
+    ret = check_lease_conflict(frame, oldloc->inode, lease_id, fop_flags);
+    if (ret < 0)
+        goto err;
+    else if (ret == BLOCK_FOP)
+        goto block;
+    else if (ret == WIND_FOP)
+        goto out;
+
+block:
+    LEASE_BLOCK_FOP(oldloc->inode, link, frame, this, oldloc, newloc, xdata);
+    return 0;
+out:
+    STACK_WIND(frame, leases_link_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->link, oldloc, newloc, xdata);
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(link, frame, -1, errno, NULL, NULL, NULL, NULL, NULL);
+    return 0;
+}
+
+int32_t
+leases_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                  int op_errno, fd_t *fd, inode_t *inode, struct iatt *stbuf,
+                  struct iatt *preparent, struct iatt *postparent,
+                  dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(create, frame, op_ret, op_errno, fd, inode, stbuf,
+                        preparent, postparent, xdata);
+
+    return 0;
+}
+
+int32_t
+leases_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+              mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+    uint32_t fop_flags = 0;
+    char *lease_id = NULL;
+    int ret = 0;
+
+    EXIT_IF_LEASES_OFF(this, out);
+    EXIT_IF_INTERNAL_FOP(frame, xdata, out);
+
+    GET_LEASE_ID(xdata, lease_id, frame->root->client->client_uid);
+    GET_FLAGS(frame->root->op, flags);
+
+    ret = check_lease_conflict(frame, fd->inode, lease_id, fop_flags);
+    if (ret < 0)
+        goto err;
+    else if (ret == BLOCK_FOP)
+        goto block;
+    else if (ret == WIND_FOP)
+        goto out;
+
+block:
+    LEASE_BLOCK_FOP(fd->inode, create, frame, this, loc, flags, mode, umask, fd,
+                    xdata);
+    return 0;
+
+out:
+    STACK_WIND(frame, leases_create_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->create, loc, flags, mode, umask, fd,
+               xdata);
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(create, frame, -1, errno, NULL, NULL, NULL, NULL, NULL,
+                        NULL);
+    return 0;
+}
+
+int32_t
+leases_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                 struct iatt *postbuf, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(fsync, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+    return 0;
+}
+
+int32_t
+leases_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+             dict_t *xdata)
+{
+    uint32_t fop_flags = 0;
+    char *lease_id = NULL;
+    int ret = 0;
+
+    EXIT_IF_LEASES_OFF(this, out);
+    EXIT_IF_INTERNAL_FOP(frame, xdata, out);
+
+    GET_LEASE_ID(xdata, lease_id, frame->root->client->client_uid);
+    GET_FLAGS(frame->root->op, fd->flags);
+
+    ret = check_lease_conflict(frame, fd->inode, lease_id, fop_flags);
+    if (ret < 0)
+        goto err;
+    else if (ret == BLOCK_FOP)
+        goto block;
+    else if (ret == WIND_FOP)
+        goto out;
+
+block:
+    LEASE_BLOCK_FOP(fd->inode, fsync, frame, this, fd, flags, xdata);
+    return 0;
+
+out:
+    STACK_WIND(frame, leases_fsync_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsync, fd, flags, xdata);
+    return 0;
+err:
+    STACK_UNWIND_STRICT(fsync, frame, -1, errno, NULL, NULL, NULL);
+    return 0;
+}
+
+int32_t
+leases_ftruncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(ftruncate, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+    return 0;
+}
+
+int32_t
+leases_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                 dict_t *xdata)
+{
+    uint32_t fop_flags = 0;
+    char *lease_id = NULL;
+    int ret = 0;
+
+    EXIT_IF_LEASES_OFF(this, out);
+    EXIT_IF_INTERNAL_FOP(frame, xdata, out);
+
+    GET_LEASE_ID(xdata, lease_id, frame->root->client->client_uid);
+    GET_FLAGS(frame->root->op, 0); /* TODO:fd->flags?*/
+
+    ret = check_lease_conflict(frame, fd->inode, lease_id, fop_flags);
+    if (ret < 0)
+        goto err;
+    else if (ret == BLOCK_FOP)
+        goto block;
+    else if (ret == WIND_FOP)
+        goto out;
+
+block:
+    LEASE_BLOCK_FOP(fd->inode, ftruncate, frame, this, fd, offset, xdata);
+    return 0;
+
+out:
+    STACK_WIND(frame, leases_ftruncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(ftruncate, frame, -1, errno, NULL, NULL, NULL);
+    return 0;
+}
+
+int32_t
+leases_fsetattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *statpre,
+                    struct iatt *statpost, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(fsetattr, frame, op_ret, op_errno, statpre, statpost,
+                        xdata);
+    return 0;
+}
+
+int32_t
+leases_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    uint32_t fop_flags = 0;
+    char *lease_id = NULL;
+    int ret = 0;
+
+    EXIT_IF_LEASES_OFF(this, out);
+    EXIT_IF_INTERNAL_FOP(frame, xdata, out);
+
+    GET_LEASE_ID(xdata, lease_id, frame->root->client->client_uid);
+    GET_FLAGS(frame->root->op, fd->flags);
+
+    ret = check_lease_conflict(frame, fd->inode, lease_id, fop_flags);
+    if (ret < 0)
+        goto err;
+    else if (ret == BLOCK_FOP)
+        goto block;
+    else if (ret == WIND_FOP)
+        goto out;
+
+block:
+    LEASE_BLOCK_FOP(fd->inode, fsetattr, frame, this, fd, stbuf, valid, xdata);
+    return 0;
+
+out:
+    STACK_WIND(frame, leases_fsetattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetattr, fd, stbuf, valid, xdata);
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(fsetattr, frame, -1, errno, NULL, NULL, NULL);
+    return 0;
+}
+
+int32_t
+leases_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *pre,
+                     struct iatt *post, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(fallocate, frame, op_ret, op_errno, pre, post, xdata);
+
+    return 0;
+}
+
+int32_t
+leases_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+                 off_t offset, size_t len, dict_t *xdata)
+{
+    uint32_t fop_flags = 0;
+    char *lease_id = NULL;
+    int ret = 0;
+
+    EXIT_IF_LEASES_OFF(this, out);
+    EXIT_IF_INTERNAL_FOP(frame, xdata, out);
+
+    GET_LEASE_ID(xdata, lease_id, frame->root->client->client_uid);
+    GET_FLAGS(frame->root->op, fd->flags);
+
+    ret = check_lease_conflict(frame, fd->inode, lease_id, fop_flags);
+    if (ret < 0)
+        goto err;
+    else if (ret == BLOCK_FOP)
+        goto block;
+    else if (ret == WIND_FOP)
+        goto out;
+
+block:
+    LEASE_BLOCK_FOP(fd->inode, fallocate, frame, this, fd, mode, offset, len,
+                    xdata);
+    return 0;
+
+out:
+    STACK_WIND(frame, leases_fallocate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fallocate, fd, mode, offset, len,
+               xdata);
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(fallocate, frame, -1, errno, NULL, NULL, NULL);
+    return 0;
+}
+
+int32_t
+leases_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *pre,
+                   struct iatt *post, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(discard, frame, op_ret, op_errno, pre, post, xdata);
+
+    return 0;
+}
+
+int32_t
+leases_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+               size_t len, dict_t *xdata)
+{
+    uint32_t fop_flags = 0;
+    char *lease_id = NULL;
+    int ret = 0;
+
+    EXIT_IF_LEASES_OFF(this, out);
+    EXIT_IF_INTERNAL_FOP(frame, xdata, out);
+
+    GET_LEASE_ID(xdata, lease_id, frame->root->client->client_uid);
+    GET_FLAGS(frame->root->op, fd->flags);
+
+    ret = check_lease_conflict(frame, fd->inode, lease_id, fop_flags);
+    if (ret < 0)
+        goto err;
+    else if (ret == BLOCK_FOP)
+        goto block;
+    else if (ret == WIND_FOP)
+        goto out;
+
+block:
+    LEASE_BLOCK_FOP(fd->inode, discard, frame, this, fd, offset, len, xdata);
+    return 0;
+
+out:
+    STACK_WIND(frame, leases_discard_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->discard, fd, offset, len, xdata);
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(discard, frame, -1, errno, NULL, NULL, NULL);
+    return 0;
+}
+
+int32_t
+leases_zerofill_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *pre,
+                    struct iatt *post, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(zerofill, frame, op_ret, op_errno, pre, post, xdata);
+
+    return 0;
+}
+
+int
+leases_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                off_t len, dict_t *xdata)
+{
+    uint32_t fop_flags = 0;
+    char *lease_id = NULL;
+    int ret = 0;
+
+    EXIT_IF_LEASES_OFF(this, out);
+    EXIT_IF_INTERNAL_FOP(frame, xdata, out);
+
+    GET_LEASE_ID(xdata, lease_id, frame->root->client->client_uid);
+    GET_FLAGS(frame->root->op, fd->flags);
+
+    ret = check_lease_conflict(frame, fd->inode, lease_id, fop_flags);
+    if (ret < 0)
+        goto err;
+    else if (ret == BLOCK_FOP)
+        goto block;
+    else if (ret == WIND_FOP)
+        goto out;
+
+block:
+    LEASE_BLOCK_FOP(fd->inode, zerofill, frame, this, fd, offset, len, xdata);
+    return 0;
+
+out:
+    STACK_WIND(frame, leases_zerofill_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->zerofill, fd, offset, len, xdata);
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(zerofill, frame, -1, errno, NULL, NULL, NULL);
+    return 0;
+}
+
+int
+leases_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(flush, frame, op_ret, op_errno, xdata);
+
+    return 0;
+}
+
+int
+leases_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    uint32_t fop_flags = 0;
+    char *lease_id = NULL;
+    int ret = 0;
+    lease_fd_ctx_t *fd_ctx = NULL;
+    uint64_t ctx = 0;
+
+    EXIT_IF_LEASES_OFF(this, out);
+    EXIT_IF_INTERNAL_FOP(frame, xdata, out);
+
+    GET_LEASE_ID(xdata, lease_id, frame->root->client->client_uid);
+    GET_FLAGS(frame->root->op, fd->flags);
+
+    ret = check_lease_conflict(frame, fd->inode, lease_id, fop_flags);
+    if (ret < 0)
+        goto err;
+    else if (ret == BLOCK_FOP)
+        goto block;
+    else if (ret == WIND_FOP)
+        goto out;
+
+block:
+    LEASE_BLOCK_FOP(fd->inode, flush, frame, this, fd, xdata);
+    return 0;
+
+out:
+    /* *
+     * currently release is not called after the close fop from the
+     * application. Hence lease fd ctx is reset on here.
+     * This is actually not the right way, since flush can be called
+     * not only from the close op.
+     * TODO :
+     *     - Either identify the flush is called from close call on fd from
+     *     from the application.
+     *                      OR
+     *     - Find why release is not called post the last close call
+     */
+    ret = fd_ctx_get(fd, this, &ctx);
+    if (ret == 0) {
+        fd_ctx = (lease_fd_ctx_t *)(long)ctx;
+        if (fd_ctx->client_uid) {
+            GF_FREE(fd_ctx->client_uid);
+            fd_ctx->client_uid = NULL;
+        }
+        memset(fd_ctx->lease_id, 0, LEASE_ID_SIZE);
+    }
+    STACK_WIND(frame, leases_flush_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->flush, fd, xdata);
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(create, frame, -1, errno, NULL, NULL, NULL, NULL, NULL,
+                        NULL);
+    return 0;
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    if (!this)
+        return ret;
+
+    ret = xlator_mem_acct_init(this, gf_leases_mt_end + 1);
+
+    if (ret != 0) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, LEASE_MSG_NO_MEM,
+               "mem account init failed");
+        return ret;
+    }
+
+    return ret;
+}
+
+static int
+leases_init_priv(xlator_t *this)
+{
+    int ret = 0;
+    leases_private_t *priv = NULL;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    if (!priv->timer_wheel) {
+        priv->timer_wheel = glusterfs_ctx_tw_get(this->ctx);
+        if (!priv->timer_wheel) {
+            ret = -1;
+            goto out;
+        }
+    }
+
+    if (!priv->inited_recall_thr) {
+        ret = gf_thread_create(&priv->recall_thr, NULL, expired_recall_cleanup,
+                               this, "leasercl");
+        if (!ret)
+            priv->inited_recall_thr = _gf_true;
+    }
+
+out:
+    return ret;
+}
+
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    leases_private_t *priv = NULL;
+    int ret = -1;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    /* TODO: In case of reconfigure, if its enabling the leases
+     * its not an issue, but if its disabling the leases, there
+     * is more to it, like recall all the existing leases, wait
+     * for unlock of all the leases etc., hence not supporting the
+     * reconfigure for now.
+
+    GF_OPTION_RECONF ("leases", priv->leases_enabled,
+                      options, bool, out);
+
+    if (priv->leases_enabled) {
+            ret = leases_init_priv (this);
+            if (ret)
+                    goto out;
+    }
+    */
+
+    GF_OPTION_RECONF("lease-lock-recall-timeout", priv->recall_lease_timeout,
+                     options, int32, out);
+
+    ret = 0;
+out:
+    return ret;
+}
+
+int
+init(xlator_t *this)
+{
+    int ret = -1;
+    leases_private_t *priv = NULL;
+
+    priv = GF_CALLOC(1, sizeof(*priv), gf_leases_mt_private_t);
+    if (!priv) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, LEASE_MSG_NO_MEM,
+               "Leases init failed");
+        goto out;
+    }
+
+    GF_OPTION_INIT("leases", priv->leases_enabled, bool, out);
+    GF_OPTION_INIT("lease-lock-recall-timeout", priv->recall_lease_timeout,
+                   int32, out);
+    pthread_mutex_init(&priv->mutex, NULL);
+    INIT_LIST_HEAD(&priv->client_list);
+    INIT_LIST_HEAD(&priv->recall_list);
+
+    this->private = priv;
+
+    if (priv->leases_enabled) {
+        ret = leases_init_priv(this);
+        if (ret)
+            goto out;
+    }
+
+    ret = 0;
+
+out:
+    if (ret) {
+        GF_FREE(priv);
+        this->private = NULL;
+    }
+
+    return ret;
+}
+
+void
+fini(xlator_t *this)
+{
+    leases_private_t *priv = NULL;
+
+    priv = this->private;
+    if (!priv) {
+        return;
+    }
+    this->private = NULL;
+
+    priv->fini = _gf_true;
+    pthread_cond_broadcast(&priv->cond);
+    if (priv->recall_thr) {
+        gf_thread_cleanup_xint(priv->recall_thr);
+        priv->recall_thr = 0;
+        priv->inited_recall_thr = _gf_false;
+    }
+
+    if (priv->timer_wheel) {
+        glusterfs_ctx_tw_put(this->ctx);
+    }
+
+    GF_FREE(priv);
+    return;
+}
+
+static int
+leases_forget(xlator_t *this, inode_t *inode)
+{
+    /* TODO:leases_cleanup_inode_ctx (this, inode); */
+    return 0;
+}
+
+static int
+leases_release(xlator_t *this, fd_t *fd)
+{
+    int ret = -1;
+    uint64_t tmp = 0;
+    lease_fd_ctx_t *fd_ctx = NULL;
+
+    if (fd == NULL) {
+        goto out;
+    }
+
+    gf_log(this->name, GF_LOG_TRACE, "Releasing all leases with fd %p", fd);
+
+    ret = fd_ctx_del(fd, this, &tmp);
+    if (ret) {
+        gf_log(this->name, GF_LOG_DEBUG, "Could not get fdctx");
+        goto out;
+    }
+
+    fd_ctx = (lease_fd_ctx_t *)(long)tmp;
+    if (fd_ctx)
+        GF_FREE(fd_ctx);
+out:
+    return ret;
+}
+
+static int
+leases_clnt_disconnect_cbk(xlator_t *this, client_t *client)
+{
+    int ret = 0;
+
+    EXIT_IF_LEASES_OFF(this, out);
+
+    ret = cleanup_client_leases(this, client->client_uid);
+out:
+    return ret;
+}
+
+struct xlator_fops fops = {
+    /* Metadata modifying fops */
+    .fsetattr = leases_fsetattr,
+    .setattr = leases_setattr,
+
+    /* File Data reading fops */
+    .open = leases_open,
+    .readv = leases_readv,
+
+    /* File Data modifying fops */
+    .truncate = leases_truncate,
+    .ftruncate = leases_ftruncate,
+    .writev = leases_writev,
+    .zerofill = leases_zerofill,
+    .fallocate = leases_fallocate,
+    .discard = leases_discard,
+    .lk = leases_lk,
+    .fsync = leases_fsync,
+    .flush = leases_flush,
+    .lease = leases_lease,
+
+    /* Directory Data modifying fops */
+    .create = leases_create,
+    .rename = leases_rename,
+    .unlink = leases_unlink,
+    .link = leases_link,
+
+#ifdef NOT_SUPPORTED
+    /* internal lk fops */
+    .inodelk = leases_inodelk,
+    .finodelk = leases_finodelk,
+    .entrylk = leases_entrylk,
+    .fentrylk = leases_fentrylk,
+
+    /* Internal special fops*/
+    .xattrop = leases_xattrop,
+    .fxattrop = leases_fxattrop,
+#endif
+};
+
+struct xlator_cbks cbks = {
+    .forget = leases_forget,
+    .release = leases_release,
+    .client_disconnect = leases_clnt_disconnect_cbk,
+};
+
+struct volume_options options[] = {
+    {.key = {"leases"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .op_version = {GD_OP_VERSION_3_8_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .description = "When \"on\", enables leases support"},
+    {.key = {"lease-lock-recall-timeout"},
+     .type = GF_OPTION_TYPE_INT,
+     .default_value = RECALL_LEASE_LK_TIMEOUT,
+     .op_version = {GD_OP_VERSION_3_8_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .description = "After 'timeout' seconds since the recall_lease"
+                    " request has been sent to the client, the lease lock"
+                    " will be forcefully purged by the server."},
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "leases",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/features/leases/src/leases.h b/xlators/features/leases/src/leases.h
new file mode 100644
index 00000000000..a6e8a6824cc
--- /dev/null
+++ b/xlators/features/leases/src/leases.h
@@ -0,0 +1,259 @@
+/*
+  Copyright (c) 2015-2016 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _LEASES_H
+#define _LEASES_H
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include <glusterfs/common-utils.h>
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/call-stub.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/client_t.h>
+#include <glusterfs/lkowner.h>
+#include <glusterfs/locking.h>
+#include <glusterfs/upcall-utils.h>
+#include "timer-wheel.h"
+#include "leases-mem-types.h"
+#include "leases-messages.h"
+
+/* The time period for which a client lease lock will be stored after its been
+ * recalled for the first time. */
+#define RECALL_LEASE_LK_TIMEOUT "60"
+
+#define DATA_MODIFY_FOP 0x0001
+#define BLOCKING_FOP 0x0002
+
+#define BLOCK_FOP 0x0001
+#define WIND_FOP 0x0002
+
+#define EXIT_IF_LEASES_OFF(this, label)                                        \
+    do {                                                                       \
+        if (!is_leases_enabled(this))                                          \
+            goto label;                                                        \
+    } while (0)
+
+#define EXIT_IF_INTERNAL_FOP(frame, xdata, label)                              \
+    do {                                                                       \
+        if (frame->root->pid < 0)                                              \
+            goto label;                                                        \
+        if (xdata && dict_get(xdata, GLUSTERFS_INTERNAL_FOP_KEY))              \
+            goto label;                                                        \
+    } while (0)
+
+#define GET_LEASE_ID(xdata, lease_id, client_uid)                              \
+    do {                                                                       \
+        int ret_val = -1;                                                      \
+        ret_val = dict_get_bin(xdata, "lease-id", (void **)&lease_id);         \
+        if (ret_val) {                                                         \
+            ret_val = 0;                                                       \
+            gf_msg_debug("leases", 0, "Lease id is not set for client:%s",     \
+                         client_uid);                                          \
+        }                                                                      \
+    } while (0)
+
+#define GET_FLAGS(fop, fd_flags)                                               \
+    do {                                                                       \
+        if ((fd_flags & (O_WRONLY | O_RDWR)) && fop == GF_FOP_OPEN)            \
+            fop_flags = DATA_MODIFY_FOP;                                       \
+                                                                               \
+        if (fop == GF_FOP_UNLINK || fop == GF_FOP_RENAME ||                    \
+            fop == GF_FOP_TRUNCATE || fop == GF_FOP_FTRUNCATE ||               \
+            fop == GF_FOP_FLUSH || fop == GF_FOP_FSYNC ||                      \
+            fop == GF_FOP_WRITE || fop == GF_FOP_FALLOCATE ||                  \
+            fop == GF_FOP_DISCARD || fop == GF_FOP_ZEROFILL ||                 \
+            fop == GF_FOP_SETATTR || fop == GF_FOP_FSETATTR ||                 \
+            fop == GF_FOP_LINK)                                                \
+            fop_flags = DATA_MODIFY_FOP;                                       \
+                                                                               \
+        if (!(fd_flags & (O_NONBLOCK | O_NDELAY)))                             \
+            fop_flags |= BLOCKING_FOP;                                         \
+                                                                               \
+    } while (0)
+
+#define GET_FLAGS_LK(cmd, l_type, fd_flags)                                    \
+    do {                                                                       \
+        /* TODO: handle F_RESLK_LCK and other glusterfs_lk_recovery_cmds_t */  \
+        if ((cmd == F_SETLKW || cmd == F_SETLKW64 || cmd == F_SETLK ||         \
+             cmd == F_SETLK64) &&                                              \
+            l_type == F_WRLCK)                                                 \
+            fop_flags = DATA_MODIFY_FOP;                                       \
+                                                                               \
+        if (fd_flags & (O_NONBLOCK | O_NDELAY) &&                              \
+            (cmd == F_SETLKW || cmd == F_SETLKW64))                            \
+            fop_flags |= BLOCKING_FOP;                                         \
+                                                                               \
+    } while (0)
+
+#define LEASE_BLOCK_FOP(inode, fop_name, frame, this, params...)               \
+    do {                                                                       \
+        call_stub_t *__stub = NULL;                                            \
+        fop_stub_t *blk_fop = NULL;                                            \
+        lease_inode_ctx_t *lease_ctx = NULL;                                   \
+                                                                               \
+        __stub = fop_##fop_name##_stub(frame, default_##fop_name##_resume,     \
+                                       params);                                \
+        if (!__stub) {                                                         \
+            gf_msg(this->name, GF_LOG_WARNING, ENOMEM, LEASE_MSG_NO_MEM,       \
+                   "Unable to create stub");                                   \
+            ret = -ENOMEM;                                                     \
+            goto __out;                                                        \
+        }                                                                      \
+                                                                               \
+        blk_fop = GF_CALLOC(1, sizeof(*blk_fop), gf_leases_mt_fop_stub_t);     \
+        if (!blk_fop) {                                                        \
+            gf_msg(this->name, GF_LOG_WARNING, ENOMEM, LEASE_MSG_NO_MEM,       \
+                   "Unable to create lease fop stub");                         \
+            ret = -ENOMEM;                                                     \
+            goto __out;                                                        \
+        }                                                                      \
+                                                                               \
+        lease_ctx = lease_ctx_get(inode, this);                                \
+        if (!lease_ctx) {                                                      \
+            gf_msg(this->name, GF_LOG_WARNING, ENOMEM, LEASE_MSG_NO_MEM,       \
+                   "Unable to create/get inode ctx");                          \
+            ret = -ENOMEM;                                                     \
+            goto __out;                                                        \
+        }                                                                      \
+                                                                               \
+        blk_fop->stub = __stub;                                                \
+        pthread_mutex_lock(&lease_ctx->lock);                                  \
+        {                                                                      \
+            /*TODO: If the lease is unlocked btw check lease conflict and      \
+             * by now, then this fop shouldn't be add to the blocked fop       \
+             * list, can use generation number for the same?*/                 \
+            list_add_tail(&blk_fop->list, &lease_ctx->blocked_list);           \
+        }                                                                      \
+        pthread_mutex_unlock(&lease_ctx->lock);                                \
+                                                                               \
+    __out:                                                                     \
+        if (ret < 0) {                                                         \
+            gf_msg(this->name, GF_LOG_WARNING, ENOMEM, LEASE_MSG_NO_MEM,       \
+                   "Unable to create stub for blocking the fop:%s (%s)",       \
+                   gf_fop_list[frame->root->op], strerror(ENOMEM));            \
+            if (__stub != NULL) {                                              \
+                call_stub_destroy(__stub);                                     \
+            }                                                                  \
+            GF_FREE(blk_fop);                                                  \
+            goto err;                                                          \
+        }                                                                      \
+    } while (0)
+
+struct _leases_private {
+    struct list_head client_list;
+    struct list_head recall_list;
+    struct tvec_base *timer_wheel; /* timer wheel where the recall request
+                                      is qued and waits for unlock/expiry */
+    pthread_t recall_thr;
+    pthread_mutex_t mutex;
+    pthread_cond_t cond;
+    int32_t recall_lease_timeout;
+    gf_boolean_t inited_recall_thr;
+    gf_boolean_t fini;
+    gf_boolean_t leases_enabled;
+
+    char _pad[1]; /* manual padding */
+};
+typedef struct _leases_private leases_private_t;
+
+struct _lease_client {
+    char *client_uid;
+    struct list_head client_list;
+    struct list_head inode_list;
+};
+typedef struct _lease_client lease_client_t;
+
+struct _lease_inode {
+    inode_t *inode;
+    struct list_head
+        list; /* This can be part of both inode_list and recall_list */
+};
+typedef struct _lease_inode lease_inode_t;
+
+struct _lease_fd_ctx {
+    char *client_uid;
+    char lease_id[LEASE_ID_SIZE];
+};
+typedef struct _lease_fd_ctx lease_fd_ctx_t;
+
+struct _lease_inode_ctx {
+    struct list_head lease_id_list; /* clients that have taken leases */
+    int lease_type_cnt[GF_LEASE_MAX_TYPE + 1];
+    uint64_t lease_cnt;            /* Total number of leases on this inode */
+    uint64_t openfd_cnt;           /* number of fds open */
+    struct list_head blocked_list; /* List of fops blocked until the
+                                      lease recall is complete */
+    inode_t *inode;                /* this represents the inode on which the
+                                      lock was taken, required mainly during
+                                      disconnect cleanup */
+    struct gf_tw_timer_list *timer;
+    pthread_mutex_t lock;
+    int lease_type;                  /* Types of leases acquired */
+    gf_boolean_t recall_in_progress; /* if lease recall is sent on this inode */
+    gf_boolean_t blocked_fops_resuming; /* if blocked fops are being resumed */
+
+    char _pad[2]; /* manual padding */
+};
+typedef struct _lease_inode_ctx lease_inode_ctx_t;
+
+struct _lease_id_entry {
+    struct list_head lease_id_list;
+    char lease_id[LEASE_ID_SIZE];
+    char *client_uid;                          /* uid of the client that has
+                                                  taken the lease */
+    int lease_type_cnt[GF_LEASE_MAX_TYPE + 1]; /* count of each lease type */
+    uint64_t lease_cnt; /* Number of leases taken under the
+                           given lease id */
+    time_t recall_time; /* time @ which recall was sent */
+    int lease_type;     /* Union of all the leases taken
+                           under the given lease id */
+    char _pad[4];       /* manual padding */
+};
+typedef struct _lease_id_entry lease_id_entry_t;
+
+/* Required? as stub itself will have list */
+struct __fop_stub {
+    struct list_head list;
+    call_stub_t *stub;
+};
+typedef struct __fop_stub fop_stub_t;
+
+struct __lease_timer_data {
+    inode_t *inode;
+    xlator_t *this;
+};
+typedef struct __lease_timer_data lease_timer_data_t;
+
+gf_boolean_t
+is_leases_enabled(xlator_t *this);
+
+lease_inode_ctx_t *
+lease_ctx_get(inode_t *inode, xlator_t *this);
+
+int
+process_lease_req(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                  struct gf_lease *lease);
+
+int
+check_lease_conflict(call_frame_t *frame, inode_t *inode, const char *lease_id,
+                     uint32_t fop_flags);
+
+int
+cleanup_client_leases(xlator_t *this, const char *client_uid);
+
+void *
+expired_recall_cleanup(void *data);
+
+#endif /* _LEASES_H */
diff --git a/xlators/features/locks/src/Makefile.am b/xlators/features/locks/src/Makefile.am
index 53dd3aa5da5..0b174c19d2d 100644
--- a/xlators/features/locks/src/Makefile.am
+++ b/xlators/features/locks/src/Makefile.am
@@ -1,20 +1,29 @@
+if WITH_SERVER
 xlator_LTLIBRARIES = locks.la
+endif
 xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
 
-locks_la_LDFLAGS = -module -avoidversion
+locks_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
 
-locks_la_SOURCES = common.c posix.c entrylk.c inodelk.c reservelk.c
-locks_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la 
+locks_la_SOURCES = common.c posix.c entrylk.c inodelk.c reservelk.c \
+	clear.c
 
-noinst_HEADERS = locks.h common.h locks-mem-types.h
+locks_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -fno-strict-aliasing -D$(GF_HOST_OS) \
-	-I$(top_srcdir)/libglusterfs/src $(GF_CFLAGS) -shared -nostartfiles
+noinst_HEADERS = locks.h common.h locks-mem-types.h clear.h pl-messages.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+
+
+AM_CFLAGS = -Wall -fno-strict-aliasing $(GF_CFLAGS)
 
 CLEANFILES = 
 
+if WITH_SERVER
 uninstall-local:
 	rm -f $(DESTDIR)$(xlatordir)/posix-locks.so
 
 install-data-hook:
-	ln -sf locks.so $(DESTDIR)$(xlatordir)/posix-locks.so
\ No newline at end of file
+	ln -sf locks.so $(DESTDIR)$(xlatordir)/posix-locks.so
+endif
diff --git a/xlators/features/locks/src/clear.c b/xlators/features/locks/src/clear.c
new file mode 100644
index 00000000000..ab1eac68a53
--- /dev/null
+++ b/xlators/features/locks/src/clear.c
@@ -0,0 +1,460 @@
+/*
+   Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include <unistd.h>
+#include <fcntl.h>
+#include <limits.h>
+#include <pthread.h>
+
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/common-utils.h>
+
+#include "locks.h"
+#include "common.h"
+#include <glusterfs/statedump.h>
+#include "clear.h"
+
+const char *clrlk_type_names[CLRLK_TYPE_MAX] = {
+    [CLRLK_INODE] = "inode",
+    [CLRLK_ENTRY] = "entry",
+    [CLRLK_POSIX] = "posix",
+};
+
+int
+clrlk_get_kind(char *kind)
+{
+    char *clrlk_kinds[CLRLK_KIND_MAX] = {"dummy", "blocked", "granted", "all"};
+    int ret_kind = CLRLK_KIND_MAX;
+    int i = 0;
+
+    for (i = CLRLK_BLOCKED; i < CLRLK_KIND_MAX; i++) {
+        if (!strcmp(clrlk_kinds[i], kind)) {
+            ret_kind = i;
+            break;
+        }
+    }
+
+    return ret_kind;
+}
+
+int
+clrlk_get_type(char *type)
+{
+    char *clrlk_types[CLRLK_TYPE_MAX] = {"inode", "entry", "posix"};
+    int ret_type = CLRLK_TYPE_MAX;
+    int i = 0;
+
+    for (i = CLRLK_INODE; i < CLRLK_TYPE_MAX; i++) {
+        if (!strcmp(clrlk_types[i], type)) {
+            ret_type = i;
+            break;
+        }
+    }
+
+    return ret_type;
+}
+
+int
+clrlk_get_lock_range(char *range_str, struct gf_flock *ulock,
+                     gf_boolean_t *chk_range)
+{
+    int ret = -1;
+
+    if (!chk_range)
+        goto out;
+
+    if (!range_str) {
+        ret = 0;
+        *chk_range = _gf_false;
+        goto out;
+    }
+
+    if (sscanf(range_str,
+               "%hd,%" PRId64 "-"
+               "%" PRId64,
+               &ulock->l_whence, &ulock->l_start, &ulock->l_len) != 3) {
+        goto out;
+    }
+
+    ret = 0;
+    *chk_range = _gf_true;
+out:
+    return ret;
+}
+
+int
+clrlk_parse_args(const char *cmd, clrlk_args *args)
+{
+    char *opts = NULL;
+    char *cur = NULL;
+    char *tok = NULL;
+    char *sptr = NULL;
+    char *free_ptr = NULL;
+    char kw[KW_MAX] = {
+        [KW_TYPE] = 't',
+        [KW_KIND] = 'k',
+    };
+    int ret = -1;
+    int i = 0;
+
+    GF_ASSERT(cmd);
+    free_ptr = opts = GF_CALLOC(1, strlen(cmd), gf_common_mt_char);
+    if (!opts)
+        goto out;
+
+    if (sscanf(cmd, GF_XATTR_CLRLK_CMD ".%s", opts) < 1) {
+        ret = -1;
+        goto out;
+    }
+
+    /*clr_lk_prefix.ttype.kkind.args, args - type specific*/
+    cur = opts;
+    for (i = 0; i < KW_MAX && (tok = strtok_r(cur, ".", &sptr));
+         cur = NULL, i++) {
+        if (tok[0] != kw[i]) {
+            ret = -1;
+            goto out;
+        }
+        if (i == KW_TYPE)
+            args->type = clrlk_get_type(tok + 1);
+        if (i == KW_KIND)
+            args->kind = clrlk_get_kind(tok + 1);
+    }
+
+    if ((args->type == CLRLK_TYPE_MAX) || (args->kind == CLRLK_KIND_MAX))
+        goto out;
+
+    /*optional args, neither range nor basename can 'legally' contain
+     * "/" in them*/
+    tok = strtok_r(NULL, "/", &sptr);
+    if (tok)
+        args->opts = gf_strdup(tok);
+
+    ret = 0;
+out:
+    GF_FREE(free_ptr);
+    return ret;
+}
+
+int
+clrlk_clear_posixlk(xlator_t *this, pl_inode_t *pl_inode, clrlk_args *args,
+                    int *blkd, int *granted, int *op_errno)
+{
+    posix_lock_t *plock = NULL;
+    posix_lock_t *tmp = NULL;
+    struct gf_flock ulock = {
+        0,
+    };
+    int ret = -1;
+    int bcount = 0;
+    int gcount = 0;
+    gf_boolean_t chk_range = _gf_false;
+
+    if (clrlk_get_lock_range(args->opts, &ulock, &chk_range)) {
+        *op_errno = EINVAL;
+        goto out;
+    }
+
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        list_for_each_entry_safe(plock, tmp, &pl_inode->ext_list, list)
+        {
+            if ((plock->blocked && !(args->kind & CLRLK_BLOCKED)) ||
+                (!plock->blocked && !(args->kind & CLRLK_GRANTED)))
+                continue;
+
+            if (chk_range && (plock->user_flock.l_whence != ulock.l_whence ||
+                              plock->user_flock.l_start != ulock.l_start ||
+                              plock->user_flock.l_len != ulock.l_len))
+                continue;
+
+            list_del_init(&plock->list);
+            if (plock->blocked) {
+                bcount++;
+                pl_trace_out(this, plock->frame, NULL, NULL, F_SETLKW,
+                             &plock->user_flock, -1, EINTR, NULL);
+
+                STACK_UNWIND_STRICT(lk, plock->frame, -1, EINTR,
+                                    &plock->user_flock, NULL);
+
+            } else {
+                gcount++;
+            }
+            __destroy_lock(plock);
+        }
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+    grant_blocked_locks(this, pl_inode);
+    ret = 0;
+out:
+    *blkd = bcount;
+    *granted = gcount;
+    return ret;
+}
+
+/* Returns 0 on success and -1 on failure */
+int
+clrlk_clear_inodelk(xlator_t *this, pl_inode_t *pl_inode, pl_dom_list_t *dom,
+                    clrlk_args *args, int *blkd, int *granted, int *op_errno)
+{
+    posix_locks_private_t *priv;
+    pl_inode_lock_t *ilock = NULL;
+    pl_inode_lock_t *tmp = NULL;
+    struct gf_flock ulock = {
+        0,
+    };
+    int ret = -1;
+    int bcount = 0;
+    int gcount = 0;
+    gf_boolean_t chk_range = _gf_false;
+    struct list_head *pcontend = NULL;
+    struct list_head released;
+    struct list_head contend;
+    struct timespec now = {};
+
+    INIT_LIST_HEAD(&released);
+
+    priv = this->private;
+    if (priv->notify_contention) {
+        pcontend = &contend;
+        INIT_LIST_HEAD(pcontend);
+        timespec_now(&now);
+    }
+
+    if (clrlk_get_lock_range(args->opts, &ulock, &chk_range)) {
+        *op_errno = EINVAL;
+        goto out;
+    }
+
+    if (args->kind & CLRLK_BLOCKED)
+        goto blkd;
+
+    if (args->kind & CLRLK_GRANTED)
+        goto granted;
+
+blkd:
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        list_for_each_entry_safe(ilock, tmp, &dom->blocked_inodelks,
+                                 blocked_locks)
+        {
+            if (chk_range && (ilock->user_flock.l_whence != ulock.l_whence ||
+                              ilock->user_flock.l_start != ulock.l_start ||
+                              ilock->user_flock.l_len != ulock.l_len))
+                continue;
+
+            bcount++;
+            list_del_init(&ilock->client_list);
+            list_del_init(&ilock->blocked_locks);
+            list_add(&ilock->blocked_locks, &released);
+        }
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+
+    if (!list_empty(&released)) {
+        list_for_each_entry_safe(ilock, tmp, &released, blocked_locks)
+        {
+            list_del_init(&ilock->blocked_locks);
+            pl_trace_out(this, ilock->frame, NULL, NULL, F_SETLKW,
+                         &ilock->user_flock, -1, EAGAIN, ilock->volume);
+            STACK_UNWIND_STRICT(inodelk, ilock->frame, -1, EAGAIN, NULL);
+            // No need to take lock as the locks are only in one list
+            __pl_inodelk_unref(ilock);
+        }
+    }
+
+    if (!(args->kind & CLRLK_GRANTED)) {
+        ret = 0;
+        goto out;
+    }
+
+granted:
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        list_for_each_entry_safe(ilock, tmp, &dom->inodelk_list, list)
+        {
+            if (chk_range && (ilock->user_flock.l_whence != ulock.l_whence ||
+                              ilock->user_flock.l_start != ulock.l_start ||
+                              ilock->user_flock.l_len != ulock.l_len))
+                continue;
+
+            gcount++;
+            list_del_init(&ilock->client_list);
+            list_del_init(&ilock->list);
+            list_add(&ilock->list, &released);
+        }
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+
+    list_for_each_entry_safe(ilock, tmp, &released, list)
+    {
+        list_del_init(&ilock->list);
+        // No need to take lock as the locks are only in one list
+        __pl_inodelk_unref(ilock);
+    }
+
+    ret = 0;
+out:
+    grant_blocked_inode_locks(this, pl_inode, dom, &now, pcontend);
+    if (pcontend != NULL) {
+        inodelk_contention_notify(this, pcontend);
+    }
+    *blkd = bcount;
+    *granted = gcount;
+    return ret;
+}
+
+/* Returns 0 on success and -1 on failure */
+int
+clrlk_clear_entrylk(xlator_t *this, pl_inode_t *pl_inode, pl_dom_list_t *dom,
+                    clrlk_args *args, int *blkd, int *granted, int *op_errno)
+{
+    posix_locks_private_t *priv;
+    pl_entry_lock_t *elock = NULL;
+    pl_entry_lock_t *tmp = NULL;
+    int bcount = 0;
+    int gcount = 0;
+    int ret = -1;
+    struct list_head *pcontend = NULL;
+    struct list_head removed;
+    struct list_head released;
+    struct list_head contend;
+    struct timespec now;
+
+    INIT_LIST_HEAD(&released);
+
+    priv = this->private;
+    if (priv->notify_contention) {
+        pcontend = &contend;
+        INIT_LIST_HEAD(pcontend);
+        timespec_now(&now);
+    }
+
+    if (args->kind & CLRLK_BLOCKED)
+        goto blkd;
+
+    if (args->kind & CLRLK_GRANTED)
+        goto granted;
+
+blkd:
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        list_for_each_entry_safe(elock, tmp, &dom->blocked_entrylks,
+                                 blocked_locks)
+        {
+            if (args->opts) {
+                if (!elock->basename || strcmp(elock->basename, args->opts))
+                    continue;
+            }
+
+            bcount++;
+
+            list_del_init(&elock->client_list);
+            list_del_init(&elock->blocked_locks);
+            list_add_tail(&elock->blocked_locks, &released);
+        }
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+
+    if (!list_empty(&released)) {
+        list_for_each_entry_safe(elock, tmp, &released, blocked_locks)
+        {
+            list_del_init(&elock->blocked_locks);
+            entrylk_trace_out(this, elock->frame, elock->volume, NULL, NULL,
+                              elock->basename, ENTRYLK_LOCK, elock->type, -1,
+                              EAGAIN);
+            STACK_UNWIND_STRICT(entrylk, elock->frame, -1, EAGAIN, NULL);
+
+            __pl_entrylk_unref(elock);
+        }
+    }
+
+    if (!(args->kind & CLRLK_GRANTED)) {
+        ret = 0;
+        goto out;
+    }
+
+granted:
+    INIT_LIST_HEAD(&removed);
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        list_for_each_entry_safe(elock, tmp, &dom->entrylk_list, domain_list)
+        {
+            if (args->opts) {
+                if (!elock->basename || strcmp(elock->basename, args->opts))
+                    continue;
+            }
+
+            gcount++;
+            list_del_init(&elock->client_list);
+            list_del_init(&elock->domain_list);
+            list_add_tail(&elock->domain_list, &removed);
+
+            __pl_entrylk_unref(elock);
+        }
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+
+    grant_blocked_entry_locks(this, pl_inode, dom, &now, pcontend);
+    if (pcontend != NULL) {
+        entrylk_contention_notify(this, pcontend);
+    }
+
+    ret = 0;
+out:
+    *blkd = bcount;
+    *granted = gcount;
+    return ret;
+}
+
+int
+clrlk_clear_lks_in_all_domains(xlator_t *this, pl_inode_t *pl_inode,
+                               clrlk_args *args, int *blkd, int *granted,
+                               int *op_errno)
+{
+    pl_dom_list_t *dom = NULL;
+    int ret = -1;
+    int tmp_bcount = 0;
+    int tmp_gcount = 0;
+
+    if (list_empty(&pl_inode->dom_list)) {
+        ret = 0;
+        goto out;
+    }
+
+    list_for_each_entry(dom, &pl_inode->dom_list, inode_list)
+    {
+        tmp_bcount = tmp_gcount = 0;
+
+        switch (args->type) {
+            case CLRLK_INODE:
+                ret = clrlk_clear_inodelk(this, pl_inode, dom, args,
+                                          &tmp_bcount, &tmp_gcount, op_errno);
+                if (ret)
+                    goto out;
+                break;
+            case CLRLK_ENTRY:
+                ret = clrlk_clear_entrylk(this, pl_inode, dom, args,
+                                          &tmp_bcount, &tmp_gcount, op_errno);
+                if (ret)
+                    goto out;
+                break;
+        }
+
+        *blkd += tmp_bcount;
+        *granted += tmp_gcount;
+    }
+
+    ret = 0;
+out:
+    return ret;
+}
diff --git a/xlators/features/locks/src/clear.h b/xlators/features/locks/src/clear.h
new file mode 100644
index 00000000000..bc118cb1b81
--- /dev/null
+++ b/xlators/features/locks/src/clear.h
@@ -0,0 +1,73 @@
+/*
+   Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#ifndef __CLEAR_H__
+#define __CLEAR_H__
+
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/stack.h>
+#include <glusterfs/call-stub.h>
+#include "locks.h"
+
+typedef enum {
+    CLRLK_INODE,
+    CLRLK_ENTRY,
+    CLRLK_POSIX,
+    CLRLK_TYPE_MAX
+} clrlk_type;
+
+extern const char *clrlk_type_names[];
+
+typedef enum {
+    CLRLK_BLOCKED = 1,
+    CLRLK_GRANTED,
+    CLRLK_ALL,
+    CLRLK_KIND_MAX
+} clrlk_kind;
+
+typedef enum {
+    KW_TYPE,
+    KW_KIND,
+    /*add new keywords here*/
+    KW_MAX
+} clrlk_opts;
+
+struct _clrlk_args;
+typedef struct _clrlk_args clrlk_args;
+
+struct _clrlk_args {
+    int type;
+    int kind;
+    char *opts;
+};
+
+int
+clrlk_get__kind(char *kind);
+int
+clrlk_get_type(char *type);
+int
+clrlk_get_lock_range(char *range_str, struct gf_flock *ulock,
+                     gf_boolean_t *chk_range);
+int
+clrlk_parse_args(const char *cmd, clrlk_args *args);
+
+int
+clrlk_clear_posixlk(xlator_t *this, pl_inode_t *pl_inode, clrlk_args *args,
+                    int *blkd, int *granted, int *op_errno);
+int
+clrlk_clear_inodelk(xlator_t *this, pl_inode_t *pl_inode, pl_dom_list_t *dom,
+                    clrlk_args *args, int *blkd, int *granted, int *op_errno);
+int
+clrlk_clear_entrylk(xlator_t *this, pl_inode_t *pl_inode, pl_dom_list_t *dom,
+                    clrlk_args *args, int *blkd, int *granted, int *op_errno);
+int
+clrlk_clear_lks_in_all_domains(xlator_t *this, pl_inode_t *pl_inode,
+                               clrlk_args *args, int *blkd, int *granted,
+                               int *op_errno);
+#endif /* __CLEAR_H__ */
diff --git a/xlators/features/locks/src/common.c b/xlators/features/locks/src/common.c
index e295e4fd518..a2c6be93e03 100644
--- a/xlators/features/locks/src/common.c
+++ b/xlators/features/locks/src/common.c
@@ -1,708 +1,786 @@
 /*
-  Copyright (c) 2006, 2007, 2008 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
+   Copyright (c) 2006-2012, 2015-2016 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
 #include <unistd.h>
 #include <fcntl.h>
 #include <limits.h>
 #include <pthread.h>
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "compat.h"
-#include "xlator.h"
-#include "inode.h"
-#include "logging.h"
-#include "common-utils.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/syncop.h>
 
 #include "locks.h"
 #include "common.h"
 
-
 static int
-__is_lock_grantable (pl_inode_t *pl_inode, posix_lock_t *lock);
+__is_lock_grantable(pl_inode_t *pl_inode, posix_lock_t *lock);
 static void
-__insert_and_merge (pl_inode_t *pl_inode, posix_lock_t *lock);
+__insert_and_merge(pl_inode_t *pl_inode, posix_lock_t *lock);
 static int
-pl_send_prelock_unlock (xlator_t *this, pl_inode_t *pl_inode,
-                        posix_lock_t *old_lock);
+pl_send_prelock_unlock(xlator_t *this, pl_inode_t *pl_inode,
+                       posix_lock_t *old_lock);
+
 static pl_dom_list_t *
-allocate_domain (const char *volume)
+__allocate_domain(const char *volume)
 {
-        pl_dom_list_t *dom = NULL;
+    pl_dom_list_t *dom = NULL;
 
-        dom = GF_CALLOC (1, sizeof (*dom),
-                         gf_locks_mt_pl_dom_list_t);
-        if (!dom)
-                return NULL;
+    dom = GF_CALLOC(1, sizeof(*dom), gf_locks_mt_pl_dom_list_t);
+    if (!dom)
+        goto out;
 
+    dom->domain = gf_strdup(volume);
+    if (!dom->domain)
+        goto out;
 
-        dom->domain = gf_strdup(volume);
-	if (!dom->domain) {
-		gf_log ("posix-locks", GF_LOG_TRACE,
-			"Out of Memory");
-		return NULL;
-	}
+    gf_log("posix-locks", GF_LOG_TRACE, "New domain allocated: %s",
+           dom->domain);
 
-        gf_log ("posix-locks", GF_LOG_TRACE,
-                "New domain allocated: %s", dom->domain);
+    INIT_LIST_HEAD(&dom->inode_list);
+    INIT_LIST_HEAD(&dom->entrylk_list);
+    INIT_LIST_HEAD(&dom->blocked_entrylks);
+    INIT_LIST_HEAD(&dom->inodelk_list);
+    INIT_LIST_HEAD(&dom->blocked_inodelks);
 
-        INIT_LIST_HEAD (&dom->inode_list);
-        INIT_LIST_HEAD (&dom->entrylk_list);
-        INIT_LIST_HEAD (&dom->blocked_entrylks);
-        INIT_LIST_HEAD (&dom->inodelk_list);
-        INIT_LIST_HEAD (&dom->blocked_inodelks);
+out:
+    if (dom && (NULL == dom->domain)) {
+        GF_FREE(dom);
+        dom = NULL;
+    }
 
-        return dom;
+    return dom;
 }
 
 /* Returns domain for the lock. If domain is not present,
  * allocates a domain and returns it
  */
 pl_dom_list_t *
-get_domain (pl_inode_t *pl_inode, const char *volume)
+get_domain(pl_inode_t *pl_inode, const char *volume)
 {
-        pl_dom_list_t *dom = NULL;
-
-        list_for_each_entry (dom, &pl_inode->dom_list, inode_list) {
-                if (strcmp (dom->domain, volume) == 0)
-                        goto found;
+    pl_dom_list_t *dom = NULL;
 
+    GF_VALIDATE_OR_GOTO("posix-locks", pl_inode, out);
+    GF_VALIDATE_OR_GOTO("posix-locks", volume, out);
 
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        list_for_each_entry(dom, &pl_inode->dom_list, inode_list)
+        {
+            if (strcmp(dom->domain, volume) == 0)
+                goto unlock;
         }
 
-        dom = allocate_domain(volume);
-
+        dom = __allocate_domain(volume);
         if (dom)
-                list_add (&dom->inode_list, &pl_inode->dom_list);
-found:
-
-        return dom;
+            list_add(&dom->inode_list, &pl_inode->dom_list);
+    }
+unlock:
+    pthread_mutex_unlock(&pl_inode->mutex);
+    if (dom) {
+        gf_log("posix-locks", GF_LOG_TRACE, "Domain %s found", volume);
+    } else {
+        gf_log("posix-locks", GF_LOG_TRACE, "Domain %s not found", volume);
+    }
+out:
+    return dom;
 }
 
 unsigned long
-fd_to_fdnum (fd_t *fd)
+fd_to_fdnum(fd_t *fd)
 {
-        return ((unsigned long) fd);
+    return ((unsigned long)fd);
 }
 
 fd_t *
-fd_from_fdnum (posix_lock_t *lock)
+fd_from_fdnum(posix_lock_t *lock)
 {
-        return ((fd_t *) lock->fd_num);
+    return ((fd_t *)lock->fd_num);
 }
 
 int
-__pl_inode_is_empty (pl_inode_t *pl_inode)
+__pl_inode_is_empty(pl_inode_t *pl_inode)
 {
-        pl_dom_list_t *dom = NULL;
-        int            is_empty = 1;
-
-        if (!list_empty (&pl_inode->ext_list))
-                is_empty = 0;
-
-        list_for_each_entry (dom, &pl_inode->dom_list, inode_list) {
-                if (!list_empty (&dom->entrylk_list))
-                        is_empty = 0;
-
-                if (!list_empty (&dom->inodelk_list))
-                        is_empty = 0;
-        }
-
-        return is_empty;
+    return (list_empty(&pl_inode->ext_list));
 }
 
 void
-pl_print_locker (char *str, int size, xlator_t *this, call_frame_t *frame)
+pl_print_locker(char *str, int size, xlator_t *this, call_frame_t *frame)
 {
-        snprintf (str, size, "Pid=%llu, lk-owner=%llu, Transport=%p, Frame=%llu",
-                  (unsigned long long) frame->root->pid,
-                  (unsigned long long) frame->root->lk_owner,
-                  (void *)frame->root->trans,
-                  (unsigned long long) frame->root->unique);
+    snprintf(str, size, "Pid=%llu, lk-owner=%s, Client=%p, Frame=%llu",
+             (unsigned long long)frame->root->pid,
+             lkowner_utoa(&frame->root->lk_owner), frame->root->client,
+             (unsigned long long)frame->root->unique);
 }
 
-
 void
-pl_print_lockee (char *str, int size, fd_t *fd, loc_t *loc)
+pl_print_lockee(char *str, int size, fd_t *fd, loc_t *loc)
 {
-        inode_t  *inode = NULL;
-        char     *ipath = NULL;
-        int       ret = 0;
+    inode_t *inode = NULL;
+    char *ipath = NULL;
+    int ret = 0;
 
-        if (fd)
-                inode = fd->inode;
-        if (loc)
-                inode = loc->inode;
+    if (fd)
+        inode = fd->inode;
+    if (loc)
+        inode = loc->inode;
 
-        if (!inode) {
-                snprintf (str, size, "<nul>");
-                return;
-        }
+    if (!inode) {
+        snprintf(str, size, "<nul>");
+        return;
+    }
 
-        if (loc && loc->path) {
-                ipath = gf_strdup (loc->path);
-        } else {
-                ret = inode_path (inode, NULL, &ipath);
-                if (ret <= 0)
-                        ipath = NULL;
-        }
+    if (loc && loc->path) {
+        ipath = gf_strdup(loc->path);
+    } else {
+        ret = inode_path(inode, NULL, &ipath);
+        if (ret <= 0)
+            ipath = NULL;
+    }
 
-        snprintf (str, size, "ino=%llu, fd=%p, path=%s",
-                  (unsigned long long) inode->ino, fd,
-                  ipath ? ipath : "<nul>");
+    snprintf(str, size, "gfid=%s, fd=%p, path=%s", uuid_utoa(inode->gfid), fd,
+             ipath ? ipath : "<nul>");
 
-        if (ipath)
-                GF_FREE (ipath);
+    GF_FREE(ipath);
 }
 
-
 void
-pl_print_lock (char *str, int size, int cmd,
-               struct gf_flock *flock, uint64_t owner)
+pl_print_lock(char *str, int size, int cmd, struct gf_flock *flock,
+              gf_lkowner_t *owner)
 {
-        char *cmd_str = NULL;
-        char *type_str = NULL;
+    char *cmd_str = NULL;
+    char *type_str = NULL;
 
-        switch (cmd) {
+    switch (cmd) {
 #if F_GETLK != F_GETLK64
         case F_GETLK64:
 #endif
         case F_GETLK:
-                cmd_str = "GETLK";
-                break;
+            cmd_str = "GETLK";
+            break;
 
 #if F_SETLK != F_SETLK64
         case F_SETLK64:
 #endif
         case F_SETLK:
-                cmd_str = "SETLK";
-                break;
+            cmd_str = "SETLK";
+            break;
 
 #if F_SETLKW != F_SETLKW64
         case F_SETLKW64:
 #endif
         case F_SETLKW:
-                cmd_str = "SETLKW";
-                break;
+            cmd_str = "SETLKW";
+            break;
 
         default:
-                cmd_str = "UNKNOWN";
-                break;
-        }
+            cmd_str = "UNKNOWN";
+            break;
+    }
 
-        switch (flock->l_type) {
+    switch (flock->l_type) {
         case F_RDLCK:
-                type_str = "READ";
-                break;
+            type_str = "READ";
+            break;
         case F_WRLCK:
-                type_str = "WRITE";
-                break;
+            type_str = "WRITE";
+            break;
         case F_UNLCK:
-                type_str = "UNLOCK";
-                break;
+            type_str = "UNLOCK";
+            break;
         default:
-                type_str = "UNKNOWN";
-                break;
-        }
-
-        snprintf (str, size, "lock=FCNTL, cmd=%s, type=%s, "
-                  "start=%llu, len=%llu, pid=%llu, lk-owner=%llu",
-                  cmd_str, type_str, (unsigned long long) flock->l_start,
-                  (unsigned long long) flock->l_len,
-                  (unsigned long long) flock->l_pid,
-                  (unsigned long long) owner);
+            type_str = "UNKNOWN";
+            break;
+    }
+
+    snprintf(str, size,
+             "lock=FCNTL, cmd=%s, type=%s, "
+             "start=%llu, len=%llu, pid=%llu, lk-owner=%s",
+             cmd_str, type_str, (unsigned long long)flock->l_start,
+             (unsigned long long)flock->l_len, (unsigned long long)flock->l_pid,
+             lkowner_utoa(owner));
 }
 
-
 void
-pl_trace_in (xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc,
-             int cmd, struct gf_flock *flock, const char *domain)
+pl_trace_in(xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc, int cmd,
+            struct gf_flock *flock, const char *domain)
 {
-        posix_locks_private_t  *priv = NULL;
-        char                    pl_locker[256];
-        char                    pl_lockee[256];
-        char                    pl_lock[256];
+    posix_locks_private_t *priv = this->private;
+    char pl_locker[256];
+    char pl_lockee[256];
+    char pl_lock[256];
 
-        priv = this->private;
-
-        if (!priv->trace)
-                return;
+    if (!priv->trace)
+        return;
 
-        pl_print_locker (pl_locker, 256, this, frame);
-        pl_print_lockee (pl_lockee, 256, fd, loc);
-        if (domain)
-                pl_print_inodelk (pl_lock, 256, cmd, flock, domain);
-        else
-                pl_print_lock (pl_lock, 256, cmd, flock, frame->root->lk_owner);
+    pl_print_locker(pl_locker, 256, this, frame);
+    pl_print_lockee(pl_lockee, 256, fd, loc);
+    if (domain)
+        pl_print_inodelk(pl_lock, 256, cmd, flock, domain);
+    else
+        pl_print_lock(pl_lock, 256, cmd, flock, &frame->root->lk_owner);
 
-        gf_log (this->name, GF_LOG_NORMAL,
-                "[REQUEST] Locker = {%s} Lockee = {%s} Lock = {%s}",
-                pl_locker, pl_lockee, pl_lock);
+    gf_log(this->name, GF_LOG_INFO,
+           "[REQUEST] Locker = {%s} Lockee = {%s} Lock = {%s}", pl_locker,
+           pl_lockee, pl_lock);
 }
 
-
 void
-pl_print_verdict (char *str, int size, int op_ret, int op_errno)
+pl_print_verdict(char *str, int size, int op_ret, int op_errno)
 {
-        char *verdict = NULL;
-
-        if (op_ret == 0) {
-                verdict = "GRANTED";
-        } else {
-                switch (op_errno) {
-                case EAGAIN:
-                        verdict = "TRYAGAIN";
-                        break;
-                default:
-                        verdict = strerror (op_errno);
-                }
+    char *verdict = NULL;
+
+    if (op_ret == 0) {
+        verdict = "GRANTED";
+    } else {
+        switch (op_errno) {
+            case EAGAIN:
+                verdict = "TRYAGAIN";
+                break;
+            default:
+                verdict = strerror(op_errno);
         }
+    }
 
-        snprintf (str, size, "%s", verdict);
+    snprintf(str, size, "%s", verdict);
 }
 
-
 void
-pl_trace_out (xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc,
-              int cmd, struct gf_flock *flock, int op_ret, int op_errno, const char *domain)
+pl_trace_out(xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc, int cmd,
+             struct gf_flock *flock, int op_ret, int op_errno,
+             const char *domain)
 
 {
-        posix_locks_private_t  *priv = NULL;
-        char                    pl_locker[256];
-        char                    pl_lockee[256];
-        char                    pl_lock[256];
-        char                    verdict[32];
+    posix_locks_private_t *priv = NULL;
+    char pl_locker[256];
+    char pl_lockee[256];
+    char pl_lock[256];
+    char verdict[32];
 
-        priv = this->private;
+    priv = this->private;
 
-        if (!priv->trace)
-                return;
+    if (!priv->trace)
+        return;
 
-        pl_print_locker (pl_locker, 256, this, frame);
-        pl_print_lockee (pl_lockee, 256, fd, loc);
-        if (domain)
-                pl_print_inodelk (pl_lock, 256, cmd, flock, domain);
-        else
-                pl_print_lock (pl_lock, 256, cmd, flock, frame->root->lk_owner);
+    pl_print_locker(pl_locker, 256, this, frame);
+    pl_print_lockee(pl_lockee, 256, fd, loc);
+    if (domain)
+        pl_print_inodelk(pl_lock, 256, cmd, flock, domain);
+    else
+        pl_print_lock(pl_lock, 256, cmd, flock, &frame->root->lk_owner);
 
-        pl_print_verdict (verdict, 32, op_ret, op_errno);
+    pl_print_verdict(verdict, 32, op_ret, op_errno);
 
-        gf_log (this->name, GF_LOG_NORMAL,
-                "[%s] Locker = {%s} Lockee = {%s} Lock = {%s}",
-                verdict, pl_locker, pl_lockee, pl_lock);
+    gf_log(this->name, GF_LOG_INFO,
+           "[%s] Locker = {%s} Lockee = {%s} Lock = {%s}", verdict, pl_locker,
+           pl_lockee, pl_lock);
 }
 
-
 void
-pl_trace_block (xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc,
-                int cmd, struct gf_flock *flock, const char *domain)
+pl_trace_block(xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc,
+               int cmd, struct gf_flock *flock, const char *domain)
 
 {
-        posix_locks_private_t  *priv = NULL;
-        char                    pl_locker[256];
-        char                    pl_lockee[256];
-        char                    pl_lock[256];
+    posix_locks_private_t *priv = this->private;
+    char pl_locker[256];
+    char pl_lockee[256];
+    char pl_lock[256];
 
-        priv = this->private;
-
-        if (!priv->trace)
-                return;
+    if (!priv->trace)
+        return;
 
-        pl_print_locker (pl_locker, 256, this, frame);
-        pl_print_lockee (pl_lockee, 256, fd, loc);
-        if (domain)
-                pl_print_inodelk (pl_lock, 256, cmd, flock, domain);
-        else
-                pl_print_lock (pl_lock, 256, cmd, flock, frame->root->lk_owner);
+    pl_print_locker(pl_locker, 256, this, frame);
+    pl_print_lockee(pl_lockee, 256, fd, loc);
+    if (domain)
+        pl_print_inodelk(pl_lock, 256, cmd, flock, domain);
+    else
+        pl_print_lock(pl_lock, 256, cmd, flock, &frame->root->lk_owner);
 
-        gf_log (this->name, GF_LOG_NORMAL,
-                "[BLOCKED] Locker = {%s} Lockee = {%s} Lock = {%s}",
-                pl_locker, pl_lockee, pl_lock);
+    gf_log(this->name, GF_LOG_INFO,
+           "[BLOCKED] Locker = {%s} Lockee = {%s} Lock = {%s}", pl_locker,
+           pl_lockee, pl_lock);
 }
 
-
 void
-pl_trace_flush (xlator_t *this, call_frame_t *frame, fd_t *fd)
+pl_trace_flush(xlator_t *this, call_frame_t *frame, fd_t *fd)
 {
-        posix_locks_private_t  *priv = NULL;
-        char                    pl_locker[256];
-        char                    pl_lockee[256];
-        pl_inode_t             *pl_inode = NULL;
+    posix_locks_private_t *priv = NULL;
+    char pl_locker[256];
+    char pl_lockee[256];
+    pl_inode_t *pl_inode = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (!priv->trace)
-                return;
+    if (!priv->trace)
+        return;
 
-        pl_inode = pl_inode_get (this, fd->inode);
+    pl_inode = pl_inode_get(this, fd->inode, NULL);
 
-        if (pl_inode && __pl_inode_is_empty (pl_inode))
-                return;
+    if (pl_inode && __pl_inode_is_empty(pl_inode))
+        return;
 
-        pl_print_locker (pl_locker, 256, this, frame);
-        pl_print_lockee (pl_lockee, 256, fd, NULL);
+    pl_print_locker(pl_locker, 256, this, frame);
+    pl_print_lockee(pl_lockee, 256, fd, NULL);
 
-        gf_log (this->name, GF_LOG_NORMAL,
-                "[FLUSH] Locker = {%s} Lockee = {%s}",
-                pl_locker, pl_lockee);
+    gf_log(this->name, GF_LOG_INFO, "[FLUSH] Locker = {%s} Lockee = {%s}",
+           pl_locker, pl_lockee);
 }
 
 void
-pl_trace_release (xlator_t *this, fd_t *fd)
+pl_trace_release(xlator_t *this, fd_t *fd)
 {
-        posix_locks_private_t  *priv = NULL;
-        char                    pl_lockee[256];
+    posix_locks_private_t *priv = NULL;
+    char pl_lockee[256];
 
-        priv = this->private;
+    priv = this->private;
 
-        if (!priv->trace)
-                return;
+    if (!priv->trace)
+        return;
 
-        pl_print_lockee (pl_lockee, 256, fd, NULL);
+    pl_print_lockee(pl_lockee, 256, fd, NULL);
 
-        gf_log (this->name, GF_LOG_NORMAL,
-                "[RELEASE] Lockee = {%s}", pl_lockee);
+    gf_log(this->name, GF_LOG_INFO, "[RELEASE] Lockee = {%s}", pl_lockee);
 }
 
-
 void
-pl_update_refkeeper (xlator_t *this, inode_t *inode)
+pl_update_refkeeper(xlator_t *this, inode_t *inode)
 {
-        pl_inode_t *pl_inode  = NULL;
-        int         is_empty  = 0;
-        int         need_unref = 0;
+    pl_inode_t *pl_inode = NULL;
+    int is_empty = 0;
+    int need_unref = 0;
+    int need_ref = 0;
 
-        pl_inode = pl_inode_get (this, inode);
+    pl_inode = pl_inode_get(this, inode, NULL);
+    if (!pl_inode)
+        return;
 
-        pthread_mutex_lock (&pl_inode->mutex);
-        {
-                is_empty = __pl_inode_is_empty (pl_inode);
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        is_empty = __pl_inode_is_empty(pl_inode);
 
-                if (is_empty && pl_inode->refkeeper) {
-                        need_unref = 1;
-                        pl_inode->refkeeper = NULL;
-                }
+        if (is_empty && pl_inode->refkeeper) {
+            need_unref = 1;
+            pl_inode->refkeeper = NULL;
+        }
 
-                if (!is_empty && !pl_inode->refkeeper) {
-                        pl_inode->refkeeper = inode_ref (inode);
-                }
+        if (!is_empty && !pl_inode->refkeeper) {
+            need_ref = 1;
+            pl_inode->refkeeper = inode;
         }
-        pthread_mutex_unlock (&pl_inode->mutex);
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
 
-        if (need_unref)
-                inode_unref (inode);
+    if (need_unref)
+        inode_unref(inode);
+
+    if (need_ref)
+        inode_ref(inode);
 }
 
+/* Get lock enforcement info from disk */
+int
+pl_fetch_mlock_info_from_disk(xlator_t *this, pl_inode_t *pl_inode,
+                              pl_local_t *local)
+{
+    dict_t *xdata_rsp = NULL;
+    int ret = 0;
+    int op_ret = 0;
+
+    if (!local) {
+        return -1;
+    }
+
+    if (local->fd) {
+        op_ret = syncop_fgetxattr(this, local->fd, &xdata_rsp,
+                                  GF_ENFORCE_MANDATORY_LOCK, NULL, NULL);
+    } else {
+        op_ret = syncop_getxattr(this, &local->loc[0], &xdata_rsp,
+                                 GF_ENFORCE_MANDATORY_LOCK, NULL, NULL);
+    }
+
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        if (op_ret >= 0) {
+            pl_inode->mlock_enforced = _gf_true;
+            pl_inode->check_mlock_info = _gf_false;
+        } else {
+            gf_msg(this->name, GF_LOG_WARNING, -op_ret, 0,
+                   "getxattr failed with %d", op_ret);
+            pl_inode->mlock_enforced = _gf_false;
+
+            if (-op_ret == ENODATA) {
+                pl_inode->check_mlock_info = _gf_false;
+            } else {
+                pl_inode->check_mlock_info = _gf_true;
+            }
+        }
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+
+    return ret;
+}
 
 pl_inode_t *
-pl_inode_get (xlator_t *this, inode_t *inode)
+pl_inode_get(xlator_t *this, inode_t *inode, pl_local_t *local)
 {
-        uint64_t    tmp_pl_inode = 0;
-	pl_inode_t *pl_inode = NULL;
-//	mode_t      st_mode = 0;
-	int         ret = 0;
-
-	ret = inode_ctx_get (inode, this,&tmp_pl_inode);
-	if (ret == 0) {
-                pl_inode = (pl_inode_t *)(long)tmp_pl_inode;
-		goto out;
+    uint64_t tmp_pl_inode = 0;
+    pl_inode_t *pl_inode = NULL;
+    int ret = 0;
+
+    LOCK(&inode->lock);
+    {
+        ret = __inode_ctx_get(inode, this, &tmp_pl_inode);
+        if (ret == 0) {
+            pl_inode = (pl_inode_t *)(long)tmp_pl_inode;
+            goto unlock;
         }
-	pl_inode = GF_CALLOC (1, sizeof (*pl_inode),
-                              gf_locks_mt_pl_inode_t);
-	if (!pl_inode) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		goto out;
-	}
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "Allocating new pl inode");
+        pl_inode = GF_CALLOC(1, sizeof(*pl_inode), gf_locks_mt_pl_inode_t);
+        if (!pl_inode) {
+            goto unlock;
+        }
 
-/*
-	st_mode  = inode->st_mode;
-	if ((st_mode & S_ISGID) && !(st_mode & S_IXGRP))
-		pl_inode->mandatory = 1;
-*/
+        gf_log(this->name, GF_LOG_TRACE, "Allocating new pl inode");
+
+        pthread_mutex_init(&pl_inode->mutex, NULL);
+        pthread_cond_init(&pl_inode->check_fop_wind_count, 0);
+
+        INIT_LIST_HEAD(&pl_inode->dom_list);
+        INIT_LIST_HEAD(&pl_inode->ext_list);
+        INIT_LIST_HEAD(&pl_inode->rw_list);
+        INIT_LIST_HEAD(&pl_inode->reservelk_list);
+        INIT_LIST_HEAD(&pl_inode->blocked_reservelks);
+        INIT_LIST_HEAD(&pl_inode->blocked_calls);
+        INIT_LIST_HEAD(&pl_inode->metalk_list);
+        INIT_LIST_HEAD(&pl_inode->queued_locks);
+        INIT_LIST_HEAD(&pl_inode->waiting);
+        gf_uuid_copy(pl_inode->gfid, inode->gfid);
+
+        pl_inode->check_mlock_info = _gf_true;
+        pl_inode->mlock_enforced = _gf_false;
+
+        /* -2 means never looked up. -1 means something went wrong and link
+         * tracking is disabled. */
+        pl_inode->links = -2;
+
+        ret = __inode_ctx_put(inode, this, (uint64_t)(long)(pl_inode));
+        if (ret) {
+            pthread_mutex_destroy(&pl_inode->mutex);
+            GF_FREE(pl_inode);
+            pl_inode = NULL;
+            goto unlock;
+        }
+    }
+unlock:
+    UNLOCK(&inode->lock);
+
+    if ((pl_inode != NULL) && pl_is_mandatory_locking_enabled(pl_inode) &&
+        pl_inode->check_mlock_info && local) {
+        /* Note: The lock enforcement information per file can be stored in the
+           attribute flag of stat(x) in posix. With that there won't be a need
+           for doing getxattr post a reboot
+        */
+        pl_fetch_mlock_info_from_disk(this, pl_inode, local);
+    }
+
+    return pl_inode;
+}
 
-	pthread_mutex_init (&pl_inode->mutex, NULL);
+/* Create a new posix_lock_t */
+posix_lock_t *
+new_posix_lock(struct gf_flock *flock, client_t *client, pid_t client_pid,
+               gf_lkowner_t *owner, fd_t *fd, uint32_t lk_flags, int blocking,
+               int32_t *op_errno)
+{
+    posix_lock_t *lock = NULL;
 
-	INIT_LIST_HEAD (&pl_inode->dom_list);
-	INIT_LIST_HEAD (&pl_inode->ext_list);
-	INIT_LIST_HEAD (&pl_inode->rw_list);
-	INIT_LIST_HEAD (&pl_inode->reservelk_list);
-	INIT_LIST_HEAD (&pl_inode->blocked_reservelks);
-        INIT_LIST_HEAD (&pl_inode->blocked_calls);
+    GF_VALIDATE_OR_GOTO("posix-locks", flock, out);
+    GF_VALIDATE_OR_GOTO("posix-locks", client, out);
+    GF_VALIDATE_OR_GOTO("posix-locks", fd, out);
 
-	inode_ctx_put (inode, this, (uint64_t)(long)(pl_inode));
+    if (!pl_is_lk_owner_valid(owner, client)) {
+        *op_errno = EINVAL;
+        goto out;
+    }
 
-out:
-	return pl_inode;
-}
+    lock = GF_CALLOC(1, sizeof(posix_lock_t), gf_locks_mt_posix_lock_t);
+    if (!lock) {
+        *op_errno = ENOMEM;
+        goto out;
+    }
 
+    lock->fl_start = flock->l_start;
+    lock->fl_type = flock->l_type;
 
-/* Create a new posix_lock_t */
-posix_lock_t *
-new_posix_lock (struct gf_flock *flock, void *transport, pid_t client_pid,
-                uint64_t owner, fd_t *fd)
-{
-	posix_lock_t *lock = NULL;
+    if (flock->l_len == 0)
+        lock->fl_end = LLONG_MAX;
+    else
+        lock->fl_end = flock->l_start + flock->l_len - 1;
 
-	lock = GF_CALLOC (1, sizeof (posix_lock_t),
-                          gf_locks_mt_posix_lock_t);
-	if (!lock) {
-		return NULL;
-	}
+    lock->client = client;
 
-	lock->fl_start = flock->l_start;
-	lock->fl_type  = flock->l_type;
+    lock->client_uid = gf_strdup(client->client_uid);
+    if (lock->client_uid == NULL) {
+        GF_FREE(lock);
+        lock = NULL;
+        *op_errno = ENOMEM;
+        goto out;
+    }
 
-	if (flock->l_len == 0)
-		lock->fl_end = LLONG_MAX;
-	else
-		lock->fl_end = flock->l_start + flock->l_len - 1;
+    lock->fd_num = fd_to_fdnum(fd);
+    lock->fd = fd;
+    lock->client_pid = client_pid;
+    lock->owner = *owner;
+    lock->lk_flags = lk_flags;
 
-	lock->transport  = transport;
-        lock->fd_num     = fd_to_fdnum (fd);
-        lock->fd         = fd;
-	lock->client_pid = client_pid;
-        lock->owner      = owner;
+    lock->blocking = blocking;
+    memcpy(&lock->user_flock, flock, sizeof(lock->user_flock));
 
-	INIT_LIST_HEAD (&lock->list);
+    INIT_LIST_HEAD(&lock->list);
 
-	return lock;
+out:
+    return lock;
 }
 
-
 /* Delete a lock from the inode's lock list */
 void
-__delete_lock (pl_inode_t *pl_inode, posix_lock_t *lock)
+__delete_lock(posix_lock_t *lock)
 {
-	list_del_init (&lock->list);
+    list_del_init(&lock->list);
 }
 
-
 /* Destroy a posix_lock */
 void
-__destroy_lock (posix_lock_t *lock)
+__destroy_lock(posix_lock_t *lock)
 {
-	GF_FREE (lock);
+    GF_FREE(lock->client_uid);
+    GF_FREE(lock);
 }
 
+static posix_lock_t *
+__copy_lock(posix_lock_t *src)
+{
+    posix_lock_t *dst;
+
+    dst = GF_MALLOC(sizeof(posix_lock_t), gf_locks_mt_posix_lock_t);
+    if (dst != NULL) {
+        memcpy(dst, src, sizeof(posix_lock_t));
+        dst->client_uid = gf_strdup(src->client_uid);
+        if (dst->client_uid == NULL) {
+            GF_FREE(dst);
+            dst = NULL;
+        }
+
+        if (dst != NULL)
+            INIT_LIST_HEAD(&dst->list);
+    }
+
+    return dst;
+}
 
 /* Convert a posix_lock to a struct gf_flock */
 void
-posix_lock_to_flock (posix_lock_t *lock, struct gf_flock *flock)
+posix_lock_to_flock(posix_lock_t *lock, struct gf_flock *flock)
 {
-	flock->l_pid   = lock->client_pid;
-	flock->l_type  = lock->fl_type;
-	flock->l_start = lock->fl_start;
-        flock->l_owner = lock->owner;
-
-	if (lock->fl_end == LLONG_MAX)
-		flock->l_len = 0;
-	else
-		flock->l_len = lock->fl_end - lock->fl_start + 1;
+    flock->l_pid = lock->user_flock.l_pid;
+    flock->l_type = lock->fl_type;
+    flock->l_start = lock->fl_start;
+    flock->l_owner = lock->owner;
+
+    if (lock->fl_end == LLONG_MAX)
+        flock->l_len = 0;
+    else
+        flock->l_len = lock->fl_end - lock->fl_start + 1;
 }
 
 /* Insert the lock into the inode's lock list */
 static void
-__insert_lock (pl_inode_t *pl_inode, posix_lock_t *lock)
+__insert_lock(pl_inode_t *pl_inode, posix_lock_t *lock)
 {
-        list_add_tail (&lock->list, &pl_inode->ext_list);
+    if (lock->blocked)
+        lock->blkd_time = gf_time();
+    else
+        lock->granted_time = gf_time();
 
-	return;
+    list_add_tail(&lock->list, &pl_inode->ext_list);
 }
 
-
 /* Return true if the locks overlap, false otherwise */
 int
-locks_overlap (posix_lock_t *l1, posix_lock_t *l2)
+locks_overlap(posix_lock_t *l1, posix_lock_t *l2)
 {
-	/* 
-	   Note:
-	   FUSE always gives us absolute offsets, so no need to worry 
-	   about SEEK_CUR or SEEK_END
-	*/
-
-	return ((l1->fl_end >= l2->fl_start) &&
-		(l2->fl_end >= l1->fl_start));
-}
+    /*
+       Note:
+       FUSE always gives us absolute offsets, so no need to worry
+       about SEEK_CUR or SEEK_END
+    */
 
+    return ((l1->fl_end >= l2->fl_start) && (l2->fl_end >= l1->fl_start));
+}
 
 /* Return true if the locks have the same owner */
 int
-same_owner (posix_lock_t *l1, posix_lock_t *l2)
+same_owner(posix_lock_t *l1, posix_lock_t *l2)
 {
-
-                return ((l1->owner == l2->owner) &&
-                        (l1->transport  == l2->transport));
-
+    return (is_same_lkowner(&l1->owner, &l2->owner) &&
+            (l1->client == l2->client));
 }
 
-
 /* Delete all F_UNLCK locks */
 void
-__delete_unlck_locks (pl_inode_t *pl_inode)
+__delete_unlck_locks(pl_inode_t *pl_inode)
 {
-	posix_lock_t *l = NULL;
-	posix_lock_t *tmp = NULL;
-
-	list_for_each_entry_safe (l, tmp, &pl_inode->ext_list, list) {
-		if (l->fl_type == F_UNLCK) {
-			__delete_lock (pl_inode, l);
-			__destroy_lock (l);
-		}
-	}
+    posix_lock_t *l = NULL;
+    posix_lock_t *tmp = NULL;
+
+    list_for_each_entry_safe(l, tmp, &pl_inode->ext_list, list)
+    {
+        if (l->fl_type == F_UNLCK) {
+            __delete_lock(l);
+            __destroy_lock(l);
+        }
+    }
 }
 
-
 /* Add two locks */
 static posix_lock_t *
-add_locks (posix_lock_t *l1, posix_lock_t *l2)
+add_locks(posix_lock_t *l1, posix_lock_t *l2, posix_lock_t *dst)
 {
-	posix_lock_t *sum = NULL;
+    posix_lock_t *sum = NULL;
+
+    sum = __copy_lock(dst);
+    if (!sum)
+        return NULL;
 
-	sum = GF_CALLOC (1, sizeof (posix_lock_t),
-                         gf_locks_mt_posix_lock_t);
-	if (!sum)
-		return NULL;
+    sum->fl_start = min(l1->fl_start, l2->fl_start);
+    sum->fl_end = max(l1->fl_end, l2->fl_end);
 
-	sum->fl_start = min (l1->fl_start, l2->fl_start);
-	sum->fl_end   = max (l1->fl_end, l2->fl_end);
+    posix_lock_to_flock(sum, &sum->user_flock);
 
-	return sum;
+    return sum;
 }
 
 /* Subtract two locks */
 struct _values {
-	posix_lock_t *locks[3];
+    posix_lock_t *locks[3];
 };
 
 /* {big} must always be contained inside {small} */
 static struct _values
-subtract_locks (posix_lock_t *big, posix_lock_t *small)
+subtract_locks(posix_lock_t *big, posix_lock_t *small)
 {
+    struct _values v = {.locks = {0, 0, 0}};
 
-	struct _values v = { .locks = {0, 0, 0} };
-
-	if ((big->fl_start == small->fl_start) &&
-	    (big->fl_end   == small->fl_end)) {
-		/* both edges coincide with big */
-		v.locks[0] = GF_CALLOC (1, sizeof (posix_lock_t),
-                                        gf_locks_mt_posix_lock_t);
-                if (!v.locks[0])
-                        goto out;
-		memcpy (v.locks[0], big, sizeof (posix_lock_t));
-		v.locks[0]->fl_type = small->fl_type;
-                goto done;
-	}
-
-	if ((small->fl_start > big->fl_start) &&
-            (small->fl_end   < big->fl_end)) {
-		/* both edges lie inside big */
-		v.locks[0] = GF_CALLOC (1, sizeof (posix_lock_t),
-                                        gf_locks_mt_posix_lock_t);
-                if (!v.locks[0])
-                        goto out;
-
-		v.locks[1] = GF_CALLOC (1, sizeof (posix_lock_t),
-                                        gf_locks_mt_posix_lock_t);
-                if (!v.locks[1])
-                        goto out;
-
-		v.locks[2] = GF_CALLOC (1, sizeof (posix_lock_t),
-                                        gf_locks_mt_posix_lock_t);
-                if (!v.locks[1])
-                        goto out;
-
-		memcpy (v.locks[0], big, sizeof (posix_lock_t));
-		v.locks[0]->fl_end = small->fl_start - 1;
-
-		memcpy (v.locks[1], small, sizeof (posix_lock_t));
-
-		memcpy (v.locks[2], big, sizeof (posix_lock_t));
-		v.locks[2]->fl_start = small->fl_end + 1;
-                goto done;
-
-	}
-
-	/* one edge coincides with big */
-        if (small->fl_start == big->fl_start) {
-		v.locks[0] = GF_CALLOC (1, sizeof (posix_lock_t),
-                                        gf_locks_mt_posix_lock_t);
-                if (!v.locks[0])
-                        goto out;
-
-		v.locks[1] = GF_CALLOC (1, sizeof (posix_lock_t),
-                                        gf_locks_mt_posix_lock_t);
-                if (!v.locks[1])
-                        goto out;
-
-		memcpy (v.locks[0], big, sizeof (posix_lock_t));
-		v.locks[0]->fl_start = small->fl_end + 1;
-
-		memcpy (v.locks[1], small, sizeof (posix_lock_t));
-                goto done;
-	}
-
-	if (small->fl_end  == big->fl_end) {
-		v.locks[0] = GF_CALLOC (1, sizeof (posix_lock_t),
-                                        gf_locks_mt_posix_lock_t);
-                if (!v.locks[0])
-                        goto out;
-
-		v.locks[1] = GF_CALLOC (1, sizeof (posix_lock_t),
-                                        gf_locks_mt_posix_lock_t);
-                if (!v.locks[1])
-                        goto out;
-
-		memcpy (v.locks[0], big, sizeof (posix_lock_t));
-		v.locks[0]->fl_end = small->fl_start - 1;
-
-		memcpy (v.locks[1], small, sizeof (posix_lock_t));
-                goto done;
-	}
-
-        gf_log ("posix-locks", GF_LOG_ERROR,
-                "Unexpected case in subtract_locks. Please send "
-                "a bug report to gluster-devel@nongnu.org");
+    if ((big->fl_start == small->fl_start) && (big->fl_end == small->fl_end)) {
+        /* both edges coincide with big */
+        v.locks[0] = __copy_lock(big);
+        if (!v.locks[0]) {
+            goto out;
+        }
 
-out:
-        if (v.locks[0]) {
-                GF_FREE (v.locks[0]);
-                v.locks[0] = NULL;
+        v.locks[0]->fl_type = small->fl_type;
+        v.locks[0]->user_flock.l_type = small->fl_type;
+        goto done;
+    }
+
+    if ((small->fl_start > big->fl_start) && (small->fl_end < big->fl_end)) {
+        /* both edges lie inside big */
+        v.locks[0] = __copy_lock(big);
+        v.locks[1] = __copy_lock(small);
+        v.locks[2] = __copy_lock(big);
+        if ((v.locks[0] == NULL) || (v.locks[1] == NULL) ||
+            (v.locks[2] == NULL)) {
+            goto out;
         }
-        if (v.locks[1]) {
-                GF_FREE (v.locks[1]);
-                v.locks[1] = NULL;
+
+        v.locks[0]->fl_end = small->fl_start - 1;
+        v.locks[2]->fl_start = small->fl_end + 1;
+        posix_lock_to_flock(v.locks[0], &v.locks[0]->user_flock);
+        posix_lock_to_flock(v.locks[2], &v.locks[2]->user_flock);
+        goto done;
+    }
+
+    /* one edge coincides with big */
+    if (small->fl_start == big->fl_start) {
+        v.locks[0] = __copy_lock(big);
+        v.locks[1] = __copy_lock(small);
+        if ((v.locks[0] == NULL) || (v.locks[1] == NULL)) {
+            goto out;
         }
-        if (v.locks[2]) {
-                GF_FREE (v.locks[2]);
-                v.locks[2] = NULL;
+
+        v.locks[0]->fl_start = small->fl_end + 1;
+        posix_lock_to_flock(v.locks[0], &v.locks[0]->user_flock);
+        goto done;
+    }
+
+    if (small->fl_end == big->fl_end) {
+        v.locks[0] = __copy_lock(big);
+        v.locks[1] = __copy_lock(small);
+        if ((v.locks[0] == NULL) || (v.locks[1] == NULL)) {
+            goto out;
         }
 
+        v.locks[0]->fl_end = small->fl_start - 1;
+        posix_lock_to_flock(v.locks[0], &v.locks[0]->user_flock);
+        goto done;
+    }
+
+    GF_ASSERT(0);
+    gf_log("posix-locks", GF_LOG_ERROR, "Unexpected case in subtract_locks");
+
+out:
+    if (v.locks[0]) {
+        __destroy_lock(v.locks[0]);
+        v.locks[0] = NULL;
+    }
+    if (v.locks[1]) {
+        __destroy_lock(v.locks[1]);
+        v.locks[1] = NULL;
+    }
+    if (v.locks[2]) {
+        __destroy_lock(v.locks[2]);
+        v.locks[2] = NULL;
+    }
+
 done:
-        return v;
+    return v;
+}
+
+static posix_lock_t *
+first_conflicting_overlap(pl_inode_t *pl_inode, posix_lock_t *lock)
+{
+    posix_lock_t *l = NULL;
+    posix_lock_t *conf = NULL;
+
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        list_for_each_entry(l, &pl_inode->ext_list, list)
+        {
+            if (l->blocked)
+                continue;
+
+            if (locks_overlap(l, lock)) {
+                if (same_owner(l, lock))
+                    continue;
+
+                if ((l->fl_type == F_WRLCK) || (lock->fl_type == F_WRLCK)) {
+                    conf = l;
+                    goto unlock;
+                }
+            }
+        }
+    }
+unlock:
+    pthread_mutex_unlock(&pl_inode->mutex);
+
+    return conf;
 }
 
 /*
@@ -711,343 +789,803 @@ done:
   If {begin} is NULL, then start from the beginning of the list
 */
 static posix_lock_t *
-first_overlap (pl_inode_t *pl_inode, posix_lock_t *lock)
+first_overlap(pl_inode_t *pl_inode, posix_lock_t *lock)
 {
-        posix_lock_t *l = NULL;
+    posix_lock_t *l = NULL;
 
-        list_for_each_entry (l, &pl_inode->ext_list, list) {
-                if (l->blocked)
-                        continue;
+    list_for_each_entry(l, &pl_inode->ext_list, list)
+    {
+        if (l->blocked)
+            continue;
 
-                if (locks_overlap (l, lock))
-                        return l;
-        }
+        if (locks_overlap(l, lock))
+            return l;
+    }
 
-        return NULL;
+    return NULL;
 }
 
-
-
 /* Return true if lock is grantable */
 static int
-__is_lock_grantable (pl_inode_t *pl_inode, posix_lock_t *lock)
+__is_lock_grantable(pl_inode_t *pl_inode, posix_lock_t *lock)
 {
-        posix_lock_t *l = NULL;
-        int           ret = 1;
-
-        list_for_each_entry (l, &pl_inode->ext_list, list) {
-                if (!l->blocked && locks_overlap (lock, l)) {
-                        if (((l->fl_type == F_WRLCK)
-                             || (lock->fl_type == F_WRLCK))
-                            && (lock->fl_type != F_UNLCK)
-                            && !same_owner (l, lock)) {
-                                ret = 0;
-                                break;
-                        }
-                }
+    posix_lock_t *l = NULL;
+    int ret = 1;
+
+    list_for_each_entry(l, &pl_inode->ext_list, list)
+    {
+        if (!l->blocked && locks_overlap(lock, l)) {
+            if (((l->fl_type == F_WRLCK) || (lock->fl_type == F_WRLCK)) &&
+                (lock->fl_type != F_UNLCK) && !same_owner(l, lock)) {
+                ret = 0;
+                break;
+            }
         }
-        return ret;
+    }
+    return ret;
 }
 
-
-extern void do_blocked_rw (pl_inode_t *);
-
+extern void
+do_blocked_rw(pl_inode_t *);
 
 static void
-__insert_and_merge (pl_inode_t *pl_inode, posix_lock_t *lock)
+__insert_and_merge(pl_inode_t *pl_inode, posix_lock_t *lock)
 {
-        posix_lock_t  *conf = NULL;
-        posix_lock_t  *t = NULL;
-        posix_lock_t  *sum = NULL;
-        int            i = 0;
-        struct _values v = { .locks = {0, 0, 0} };
-
-        list_for_each_entry_safe (conf, t, &pl_inode->ext_list, list) {
-                if (!locks_overlap (conf, lock))
-                        continue;
+    posix_lock_t *conf = NULL;
+    posix_lock_t *t = NULL;
+    posix_lock_t *sum = NULL;
+    int i = 0;
+    struct _values v = {.locks = {0, 0, 0}};
+
+    list_for_each_entry_safe(conf, t, &pl_inode->ext_list, list)
+    {
+        if (conf->blocked)
+            continue;
+        if (!locks_overlap(conf, lock))
+            continue;
+
+        if (same_owner(conf, lock)) {
+            if (conf->fl_type == lock->fl_type &&
+                conf->lk_flags == lock->lk_flags) {
+                sum = add_locks(lock, conf, lock);
+
+                __delete_lock(conf);
+                __destroy_lock(conf);
+
+                __destroy_lock(lock);
+                INIT_LIST_HEAD(&sum->list);
+                posix_lock_to_flock(sum, &sum->user_flock);
+                __insert_and_merge(pl_inode, sum);
+
+                return;
+            } else {
+                sum = add_locks(lock, conf, conf);
+
+                v = subtract_locks(sum, lock);
 
-                if (same_owner (conf, lock)) {
-                        if (conf->fl_type == lock->fl_type) {
-                                sum = add_locks (lock, conf);
+                __delete_lock(conf);
+                __destroy_lock(conf);
 
-                                sum->fl_type    = lock->fl_type;
-                                sum->transport  = lock->transport;
-                                sum->fd_num     = lock->fd_num;
-                                sum->client_pid = lock->client_pid;
-                                sum->owner      = lock->owner;
+                __delete_lock(lock);
+                __destroy_lock(lock);
 
-                                __delete_lock (pl_inode, conf);
-                                __destroy_lock (conf);
+                __destroy_lock(sum);
 
-                                __destroy_lock (lock);
-                                __insert_and_merge (pl_inode, sum);
+                for (i = 0; i < 3; i++) {
+                    if (!v.locks[i])
+                        continue;
+
+                    __insert_and_merge(pl_inode, v.locks[i]);
+                }
 
-                                return;
-                        } else {
-                                sum = add_locks (lock, conf);
+                __delete_unlck_locks(pl_inode);
+                return;
+            }
+        }
 
-                                sum->fl_type    = conf->fl_type;
-                                sum->transport  = conf->transport;
-                                sum->fd_num     = conf->fd_num;
-                                sum->client_pid = conf->client_pid;
-                                sum->owner      = conf->owner;
+        if (lock->fl_type == F_UNLCK) {
+            continue;
+        }
 
-                                v = subtract_locks (sum, lock);
+        if ((conf->fl_type == F_RDLCK) && (lock->fl_type == F_RDLCK)) {
+            __insert_lock(pl_inode, lock);
+            return;
+        }
+    }
+
+    /* no conflicts, so just insert */
+    if (lock->fl_type != F_UNLCK) {
+        __insert_lock(pl_inode, lock);
+    } else {
+        __destroy_lock(lock);
+    }
+}
 
-                                __delete_lock (pl_inode, conf);
-                                __destroy_lock (conf);
+void
+__grant_blocked_locks(xlator_t *this, pl_inode_t *pl_inode,
+                      struct list_head *granted)
+{
+    struct list_head tmp_list;
+    posix_lock_t *l = NULL;
+    posix_lock_t *tmp = NULL;
+    posix_lock_t *conf = NULL;
+
+    INIT_LIST_HEAD(&tmp_list);
+
+    list_for_each_entry_safe(l, tmp, &pl_inode->ext_list, list)
+    {
+        if (l->blocked) {
+            conf = first_overlap(pl_inode, l);
+            if (conf)
+                continue;
+
+            l->blocked = 0;
+            list_move_tail(&l->list, &tmp_list);
+        }
+    }
 
-                                __delete_lock (pl_inode, lock);
-                                __destroy_lock (lock);
+    list_for_each_entry_safe(l, tmp, &tmp_list, list)
+    {
+        list_del_init(&l->list);
 
-                                __destroy_lock (sum);
+        if (__is_lock_grantable(pl_inode, l)) {
+            conf = GF_CALLOC(1, sizeof(*conf), gf_locks_mt_posix_lock_t);
 
-                                for (i = 0; i < 3; i++) {
-                                        if (!v.locks[i])
-                                                continue;
+            if (!conf) {
+                l->blocked = 1;
+                __insert_lock(pl_inode, l);
+                continue;
+            }
 
-                                        INIT_LIST_HEAD (&v.locks[i]->list);
-                                        __insert_and_merge (pl_inode,
-                                                            v.locks[i]);
-                                }
+            conf->frame = l->frame;
+            l->frame = NULL;
 
-                                __delete_unlck_locks (pl_inode);
-                                return;
-                        }
-                }
+            posix_lock_to_flock(l, &conf->user_flock);
 
-                if (lock->fl_type == F_UNLCK) {
-                        continue;
-                }
+            gf_log(this->name, GF_LOG_TRACE,
+                   "%s (pid=%d) lk-owner:%s %" PRId64 " - %" PRId64
+                   " => Granted",
+                   l->fl_type == F_UNLCK ? "Unlock" : "Lock", l->client_pid,
+                   lkowner_utoa(&l->owner), l->user_flock.l_start,
+                   l->user_flock.l_len);
 
-                if ((conf->fl_type == F_RDLCK) && (lock->fl_type == F_RDLCK)) {
-                        __insert_lock (pl_inode, lock);
-                        return;
-                }
-        }
+            __insert_and_merge(pl_inode, l);
 
-        /* no conflicts, so just insert */
-        if (lock->fl_type != F_UNLCK) {
-                __insert_lock (pl_inode, lock);
+            list_add(&conf->list, granted);
         } else {
-                __destroy_lock (lock);
+            l->blocked = 1;
+            __insert_lock(pl_inode, l);
         }
+    }
 }
 
-
 void
-__grant_blocked_locks (xlator_t *this, pl_inode_t *pl_inode, struct list_head *granted)
+grant_blocked_locks(xlator_t *this, pl_inode_t *pl_inode)
 {
-        struct list_head  tmp_list;
-        posix_lock_t     *l = NULL;
-        posix_lock_t     *tmp = NULL;
-        posix_lock_t     *conf = NULL;
+    struct list_head granted_list;
+    posix_lock_t *tmp = NULL;
+    posix_lock_t *lock = NULL;
+    pl_local_t *local = NULL;
+    INIT_LIST_HEAD(&granted_list);
+
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        __grant_blocked_locks(this, pl_inode, &granted_list);
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+
+    list_for_each_entry_safe(lock, tmp, &granted_list, list)
+    {
+        list_del_init(&lock->list);
+
+        pl_trace_out(this, lock->frame, NULL, NULL, F_SETLKW, &lock->user_flock,
+                     0, 0, NULL);
+        local = lock->frame->local;
+        PL_STACK_UNWIND_AND_FREE(local, lk, lock->frame, 0, 0,
+                                 &lock->user_flock, NULL);
+        __destroy_lock(lock);
+    }
+
+    return;
+}
 
-        INIT_LIST_HEAD (&tmp_list);
+static int
+pl_send_prelock_unlock(xlator_t *this, pl_inode_t *pl_inode,
+                       posix_lock_t *old_lock)
+{
+    struct gf_flock flock = {
+        0,
+    };
+    posix_lock_t *unlock_lock = NULL;
+    int32_t op_errno = 0;
 
-        list_for_each_entry_safe (l, tmp, &pl_inode->ext_list, list) {
-                if (l->blocked) {
-                        conf = first_overlap (pl_inode, l);
-                        if (conf)
-                                continue;
+    struct list_head granted_list;
+    posix_lock_t *tmp = NULL;
+    posix_lock_t *lock = NULL;
+    pl_local_t *local = NULL;
 
-                        l->blocked = 0;
-                        list_move_tail (&l->list, &tmp_list);
-                }
-        }
+    int ret = -1;
 
-        list_for_each_entry_safe (l, tmp, &tmp_list, list) {
-                list_del_init (&l->list);
+    INIT_LIST_HEAD(&granted_list);
 
-                if (__is_lock_grantable (pl_inode, l)) {
-			conf = GF_CALLOC (1, sizeof (*conf),
-                                          gf_locks_mt_posix_lock_t);
+    flock.l_type = F_UNLCK;
+    flock.l_whence = old_lock->user_flock.l_whence;
+    flock.l_start = old_lock->user_flock.l_start;
+    flock.l_len = old_lock->user_flock.l_len;
+    flock.l_pid = old_lock->user_flock.l_pid;
 
-                        if (!conf) {
-                                l->blocked = 1;
-                                __insert_lock (pl_inode, l);
-                                continue;
-                        }
+    unlock_lock = new_posix_lock(&flock, old_lock->client, old_lock->client_pid,
+                                 &old_lock->owner, old_lock->fd,
+                                 old_lock->lk_flags, 0, &op_errno);
+    GF_VALIDATE_OR_GOTO(this->name, unlock_lock, out);
+    ret = 0;
 
-                        conf->frame = l->frame;
-                        l->frame = NULL;
+    __insert_and_merge(pl_inode, unlock_lock);
 
-                        posix_lock_to_flock (l, &conf->user_flock);
+    __grant_blocked_locks(this, pl_inode, &granted_list);
 
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "%s (pid=%d) lk-owner:%"PRIu64" %"PRId64" - %"PRId64" => Granted",
-                                l->fl_type == F_UNLCK ? "Unlock" : "Lock",
-                                l->client_pid,
-                                l->owner,
-                                l->user_flock.l_start,
-                                l->user_flock.l_len);
+    list_for_each_entry_safe(lock, tmp, &granted_list, list)
+    {
+        list_del_init(&lock->list);
 
-                        __insert_and_merge (pl_inode, l);
+        pl_trace_out(this, lock->frame, NULL, NULL, F_SETLKW, &lock->user_flock,
+                     0, 0, NULL);
+        local = lock->frame->local;
+        PL_STACK_UNWIND_AND_FREE(local, lk, lock->frame, 0, 0,
+                                 &lock->user_flock, NULL);
+        __destroy_lock(lock);
+    }
 
-                        list_add (&conf->list, granted);
-                } else {
-                        l->blocked = 1;
-                        __insert_lock (pl_inode, l);
-                }
+out:
+    return ret;
+}
+
+int
+pl_setlk(xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock,
+         int can_block)
+{
+    int ret = 0;
+
+    errno = 0;
+
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        /* Send unlock before the actual lock to
+           prevent lock upgrade / downgrade
+           problems only if:
+           - it is a blocking call
+           - it has other conflicting locks
+        */
+
+        if (can_block && !(__is_lock_grantable(pl_inode, lock))) {
+            ret = pl_send_prelock_unlock(this, pl_inode, lock);
+            if (ret)
+                gf_log(this->name, GF_LOG_DEBUG,
+                       "Could not send pre-lock "
+                       "unlock");
+        }
+
+        if (__is_lock_grantable(pl_inode, lock)) {
+            if (pl_metalock_is_active(pl_inode)) {
+                __pl_queue_lock(pl_inode, lock);
+                pthread_mutex_unlock(&pl_inode->mutex);
+                ret = -2;
+                goto out;
+            }
+            gf_log(this->name, GF_LOG_TRACE,
+                   "%s (pid=%d) lk-owner:%s %" PRId64 " - %" PRId64 " => OK",
+                   lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
+                   lock->client_pid, lkowner_utoa(&lock->owner),
+                   lock->user_flock.l_start, lock->user_flock.l_len);
+            __insert_and_merge(pl_inode, lock);
+        } else if (can_block) {
+            if (pl_metalock_is_active(pl_inode)) {
+                __pl_queue_lock(pl_inode, lock);
+                pthread_mutex_unlock(&pl_inode->mutex);
+                ret = -2;
+                goto out;
+            }
+            gf_log(this->name, GF_LOG_TRACE,
+                   "%s (pid=%d) lk-owner:%s %" PRId64 " - %" PRId64
+                   " => Blocked",
+                   lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
+                   lock->client_pid, lkowner_utoa(&lock->owner),
+                   lock->user_flock.l_start, lock->user_flock.l_len);
+
+            pl_trace_block(this, lock->frame, NULL, NULL, F_SETLKW,
+                           &lock->user_flock, NULL);
+
+            lock->blocked = 1;
+            __insert_lock(pl_inode, lock);
+            ret = -1;
+        } else {
+            gf_log(this->name, GF_LOG_TRACE,
+                   "%s (pid=%d) lk-owner:%s %" PRId64 " - %" PRId64 " => NOK",
+                   lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
+                   lock->client_pid, lkowner_utoa(&lock->owner),
+                   lock->user_flock.l_start, lock->user_flock.l_len);
+            errno = EAGAIN;
+            ret = -1;
         }
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+
+    grant_blocked_locks(this, pl_inode);
+
+    do_blocked_rw(pl_inode);
+
+out:
+    return ret;
 }
 
+posix_lock_t *
+pl_getlk(pl_inode_t *pl_inode, posix_lock_t *lock)
+{
+    posix_lock_t *conf = first_conflicting_overlap(pl_inode, lock);
+    if (conf == NULL) {
+        lock->fl_type = F_UNLCK;
+        return lock;
+    }
 
-void
-grant_blocked_locks (xlator_t *this, pl_inode_t *pl_inode)
+    return conf;
+}
+
+gf_boolean_t
+pl_does_monkey_want_stuck_lock()
+{
+    long int monkey_unlock_rand = 0;
+    long int monkey_unlock_rand_rem = 0;
+
+    /* coverity[DC.WEAK_CRYPTO] */
+    monkey_unlock_rand = random();
+    monkey_unlock_rand_rem = monkey_unlock_rand % 100;
+    if (monkey_unlock_rand_rem == 0)
+        return _gf_true;
+    return _gf_false;
+}
+
+int
+pl_lock_preempt(pl_inode_t *pl_inode, posix_lock_t *reqlock)
 {
-        struct list_head granted_list;
-        posix_lock_t     *tmp = NULL;
-        posix_lock_t     *lock = NULL;
+    posix_lock_t *lock = NULL;
+    posix_lock_t *i = NULL;
+    pl_rw_req_t *rw = NULL;
+    pl_rw_req_t *itr = NULL;
+    struct list_head unwind_blist = {
+        0,
+    };
+    struct list_head unwind_rw_list = {
+        0,
+    };
+    int ret = 0;
+
+    INIT_LIST_HEAD(&unwind_blist);
+    INIT_LIST_HEAD(&unwind_rw_list);
+
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        /*
+            - go through the lock list
+            - remove all locks from different owners
+            - same owner locks will be added or substracted based on
+              the new request
+            - add the new lock
+        */
+        list_for_each_entry_safe(lock, i, &pl_inode->ext_list, list)
+        {
+            if (lock->blocked) {
+                list_del_init(&lock->list);
+                list_add(&lock->list, &unwind_blist);
+                continue;
+            }
+
+            if (locks_overlap(lock, reqlock)) {
+                if (same_owner(lock, reqlock))
+                    continue;
+
+                /* remove conflicting locks */
+                list_del_init(&lock->list);
+                __delete_lock(lock);
+                __destroy_lock(lock);
+            }
+        }
 
-        INIT_LIST_HEAD (&granted_list);
+        __insert_and_merge(pl_inode, reqlock);
 
-        pthread_mutex_lock (&pl_inode->mutex);
+        list_for_each_entry_safe(rw, itr, &pl_inode->rw_list, list)
         {
-                __grant_blocked_locks (this, pl_inode, &granted_list);
+            list_del_init(&rw->list);
+            list_add(&rw->list, &unwind_rw_list);
         }
-        pthread_mutex_unlock (&pl_inode->mutex);
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+
+    /* unwind blocked locks */
+    list_for_each_entry_safe(lock, i, &unwind_blist, list)
+    {
+        PL_STACK_UNWIND_AND_FREE(((pl_local_t *)lock->frame->local), lk,
+                                 lock->frame, -1, EBUSY, &lock->user_flock,
+                                 NULL);
+        __destroy_lock(lock);
+    }
+
+    /* unwind blocked IOs */
+    list_for_each_entry_safe(rw, itr, &unwind_rw_list, list)
+    {
+        pl_clean_local(rw->stub->frame->local);
+        call_unwind_error(rw->stub, -1, EBUSY);
+    }
+
+    return ret;
+}
+
+/* Return true in case we need to ensure mandatory-locking
+ * semantics under different modes.
+ */
+gf_boolean_t
+pl_is_mandatory_locking_enabled(pl_inode_t *pl_inode)
+{
+    posix_locks_private_t *priv = THIS->private;
 
-        list_for_each_entry_safe (lock, tmp, &granted_list, list) {
-                list_del_init (&lock->list);
+    if (priv->mandatory_mode == MLK_FILE_BASED && pl_inode->mandatory)
+        return _gf_true;
+    else if (priv->mandatory_mode == MLK_FORCED ||
+             priv->mandatory_mode == MLK_OPTIMAL)
+        return _gf_true;
 
-                pl_trace_out (this, lock->frame, NULL, NULL, F_SETLKW,
-                              &lock->user_flock, 0, 0, NULL);
+    return _gf_false;
+}
+
+void
+pl_clean_local(pl_local_t *local)
+{
+    if (!local)
+        return;
 
-                STACK_UNWIND (lock->frame, 0, 0, &lock->user_flock);
+    if (local->inodelk_dom_count_req)
+        data_unref(local->inodelk_dom_count_req);
+    loc_wipe(&local->loc[0]);
+    loc_wipe(&local->loc[1]);
+    if (local->fd)
+        fd_unref(local->fd);
+    if (local->inode)
+        inode_unref(local->inode);
+    mem_put(local);
+}
 
-                GF_FREE (lock);
+/*
+TODO: detach local initialization from PL_LOCAL_GET_REQUESTS and add it here
+*/
+int
+pl_local_init(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd)
+{
+    pl_local_t *local = NULL;
+
+    if (!loc && !fd) {
+        return -1;
+    }
+
+    if (!frame->local) {
+        local = mem_get0(this->local_pool);
+        if (!local) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, 0,
+                   "mem allocation failed");
+            return -1;
         }
 
-        return;
+        local->inode = (loc ? inode_ref(loc->inode) : inode_ref(fd->inode));
+
+        frame->local = local;
+    }
+
+    return 0;
 }
 
-static int
-pl_send_prelock_unlock (xlator_t *this, pl_inode_t *pl_inode,
-                        posix_lock_t *old_lock)
+gf_boolean_t
+pl_is_lk_owner_valid(gf_lkowner_t *owner, client_t *client)
 {
-        struct gf_flock  flock       = {0,};
-        posix_lock_t *unlock_lock = NULL;
+    if (client && (client->opversion < GD_OP_VERSION_7_0)) {
+        return _gf_true;
+    }
+
+    if (is_lk_owner_null(owner)) {
+        return _gf_false;
+    }
+    return _gf_true;
+}
 
-        struct list_head granted_list;
-        posix_lock_t     *tmp = NULL;
-        posix_lock_t     *lock = NULL;
+static int32_t
+pl_inode_from_loc(loc_t *loc, inode_t **pinode)
+{
+    inode_t *inode = NULL;
+    int32_t error = 0;
+
+    if (loc->inode != NULL) {
+        inode = inode_ref(loc->inode);
+        goto done;
+    }
+
+    if (loc->parent == NULL) {
+        error = EINVAL;
+        goto done;
+    }
+
+    if (!gf_uuid_is_null(loc->gfid)) {
+        inode = inode_find(loc->parent->table, loc->gfid);
+        if (inode != NULL) {
+            goto done;
+        }
+    }
 
-        int ret = 0;
+    if (loc->name == NULL) {
+        error = EINVAL;
+        goto done;
+    }
 
-        INIT_LIST_HEAD (&granted_list);
+    inode = inode_grep(loc->parent->table, loc->parent, loc->name);
+    if (inode == NULL) {
+        /* We haven't found any inode. This means that the file doesn't exist
+         * or that even if it exists, we don't have any knowledge about it, so
+         * we don't have locks on it either, which is fine for our purposes. */
+        goto done;
+    }
 
-        flock.l_type   = F_UNLCK;
-        flock.l_whence = old_lock->user_flock.l_whence;
-        flock.l_start  = old_lock->user_flock.l_start;
-        flock.l_len    = old_lock->user_flock.l_len;
+done:
+    *pinode = inode;
 
+    return error;
+}
 
-        unlock_lock = new_posix_lock (&flock, old_lock->transport,
-                                      old_lock->client_pid, old_lock->owner,
-                                      old_lock->fd);
-        if (!unlock_lock) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory");
-                ret = -1;
-                goto out;
+static gf_boolean_t
+pl_inode_has_owners(xlator_t *xl, client_t *client, pl_inode_t *pl_inode,
+                    struct timespec *now, struct list_head *contend)
+{
+    pl_dom_list_t *dom;
+    pl_inode_lock_t *lock;
+    gf_boolean_t has_owners = _gf_false;
+
+    list_for_each_entry(dom, &pl_inode->dom_list, inode_list)
+    {
+        list_for_each_entry(lock, &dom->inodelk_list, list)
+        {
+            /* If the lock belongs to the same client, we assume it's related
+             * to the same operation, so we allow the removal to continue. */
+            if (lock->client == client) {
+                continue;
+            }
+            /* If the lock belongs to an internal process, we don't block the
+             * removal. */
+            if (lock->client_pid < 0) {
+                continue;
+            }
+            if (contend == NULL) {
+                return _gf_true;
+            }
+            has_owners = _gf_true;
+            inodelk_contention_notify_check(xl, lock, now, contend);
         }
+    }
+
+    return has_owners;
+}
+
+int32_t
+pl_inode_remove_prepare(xlator_t *xl, call_frame_t *frame, loc_t *loc,
+                        pl_inode_t **ppl_inode, struct list_head *contend)
+{
+    struct timespec now;
+    inode_t *inode;
+    pl_inode_t *pl_inode;
+    int32_t error;
+
+    pl_inode = NULL;
 
-        __insert_and_merge (pl_inode, unlock_lock);
+    error = pl_inode_from_loc(loc, &inode);
+    if ((error != 0) || (inode == NULL)) {
+        goto done;
+    }
 
-        __grant_blocked_locks (this, pl_inode, &granted_list);
+    pl_inode = pl_inode_get(xl, inode, NULL);
+    if (pl_inode == NULL) {
+        inode_unref(inode);
+        error = ENOMEM;
+        goto done;
+    }
 
-        list_for_each_entry_safe (lock, tmp, &granted_list, list) {
-                list_del_init (&lock->list);
+    /* pl_inode_from_loc() already increments ref count for inode, so
+     * we only assign here our reference. */
+    pl_inode->inode = inode;
 
-                pl_trace_out (this, lock->frame, NULL, NULL, F_SETLKW,
-                              &lock->user_flock, 0, 0, NULL);
+    timespec_now(&now);
 
-                STACK_UNWIND (lock->frame, 0, 0, &lock->user_flock);
+    pthread_mutex_lock(&pl_inode->mutex);
+
+    if (pl_inode->removed) {
+        error = ESTALE;
+        goto unlock;
+    }
+
+    if (pl_inode_has_owners(xl, frame->root->client, pl_inode, &now, contend)) {
+        error = -1;
+        /* We skip the unlock here because the caller must create a stub when
+         * we return -1 and do a call to pl_inode_remove_complete(), which
+         * assumes the lock is still acquired and will release it once
+         * everything else is prepared. */
+        goto done;
+    }
+
+    pl_inode->is_locked = _gf_true;
+    pl_inode->remove_running++;
+
+unlock:
+    pthread_mutex_unlock(&pl_inode->mutex);
+
+done:
+    *ppl_inode = pl_inode;
+
+    return error;
+}
 
-                GF_FREE (lock);
+int32_t
+pl_inode_remove_complete(xlator_t *xl, pl_inode_t *pl_inode, call_stub_t *stub,
+                         struct list_head *contend)
+{
+    pl_inode_lock_t *lock;
+    int32_t error = -1;
+
+    if (stub != NULL) {
+        list_add_tail(&stub->list, &pl_inode->waiting);
+        pl_inode->is_locked = _gf_true;
+    } else {
+        error = ENOMEM;
+
+        while (!list_empty(contend)) {
+            lock = list_first_entry(contend, pl_inode_lock_t, list);
+            list_del_init(&lock->list);
+            __pl_inodelk_unref(lock);
         }
+    }
 
-out:
+    pthread_mutex_unlock(&pl_inode->mutex);
+
+    if (error < 0) {
+        inodelk_contention_notify(xl, contend);
+    }
+
+    inode_unref(pl_inode->inode);
 
-        return ret;
+    return error;
 }
 
-int
-pl_setlk (xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock,
-          int can_block)
+void
+pl_inode_remove_wake(struct list_head *list)
+{
+    call_stub_t *stub;
+
+    while (!list_empty(list)) {
+        stub = list_first_entry(list, call_stub_t, list);
+        list_del_init(&stub->list);
+
+        call_resume(stub);
+    }
+}
+
+void
+pl_inode_remove_cbk(xlator_t *xl, pl_inode_t *pl_inode, int32_t error)
 {
-        int              ret = 0;
+    struct list_head contend, granted;
+    struct timespec now;
+    pl_dom_list_t *dom;
+
+    if (pl_inode == NULL) {
+        return;
+    }
+
+    INIT_LIST_HEAD(&contend);
+    INIT_LIST_HEAD(&granted);
+    timespec_now(&now);
+
+    pthread_mutex_lock(&pl_inode->mutex);
 
-        errno = 0;
+    if (error == 0) {
+        if (pl_inode->links >= 0) {
+            pl_inode->links--;
+        }
+        if (pl_inode->links == 0) {
+            pl_inode->removed = _gf_true;
+        }
+    }
+
+    pl_inode->remove_running--;
+
+    if ((pl_inode->remove_running == 0) && list_empty(&pl_inode->waiting)) {
+        pl_inode->is_locked = _gf_false;
 
-        pthread_mutex_lock (&pl_inode->mutex);
+        list_for_each_entry(dom, &pl_inode->dom_list, inode_list)
         {
-                /* Send unlock before the actual lock to
-                   prevent lock upgrade / downgrade
-                   problems
-                */
-
-                ret = pl_send_prelock_unlock (this, pl_inode,
-                                              lock);
-                if (ret)
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Could not send pre-lock "
-                                "unlock");
-
-                if (__is_lock_grantable (pl_inode, lock)) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "%s (pid=%d) lk-owner:%"PRIu64" %"PRId64" - %"PRId64" => OK",
-                                lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
-                                lock->client_pid,
-                                lock->owner,
-                                lock->user_flock.l_start,
-                                lock->user_flock.l_len);
-                        __insert_and_merge (pl_inode, lock);
-                } else if (can_block) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "%s (pid=%d) lk-owner:%"PRIu64" %"PRId64" - %"PRId64" => Blocked",
-                                lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
-                                lock->client_pid,
-                                lock->owner,
-                                lock->user_flock.l_start,
-                                lock->user_flock.l_len);
-                        lock->blocked = 1;
-                        __insert_lock (pl_inode, lock);
-                        ret = -1;
-                } else {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "%s (pid=%d) lk-owner:%"PRIu64" %"PRId64" - %"PRId64" => NOK",
-                                lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
-                                lock->client_pid,
-                                lock->owner,
-                                lock->user_flock.l_start,
-                                lock->user_flock.l_len);
-                        errno = EAGAIN;
-                        ret = -1;
-                }
+            __grant_blocked_inode_locks(xl, pl_inode, &granted, dom, &now,
+                                        &contend);
         }
-        pthread_mutex_unlock (&pl_inode->mutex);
+    }
 
-        grant_blocked_locks (this, pl_inode);
+    pthread_mutex_unlock(&pl_inode->mutex);
 
-        do_blocked_rw (pl_inode);
+    unwind_granted_inodes(xl, pl_inode, &granted);
 
-        return ret;
-}
+    inodelk_contention_notify(xl, &contend);
 
+    inode_unref(pl_inode->inode);
+}
 
-posix_lock_t *
-pl_getlk (pl_inode_t *pl_inode, posix_lock_t *lock)
+void
+pl_inode_remove_unlocked(xlator_t *xl, pl_inode_t *pl_inode,
+                         struct list_head *list)
 {
-        posix_lock_t *conf = NULL;
+    call_stub_t *stub, *tmp;
+
+    if (!pl_inode->is_locked) {
+        return;
+    }
 
-        conf = first_overlap (pl_inode, lock);
+    list_for_each_entry_safe(stub, tmp, &pl_inode->waiting, list)
+    {
+        if (!pl_inode_has_owners(xl, stub->frame->root->client, pl_inode, NULL,
+                                 NULL)) {
+            list_move_tail(&stub->list, list);
+        }
+    }
+}
 
-        if (conf == NULL) {
-                lock->fl_type = F_UNLCK;
-                return lock;
+/* This function determines if an inodelk attempt can be done now or it needs
+ * to wait.
+ *
+ * Possible return values:
+ *   < 0: An error occurred. Currently only -ESTALE can be returned if the
+ *        inode has been deleted previously by unlink/rmdir/rename
+ *   = 0: The lock can be attempted.
+ *   > 0: The lock needs to wait because a conflicting remove operation is
+ *        ongoing.
+ */
+int32_t
+pl_inode_remove_inodelk(pl_inode_t *pl_inode, pl_inode_lock_t *lock)
+{
+    pl_dom_list_t *dom;
+    pl_inode_lock_t *ilock;
+
+    /* If the inode has been deleted, we won't allow any lock. */
+    if (pl_inode->removed) {
+        return -ESTALE;
+    }
+
+    /* We only synchronize with locks made for regular operations coming from
+     * the user. Locks done for internal purposes are hard to control and could
+     * lead to long delays or deadlocks quite easily. */
+    if (lock->client_pid < 0) {
+        return 0;
+    }
+    if (!pl_inode->is_locked) {
+        return 0;
+    }
+    if (pl_inode->remove_running > 0) {
+        return 1;
+    }
+
+    list_for_each_entry(dom, &pl_inode->dom_list, inode_list)
+    {
+        list_for_each_entry(ilock, &dom->inodelk_list, list)
+        {
+            /* If a lock from the same client is already granted, we allow this
+             * one to continue. This is necessary to prevent deadlocks when
+             * multiple locks are taken for the same operation.
+             *
+             * On the other side it's unlikely that the same client sends
+             * completely unrelated locks for the same inode.
+             */
+            if (ilock->client == lock->client) {
+                return 0;
+            }
         }
+    }
 
-        return conf;
+    return 1;
 }
diff --git a/xlators/features/locks/src/common.h b/xlators/features/locks/src/common.h
index b13f44bc278..281223bf3b8 100644
--- a/xlators/features/locks/src/common.h
+++ b/xlators/features/locks/src/common.h
@@ -1,133 +1,262 @@
 /*
-  Copyright (c) 2006, 2007, 2008 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
+   Copyright (c) 2006-2012, 2015-2016 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
 #ifndef __COMMON_H__
 #define __COMMON_H__
 
+/*dump locks format strings */
+#define RANGE_FMT "type=%s, whence=%hd, start=%llu, len=%llu"
+#define ENTRY_FMT "type=%s on basename=%s"
+#define DUMP_GEN_FMT "pid = %llu, owner=%s, client=%p"
+#define GRNTD_AT "granted at %s"
+#define BLKD_AT "blocked at %s"
+#define CONN_ID "connection-id=%s"
+#define DUMP_BLKD_FMT DUMP_GEN_FMT ", " CONN_ID ", " BLKD_AT
+#define DUMP_GRNTD_FMT DUMP_GEN_FMT ", " CONN_ID ", " GRNTD_AT
+#define DUMP_BLKD_GRNTD_FMT DUMP_GEN_FMT ", " CONN_ID ", " BLKD_AT ", " GRNTD_AT
+
+#define ENTRY_BLKD_FMT ENTRY_FMT ", " DUMP_BLKD_FMT
+#define ENTRY_GRNTD_FMT ENTRY_FMT ", " DUMP_GRNTD_FMT
+#define ENTRY_BLKD_GRNTD_FMT ENTRY_FMT ", " DUMP_BLKD_GRNTD_FMT
+
+#define RANGE_BLKD_FMT RANGE_FMT ", " DUMP_BLKD_FMT
+#define RANGE_GRNTD_FMT RANGE_FMT ", " DUMP_GRNTD_FMT
+#define RANGE_BLKD_GRNTD_FMT RANGE_FMT ", " DUMP_BLKD_GRNTD_FMT
+
+#define SET_FLOCK_PID(flock, lock) ((flock)->l_pid = lock->client_pid)
+
+#define PL_STACK_UNWIND_AND_FREE(__local, fop, frame, op_ret, params...)       \
+    do {                                                                       \
+        frame->local = NULL;                                                   \
+        STACK_UNWIND_STRICT(fop, frame, op_ret, params);                       \
+        if (__local) {                                                         \
+            if (__local->inodelk_dom_count_req)                                \
+                data_unref(__local->inodelk_dom_count_req);                    \
+            loc_wipe(&__local->loc[0]);                                        \
+            loc_wipe(&__local->loc[1]);                                        \
+            if (__local->fd)                                                   \
+                fd_unref(__local->fd);                                         \
+            if (__local->inode)                                                \
+                inode_unref(__local->inode);                                   \
+            if (__local->xdata) {                                              \
+                dict_unref(__local->xdata);                                    \
+                __local->xdata = NULL;                                         \
+            }                                                                  \
+            mem_put(__local);                                                  \
+        }                                                                      \
+    } while (0)
+
 posix_lock_t *
-new_posix_lock (struct gf_flock *flock, void *transport, pid_t client_pid,
-                uint64_t owner, fd_t *fd);
+new_posix_lock(struct gf_flock *flock, client_t *client, pid_t client_pid,
+               gf_lkowner_t *owner, fd_t *fd, uint32_t lk_flags, int blocking,
+               int32_t *op_errno);
 
 pl_inode_t *
-pl_inode_get (xlator_t *this, inode_t *inode);
+pl_inode_get(xlator_t *this, inode_t *inode, pl_local_t *local);
 
 posix_lock_t *
-pl_getlk (pl_inode_t *inode, posix_lock_t *lock);
+pl_getlk(pl_inode_t *inode, posix_lock_t *lock);
+
+int
+pl_setlk(xlator_t *this, pl_inode_t *inode, posix_lock_t *lock, int can_block);
 
 int
-pl_setlk (xlator_t *this, pl_inode_t *inode, posix_lock_t *lock,
-          int can_block);
+pl_lock_preempt(pl_inode_t *pl_inode, posix_lock_t *reqlock);
 
 void
-grant_blocked_locks (xlator_t *this, pl_inode_t *inode);
+grant_blocked_locks(xlator_t *this, pl_inode_t *inode);
 
 void
-posix_lock_to_flock (posix_lock_t *lock, struct gf_flock *flock);
+posix_lock_to_flock(posix_lock_t *lock, struct gf_flock *flock);
 
 int
-locks_overlap (posix_lock_t *l1, posix_lock_t *l2);
+locks_overlap(posix_lock_t *l1, posix_lock_t *l2);
 
 int
-same_owner (posix_lock_t *l1, posix_lock_t *l2);
+same_owner(posix_lock_t *l1, posix_lock_t *l2);
 
-void __delete_lock (pl_inode_t *, posix_lock_t *);
+void
+__delete_lock(posix_lock_t *);
 
-void __destroy_lock (posix_lock_t *);
+void
+__destroy_lock(posix_lock_t *);
 
 pl_dom_list_t *
-get_domain (pl_inode_t *pl_inode, const char *volume);
+get_domain(pl_inode_t *pl_inode, const char *volume);
 
 void
-grant_blocked_inode_locks (xlator_t *this, pl_inode_t *pl_inode, pl_dom_list_t *dom);
+grant_blocked_inode_locks(xlator_t *this, pl_inode_t *pl_inode,
+                          pl_dom_list_t *dom, struct timespec *now,
+                          struct list_head *contend);
 
 void
-__delete_inode_lock (pl_inode_lock_t *lock);
+inodelk_contention_notify(xlator_t *this, struct list_head *contend);
 
 void
-__destroy_inode_lock (pl_inode_lock_t *lock);
+__delete_inode_lock(pl_inode_lock_t *lock);
 
 void
-grant_blocked_entry_locks (xlator_t *this, pl_inode_t *pl_inode,
-                           pl_entry_lock_t *unlocked, pl_dom_list_t *dom);
+__pl_inodelk_unref(pl_inode_lock_t *lock);
 
-void pl_update_refkeeper (xlator_t *this, inode_t *inode);
+void
+__grant_blocked_inode_locks(xlator_t *this, pl_inode_t *pl_inode,
+                            struct list_head *granted, pl_dom_list_t *dom,
+                            struct timespec *now, struct list_head *contend);
 
+void
+unwind_granted_inodes(xlator_t *this, pl_inode_t *pl_inode,
+                      struct list_head *granted);
+
+void
+grant_blocked_entry_locks(xlator_t *this, pl_inode_t *pl_inode,
+                          pl_dom_list_t *dom, struct timespec *now,
+                          struct list_head *contend);
+
+void
+entrylk_contention_notify(xlator_t *this, struct list_head *contend);
+
+void
+pl_update_refkeeper(xlator_t *this, inode_t *inode);
+
+int32_t
+__get_inodelk_count(xlator_t *this, pl_inode_t *pl_inode, char *domname);
 int32_t
-get_inodelk_count (xlator_t *this, inode_t *inode);
+get_inodelk_count(xlator_t *this, inode_t *inode, char *domname);
 
 int32_t
-get_entrylk_count (xlator_t *this, inode_t *inode);
+__get_entrylk_count(xlator_t *this, pl_inode_t *pl_inode);
+int32_t
+get_entrylk_count(xlator_t *this, inode_t *inode);
 
-void pl_trace_in (xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc,
-                  int cmd, struct gf_flock *flock, const char *domain);
+void
+pl_trace_in(xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc, int cmd,
+            struct gf_flock *flock, const char *domain);
 
-void pl_trace_out (xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc,
-                   int cmd, struct gf_flock *flock, int op_ret, int op_errno, const char *domain);
+void
+pl_trace_out(xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc, int cmd,
+             struct gf_flock *flock, int op_ret, int op_errno,
+             const char *domain);
 
-void pl_trace_block (xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc,
-                     int cmd, struct gf_flock *flock, const char *domain);
+void
+pl_trace_block(xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc,
+               int cmd, struct gf_flock *flock, const char *domain);
 
-void pl_trace_flush (xlator_t *this, call_frame_t *frame, fd_t *fd);
+void
+pl_trace_flush(xlator_t *this, call_frame_t *frame, fd_t *fd);
 
-void entrylk_trace_in (xlator_t *this, call_frame_t *frame, const char *volume,
-                       fd_t *fd, loc_t *loc, const char *basename,
-                       entrylk_cmd cmd, entrylk_type type);
+void
+entrylk_trace_in(xlator_t *this, call_frame_t *frame, const char *volume,
+                 fd_t *fd, loc_t *loc, const char *basename, entrylk_cmd cmd,
+                 entrylk_type type);
 
-void entrylk_trace_out (xlator_t *this, call_frame_t *frame, const char *volume,
-                        fd_t *fd, loc_t *loc, const char *basename,
-                        entrylk_cmd cmd, entrylk_type type,
-                        int op_ret, int op_errno);
+void
+entrylk_trace_out(xlator_t *this, call_frame_t *frame, const char *volume,
+                  fd_t *fd, loc_t *loc, const char *basename, entrylk_cmd cmd,
+                  entrylk_type type, int op_ret, int op_errno);
 
-void entrylk_trace_block (xlator_t *this, call_frame_t *frame, const char *volume,
-                          fd_t *fd, loc_t *loc, const char *basename,
-                          entrylk_cmd cmd, entrylk_type type);
+void
+entrylk_trace_block(xlator_t *this, call_frame_t *frame, const char *volume,
+                    fd_t *fd, loc_t *loc, const char *basename, entrylk_cmd cmd,
+                    entrylk_type type);
 
 void
-pl_print_verdict (char *str, int size, int op_ret, int op_errno);
+pl_print_verdict(char *str, int size, int op_ret, int op_errno);
 
 void
-pl_print_lockee (char *str, int size, fd_t *fd, loc_t *loc);
+pl_print_lockee(char *str, int size, fd_t *fd, loc_t *loc);
 
 void
-pl_print_locker (char *str, int size, xlator_t *this, call_frame_t *frame);
+pl_print_locker(char *str, int size, xlator_t *this, call_frame_t *frame);
 
 void
-pl_print_inodelk (char *str, int size, int cmd, struct gf_flock *flock, const char *domain);
+pl_print_inodelk(char *str, int size, int cmd, struct gf_flock *flock,
+                 const char *domain);
 
 void
-pl_trace_release (xlator_t *this, fd_t *fd);
+pl_trace_release(xlator_t *this, fd_t *fd);
 
 unsigned long
-fd_to_fdnum (fd_t *fd);
+fd_to_fdnum(fd_t *fd);
 
 fd_t *
-fd_from_fdnum (posix_lock_t *lock);
+fd_from_fdnum(posix_lock_t *lock);
 
 int
-pl_reserve_setlk (xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock,
-                  int can_block);
+pl_reserve_setlk(xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock,
+                 int can_block);
 int
-reservelks_equal (posix_lock_t *l1, posix_lock_t *l2);
+reservelks_equal(posix_lock_t *l1, posix_lock_t *l2);
+
+int
+pl_verify_reservelk(xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock,
+                    int can_block);
+int
+pl_reserve_unlock(xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *reqlock);
+
+int32_t
+check_entrylk_on_basename(xlator_t *this, inode_t *parent, char *basename);
+
+void
+__pl_inodelk_unref(pl_inode_lock_t *lock);
+void
+__pl_entrylk_unref(pl_entry_lock_t *lock);
 
 int
-pl_verify_reservelk (xlator_t *this, pl_inode_t *pl_inode,
-                     posix_lock_t *lock, int can_block);
+pl_metalock_is_active(pl_inode_t *pl_inode);
+
+void
+__pl_queue_lock(pl_inode_t *pl_inode, posix_lock_t *reqlock);
+
+void
+inodelk_contention_notify_check(xlator_t *xl, pl_inode_lock_t *lock,
+                                struct timespec *now,
+                                struct list_head *contend);
+
+void
+entrylk_contention_notify_check(xlator_t *xl, pl_entry_lock_t *lock,
+                                struct timespec *now,
+                                struct list_head *contend);
+
+gf_boolean_t
+pl_does_monkey_want_stuck_lock();
+
+gf_boolean_t
+pl_is_mandatory_locking_enabled(pl_inode_t *pl_inode);
+
+void
+pl_clean_local(pl_local_t *local);
+
 int
-pl_reserve_unlock (xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *reqlock);
+pl_local_init(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd);
+
+gf_boolean_t
+pl_is_lk_owner_valid(gf_lkowner_t *owner, client_t *client);
+
+int32_t
+pl_inode_remove_prepare(xlator_t *xl, call_frame_t *frame, loc_t *loc,
+                        pl_inode_t **ppl_inode, struct list_head *contend);
+
+int32_t
+pl_inode_remove_complete(xlator_t *xl, pl_inode_t *pl_inode, call_stub_t *stub,
+                         struct list_head *contend);
+
+void
+pl_inode_remove_wake(struct list_head *list);
+
+void
+pl_inode_remove_cbk(xlator_t *xl, pl_inode_t *pl_inode, int32_t error);
+
+void
+pl_inode_remove_unlocked(xlator_t *xl, pl_inode_t *pl_inode,
+                         struct list_head *list);
+
+int32_t
+pl_inode_remove_inodelk(pl_inode_t *pl_inode, pl_inode_lock_t *lock);
+
 #endif /* __COMMON_H__ */
diff --git a/xlators/features/locks/src/entrylk.c b/xlators/features/locks/src/entrylk.c
index 827b87d20d1..fd772c850dd 100644
--- a/xlators/features/locks/src/entrylk.c
+++ b/xlators/features/locks/src/entrylk.c
@@ -1,66 +1,83 @@
 /*
-  Copyright (c) 2006, 2007, 2008 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+   Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
-#include "glusterfs.h"
-#include "compat.h"
-#include "xlator.h"
-#include "inode.h"
-#include "logging.h"
-#include "common-utils.h"
-#include "list.h"
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/list.h>
+#include <glusterfs/upcall-utils.h>
 
 #include "locks.h"
+#include "clear.h"
 #include "common.h"
+#include "pl-messages.h"
 
-static pl_entry_lock_t *
-new_entrylk_lock (pl_inode_t *pinode, const char *basename, entrylk_type type,
-		  void *trans, pid_t client_pid, uint64_t owner, const char *volume)
-
+void
+__pl_entrylk_unref(pl_entry_lock_t *lock)
 {
-	pl_entry_lock_t *newlock = NULL;
-
-	newlock = GF_CALLOC (1, sizeof (pl_entry_lock_t),
-                             gf_locks_mt_pl_entry_lock_t);
-	if (!newlock) {
-		goto out;
-	}
-
-	newlock->basename       = basename ? gf_strdup (basename) : NULL;
-	newlock->type           = type;
-	newlock->trans          = trans;
-	newlock->volume         = volume;
-        newlock->client_pid     = client_pid;
-        newlock->owner          = owner;
+    lock->ref--;
+    if (!lock->ref) {
+        GF_FREE((char *)lock->basename);
+        GF_FREE(lock->connection_id);
+        GF_FREE(lock);
+    }
+}
 
-	INIT_LIST_HEAD (&newlock->domain_list);
-	INIT_LIST_HEAD (&newlock->blocked_locks);
+static void
+__pl_entrylk_ref(pl_entry_lock_t *lock)
+{
+    lock->ref++;
+}
 
+static pl_entry_lock_t *
+new_entrylk_lock(pl_inode_t *pinode, const char *basename, entrylk_type type,
+                 const char *domain, call_frame_t *frame, char *conn_id,
+                 int32_t *op_errno)
+{
+    pl_entry_lock_t *newlock = NULL;
+
+    if (!pl_is_lk_owner_valid(&frame->root->lk_owner, frame->root->client)) {
+        *op_errno = EINVAL;
+        goto out;
+    }
+
+    newlock = GF_CALLOC(1, sizeof(pl_entry_lock_t),
+                        gf_locks_mt_pl_entry_lock_t);
+    if (!newlock) {
+        *op_errno = ENOMEM;
+        goto out;
+    }
+
+    newlock->basename = basename ? gf_strdup(basename) : NULL;
+    newlock->type = type;
+    newlock->client = frame->root->client;
+    newlock->client_pid = frame->root->pid;
+    newlock->volume = domain;
+    newlock->owner = frame->root->lk_owner;
+    newlock->frame = frame;
+    newlock->this = frame->this;
+
+    if (conn_id) {
+        newlock->connection_id = gf_strdup(conn_id);
+    }
+
+    INIT_LIST_HEAD(&newlock->domain_list);
+    INIT_LIST_HEAD(&newlock->blocked_locks);
+    INIT_LIST_HEAD(&newlock->client_list);
+
+    __pl_entrylk_ref(newlock);
 out:
-	return newlock;
+    return newlock;
 }
 
-
 /**
  * all_names - does a basename represent all names?
  * @basename: name to check
@@ -75,206 +92,411 @@ out:
  */
 
 static int
-names_conflict (const char *n1, const char *n2)
+names_conflict(const char *n1, const char *n2)
 {
-	return all_names (n1) || all_names (n2) || !strcmp (n1, n2);
+    return all_names(n1) || all_names(n2) || !strcmp(n1, n2);
 }
 
+static int
+__same_entrylk_owner(pl_entry_lock_t *l1, pl_entry_lock_t *l2)
+{
+    return (is_same_lkowner(&l1->owner, &l2->owner) &&
+            (l1->client == l2->client));
+}
 
+/* Just as in inodelk, allow conflicting name locks from same (lk_owner, conn)*/
 static int
-__same_entrylk_owner (pl_entry_lock_t *l1, pl_entry_lock_t *l2)
+__conflicting_entrylks(pl_entry_lock_t *l1, pl_entry_lock_t *l2)
+{
+    if (names_conflict(l1->basename, l2->basename) &&
+        !__same_entrylk_owner(l1, l2))
+        return 1;
+
+    return 0;
+}
+
+/* See comments in inodelk.c for details */
+static inline gf_boolean_t
+__stale_entrylk(xlator_t *this, pl_entry_lock_t *candidate_lock,
+                pl_entry_lock_t *requested_lock, time_t *lock_age_sec)
+{
+    posix_locks_private_t *priv = NULL;
+
+    priv = this->private;
+
+    /* Question: Should we just prune them all given the
+     * chance?  Or just the locks we are attempting to acquire?
+     */
+    if (names_conflict(candidate_lock->basename, requested_lock->basename)) {
+        *lock_age_sec = gf_time() - candidate_lock->granted_time;
+        if (*lock_age_sec > priv->revocation_secs)
+            return _gf_true;
+    }
+    return _gf_false;
+}
+
+/* See comments in inodelk.c for details */
+static gf_boolean_t
+__entrylk_prune_stale(xlator_t *this, pl_inode_t *pinode, pl_dom_list_t *dom,
+                      pl_entry_lock_t *lock)
 {
+    posix_locks_private_t *priv = NULL;
+    pl_entry_lock_t *tmp = NULL;
+    pl_entry_lock_t *lk = NULL;
+    gf_boolean_t revoke_lock = _gf_false;
+    int bcount = 0;
+    int gcount = 0;
+    int op_errno = 0;
+    clrlk_args args;
+    args.opts = NULL;
+    time_t lk_age_sec = 0;
+    uint32_t max_blocked = 0;
+    char *reason_str = NULL;
+
+    priv = this->private;
+    args.type = CLRLK_ENTRY;
+    if (priv->revocation_clear_all == _gf_true)
+        args.kind = CLRLK_ALL;
+    else
+        args.kind = CLRLK_GRANTED;
+
+    if (list_empty(&dom->entrylk_list))
+        goto out;
+
+    pthread_mutex_lock(&pinode->mutex);
+    lock->pinode = pinode;
+    list_for_each_entry_safe(lk, tmp, &dom->entrylk_list, domain_list)
+    {
+        if (__stale_entrylk(this, lk, lock, &lk_age_sec) == _gf_true) {
+            revoke_lock = _gf_true;
+            reason_str = "age";
+            break;
+        }
+    }
+    max_blocked = priv->revocation_max_blocked;
+    if (max_blocked != 0 && revoke_lock == _gf_false) {
+        list_for_each_entry_safe(lk, tmp, &dom->blocked_entrylks, blocked_locks)
+        {
+            max_blocked--;
+            if (max_blocked == 0) {
+                revoke_lock = _gf_true;
+                reason_str = "max blocked";
+                break;
+            }
+        }
+    }
+    pthread_mutex_unlock(&pinode->mutex);
 
-        return ((l1->owner == l2->owner) &&
-		(l1->trans  == l2->trans));
+out:
+    if (revoke_lock == _gf_true) {
+        clrlk_clear_entrylk(this, pinode, dom, &args, &bcount, &gcount,
+                            &op_errno);
+        gf_log(this->name, GF_LOG_WARNING,
+               "Lock revocation [reason: %s; gfid: %s; domain: %s; "
+               "age: %ld sec] - Entry lock revoked:  %d granted & %d "
+               "blocked locks cleared",
+               reason_str, uuid_utoa(pinode->gfid), dom->domain, lk_age_sec,
+               gcount, bcount);
+    }
+
+    return revoke_lock;
+}
+
+void
+entrylk_contention_notify_check(xlator_t *this, pl_entry_lock_t *lock,
+                                struct timespec *now, struct list_head *contend)
+{
+    posix_locks_private_t *priv;
+    int64_t elapsed;
+
+    priv = this->private;
+
+    /* If this lock is in a list, it means that we are about to send a
+     * notification for it, so no need to do anything else. */
+    if (!list_empty(&lock->contend)) {
+        return;
+    }
+
+    elapsed = now->tv_sec;
+    elapsed -= lock->contention_time.tv_sec;
+    if (now->tv_nsec < lock->contention_time.tv_nsec) {
+        elapsed--;
+    }
+    if (elapsed < priv->notify_contention_delay) {
+        return;
+    }
+
+    /* All contention notifications will be sent outside of the locked
+     * region. This means that currently granted locks might have already
+     * been unlocked by that time. To avoid the lock or the inode to be
+     * destroyed before we process them, we take an additional reference
+     * on both. */
+    inode_ref(lock->pinode->inode);
+    __pl_entrylk_ref(lock);
+
+    lock->contention_time = *now;
+
+    list_add_tail(&lock->contend, contend);
 }
 
+void
+entrylk_contention_notify(xlator_t *this, struct list_head *contend)
+{
+    struct gf_upcall up;
+    struct gf_upcall_entrylk_contention lc;
+    pl_entry_lock_t *lock;
+    pl_inode_t *pl_inode;
+    client_t *client;
+    gf_boolean_t notify;
+
+    while (!list_empty(contend)) {
+        lock = list_first_entry(contend, pl_entry_lock_t, contend);
+
+        pl_inode = lock->pinode;
+
+        pthread_mutex_lock(&pl_inode->mutex);
+
+        /* If the lock has already been released, no notification is
+         * sent. We clear the notification time in this case. */
+        notify = !list_empty(&lock->domain_list);
+        if (!notify) {
+            lock->contention_time.tv_sec = 0;
+            lock->contention_time.tv_nsec = 0;
+        } else {
+            lc.type = lock->type;
+            lc.name = lock->basename;
+            lc.pid = lock->client_pid;
+            lc.domain = lock->volume;
+            lc.xdata = NULL;
+
+            gf_uuid_copy(up.gfid, lock->pinode->gfid);
+            client = (client_t *)lock->client;
+            if (client == NULL) {
+                /* A NULL client can be found if the entrylk
+                 * was issued by a server side xlator. */
+                up.client_uid = NULL;
+            } else {
+                up.client_uid = client->client_uid;
+            }
+        }
+
+        pthread_mutex_unlock(&pl_inode->mutex);
+
+        if (notify) {
+            up.event_type = GF_UPCALL_ENTRYLK_CONTENTION;
+            up.data = &lc;
+
+            if (this->notify(this, GF_EVENT_UPCALL, &up) < 0) {
+                gf_msg_debug(this->name, 0,
+                             "Entrylk contention notification "
+                             "failed");
+            } else {
+                gf_msg_debug(this->name, 0,
+                             "Entrylk contention notification "
+                             "sent");
+            }
+        }
+
+        pthread_mutex_lock(&pl_inode->mutex);
+
+        list_del_init(&lock->contend);
+        __pl_entrylk_unref(lock);
+
+        pthread_mutex_unlock(&pl_inode->mutex);
+
+        inode_unref(pl_inode->inode);
+    }
+}
 
 /**
- * lock_grantable - is this lock grantable?
+ * entrylk_grantable - is this lock grantable?
  * @inode: inode in which to look
  * @basename: name we're trying to lock
  * @type: type of lock
  */
 static pl_entry_lock_t *
-__lock_grantable (pl_dom_list_t *dom, const char *basename, entrylk_type type)
+__entrylk_grantable(xlator_t *this, pl_dom_list_t *dom, pl_entry_lock_t *lock,
+                    struct timespec *now, struct list_head *contend)
 {
-	pl_entry_lock_t *lock = NULL;
-
-	if (list_empty (&dom->entrylk_list))
-		return NULL;
-
-	list_for_each_entry (lock, &dom->entrylk_list, domain_list) {
-		if (names_conflict (lock->basename, basename))
-			return lock;
-	}
+    pl_entry_lock_t *tmp = NULL;
+    pl_entry_lock_t *ret = NULL;
+
+    list_for_each_entry(tmp, &dom->entrylk_list, domain_list)
+    {
+        if (__conflicting_entrylks(tmp, lock)) {
+            if (ret == NULL) {
+                ret = tmp;
+                if (contend == NULL) {
+                    break;
+                }
+            }
+            entrylk_contention_notify_check(this, tmp, now, contend);
+        }
+    }
 
-	return NULL;
+    return ret;
 }
 
 static pl_entry_lock_t *
-__blocked_lock_conflict (pl_dom_list_t *dom, const char *basename, entrylk_type type)
+__blocked_entrylk_conflict(pl_dom_list_t *dom, pl_entry_lock_t *lock)
 {
-	pl_entry_lock_t *lock = NULL;
+    pl_entry_lock_t *tmp = NULL;
 
-	if (list_empty (&dom->blocked_entrylks))
-		return NULL;
+    list_for_each_entry(tmp, &dom->blocked_entrylks, blocked_locks)
+    {
+        if (names_conflict(tmp->basename, lock->basename))
+            return lock;
+    }
 
-	list_for_each_entry (lock, &dom->blocked_entrylks, blocked_locks) {
-		if (names_conflict (lock->basename, basename))
-			return lock;
-	}
-
-	return NULL;
+    return NULL;
 }
 
 static int
-__owner_has_lock (pl_dom_list_t *dom, pl_entry_lock_t *newlock)
+__owner_has_lock(pl_dom_list_t *dom, pl_entry_lock_t *newlock)
 {
-        pl_entry_lock_t *lock = NULL;
+    pl_entry_lock_t *lock = NULL;
 
-	list_for_each_entry (lock, &dom->entrylk_list, domain_list) {
-		if (__same_entrylk_owner (lock, newlock))
-			return 1;
-	}
+    list_for_each_entry(lock, &dom->entrylk_list, domain_list)
+    {
+        if (__same_entrylk_owner(lock, newlock))
+            return 1;
+    }
 
-	list_for_each_entry (lock, &dom->blocked_entrylks, blocked_locks) {
-		if (__same_entrylk_owner (lock, newlock))
-			return 1;
-	}
+    list_for_each_entry(lock, &dom->blocked_entrylks, blocked_locks)
+    {
+        if (__same_entrylk_owner(lock, newlock))
+            return 1;
+    }
 
-	return 0;
+    return 0;
 }
 
 static int
-names_equal (const char *n1, const char *n2)
+names_equal(const char *n1, const char *n2)
 {
-	return (n1 == NULL && n2 == NULL) || (n1 && n2 && !strcmp (n1, n2));
+    return (n1 == NULL && n2 == NULL) || (n1 && n2 && !strcmp(n1, n2));
 }
 
 void
-pl_print_entrylk (char *str, int size, entrylk_cmd cmd, entrylk_type type,
-                  const char *basename, const char *domain)
+pl_print_entrylk(char *str, int size, entrylk_cmd cmd, entrylk_type type,
+                 const char *basename, const char *domain)
 {
-        char *cmd_str = NULL;
-        char *type_str = NULL;
+    char *cmd_str = NULL;
+    char *type_str = NULL;
 
-        switch (cmd) {
+    switch (cmd) {
         case ENTRYLK_LOCK:
-                cmd_str = "LOCK";
-                break;
+            cmd_str = "LOCK";
+            break;
 
         case ENTRYLK_LOCK_NB:
-                cmd_str = "LOCK_NB";
-                break;
+            cmd_str = "LOCK_NB";
+            break;
 
         case ENTRYLK_UNLOCK:
-                cmd_str = "UNLOCK";
-                break;
+            cmd_str = "UNLOCK";
+            break;
 
         default:
-                cmd_str = "UNKNOWN";
-                break;
-        }
+            cmd_str = "UNKNOWN";
+            break;
+    }
 
-        switch (type) {
+    switch (type) {
         case ENTRYLK_RDLCK:
-                type_str = "READ";
-                break;
+            type_str = "READ";
+            break;
         case ENTRYLK_WRLCK:
-                type_str = "WRITE";
-                break;
+            type_str = "WRITE";
+            break;
         default:
-                type_str = "UNKNOWN";
-                break;
-        }
+            type_str = "UNKNOWN";
+            break;
+    }
 
-        snprintf (str, size, "lock=ENTRYLK, cmd=%s, type=%s, basename=%s, domain: %s",
-                  cmd_str, type_str, basename, domain);
+    snprintf(str, size,
+             "lock=ENTRYLK, cmd=%s, type=%s, basename=%s, domain: %s", cmd_str,
+             type_str, basename, domain);
 }
 
-
 void
-entrylk_trace_in (xlator_t *this, call_frame_t *frame, const char *domain,
-                  fd_t *fd, loc_t *loc, const char *basename,
-                  entrylk_cmd cmd, entrylk_type type)
+entrylk_trace_in(xlator_t *this, call_frame_t *frame, const char *domain,
+                 fd_t *fd, loc_t *loc, const char *basename, entrylk_cmd cmd,
+                 entrylk_type type)
 {
-        posix_locks_private_t  *priv = NULL;
-        char                    pl_locker[256];
-        char                    pl_lockee[256];
-        char                    pl_entrylk[256];
+    posix_locks_private_t *priv = NULL;
+    char pl_locker[256];
+    char pl_lockee[256];
+    char pl_entrylk[256];
 
-        priv = this->private;
+    priv = this->private;
 
-        if (!priv->trace)
-                return;
+    if (!priv->trace)
+        return;
 
-        pl_print_locker (pl_locker, 256, this, frame);
-        pl_print_lockee (pl_lockee, 256, fd, loc);
-        pl_print_entrylk (pl_entrylk, 256, cmd, type, basename, domain);
+    pl_print_locker(pl_locker, 256, this, frame);
+    pl_print_lockee(pl_lockee, 256, fd, loc);
+    pl_print_entrylk(pl_entrylk, 256, cmd, type, basename, domain);
 
-        gf_log (this->name, GF_LOG_NORMAL,
-                "[REQUEST] Locker = {%s} Lockee = {%s} Lock = {%s}",
-                pl_locker, pl_lockee, pl_entrylk);
+    gf_log(this->name, GF_LOG_INFO,
+           "[REQUEST] Locker = {%s} Lockee = {%s} Lock = {%s}", pl_locker,
+           pl_lockee, pl_entrylk);
 }
 
-
 void
-entrylk_trace_out (xlator_t *this, call_frame_t *frame, const char *domain,
-                   fd_t *fd, loc_t *loc, const char *basename,
-                   entrylk_cmd cmd, entrylk_type type, int op_ret, int op_errno)
+entrylk_trace_out(xlator_t *this, call_frame_t *frame, const char *domain,
+                  fd_t *fd, loc_t *loc, const char *basename, entrylk_cmd cmd,
+                  entrylk_type type, int op_ret, int op_errno)
 {
-        posix_locks_private_t  *priv = NULL;
-        char                    pl_locker[256];
-        char                    pl_lockee[256];
-        char                    pl_entrylk[256];
-        char                    verdict[32];
+    posix_locks_private_t *priv = NULL;
+    char pl_locker[256];
+    char pl_lockee[256];
+    char pl_entrylk[256];
+    char verdict[32];
 
-        priv = this->private;
+    priv = this->private;
 
-        if (!priv->trace)
-                return;
+    if (!priv->trace)
+        return;
 
-        pl_print_locker (pl_locker, 256, this, frame);
-        pl_print_lockee (pl_lockee, 256, fd, loc);
-        pl_print_entrylk (pl_entrylk, 256, cmd, type, basename, domain);
-        pl_print_verdict (verdict, 32, op_ret, op_errno);
+    pl_print_locker(pl_locker, 256, this, frame);
+    pl_print_lockee(pl_lockee, 256, fd, loc);
+    pl_print_entrylk(pl_entrylk, 256, cmd, type, basename, domain);
+    pl_print_verdict(verdict, 32, op_ret, op_errno);
 
-        gf_log (this->name, GF_LOG_NORMAL,
-                "[%s] Locker = {%s} Lockee = {%s} Lock = {%s}",
-                verdict, pl_locker, pl_lockee, pl_entrylk);
+    gf_log(this->name, GF_LOG_INFO,
+           "[%s] Locker = {%s} Lockee = {%s} Lock = {%s}", verdict, pl_locker,
+           pl_lockee, pl_entrylk);
 }
 
-
 void
-entrylk_trace_block (xlator_t *this, call_frame_t *frame, const char *volume,
-                     fd_t *fd, loc_t *loc, const char *basename,
-                     entrylk_cmd cmd, entrylk_type type)
+entrylk_trace_block(xlator_t *this, call_frame_t *frame, const char *volume,
+                    fd_t *fd, loc_t *loc, const char *basename, entrylk_cmd cmd,
+                    entrylk_type type)
 
 {
-        posix_locks_private_t  *priv = NULL;
-        char                    pl_locker[256];
-        char                    pl_lockee[256];
-        char                    pl_entrylk[256];
+    posix_locks_private_t *priv = NULL;
+    char pl_locker[256];
+    char pl_lockee[256];
+    char pl_entrylk[256];
 
-        priv = this->private;
+    priv = this->private;
 
-        if (!priv->trace)
-                return;
+    if (!priv->trace)
+        return;
 
-        pl_print_locker (pl_locker, 256, this, frame);
-        pl_print_lockee (pl_lockee, 256, fd, loc);
-        pl_print_entrylk (pl_entrylk, 256, cmd, type, basename, volume);
+    pl_print_locker(pl_locker, 256, this, frame);
+    pl_print_lockee(pl_lockee, 256, fd, loc);
+    pl_print_entrylk(pl_entrylk, 256, cmd, type, basename, volume);
 
-        gf_log (this->name, GF_LOG_NORMAL,
-                "[BLOCKED] Locker = {%s} Lockee = {%s} Lock = {%s}",
-                pl_locker, pl_lockee, pl_entrylk);
+    gf_log(this->name, GF_LOG_INFO,
+           "[BLOCKED] Locker = {%s} Lockee = {%s} Lock = {%s}", pl_locker,
+           pl_lockee, pl_entrylk);
 }
 
 /**
- * __find_most_matching_lock - find the lock struct which most matches in order of:
- *                           lock on the exact basename ||
- *                           an all_names lock
+ * __find_most_matching_lock - find the lock struct which most matches in order
+ * of: lock on the exact basename || an all_names lock
  *
  *
  * @inode: inode in which to look
@@ -282,27 +504,61 @@ entrylk_trace_block (xlator_t *this, call_frame_t *frame, const char *volume,
  */
 
 static pl_entry_lock_t *
-__find_most_matching_lock (pl_dom_list_t *dom, const char *basename)
+__find_most_matching_lock(pl_dom_list_t *dom, const char *basename)
+{
+    pl_entry_lock_t *lock;
+    pl_entry_lock_t *all = NULL;
+    pl_entry_lock_t *exact = NULL;
+
+    if (list_empty(&dom->entrylk_list))
+        return NULL;
+
+    list_for_each_entry(lock, &dom->entrylk_list, domain_list)
+    {
+        if (all_names(lock->basename))
+            all = lock;
+        else if (names_equal(lock->basename, basename))
+            exact = lock;
+    }
+
+    return (exact ? exact : all);
+}
+
+static pl_entry_lock_t *
+__find_matching_lock(pl_dom_list_t *dom, pl_entry_lock_t *lock)
+{
+    pl_entry_lock_t *tmp = NULL;
+
+    list_for_each_entry(tmp, &dom->entrylk_list, domain_list)
+    {
+        if (names_equal(lock->basename, tmp->basename) &&
+            __same_entrylk_owner(lock, tmp) && (lock->type == tmp->type))
+            return tmp;
+    }
+    return NULL;
+}
+
+static int
+__lock_blocked_add(xlator_t *this, pl_inode_t *pinode, pl_dom_list_t *dom,
+                   pl_entry_lock_t *lock, int nonblock)
 {
-	pl_entry_lock_t *lock;
-	pl_entry_lock_t *all = NULL;
-	pl_entry_lock_t *exact = NULL;
+    if (nonblock)
+        goto out;
 
-	if (list_empty (&dom->entrylk_list))
-		return NULL;
+    lock->blkd_time = gf_time();
+    list_add_tail(&lock->blocked_locks, &dom->blocked_entrylks);
 
-	list_for_each_entry (lock, &dom->entrylk_list, domain_list) {
-		if (all_names (lock->basename))
-			all = lock;
-		else if (names_equal (lock->basename, basename))
-			exact = lock;
-	}
+    gf_msg_trace(this->name, 0, "Blocking lock: {pinode=%p, basename=%s}",
+                 pinode, lock->basename);
 
-	return (exact ? exact : all);
+    entrylk_trace_block(this, lock->frame, NULL, NULL, NULL, lock->basename,
+                        ENTRYLK_LOCK, lock->type);
+out:
+    return -EAGAIN;
 }
 
 /**
- * __lock_name - lock a name in a directory
+ * __lock_entrylk - lock a name in a directory
  * @inode: inode for the directory in which to lock
  * @basename: name of the entry to lock
  *            if null, lock the entire directory
@@ -313,416 +569,375 @@ __find_most_matching_lock (pl_dom_list_t *dom, const char *basename)
  */
 
 int
-__lock_name (pl_inode_t *pinode, const char *basename, entrylk_type type,
-	     call_frame_t *frame, pl_dom_list_t *dom, xlator_t *this, int nonblock)
-{
-	pl_entry_lock_t *lock       = NULL;
-	pl_entry_lock_t *conf       = NULL;
-	void            *trans      = NULL;
-        pid_t            client_pid = 0;
-        uint64_t         owner      = 0;
-
-	int ret = -EINVAL;
-
-	trans = frame->root->trans;
-        client_pid = frame->root->pid;
-        owner      = frame->root->lk_owner;
-
-	lock = new_entrylk_lock (pinode, basename, type, trans, client_pid, owner, dom->domain);
-	if (!lock) {
-		ret = -ENOMEM;
-		goto out;
-	}
-
-        lock->frame   = frame;
-        lock->this    = this;
-        lock->trans   = trans;
-
-	conf = __lock_grantable (dom, basename, type);
-	if (conf) {
-		ret = -EAGAIN;
-		if (nonblock){
-			if (lock->basename)
-				GF_FREE ((char *)lock->basename);
-                        GF_FREE (lock);
-                        goto out;
-
-                }
-
-		list_add_tail (&lock->blocked_locks, &dom->blocked_entrylks);
-
-		gf_log (this->name, GF_LOG_TRACE,
-			"Blocking lock: {pinode=%p, basename=%s}",
-			pinode, basename);
-
-		goto out;
-	}
-
-        if ( __blocked_lock_conflict (dom, basename, type) && !(__owner_has_lock (dom, lock))) {
-                ret = -EAGAIN;
-                if (nonblock) {
-			if (lock->basename)
-				GF_FREE ((char *) lock->basename);
-                        GF_FREE (lock);
-                        goto out;
-
-                }
-                lock->frame     = frame;
-                lock->this      = this;
-
-                list_add_tail (&lock->blocked_locks, &dom->blocked_entrylks);
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "Lock is grantable, but blocking to prevent starvation");
-		gf_log (this->name, GF_LOG_TRACE,
-			"Blocking lock: {pinode=%p, basename=%s}",
-			pinode, basename);
-
-                ret = -EAGAIN;
-		goto out;
+__lock_entrylk(xlator_t *this, pl_inode_t *pinode, pl_entry_lock_t *lock,
+               int nonblock, pl_dom_list_t *dom, struct timespec *now,
+               struct list_head *contend)
+{
+    pl_entry_lock_t *conf = NULL;
+    int ret = -EAGAIN;
+
+    conf = __entrylk_grantable(this, dom, lock, now, contend);
+    if (conf) {
+        ret = __lock_blocked_add(this, pinode, dom, lock, nonblock);
+        goto out;
+    }
+
+    /* To prevent blocked locks starvation, check if there are any blocked
+     * locks thay may conflict with this lock. If there is then don't grant
+     * the lock. BUT grant the lock if the owner already has lock to allow
+     * nested locks.
+     * Example: SHD from Machine1 takes (gfid, basename=257-length-name)
+     * and is granted.
+     * SHD from machine2 takes (gfid, basename=NULL) and is blocked.
+     * When SHD from Machine1 takes (gfid, basename=NULL) it needs to be
+     * granted, without which self-heal can't progress.
+     * TODO: Find why 'owner_has_lock' is checked even for blocked locks.
+     */
+    if (__blocked_entrylk_conflict(dom, lock) &&
+        !(__owner_has_lock(dom, lock))) {
+        if (nonblock == 0) {
+            gf_log(this->name, GF_LOG_DEBUG,
+                   "Lock is grantable, but blocking to prevent "
+                   "starvation");
         }
-        switch (type) {
 
-	case ENTRYLK_WRLCK:
-		list_add_tail (&lock->domain_list, &dom->entrylk_list);
-		break;
+        ret = __lock_blocked_add(this, pinode, dom, lock, nonblock);
+        goto out;
+    }
 
-        default:
+    __pl_entrylk_ref(lock);
+    lock->granted_time = gf_time();
+    list_add(&lock->domain_list, &dom->entrylk_list);
 
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Invalid type for entrylk specified: %d", type);
-                ret = -EINVAL;
-                goto out;
-	}
-
-	ret = 0;
+    ret = 0;
 out:
-	return ret;
+    return ret;
 }
 
 /**
- * __unlock_name - unlock a name in a directory
+ * __unlock_entrylk - unlock a name in a directory
  * @inode: inode for the directory to unlock in
  * @basename: name of the entry to unlock
  *            if null, unlock the entire directory
  */
 
 pl_entry_lock_t *
-__unlock_name (pl_dom_list_t *dom, const char *basename, entrylk_type type)
+__unlock_entrylk(pl_dom_list_t *dom, pl_entry_lock_t *lock)
 {
-	pl_entry_lock_t *lock = NULL;
-	pl_entry_lock_t *ret_lock = NULL;
+    pl_entry_lock_t *ret_lock = NULL;
 
-	lock = __find_most_matching_lock (dom, basename);
+    ret_lock = __find_matching_lock(dom, lock);
 
-	if (!lock) {
-		gf_log ("locks", GF_LOG_DEBUG,
-			"unlock on %s (type=ENTRYLK_WRLCK) attempted but no matching lock found",
-			basename);
-		goto out;
-	}
+    if (ret_lock) {
+        list_del_init(&ret_lock->domain_list);
+    } else {
+        gf_log("locks", GF_LOG_ERROR,
+               "unlock on %s "
+               "(type=ENTRYLK_WRLCK) attempted but no matching lock "
+               "found",
+               lock->basename);
+    }
 
-	if (names_equal (lock->basename, basename)
-	    && lock->type == type) {
+    return ret_lock;
+}
 
-		if (type == ENTRYLK_WRLCK) {
-			list_del_init (&lock->domain_list);
-			ret_lock = lock;
-		}
-	} else {
-		gf_log ("locks", GF_LOG_DEBUG,
-			"Unlock for a non-existing lock!");
-		goto out;
-	}
+int32_t
+check_entrylk_on_basename(xlator_t *this, inode_t *parent, char *basename)
+{
+    int32_t entrylk = 0;
+    pl_dom_list_t *dom = NULL;
+    pl_entry_lock_t *conf = NULL;
+
+    pl_inode_t *pinode = pl_inode_get(this, parent, NULL);
+    if (!pinode)
+        goto out;
+    pthread_mutex_lock(&pinode->mutex);
+    {
+        list_for_each_entry(dom, &pinode->dom_list, inode_list)
+        {
+            conf = __find_most_matching_lock(dom, basename);
+            if (conf && conf->basename) {
+                entrylk = 1;
+                break;
+            }
+        }
+    }
+    pthread_mutex_unlock(&pinode->mutex);
 
 out:
-	return ret_lock;
+    return entrylk;
 }
 
-
 void
-__grant_blocked_entry_locks (xlator_t *this, pl_inode_t *pl_inode,
-			     pl_dom_list_t *dom, struct list_head *granted)
+__grant_blocked_entry_locks(xlator_t *this, pl_inode_t *pl_inode,
+                            pl_dom_list_t *dom, struct list_head *granted,
+                            struct timespec *now, struct list_head *contend)
 {
-	int              bl_ret = 0;
-	pl_entry_lock_t *bl   = NULL;
-	pl_entry_lock_t *tmp  = NULL;
+    int bl_ret = 0;
+    pl_entry_lock_t *bl = NULL;
+    pl_entry_lock_t *tmp = NULL;
 
-        struct list_head blocked_list;
+    struct list_head blocked_list;
 
-        INIT_LIST_HEAD (&blocked_list);
-        list_splice_init (&dom->blocked_entrylks, &blocked_list);
-        
-	list_for_each_entry_safe (bl, tmp, &blocked_list,
-				  blocked_locks) {
+    INIT_LIST_HEAD(&blocked_list);
+    list_splice_init(&dom->blocked_entrylks, &blocked_list);
 
-		list_del_init (&bl->blocked_locks);
+    list_for_each_entry_safe(bl, tmp, &blocked_list, blocked_locks)
+    {
+        list_del_init(&bl->blocked_locks);
 
+        bl_ret = __lock_entrylk(bl->this, pl_inode, bl, 0, dom, now, contend);
 
-		gf_log ("locks", GF_LOG_TRACE,
-			"Trying to unblock: {pinode=%p, basename=%s}",
-			pl_inode, bl->basename);
-
-		bl_ret = __lock_name (pl_inode, bl->basename, bl->type,
-				      bl->frame, dom, bl->this, 0);
-
-		if (bl_ret == 0) {
-			list_add (&bl->blocked_locks, granted);
-		} else {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "should never happen");
-			if (bl->basename)
-				GF_FREE ((char *)bl->basename);
-			GF_FREE (bl);
-		}
-	}
-	return;
+        if (bl_ret == 0) {
+            list_add_tail(&bl->blocked_locks, granted);
+        }
+    }
 }
 
 /* Grants locks if possible which are blocked on a lock */
 void
-grant_blocked_entry_locks (xlator_t *this, pl_inode_t *pl_inode,
-			   pl_entry_lock_t *unlocked, pl_dom_list_t *dom)
+grant_blocked_entry_locks(xlator_t *this, pl_inode_t *pl_inode,
+                          pl_dom_list_t *dom, struct timespec *now,
+                          struct list_head *contend)
 {
-	struct list_head  granted_list;
-	pl_entry_lock_t  *tmp = NULL;
-	pl_entry_lock_t  *lock = NULL;
-
-	INIT_LIST_HEAD (&granted_list);
-
-	pthread_mutex_lock (&pl_inode->mutex);
-	{
-		__grant_blocked_entry_locks (this, pl_inode, dom, &granted_list);
-	}
-	pthread_mutex_unlock (&pl_inode->mutex);
-
-	list_for_each_entry_safe (lock, tmp, &granted_list, blocked_locks) {
-		list_del_init (&lock->blocked_locks);
-
-                entrylk_trace_out (this, lock->frame, NULL, NULL, NULL,
-                                   lock->basename, ENTRYLK_LOCK, lock->type,
-                                   0, 0);
-
-		STACK_UNWIND_STRICT (entrylk, lock->frame, 0, 0);
-
-	}
-
-	GF_FREE ((char *)unlocked->basename);
-	GF_FREE (unlocked);
-
-	return;
+    struct list_head granted_list;
+    pl_entry_lock_t *tmp = NULL;
+    pl_entry_lock_t *lock = NULL;
+
+    INIT_LIST_HEAD(&granted_list);
+
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        __grant_blocked_entry_locks(this, pl_inode, dom, &granted_list, now,
+                                    contend);
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+
+    list_for_each_entry_safe(lock, tmp, &granted_list, blocked_locks)
+    {
+        entrylk_trace_out(this, lock->frame, NULL, NULL, NULL, lock->basename,
+                          ENTRYLK_LOCK, lock->type, 0, 0);
+
+        STACK_UNWIND_STRICT(entrylk, lock->frame, 0, 0, NULL);
+        lock->frame = NULL;
+    }
+
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        list_for_each_entry_safe(lock, tmp, &granted_list, blocked_locks)
+        {
+            list_del_init(&lock->blocked_locks);
+            __pl_entrylk_unref(lock);
+        }
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
 }
 
-/**
- * release_entry_locks_for_transport: release all entry locks from this
- * transport for this loc_t
- */
+/* Common entrylk code called by pl_entrylk and pl_fentrylk */
+int
+pl_common_entrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+                  inode_t *inode, const char *basename, entrylk_cmd cmd,
+                  entrylk_type type, loc_t *loc, fd_t *fd, dict_t *xdata)
 
-static int
-release_entry_locks_for_transport (xlator_t *this, pl_inode_t *pinode,
-				   pl_dom_list_t *dom, void *trans)
 {
-	pl_entry_lock_t  *lock = NULL;
-	pl_entry_lock_t  *tmp = NULL;
-	struct list_head  granted;
-        struct list_head  released;
-
-	INIT_LIST_HEAD (&granted);
-        INIT_LIST_HEAD (&released);
-
-	pthread_mutex_lock (&pinode->mutex);
-	{
-                list_for_each_entry_safe (lock, tmp, &dom->blocked_entrylks,
-                                          blocked_locks) {
-                        if (lock->trans != trans)
-                                continue;
-
-                        list_del_init (&lock->blocked_locks);
-
-			gf_log (this->name, GF_LOG_TRACE,
-                                "releasing lock on  held by "
-                                "{transport=%p}",trans);
-
-                        list_add (&lock->blocked_locks, &released);
-
+    int32_t op_ret = -1;
+    int32_t op_errno = 0;
+    int ret = -1;
+    char unwind = 1;
+    GF_UNUSED int dict_ret = -1;
+    pl_inode_t *pinode = NULL;
+    pl_entry_lock_t *reqlock = NULL;
+    pl_entry_lock_t *unlocked = NULL;
+    pl_dom_list_t *dom = NULL;
+    char *conn_id = NULL;
+    pl_ctx_t *ctx = NULL;
+    int nonblock = 0;
+    gf_boolean_t need_inode_unref = _gf_false;
+    posix_locks_private_t *priv = NULL;
+    struct list_head *pcontend = NULL;
+    struct list_head contend;
+    struct timespec now = {};
+
+    priv = this->private;
+
+    if (priv->notify_contention) {
+        pcontend = &contend;
+        INIT_LIST_HEAD(pcontend);
+        timespec_now(&now);
+    }
+
+    if (xdata)
+        dict_ret = dict_get_str(xdata, "connection-id", &conn_id);
+
+    pinode = pl_inode_get(this, inode, NULL);
+    if (!pinode) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    if (frame->root->client) {
+        ctx = pl_ctx_get(frame->root->client, this);
+        if (!ctx) {
+            op_errno = ENOMEM;
+            gf_log(this->name, GF_LOG_INFO, "pl_ctx_get() failed");
+            goto unwind;
+        }
+    }
+
+    dom = get_domain(pinode, volume);
+    if (!dom) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    entrylk_trace_in(this, frame, volume, fd, loc, basename, cmd, type);
+
+    reqlock = new_entrylk_lock(pinode, basename, type, dom->domain, frame,
+                               conn_id, &op_errno);
+    if (!reqlock) {
+        op_ret = -1;
+        goto unwind;
+    }
+
+    /* Ideally, AFTER a successful lock (both blocking and non-blocking) or
+     * an unsuccessful blocking lock operation, the inode needs to be ref'd.
+     *
+     * But doing so might give room to a race where the lock-requesting
+     * client could send a DISCONNECT just before this thread refs the inode
+     * after the locking is done, and the epoll thread could unref the inode
+     * in cleanup which means the inode's refcount would come down to 0, and
+     * the call to pl_forget() at this point destroys @pinode. Now when
+     * the io-thread executing this function tries to access pinode,
+     * it could crash on account of illegal memory access.
+     *
+     * To get around this problem, the inode is ref'd once even before
+     * adding the lock into client_list as a precautionary measure.
+     * This way even if there are DISCONNECTs, there will always be 1 extra
+     * ref on the inode, so @pinode is still alive until after the
+     * current stack unwinds.
+     */
+    pinode->inode = inode_ref(inode);
+    if (priv->revocation_secs != 0) {
+        if (cmd != ENTRYLK_UNLOCK) {
+            __entrylk_prune_stale(this, pinode, dom, reqlock);
+        } else if (priv->monkey_unlocking == _gf_true) {
+            if (pl_does_monkey_want_stuck_lock()) {
+                gf_log(this->name, GF_LOG_WARNING,
+                       "MONKEY LOCKING (forcing stuck lock)!");
+                op_ret = 0;
+                need_inode_unref = _gf_true;
+                pthread_mutex_lock(&pinode->mutex);
+                {
+                    __pl_entrylk_unref(reqlock);
                 }
-
-		list_for_each_entry_safe (lock, tmp, &dom->entrylk_list,
-					  domain_list) {
-			if (lock->trans != trans)
-				continue;
-
-			list_del_init (&lock->domain_list);
-
-			gf_log (this->name, GF_LOG_TRACE,
-                                "releasing lock on  held by "
-                                "{transport=%p}",trans);
-
-			GF_FREE ((char *)lock->basename);
-			GF_FREE (lock);
-		}
-
-		__grant_blocked_entry_locks (this, pinode, dom, &granted);
-
-	}
-
-	pthread_mutex_unlock (&pinode->mutex);
-
-        list_for_each_entry_safe (lock, tmp, &released, blocked_locks) {
-                list_del_init (&lock->blocked_locks);
-
-                STACK_UNWIND_STRICT (entrylk, lock->frame, -1, EAGAIN);
-
-                if (lock->basename)
-                        GF_FREE ((char *)lock->basename);
-                GF_FREE (lock);
-
+                pthread_mutex_unlock(&pinode->mutex);
+                goto out;
+            }
         }
+    }
 
-	list_for_each_entry_safe (lock, tmp, &granted, blocked_locks) {
-		list_del_init (&lock->blocked_locks);
-
-		STACK_UNWIND_STRICT (entrylk, lock->frame, 0, 0);
-
-		if (lock->basename)
-			GF_FREE ((char *)lock->basename);
-		GF_FREE (lock);
-	}
+    switch (cmd) {
+        case ENTRYLK_LOCK_NB:
+            nonblock = 1;
+            /* fall through */
+        case ENTRYLK_LOCK:
+            if (ctx)
+                pthread_mutex_lock(&ctx->lock);
+            pthread_mutex_lock(&pinode->mutex);
+            {
+                reqlock->pinode = pinode;
+
+                ret = __lock_entrylk(this, pinode, reqlock, nonblock, dom, &now,
+                                     pcontend);
+                if (ret == 0) {
+                    reqlock->frame = NULL;
+                    op_ret = 0;
+                } else {
+                    op_errno = -ret;
+                }
 
-	return 0;
-}
+                if (ctx && (!ret || !nonblock))
+                    list_add(&reqlock->client_list, &ctx->entrylk_lockers);
 
-/* Common entrylk code called by pl_entrylk and pl_fentrylk */
-int
-pl_common_entrylk (call_frame_t *frame, xlator_t *this,
-                   const char *volume, inode_t *inode, const char *basename,
-                   entrylk_cmd cmd, entrylk_type type, loc_t *loc, fd_t *fd)
-{
-        uint64_t owner    = 0;
-	int32_t  op_ret   = -1;
-	int32_t  op_errno = 0;
-
-	void *        transport = NULL;
-	pid_t         pid       = -1;
-
-	pl_inode_t *     pinode   = NULL;
-	int              ret      = -1;
-	pl_entry_lock_t *unlocked = NULL;
-	char             unwind   = 1;
-
-	pl_dom_list_t	  *dom = NULL;
-
-	pinode = pl_inode_get (this, inode);
-	if (!pinode) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		op_errno = ENOMEM;
-		goto out;
-	}
-
-	dom = get_domain (pinode, volume);
-	if (!dom){
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory");
-		op_errno = ENOMEM;
-		goto out;
-	}
-
-        entrylk_trace_in (this, frame, volume, fd, loc, basename, cmd, type);
-
-	pid       = frame->root->pid;
-        owner     = frame->root->lk_owner;
-	transport = frame->root->trans;
-
-	if (owner == 0) {
-		/*
-                  this is a special case that means release
-                  all locks from this transport
-		*/
-
-		gf_log (this->name, GF_LOG_TRACE,
-			"Releasing locks for transport %p", transport);
-
-		release_entry_locks_for_transport (this, pinode, dom, transport);
-		op_ret = 0;
-
-		goto out;
-	}
-
-	switch (cmd) {
-	case ENTRYLK_LOCK:
-		pthread_mutex_lock (&pinode->mutex);
-		{
-			ret = __lock_name (pinode, basename, type,
-					   frame, dom, this, 0);
-		}
-		pthread_mutex_unlock (&pinode->mutex);
-
-                op_errno = -ret;
-		if (ret < 0) {
-                        if (ret == -EAGAIN)
-                                unwind = 0;
-                        else
-                                unwind = 1;
-			goto out;
-		} else {
-                        op_ret = 0;
-                        op_errno = 0;
-                        unwind = 1;
-                        goto out;
+                if (ret == -EAGAIN && !nonblock) {
+                    /* blocked */
+                    unwind = 0;
+                } else {
+                    __pl_entrylk_unref(reqlock);
                 }
 
-		break;
-
-	case ENTRYLK_LOCK_NB:
-		pthread_mutex_lock (&pinode->mutex);
-		{
-			ret = __lock_name (pinode, basename, type,
-					   frame, dom, this, 1);
-		}
-		pthread_mutex_unlock (&pinode->mutex);
+                /* For all but the case where a non-blocking lock
+                 * attempt fails, the extra ref taken before the switch
+                 * block must be negated.
+                 */
+                if ((ret == -EAGAIN) && (nonblock))
+                    need_inode_unref = _gf_true;
+            }
+            pthread_mutex_unlock(&pinode->mutex);
+            if (ctx)
+                pthread_mutex_unlock(&ctx->lock);
+            break;
 
-		if (ret < 0)
-			op_errno = -ret;
+        case ENTRYLK_UNLOCK:
+            if (ctx)
+                pthread_mutex_lock(&ctx->lock);
+            pthread_mutex_lock(&pinode->mutex);
+            {
+                /* Irrespective of whether unlock succeeds or not,
+                 * the extra inode ref that was done before the switch
+                 * block must be negated. Towards this,
+                 * @need_inode_unref flag is set unconditionally here.
+                 */
+                need_inode_unref = _gf_true;
+                unlocked = __unlock_entrylk(dom, reqlock);
+                if (unlocked) {
+                    list_del_init(&unlocked->client_list);
+                    __pl_entrylk_unref(unlocked);
+                    op_ret = 0;
+                } else {
+                    op_errno = EINVAL;
+                }
+                __pl_entrylk_unref(reqlock);
+            }
+            pthread_mutex_unlock(&pinode->mutex);
+            if (ctx)
+                pthread_mutex_unlock(&ctx->lock);
 
-                unwind = 1;
-                goto out;
-		break;
+            grant_blocked_entry_locks(this, pinode, dom, &now, pcontend);
 
-	case ENTRYLK_UNLOCK:
-		pthread_mutex_lock (&pinode->mutex);
-		{
-                        unlocked = __unlock_name (dom, basename, type);
-		}
-		pthread_mutex_unlock (&pinode->mutex);
+            break;
 
-		if (unlocked)
-			grant_blocked_entry_locks (this, pinode, unlocked, dom);
+        default:
+            need_inode_unref = _gf_true;
+            gf_log(this->name, GF_LOG_ERROR,
+                   "Unexpected case in entrylk (cmd=%d). Please file"
+                   "a bug report at http://bugs.gluster.com",
+                   cmd);
+            goto out;
+    }
+    /* The following (extra) unref corresponds to the ref that
+     * was done at the time the lock was granted.
+     */
+    if ((cmd == ENTRYLK_UNLOCK) && (op_ret == 0))
+        inode_unref(pinode->inode);
 
-		break;
+out:
 
-	default:
-		gf_log (this->name, GF_LOG_ERROR,
-			"Unexpected case in entrylk (cmd=%d). Please file"
-                        "a bug report at http://bugs.gluster.com", cmd);
-		goto out;
-	}
+    if (need_inode_unref)
+        inode_unref(pinode->inode);
 
-	op_ret = 0;
-out:
-        pl_update_refkeeper (this, inode);
-	if (unwind) {
-                entrylk_trace_out (this, frame, volume, fd, loc, basename,
-                                   cmd, type, op_ret, op_errno);
-
-		STACK_UNWIND_STRICT (entrylk, frame, op_ret, op_errno);
-	} else {
-                entrylk_trace_block (this, frame, volume, fd, loc, basename,
-                                     cmd, type);
-        }
+    if (unwind) {
+        entrylk_trace_out(this, frame, volume, fd, loc, basename, cmd, type,
+                          op_ret, op_errno);
+    unwind:
+        STACK_UNWIND_STRICT(entrylk, frame, op_ret, op_errno, NULL);
+    }
 
+    if (pcontend != NULL) {
+        entrylk_contention_notify(this, pcontend);
+    }
 
-	return 0;
+    return 0;
 }
 
 /**
@@ -732,17 +947,16 @@ out:
  */
 
 int
-pl_entrylk (call_frame_t *frame, xlator_t *this,
-	    const char *volume, loc_t *loc, const char *basename,
-	    entrylk_cmd cmd, entrylk_type type)
+pl_entrylk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+           const char *basename, entrylk_cmd cmd, entrylk_type type,
+           dict_t *xdata)
 {
+    pl_common_entrylk(frame, this, volume, loc->inode, basename, cmd, type, loc,
+                      NULL, xdata);
 
-        pl_common_entrylk (frame, this, volume, loc->inode, basename, cmd, type, loc, NULL);
-
-        return 0;
+    return 0;
 }
 
-
 /**
  * pl_fentrylk:
  *
@@ -750,73 +964,190 @@ pl_entrylk (call_frame_t *frame, xlator_t *this,
  */
 
 int
-pl_fentrylk (call_frame_t *frame, xlator_t *this,
-             const char *volume, fd_t *fd, const char *basename,
-             entrylk_cmd cmd, entrylk_type type)
+pl_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+            const char *basename, entrylk_cmd cmd, entrylk_type type,
+            dict_t *xdata)
 {
+    pl_common_entrylk(frame, this, volume, fd->inode, basename, cmd, type, NULL,
+                      fd, xdata);
 
-        pl_common_entrylk (frame, this, volume, fd->inode, basename, cmd, type, NULL, fd);
-
-        return 0;
+    return 0;
 }
 
-
-static int32_t
-__get_entrylk_count (xlator_t *this, pl_inode_t *pl_inode)
+static void
+pl_entrylk_log_cleanup(pl_entry_lock_t *lock)
 {
-        int32_t            count = 0;
-        pl_entry_lock_t   *lock  = NULL;
-        pl_dom_list_t     *dom   = NULL;
+    pl_inode_t *pinode = NULL;
 
-        list_for_each_entry (dom, &pl_inode->dom_list, inode_list) {
-                list_for_each_entry (lock, &dom->entrylk_list, domain_list) {
+    pinode = lock->pinode;
 
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                " XATTR DEBUG"
-                                " domain: %s  %s on %s state = Active",
-                                dom->domain,
-                                lock->type == ENTRYLK_RDLCK ? "ENTRYLK_RDLCK" : 
-                                "ENTRYLK_WRLCK", lock->basename);
-                        count++;
+    gf_log(THIS->name, GF_LOG_WARNING,
+           "releasing lock on %s held by "
+           "{client=%p, pid=%" PRId64 " lk-owner=%s}",
+           uuid_utoa(pinode->gfid), lock->client, (uint64_t)lock->client_pid,
+           lkowner_utoa(&lock->owner));
+}
+
+/* Release all entrylks from this client */
+int
+pl_entrylk_client_cleanup(xlator_t *this, pl_ctx_t *ctx)
+{
+    posix_locks_private_t *priv;
+    pl_entry_lock_t *tmp = NULL;
+    pl_entry_lock_t *l = NULL;
+    pl_dom_list_t *dom = NULL;
+    pl_inode_t *pinode = NULL;
+    struct list_head *pcontend = NULL;
+    struct list_head released;
+    struct list_head unwind;
+    struct list_head contend;
+    struct timespec now = {};
+
+    INIT_LIST_HEAD(&released);
+    INIT_LIST_HEAD(&unwind);
+
+    priv = this->private;
+    if (priv->notify_contention) {
+        pcontend = &contend;
+        INIT_LIST_HEAD(pcontend);
+        timespec_now(&now);
+    }
+
+    pthread_mutex_lock(&ctx->lock);
+    {
+        list_for_each_entry_safe(l, tmp, &ctx->entrylk_lockers, client_list)
+        {
+            pl_entrylk_log_cleanup(l);
+
+            pinode = l->pinode;
+
+            pthread_mutex_lock(&pinode->mutex);
+            {
+                /* If the entrylk object is part of granted list but not
+                 * blocked list, then perform the following actions:
+                 * i.   delete the object from granted list;
+                 * ii.  grant other locks (from other clients) that may
+                 *      have been blocked on this entrylk; and
+                 * iii. unref the object.
+                 *
+                 * If the entrylk object (L1) is part of both granted
+                 * and blocked lists, then this means that a parallel
+                 * unlock on another entrylk (L2 say) may have 'granted'
+                 * L1 and added it to 'granted' list in
+                 * __grant_blocked_entry_locks() (although using the
+                 * 'blocked_locks' member). In that case, the cleanup
+                 * codepath must try and grant other overlapping
+                 * blocked entrylks from other clients, now that L1 is
+                 * out of their way and then unref L1 in the end, and
+                 * leave it to the other thread (the one executing
+                 * unlock codepath) to unwind L1's frame, delete it from
+                 * blocked_locks list, and perform the last unref on L1.
+                 *
+                 * If the entrylk object (L1) is part of blocked list
+                 * only, the cleanup code path must:
+                 * i.   delete it from the blocked_locks list inside
+                 *      this critical section,
+                 * ii.  unwind its frame with EAGAIN,
+                 * iii. try and grant blocked entry locks from other
+                 *      clients that were otherwise grantable, but were
+                 *      blocked to avoid leaving L1 to starve forever.
+                 * iv.  unref the object.
+                 */
+                list_del_init(&l->client_list);
+
+                if (!list_empty(&l->domain_list)) {
+                    list_del_init(&l->domain_list);
+                    list_add_tail(&l->client_list, &released);
+                } else {
+                    list_del_init(&l->blocked_locks);
+                    list_add_tail(&l->client_list, &unwind);
                 }
+            }
+            pthread_mutex_unlock(&pinode->mutex);
+        }
+    }
+    pthread_mutex_unlock(&ctx->lock);
 
-                list_for_each_entry (lock, &dom->blocked_entrylks, blocked_locks) {
+    if (!list_empty(&unwind)) {
+        list_for_each_entry_safe(l, tmp, &unwind, client_list)
+        {
+            list_del_init(&l->client_list);
 
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                " XATTR DEBUG"
-                                " domain: %s  %s on %s state = Blocked",
-                                dom->domain,
-                                lock->type == ENTRYLK_RDLCK ? "ENTRYLK_RDLCK" : 
-                                "ENTRYLK_WRLCK", lock->basename);
-                        count++;
-                }
+            if (l->frame)
+                STACK_UNWIND_STRICT(entrylk, l->frame, -1, EAGAIN, NULL);
+            list_add_tail(&l->client_list, &released);
+        }
+    }
+
+    if (!list_empty(&released)) {
+        list_for_each_entry_safe(l, tmp, &released, client_list)
+        {
+            list_del_init(&l->client_list);
+
+            pinode = l->pinode;
 
+            dom = get_domain(pinode, l->volume);
+
+            grant_blocked_entry_locks(this, pinode, dom, &now, pcontend);
+
+            pthread_mutex_lock(&pinode->mutex);
+            {
+                __pl_entrylk_unref(l);
+            }
+            pthread_mutex_unlock(&pinode->mutex);
+
+            inode_unref(pinode->inode);
         }
+    }
 
-        return count;
+    if (pcontend != NULL) {
+        entrylk_contention_notify(this, pcontend);
+    }
+
+    return 0;
 }
 
 int32_t
-get_entrylk_count (xlator_t *this, inode_t *inode)
+__get_entrylk_count(xlator_t *this, pl_inode_t *pl_inode)
 {
-        pl_inode_t   *pl_inode = NULL;
-        uint64_t      tmp_pl_inode = 0;
-        int           ret      = 0;
-        int32_t       count    = 0;
+    int32_t count = 0;
+    pl_entry_lock_t *lock = NULL;
+    pl_dom_list_t *dom = NULL;
 
-        ret = inode_ctx_get (inode, this, &tmp_pl_inode);
-        if (ret != 0) {
-                goto out;
-        }
-
-        pl_inode = (pl_inode_t *)(long) tmp_pl_inode;
+    list_for_each_entry(dom, &pl_inode->dom_list, inode_list)
+    {
+        list_for_each_entry(lock, &dom->entrylk_list, domain_list) { count++; }
 
-        pthread_mutex_lock (&pl_inode->mutex);
+        list_for_each_entry(lock, &dom->blocked_entrylks, blocked_locks)
         {
-                count = __get_entrylk_count (this, pl_inode);
+            count++;
         }
-        pthread_mutex_unlock (&pl_inode->mutex);
+    }
+
+    return count;
+}
+
+int32_t
+get_entrylk_count(xlator_t *this, inode_t *inode)
+{
+    pl_inode_t *pl_inode = NULL;
+    uint64_t tmp_pl_inode = 0;
+    int ret = 0;
+    int32_t count = 0;
+
+    ret = inode_ctx_get(inode, this, &tmp_pl_inode);
+    if (ret != 0) {
+        goto out;
+    }
+
+    pl_inode = (pl_inode_t *)(long)tmp_pl_inode;
+
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        count = __get_entrylk_count(this, pl_inode);
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
 
 out:
-        return count;
+    return count;
 }
diff --git a/xlators/features/locks/src/inodelk.c b/xlators/features/locks/src/inodelk.c
index bbbd03398fe..d4e51d6e0a1 100644
--- a/xlators/features/locks/src/inodelk.c
+++ b/xlators/features/locks/src/inodelk.c
@@ -1,744 +1,1174 @@
 /*
-  Copyright (c) 2006, 2007, 2008 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+   Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
-#include "glusterfs.h"
-#include "compat.h"
-#include "xlator.h"
-#include "inode.h"
-#include "logging.h"
-#include "common-utils.h"
-#include "list.h"
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/dict.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/list.h>
+#include <glusterfs/upcall-utils.h>
 
 #include "locks.h"
+#include "clear.h"
 #include "common.h"
 
 void
-__delete_inode_lock (pl_inode_lock_t *lock)
+__delete_inode_lock(pl_inode_lock_t *lock)
 {
-	list_del (&lock->list);
+    list_del_init(&lock->list);
+}
+
+static void
+__pl_inodelk_ref(pl_inode_lock_t *lock)
+{
+    lock->ref++;
 }
 
 void
-__destroy_inode_lock (pl_inode_lock_t *lock)
+__pl_inodelk_unref(pl_inode_lock_t *lock)
 {
-	GF_FREE (lock);
+    lock->ref--;
+    if (!lock->ref) {
+        GF_FREE(lock->connection_id);
+        GF_FREE(lock);
+    }
 }
 
-/* Check if 2 inodelks are conflicting on type. Only 2 shared locks don't conflict */
+/* Check if 2 inodelks are conflicting on type. Only 2 shared locks don't
+ * conflict */
 static int
-inodelk_type_conflict (pl_inode_lock_t *l1, pl_inode_lock_t *l2)
+inodelk_type_conflict(pl_inode_lock_t *l1, pl_inode_lock_t *l2)
 {
-	if (l2->fl_type == F_WRLCK || l1->fl_type == F_WRLCK)
-		return 1;
+    if (l2->fl_type == F_WRLCK || l1->fl_type == F_WRLCK)
+        return 1;
 
-	return 0;
+    return 0;
 }
 
 void
-pl_print_inodelk (char *str, int size, int cmd, struct gf_flock *flock, const char *domain)
+pl_print_inodelk(char *str, int size, int cmd, struct gf_flock *flock,
+                 const char *domain)
 {
-        char *cmd_str = NULL;
-        char *type_str = NULL;
+    char *cmd_str = NULL;
+    char *type_str = NULL;
 
-        switch (cmd) {
+    switch (cmd) {
 #if F_GETLK != F_GETLK64
         case F_GETLK64:
 #endif
         case F_GETLK:
-                cmd_str = "GETLK";
-                break;
+            cmd_str = "GETLK";
+            break;
 
 #if F_SETLK != F_SETLK64
         case F_SETLK64:
 #endif
         case F_SETLK:
-                cmd_str = "SETLK";
-                break;
+            cmd_str = "SETLK";
+            break;
 
 #if F_SETLKW != F_SETLKW64
         case F_SETLKW64:
 #endif
         case F_SETLKW:
-                cmd_str = "SETLKW";
-                break;
+            cmd_str = "SETLKW";
+            break;
 
         default:
-                cmd_str = "UNKNOWN";
-                break;
-        }
+            cmd_str = "UNKNOWN";
+            break;
+    }
 
-        switch (flock->l_type) {
+    switch (flock->l_type) {
         case F_RDLCK:
-                type_str = "READ";
-                break;
+            type_str = "READ";
+            break;
         case F_WRLCK:
-                type_str = "WRITE";
-                break;
+            type_str = "WRITE";
+            break;
         case F_UNLCK:
-                type_str = "UNLOCK";
-                break;
+            type_str = "UNLOCK";
+            break;
         default:
-                type_str = "UNKNOWN";
-                break;
-        }
-
-        snprintf (str, size, "lock=INODELK, cmd=%s, type=%s, "
-                  "domain: %s, start=%llu, len=%llu, pid=%llu",
-                  cmd_str, type_str, domain,
-                  (unsigned long long) flock->l_start,
-                  (unsigned long long) flock->l_len,
-                  (unsigned long long) flock->l_pid);
+            type_str = "UNKNOWN";
+            break;
+    }
+
+    snprintf(str, size,
+             "lock=INODELK, cmd=%s, type=%s, "
+             "domain: %s, start=%llu, len=%llu, pid=%llu",
+             cmd_str, type_str, domain, (unsigned long long)flock->l_start,
+             (unsigned long long)flock->l_len,
+             (unsigned long long)flock->l_pid);
 }
 
 /* Determine if the two inodelks overlap reach other's lock regions */
 static int
-inodelk_overlap (pl_inode_lock_t *l1, pl_inode_lock_t *l2)
+inodelk_overlap(pl_inode_lock_t *l1, pl_inode_lock_t *l2)
 {
-	return ((l1->fl_end >= l2->fl_start) &&
-		(l2->fl_end >= l1->fl_start));
+    return ((l1->fl_end >= l2->fl_start) && (l2->fl_end >= l1->fl_start));
 }
 
 /* Returns true if the 2 inodelks have the same owner */
-static int same_inodelk_owner (pl_inode_lock_t *l1, pl_inode_lock_t *l2)
+static int
+same_inodelk_owner(pl_inode_lock_t *l1, pl_inode_lock_t *l2)
 {
-	return ((l1->owner == l2->owner) &&
-                (l1->transport  == l2->transport));
+    return (is_same_lkowner(&l1->owner, &l2->owner) &&
+            (l1->client == l2->client));
 }
 
 /* Returns true if the 2 inodelks conflict with each other */
 static int
-inodelk_conflict (pl_inode_lock_t *l1, pl_inode_lock_t *l2)
+inodelk_conflict(pl_inode_lock_t *l1, pl_inode_lock_t *l2)
 {
-        return (inodelk_overlap (l1, l2) &&
-                inodelk_type_conflict (l1, l2));
+    return (inodelk_overlap(l1, l2) && inodelk_type_conflict(l1, l2));
 }
 
-/* Determine if lock is grantable or not */
-static pl_inode_lock_t *
-__inodelk_grantable (pl_dom_list_t *dom, pl_inode_lock_t *lock)
-{
-	pl_inode_lock_t *l = NULL;
-	pl_inode_lock_t *ret = NULL;
-	if (list_empty (&dom->inodelk_list))
-		goto out;
-	list_for_each_entry (l, &dom->inodelk_list, list){
-		if (inodelk_conflict (lock, l)) {
-			ret = l;
-			goto out;
-		}
-	}
-out:
-	return ret;
+/*
+ * Check to see if the candidate lock overlaps/conflicts with the
+ * requested lock.  If so, determine how old the lock is and return
+ * true if it exceeds the configured threshold, false otherwise.
+ */
+static inline gf_boolean_t
+__stale_inodelk(xlator_t *this, pl_inode_lock_t *candidate_lock,
+                pl_inode_lock_t *requested_lock, time_t *lock_age_sec)
+{
+    posix_locks_private_t *priv = NULL;
+
+    priv = this->private;
+    /* Question: Should we just prune them all given the
+     * chance?  Or just the locks we are attempting to acquire?
+     */
+    if (inodelk_conflict(candidate_lock, requested_lock)) {
+        *lock_age_sec = gf_time() - candidate_lock->granted_time;
+        if (*lock_age_sec > priv->revocation_secs)
+            return _gf_true;
+    }
+    return _gf_false;
 }
 
-static pl_inode_lock_t *
-__blocked_lock_conflict (pl_dom_list_t *dom, pl_inode_lock_t *lock)
+/* Examine any locks held on this inode and potentially revoke the lock
+ * if the age exceeds revocation_secs.  We will clear _only_ those locks
+ * which are granted, and then grant those locks which are blocked.
+ *
+ * Depending on how this patch works in the wild, we may expand this and
+ * introduce a heuristic which clears blocked locks as well if they
+ * are beyond a threshold.
+ */
+static gf_boolean_t
+__inodelk_prune_stale(xlator_t *this, pl_inode_t *pinode, pl_dom_list_t *dom,
+                      pl_inode_lock_t *lock)
 {
-        pl_inode_lock_t *l   = NULL;
-        pl_inode_lock_t *ret = NULL;
-
-	if (list_empty (&dom->blocked_entrylks))
-		return NULL;
+    posix_locks_private_t *priv = NULL;
+    pl_inode_lock_t *tmp = NULL;
+    pl_inode_lock_t *lk = NULL;
+    gf_boolean_t revoke_lock = _gf_false;
+    int bcount = 0;
+    int gcount = 0;
+    int op_errno = 0;
+    clrlk_args args;
+    args.opts = NULL;
+    time_t lk_age_sec = 0;
+    uint32_t max_blocked = 0;
+    char *reason_str = NULL;
+
+    priv = this->private;
+
+    args.type = CLRLK_INODE;
+    if (priv->revocation_clear_all == _gf_true)
+        args.kind = CLRLK_ALL;
+    else
+        args.kind = CLRLK_GRANTED;
+
+    if (list_empty(&dom->inodelk_list))
+        goto out;
+
+    pthread_mutex_lock(&pinode->mutex);
+    list_for_each_entry_safe(lk, tmp, &dom->inodelk_list, list)
+    {
+        if (__stale_inodelk(this, lk, lock, &lk_age_sec) == _gf_true) {
+            revoke_lock = _gf_true;
+            reason_str = "age";
+            break;
+        }
+    }
 
-	list_for_each_entry (l, &dom->blocked_inodelks, blocked_locks) {
-		if (inodelk_conflict (lock, l)) {
-			ret = l;
-			goto out;
-                }
-	}
+    max_blocked = priv->revocation_max_blocked;
+    if (max_blocked != 0 && revoke_lock == _gf_false) {
+        list_for_each_entry_safe(lk, tmp, &dom->blocked_inodelks, blocked_locks)
+        {
+            max_blocked--;
+            if (max_blocked == 0) {
+                revoke_lock = _gf_true;
+                reason_str = "max blocked";
+                break;
+            }
+        }
+    }
+    pthread_mutex_unlock(&pinode->mutex);
 
 out:
-	return ret;
+    if (revoke_lock == _gf_true) {
+        clrlk_clear_inodelk(this, pinode, dom, &args, &bcount, &gcount,
+                            &op_errno);
+        gf_log(this->name, GF_LOG_WARNING,
+               "Lock revocation [reason: %s; gfid: %s; domain: %s; "
+               "age: %ld sec] - Inode lock revoked:  %d granted & %d "
+               "blocked locks cleared",
+               reason_str, uuid_utoa(pinode->gfid), dom->domain, lk_age_sec,
+               gcount, bcount);
+    }
+    return revoke_lock;
 }
 
-static int
-__owner_has_lock (pl_dom_list_t *dom, pl_inode_lock_t *newlock)
+void
+inodelk_contention_notify_check(xlator_t *this, pl_inode_lock_t *lock,
+                                struct timespec *now, struct list_head *contend)
 {
-        pl_inode_lock_t *lock = NULL;
+    posix_locks_private_t *priv;
+    int64_t elapsed;
+
+    priv = this->private;
+
+    /* If this lock is in a list, it means that we are about to send a
+     * notification for it, so no need to do anything else. */
+    if (!list_empty(&lock->contend)) {
+        return;
+    }
+
+    elapsed = now->tv_sec;
+    elapsed -= lock->contention_time.tv_sec;
+    if (now->tv_nsec < lock->contention_time.tv_nsec) {
+        elapsed--;
+    }
+    if (elapsed < priv->notify_contention_delay) {
+        return;
+    }
+
+    /* All contention notifications will be sent outside of the locked
+     * region. This means that currently granted locks might have already
+     * been unlocked by that time. To avoid the lock or the inode to be
+     * destroyed before we process them, we take an additional reference
+     * on both. */
+    inode_ref(lock->pl_inode->inode);
+    __pl_inodelk_ref(lock);
+
+    lock->contention_time = *now;
+
+    list_add_tail(&lock->contend, contend);
+}
 
-	list_for_each_entry (lock, &dom->entrylk_list, list) {
-		if (same_inodelk_owner (lock, newlock))
-			return 1;
-	}
+void
+inodelk_contention_notify(xlator_t *this, struct list_head *contend)
+{
+    struct gf_upcall up;
+    struct gf_upcall_inodelk_contention lc;
+    pl_inode_lock_t *lock;
+    pl_inode_t *pl_inode;
+    client_t *client;
+    gf_boolean_t notify;
+
+    while (!list_empty(contend)) {
+        lock = list_first_entry(contend, pl_inode_lock_t, contend);
+
+        pl_inode = lock->pl_inode;
+
+        pthread_mutex_lock(&pl_inode->mutex);
+
+        /* If the lock has already been released, no notification is
+         * sent. We clear the notification time in this case. */
+        notify = !list_empty(&lock->list);
+        if (!notify) {
+            lock->contention_time.tv_sec = 0;
+            lock->contention_time.tv_nsec = 0;
+        } else {
+            memcpy(&lc.flock, &lock->user_flock, sizeof(lc.flock));
+            lc.pid = lock->client_pid;
+            lc.domain = lock->volume;
+            lc.xdata = NULL;
+
+            gf_uuid_copy(up.gfid, lock->pl_inode->gfid);
+            client = (client_t *)lock->client;
+            if (client == NULL) {
+                /* A NULL client can be found if the inodelk
+                 * was issued by a server side xlator. */
+                up.client_uid = NULL;
+            } else {
+                up.client_uid = client->client_uid;
+            }
+        }
 
-	list_for_each_entry (lock, &dom->blocked_entrylks, blocked_locks) {
-		if (same_inodelk_owner (lock, newlock))
-			return 1;
-	}
+        pthread_mutex_unlock(&pl_inode->mutex);
+
+        if (notify) {
+            up.event_type = GF_UPCALL_INODELK_CONTENTION;
+            up.data = &lc;
+
+            if (this->notify(this, GF_EVENT_UPCALL, &up) < 0) {
+                gf_msg_debug(this->name, 0,
+                             "Inodelk contention notification "
+                             "failed");
+            } else {
+                gf_msg_debug(this->name, 0,
+                             "Inodelk contention notification "
+                             "sent");
+            }
+        }
 
-	return 0;
-}
+        pthread_mutex_lock(&pl_inode->mutex);
 
+        list_del_init(&lock->contend);
+        __pl_inodelk_unref(lock);
 
-/* Determines if lock can be granted and adds the lock. If the lock
- * is blocking, adds it to the blocked_inodelks list of the domain.
- */
-static int
-__lock_inodelk (xlator_t *this, pl_inode_t *pl_inode, pl_inode_lock_t *lock,
-		int can_block,  pl_dom_list_t *dom)
+        pthread_mutex_unlock(&pl_inode->mutex);
+
+        inode_unref(pl_inode->inode);
+    }
+}
+
+/* Determine if lock is grantable or not */
+static pl_inode_lock_t *
+__inodelk_grantable(xlator_t *this, pl_dom_list_t *dom, pl_inode_lock_t *lock,
+                    struct timespec *now, struct list_head *contend)
 {
-	pl_inode_lock_t *conf = NULL;
-	int ret = -EINVAL;
+    pl_inode_lock_t *l = NULL;
+    pl_inode_lock_t *ret = NULL;
+
+    list_for_each_entry(l, &dom->inodelk_list, list)
+    {
+        if (inodelk_conflict(lock, l) && !same_inodelk_owner(lock, l)) {
+            if (ret == NULL) {
+                ret = l;
+                if (contend == NULL) {
+                    break;
+                }
+            }
+            inodelk_contention_notify_check(this, l, now, contend);
+        }
+    }
 
-	conf = __inodelk_grantable (dom, lock);
-	if (conf){
-		ret = -EAGAIN;
-		if (can_block == 0)
-			goto out;
+    return ret;
+}
 
-		list_add_tail (&lock->blocked_locks, &dom->blocked_inodelks);
+static pl_inode_lock_t *
+__blocked_lock_conflict(pl_dom_list_t *dom, pl_inode_lock_t *lock)
+{
+    pl_inode_lock_t *l = NULL;
 
-                gf_log (this->name, GF_LOG_TRACE,
-                        "%s (pid=%d) lk-owner:%"PRIu64" %"PRId64" - %"PRId64" => Blocked",
-                        lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
-                        lock->client_pid,
-                        lock->owner,
-                        lock->user_flock.l_start,
-                        lock->user_flock.l_len);
+    list_for_each_entry(l, &dom->blocked_inodelks, blocked_locks)
+    {
+        if (inodelk_conflict(lock, l)) {
+            return l;
+        }
+    }
 
+    return NULL;
+}
 
-		goto out;
-	}
+static int
+__owner_has_lock(pl_dom_list_t *dom, pl_inode_lock_t *newlock)
+{
+    pl_inode_lock_t *lock = NULL;
 
-        if (__blocked_lock_conflict (dom, lock) && !(__owner_has_lock (dom, lock))) {
-                ret = -EAGAIN;
-                if (can_block == 0)
-                        goto out;
+    list_for_each_entry(lock, &dom->inodelk_list, list)
+    {
+        if (same_inodelk_owner(lock, newlock))
+            return 1;
+    }
 
-                list_add_tail (&lock->blocked_locks, &dom->blocked_inodelks);
+    list_for_each_entry(lock, &dom->blocked_inodelks, blocked_locks)
+    {
+        if (same_inodelk_owner(lock, newlock))
+            return 1;
+    }
 
-                gf_log (this->name, GF_LOG_TRACE,
-                        "Lock is grantable, but blocking to prevent starvation");
-		gf_log (this->name, GF_LOG_TRACE,
-                        "%s (pid=%d) (lk-owner=%"PRIu64") %"PRId64" - %"PRId64" => Blocked",
-                        lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
-                        lock->client_pid,
-                        lock->owner,
-                        lock->user_flock.l_start,
-                        lock->user_flock.l_len);
+    return 0;
+}
 
+static int
+__lock_blocked_add(xlator_t *this, pl_dom_list_t *dom, pl_inode_lock_t *lock,
+                   int can_block)
+{
+    if (can_block == 0) {
+        goto out;
+    }
+
+    lock->blkd_time = gf_time();
+    list_add_tail(&lock->blocked_locks, &dom->blocked_inodelks);
+
+    gf_msg_trace(this->name, 0,
+                 "%s (pid=%d) (lk-owner=%s) %" PRId64
+                 " - "
+                 "%" PRId64 " => Blocked",
+                 lock->fl_type == F_UNLCK ? "Unlock" : "Lock", lock->client_pid,
+                 lkowner_utoa(&lock->owner), lock->user_flock.l_start,
+                 lock->user_flock.l_len);
+
+    pl_trace_block(this, lock->frame, NULL, NULL, F_SETLKW, &lock->user_flock,
+                   lock->volume);
+out:
+    return -EAGAIN;
+}
 
-		goto out;
+/* Determines if lock can be granted and adds the lock. If the lock
+ * is blocking, adds it to the blocked_inodelks list of the domain.
+ */
+static int
+__lock_inodelk(xlator_t *this, pl_inode_t *pl_inode, pl_inode_lock_t *lock,
+               int can_block, pl_dom_list_t *dom, struct timespec *now,
+               struct list_head *contend)
+{
+    pl_inode_lock_t *conf = NULL;
+    int ret;
+
+    ret = pl_inode_remove_inodelk(pl_inode, lock);
+    if (ret < 0) {
+        return ret;
+    }
+    if (ret == 0) {
+        conf = __inodelk_grantable(this, dom, lock, now, contend);
+    }
+    if ((ret > 0) || (conf != NULL)) {
+        return __lock_blocked_add(this, dom, lock, can_block);
+    }
+
+    /* To prevent blocked locks starvation, check if there are any blocked
+     * locks thay may conflict with this lock. If there is then don't grant
+     * the lock. BUT grant the lock if the owner already has lock to allow
+     * nested locks.
+     * Example:
+     * SHD from Machine1 takes (gfid, 0-infinity) and is granted.
+     * SHD from machine2 takes (gfid, 0-infinity) and is blocked.
+     * When SHD from Machine1 takes (gfid, 0-128KB) it
+     * needs to be granted, without which the earlier lock on 0-infinity
+     * will not be unlocked by SHD from Machine1.
+     * TODO: Find why 'owner_has_lock' is checked even for blocked locks.
+     */
+    if (__blocked_lock_conflict(dom, lock) && !(__owner_has_lock(dom, lock))) {
+        if (can_block != 0) {
+            gf_log(this->name, GF_LOG_DEBUG,
+                   "Lock is grantable, but blocking to prevent "
+                   "starvation");
         }
-	list_add (&lock->list, &dom->inodelk_list);
 
-	ret = 0;
+        return __lock_blocked_add(this, dom, lock, can_block);
+    }
+    __pl_inodelk_ref(lock);
+    lock->granted_time = gf_time();
+    list_add(&lock->list, &dom->inodelk_list);
 
-out:
-	return ret;
+    return 0;
 }
 
 /* Return true if the two inodelks have exactly same lock boundaries */
 static int
-inodelks_equal (pl_inode_lock_t *l1, pl_inode_lock_t *l2)
+inodelks_equal(pl_inode_lock_t *l1, pl_inode_lock_t *l2)
 {
-	if ((l1->fl_start == l2->fl_start) &&
-	    (l1->fl_end == l2->fl_end))
-		return 1;
+    if ((l1->fl_start == l2->fl_start) && (l1->fl_end == l2->fl_end))
+        return 1;
 
-	return 0;
+    return 0;
 }
 
-
 static pl_inode_lock_t *
-find_matching_inodelk (pl_inode_lock_t *lock, pl_dom_list_t *dom)
+find_matching_inodelk(pl_inode_lock_t *lock, pl_dom_list_t *dom)
 {
-	pl_inode_lock_t *l = NULL;
-	list_for_each_entry (l, &dom->inodelk_list, list) {
-		if (inodelks_equal (l, lock) &&
-                    same_inodelk_owner (l, lock))
-			return l;
-	}
-	return NULL;
+    pl_inode_lock_t *l = NULL;
+    list_for_each_entry(l, &dom->inodelk_list, list)
+    {
+        if (inodelks_equal(l, lock) && same_inodelk_owner(l, lock))
+            return l;
+    }
+    return NULL;
 }
 
 /* Set F_UNLCK removes a lock which has the exact same lock boundaries
  * as the UNLCK lock specifies. If such a lock is not found, returns invalid
  */
 static pl_inode_lock_t *
-__inode_unlock_lock (xlator_t *this, pl_inode_lock_t *lock, pl_dom_list_t *dom)
+__inode_unlock_lock(xlator_t *this, pl_inode_lock_t *lock, pl_dom_list_t *dom)
 {
-
-	pl_inode_lock_t *conf = NULL;
-
-	conf = find_matching_inodelk (lock, dom);
-	if (!conf) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        " Matching lock not found for unlock");
-		goto out;
-        }
-	__delete_inode_lock (conf);
-        gf_log (this->name, GF_LOG_DEBUG,
-                " Matching lock found for unlock");
-        __destroy_inode_lock (lock);
-
+    pl_inode_lock_t *conf = NULL;
+    inode_t *inode = NULL;
+
+    inode = lock->pl_inode->inode;
+
+    conf = find_matching_inodelk(lock, dom);
+    if (!conf) {
+        gf_log(this->name, GF_LOG_ERROR,
+               " Matching lock not found for unlock %llu-%llu, by %s "
+               "on %p for gfid:%s",
+               (unsigned long long)lock->fl_start,
+               (unsigned long long)lock->fl_end, lkowner_utoa(&lock->owner),
+               lock->client, inode ? uuid_utoa(inode->gfid) : "UNKNOWN");
+        goto out;
+    }
+    __delete_inode_lock(conf);
+    gf_log(this->name, GF_LOG_DEBUG,
+           " Matching lock found for unlock %llu-%llu, by %s on %p for gfid:%s",
+           (unsigned long long)lock->fl_start, (unsigned long long)lock->fl_end,
+           lkowner_utoa(&lock->owner), lock->client,
+           inode ? uuid_utoa(inode->gfid) : "UNKNOWN");
 
 out:
-	return conf;
-
-
+    return conf;
 }
-static void
-__grant_blocked_inode_locks (xlator_t *this, pl_inode_t *pl_inode, 
-			     struct list_head *granted, pl_dom_list_t *dom)
-{
-	int	      bl_ret = 0;
-	pl_inode_lock_t *bl = NULL;
-	pl_inode_lock_t *tmp = NULL;
 
-        struct list_head blocked_list;
+void
+__grant_blocked_inode_locks(xlator_t *this, pl_inode_t *pl_inode,
+                            struct list_head *granted, pl_dom_list_t *dom,
+                            struct timespec *now, struct list_head *contend)
+{
+    pl_inode_lock_t *bl = NULL;
+    pl_inode_lock_t *tmp = NULL;
 
-        INIT_LIST_HEAD (&blocked_list);
-        list_splice_init (&dom->blocked_inodelks, &blocked_list);
+    struct list_head blocked_list;
 
-	list_for_each_entry_safe (bl, tmp, &blocked_list, blocked_locks) {
+    INIT_LIST_HEAD(&blocked_list);
+    list_splice_init(&dom->blocked_inodelks, &blocked_list);
 
-		list_del_init (&bl->blocked_locks);
+    list_for_each_entry_safe(bl, tmp, &blocked_list, blocked_locks)
+    {
+        list_del_init(&bl->blocked_locks);
 
-		bl_ret = __lock_inodelk (this, pl_inode, bl, 1, dom);
+        bl->status = __lock_inodelk(this, pl_inode, bl, 1, dom, now, contend);
 
-		if (bl_ret == 0) {
-			list_add (&bl->blocked_locks, granted);
-                }
+        if (bl->status != -EAGAIN) {
+            list_add_tail(&bl->blocked_locks, granted);
         }
-	return;
+    }
 }
 
-/* Grant all inodelks blocked on a lock */
 void
-grant_blocked_inode_locks (xlator_t *this, pl_inode_t *pl_inode, pl_dom_list_t *dom)
+unwind_granted_inodes(xlator_t *this, pl_inode_t *pl_inode,
+                      struct list_head *granted)
 {
-	struct list_head granted;
-	pl_inode_lock_t *lock;
-	pl_inode_lock_t *tmp;
+    pl_inode_lock_t *lock;
+    pl_inode_lock_t *tmp;
+    int32_t op_ret;
+    int32_t op_errno;
+
+    list_for_each_entry_safe(lock, tmp, granted, blocked_locks)
+    {
+        if (lock->status == 0) {
+            op_ret = 0;
+            op_errno = 0;
+            gf_log(this->name, GF_LOG_TRACE,
+                   "%s (pid=%d) (lk-owner=%s) %" PRId64 " - %" PRId64
+                   " => Granted",
+                   lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
+                   lock->client_pid, lkowner_utoa(&lock->owner),
+                   lock->user_flock.l_start, lock->user_flock.l_len);
+        } else {
+            op_ret = -1;
+            op_errno = -lock->status;
+        }
+        pl_trace_out(this, lock->frame, NULL, NULL, F_SETLKW, &lock->user_flock,
+                     op_ret, op_errno, lock->volume);
 
-	INIT_LIST_HEAD (&granted);
+        STACK_UNWIND_STRICT(inodelk, lock->frame, op_ret, op_errno, NULL);
+        lock->frame = NULL;
+    }
 
-        if (list_empty (&dom->blocked_inodelks)) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "No blocked locks to be granted for domain: %s", dom->domain);
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        list_for_each_entry_safe(lock, tmp, granted, blocked_locks)
+        {
+            list_del_init(&lock->blocked_locks);
+            __pl_inodelk_unref(lock);
         }
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+}
 
-        pthread_mutex_lock (&pl_inode->mutex);
-	{
-		__grant_blocked_inode_locks (this, pl_inode, &granted, dom);
-	}
-        pthread_mutex_unlock (&pl_inode->mutex);
-
-	list_for_each_entry_safe (lock, tmp, &granted, blocked_locks) {
-                gf_log (this->name, GF_LOG_TRACE,
-			"%s (pid=%d) (lk-owner=%"PRIu64") %"PRId64" - %"PRId64" => Granted",
-			lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
-			lock->client_pid,
-			lock->owner,
-			lock->user_flock.l_start,
-			lock->user_flock.l_len);
+/* Grant all inodelks blocked on a lock */
+void
+grant_blocked_inode_locks(xlator_t *this, pl_inode_t *pl_inode,
+                          pl_dom_list_t *dom, struct timespec *now,
+                          struct list_head *contend)
+{
+    struct list_head granted;
 
-		pl_trace_out (this, lock->frame, NULL, NULL, F_SETLKW,
-			      &lock->user_flock, 0, 0, lock->volume);
+    INIT_LIST_HEAD(&granted);
 
-		STACK_UNWIND_STRICT (inodelk, lock->frame, 0, 0);
-	}
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        __grant_blocked_inode_locks(this, pl_inode, &granted, dom, now,
+                                    contend);
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
 
+    unwind_granted_inodes(this, pl_inode, &granted);
 }
 
-/* Release all inodelks from this transport */
-static int
-release_inode_locks_of_transport (xlator_t *this, pl_dom_list_t *dom,
-                                  inode_t *inode, void *trans)
+static void
+pl_inodelk_log_cleanup(pl_inode_lock_t *lock)
 {
-	pl_inode_lock_t *tmp = NULL;
-	pl_inode_lock_t *l = NULL;
-
-        pl_inode_t * pinode = NULL;
-
-        struct list_head granted;
-        struct list_head released;
+    pl_inode_t *pl_inode = NULL;
 
-        char *path = NULL;
+    pl_inode = lock->pl_inode;
 
-        INIT_LIST_HEAD (&granted);
-        INIT_LIST_HEAD (&released);
-
-        pinode = pl_inode_get (this, inode);
+    gf_log(THIS->name, GF_LOG_WARNING,
+           "releasing lock on %s held by "
+           "{client=%p, pid=%" PRId64 " lk-owner=%s}",
+           uuid_utoa(pl_inode->gfid), lock->client, (uint64_t)lock->client_pid,
+           lkowner_utoa(&lock->owner));
+}
 
-        pthread_mutex_lock (&pinode->mutex);
+/* Release all inodelks from this client */
+int
+pl_inodelk_client_cleanup(xlator_t *this, pl_ctx_t *ctx)
+{
+    posix_locks_private_t *priv;
+    pl_inode_lock_t *tmp = NULL;
+    pl_inode_lock_t *l = NULL;
+    pl_dom_list_t *dom = NULL;
+    pl_inode_t *pl_inode = NULL;
+    struct list_head *pcontend = NULL;
+    struct list_head released;
+    struct list_head unwind;
+    struct list_head contend;
+    struct timespec now = {};
+
+    priv = this->private;
+
+    INIT_LIST_HEAD(&released);
+    INIT_LIST_HEAD(&unwind);
+
+    if (priv->notify_contention) {
+        pcontend = &contend;
+        INIT_LIST_HEAD(pcontend);
+        timespec_now(&now);
+    }
+
+    pthread_mutex_lock(&ctx->lock);
+    {
+        list_for_each_entry_safe(l, tmp, &ctx->inodelk_lockers, client_list)
         {
-
-                list_for_each_entry_safe (l, tmp, &dom->blocked_inodelks, blocked_locks) {
-                        if (l->transport != trans)
-                                continue;
-
-                        list_del_init (&l->blocked_locks);
-
-                        if (inode_path (inode, NULL, &path) < 0) {
-                                gf_log (this->name, GF_LOG_TRACE,
-                                        "inode_path failed");
-                                goto unlock;
-                        }
-
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "releasing lock on %s held by "
-                                "{transport=%p, pid=%"PRId64" lk-owner=%"PRIu64"}",
-                                path, trans,
-                                (uint64_t) l->client_pid,
-                                l->owner);
-
-                        list_add (&l->blocked_locks, &released);
-
+            pl_inodelk_log_cleanup(l);
+
+            pl_inode = l->pl_inode;
+
+            pthread_mutex_lock(&pl_inode->mutex);
+            {
+                /* If the inodelk object is part of granted list but not
+                 * blocked list, then perform the following actions:
+                 * i.   delete the object from granted list;
+                 * ii.  grant other locks (from other clients) that may
+                 *      have been blocked on this inodelk; and
+                 * iii. unref the object.
+                 *
+                 * If the inodelk object (L1) is part of both granted
+                 * and blocked lists, then this means that a parallel
+                 * unlock on another inodelk (L2 say) may have 'granted'
+                 * L1 and added it to 'granted' list in
+                 * __grant_blocked_inode_locks() (although using the
+                 * 'blocked_locks' member). In that case, the cleanup
+                 * codepath must try and grant other overlapping
+                 * blocked inodelks from other clients, now that L1 is
+                 * out of their way and then unref L1 in the end, and
+                 * leave it to the other thread (the one executing
+                 * unlock codepath) to unwind L1's frame, delete it from
+                 * blocked_locks list, and perform the last unref on L1.
+                 *
+                 * If the inodelk object (L1) is part of blocked list
+                 * only, the cleanup code path must:
+                 * i.   delete it from the blocked_locks list inside
+                 *      this critical section,
+                 * ii.  unwind its frame with EAGAIN,
+                 * iii. try and grant blocked inode locks from other
+                 *      clients that were otherwise grantable, but just
+                 *      got blocked to avoid leaving L1 to starve
+                 *      forever.
+                 * iv.  unref the object.
+                 */
+                list_del_init(&l->client_list);
+
+                if (!list_empty(&l->list)) {
+                    __delete_inode_lock(l);
+                    list_add_tail(&l->client_list, &released);
+                } else {
+                    list_del_init(&l->blocked_locks);
+                    list_add_tail(&l->client_list, &unwind);
                 }
+            }
+            pthread_mutex_unlock(&pl_inode->mutex);
+        }
+    }
+    pthread_mutex_unlock(&ctx->lock);
 
-                list_for_each_entry_safe (l, tmp, &dom->inodelk_list, list) {
-                        if (l->transport != trans)
-                                continue;
+    if (!list_empty(&unwind)) {
+        list_for_each_entry_safe(l, tmp, &unwind, client_list)
+        {
+            list_del_init(&l->client_list);
 
-                        __delete_inode_lock (l);
-			__destroy_inode_lock (l);
+            if (l->frame)
+                STACK_UNWIND_STRICT(inodelk, l->frame, -1, EAGAIN, NULL);
+            list_add_tail(&l->client_list, &released);
+        }
+    }
 
+    if (!list_empty(&released)) {
+        list_for_each_entry_safe(l, tmp, &released, client_list)
+        {
+            list_del_init(&l->client_list);
 
-                        if (inode_path (inode, NULL, &path) < 0) {
-                                gf_log (this->name, GF_LOG_TRACE,
-                                        "inode_path failed");
-                                goto unlock;
-                        }
+            pl_inode = l->pl_inode;
 
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "releasing lock on %s held by "
-                                "{transport=%p, pid=%"PRId64" lk-owner=%"PRIu64"}",
-                                path, trans,
-                                (uint64_t) l->client_pid,
-                                l->owner);
+            dom = get_domain(pl_inode, l->volume);
 
+            grant_blocked_inode_locks(this, pl_inode, dom, &now, pcontend);
 
-                }
+            pthread_mutex_lock(&pl_inode->mutex);
+            {
+                __pl_inodelk_unref(l);
+            }
+            pthread_mutex_unlock(&pl_inode->mutex);
+            inode_unref(pl_inode->inode);
         }
-unlock:
-        if (path)
-                GF_FREE (path);
-
-        pthread_mutex_unlock (&pinode->mutex);
+    }
 
-        list_for_each_entry_safe (l, tmp, &released, blocked_locks) {
-                list_del_init (&l->blocked_locks);
+    if (pcontend != NULL) {
+        inodelk_contention_notify(this, pcontend);
+    }
 
-                STACK_UNWIND_STRICT (inodelk, l->frame, -1, EAGAIN);
-                GF_FREE (l);
-        }
-
-	grant_blocked_inode_locks (this, pinode, dom);
-	return 0;
+    return 0;
 }
 
-
 static int
-pl_inode_setlk (xlator_t *this, pl_inode_t *pl_inode, pl_inode_lock_t *lock,
-		int can_block,  pl_dom_list_t *dom)
-{
-	int ret = -EINVAL;
-        pl_inode_lock_t *retlock = NULL;
-
-	pthread_mutex_lock (&pl_inode->mutex);
-	{
-		if (lock->fl_type != F_UNLCK) {
-			ret = __lock_inodelk (this, pl_inode, lock, can_block, dom);
-			if (ret == 0)
-				gf_log (this->name, GF_LOG_TRACE,
-                                        "%s (pid=%d) (lk-owner=%"PRIu64") %"PRId64" - %"PRId64" => OK",
-                                        lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
-                                        lock->client_pid,
-                                        lock->owner,
-                                        lock->fl_start,
-                                        lock->fl_end);
-
-			if (ret == -EAGAIN)
-				gf_log (this->name, GF_LOG_TRACE,
-                                        "%s (pid=%d) (lk-owner=%"PRIu64") %"PRId64" - %"PRId64" => NOK",
-					lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
-					lock->client_pid,
-                                        lock->owner,
-					lock->user_flock.l_start,
-					lock->user_flock.l_len);
-
-			goto out;
-		}
-
-
-		retlock = __inode_unlock_lock (this, lock, dom);
-		if (!retlock) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"Bad Unlock issued on Inode lock");
-                        ret = -EINVAL;
-                        goto out;
+pl_inode_setlk(xlator_t *this, pl_ctx_t *ctx, pl_inode_t *pl_inode,
+               pl_inode_lock_t *lock, int can_block, pl_dom_list_t *dom,
+               inode_t *inode)
+{
+    posix_locks_private_t *priv = NULL;
+    int ret = -EINVAL;
+    pl_inode_lock_t *retlock = NULL;
+    gf_boolean_t unref = _gf_true;
+    gf_boolean_t need_inode_unref = _gf_false;
+    struct list_head *pcontend = NULL;
+    struct list_head contend;
+    struct list_head wake;
+    struct timespec now = {};
+    short fl_type;
+
+    lock->pl_inode = pl_inode;
+    fl_type = lock->fl_type;
+
+    priv = this->private;
+
+    /* Ideally, AFTER a successful lock (both blocking and non-blocking) or
+     * an unsuccessful blocking lock operation, the inode needs to be ref'd.
+     *
+     * But doing so might give room to a race where the lock-requesting
+     * client could send a DISCONNECT just before this thread refs the inode
+     * after the locking is done, and the epoll thread could unref the inode
+     * in cleanup which means the inode's refcount would come down to 0, and
+     * the call to pl_forget() at this point destroys @pl_inode. Now when
+     * the io-thread executing this function tries to access pl_inode,
+     * it could crash on account of illegal memory access.
+     *
+     * To get around this problem, the inode is ref'd once even before
+     * adding the lock into client_list as a precautionary measure.
+     * This way even if there are DISCONNECTs, there will always be 1 extra
+     * ref on the inode, so @pl_inode is still alive until after the
+     * current stack unwinds.
+     */
+    pl_inode->inode = inode_ref(inode);
+
+    if (priv->revocation_secs != 0) {
+        if (lock->fl_type != F_UNLCK) {
+            __inodelk_prune_stale(this, pl_inode, dom, lock);
+        } else if (priv->monkey_unlocking == _gf_true) {
+            if (pl_does_monkey_want_stuck_lock()) {
+                pthread_mutex_lock(&pl_inode->mutex);
+                {
+                    __pl_inodelk_unref(lock);
                 }
-		__destroy_inode_lock (retlock);
-
-                ret = 0;
+                pthread_mutex_unlock(&pl_inode->mutex);
+                inode_unref(pl_inode->inode);
+                gf_log(this->name, GF_LOG_WARNING,
+                       "MONKEY LOCKING (forcing stuck lock)!");
+                return 0;
+            }
+        }
+    }
+
+    if (priv->notify_contention) {
+        pcontend = &contend;
+        INIT_LIST_HEAD(pcontend);
+        timespec_now(&now);
+    }
+
+    INIT_LIST_HEAD(&wake);
+
+    if (ctx)
+        pthread_mutex_lock(&ctx->lock);
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        if (lock->fl_type != F_UNLCK) {
+            ret = __lock_inodelk(this, pl_inode, lock, can_block, dom, &now,
+                                 pcontend);
+            if (ret == 0) {
+                lock->frame = NULL;
+                gf_log(this->name, GF_LOG_TRACE,
+                       "%s (pid=%d) (lk-owner=%s) %" PRId64 " - %" PRId64
+                       " => OK",
+                       lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
+                       lock->client_pid, lkowner_utoa(&lock->owner),
+                       lock->fl_start, lock->fl_end);
+            } else if (ret == -EAGAIN) {
+                gf_log(this->name, GF_LOG_TRACE,
+                       "%s (pid=%d) (lk-owner=%s) %" PRId64 " - %" PRId64
+                       " => NOK",
+                       lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
+                       lock->client_pid, lkowner_utoa(&lock->owner),
+                       lock->user_flock.l_start, lock->user_flock.l_len);
+                if (can_block) {
+                    unref = _gf_false;
+                }
+            }
+            /* For all but the case where a non-blocking lock attempt fails
+             * with -EAGAIN, the extra ref taken at the start of this function
+             * must be negated. */
+            need_inode_unref = (ret != 0) && ((ret != -EAGAIN) || !can_block);
+            if (ctx && !need_inode_unref) {
+                list_add_tail(&lock->client_list, &ctx->inodelk_lockers);
+            }
+        } else {
+            /* Irrespective of whether unlock succeeds or not,
+             * the extra inode ref that was done at the start of
+             * this function must be negated. Towards this,
+             * @need_inode_unref flag is set unconditionally here.
+             */
+            need_inode_unref = _gf_true;
+            retlock = __inode_unlock_lock(this, lock, dom);
+            if (!retlock) {
+                gf_log(this->name, GF_LOG_DEBUG,
+                       "Bad Unlock issued on Inode lock");
+                ret = -EINVAL;
+                goto out;
+            }
+            list_del_init(&retlock->client_list);
+            __pl_inodelk_unref(retlock);
 
+            pl_inode_remove_unlocked(this, pl_inode, &wake);
 
-	}
-out:
-	pthread_mutex_unlock (&pl_inode->mutex);
-	grant_blocked_inode_locks (this, pl_inode, dom);
-        return ret;
+            ret = 0;
+        }
+    out:
+        if (unref)
+            __pl_inodelk_unref(lock);
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+    if (ctx)
+        pthread_mutex_unlock(&ctx->lock);
+
+    pl_inode_remove_wake(&wake);
+
+    /* The following (extra) unref corresponds to the ref that
+     * was done at the time the lock was granted.
+     */
+    if ((fl_type == F_UNLCK) && (ret == 0)) {
+        inode_unref(pl_inode->inode);
+        grant_blocked_inode_locks(this, pl_inode, dom, &now, pcontend);
+    }
+
+    if (need_inode_unref) {
+        inode_unref(pl_inode->inode);
+    }
+
+    if (pcontend != NULL) {
+        inodelk_contention_notify(this, pcontend);
+    }
+
+    return ret;
 }
 
 /* Create a new inode_lock_t */
-pl_inode_lock_t *
-new_inode_lock (struct gf_flock *flock, void *transport, pid_t client_pid,
-                uint64_t owner, const char *volume)
+static pl_inode_lock_t *
+new_inode_lock(struct gf_flock *flock, client_t *client, pid_t client_pid,
+               call_frame_t *frame, xlator_t *this, const char *volume,
+               char *conn_id, int32_t *op_errno)
 
 {
-	pl_inode_lock_t *lock = NULL;
+    pl_inode_lock_t *lock = NULL;
+
+    if (!pl_is_lk_owner_valid(&frame->root->lk_owner, frame->root->client)) {
+        *op_errno = EINVAL;
+        goto out;
+    }
+
+    lock = GF_CALLOC(1, sizeof(*lock), gf_locks_mt_pl_inode_lock_t);
+    if (!lock) {
+        *op_errno = ENOMEM;
+        goto out;
+    }
+
+    lock->fl_start = flock->l_start;
+    lock->fl_type = flock->l_type;
+
+    if (flock->l_len == 0)
+        lock->fl_end = LLONG_MAX;
+    else
+        lock->fl_end = flock->l_start + flock->l_len - 1;
+
+    lock->client = client;
+    lock->client_pid = client_pid;
+    lock->volume = volume;
+    lock->owner = frame->root->lk_owner;
+    lock->frame = frame;
+    lock->this = this;
+
+    if (conn_id) {
+        lock->connection_id = gf_strdup(conn_id);
+    }
+
+    INIT_LIST_HEAD(&lock->list);
+    INIT_LIST_HEAD(&lock->blocked_locks);
+    INIT_LIST_HEAD(&lock->client_list);
+    INIT_LIST_HEAD(&lock->contend);
+    __pl_inodelk_ref(lock);
+
+out:
+    return lock;
+}
 
-	lock = GF_CALLOC (1, sizeof (*lock),
-                          gf_locks_mt_pl_inode_lock_t);
-	if (!lock) {
-		return NULL;
-	}
+int32_t
+_pl_convert_volume(const char *volume, char **res)
+{
+    char *mdata_vol = NULL;
+    int ret = 0;
 
-	lock->fl_start = flock->l_start;
-	lock->fl_type  = flock->l_type;
+    mdata_vol = strrchr(volume, ':');
+    // if the volume already ends with :metadata don't bother
+    if (mdata_vol && (strcmp(mdata_vol, ":metadata") == 0))
+        return 0;
 
-	if (flock->l_len == 0)
-		lock->fl_end = LLONG_MAX;
-	else
-		lock->fl_end = flock->l_start + flock->l_len - 1;
+    ret = gf_asprintf(res, "%s:metadata", volume);
+    if (ret <= 0)
+        return ENOMEM;
+    return 0;
+}
 
-	lock->transport  = transport;
-	lock->client_pid = client_pid;
-        lock->owner      = owner;
-	lock->volume     = volume;
+int32_t
+_pl_convert_volume_for_special_range(struct gf_flock *flock, const char *volume,
+                                     char **res)
+{
+    int32_t ret = 0;
 
-	INIT_LIST_HEAD (&lock->list);
-	INIT_LIST_HEAD (&lock->blocked_locks);
+    if ((flock->l_start == LLONG_MAX - 1) && (flock->l_len == 0)) {
+        ret = _pl_convert_volume(volume, res);
+    }
 
-	return lock;
+    return ret;
 }
 
 /* Common inodelk code called from pl_inodelk and pl_finodelk */
 int
-pl_common_inodelk (call_frame_t *frame, xlator_t *this,
-                   const char *volume, inode_t *inode, int32_t cmd,
-                   struct gf_flock *flock, loc_t *loc, fd_t *fd)
-{
-	int32_t op_ret   = -1;
-	int32_t op_errno = 0;
-	int     ret      = -1;
-	int     can_block = 0;
-	void *                  transport  = NULL;
-	pid_t                   client_pid = -1;
-        uint64_t                owner      = -1;
-	pl_inode_t *            pinode     = NULL;
-	pl_inode_lock_t *       reqlock    = NULL;
-	pl_dom_list_t *		dom	   = NULL;
-
-	VALIDATE_OR_GOTO (frame, out);
-	VALIDATE_OR_GOTO (inode, unwind);
-	VALIDATE_OR_GOTO (flock, unwind);
-
-	if ((flock->l_start < 0) || (flock->l_len < 0)) {
-		op_errno = EINVAL;
-		goto unwind;
-	}
-
-        pl_trace_in (this, frame, fd, loc, cmd, flock, volume);
-
-	transport  = frame->root->trans;
-	client_pid = frame->root->pid;
-        owner      = frame->root->lk_owner;
-
-	pinode = pl_inode_get (this, inode);
-	if (!pinode) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		op_errno = ENOMEM;
-		goto unwind;
-	}
-
-	dom = get_domain (pinode, volume);
-
-	if (owner == 0) {
-		/*
-                  special case: this means release all locks
-                  from this transport
-		*/
-		gf_log (this->name, GF_LOG_TRACE,
-			"Releasing all locks from transport %p", transport);
-
-		release_inode_locks_of_transport (this, dom, inode, transport);
-		goto unwind;
-	}
-
-	reqlock = new_inode_lock (flock, transport, client_pid, owner, volume);
-
-	if (!reqlock) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"Out of memory.");
-		op_ret = -1;
-		op_errno = ENOMEM;
-		goto unwind;
-	}
-
-        reqlock->frame = frame;
-        reqlock->this  = this;
-
-	switch (cmd) {
-	case F_SETLKW:
-		can_block = 1;
-		reqlock->frame = frame;
-		reqlock->this  = this;
-
-		/* fall through */
-
-	case F_SETLK:
-		memcpy (&reqlock->user_flock, flock, sizeof (struct gf_flock));
-		ret = pl_inode_setlk (this, pinode, reqlock,
-                                      can_block, dom);
-
-		if (ret < 0) {
-                        if (can_block) {
-                                pl_trace_block (this, frame, fd, loc,
-                                                cmd, flock, volume);
-				goto out;
-                        }
-			gf_log (this->name, GF_LOG_TRACE, "returning EAGAIN");
-			op_errno = -ret;
-			__destroy_inode_lock (reqlock);
-			goto unwind;
-		}
-		break;
-
-	default:
-		op_errno = ENOTSUP;
-		gf_log (this->name, GF_LOG_DEBUG,
-                        "Lock command F_GETLK not supported for [f]inodelk "
-                        "(cmd=%d)",
-                        cmd);
-                goto unwind;
+pl_common_inodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+                  inode_t *inode, int32_t cmd, struct gf_flock *flock,
+                  loc_t *loc, fd_t *fd, dict_t *xdata)
+{
+    int32_t op_ret = -1;
+    int32_t op_errno = 0;
+    int ret = -1;
+    GF_UNUSED int dict_ret = -1;
+    int can_block = 0;
+    short lock_type = 0;
+    pl_inode_t *pinode = NULL;
+    pl_inode_lock_t *reqlock = NULL;
+    pl_dom_list_t *dom = NULL;
+    char *res = NULL;
+    char *res1 = NULL;
+    char *conn_id = NULL;
+    pl_ctx_t *ctx = NULL;
+
+    if (xdata)
+        dict_ret = dict_get_str(xdata, "connection-id", &conn_id);
+
+    VALIDATE_OR_GOTO(frame, out);
+    VALIDATE_OR_GOTO(inode, unwind);
+    VALIDATE_OR_GOTO(flock, unwind);
+
+    if ((flock->l_start < 0) || (flock->l_len < 0)) {
+        op_errno = EINVAL;
+        goto unwind;
+    }
+
+    op_errno = _pl_convert_volume_for_special_range(flock, volume, &res);
+    if (op_errno)
+        goto unwind;
+    if (res)
+        volume = res;
+
+    pl_trace_in(this, frame, fd, loc, cmd, flock, volume);
+
+    if (frame->root->client) {
+        ctx = pl_ctx_get(frame->root->client, this);
+        if (!ctx) {
+            op_errno = ENOMEM;
+            gf_log(this->name, GF_LOG_INFO, "pl_ctx_get() failed");
+            goto unwind;
         }
+    }
+
+    pinode = pl_inode_get(this, inode, NULL);
+    if (!pinode) {
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+
+    dom = get_domain(pinode, volume);
+    if (!dom) {
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+
+    reqlock = new_inode_lock(flock, frame->root->client, frame->root->pid,
+                             frame, this, dom->domain, conn_id, &op_errno);
+
+    if (!reqlock) {
+        op_ret = -1;
+        goto unwind;
+    }
+
+    switch (cmd) {
+        case F_SETLKW:
+            can_block = 1;
+
+            /* fall through */
+
+        case F_SETLK:
+            lock_type = flock->l_type;
+            memcpy(&reqlock->user_flock, flock, sizeof(struct gf_flock));
+            ret = pl_inode_setlk(this, ctx, pinode, reqlock, can_block, dom,
+                                 inode);
+
+            if (ret < 0) {
+                if (ret == -EAGAIN) {
+                    if (can_block && (F_UNLCK != lock_type)) {
+                        goto out;
+                    }
+                    gf_log(this->name, GF_LOG_TRACE, "returning EAGAIN");
+                } else {
+                    gf_log(this->name, GF_LOG_TRACE, "returning %d", ret);
+                }
+                op_errno = -ret;
+                goto unwind;
+            }
+            break;
+
+        default:
+            op_errno = ENOTSUP;
+            gf_log(this->name, GF_LOG_DEBUG,
+                   "Lock command F_GETLK not supported for [f]inodelk "
+                   "(cmd=%d)",
+                   cmd);
+            goto unwind;
+    }
 
-        op_ret = 0;
+    op_ret = 0;
 
 unwind:
-	if ((inode != NULL) && (flock !=NULL)) {
-		pl_update_refkeeper (this, inode);
-		pl_trace_out (this, frame, fd, loc, cmd, flock, op_ret, op_errno, volume);
-	}
+    if (flock != NULL)
+        pl_trace_out(this, frame, fd, loc, cmd, flock, op_ret, op_errno,
+                     volume);
 
-        STACK_UNWIND_STRICT (inodelk, frame, op_ret, op_errno);
+    STACK_UNWIND_STRICT(inodelk, frame, op_ret, op_errno, NULL);
 out:
-        return 0;
+    GF_FREE(res);
+    GF_FREE(res1);
+    return 0;
 }
 
 int
-pl_inodelk (call_frame_t *frame, xlator_t *this,
-            const char *volume, loc_t *loc, int32_t cmd, struct gf_flock *flock)
+pl_inodelk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+           int32_t cmd, struct gf_flock *flock, dict_t *xdata)
 {
+    pl_common_inodelk(frame, this, volume, loc->inode, cmd, flock, loc, NULL,
+                      xdata);
 
-        pl_common_inodelk (frame, this, volume, loc->inode, cmd, flock, loc, NULL);
-
-        return 0;
+    return 0;
 }
 
 int
-pl_finodelk (call_frame_t *frame, xlator_t *this,
-             const char *volume, fd_t *fd, int32_t cmd, struct gf_flock *flock)
+pl_finodelk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+            int32_t cmd, struct gf_flock *flock, dict_t *xdata)
 {
+    pl_common_inodelk(frame, this, volume, fd->inode, cmd, flock, NULL, fd,
+                      xdata);
 
-        pl_common_inodelk (frame, this, volume, fd->inode, cmd, flock, NULL, fd);
-
-        return 0;
-
+    return 0;
 }
 
-
 static int32_t
-__get_inodelk_count (xlator_t *this, pl_inode_t *pl_inode)
-{
-        int32_t            count  = 0;
-        pl_inode_lock_t   *lock   = NULL;
-        pl_dom_list_t     *dom    = NULL;
-
-        list_for_each_entry (dom, &pl_inode->dom_list, inode_list) {
-                list_for_each_entry (lock, &dom->inodelk_list, list) {
-
-			gf_log (this->name, GF_LOG_DEBUG,
-                                " XATTR DEBUG"
-				" domain: %s %s (pid=%d) (lk-owner=%"PRIu64") %"PRId64" - %"PRId64" "
-                                "state = Active",
-                                dom->domain,
-				lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
-				lock->client_pid,
-                                lock->owner,
-				lock->user_flock.l_start,
-				lock->user_flock.l_len);
-
-                        count++;
-                }
-
-                list_for_each_entry (lock, &dom->blocked_inodelks, blocked_locks) {
-
-			gf_log (this->name, GF_LOG_DEBUG,
-                                " XATTR DEBUG"
-				" domain: %s %s (pid=%d) (lk-owner=%"PRIu64") %"PRId64" - %"PRId64" "
-                                "state = Blocked",
-                                dom->domain,
-				lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
-				lock->client_pid,
-                                lock->owner,
-				lock->user_flock.l_start,
-				lock->user_flock.l_len);
+__get_inodelk_dom_count(pl_dom_list_t *dom)
+{
+    pl_inode_lock_t *lock = NULL;
+    int32_t count = 0;
+
+    list_for_each_entry(lock, &dom->inodelk_list, list) { count++; }
+    list_for_each_entry(lock, &dom->blocked_inodelks, blocked_locks)
+    {
+        count++;
+    }
+    return count;
+}
 
-                        count++;
-                }
+/* Returns the no. of locks (blocked/granted) held on a given domain name
+ * If @domname is NULL, returns the no. of locks in all the domains present.
+ * If @domname is non-NULL and non-existent, returns 0 */
+int32_t
+__get_inodelk_count(xlator_t *this, pl_inode_t *pl_inode, char *domname)
+{
+    int32_t count = 0;
+    pl_dom_list_t *dom = NULL;
+
+    list_for_each_entry(dom, &pl_inode->dom_list, inode_list)
+    {
+        if (domname) {
+            if (strcmp(domname, dom->domain) == 0) {
+                count = __get_inodelk_dom_count(dom);
+                goto out;
+            }
 
+        } else {
+            /* Counting locks from all domains */
+            count += __get_inodelk_dom_count(dom);
         }
+    }
 
-        return count;
+out:
+    return count;
 }
 
 int32_t
-get_inodelk_count (xlator_t *this, inode_t *inode)
+get_inodelk_count(xlator_t *this, inode_t *inode, char *domname)
 {
-        pl_inode_t   *pl_inode = NULL;
-        uint64_t      tmp_pl_inode = 0;
-        int           ret      = 0;
-        int32_t       count    = 0;
+    pl_inode_t *pl_inode = NULL;
+    uint64_t tmp_pl_inode = 0;
+    int ret = 0;
+    int32_t count = 0;
 
-        ret = inode_ctx_get (inode, this, &tmp_pl_inode);
-        if (ret != 0) {
-                goto out;
-        }
+    ret = inode_ctx_get(inode, this, &tmp_pl_inode);
+    if (ret != 0) {
+        goto out;
+    }
 
-        pl_inode = (pl_inode_t *)(long) tmp_pl_inode;
+    pl_inode = (pl_inode_t *)(long)tmp_pl_inode;
 
-        pthread_mutex_lock (&pl_inode->mutex);
-        {
-                count = __get_inodelk_count (this, pl_inode);
-        }
-        pthread_mutex_unlock (&pl_inode->mutex);
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        count = __get_inodelk_count(this, pl_inode, domname);
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
 
 out:
-        return count;
+    return count;
 }
diff --git a/xlators/features/locks/src/locks-mem-types.h b/xlators/features/locks/src/locks-mem-types.h
index b4f7e28f18e..a76605027b3 100644
--- a/xlators/features/locks/src/locks-mem-types.h
+++ b/xlators/features/locks/src/locks-mem-types.h
@@ -1,40 +1,28 @@
 /*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
    This file is part of GlusterFS.
 
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
 */
 
-
 #ifndef __LOCKS_MEM_TYPES_H__
 #define __LOCKS_MEM_TYPES_H__
 
-#include "mem-types.h"
+#include <glusterfs/mem-types.h>
 
 enum gf_locks_mem_types_ {
-        gf_locks_mt_pl_dom_list_t = gf_common_mt_end + 1,
-        gf_locks_mt_pl_inode_t,
-        gf_locks_mt_posix_lock_t,
-        gf_locks_mt_pl_entry_lock_t,
-        gf_locks_mt_pl_inode_lock_t,
-        gf_locks_mt_truncate_ops,
-        gf_locks_mt_pl_rw_req_t,
-        gf_locks_mt_posix_locks_private_t,
-        gf_locks_mt_pl_local_t,
-        gf_locks_mt_pl_fdctx_t,
-        gf_locks_mt_end
+    gf_locks_mt_pl_dom_list_t = gf_common_mt_end + 1,
+    gf_locks_mt_pl_inode_t,
+    gf_locks_mt_posix_lock_t,
+    gf_locks_mt_pl_entry_lock_t,
+    gf_locks_mt_pl_inode_lock_t,
+    gf_locks_mt_pl_rw_req_t,
+    gf_locks_mt_posix_locks_private_t,
+    gf_locks_mt_pl_fdctx_t,
+    gf_locks_mt_pl_meta_lock_t,
+    gf_locks_mt_end
 };
 #endif
-
diff --git a/xlators/features/locks/src/locks.h b/xlators/features/locks/src/locks.h
index 614bddb6451..c868eb494a2 100644
--- a/xlators/features/locks/src/locks.h
+++ b/xlators/features/locks/src/locks.h
@@ -1,161 +1,292 @@
 /*
-  Copyright (c) 2006, 2007, 2008 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
+   Copyright (c) 2006-2012, 2015-2016 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
 #ifndef __POSIX_LOCKS_H__
 #define __POSIX_LOCKS_H__
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "compat-errno.h"
-#include "stack.h"
-#include "call-stub.h"
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/stack.h>
+#include <glusterfs/call-stub.h>
 #include "locks-mem-types.h"
+#include <glusterfs/client_t.h>
+
+#include <glusterfs/lkowner.h>
+
+typedef enum {
+    MLK_NONE,
+    MLK_FILE_BASED,
+    MLK_FORCED,
+    MLK_OPTIMAL
+} mlk_mode_t; /* defines different mandatory locking modes*/
 
 struct __pl_fd;
 
 struct __posix_lock {
-        struct list_head   list;
+    struct list_head list;
+
+    off_t fl_start;
+    off_t fl_end;
+    uint32_t lk_flags;
+
+    short fl_type;
+    short blocked;              /* waiting to acquire */
+    struct gf_flock user_flock; /* the flock supplied by the user */
+    xlator_t *this;             /* required for blocked locks */
+    unsigned long fd_num;
 
-        short              fl_type;
-        off_t              fl_start;
-        off_t              fl_end;
+    fd_t *fd;
+    call_frame_t *frame;
 
-        short              blocked;    /* waiting to acquire */
-        struct gf_flock       user_flock; /* the flock supplied by the user */
-        xlator_t          *this;       /* required for blocked locks */
-        unsigned long      fd_num;
+    time_t blkd_time;    /* time at which lock was queued into blkd list */
+    time_t granted_time; /* time at which lock was queued into active list */
 
-        fd_t              *fd;
-        call_frame_t      *frame;
+    /* These two together serve to uniquely identify each process
+       across nodes */
 
-        /* These two together serve to uniquely identify each process
-           across nodes */
+    void *client; /* to identify client node */
 
-        void              *transport;     /* to identify client node */
-        pid_t              client_pid;    /* pid of client process */
-        uint64_t           owner;         /* lock owner from fuse */
+    /* This field uniquely identifies the client the lock belongs to.  As
+     * lock migration is handled by rebalance, the client_t object will be
+     * overwritten by rebalance and can't be deemed as the owner of the
+     * lock on destination. Hence, the below field is migrated from
+     * source to destination by lock_migration_info_t and updated on the
+     * destination. So that on client-server disconnection, server can
+     * cleanup the locks proper;y.  */
+
+    char *client_uid;
+    gf_lkowner_t owner;
+    pid_t client_pid; /* pid of client process */
+
+    int blocking;
 };
 typedef struct __posix_lock posix_lock_t;
 
 struct __pl_inode_lock {
-        struct list_head   list;
-        struct list_head   blocked_locks; /* list_head pointing to blocked_inodelks */
+    struct list_head list;
+    struct list_head blocked_locks; /* list_head pointing to blocked_inodelks */
+    struct list_head contend;       /* list of contending locks */
+    int ref;
+
+    off_t fl_start;
+    off_t fl_end;
+
+    const char *volume;
+
+    struct gf_flock user_flock; /* the flock supplied by the user */
+    xlator_t *this;             /* required for blocked locks */
+    struct __pl_inode *pl_inode;
+
+    call_frame_t *frame;
+
+    time_t blkd_time;    /* time at which lock was queued into blkd list */
+    time_t granted_time; /* time at which lock was queued into active list */
 
-        short              fl_type;
-        off_t              fl_start;
-        off_t              fl_end;
+    /*last time at which lock contention was detected and notified*/
+    struct timespec contention_time;
 
-        const char        *volume;
+    /* These two together serve to uniquely identify each process
+       across nodes */
 
-        struct gf_flock       user_flock; /* the flock supplied by the user */
-        xlator_t          *this;       /* required for blocked locks */
-        fd_t              *fd;
+    void *client; /* to identify client node */
+    gf_lkowner_t owner;
+    pid_t client_pid; /* pid of client process */
 
-        call_frame_t      *frame;
+    char *connection_id; /* stores the client connection id */
 
-        /* These two together serve to uniquely identify each process
-           across nodes */
+    struct list_head client_list; /* list of all locks from a client */
+    short fl_type;
 
-        void              *transport;     /* to identify client node */
-        pid_t              client_pid;    /* pid of client process */
-        uint64_t           owner;
+    int32_t status; /* Error code when we try to grant a lock in blocked
+                       state */
 };
 typedef struct __pl_inode_lock pl_inode_lock_t;
 
-struct __pl_rw_req_t {
-        struct list_head      list;
-        call_stub_t          *stub;
-        posix_lock_t          region;
+struct _pl_rw_req {
+    struct list_head list;
+    call_stub_t *stub;
+    posix_lock_t region;
 };
-typedef struct __pl_rw_req_t pl_rw_req_t;
-
-struct __pl_dom_list_t {
-        struct list_head   inode_list;       /* list_head back to pl_inode_t */
-        const char        *domain;
-        struct list_head   entrylk_list;     /* List of entry locks */
-        struct list_head   blocked_entrylks; /* List of all blocked entrylks */
-        struct list_head   inodelk_list;     /* List of inode locks */
-        struct list_head   blocked_inodelks; /* List of all blocked inodelks */
+typedef struct _pl_rw_req pl_rw_req_t;
+
+struct _pl_dom_list {
+    struct list_head inode_list; /* list_head back to pl_inode_t */
+    const char *domain;
+    struct list_head entrylk_list;     /* List of entry locks */
+    struct list_head blocked_entrylks; /* List of all blocked entrylks */
+    struct list_head inodelk_list;     /* List of inode locks */
+    struct list_head blocked_inodelks; /* List of all blocked inodelks */
 };
-typedef struct __pl_dom_list_t pl_dom_list_t;
+typedef struct _pl_dom_list pl_dom_list_t;
 
 struct __entry_lock {
-        struct list_head  domain_list;    /* list_head back to pl_dom_list_t */
-        struct list_head  blocked_locks; /* list_head back to blocked_entrylks */
+    struct list_head domain_list;   /* list_head back to pl_dom_list_t */
+    struct list_head blocked_locks; /* list_head back to blocked_entrylks */
+    struct list_head contend;       /* list of contending locks */
+    int ref;
 
-        call_frame_t     *frame;
-        xlator_t         *this;
+    call_frame_t *frame;
+    xlator_t *this;
+    struct __pl_inode *pinode;
 
-        const char       *volume;
+    const char *volume;
 
-        const char       *basename;
-        entrylk_type      type;
+    const char *basename;
 
-        void      *trans;
-        pid_t             client_pid;    /* pid of client process */
-        uint64_t          owner;
+    time_t blkd_time;    /* time at which lock was queued into blkd list */
+    time_t granted_time; /* time at which lock was queued into active list */
+
+    /*last time at which lock contention was detected and notified*/
+    struct timespec contention_time;
+
+    void *client;
+    gf_lkowner_t owner;
+    pid_t client_pid; /* pid of client process */
+
+    char *connection_id; /* stores the client connection id */
+
+    struct list_head client_list; /* list of all locks from a client */
+    entrylk_type type;
 };
 typedef struct __entry_lock pl_entry_lock_t;
 
-
 /* The "simulated" inode. This contains a list of all the locks associated
    with this file */
 
 struct __pl_inode {
-        pthread_mutex_t  mutex;
-
-        struct list_head dom_list;       /* list of domains */
-        struct list_head ext_list;       /* list of fcntl locks */
-        struct list_head rw_list;        /* list of waiting r/w requests */
-        struct list_head reservelk_list;        /* list of reservelks */
-        struct list_head blocked_reservelks;        /* list of blocked reservelks */
-        struct list_head blocked_calls;  /* List of blocked lock calls while a reserve is held*/
-        int              mandatory;      /* if mandatory locking is enabled */
-
-        inode_t          *refkeeper;     /* hold refs on an inode while locks are
-                                            held to prevent pruning */
+    pthread_mutex_t mutex;
+
+    struct list_head dom_list;           /* list of domains */
+    struct list_head ext_list;           /* list of fcntl locks */
+    struct list_head rw_list;            /* list of waiting r/w requests */
+    struct list_head reservelk_list;     /* list of reservelks */
+    struct list_head blocked_reservelks; /* list of blocked reservelks */
+    struct list_head blocked_calls;      /* List of blocked lock calls while a
+                                            reserve is held*/
+    struct list_head metalk_list;        /* Meta lock list */
+    struct list_head queued_locks;       /* This is to store the incoming lock
+                                            requests while meta lock is enabled */
+    struct list_head waiting; /* List of pending fops waiting to unlink/rmdir
+                                 the inode. */
+    int mandatory;            /* if mandatory locking is enabled */
+
+    inode_t *refkeeper; /* hold refs on an inode while locks are
+                           held to prevent pruning */
+    uuid_t gfid;        /* placeholder for gfid of the inode */
+    inode_t *inode;     /* pointer to be used for ref and unref
+                           of inode_t as long as there are
+                           locks on it */
+    gf_boolean_t migrated;
+
+    /* Flag to indicate whether to read mlock-enforce xattr from disk */
+    gf_boolean_t check_mlock_info;
+
+    /* Mandatory_lock enforce: IO will be allowed if and only if the lkowner has
+       held the lock.
+
+       Note: An xattr is set on the file to recover this information post
+       reboot. If client does not want mandatory lock to be enforced, then it
+       should remove this xattr explicitly
+    */
+    gf_boolean_t mlock_enforced;
+    /* There are scenarios where mandatory lock is granted but there are IOs
+       pending at posix level. To avoid this before preempting the previous lock
+       owner, we wait for all the fops to be unwound.
+    */
+    int fop_wind_count;
+    pthread_cond_t check_fop_wind_count;
+
+    gf_boolean_t track_fop_wind_count;
+
+    int32_t links;           /* Number of hard links the inode has. */
+    uint32_t remove_running; /* Number of remove operations running. */
+    gf_boolean_t is_locked;  /* Regular locks will be blocked. */
+    gf_boolean_t removed;    /* The inode has been deleted. */
 };
 typedef struct __pl_inode pl_inode_t;
 
+struct __pl_metalk {
+    pthread_mutex_t mutex;
+    /* For pl_inode meta lock list */
+    struct list_head list;
+    /* For pl_ctx_t list */
+    struct list_head client_list;
+    char *client_uid;
 
-struct __pl_fd {
-        gf_boolean_t nonblocking;       /* whether O_NONBLOCK has been set */
+    pl_inode_t *pl_inode;
+    int ref;
 };
-typedef struct __pl_fd pl_fd_t;
-
+typedef struct __pl_metalk pl_meta_lock_t;
 
 typedef struct {
-        gf_boolean_t    mandatory;      /* if mandatory locking is enabled */
-        gf_boolean_t    trace;          /* trace lock requests in and out */
+    char *brickname;
+    uint32_t revocation_secs;
+    uint32_t revocation_max_blocked;
+    uint32_t notify_contention_delay;
+    mlk_mode_t mandatory_mode; /* holds current mandatory locking mode */
+    gf_boolean_t trace;        /* trace lock requests in and out */
+    gf_boolean_t monkey_unlocking;
+    gf_boolean_t revocation_clear_all;
+    gf_boolean_t notify_contention;
+    gf_boolean_t mlock_enforced;
 } posix_locks_private_t;
 
 typedef struct {
-        gf_boolean_t   entrylk_count_req;
-        gf_boolean_t   inodelk_count_req;
-        gf_boolean_t   posixlk_count_req;
+    data_t *inodelk_dom_count_req;
+
+    dict_t *xdata;
+    loc_t loc[2];
+    fd_t *fd;
+    inode_t *inode;
+    off_t offset;
+    glusterfs_fop_t op;
+    gf_boolean_t entrylk_count_req;
+    gf_boolean_t inodelk_count_req;
+    gf_boolean_t posixlk_count_req;
+    gf_boolean_t parent_entrylk_req;
+    gf_boolean_t multiple_dom_lk_requests;
+    int update_mlock_enforced_flag;
 } pl_local_t;
 
 typedef struct {
-        struct list_head locks_list;
+    struct list_head locks_list;
 } pl_fdctx_t;
 
+struct _locker {
+    struct list_head lockers;
+    char *volume;
+    inode_t *inode;
+    gf_lkowner_t owner;
+};
+
+typedef struct _locks_ctx {
+    pthread_mutex_t lock;
+    struct list_head inodelk_lockers;
+    struct list_head entrylk_lockers;
+    struct list_head metalk_list;
+} pl_ctx_t;
+
+typedef struct _multi_dom_lk_data {
+    xlator_t *this;
+    inode_t *inode;
+    dict_t *xdata_rsp;
+    gf_boolean_t keep_max;
+} multi_dom_lk_data;
+
+typedef enum { DECREMENT, INCREMENT } pl_count_op_t;
+
+pl_ctx_t *
+pl_ctx_get(client_t *client, xlator_t *xlator);
+
+int
+pl_inodelk_client_cleanup(xlator_t *this, pl_ctx_t *ctx);
+
+int
+pl_entrylk_client_cleanup(xlator_t *this, pl_ctx_t *ctx);
+
 #endif /* __POSIX_LOCKS_H__ */
diff --git a/xlators/features/locks/src/pl-messages.h b/xlators/features/locks/src/pl-messages.h
new file mode 100644
index 00000000000..e2d3d7ca974
--- /dev/null
+++ b/xlators/features/locks/src/pl-messages.h
@@ -0,0 +1,29 @@
+/*
+  Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _PL_MESSAGES_H_
+#define _PL_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(PL, PL_MSG_LOCK_NUMBER, PL_MSG_INODELK_CONTENTION_FAILED,
+           PL_MSG_ENTRYLK_CONTENTION_FAILED);
+
+#endif /* !_PL_MESSAGES_H_ */
diff --git a/xlators/features/locks/src/posix.c b/xlators/features/locks/src/posix.c
index 36c399c115c..cf0ae4c57dd 100644
--- a/xlators/features/locks/src/posix.c
+++ b/xlators/features/locks/src/posix.c
@@ -1,1947 +1,5095 @@
 /*
-  Copyright (c) 2006, 2007, 2008 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
+   Copyright (c) 2006-2012, 2016 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
 #include <unistd.h>
 #include <fcntl.h>
 #include <limits.h>
 #include <pthread.h>
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "compat.h"
-#include "xlator.h"
-#include "inode.h"
-#include "logging.h"
-#include "common-utils.h"
+#include <glusterfs/compat.h>
+#include <glusterfs/logging.h>
 
 #include "locks.h"
 #include "common.h"
-#include "statedump.h"
+#include <glusterfs/statedump.h>
+#include "clear.h"
+#include <glusterfs/defaults.h>
+#include <glusterfs/syncop.h>
 
 #ifndef LLONG_MAX
 #define LLONG_MAX LONG_LONG_MAX /* compat with old gcc */
-#endif /* LLONG_MAX */
+#endif                          /* LLONG_MAX */
 
 /* Forward declarations */
 
+void
+do_blocked_rw(pl_inode_t *);
+static int
+__rw_allowable(pl_inode_t *, posix_lock_t *, glusterfs_fop_t);
+static int
+format_brickname(char *);
+int
+pl_lockinfo_get_brickname(xlator_t *, inode_t *, int32_t *);
+static int
+fetch_pathinfo(xlator_t *, inode_t *, int32_t *, char **);
 
-void do_blocked_rw (pl_inode_t *);
-static int __rw_allowable (pl_inode_t *, posix_lock_t *, glusterfs_fop_t);
+/*
+ * The client is always requesting data, but older
+ * servers were not returning it. Newer ones are, so
+ * the client is receiving a mix of NULL and non-NULL
+ * xdata in the answers when bricks are of different
+ * versions. This triggers a bug in older clients.
+ * To prevent that, we avoid returning extra xdata to
+ * older clients (making the newer brick to behave as
+ * an old brick).
+ */
+#define PL_STACK_UNWIND_FOR_CLIENT(fop, xdata, frame, op_ret, params...)       \
+    do {                                                                       \
+        pl_local_t *__local = NULL;                                            \
+        if (frame->root->client &&                                             \
+            (frame->root->client->opversion < GD_OP_VERSION_3_10_0)) {         \
+            __local = frame->local;                                            \
+            PL_STACK_UNWIND_AND_FREE(__local, fop, frame, op_ret, params);     \
+        } else {                                                               \
+            PL_STACK_UNWIND(fop, xdata, frame, op_ret, params);                \
+        }                                                                      \
+    } while (0)
+
+#define PL_STACK_UNWIND(fop, xdata, frame, op_ret, params...)                  \
+    do {                                                                       \
+        pl_local_t *__local = NULL;                                            \
+        inode_t *__parent = NULL;                                              \
+        inode_t *__inode = NULL;                                               \
+        char *__name = NULL;                                                   \
+        dict_t *__unref = NULL;                                                \
+        int __i = 0;                                                           \
+        __local = frame->local;                                                \
+        if (op_ret >= 0 && pl_needs_xdata_response(frame->local)) {            \
+            if (xdata)                                                         \
+                dict_ref(xdata);                                               \
+            else                                                               \
+                xdata = dict_new();                                            \
+            if (xdata) {                                                       \
+                __unref = xdata;                                               \
+                while (__local->fd || __local->loc[__i].inode) {               \
+                    pl_get_xdata_rsp_args(__local, #fop, &__parent, &__inode,  \
+                                          &__name, __i);                       \
+                    pl_set_xdata_response(frame->this, __local, __parent,      \
+                                          __inode, __name, xdata, __i > 0);    \
+                    if (__local->fd || __i == 1)                               \
+                        break;                                                 \
+                    __i++;                                                     \
+                }                                                              \
+            }                                                                  \
+        }                                                                      \
+        PL_STACK_UNWIND_AND_FREE(__local, fop, frame, op_ret, params);         \
+        if (__unref)                                                           \
+            dict_unref(__unref);                                               \
+    } while (0)
+
+#define PL_LOCAL_GET_REQUESTS(frame, this, xdata, __fd, __loc, __newloc)       \
+    do {                                                                       \
+        if (pl_has_xdata_requests(xdata)) {                                    \
+            if (!frame->local)                                                 \
+                frame->local = mem_get0(this->local_pool);                     \
+            pl_local_t *__local = frame->local;                                \
+            if (__local) {                                                     \
+                if (__fd) {                                                    \
+                    __local->fd = fd_ref(__fd);                                \
+                    __local->inode = inode_ref(__fd->inode);                   \
+                } else {                                                       \
+                    if (__loc)                                                 \
+                        loc_copy(&__local->loc[0], __loc);                     \
+                    if (__newloc)                                              \
+                        loc_copy(&__local->loc[1], __newloc);                  \
+                    __local->inode = inode_ref(__local->loc[0].inode);         \
+                }                                                              \
+                pl_get_xdata_requests(__local, xdata);                         \
+            }                                                                  \
+        }                                                                      \
+    } while (0)
+
+#define PL_CHECK_LOCK_ENFORCE_KEY(frame, dict, name, this, loc, fd, priv)      \
+    do {                                                                       \
+        if ((dict && (dict_get(dict, GF_ENFORCE_MANDATORY_LOCK))) ||           \
+            (name && (strcmp(name, GF_ENFORCE_MANDATORY_LOCK) == 0))) {        \
+            inode_t *__inode = (loc ? loc->inode : fd->inode);                 \
+            pl_inode_t *__pl_inode = pl_inode_get(this, __inode, NULL);        \
+            if (__pl_inode == NULL) {                                          \
+                op_ret = -1;                                                   \
+                op_errno = ENOMEM;                                             \
+                goto unwind;                                                   \
+            }                                                                  \
+            if (!pl_is_mandatory_locking_enabled(__pl_inode) ||                \
+                !priv->mlock_enforced) {                                       \
+                op_ret = -1;                                                   \
+                gf_msg(this->name, GF_LOG_DEBUG, EINVAL, 0,                    \
+                       "option %s would need mandatory lock to be enabled "    \
+                       "and feature.enforce-mandatory-lock option to be set "  \
+                       "to on",                                                \
+                       GF_ENFORCE_MANDATORY_LOCK);                             \
+                op_errno = EINVAL;                                             \
+                goto unwind;                                                   \
+            }                                                                  \
+                                                                               \
+            op_ret = pl_local_init(frame, this, loc, fd);                      \
+            if (op_ret) {                                                      \
+                op_errno = ENOMEM;                                             \
+                goto unwind;                                                   \
+            }                                                                  \
+                                                                               \
+            ((pl_local_t *)(frame->local))->update_mlock_enforced_flag = 1;    \
+        }                                                                      \
+    } while (0)
+
+#define PL_INODE_REMOVE(_fop, _frame, _xl, _loc1, _loc2, _cont, _cbk,          \
+                        _args...)                                              \
+    ({                                                                         \
+        struct list_head contend;                                              \
+        pl_inode_t *__pl_inode;                                                \
+        call_stub_t *__stub;                                                   \
+        int32_t __error;                                                       \
+        INIT_LIST_HEAD(&contend);                                              \
+        __error = pl_inode_remove_prepare(_xl, _frame, _loc2 ? _loc2 : _loc1,  \
+                                          &__pl_inode, &contend);              \
+        if (__error < 0) {                                                     \
+            __stub = fop_##_fop##_stub(_frame, _cont, ##_args);                \
+            __error = pl_inode_remove_complete(_xl, __pl_inode, __stub,        \
+                                               &contend);                      \
+        } else if (__error == 0) {                                             \
+            PL_LOCAL_GET_REQUESTS(_frame, _xl, xdata, ((fd_t *)NULL), _loc1,   \
+                                  _loc2);                                      \
+            STACK_WIND_COOKIE(_frame, _cbk, __pl_inode, FIRST_CHILD(_xl),      \
+                              FIRST_CHILD(_xl)->fops->_fop, ##_args);          \
+        }                                                                      \
+        __error;                                                               \
+    })
+
+gf_boolean_t
+pl_has_xdata_requests(dict_t *xdata)
+{
+    static char *reqs[] = {GLUSTERFS_ENTRYLK_COUNT,
+                           GLUSTERFS_INODELK_COUNT,
+                           GLUSTERFS_INODELK_DOM_COUNT,
+                           GLUSTERFS_POSIXLK_COUNT,
+                           GLUSTERFS_PARENT_ENTRYLK,
+                           GLUSTERFS_MULTIPLE_DOM_LK_CNT_REQUESTS,
+                           NULL};
+    static int reqs_size[] = {SLEN(GLUSTERFS_ENTRYLK_COUNT),
+                              SLEN(GLUSTERFS_INODELK_COUNT),
+                              SLEN(GLUSTERFS_INODELK_DOM_COUNT),
+                              SLEN(GLUSTERFS_POSIXLK_COUNT),
+                              SLEN(GLUSTERFS_PARENT_ENTRYLK),
+                              SLEN(GLUSTERFS_MULTIPLE_DOM_LK_CNT_REQUESTS),
+                              0};
+    int i = 0;
+
+    if (!xdata)
+        return _gf_false;
+
+    for (i = 0; reqs[i]; i++)
+        if (dict_getn(xdata, reqs[i], reqs_size[i]))
+            return _gf_true;
+
+    return _gf_false;
+}
 
-struct _truncate_ops {
-        loc_t  loc;
-        fd_t  *fd;
-        off_t  offset;
-        enum {TRUNCATE, FTRUNCATE} op;
-};
+static int
+dict_delete_domain_key(dict_t *dict, char *key, data_t *value, void *data)
+{
+    dict_del(dict, key);
+    return 0;
+}
 
-static pl_fdctx_t *
-pl_new_fdctx ()
+void
+pl_get_xdata_requests(pl_local_t *local, dict_t *xdata)
 {
-        pl_fdctx_t *fdctx = NULL;
+    if (!local || !xdata)
+        return;
+
+    GF_ASSERT(local->xdata == NULL);
+    local->xdata = dict_copy_with_ref(xdata, NULL);
+
+    if (dict_get_sizen(xdata, GLUSTERFS_ENTRYLK_COUNT)) {
+        local->entrylk_count_req = 1;
+        dict_del_sizen(xdata, GLUSTERFS_ENTRYLK_COUNT);
+    }
+    if (dict_get_sizen(xdata, GLUSTERFS_INODELK_COUNT)) {
+        local->inodelk_count_req = 1;
+        dict_del_sizen(xdata, GLUSTERFS_INODELK_COUNT);
+    }
+    if (dict_get_sizen(xdata, GLUSTERFS_MULTIPLE_DOM_LK_CNT_REQUESTS)) {
+        local->multiple_dom_lk_requests = 1;
+        dict_del_sizen(xdata, GLUSTERFS_MULTIPLE_DOM_LK_CNT_REQUESTS);
+        dict_foreach_fnmatch(xdata, GLUSTERFS_INODELK_DOM_PREFIX "*",
+                             dict_delete_domain_key, NULL);
+    }
+
+    local->inodelk_dom_count_req = dict_get_sizen(xdata,
+                                                  GLUSTERFS_INODELK_DOM_COUNT);
+    if (local->inodelk_dom_count_req) {
+        data_ref(local->inodelk_dom_count_req);
+        dict_del_sizen(xdata, GLUSTERFS_INODELK_DOM_COUNT);
+    }
+
+    if (dict_get_sizen(xdata, GLUSTERFS_POSIXLK_COUNT)) {
+        local->posixlk_count_req = 1;
+        dict_del_sizen(xdata, GLUSTERFS_POSIXLK_COUNT);
+    }
+
+    if (dict_get_sizen(xdata, GLUSTERFS_PARENT_ENTRYLK)) {
+        local->parent_entrylk_req = 1;
+        dict_del_sizen(xdata, GLUSTERFS_PARENT_ENTRYLK);
+    }
+}
 
-        fdctx = GF_CALLOC (1, sizeof (*fdctx),
-                           gf_locks_mt_pl_fdctx_t);
-        if (!fdctx)
-                return NULL;
+gf_boolean_t
+pl_needs_xdata_response(pl_local_t *local)
+{
+    if (!local)
+        return _gf_false;
 
-        INIT_LIST_HEAD (&fdctx->locks_list);
+    if (local->parent_entrylk_req || local->entrylk_count_req ||
+        local->inodelk_dom_count_req || local->inodelk_count_req ||
+        local->posixlk_count_req || local->multiple_dom_lk_requests)
+        return _gf_true;
 
-        return fdctx;
+    return _gf_false;
 }
 
-static pl_fdctx_t *
-pl_get_fdctx (xlator_t *this, fd_t *fd)
+void
+pl_get_xdata_rsp_args(pl_local_t *local, char *fop, inode_t **parent,
+                      inode_t **inode, char **name, int i)
 {
-        int ret = 0;
-        uint64_t tmp = 0;
+    if (strcmp(fop, "lookup") == 0) {
+        *parent = local->loc[0].parent;
+        *inode = local->loc[0].inode;
+        *name = (char *)local->loc[0].name;
+    } else {
+        if (local->fd) {
+            *inode = local->fd->inode;
+        } else {
+            *inode = local->loc[i].parent;
+        }
+    }
+}
 
-        ret = fd_ctx_get (fd, this, &tmp);
-        if (ret) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Could not get fdctx");
-                goto out;
+static inline int
+pl_track_io_fop_count(pl_local_t *local, xlator_t *this, pl_count_op_t op)
+{
+    pl_inode_t *pl_inode = NULL;
+
+    if (!local)
+        return -1;
+
+    pl_inode = pl_inode_get(this, local->inode, NULL);
+    if (!pl_inode)
+        return -1;
+
+    if (pl_inode->mlock_enforced && pl_inode->track_fop_wind_count) {
+        pthread_mutex_lock(&pl_inode->mutex);
+        {
+            if (op == DECREMENT) {
+                pl_inode->fop_wind_count--;
+                /* fop_wind_count can go negative when lock enforcement is
+                 * enabled on unwind path of an IO. Hence the "<" comparision.
+                 */
+                if (pl_inode->fop_wind_count <= 0) {
+                    pthread_cond_broadcast(&pl_inode->check_fop_wind_count);
+                    pl_inode->track_fop_wind_count = _gf_false;
+                    pl_inode->fop_wind_count = 0;
+                }
+            } else {
+                pl_inode->fop_wind_count++;
+            }
         }
+        pthread_mutex_unlock(&pl_inode->mutex);
+    }
+
+    return 0;
+}
+
+static int32_t
+__get_posixlk_count(pl_inode_t *pl_inode)
+{
+    posix_lock_t *lock = NULL;
+    int32_t count = 0;
+
+    list_for_each_entry(lock, &pl_inode->ext_list, list) { count++; }
+
+    return count;
+}
+
+int32_t
+get_posixlk_count(xlator_t *this, inode_t *inode)
+{
+    pl_inode_t *pl_inode = NULL;
+    uint64_t tmp_pl_inode = 0;
+    int32_t count = 0;
+
+    int ret = inode_ctx_get(inode, this, &tmp_pl_inode);
+    if (ret != 0) {
+        goto out;
+    }
+
+    pl_inode = (pl_inode_t *)(long)tmp_pl_inode;
+
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        count = __get_posixlk_count(pl_inode);
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
 
-        return ((pl_fdctx_t *) (long) tmp);
+out:
+    return count;
+}
 
+void
+pl_parent_entrylk_xattr_fill(xlator_t *this, inode_t *parent, char *basename,
+                             dict_t *dict, gf_boolean_t keep_max)
+{
+    int32_t entrylk = 0;
+    int32_t maxcount = -1;
+    int ret = -1;
+
+    if (!parent || !basename)
+        goto out;
+    if (keep_max) {
+        ret = dict_get_int32_sizen(dict, GLUSTERFS_PARENT_ENTRYLK, &maxcount);
+        if (ret < 0)
+            gf_msg_debug(this->name, 0, " Failed to fetch the value for key %s",
+                         GLUSTERFS_PARENT_ENTRYLK);
+    }
+    entrylk = check_entrylk_on_basename(this, parent, basename);
+    if (maxcount >= entrylk)
+        return;
 out:
-        return NULL;
+    ret = dict_set_int32_sizen(dict, GLUSTERFS_PARENT_ENTRYLK, entrylk);
+    if (ret < 0) {
+        gf_msg_debug(this->name, 0, " dict_set failed on key %s",
+                     GLUSTERFS_PARENT_ENTRYLK);
+    }
+}
+
+void
+pl_entrylk_xattr_fill(xlator_t *this, inode_t *inode, dict_t *dict,
+                      gf_boolean_t keep_max)
+{
+    int32_t count = 0;
+    int32_t maxcount = -1;
+    int ret = -1;
+
+    if (keep_max) {
+        ret = dict_get_int32_sizen(dict, GLUSTERFS_ENTRYLK_COUNT, &maxcount);
+        if (ret < 0)
+            gf_msg_debug(this->name, 0, " Failed to fetch the value for key %s",
+                         GLUSTERFS_ENTRYLK_COUNT);
+    }
+    count = get_entrylk_count(this, inode);
+    if (maxcount >= count)
+        return;
 
+    ret = dict_set_int32_sizen(dict, GLUSTERFS_ENTRYLK_COUNT, count);
+    if (ret < 0) {
+        gf_msg_debug(this->name, 0, " dict_set failed on key %s",
+                     GLUSTERFS_ENTRYLK_COUNT);
+    }
+}
+
+void
+pl_inodelk_xattr_fill(xlator_t *this, inode_t *inode, dict_t *dict,
+                      char *domname, gf_boolean_t keep_max)
+{
+    int32_t count = 0;
+    int32_t maxcount = -1;
+    int ret = -1;
+
+    if (keep_max) {
+        ret = dict_get_int32_sizen(dict, GLUSTERFS_INODELK_COUNT, &maxcount);
+        if (ret < 0)
+            gf_msg_debug(this->name, 0, " Failed to fetch the value for key %s",
+                         GLUSTERFS_INODELK_COUNT);
+    }
+    count = get_inodelk_count(this, inode, domname);
+    if (maxcount >= count)
+        return;
+
+    ret = dict_set_int32_sizen(dict, GLUSTERFS_INODELK_COUNT, count);
+    if (ret < 0) {
+        gf_msg_debug(this->name, 0,
+                     "Failed to set count for "
+                     "key %s",
+                     GLUSTERFS_INODELK_COUNT);
+    }
+
+    return;
+}
+
+void
+pl_posixlk_xattr_fill(xlator_t *this, inode_t *inode, dict_t *dict,
+                      gf_boolean_t keep_max)
+{
+    int32_t count = 0;
+    int32_t maxcount = -1;
+    int ret = -1;
+
+    if (keep_max) {
+        ret = dict_get_int32_sizen(dict, GLUSTERFS_POSIXLK_COUNT, &maxcount);
+        if (ret < 0)
+            gf_msg_debug(this->name, 0, " Failed to fetch the value for key %s",
+                         GLUSTERFS_POSIXLK_COUNT);
+    }
+    count = get_posixlk_count(this, inode);
+    if (maxcount >= count)
+        return;
+
+    ret = dict_set_int32_sizen(dict, GLUSTERFS_POSIXLK_COUNT, count);
+    if (ret < 0) {
+        gf_msg_debug(this->name, 0, " dict_set failed on key %s",
+                     GLUSTERFS_POSIXLK_COUNT);
+    }
+}
+
+void
+pl_inodelk_xattr_fill_each(xlator_t *this, inode_t *inode, dict_t *dict,
+                           char *domname, gf_boolean_t keep_max, char *key)
+{
+    int32_t count = 0;
+    int32_t maxcount = -1;
+    int ret = -1;
+
+    if (keep_max) {
+        ret = dict_get_int32(dict, key, &maxcount);
+        if (ret < 0)
+            gf_msg_debug(this->name, 0, " Failed to fetch the value for key %s",
+                         GLUSTERFS_INODELK_COUNT);
+    }
+    count = get_inodelk_count(this, inode, domname);
+    if (maxcount >= count)
+        return;
+
+    ret = dict_set_int32(dict, key, count);
+    if (ret < 0) {
+        gf_msg_debug(this->name, 0,
+                     "Failed to set count for "
+                     "key %s",
+                     key);
+    }
+
+    return;
 }
 
 static int
-pl_set_fdctx (xlator_t *this, fd_t *fd,
-              pl_fdctx_t *fdctx)
+pl_inodelk_xattr_fill_multiple(dict_t *this, char *key, data_t *value,
+                               void *data)
+{
+    multi_dom_lk_data *d = data;
+    char *tmp_key = NULL;
+    char *save_ptr = NULL;
+
+    tmp_key = gf_strdup(key);
+    if (!tmp_key)
+        return -1;
+
+    strtok_r(tmp_key, ":", &save_ptr);
+    if (!*save_ptr) {
+        if (tmp_key)
+            GF_FREE(tmp_key);
+        gf_msg(THIS->name, GF_LOG_ERROR, 0, EINVAL,
+               "Could not tokenize domain string from key %s", key);
+        return -1;
+    }
+
+    pl_inodelk_xattr_fill_each(d->this, d->inode, d->xdata_rsp, save_ptr,
+                               d->keep_max, key);
+    if (tmp_key)
+        GF_FREE(tmp_key);
+
+    return 0;
+}
+
+void
+pl_fill_multiple_dom_lk_requests(xlator_t *this, pl_local_t *local,
+                                 inode_t *inode, dict_t *dict,
+                                 gf_boolean_t keep_max)
 {
-        int ret = 0;
+    multi_dom_lk_data data;
 
-        ret = fd_ctx_set (fd, this,
-                          (uint64_t) (unsigned long) (fdctx));
+    data.this = this;
+    data.inode = inode;
+    data.xdata_rsp = dict;
+    data.keep_max = keep_max;
 
-        if (ret) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "Could not set fdctx");
-                goto out;
+    dict_foreach_fnmatch(local->xdata, GLUSTERFS_INODELK_DOM_PREFIX "*",
+                         pl_inodelk_xattr_fill_multiple, &data);
+}
+
+void
+pl_set_xdata_response(xlator_t *this, pl_local_t *local, inode_t *parent,
+                      inode_t *inode, char *name, dict_t *xdata,
+                      gf_boolean_t max_lock)
+{
+    if (!xdata || !local)
+        return;
+
+    if (local->parent_entrylk_req && parent && name && name[0] != '\0')
+        pl_parent_entrylk_xattr_fill(this, parent, name, xdata, max_lock);
+
+    if (!inode)
+        return;
+
+    if (local->entrylk_count_req)
+        pl_entrylk_xattr_fill(this, inode, xdata, max_lock);
+
+    if (local->inodelk_dom_count_req)
+        pl_inodelk_xattr_fill(this, inode, xdata,
+                              data_to_str(local->inodelk_dom_count_req),
+                              max_lock);
+
+    if (local->inodelk_count_req)
+        pl_inodelk_xattr_fill(this, inode, xdata, NULL, max_lock);
+
+    if (local->posixlk_count_req)
+        pl_posixlk_xattr_fill(this, inode, xdata, max_lock);
+
+    if (local->multiple_dom_lk_requests)
+        pl_fill_multiple_dom_lk_requests(this, local, inode, xdata, max_lock);
+}
+
+/* Checks whether the region where fop is acting upon conflicts
+ * with existing locks. If there is no conflict function returns
+ * 1 else returns 0 with can_block boolean set accordingly to
+ * indicate block/fail the fop.
+ */
+int
+pl_is_fop_allowed(pl_inode_t *pl_inode, posix_lock_t *region, fd_t *fd,
+                  glusterfs_fop_t op, gf_boolean_t *can_block)
+{
+    int ret = 0;
+
+    if (!__rw_allowable(pl_inode, region, op)) {
+        if (pl_inode->mlock_enforced) {
+            *can_block = _gf_false;
+        } else if ((!fd) || (fd && (fd->flags & O_NONBLOCK))) {
+            gf_log("locks", GF_LOG_TRACE,
+                   "returning EAGAIN"
+                   " because fd is O_NONBLOCK");
+            *can_block = _gf_false;
+        } else {
+            *can_block = _gf_true;
         }
+    } else {
+        ret = 1;
+    }
+
+    return ret;
+}
+
+static pl_fdctx_t *
+pl_new_fdctx()
+{
+    pl_fdctx_t *fdctx = GF_MALLOC(sizeof(*fdctx), gf_locks_mt_pl_fdctx_t);
+    GF_VALIDATE_OR_GOTO("posix-locks", fdctx, out);
+
+    INIT_LIST_HEAD(&fdctx->locks_list);
 
 out:
-        return ret;
+    return fdctx;
+}
+
+static pl_fdctx_t *
+pl_check_n_create_fdctx(xlator_t *this, fd_t *fd)
+{
+    int ret = 0;
+    uint64_t tmp = 0;
+    pl_fdctx_t *fdctx = NULL;
+
+    GF_VALIDATE_OR_GOTO("posix-locks", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    LOCK(&fd->lock);
+    {
+        ret = __fd_ctx_get(fd, this, &tmp);
+        if ((ret != 0) || (tmp == 0)) {
+            fdctx = pl_new_fdctx();
+            if (fdctx == NULL) {
+                goto unlock;
+            }
+        }
+
+        ret = __fd_ctx_set(fd, this, (uint64_t)(long)fdctx);
+        if (ret != 0) {
+            GF_FREE(fdctx);
+            fdctx = NULL;
+            UNLOCK(&fd->lock);
+            gf_log(this->name, GF_LOG_DEBUG, "failed to set fd ctx");
+            goto out;
+        }
+    }
+unlock:
+    UNLOCK(&fd->lock);
+
+out:
+    return fdctx;
+}
 
+int32_t
+pl_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+               struct iatt *postbuf, dict_t *xdata)
+{
+    pl_track_io_fop_count(frame->local, this, DECREMENT);
+
+    PL_STACK_UNWIND(discard, xdata, frame, op_ret, op_errno, prebuf, postbuf,
+                    xdata);
+    return 0;
 }
 
 int
-pl_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                 int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                 struct iatt *postbuf)
+pl_discard_cont(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                size_t len, dict_t *xdata)
 {
-        struct _truncate_ops *local = NULL;
+    pl_track_io_fop_count(frame->local, this, INCREMENT);
 
-        local = frame->local;
+    STACK_WIND(frame, pl_discard_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->discard, fd, offset, len, xdata);
+    return 0;
+}
+
+int32_t
+pl_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+           size_t len, dict_t *xdata)
+{
+    pl_local_t *local = NULL;
+    pl_inode_t *pl_inode = NULL;
+    pl_rw_req_t *rw = NULL;
+    posix_lock_t region = {
+        .list =
+            {
+                0,
+            },
+    };
+    gf_boolean_t enabled = _gf_false;
+    gf_boolean_t can_block = _gf_true;
+    int op_ret = 0;
+    int op_errno = 0;
+    int allowed = 1;
+
+    GF_VALIDATE_OR_GOTO("locks", this, unwind);
+
+    local = mem_get0(this->local_pool);
+    if (!local) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+
+    frame->local = local;
+    local->inode = inode_ref(fd->inode);
+    local->fd = fd_ref(fd);
+
+    pl_inode = pl_inode_get(this, fd->inode, local);
+    if (!pl_inode) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+
+    if (frame->root->pid < 0)
+        enabled = _gf_false;
+    else
+        enabled = pl_is_mandatory_locking_enabled(pl_inode);
+
+    if (enabled) {
+        region.fl_start = offset;
+        region.fl_end = offset + len - 1;
+        region.client = frame->root->client;
+        region.fd_num = fd_to_fdnum(fd);
+        region.client_pid = frame->root->pid;
+        region.owner = frame->root->lk_owner;
+
+        pthread_mutex_lock(&pl_inode->mutex);
+        {
+            allowed = pl_is_fop_allowed(pl_inode, &region, fd, GF_FOP_DISCARD,
+                                        &can_block);
+            if (allowed == 1) {
+                if (pl_inode->mlock_enforced &&
+                    pl_inode->track_fop_wind_count) {
+                    pl_inode->fop_wind_count++;
+                }
+                goto unlock;
+            } else if (!can_block) {
+                op_errno = EAGAIN;
+                op_ret = -1;
+                goto unlock;
+            }
 
-        if (local->op == TRUNCATE)
-                loc_wipe (&local->loc);
+            rw = GF_MALLOC(sizeof(*rw), gf_locks_mt_pl_rw_req_t);
+            if (!rw) {
+                op_errno = ENOMEM;
+                op_ret = -1;
+                goto unlock;
+            }
 
-        STACK_UNWIND_STRICT (truncate, frame, op_ret, op_errno,
-                             prebuf, postbuf);
-        return 0;
+            rw->stub = fop_discard_stub(frame, pl_discard_cont, fd, offset, len,
+                                        xdata);
+            if (!rw->stub) {
+                op_errno = ENOMEM;
+                op_ret = -1;
+                GF_FREE(rw);
+                goto unlock;
+            }
+
+            rw->region = region;
+
+            list_add_tail(&rw->list, &pl_inode->rw_list);
+        }
+    unlock:
+        pthread_mutex_unlock(&pl_inode->mutex);
+    }
+
+    if (allowed == 1)
+        STACK_WIND(frame, pl_discard_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->discard, fd, offset, len, xdata);
+unwind:
+    if (op_ret == -1)
+        PL_STACK_UNWIND(discard, xdata, frame, op_ret, op_errno, NULL, NULL,
+                        NULL);
+
+    return 0;
 }
 
+int32_t
+pl_zerofill_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                struct iatt *postbuf, dict_t *xdata)
+{
+    pl_track_io_fop_count(frame->local, this, DECREMENT);
 
-static int
-truncate_allowed (pl_inode_t *pl_inode,
-                  void *transport, pid_t client_pid,
-                  uint64_t owner, off_t offset)
+    PL_STACK_UNWIND(zerofill, xdata, frame, op_ret, op_errno, prebuf, postbuf,
+                    xdata);
+    return 0;
+}
+
+int
+pl_zerofill_cont(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                 off_t len, dict_t *xdata)
 {
-        posix_lock_t *l = NULL;
-        posix_lock_t  region = {.list = {0, }, };
-        int           ret = 1;
+    pl_track_io_fop_count(frame->local, this, INCREMENT);
 
-        region.fl_start   = offset;
-        region.fl_end     = LLONG_MAX;
-        region.transport  = transport;
-        region.client_pid = client_pid;
-        region.owner      = owner;
+    STACK_WIND(frame, pl_zerofill_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->zerofill, fd, offset, len, xdata);
+    return 0;
+}
 
-        pthread_mutex_lock (&pl_inode->mutex);
+int32_t
+pl_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+            off_t len, dict_t *xdata)
+{
+    pl_local_t *local = NULL;
+    pl_inode_t *pl_inode = NULL;
+    pl_rw_req_t *rw = NULL;
+    posix_lock_t region = {
+        .list =
+            {
+                0,
+            },
+    };
+    gf_boolean_t enabled = _gf_false;
+    gf_boolean_t can_block = _gf_true;
+    int op_ret = 0;
+    int op_errno = 0;
+    int allowed = 1;
+
+    GF_VALIDATE_OR_GOTO("locks", this, unwind);
+
+    local = mem_get0(this->local_pool);
+    if (!local) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+
+    frame->local = local;
+    local->inode = inode_ref(fd->inode);
+    local->fd = fd_ref(fd);
+
+    pl_inode = pl_inode_get(this, fd->inode, local);
+    if (!pl_inode) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+
+    if (frame->root->pid < 0)
+        enabled = _gf_false;
+    else
+        enabled = pl_is_mandatory_locking_enabled(pl_inode);
+
+    if (enabled) {
+        region.fl_start = offset;
+        region.fl_end = offset + len - 1;
+        region.client = frame->root->client;
+        region.fd_num = fd_to_fdnum(fd);
+        region.client_pid = frame->root->pid;
+        region.owner = frame->root->lk_owner;
+
+        pthread_mutex_lock(&pl_inode->mutex);
         {
-                list_for_each_entry (l, &pl_inode->ext_list, list) {
-                        if (!l->blocked
-                            && locks_overlap (&region, l)
-                            && !same_owner (&region, l)) {
-                                ret = 0;
-                                break;
-                        }
+            allowed = pl_is_fop_allowed(pl_inode, &region, fd, GF_FOP_ZEROFILL,
+                                        &can_block);
+            if (allowed == 1) {
+                if (pl_inode->mlock_enforced &&
+                    pl_inode->track_fop_wind_count) {
+                    pl_inode->fop_wind_count++;
                 }
+                goto unlock;
+            } else if (!can_block) {
+                op_errno = EAGAIN;
+                op_ret = -1;
+                goto unlock;
+            }
+
+            rw = GF_MALLOC(sizeof(*rw), gf_locks_mt_pl_rw_req_t);
+            if (!rw) {
+                op_errno = ENOMEM;
+                op_ret = -1;
+                goto unlock;
+            }
+
+            rw->stub = fop_zerofill_stub(frame, pl_zerofill_cont, fd, offset,
+                                         len, xdata);
+            if (!rw->stub) {
+                op_errno = ENOMEM;
+                op_ret = -1;
+                GF_FREE(rw);
+                goto unlock;
+            }
+
+            rw->region = region;
+
+            list_add_tail(&rw->list, &pl_inode->rw_list);
         }
-        pthread_mutex_unlock (&pl_inode->mutex);
+    unlock:
+        pthread_mutex_unlock(&pl_inode->mutex);
+    }
 
-        return ret;
+    if (allowed == 1)
+        STACK_WIND(frame, pl_zerofill_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->zerofill, fd, offset, len, xdata);
+unwind:
+    if (op_ret == -1)
+        PL_STACK_UNWIND(zerofill, xdata, frame, op_ret, op_errno, NULL, NULL,
+                        NULL);
+
+    return 0;
 }
 
+int
+pl_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                struct iatt *postbuf, dict_t *xdata)
+{
+    pl_local_t *local = frame->local;
+
+    pl_track_io_fop_count(local, this, DECREMENT);
 
-static int
-truncate_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno, struct iatt *buf)
+    if (local->op == GF_FOP_TRUNCATE)
+        PL_STACK_UNWIND(truncate, xdata, frame, op_ret, op_errno, prebuf,
+                        postbuf, xdata);
+    else
+        PL_STACK_UNWIND(ftruncate, xdata, frame, op_ret, op_errno, prebuf,
+                        postbuf, xdata);
+    return 0;
+}
+
+int
+pl_ftruncate_cont(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                  dict_t *xdata)
 {
-        posix_locks_private_t *priv = NULL;
-        struct _truncate_ops  *local = NULL;
-        inode_t               *inode = NULL;
-        pl_inode_t            *pl_inode = NULL;
+    pl_track_io_fop_count(frame->local, this, INCREMENT);
 
+    STACK_WIND(frame, pl_truncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+    return 0;
+}
 
-        priv = this->private;
-        local = frame->local;
+int
+pl_truncate_cont(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+                 dict_t *xdata)
+{
+    pl_track_io_fop_count(frame->local, this, INCREMENT);
 
-        if (op_ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "got error (errno=%d, stderror=%s) from child",
-                        op_errno, strerror (op_errno));
-                goto unwind;
-        }
+    STACK_WIND(frame, pl_truncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+    return 0;
+}
 
-        if (local->op == TRUNCATE)
-                inode = local->loc.inode;
-        else
-                inode = local->fd->inode;
+static int
+truncate_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                  dict_t *xdata)
+{
+    pl_local_t *local = frame->local;
+    inode_t *inode = NULL;
+    pl_inode_t *pl_inode = NULL;
+    pl_rw_req_t *rw = NULL;
+    posix_lock_t region = {
+        .list =
+            {
+                0,
+            },
+    };
+    gf_boolean_t enabled = _gf_false;
+    gf_boolean_t can_block = _gf_true;
+    int allowed = 1;
+
+    GF_VALIDATE_OR_GOTO("locks", this, unwind);
+
+    if (op_ret != 0) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "got error (errno=%d, stderror=%s) from child", op_errno,
+               strerror(op_errno));
+        goto unwind;
+    }
+
+    if (local->op == GF_FOP_TRUNCATE)
+        inode = local->loc[0].inode;
+    else
+        inode = local->fd->inode;
+
+    local->inode = inode_ref(inode);
+
+    pl_inode = pl_inode_get(this, inode, local);
+    if (!pl_inode) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+
+    if (frame->root->pid < 0)
+        enabled = _gf_false;
+    else
+        enabled = pl_is_mandatory_locking_enabled(pl_inode);
+
+    if (enabled) {
+        region.fl_start = local->offset;
+        region.fl_end = LLONG_MAX;
+        region.client = frame->root->client;
+        region.fd_num = fd_to_fdnum(local->fd);
+        region.client_pid = frame->root->pid;
+        region.owner = frame->root->lk_owner;
+        pthread_mutex_lock(&pl_inode->mutex);
+        {
+            allowed = pl_is_fop_allowed(pl_inode, &region, local->fd, local->op,
+                                        &can_block);
 
-        pl_inode = pl_inode_get (this, inode);
-        if (!pl_inode) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory.");
-                op_ret   = -1;
+            if (allowed == 1) {
+                if (pl_inode->mlock_enforced &&
+                    pl_inode->track_fop_wind_count) {
+                    pl_inode->fop_wind_count++;
+                }
+                goto unlock;
+            } else if (!can_block) {
+                op_errno = EAGAIN;
+                op_ret = -1;
+                goto unlock;
+            }
+
+            rw = GF_MALLOC(sizeof(*rw), gf_locks_mt_pl_rw_req_t);
+            if (!rw) {
                 op_errno = ENOMEM;
-                goto unwind;
-        }
+                op_ret = -1;
+                goto unlock;
+            }
+
+            if (local->op == GF_FOP_TRUNCATE)
+                rw->stub = fop_truncate_stub(frame, pl_truncate_cont,
+                                             &local->loc[0], local->offset,
+                                             local->xdata);
+            else
+                rw->stub = fop_ftruncate_stub(frame, pl_ftruncate_cont,
+                                              local->fd, local->offset,
+                                              local->xdata);
+            if (!rw->stub) {
+                op_errno = ENOMEM;
+                op_ret = -1;
+                GF_FREE(rw);
+                goto unlock;
+            }
 
-        if (priv->mandatory
-            && pl_inode->mandatory
-            && !truncate_allowed (pl_inode, frame->root->trans,
-                                  frame->root->pid, frame->root->lk_owner,
-                                  local->offset)) {
-                op_ret   = -1;
-                op_errno = EAGAIN;
-                goto unwind;
+            rw->region = region;
+
+            list_add_tail(&rw->list, &pl_inode->rw_list);
         }
+    unlock:
+        pthread_mutex_unlock(&pl_inode->mutex);
+    }
 
+    if (allowed == 1) {
         switch (local->op) {
-        case TRUNCATE:
-                STACK_WIND (frame, pl_truncate_cbk, FIRST_CHILD (this),
-                            FIRST_CHILD (this)->fops->truncate,
-                            &local->loc, local->offset);
+            case GF_FOP_TRUNCATE:
+                STACK_WIND(frame, pl_truncate_cbk, FIRST_CHILD(this),
+                           FIRST_CHILD(this)->fops->truncate, &local->loc[0],
+                           local->offset, local->xdata);
+                break;
+            case GF_FOP_FTRUNCATE:
+                STACK_WIND(frame, pl_truncate_cbk, FIRST_CHILD(this),
+                           FIRST_CHILD(this)->fops->ftruncate, local->fd,
+                           local->offset, local->xdata);
                 break;
-        case FTRUNCATE:
-                STACK_WIND (frame, pl_truncate_cbk, FIRST_CHILD (this),
-                            FIRST_CHILD (this)->fops->ftruncate,
-                            local->fd, local->offset);
+            default:
                 break;
         }
-
-        return 0;
-
+    }
 unwind:
-        if (local->op == TRUNCATE)
-                loc_wipe (&local->loc);
+    if (op_ret == -1) {
+        gf_log(this ? this->name : "locks", GF_LOG_ERROR,
+               "truncate failed with "
+               "ret: %d, error: %s",
+               op_ret, strerror(op_errno));
 
-        STACK_UNWIND_STRICT (truncate, frame, op_ret, op_errno, buf, NULL);
-        return 0;
+        switch (local->op) {
+            case GF_FOP_TRUNCATE:
+                PL_STACK_UNWIND(truncate, xdata, frame, op_ret, op_errno, buf,
+                                NULL, xdata);
+                break;
+            case GF_FOP_FTRUNCATE:
+                PL_STACK_UNWIND(ftruncate, xdata, frame, op_ret, op_errno, buf,
+                                NULL, xdata);
+                break;
+            default:
+                break;
+        }
+    }
+    return 0;
 }
 
-
 int
-pl_truncate (call_frame_t *frame, xlator_t *this,
-             loc_t *loc, off_t offset)
+pl_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+            dict_t *xdata)
 {
-        struct _truncate_ops *local = NULL;
+    pl_local_t *local = NULL;
+    int ret = -1;
 
-	local = GF_CALLOC (1, sizeof (struct _truncate_ops),
-                           gf_locks_mt_truncate_ops);
-        if (!local) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory.");
-                goto unwind;
-        }
+    GF_VALIDATE_OR_GOTO("locks", this, unwind);
 
-        local->op         = TRUNCATE;
-        local->offset     = offset;
-        loc_copy (&local->loc, loc);
+    local = mem_get0(this->local_pool);
+    GF_VALIDATE_OR_GOTO(this->name, local, unwind);
 
-        frame->local = local;
+    local->op = GF_FOP_TRUNCATE;
+    local->offset = offset;
+    loc_copy(&local->loc[0], loc);
+    if (xdata)
+        local->xdata = dict_ref(xdata);
 
-        STACK_WIND (frame, truncate_stat_cbk, FIRST_CHILD (this),
-                    FIRST_CHILD (this)->fops->stat, loc);
+    frame->local = local;
 
-        return 0;
+    STACK_WIND(frame, truncate_stat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->stat, loc, NULL);
+    ret = 0;
 
 unwind:
-        STACK_UNWIND_STRICT (truncate, frame, -1, ENOMEM, NULL, NULL);
-
-        return 0;
+    if (ret == -1) {
+        gf_log(this ? this->name : "locks", GF_LOG_ERROR,
+               "truncate on %s failed with"
+               " ret: %d, error: %s",
+               loc->path, -1, strerror(ENOMEM));
+        STACK_UNWIND_STRICT(truncate, frame, -1, ENOMEM, NULL, NULL, NULL);
+    }
+    return 0;
 }
 
-
 int
-pl_ftruncate (call_frame_t *frame, xlator_t *this,
-              fd_t *fd, off_t offset)
+pl_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             dict_t *xdata)
 {
-        struct _truncate_ops *local = NULL;
-
-	local = GF_CALLOC (1, sizeof (struct _truncate_ops),
-                           gf_locks_mt_truncate_ops);
-        if (!local) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory.");
-                goto unwind;
-        }
+    pl_local_t *local = NULL;
+    int ret = -1;
 
-        local->op         = FTRUNCATE;
-        local->offset     = offset;
-        local->fd         = fd;
+    GF_VALIDATE_OR_GOTO("locks", this, unwind);
+    local = mem_get0(this->local_pool);
+    GF_VALIDATE_OR_GOTO(this->name, local, unwind);
 
-        frame->local = local;
+    local->op = GF_FOP_FTRUNCATE;
+    local->offset = offset;
+    local->fd = fd_ref(fd);
+    if (xdata)
+        local->xdata = dict_ref(xdata);
 
-        STACK_WIND (frame, truncate_stat_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fstat, fd);
-        return 0;
+    frame->local = local;
 
+    STACK_WIND(frame, truncate_stat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fstat, fd, xdata);
+    ret = 0;
 unwind:
-        STACK_UNWIND_STRICT (ftruncate, frame, -1, ENOMEM, NULL, NULL);
-
-        return 0;
+    if (ret == -1) {
+        gf_log(this ? this->name : "locks", GF_LOG_ERROR,
+               "ftruncate failed with"
+               " ret: %d, error: %s",
+               -1, strerror(ENOMEM));
+        STACK_UNWIND_STRICT(ftruncate, frame, -1, ENOMEM, NULL, NULL, NULL);
+    }
+    return 0;
 }
 
-static void
-delete_locks_of_fd (xlator_t *this, pl_inode_t *pl_inode, fd_t *fd)
+int
+pl_locks_by_fd(pl_inode_t *pl_inode, fd_t *fd)
 {
-       posix_lock_t *tmp = NULL;
-       posix_lock_t *l = NULL;
+    posix_lock_t *l = NULL;
+    int found = 0;
 
-       struct list_head blocked_list;
-
-       INIT_LIST_HEAD (&blocked_list);
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        list_for_each_entry(l, &pl_inode->ext_list, list)
+        {
+            if (l->fd_num == fd_to_fdnum(fd)) {
+                found = 1;
+                break;
+            }
+        }
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+    return found;
+}
 
-       pthread_mutex_lock (&pl_inode->mutex);
-       {
+static void
+delete_locks_of_fd(xlator_t *this, pl_inode_t *pl_inode, fd_t *fd)
+{
+    posix_lock_t *tmp = NULL;
+    posix_lock_t *l = NULL;
 
-               list_for_each_entry_safe (l, tmp, &pl_inode->ext_list, list) {
-                       if ((l->fd_num == fd_to_fdnum(fd))) {
-                               if (l->blocked) {
-                                       list_move_tail (&l->list, &blocked_list);
-                                       continue;
-                               }
-                               __delete_lock (pl_inode, l);
-                               __destroy_lock (l);
-                       }
-               }
+    struct list_head blocked_list;
 
-       }
-       pthread_mutex_unlock (&pl_inode->mutex);
+    INIT_LIST_HEAD(&blocked_list);
 
-       list_for_each_entry_safe (l, tmp, &blocked_list, list) {
-               list_del_init(&l->list);
-               STACK_UNWIND_STRICT (lk, l->frame, -1, EAGAIN, &l->user_flock);
-               __destroy_lock (l);
-       }
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        list_for_each_entry_safe(l, tmp, &pl_inode->ext_list, list)
+        {
+            if (l->fd_num == fd_to_fdnum(fd)) {
+                if (l->blocked) {
+                    list_move_tail(&l->list, &blocked_list);
+                    continue;
+                }
+                __delete_lock(l);
+                __destroy_lock(l);
+            }
+        }
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
 
-        grant_blocked_locks (this, pl_inode);
+    list_for_each_entry_safe(l, tmp, &blocked_list, list)
+    {
+        list_del_init(&l->list);
+        STACK_UNWIND_STRICT(lk, l->frame, -1, EAGAIN, &l->user_flock, NULL);
+        __destroy_lock(l);
+    }
 
-        do_blocked_rw (pl_inode);
+    grant_blocked_locks(this, pl_inode);
 
+    do_blocked_rw(pl_inode);
 }
 
 static void
-__delete_locks_of_owner (pl_inode_t *pl_inode,
-                         void *transport, uint64_t owner)
-{
-        posix_lock_t *tmp = NULL;
-        posix_lock_t *l = NULL;
-
-        /* TODO: what if it is a blocked lock with pending l->frame */
-
-        list_for_each_entry_safe (l, tmp, &pl_inode->ext_list, list) {
-                if ((l->transport == transport)
-                    && (l->owner == owner)) {
-			gf_log ("posix-locks", GF_LOG_TRACE,
-                                " Flushing lock"
-				"%s (pid=%d) (lk-owner=%"PRIu64") %"PRId64" - %"PRId64" state: %s",
-				l->fl_type == F_UNLCK ? "Unlock" : "Lock",
-				l->client_pid,
-                                l->owner,
-				l->user_flock.l_start,
-				l->user_flock.l_len,
-                                l->blocked == 1 ? "Blocked" : "Active");
-
-                        __delete_lock (pl_inode, l);
-                        __destroy_lock (l);
-                }
+__delete_locks_of_owner(pl_inode_t *pl_inode, client_t *client,
+                        gf_lkowner_t *owner)
+{
+    posix_lock_t *tmp = NULL;
+    posix_lock_t *l = NULL;
+
+    /* TODO: what if it is a blocked lock with pending l->frame */
+
+    list_for_each_entry_safe(l, tmp, &pl_inode->ext_list, list)
+    {
+        if (l->blocked)
+            continue;
+        if ((l->client == client) && is_same_lkowner(&l->owner, owner)) {
+            gf_log("posix-locks", GF_LOG_TRACE,
+                   " Flushing lock"
+                   "%s (pid=%d) (lk-owner=%s) %" PRId64 " - %" PRId64
+                   " state: %s",
+                   l->fl_type == F_UNLCK ? "Unlock" : "Lock", l->client_pid,
+                   lkowner_utoa(&l->owner), l->user_flock.l_start,
+                   l->user_flock.l_len, l->blocked == 1 ? "Blocked" : "Active");
+
+            __delete_lock(l);
+            __destroy_lock(l);
         }
+    }
 
-        return;
+    return;
 }
 
 int32_t
-pl_opendir_cbk (call_frame_t *frame,
-		     void *cookie,
-		     xlator_t *this,
-		     int32_t op_ret,
-		     int32_t op_errno,
-		     fd_t *fd)
+pl_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
 {
-        pl_fdctx_t *fdctx = NULL;
-        int ret = -1;
+    STACK_UNWIND_STRICT(getxattr, frame, op_ret, op_errno, dict, xdata);
+    return 0;
+}
 
-        if (op_ret < 0)
-                goto unwind;
+static int32_t
+pl_getxattr_clrlk(xlator_t *this, const char *name, inode_t *inode,
+                  dict_t **dict, int32_t *op_errno)
+{
+    int32_t bcount = 0;
+    int32_t gcount = 0;
+    char *key = NULL;
+    char *lk_summary = NULL;
+    pl_inode_t *pl_inode = NULL;
+    clrlk_args args = {
+        0,
+    };
+    char *brickname = NULL;
+    int32_t op_ret = -1;
+
+    *op_errno = EINVAL;
+
+    if (clrlk_parse_args(name, &args)) {
+        *op_errno = EINVAL;
+        goto out;
+    }
+
+    *dict = dict_new();
+    if (!*dict) {
+        *op_errno = ENOMEM;
+        goto out;
+    }
+
+    pl_inode = pl_inode_get(this, inode, NULL);
+    if (!pl_inode) {
+        *op_errno = ENOMEM;
+        goto out;
+    }
+
+    switch (args.type) {
+        case CLRLK_INODE:
+        case CLRLK_ENTRY:
+            op_ret = clrlk_clear_lks_in_all_domains(this, pl_inode, &args,
+                                                    &bcount, &gcount, op_errno);
+            break;
+        case CLRLK_POSIX:
+            op_ret = clrlk_clear_posixlk(this, pl_inode, &args, &bcount,
+                                         &gcount, op_errno);
+            break;
+        default:
+            op_ret = -1;
+            *op_errno = EINVAL;
+    }
+    if (op_ret) {
+        if (args.type >= CLRLK_TYPE_MAX) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "clear locks: invalid lock type %d", args.type);
+        } else {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "clear locks of type %s failed: %s",
+                   clrlk_type_names[args.type], strerror(*op_errno));
+        }
 
-        fdctx = pl_get_fdctx (this, fd);
-        if (!fdctx) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "fdctx not present");
-                fdctx = pl_new_fdctx ();
-                if (!fdctx) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Out of memory");
-                        op_errno = ENOMEM;
-                        op_ret   = -1;
-                        goto unwind;
-                }
+        goto out;
+    }
 
-                ret = pl_set_fdctx (this, fd, fdctx);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "could not set fdctx");
-                        op_ret = -1;
-                        op_errno = ENOMEM;
-                        goto unwind;
-                }
+    op_ret = fetch_pathinfo(this, inode, op_errno, &brickname);
+    if (op_ret) {
+        gf_log(this->name, GF_LOG_WARNING, "Couldn't get brickname");
+    } else {
+        op_ret = format_brickname(brickname);
+        if (op_ret) {
+            gf_log(this->name, GF_LOG_WARNING, "Couldn't format brickname");
+            GF_FREE(brickname);
+            brickname = NULL;
         }
+    }
 
-unwind:
-	STACK_UNWIND_STRICT (opendir,
-                             frame,
-                             op_ret,
-                             op_errno,
-                             fd);
-	return 0;
+    if (!gcount && !bcount) {
+        if (gf_asprintf(&lk_summary, "No locks cleared.") == -1) {
+            op_ret = -1;
+            *op_errno = ENOMEM;
+            goto out;
+        }
+    } else if (gf_asprintf(&lk_summary,
+                           "%s: %s blocked locks=%d "
+                           "granted locks=%d",
+                           (brickname == NULL) ? this->name : brickname,
+                           clrlk_type_names[args.type], bcount, gcount) == -1) {
+        op_ret = -1;
+        *op_errno = ENOMEM;
+        goto out;
+    }
+    gf_log(this->name, GF_LOG_DEBUG, "%s", lk_summary);
+
+    key = gf_strdup(name);
+    if (!key) {
+        op_ret = -1;
+        goto out;
+    }
+    if (dict_set_dynstr(*dict, key, lk_summary)) {
+        op_ret = -1;
+        *op_errno = ENOMEM;
+        goto out;
+    }
+
+    op_ret = 0;
+
+out:
+    GF_FREE(brickname);
+    GF_FREE(args.opts);
+    GF_FREE(key);
+    if (op_ret) {
+        GF_FREE(lk_summary);
+    }
+
+    return op_ret;
 }
 
-int32_t 
-pl_opendir (call_frame_t *frame, xlator_t *this,
-	     loc_t *loc, fd_t *fd)
+int32_t
+pl_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, const char *name,
+            dict_t *xdata)
 {
-	STACK_WIND (frame,
-		    pl_opendir_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->opendir,
-		    loc, fd);
-	return 0;
+    int32_t op_errno = EINVAL;
+    int32_t op_ret = -1;
+    dict_t *dict = NULL;
+
+    if (!name)
+        goto usual;
+
+    if (strncmp(name, GF_XATTR_CLRLK_CMD, SLEN(GF_XATTR_CLRLK_CMD)))
+        goto usual;
+
+    op_ret = pl_getxattr_clrlk(this, name, loc->inode, &dict, &op_errno);
+
+    STACK_UNWIND_STRICT(getxattr, frame, op_ret, op_errno, dict, xdata);
 
+    if (dict)
+        dict_unref(dict);
+    return 0;
+
+usual:
+    STACK_WIND(frame, pl_getxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->getxattr, loc, name, xdata);
+    return 0;
 }
 
-int
-pl_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-              int32_t op_ret, int32_t op_errno)
+static int
+format_brickname(char *brickname)
 {
-        STACK_UNWIND_STRICT (flush, frame, op_ret, op_errno);
+    int ret = -1;
+    char *hostname = NULL;
+    char *volume = NULL;
+    char *saveptr = NULL;
 
-        return 0;
+    if (!brickname)
+        goto out;
+
+    strtok_r(brickname, ":", &saveptr);
+    hostname = gf_strdup(strtok_r(NULL, ":", &saveptr));
+    if (hostname == NULL)
+        goto out;
+    volume = gf_strdup(strtok_r(NULL, ".", &saveptr));
+    if (volume == NULL)
+        goto out;
+
+    sprintf(brickname, "%s:%s", hostname, volume);
+
+    ret = 0;
+out:
+    GF_FREE(hostname);
+    GF_FREE(volume);
+    return ret;
 }
 
+static int
+fetch_pathinfo(xlator_t *this, inode_t *inode, int32_t *op_errno,
+               char **brickname)
+{
+    int ret = -1;
+    loc_t loc = {
+        0,
+    };
+    dict_t *dict = NULL;
+
+    if (!brickname)
+        goto out;
+
+    if (!op_errno)
+        goto out;
+
+    gf_uuid_copy(loc.gfid, inode->gfid);
+    loc.inode = inode_ref(inode);
+
+    ret = syncop_getxattr(FIRST_CHILD(this), &loc, &dict, GF_XATTR_PATHINFO_KEY,
+                          NULL, NULL);
+    if (ret < 0) {
+        *op_errno = -ret;
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_get_str_sizen(dict, GF_XATTR_PATHINFO_KEY, brickname);
+    if (ret)
+        goto out;
+
+    *brickname = gf_strdup(*brickname);
+    if (*brickname == NULL) {
+        ret = -1;
+        goto out;
+    }
+
+    ret = 0;
+out:
+    if (dict != NULL) {
+        dict_unref(dict);
+    }
+    loc_wipe(&loc);
+
+    return ret;
+}
 
 int
-pl_flush (call_frame_t *frame, xlator_t *this,
-          fd_t *fd)
+pl_lockinfo_get_brickname(xlator_t *this, inode_t *inode, int32_t *op_errno)
 {
-        pl_inode_t            *pl_inode = NULL;
-        uint64_t              owner     = -1;
-
-        owner = frame->root->lk_owner;
+    posix_locks_private_t *priv = this->private;
+    char *brickname = NULL;
+    char *end = NULL;
+    char *tmp = NULL;
+
+    int ret = fetch_pathinfo(this, inode, op_errno, &brickname);
+    if (ret)
+        goto out;
+
+    end = strrchr(brickname, ':');
+    if (!end) {
+        GF_FREE(brickname);
+        ret = -1;
+        goto out;
+    }
+
+    tmp = brickname;
+    brickname = gf_strndup(brickname, (end - brickname));
+    if (brickname == NULL) {
+        ret = -1;
+        goto out;
+    }
+
+    priv->brickname = brickname;
+    ret = 0;
+out:
+    GF_FREE(tmp);
+    return ret;
+}
 
-        pl_inode = pl_inode_get (this, fd->inode);
+char *
+pl_lockinfo_key(xlator_t *this, inode_t *inode, int32_t *op_errno)
+{
+    posix_locks_private_t *priv = this->private;
+    char *key = NULL;
+    int ret = 0;
 
-        if (!pl_inode) {
-                gf_log (this->name, GF_LOG_DEBUG, "Could not get inode.");
-                STACK_UNWIND_STRICT (flush, frame, -1, EBADFD);
-                return 0;
+    if (priv->brickname == NULL) {
+        ret = pl_lockinfo_get_brickname(this, inode, op_errno);
+        if (ret < 0) {
+            gf_log(this->name, GF_LOG_WARNING, "cannot get brickname");
+            goto out;
         }
+    }
 
-        pl_trace_flush (this, frame, fd);
+    key = priv->brickname;
+out:
+    return key;
+}
 
-        if (owner == 0) {
-                /* Handle special case when protocol/server sets lk-owner to zero.
-                 * This usually happens due to a client disconnection. Hence, free
-                 * all locks opened with this fd.
-                 */
-                gf_log (this->name, GF_LOG_TRACE,
-			"Releasing all locks with fd %p", fd);
-                delete_locks_of_fd (this, pl_inode, fd);
-                goto wind;
+int32_t
+pl_fgetxattr_handle_lockinfo(xlator_t *this, fd_t *fd, dict_t *dict,
+                             int32_t *op_errno)
+{
+    char *key = NULL, *buf = NULL;
+    int32_t op_ret = 0;
+    unsigned long fdnum = 0;
+    int32_t len = 0;
+    dict_t *tmp = NULL;
+
+    pl_inode_t *pl_inode = pl_inode_get(this, fd->inode, NULL);
+
+    if (!pl_inode) {
+        gf_log(this->name, GF_LOG_DEBUG, "Could not get inode.");
+        *op_errno = EBADFD;
+        op_ret = -1;
+        goto out;
+    }
+
+    if (!pl_locks_by_fd(pl_inode, fd)) {
+        op_ret = 0;
+        goto out;
+    }
+
+    fdnum = fd_to_fdnum(fd);
+
+    key = pl_lockinfo_key(this, fd->inode, op_errno);
+    if (key == NULL) {
+        op_ret = -1;
+        goto out;
+    }
+
+    tmp = dict_new();
+    if (tmp == NULL) {
+        op_ret = -1;
+        *op_errno = ENOMEM;
+        goto out;
+    }
+
+    op_ret = dict_set_uint64(tmp, key, fdnum);
+    if (op_ret < 0) {
+        *op_errno = -op_ret;
+        op_ret = -1;
+        gf_log(this->name, GF_LOG_WARNING,
+               "setting lockinfo value "
+               "(%lu) for fd (ptr:%p inode-gfid:%s) failed (%s)",
+               fdnum, fd, uuid_utoa(fd->inode->gfid), strerror(*op_errno));
+        goto out;
+    }
+
+    op_ret = dict_allocate_and_serialize(tmp, (char **)&buf,
+                                         (unsigned int *)&len);
+    if (op_ret != 0) {
+        *op_errno = -op_ret;
+        op_ret = -1;
+        gf_log(this->name, GF_LOG_WARNING,
+               "dict_serialized_length failed (%s) while handling "
+               "lockinfo for fd (ptr:%p inode-gfid:%s)",
+               strerror(*op_errno), fd, uuid_utoa(fd->inode->gfid));
+        goto out;
+    }
+
+    op_ret = dict_set_dynptr(dict, GF_XATTR_LOCKINFO_KEY, buf, len);
+    if (op_ret < 0) {
+        *op_errno = -op_ret;
+        op_ret = -1;
+        gf_log(this->name, GF_LOG_WARNING,
+               "setting lockinfo value "
+               "(%lu) for fd (ptr:%p inode-gfid:%s) failed (%s)",
+               fdnum, fd, uuid_utoa(fd->inode->gfid), strerror(*op_errno));
+        goto out;
+    }
+
+    buf = NULL;
+out:
+    if (tmp != NULL) {
+        dict_unref(tmp);
+    }
 
+    if (buf != NULL) {
+        GF_FREE(buf);
+    }
+
+    return op_ret;
+}
+
+int32_t
+pl_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+             dict_t *xdata)
+{
+    int32_t op_ret = 0, op_errno = 0;
+    dict_t *dict = NULL;
+
+    if (!name) {
+        goto usual;
+    }
+
+    if (strcmp(name, GF_XATTR_LOCKINFO_KEY) == 0) {
+        dict = dict_new();
+        if (dict == NULL) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto unwind;
         }
-        pthread_mutex_lock (&pl_inode->mutex);
-        {
-                __delete_locks_of_owner (pl_inode, frame->root->trans,
-                                         owner);
+
+        op_ret = pl_fgetxattr_handle_lockinfo(this, fd, dict, &op_errno);
+        if (op_ret < 0) {
+            gf_log(this->name, GF_LOG_WARNING,
+                   "getting lockinfo on fd (ptr:%p inode-gfid:%s) "
+                   "failed (%s)",
+                   fd, uuid_utoa(fd->inode->gfid), strerror(op_errno));
         }
-        pthread_mutex_unlock (&pl_inode->mutex);
 
-        grant_blocked_locks (this, pl_inode);
+        goto unwind;
+    } else if (strncmp(name, GF_XATTR_CLRLK_CMD, SLEN(GF_XATTR_CLRLK_CMD)) ==
+               0) {
+        op_ret = pl_getxattr_clrlk(this, name, fd->inode, &dict, &op_errno);
 
-        do_blocked_rw (pl_inode);
+        goto unwind;
+    } else {
+        goto usual;
+    }
 
-wind:
-        STACK_WIND (frame, pl_flush_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->flush, fd);
-        return 0;
+unwind:
+    STACK_UNWIND_STRICT(fgetxattr, frame, op_ret, op_errno, dict, NULL);
+    if (dict != NULL) {
+        dict_unref(dict);
+    }
+
+    return 0;
+
+usual:
+    STACK_WIND(frame, default_fgetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fgetxattr, fd, name, xdata);
+    return 0;
 }
 
+int32_t
+pl_migrate_locks(call_frame_t *frame, fd_t *newfd, uint64_t oldfd_num,
+                 int32_t *op_errno)
+{
+    posix_lock_t *l = NULL;
+    int32_t op_ret = 0;
+    uint64_t newfd_num = fd_to_fdnum(newfd);
+
+    pl_inode_t *pl_inode = pl_inode_get(frame->this, newfd->inode, NULL);
+    if (pl_inode == NULL) {
+        op_ret = -1;
+        *op_errno = EBADFD;
+        goto out;
+    }
+
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        list_for_each_entry(l, &pl_inode->ext_list, list)
+        {
+            if (l->fd_num == oldfd_num) {
+                l->fd_num = newfd_num;
+                l->client = frame->root->client;
+            }
+        }
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+
+    op_ret = 0;
+out:
+    return op_ret;
+}
 
-int
-pl_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-             int32_t op_ret, int32_t op_errno, fd_t *fd)
+int32_t
+pl_fsetxattr_handle_lockinfo(call_frame_t *frame, fd_t *fd, char *lockinfo_buf,
+                             int len, int32_t *op_errno)
 {
-        pl_fdctx_t *fdctx = NULL;
-        int ret = -1;
+    int32_t op_ret = -1;
+    uint64_t oldfd_num = 0;
+    char *key = NULL;
+
+    dict_t *lockinfo = dict_new();
+    if (lockinfo == NULL) {
+        op_ret = -1;
+        *op_errno = ENOMEM;
+        goto out;
+    }
+
+    op_ret = dict_unserialize(lockinfo_buf, len, &lockinfo);
+    if (op_ret < 0) {
+        *op_errno = -op_ret;
+        op_ret = -1;
+        goto out;
+    }
+
+    key = pl_lockinfo_key(frame->this, fd->inode, op_errno);
+    if (key == NULL) {
+        op_ret = -1;
+        goto out;
+    }
+
+    op_ret = dict_get_uint64(lockinfo, key, &oldfd_num);
+
+    if (oldfd_num == 0) {
+        op_ret = 0;
+        goto out;
+    }
+
+    op_ret = pl_migrate_locks(frame, fd, oldfd_num, op_errno);
+    if (op_ret < 0) {
+        gf_log(frame->this->name, GF_LOG_WARNING,
+               "migration of locks from oldfd (ptr:%p) to newfd "
+               "(ptr:%p) (inode-gfid:%s)",
+               (void *)(uintptr_t)oldfd_num, fd, uuid_utoa(fd->inode->gfid));
+        goto out;
+    }
 
-        if (op_ret < 0)
-                goto unwind;
+out:
+    dict_unref(lockinfo);
 
-        fdctx = pl_get_fdctx (this, fd);
-        if (!fdctx) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "fdctx not present");
-                fdctx = pl_new_fdctx ();
-                if (!fdctx) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Out of memory");
-                        op_errno = ENOMEM;
-                        op_ret   = -1;
-                        goto unwind;
-                }
+    return op_ret;
+}
 
-                ret = pl_set_fdctx (this, fd, fdctx);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "could not set fdctx");
-                        op_ret = -1;
-                        op_errno = ENOMEM;
-                        goto unwind;
-                }
+int32_t
+pl_fsetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    pl_local_t *local = NULL;
+    pl_inode_t *pl_inode = NULL;
+
+    local = frame->local;
+    if (local && local->update_mlock_enforced_flag && op_ret != -1) {
+        pl_inode = pl_inode_get(this, local->inode, NULL);
+        if (!pl_inode) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto unwind;
+        }
+
+        pthread_mutex_lock(&pl_inode->mutex);
+        {
+            pl_inode->mlock_enforced = _gf_true;
+            pl_inode->check_mlock_info = _gf_false;
         }
+        pthread_mutex_unlock(&pl_inode->mutex);
+    }
 
 unwind:
-        STACK_UNWIND_STRICT (open, frame, op_ret, op_errno, fd);
+    PL_STACK_UNWIND_FOR_CLIENT(fsetxattr, xdata, frame, op_ret, op_errno,
+                               xdata);
+    return 0;
+}
 
-        return 0;
+int32_t
+pl_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+             int32_t flags, dict_t *xdata)
+{
+    int32_t op_errno = 0;
+    void *lockinfo_buf = NULL;
+    int len = 0;
+    char *name = NULL;
+    posix_locks_private_t *priv = this->private;
+
+    int32_t op_ret = dict_get_ptr_and_len(dict, GF_XATTR_LOCKINFO_KEY,
+                                          &lockinfo_buf, &len);
+    if (lockinfo_buf == NULL) {
+        goto usual;
+    }
+
+    op_ret = pl_fsetxattr_handle_lockinfo(frame, fd, lockinfo_buf, len,
+                                          &op_errno);
+    if (op_ret < 0) {
+        goto unwind;
+    }
+
+usual:
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+
+    PL_CHECK_LOCK_ENFORCE_KEY(frame, dict, name, this, ((loc_t *)NULL), fd,
+                              priv);
+
+    STACK_WIND(frame, pl_fsetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
+    return 0;
+
+unwind:
+    PL_STACK_UNWIND_FOR_CLIENT(fsetxattr, xdata, frame, op_ret, op_errno, NULL);
+
+    return 0;
+}
+
+int32_t
+pl_opendir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
+{
+    pl_fdctx_t *fdctx = NULL;
+
+    if (op_ret < 0)
+        goto unwind;
+
+    fdctx = pl_check_n_create_fdctx(this, fd);
+    if (!fdctx) {
+        op_errno = ENOMEM;
+        op_ret = -1;
+        goto unwind;
+    }
+
+unwind:
+    PL_STACK_UNWIND(opendir, xdata, frame, op_ret, op_errno, fd, xdata);
+
+    return 0;
 }
 
+int32_t
+pl_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+           dict_t *xdata)
+{
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+    STACK_WIND(frame, pl_opendir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->opendir, loc, fd, xdata);
+    return 0;
+}
 
 int
-pl_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
-         fd_t *fd, int32_t wbflags)
+pl_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, dict_t *xdata)
 {
-        /* why isn't O_TRUNC being handled ? */
-        STACK_WIND (frame, pl_open_cbk,
-                    FIRST_CHILD(this), FIRST_CHILD(this)->fops->open,
-                    loc, flags & ~O_TRUNC, fd, wbflags);
+    PL_STACK_UNWIND_FOR_CLIENT(flush, xdata, frame, op_ret, op_errno, xdata);
 
-        return 0;
+    return 0;
 }
 
+int
+pl_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    pl_inode_t *pl_inode = pl_inode_get(this, fd->inode, NULL);
+    if (!pl_inode) {
+        gf_log(this->name, GF_LOG_DEBUG, "Could not get inode.");
+        STACK_UNWIND_STRICT(flush, frame, -1, EBADFD, NULL);
+        return 0;
+    }
+
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        if (pl_inode->migrated) {
+            pthread_mutex_unlock(&pl_inode->mutex);
+            STACK_UNWIND_STRICT(flush, frame, -1, EREMOTE, NULL);
+            return 0;
+        }
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+
+    pl_trace_flush(this, frame, fd);
+
+    if (frame->root->lk_owner.len == 0) {
+        /* Handle special case when protocol/server sets lk-owner to zero.
+         * This usually happens due to a client disconnection. Hence, free
+         * all locks opened with this fd.
+         */
+        gf_log(this->name, GF_LOG_TRACE, "Releasing all locks with fd %p", fd);
+        delete_locks_of_fd(this, pl_inode, fd);
+        goto wind;
+    }
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        __delete_locks_of_owner(pl_inode, frame->root->client,
+                                &frame->root->lk_owner);
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+
+    grant_blocked_locks(this, pl_inode);
+
+    do_blocked_rw(pl_inode);
+
+wind:
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+    STACK_WIND(frame, pl_flush_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->flush, fd, xdata);
+    return 0;
+}
 
 int
-pl_create_cbk (call_frame_t *frame, void *cookie,
-               xlator_t *this, int32_t op_ret, int32_t op_errno,
-               fd_t *fd, inode_t *inode, struct iatt *buf,
-               struct iatt *preparent, struct iatt *postparent)
+pl_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, fd_t *fd, dict_t *xdata)
 {
-        pl_fdctx_t *fdctx = NULL;
-        int ret = -1;
+    pl_fdctx_t *fdctx = NULL;
 
-        if (op_ret < 0)
-                goto unwind;
+    if (op_ret < 0)
+        goto unwind;
 
-        fdctx = pl_get_fdctx (this, fd);
-        if (!fdctx) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "fdctx not present");
-                fdctx = pl_new_fdctx ();
-                if (!fdctx) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "Out of memory");
-                        op_errno = ENOMEM;
-                        op_ret   = -1;
-                        goto unwind;
-                }
+    fdctx = pl_check_n_create_fdctx(this, fd);
+    if (!fdctx) {
+        op_errno = ENOMEM;
+        op_ret = -1;
+        goto unwind;
+    }
+
+unwind:
+    STACK_UNWIND_STRICT(open, frame, op_ret, op_errno, fd, xdata);
+
+    return 0;
+}
 
-                ret = pl_set_fdctx (this, fd, fdctx);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "could not set fdctx");
+int
+pl_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+        fd_t *fd, dict_t *xdata)
+{
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    pl_inode_t *pl_inode = NULL;
+    posix_lock_t *l = NULL;
+    posix_locks_private_t *priv = this->private;
+
+    GF_VALIDATE_OR_GOTO("locks", this, unwind);
+
+    op_ret = 0, op_errno = 0;
+    pl_inode = pl_inode_get(this, fd->inode, NULL);
+    if (!pl_inode) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, ENOMEM, "Could not get inode");
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+
+    /* As per design, under forced and file-based mandatory locking modes
+     * it doesn't matter whether inodes's lock list contain advisory or
+     * mandatory type locks. So we just check whether inode's lock list is
+     * empty or not to make sure that no locks are being held for the file.
+     * Whereas under optimal mandatory locking mode, we strictly fail open
+     * if and only if lock list contain mandatory locks.
+     */
+    if (((priv->mandatory_mode == MLK_FILE_BASED) && pl_inode->mandatory) ||
+        priv->mandatory_mode == MLK_FORCED) {
+        if (fd->flags & O_TRUNC) {
+            pthread_mutex_lock(&pl_inode->mutex);
+            {
+                if (!list_empty(&pl_inode->ext_list)) {
+                    op_ret = -1;
+                    op_errno = EAGAIN;
+                }
+            }
+            pthread_mutex_unlock(&pl_inode->mutex);
+        }
+    } else if (priv->mandatory_mode == MLK_OPTIMAL) {
+        if (fd->flags & O_TRUNC) {
+            pthread_mutex_lock(&pl_inode->mutex);
+            {
+                list_for_each_entry(l, &pl_inode->ext_list, list)
+                {
+                    if ((l->lk_flags & GF_LK_MANDATORY)) {
                         op_ret = -1;
-                        op_errno = ENOMEM;
-                        goto unwind;
+                        op_errno = EAGAIN;
+                        break;
+                    }
                 }
+            }
+            pthread_mutex_unlock(&pl_inode->mutex);
         }
+    }
 
 unwind:
-        STACK_UNWIND_STRICT (create, frame, op_ret, op_errno, fd, inode, buf,
-                             preparent, postparent);
-
-        return 0;
+    if (op_ret == -1)
+        STACK_UNWIND_STRICT(open, frame, op_ret, op_errno, NULL, NULL);
+    else
+        STACK_WIND(frame, pl_open_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->open, loc, flags, fd, xdata);
+    return 0;
 }
 
-
 int
-pl_create (call_frame_t *frame, xlator_t *this,
-           loc_t *loc, int32_t flags, mode_t mode, fd_t *fd, dict_t *params)
+pl_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, fd_t *fd, inode_t *inode, struct iatt *buf,
+              struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
 {
-        STACK_WIND (frame, pl_create_cbk,
-                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->create,
-                    loc, flags, mode, fd, params);
-        return 0;
-}
+    pl_fdctx_t *fdctx = NULL;
+
+    if (op_ret < 0)
+        goto unwind;
 
+    fdctx = pl_check_n_create_fdctx(this, fd);
+    if (!fdctx) {
+        op_errno = ENOMEM;
+        op_ret = -1;
+        goto unwind;
+    }
+
+unwind:
+    PL_STACK_UNWIND(create, xdata, frame, op_ret, op_errno, fd, inode, buf,
+                    preparent, postparent, xdata);
+
+    return 0;
+}
 
 int
-pl_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-              int32_t op_ret, int32_t op_errno,
-              struct iovec *vector, int32_t count, struct iatt *stbuf,
-              struct iobref *iobref)
+pl_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+          mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
 {
-        STACK_UNWIND_STRICT (readv, frame, op_ret, op_errno,
-                             vector, count, stbuf, iobref);
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
 
-        return 0;
+    STACK_WIND(frame, pl_create_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->create, loc, flags, mode, umask, fd,
+               xdata);
+    return 0;
 }
 
 int
-pl_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-               int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-               struct iatt *postbuf)
+pl_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, struct iovec *vector, int32_t count,
+             struct iatt *stbuf, struct iobref *iobref, dict_t *xdata)
 {
-        STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, prebuf, postbuf);
+    pl_track_io_fop_count(frame->local, this, DECREMENT);
 
-        return 0;
+    PL_STACK_UNWIND(readv, xdata, frame, op_ret, op_errno, vector, count, stbuf,
+                    iobref, xdata);
+
+    return 0;
 }
 
+int
+pl_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct iatt *prebuf, struct iatt *postbuf,
+              dict_t *xdata)
+{
+    pl_track_io_fop_count(frame->local, this, DECREMENT);
+
+    PL_STACK_UNWIND(writev, xdata, frame, op_ret, op_errno, prebuf, postbuf,
+                    xdata);
+
+    return 0;
+}
 
 void
-do_blocked_rw (pl_inode_t *pl_inode)
+do_blocked_rw(pl_inode_t *pl_inode)
 {
-        struct list_head  wind_list;
-        pl_rw_req_t      *rw = NULL;
-        pl_rw_req_t      *tmp = NULL;
+    struct list_head wind_list;
+    pl_rw_req_t *rw = NULL;
+    pl_rw_req_t *tmp = NULL;
 
-        INIT_LIST_HEAD (&wind_list);
+    INIT_LIST_HEAD(&wind_list);
 
-        pthread_mutex_lock (&pl_inode->mutex);
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        list_for_each_entry_safe(rw, tmp, &pl_inode->rw_list, list)
         {
-                list_for_each_entry_safe (rw, tmp, &pl_inode->rw_list, list) {
-                        if (__rw_allowable (pl_inode, &rw->region,
-                                            rw->stub->fop)) {
-                                list_del_init (&rw->list);
-                                list_add_tail (&rw->list, &wind_list);
-                        }
+            if (__rw_allowable(pl_inode, &rw->region, rw->stub->fop)) {
+                list_del_init(&rw->list);
+                list_add_tail(&rw->list, &wind_list);
+                if (pl_inode->mlock_enforced &&
+                    pl_inode->track_fop_wind_count) {
+                    pl_inode->fop_wind_count++;
                 }
+            }
         }
-        pthread_mutex_unlock (&pl_inode->mutex);
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
 
-        list_for_each_entry_safe (rw, tmp, &wind_list, list) {
-                list_del_init (&rw->list);
-                call_resume (rw->stub);
-                GF_FREE (rw);
-        }
+    list_for_each_entry_safe(rw, tmp, &wind_list, list)
+    {
+        list_del_init(&rw->list);
+        call_resume(rw->stub);
+        GF_FREE(rw);
+    }
 
-        return;
+    return;
 }
 
+/* when mandatory lock is enforced:
+    If an IO request comes on a region which is out of the boundary of the
+    granted mandatory lock, it will be rejected.
+
+    Note: There is no IO blocking with mandatory lock enforced as it may be
+    a stale data from an old client.
+ */
+gf_boolean_t static within_range(posix_lock_t *existing, posix_lock_t *new)
+{
+    if (existing->fl_start <= new->fl_start && existing->fl_end >= new->fl_end)
+        return _gf_true;
+
+    return _gf_false;
+}
 
 static int
-__rw_allowable (pl_inode_t *pl_inode, posix_lock_t *region,
-                glusterfs_fop_t op)
+__rw_allowable(pl_inode_t *pl_inode, posix_lock_t *region, glusterfs_fop_t op)
 {
-        posix_lock_t *l = NULL;
-        int           ret = 1;
+    posix_lock_t *l = NULL;
+    posix_locks_private_t *priv = THIS->private;
+    int ret = 1;
 
-        list_for_each_entry (l, &pl_inode->ext_list, list) {
-                if (locks_overlap (l, region) && !same_owner (l, region)) {
-                        if ((op == GF_FOP_READ) && (l->fl_type != F_WRLCK))
-                                continue;
-                        ret = 0;
-                        break;
+    if (pl_inode->mlock_enforced) {
+        list_for_each_entry(l, &pl_inode->ext_list, list)
+        {
+            /*
+                with lock enforced (fencing) there should not be any blocking
+                lock coexisting.
+            */
+            if (same_owner(l, region)) {
+                /* Should range check be strict for same owner with fencing? */
+                if (locks_overlap(l, region)) {
+                    if (within_range(l, region)) {
+                        return 1;
+                    } else {
+                        /*
+                        Should we allow read fop if it does not fit it in the
+                        range?
+                        if (op == GF_FOP_READ && l->fl_type != F_WRLCK) {
+                            return 1;
+                        }
+                        */
+                        return 0;
+                    }
                 }
+            } else {
+                if (locks_overlap(l, region)) {
+                    /*
+                    with fencing should a read from a different owner be
+                    allowed if the mandatory lock taken is F_RDLCK?
+                    if (op == GF_FOP_READ && l->fl_type != F_WRLCK) {
+                        return 1;
+                    }
+                    */
+                    return 0;
+                }
+            }
         }
 
-        return ret;
-}
+        /* No lock has been taken by this owner */
+        return 0;
+    }
+
+    list_for_each_entry(l, &pl_inode->ext_list, list)
+    {
+        if (!l->blocked && locks_overlap(l, region) && !same_owner(l, region)) {
+            if ((op == GF_FOP_READ) && (l->fl_type != F_WRLCK))
+                continue;
+            /* Check for mandatory lock under optimal
+             * mandatory-locking mode */
+            if (priv->mandatory_mode == MLK_OPTIMAL &&
+                !(l->lk_flags & GF_LK_MANDATORY))
+                continue;
+            ret = 0;
+            break;
+        }
+    }
 
+    return ret;
+}
 
 int
-pl_readv_cont (call_frame_t *frame, xlator_t *this,
-               fd_t *fd, size_t size, off_t offset)
+pl_readv_cont(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+              off_t offset, uint32_t flags, dict_t *xdata)
 {
-        STACK_WIND (frame, pl_readv_cbk,
-                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->readv,
-                    fd, size, offset);
+    pl_track_io_fop_count(frame->local, this, INCREMENT);
 
-        return 0;
-}
+    STACK_WIND(frame, pl_readv_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readv, fd, size, offset, flags, xdata);
 
+    return 0;
+}
 
 int
-pl_readv (call_frame_t *frame, xlator_t *this,
-          fd_t *fd, size_t size, off_t offset)
-{
-        posix_locks_private_t *priv = NULL;
-        pl_inode_t            *pl_inode = NULL;
-        pl_rw_req_t           *rw = NULL;
-        posix_lock_t           region = {.list = {0, }, };
-        int                    op_ret = 0;
-        int                    op_errno = 0;
-        char                   wind_needed = 1;
-
-
-        priv = this->private;
-        pl_inode = pl_inode_get (this, fd->inode);
-
-        if (priv->mandatory && pl_inode->mandatory) {
-                region.fl_start   = offset;
-                region.fl_end     = offset + size - 1;
-                region.transport  = frame->root->trans;
-                region.fd_num     = fd_to_fdnum(fd);
-                region.client_pid = frame->root->pid;
-                region.owner      = frame->root->lk_owner;
-
-                pthread_mutex_lock (&pl_inode->mutex);
-                {
-                        wind_needed = __rw_allowable (pl_inode, &region,
-                                                      GF_FOP_READ);
-                        if (wind_needed) {
-                                goto unlock;
-                        }
-
-                        if (fd->flags & O_NONBLOCK) {
-                                gf_log (this->name, GF_LOG_TRACE,
-                                        "returning EAGAIN as fd is O_NONBLOCK");
-                                op_errno = EAGAIN;
-                                op_ret = -1;
-                                goto unlock;
-                        }
-
-			rw = GF_CALLOC (1, sizeof (*rw),
-                                        gf_locks_mt_pl_rw_req_t);
-                        if (!rw) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "Out of memory.");
-                                op_errno = ENOMEM;
-                                op_ret = -1;
-                                goto unlock;
-                        }
-
-                        rw->stub = fop_readv_stub (frame, pl_readv_cont,
-                                                   fd, size, offset);
-                        if (!rw->stub) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "Out of memory.");
-                                op_errno = ENOMEM;
-                                op_ret = -1;
-                                GF_FREE (rw);
-                                goto unlock;
-                        }
+pl_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+         off_t offset, uint32_t flags, dict_t *xdata)
+{
+    pl_local_t *local = NULL;
+    pl_inode_t *pl_inode = NULL;
+    pl_rw_req_t *rw = NULL;
+    posix_lock_t region = {
+        .list =
+            {
+                0,
+            },
+    };
+    gf_boolean_t enabled = _gf_false;
+    gf_boolean_t can_block = _gf_true;
+    int op_ret = 0;
+    int op_errno = 0;
+    int allowed = 1;
+
+    GF_VALIDATE_OR_GOTO("locks", this, unwind);
+
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+
+    if (!frame->local) {
+        frame->local = mem_get0(this->local_pool);
+        local = frame->local;
+        local->inode = inode_ref(fd->inode);
+        local->fd = fd_ref(fd);
+    }
+
+    pl_inode = pl_inode_get(this, fd->inode, local);
+    if (!pl_inode) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+
+    if (frame->root->pid < 0)
+        enabled = _gf_false;
+    else
+        enabled = pl_is_mandatory_locking_enabled(pl_inode);
+
+    if (enabled) {
+        region.fl_start = offset;
+        region.fl_end = offset + size - 1;
+        region.client = frame->root->client;
+        region.fd_num = fd_to_fdnum(fd);
+        region.client_pid = frame->root->pid;
+        region.owner = frame->root->lk_owner;
+
+        pthread_mutex_lock(&pl_inode->mutex);
+        {
+            allowed = pl_is_fop_allowed(pl_inode, &region, fd, GF_FOP_READ,
+                                        &can_block);
+            if (allowed == 1) {
+                if (pl_inode->mlock_enforced &&
+                    pl_inode->track_fop_wind_count) {
+                    pl_inode->fop_wind_count++;
+                }
+                goto unlock;
+            } else if (!can_block) {
+                op_errno = EAGAIN;
+                op_ret = -1;
+                goto unlock;
+            }
 
-                        rw->region = region;
+            rw = GF_MALLOC(sizeof(*rw), gf_locks_mt_pl_rw_req_t);
+            if (!rw) {
+                op_errno = ENOMEM;
+                op_ret = -1;
+                goto unlock;
+            }
 
-                        list_add_tail (&rw->list, &pl_inode->rw_list);
-                }
-        unlock:
-                pthread_mutex_unlock (&pl_inode->mutex);
-        }
+            rw->stub = fop_readv_stub(frame, pl_readv_cont, fd, size, offset,
+                                      flags, xdata);
+            if (!rw->stub) {
+                op_errno = ENOMEM;
+                op_ret = -1;
+                GF_FREE(rw);
+                goto unlock;
+            }
 
+            rw->region = region;
 
-        if (wind_needed) {
-                STACK_WIND (frame, pl_readv_cbk,
-                            FIRST_CHILD (this), FIRST_CHILD (this)->fops->readv,
-                            fd, size, offset);
+            list_add_tail(&rw->list, &pl_inode->rw_list);
         }
+    unlock:
+        pthread_mutex_unlock(&pl_inode->mutex);
+    }
+
+    if (allowed == 1) {
+        STACK_WIND(frame, pl_readv_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->readv, fd, size, offset, flags,
+                   xdata);
+    }
+unwind:
+    if (op_ret == -1)
+        PL_STACK_UNWIND(readv, xdata, frame, op_ret, op_errno, NULL, 0, NULL,
+                        NULL, NULL);
 
-        if (op_ret == -1)
-                STACK_UNWIND_STRICT (readv, frame, -1, op_errno,
-                                     NULL, 0, NULL, NULL);
-
-        return 0;
+    return 0;
 }
 
-
 int
-pl_writev_cont (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                struct iovec *vector, int count, off_t offset,
-                struct iobref *iobref)
+pl_writev_cont(call_frame_t *frame, xlator_t *this, fd_t *fd,
+               struct iovec *vector, int count, off_t offset, uint32_t flags,
+               struct iobref *iobref, dict_t *xdata)
 {
-        STACK_WIND (frame, pl_writev_cbk,
-                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->writev,
-                    fd, vector, count, offset, iobref);
+    pl_track_io_fop_count(frame->local, this, INCREMENT);
 
-        return 0;
-}
+    STACK_WIND(frame, pl_writev_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->writev, fd, vector, count, offset,
+               flags, iobref, xdata);
 
+    return 0;
+}
 
 int
-pl_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
-           struct iovec *vector, int32_t count, off_t offset,
-           struct iobref *iobref)
-{
-        posix_locks_private_t *priv = NULL;
-        pl_inode_t            *pl_inode = NULL;
-        pl_rw_req_t           *rw = NULL;
-        posix_lock_t           region = {.list = {0, }, };
-        int                    op_ret = 0;
-        int                    op_errno = 0;
-        char                   wind_needed = 1;
-
-
-        priv = this->private;
-        pl_inode = pl_inode_get (this, fd->inode);
-
-        if (priv->mandatory && pl_inode->mandatory) {
-                region.fl_start   = offset;
-                region.fl_end     = offset + iov_length (vector, count) - 1;
-                region.transport  = frame->root->trans;
-                region.fd_num     = fd_to_fdnum(fd);
-                region.client_pid = frame->root->pid;
-                region.owner      = frame->root->lk_owner;
-
-                pthread_mutex_lock (&pl_inode->mutex);
-                {
-                        wind_needed = __rw_allowable (pl_inode, &region,
-                                                      GF_FOP_WRITE);
-                        if (wind_needed)
-                                goto unlock;
-
-                        if (fd->flags & O_NONBLOCK) {
-                                gf_log (this->name, GF_LOG_TRACE,
-                                        "returning EAGAIN because fd is "
-                                        "O_NONBLOCK");
-                                op_errno = EAGAIN;
-                                op_ret = -1;
-                                goto unlock;
-                        }
-
-			rw = GF_CALLOC (1, sizeof (*rw),
-                                        gf_locks_mt_pl_rw_req_t);
-                        if (!rw) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "Out of memory.");
-                                op_errno = ENOMEM;
-                                op_ret = -1;
-                                goto unlock;
-                        }
-
-                        rw->stub = fop_writev_stub (frame, pl_writev_cont,
-                                                    fd, vector, count, offset,
-                                                    iobref);
-                        if (!rw->stub) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "Out of memory.");
-                                op_errno = ENOMEM;
-                                op_ret = -1;
-                                GF_FREE (rw);
-                                goto unlock;
-                        }
+pl_writev(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
+          int32_t count, off_t offset, uint32_t flags, struct iobref *iobref,
+          dict_t *xdata)
+{
+    pl_local_t *local = NULL;
+    pl_inode_t *pl_inode = NULL;
+    pl_rw_req_t *rw = NULL;
+    posix_lock_t region = {
+        .list =
+            {
+                0,
+            },
+    };
+    gf_boolean_t enabled = _gf_false;
+    gf_boolean_t can_block = _gf_true;
+    int op_ret = 0;
+    int op_errno = 0;
+    int allowed = 1;
+
+    GF_VALIDATE_OR_GOTO("locks", this, unwind);
+
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+
+    if (!frame->local) {
+        frame->local = mem_get0(this->local_pool);
+        local = frame->local;
+        local->inode = inode_ref(fd->inode);
+        local->fd = fd_ref(fd);
+    }
+
+    pl_inode = pl_inode_get(this, fd->inode, local);
+    if (!pl_inode) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+
+    if (frame->root->pid < 0)
+        enabled = _gf_false;
+    else
+        enabled = pl_is_mandatory_locking_enabled(pl_inode);
+
+    if (enabled) {
+        region.fl_start = offset;
+        region.fl_end = offset + iov_length(vector, count) - 1;
+        region.client = frame->root->client;
+        region.fd_num = fd_to_fdnum(fd);
+        region.client_pid = frame->root->pid;
+        region.owner = frame->root->lk_owner;
+
+        pthread_mutex_lock(&pl_inode->mutex);
+        {
+            allowed = pl_is_fop_allowed(pl_inode, &region, fd, GF_FOP_WRITE,
+                                        &can_block);
+            if (allowed == 1) {
+                if (pl_inode->mlock_enforced &&
+                    pl_inode->track_fop_wind_count) {
+                    pl_inode->fop_wind_count++;
+                }
+                goto unlock;
+            } else if (!can_block) {
+                if (pl_inode->mlock_enforced) {
+                    op_errno = EBUSY;
+                } else {
+                    op_errno = EAGAIN;
+                }
 
-                        rw->region = region;
+                op_ret = -1;
+                goto unlock;
+            }
 
-                        list_add_tail (&rw->list, &pl_inode->rw_list);
-                }
-        unlock:
-                pthread_mutex_unlock (&pl_inode->mutex);
-        }
+            rw = GF_MALLOC(sizeof(*rw), gf_locks_mt_pl_rw_req_t);
+            if (!rw) {
+                op_errno = ENOMEM;
+                op_ret = -1;
+                goto unlock;
+            }
 
+            rw->stub = fop_writev_stub(frame, pl_writev_cont, fd, vector, count,
+                                       offset, flags, iobref, xdata);
+            if (!rw->stub) {
+                op_errno = ENOMEM;
+                op_ret = -1;
+                GF_FREE(rw);
+                goto unlock;
+            }
 
-        if (wind_needed)
-                STACK_WIND (frame, pl_writev_cbk,
-                            FIRST_CHILD (this), FIRST_CHILD (this)->fops->writev,
-                            fd, vector, count, offset, iobref);
+            rw->region = region;
 
-        if (op_ret == -1)
-                STACK_UNWIND_STRICT (writev, frame, -1, op_errno, NULL, NULL);
+            list_add_tail(&rw->list, &pl_inode->rw_list);
+        }
+    unlock:
+        pthread_mutex_unlock(&pl_inode->mutex);
+    }
+
+    if (allowed == 1) {
+        STACK_WIND(frame, pl_writev_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->writev, fd, vector, count, offset,
+                   flags, iobref, xdata);
+    }
+unwind:
+    if (op_ret == -1)
+        PL_STACK_UNWIND(writev, xdata, frame, op_ret, op_errno, NULL, NULL,
+                        NULL);
 
-        return 0;
+    return 0;
 }
 
 static int
-__fd_has_locks (pl_inode_t *pl_inode, fd_t *fd)
+__fd_has_locks(pl_inode_t *pl_inode, fd_t *fd)
 {
-        int           found = 0;
-        posix_lock_t *l     = NULL;
+    posix_lock_t *l = NULL;
 
-        list_for_each_entry (l, &pl_inode->ext_list, list) {
-                if ((l->fd_num == fd_to_fdnum(fd))) {
-                        found = 1;
-                        break;
-                }
+    list_for_each_entry(l, &pl_inode->ext_list, list)
+    {
+        if (l->fd_num == fd_to_fdnum(fd)) {
+            return 1;
         }
+    }
 
-        return found;
+    return 0;
 }
 
 static posix_lock_t *
-lock_dup (posix_lock_t *lock)
+lock_dup(posix_lock_t *lock)
 {
-        posix_lock_t *new_lock = NULL;
-
-        new_lock = new_posix_lock (&lock->user_flock, lock->transport,
-                                   lock->client_pid, lock->owner,
-                                   (fd_t *)lock->fd_num);
-        return new_lock;
+    int32_t op_errno = 0;
+    return new_posix_lock(&lock->user_flock, lock->client, lock->client_pid,
+                          &lock->owner, (fd_t *)lock->fd_num, lock->lk_flags,
+                          lock->blocking, &op_errno);
 }
 
 static int
-__dup_locks_to_fdctx (pl_inode_t *pl_inode, fd_t *fd,
-                      pl_fdctx_t *fdctx)
-{
-        posix_lock_t *l        = NULL;
-        posix_lock_t *duplock = NULL;
-        int ret = 0;
-
-        list_for_each_entry (l, &pl_inode->ext_list, list) {
-                if ((l->fd_num == fd_to_fdnum(fd))) {
-                        duplock = lock_dup (l);
-                        if (!duplock) {
-                                gf_log (THIS->name, GF_LOG_DEBUG,
-                                        "Out of memory");
-                                ret = -1;
-                                break;
-                        }
+__dup_locks_to_fdctx(pl_inode_t *pl_inode, fd_t *fd, pl_fdctx_t *fdctx)
+{
+    posix_lock_t *l = NULL;
+    posix_lock_t *duplock = NULL;
+    int ret = 0;
+
+    list_for_each_entry(l, &pl_inode->ext_list, list)
+    {
+        if (l->fd_num == fd_to_fdnum(fd)) {
+            duplock = lock_dup(l);
+            if (!duplock) {
+                ret = -1;
+                break;
+            }
 
-                        list_add_tail (&duplock->list, &fdctx->locks_list);
-                }
+            list_add_tail(&duplock->list, &fdctx->locks_list);
         }
+    }
 
-        return ret;
+    return ret;
 }
 
 static int
-__copy_locks_to_fdctx (pl_inode_t *pl_inode, fd_t *fd,
-                      pl_fdctx_t *fdctx)
+__copy_locks_to_fdctx(pl_inode_t *pl_inode, fd_t *fd, pl_fdctx_t *fdctx)
 {
-        int ret = 0;
-
-        ret = __dup_locks_to_fdctx (pl_inode, fd, fdctx);
-        if (ret)
-                goto out;
-
-out:
-        return ret;
-
+    return __dup_locks_to_fdctx(pl_inode, fd, fdctx);
 }
 
 static void
-pl_mark_eol_lock (posix_lock_t *lock)
+pl_mark_eol_lock(posix_lock_t *lock)
 {
-        lock->user_flock.l_type = GF_LK_EOL;
-        return;
+    lock->user_flock.l_type = GF_LK_EOL;
+    return;
 }
 
 static posix_lock_t *
-__get_next_fdctx_lock (pl_fdctx_t *fdctx)
+__get_next_fdctx_lock(pl_fdctx_t *fdctx)
 {
-        posix_lock_t *lock = NULL;
+    posix_lock_t *lock = NULL;
 
-        GF_ASSERT (fdctx);
+    GF_ASSERT(fdctx);
 
-        if (list_empty (&fdctx->locks_list)) {
-                gf_log (THIS->name, GF_LOG_DEBUG,
-                        "fdctx lock list empty");
-                goto out;
-        }
+    if (list_empty(&fdctx->locks_list)) {
+        gf_log(THIS->name, GF_LOG_DEBUG, "fdctx lock list empty");
+        goto out;
+    }
 
-        lock = list_entry (fdctx->locks_list.next, typeof (*lock),
-                           list);
+    lock = list_entry(fdctx->locks_list.next, typeof(*lock), list);
 
-        GF_ASSERT (lock);
+    GF_ASSERT(lock);
 
-        list_del_init (&lock->list);
+    list_del_init(&lock->list);
 
 out:
-        return lock;
+    return lock;
 }
 
 static int
-__set_next_lock_fd (pl_fdctx_t *fdctx, posix_lock_t *reqlock)
+__set_next_lock_fd(pl_fdctx_t *fdctx, posix_lock_t *reqlock)
 {
-        posix_lock_t *lock  = NULL;
-        int           ret   = 0;
+    posix_lock_t *lock = NULL;
+    int ret = 0;
 
-        GF_ASSERT (fdctx);
+    GF_ASSERT(fdctx);
 
-        lock = __get_next_fdctx_lock (fdctx);
-        if (!lock) {
-                gf_log (THIS->name, GF_LOG_DEBUG,
-                        "marking EOL in reqlock");
-                pl_mark_eol_lock (reqlock);
-                goto out;
-        }
+    lock = __get_next_fdctx_lock(fdctx);
+    if (!lock) {
+        gf_log(THIS->name, GF_LOG_DEBUG, "marking EOL in reqlock");
+        pl_mark_eol_lock(reqlock);
+        goto out;
+    }
 
-        reqlock->user_flock  = lock->user_flock;
-        reqlock->fl_start    = lock->fl_start;
-        reqlock->fl_type     = lock->fl_type;
-        reqlock->fl_end      = lock->fl_end;
-        reqlock->owner       = lock->owner;
+    reqlock->user_flock = lock->user_flock;
+    reqlock->fl_start = lock->fl_start;
+    reqlock->fl_type = lock->fl_type;
+    reqlock->fl_end = lock->fl_end;
+    reqlock->owner = lock->owner;
 
 out:
-        if (lock)
-                __destroy_lock (lock);
+    if (lock)
+        __destroy_lock(lock);
 
-        return ret;
+    return ret;
 }
+
 static int
-pl_getlk_fd (xlator_t *this, pl_inode_t *pl_inode,
-             fd_t *fd, posix_lock_t *reqlock)
+pl_getlk_fd(xlator_t *this, pl_inode_t *pl_inode, fd_t *fd,
+            posix_lock_t *reqlock)
 {
-        uint64_t    tmp   = 0;
-        pl_fdctx_t *fdctx = NULL;
-        int         ret   = 0;
-
-        pthread_mutex_lock (&pl_inode->mutex);
-        {
-                if (!__fd_has_locks (pl_inode, fd)) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "fd=%p has no active locks", fd);
-                        ret = 0;
-                        goto unlock;
-                }
+    uint64_t tmp = 0;
+    pl_fdctx_t *fdctx = NULL;
+    int ret = 0;
+
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        if (!__fd_has_locks(pl_inode, fd)) {
+            pthread_mutex_unlock(&pl_inode->mutex);
+            gf_log(this->name, GF_LOG_DEBUG, "fd=%p has no active locks", fd);
+            ret = 0;
+            goto out;
+        }
 
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "There are active locks on fd");
+        gf_log(this->name, GF_LOG_DEBUG, "There are active locks on fd");
 
-                ret = fd_ctx_get (fd, this, &tmp);
-                fdctx = (pl_fdctx_t *) tmp;
+        ret = fd_ctx_get(fd, this, &tmp);
+        fdctx = (pl_fdctx_t *)(long)tmp;
 
-                if (list_empty (&fdctx->locks_list)) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "no fdctx -> copying all locks on fd");
+        if (list_empty(&fdctx->locks_list)) {
+            gf_log(this->name, GF_LOG_TRACE,
+                   "no fdctx -> copying all locks on fd");
 
-                        ret = __copy_locks_to_fdctx (pl_inode, fd, fdctx);
-                        if (ret) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "Out of memory");
-                                goto unlock;
-                        }
+            ret = __copy_locks_to_fdctx(pl_inode, fd, fdctx);
+            if (ret) {
+                goto unlock;
+            }
 
-                        ret = __set_next_lock_fd (fdctx, reqlock);
+            ret = __set_next_lock_fd(fdctx, reqlock);
 
-                } else {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "fdctx present -> returning the next lock");
-                        ret = __set_next_lock_fd (fdctx, reqlock);
-                        if (ret) {
-                                gf_log (this->name, GF_LOG_DEBUG,
-                                        "could not get next lock of fd");
-                                goto unlock;
-                        }
-                }
+        } else {
+            gf_log(this->name, GF_LOG_TRACE,
+                   "fdctx present -> returning the next lock");
+            ret = __set_next_lock_fd(fdctx, reqlock);
+            if (ret) {
+                pthread_mutex_unlock(&pl_inode->mutex);
+                gf_log(this->name, GF_LOG_DEBUG,
+                       "could not get next lock of fd");
+                goto out;
+            }
         }
+    }
 
 unlock:
-        pthread_mutex_unlock (&pl_inode->mutex);
-        return ret;
-
+    pthread_mutex_unlock(&pl_inode->mutex);
+out:
+    return ret;
 }
 
 int
-pl_lk (call_frame_t *frame, xlator_t *this,
-       fd_t *fd, int32_t cmd, struct gf_flock *flock)
-{
-        void                  *transport = NULL;
-        pid_t                  client_pid = 0;
-        uint64_t               owner      = 0;
-        pl_inode_t            *pl_inode = NULL;
-        int                    op_ret = 0;
-        int                    op_errno = 0;
-        int                    can_block = 0;
-        posix_lock_t          *reqlock = NULL;
-        posix_lock_t          *conf = NULL;
-        int                    ret = 0;
-
-        transport  = frame->root->trans;
-        client_pid = frame->root->pid;
-        owner      = frame->root->lk_owner;
-
-        if ((flock->l_start < 0) || (flock->l_len < 0)) {
-                op_ret = -1;
-                op_errno = EINVAL;
-                goto unwind;
-        }
+pl_metalock_is_active(pl_inode_t *pl_inode)
+{
+    if (list_empty(&pl_inode->metalk_list))
+        return 0;
+    else
+        return 1;
+}
 
-        pl_inode = pl_inode_get (this, fd->inode);
-        if (!pl_inode) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory.");
-                op_ret = -1;
-                op_errno = ENOMEM;
-                goto unwind;
-        }
+void
+__pl_queue_lock(pl_inode_t *pl_inode, posix_lock_t *reqlock)
+{
+    list_add_tail(&reqlock->list, &pl_inode->queued_locks);
+}
 
-        reqlock = new_posix_lock (flock, transport, client_pid,
-                                  owner, fd);
+int
+pl_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+      struct gf_flock *flock, dict_t *xdata)
+{
+    pl_inode_t *pl_inode = NULL;
+    int op_ret = 0;
+    int op_errno = 0;
+    int can_block = 0;
+    posix_lock_t *reqlock = NULL;
+    posix_lock_t *conf = NULL;
+    uint32_t lk_flags = 0;
+    posix_locks_private_t *priv = this->private;
+    pl_local_t *local = NULL;
+    short lock_type = 0;
+
+    int ret = dict_get_uint32(xdata, GF_LOCK_MODE, &lk_flags);
+    if (ret == 0) {
+        if (priv->mandatory_mode == MLK_NONE)
+            gf_log(this->name, GF_LOG_DEBUG,
+                   "Lock flags received "
+                   "in a non-mandatory locking environment, "
+                   "continuing");
+        else
+            gf_log(this->name, GF_LOG_DEBUG,
+                   "Lock flags received, "
+                   "continuing");
+    }
+
+    if ((flock->l_start < 0) || ((flock->l_start + flock->l_len) < 0)) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto unwind;
+    }
+
+    /* As per 'man 3 fcntl', the value of l_len may be
+     * negative. In such cases, lock request should be
+     * considered for the range starting at 'l_start+l_len'
+     * and ending at 'l_start-1'. Update the fields accordingly.
+     */
+    if (flock->l_len < 0) {
+        flock->l_start += flock->l_len;
+        flock->l_len = labs(flock->l_len);
+    }
+
+    local = mem_get0(this->local_pool);
+    if (!local) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    } else {
+        frame->local = local;
+        local->fd = fd_ref(fd);
+    }
 
-        if (!reqlock) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "Out of memory.");
-                op_ret = -1;
-                op_errno = ENOMEM;
-                goto unwind;
-        }
+    pl_inode = pl_inode_get(this, fd->inode, local);
+    if (!pl_inode) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+
+    reqlock = new_posix_lock(flock, frame->root->client, frame->root->pid,
+                             &frame->root->lk_owner, fd, lk_flags, can_block,
+                             &op_errno);
 
-        pl_trace_in (this, frame, fd, NULL, cmd, flock, NULL);
+    if (!reqlock) {
+        op_ret = -1;
+        goto unwind;
+    }
 
-        switch (cmd) {
+    pl_trace_in(this, frame, fd, NULL, cmd, flock, NULL);
 
+    switch (cmd) {
         case F_RESLK_LCKW:
-                can_block = 1;
+            can_block = 1;
 
-                /* fall through */
+            /* fall through */
         case F_RESLK_LCK:
-                memcpy (&reqlock->user_flock, flock, sizeof (struct gf_flock));
-                reqlock->frame = frame;
-                reqlock->this = this;
+            reqlock->frame = frame;
+            reqlock->this = this;
 
-                ret = pl_reserve_setlk (this, pl_inode, reqlock,
-                                        can_block);
-                if (ret < 0) {
-                        if (can_block)
-                                goto out;
+            ret = pl_reserve_setlk(this, pl_inode, reqlock, can_block);
+            if (ret < 0) {
+                if (can_block)
+                    goto out;
 
-                        op_ret = -1;
-                        op_errno = -ret;
-                        __destroy_lock (reqlock);
-                        goto unwind;
-                }
-                /* Finally a getlk and return the call */
-                conf = pl_getlk (pl_inode, reqlock);
-                if (conf)
-                        posix_lock_to_flock (conf, flock);
-                break;
+                op_ret = -1;
+                op_errno = -ret;
+                __destroy_lock(reqlock);
+                goto unwind;
+            }
+            /* Finally a getlk and return the call */
+            conf = pl_getlk(pl_inode, reqlock);
+            if (conf)
+                posix_lock_to_flock(conf, flock);
+            break;
 
         case F_RESLK_UNLCK:
-                reqlock->frame = frame;
-                reqlock->this = this;
-                ret = pl_reserve_unlock (this, pl_inode, reqlock);
-                if (ret < 0) {
-                        op_ret = -1;
-                        op_errno = -ret;
-                }
-                __destroy_lock (reqlock);
-                goto unwind;
+            reqlock->frame = frame;
+            reqlock->this = this;
+            ret = pl_reserve_unlock(this, pl_inode, reqlock);
+            if (ret < 0) {
+                op_ret = -1;
+                op_errno = -ret;
+            }
+            __destroy_lock(reqlock);
+            goto unwind;
 
-                break;
+            break;
 
         case F_GETLK_FD:
-                reqlock->frame = frame;
-                reqlock->this = this;
-                ret = pl_verify_reservelk (this, pl_inode, reqlock, can_block);
-                GF_ASSERT (ret >= 0);
-
-                ret = pl_getlk_fd (this, pl_inode, fd, reqlock);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "getting locks on fd failed");
-                        op_ret = -1;
-                        op_errno = ENOLCK;
-                        goto unwind;
-                }
+            reqlock->frame = frame;
+            reqlock->this = this;
+            ret = pl_verify_reservelk(this, pl_inode, reqlock, can_block);
+            GF_ASSERT(ret >= 0);
+
+            ret = pl_getlk_fd(this, pl_inode, fd, reqlock);
+            if (ret < 0) {
+                gf_log(this->name, GF_LOG_DEBUG, "getting locks on fd failed");
+                op_ret = -1;
+                op_errno = ENOLCK;
+                goto unwind;
+            }
 
-                gf_log (this->name, GF_LOG_TRACE,
-                        "Replying with a lock on fd for healing");
+            gf_log(this->name, GF_LOG_TRACE,
+                   "Replying with a lock on fd for healing");
 
-                posix_lock_to_flock (reqlock, flock);
-                __destroy_lock (reqlock);
+            posix_lock_to_flock(reqlock, flock);
+            __destroy_lock(reqlock);
 
-                break;
+            break;
 
 #if F_GETLK != F_GETLK64
         case F_GETLK64:
 #endif
         case F_GETLK:
-                conf = pl_getlk (pl_inode, reqlock);
-                posix_lock_to_flock (conf, flock);
-                __destroy_lock (reqlock);
+            conf = pl_getlk(pl_inode, reqlock);
+            posix_lock_to_flock(conf, flock);
+            __destroy_lock(reqlock);
 
-                break;
+            break;
 
 #if F_SETLKW != F_SETLKW64
         case F_SETLKW64:
 #endif
         case F_SETLKW:
-                can_block = 1;
-                reqlock->frame  = frame;
-                reqlock->this   = this;
-
-                /* fall through */
+            can_block = 1;
+            reqlock->frame = frame;
+            reqlock->this = this;
+            reqlock->blocking = can_block;
+            /* fall through */
 
 #if F_SETLK != F_SETLK64
         case F_SETLK64:
 #endif
         case F_SETLK:
-                memcpy (&reqlock->user_flock, flock, sizeof (struct gf_flock));
-                ret = pl_verify_reservelk (this, pl_inode, reqlock, can_block);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "Lock blocked due to conflicting reserve lock");
-                        goto out;
+            reqlock->frame = frame;
+            reqlock->this = this;
+            lock_type = flock->l_type;
+
+            pthread_mutex_lock(&pl_inode->mutex);
+            {
+                if (pl_inode->migrated) {
+                    op_errno = EREMOTE;
+                    pthread_mutex_unlock(&pl_inode->mutex);
+                    STACK_UNWIND_STRICT(lk, frame, -1, op_errno, flock, xdata);
+
+                    __destroy_lock(reqlock);
+                    goto out;
                 }
-                ret = pl_setlk (this, pl_inode, reqlock,
-                                can_block);
+            }
+            pthread_mutex_unlock(&pl_inode->mutex);
 
+            ret = pl_verify_reservelk(this, pl_inode, reqlock, can_block);
+            if (ret < 0) {
+                gf_log(this->name, GF_LOG_TRACE,
+                       "Lock blocked due to conflicting reserve lock");
+                goto out;
+            }
+
+            if (reqlock->fl_type != F_UNLCK && pl_inode->mlock_enforced) {
+                ret = pl_lock_preempt(pl_inode, reqlock);
                 if (ret == -1) {
-                        if (can_block) {
-                                pl_trace_block (this, frame, fd, NULL, cmd, flock, NULL);
-                                goto out;
-                        }
-                        gf_log (this->name, GF_LOG_DEBUG, "returning EAGAIN");
-                        op_ret = -1;
-                        op_errno = EAGAIN;
-                        __destroy_lock (reqlock);
+                    gf_log(this->name, GF_LOG_ERROR, "lock preempt failed");
+                    op_ret = -1;
+                    op_errno = EAGAIN;
+                    __destroy_lock(reqlock);
+                    goto out;
                 }
-        }
+
+                pl_trace_block(this, frame, fd, NULL, cmd, flock, NULL);
+                goto unwind;
+            }
+
+            ret = pl_setlk(this, pl_inode, reqlock, can_block);
+            if (ret == -1) {
+                if ((can_block) && (F_UNLCK != lock_type)) {
+                    goto out;
+                }
+                gf_log(this->name, GF_LOG_DEBUG, "returning EAGAIN");
+                op_ret = -1;
+                op_errno = EAGAIN;
+                __destroy_lock(reqlock);
+            } else if (ret == -2) {
+                goto out;
+            } else if ((0 == ret) && (F_UNLCK == flock->l_type)) {
+                /* For NLM's last "unlock on fd" detection */
+                if (pl_locks_by_fd(pl_inode, fd))
+                    flock->l_type = F_RDLCK;
+                else
+                    flock->l_type = F_UNLCK;
+            }
+    }
 
 unwind:
-        pl_trace_out (this, frame, fd, NULL, cmd, flock, op_ret, op_errno, NULL);
-        pl_update_refkeeper (this, fd->inode);
-        STACK_UNWIND_STRICT (lk, frame, op_ret, op_errno, flock);
+    pl_trace_out(this, frame, fd, NULL, cmd, flock, op_ret, op_errno, NULL);
+    pl_update_refkeeper(this, fd->inode);
+
+    PL_STACK_UNWIND(lk, xdata, frame, op_ret, op_errno, flock, xdata);
 out:
-        return 0;
+    return 0;
 }
 
-
 /* TODO: this function just logs, no action required?? */
 int
-pl_forget (xlator_t *this,
-           inode_t *inode)
+pl_forget(xlator_t *this, inode_t *inode)
 {
-        pl_inode_t   *pl_inode = NULL;
+    pl_inode_t *pl_inode = NULL;
 
-        posix_lock_t *ext_tmp = NULL;
-        posix_lock_t *ext_l   = NULL;
-	struct list_head posixlks_released;
+    posix_lock_t *ext_tmp = NULL;
+    posix_lock_t *ext_l = NULL;
+    struct list_head posixlks_released;
 
-        pl_inode_lock_t *ino_tmp = NULL;
-        pl_inode_lock_t *ino_l   = NULL;
-	struct list_head inodelks_released;
+    pl_inode_lock_t *ino_tmp = NULL;
+    pl_inode_lock_t *ino_l = NULL;
+    struct list_head inodelks_released;
 
-        pl_rw_req_t *rw_tmp = NULL;
-        pl_rw_req_t *rw_req = NULL;
+    pl_rw_req_t *rw_tmp = NULL;
+    pl_rw_req_t *rw_req = NULL;
 
-        pl_entry_lock_t *entry_tmp = NULL;
-        pl_entry_lock_t *entry_l   = NULL;
-	struct list_head entrylks_released;
+    pl_entry_lock_t *entry_tmp = NULL;
+    pl_entry_lock_t *entry_l = NULL;
+    struct list_head entrylks_released;
 
-        pl_dom_list_t *dom = NULL;
-        pl_dom_list_t *dom_tmp = NULL;
+    pl_dom_list_t *dom = NULL;
+    pl_dom_list_t *dom_tmp = NULL;
 
-	INIT_LIST_HEAD (&posixlks_released);
-	INIT_LIST_HEAD (&inodelks_released);
-	INIT_LIST_HEAD (&entrylks_released);
+    INIT_LIST_HEAD(&posixlks_released);
+    INIT_LIST_HEAD(&inodelks_released);
+    INIT_LIST_HEAD(&entrylks_released);
 
-        pl_inode = pl_inode_get (this, inode);
-
-	pthread_mutex_lock (&pl_inode->mutex);
-	{
+    pl_inode = pl_inode_get(this, inode, NULL);
+    if (!pl_inode)
+        return 0;
 
-		if (!list_empty (&pl_inode->rw_list)) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"Pending R/W requests found, releasing.");
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        if (!list_empty(&pl_inode->rw_list)) {
+            gf_log(this->name, GF_LOG_WARNING,
+                   "Pending R/W requests found, releasing.");
+
+            list_for_each_entry_safe(rw_req, rw_tmp, &pl_inode->rw_list, list)
+            {
+                list_del(&rw_req->list);
+                call_stub_destroy(rw_req->stub);
+                GF_FREE(rw_req);
+            }
+        }
 
-			list_for_each_entry_safe (rw_req, rw_tmp, &pl_inode->rw_list,
-						  list) {
+        if (!list_empty(&pl_inode->ext_list)) {
+            gf_log(this->name, GF_LOG_WARNING,
+                   "Pending fcntl locks found, releasing.");
 
-				list_del (&rw_req->list);
-				GF_FREE (rw_req);
-			}
-		}
+            list_for_each_entry_safe(ext_l, ext_tmp, &pl_inode->ext_list, list)
+            {
+                __delete_lock(ext_l);
+                if (ext_l->blocked) {
+                    list_add_tail(&ext_l->list, &posixlks_released);
+                    continue;
+                }
+                __destroy_lock(ext_l);
+            }
+        }
 
-		if (!list_empty (&pl_inode->ext_list)) {
-			gf_log (this->name, GF_LOG_WARNING,
-				"Pending fcntl locks found, releasing.");
+        list_for_each_entry_safe(dom, dom_tmp, &pl_inode->dom_list, inode_list)
+        {
+            if (!list_empty(&dom->inodelk_list)) {
+                gf_log(this->name, GF_LOG_WARNING,
+                       "Pending inode locks found, releasing.");
 
-			list_for_each_entry_safe (ext_l, ext_tmp, &pl_inode->ext_list,
-						  list) {
+                list_for_each_entry_safe(ino_l, ino_tmp, &dom->inodelk_list,
+                                         list)
+                {
+                    __delete_inode_lock(ino_l);
+                    __pl_inodelk_unref(ino_l);
+                }
 
-				__delete_lock (pl_inode, ext_l);
-				if (ext_l->blocked) {
-					list_add_tail (&ext_l->list, &posixlks_released);
-					continue;
-				}
-				__destroy_lock (ext_l);
-			}
-		}
+                list_splice_init(&dom->blocked_inodelks, &inodelks_released);
+            }
+            if (!list_empty(&dom->entrylk_list)) {
+                gf_log(this->name, GF_LOG_WARNING,
+                       "Pending entry locks found, releasing.");
 
+                list_for_each_entry_safe(entry_l, entry_tmp, &dom->entrylk_list,
+                                         domain_list)
+                {
+                    list_del_init(&entry_l->domain_list);
 
-		list_for_each_entry_safe (dom, dom_tmp, &pl_inode->dom_list, inode_list) {
+                    GF_FREE((char *)entry_l->basename);
+                    GF_FREE(entry_l->connection_id);
+                    GF_FREE(entry_l);
+                }
 
-			if (!list_empty (&dom->inodelk_list)) {
-				gf_log (this->name, GF_LOG_WARNING,
-					"Pending inode locks found, releasing.");
+                list_splice_init(&dom->blocked_entrylks, &entrylks_released);
+            }
 
-				list_for_each_entry_safe (ino_l, ino_tmp, &dom->inodelk_list, list) {
-					__delete_inode_lock (ino_l);
-					__destroy_inode_lock (ino_l);
-				}
+            list_del(&dom->inode_list);
+            gf_log("posix-locks", GF_LOG_TRACE, " Cleaning up domain: %s",
+                   dom->domain);
+            GF_FREE((char *)(dom->domain));
+            GF_FREE(dom);
+        }
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
 
-				list_splice_init (&dom->blocked_inodelks, &inodelks_released);
-				
+    if (!list_empty(&posixlks_released)) {
+        list_for_each_entry_safe(ext_l, ext_tmp, &posixlks_released, list)
+        {
+            STACK_UNWIND_STRICT(lk, ext_l->frame, -1, 0, &ext_l->user_flock,
+                                NULL);
+            __destroy_lock(ext_l);
+        }
+    }
 
-			}
-			if (!list_empty (&dom->entrylk_list)) {
-				gf_log (this->name, GF_LOG_WARNING,
-					"Pending entry locks found, releasing.");
+    if (!list_empty(&inodelks_released)) {
+        list_for_each_entry_safe(ino_l, ino_tmp, &inodelks_released,
+                                 blocked_locks)
+        {
+            STACK_UNWIND_STRICT(inodelk, ino_l->frame, -1, 0, NULL);
+            __pl_inodelk_unref(ino_l);
+        }
+    }
 
-				list_for_each_entry_safe (entry_l, entry_tmp, &dom->entrylk_list, domain_list) {
-					list_del_init (&entry_l->domain_list);
+    if (!list_empty(&entrylks_released)) {
+        list_for_each_entry_safe(entry_l, entry_tmp, &entrylks_released,
+                                 blocked_locks)
+        {
+            STACK_UNWIND_STRICT(entrylk, entry_l->frame, -1, 0, NULL);
+            GF_FREE((char *)entry_l->basename);
+            GF_FREE(entry_l->connection_id);
+            GF_FREE(entry_l);
+        }
+    }
 
-					if (entry_l->basename)
-						GF_FREE ((char *)entry_l->basename);
-					GF_FREE (entry_l);
-				}
+    pthread_mutex_destroy(&pl_inode->mutex);
 
-				list_splice_init (&dom->blocked_entrylks, &entrylks_released);
-			}
+    GF_FREE(pl_inode);
 
-			list_del (&dom->inode_list);
-			gf_log ("posix-locks", GF_LOG_TRACE,
-				" Cleaning up domain: %s", dom->domain);
-			GF_FREE ((char *)(dom->domain));
-			GF_FREE (dom);
-		}
+    return 0;
+}
 
-	}
-	pthread_mutex_unlock (&pl_inode->mutex);
+int
+pl_release(xlator_t *this, fd_t *fd)
+{
+    pl_inode_t *pl_inode = NULL;
+    uint64_t tmp_pl_inode = 0;
+    int ret = -1;
+    uint64_t tmp = 0;
+    pl_fdctx_t *fdctx = NULL;
 
-	list_for_each_entry_safe (ext_l, ext_tmp, &posixlks_released, list) {
+    if (fd == NULL) {
+        goto out;
+    }
 
-		STACK_UNWIND_STRICT (lk, ext_l->frame, -1, 0, &ext_l->user_flock);
-		__destroy_lock (ext_l);
-	}
+    ret = inode_ctx_get(fd->inode, this, &tmp_pl_inode);
+    if (ret != 0)
+        goto clean;
 
-	list_for_each_entry_safe (ino_l, ino_tmp, &inodelks_released, blocked_locks) {
+    pl_inode = (pl_inode_t *)(long)tmp_pl_inode;
 
-		STACK_UNWIND_STRICT (inodelk, ino_l->frame, -1, 0);
-		__destroy_inode_lock (ino_l);
-	}
+    pl_trace_release(this, fd);
 
-	list_for_each_entry_safe (entry_l, entry_tmp, &entrylks_released, blocked_locks) {
+    gf_log(this->name, GF_LOG_TRACE, "Releasing all locks with fd %p", fd);
 
-		STACK_UNWIND_STRICT (entrylk, entry_l->frame, -1, 0);
-		if (entry_l->basename)
-			GF_FREE ((char *)entry_l->basename);
-		GF_FREE (entry_l);
+    delete_locks_of_fd(this, pl_inode, fd);
+    pl_update_refkeeper(this, fd->inode);
 
-	}
+clean:
+    ret = fd_ctx_del(fd, this, &tmp);
+    if (ret) {
+        gf_log(this->name, GF_LOG_DEBUG, "Could not get fdctx");
+        goto out;
+    }
 
-        GF_FREE (pl_inode);
+    fdctx = (pl_fdctx_t *)(long)tmp;
 
-        return 0;
+    GF_FREE(fdctx);
+out:
+    return ret;
 }
 
 int
-pl_release (xlator_t *this, fd_t *fd)
+pl_releasedir(xlator_t *this, fd_t *fd)
 {
-        pl_inode_t *pl_inode     = NULL;
-        uint64_t    tmp_pl_inode = 0;
-        int         ret          = -1;
+    int ret = -1;
+    uint64_t tmp = 0;
+    pl_fdctx_t *fdctx = NULL;
 
-	ret = inode_ctx_get (fd->inode, this, &tmp_pl_inode);
-        if (ret != 0)
-                goto out;
+    if (fd == NULL) {
+        goto out;
+    }
 
-        pl_inode = (pl_inode_t *)(long)tmp_pl_inode;
+    ret = fd_ctx_del(fd, this, &tmp);
+    if (ret) {
+        gf_log(this->name, GF_LOG_DEBUG, "Could not get fdctx");
+        goto out;
+    }
 
-        pl_trace_release (this, fd);
+    fdctx = (pl_fdctx_t *)(long)tmp;
 
-        gf_log (this->name, GF_LOG_TRACE,
-                "Releasing all locks with fd %p", fd);
+    GF_FREE(fdctx);
+out:
+    return ret;
+}
 
-        delete_locks_of_fd (this, pl_inode, fd);
+static int32_t
+pl_request_link_count(dict_t **pxdata)
+{
+    dict_t *xdata;
 
-out:
-        return ret;
+    xdata = *pxdata;
+    if (xdata == NULL) {
+        xdata = dict_new();
+        if (xdata == NULL) {
+            return ENOMEM;
+        }
+    } else {
+        dict_ref(xdata);
+    }
+
+    if (dict_set_uint32(xdata, GET_LINK_COUNT, 0) != 0) {
+        dict_unref(xdata);
+        return ENOMEM;
+    }
+
+    *pxdata = xdata;
+
+    return 0;
 }
+
 static int32_t
-__get_posixlk_count (xlator_t *this, pl_inode_t *pl_inode)
+pl_check_link_count(dict_t *xdata)
 {
-        posix_lock_t *lock   = NULL;
-        int32_t       count  = 0;
+    int32_t count;
 
-        list_for_each_entry (lock, &pl_inode->ext_list, list) {
+    /* In case we are unable to read the link count from xdata, we take a
+     * conservative approach and return -2, which will prevent the inode from
+     * being considered deleted. In fact it will cause link tracking for this
+     * inode to be disabled completely to avoid races. */
 
-			gf_log (this->name, GF_LOG_DEBUG,
-                                " XATTR DEBUG"
-				"%s (pid=%d) (lk-owner=%"PRIu64") %"PRId64" - %"PRId64" state: %s",
-				lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
-				lock->client_pid,
-                                lock->owner,
-				lock->user_flock.l_start,
-				lock->user_flock.l_len,
-                                lock->blocked == 1 ? "Blocked" : "Active");
+    if (xdata == NULL) {
+        return -2;
+    }
 
-                count++;
-        }
+    if (dict_get_int32(xdata, GET_LINK_COUNT, &count) != 0) {
+        return -2;
+    }
 
-        return count;
+    return count;
 }
 
 int32_t
-get_posixlk_count (xlator_t *this, inode_t *inode)
+pl_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, inode_t *inode, struct iatt *buf, dict_t *xdata,
+              struct iatt *postparent)
 {
-        pl_inode_t   *pl_inode = NULL;
-        uint64_t      tmp_pl_inode = 0;
-        int           ret      = 0;
-        int32_t       count    = 0;
+    pl_inode_t *pl_inode;
+
+    if (op_ret >= 0) {
+        pl_inode = pl_inode_get(this, inode, NULL);
+        if (pl_inode == NULL) {
+            PL_STACK_UNWIND(lookup, xdata, frame, -1, ENOMEM, NULL, NULL, NULL,
+                            NULL);
+            return 0;
+        }
 
-        ret = inode_ctx_get (inode, this, &tmp_pl_inode);
-        if (ret != 0) {
-                goto out;
+        pthread_mutex_lock(&pl_inode->mutex);
+
+        /* We only update the link count if we previously didn't know it.
+         * Doing it always can lead to races since lookup is not executed
+         * atomically most of the times. */
+        if (pl_inode->links == -2) {
+            pl_inode->links = pl_check_link_count(xdata);
+            if (buf->ia_type == IA_IFDIR) {
+                /* Directories have at least 2 links. To avoid special handling
+                 * for directories, we simply decrement the value here to make
+                 * them equivalent to regular files. */
+                pl_inode->links--;
+            }
         }
 
-        pl_inode = (pl_inode_t *)(long) tmp_pl_inode;
+        pthread_mutex_unlock(&pl_inode->mutex);
+    }
+
+    PL_STACK_UNWIND(lookup, xdata, frame, op_ret, op_errno, inode, buf, xdata,
+                    postparent);
+    return 0;
+}
+
+int32_t
+pl_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    int32_t error;
+
+    error = pl_request_link_count(&xdata);
+    if (error == 0) {
+        PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+        STACK_WIND(frame, pl_lookup_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->lookup, loc, xdata);
+        dict_unref(xdata);
+    } else {
+        STACK_UNWIND_STRICT(lookup, frame, -1, error, NULL, NULL, NULL, NULL);
+    }
+    return 0;
+}
 
-        pthread_mutex_lock (&pl_inode->mutex);
+int32_t
+pl_fstat_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, struct iatt *buf, dict_t *xdata)
+{
+    PL_STACK_UNWIND(fstat, xdata, frame, op_ret, op_errno, buf, xdata);
+    return 0;
+}
+
+int32_t
+pl_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+    STACK_WIND(frame, pl_fstat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fstat, fd, xdata);
+    return 0;
+}
+
+int
+pl_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                int op_errno, gf_dirent_t *entries, dict_t *xdata)
+{
+    pl_local_t *local = NULL;
+    gf_dirent_t *entry = NULL;
+
+    if (op_ret <= 0)
+        goto unwind;
+
+    local = frame->local;
+    if (!local)
+        goto unwind;
+
+    list_for_each_entry(entry, &entries->list, list)
+    {
+        pl_set_xdata_response(this, local, local->fd->inode, entry->inode,
+                              entry->d_name, entry->dict, 0);
+    }
+
+unwind:
+    PL_STACK_UNWIND(readdirp, xdata, frame, op_ret, op_errno, entries, xdata);
+
+    return 0;
+}
+
+int
+pl_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t offset, dict_t *xdata)
+{
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+    STACK_WIND(frame, pl_readdirp_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdirp, fd, size, offset, xdata);
+
+    return 0;
+}
+
+lock_migration_info_t *
+gf_mig_info_for_lock(posix_lock_t *lock)
+{
+    lock_migration_info_t *new = GF_MALLOC(sizeof(lock_migration_info_t),
+                                           gf_common_mt_lock_mig);
+    if (new == NULL) {
+        goto out;
+    }
+
+    INIT_LIST_HEAD(&new->list);
+
+    posix_lock_to_flock(lock, &new->flock);
+
+    new->lk_flags = lock->lk_flags;
+
+    new->client_uid = gf_strdup(lock->client_uid);
+
+out:
+    return new;
+}
+
+int
+pl_fill_active_locks(pl_inode_t *pl_inode, lock_migration_info_t *lmi)
+{
+    posix_lock_t *temp = NULL;
+    lock_migration_info_t *newlock = NULL;
+    int count = 0;
+
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        if (list_empty(&pl_inode->ext_list)) {
+            count = 0;
+            goto unlock;
+        }
+
+        list_for_each_entry(temp, &pl_inode->ext_list, list)
         {
-                count =__get_posixlk_count (this, pl_inode);
+            if (temp->blocked)
+                continue;
+
+            newlock = gf_mig_info_for_lock(temp);
+            if (!newlock) {
+                pthread_mutex_unlock(&pl_inode->mutex);
+                gf_msg(THIS->name, GF_LOG_ERROR, 0, 0, "lock_dup failed");
+                count = -1;
+                goto out;
+            }
+
+            list_add_tail(&newlock->list, &lmi->list);
+            count++;
         }
-        pthread_mutex_unlock (&pl_inode->mutex);
+    }
 
+unlock:
+    pthread_mutex_unlock(&pl_inode->mutex);
 out:
-        return count;
+    return count;
 }
 
-void
-pl_entrylk_xattr_fill (xlator_t *this, inode_t *inode,
-                       dict_t *dict)
+/* This function reads only active locks */
+static int
+pl_getactivelk(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-        int32_t     count = 0;
-        int         ret   = -1;
+    pl_inode_t *pl_inode = NULL;
+    lock_migration_info_t locks;
+    int op_ret = 0;
+    int op_errno = 0;
+    int count = 0;
 
-        count = get_entrylk_count (this, inode);
-        ret = dict_set_int32 (dict, GLUSTERFS_ENTRYLK_COUNT, count);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        " dict_set failed on key %s", GLUSTERFS_ENTRYLK_COUNT);
-        }
+    INIT_LIST_HEAD(&locks.list);
+
+    pl_inode = pl_inode_get(this, loc->inode, NULL);
+    if (!pl_inode) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "pl_inode_get failed");
+
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    count = pl_fill_active_locks(pl_inode, &locks);
+
+    op_ret = count;
+
+out:
+    STACK_UNWIND_STRICT(getactivelk, frame, op_ret, op_errno, &locks, NULL);
+
+    gf_free_mig_locks(&locks);
 
+    return 0;
 }
 
 void
-pl_inodelk_xattr_fill (xlator_t *this, inode_t *inode,
-                       dict_t *dict)
+pl_metalk_unref(pl_meta_lock_t *lock)
 {
-        int32_t     count = 0;
-        int         ret   = -1;
+    lock->ref--;
+    if (!lock->ref) {
+        GF_FREE(lock->client_uid);
+        GF_FREE(lock);
+    }
+}
 
-        count = get_inodelk_count (this, inode);
-        ret = dict_set_int32 (dict, GLUSTERFS_INODELK_COUNT, count);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        " dict_set failed on key %s", GLUSTERFS_INODELK_COUNT);
-        }
+void
+__pl_metalk_ref(pl_meta_lock_t *lock)
+{
+    lock->ref++;
+}
 
+pl_meta_lock_t *
+new_meta_lock(call_frame_t *frame, xlator_t *this)
+{
+    pl_meta_lock_t *lock = GF_CALLOC(1, sizeof(*lock),
+                                     gf_locks_mt_pl_meta_lock_t);
+
+    if (!lock) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, ENOMEM,
+               "mem allocation"
+               " failed for meta lock");
+        goto out;
+    }
+
+    INIT_LIST_HEAD(&lock->list);
+    INIT_LIST_HEAD(&lock->client_list);
+
+    lock->client_uid = gf_strdup(frame->root->client->client_uid);
+    if (!lock->client_uid) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, ENOMEM,
+               "mem allocation"
+               " failed for client_uid");
+        GF_FREE(lock);
+        lock = NULL;
+        goto out;
+    }
+
+    __pl_metalk_ref(lock);
+out:
+    return lock;
 }
 
-void
-pl_posixlk_xattr_fill (xlator_t *this, inode_t *inode,
-                       dict_t *dict)
+int
+pl_insert_metalk(pl_inode_t *pl_inode, pl_ctx_t *ctx, pl_meta_lock_t *lock)
 {
-        int32_t     count = 0;
-        int         ret   = -1;
+    int ret = 0;
 
-        count = get_posixlk_count (this, inode);
-        ret = dict_set_int32 (dict, GLUSTERFS_POSIXLK_COUNT, count);
-        if (ret < 0) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        " dict_set failed on key %s", GLUSTERFS_POSIXLK_COUNT);
+    if (!pl_inode || !ctx || !lock) {
+        gf_msg(THIS->name, GF_LOG_INFO, 0, 0, "NULL parameter");
+        ret = -1;
+        goto out;
+    }
+
+    lock->pl_inode = pl_inode;
+
+    /* refer function pl_inode_setlk for more info for this ref.
+     * This should be unrefed on meta-unlock triggered by rebalance or
+     * in cleanup with client disconnect*/
+    /*TODO: unref this in  cleanup code for disconnect and meta-unlock*/
+    pl_inode->inode = inode_ref(pl_inode->inode);
+
+    /* NOTE:In case of a client-server disconnect we need to cleanup metalk.
+     * Hence, adding the metalk to pl_ctx_t as well. The mutex lock order
+     * should always be on ctx and then on pl_inode*/
+
+    pthread_mutex_lock(&ctx->lock);
+    {
+        pthread_mutex_lock(&pl_inode->mutex);
+        {
+            list_add_tail(&lock->list, &pl_inode->metalk_list);
         }
+        pthread_mutex_unlock(&pl_inode->mutex);
+
+        list_add_tail(&lock->client_list, &ctx->metalk_list);
+    }
+    pthread_mutex_unlock(&ctx->lock);
 
+out:
+    return ret;
 }
 
 int32_t
-pl_lookup_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    inode_t *inode,
-		    struct iatt *buf,
-                    dict_t *dict,
-                    struct iatt *postparent)
-{
-        pl_local_t *local = NULL;
-
-        if (!frame->local) {
-                goto out;
+pl_metalk(call_frame_t *frame, xlator_t *this, inode_t *inode)
+{
+    pl_inode_t *pl_inode = NULL;
+    int ret = 0;
+    pl_meta_lock_t *reqlk = NULL;
+    pl_ctx_t *ctx = NULL;
+
+    pl_inode = pl_inode_get(this, inode, NULL);
+    if (!pl_inode) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, ENOMEM,
+               "pl_inode mem allocation failedd");
+
+        ret = -1;
+        goto out;
+    }
+
+    /* Non rebalance process trying to do metalock */
+    if (frame->root->pid != GF_CLIENT_PID_DEFRAG) {
+        ret = -1;
+        goto out;
+    }
+
+    /* Note: In the current scheme of glusterfs where lock migration is
+     * experimental, (ideally) the rebalance process which is migrating
+     * the file should request for a metalock. Hence, the metalock count
+     * should not be more than one for an inode. In future, if there is a
+     * need for meta-lock from other clients, the following block can be
+     * removed.
+     *
+     * Since pl_metalk is called as part of setxattr operation, any client
+     * process(non-rebalance) residing outside trusted network can exhaust
+     * memory of the server node by issuing setxattr repetitively on the
+     * metalock key. The following code makes sure that more than
+     * one metalock cannot be granted on an inode*/
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        if (pl_metalock_is_active(pl_inode)) {
+            ret = -1;
+        }
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+
+    if (ret == -1) {
+        gf_msg(this->name, GF_LOG_WARNING, EINVAL, 0,
+               "More than one meta-lock cannot be granted on"
+               " the inode");
+        goto out;
+    }
+
+    if (frame->root->client) {
+        ctx = pl_ctx_get(frame->root->client, this);
+        if (!ctx) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, 0, "pl_ctx_get failed");
+
+            ret = -1;
+            goto out;
         }
+    } else {
+        gf_msg(this->name, GF_LOG_INFO, 0, 0,
+               "frame-root-client "
+               "is NULL");
+
+        ret = -1;
+        goto out;
+    }
+
+    reqlk = new_meta_lock(frame, this);
+    if (!reqlk) {
+        ret = -1;
+        goto out;
+    }
+
+    ret = pl_insert_metalk(pl_inode, ctx, reqlk);
+    if (ret < 0) {
+        pl_metalk_unref(reqlk);
+    }
 
-        if (op_ret) {
-                goto out;
+out:
+    return ret;
+}
+
+static void
+__unwind_queued_locks(pl_inode_t *pl_inode, struct list_head *tmp_list)
+{
+    if (list_empty(&pl_inode->queued_locks))
+        return;
+
+    list_splice_init(&pl_inode->queued_locks, tmp_list);
+}
+
+static void
+__unwind_blocked_locks(pl_inode_t *pl_inode, struct list_head *tmp_list)
+{
+    posix_lock_t *lock = NULL;
+    posix_lock_t *tmp = NULL;
+
+    if (list_empty(&pl_inode->ext_list))
+        return;
+
+    list_for_each_entry_safe(lock, tmp, &pl_inode->ext_list, list)
+    {
+        if (!lock->blocking)
+            continue;
+
+        list_del_init(&lock->list);
+        list_add_tail(&lock->list, tmp_list);
+    }
+}
+
+int
+pl_metaunlock(call_frame_t *frame, xlator_t *this, inode_t *inode, dict_t *dict)
+{
+    pl_inode_t *pl_inode = NULL;
+    int ret = 0;
+    pl_meta_lock_t *meta_lock = NULL;
+    pl_meta_lock_t *tmp_metalk = NULL;
+    pl_ctx_t *ctx = NULL;
+    posix_lock_t *posix_lock = NULL;
+    posix_lock_t *tmp_posixlk = NULL;
+    struct list_head tmp_posixlk_list;
+
+    INIT_LIST_HEAD(&tmp_posixlk_list);
+
+    if (frame->root->client) {
+        ctx = pl_ctx_get(frame->root->client, this);
+        if (!ctx) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, 0, "pl_ctx_get failed");
+
+            ret = -1;
+            goto out;
         }
+    } else {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+               "frame-root-client is "
+               "NULL");
+        ret = -1;
+        goto out;
+    }
+
+    pl_inode = pl_inode_get(this, inode, NULL);
+    if (!pl_inode) {
+        ret = -1;
+        goto out;
+    }
+
+    pthread_mutex_lock(&ctx->lock);
+    {
+        pthread_mutex_lock(&pl_inode->mutex);
+        {
+            /* Unwind queued locks regardless of migration status */
+            __unwind_queued_locks(pl_inode, &tmp_posixlk_list);
 
-        local = frame->local;
+            /* Unwind blocked locks only for successful migration */
+            if (dict_get_sizen(dict, "status")) {
+                /* unwind all blocked locks */
+                __unwind_blocked_locks(pl_inode, &tmp_posixlk_list);
+            }
+
+            /* unlock metalk */
+            /* if this list is empty then pl_inode->metalk_list
+             * should be empty too. meta lock should in all cases
+             * be added/removed from both pl_ctx_t and pl_inode */
+
+            if (list_empty(&ctx->metalk_list))
+                goto unlock;
+
+            list_for_each_entry_safe(meta_lock, tmp_metalk, &ctx->metalk_list,
+                                     client_list)
+            {
+                list_del_init(&meta_lock->client_list);
 
-        if (local->entrylk_count_req)
-                pl_entrylk_xattr_fill (this, inode, dict);
-        if (local->inodelk_count_req)
-                pl_inodelk_xattr_fill (this, inode, dict);
-        if (local->posixlk_count_req)
-                pl_posixlk_xattr_fill (this, inode, dict);
+                pl_inode = meta_lock->pl_inode;
 
+                list_del_init(&meta_lock->list);
 
-        frame->local = NULL;
+                pl_metalk_unref(meta_lock);
 
-        if (local != NULL)
-                GF_FREE (local);
+                /* The corresponding ref is taken in
+                 * pl_insert_metalk*/
+                inode_unref(pl_inode->inode);
+            }
+
+            if (dict_get_sizen(dict, "status"))
+                pl_inode->migrated = _gf_true;
+            else
+                pl_inode->migrated = _gf_false;
+        }
+    unlock:
+
+        pthread_mutex_unlock(&pl_inode->mutex);
+    }
+    pthread_mutex_unlock(&ctx->lock);
 
 out:
-	STACK_UNWIND_STRICT (
-                     lookup,
-                     frame,
-		      op_ret,
-		      op_errno,
-		      inode,
-		      buf,
-                      dict,
-                      postparent);
-	return 0;
+    list_for_each_entry_safe(posix_lock, tmp_posixlk, &tmp_posixlk_list, list)
+    {
+        list_del_init(&posix_lock->list);
+
+        STACK_UNWIND_STRICT(lk, posix_lock->frame, -1, EREMOTE,
+                            &posix_lock->user_flock, NULL);
+
+        __destroy_lock(posix_lock);
+    }
+
+    return ret;
 }
 
 int32_t
-pl_lookup (call_frame_t *frame,
-		xlator_t *this,
-		loc_t *loc,
-		dict_t *xattr_req)
+pl_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        pl_local_t *local  = NULL;
-        int         ret    = -1;
-
-        VALIDATE_OR_GOTO (frame, out);
-        VALIDATE_OR_GOTO (this, out);
-        VALIDATE_OR_GOTO (loc, out);
+    pl_local_t *local = NULL;
+    pl_inode_t *pl_inode = NULL;
+    local = frame->local;
+    if (local && local->update_mlock_enforced_flag && op_ret != -1) {
+        pl_inode = pl_inode_get(this, local->inode, NULL);
+        if (!pl_inode) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto unwind;
+        }
 
-        local = GF_CALLOC (1, sizeof (*local), gf_locks_mt_pl_local_t);
-        if (!local) {
-                ret = -1;
-                gf_log (this->name, GF_LOG_ERROR,
-                        " Out of memory");
-                goto out;
+        pthread_mutex_lock(&pl_inode->mutex);
+        {
+            while (pl_inode->fop_wind_count > 0) {
+                gf_msg(this->name, GF_LOG_INFO, 0, 0,
+                       "waiting for existing fops (count %d) to drain for "
+                       "gfid %s",
+                       pl_inode->fop_wind_count, uuid_utoa(pl_inode->gfid));
+                pthread_cond_wait(&pl_inode->check_fop_wind_count,
+                                  &pl_inode->mutex);
+            }
+            pl_inode->mlock_enforced = _gf_true;
+            pl_inode->check_mlock_info = _gf_false;
         }
+        pthread_mutex_unlock(&pl_inode->mutex);
+    }
 
-        if (dict_get (xattr_req, GLUSTERFS_ENTRYLK_COUNT))
-                local->entrylk_count_req = 1;
-        if (dict_get (xattr_req, GLUSTERFS_INODELK_COUNT))
-                local->inodelk_count_req = 1;
-        if (dict_get (xattr_req, GLUSTERFS_POSIXLK_COUNT))
-                local->posixlk_count_req = 1;
+unwind:
+    PL_STACK_UNWIND_FOR_CLIENT(setxattr, xdata, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
-        frame->local = local;
+int32_t
+pl_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+            int flags, dict_t *xdata)
+{
+    int op_ret = 0;
+    int op_errno = EINVAL;
+    dict_t *xdata_rsp = NULL;
+    char *name = NULL;
+    posix_locks_private_t *priv = this->private;
 
-	STACK_WIND (frame,
-		    pl_lookup_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->lookup,
-		    loc,
-		    xattr_req);
-        ret = 0;
-out:
-        if (ret == -1)
-                STACK_UNWIND_STRICT (lookup, frame, -1, 0, NULL, NULL, NULL, NULL);
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+
+    if (dict_get_sizen(dict, GF_META_LOCK_KEY)) {
+        op_ret = pl_metalk(frame, this, loc->inode);
+
+    } else if (dict_get_sizen(dict, GF_META_UNLOCK_KEY)) {
+        op_ret = pl_metaunlock(frame, this, loc->inode, dict);
+    } else {
+        goto usual;
+    }
+
+    PL_STACK_UNWIND_FOR_CLIENT(setxattr, xdata_rsp, frame, op_ret, op_errno,
+                               xdata_rsp);
+    return 0;
+
+usual:
+    PL_CHECK_LOCK_ENFORCE_KEY(frame, dict, name, this, loc, ((fd_t *)NULL),
+                              priv);
+
+    STACK_WIND(frame, pl_setxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setxattr, loc, dict, flags, xdata);
+    return 0;
+
+unwind:
+    PL_STACK_UNWIND_FOR_CLIENT(setxattr, xdata, frame, op_ret, op_errno, xdata);
 
-	return 0;
+    return 0;
 }
 
 void
-pl_dump_lock (char *str, int size, struct gf_flock *flock,
-              uint64_t owner, void *trans)
+pl_dump_lock(char *str, int size, struct gf_flock *flock, gf_lkowner_t *owner,
+             void *trans, char *conn_id, time_t *granted_time,
+             time_t *blkd_time, gf_boolean_t active)
 {
-        char *type_str = NULL;
-
-        switch (flock->l_type) {
+    char *type_str = NULL;
+    char granted[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    char blocked[GF_TIMESTR_SIZE] = {
+        0,
+    };
+
+    if (granted_time)
+        gf_time_fmt(granted, sizeof(granted), *granted_time, gf_timefmt_FT);
+    if (blkd_time)
+        gf_time_fmt(blocked, sizeof(blocked), *blkd_time, gf_timefmt_FT);
+    switch (flock->l_type) {
         case F_RDLCK:
-                type_str = "READ";
-                break;
+            type_str = "READ";
+            break;
         case F_WRLCK:
-                type_str = "WRITE";
-                break;
+            type_str = "WRITE";
+            break;
         case F_UNLCK:
-                type_str = "UNLOCK";
-                break;
+            type_str = "UNLOCK";
+            break;
         default:
-                type_str = "UNKNOWN";
-                break;
+            type_str = "UNKNOWN";
+            break;
+    }
+
+    if (active) {
+        if (blkd_time && *blkd_time == 0) {
+            snprintf(str, size, RANGE_GRNTD_FMT, type_str, flock->l_whence,
+                     (unsigned long long)flock->l_start,
+                     (unsigned long long)flock->l_len,
+                     (unsigned long long)flock->l_pid, lkowner_utoa(owner),
+                     trans, conn_id, granted);
+        } else {
+            snprintf(str, size, RANGE_BLKD_GRNTD_FMT, type_str, flock->l_whence,
+                     (unsigned long long)flock->l_start,
+                     (unsigned long long)flock->l_len,
+                     (unsigned long long)flock->l_pid, lkowner_utoa(owner),
+                     trans, conn_id, blocked, granted);
         }
+    } else {
+        snprintf(str, size, RANGE_BLKD_FMT, type_str, flock->l_whence,
+                 (unsigned long long)flock->l_start,
+                 (unsigned long long)flock->l_len,
+                 (unsigned long long)flock->l_pid, lkowner_utoa(owner), trans,
+                 conn_id, blocked);
+    }
+}
 
-        snprintf (str, size, "type=%s, start=%llu, len=%llu, pid=%llu, lk-owner=%llu, transport=%p",
-                  type_str, (unsigned long long) flock->l_start,
-                  (unsigned long long) flock->l_len,
-                  (unsigned long long) flock->l_pid,
-                  (unsigned long long) owner,
-                  trans);
+void
+__dump_entrylks(pl_inode_t *pl_inode)
+{
+    pl_dom_list_t *dom = NULL;
+    pl_entry_lock_t *lock = NULL;
+    char blocked[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    char granted[GF_TIMESTR_SIZE] = {
+        0,
+    };
+    int count = 0;
+    char key[GF_DUMP_MAX_BUF_LEN] = {
+        0,
+    };
+    char *k = "xlator.feature.locks.lock-dump.domain.entrylk";
+
+    char tmp[4098];
+
+    list_for_each_entry(dom, &pl_inode->dom_list, inode_list)
+    {
+        count = 0;
+
+        gf_proc_dump_build_key(key, "lock-dump.domain", "domain");
+        gf_proc_dump_write(key, "%s", dom->domain);
+
+        list_for_each_entry(lock, &dom->entrylk_list, domain_list)
+        {
+            gf_time_fmt(granted, sizeof(granted), lock->granted_time,
+                        gf_timefmt_FT);
+            gf_proc_dump_build_key(key, k, "entrylk[%d](ACTIVE)", count);
+            if (lock->blkd_time == 0) {
+                snprintf(tmp, sizeof(tmp), ENTRY_GRNTD_FMT,
+                         lock->type == ENTRYLK_RDLCK ? "ENTRYLK_RDLCK"
+                                                     : "ENTRYLK_WRLCK",
+                         lock->basename, (unsigned long long)lock->client_pid,
+                         lkowner_utoa(&lock->owner), lock->client,
+                         lock->connection_id, granted);
+            } else {
+                gf_time_fmt(blocked, sizeof(blocked), lock->blkd_time,
+                            gf_timefmt_FT);
+                snprintf(tmp, sizeof(tmp), ENTRY_BLKD_GRNTD_FMT,
+                         lock->type == ENTRYLK_RDLCK ? "ENTRYLK_RDLCK"
+                                                     : "ENTRYLK_WRLCK",
+                         lock->basename, (unsigned long long)lock->client_pid,
+                         lkowner_utoa(&lock->owner), lock->client,
+                         lock->connection_id, blocked, granted);
+            }
+
+            gf_proc_dump_write(key, "%s", tmp);
+
+            count++;
+        }
+
+        list_for_each_entry(lock, &dom->blocked_entrylks, blocked_locks)
+        {
+            gf_time_fmt(blocked, sizeof(blocked), lock->blkd_time,
+                        gf_timefmt_FT);
+
+            gf_proc_dump_build_key(key, k, "entrylk[%d](BLOCKED)", count);
+            snprintf(
+                tmp, sizeof(tmp), ENTRY_BLKD_FMT,
+                lock->type == ENTRYLK_RDLCK ? "ENTRYLK_RDLCK" : "ENTRYLK_WRLCK",
+                lock->basename, (unsigned long long)lock->client_pid,
+                lkowner_utoa(&lock->owner), lock->client, lock->connection_id,
+                blocked);
 
+            gf_proc_dump_write(key, "%s", tmp);
 
+            count++;
+        }
+    }
 }
 
 void
-__dump_entrylks (pl_inode_t *pl_inode)
+dump_entrylks(pl_inode_t *pl_inode)
 {
-        pl_dom_list_t   *dom  = NULL;
-        pl_entry_lock_t *lock = NULL;
-        int             count = 0;
-        char            key[GF_DUMP_MAX_BUF_LEN];
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        __dump_entrylks(pl_inode);
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+}
 
-        char tmp[256];
+void
+__dump_inodelks(pl_inode_t *pl_inode)
+{
+    pl_dom_list_t *dom = NULL;
+    pl_inode_lock_t *lock = NULL;
+    int count = 0;
+    char key[GF_DUMP_MAX_BUF_LEN];
 
-        list_for_each_entry (dom, &pl_inode->dom_list, inode_list) {
+    char tmp[4098];
 
-                count = 0;
+    list_for_each_entry(dom, &pl_inode->dom_list, inode_list)
+    {
+        count = 0;
 
-                gf_proc_dump_build_key(key,
-                                       "xlator.feature.locks.lock-dump.domain",
-                                       "domain");
-                gf_proc_dump_write(key, "%s", dom->domain);
+        gf_proc_dump_build_key(key, "lock-dump.domain", "domain");
+        gf_proc_dump_write(key, "%s", dom->domain);
 
-                list_for_each_entry (lock, &dom->entrylk_list, domain_list) {
+        list_for_each_entry(lock, &dom->inodelk_list, list)
+        {
+            gf_proc_dump_build_key(key, "inodelk", "inodelk[%d](ACTIVE)",
+                                   count);
 
-                        gf_proc_dump_build_key(key,
-                                               "xlator.feature.locks.lock-dump.domain.entrylk",
-                                               "entrylk[%d](ACTIVE)",count );
-                        snprintf (tmp, 256," %s on %s owner=%llu, transport=%p",
-                                  lock->type == ENTRYLK_RDLCK ? "ENTRYLK_RDLCK" : 
-                                  "ENTRYLK_WRLCK", lock->basename,
-                                  (unsigned long long) lock->owner,
-                                  lock->trans);
+            SET_FLOCK_PID(&lock->user_flock, lock);
+            pl_dump_lock(tmp, sizeof(tmp), &lock->user_flock, &lock->owner,
+                         lock->client, lock->connection_id, &lock->granted_time,
+                         &lock->blkd_time, _gf_true);
+            gf_proc_dump_write(key, "%s", tmp);
 
-                        gf_proc_dump_write(key, tmp);
+            count++;
+        }
 
-                        count++;
-                }
+        list_for_each_entry(lock, &dom->blocked_inodelks, blocked_locks)
+        {
+            gf_proc_dump_build_key(key, "inodelk", "inodelk[%d](BLOCKED)",
+                                   count);
+            SET_FLOCK_PID(&lock->user_flock, lock);
+            pl_dump_lock(tmp, sizeof(tmp), &lock->user_flock, &lock->owner,
+                         lock->client, lock->connection_id, 0, &lock->blkd_time,
+                         _gf_false);
+            gf_proc_dump_write(key, "%s", tmp);
+
+            count++;
+        }
+    }
+}
 
-                list_for_each_entry (lock, &dom->blocked_entrylks, blocked_locks) {
+void
+dump_inodelks(pl_inode_t *pl_inode)
+{
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        __dump_inodelks(pl_inode);
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+}
+
+void
+__dump_posixlks(pl_inode_t *pl_inode)
+{
+    posix_lock_t *lock = NULL;
+    int count = 0;
+    char key[GF_DUMP_MAX_BUF_LEN];
+
+    char tmp[4098];
+
+    list_for_each_entry(lock, &pl_inode->ext_list, list)
+    {
+        SET_FLOCK_PID(&lock->user_flock, lock);
+        gf_proc_dump_build_key(key, "posixlk", "posixlk[%d](%s)", count,
+                               lock->blocked ? "BLOCKED" : "ACTIVE");
+        pl_dump_lock(tmp, sizeof(tmp), &lock->user_flock, &lock->owner,
+                     lock->client, lock->client_uid, &lock->granted_time,
+                     &lock->blkd_time, (lock->blocked) ? _gf_false : _gf_true);
+        gf_proc_dump_write(key, "%s", tmp);
+
+        count++;
+    }
+}
 
-                        gf_proc_dump_build_key(key,
-                                               "xlator.feature.locks.lock-dump.domain.entrylk",
-                                               "entrylk[%d](BLOCKED)",count );
-                        snprintf (tmp, 256," %s on %s state = Blocked",
-                                  lock->type == ENTRYLK_RDLCK ? "ENTRYLK_RDLCK" : 
-                                  "ENTRYLK_WRLCK", lock->basename);
+void
+dump_posixlks(pl_inode_t *pl_inode)
+{
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        __dump_posixlks(pl_inode);
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+}
 
-                        gf_proc_dump_write(key, tmp);
+int32_t
+pl_dump_inode_priv(xlator_t *this, inode_t *inode)
+{
+    int ret = -1;
+    uint64_t tmp_pl_inode = 0;
+    pl_inode_t *pl_inode = NULL;
+    char *pathname = NULL;
+    gf_boolean_t section_added = _gf_false;
+
+    int count = 0;
+
+    if (!inode) {
+        errno = EINVAL;
+        goto out;
+    }
+
+    ret = TRY_LOCK(&inode->lock);
+    if (ret)
+        goto out;
+    {
+        ret = __inode_ctx_get(inode, this, &tmp_pl_inode);
+        if (ret)
+            goto unlock;
+    }
+unlock:
+    UNLOCK(&inode->lock);
+    if (ret)
+        goto out;
+
+    pl_inode = (pl_inode_t *)(long)tmp_pl_inode;
+    if (!pl_inode) {
+        ret = -1;
+        goto out;
+    }
+
+    gf_proc_dump_add_section("xlator.features.locks.%s.inode", this->name);
+    section_added = _gf_true;
+
+    /*We are safe to call __inode_path since we have the
+     * inode->table->lock */
+    __inode_path(inode, NULL, &pathname);
+    if (pathname)
+        gf_proc_dump_write("path", "%s", pathname);
+
+    gf_proc_dump_write("mandatory", "%d", pl_inode->mandatory);
+
+    ret = pthread_mutex_trylock(&pl_inode->mutex);
+    if (ret)
+        goto out;
+    {
+        count = __get_entrylk_count(this, pl_inode);
+        if (count) {
+            gf_proc_dump_write("entrylk-count", "%d", count);
+            __dump_entrylks(pl_inode);
+        }
 
-                        count++;
-                }
+        count = __get_inodelk_count(this, pl_inode, NULL);
+        if (count) {
+            gf_proc_dump_write("inodelk-count", "%d", count);
+            __dump_inodelks(pl_inode);
+        }
 
+        count = __get_posixlk_count(pl_inode);
+        if (count) {
+            gf_proc_dump_write("posixlk-count", "%d", count);
+            __dump_posixlks(pl_inode);
         }
 
+        gf_proc_dump_write("links", "%d", pl_inode->links);
+        gf_proc_dump_write("removes_pending", "%u", pl_inode->remove_running);
+        gf_proc_dump_write("removed", "%u", pl_inode->removed);
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+
+out:
+    GF_FREE(pathname);
+
+    if (ret && inode) {
+        if (!section_added)
+            gf_proc_dump_add_section(
+                "xlator.features.locks.%s."
+                "inode",
+                this->name);
+        gf_proc_dump_write("Unable to print lock state",
+                           "(Lock "
+                           "acquisition failure) %s",
+                           uuid_utoa(inode->gfid));
+    }
+    return ret;
 }
 
-void
-dump_entrylks (pl_inode_t *pl_inode)
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    if (!this)
+        return ret;
+
+    ret = xlator_mem_acct_init(this, gf_locks_mt_end + 1);
+
+    if (ret != 0) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Memory accounting init"
+               "failed");
+        return ret;
+    }
+
+    return ret;
+}
+
+pl_ctx_t *
+pl_ctx_get(client_t *client, xlator_t *xlator)
+{
+    void *tmp = NULL;
+    pl_ctx_t *ctx = NULL;
+    pl_ctx_t *setted_ctx = NULL;
+
+    client_ctx_get(client, xlator, &tmp);
+
+    ctx = tmp;
+
+    if (ctx != NULL)
+        goto out;
+
+    ctx = GF_CALLOC(1, sizeof(pl_ctx_t), gf_locks_mt_posix_lock_t);
+
+    if (ctx == NULL)
+        goto out;
+
+    pthread_mutex_init(&ctx->lock, NULL);
+    INIT_LIST_HEAD(&ctx->inodelk_lockers);
+    INIT_LIST_HEAD(&ctx->entrylk_lockers);
+    INIT_LIST_HEAD(&ctx->metalk_list);
+
+    setted_ctx = client_ctx_set(client, xlator, ctx);
+    if (ctx != setted_ctx) {
+        pthread_mutex_destroy(&ctx->lock);
+        GF_FREE(ctx);
+        ctx = setted_ctx;
+    }
+out:
+    return ctx;
+}
+
+int
+pl_metalk_client_cleanup(xlator_t *this, pl_ctx_t *ctx)
 {
-        pthread_mutex_lock (&pl_inode->mutex);
+    pl_meta_lock_t *meta_lock = NULL;
+    pl_meta_lock_t *tmp_metalk = NULL;
+    pl_inode_t *pl_inode = NULL;
+    posix_lock_t *posix_lock = NULL;
+    posix_lock_t *tmp_posixlk = NULL;
+    struct list_head tmp_posixlk_list;
+
+    INIT_LIST_HEAD(&tmp_posixlk_list);
+
+    pthread_mutex_lock(&ctx->lock);
+    {
+        /* if this list is empty then pl_inode->metalk_list should be
+         * empty too. meta lock should in all cases be added/removed
+         * from both pl_ctx_t and pl_inode */
+        if (list_empty(&ctx->metalk_list))
+            goto unlock;
+
+        list_for_each_entry_safe(meta_lock, tmp_metalk, &ctx->metalk_list,
+                                 client_list)
         {
-                __dump_entrylks (pl_inode);
+            list_del_init(&meta_lock->client_list);
+
+            pl_inode = meta_lock->pl_inode;
+
+            pthread_mutex_lock(&pl_inode->mutex);
+
+            {
+                /* Since the migration status is unknown here
+                 * unwind all queued and blocked locks to check
+                 * migration status and find the correct
+                 * destination */
+                __unwind_queued_locks(pl_inode, &tmp_posixlk_list);
+
+                __unwind_blocked_locks(pl_inode, &tmp_posixlk_list);
+
+                list_del_init(&meta_lock->list);
+
+                pl_metalk_unref(meta_lock);
+            }
+            pthread_mutex_unlock(&pl_inode->mutex);
+
+            /* The corresponding ref is taken in
+             * pl_insert_metalk*/
+            inode_unref(pl_inode->inode);
         }
-        pthread_mutex_unlock (&pl_inode->mutex);
+    }
+
+unlock:
+    pthread_mutex_unlock(&ctx->lock);
+
+    list_for_each_entry_safe(posix_lock, tmp_posixlk, &tmp_posixlk_list, list)
+    {
+        list_del_init(&posix_lock->list);
+
+        STACK_UNWIND_STRICT(lk, posix_lock->frame, -1, EREMOTE,
+                            &posix_lock->user_flock, NULL);
 
+        __destroy_lock(posix_lock);
+    }
+    return 0;
 }
 
-void
-__dump_inodelks (pl_inode_t *pl_inode)
+static int
+pl_client_disconnect_cbk(xlator_t *this, client_t *client)
 {
-        pl_dom_list_t   *dom  = NULL;
-        pl_inode_lock_t *lock = NULL;
-        int             count = 0;
-        char            key[GF_DUMP_MAX_BUF_LEN];
+    pl_ctx_t *pl_ctx = pl_ctx_get(client, this);
+    if (pl_ctx) {
+        pl_inodelk_client_cleanup(this, pl_ctx);
+        pl_entrylk_client_cleanup(this, pl_ctx);
+        pl_metalk_client_cleanup(this, pl_ctx);
+    }
+
+    return 0;
+}
 
-        char tmp[256];
+static int
+pl_client_destroy_cbk(xlator_t *this, client_t *client)
+{
+    void *tmp = NULL;
+    pl_ctx_t *pl_ctx = NULL;
 
-        list_for_each_entry (dom, &pl_inode->dom_list, inode_list) {
+    pl_client_disconnect_cbk(this, client);
 
-                count = 0;
+    client_ctx_del(client, this, &tmp);
 
-                gf_proc_dump_build_key(key,
-                                       "xlator.feature.locks.lock-dump.domain",
-                                       "domain");
-                gf_proc_dump_write(key, "%s", dom->domain);
+    if (tmp == NULL)
+        return 0;
 
-                list_for_each_entry (lock, &dom->inodelk_list, list) {
+    pl_ctx = tmp;
 
-                        gf_proc_dump_build_key(key,
-                                               "xlator.feature.locks.lock-dump.domain.inodelk",
-                                               "inodelk[%d](ACTIVE)",count );
+    GF_ASSERT(list_empty(&pl_ctx->inodelk_lockers));
+    GF_ASSERT(list_empty(&pl_ctx->entrylk_lockers));
 
-                        pl_dump_lock (tmp, 256, &lock->user_flock,
-                                      lock->owner, lock->transport);
-                        gf_proc_dump_write(key, tmp);
+    pthread_mutex_destroy(&pl_ctx->lock);
+    GF_FREE(pl_ctx);
 
-                        count++;
-                }
+    return 0;
+}
 
-                list_for_each_entry (lock, &dom->blocked_inodelks, blocked_locks) {
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    posix_locks_private_t *priv = this->private;
+    int ret = -1;
+    char *tmp_str = NULL;
 
-                        gf_proc_dump_build_key(key,
-                                               "xlator.feature.locks.lock-dump.domain.inodelk",
-                                               "inodelk[%d](BLOCKED)",count );
-                        pl_dump_lock (tmp, 256, &lock->user_flock,
-                                      lock->owner, lock->transport);
-                        gf_proc_dump_write(key, tmp);
+    GF_OPTION_RECONF("trace", priv->trace, options, bool, out);
 
-                        count++;
-                }
+    GF_OPTION_RECONF("monkey-unlocking", priv->monkey_unlocking, options, bool,
+                     out);
 
-        }
+    GF_OPTION_RECONF("revocation-secs", priv->revocation_secs, options, uint32,
+                     out);
 
+    GF_OPTION_RECONF("revocation-clear-all", priv->revocation_clear_all,
+                     options, bool, out);
+
+    GF_OPTION_RECONF("revocation-max-blocked", priv->revocation_max_blocked,
+                     options, uint32, out);
+
+    GF_OPTION_RECONF("notify-contention", priv->notify_contention, options,
+                     bool, out);
+
+    GF_OPTION_RECONF("notify-contention-delay", priv->notify_contention_delay,
+                     options, uint32, out);
+
+    GF_OPTION_RECONF("mandatory-locking", tmp_str, options, str, out);
+
+    GF_OPTION_RECONF("enforce-mandatory-lock", priv->mlock_enforced, options,
+                     bool, out);
+
+    if (!strcmp(tmp_str, "forced"))
+        priv->mandatory_mode = MLK_FORCED;
+    else if (!strcmp(tmp_str, "file"))
+        priv->mandatory_mode = MLK_FILE_BASED;
+    else if (!strcmp(tmp_str, "optimal"))
+        priv->mandatory_mode = MLK_OPTIMAL;
+    else
+        priv->mandatory_mode = MLK_NONE;
+
+    ret = 0;
+
+out:
+    return ret;
 }
 
-void
-dump_inodelks (pl_inode_t *pl_inode)
+int
+init(xlator_t *this)
 {
-        pthread_mutex_lock (&pl_inode->mutex);
-        {
-                __dump_inodelks (pl_inode);
-        }
-        pthread_mutex_unlock (&pl_inode->mutex);
+    posix_locks_private_t *priv = NULL;
+    xlator_list_t *trav = NULL;
+    char *tmp_str = NULL;
+    int ret = -1;
+
+    if (!this->children || this->children->next) {
+        gf_log(this->name, GF_LOG_CRITICAL,
+               "FATAL: posix-locks should have exactly one child");
+        goto out;
+    }
+
+    if (!this->parents) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "Volume is dangling. Please check the volume file.");
+    }
+
+    trav = this->children;
+    while (trav->xlator->children)
+        trav = trav->xlator->children;
 
+    if (strncmp("storage/", trav->xlator->type, 8)) {
+        gf_log(this->name, GF_LOG_CRITICAL,
+               "'locks' translator is not loaded over a storage "
+               "translator");
+        goto out;
+    }
+
+    priv = GF_CALLOC(1, sizeof(*priv), gf_locks_mt_posix_locks_private_t);
+
+    GF_OPTION_INIT("mandatory-locking", tmp_str, str, out);
+    if (!strcmp(tmp_str, "forced"))
+        priv->mandatory_mode = MLK_FORCED;
+    else if (!strcmp(tmp_str, "file"))
+        priv->mandatory_mode = MLK_FILE_BASED;
+    else if (!strcmp(tmp_str, "optimal"))
+        priv->mandatory_mode = MLK_OPTIMAL;
+    else
+        priv->mandatory_mode = MLK_NONE;
+
+    tmp_str = NULL;
+
+    GF_OPTION_INIT("trace", priv->trace, bool, out);
+
+    GF_OPTION_INIT("monkey-unlocking", priv->monkey_unlocking, bool, out);
+
+    GF_OPTION_INIT("revocation-secs", priv->revocation_secs, uint32, out);
+
+    GF_OPTION_INIT("revocation-clear-all", priv->revocation_clear_all, bool,
+                   out);
+
+    GF_OPTION_INIT("revocation-max-blocked", priv->revocation_max_blocked,
+                   uint32, out);
+
+    GF_OPTION_INIT("notify-contention", priv->notify_contention, bool, out);
+
+    GF_OPTION_INIT("notify-contention-delay", priv->notify_contention_delay,
+                   uint32, out);
+
+    GF_OPTION_INIT("enforce-mandatory-lock", priv->mlock_enforced, bool, out);
+
+    this->local_pool = mem_pool_new(pl_local_t, 32);
+    if (!this->local_pool) {
+        ret = -1;
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to create local_t's memory pool");
+        goto out;
+    }
+
+    this->private = priv;
+    ret = 0;
+
+out:
+    if (ret) {
+        GF_FREE(priv);
+    }
+    return ret;
 }
 
 void
-__dump_posixlks (pl_inode_t *pl_inode)
+fini(xlator_t *this)
 {
-        posix_lock_t    *lock = NULL;
-        int             count = 0;
-        char            key[GF_DUMP_MAX_BUF_LEN];
+    posix_locks_private_t *priv = this->private;
+    if (!priv)
+        return;
+    this->private = NULL;
+    if (this->local_pool) {
+        mem_pool_destroy(this->local_pool);
+        this->local_pool = NULL;
+    }
+    GF_FREE(priv->brickname);
+    GF_FREE(priv);
+
+    return;
+}
+
+int
+pl_inodelk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+           int32_t cmd, struct gf_flock *flock, dict_t *xdata);
 
-        char tmp[256];
+int
+pl_finodelk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+            int32_t cmd, struct gf_flock *flock, dict_t *xdata);
 
-      list_for_each_entry (lock, &pl_inode->ext_list, list) {
+int
+pl_entrylk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+           const char *basename, entrylk_cmd cmd, entrylk_type type,
+           dict_t *xdata);
 
-              gf_proc_dump_build_key(key,
-                                     "xlator.feature.locks.lock-dump.domain.posixlk",
-                                     "posixlk[%d](%s)",
-                                     count,
-                                     lock->blocked ? "BLOCKED" : "ACTIVE");
-              pl_dump_lock (tmp, 256, &lock->user_flock,
-                            lock->owner, lock->transport);
-              gf_proc_dump_write(key, tmp);
+int
+pl_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+            const char *basename, entrylk_cmd cmd, entrylk_type type,
+            dict_t *xdata);
 
-              count++;
-        }
+int32_t
+pl_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct iatt *buf, struct iatt *preoldparent,
+              struct iatt *postoldparent, struct iatt *prenewparent,
+              struct iatt *postnewparent, dict_t *xdata)
+{
+    pl_inode_remove_cbk(this, cookie, op_ret < 0 ? op_errno : 0);
+
+    PL_STACK_UNWIND(rename, xdata, frame, op_ret, op_errno, buf, preoldparent,
+                    postoldparent, prenewparent, postnewparent, xdata);
+
+    return 0;
+}
 
+int32_t
+pl_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+          dict_t *xdata)
+{
+    int32_t error;
 
+    error = PL_INODE_REMOVE(rename, frame, this, oldloc, newloc, pl_rename,
+                            pl_rename_cbk, oldloc, newloc, xdata);
+    if (error > 0) {
+        STACK_UNWIND_STRICT(rename, frame, -1, error, NULL, NULL, NULL, NULL,
+                            NULL, NULL);
+    }
 
+    return 0;
 }
 
-void
-dump_posixlks (pl_inode_t *pl_inode)
+posix_lock_t *
+gf_lkmig_info_to_posix_lock(call_frame_t *frame, lock_migration_info_t *lmi)
 {
-        pthread_mutex_lock (&pl_inode->mutex);
+    posix_lock_t *lock = GF_CALLOC(1, sizeof(posix_lock_t),
+                                   gf_locks_mt_posix_lock_t);
+    if (!lock)
+        goto out;
+
+    lock->fl_start = lmi->flock.l_start;
+    lock->fl_type = lmi->flock.l_type;
+
+    if (lmi->flock.l_len == 0)
+        lock->fl_end = LLONG_MAX;
+    else
+        lock->fl_end = lmi->flock.l_start + lmi->flock.l_len - 1;
+
+    lock->client = frame->root->client;
+
+    lock->lk_flags = lmi->lk_flags;
+
+    lock->client_uid = gf_strdup(lmi->client_uid);
+    if (lock->client_uid == NULL) {
+        GF_FREE(lock);
+        lock = NULL;
+        goto out;
+    }
+
+    lock->client_pid = lmi->flock.l_pid;
+    lock->owner = lmi->flock.l_owner;
+
+    INIT_LIST_HEAD(&lock->list);
+
+out:
+    return lock;
+}
+
+/* This function is supposed to write the active locks from the source brick(in
+ * rebalance context) and write here. Hence, will add the locks directly to the
+ * pl_inode->ext_list*/
+int
+pl_write_active_locks(call_frame_t *frame, pl_inode_t *pl_inode,
+                      lock_migration_info_t *locklist)
+{
+    posix_lock_t *newlock = NULL;
+    lock_migration_info_t *temp = NULL;
+    int ret = 0;
+
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        /* Just making sure the activelk list is empty. Should not
+         * happen though*/
+        if (!list_empty(&pl_inode->ext_list)) {
+            pthread_mutex_unlock(&pl_inode->mutex);
+            gf_msg(THIS->name, GF_LOG_ERROR, 0, 0, "invalid locks found");
+
+            ret = -1;
+            goto out;
+        }
+
+        /* This list also should not be empty */
+        if (list_empty(&locklist->list)) {
+            pthread_mutex_unlock(&pl_inode->mutex);
+            gf_msg(THIS->name, GF_LOG_ERROR, 0, 0, "empty lock list");
+
+            ret = -1;
+            goto out;
+        }
+
+        list_for_each_entry(temp, &locklist->list, list)
         {
-                __dump_posixlks (pl_inode);
+            newlock = gf_lkmig_info_to_posix_lock(frame, temp);
+            if (!newlock) {
+                pthread_mutex_unlock(&pl_inode->mutex);
+                gf_msg(THIS->name, GF_LOG_ERROR, 0, 0,
+                       "mem allocation failed for newlock");
+
+                ret = -1;
+                goto out;
+            }
+            list_add_tail(&newlock->list, &pl_inode->ext_list);
         }
-        pthread_mutex_unlock (&pl_inode->mutex);
+    }
+    /*TODO: What if few lock add failed with ENOMEM. Should the already
+     *      added locks be clearted */
+    pthread_mutex_unlock(&pl_inode->mutex);
+out:
+    return ret;
+}
+
+static int
+pl_setactivelk(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               lock_migration_info_t *locklist, dict_t *xdata)
+{
+    int op_ret = 0;
+    int op_errno = 0;
+    int ret = 0;
+
+    pl_inode_t *pl_inode = pl_inode_get(this, loc->inode, NULL);
+    if (!pl_inode) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, 0, "pl_inode_get failed");
+
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto out;
+    }
+    ret = pl_write_active_locks(frame, pl_inode, locklist);
+
+    op_ret = ret;
 
+out:
+    STACK_UNWIND_STRICT(setactivelk, frame, op_ret, op_errno, NULL);
+
+    return 0;
 }
 
 int32_t
-pl_dump_inode_priv (xlator_t *this, inode_t *inode)
+pl_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct iatt *preparent, struct iatt *postparent,
+              dict_t *xdata)
 {
+    pl_inode_remove_cbk(this, cookie, op_ret < 0 ? op_errno : 0);
 
-        int             ret = -1;
-        uint64_t        tmp_pl_inode = 0;
-        pl_inode_t      *pl_inode = NULL;
-        char            key[GF_DUMP_MAX_BUF_LEN];
+    PL_STACK_UNWIND(unlink, xdata, frame, op_ret, op_errno, preparent,
+                    postparent, xdata);
 
-        int count      = 0;
+    return 0;
+}
 
-        if (!inode)
-                return -1;
+int32_t
+pl_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+          dict_t *xdata)
+{
+    int32_t error;
 
-        ret = inode_ctx_get (inode, this, &tmp_pl_inode);
+    error = PL_INODE_REMOVE(unlink, frame, this, loc, NULL, pl_unlink,
+                            pl_unlink_cbk, loc, xflag, xdata);
+    if (error > 0) {
+        STACK_UNWIND_STRICT(unlink, frame, -1, error, NULL, NULL, NULL);
+    }
 
-        if (ret != 0)
-                return ret;
+    return 0;
+}
 
-        pl_inode = (pl_inode_t *)(long)tmp_pl_inode;
+int32_t
+pl_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, inode_t *inode, struct iatt *buf,
+             struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    PL_STACK_UNWIND_FOR_CLIENT(mkdir, xdata, frame, op_ret, op_errno, inode,
+                               buf, preparent, postparent, xdata);
+    return 0;
+}
 
-        if (!pl_inode)
-                return -1;
+int
+pl_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+         mode_t umask, dict_t *xdata)
+{
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+    STACK_WIND(frame, pl_mkdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mkdir, loc, mode, umask, xdata);
+    return 0;
+}
 
-        gf_proc_dump_build_key(key,
-                               "xlator.feature.locks.inode",
-                               "%ld.mandatory",inode->ino);
-        gf_proc_dump_write(key, "%d", pl_inode->mandatory);
+int32_t
+pl_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, struct iatt *buf, dict_t *xdata)
+{
+    PL_STACK_UNWIND_FOR_CLIENT(stat, xdata, frame, op_ret, op_errno, buf,
+                               xdata);
+    return 0;
+}
 
+int
+pl_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+    STACK_WIND(frame, pl_stat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->stat, loc, xdata);
+    return 0;
+}
 
-                count = get_entrylk_count (this, inode);
-                gf_proc_dump_build_key(key,
-                                       "xlator.feature.locks.entrylk-count",
-                                       "%ld.entrylk-count", inode->ino);
-                gf_proc_dump_write(key, "%d", count);
+int32_t
+pl_mknod_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, inode_t *inode, struct iatt *buf,
+             struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    PL_STACK_UNWIND_FOR_CLIENT(mknod, xdata, frame, op_ret, op_errno, inode,
+                               buf, preparent, postparent, xdata);
+    return 0;
+}
 
-                dump_entrylks(pl_inode);
+int
+pl_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+         dev_t rdev, mode_t umask, dict_t *xdata)
+{
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+    STACK_WIND(frame, pl_mknod_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mknod, loc, mode, rdev, umask, xdata);
+    return 0;
+}
 
-                count = get_inodelk_count (this, inode);
-                gf_proc_dump_build_key(key,
-                                       "xlator.feature.locks.inodelk-count",
-                                       "%ld.inodelk-count", inode->ino);
-                gf_proc_dump_write(key, "%d", count);
+int32_t
+pl_rmdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, struct iatt *preparent, struct iatt *postparent,
+             dict_t *xdata)
+{
+    pl_inode_remove_cbk(this, cookie, op_ret < 0 ? op_errno : 0);
 
-                dump_inodelks(pl_inode);
+    PL_STACK_UNWIND_FOR_CLIENT(rmdir, xdata, frame, op_ret, op_errno, preparent,
+                               postparent, xdata);
 
-                count = get_posixlk_count (this, inode);
-                gf_proc_dump_build_key(key,
-                                       "xlator.feature.locks.posixlk-count",
-                                       "%ld.posixlk-count", inode->ino);
-                gf_proc_dump_write(key, "%d", count);
+    return 0;
+}
 
-                dump_posixlks(pl_inode);
+int
+pl_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflags,
+         dict_t *xdata)
+{
+    int32_t error;
 
+    error = PL_INODE_REMOVE(rmdir, frame, this, loc, NULL, pl_rmdir,
+                            pl_rmdir_cbk, loc, xflags, xdata);
+    if (error > 0) {
+        STACK_UNWIND_STRICT(rmdir, frame, -1, error, NULL, NULL, NULL);
+    }
 
-        return 0;
+    return 0;
 }
 
+int32_t
+pl_symlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, inode_t *inode,
+               struct iatt *buf, struct iatt *preparent,
+               struct iatt *postparent, dict_t *xdata)
+{
+    PL_STACK_UNWIND_FOR_CLIENT(symlink, xdata, frame, op_ret, op_errno, inode,
+                               buf, preparent, postparent, xdata);
+    return 0;
+}
 
-
-/*
- * pl_dump_inode - inode dump function for posix locks
- *
- */
 int
-pl_dump_inode (xlator_t *this)
+pl_symlink(call_frame_t *frame, xlator_t *this, const char *linkname,
+           loc_t *loc, mode_t umask, dict_t *xdata)
 {
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+    STACK_WIND(frame, pl_symlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->symlink, linkname, loc, umask, xdata);
+    return 0;
+}
+
+int32_t
+pl_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, inode_t *inode, struct iatt *buf,
+            struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    pl_inode_t *pl_inode = (pl_inode_t *)cookie;
 
-        assert(this);
+    if (op_ret >= 0) {
+        pthread_mutex_lock(&pl_inode->mutex);
 
-        if (this->itable) {
-                inode_table_dump(this->itable,
-                                 "xlator.features.locks.inode_table");
+        /* TODO: can happen pl_inode->links == 0 ? */
+        if (pl_inode->links >= 0) {
+            pl_inode->links++;
         }
 
+        pthread_mutex_unlock(&pl_inode->mutex);
+    }
+
+    PL_STACK_UNWIND_FOR_CLIENT(link, xdata, frame, op_ret, op_errno, inode, buf,
+                               preparent, postparent, xdata);
+    return 0;
+}
+
+int
+pl_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+        dict_t *xdata)
+{
+    pl_inode_t *pl_inode;
+
+    pl_inode = pl_inode_get(this, oldloc->inode, NULL);
+    if (pl_inode == NULL) {
+        STACK_UNWIND_STRICT(link, frame, -1, ENOMEM, NULL, NULL, NULL, NULL,
+                            NULL);
         return 0;
+    }
+
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), oldloc, newloc);
+    STACK_WIND_COOKIE(frame, pl_link_cbk, pl_inode, FIRST_CHILD(this),
+                      FIRST_CHILD(this)->fops->link, oldloc, newloc, xdata);
+    return 0;
 }
 
 int32_t
-mem_acct_init (xlator_t *this)
+pl_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, struct iatt *prebuf, struct iatt *postbuf,
+             dict_t *xdata)
 {
-        int     ret = -1;
+    PL_STACK_UNWIND_FOR_CLIENT(fsync, xdata, frame, op_ret, op_errno, prebuf,
+                               postbuf, xdata);
+    return 0;
+}
 
-        if (!this)
-                return ret;
+int
+pl_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+         dict_t *xdata)
+{
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+    STACK_WIND(frame, pl_fsync_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsync, fd, datasync, xdata);
+    return 0;
+}
 
-        ret = xlator_mem_acct_init (this, gf_locks_mt_end + 1);
-        
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
-                                "failed");
-                return ret;
-        }
+int32_t
+pl_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, gf_dirent_t *entries,
+               dict_t *xdata)
+{
+    PL_STACK_UNWIND_FOR_CLIENT(readdir, xdata, frame, op_ret, op_errno, entries,
+                               xdata);
+    return 0;
+}
 
-        return ret;
+int
+pl_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+           off_t offset, dict_t *xdata)
+{
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+    STACK_WIND(frame, pl_readdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdir, fd, size, offset, xdata);
+    return 0;
+}
+
+int32_t
+pl_fsyncdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    PL_STACK_UNWIND_FOR_CLIENT(fsyncdir, xdata, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+int
+pl_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+            dict_t *xdata)
+{
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+    STACK_WIND(frame, pl_fsyncdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsyncdir, fd, datasync, xdata);
+    return 0;
+}
+
+int32_t
+pl_statfs_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct statvfs *buf, dict_t *xdata)
+{
+    PL_STACK_UNWIND_FOR_CLIENT(statfs, xdata, frame, op_ret, op_errno, buf,
+                               xdata);
+    return 0;
 }
 
 int
-init (xlator_t *this)
+pl_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+    STACK_WIND(frame, pl_statfs_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->statfs, loc, xdata);
+    return 0;
+}
+
+int32_t
+pl_removexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-        posix_locks_private_t *priv = NULL;
-        xlator_list_t         *trav = NULL;
-        data_t                *mandatory = NULL;
-	data_t                *trace = NULL;
+    pl_local_t *local = NULL;
+    pl_inode_t *pl_inode = NULL;
 
-        if (!this->children || this->children->next) {
-                gf_log (this->name, GF_LOG_CRITICAL,
-                        "FATAL: posix-locks should have exactly one child");
-                return -1;
+    local = frame->local;
+    if (local && local->update_mlock_enforced_flag && op_ret != -1) {
+        pl_inode = pl_inode_get(this, local->inode, NULL);
+        if (!pl_inode) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto unwind;
         }
 
-        if (!this->parents) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "Volume is dangling. Please check the volume file.");
+        pthread_mutex_lock(&pl_inode->mutex);
+        {
+            pl_inode->mlock_enforced = _gf_false;
+            pl_inode->check_mlock_info = _gf_false;
+            pl_inode->track_fop_wind_count = _gf_true;
         }
+        pthread_mutex_unlock(&pl_inode->mutex);
+    }
+
+unwind:
+    PL_STACK_UNWIND_FOR_CLIENT(removexattr, xdata, frame, op_ret, op_errno,
+                               xdata);
+    return 0;
+}
+
+int
+pl_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               const char *name, dict_t *xdata)
+{
+    int op_ret = 0;
+    int op_errno = EINVAL;
+    posix_locks_private_t *priv = this->private;
+
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
 
-        trav = this->children;
-        while (trav->xlator->children)
-                trav = trav->xlator->children;
+    PL_CHECK_LOCK_ENFORCE_KEY(frame, ((dict_t *)NULL), name, this, loc,
+                              ((fd_t *)NULL), priv);
 
-        if (strncmp ("storage/", trav->xlator->type, 8)) {
-                gf_log (this->name, GF_LOG_CRITICAL,
-                        "'locks' translator is not loaded over a storage "
-                        "translator");
-                return -1;
+    STACK_WIND(frame, pl_removexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->removexattr, loc, name, xdata);
+    return 0;
+
+unwind:
+    PL_STACK_UNWIND_FOR_CLIENT(removexattr, xdata, frame, op_ret, op_errno,
+                               NULL);
+
+    return 0;
+}
+
+int32_t
+pl_fremovexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    pl_local_t *local = NULL;
+    pl_inode_t *pl_inode = NULL;
+
+    local = frame->local;
+    if (local && local->update_mlock_enforced_flag && op_ret != -1) {
+        pl_inode = pl_inode_get(this, local->inode, NULL);
+        if (!pl_inode) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto unwind;
         }
 
-	priv = GF_CALLOC (1, sizeof (*priv),
-                          gf_locks_mt_posix_locks_private_t);
+        pthread_mutex_lock(&pl_inode->mutex);
+        {
+            pl_inode->mlock_enforced = _gf_false;
+            pl_inode->check_mlock_info = _gf_false;
+        }
+        pthread_mutex_unlock(&pl_inode->mutex);
+    }
 
-        mandatory = dict_get (this->options, "mandatory-locks");
-        if (mandatory)
-                gf_log (this->name, GF_LOG_WARNING,
-                        "mandatory locks not supported in this minor release.");
+unwind:
+    PL_STACK_UNWIND_FOR_CLIENT(fremovexattr, xdata, frame, op_ret, op_errno,
+                               xdata);
+    return 0;
+}
 
-	trace = dict_get (this->options, "trace");
-	if (trace) {
-		if (gf_string2boolean (trace->data,
-				       &priv->trace) == -1) {
-			gf_log (this->name, GF_LOG_ERROR,
-				"'trace' takes on only boolean values.");
-			return -1;
-		}
-	}
+int
+pl_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+                dict_t *xdata)
+{
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    posix_locks_private_t *priv = this->private;
 
-        this->private = priv;
-        return 0;
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+
+    PL_CHECK_LOCK_ENFORCE_KEY(frame, ((dict_t *)NULL), name, this,
+                              ((loc_t *)NULL), fd, priv);
+
+    STACK_WIND(frame, pl_fremovexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fremovexattr, fd, name, xdata);
+    return 0;
+
+unwind:
+    PL_STACK_UNWIND_FOR_CLIENT(fremovexattr, xdata, frame, op_ret, op_errno,
+                               NULL);
+    return 0;
 }
 
+int32_t
+pl_rchecksum_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, uint32_t weak_cksum,
+                 uint8_t *strong_cksum, dict_t *xdata)
+{
+    PL_STACK_UNWIND_FOR_CLIENT(rchecksum, xdata, frame, op_ret, op_errno,
+                               weak_cksum, strong_cksum, xdata);
+    return 0;
+}
 
 int
-fini (xlator_t *this)
+pl_rchecksum(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             int32_t len, dict_t *xdata)
+{
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+    STACK_WIND(frame, pl_rchecksum_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rchecksum, fd, offset, len, xdata);
+    return 0;
+}
+
+int32_t
+pl_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
 {
-        posix_locks_private_t *priv = NULL;
+    PL_STACK_UNWIND_FOR_CLIENT(xattrop, xdata, frame, op_ret, op_errno, dict,
+                               xdata);
+    return 0;
+}
 
-        priv = this->private;
-        GF_FREE (priv);
+int
+pl_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+           gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+    STACK_WIND(frame, pl_xattrop_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->xattrop, loc, optype, xattr, xdata);
+    return 0;
+}
 
-        return 0;
+int32_t
+pl_fxattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+    PL_STACK_UNWIND_FOR_CLIENT(fxattrop, xdata, frame, op_ret, op_errno, dict,
+                               xdata);
+    return 0;
+}
+
+int
+pl_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+            gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+    STACK_WIND(frame, pl_fxattrop_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fxattrop, fd, optype, xattr, xdata);
+    return 0;
+}
+
+int32_t
+pl_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, struct iatt *statpre,
+               struct iatt *statpost, dict_t *xdata)
+{
+    PL_STACK_UNWIND_FOR_CLIENT(setattr, xdata, frame, op_ret, op_errno, statpre,
+                               statpost, xdata);
+    return 0;
+}
+
+int
+pl_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *stbuf,
+           int32_t valid, dict_t *xdata)
+{
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+    STACK_WIND(frame, pl_setattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setattr, loc, stbuf, valid, xdata);
+    return 0;
 }
 
+int32_t
+pl_fsetattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *statpre,
+                struct iatt *statpost, dict_t *xdata)
+{
+    PL_STACK_UNWIND_FOR_CLIENT(fsetattr, xdata, frame, op_ret, op_errno,
+                               statpre, statpost, xdata);
+    return 0;
+}
 
 int
-pl_inodelk (call_frame_t *frame, xlator_t *this,
-            const char *volume, loc_t *loc, int32_t cmd, struct gf_flock *flock);
+pl_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
+            int32_t valid, dict_t *xdata)
+{
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+    STACK_WIND(frame, pl_fsetattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetattr, fd, stbuf, valid, xdata);
+    return 0;
+}
+
+int32_t
+pl_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct iatt *pre,
+                 struct iatt *post, dict_t *xdata)
+{
+    PL_STACK_UNWIND_FOR_CLIENT(fallocate, xdata, frame, op_ret, op_errno, pre,
+                               post, xdata);
+    return 0;
+}
 
 int
-pl_finodelk (call_frame_t *frame, xlator_t *this,
-             const char *volume, fd_t *fd, int32_t cmd, struct gf_flock *flock);
+pl_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t keep_size,
+             off_t offset, size_t len, dict_t *xdata)
+{
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+    STACK_WIND(frame, pl_fallocate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fallocate, fd, keep_size, offset, len,
+               xdata);
+    return 0;
+}
+
+int32_t
+pl_readlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, const char *path,
+                struct iatt *buf, dict_t *xdata)
+{
+    PL_STACK_UNWIND_FOR_CLIENT(readlink, xdata, frame, op_ret, op_errno, path,
+                               buf, xdata);
+    return 0;
+}
 
 int
-pl_entrylk (call_frame_t *frame, xlator_t *this,
-            const char *volume, loc_t *loc, const char *basename,
-            entrylk_cmd cmd, entrylk_type type);
+pl_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+            dict_t *xdata)
+{
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+    STACK_WIND(frame, pl_readlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readlink, loc, size, xdata);
+    return 0;
+}
+
+int32_t
+pl_access_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, dict_t *xdata)
+{
+    PL_STACK_UNWIND_FOR_CLIENT(access, xdata, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
 int
-pl_fentrylk (call_frame_t *frame, xlator_t *this,
-             const char *volume, fd_t *fd, const char *basename,
-             entrylk_cmd cmd, entrylk_type type);
+pl_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+          dict_t *xdata)
+{
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+    STACK_WIND(frame, pl_access_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->access, loc, mask, xdata);
+    return 0;
+}
+
+int32_t
+pl_seek_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, off_t offset, dict_t *xdata)
+{
+    PL_STACK_UNWIND_FOR_CLIENT(seek, xdata, frame, op_ret, op_errno, offset,
+                               xdata);
+    return 0;
+}
+
+int32_t
+pl_seek(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+        gf_seek_what_t what, dict_t *xdata)
+{
+    PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+    STACK_WIND(frame, pl_seek_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->seek, fd, offset, what, xdata);
+    return 0;
+}
 
 struct xlator_fops fops = {
-        .lookup      = pl_lookup,
-        .create      = pl_create,
-        .truncate    = pl_truncate,
-        .ftruncate   = pl_ftruncate,
-        .open        = pl_open,
-        .readv       = pl_readv,
-        .writev      = pl_writev,
-        .lk          = pl_lk,
-        .inodelk     = pl_inodelk,
-        .finodelk    = pl_finodelk,
-        .entrylk     = pl_entrylk,
-        .fentrylk    = pl_fentrylk,
-        .flush       = pl_flush,
-        .opendir     = pl_opendir,
+    .lookup = pl_lookup,
+    .create = pl_create,
+    .fstat = pl_fstat,
+    .truncate = pl_truncate,
+    .ftruncate = pl_ftruncate,
+    .discard = pl_discard,
+    .zerofill = pl_zerofill,
+    .open = pl_open,
+    .readv = pl_readv,
+    .writev = pl_writev,
+    .lk = pl_lk,
+    .inodelk = pl_inodelk,
+    .finodelk = pl_finodelk,
+    .entrylk = pl_entrylk,
+    .fentrylk = pl_fentrylk,
+    .flush = pl_flush,
+    .opendir = pl_opendir,
+    .readdirp = pl_readdirp,
+    .setxattr = pl_setxattr,
+    .fsetxattr = pl_fsetxattr,
+    .getxattr = pl_getxattr,
+    .fgetxattr = pl_fgetxattr,
+    .removexattr = pl_removexattr,
+    .fremovexattr = pl_fremovexattr,
+    .rename = pl_rename,
+    .getactivelk = pl_getactivelk,
+    .setactivelk = pl_setactivelk,
+    .unlink = pl_unlink,
+    .access = pl_access,
+    .readlink = pl_readlink,
+    .fallocate = pl_fallocate,
+    .fsetattr = pl_fsetattr,
+    .setattr = pl_setattr,
+    .fxattrop = pl_fxattrop,
+    .xattrop = pl_xattrop,
+    .rchecksum = pl_rchecksum,
+    .statfs = pl_statfs,
+    .fsyncdir = pl_fsyncdir,
+    .readdir = pl_readdir,
+    .symlink = pl_symlink,
+    .link = pl_link,
+    .rmdir = pl_rmdir,
+    .mknod = pl_mknod,
+    .stat = pl_stat,
+    .seek = pl_seek,
 };
 
 struct xlator_dumpops dumpops = {
-        .inodectx    = pl_dump_inode_priv,
+    .inodectx = pl_dump_inode_priv,
 };
 
 struct xlator_cbks cbks = {
-        .forget      = pl_forget,
-        .release     = pl_release,
+    .forget = pl_forget,
+    .release = pl_release,
+    .releasedir = pl_releasedir,
+    .client_destroy = pl_client_destroy_cbk,
+    .client_disconnect = pl_client_disconnect_cbk,
 };
 
-
 struct volume_options options[] = {
-        { .key  = { "mandatory-locks", "mandatory" },
-          .type = GF_OPTION_TYPE_BOOL
-        },
-	{ .key  = { "trace" },
-	  .type = GF_OPTION_TYPE_BOOL
-	},
-        { .key = {NULL} },
+    {.key = {"mandatory-locking"},
+     .type = GF_OPTION_TYPE_STR,
+     .default_value = "off",
+     .op_version = {GD_OP_VERSION_3_8_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"locks"},
+     .description = "Specifies the mandatory-locking mode. Valid options "
+                    "are 'file' to use linux style mandatory locks, "
+                    "'forced' to use volume strictly under mandatory lock "
+                    "semantics only and 'optimal' to treat advisory and "
+                    "mandatory locks separately on their own."},
+    {.key = {"trace"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .op_version = {GD_OP_VERSION_3_7_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"locks"},
+     .description = "Trace the different lock requests "
+                    "to logs."},
+    {.key = {"monkey-unlocking"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "false",
+     .op_version = {GD_OP_VERSION_3_9_0},
+     .flags = OPT_FLAG_SETTABLE,
+     .tags = {"locks"},
+     .description = "Ignore a random number of unlock requests.  Useful "
+                    "for testing/creating robust lock recovery mechanisms."},
+    {
+        .key = {"revocation-secs"},
+        .type = GF_OPTION_TYPE_INT,
+        .min = 0,
+        .max = INT_MAX,
+        .default_value = "0",
+        .op_version = {GD_OP_VERSION_3_9_0},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"locks"},
+        .description = "Maximum time a lock can be taken out, before"
+                       "being revoked.",
+    },
+    {
+        .key = {"revocation-clear-all"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "false",
+        .op_version = {GD_OP_VERSION_3_9_0},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"locks"},
+        .description = "If set to true, will revoke BOTH granted and blocked "
+                       "(pending) lock requests if a revocation threshold is "
+                       "hit.",
+    },
+    {.key = {"revocation-max-blocked"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 0,
+     .max = INT_MAX,
+     .default_value = "0",
+     .op_version = {GD_OP_VERSION_3_9_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"locks"},
+     .description = "A number of blocked lock requests after which a lock "
+                    "will be revoked to allow the others to proceed.  Can "
+                    "be used in conjunction w/ revocation-clear-all."},
+    {.key = {"notify-contention"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "yes",
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .op_version = {GD_OP_VERSION_4_0_0},
+     .tags = {"locks", "contention"},
+     .description = "When this option is enabled and a lock request "
+                    "conflicts with a currently granted lock, an upcall "
+                    "notification will be sent to the current owner of "
+                    "the lock to request it to be released as soon as "
+                    "possible."},
+    {.key = {"notify-contention-delay"},
+     .type = GF_OPTION_TYPE_INT,
+     .min = 0, /* An upcall notification is sent every time a conflict is
+                * detected. */
+     .max = 60,
+     .default_value = "5",
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .op_version = {GD_OP_VERSION_4_0_0},
+     .tags = {"locks", "contention", "timeout"},
+     .description = "This value determines the minimum amount of time "
+                    "(in seconds) between upcall contention notifications "
+                    "on the same inode. If multiple lock requests are "
+                    "received during this period, only one upcall will "
+                    "be sent."},
+    {.key = {"enforce-mandatory-lock"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     .flags = OPT_FLAG_SETTABLE,
+     .op_version = {GD_OP_VERSION_6_0},
+     .description = "option to enable lock enforcement"},
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .dumpops = &dumpops,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "locks",
+    .category = GF_MAINTAINED,
 };
diff --git a/xlators/features/locks/src/reservelk.c b/xlators/features/locks/src/reservelk.c
index e2b88140aed..604691fd887 100644
--- a/xlators/features/locks/src/reservelk.c
+++ b/xlators/features/locks/src/reservelk.c
@@ -1,450 +1,382 @@
 /*
-  Copyright (c) 2006, 2007, 2008 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+   Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
-#include "glusterfs.h"
-#include "compat.h"
-#include "xlator.h"
-#include "inode.h"
-#include "logging.h"
-#include "common-utils.h"
-#include "list.h"
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/list.h>
 
 #include "locks.h"
 #include "common.h"
 
-void
-__delete_reserve_lock (posix_lock_t *lock)
-{
-	list_del (&lock->list);
-}
-
-void
-__destroy_reserve_lock (posix_lock_t *lock)
-{
-	GF_FREE (lock);
-}
-
 /* Return true if the two reservelks have exactly same lock boundaries */
 int
-reservelks_equal (posix_lock_t *l1, posix_lock_t *l2)
+reservelks_equal(posix_lock_t *l1, posix_lock_t *l2)
 {
-	if ((l1->fl_start == l2->fl_start) &&
-	    (l1->fl_end == l2->fl_end))
-		return 1;
+    if ((l1->fl_start == l2->fl_start) && (l1->fl_end == l2->fl_end))
+        return 1;
 
-	return 0;
+    return 0;
 }
 
 /* Determine if lock is grantable or not */
 static posix_lock_t *
-__reservelk_grantable (pl_inode_t *pl_inode, posix_lock_t *lock)
+__reservelk_grantable(pl_inode_t *pl_inode, posix_lock_t *lock)
 {
-        xlator_t     *this     = NULL;
-	posix_lock_t *l        = NULL;
-        posix_lock_t *ret_lock = NULL;
-
-        this = THIS;
-
-	if (list_empty (&pl_inode->reservelk_list)) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "No reservelks in list");
-		goto out;
+    xlator_t *this = THIS;
+    posix_lock_t *l = NULL;
+    posix_lock_t *ret_lock = NULL;
+
+    if (list_empty(&pl_inode->reservelk_list)) {
+        gf_log(this->name, GF_LOG_TRACE, "No reservelks in list");
+        goto out;
+    }
+    list_for_each_entry(l, &pl_inode->reservelk_list, list)
+    {
+        if (reservelks_equal(lock, l)) {
+            ret_lock = l;
+            break;
         }
-	list_for_each_entry (l, &pl_inode->reservelk_list, list){
-		if (reservelks_equal (lock, l)) {
-                        ret_lock = l;
-                        break;
-		}
-	}
+    }
 out:
-	return ret_lock;
+    return ret_lock;
 }
 
 static int
-__same_owner_reservelk (posix_lock_t *l1, posix_lock_t *l2)
+__same_owner_reservelk(posix_lock_t *l1, posix_lock_t *l2)
 {
-        return ((l1->owner == l2->owner));
-
+    return (is_same_lkowner(&l1->owner, &l2->owner));
 }
 
 static posix_lock_t *
-__matching_reservelk (pl_inode_t *pl_inode, posix_lock_t *lock)
+__matching_reservelk(pl_inode_t *pl_inode, posix_lock_t *lock)
 {
-        posix_lock_t *l = NULL;
-
-        if (list_empty (&pl_inode->reservelk_list)) {
-            gf_log ("posix-locks", GF_LOG_TRACE,
-                    "reservelk list empty");
-            return NULL;
+    posix_lock_t *l = NULL;
+
+    if (list_empty(&pl_inode->reservelk_list)) {
+        gf_log("posix-locks", GF_LOG_TRACE, "reservelk list empty");
+        return NULL;
+    }
+
+    list_for_each_entry(l, &pl_inode->reservelk_list, list)
+    {
+        if (reservelks_equal(l, lock)) {
+            gf_log("posix-locks", GF_LOG_TRACE, "equal reservelk found");
+            break;
         }
+    }
 
-        list_for_each_entry (l, &pl_inode->reservelk_list, list) {
-                if (reservelks_equal (l, lock)) {
-                        gf_log ("posix-locks", GF_LOG_TRACE,
-                                "equal reservelk found");
-                        break;
-                }
-        }
-
-        return l;
+    return l;
 }
 
 static int
-__reservelk_conflict (xlator_t *this, pl_inode_t *pl_inode,
-                      posix_lock_t *lock)
+__reservelk_conflict(xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock)
 {
-        posix_lock_t *conf = NULL;
-        int ret = 0;
-
-        conf = __matching_reservelk (pl_inode, lock);
-        if (conf) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "Matching reservelk found");
-                if (__same_owner_reservelk (lock, conf)) {
-                        list_del_init (&conf->list);
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "Removing the matching reservelk for setlk to progress");
-                        GF_FREE (conf);
-                        ret = 0;
-                } else {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "Conflicting reservelk found");
-                        ret = 1;
-                }
-
+    int ret = 0;
+
+    posix_lock_t *conf = __matching_reservelk(pl_inode, lock);
+    if (conf) {
+        gf_log(this->name, GF_LOG_TRACE, "Matching reservelk found");
+        if (__same_owner_reservelk(lock, conf)) {
+            list_del_init(&conf->list);
+            gf_log(this->name, GF_LOG_TRACE,
+                   "Removing the matching reservelk for setlk to progress");
+            __destroy_lock(conf);
+            ret = 0;
+        } else {
+            gf_log(this->name, GF_LOG_TRACE, "Conflicting reservelk found");
+            ret = 1;
         }
-        return ret;
-
+    }
+    return ret;
 }
 
 int
-pl_verify_reservelk (xlator_t *this, pl_inode_t *pl_inode,
-                     posix_lock_t *lock, int can_block)
+pl_verify_reservelk(xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock,
+                    const int can_block)
 {
-        int ret = 0;
-
-        pthread_mutex_lock (&pl_inode->mutex);
-        {
-                if (__reservelk_conflict (this, pl_inode, lock)) {
-                        gf_log (this->name, GF_LOG_TRACE,
-                                "Found conflicting reservelk. Blocking until reservelk is unlocked.");
-                        lock->blocked = can_block;
-                        list_add_tail (&lock->list, &pl_inode->blocked_calls);
-                        ret = -1;
-                        goto unlock;
-                }
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "no conflicting reservelk found. Call continuing");
-                ret = 0;
-
+    int ret = 0;
+
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        if (__reservelk_conflict(this, pl_inode, lock)) {
+            lock->blocked = can_block;
+            list_add_tail(&lock->list, &pl_inode->blocked_calls);
+            pthread_mutex_unlock(&pl_inode->mutex);
+            gf_log(this->name, GF_LOG_TRACE,
+                   "Found conflicting reservelk. Blocking until reservelk is "
+                   "unlocked.");
+            ret = -1;
+            goto out;
         }
-unlock:
-        pthread_mutex_unlock (&pl_inode->mutex);
-
-        return ret;
-
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+    gf_log(this->name, GF_LOG_TRACE,
+           "no conflicting reservelk found. Call continuing");
+    ret = 0;
+out:
+    return ret;
 }
 
-
 /* Determines if lock can be granted and adds the lock. If the lock
  * is blocking, adds it to the blocked_reservelks.
  */
 static int
-__lock_reservelk (xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock,
-                  int can_block)
+__lock_reservelk(xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock,
+                 const int can_block)
 {
-	posix_lock_t *conf = NULL;
-	int ret = -EINVAL;
+    int ret = -EINVAL;
 
-	conf = __reservelk_grantable (pl_inode, lock);
-	if (conf){
-		ret = -EAGAIN;
-		if (can_block == 0)
-			goto out;
+    posix_lock_t *conf = __reservelk_grantable(pl_inode, lock);
+    if (conf) {
+        ret = -EAGAIN;
+        if (can_block == 0)
+            goto out;
 
-		list_add_tail (&lock->list, &pl_inode->blocked_reservelks);
+        list_add_tail(&lock->list, &pl_inode->blocked_reservelks);
 
-                gf_log (this->name, GF_LOG_TRACE,
-                        "%s (pid=%d) lk-owner:%"PRIu64" %"PRId64" - %"PRId64" => Blocked",
-                        lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
-                        lock->client_pid,
-                        lock->owner,
-                        lock->user_flock.l_start,
-                        lock->user_flock.l_len);
+        gf_log(this->name, GF_LOG_TRACE,
+               "%s (pid=%d) lk-owner:%s %" PRId64 " - %" PRId64 " => Blocked",
+               lock->fl_type == F_UNLCK ? "Unlock" : "Lock", lock->client_pid,
+               lkowner_utoa(&lock->owner), lock->user_flock.l_start,
+               lock->user_flock.l_len);
 
+        goto out;
+    }
 
-		goto out;
-	}
+    list_add(&lock->list, &pl_inode->reservelk_list);
 
-	list_add (&lock->list, &pl_inode->reservelk_list);
-
-	ret = 0;
+    ret = 0;
 
 out:
-	return ret;
+    return ret;
 }
 
 static posix_lock_t *
-find_matching_reservelk (posix_lock_t *lock, pl_inode_t *pl_inode)
+find_matching_reservelk(posix_lock_t *lock, pl_inode_t *pl_inode)
 {
-	posix_lock_t *l = NULL;
-	list_for_each_entry (l, &pl_inode->reservelk_list, list) {
-		if (reservelks_equal (l, lock))
-			return l;
-	}
-	return NULL;
+    posix_lock_t *l = NULL;
+    list_for_each_entry(l, &pl_inode->reservelk_list, list)
+    {
+        if (reservelks_equal(l, lock))
+            return l;
+    }
+    return NULL;
 }
 
 /* Set F_UNLCK removes a lock which has the exact same lock boundaries
  * as the UNLCK lock specifies. If such a lock is not found, returns invalid
  */
 static posix_lock_t *
-__reserve_unlock_lock (xlator_t *this, posix_lock_t *lock, pl_inode_t *pl_inode)
+__reserve_unlock_lock(xlator_t *this, posix_lock_t *lock, pl_inode_t *pl_inode)
 {
-
-	posix_lock_t *conf = NULL;
-
-	conf = find_matching_reservelk (lock, pl_inode);
-	if (!conf) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        " Matching lock not found for unlock");
-		goto out;
-        }
-	__delete_reserve_lock (conf);
-        gf_log (this->name, GF_LOG_DEBUG,
-                " Matching lock found for unlock");
+    posix_lock_t *conf = find_matching_reservelk(lock, pl_inode);
+    if (!conf) {
+        gf_log(this->name, GF_LOG_DEBUG, " Matching lock not found for unlock");
+        goto out;
+    }
+    __delete_lock(conf);
+    gf_log(this->name, GF_LOG_DEBUG, " Matching lock found for unlock");
 
 out:
-	return conf;
-
-
+    return conf;
 }
 
 static void
-__grant_blocked_reserve_locks (xlator_t *this, pl_inode_t *pl_inode,
-                               struct list_head *granted)
+__grant_blocked_reserve_locks(xlator_t *this, pl_inode_t *pl_inode,
+                              struct list_head *granted)
 {
-	int	      bl_ret = 0;
-	posix_lock_t *bl = NULL;
-	posix_lock_t *tmp = NULL;
+    int bl_ret = 0;
+    posix_lock_t *bl = NULL;
+    posix_lock_t *tmp = NULL;
 
-        struct list_head blocked_list;
+    struct list_head blocked_list;
 
-        INIT_LIST_HEAD (&blocked_list);
-        list_splice_init (&pl_inode->blocked_reservelks, &blocked_list);
+    INIT_LIST_HEAD(&blocked_list);
+    list_splice_init(&pl_inode->blocked_reservelks, &blocked_list);
 
-	list_for_each_entry_safe (bl, tmp, &blocked_list, list) {
+    list_for_each_entry_safe(bl, tmp, &blocked_list, list)
+    {
+        list_del_init(&bl->list);
 
-		list_del_init (&bl->list);
+        bl_ret = __lock_reservelk(this, pl_inode, bl, 1);
 
-		bl_ret = __lock_reservelk (this, pl_inode, bl, 1);
-
-		if (bl_ret == 0) {
-			list_add (&bl->list, granted);
-                }
+        if (bl_ret == 0) {
+            list_add(&bl->list, granted);
         }
-	return;
+    }
+    return;
 }
 
 /* Grant all reservelks blocked on lock(s) */
 void
-grant_blocked_reserve_locks (xlator_t *this, pl_inode_t *pl_inode)
+grant_blocked_reserve_locks(xlator_t *this, pl_inode_t *pl_inode)
 {
-	struct list_head granted;
-	posix_lock_t *lock = NULL;
-	posix_lock_t *tmp = NULL;
-
-	INIT_LIST_HEAD (&granted);
-
-        if (list_empty (&pl_inode->blocked_reservelks)) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "No blocked locks to be granted");
-                return;
-        }
-
-        pthread_mutex_lock (&pl_inode->mutex);
-	{
-		__grant_blocked_reserve_locks (this, pl_inode, &granted);
-	}
-        pthread_mutex_unlock (&pl_inode->mutex);
-
-	list_for_each_entry_safe (lock, tmp, &granted, list) {
-                gf_log (this->name, GF_LOG_TRACE,
-			"%s (pid=%d) (lk-owner=%"PRIu64") %"PRId64" - %"PRId64" => Granted",
-			lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
-			lock->client_pid,
-			lock->owner,
-			lock->user_flock.l_start,
-			lock->user_flock.l_len);
-
-		STACK_UNWIND_STRICT (lk, lock->frame, 0, 0, &lock->user_flock);
-	}
-
+    struct list_head granted;
+    posix_lock_t *lock = NULL;
+    posix_lock_t *tmp = NULL;
+
+    INIT_LIST_HEAD(&granted);
+
+    if (list_empty(&pl_inode->blocked_reservelks)) {
+        gf_log(this->name, GF_LOG_TRACE, "No blocked locks to be granted");
+        return;
+    }
+
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        __grant_blocked_reserve_locks(this, pl_inode, &granted);
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+
+    list_for_each_entry_safe(lock, tmp, &granted, list)
+    {
+        gf_log(this->name, GF_LOG_TRACE,
+               "%s (pid=%d) (lk-owner=%s) %" PRId64 " - %" PRId64 " => Granted",
+               lock->fl_type == F_UNLCK ? "Unlock" : "Lock", lock->client_pid,
+               lkowner_utoa(&lock->owner), lock->user_flock.l_start,
+               lock->user_flock.l_len);
+
+        STACK_UNWIND_STRICT(lk, lock->frame, 0, 0, &lock->user_flock, NULL);
+    }
 }
 
 static void
-__grant_blocked_lock_calls (xlator_t *this, pl_inode_t *pl_inode,
-                            struct list_head *granted)
+__grant_blocked_lock_calls(xlator_t *this, pl_inode_t *pl_inode,
+                           struct list_head *granted)
 {
-	int	      bl_ret = 0;
-	posix_lock_t *bl = NULL;
-	posix_lock_t *tmp = NULL;
-
-        struct list_head blocked_list;
+    int bl_ret = 0;
+    posix_lock_t *bl = NULL;
+    posix_lock_t *tmp = NULL;
 
-        INIT_LIST_HEAD (&blocked_list);
-        list_splice_init (&pl_inode->blocked_reservelks, &blocked_list);
+    struct list_head blocked_list;
 
-	list_for_each_entry_safe (bl, tmp, &blocked_list, list) {
+    INIT_LIST_HEAD(&blocked_list);
+    list_splice_init(&pl_inode->blocked_reservelks, &blocked_list);
 
-		list_del_init (&bl->list);
+    list_for_each_entry_safe(bl, tmp, &blocked_list, list)
+    {
+        list_del_init(&bl->list);
 
-		bl_ret = pl_verify_reservelk (this, pl_inode, bl, bl->blocked);
+        bl_ret = pl_verify_reservelk(this, pl_inode, bl, bl->blocked);
 
-		if (bl_ret == 0) {
-                        list_add_tail (&bl->list, granted);
-                }
+        if (bl_ret == 0) {
+            list_add_tail(&bl->list, granted);
         }
-	return;
+    }
+    return;
 }
 
 void
-grant_blocked_lock_calls (xlator_t *this, pl_inode_t *pl_inode)
+grant_blocked_lock_calls(xlator_t *this, pl_inode_t *pl_inode)
 {
-	struct list_head granted;
-	posix_lock_t *lock = NULL;
-	posix_lock_t *tmp = NULL;
-        fd_t         *fd  = NULL;
-
-        int can_block = 0;
-        int32_t cmd = 0;
-        int ret = 0;
-
-        if (list_empty (&pl_inode->blocked_calls)) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "No blocked lock calls to be granted");
-                return;
-        }
-
-        pthread_mutex_lock (&pl_inode->mutex);
-	{
-		__grant_blocked_lock_calls (this, pl_inode, &granted);
-	}
-        pthread_mutex_unlock (&pl_inode->mutex);
-
-        list_for_each_entry_safe (lock, tmp, &granted, list) {
-                fd = fd_from_fdnum (lock);
-
-                if (lock->blocked) {
-                        can_block = 1;
-                        cmd = F_SETLKW;
-                }
-                else
-                        cmd = F_SETLK;
-
-                lock->blocked = 0;
-                ret = pl_setlk (this, pl_inode, lock, can_block);
-                if (ret == -1) {
-                        if (can_block) {
-                                pl_trace_block (this, lock->frame, fd, NULL,
-                                                cmd, &lock->user_flock, NULL);
-                                continue;
-                        } else {
-                                gf_log (this->name, GF_LOG_DEBUG, "returning EAGAIN");
-                                pl_trace_out (this, lock->frame, fd, NULL, cmd,
-                                              &lock->user_flock, -1, EAGAIN, NULL);
-                                pl_update_refkeeper (this, fd->inode);
-                                STACK_UNWIND_STRICT (lk, lock->frame, -1, EAGAIN, &lock->user_flock);
-                                __destroy_lock (lock);
-                        }
-                }
-
+    struct list_head granted;
+    posix_lock_t *lock = NULL;
+    posix_lock_t *tmp = NULL;
+    fd_t *fd = NULL;
+
+    int can_block = 0;
+    int32_t cmd = 0;
+    int ret = 0;
+
+    if (list_empty(&pl_inode->blocked_calls)) {
+        gf_log(this->name, GF_LOG_TRACE, "No blocked lock calls to be granted");
+        return;
+    }
+
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        __grant_blocked_lock_calls(this, pl_inode, &granted);
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+
+    list_for_each_entry_safe(lock, tmp, &granted, list)
+    {
+        fd = fd_from_fdnum(lock);
+
+        if (lock->blocked) {
+            can_block = 1;
+            cmd = F_SETLKW;
+        } else
+            cmd = F_SETLK;
+
+        lock->blocked = 0;
+        ret = pl_setlk(this, pl_inode, lock, can_block);
+        if (ret == -1) {
+            if (can_block) {
+                continue;
+            } else {
+                gf_log(this->name, GF_LOG_DEBUG, "returning EAGAIN");
+                pl_trace_out(this, lock->frame, fd, NULL, cmd,
+                             &lock->user_flock, -1, EAGAIN, NULL);
+                pl_update_refkeeper(this, fd->inode);
+                STACK_UNWIND_STRICT(lk, lock->frame, -1, EAGAIN,
+                                    &lock->user_flock, NULL);
+                __destroy_lock(lock);
+            }
         }
-
+    }
 }
 
-
 int
-pl_reserve_unlock (xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock)
+pl_reserve_unlock(xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock)
 {
-        posix_lock_t *retlock = NULL;
-        int ret = -1;
-
-        pthread_mutex_lock (&pl_inode->mutex);
-	{
-		retlock = __reserve_unlock_lock (this, lock, pl_inode);
-		if (!retlock) {
-			gf_log (this->name, GF_LOG_DEBUG,
-				"Bad Unlock issued on Inode lock");
-                        ret = -EINVAL;
-                        goto out;
-                }
-
-                gf_log (this->name, GF_LOG_TRACE,
-                        "Reservelk Unlock successful");
-		__destroy_reserve_lock (retlock);
-                ret = 0;
+    posix_lock_t *retlock = NULL;
+    int ret = -1;
+
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        retlock = __reserve_unlock_lock(this, lock, pl_inode);
+        if (!retlock) {
+            pthread_mutex_unlock(&pl_inode->mutex);
+            gf_log(this->name, GF_LOG_DEBUG, "Bad Unlock issued on Inode lock");
+            ret = -EINVAL;
+            goto out;
         }
-out:
-        pthread_mutex_unlock (&pl_inode->mutex);
 
-        grant_blocked_reserve_locks (this, pl_inode);
-        grant_blocked_lock_calls (this, pl_inode);
-
-        return ret;
+        gf_log(this->name, GF_LOG_TRACE, "Reservelk Unlock successful");
+        __destroy_lock(retlock);
+        ret = 0;
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+out:
+    grant_blocked_reserve_locks(this, pl_inode);
+    grant_blocked_lock_calls(this, pl_inode);
 
+    return ret;
 }
 
 int
-pl_reserve_setlk (xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock,
-                  int can_block)
+pl_reserve_setlk(xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock,
+                 int can_block)
 {
-	int ret = -EINVAL;
-
-	pthread_mutex_lock (&pl_inode->mutex);
-	{
-
-			ret = __lock_reservelk (this, pl_inode, lock, can_block);
-                        if (ret < 0)
-				gf_log (this->name, GF_LOG_TRACE,
-                                        "%s (pid=%d) (lk-owner=%"PRIu64") %"PRId64" - %"PRId64" => NOK",
-					lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
-					lock->client_pid,
-                                        lock->owner,
-					lock->user_flock.l_start,
-					lock->user_flock.l_len);
-                        else
-				gf_log (this->name, GF_LOG_TRACE,
-                                        "%s (pid=%d) (lk-owner=%"PRIu64") %"PRId64" - %"PRId64" => OK",
-                                        lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
-                                        lock->client_pid,
-                                        lock->owner,
-                                        lock->fl_start,
-                                        lock->fl_end);
-
-	}
-	pthread_mutex_unlock (&pl_inode->mutex);
-        return ret;
+    int ret = -EINVAL;
+
+    pthread_mutex_lock(&pl_inode->mutex);
+    {
+        ret = __lock_reservelk(this, pl_inode, lock, can_block);
+    }
+    pthread_mutex_unlock(&pl_inode->mutex);
+
+    if (ret < 0)
+        gf_log(this->name, GF_LOG_TRACE,
+               "%s (pid=%d) (lk-owner=%s) %" PRId64 " - %" PRId64 " => NOK",
+               lock->fl_type == F_UNLCK ? "Unlock" : "Lock", lock->client_pid,
+               lkowner_utoa(&lock->owner), lock->user_flock.l_start,
+               lock->user_flock.l_len);
+    else
+        gf_log(this->name, GF_LOG_TRACE,
+               "%s (pid=%d) (lk-owner=%s) %" PRId64 " - %" PRId64 " => OK",
+               lock->fl_type == F_UNLCK ? "Unlock" : "Lock", lock->client_pid,
+               lkowner_utoa(&lock->owner), lock->fl_start, lock->fl_end);
+
+    return ret;
 }
diff --git a/xlators/features/locks/tests/unit-test.c b/xlators/features/locks/tests/unit-test.c
index 889c304da1a..d285b12b5aa 100644
--- a/xlators/features/locks/tests/unit-test.c
+++ b/xlators/features/locks/tests/unit-test.c
@@ -1,75 +1,77 @@
 /*
-  Copyright (c) 2006-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
+   Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
 */
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "compat.h"
-#include "xlator.h"
-#include "inode.h"
-#include "logging.h"
-#include "common-utils.h"
-#include "list.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/list.h>
 
 #include "locks.h"
 #include "common.h"
 
-#define expect(cond) if (!(cond)) { goto out; }
+#define expect(cond)                                                           \
+    if (!(cond)) {                                                             \
+        goto out;                                                              \
+    }
 
-extern int lock_name (pl_inode_t *, const char *, entrylk_type);
-extern int unlock_name (pl_inode_t *, const char *, entrylk_type);
+extern int
+lock_name(pl_inode_t *, const char *, entrylk_type);
+extern int
+unlock_name(pl_inode_t *, const char *, entrylk_type);
 
-int main (int argc, char **argv)
+int
+main(int argc, char **argv)
 {
-	int ret = 1;
-	int r = -1;
+    int ret = 1;
+    int r = -1;
+
+    pl_inode_t *pinode = CALLOC(sizeof(pl_inode_t), 1);
+    pthread_mutex_init(&pinode->dir_lock_mutex, NULL);
+    INIT_LIST_HEAD(&pinode->gf_dir_locks);
 
-	pl_inode_t *pinode = CALLOC (sizeof (pl_inode_t), 1);
-	pthread_mutex_init (&pinode->dir_lock_mutex, NULL);
-	INIT_LIST_HEAD (&pinode->gf_dir_locks);
+    r = lock_name(pinode, NULL, ENTRYLK_WRLCK);
+    expect(r == 0);
+    {
+        r = lock_name(pinode, "foo", ENTRYLK_WRLCK);
+        expect(r == -EAGAIN);
+    }
+    r = unlock_name(pinode, NULL, ENTRYLK_WRLCK);
+    expect(r == 0);
 
-	r = lock_name (pinode, NULL, ENTRYLK_WRLCK); expect (r == 0);
-	{
-		r = lock_name (pinode, "foo", ENTRYLK_WRLCK); expect (r == -EAGAIN);
-	}
-	r = unlock_name (pinode, NULL, ENTRYLK_WRLCK); expect (r == 0);
+    r = lock_name(pinode, "foo", ENTRYLK_RDLCK);
+    expect(r == 0);
+    {
+        r = lock_name(pinode, "foo", ENTRYLK_RDLCK);
+        expect(r == 0);
+        {
+            r = lock_name(pinode, "foo", ENTRYLK_WRLCK);
+            expect(r == -EAGAIN);
+        }
+        r = unlock_name(pinode, "foo", ENTRYLK_RDLCK);
+        expect(r == 0);
+    }
+    r = unlock_name(pinode, "foo", ENTRYLK_RDLCK);
+    expect(r == 0);
 
-	r = lock_name (pinode, "foo", ENTRYLK_RDLCK); expect (r == 0);
-	{
-		r = lock_name (pinode, "foo", ENTRYLK_RDLCK); expect (r == 0);
-		{
-			r = lock_name (pinode, "foo", ENTRYLK_WRLCK); expect (r == -EAGAIN);
-		}
-		r = unlock_name (pinode, "foo", ENTRYLK_RDLCK); expect (r == 0);
-	}
-	r = unlock_name (pinode, "foo", ENTRYLK_RDLCK); expect (r == 0);
-	
-	r = lock_name (pinode, "foo", ENTRYLK_WRLCK); expect (r == 0);
-	r = unlock_name (pinode, "foo", ENTRYLK_WRLCK); expect (r == 0);
+    r = lock_name(pinode, "foo", ENTRYLK_WRLCK);
+    expect(r == 0);
+    r = unlock_name(pinode, "foo", ENTRYLK_WRLCK);
+    expect(r == 0);
 
-	r = lock_name (pinode, "baz", ENTRYLK_WRLCK); expect (r == 0);
-	r = lock_name (pinode, "baz", ENTRYLK_RDLCK); expect (r == -EAGAIN);
+    r = lock_name(pinode, "baz", ENTRYLK_WRLCK);
+    expect(r == 0);
+    r = lock_name(pinode, "baz", ENTRYLK_RDLCK);
+    expect(r == -EAGAIN);
 
-	ret = 0;
+    ret = 0;
 out:
-	return ret;
+    return ret;
 }
diff --git a/xlators/features/mac-compat/Makefile.am b/xlators/features/mac-compat/Makefile.am
deleted file mode 100644
index d471a3f9243..00000000000
--- a/xlators/features/mac-compat/Makefile.am
+++ /dev/null
@@ -1,3 +0,0 @@
-SUBDIRS = src
-
-CLEANFILES = 
diff --git a/xlators/features/mac-compat/src/Makefile.am b/xlators/features/mac-compat/src/Makefile.am
deleted file mode 100644
index 915c13e308f..00000000000
--- a/xlators/features/mac-compat/src/Makefile.am
+++ /dev/null
@@ -1,13 +0,0 @@
-xlator_LTLIBRARIES = mac-compat.la
-xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
-
-mac_compat_la_LDFLAGS = -module -avoidversion
-
-mac_compat_la_SOURCES = mac-compat.c
-mac_compat_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
-
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS) \
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
-
-CLEANFILES = 
-
diff --git a/xlators/features/mac-compat/src/mac-compat.c b/xlators/features/mac-compat/src/mac-compat.c
deleted file mode 100644
index b1277ec1610..00000000000
--- a/xlators/features/mac-compat/src/mac-compat.c
+++ /dev/null
@@ -1,247 +0,0 @@
-/*
-  Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "xlator.h"
-#include "defaults.h"
-#include "compat-errno.h"
-
-
-enum apple_xattr {
-        GF_FINDER_INFO_XATTR,
-        GF_RESOURCE_FORK_XATTR,
-        GF_XATTR_ALL,
-        GF_XATTR_NONE
-};
-
-static char *apple_xattr_name[] = {
-        [GF_FINDER_INFO_XATTR]   = "com.apple.FinderInfo",
-        [GF_RESOURCE_FORK_XATTR] = "com.apple.ResourceFork"
-};
-
-static const char *apple_xattr_value[] = {
-        [GF_FINDER_INFO_XATTR]   =
-        /* 1 2 3 4 5 6 7 8 */
-         "\0\0\0\0\0\0\0\0"
-         "\0\0\0\0\0\0\0\0"
-         "\0\0\0\0\0\0\0\0"
-         "\0\0\0\0\0\0\0\0",
-        [GF_RESOURCE_FORK_XATTR] = ""
-};
-
-static int32_t apple_xattr_len[] = {
-        [GF_FINDER_INFO_XATTR]   = 32,
-        [GF_RESOURCE_FORK_XATTR] = 1
-};
-
-
-int32_t
-maccomp_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno, dict_t *dict)
-{
-        intptr_t ax = (intptr_t)this->private;
-        int i = 0;
-
-        if ((ax == GF_XATTR_ALL && op_ret >= 0) || ax != GF_XATTR_NONE) {
-                op_ret = op_errno = 0;
-
-                for (i = 0; i < GF_XATTR_ALL; i++) {
-                        if (dict_get (dict, apple_xattr_name[i]))
-                                continue;
-
-                        if (dict_set (dict, apple_xattr_name[i],
-                                      bin_to_data ((void *)apple_xattr_value[i],
-                                                   apple_xattr_len[i])) == -1) {
-                                op_ret = -1;
-                                op_errno = ENOMEM;
-
-                                break;
-                        }
-                }
-         }
-
-        STACK_UNWIND_STRICT (getxattr, frame, op_ret, op_errno, dict);
-
-        return 0;
-}
-
-
-int32_t
-maccomp_getxattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                  const char *name)
-{
-        intptr_t ax = GF_XATTR_NONE;
-        int i = 0;
-
-        if (name) {
-                for (i = 0; i < GF_XATTR_ALL; i++) {
-                        if (strcmp (apple_xattr_name[i], name) == 0) {
-                                ax = i;
-
-                                break;
-                        }
-                }
-        } else
-                ax = GF_XATTR_ALL;
-
-        this->private = (void *)ax;
-
-        STACK_WIND (frame, maccomp_getxattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->getxattr,
-                    loc, name);
-        return 0;
-}
-
-
-int32_t
-maccomp_fgetxattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
-                   const char *name)
-{
-        intptr_t ax = GF_XATTR_NONE;
-        int i = 0;
-
-        if (name) {
-                for (i = 0; i < GF_XATTR_ALL; i++) {
-                        if (strcmp (apple_xattr_name[i], name) == 0) {
-                                ax = i;
-
-                                break;
-                        }
-                }
-        } else
-                ax = GF_XATTR_ALL;
-
-        this->private = (void *)ax;
-
-        STACK_WIND (frame, maccomp_getxattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fgetxattr,
-                    fd, name);
-        return 0;
-}
-
-
-int32_t
-maccomp_setxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                      int32_t op_ret, int32_t op_errno)
-{
-        intptr_t ax = (intptr_t)this->private;
-
-        if (op_ret == -1 && ax != GF_XATTR_NONE)
-                op_ret = op_errno = 0;
-
-        STACK_UNWIND_STRICT (setxattr, frame, op_ret, op_errno);
-
-        return 0;
-}
-
-
-int32_t
-maccomp_setxattr (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
-                  int32_t flags)
-{
-        intptr_t ax = GF_XATTR_NONE;
-        int i = 0;
-
-        for (i = 0; i < GF_XATTR_ALL; i++) {
-                if (dict_get (dict, apple_xattr_name[i])) {
-                        ax = i;
-
-                        break;
-                }
-        }
-
-        this->private = (void *)ax;
-
-        STACK_WIND (frame, maccomp_setxattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->setxattr,
-                    loc, dict, flags);
-        return 0;
-}
-
-
-int32_t
-maccomp_fsetxattr (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
-                   int32_t flags)
-{
-        intptr_t ax = GF_XATTR_NONE;
-        int i = 0;
-
-        for (i = 0; i < GF_XATTR_ALL; i++) {
-                if (dict_get (dict, apple_xattr_name[i])) {
-                        ax = i;
-
-                        break;
-                }
-        }
-
-        this->private = (void *)ax;
-
-        STACK_WIND (frame, maccomp_setxattr_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->fsetxattr,
-                    fd, dict, flags);
-        return 0;
-}
-
-
-int32_t
-init (xlator_t *this)
-{
-        if (!this->children || this->children->next) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "translator not configured with exactly one child");
-                return -1;
-        }
-
-        if (!this->parents) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "dangling volume. check volfile ");
-        }
-
-        return 0;
-}
-
-
-void
-fini (xlator_t *this)
-{
-        return;
-}
-
-
-struct xlator_fops fops = {
-        .getxattr    = maccomp_getxattr,
-        .fgetxattr   = maccomp_fgetxattr,
-        .setxattr    = maccomp_setxattr,
-        .fsetxattr   = maccomp_fsetxattr,
-};
-
-struct xlator_cbks cbks = {
-};
-
-struct volume_options options[] = {
-        { .key = {NULL} },
-};
diff --git a/xlators/features/marker/Makefile.am b/xlators/features/marker/Makefile.am
new file mode 100644
index 00000000000..a985f42a877
--- /dev/null
+++ b/xlators/features/marker/Makefile.am
@@ -0,0 +1,3 @@
+SUBDIRS = src
+
+CLEANFILES =
diff --git a/xlators/features/marker/src/Makefile.am b/xlators/features/marker/src/Makefile.am
new file mode 100644
index 00000000000..58056b36511
--- /dev/null
+++ b/xlators/features/marker/src/Makefile.am
@@ -0,0 +1,24 @@
+if WITH_SERVER
+xlator_LTLIBRARIES = marker.la
+endif
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+marker_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+marker_la_SOURCES = marker.c marker-quota.c marker-quota-helper.c \
+	marker-common.c
+
+marker_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = marker-mem-types.h marker.h marker-quota.h \
+	marker-quota-helper.h marker-common.h \
+	$(top_builddir)/xlators/lib/src/libxlator.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src \
+	-I$(top_srcdir)/xlators/lib/src
+
+AM_CFLAGS = -Wall -fno-strict-aliasing $(GF_CFLAGS)
+
+CLEANFILES =
+
diff --git a/xlators/features/marker/src/marker-common.c b/xlators/features/marker/src/marker-common.c
new file mode 100644
index 00000000000..9c9047005d6
--- /dev/null
+++ b/xlators/features/marker/src/marker-common.c
@@ -0,0 +1,57 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include <fnmatch.h>
+#include "marker-common.h"
+
+marker_inode_ctx_t *
+marker_inode_ctx_new()
+{
+    marker_inode_ctx_t *ctx = NULL;
+
+    ctx = GF_CALLOC(1, sizeof(marker_inode_ctx_t),
+                    gf_marker_mt_marker_inode_ctx_t);
+    if (ctx == NULL)
+        goto out;
+
+    ctx->quota_ctx = NULL;
+out:
+    return ctx;
+}
+
+int32_t
+marker_force_inode_ctx_get(inode_t *inode, xlator_t *this,
+                           marker_inode_ctx_t **ctx)
+{
+    int32_t ret = -1;
+    uint64_t ctx_int = 0;
+
+    LOCK(&inode->lock);
+    {
+        ret = __inode_ctx_get(inode, this, &ctx_int);
+        if (ret == 0)
+            *ctx = (marker_inode_ctx_t *)(unsigned long)ctx_int;
+        else {
+            *ctx = marker_inode_ctx_new();
+            if (*ctx == NULL)
+                goto unlock;
+
+            ret = __inode_ctx_put(inode, this, (uint64_t)(unsigned long)*ctx);
+            if (ret == -1) {
+                GF_FREE(*ctx);
+                goto unlock;
+            }
+            ret = 0;
+        }
+    }
+unlock:
+    UNLOCK(&inode->lock);
+
+    return ret;
+}
diff --git a/xlators/features/marker/src/marker-common.h b/xlators/features/marker/src/marker-common.h
new file mode 100644
index 00000000000..7f8cffe7d35
--- /dev/null
+++ b/xlators/features/marker/src/marker-common.h
@@ -0,0 +1,19 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#ifndef _MARKER_COMMON_H
+#define _MARKER_COMMON_H
+
+#include <glusterfs/xlator.h>
+#include "marker.h"
+
+int32_t
+marker_force_inode_ctx_get(inode_t *, xlator_t *, marker_inode_ctx_t **);
+
+#endif
diff --git a/xlators/features/marker/src/marker-mem-types.h b/xlators/features/marker/src/marker-mem-types.h
new file mode 100644
index 00000000000..aedfdb4a1b7
--- /dev/null
+++ b/xlators/features/marker/src/marker-mem-types.h
@@ -0,0 +1,28 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#ifndef __MARKER_MEM_TYPES_H__
+#define __MARKER_MEM_TYPES_H__
+
+#include <glusterfs/mem-types.h>
+
+enum gf_marker_mem_types_ {
+    /* Those are used by ALLOCATE_OR_GOTO macro */
+    gf_marker_mt_marker_conf_t = gf_common_mt_end + 1,
+    gf_marker_mt_loc_t,
+    gf_marker_mt_volume_mark,
+    gf_marker_mt_int64_t,
+    gf_marker_mt_quota_inode_ctx_t,
+    gf_marker_mt_marker_inode_ctx_t,
+    gf_marker_mt_inode_contribution_t,
+    gf_marker_mt_quota_meta_t,
+    gf_marker_mt_quota_synctask_t,
+    gf_marker_mt_end
+};
+#endif
diff --git a/xlators/features/marker/src/marker-quota-helper.c b/xlators/features/marker/src/marker-quota-helper.c
new file mode 100644
index 00000000000..ecd85d67b2b
--- /dev/null
+++ b/xlators/features/marker/src/marker-quota-helper.c
@@ -0,0 +1,380 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include <glusterfs/locking.h>
+#include "marker-quota.h"
+#include "marker-common.h"
+#include "marker-quota-helper.h"
+#include "marker-mem-types.h"
+
+int
+mq_loc_fill(loc_t *loc, inode_t *inode, inode_t *parent, char *path)
+{
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("marker", loc, out);
+    GF_VALIDATE_OR_GOTO("marker", inode, out);
+    GF_VALIDATE_OR_GOTO("marker", path, out);
+    /* Not checking for parent because while filling
+     * loc of root, parent will be NULL
+     */
+
+    if (inode) {
+        loc->inode = inode_ref(inode);
+    }
+
+    if (parent)
+        loc->parent = inode_ref(parent);
+
+    if (!gf_uuid_is_null(inode->gfid))
+        gf_uuid_copy(loc->gfid, inode->gfid);
+
+    loc->path = gf_strdup(path);
+    if (!loc->path) {
+        gf_log("loc fill", GF_LOG_ERROR, "strdup failed");
+        goto out;
+    }
+
+    loc->name = strrchr(loc->path, '/');
+    if (loc->name)
+        loc->name++;
+    else
+        goto out;
+
+    ret = 0;
+
+out:
+    if (ret < 0)
+        loc_wipe(loc);
+
+    return ret;
+}
+
+int32_t
+mq_inode_loc_fill(const char *parent_gfid, inode_t *inode, loc_t *loc)
+{
+    char *resolvedpath = NULL;
+    inode_t *parent = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    xlator_t *this = NULL;
+    int ret = -1;
+
+    this = THIS;
+
+    if (inode == NULL) {
+        gf_log_callingfn("marker", GF_LOG_ERROR,
+                         "loc fill failed, "
+                         "inode is NULL");
+        return ret;
+    }
+
+    if (loc == NULL)
+        return ret;
+
+    if ((inode) && __is_root_gfid(inode->gfid)) {
+        loc->parent = NULL;
+        goto ignore_parent;
+    }
+
+    if (parent_gfid == NULL)
+        parent = inode_parent(inode, 0, NULL);
+    else
+        parent = inode_find(inode->table, (unsigned char *)parent_gfid);
+
+    if (parent == NULL) {
+        gf_log("marker", GF_LOG_ERROR, "parent is NULL for %s",
+               uuid_utoa(inode->gfid));
+        goto err;
+    }
+
+ignore_parent:
+    ret = inode_path(inode, NULL, &resolvedpath);
+    if (ret < 0) {
+        gf_log("marker", GF_LOG_ERROR, "failed to resolve path for %s",
+               uuid_utoa(inode->gfid));
+        goto err;
+    }
+
+    ret = mq_loc_fill(loc, inode, parent, resolvedpath);
+    if (ret < 0)
+        goto err;
+
+    ret = mq_inode_ctx_get(inode, this, &ctx);
+    if (ret < 0 || ctx == NULL)
+        ctx = mq_inode_ctx_new(inode, this);
+    if (ctx == NULL) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "mq_inode_ctx_new "
+               "failed for %s",
+               uuid_utoa(inode->gfid));
+        ret = -1;
+        goto err;
+    }
+    ret = 0;
+
+err:
+    if (parent)
+        inode_unref(parent);
+
+    GF_FREE(resolvedpath);
+
+    return ret;
+}
+
+quota_inode_ctx_t *
+mq_alloc_inode_ctx()
+{
+    int32_t ret = -1;
+    quota_inode_ctx_t *ctx = NULL;
+
+    QUOTA_ALLOC(ctx, quota_inode_ctx_t, ret);
+    if (ret == -1)
+        goto out;
+
+    ctx->size = 0;
+    ctx->dirty = 0;
+    ctx->updation_status = _gf_false;
+    LOCK_INIT(&ctx->lock);
+    INIT_LIST_HEAD(&ctx->contribution_head);
+out:
+    return ctx;
+}
+
+static void
+mq_contri_fini(inode_contribution_t *contri)
+{
+    LOCK_DESTROY(&contri->lock);
+    GF_FREE(contri);
+}
+
+inode_contribution_t *
+mq_contri_init(inode_t *inode)
+{
+    inode_contribution_t *contri = NULL;
+    int32_t ret = 0;
+
+    QUOTA_ALLOC(contri, inode_contribution_t, ret);
+    if (ret == -1)
+        goto out;
+
+    GF_REF_INIT(contri, mq_contri_fini);
+
+    contri->contribution = 0;
+    contri->file_count = 0;
+    contri->dir_count = 0;
+    gf_uuid_copy(contri->gfid, inode->gfid);
+
+    LOCK_INIT(&contri->lock);
+    INIT_LIST_HEAD(&contri->contri_list);
+
+out:
+    return contri;
+}
+
+inode_contribution_t *
+mq_get_contribution_node(inode_t *inode, quota_inode_ctx_t *ctx)
+{
+    inode_contribution_t *contri = NULL;
+    inode_contribution_t *temp = NULL;
+
+    if (!inode || !ctx)
+        goto out;
+
+    LOCK(&ctx->lock);
+    {
+        if (list_empty(&ctx->contribution_head))
+            goto unlock;
+
+        list_for_each_entry(temp, &ctx->contribution_head, contri_list)
+        {
+            if (gf_uuid_compare(temp->gfid, inode->gfid) == 0) {
+                contri = temp;
+                GF_REF_GET(contri);
+                break;
+            }
+        }
+    }
+unlock:
+    UNLOCK(&ctx->lock);
+
+out:
+    return contri;
+}
+
+inode_contribution_t *
+__mq_add_new_contribution_node(xlator_t *this, quota_inode_ctx_t *ctx,
+                               loc_t *loc)
+{
+    inode_contribution_t *contribution = NULL;
+
+    if (!loc->parent) {
+        if (!gf_uuid_is_null(loc->pargfid))
+            loc->parent = inode_find(loc->inode->table, loc->pargfid);
+
+        if (!loc->parent)
+            loc->parent = inode_parent(loc->inode, loc->pargfid, loc->name);
+        if (!loc->parent)
+            goto out;
+    }
+
+    list_for_each_entry(contribution, &ctx->contribution_head, contri_list)
+    {
+        if (loc->parent &&
+            gf_uuid_compare(contribution->gfid, loc->parent->gfid) == 0) {
+            goto out;
+        }
+    }
+
+    contribution = mq_contri_init(loc->parent);
+    if (contribution == NULL)
+        goto out;
+
+    list_add_tail(&contribution->contri_list, &ctx->contribution_head);
+
+out:
+    return contribution;
+}
+
+inode_contribution_t *
+mq_add_new_contribution_node(xlator_t *this, quota_inode_ctx_t *ctx, loc_t *loc)
+{
+    inode_contribution_t *contribution = NULL;
+
+    if ((ctx == NULL) || (loc == NULL))
+        return NULL;
+
+    if (((loc->path) && (strcmp(loc->path, "/") == 0)) ||
+        (!loc->path && gf_uuid_is_null(loc->pargfid)))
+        return NULL;
+
+    LOCK(&ctx->lock);
+    {
+        contribution = __mq_add_new_contribution_node(this, ctx, loc);
+        if (contribution)
+            GF_REF_GET(contribution);
+    }
+    UNLOCK(&ctx->lock);
+
+    return contribution;
+}
+
+int32_t
+mq_dict_set_contribution(xlator_t *this, dict_t *dict, loc_t *loc, uuid_t gfid,
+                         char *contri_key)
+{
+    int32_t ret = -1;
+    char key[QUOTA_KEY_MAX] = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("marker", this, out);
+    GF_VALIDATE_OR_GOTO("marker", dict, out);
+    GF_VALIDATE_OR_GOTO("marker", loc, out);
+
+    if (gfid && !gf_uuid_is_null(gfid)) {
+        GET_CONTRI_KEY(this, key, gfid, ret);
+    } else if (loc->parent) {
+        GET_CONTRI_KEY(this, key, loc->parent->gfid, ret);
+    } else {
+        /* nameless lookup, fetch contributions to all parents */
+        GET_CONTRI_KEY(this, key, NULL, ret);
+    }
+
+    if (ret < 0)
+        goto out;
+
+    ret = dict_set_int64(dict, key, 0);
+    if (ret < 0)
+        goto out;
+
+    if (contri_key)
+        if (snprintf(contri_key, QUOTA_KEY_MAX, "%s", key) >= QUOTA_KEY_MAX) {
+            ret = -1;
+            goto out;
+        }
+
+out:
+    if (ret < 0)
+        gf_log_callingfn(this ? this->name : "Marker", GF_LOG_ERROR,
+                         "dict set failed");
+
+    return ret;
+}
+
+int32_t
+mq_inode_ctx_get(inode_t *inode, xlator_t *this, quota_inode_ctx_t **ctx)
+{
+    int32_t ret = -1;
+    uint64_t ctx_int = 0;
+    marker_inode_ctx_t *mark_ctx = NULL;
+
+    GF_VALIDATE_OR_GOTO("marker", inode, out);
+    GF_VALIDATE_OR_GOTO("marker", this, out);
+    GF_VALIDATE_OR_GOTO("marker", ctx, out);
+
+    ret = inode_ctx_get(inode, this, &ctx_int);
+    if (ret < 0) {
+        ret = -1;
+        *ctx = NULL;
+        goto out;
+    }
+
+    mark_ctx = (marker_inode_ctx_t *)(unsigned long)ctx_int;
+    if (mark_ctx->quota_ctx == NULL) {
+        ret = -1;
+        goto out;
+    }
+
+    *ctx = mark_ctx->quota_ctx;
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+quota_inode_ctx_t *
+__mq_inode_ctx_new(inode_t *inode, xlator_t *this)
+{
+    int32_t ret = -1;
+    quota_inode_ctx_t *quota_ctx = NULL;
+    marker_inode_ctx_t *mark_ctx = NULL;
+
+    ret = marker_force_inode_ctx_get(inode, this, &mark_ctx);
+    if (ret < 0) {
+        gf_log(this->name, GF_LOG_ERROR, "marker_force_inode_ctx_get() failed");
+        goto out;
+    }
+
+    LOCK(&inode->lock);
+    {
+        if (mark_ctx->quota_ctx == NULL) {
+            quota_ctx = mq_alloc_inode_ctx();
+            if (quota_ctx == NULL) {
+                ret = -1;
+                goto unlock;
+            }
+            mark_ctx->quota_ctx = quota_ctx;
+        } else {
+            quota_ctx = mark_ctx->quota_ctx;
+        }
+
+        ret = 0;
+    }
+unlock:
+    UNLOCK(&inode->lock);
+out:
+    return quota_ctx;
+}
+
+quota_inode_ctx_t *
+mq_inode_ctx_new(inode_t *inode, xlator_t *this)
+{
+    return __mq_inode_ctx_new(inode, this);
+}
diff --git a/xlators/features/marker/src/marker-quota-helper.h b/xlators/features/marker/src/marker-quota-helper.h
new file mode 100644
index 00000000000..d4091dd2180
--- /dev/null
+++ b/xlators/features/marker/src/marker-quota-helper.h
@@ -0,0 +1,66 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef _MARKER_QUOTA_HELPER_H
+#define _MARKER_QUOTA_HELPER_H
+
+#include "marker.h"
+
+#define QUOTA_FREE_CONTRIBUTION_NODE(ctx, _contribution)                       \
+    do {                                                                       \
+        LOCK(&ctx->lock);                                                      \
+        {                                                                      \
+            list_del_init(&_contribution->contri_list);                        \
+            GF_REF_PUT(_contribution);                                         \
+        }                                                                      \
+        UNLOCK(&ctx->lock);                                                    \
+    } while (0)
+
+#define QUOTA_SAFE_INCREMENT(lock, var)                                        \
+    do {                                                                       \
+        LOCK(lock);                                                            \
+        var++;                                                                 \
+        UNLOCK(lock);                                                          \
+    } while (0)
+
+#define QUOTA_SAFE_DECREMENT(lock, var, value)                                 \
+    do {                                                                       \
+        LOCK(lock);                                                            \
+        {                                                                      \
+            value = --var;                                                     \
+        }                                                                      \
+        UNLOCK(lock);                                                          \
+    } while (0)
+
+inode_contribution_t *
+mq_add_new_contribution_node(xlator_t *, quota_inode_ctx_t *, loc_t *);
+
+int32_t
+mq_dict_set_contribution(xlator_t *, dict_t *, loc_t *, uuid_t, char *);
+
+quota_inode_ctx_t *
+mq_inode_ctx_new(inode_t *, xlator_t *);
+
+int32_t
+mq_inode_ctx_get(inode_t *, xlator_t *, quota_inode_ctx_t **);
+
+int32_t
+mq_delete_contribution_node(dict_t *, char *, inode_contribution_t *);
+
+int32_t
+mq_inode_loc_fill(const char *, inode_t *, loc_t *);
+
+inode_contribution_t *
+mq_contri_init(inode_t *inode);
+
+inode_contribution_t *
+mq_get_contribution_node(inode_t *, quota_inode_ctx_t *);
+
+#endif
diff --git a/xlators/features/marker/src/marker-quota.c b/xlators/features/marker/src/marker-quota.c
new file mode 100644
index 00000000000..3de2ea1c92c
--- /dev/null
+++ b/xlators/features/marker/src/marker-quota.c
@@ -0,0 +1,2297 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include <glusterfs/dict.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+#include "libxlator.h"
+#include <glusterfs/common-utils.h>
+#include <glusterfs/byte-order.h>
+#include "marker-quota.h"
+#include "marker-quota-helper.h"
+#include <glusterfs/syncop.h>
+#include <glusterfs/quota-common-utils.h>
+
+int
+mq_loc_copy(loc_t *dst, loc_t *src)
+{
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("marker", dst, out);
+    GF_VALIDATE_OR_GOTO("marker", src, out);
+
+    if (src->inode == NULL ||
+        ((src->parent == NULL) && (gf_uuid_is_null(src->pargfid)) &&
+         !__is_root_gfid(src->inode->gfid))) {
+        gf_log("marker", GF_LOG_WARNING, "src loc is not valid");
+        goto out;
+    }
+
+    ret = loc_copy(dst, src);
+out:
+    return ret;
+}
+
+static void
+mq_set_ctx_status(quota_inode_ctx_t *ctx, gf_boolean_t *flag,
+                  gf_boolean_t status)
+{
+    LOCK(&ctx->lock);
+    {
+        *flag = status;
+    }
+    UNLOCK(&ctx->lock);
+}
+
+static void
+mq_test_and_set_ctx_status(quota_inode_ctx_t *ctx, gf_boolean_t *flag,
+                           gf_boolean_t *status)
+{
+    gf_boolean_t temp = _gf_false;
+
+    LOCK(&ctx->lock);
+    {
+        temp = *status;
+        *status = *flag;
+        *flag = temp;
+    }
+    UNLOCK(&ctx->lock);
+}
+
+static void
+mq_get_ctx_status(quota_inode_ctx_t *ctx, gf_boolean_t *flag,
+                  gf_boolean_t *status)
+{
+    LOCK(&ctx->lock);
+    {
+        *status = *flag;
+    }
+    UNLOCK(&ctx->lock);
+}
+
+int32_t
+mq_get_ctx_updation_status(quota_inode_ctx_t *ctx, gf_boolean_t *status)
+{
+    GF_VALIDATE_OR_GOTO("marker", ctx, out);
+    GF_VALIDATE_OR_GOTO("marker", status, out);
+
+    mq_get_ctx_status(ctx, &ctx->updation_status, status);
+    return 0;
+out:
+    return -1;
+}
+
+int32_t
+mq_set_ctx_updation_status(quota_inode_ctx_t *ctx, gf_boolean_t status)
+{
+    GF_VALIDATE_OR_GOTO("marker", ctx, out);
+
+    mq_set_ctx_status(ctx, &ctx->updation_status, status);
+    return 0;
+out:
+    return -1;
+}
+
+int32_t
+mq_test_and_set_ctx_updation_status(quota_inode_ctx_t *ctx,
+                                    gf_boolean_t *status)
+{
+    GF_VALIDATE_OR_GOTO("marker", ctx, out);
+    GF_VALIDATE_OR_GOTO("marker", status, out);
+
+    mq_test_and_set_ctx_status(ctx, &ctx->updation_status, status);
+    return 0;
+out:
+    return -1;
+}
+
+int32_t
+mq_set_ctx_create_status(quota_inode_ctx_t *ctx, gf_boolean_t status)
+{
+    GF_VALIDATE_OR_GOTO("marker", ctx, out);
+
+    mq_set_ctx_status(ctx, &ctx->create_status, status);
+    return 0;
+out:
+    return -1;
+}
+
+int32_t
+mq_test_and_set_ctx_create_status(quota_inode_ctx_t *ctx, gf_boolean_t *status)
+{
+    GF_VALIDATE_OR_GOTO("marker", ctx, out);
+    GF_VALIDATE_OR_GOTO("marker", status, out);
+
+    mq_test_and_set_ctx_status(ctx, &ctx->create_status, status);
+    return 0;
+out:
+    return -1;
+}
+
+static void
+mq_set_ctx_dirty_status(quota_inode_ctx_t *ctx, gf_boolean_t status)
+{
+    GF_VALIDATE_OR_GOTO("marker", ctx, out);
+
+    mq_set_ctx_status(ctx, &ctx->dirty_status, status);
+out:
+    return;
+}
+
+int
+mq_build_ancestry(xlator_t *this, loc_t *loc)
+{
+    int32_t ret = -1;
+    fd_t *fd = NULL;
+    gf_dirent_t entries;
+    gf_dirent_t *entry = NULL;
+    dict_t *xdata = NULL;
+    inode_t *tmp_parent = NULL;
+    inode_t *tmp_inode = NULL;
+    inode_t *linked_inode = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+
+    INIT_LIST_HEAD(&entries.list);
+
+    xdata = dict_new();
+    if (xdata == NULL) {
+        gf_log(this->name, GF_LOG_ERROR, "dict_new failed");
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ret = dict_set_int8(xdata, GET_ANCESTRY_DENTRY_KEY, 1);
+    if (ret < 0)
+        goto out;
+
+    fd = fd_anonymous(loc->inode);
+    if (fd == NULL) {
+        gf_log(this->name, GF_LOG_ERROR, "fd creation failed");
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    fd_bind(fd);
+
+    ret = syncop_readdirp(this, fd, 131072, 0, &entries, xdata, NULL);
+    if (ret < 0) {
+        gf_log(this->name,
+               (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR,
+               "readdirp failed "
+               "for %s: %s",
+               loc->path, strerror(-ret));
+        goto out;
+    }
+
+    if (list_empty(&entries.list)) {
+        ret = -1;
+        goto out;
+    }
+
+    list_for_each_entry(entry, &entries.list, list)
+    {
+        if (__is_root_gfid(entry->inode->gfid)) {
+            /* The list contains a sub-list for each possible path
+             * to the target inode. Each sub-list starts with the
+             * root entry of the tree and is followed by the child
+             * entries for a particular path to the target entry.
+             * The root entry is an implied sub-list delimiter,
+             * as it denotes we have started processing a new path.
+             * Reset the parent pointer and continue
+             */
+
+            tmp_parent = NULL;
+        } else {
+            linked_inode = inode_link(entry->inode, tmp_parent, entry->d_name,
+                                      &entry->d_stat);
+            if (linked_inode) {
+                tmp_inode = entry->inode;
+                entry->inode = linked_inode;
+                inode_unref(tmp_inode);
+            } else {
+                gf_log(this->name, GF_LOG_ERROR, "inode link failed");
+                ret = -EINVAL;
+                goto out;
+            }
+        }
+
+        ctx = mq_inode_ctx_new(entry->inode, this);
+        if (ctx == NULL) {
+            gf_log(this->name, GF_LOG_WARNING,
+                   "mq_inode_ctx_new "
+                   "failed for %s",
+                   uuid_utoa(entry->inode->gfid));
+            ret = -ENOMEM;
+            goto out;
+        }
+
+        /* For non-directory, posix_get_ancestry_non_directory returns
+         * all hard-links that are represented by nodes adjacent to
+         * each other in the dentry-list.
+         * (Unlike the directory case where adjacent nodes either have
+         *  a parent/child relationship or belong to different paths).
+         */
+        if (entry->inode->ia_type == IA_IFDIR)
+            tmp_parent = entry->inode;
+    }
+
+    if (loc->parent)
+        inode_unref(loc->parent);
+
+    loc->parent = inode_parent(loc->inode, 0, NULL);
+    if (loc->parent == NULL) {
+        ret = -1;
+        goto out;
+    }
+
+    ret = 0;
+
+out:
+    gf_dirent_free(&entries);
+
+    if (fd)
+        fd_unref(fd);
+
+    if (xdata)
+        dict_unref(xdata);
+
+    return ret;
+}
+
+/* This function should be used only in inspect_directory and inspect_file
+ * function to heal quota xattrs.
+ * Inode quota feature is introduced in 3.7.
+ * If gluster setup is upgraded from 3.6 to 3.7, there can be a
+ * getxattr and setxattr spikes with quota heal as inode quota is missing.
+ * So this wrapper function is to avoid xattrs spikes during upgrade.
+ * This function returns success even is inode-quota xattrs are missing and
+ * hence no healing performed.
+ */
+static int32_t
+_quota_dict_get_meta(xlator_t *this, dict_t *dict, char *key, const int keylen,
+                     quota_meta_t *meta, ia_type_t ia_type,
+                     gf_boolean_t add_delta)
+{
+    int32_t ret = 0;
+    marker_conf_t *priv = NULL;
+
+    priv = this->private;
+
+    ret = quota_dict_get_inode_meta(dict, key, keylen, meta);
+    if (ret == -2 && (priv->feature_enabled & GF_INODE_QUOTA) == 0) {
+        /* quota_dict_get_inode_meta returns -2 if
+         * inode quota xattrs are not present.
+         * if inode quota self heal is turned off,
+         * then we should skip healing inode quotas
+         */
+
+        gf_log(this->name, GF_LOG_DEBUG,
+               "inode quota disabled. "
+               "inode quota self heal will not be performed");
+        ret = 0;
+        if (add_delta) {
+            if (ia_type == IA_IFDIR)
+                meta->dir_count = 1;
+            else
+                meta->file_count = 1;
+        }
+    }
+
+    return ret;
+}
+
+int32_t
+quota_dict_set_size_meta(xlator_t *this, dict_t *dict, const quota_meta_t *meta)
+{
+    int32_t ret = -ENOMEM;
+    quota_meta_t *value = NULL;
+    char size_key[QUOTA_KEY_MAX] = {
+        0,
+    };
+
+    value = GF_MALLOC(2 * sizeof(quota_meta_t), gf_common_quota_meta_t);
+    if (value == NULL) {
+        goto out;
+    }
+    value[0].size = hton64(meta->size);
+    value[0].file_count = hton64(meta->file_count);
+    value[0].dir_count = hton64(meta->dir_count);
+
+    value[1].size = 0;
+    value[1].file_count = 0;
+    value[1].dir_count = hton64(1);
+
+    GET_SIZE_KEY(this, size_key, ret);
+    if (ret < 0)
+        goto out;
+    ret = dict_set_bin(dict, size_key, value, (sizeof(quota_meta_t) * 2));
+    if (ret < 0) {
+        gf_log_callingfn("quota", GF_LOG_ERROR, "dict set failed");
+        GF_FREE(value);
+    }
+out:
+    return ret;
+}
+
+void
+mq_compute_delta(quota_meta_t *delta, const quota_meta_t *op1,
+                 const quota_meta_t *op2)
+{
+    delta->size = op1->size - op2->size;
+    delta->file_count = op1->file_count - op2->file_count;
+    delta->dir_count = op1->dir_count - op2->dir_count;
+}
+
+void
+mq_add_meta(quota_meta_t *dst, const quota_meta_t *src)
+{
+    dst->size += src->size;
+    dst->file_count += src->file_count;
+    dst->dir_count += src->dir_count;
+}
+
+void
+mq_sub_meta(quota_meta_t *dst, const quota_meta_t *src)
+{
+    if (src == NULL) {
+        dst->size = -dst->size;
+        dst->file_count = -dst->file_count;
+        dst->dir_count = -dst->dir_count;
+    } else {
+        dst->size = src->size - dst->size;
+        dst->file_count = src->file_count - dst->file_count;
+        dst->dir_count = src->dir_count - dst->dir_count;
+    }
+}
+
+int32_t
+mq_are_xattrs_set(xlator_t *this, loc_t *loc, gf_boolean_t *contri_set,
+                  gf_boolean_t *size_set)
+{
+    int32_t ret = -1;
+    char contri_key[QUOTA_KEY_MAX] = {
+        0,
+    };
+    char size_key[QUOTA_KEY_MAX] = {
+        0,
+    };
+    quota_meta_t meta = {
+        0,
+    };
+    struct iatt stbuf = {
+        0,
+    };
+    dict_t *dict = NULL;
+    dict_t *rsp_dict = NULL;
+
+    dict = dict_new();
+    if (dict == NULL) {
+        gf_log(this->name, GF_LOG_ERROR, "dict_new failed");
+        goto out;
+    }
+
+    ret = mq_req_xattr(this, loc, dict, contri_key, size_key);
+    if (ret < 0)
+        goto out;
+
+    ret = syncop_lookup(FIRST_CHILD(this), loc, &stbuf, NULL, dict, &rsp_dict);
+    if (ret < 0) {
+        gf_log_callingfn(
+            this->name,
+            (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR,
+            "lookup failed "
+            "for %s: %s",
+            loc->path, strerror(-ret));
+        goto out;
+    }
+
+    if (rsp_dict == NULL)
+        goto out;
+
+    *contri_set = _gf_true;
+    *size_set = _gf_true;
+    if (loc->inode->ia_type == IA_IFDIR) {
+        ret = quota_dict_get_inode_meta(rsp_dict, size_key, strlen(size_key),
+                                        &meta);
+        if (ret < 0 || meta.dir_count == 0)
+            *size_set = _gf_false;
+    }
+
+    if (!loc_is_root(loc)) {
+        ret = quota_dict_get_inode_meta(rsp_dict, contri_key,
+                                        strlen(contri_key), &meta);
+        if (ret < 0)
+            *contri_set = _gf_false;
+    }
+
+    ret = 0;
+out:
+    if (dict)
+        dict_unref(dict);
+
+    if (rsp_dict)
+        dict_unref(rsp_dict);
+
+    return ret;
+}
+
+int32_t
+mq_create_size_xattrs(xlator_t *this, quota_inode_ctx_t *ctx, loc_t *loc)
+{
+    int32_t ret = -1;
+    quota_meta_t size = {
+        0,
+    };
+    dict_t *dict = NULL;
+
+    GF_VALIDATE_OR_GOTO("marker", loc, out);
+    GF_VALIDATE_OR_GOTO("marker", loc->inode, out);
+
+    if (loc->inode->ia_type != IA_IFDIR) {
+        ret = 0;
+        goto out;
+    }
+
+    dict = dict_new();
+    if (!dict) {
+        gf_log(this->name, GF_LOG_ERROR, "dict_new failed");
+        ret = -1;
+        goto out;
+    }
+
+    ret = quota_dict_set_size_meta(this, dict, &size);
+    if (ret < 0)
+        goto out;
+
+    ret = syncop_xattrop(FIRST_CHILD(this), loc,
+                         GF_XATTROP_ADD_ARRAY64_WITH_DEFAULT, dict, NULL, NULL,
+                         NULL);
+
+    if (ret < 0) {
+        gf_log_callingfn(
+            this->name,
+            (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR,
+            "xattrop failed "
+            "for %s: %s",
+            loc->path, strerror(-ret));
+        goto out;
+    }
+
+out:
+    if (dict)
+        dict_unref(dict);
+
+    return ret;
+}
+
+int32_t
+mq_lock(xlator_t *this, loc_t *loc, short l_type)
+{
+    struct gf_flock lock = {
+        0,
+    };
+    int32_t ret = -1;
+
+    GF_VALIDATE_OR_GOTO("marker", loc, out);
+    GF_VALIDATE_OR_GOTO("marker", loc->inode, out);
+
+    gf_log(this->name, GF_LOG_DEBUG, "set lock type %d on %s", l_type,
+           loc->path);
+
+    lock.l_len = 0;
+    lock.l_start = 0;
+    lock.l_type = l_type;
+    lock.l_whence = SEEK_SET;
+
+    ret = syncop_inodelk(FIRST_CHILD(this), this->name, loc, F_SETLKW, &lock,
+                         NULL, NULL);
+    if (ret < 0)
+        gf_log_callingfn(
+            this->name,
+            (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR,
+            "inodelk failed "
+            "for %s: %s",
+            loc->path, strerror(-ret));
+
+out:
+
+    return ret;
+}
+
+int32_t
+mq_get_dirty(xlator_t *this, loc_t *loc, int32_t *dirty)
+{
+    int32_t ret = -1;
+    int8_t value = 0;
+    dict_t *dict = NULL;
+    dict_t *rsp_dict = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+
+    dict = dict_new();
+    if (dict == NULL) {
+        gf_log(this->name, GF_LOG_ERROR, "dict_new failed");
+        goto out;
+    }
+
+    ret = dict_set_int64(dict, QUOTA_DIRTY_KEY, 0);
+    if (ret < 0) {
+        gf_log(this->name, GF_LOG_WARNING, "dict set failed");
+        goto out;
+    }
+
+    ret = syncop_lookup(FIRST_CHILD(this), loc, &stbuf, NULL, dict, &rsp_dict);
+    if (ret < 0) {
+        gf_log_callingfn(
+            this->name,
+            (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR,
+            "lookup failed "
+            "for %s: %s",
+            loc->path, strerror(-ret));
+        goto out;
+    }
+
+    ret = dict_get_int8(rsp_dict, QUOTA_DIRTY_KEY, &value);
+    if (ret < 0)
+        goto out;
+
+    *dirty = value;
+
+out:
+    if (dict)
+        dict_unref(dict);
+
+    if (rsp_dict)
+        dict_unref(rsp_dict);
+
+    return ret;
+}
+
+int32_t
+mq_get_set_dirty(xlator_t *this, loc_t *loc, int32_t dirty, int32_t *prev_dirty)
+{
+    int32_t ret = -1;
+    int8_t value = 0;
+    quota_inode_ctx_t *ctx = NULL;
+    dict_t *dict = NULL;
+    dict_t *rsp_dict = NULL;
+
+    GF_VALIDATE_OR_GOTO("marker", loc, out);
+    GF_VALIDATE_OR_GOTO("marker", loc->inode, out);
+    GF_VALIDATE_OR_GOTO("marker", prev_dirty, out);
+
+    ret = mq_inode_ctx_get(loc->inode, this, &ctx);
+    if (ret < 0) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to get inode ctx for "
+               "%s",
+               loc->path);
+        goto out;
+    }
+
+    dict = dict_new();
+    if (!dict) {
+        gf_log(this->name, GF_LOG_ERROR, "dict_new failed");
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_set_int8(dict, QUOTA_DIRTY_KEY, dirty);
+    if (ret < 0) {
+        gf_log(this->name, GF_LOG_ERROR, "dict_set failed");
+        goto out;
+    }
+
+    ret = syncop_xattrop(FIRST_CHILD(this), loc, GF_XATTROP_GET_AND_SET, dict,
+                         NULL, NULL, &rsp_dict);
+    if (ret < 0) {
+        gf_log_callingfn(
+            this->name,
+            (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR,
+            "xattrop failed "
+            "for %s: %s",
+            loc->path, strerror(-ret));
+        goto out;
+    }
+
+    *prev_dirty = 0;
+    if (rsp_dict) {
+        ret = dict_get_int8(rsp_dict, QUOTA_DIRTY_KEY, &value);
+        if (ret == 0)
+            *prev_dirty = value;
+    }
+
+    LOCK(&ctx->lock);
+    {
+        ctx->dirty = dirty;
+    }
+    UNLOCK(&ctx->lock);
+    ret = 0;
+out:
+    if (dict)
+        dict_unref(dict);
+
+    if (rsp_dict)
+        dict_unref(rsp_dict);
+
+    return ret;
+}
+
+int32_t
+mq_mark_dirty(xlator_t *this, loc_t *loc, int32_t dirty)
+{
+    int32_t ret = -1;
+    dict_t *dict = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+
+    GF_VALIDATE_OR_GOTO("marker", loc, out);
+    GF_VALIDATE_OR_GOTO("marker", loc->inode, out);
+
+    ret = mq_inode_ctx_get(loc->inode, this, &ctx);
+    if (ret < 0) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to get inode ctx for "
+               "%s",
+               loc->path);
+        ret = 0;
+        goto out;
+    }
+
+    dict = dict_new();
+    if (!dict) {
+        ret = -1;
+        gf_log(this->name, GF_LOG_ERROR, "dict_new failed");
+        goto out;
+    }
+
+    ret = dict_set_int8(dict, QUOTA_DIRTY_KEY, dirty);
+    if (ret < 0) {
+        gf_log(this->name, GF_LOG_ERROR, "dict_set failed");
+        goto out;
+    }
+
+    ret = syncop_setxattr(FIRST_CHILD(this), loc, dict, 0, NULL, NULL);
+    if (ret < 0) {
+        gf_log_callingfn(
+            this->name,
+            (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR,
+            "setxattr dirty = %d "
+            "failed for %s: %s",
+            dirty, loc->path, strerror(-ret));
+        goto out;
+    }
+
+    LOCK(&ctx->lock);
+    {
+        ctx->dirty = dirty;
+    }
+    UNLOCK(&ctx->lock);
+
+out:
+    if (dict)
+        dict_unref(dict);
+
+    return ret;
+}
+
+int32_t
+_mq_get_metadata(xlator_t *this, loc_t *loc, quota_meta_t *contri,
+                 quota_meta_t *size, uuid_t contri_gfid)
+{
+    int32_t ret = -1;
+    quota_meta_t meta = {
+        0,
+    };
+    char contri_key[QUOTA_KEY_MAX] = {
+        0,
+    };
+    char size_key[QUOTA_KEY_MAX] = {
+        0,
+    };
+    int keylen = 0;
+    dict_t *dict = NULL;
+    dict_t *rsp_dict = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("marker", loc, out);
+    GF_VALIDATE_OR_GOTO("marker", loc->inode, out);
+
+    if (size == NULL && contri == NULL)
+        goto out;
+
+    dict = dict_new();
+    if (dict == NULL) {
+        gf_log(this->name, GF_LOG_ERROR, "dict_new failed");
+        goto out;
+    }
+
+    if (size && loc->inode->ia_type == IA_IFDIR) {
+        GET_SIZE_KEY(this, size_key, keylen);
+        if (keylen < 0)
+            goto out;
+        ret = dict_set_int64(dict, size_key, 0);
+        if (ret < 0) {
+            gf_log(this->name, GF_LOG_ERROR, "dict_set failed.");
+            goto out;
+        }
+    }
+
+    if (contri && !loc_is_root(loc)) {
+        ret = mq_dict_set_contribution(this, dict, loc, contri_gfid,
+                                       contri_key);
+        if (ret < 0)
+            goto out;
+    }
+
+    ret = syncop_lookup(FIRST_CHILD(this), loc, &stbuf, NULL, dict, &rsp_dict);
+    if (ret < 0) {
+        gf_log_callingfn(
+            this->name,
+            (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR,
+            "lookup failed "
+            "for %s: %s",
+            loc->path, strerror(-ret));
+        goto out;
+    }
+
+    if (size) {
+        if (loc->inode->ia_type == IA_IFDIR) {
+            ret = quota_dict_get_meta(rsp_dict, size_key, keylen, &meta);
+            if (ret < 0) {
+                gf_log(this->name, GF_LOG_ERROR, "dict_get failed.");
+                goto out;
+            }
+
+            size->size = meta.size;
+            size->file_count = meta.file_count;
+            size->dir_count = meta.dir_count;
+        } else {
+            size->size = stbuf.ia_blocks * 512;
+            size->file_count = 1;
+            size->dir_count = 0;
+        }
+    }
+
+    if (contri && !loc_is_root(loc)) {
+        ret = quota_dict_get_meta(rsp_dict, contri_key, strlen(contri_key),
+                                  &meta);
+        if (ret < 0) {
+            contri->size = 0;
+            contri->file_count = 0;
+            contri->dir_count = 0;
+        } else {
+            contri->size = meta.size;
+            contri->file_count = meta.file_count;
+            contri->dir_count = meta.dir_count;
+        }
+    }
+
+    ret = 0;
+
+out:
+    if (dict)
+        dict_unref(dict);
+
+    if (rsp_dict)
+        dict_unref(rsp_dict);
+
+    return ret;
+}
+
+int32_t
+mq_get_metadata(xlator_t *this, loc_t *loc, quota_meta_t *contri,
+                quota_meta_t *size, quota_inode_ctx_t *ctx,
+                inode_contribution_t *contribution)
+{
+    int32_t ret = -1;
+
+    GF_VALIDATE_OR_GOTO("marker", loc, out);
+    GF_VALIDATE_OR_GOTO("marker", loc->inode, out);
+    GF_VALIDATE_OR_GOTO("marker", ctx, out);
+    GF_VALIDATE_OR_GOTO("marker", contribution, out);
+
+    if (size == NULL && contri == NULL) {
+        ret = 0;
+        goto out;
+    }
+
+    ret = _mq_get_metadata(this, loc, contri, size, contribution->gfid);
+    if (ret < 0)
+        goto out;
+
+    if (size) {
+        LOCK(&ctx->lock);
+        {
+            ctx->size = size->size;
+            ctx->file_count = size->file_count;
+            ctx->dir_count = size->dir_count;
+        }
+        UNLOCK(&ctx->lock);
+    }
+
+    if (contri) {
+        LOCK(&contribution->lock);
+        {
+            contribution->contribution = contri->size;
+            contribution->file_count = contri->file_count;
+            contribution->dir_count = contri->dir_count;
+        }
+        UNLOCK(&contribution->lock);
+    }
+
+out:
+    return ret;
+}
+
+int32_t
+mq_get_delta(xlator_t *this, loc_t *loc, quota_meta_t *delta,
+             quota_inode_ctx_t *ctx, inode_contribution_t *contribution)
+{
+    int32_t ret = -1;
+    quota_meta_t size = {
+        0,
+    };
+    quota_meta_t contri = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("marker", loc, out);
+    GF_VALIDATE_OR_GOTO("marker", loc->inode, out);
+    GF_VALIDATE_OR_GOTO("marker", ctx, out);
+    GF_VALIDATE_OR_GOTO("marker", contribution, out);
+
+    ret = mq_get_metadata(this, loc, &contri, &size, ctx, contribution);
+    if (ret < 0)
+        goto out;
+
+    mq_compute_delta(delta, &size, &contri);
+
+out:
+    return ret;
+}
+
+int32_t
+mq_remove_contri(xlator_t *this, loc_t *loc, quota_inode_ctx_t *ctx,
+                 inode_contribution_t *contri, quota_meta_t *delta,
+                 uint32_t nlink)
+{
+    int32_t ret = -1;
+    char contri_key[QUOTA_KEY_MAX] = {
+        0,
+    };
+
+    if (nlink == 1) {
+        /*File was a last link and has been deleted */
+        ret = 0;
+        goto done;
+    }
+
+    GET_CONTRI_KEY(this, contri_key, contri->gfid, ret);
+    if (ret < 0) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "get contri_key "
+               "failed for %s",
+               uuid_utoa(contri->gfid));
+        goto out;
+    }
+
+    ret = syncop_removexattr(FIRST_CHILD(this), loc, contri_key, 0, NULL);
+    if (ret < 0) {
+        if (-ret == ENOENT || -ret == ESTALE || -ret == ENODATA ||
+            -ret == ENOATTR) {
+            /* Remove contri in done when unlink operation is
+             * performed, so return success on ENOENT/ESTSLE
+             * rename operation removes xattr earlier,
+             * so return success on ENODATA
+             */
+            ret = 0;
+        } else {
+            gf_log_callingfn(this->name, GF_LOG_ERROR,
+                             "removexattr %s failed for %s: %s", contri_key,
+                             loc->path, strerror(-ret));
+            goto out;
+        }
+    }
+
+done:
+    LOCK(&contri->lock);
+    {
+        contri->contribution += delta->size;
+        contri->file_count += delta->file_count;
+        contri->dir_count += delta->dir_count;
+    }
+    UNLOCK(&contri->lock);
+
+    ret = 0;
+
+out:
+    QUOTA_FREE_CONTRIBUTION_NODE(ctx, contri);
+
+    return ret;
+}
+
+int32_t
+mq_update_contri(xlator_t *this, loc_t *loc, inode_contribution_t *contri,
+                 quota_meta_t *delta)
+{
+    int32_t ret = -1;
+    char contri_key[QUOTA_KEY_MAX] = {
+        0,
+    };
+    dict_t *dict = NULL;
+
+    GF_VALIDATE_OR_GOTO("marker", loc, out);
+    GF_VALIDATE_OR_GOTO("marker", loc->inode, out);
+    GF_VALIDATE_OR_GOTO("marker", delta, out);
+    GF_VALIDATE_OR_GOTO("marker", contri, out);
+
+    if (quota_meta_is_null(delta)) {
+        ret = 0;
+        goto out;
+    }
+
+    dict = dict_new();
+    if (!dict) {
+        gf_log(this->name, GF_LOG_ERROR, "dict_new failed");
+        ret = -1;
+        goto out;
+    }
+
+    GET_CONTRI_KEY(this, contri_key, contri->gfid, ret);
+    if (ret < 0) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "get contri_key "
+               "failed for %s",
+               uuid_utoa(contri->gfid));
+        goto out;
+    }
+
+    ret = quota_dict_set_meta(dict, contri_key, delta, loc->inode->ia_type);
+    if (ret < 0)
+        goto out;
+
+    ret = syncop_xattrop(FIRST_CHILD(this), loc, GF_XATTROP_ADD_ARRAY64, dict,
+                         NULL, NULL, NULL);
+    if (ret < 0) {
+        gf_log_callingfn(
+            this->name,
+            (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR,
+            "xattrop failed "
+            "for %s: %s",
+            loc->path, strerror(-ret));
+        goto out;
+    }
+
+    LOCK(&contri->lock);
+    {
+        contri->contribution += delta->size;
+        contri->file_count += delta->file_count;
+        contri->dir_count += delta->dir_count;
+    }
+    UNLOCK(&contri->lock);
+
+out:
+    if (dict)
+        dict_unref(dict);
+
+    return ret;
+}
+
+int32_t
+mq_update_size(xlator_t *this, loc_t *loc, quota_meta_t *delta)
+{
+    int32_t ret = -1;
+    quota_inode_ctx_t *ctx = NULL;
+    dict_t *dict = NULL;
+
+    GF_VALIDATE_OR_GOTO("marker", loc, out);
+    GF_VALIDATE_OR_GOTO("marker", loc->inode, out);
+    GF_VALIDATE_OR_GOTO("marker", delta, out);
+
+    if (quota_meta_is_null(delta)) {
+        ret = 0;
+        goto out;
+    }
+
+    ret = mq_inode_ctx_get(loc->inode, this, &ctx);
+    if (ret < 0) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to get inode ctx for "
+               "%s",
+               loc->path);
+        goto out;
+    }
+
+    dict = dict_new();
+    if (!dict) {
+        gf_log(this->name, GF_LOG_ERROR, "dict_new failed");
+        ret = -1;
+        goto out;
+    }
+
+    ret = quota_dict_set_size_meta(this, dict, delta);
+    if (ret < 0)
+        goto out;
+
+    ret = syncop_xattrop(FIRST_CHILD(this), loc,
+                         GF_XATTROP_ADD_ARRAY64_WITH_DEFAULT, dict, NULL, NULL,
+                         NULL);
+    if (ret < 0) {
+        gf_log_callingfn(
+            this->name,
+            (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR,
+            "xattrop failed "
+            "for %s: %s",
+            loc->path, strerror(-ret));
+        goto out;
+    }
+
+    LOCK(&ctx->lock);
+    {
+        ctx->size += delta->size;
+        ctx->file_count += delta->file_count;
+        if (ctx->dir_count == 0)
+            ctx->dir_count += delta->dir_count + 1;
+        else
+            ctx->dir_count += delta->dir_count;
+    }
+    UNLOCK(&ctx->lock);
+
+out:
+    if (dict)
+        dict_unref(dict);
+
+    return ret;
+}
+
+int
+mq_synctask_cleanup(int ret, call_frame_t *frame, void *opaque)
+{
+    quota_synctask_t *args = NULL;
+
+    GF_ASSERT(opaque);
+
+    args = (quota_synctask_t *)opaque;
+    loc_wipe(&args->loc);
+
+    if (args->stub)
+        call_resume(args->stub);
+
+    if (!args->is_static)
+        GF_FREE(args);
+
+    return 0;
+}
+
+int
+mq_synctask1(xlator_t *this, synctask_fn_t task, gf_boolean_t spawn, loc_t *loc,
+             quota_meta_t *contri, uint32_t nlink, call_stub_t *stub)
+{
+    int32_t ret = -1;
+    quota_synctask_t *args = NULL;
+    quota_synctask_t static_args = {
+        0,
+    };
+
+    if (spawn) {
+        QUOTA_ALLOC_OR_GOTO(args, quota_synctask_t, ret, out);
+        args->is_static = _gf_false;
+    } else {
+        args = &static_args;
+        args->is_static = _gf_true;
+    }
+
+    args->this = this;
+    args->stub = stub;
+    loc_copy(&args->loc, loc);
+    args->ia_nlink = nlink;
+
+    if (contri) {
+        args->contri = *contri;
+    } else {
+        args->contri.size = -1;
+        args->contri.file_count = -1;
+        args->contri.dir_count = -1;
+    }
+
+    if (spawn) {
+        ret = synctask_new1(this->ctx->env, 1024 * 16, task,
+                            mq_synctask_cleanup, NULL, args);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "Failed to spawn "
+                   "new synctask");
+            mq_synctask_cleanup(ret, NULL, args);
+        }
+    } else {
+        ret = task(args);
+        mq_synctask_cleanup(ret, NULL, args);
+    }
+
+out:
+    return ret;
+}
+
+int
+mq_synctask(xlator_t *this, synctask_fn_t task, gf_boolean_t spawn, loc_t *loc)
+{
+    return mq_synctask1(this, task, spawn, loc, NULL, -1, NULL);
+}
+
+int32_t
+mq_prevalidate_txn(xlator_t *this, loc_t *origin_loc, loc_t *loc,
+                   quota_inode_ctx_t **ctx, struct iatt *buf)
+{
+    int32_t ret = -1;
+    quota_inode_ctx_t *ctxtmp = NULL;
+
+    if (buf) {
+        if (buf->ia_type == IA_IFREG && IS_DHT_LINKFILE_MODE(buf))
+            goto out;
+
+        if (buf->ia_type != IA_IFREG && buf->ia_type != IA_IFLNK &&
+            buf->ia_type != IA_IFDIR)
+            goto out;
+    }
+
+    if (origin_loc == NULL || origin_loc->inode == NULL ||
+        gf_uuid_is_null(origin_loc->inode->gfid))
+        goto out;
+
+    loc_copy(loc, origin_loc);
+
+    if (gf_uuid_is_null(loc->gfid))
+        gf_uuid_copy(loc->gfid, loc->inode->gfid);
+
+    if (!loc_is_root(loc) && loc->parent == NULL)
+        loc->parent = inode_parent(loc->inode, 0, NULL);
+
+    ret = mq_inode_ctx_get(loc->inode, this, &ctxtmp);
+    if (ret < 0) {
+        gf_log_callingfn(this->name, GF_LOG_WARNING,
+                         "inode ctx for "
+                         "is NULL for %s",
+                         loc->path);
+        goto out;
+    }
+    if (ctx)
+        *ctx = ctxtmp;
+
+    ret = 0;
+out:
+    return ret;
+}
+
+int
+mq_create_xattrs_task(void *opaque)
+{
+    int32_t ret = -1;
+    gf_boolean_t locked = _gf_false;
+    gf_boolean_t contri_set = _gf_false;
+    gf_boolean_t size_set = _gf_false;
+    gf_boolean_t need_txn = _gf_false;
+    quota_synctask_t *args = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    xlator_t *this = NULL;
+    loc_t *loc = NULL;
+    gf_boolean_t status = _gf_false;
+
+    GF_ASSERT(opaque);
+
+    args = (quota_synctask_t *)opaque;
+    loc = &args->loc;
+    this = args->this;
+    THIS = this;
+
+    ret = mq_inode_ctx_get(loc->inode, this, &ctx);
+    if (ret < 0) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "Failed to"
+               "get inode ctx, aborting quota create txn");
+        goto out;
+    }
+
+    if (loc->inode->ia_type == IA_IFDIR) {
+        /* lock not required for files */
+        ret = mq_lock(this, loc, F_WRLCK);
+        if (ret < 0)
+            goto out;
+        locked = _gf_true;
+    }
+
+    ret = mq_are_xattrs_set(this, loc, &contri_set, &size_set);
+    if (ret < 0 || (contri_set && size_set))
+        goto out;
+
+    mq_set_ctx_create_status(ctx, _gf_false);
+    status = _gf_true;
+
+    if (loc->inode->ia_type == IA_IFDIR && size_set == _gf_false) {
+        ret = mq_create_size_xattrs(this, ctx, loc);
+        if (ret < 0)
+            goto out;
+    }
+
+    need_txn = _gf_true;
+out:
+    if (locked)
+        ret = mq_lock(this, loc, F_UNLCK);
+
+    if (status == _gf_false)
+        mq_set_ctx_create_status(ctx, _gf_false);
+
+    if (need_txn)
+        ret = mq_initiate_quota_blocking_txn(this, loc, NULL);
+
+    return ret;
+}
+
+static int
+_mq_create_xattrs_txn(xlator_t *this, loc_t *origin_loc, struct iatt *buf,
+                      gf_boolean_t spawn)
+{
+    int32_t ret = -1;
+    quota_inode_ctx_t *ctx = NULL;
+    gf_boolean_t status = _gf_true;
+    loc_t loc = {
+        0,
+    };
+    inode_contribution_t *contribution = NULL;
+
+    ret = mq_prevalidate_txn(this, origin_loc, &loc, &ctx, buf);
+    if (ret < 0)
+        goto out;
+
+    ret = mq_test_and_set_ctx_create_status(ctx, &status);
+    if (ret < 0 || status == _gf_true)
+        goto out;
+
+    if (!loc_is_root(&loc) && loc.parent) {
+        contribution = mq_add_new_contribution_node(this, ctx, &loc);
+        if (contribution == NULL) {
+            gf_log(this->name, GF_LOG_WARNING,
+                   "cannot add a new contribution node "
+                   "(%s)",
+                   uuid_utoa(loc.gfid));
+            ret = -1;
+            goto out;
+        } else {
+            GF_REF_PUT(contribution);
+        }
+    }
+
+    ret = mq_synctask(this, mq_create_xattrs_task, spawn, &loc);
+out:
+    if (ret < 0 && status == _gf_false)
+        mq_set_ctx_create_status(ctx, _gf_false);
+
+    loc_wipe(&loc);
+    return ret;
+}
+
+int
+mq_create_xattrs_txn(xlator_t *this, loc_t *loc, struct iatt *buf)
+{
+    int32_t ret = -1;
+
+    GF_VALIDATE_OR_GOTO("marker", loc, out);
+    GF_VALIDATE_OR_GOTO("marker", loc->inode, out);
+
+    ret = _mq_create_xattrs_txn(this, loc, buf, _gf_true);
+out:
+    return ret;
+}
+
+int32_t
+mq_reduce_parent_size_task(void *opaque)
+{
+    int32_t ret = -1;
+    int32_t prev_dirty = 0;
+    quota_inode_ctx_t *ctx = NULL;
+    quota_inode_ctx_t *parent_ctx = NULL;
+    inode_contribution_t *contribution = NULL;
+    quota_meta_t delta = {
+        0,
+    };
+    quota_meta_t contri = {
+        0,
+    };
+    loc_t parent_loc = {
+        0,
+    };
+    gf_boolean_t locked = _gf_false;
+    gf_boolean_t dirty = _gf_false;
+    quota_synctask_t *args = NULL;
+    xlator_t *this = NULL;
+    loc_t *loc = NULL;
+    gf_boolean_t remove_xattr = _gf_true;
+    uint32_t nlink = 0;
+
+    GF_ASSERT(opaque);
+
+    args = (quota_synctask_t *)opaque;
+    loc = &args->loc;
+    contri = args->contri;
+    nlink = args->ia_nlink;
+    this = args->this;
+    THIS = this;
+
+    ret = mq_inode_loc_fill(NULL, loc->parent, &parent_loc);
+    if (ret < 0) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "parent_loc fill failed for "
+               "child inode %s: ",
+               uuid_utoa(loc->inode->gfid));
+        goto out;
+    }
+
+    ret = mq_lock(this, &parent_loc, F_WRLCK);
+    if (ret < 0)
+        goto out;
+    locked = _gf_true;
+
+    if (contri.size >= 0) {
+        /* contri parameter is supplied only for rename operation.
+         * remove xattr is alreday performed, we need to skip
+         * removexattr for rename operation
+         */
+        remove_xattr = _gf_false;
+        delta.size = contri.size;
+        delta.file_count = contri.file_count;
+        delta.dir_count = contri.dir_count;
+    } else {
+        remove_xattr = _gf_true;
+
+        ret = mq_inode_ctx_get(loc->inode, this, &ctx);
+        if (ret < 0) {
+            gf_log_callingfn(this->name, GF_LOG_WARNING,
+                             "ctx for"
+                             " the node %s is NULL",
+                             loc->path);
+            goto out;
+        }
+
+        contribution = mq_get_contribution_node(loc->parent, ctx);
+        if (contribution == NULL) {
+            ret = -1;
+            gf_log(this->name, GF_LOG_DEBUG,
+                   "contribution for the node %s is NULL", loc->path);
+            goto out;
+        }
+
+        LOCK(&contribution->lock);
+        {
+            delta.size = contribution->contribution;
+            delta.file_count = contribution->file_count;
+            delta.dir_count = contribution->dir_count;
+        }
+        UNLOCK(&contribution->lock);
+    }
+
+    ret = mq_get_set_dirty(this, &parent_loc, 1, &prev_dirty);
+    if (ret < 0)
+        goto out;
+    dirty = _gf_true;
+
+    mq_sub_meta(&delta, NULL);
+
+    if (remove_xattr) {
+        ret = mq_remove_contri(this, loc, ctx, contribution, &delta, nlink);
+        if (ret < 0)
+            goto out;
+    }
+
+    if (quota_meta_is_null(&delta))
+        goto out;
+
+    ret = mq_update_size(this, &parent_loc, &delta);
+    if (ret < 0)
+        goto out;
+
+out:
+    if (dirty) {
+        if (ret < 0 || prev_dirty) {
+            /* On failure clear dirty status flag.
+             * In the next lookup inspect_directory_xattr
+             * can set the status flag and fix the
+             * dirty directory.
+             * Do the same if dir was dirty before
+             * the txn
+             */
+            ret = mq_inode_ctx_get(parent_loc.inode, this, &parent_ctx);
+            if (ret == 0)
+                mq_set_ctx_dirty_status(parent_ctx, _gf_false);
+        } else {
+            ret = mq_mark_dirty(this, &parent_loc, 0);
+        }
+    }
+
+    if (locked)
+        ret = mq_lock(this, &parent_loc, F_UNLCK);
+
+    if (ret >= 0)
+        ret = mq_initiate_quota_blocking_txn(this, &parent_loc, NULL);
+
+    loc_wipe(&parent_loc);
+
+    if (contribution)
+        GF_REF_PUT(contribution);
+
+    return ret;
+}
+
+int32_t
+mq_reduce_parent_size_txn(xlator_t *this, loc_t *origin_loc,
+                          quota_meta_t *contri, uint32_t nlink,
+                          call_stub_t *stub)
+{
+    int32_t ret = -1;
+    loc_t loc = {
+        0,
+    };
+    gf_boolean_t resume_stub = _gf_true;
+
+    GF_VALIDATE_OR_GOTO("marker", this, out);
+    GF_VALIDATE_OR_GOTO("marker", origin_loc, out);
+
+    ret = mq_prevalidate_txn(this, origin_loc, &loc, NULL, NULL);
+    if (ret < 0)
+        goto out;
+
+    if (loc_is_root(&loc)) {
+        ret = 0;
+        goto out;
+    }
+
+    resume_stub = _gf_false;
+    ret = mq_synctask1(this, mq_reduce_parent_size_task, _gf_true, &loc, contri,
+                       nlink, stub);
+out:
+    loc_wipe(&loc);
+
+    if (resume_stub && stub)
+        call_resume(stub);
+
+    if (ret)
+        gf_log_callingfn(this ? this->name : "Marker", GF_LOG_ERROR,
+                         "mq_reduce_parent_size_txn failed");
+
+    return ret;
+}
+
+int
+mq_initiate_quota_task(void *opaque)
+{
+    int32_t ret = -1;
+    int32_t prev_dirty = 0;
+    loc_t child_loc = {
+        0,
+    };
+    loc_t parent_loc = {
+        0,
+    };
+    gf_boolean_t locked = _gf_false;
+    gf_boolean_t dirty = _gf_false;
+    gf_boolean_t status = _gf_false;
+    quota_meta_t delta = {
+        0,
+    };
+    quota_synctask_t *args = NULL;
+    xlator_t *this = NULL;
+    loc_t *loc = NULL;
+    inode_contribution_t *contri = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    quota_inode_ctx_t *parent_ctx = NULL;
+    inode_t *tmp_parent = NULL;
+
+    GF_VALIDATE_OR_GOTO("marker", opaque, out);
+
+    args = (quota_synctask_t *)opaque;
+    loc = &args->loc;
+    this = args->this;
+
+    GF_VALIDATE_OR_GOTO("marker", this, out);
+    THIS = this;
+
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    ret = mq_loc_copy(&child_loc, loc);
+    if (ret < 0) {
+        gf_log(this->name, GF_LOG_ERROR, "loc copy failed");
+        goto out;
+    }
+
+    while (!__is_root_gfid(child_loc.gfid)) {
+        ret = mq_inode_ctx_get(child_loc.inode, this, &ctx);
+        if (ret < 0) {
+            gf_log(this->name, GF_LOG_WARNING,
+                   "inode ctx get failed for %s, "
+                   "aborting update txn",
+                   child_loc.path);
+            goto out;
+        }
+
+        /* To improve performance, abort current transaction
+         * if one is already in progress for same inode
+         */
+        if (status == _gf_true) {
+            /* status will already set before txn start,
+             * so it should not be set in first
+             * loop iteration
+             */
+            ret = mq_test_and_set_ctx_updation_status(ctx, &status);
+            if (ret < 0 || status == _gf_true)
+                goto out;
+        }
+
+        if (child_loc.parent == NULL) {
+            ret = mq_build_ancestry(this, &child_loc);
+            if (ret < 0 || child_loc.parent == NULL) {
+                /* If application performs parallel remove
+                 * operations on same set of files/directories
+                 * then we may get ENOENT/ESTALE
+                 */
+                gf_log(this->name,
+                       (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG
+                                                          : GF_LOG_ERROR,
+                       "build ancestry failed for inode %s",
+                       uuid_utoa(child_loc.inode->gfid));
+                ret = -1;
+                goto out;
+            }
+        }
+
+        ret = mq_inode_loc_fill(NULL, child_loc.parent, &parent_loc);
+        if (ret < 0) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "parent_loc fill "
+                   "failed for child inode %s: ",
+                   uuid_utoa(child_loc.inode->gfid));
+            goto out;
+        }
+
+        ret = mq_lock(this, &parent_loc, F_WRLCK);
+        if (ret < 0)
+            goto out;
+        locked = _gf_true;
+
+        mq_set_ctx_updation_status(ctx, _gf_false);
+        status = _gf_true;
+
+        /* Contribution node can be NULL in below scenarios and
+           create if needed:
+
+           Scenario 1)
+           In this case create a new contribution node
+           Suppose hard link for a file f1 present in a directory d1 is
+           created in the directory d2 (as f2). Now, since d2's
+           contribution is not there in f1's inode ctx, d2's
+           contribution xattr won't be created and will create problems
+           for quota operations.
+
+           Don't create contribution if parent has been changed after
+           taking a lock, this can happen when rename is performed
+           and writes is still in-progress for the same file
+
+           Scenario 2)
+           When a rename operation is performed, contribution node
+           for olp path will be removed.
+
+           Create contribution node only if oldparent is same as
+           newparent.
+           Consider below example
+           1) rename FOP invoked on file 'x'
+           2) write is still in progress for file 'x'
+           3) rename takes a lock on old-parent
+           4) write-update txn blocked on old-parent to acquire lock
+           5) in rename_cbk, contri xattrs are removed and contribution
+              is deleted and lock is released
+           6) now write-update txn gets the lock and updates the
+              wrong parent as it was holding lock on old parent
+              so validate parent once the lock is acquired
+
+             For more information on this problem, please see
+             doc for marker_rename in file marker.c
+        */
+        contri = mq_get_contribution_node(child_loc.parent, ctx);
+        if (contri == NULL) {
+            tmp_parent = inode_parent(child_loc.inode, 0, NULL);
+            if (tmp_parent == NULL) {
+                /* This can happen if application performs
+                 * parallel remove operations on same set
+                 * of files/directories
+                 */
+                gf_log(this->name, GF_LOG_WARNING,
+                       "parent is "
+                       "NULL for inode %s",
+                       uuid_utoa(child_loc.inode->gfid));
+                ret = -1;
+                goto out;
+            }
+            if (gf_uuid_compare(tmp_parent->gfid, parent_loc.gfid)) {
+                /* abort txn if parent has changed */
+                ret = 0;
+                goto out;
+            }
+
+            inode_unref(tmp_parent);
+            tmp_parent = NULL;
+
+            contri = mq_add_new_contribution_node(this, ctx, &child_loc);
+            if (contri == NULL) {
+                gf_log(this->name, GF_LOG_ERROR,
+                       "Failed to "
+                       "create contribution node for %s, "
+                       "abort update txn",
+                       child_loc.path);
+                ret = -1;
+                goto out;
+            }
+        }
+
+        ret = mq_get_delta(this, &child_loc, &delta, ctx, contri);
+        if (ret < 0)
+            goto out;
+
+        if (quota_meta_is_null(&delta))
+            goto out;
+
+        ret = mq_get_set_dirty(this, &parent_loc, 1, &prev_dirty);
+        if (ret < 0)
+            goto out;
+        dirty = _gf_true;
+
+        ret = mq_update_contri(this, &child_loc, contri, &delta);
+        if (ret < 0)
+            goto out;
+
+        ret = mq_update_size(this, &parent_loc, &delta);
+        if (ret < 0) {
+            gf_log(this->name, GF_LOG_DEBUG,
+                   "rollback "
+                   "contri updation");
+            mq_sub_meta(&delta, NULL);
+            mq_update_contri(this, &child_loc, contri, &delta);
+            goto out;
+        }
+
+        if (prev_dirty == 0) {
+            ret = mq_mark_dirty(this, &parent_loc, 0);
+        } else {
+            ret = mq_inode_ctx_get(parent_loc.inode, this, &parent_ctx);
+            if (ret == 0)
+                mq_set_ctx_dirty_status(parent_ctx, _gf_false);
+        }
+        dirty = _gf_false;
+        prev_dirty = 0;
+
+        ret = mq_lock(this, &parent_loc, F_UNLCK);
+        locked = _gf_false;
+
+        if (__is_root_gfid(parent_loc.gfid))
+            break;
+
+        /* Repeate above steps upwards till the root */
+        loc_wipe(&child_loc);
+        ret = mq_loc_copy(&child_loc, &parent_loc);
+        if (ret < 0)
+            goto out;
+
+        loc_wipe(&parent_loc);
+        GF_REF_PUT(contri);
+        contri = NULL;
+    }
+
+out:
+    if ((dirty) && (ret < 0)) {
+        /* On failure clear dirty status flag.
+         * In the next lookup inspect_directory_xattr
+         * can set the status flag and fix the
+         * dirty directory.
+         * Do the same if the dir was dirty before
+         * txn
+         */
+        ret = mq_inode_ctx_get(parent_loc.inode, this, &parent_ctx);
+        if (ret == 0)
+            mq_set_ctx_dirty_status(parent_ctx, _gf_false);
+    }
+
+    if (locked)
+        ret = mq_lock(this, &parent_loc, F_UNLCK);
+
+    if (ctx && status == _gf_false)
+        mq_set_ctx_updation_status(ctx, _gf_false);
+
+    loc_wipe(&child_loc);
+    loc_wipe(&parent_loc);
+
+    if (tmp_parent)
+        inode_unref(tmp_parent);
+
+    if (contri)
+        GF_REF_PUT(contri);
+
+    return 0;
+}
+
+int
+_mq_initiate_quota_txn(xlator_t *this, loc_t *origin_loc, struct iatt *buf,
+                       gf_boolean_t spawn)
+{
+    int32_t ret = -1;
+    quota_inode_ctx_t *ctx = NULL;
+    gf_boolean_t status = _gf_true;
+    loc_t loc = {
+        0,
+    };
+
+    ret = mq_prevalidate_txn(this, origin_loc, &loc, &ctx, buf);
+    if (ret < 0)
+        goto out;
+
+    if (loc_is_root(&loc)) {
+        ret = 0;
+        goto out;
+    }
+
+    ret = mq_test_and_set_ctx_updation_status(ctx, &status);
+    if (ret < 0 || status == _gf_true)
+        goto out;
+
+    ret = mq_synctask(this, mq_initiate_quota_task, spawn, &loc);
+
+out:
+    if (ret < 0 && status == _gf_false)
+        mq_set_ctx_updation_status(ctx, _gf_false);
+
+    loc_wipe(&loc);
+    return ret;
+}
+
+int
+mq_initiate_quota_txn(xlator_t *this, loc_t *loc, struct iatt *buf)
+{
+    int32_t ret = -1;
+
+    GF_VALIDATE_OR_GOTO("marker", this, out);
+    GF_VALIDATE_OR_GOTO("marker", loc, out);
+    GF_VALIDATE_OR_GOTO("marker", loc->inode, out);
+
+    ret = _mq_initiate_quota_txn(this, loc, buf, _gf_true);
+out:
+    return ret;
+}
+
+int
+mq_initiate_quota_blocking_txn(xlator_t *this, loc_t *loc, struct iatt *buf)
+{
+    int32_t ret = -1;
+
+    GF_VALIDATE_OR_GOTO("marker", this, out);
+    GF_VALIDATE_OR_GOTO("marker", loc, out);
+    GF_VALIDATE_OR_GOTO("marker", loc->inode, out);
+
+    ret = _mq_initiate_quota_txn(this, loc, buf, _gf_false);
+out:
+    return ret;
+}
+
+int
+mq_update_dirty_inode_task(void *opaque)
+{
+    int32_t ret = -1;
+    fd_t *fd = NULL;
+    off_t offset = 0;
+    gf_dirent_t entries;
+    gf_dirent_t *entry = NULL;
+    gf_boolean_t locked = _gf_false;
+    gf_boolean_t updated = _gf_false;
+    int32_t dirty = 0;
+    quota_meta_t contri = {
+        0,
+    };
+    quota_meta_t size = {
+        0,
+    };
+    quota_meta_t contri_sum = {
+        0,
+    };
+    quota_meta_t delta = {
+        0,
+    };
+    quota_synctask_t *args = NULL;
+    xlator_t *this = NULL;
+    loc_t *loc = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    dict_t *xdata = NULL;
+    char contri_key[QUOTA_KEY_MAX] = {
+        0,
+    };
+    int keylen = 0;
+
+    GF_ASSERT(opaque);
+
+    args = (quota_synctask_t *)opaque;
+    loc = &args->loc;
+    this = args->this;
+    THIS = this;
+    INIT_LIST_HEAD(&entries.list);
+
+    ret = mq_inode_ctx_get(loc->inode, this, &ctx);
+    if (ret < 0)
+        goto out;
+
+    GET_CONTRI_KEY(this, contri_key, loc->gfid, keylen);
+    if (keylen < 0) {
+        ret = keylen;
+        goto out;
+    }
+
+    xdata = dict_new();
+    if (xdata == NULL) {
+        gf_log(this->name, GF_LOG_ERROR, "dict_new failed");
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_set_int64(xdata, contri_key, 0);
+    if (ret < 0) {
+        gf_log(this->name, GF_LOG_ERROR, "dict_set failed");
+        goto out;
+    }
+
+    ret = mq_lock(this, loc, F_WRLCK);
+    if (ret < 0)
+        goto out;
+    locked = _gf_true;
+
+    ret = mq_get_dirty(this, loc, &dirty);
+    if (ret < 0 || dirty == 0) {
+        ret = 0;
+        goto out;
+    }
+
+    fd = fd_create(loc->inode, 0);
+    if (!fd) {
+        gf_log(this->name, GF_LOG_ERROR, "Failed to create fd");
+        ret = -1;
+        goto out;
+    }
+
+    ret = syncop_opendir(this, loc, fd, NULL, NULL);
+    if (ret < 0) {
+        gf_log(this->name,
+               (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR,
+               "opendir failed "
+               "for %s: %s",
+               loc->path, strerror(-ret));
+        goto out;
+    }
+
+    fd_bind(fd);
+    while ((ret = syncop_readdirp(this, fd, 131072, offset, &entries, xdata,
+                                  NULL)) != 0) {
+        if (ret < 0) {
+            gf_log(this->name,
+                   (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG
+                                                      : GF_LOG_ERROR,
+                   "readdirp failed "
+                   "for %s: %s",
+                   loc->path, strerror(-ret));
+            goto out;
+        }
+
+        if (list_empty(&entries.list))
+            break;
+
+        list_for_each_entry(entry, &entries.list, list)
+        {
+            offset = entry->d_off;
+
+            if (!strcmp(entry->d_name, ".") || !strcmp(entry->d_name, ".."))
+                continue;
+
+            memset(&contri, 0, sizeof(contri));
+            quota_dict_get_meta(entry->dict, contri_key, keylen, &contri);
+            if (quota_meta_is_null(&contri))
+                continue;
+
+            mq_add_meta(&contri_sum, &contri);
+        }
+
+        gf_dirent_free(&entries);
+    }
+    /* Inculde for self */
+    contri_sum.dir_count++;
+
+    ret = _mq_get_metadata(this, loc, NULL, &size, 0);
+    if (ret < 0)
+        goto out;
+
+    mq_compute_delta(&delta, &contri_sum, &size);
+
+    if (quota_meta_is_null(&delta))
+        goto out;
+
+    gf_log(this->name, GF_LOG_INFO,
+           "calculated size = %" PRId64 ", original size = %" PRIu64
+           ", diff = %" PRIu64 ", path = %s ",
+           contri_sum.size, size.size, delta.size, loc->path);
+
+    gf_log(this->name, GF_LOG_INFO,
+           "calculated f_count = %" PRId64 ", original f_count = %" PRIu64
+           ", diff = %" PRIu64 ", path = %s ",
+           contri_sum.file_count, size.file_count, delta.file_count, loc->path);
+
+    gf_log(this->name, GF_LOG_INFO,
+           "calculated d_count = %" PRId64 ", original d_count = %" PRIu64
+           ", diff = %" PRIu64 ", path = %s ",
+           contri_sum.dir_count, size.dir_count, delta.dir_count, loc->path);
+
+    ret = mq_update_size(this, loc, &delta);
+    if (ret < 0)
+        goto out;
+
+    updated = _gf_true;
+
+out:
+    gf_dirent_free(&entries);
+
+    if (fd)
+        fd_unref(fd);
+
+    if (xdata)
+        dict_unref(xdata);
+
+    if (ret < 0) {
+        /* On failure clear dirty status flag.
+         * In the next lookup inspect_directory_xattr
+         * can set the status flag and fix the
+         * dirty directory
+         */
+        if (ctx)
+            mq_set_ctx_dirty_status(ctx, _gf_false);
+    } else if (dirty) {
+        mq_mark_dirty(this, loc, 0);
+    }
+
+    if (locked)
+        mq_lock(this, loc, F_UNLCK);
+
+    if (updated)
+        mq_initiate_quota_blocking_txn(this, loc, NULL);
+
+    return ret;
+}
+
+int32_t
+mq_update_dirty_inode_txn(xlator_t *this, loc_t *loc, quota_inode_ctx_t *ctx)
+{
+    int32_t ret = -1;
+    gf_boolean_t status = _gf_true;
+
+    GF_VALIDATE_OR_GOTO("marker", loc, out);
+    GF_VALIDATE_OR_GOTO("marker", loc->inode, out);
+
+    mq_test_and_set_ctx_status(ctx, &ctx->dirty_status, &status);
+    if (status == _gf_true)
+        goto out;
+
+    ret = mq_synctask(this, mq_update_dirty_inode_task, _gf_true, loc);
+out:
+    if (ret < 0 && status == _gf_false)
+        mq_set_ctx_dirty_status(ctx, _gf_false);
+
+    return ret;
+}
+
+int32_t
+mq_inspect_directory_xattr(xlator_t *this, quota_inode_ctx_t *ctx,
+                           inode_contribution_t *contribution, loc_t *loc,
+                           dict_t *dict)
+{
+    int32_t ret = -1;
+    int8_t dirty = -1;
+    quota_meta_t size = {
+        0,
+    };
+    quota_meta_t contri = {
+        0,
+    };
+    quota_meta_t delta = {
+        0,
+    };
+    char contri_key[QUOTA_KEY_MAX] = {
+        0,
+    };
+    char size_key[QUOTA_KEY_MAX] = {
+        0,
+    };
+    int keylen = 0;
+    gf_boolean_t status = _gf_false;
+
+    ret = dict_get_int8(dict, QUOTA_DIRTY_KEY, &dirty);
+    if (ret < 0) {
+        /* dirty is set only on the first file write operation
+         * so ignore this error
+         */
+        ret = 0;
+        dirty = 0;
+    }
+
+    GET_SIZE_KEY(this, size_key, keylen);
+    if (keylen < 0) {
+        ret = -1;
+        goto out;
+    }
+    ret = _quota_dict_get_meta(this, dict, size_key, keylen, &size, IA_IFDIR,
+                               _gf_false);
+    if (ret < 0)
+        goto create_xattr;
+
+    if (!contribution)
+        goto create_xattr;
+
+    if (!loc_is_root(loc)) {
+        GET_CONTRI_KEY(this, contri_key, contribution->gfid, keylen);
+        if (keylen < 0) {
+            ret = -1;
+            goto out;
+        }
+        ret = _quota_dict_get_meta(this, dict, contri_key, keylen, &contri,
+                                   IA_IFDIR, _gf_false);
+        if (ret < 0)
+            goto create_xattr;
+
+        LOCK(&contribution->lock);
+        {
+            contribution->contribution = contri.size;
+            contribution->file_count = contri.file_count;
+            contribution->dir_count = contri.dir_count;
+        }
+        UNLOCK(&contribution->lock);
+    }
+
+    LOCK(&ctx->lock);
+    {
+        ctx->size = size.size;
+        ctx->file_count = size.file_count;
+        ctx->dir_count = size.dir_count;
+        ctx->dirty = dirty;
+    }
+    UNLOCK(&ctx->lock);
+
+    ret = mq_get_ctx_updation_status(ctx, &status);
+    if (ret < 0 || status == _gf_true) {
+        /* If the update txn is in progress abort inspection */
+        ret = 0;
+        goto out;
+    }
+
+    mq_compute_delta(&delta, &size, &contri);
+
+    if (dirty) {
+        ret = mq_update_dirty_inode_txn(this, loc, ctx);
+        goto out;
+    }
+
+    if (!loc_is_root(loc) && !quota_meta_is_null(&delta))
+        mq_initiate_quota_txn(this, loc, NULL);
+
+    ret = 0;
+    goto out;
+
+create_xattr:
+    if (ret < 0)
+        ret = mq_create_xattrs_txn(this, loc, NULL);
+
+out:
+    return ret;
+}
+
+int32_t
+mq_inspect_file_xattr(xlator_t *this, quota_inode_ctx_t *ctx,
+                      inode_contribution_t *contribution, loc_t *loc,
+                      dict_t *dict, struct iatt *buf)
+{
+    int32_t ret = -1;
+    quota_meta_t size = {
+        0,
+    };
+    quota_meta_t contri = {
+        0,
+    };
+    quota_meta_t delta = {
+        0,
+    };
+    char contri_key[QUOTA_KEY_MAX] = {
+        0,
+    };
+    int keylen = 0;
+    gf_boolean_t status = _gf_false;
+
+    if (!buf || !contribution || !ctx)
+        goto out;
+
+    LOCK(&ctx->lock);
+    {
+        ctx->size = 512 * buf->ia_blocks;
+        ctx->file_count = 1;
+        ctx->dir_count = 0;
+
+        size.size = ctx->size;
+        size.file_count = ctx->file_count;
+        size.dir_count = ctx->dir_count;
+    }
+    UNLOCK(&ctx->lock);
+
+    GET_CONTRI_KEY(this, contri_key, contribution->gfid, keylen);
+    if (keylen < 0) {
+        ret = -1;
+        goto out;
+    }
+
+    ret = _quota_dict_get_meta(this, dict, contri_key, keylen, &contri,
+                               IA_IFREG, _gf_true);
+    if (ret < 0) {
+        ret = mq_create_xattrs_txn(this, loc, NULL);
+    } else {
+        LOCK(&contribution->lock);
+        {
+            contribution->contribution = contri.size;
+            contribution->file_count = contri.file_count;
+            contribution->dir_count = contri.dir_count;
+        }
+        UNLOCK(&contribution->lock);
+
+        ret = mq_get_ctx_updation_status(ctx, &status);
+        if (ret < 0 || status == _gf_true) {
+            /* If the update txn is in progress abort inspection */
+            ret = 0;
+            goto out;
+        }
+
+        mq_compute_delta(&delta, &size, &contri);
+        if (!quota_meta_is_null(&delta))
+            mq_initiate_quota_txn(this, loc, NULL);
+    }
+    /* TODO: revist this code when fixing hardlinks */
+
+out:
+    return ret;
+}
+
+int32_t
+mq_xattr_state(xlator_t *this, loc_t *origin_loc, dict_t *dict,
+               struct iatt *buf)
+{
+    int32_t ret = -1;
+    quota_inode_ctx_t *ctx = NULL;
+    loc_t loc = {
+        0,
+    };
+    inode_contribution_t *contribution = NULL;
+
+    ret = mq_prevalidate_txn(this, origin_loc, &loc, &ctx, buf);
+    if (ret < 0 || loc.parent == NULL)
+        goto out;
+
+    if (!loc_is_root(&loc)) {
+        contribution = mq_add_new_contribution_node(this, ctx, &loc);
+        if (contribution == NULL) {
+            if (!gf_uuid_is_null(loc.inode->gfid))
+                gf_log(this->name, GF_LOG_WARNING,
+                       "cannot add a new contribution node "
+                       "(%s)",
+                       uuid_utoa(loc.gfid));
+            ret = -1;
+            goto out;
+        }
+        if (buf->ia_type == IA_IFDIR)
+            mq_inspect_directory_xattr(this, ctx, contribution, &loc, dict);
+        else
+            mq_inspect_file_xattr(this, ctx, contribution, &loc, dict, buf);
+    } else {
+        mq_inspect_directory_xattr(this, ctx, 0, &loc, dict);
+    }
+
+out:
+    loc_wipe(&loc);
+
+    if (contribution)
+        GF_REF_PUT(contribution);
+
+    return ret;
+}
+
+int32_t
+mq_req_xattr(xlator_t *this, loc_t *loc, dict_t *dict, char *contri_key,
+             char *size_key)
+{
+    int32_t ret = -1;
+    char key[QUOTA_KEY_MAX] = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("marker", this, out);
+    GF_VALIDATE_OR_GOTO("marker", loc, out);
+    GF_VALIDATE_OR_GOTO("marker", dict, out);
+
+    if (!loc_is_root(loc)) {
+        ret = mq_dict_set_contribution(this, dict, loc, NULL, contri_key);
+        if (ret < 0)
+            goto out;
+    }
+
+    GET_SIZE_KEY(this, key, ret);
+    if (ret < 0)
+        goto out;
+    if (size_key)
+        if (snprintf(size_key, QUOTA_KEY_MAX, "%s", key) >= QUOTA_KEY_MAX) {
+            ret = -1;
+            goto out;
+        }
+
+    ret = dict_set_uint64(dict, key, 0);
+    if (ret < 0)
+        goto out;
+
+    ret = dict_set_int8(dict, QUOTA_DIRTY_KEY, 0);
+
+out:
+    if (ret < 0)
+        gf_log_callingfn(this ? this->name : "Marker", GF_LOG_ERROR,
+                         "dict set failed");
+    return ret;
+}
+
+int32_t
+mq_forget(xlator_t *this, quota_inode_ctx_t *ctx)
+{
+    inode_contribution_t *contri = NULL;
+    inode_contribution_t *next = NULL;
+
+    GF_VALIDATE_OR_GOTO("marker", this, out);
+    GF_VALIDATE_OR_GOTO("marker", ctx, out);
+
+    list_for_each_entry_safe(contri, next, &ctx->contribution_head, contri_list)
+    {
+        list_del_init(&contri->contri_list);
+        GF_REF_PUT(contri);
+    }
+
+    LOCK_DESTROY(&ctx->lock);
+    GF_FREE(ctx);
+out:
+    return 0;
+}
diff --git a/xlators/features/marker/src/marker-quota.h b/xlators/features/marker/src/marker-quota.h
new file mode 100644
index 00000000000..4bbf6878b22
--- /dev/null
+++ b/xlators/features/marker/src/marker-quota.h
@@ -0,0 +1,140 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#ifndef _MARKER_QUOTA_H
+#define _MARKER_QUOTA_H
+
+#include <glusterfs/xlator.h>
+#include "marker-mem-types.h"
+#include <glusterfs/refcount.h>
+#include <glusterfs/quota-common-utils.h>
+#include <glusterfs/call-stub.h>
+
+#define QUOTA_XATTR_PREFIX "trusted.glusterfs"
+#define QUOTA_DIRTY_KEY "trusted.glusterfs.quota.dirty"
+
+#define CONTRIBUTION "contri"
+#define QUOTA_KEY_MAX 512
+#define READDIR_BUF 4096
+
+#define QUOTA_ALLOC(var, type, ret)                                            \
+    do {                                                                       \
+        ret = 0;                                                               \
+        var = GF_CALLOC(sizeof(type), 1, gf_marker_mt_##type);                 \
+        if (!var) {                                                            \
+            ret = -1;                                                          \
+        }                                                                      \
+    } while (0);
+
+#define QUOTA_ALLOC_OR_GOTO(var, type, ret, label)                             \
+    do {                                                                       \
+        var = GF_CALLOC(sizeof(type), 1, gf_marker_mt_##type);                 \
+        if (!var) {                                                            \
+            gf_log("", GF_LOG_ERROR, "out of memory");                         \
+            ret = -1;                                                          \
+            goto label;                                                        \
+        }                                                                      \
+        ret = 0;                                                               \
+    } while (0);
+
+#define GET_QUOTA_KEY(_this, var, key, _ret)                                   \
+    do {                                                                       \
+        marker_conf_t *_priv = _this->private;                                 \
+        if (_priv->version > 0)                                                \
+            _ret = snprintf(var, QUOTA_KEY_MAX, "%s.%d", key, _priv->version); \
+        else                                                                   \
+            _ret = snprintf(var, QUOTA_KEY_MAX, "%s", key);                    \
+    } while (0)
+
+#define GET_CONTRI_KEY(_this, var, _gfid, _ret)                                \
+    do {                                                                       \
+        char _tmp_var[QUOTA_KEY_MAX] = {                                       \
+            0,                                                                 \
+        };                                                                     \
+        if (_gfid != NULL) {                                                   \
+            char _gfid_unparsed[40];                                           \
+            gf_uuid_unparse(_gfid, _gfid_unparsed);                            \
+            _ret = snprintf(_tmp_var, QUOTA_KEY_MAX,                           \
+                            QUOTA_XATTR_PREFIX ".%s.%s." CONTRIBUTION,         \
+                            "quota", _gfid_unparsed);                          \
+        } else {                                                               \
+            _ret = snprintf(_tmp_var, QUOTA_KEY_MAX,                           \
+                            QUOTA_XATTR_PREFIX ".%s.." CONTRIBUTION, "quota"); \
+        }                                                                      \
+        GET_QUOTA_KEY(_this, var, _tmp_var, _ret);                             \
+    } while (0)
+
+#define GET_SIZE_KEY(_this, var, _ret)                                         \
+    {                                                                          \
+        GET_QUOTA_KEY(_this, var, QUOTA_SIZE_KEY, _ret);                       \
+    }
+
+#define QUOTA_SAFE_INCREMENT(lock, var)                                        \
+    do {                                                                       \
+        LOCK(lock);                                                            \
+        var++;                                                                 \
+        UNLOCK(lock);                                                          \
+    } while (0)
+
+struct quota_inode_ctx {
+    int64_t size;
+    int64_t file_count;
+    int64_t dir_count;
+    int8_t dirty;
+    gf_boolean_t create_status;
+    gf_boolean_t updation_status;
+    gf_boolean_t dirty_status;
+    gf_lock_t lock;
+    struct list_head contribution_head;
+};
+typedef struct quota_inode_ctx quota_inode_ctx_t;
+
+struct quota_synctask {
+    xlator_t *this;
+    loc_t loc;
+    quota_meta_t contri;
+    gf_boolean_t is_static;
+    uint32_t ia_nlink;
+    call_stub_t *stub;
+};
+typedef struct quota_synctask quota_synctask_t;
+
+struct inode_contribution {
+    struct list_head contri_list;
+    int64_t contribution;
+    int64_t file_count;
+    int64_t dir_count;
+    uuid_t gfid;
+    gf_lock_t lock;
+    GF_REF_DECL;
+};
+typedef struct inode_contribution inode_contribution_t;
+
+int32_t
+mq_req_xattr(xlator_t *, loc_t *, dict_t *, char *, char *);
+
+int32_t
+mq_xattr_state(xlator_t *, loc_t *, dict_t *, struct iatt *);
+
+int
+mq_initiate_quota_txn(xlator_t *, loc_t *, struct iatt *);
+
+int
+mq_initiate_quota_blocking_txn(xlator_t *, loc_t *, struct iatt *);
+
+int
+mq_create_xattrs_txn(xlator_t *this, loc_t *loc, struct iatt *buf);
+
+int32_t
+mq_reduce_parent_size_txn(xlator_t *, loc_t *, quota_meta_t *, uint32_t nlink,
+                          call_stub_t *stub);
+
+int32_t
+mq_forget(xlator_t *, quota_inode_ctx_t *);
+#endif
diff --git a/xlators/features/marker/src/marker.c b/xlators/features/marker/src/marker.c
new file mode 100644
index 00000000000..1375ccc498c
--- /dev/null
+++ b/xlators/features/marker/src/marker.c
@@ -0,0 +1,3568 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+#include "libxlator.h"
+#include "marker.h"
+#include "marker-mem-types.h"
+#include "marker-quota.h"
+#include "marker-quota-helper.h"
+#include "marker-common.h"
+#include <glusterfs/byte-order.h>
+#include <glusterfs/syncop.h>
+#include <glusterfs/syscall.h>
+
+#include <fnmatch.h>
+
+#define _GF_UID_GID_CHANGED 1
+
+static char *mq_ext_xattrs[] = {
+    QUOTA_SIZE_KEY,
+    QUOTA_LIMIT_KEY,
+    QUOTA_LIMIT_OBJECTS_KEY,
+    NULL,
+};
+
+void
+fini(xlator_t *this);
+
+int32_t
+marker_start_setxattr(call_frame_t *, xlator_t *);
+
+/* When client/quotad request for quota xattrs,
+ * replace the key-name by adding the version number
+ * in end of the key-name.
+ * In the cbk, result value of xattrs for original
+ * key-name.
+ * Below function marker_key_replace_with_ver and
+ * marker_key_set_ver is used for setting/removing
+ * version for the key-name
+ */
+int
+marker_key_replace_with_ver(xlator_t *this, dict_t *dict)
+{
+    int ret = -1;
+    int i = 0;
+    marker_conf_t *priv = NULL;
+    char key[QUOTA_KEY_MAX] = {
+        0,
+    };
+
+    priv = this->private;
+
+    if (dict == NULL || priv->version <= 0) {
+        ret = 0;
+        goto out;
+    }
+
+    for (i = 0; mq_ext_xattrs[i]; i++) {
+        if (dict_get(dict, mq_ext_xattrs[i])) {
+            GET_QUOTA_KEY(this, key, mq_ext_xattrs[i], ret);
+            if (ret < 0)
+                goto out;
+
+            ret = dict_set(dict, key, dict_get(dict, mq_ext_xattrs[i]));
+            if (ret < 0)
+                goto out;
+
+            dict_del(dict, mq_ext_xattrs[i]);
+        }
+    }
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+int
+marker_key_set_ver(xlator_t *this, dict_t *dict)
+{
+    int ret = -1;
+    int i = -1;
+    marker_conf_t *priv = NULL;
+    char key[QUOTA_KEY_MAX] = {
+        0,
+    };
+
+    priv = this->private;
+
+    if (dict == NULL || priv->version <= 0) {
+        ret = 0;
+        goto out;
+    }
+
+    for (i = 0; mq_ext_xattrs[i]; i++) {
+        GET_QUOTA_KEY(this, key, mq_ext_xattrs[i], ret);
+        if (ret < 0)
+            goto out;
+
+        if (dict_get(dict, key))
+            dict_set(dict, mq_ext_xattrs[i], dict_get(dict, key));
+    }
+
+    ret = 0;
+out:
+    return ret;
+}
+
+marker_local_t *
+marker_local_ref(marker_local_t *local)
+{
+    GF_VALIDATE_OR_GOTO("marker", local, err);
+
+    LOCK(&local->lock);
+    {
+        local->ref++;
+    }
+    UNLOCK(&local->lock);
+
+    return local;
+err:
+    return NULL;
+}
+
+int
+marker_loc_fill(loc_t *loc, inode_t *inode, inode_t *parent, char *path)
+{
+    int ret = -1;
+
+    if (!loc)
+        return ret;
+
+    if (inode) {
+        loc->inode = inode_ref(inode);
+        if (gf_uuid_is_null(loc->gfid)) {
+            gf_uuid_copy(loc->gfid, loc->inode->gfid);
+        }
+    }
+
+    if (parent)
+        loc->parent = inode_ref(parent);
+
+    if (path) {
+        loc->path = gf_strdup(path);
+        if (!loc->path) {
+            gf_log("loc fill", GF_LOG_ERROR, "strdup failed");
+            goto loc_wipe;
+        }
+
+        loc->name = strrchr(loc->path, '/');
+        if (loc->name)
+            loc->name++;
+    }
+
+    ret = 0;
+loc_wipe:
+    if (ret < 0)
+        loc_wipe(loc);
+
+    return ret;
+}
+
+int
+_marker_inode_loc_fill(inode_t *inode, inode_t *parent, char *name, loc_t *loc)
+{
+    char *resolvedpath = NULL;
+    int ret = -1;
+    gf_boolean_t free_parent = _gf_false;
+
+    if ((!inode) || (!loc))
+        return ret;
+
+    if (parent && name)
+        ret = inode_path(parent, name, &resolvedpath);
+    else
+        ret = inode_path(inode, NULL, &resolvedpath);
+    if (ret < 0)
+        goto err;
+
+    if (parent == NULL) {
+        parent = inode_parent(inode, NULL, NULL);
+        free_parent = _gf_true;
+    }
+
+    ret = marker_loc_fill(loc, inode, parent, resolvedpath);
+    if (ret < 0)
+        goto err;
+
+err:
+    if (free_parent)
+        inode_unref(parent);
+
+    GF_FREE(resolvedpath);
+
+    return ret;
+}
+
+int
+marker_inode_loc_fill(inode_t *inode, loc_t *loc)
+{
+    return _marker_inode_loc_fill(inode, NULL, NULL, loc);
+}
+
+int32_t
+marker_trav_parent(marker_local_t *local)
+{
+    int32_t ret = 0;
+    loc_t loc = {
+        0,
+    };
+    inode_t *parent = NULL;
+    int8_t need_unref = 0;
+
+    if (!local->loc.parent) {
+        parent = inode_parent(local->loc.inode, NULL, NULL);
+        if (parent)
+            need_unref = 1;
+    } else
+        parent = local->loc.parent;
+
+    ret = marker_inode_loc_fill(parent, &loc);
+
+    if (ret < 0) {
+        ret = -1;
+        goto out;
+    }
+
+    loc_wipe(&local->loc);
+
+    local->loc = loc;
+out:
+    if (need_unref)
+        inode_unref(parent);
+
+    return ret;
+}
+
+void
+marker_error_handler(xlator_t *this, marker_local_t *local, int32_t op_errno)
+{
+    marker_conf_t *priv = (marker_conf_t *)this->private;
+    const char *path = local ? ((local->loc.path) ? local->loc.path
+                                                  : uuid_utoa(local->loc.gfid))
+                             : "<nul>";
+
+    gf_log(this->name, GF_LOG_CRITICAL,
+           "Indexing gone corrupt at %s (reason: %s)."
+           " Geo-replication slave content needs to be revalidated",
+           path, strerror(op_errno));
+    sys_unlink(priv->timestamp_file);
+}
+
+int32_t
+marker_local_unref(marker_local_t *local)
+{
+    int32_t var = 0;
+
+    if (local == NULL)
+        return -1;
+
+    LOCK(&local->lock);
+    {
+        var = --local->ref;
+    }
+    UNLOCK(&local->lock);
+
+    if (var != 0)
+        goto out;
+
+    loc_wipe(&local->loc);
+    loc_wipe(&local->parent_loc);
+    if (local->xdata)
+        dict_unref(local->xdata);
+
+    if (local->lk_frame) {
+        STACK_DESTROY(local->lk_frame->root);
+        local->lk_frame = NULL;
+    }
+
+    if (local->oplocal) {
+        marker_local_unref(local->oplocal);
+        local->oplocal = NULL;
+    }
+    mem_put(local);
+out:
+    return 0;
+}
+
+int32_t
+stat_stampfile(xlator_t *this, marker_conf_t *priv, struct volume_mark **status)
+{
+    struct stat buf = {
+        0,
+    };
+    struct volume_mark *vol_mark = NULL;
+
+    vol_mark = GF_CALLOC(sizeof(struct volume_mark), 1,
+                         gf_marker_mt_volume_mark);
+
+    vol_mark->major = 1;
+    vol_mark->minor = 0;
+
+    GF_ASSERT(sizeof(priv->volume_uuid_bin) == 16);
+    memcpy(vol_mark->uuid, priv->volume_uuid_bin, 16);
+
+    if (sys_stat(priv->timestamp_file, &buf) != -1) {
+        vol_mark->retval = 0;
+        vol_mark->sec = htonl(buf.st_mtime);
+        vol_mark->usec = htonl(ST_MTIM_NSEC(&buf) / 1000);
+    } else
+        vol_mark->retval = 1;
+
+    *status = vol_mark;
+
+    return 0;
+}
+
+int32_t
+marker_getxattr_stampfile_cbk(call_frame_t *frame, xlator_t *this,
+                              const char *name, struct volume_mark *vol_mark,
+                              dict_t *xdata)
+{
+    int32_t ret = -1;
+    dict_t *dict = NULL;
+
+    if (vol_mark == NULL) {
+        STACK_UNWIND_STRICT(getxattr, frame, -1, ENOMEM, NULL, NULL);
+
+        goto out;
+    }
+
+    dict = dict_new();
+
+    ret = dict_set_bin(dict, (char *)name, vol_mark,
+                       sizeof(struct volume_mark));
+    if (ret) {
+        GF_FREE(vol_mark);
+        gf_log(this->name, GF_LOG_WARNING, "failed to set key %s", name);
+    }
+
+    STACK_UNWIND_STRICT(getxattr, frame, 0, 0, dict, xdata);
+
+    if (dict)
+        dict_unref(dict);
+out:
+    return 0;
+}
+
+gf_boolean_t
+call_from_special_client(call_frame_t *frame, xlator_t *this, const char *name)
+{
+    struct volume_mark *vol_mark = NULL;
+    marker_conf_t *priv = NULL;
+    gf_boolean_t is_true = _gf_true;
+
+    priv = (marker_conf_t *)this->private;
+
+    if (frame->root->pid != GF_CLIENT_PID_GSYNCD || name == NULL ||
+        strcmp(name, MARKER_XATTR_PREFIX "." VOLUME_MARK) != 0) {
+        is_true = _gf_false;
+        goto out;
+    }
+
+    stat_stampfile(this, priv, &vol_mark);
+
+    marker_getxattr_stampfile_cbk(frame, this, name, vol_mark, NULL);
+out:
+    return is_true;
+}
+
+static gf_boolean_t
+_is_quota_internal_xattr(dict_t *d, char *k, data_t *v, void *data)
+{
+    int i = 0;
+    char **external_xattrs = data;
+
+    for (i = 0; external_xattrs && external_xattrs[i]; i++) {
+        if (strcmp(k, external_xattrs[i]) == 0)
+            return _gf_false;
+    }
+
+    if (fnmatch("trusted.glusterfs.quota*", k, 0) == 0)
+        return _gf_true;
+
+    /* It would be nice if posix filters pgfid xattrs. But since marker
+     * also takes up responsibility to clean these up, adding the filtering
+     * here (Check 'quota_xattr_cleaner')
+     */
+    if (fnmatch(PGFID_XATTR_KEY_PREFIX "*", k, 0) == 0)
+        return _gf_true;
+
+    return _gf_false;
+}
+
+static void
+marker_filter_internal_xattrs(xlator_t *this, dict_t *xattrs)
+{
+    marker_conf_t *priv = NULL;
+    char **ext = NULL;
+
+    priv = this->private;
+    if (priv->feature_enabled & GF_QUOTA)
+        ext = mq_ext_xattrs;
+
+    dict_foreach_match(xattrs, _is_quota_internal_xattr, ext,
+                       dict_remove_foreach_fn, NULL);
+}
+
+static void
+marker_filter_gsyncd_xattrs(call_frame_t *frame, xlator_t *this, dict_t *xattrs)
+{
+    marker_conf_t *priv = NULL;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+    GF_ASSERT(frame);
+
+    if (xattrs && frame->root->pid != GF_CLIENT_PID_GSYNCD) {
+        GF_REMOVE_INTERNAL_XATTR(GF_XATTR_XTIME_PATTERN, xattrs);
+    }
+    return;
+}
+
+int32_t
+marker_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *dict,
+                    dict_t *xdata)
+{
+    int32_t ret = -1;
+    if (op_ret < 0)
+        goto unwind;
+
+    ret = marker_key_set_ver(this, dict);
+    if (ret < 0) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+
+    if (cookie) {
+        gf_log(this->name, GF_LOG_DEBUG,
+               "Filtering the quota extended attributes");
+
+        /* If the getxattr is from a non special client, then do not
+           copy the quota related xattrs (except the quota limit key
+           i.e trusted.glusterfs.quota.limit-set which has been set by
+           glusterd on the directory on which quota limit is set.) for
+           directories. Let the healing of xattrs happen upon lookup.
+           NOTE: setting of trusted.glusterfs.quota.limit-set as of now
+           happens from glusterd. It should be moved to quotad. Also
+           trusted.glusterfs.quota.limit-set is set on directory which
+           is permanent till quota is removed on that directory or limit
+           is changed. So let that xattr be healed by other xlators
+           properly whenever directory healing is done.
+        */
+        /*
+         * Except limit-set xattr, rest of the xattrs are maintained
+         * by quota xlator. Don't expose them to other xlators.
+         * This filter makes sure quota xattrs are not healed as part of
+         * metadata self-heal
+         */
+        marker_filter_internal_xattrs(frame->this, dict);
+    }
+
+    /* Filter gsyncd xtime xattr for non gsyncd clients */
+    marker_filter_gsyncd_xattrs(frame, frame->this, dict);
+
+unwind:
+    MARKER_STACK_UNWIND(getxattr, frame, op_ret, op_errno, dict, xdata);
+    return 0;
+}
+
+int32_t
+marker_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                const char *name, dict_t *xdata)
+{
+    gf_boolean_t is_true = _gf_false;
+    marker_conf_t *priv = NULL;
+    unsigned long cookie = 0;
+    marker_local_t *local = NULL;
+    char key[QUOTA_KEY_MAX] = {
+        0,
+    };
+    int32_t ret = -1;
+    int32_t i = 0;
+
+    priv = this->private;
+
+    if (name) {
+        for (i = 0; mq_ext_xattrs[i]; i++) {
+            if (strcmp(name, mq_ext_xattrs[i]))
+                continue;
+
+            GET_QUOTA_KEY(this, key, mq_ext_xattrs[i], ret);
+            if (ret < 0)
+                goto out;
+            name = key;
+            break;
+        }
+    }
+
+    frame->local = mem_get0(this->local_pool);
+    local = frame->local;
+    if (local == NULL)
+        goto out;
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    if ((loc_copy(&local->loc, loc)) < 0)
+        goto out;
+
+    gf_log(this->name, GF_LOG_DEBUG, "USER:PID = %d", frame->root->pid);
+
+    if (priv && priv->feature_enabled & GF_XTIME)
+        is_true = call_from_special_client(frame, this, name);
+
+    if (is_true == _gf_false) {
+        if (name == NULL) {
+            /* Signifies that marker translator
+             * has to filter the quota's xattr's,
+             * this is to prevent afr from performing
+             * self healing on marker-quota xattrs'
+             */
+            cookie = 1;
+        }
+        STACK_WIND_COOKIE(frame, marker_getxattr_cbk, (void *)cookie,
+                          FIRST_CHILD(this), FIRST_CHILD(this)->fops->getxattr,
+                          loc, name, xdata);
+    }
+
+    return 0;
+out:
+    MARKER_STACK_UNWIND(getxattr, frame, -1, ENOMEM, NULL, NULL);
+    return 0;
+}
+
+int32_t
+marker_setxattr_done(call_frame_t *frame)
+{
+    marker_local_t *local = NULL;
+
+    local = (marker_local_t *)frame->local;
+
+    frame->local = NULL;
+
+    STACK_DESTROY(frame->root);
+
+    marker_local_unref(local);
+
+    return 0;
+}
+
+int
+marker_specific_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                             int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    int32_t ret = 0;
+    int32_t done = 1;
+    marker_local_t *local = NULL;
+
+    local = (marker_local_t *)frame->local;
+
+    if (op_ret == -1 && op_errno == ENOSPC) {
+        marker_error_handler(this, local, op_errno);
+        goto out;
+    }
+
+    if (local) {
+        if (local->loc.path && strcmp(local->loc.path, "/") == 0) {
+            goto out;
+        }
+        if (__is_root_gfid(local->loc.gfid)) {
+            goto out;
+        }
+    }
+
+    ret = (local) ? marker_trav_parent(local) : -1;
+
+    if (ret == -1) {
+        gf_log(this->name, GF_LOG_DEBUG,
+               "Error occurred "
+               "while traversing to the parent, stopping marker");
+        goto out;
+    }
+
+    marker_start_setxattr(frame, this);
+    done = 0;
+out:
+    if (done) {
+        marker_setxattr_done(frame);
+    }
+
+    return 0;
+}
+
+int32_t
+marker_start_setxattr(call_frame_t *frame, xlator_t *this)
+{
+    int32_t ret = -1;
+    dict_t *dict = NULL;
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    priv = this->private;
+
+    local = (marker_local_t *)frame->local;
+
+    if (!local)
+        goto out;
+
+    dict = dict_new();
+
+    if (!dict)
+        goto out;
+
+    if (local->loc.inode && gf_uuid_is_null(local->loc.gfid))
+        gf_uuid_copy(local->loc.gfid, local->loc.inode->gfid);
+
+    GF_UUID_ASSERT(local->loc.gfid);
+
+    ret = dict_set_static_bin(dict, priv->marker_xattr, (void *)local->timebuf,
+                              8);
+    if (ret) {
+        gf_log(this->name, GF_LOG_WARNING, "failed to set marker xattr (%s)",
+               local->loc.path);
+        goto out;
+    }
+
+    STACK_WIND(frame, marker_specific_setxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setxattr, &local->loc, dict, 0, NULL);
+
+    ret = 0;
+out:
+    if (dict)
+        dict_unref(dict);
+
+    return ret;
+}
+
+void
+marker_gettimeofday(marker_local_t *local)
+{
+    struct timeval tv = {
+        0,
+    };
+
+    gettimeofday(&tv, NULL);
+
+    local->timebuf[0] = htonl(tv.tv_sec);
+    local->timebuf[1] = htonl(tv.tv_usec);
+
+    return;
+}
+
+int32_t
+marker_create_frame(xlator_t *this, marker_local_t *local)
+{
+    call_frame_t *frame = NULL;
+
+    frame = create_frame(this, this->ctx->pool);
+
+    if (!frame)
+        return -1;
+
+    frame->local = (void *)local;
+
+    marker_start_setxattr(frame, this);
+
+    return 0;
+}
+
+int32_t
+marker_xtime_update_marks(xlator_t *this, marker_local_t *local)
+{
+    marker_conf_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("marker", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, local, out);
+
+    priv = this->private;
+
+    if ((local->pid == GF_CLIENT_PID_GSYNCD &&
+         !(priv->feature_enabled & GF_XTIME_GSYNC_FORCE)) ||
+        (local->pid == GF_CLIENT_PID_DEFRAG))
+        goto out;
+
+    marker_gettimeofday(local);
+
+    marker_local_ref(local);
+
+    marker_create_frame(this, local);
+out:
+    return 0;
+}
+
+int32_t
+marker_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, inode_t *inode,
+                 struct iatt *buf, struct iatt *preparent,
+                 struct iatt *postparent, dict_t *xdata)
+{
+    marker_conf_t *priv = NULL;
+    marker_local_t *local = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_TRACE,
+               "error occurred "
+               "while creating directory %s",
+               strerror(op_errno));
+    }
+
+    local = (marker_local_t *)frame->local;
+
+    frame->local = NULL;
+    priv = this->private;
+
+    if (op_ret >= 0 && inode && (priv->feature_enabled & GF_QUOTA)) {
+        ctx = mq_inode_ctx_new(inode, this);
+        if (ctx == NULL) {
+            gf_log(this->name, GF_LOG_WARNING,
+                   "mq_inode_ctx_new "
+                   "failed for %s",
+                   uuid_utoa(inode->gfid));
+            op_ret = -1;
+            op_errno = ENOMEM;
+        }
+    }
+
+    STACK_UNWIND_STRICT(mkdir, frame, op_ret, op_errno, inode, buf, preparent,
+                        postparent, xdata);
+
+    if (op_ret == -1 || local == NULL)
+        goto out;
+
+    if (gf_uuid_is_null(local->loc.gfid))
+        gf_uuid_copy(local->loc.gfid, buf->ia_gfid);
+
+    if (priv->feature_enabled & GF_QUOTA)
+        mq_create_xattrs_txn(this, &local->loc, NULL);
+
+    if (priv->feature_enabled & GF_XTIME)
+        marker_xtime_update_marks(this, local);
+
+out:
+    marker_local_unref(local);
+
+    return 0;
+}
+
+int
+marker_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+             mode_t umask, dict_t *xdata)
+{
+    int32_t ret = 0;
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    priv = this->private;
+
+    if (priv->feature_enabled == 0)
+        goto wind;
+
+    local = mem_get0(this->local_pool);
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    ret = loc_copy(&local->loc, loc);
+
+    if (ret == -1)
+        goto err;
+wind:
+    STACK_WIND(frame, marker_mkdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mkdir, loc, mode, umask, xdata);
+
+    return 0;
+err:
+    MARKER_STACK_UNWIND(mkdir, frame, -1, ENOMEM, NULL, NULL, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int32_t
+marker_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, fd_t *fd, inode_t *inode,
+                  struct iatt *buf, struct iatt *preparent,
+                  struct iatt *postparent, dict_t *xdata)
+{
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_TRACE,
+               "error occurred "
+               "while creating file %s",
+               strerror(op_errno));
+    }
+
+    local = (marker_local_t *)frame->local;
+
+    frame->local = NULL;
+    priv = this->private;
+
+    if (op_ret >= 0 && inode && (priv->feature_enabled & GF_QUOTA)) {
+        ctx = mq_inode_ctx_new(inode, this);
+        if (ctx == NULL) {
+            gf_log(this->name, GF_LOG_WARNING,
+                   "mq_inode_ctx_new "
+                   "failed for %s",
+                   uuid_utoa(inode->gfid));
+            op_ret = -1;
+            op_errno = ENOMEM;
+        }
+    }
+
+    STACK_UNWIND_STRICT(create, frame, op_ret, op_errno, fd, inode, buf,
+                        preparent, postparent, xdata);
+
+    if (op_ret == -1 || local == NULL)
+        goto out;
+
+    if (gf_uuid_is_null(local->loc.gfid))
+        gf_uuid_copy(local->loc.gfid, buf->ia_gfid);
+
+    if (priv->feature_enabled & GF_QUOTA)
+        mq_create_xattrs_txn(this, &local->loc, buf);
+
+    if (priv->feature_enabled & GF_XTIME)
+        marker_xtime_update_marks(this, local);
+
+out:
+    marker_local_unref(local);
+
+    return 0;
+}
+
+int32_t
+marker_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+              mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+    int32_t ret = 0;
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    priv = this->private;
+
+    if (priv->feature_enabled == 0)
+        goto wind;
+
+    local = mem_get0(this->local_pool);
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    ret = loc_copy(&local->loc, loc);
+
+    if (ret == -1)
+        goto err;
+wind:
+    STACK_WIND(frame, marker_create_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->create, loc, flags, mode, umask, fd,
+               xdata);
+    return 0;
+err:
+    MARKER_STACK_UNWIND(create, frame, -1, ENOMEM, NULL, NULL, NULL, NULL, NULL,
+                        NULL);
+
+    return 0;
+}
+
+int32_t
+marker_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                  struct iatt *postbuf, dict_t *xdata)
+{
+    marker_conf_t *priv = NULL;
+    marker_local_t *local = NULL;
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_TRACE,
+               "error occurred "
+               "while write, %s",
+               strerror(op_errno));
+    }
+
+    local = (marker_local_t *)frame->local;
+
+    frame->local = NULL;
+
+    STACK_UNWIND_STRICT(writev, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+
+    if (op_ret == -1 || local == NULL)
+        goto out;
+
+    priv = this->private;
+
+    if (priv->feature_enabled & GF_QUOTA)
+        mq_initiate_quota_txn(this, &local->loc, postbuf);
+
+    if (priv->feature_enabled & GF_XTIME)
+        marker_xtime_update_marks(this, local);
+
+out:
+    marker_local_unref(local);
+
+    return 0;
+}
+
+int32_t
+marker_writev(call_frame_t *frame, xlator_t *this, fd_t *fd,
+              struct iovec *vector, int32_t count, off_t offset, uint32_t flags,
+              struct iobref *iobref, dict_t *xdata)
+{
+    int32_t ret = 0;
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    priv = this->private;
+
+    if (priv->feature_enabled == 0)
+        goto wind;
+
+    local = mem_get0(this->local_pool);
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    ret = marker_inode_loc_fill(fd->inode, &local->loc);
+
+    if (ret == -1)
+        goto err;
+wind:
+    STACK_WIND(frame, marker_writev_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->writev, fd, vector, count, offset,
+               flags, iobref, xdata);
+    return 0;
+err:
+    MARKER_STACK_UNWIND(writev, frame, -1, ENOMEM, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int32_t
+marker_rmdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                 struct iatt *postparent, dict_t *xdata)
+{
+    marker_conf_t *priv = NULL;
+    marker_local_t *local = NULL;
+    call_stub_t *stub = NULL;
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_TRACE,
+               "error occurred "
+               "rmdir %s",
+               strerror(op_errno));
+    }
+
+    local = (marker_local_t *)frame->local;
+
+    frame->local = NULL;
+    priv = this->private;
+
+    if (op_ret == -1 || local == NULL)
+        goto out;
+
+    if (priv->feature_enabled & GF_XTIME)
+        marker_xtime_update_marks(this, local);
+
+    if (priv->feature_enabled & GF_QUOTA) {
+        /* If a 'rm -rf' is performed by a client, rmdir can be faster
+           than marker background mq_reduce_parent_size_txn.
+           In this case, as part of rmdir parent child association
+           will be removed in the server protocol.
+           This can lead to mq_reduce_parent_size_txn failures.
+
+           So perform mq_reduce_parent_size_txn in foreground
+           and unwind to server once txn is complete
+         */
+
+        stub = fop_rmdir_cbk_stub(frame, default_rmdir_cbk, op_ret, op_errno,
+                                  preparent, postparent, xdata);
+        mq_reduce_parent_size_txn(this, &local->loc, NULL, 1, stub);
+
+        if (stub) {
+            marker_local_unref(local);
+            return 0;
+        }
+    }
+
+out:
+    STACK_UNWIND_STRICT(rmdir, frame, op_ret, op_errno, preparent, postparent,
+                        xdata);
+
+    marker_local_unref(local);
+
+    return 0;
+}
+
+int32_t
+marker_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+             dict_t *xdata)
+{
+    int32_t ret = 0;
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    priv = this->private;
+
+    if (priv->feature_enabled == 0)
+        goto wind;
+
+    local = mem_get0(this->local_pool);
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    ret = loc_copy(&local->loc, loc);
+
+    if (ret == -1)
+        goto err;
+wind:
+    STACK_WIND(frame, marker_rmdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rmdir, loc, flags, xdata);
+    return 0;
+err:
+    MARKER_STACK_UNWIND(rmdir, frame, -1, ENOMEM, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int32_t
+marker_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                  struct iatt *postparent, dict_t *xdata)
+{
+    marker_conf_t *priv = NULL;
+    marker_local_t *local = NULL;
+    uint32_t nlink = -1;
+    GF_UNUSED int32_t ret = 0;
+    call_stub_t *stub = NULL;
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_TRACE, "%s occurred in unlink",
+               strerror(op_errno));
+    }
+
+    local = (marker_local_t *)frame->local;
+
+    frame->local = NULL;
+    priv = this->private;
+
+    if (op_ret == -1 || local == NULL)
+        goto out;
+
+    if (priv->feature_enabled & GF_XTIME)
+        marker_xtime_update_marks(this, local);
+
+    if (priv->feature_enabled & GF_QUOTA) {
+        if (local->skip_txn)
+            goto out;
+
+        if (xdata) {
+            ret = dict_get_uint32(xdata, GF_RESPONSE_LINK_COUNT_XDATA, &nlink);
+            if (ret) {
+                gf_log(this->name, GF_LOG_TRACE, "dict get failed %s ",
+                       strerror(-ret));
+            }
+        }
+
+        /* If a 'rm -rf' is performed by a client, unlink can be faster
+           than marker background mq_reduce_parent_size_txn.
+           In this case, as part of unlink parent child association
+           will be removed in the server protocol.
+           This can lead to mq_reduce_parent_size_txn failures.
+
+          So perform mq_reduce_parent_size_txn in foreground
+          and unwind to server once txn is complete
+        */
+
+        stub = fop_unlink_cbk_stub(frame, default_unlink_cbk, op_ret, op_errno,
+                                   preparent, postparent, xdata);
+        mq_reduce_parent_size_txn(this, &local->loc, NULL, nlink, stub);
+
+        if (stub) {
+            marker_local_unref(local);
+            return 0;
+        }
+    }
+
+out:
+    STACK_UNWIND_STRICT(unlink, frame, op_ret, op_errno, preparent, postparent,
+                        xdata);
+
+    marker_local_unref(local);
+
+    return 0;
+}
+
+int32_t
+marker_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+              dict_t *xdata)
+{
+    int32_t ret = 0;
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+    gf_boolean_t dict_free = _gf_false;
+
+    priv = this->private;
+
+    if (priv->feature_enabled == 0)
+        goto unlink_wind;
+
+    local = mem_get0(this->local_pool);
+    local->xflag = xflag;
+    if (xdata)
+        local->xdata = dict_ref(xdata);
+    MARKER_INIT_LOCAL(frame, local);
+
+    ret = loc_copy(&local->loc, loc);
+
+    if (ret == -1)
+        goto err;
+
+    if (xdata && dict_get(xdata, GLUSTERFS_MARKER_DONT_ACCOUNT_KEY)) {
+        local->skip_txn = 1;
+        goto unlink_wind;
+    }
+
+    if (xdata == NULL) {
+        xdata = dict_new();
+        dict_free = _gf_true;
+    }
+
+    ret = dict_set_int32(xdata, GF_REQUEST_LINK_COUNT_XDATA, 1);
+    if (ret < 0)
+        goto err;
+
+unlink_wind:
+    STACK_WIND(frame, marker_unlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->unlink, loc, xflag, xdata);
+    goto out;
+
+err:
+    MARKER_STACK_UNWIND(unlink, frame, -1, ENOMEM, NULL, NULL, NULL);
+
+out:
+    if (dict_free)
+        dict_unref(xdata);
+    return 0;
+}
+
+int32_t
+marker_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, inode_t *inode,
+                struct iatt *buf, struct iatt *preparent,
+                struct iatt *postparent, dict_t *xdata)
+{
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_TRACE,
+               "%s occurred while "
+               "linking a file ",
+               strerror(op_errno));
+    }
+
+    local = (marker_local_t *)frame->local;
+
+    frame->local = NULL;
+
+    STACK_UNWIND_STRICT(link, frame, op_ret, op_errno, inode, buf, preparent,
+                        postparent, xdata);
+
+    if (op_ret == -1 || local == NULL)
+        goto out;
+
+    priv = this->private;
+
+    if (priv->feature_enabled & GF_QUOTA) {
+        if (!local->skip_txn)
+            mq_create_xattrs_txn(this, &local->loc, buf);
+    }
+
+    if (priv->feature_enabled & GF_XTIME)
+        marker_xtime_update_marks(this, local);
+out:
+    marker_local_unref(local);
+
+    return 0;
+}
+
+int32_t
+marker_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+            dict_t *xdata)
+{
+    int32_t ret = 0;
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    priv = this->private;
+
+    if (priv->feature_enabled == 0)
+        goto wind;
+
+    local = mem_get0(this->local_pool);
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    ret = loc_copy(&local->loc, newloc);
+
+    if (ret == -1)
+        goto err;
+
+    if (xdata && dict_get(xdata, GLUSTERFS_MARKER_DONT_ACCOUNT_KEY))
+        local->skip_txn = 1;
+wind:
+    STACK_WIND(frame, marker_link_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->link, oldloc, newloc, xdata);
+    return 0;
+err:
+    MARKER_STACK_UNWIND(link, frame, -1, ENOMEM, NULL, NULL, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int32_t
+marker_rename_done(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    marker_local_t *local = NULL, *oplocal = NULL;
+    loc_t newloc = {
+        0,
+    };
+    marker_conf_t *priv = NULL;
+
+    local = frame->local;
+    oplocal = local->oplocal;
+
+    priv = this->private;
+
+    frame->local = NULL;
+
+    if (op_ret < 0) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "inodelk (UNLOCK) failed on path:%s (gfid:%s) (%s)",
+               oplocal->parent_loc.path,
+               uuid_utoa(oplocal->parent_loc.inode->gfid), strerror(op_errno));
+    }
+
+    if (local->err != 0)
+        goto err;
+
+    mq_reduce_parent_size_txn(this, &oplocal->loc, &oplocal->contribution, -1,
+                              NULL);
+
+    if (local->loc.inode != NULL) {
+        /* If destination file exits before rename, it would have
+         * been unlinked while renaming a file
+         */
+        mq_reduce_parent_size_txn(this, &local->loc, NULL, local->ia_nlink,
+                                  NULL);
+    }
+
+    newloc.inode = inode_ref(oplocal->loc.inode);
+    newloc.path = gf_strdup(local->loc.path);
+    newloc.name = strrchr(newloc.path, '/');
+    if (newloc.name)
+        newloc.name++;
+    newloc.parent = inode_ref(local->loc.parent);
+
+    mq_create_xattrs_txn(this, &newloc, &local->buf);
+
+    loc_wipe(&newloc);
+
+    if (priv->feature_enabled & GF_XTIME) {
+        if (!local->loc.inode)
+            local->loc.inode = inode_ref(oplocal->loc.inode);
+        // update marks on oldpath
+        gf_uuid_copy(local->loc.gfid, oplocal->loc.inode->gfid);
+        marker_xtime_update_marks(this, oplocal);
+        marker_xtime_update_marks(this, local);
+    }
+
+err:
+    marker_local_unref(local);
+    marker_local_unref(oplocal);
+
+    return 0;
+}
+
+void
+marker_rename_release_oldp_lock(marker_local_t *local, xlator_t *this)
+{
+    marker_local_t *oplocal = NULL;
+    call_frame_t *lk_frame = NULL;
+    struct gf_flock lock = {
+        0,
+    };
+
+    oplocal = local->oplocal;
+    lk_frame = local->lk_frame;
+
+    if (lk_frame == NULL)
+        goto err;
+
+    lock.l_type = F_UNLCK;
+    lock.l_whence = SEEK_SET;
+    lock.l_start = 0;
+    lock.l_len = 0;
+    lock.l_pid = 0;
+
+    STACK_WIND(lk_frame, marker_rename_done, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->inodelk, this->name,
+               &oplocal->parent_loc, F_SETLKW, &lock, NULL);
+
+    return;
+
+err:
+    marker_local_unref(local);
+    marker_local_unref(oplocal);
+}
+
+int32_t
+marker_rename_unwind(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    marker_local_t *local = NULL;
+    marker_local_t *oplocal = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    inode_contribution_t *contri = NULL;
+
+    local = frame->local;
+    oplocal = local->oplocal;
+    frame->local = NULL;
+
+    // Reset frame uid and gid if set.
+    if (cookie == (void *)_GF_UID_GID_CHANGED)
+        MARKER_RESET_UID_GID(frame, frame->root, local);
+
+    if (op_ret < 0)
+        local->err = op_errno ? op_errno : EINVAL;
+
+    if (local->stub != NULL) {
+        /* Remove contribution node from in-memory even if
+         * remove-xattr has failed as the rename is already performed
+         * if local->stub is set, which means rename was successful
+         */
+        (void)mq_inode_ctx_get(oplocal->loc.inode, this, &ctx);
+        if (ctx) {
+            contri = mq_get_contribution_node(oplocal->loc.parent, ctx);
+            if (contri) {
+                QUOTA_FREE_CONTRIBUTION_NODE(ctx, contri);
+                GF_REF_PUT(contri);
+            }
+        }
+
+        call_resume(local->stub);
+        local->stub = NULL;
+        local->err = 0;
+    } else if (local->err != 0) {
+        STACK_UNWIND_STRICT(rename, frame, -1, local->err, NULL, NULL, NULL,
+                            NULL, NULL, NULL);
+    } else {
+        gf_log(this->name, GF_LOG_CRITICAL,
+               "continuation stub to unwind the call is absent, hence "
+               "call will be hung (call-stack id = %" PRIu64 ")",
+               frame->root->unique);
+    }
+
+    /* If there are in-progress writes on old-path when during rename
+     * operation, update txn will update the wrong path if lock
+     * is released before rename unwind.
+     * So release lock only after rename unwind
+     */
+    marker_rename_release_oldp_lock(local, this);
+
+    return 0;
+}
+
+int32_t
+marker_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                  struct iatt *preoldparent, struct iatt *postoldparent,
+                  struct iatt *prenewparent, struct iatt *postnewparent,
+                  dict_t *xdata)
+{
+    marker_conf_t *priv = NULL;
+    marker_local_t *local = NULL;
+    marker_local_t *oplocal = NULL;
+    call_stub_t *stub = NULL;
+    int32_t ret = 0;
+    char contri_key[QUOTA_KEY_MAX] = {
+        0,
+    };
+    loc_t newloc = {
+        0,
+    };
+
+    local = (marker_local_t *)frame->local;
+
+    if (local != NULL) {
+        oplocal = local->oplocal;
+    }
+
+    priv = this->private;
+
+    if (op_ret < 0) {
+        if (local != NULL) {
+            local->err = op_errno;
+        }
+
+        gf_log(this->name, GF_LOG_TRACE,
+               "%s occurred while "
+               "renaming a file ",
+               strerror(op_errno));
+    }
+
+    if (priv->feature_enabled & GF_QUOTA) {
+        if ((op_ret < 0) || (local == NULL)) {
+            goto quota_err;
+        }
+
+        local->ia_nlink = 0;
+        if (xdata)
+            ret = dict_get_uint32(xdata, GF_RESPONSE_LINK_COUNT_XDATA,
+                                  &local->ia_nlink);
+
+        local->buf = *buf;
+        stub = fop_rename_cbk_stub(frame, default_rename_cbk, op_ret, op_errno,
+                                   buf, preoldparent, postoldparent,
+                                   prenewparent, postnewparent, xdata);
+        if (stub == NULL) {
+            local->err = ENOMEM;
+            goto quota_err;
+        }
+
+        local->stub = stub;
+
+        GET_CONTRI_KEY(this, contri_key, oplocal->loc.parent->gfid, ret);
+        if (ret < 0) {
+            local->err = ENOMEM;
+            goto quota_err;
+        }
+
+        /* Removexattr requires uid and gid to be 0,
+         * reset them in the callback.
+         */
+        MARKER_SET_UID_GID(frame, local, frame->root);
+
+        newloc.inode = inode_ref(oplocal->loc.inode);
+        newloc.path = gf_strdup(local->loc.path);
+        newloc.name = strrchr(newloc.path, '/');
+        if (newloc.name)
+            newloc.name++;
+        newloc.parent = inode_ref(local->loc.parent);
+        gf_uuid_copy(newloc.gfid, oplocal->loc.inode->gfid);
+
+        STACK_WIND_COOKIE(
+            frame, marker_rename_unwind, frame->cookie, FIRST_CHILD(this),
+            FIRST_CHILD(this)->fops->removexattr, &newloc, contri_key, NULL);
+
+        loc_wipe(&newloc);
+    } else {
+        frame->local = NULL;
+
+        STACK_UNWIND_STRICT(rename, frame, op_ret, op_errno, buf, preoldparent,
+                            postoldparent, prenewparent, postnewparent, xdata);
+
+        if ((op_ret < 0) || (local == NULL)) {
+            goto out;
+        }
+
+        if (priv->feature_enabled & GF_XTIME) {
+            // update marks on oldpath
+            if (!local->loc.inode)
+                local->loc.inode = inode_ref(oplocal->loc.inode);
+            gf_uuid_copy(local->loc.gfid, oplocal->loc.inode->gfid);
+            marker_xtime_update_marks(this, oplocal);
+            marker_xtime_update_marks(this, local);
+        }
+    }
+
+out:
+    if (!(priv->feature_enabled & GF_QUOTA)) {
+        marker_local_unref(local);
+        marker_local_unref(oplocal);
+    }
+
+    return 0;
+
+quota_err:
+    marker_rename_unwind(frame, NULL, this, 0, 0, NULL);
+    return 0;
+}
+
+int32_t
+marker_do_rename(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+    marker_local_t *local = NULL;
+    marker_local_t *oplocal = NULL;
+    char contri_key[QUOTA_KEY_MAX] = {
+        0,
+    };
+    int keylen = 0;
+    quota_meta_t contribution = {
+        0,
+    };
+
+    local = frame->local;
+    oplocal = local->oplocal;
+
+    // Reset frame uid and gid if set.
+    if (cookie == (void *)_GF_UID_GID_CHANGED)
+        MARKER_RESET_UID_GID(frame, frame->root, local);
+
+    if ((op_ret < 0) && (op_errno != ENOATTR) && (op_errno != ENODATA)) {
+        local->err = op_errno ? op_errno : EINVAL;
+        gf_log(this->name, GF_LOG_WARNING,
+               "fetching contribution values from %s (gfid:%s) "
+               "failed (%s)",
+               oplocal->loc.path, uuid_utoa(oplocal->loc.inode->gfid),
+               strerror(op_errno));
+        goto err;
+    }
+
+    GET_CONTRI_KEY(this, contri_key, oplocal->loc.parent->gfid, keylen);
+    if (keylen < 0) {
+        local->err = errno ? errno : ENOMEM;
+        goto err;
+    }
+    quota_dict_get_meta(dict, contri_key, keylen, &contribution);
+    oplocal->contribution = contribution;
+
+    STACK_WIND(frame, marker_rename_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rename, &oplocal->loc, &local->loc,
+               local->xdata);
+
+    return 0;
+
+err:
+    marker_rename_unwind(frame, NULL, this, 0, 0, NULL);
+    return 0;
+}
+
+int32_t
+marker_get_oldpath_contribution(call_frame_t *lk_frame, void *cookie,
+                                xlator_t *this, int32_t op_ret,
+                                int32_t op_errno, dict_t *xdata)
+{
+    call_frame_t *frame = NULL;
+    marker_local_t *local = NULL;
+    marker_local_t *oplocal = NULL;
+    char contri_key[QUOTA_KEY_MAX] = {
+        0,
+    };
+    int32_t ret = 0;
+
+    local = lk_frame->local;
+    oplocal = local->oplocal;
+    frame = local->frame;
+
+    if (op_ret < 0) {
+        local->err = op_errno ? op_errno : EINVAL;
+        gf_log(this->name, GF_LOG_WARNING,
+               "cannot hold inodelk on %s (gfid:%s) (%s)", oplocal->loc.path,
+               uuid_utoa(oplocal->loc.inode->gfid), strerror(op_errno));
+        if (local->lk_frame) {
+            STACK_DESTROY(local->lk_frame->root);
+            local->lk_frame = NULL;
+        }
+        goto err;
+    }
+
+    GET_CONTRI_KEY(this, contri_key, oplocal->loc.parent->gfid, ret);
+    if (ret < 0) {
+        local->err = errno ? errno : ENOMEM;
+        goto err;
+    }
+
+    /* getxattr requires uid and gid to be 0,
+     * reset them in the callback.
+     */
+    MARKER_SET_UID_GID(frame, local, frame->root);
+
+    if (gf_uuid_is_null(oplocal->loc.gfid))
+        gf_uuid_copy(oplocal->loc.gfid, oplocal->loc.inode->gfid);
+
+    GF_UUID_ASSERT(oplocal->loc.gfid);
+
+    STACK_WIND_COOKIE(frame, marker_do_rename, frame->cookie, FIRST_CHILD(this),
+                      FIRST_CHILD(this)->fops->getxattr, &oplocal->loc,
+                      contri_key, NULL);
+
+    return 0;
+err:
+    marker_rename_unwind(frame, NULL, this, 0, 0, NULL);
+    return 0;
+}
+
+/* For a marker_rename FOP, following is the algorithm used for Quota
+ * accounting. The use-case considered is:
+ * 1. rename (src, dst)
+ * 2. both src and dst exist
+ * 3. there are parallel operations on src and dst (lets say through fds
+ *    opened on them before rename was initiated).
+ *
+ * PS: We've not thought through whether this algo works in the presence of
+ *     hardlinks to src and/or dst.
+ *
+ * Algorithm:
+ * ==========
+ *
+ * 1) set inodelk on src-parent
+ *    As part of rename operation, parent can change for the file.
+ *    We need to remove contribution (both on disk xattr and in-memory one)
+ *    to src-parent (and its ancestors) and add the contribution to dst-parent
+ *    (and its ancestors). While we are doing these operations, contribution of
+ *    the file/directory shouldn't be changing as we want to be sure that
+ *      a) what we subtract from src-parent is exactly what we add to dst-parent
+ *      b) we should subtract from src-parent exactly what we contributed to
+ *         src-parent
+ *    So, We hold a lock on src-parent to block any parallel transcations on
+ *    src-inode (since that's the one which survives rename).
+ *
+ *    If there are any parallel transactions on dst-inode they keep succeeding
+ *    till the association of dst-inode with dst-parent is broken because of an
+ *    inode_rename after unwind of rename fop from marker. Only after unwind
+ *    (and hence inode_rename), we delete and subtract the contribution of
+ *    dst-inode to dst-parent. That way we are making sure we subtract exactly
+ *    what dst-inode contributed to dst-parent.
+ *
+ * 2) lookup contribution to src-parent on src-inode.
+ *    We need to save the contribution info for use at step-8.
+ *
+ * 3) wind rename
+ *    Perform rename on disk
+ *
+ * 4) remove xattr on src-loc
+ *    After rename, parent can change, so
+ *    need to remove xattrs storing contribution to src-parent.
+ *
+ * 5) remove contribution node corresponding to src-parent from the in-memory
+ *    list.
+ *    After rename, contri gfid can change and we have
+ *    also removed xattr from file.
+ *    We need to remove in-memory contribution node to prevent updations to
+ *    src-parent even after a successful rename
+ *
+ * 6) unwind rename
+ *    This will ensure that rename is done in the server
+ *    inode table. An inode_rename disassociates src-inode from src-parent and
+ *    associates it with dst-parent. It also disassociates dst-inode from
+ *    dst-parent. After inode_rename, inode_parent on src-inode will give
+ *    dst-parent and inode_parent on dst-inode will return NULL (assuming
+ *    dst-inode doesn't have any hardlinks).
+ *
+ * 7) release inodelk on src-parent
+ *    Lock on src-parent should be released only after
+ *    rename on disk, remove xattr and rename_unwind (and hence inode_rename)
+ *    operations. If lock is released before inode_rename, a parallel
+ *    transaction on src-inode can still update src-parent (as inode_parent on
+ *    src-inode can still return src-parent). This would make the
+ *    contribution from src-inode to src-parent stored in step-2 stale.
+ *
+ * 8) Initiate mq_reduce_parent_size_txn on src-parent to remove contribution
+ *    of src-inode to src-parent. We use the contribution stored in step-2.
+ *    Since, we had acquired the lock on src-parent all along step-2 through
+ *    inode_rename, we can be sure that a parallel transaction wouldn't have
+ *    added a delta to src-parent.
+ *
+ * 9) Initiate mq_reduce_parent_size_txn on dst-parent if dst-inode exists.
+ *    The size reduced from dst-parent and its ancestors is the
+ *    size stored as contribution to dst-parent in dst-inode.
+ *    If the destination file had existed, rename will unlink the
+ *    destination file as part of its operation.
+ *    We need to reduce the size on the dest parent similarly to
+ *    unlink. Since, we are initiating reduce-parent-size transaction after
+ *    inode_rename, we can be sure that a parallel transaction wouldn't add
+ *    delta to dst-parent while we are reducing the contribution of dst-inode
+ *    from its ancestors before rename.
+ *
+ * 10) create contribution xattr to dst-parent on src-inode.
+ */
+int32_t
+marker_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+              dict_t *xdata)
+{
+    int32_t ret = 0;
+    marker_local_t *local = NULL;
+    marker_local_t *oplocal = NULL;
+    marker_conf_t *priv = NULL;
+    struct gf_flock lock = {
+        0,
+    };
+
+    priv = this->private;
+
+    if (priv->feature_enabled == 0)
+        goto rename_wind;
+
+    local = mem_get0(this->local_pool);
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    oplocal = mem_get0(this->local_pool);
+
+    MARKER_INIT_LOCAL(frame, oplocal);
+
+    frame->local = local;
+
+    local->oplocal = marker_local_ref(oplocal);
+
+    ret = loc_copy(&local->loc, newloc);
+    if (ret < 0)
+        goto err;
+
+    ret = loc_copy(&oplocal->loc, oldloc);
+    if (ret < 0)
+        goto err;
+
+    if (!(priv->feature_enabled & GF_QUOTA)) {
+        goto rename_wind;
+    }
+
+    ret = mq_inode_loc_fill(NULL, newloc->parent, &local->parent_loc);
+    if (ret < 0)
+        goto err;
+
+    ret = mq_inode_loc_fill(NULL, oldloc->parent, &oplocal->parent_loc);
+    if (ret < 0)
+        goto err;
+
+    lock.l_len = 0;
+    lock.l_start = 0;
+    lock.l_type = F_WRLCK;
+    lock.l_whence = SEEK_SET;
+
+    local->xdata = xdata ? dict_ref(xdata) : dict_new();
+    ret = dict_set_int32(local->xdata, GF_REQUEST_LINK_COUNT_XDATA, 1);
+    if (ret < 0)
+        goto err;
+
+    local->frame = frame;
+    local->lk_frame = create_frame(this, this->ctx->pool);
+    if (local->lk_frame == NULL)
+        goto err;
+
+    local->lk_frame->root->uid = 0;
+    local->lk_frame->root->gid = 0;
+    local->lk_frame->local = local;
+    set_lk_owner_from_ptr(&local->lk_frame->root->lk_owner,
+                          local->lk_frame->root);
+
+    STACK_WIND(local->lk_frame, marker_get_oldpath_contribution,
+               FIRST_CHILD(this), FIRST_CHILD(this)->fops->inodelk, this->name,
+               &oplocal->parent_loc, F_SETLKW, &lock, NULL);
+
+    return 0;
+
+rename_wind:
+    STACK_WIND(frame, marker_rename_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
+
+    return 0;
+err:
+    MARKER_STACK_UNWIND(rename, frame, -1, ENOMEM, NULL, NULL, NULL, NULL, NULL,
+                        NULL);
+    marker_local_unref(oplocal);
+
+    return 0;
+}
+
+int32_t
+marker_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                    struct iatt *postbuf, dict_t *xdata)
+{
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_TRACE,
+               "%s occurred while "
+               "truncating a file ",
+               strerror(op_errno));
+    }
+
+    local = (marker_local_t *)frame->local;
+
+    frame->local = NULL;
+
+    STACK_UNWIND_STRICT(truncate, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+
+    if (op_ret == -1 || local == NULL)
+        goto out;
+
+    priv = this->private;
+
+    if (priv->feature_enabled & GF_QUOTA) {
+        /* DHT Rebalance process, at the end of migration will
+         * first make the src file as a linkto file and then
+         * truncate the file. By doing a truncate after making the
+         * src file as linkto file, the contri which is already
+         * accounted is left over.
+         * So, we need to account for the linkto file when a truncate
+         * happens, thereby updating the contri properly.
+         * By passing NULL for postbuf, mq_prevalidate does not check
+         * for linkto file.
+         * Same happens with ftruncate as well.
+         */
+        if (postbuf && IS_DHT_LINKFILE_MODE(postbuf))
+            mq_initiate_quota_txn(this, &local->loc, NULL);
+        else
+            mq_initiate_quota_txn(this, &local->loc, postbuf);
+    }
+
+    if (priv->feature_enabled & GF_XTIME)
+        marker_xtime_update_marks(this, local);
+
+out:
+    marker_local_unref(local);
+
+    return 0;
+}
+
+int32_t
+marker_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+                dict_t *xdata)
+{
+    int32_t ret = 0;
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    priv = this->private;
+
+    if (priv->feature_enabled == 0)
+        goto wind;
+
+    local = mem_get0(this->local_pool);
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    ret = loc_copy(&local->loc, loc);
+
+    if (ret == -1)
+        goto err;
+wind:
+    STACK_WIND(frame, marker_truncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+    return 0;
+err:
+    MARKER_STACK_UNWIND(truncate, frame, -1, ENOMEM, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int32_t
+marker_ftruncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_TRACE,
+               "%s occurred while "
+               "truncating a file ",
+               strerror(op_errno));
+    }
+
+    local = (marker_local_t *)frame->local;
+
+    frame->local = NULL;
+
+    STACK_UNWIND_STRICT(ftruncate, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+
+    if (op_ret == -1 || local == NULL)
+        goto out;
+
+    priv = this->private;
+
+    if (priv->feature_enabled & GF_QUOTA) {
+        if (postbuf && IS_DHT_LINKFILE_MODE(postbuf))
+            mq_initiate_quota_txn(this, &local->loc, NULL);
+        else
+            mq_initiate_quota_txn(this, &local->loc, postbuf);
+    }
+
+    if (priv->feature_enabled & GF_XTIME)
+        marker_xtime_update_marks(this, local);
+out:
+    marker_local_unref(local);
+
+    return 0;
+}
+
+int32_t
+marker_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                 dict_t *xdata)
+{
+    int32_t ret = 0;
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    priv = this->private;
+
+    if (priv->feature_enabled == 0)
+        goto wind;
+
+    local = mem_get0(this->local_pool);
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    ret = marker_inode_loc_fill(fd->inode, &local->loc);
+
+    if (ret == -1)
+        goto err;
+wind:
+    STACK_WIND(frame, marker_ftruncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+    return 0;
+err:
+    MARKER_STACK_UNWIND(ftruncate, frame, -1, ENOMEM, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int32_t
+marker_symlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, inode_t *inode,
+                   struct iatt *buf, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
+{
+    marker_conf_t *priv = NULL;
+    marker_local_t *local = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_TRACE,
+               "%s occurred while "
+               "creating symlinks ",
+               strerror(op_errno));
+    }
+
+    local = (marker_local_t *)frame->local;
+
+    frame->local = NULL;
+    priv = this->private;
+
+    if (op_ret >= 0 && inode && (priv->feature_enabled & GF_QUOTA)) {
+        ctx = mq_inode_ctx_new(inode, this);
+        if (ctx == NULL) {
+            gf_log(this->name, GF_LOG_WARNING,
+                   "mq_inode_ctx_new "
+                   "failed for %s",
+                   uuid_utoa(inode->gfid));
+            op_ret = -1;
+            op_errno = ENOMEM;
+        }
+    }
+
+    STACK_UNWIND_STRICT(symlink, frame, op_ret, op_errno, inode, buf, preparent,
+                        postparent, xdata);
+
+    if (op_ret == -1 || local == NULL)
+        goto out;
+
+    if (gf_uuid_is_null(local->loc.gfid))
+        gf_uuid_copy(local->loc.gfid, buf->ia_gfid);
+
+    if (priv->feature_enabled & GF_QUOTA) {
+        mq_create_xattrs_txn(this, &local->loc, buf);
+    }
+
+    if (priv->feature_enabled & GF_XTIME)
+        marker_xtime_update_marks(this, local);
+out:
+    marker_local_unref(local);
+
+    return 0;
+}
+
+int
+marker_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+               loc_t *loc, mode_t umask, dict_t *xdata)
+{
+    int32_t ret = 0;
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    priv = this->private;
+
+    if (priv->feature_enabled == 0)
+        goto wind;
+
+    local = mem_get0(this->local_pool);
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    ret = loc_copy(&local->loc, loc);
+
+    if (ret == -1)
+        goto err;
+wind:
+    STACK_WIND(frame, marker_symlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->symlink, linkpath, loc, umask, xdata);
+    return 0;
+err:
+    MARKER_STACK_UNWIND(symlink, frame, -1, ENOMEM, NULL, NULL, NULL, NULL,
+                        NULL);
+
+    return 0;
+}
+
+int32_t
+marker_mknod_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, inode_t *inode,
+                 struct iatt *buf, struct iatt *preparent,
+                 struct iatt *postparent, dict_t *xdata)
+{
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_TRACE,
+               "%s occurred with "
+               "mknod ",
+               strerror(op_errno));
+    }
+
+    local = (marker_local_t *)frame->local;
+
+    frame->local = NULL;
+    priv = this->private;
+
+    if (op_ret >= 0 && inode && (priv->feature_enabled & GF_QUOTA)) {
+        ctx = mq_inode_ctx_new(inode, this);
+        if (ctx == NULL) {
+            gf_log(this->name, GF_LOG_WARNING,
+                   "mq_inode_ctx_new "
+                   "failed for %s",
+                   uuid_utoa(inode->gfid));
+            op_ret = -1;
+            op_errno = ENOMEM;
+        }
+    }
+
+    STACK_UNWIND_STRICT(mknod, frame, op_ret, op_errno, inode, buf, preparent,
+                        postparent, xdata);
+
+    if (op_ret == -1 || local == NULL)
+        goto out;
+
+    if (gf_uuid_is_null(local->loc.gfid))
+        gf_uuid_copy(local->loc.gfid, buf->ia_gfid);
+
+    if ((priv->feature_enabled & GF_QUOTA) && (S_ISREG(local->mode))) {
+        mq_create_xattrs_txn(this, &local->loc, buf);
+    }
+
+    if (priv->feature_enabled & GF_XTIME)
+        marker_xtime_update_marks(this, local);
+out:
+    marker_local_unref(local);
+
+    return 0;
+}
+
+int
+marker_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+             dev_t rdev, mode_t umask, dict_t *xdata)
+{
+    int32_t ret = 0;
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    priv = this->private;
+
+    if (priv->feature_enabled == 0)
+        goto wind;
+
+    local = mem_get0(this->local_pool);
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    ret = loc_copy(&local->loc, loc);
+
+    local->mode = mode;
+
+    if (ret == -1)
+        goto err;
+wind:
+    STACK_WIND(frame, marker_mknod_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mknod, loc, mode, rdev, umask, xdata);
+    return 0;
+err:
+    MARKER_STACK_UNWIND(mknod, frame, -1, ENOMEM, NULL, NULL, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int32_t
+marker_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                     struct iatt *postbuf, dict_t *xdata)
+{
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_TRACE,
+               "%s occurred while "
+               "fallocating a file ",
+               strerror(op_errno));
+    }
+
+    local = (marker_local_t *)frame->local;
+
+    frame->local = NULL;
+
+    STACK_UNWIND_STRICT(fallocate, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+
+    if (op_ret == -1 || local == NULL)
+        goto out;
+
+    priv = this->private;
+
+    if (priv->feature_enabled & GF_QUOTA)
+        mq_initiate_quota_txn(this, &local->loc, postbuf);
+
+    if (priv->feature_enabled & GF_XTIME)
+        marker_xtime_update_marks(this, local);
+out:
+    marker_local_unref(local);
+
+    return 0;
+}
+
+int32_t
+marker_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+                 off_t offset, size_t len, dict_t *xdata)
+{
+    int32_t ret = 0;
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    priv = this->private;
+
+    if (priv->feature_enabled == 0)
+        goto wind;
+
+    local = mem_get0(this->local_pool);
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    ret = marker_inode_loc_fill(fd->inode, &local->loc);
+
+    if (ret == -1)
+        goto err;
+wind:
+    STACK_WIND(frame, marker_fallocate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fallocate, fd, mode, offset, len,
+               xdata);
+    return 0;
+err:
+    MARKER_STACK_UNWIND(fallocate, frame, -1, ENOMEM, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int32_t
+marker_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                   struct iatt *postbuf, dict_t *xdata)
+{
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_TRACE, "%s occurred during discard",
+               strerror(op_errno));
+    }
+
+    local = (marker_local_t *)frame->local;
+
+    frame->local = NULL;
+
+    STACK_UNWIND_STRICT(discard, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+
+    if (op_ret == -1 || local == NULL)
+        goto out;
+
+    priv = this->private;
+
+    if (priv->feature_enabled & GF_QUOTA)
+        mq_initiate_quota_txn(this, &local->loc, postbuf);
+
+    if (priv->feature_enabled & GF_XTIME)
+        marker_xtime_update_marks(this, local);
+out:
+    marker_local_unref(local);
+
+    return 0;
+}
+
+int32_t
+marker_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+               size_t len, dict_t *xdata)
+{
+    int32_t ret = 0;
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    priv = this->private;
+
+    if (priv->feature_enabled == 0)
+        goto wind;
+
+    local = mem_get0(this->local_pool);
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    ret = marker_inode_loc_fill(fd->inode, &local->loc);
+
+    if (ret == -1)
+        goto err;
+wind:
+    STACK_WIND(frame, marker_discard_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->discard, fd, offset, len, xdata);
+    return 0;
+err:
+    MARKER_STACK_UNWIND(discard, frame, -1, ENOMEM, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int32_t
+marker_zerofill_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                    struct iatt *postbuf, dict_t *xdata)
+{
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_TRACE, "%s occurred during zerofill",
+               strerror(op_errno));
+    }
+
+    local = (marker_local_t *)frame->local;
+
+    frame->local = NULL;
+
+    STACK_UNWIND_STRICT(zerofill, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+
+    if (op_ret == -1 || local == NULL)
+        goto out;
+
+    priv = this->private;
+
+    if (priv->feature_enabled & GF_QUOTA)
+        mq_initiate_quota_txn(this, &local->loc, postbuf);
+
+    if (priv->feature_enabled & GF_XTIME)
+        marker_xtime_update_marks(this, local);
+out:
+    marker_local_unref(local);
+
+    return 0;
+}
+
+int32_t
+marker_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                off_t len, dict_t *xdata)
+{
+    int32_t ret = 0;
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    priv = this->private;
+
+    if (priv->feature_enabled == 0)
+        goto wind;
+
+    local = mem_get0(this->local_pool);
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    ret = marker_inode_loc_fill(fd->inode, &local->loc);
+
+    if (ret == -1)
+        goto err;
+wind:
+    STACK_WIND(frame, marker_zerofill_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->zerofill, fd, offset, len, xdata);
+    return 0;
+err:
+    MARKER_STACK_UNWIND(zerofill, frame, -1, ENOMEM, NULL, NULL, NULL);
+
+    return 0;
+}
+
+/* when a call from the special client is received on
+ * key trusted.glusterfs.volume-mark with value "RESET"
+ * or if the value is 0length, update the change the
+ * access time and modification time via touching the
+ * timestamp file.
+ */
+int32_t
+call_from_sp_client_to_reset_tmfile(call_frame_t *frame, xlator_t *this,
+                                    dict_t *dict)
+{
+    int32_t fd = 0;
+    int32_t op_ret = 0;
+    int32_t op_errno = 0;
+    data_t *data = NULL;
+    marker_conf_t *priv = NULL;
+
+    if (frame == NULL || this == NULL || dict == NULL)
+        return -1;
+
+    priv = this->private;
+
+    data = dict_get(dict, "trusted.glusterfs.volume-mark");
+    if (data == NULL)
+        return -1;
+
+    if (frame->root->pid != GF_CLIENT_PID_GSYNCD) {
+        op_ret = -1;
+        op_errno = EPERM;
+
+        goto out;
+    }
+
+    if (data->len == 0 ||
+        (data->len == 5 && memcmp(data->data, "RESET", 5) == 0)) {
+        fd = open(priv->timestamp_file, O_WRONLY | O_TRUNC);
+        if (fd != -1) {
+            /* TODO check  whether the O_TRUNC would update the
+             * timestamps on a zero length file on all machies.
+             */
+            sys_close(fd);
+        }
+
+        if (fd != -1 || errno == ENOENT) {
+            op_ret = 0;
+            op_errno = 0;
+        } else {
+            op_ret = -1;
+            op_errno = errno;
+        }
+    } else {
+        op_ret = -1;
+        op_errno = EINVAL;
+    }
+out:
+    STACK_UNWIND_STRICT(setxattr, frame, op_ret, op_errno, NULL);
+
+    return 0;
+}
+
+int32_t
+marker_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_TRACE,
+               "%s occurred in "
+               "setxattr ",
+               strerror(op_errno));
+    }
+
+    local = (marker_local_t *)frame->local;
+
+    frame->local = NULL;
+
+    STACK_UNWIND_STRICT(setxattr, frame, op_ret, op_errno, xdata);
+
+    if (op_ret == -1 || local == NULL)
+        goto out;
+
+    priv = this->private;
+
+    if (priv->feature_enabled & GF_XTIME)
+        marker_xtime_update_marks(this, local);
+out:
+    marker_local_unref(local);
+
+    return 0;
+}
+
+int
+remove_quota_keys(dict_t *dict, char *k, data_t *v, void *data)
+{
+    call_frame_t *frame = data;
+    marker_local_t *local = frame->local;
+    xlator_t *this = frame->this;
+    marker_conf_t *priv = NULL;
+    char ver_str[NAME_MAX] = {
+        0,
+    };
+    char *dot = NULL;
+    int ret = -1;
+
+    priv = this->private;
+
+    /* If quota is enabled immediately after disable.
+     * quota healing starts creating new xattrs
+     * before completing the cleanup operation.
+     * So we should check if the xattr is the new.
+     * Do not remove xattr if its xattr
+     * version is same as current version
+     */
+    if ((priv->feature_enabled & GF_QUOTA) && priv->version > 0) {
+        snprintf(ver_str, sizeof(ver_str), ".%d", priv->version);
+        dot = strrchr(k, '.');
+        if (dot && !strcmp(dot, ver_str))
+            return 0;
+    }
+
+    ret = syncop_removexattr(FIRST_CHILD(this), &local->loc, k, 0, NULL);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "%s: Failed to remove "
+               "extended attribute: %s",
+               local->loc.path, k);
+        return -1;
+    }
+    return 0;
+}
+
+int
+quota_xattr_cleaner_cbk(int ret, call_frame_t *frame, void *args)
+{
+    dict_t *xdata = args;
+    int op_ret = -1;
+    int op_errno = 0;
+
+    op_ret = (ret < 0) ? -1 : 0;
+    op_errno = -ret;
+
+    MARKER_STACK_UNWIND(setxattr, frame, op_ret, op_errno, xdata);
+    return ret;
+}
+
+int
+quota_xattr_cleaner(void *args)
+{
+    struct synctask *task = NULL;
+    call_frame_t *frame = NULL;
+    xlator_t *this = NULL;
+    marker_local_t *local = NULL;
+    dict_t *xdata = NULL;
+    int ret = -1;
+
+    task = synctask_get();
+    if (!task)
+        goto out;
+
+    frame = task->frame;
+    this = frame->this;
+    local = frame->local;
+
+    ret = syncop_listxattr(FIRST_CHILD(this), &local->loc, &xdata, NULL, NULL);
+    if (ret == -1) {
+        ret = -errno;
+        goto out;
+    }
+
+    ret = dict_foreach_fnmatch(xdata, "trusted.glusterfs.quota.*",
+                               remove_quota_keys, frame);
+    if (ret == -1) {
+        ret = -errno;
+        goto out;
+    }
+    ret = dict_foreach_fnmatch(xdata, PGFID_XATTR_KEY_PREFIX "*",
+                               remove_quota_keys, frame);
+    if (ret == -1) {
+        ret = -errno;
+        goto out;
+    }
+
+    ret = 0;
+out:
+    if (xdata)
+        dict_unref(xdata);
+
+    return ret;
+}
+
+int
+marker_do_xattr_cleanup(call_frame_t *frame, xlator_t *this, dict_t *xdata,
+                        loc_t *loc)
+{
+    int ret = -1;
+    marker_local_t *local = NULL;
+
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto out;
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    loc_copy(&local->loc, loc);
+    ret = synctask_new(this->ctx->env, quota_xattr_cleaner,
+                       quota_xattr_cleaner_cbk, frame, xdata);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Failed to create synctask "
+               "for cleaning up quota extended attributes");
+        goto out;
+    }
+
+    ret = 0;
+out:
+    if (ret)
+        MARKER_STACK_UNWIND(setxattr, frame, -1, ENOMEM, xdata);
+
+    return ret;
+}
+
+static gf_boolean_t
+marker_xattr_cleanup_cmd(dict_t *dict)
+{
+    return (dict_get(dict, VIRTUAL_QUOTA_XATTR_CLEANUP_KEY) != NULL);
+}
+
+int32_t
+marker_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+                int32_t flags, dict_t *xdata)
+{
+    int32_t ret = 0;
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+    int op_errno = ENOMEM;
+
+    priv = this->private;
+
+    if (marker_xattr_cleanup_cmd(dict)) {
+        if (frame->root->uid != 0 || frame->root->gid != 0) {
+            op_errno = EPERM;
+            ret = -1;
+            goto err;
+        }
+
+        /* The following function does the cleanup and then unwinds the
+         * corresponding call*/
+        loc_path(loc, NULL);
+        marker_do_xattr_cleanup(frame, this, xdata, loc);
+        return 0;
+    }
+
+    ret = marker_key_replace_with_ver(this, dict);
+    if (ret < 0)
+        goto err;
+
+    if (priv->feature_enabled == 0)
+        goto wind;
+
+    ret = call_from_sp_client_to_reset_tmfile(frame, this, dict);
+    if (ret == 0)
+        return 0;
+
+    local = mem_get0(this->local_pool);
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    ret = loc_copy(&local->loc, loc);
+
+    if (ret == -1)
+        goto err;
+wind:
+    STACK_WIND(frame, marker_setxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setxattr, loc, dict, flags, xdata);
+    return 0;
+err:
+    MARKER_STACK_UNWIND(setxattr, frame, -1, op_errno, NULL);
+
+    return 0;
+}
+
+int32_t
+marker_fsetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_TRACE,
+               "%s occurred in "
+               "fsetxattr",
+               strerror(op_errno));
+    }
+
+    local = (marker_local_t *)frame->local;
+
+    frame->local = NULL;
+
+    STACK_UNWIND_STRICT(fsetxattr, frame, op_ret, op_errno, xdata);
+
+    if (op_ret == -1 || local == NULL)
+        goto out;
+
+    priv = this->private;
+
+    if (priv->feature_enabled & GF_XTIME)
+        marker_xtime_update_marks(this, local);
+out:
+    marker_local_unref(local);
+
+    return 0;
+}
+
+int32_t
+marker_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+                 int32_t flags, dict_t *xdata)
+{
+    int32_t ret = 0;
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    priv = this->private;
+
+    if (priv->feature_enabled == 0)
+        goto wind;
+
+    ret = call_from_sp_client_to_reset_tmfile(frame, this, dict);
+    if (ret == 0)
+        return 0;
+
+    local = mem_get0(this->local_pool);
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    ret = marker_inode_loc_fill(fd->inode, &local->loc);
+
+    if (ret == -1)
+        goto err;
+wind:
+    STACK_WIND(frame, marker_fsetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
+    return 0;
+err:
+    MARKER_STACK_UNWIND(fsetxattr, frame, -1, ENOMEM, NULL);
+
+    return 0;
+}
+
+int32_t
+marker_fsetattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *statpre,
+                    struct iatt *statpost, dict_t *xdata)
+{
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_TRACE,
+               "%s occurred in "
+               "fsetattr ",
+               strerror(op_errno));
+    }
+
+    local = (marker_local_t *)frame->local;
+
+    frame->local = NULL;
+
+    STACK_UNWIND_STRICT(fsetattr, frame, op_ret, op_errno, statpre, statpost,
+                        xdata);
+
+    if (op_ret == -1 || local == NULL)
+        goto out;
+
+    priv = this->private;
+
+    if (priv->feature_enabled & GF_XTIME)
+        marker_xtime_update_marks(this, local);
+out:
+    marker_local_unref(local);
+
+    return 0;
+}
+
+int32_t
+marker_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    int32_t ret = 0;
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    priv = this->private;
+
+    if (priv->feature_enabled == 0)
+        goto wind;
+
+    local = mem_get0(this->local_pool);
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    ret = marker_inode_loc_fill(fd->inode, &local->loc);
+
+    if (ret == -1)
+        goto err;
+wind:
+    STACK_WIND(frame, marker_fsetattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetattr, fd, stbuf, valid, xdata);
+    return 0;
+err:
+    MARKER_STACK_UNWIND(fsetattr, frame, -1, ENOMEM, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int32_t
+marker_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *statpre,
+                   struct iatt *statpost, dict_t *xdata)
+{
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    local = (marker_local_t *)frame->local;
+
+    frame->local = NULL;
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_TRACE, "%s occurred during setattr of %s",
+               strerror(op_errno), (local ? local->loc.path : "<nul>"));
+    }
+
+    STACK_UNWIND_STRICT(setattr, frame, op_ret, op_errno, statpre, statpost,
+                        xdata);
+
+    if (op_ret == -1 || local == NULL)
+        goto out;
+
+    priv = this->private;
+
+    if (priv->feature_enabled & GF_XTIME)
+        marker_xtime_update_marks(this, local);
+out:
+    marker_local_unref(local);
+
+    return 0;
+}
+
+int32_t
+marker_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    int32_t ret = 0;
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    priv = this->private;
+
+    if (priv->feature_enabled == 0)
+        goto wind;
+
+    local = mem_get0(this->local_pool);
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    ret = loc_copy(&local->loc, loc);
+
+    if (ret == -1)
+        goto err;
+wind:
+    STACK_WIND(frame, marker_setattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setattr, loc, stbuf, valid, xdata);
+    return 0;
+err:
+    MARKER_STACK_UNWIND(setattr, frame, -1, ENOMEM, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int32_t
+marker_removexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_TRACE,
+               "%s occurred while "
+               "removing extended attribute",
+               strerror(op_errno));
+    }
+
+    local = (marker_local_t *)frame->local;
+
+    frame->local = NULL;
+
+    STACK_UNWIND_STRICT(removexattr, frame, op_ret, op_errno, xdata);
+
+    if (op_ret == -1 || local == NULL)
+        goto out;
+
+    priv = this->private;
+
+    if (priv->feature_enabled & GF_XTIME)
+        marker_xtime_update_marks(this, local);
+out:
+    marker_local_unref(local);
+
+    return 0;
+}
+
+int32_t
+marker_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   const char *name, dict_t *xdata)
+{
+    int32_t ret = -1;
+    int32_t i = 0;
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+    char key[QUOTA_KEY_MAX] = {
+        0,
+    };
+
+    priv = this->private;
+
+    if (name) {
+        for (i = 0; mq_ext_xattrs[i]; i++) {
+            if (strcmp(name, mq_ext_xattrs[i]))
+                continue;
+
+            GET_QUOTA_KEY(this, key, mq_ext_xattrs[i], ret);
+            if (ret < 0)
+                goto err;
+            name = key;
+            break;
+        }
+    }
+
+    if (priv->feature_enabled == 0)
+        goto wind;
+
+    local = mem_get0(this->local_pool);
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    ret = loc_copy(&local->loc, loc);
+
+    if (ret == -1)
+        goto err;
+wind:
+    STACK_WIND(frame, marker_removexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->removexattr, loc, name, xdata);
+    return 0;
+err:
+    MARKER_STACK_UNWIND(removexattr, frame, -1, ENOMEM, NULL);
+
+    return 0;
+}
+
+static gf_boolean_t
+__has_quota_xattrs(dict_t *xattrs)
+{
+    if (dict_foreach_match(xattrs, _is_quota_internal_xattr, NULL,
+                           dict_null_foreach_fn, NULL) > 0)
+        return _gf_true;
+
+    return _gf_false;
+}
+
+int32_t
+marker_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, inode_t *inode,
+                  struct iatt *buf, dict_t *dict, struct iatt *postparent)
+{
+    marker_conf_t *priv = NULL;
+    marker_local_t *local = NULL;
+    dict_t *xattrs = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    int32_t ret = -1;
+
+    priv = this->private;
+    local = (marker_local_t *)frame->local;
+    frame->local = NULL;
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_TRACE, "lookup failed with %s",
+               strerror(op_errno));
+        goto unwind;
+    }
+
+    ret = marker_key_set_ver(this, dict);
+    if (ret < 0) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+
+    if (dict && __has_quota_xattrs(dict)) {
+        xattrs = dict_copy_with_ref(dict, NULL);
+        if (!xattrs) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+        } else {
+            marker_filter_internal_xattrs(this, xattrs);
+        }
+    } else if (dict) {
+        xattrs = dict_ref(dict);
+    }
+
+    if (op_ret >= 0 && inode && (priv->feature_enabled & GF_QUOTA)) {
+        ctx = mq_inode_ctx_new(inode, this);
+        if (ctx == NULL) {
+            gf_log(this->name, GF_LOG_WARNING,
+                   "mq_inode_ctx_new "
+                   "failed for %s",
+                   uuid_utoa(inode->gfid));
+            op_ret = -1;
+            op_errno = ENOMEM;
+        }
+    }
+
+unwind:
+    STACK_UNWIND_STRICT(lookup, frame, op_ret, op_errno, inode, buf, xattrs,
+                        postparent);
+
+    if (op_ret == -1 || local == NULL)
+        goto out;
+
+    /* copy the gfid from the stat structure instead of inode,
+     * since if the lookup is fresh lookup, then the inode
+     * would have not yet linked to the inode table which happens
+     * in protocol/server.
+     */
+    if (gf_uuid_is_null(local->loc.gfid))
+        gf_uuid_copy(local->loc.gfid, buf->ia_gfid);
+
+    if (priv->feature_enabled & GF_QUOTA) {
+        mq_xattr_state(this, &local->loc, dict, buf);
+    }
+
+out:
+    marker_local_unref(local);
+    if (xattrs)
+        dict_unref(xattrs);
+
+    return 0;
+}
+
+int32_t
+marker_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc,
+              dict_t *xattr_req)
+{
+    int32_t ret = 0;
+    marker_local_t *local = NULL;
+    marker_conf_t *priv = NULL;
+
+    priv = this->private;
+
+    xattr_req = xattr_req ? dict_ref(xattr_req) : dict_new();
+    if (!xattr_req)
+        goto err;
+
+    ret = marker_key_replace_with_ver(this, xattr_req);
+    if (ret < 0)
+        goto err;
+
+    if (priv->feature_enabled == 0)
+        goto wind;
+
+    local = mem_get0(this->local_pool);
+    if (local == NULL)
+        goto err;
+
+    MARKER_INIT_LOCAL(frame, local);
+
+    ret = loc_copy(&local->loc, loc);
+    if (ret == -1)
+        goto err;
+
+    if ((priv->feature_enabled & GF_QUOTA))
+        mq_req_xattr(this, loc, xattr_req, NULL, NULL);
+
+wind:
+    STACK_WIND(frame, marker_lookup_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lookup, loc, xattr_req);
+
+    dict_unref(xattr_req);
+
+    return 0;
+err:
+    MARKER_STACK_UNWIND(lookup, frame, -1, ENOMEM, NULL, NULL, NULL, NULL);
+
+    if (xattr_req)
+        dict_unref(xattr_req);
+
+    return 0;
+}
+
+int
+marker_build_ancestry_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int op_ret, int op_errno, gf_dirent_t *entries,
+                          dict_t *xdata)
+{
+    gf_dirent_t *entry = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    int ret = -1;
+
+    if ((op_ret <= 0) || (entries == NULL)) {
+        goto out;
+    }
+
+    list_for_each_entry(entry, &entries->list, list)
+    {
+        if (entry->inode == NULL)
+            continue;
+
+        ret = marker_key_set_ver(this, entry->dict);
+        if (ret < 0) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            break;
+        }
+
+        ctx = mq_inode_ctx_new(entry->inode, this);
+        if (ctx == NULL)
+            gf_log(this->name, GF_LOG_WARNING,
+                   "mq_inode_ctx_new "
+                   "failed for %s",
+                   uuid_utoa(entry->inode->gfid));
+    }
+
+out:
+    STACK_UNWIND_STRICT(readdirp, frame, op_ret, op_errno, entries, xdata);
+    return 0;
+}
+
+int
+marker_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int op_ret, int op_errno, gf_dirent_t *entries,
+                    dict_t *xdata)
+{
+    gf_dirent_t *entry = NULL;
+    marker_conf_t *priv = NULL;
+    marker_local_t *local = NULL;
+    loc_t loc = {
+        0,
+    };
+    int ret = -1;
+    char *resolvedpath = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+
+    if (op_ret <= 0)
+        goto unwind;
+
+    priv = this->private;
+    local = frame->local;
+
+    if (!(priv->feature_enabled & GF_QUOTA) || (local == NULL)) {
+        goto unwind;
+    }
+
+    list_for_each_entry(entry, &entries->list, list)
+    {
+        if ((strcmp(entry->d_name, ".") == 0) ||
+            (strcmp(entry->d_name, "..") == 0) || entry->inode == NULL)
+            continue;
+
+        loc.parent = inode_ref(local->loc.inode);
+        loc.inode = inode_ref(entry->inode);
+        ret = inode_path(loc.parent, entry->d_name, &resolvedpath);
+        if (ret < 0) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "failed to get the "
+                   "path for the entry %s",
+                   entry->d_name);
+            loc_wipe(&loc);
+            continue;
+        }
+
+        loc.path = resolvedpath;
+        resolvedpath = NULL;
+
+        ctx = mq_inode_ctx_new(loc.inode, this);
+        if (ctx == NULL)
+            gf_log(this->name, GF_LOG_WARNING,
+                   "mq_inode_ctx_new "
+                   "failed for %s",
+                   uuid_utoa(loc.inode->gfid));
+
+        mq_xattr_state(this, &loc, entry->dict, &entry->d_stat);
+        loc_wipe(&loc);
+
+        ret = marker_key_set_ver(this, entry->dict);
+        if (ret < 0) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto unwind;
+        }
+    }
+
+unwind:
+    MARKER_STACK_UNWIND(readdirp, frame, op_ret, op_errno, entries, xdata);
+
+    return 0;
+}
+
+int
+marker_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                off_t offset, dict_t *dict)
+{
+    marker_conf_t *priv = NULL;
+    loc_t loc = {
+        0,
+    };
+    marker_local_t *local = NULL;
+    int ret = -1;
+
+    priv = this->private;
+
+    dict = dict ? dict_ref(dict) : dict_new();
+    if (!dict)
+        goto unwind;
+
+    ret = marker_key_replace_with_ver(this, dict);
+    if (ret < 0)
+        goto unwind;
+
+    if (dict_get(dict, GET_ANCESTRY_DENTRY_KEY)) {
+        STACK_WIND(frame, marker_build_ancestry_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->readdirp, fd, size, offset, dict);
+    } else {
+        if (priv->feature_enabled & GF_QUOTA) {
+            local = mem_get0(this->local_pool);
+
+            MARKER_INIT_LOCAL(frame, local);
+
+            loc.parent = local->loc.inode = inode_ref(fd->inode);
+
+            mq_req_xattr(this, &loc, dict, NULL, NULL);
+        }
+
+        STACK_WIND(frame, marker_readdirp_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->readdirp, fd, size, offset, dict);
+    }
+
+    dict_unref(dict);
+    return 0;
+unwind:
+    MARKER_STACK_UNWIND(readdirp, frame, -1, ENOMEM, NULL, NULL);
+    return 0;
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    if (!this)
+        return ret;
+
+    ret = xlator_mem_acct_init(this, gf_marker_mt_end + 1);
+
+    if (ret != 0) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Memory accounting init"
+               " failed");
+        return ret;
+    }
+
+    return ret;
+}
+
+int32_t
+init_xtime_priv(xlator_t *this, dict_t *options)
+{
+    int32_t ret = -1;
+    marker_conf_t *priv = NULL;
+    char *tmp_opt = NULL;
+
+    GF_VALIDATE_OR_GOTO("marker", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, options, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    priv = this->private;
+
+    ret = dict_get_str(options, "volume-uuid", &tmp_opt);
+
+    if (ret) {
+        priv->volume_uuid = NULL;
+        tmp_opt = "";
+
+        gf_log(this->name, GF_LOG_ERROR,
+               "please specify the volume-uuid"
+               "in the translator options");
+
+        return -1;
+    }
+    gf_asprintf(&priv->volume_uuid, "%s", tmp_opt);
+
+    ret = gf_uuid_parse(priv->volume_uuid, priv->volume_uuid_bin);
+
+    if (ret == -1) {
+        gf_log(this->name, GF_LOG_ERROR, "invalid volume uuid %s",
+               priv->volume_uuid);
+        goto out;
+    }
+
+    ret = gf_asprintf(&(priv->marker_xattr), "%s.%s.%s", MARKER_XATTR_PREFIX,
+                      priv->volume_uuid, XTIME);
+
+    if (ret == -1) {
+        priv->marker_xattr = NULL;
+        goto out;
+    }
+
+    gf_log(this->name, GF_LOG_DEBUG, "volume-uuid = %s", priv->volume_uuid);
+
+    ret = dict_get_str(options, "timestamp-file", &tmp_opt);
+    if (ret) {
+        priv->timestamp_file = NULL;
+        tmp_opt = "";
+
+        gf_log(this->name, GF_LOG_ERROR,
+               "please specify the timestamp-file"
+               "in the translator options");
+
+        goto out;
+    }
+
+    ret = gf_asprintf(&priv->timestamp_file, "%s", tmp_opt);
+    if (ret == -1) {
+        priv->timestamp_file = NULL;
+        goto out;
+    }
+
+    gf_log(this->name, GF_LOG_DEBUG, "the timestamp-file is = %s",
+           priv->timestamp_file);
+
+    ret = 0;
+out:
+    return ret;
+}
+
+void
+marker_xtime_priv_cleanup(xlator_t *this)
+{
+    marker_conf_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("marker", this, out);
+
+    priv = (marker_conf_t *)this->private;
+
+    GF_VALIDATE_OR_GOTO(this->name, priv, out);
+
+    GF_FREE(priv->volume_uuid);
+
+    GF_FREE(priv->timestamp_file);
+
+    GF_FREE(priv->marker_xattr);
+out:
+    return;
+}
+
+void
+marker_priv_cleanup(xlator_t *this)
+{
+    marker_conf_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("marker", this, out);
+
+    priv = (marker_conf_t *)this->private;
+
+    GF_VALIDATE_OR_GOTO(this->name, priv, out);
+
+    marker_xtime_priv_cleanup(this);
+
+    LOCK_DESTROY(&priv->lock);
+
+    GF_FREE(priv);
+
+    if (this->local_pool) {
+        mem_pool_destroy(this->local_pool);
+        this->local_pool = NULL;
+    }
+
+out:
+    return;
+}
+
+int32_t
+reconfigure(xlator_t *this, dict_t *options)
+{
+    int32_t ret = 0;
+    data_t *data = NULL;
+    gf_boolean_t flag = _gf_false;
+    marker_conf_t *priv = NULL;
+    int32_t version = 0;
+
+    GF_ASSERT(this);
+    GF_ASSERT(this->private);
+
+    priv = this->private;
+
+    priv->feature_enabled = 0;
+
+    GF_VALIDATE_OR_GOTO(this->name, options, out);
+
+    data = dict_get(options, "quota");
+    if (data) {
+        ret = gf_string2boolean(data->data, &flag);
+        if (ret == 0 && flag == _gf_true)
+            priv->feature_enabled |= GF_QUOTA;
+    }
+
+    data = dict_get(options, "inode-quota");
+    if (data) {
+        ret = gf_string2boolean(data->data, &flag);
+        if (ret == 0 && flag == _gf_true)
+            priv->feature_enabled |= GF_INODE_QUOTA;
+    }
+
+    data = dict_get(options, "quota-version");
+    if (data)
+        ret = gf_string2int32(data->data, &version);
+
+    if (priv->feature_enabled) {
+        if (version >= 0)
+            priv->version = version;
+        else
+            gf_log(this->name, GF_LOG_ERROR,
+                   "Invalid quota "
+                   "version %d",
+                   priv->version);
+    }
+
+    data = dict_get(options, "xtime");
+    if (data) {
+        ret = gf_string2boolean(data->data, &flag);
+        if (ret == 0 && flag == _gf_true) {
+            marker_xtime_priv_cleanup(this);
+
+            ret = init_xtime_priv(this, options);
+            if (ret < 0) {
+                gf_log(this->name, GF_LOG_WARNING,
+                       "failed to initialize xtime private, "
+                       "xtime updation will fail");
+            } else {
+                priv->feature_enabled |= GF_XTIME;
+                data = dict_get(options, "gsync-force-xtime");
+                if (!data)
+                    goto out;
+                ret = gf_string2boolean(data->data, &flag);
+                if (ret == 0 && flag)
+                    priv->feature_enabled |= GF_XTIME_GSYNC_FORCE;
+            }
+        }
+    }
+out:
+    return ret;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    dict_t *options = NULL;
+    data_t *data = NULL;
+    int32_t ret = 0;
+    gf_boolean_t flag = _gf_false;
+    marker_conf_t *priv = NULL;
+
+    if (!this->children) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "marker translator needs subvolume defined.");
+        return -1;
+    }
+
+    if (!this->parents) {
+        gf_log(this->name, GF_LOG_WARNING, "Volume is dangling.");
+        return -1;
+    }
+
+    options = this->options;
+
+    ALLOCATE_OR_GOTO(this->private, marker_conf_t, err);
+
+    priv = this->private;
+
+    priv->feature_enabled = 0;
+    priv->version = 0;
+
+    LOCK_INIT(&priv->lock);
+
+    data = dict_get(options, "quota");
+    if (data) {
+        ret = gf_string2boolean(data->data, &flag);
+        if (ret == 0 && flag == _gf_true)
+            priv->feature_enabled |= GF_QUOTA;
+    }
+
+    data = dict_get(options, "inode-quota");
+    if (data) {
+        ret = gf_string2boolean(data->data, &flag);
+        if (ret == 0 && flag == _gf_true)
+            priv->feature_enabled |= GF_INODE_QUOTA;
+    }
+
+    data = dict_get(options, "quota-version");
+    if (data)
+        ret = gf_string2int32(data->data, &priv->version);
+
+    if ((ret == 0) && priv->feature_enabled && priv->version < 0) {
+        gf_log(this->name, GF_LOG_ERROR, "Invalid quota version %d",
+               priv->version);
+        goto err;
+    }
+
+    data = dict_get(options, "xtime");
+    if (data) {
+        ret = gf_string2boolean(data->data, &flag);
+        if (ret == 0 && flag == _gf_true) {
+            ret = init_xtime_priv(this, options);
+            if (ret < 0)
+                goto err;
+
+            priv->feature_enabled |= GF_XTIME;
+            data = dict_get(options, "gsync-force-xtime");
+            if (!data)
+                goto cont;
+            ret = gf_string2boolean(data->data, &flag);
+            if (ret == 0 && flag)
+                priv->feature_enabled |= GF_XTIME_GSYNC_FORCE;
+        }
+    }
+
+cont:
+    this->local_pool = mem_pool_new(marker_local_t, 128);
+    if (!this->local_pool) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to create local_t's memory pool");
+        goto err;
+    }
+
+    return 0;
+err:
+    marker_priv_cleanup(this);
+
+    return -1;
+}
+
+int32_t
+marker_forget(xlator_t *this, inode_t *inode)
+{
+    marker_inode_ctx_t *ctx = NULL;
+    uint64_t value = 0;
+
+    if (inode_ctx_del(inode, this, &value) != 0)
+        goto out;
+
+    ctx = (marker_inode_ctx_t *)(unsigned long)value;
+    if (ctx == NULL) {
+        goto out;
+    }
+
+    mq_forget(this, ctx->quota_ctx);
+
+    GF_FREE(ctx);
+out:
+    return 0;
+}
+
+void
+fini(xlator_t *this)
+{
+    marker_priv_cleanup(this);
+}
+
+struct xlator_fops fops = {
+    .lookup = marker_lookup,
+    .create = marker_create,
+    .mkdir = marker_mkdir,
+    .writev = marker_writev,
+    .truncate = marker_truncate,
+    .ftruncate = marker_ftruncate,
+    .symlink = marker_symlink,
+    .link = marker_link,
+    .unlink = marker_unlink,
+    .rmdir = marker_rmdir,
+    .rename = marker_rename,
+    .mknod = marker_mknod,
+    .setxattr = marker_setxattr,
+    .fsetxattr = marker_fsetxattr,
+    .setattr = marker_setattr,
+    .fsetattr = marker_fsetattr,
+    .removexattr = marker_removexattr,
+    .getxattr = marker_getxattr,
+    .readdirp = marker_readdirp,
+    .fallocate = marker_fallocate,
+    .discard = marker_discard,
+    .zerofill = marker_zerofill,
+};
+
+struct xlator_cbks cbks = {.forget = marker_forget};
+
+struct volume_options options[] = {
+    {.key = {"volume-uuid"}, .default_value = "{{ volume.id }}"},
+    {.key = {"timestamp-file"}},
+    {
+        .key = {"quota"},
+        .op_version = {1},
+        .flags = OPT_FLAG_NONE,
+        .tags = {},
+    },
+    {
+        .key = {"inode-quota"},
+        .op_version = {1},
+        .flags = OPT_FLAG_NONE,
+        .tags = {},
+    },
+    {
+        .key = {"xtime"},
+        .op_version = {1},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_FORCE,
+        .tags = {},
+    },
+    {
+        .key = {"gsync-force-xtime"},
+        .op_version = {2},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_FORCE,
+        .tags = {},
+    },
+    {
+        .key = {"quota-version"},
+        .flags = OPT_FLAG_NONE,
+    },
+    {.key = {NULL}}};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "marker",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/features/marker/src/marker.h b/xlators/features/marker/src/marker.h
new file mode 100644
index 00000000000..4821094c14b
--- /dev/null
+++ b/xlators/features/marker/src/marker.h
@@ -0,0 +1,148 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#ifndef _MARKER_H
+#define _MARKER_H
+
+#include "marker-quota.h"
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/compat-uuid.h>
+#include <glusterfs/call-stub.h>
+
+#define MARKER_XATTR_PREFIX "trusted.glusterfs"
+#define XTIME "xtime"
+#define VOLUME_MARK "volume-mark"
+#define VOLUME_UUID "volume-uuid"
+#define TIMESTAMP_FILE "timestamp-file"
+
+enum {
+    GF_QUOTA = 1,
+    GF_XTIME = 2,
+    GF_XTIME_GSYNC_FORCE = 4,
+    GF_INODE_QUOTA = 8,
+};
+
+/*initialize the local variable*/
+#define MARKER_INIT_LOCAL(_frame, _local)                                      \
+    do {                                                                       \
+        _frame->local = _local;                                                \
+        _local->pid = _frame->root->pid;                                       \
+        memset(&_local->loc, 0, sizeof(loc_t));                                \
+        _local->ref = 1;                                                       \
+        _local->uid = -1;                                                      \
+        _local->gid = -1;                                                      \
+        LOCK_INIT(&_local->lock);                                              \
+        _local->oplocal = NULL;                                                \
+    } while (0)
+
+/* try alloc and if it fails, goto label */
+#define ALLOCATE_OR_GOTO(var, type, label)                                     \
+    do {                                                                       \
+        var = GF_CALLOC(sizeof(type), 1, gf_marker_mt_##type);                 \
+        if (!var) {                                                            \
+            gf_log(this->name, GF_LOG_ERROR, "out of memory :(");              \
+            goto label;                                                        \
+        }                                                                      \
+    } while (0)
+
+#define _MARKER_SET_UID_GID(dest, src)                                         \
+    do {                                                                       \
+        if (src->uid != -1 && src->gid != -1) {                                \
+            dest->uid = src->uid;                                              \
+            dest->gid = src->gid;                                              \
+        }                                                                      \
+    } while (0)
+
+#define MARKER_SET_UID_GID(frame, dest, src)                                   \
+    do {                                                                       \
+        _MARKER_SET_UID_GID(dest, src);                                        \
+        frame->root->uid = 0;                                                  \
+        frame->root->gid = 0;                                                  \
+        frame->cookie = (void *)_GF_UID_GID_CHANGED;                           \
+    } while (0)
+
+#define MARKER_RESET_UID_GID(frame, dest, src)                                 \
+    do {                                                                       \
+        _MARKER_SET_UID_GID(dest, src);                                        \
+        frame->cookie = NULL;                                                  \
+    } while (0)
+
+#define MARKER_STACK_UNWIND(fop, frame, params...)                             \
+    do {                                                                       \
+        quota_local_t *_local = NULL;                                          \
+        if (frame) {                                                           \
+            _local = frame->local;                                             \
+            frame->local = NULL;                                               \
+        }                                                                      \
+        STACK_UNWIND_STRICT(fop, frame, params);                               \
+        if (_local)                                                            \
+            marker_local_unref(_local);                                        \
+    } while (0)
+
+struct marker_local {
+    uint32_t timebuf[2];
+    pid_t pid;
+    loc_t loc;
+    loc_t parent_loc;
+    uid_t uid;
+    gid_t gid;
+    int32_t ref;
+    uint32_t ia_nlink;
+    struct iatt buf;
+    gf_lock_t lock;
+    mode_t mode;
+    int32_t err;
+    call_stub_t *stub;
+    call_frame_t *lk_frame;
+    quota_meta_t contribution;
+    struct marker_local *oplocal;
+
+    /* marker quota specific */
+    int64_t delta;
+    int64_t d_off;
+    int64_t sum;
+    int64_t size;
+    int32_t hl_count;
+    int32_t dentry_child_count;
+
+    fd_t *fd;
+    call_frame_t *frame;
+
+    quota_inode_ctx_t *ctx;
+    inode_contribution_t *contri;
+
+    int xflag;
+    dict_t *xdata;
+    gf_boolean_t skip_txn;
+};
+typedef struct marker_local marker_local_t;
+
+#define quota_local_t marker_local_t
+
+struct marker_inode_ctx {
+    struct quota_inode_ctx *quota_ctx;
+};
+typedef struct marker_inode_ctx marker_inode_ctx_t;
+
+struct marker_conf {
+    char feature_enabled;
+    char *size_key;
+    char *dirty_key;
+    char *volume_uuid;
+    uuid_t volume_uuid_bin;
+    char *timestamp_file;
+    char *marker_xattr;
+    uint64_t quota_lk_owner;
+    gf_lock_t lock;
+    int32_t version;
+};
+typedef struct marker_conf marker_conf_t;
+
+#endif
diff --git a/xlators/features/metadisp/Makefile.am b/xlators/features/metadisp/Makefile.am
new file mode 100644
index 00000000000..a985f42a877
--- /dev/null
+++ b/xlators/features/metadisp/Makefile.am
@@ -0,0 +1,3 @@
+SUBDIRS = src
+
+CLEANFILES =
diff --git a/xlators/features/metadisp/src/Makefile.am b/xlators/features/metadisp/src/Makefile.am
new file mode 100644
index 00000000000..1520ad8c424
--- /dev/null
+++ b/xlators/features/metadisp/src/Makefile.am
@@ -0,0 +1,38 @@
+noinst_PYTHON = gen-fops.py
+
+EXTRA_DIST = fops-tmpl.c
+
+xlator_LTLIBRARIES = metadisp.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+nodist_metadisp_la_SOURCES = fops.c
+
+BUILT_SOURCES = fops.c
+
+metadisp_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+metadisp_la_SOURCES = metadisp.c \
+		metadisp-unlink.c \
+		metadisp-stat.c \
+		metadisp-lookup.c \
+		metadisp-readdir.c \
+		metadisp-create.c \
+		metadisp-open.c \
+		metadisp-fsync.c \
+		metadisp-setattr.c \
+		backend.c
+
+metadisp_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = metadisp.h metadisp-fops.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+fops.c: fops-tmpl.c $(top_srcdir)/libglusterfs/src/generator.py gen-fops.py
+	PYTHONPATH=$(top_srcdir)/libglusterfs/src \
+	$(PYTHON) $(srcdir)/gen-fops.py $(srcdir)/fops-tmpl.c > $@
+
+CLEANFILES = $(nodist_metadisp_la_SOURCES)
diff --git a/xlators/features/metadisp/src/backend.c b/xlators/features/metadisp/src/backend.c
new file mode 100644
index 00000000000..ee2c25bfaa7
--- /dev/null
+++ b/xlators/features/metadisp/src/backend.c
@@ -0,0 +1,45 @@
+#define GFID_STR_LEN 37
+
+#include "metadisp.h"
+
+/*
+ * backend.c
+ *
+ * functions responsible for converting user-facing paths to backend-style
+ * "/$GFID" paths.
+ */
+
+int32_t
+build_backend_loc(uuid_t gfid, loc_t *src_loc, loc_t *dst_loc)
+{
+    static uuid_t root = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1};
+    char gfid_buf[GFID_STR_LEN + 1] = {
+        0,
+    };
+    char *path = NULL;
+
+    GF_VALIDATE_OR_GOTO("metadisp", src_loc, out);
+    GF_VALIDATE_OR_GOTO("metadisp", dst_loc, out);
+
+    loc_copy(dst_loc, src_loc);
+    memcpy(dst_loc->pargfid, root, sizeof(root));
+    GF_FREE((char *)dst_loc->path);  // we are overwriting path so nuke
+                                     // whatever loc_copy gave us
+
+    uuid_utoa_r(gfid, gfid_buf);
+
+    path = GF_CALLOC(GFID_STR_LEN + 1, sizeof(char),
+                     gf_common_mt_char);  // freed via loc_wipe
+
+    path[0] = '/';
+    strncpy(path + 1, gfid_buf, GFID_STR_LEN);
+    path[GFID_STR_LEN] = 0;
+    dst_loc->path = path;
+    if (src_loc->name)
+        dst_loc->name = strrchr(dst_loc->path, '/');
+    if (dst_loc->name)
+        dst_loc->name++;
+    return 0;
+out:
+    return -1;
+}
diff --git a/xlators/features/metadisp/src/fops-tmpl.c b/xlators/features/metadisp/src/fops-tmpl.c
new file mode 100644
index 00000000000..4385b7dd5b7
--- /dev/null
+++ b/xlators/features/metadisp/src/fops-tmpl.c
@@ -0,0 +1,10 @@
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include <glusterfs/xlator.h>
+#include "metadisp.h"
+#include "metadisp-fops.h"
+
+#pragma generate
diff --git a/xlators/features/metadisp/src/gen-fops.py b/xlators/features/metadisp/src/gen-fops.py
new file mode 100644
index 00000000000..8b5e120fdec
--- /dev/null
+++ b/xlators/features/metadisp/src/gen-fops.py
@@ -0,0 +1,160 @@
+#!/usr/bin/python
+
+import sys
+from generator import fop_subs, generate
+
+FN_METADATA_CHILD_GENERIC = """
+int32_t
+metadisp_@NAME@ (call_frame_t *frame, xlator_t *this,
+                 @LONG_ARGS@)
+{
+  METADISP_TRACE("@NAME@ metadata");
+  STACK_WIND (frame, default_@NAME@_cbk,
+              METADATA_CHILD(this), METADATA_CHILD(this)->fops->@NAME@,
+              @SHORT_ARGS@);
+  return 0;
+}
+"""
+
+FN_GENERIC_TEMPLATE = """
+int32_t
+metadisp_@NAME@ (call_frame_t *frame, xlator_t *this,
+                          @LONG_ARGS@)
+{
+  METADISP_TRACE("@NAME@ generic");
+  STACK_WIND (frame, default_@NAME@_cbk,
+                          DATA_CHILD(this), DATA_CHILD(this)->fops->@NAME@,
+                          @SHORT_ARGS@);
+  return 0;
+}
+"""
+
+FN_DATAFD_TEMPLATE = """
+int32_t
+metadisp_@NAME@ (call_frame_t *frame, xlator_t *this,
+                          @LONG_ARGS@)
+{
+  METADISP_TRACE("@NAME@ datafd");
+  xlator_t *child = NULL;
+  child = DATA_CHILD(this);
+  STACK_WIND (frame, default_@NAME@_cbk,
+                          child, child->fops->@NAME@,
+                          @SHORT_ARGS@);
+  return 0;
+}
+"""
+
+FN_DATALOC_TEMPLATE = """
+int32_t
+metadisp_@NAME@ (call_frame_t *frame, xlator_t *this,
+                          @LONG_ARGS@)
+{
+  METADISP_TRACE("@NAME@ dataloc");
+  loc_t backend_loc = {
+      0,
+  };
+  if (build_backend_loc(loc->gfid, loc, &backend_loc)) {
+      goto unwind;
+  }
+  xlator_t *child = NULL;
+  child = DATA_CHILD(this);
+  STACK_WIND (frame, default_@NAME@_cbk,
+                          child, child->fops->@NAME@,
+                          @SHORT_ARGS@);
+  return 0;
+
+unwind:
+  STACK_UNWIND_STRICT(lookup, frame, -1, EINVAL, NULL, NULL, NULL, NULL);
+  return 0;
+}
+"""
+
+FOPS_LINE_TEMPLATE = "\t.@NAME@ = metadisp_@NAME@,"
+
+skipped = [
+    "readdir",
+    "readdirp",
+    "lookup",
+    "fsync",
+    "stat",
+    "open",
+    "create",
+    "unlink",
+    "setattr",
+    # TODO: implement "inodelk",
+]
+
+
+def gen_fops():
+    done = skipped
+
+    #
+    # these are fops that wind to the DATA_CHILD
+    #
+    # NOTE: re-written in order from google doc:
+    #          https://docs.google.com/document/d/1KEwVtSNvDhs4qb63gWx2ulCp5GJjge77NGJk4p_Ms4Q
+    for name in [
+        "writev",
+        "readv",
+        "ftruncate",
+        "zerofill",
+        "discard",
+        "seek",
+        "fstat",
+    ]:
+        done = done + [name]
+        print(generate(FN_DATAFD_TEMPLATE, name, fop_subs))
+
+    for name in ["truncate"]:
+        done = done + [name]
+        print(generate(FN_DATALOC_TEMPLATE, name, fop_subs))
+
+    # these are fops that operate solely on dentries, folders,
+    # or extended attributes. Therefore, they must always
+    # wind to METADATA_CHILD and should never perform
+    # any path rewriting
+    #
+    # NOTE: re-written in order from google doc:
+    #          https://docs.google.com/document/d/1KEwVtSNvDhs4qb63gWx2ulCp5GJjge77NGJk4p_Ms4Q
+    for name in [
+        "mkdir",
+        "symlink",
+        "link",
+        "rename",
+        "mknod",
+        "opendir",
+        # "readdir,  # special-cased
+        # "readdirp, # special-cased
+        "fsyncdir",
+        # "setattr", # special-cased
+        "readlink",
+        "fentrylk",
+        "access",
+        # TODO: these wind to both,
+        # data for backend-attributes and metadata for the rest
+        "xattrop",
+        "setxattr",
+        "getxattr",
+        "removexattr",
+        "fgetxattr",
+        "fsetxattr",
+        "fremovexattr",
+    ]:
+
+        done = done + [name]
+        print(generate(FN_METADATA_CHILD_GENERIC, name, fop_subs))
+
+    print("struct xlator_fops fops = {")
+    for name in done:
+        print(generate(FOPS_LINE_TEMPLATE, name, fop_subs))
+
+    print("};")
+
+
+for l in open(sys.argv[1], "r").readlines():
+    if l.find("#pragma generate") != -1:
+        print("/* BEGIN GENERATED CODE - DO NOT MODIFY */")
+        gen_fops()
+        print("/* END GENERATED CODE */")
+    else:
+        print(l[:-1])
diff --git a/xlators/features/metadisp/src/metadisp-create.c b/xlators/features/metadisp/src/metadisp-create.c
new file mode 100644
index 00000000000..f8c9798dd59
--- /dev/null
+++ b/xlators/features/metadisp/src/metadisp-create.c
@@ -0,0 +1,101 @@
+#include "metadisp.h"
+#include <glusterfs/call-stub.h>
+
+/**
+ * Create, like stat, is a two-step process. We send a create
+ * to the METADATA_CHILD, then send another create to the DATA_CHILD.
+ *
+ * We do the metadata child first to ensure that the ACLs are enforced.
+ */
+
+int32_t
+metadisp_create_dentry_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno, fd_t *fd,
+                           inode_t *inode, struct iatt *buf,
+                           struct iatt *preparent, struct iatt *postparent,
+                           dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(create, frame, op_ret, op_errno, fd, inode, buf,
+                        preparent, postparent, xdata);
+    return 0;
+}
+
+int32_t
+metadisp_create_resume(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                       int32_t flags, mode_t mode, mode_t umask, fd_t *fd,
+                       dict_t *xdata)
+{
+    // create the backend data inode
+    STACK_WIND(frame, metadisp_create_dentry_cbk, DATA_CHILD(this),
+               DATA_CHILD(this)->fops->create, loc, flags, mode, umask, fd,
+               xdata);
+    return 0;
+}
+
+int32_t
+metadisp_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, fd_t *fd, inode_t *inode,
+                    struct iatt *buf, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata)
+{
+    METADISP_TRACE("%d %d", op_ret, op_errno);
+    call_stub_t *stub = cookie;
+    if (op_ret != 0) {
+        STACK_UNWIND_STRICT(create, frame, op_ret, op_errno, fd, inode, buf,
+                            preparent, postparent, xdata);
+        return 0;
+    }
+
+    if (stub == NULL) {
+        goto unwind;
+    }
+
+    if (stub->poison) {
+        call_stub_destroy(stub);
+        return 0;
+    }
+
+    call_resume(stub);
+    return 0;
+
+unwind:
+    STACK_UNWIND_STRICT(create, frame, -1, EINVAL, NULL, NULL, NULL, NULL, NULL,
+                        NULL);
+    return 0;
+}
+
+int32_t
+metadisp_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+                mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+    METADISP_TRACE(".");
+
+    loc_t backend_loc = {
+        0,
+    };
+    call_stub_t *stub = NULL;
+    uuid_t *gfid_req = NULL;
+
+    RESOLVE_GFID_REQ(xdata, gfid_req, out);
+
+    if (build_backend_loc(*gfid_req, loc, &backend_loc)) {
+        goto unwind;
+    }
+
+    frame->local = loc;
+
+    stub = fop_create_stub(frame, metadisp_create_resume, &backend_loc, flags,
+                           mode, umask, fd, xdata);
+
+    STACK_WIND_COOKIE(frame, metadisp_create_cbk, stub, METADATA_CHILD(this),
+                      METADATA_CHILD(this)->fops->create, loc, flags, mode,
+                      umask, fd, xdata);
+    return 0;
+
+unwind:
+    STACK_UNWIND_STRICT(create, frame, -1, EINVAL, NULL, NULL, NULL, NULL, NULL,
+                        NULL);
+    return 0;
+out:
+    return -1;
+}
diff --git a/xlators/features/metadisp/src/metadisp-fops.h b/xlators/features/metadisp/src/metadisp-fops.h
new file mode 100644
index 00000000000..56dd427cf34
--- /dev/null
+++ b/xlators/features/metadisp/src/metadisp-fops.h
@@ -0,0 +1,51 @@
+#ifndef GF_METADISP_FOPS_H_
+#define GF_METADISP_FOPS_H_
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/dict.h>
+#include <glusterfs/glusterfs.h>
+
+#include <sys/types.h>
+
+/* fops in here are defined in their own file. Every other fop is just defined
+ * inline of fops.c */
+
+int
+metadisp_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                 off_t off, dict_t *xdata);
+
+int
+metadisp_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                  off_t off, dict_t *dict);
+
+int
+metadisp_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata);
+
+int
+metadisp_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+                mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata);
+
+int
+metadisp_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+              fd_t *fd, dict_t *xdata);
+
+int
+metadisp_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata);
+
+int
+metadisp_inodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+                 loc_t *loc, int32_t cmd, struct gf_flock *lock, dict_t *xdata);
+
+int
+metadisp_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+               dict_t *xdata);
+
+int
+metadisp_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+                dict_t *xdata);
+
+int
+metadisp_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                 struct iatt *stbuf, int32_t valid, dict_t *xdata);
+
+#endif
diff --git a/xlators/features/metadisp/src/metadisp-fsync.c b/xlators/features/metadisp/src/metadisp-fsync.c
new file mode 100644
index 00000000000..2e46fa84eac
--- /dev/null
+++ b/xlators/features/metadisp/src/metadisp-fsync.c
@@ -0,0 +1,54 @@
+
+#include "metadisp.h"
+#include <glusterfs/call-stub.h>
+
+int32_t
+metadisp_fsync_resume(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                      int32_t flags, dict_t *xdata)
+{
+    STACK_WIND(frame, default_fsync_cbk, DATA_CHILD(this),
+               DATA_CHILD(this)->fops->fsync, fd, flags, xdata);
+    return 0;
+}
+
+int32_t
+metadisp_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                   struct iatt *postbuf, dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    if (cookie) {
+        stub = cookie;
+    }
+
+    if (op_ret != 0) {
+        goto unwind;
+    }
+
+    if (stub->poison) {
+        call_stub_destroy(stub);
+        stub = NULL;
+        return 0;
+    }
+
+    call_resume(stub);
+    return 0;
+
+unwind:
+    if (stub) {
+        call_stub_destroy(stub);
+    }
+    STACK_UNWIND_STRICT(fsync, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+    return 0;
+}
+
+int32_t
+metadisp_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+               dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    stub = fop_fsync_stub(frame, metadisp_fsync_resume, fd, flags, xdata);
+    STACK_WIND_COOKIE(frame, metadisp_fsync_cbk, stub, METADATA_CHILD(this),
+                      METADATA_CHILD(this)->fops->fsync, fd, flags, xdata);
+    return 0;
+}
diff --git a/xlators/features/metadisp/src/metadisp-lookup.c b/xlators/features/metadisp/src/metadisp-lookup.c
new file mode 100644
index 00000000000..27d90c9f746
--- /dev/null
+++ b/xlators/features/metadisp/src/metadisp-lookup.c
@@ -0,0 +1,90 @@
+#include "metadisp.h"
+#include <glusterfs/call-stub.h>
+
+/**
+ * Lookup, like stat, is a two-step process for grabbing the metadata details
+ * as well as the data details.
+ */
+
+int32_t
+metadisp_backend_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int32_t op_ret, int32_t op_errno, inode_t *inode,
+                            struct iatt *buf, dict_t *xdata,
+                            struct iatt *postparent)
+{
+    METADISP_TRACE("backend_lookup_cbk");
+    if (op_errno == ENOENT) {
+        op_errno = ENODATA;
+        op_ret = -1;
+    }
+    STACK_UNWIND_STRICT(lookup, frame, op_ret, op_errno, inode, buf, xdata,
+                        postparent);
+    return 0;
+}
+
+int32_t
+metadisp_backend_lookup_resume(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                               dict_t *xdata)
+{
+    METADISP_TRACE("backend_lookup_resume");
+    loc_t backend_loc = {
+        0,
+    };
+    if (build_backend_loc(loc->gfid, loc, &backend_loc)) {
+        goto unwind;
+    }
+
+    STACK_WIND(frame, metadisp_backend_lookup_cbk, DATA_CHILD(this),
+               DATA_CHILD(this)->fops->lookup, &backend_loc, xdata);
+    return 0;
+
+unwind:
+    STACK_UNWIND_STRICT(lookup, frame, -1, EINVAL, NULL, NULL, NULL, NULL);
+    return 0;
+}
+
+int32_t
+metadisp_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, inode_t *inode,
+                    struct iatt *buf, dict_t *xdata, struct iatt *postparent)
+{
+    METADISP_TRACE("%d %d", op_ret, op_errno);
+    call_stub_t *stub = NULL;
+    stub = cookie;
+
+    if (op_ret != 0) {
+        goto unwind;
+    }
+
+    if (!IA_ISREG(buf->ia_type)) {
+        goto unwind;
+    } else if (!stub) {
+        op_errno = EINVAL;
+        goto unwind;
+    }
+
+    METADISP_TRACE("resuming stub");
+
+    // memcpy(stub->args.loc.gfid, buf->ia_gfid, sizeof(uuid_t));
+    call_resume(stub);
+    return 0;
+unwind:
+    METADISP_TRACE("unwinding %d %d", op_ret, op_errno);
+    STACK_UNWIND_STRICT(lookup, frame, op_ret, op_errno, inode, buf, xdata,
+                        postparent);
+    if (stub) {
+        call_stub_destroy(stub);
+    }
+    return 0;
+}
+
+int32_t
+metadisp_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    METADISP_TRACE("lookup");
+    call_stub_t *stub = NULL;
+    stub = fop_lookup_stub(frame, metadisp_backend_lookup_resume, loc, xdata);
+    STACK_WIND_COOKIE(frame, metadisp_lookup_cbk, stub, METADATA_CHILD(this),
+                      METADATA_CHILD(this)->fops->lookup, loc, xdata);
+    return 0;
+}
diff --git a/xlators/features/metadisp/src/metadisp-open.c b/xlators/features/metadisp/src/metadisp-open.c
new file mode 100644
index 00000000000..64814afe636
--- /dev/null
+++ b/xlators/features/metadisp/src/metadisp-open.c
@@ -0,0 +1,70 @@
+#include <glusterfs/call-stub.h>
+#include "metadisp.h"
+
+int32_t
+metadisp_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
+{
+    METADISP_TRACE("got open results %d %d", op_ret, op_errno);
+
+    call_stub_t *stub = NULL;
+    if (cookie) {
+        stub = cookie;
+    }
+
+    if (op_ret != 0) {
+        goto unwind;
+    }
+
+    if (!stub) {
+        goto unwind;
+    }
+
+    if (stub->poison) {
+        call_stub_destroy(stub);
+        stub = NULL;
+        return 0;
+    }
+
+    call_resume(stub);
+    return 0;
+
+unwind:
+    if (stub) {
+        call_stub_destroy(stub);
+    }
+    STACK_UNWIND_STRICT(open, frame, op_ret, op_errno, fd, xdata);
+    return 0;
+}
+
+int32_t
+metadisp_open_resume(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                     int32_t flags, fd_t *fd, dict_t *xdata)
+{
+    STACK_WIND_COOKIE(frame, metadisp_open_cbk, NULL, DATA_CHILD(this),
+                      DATA_CHILD(this)->fops->open, loc, flags, fd, xdata);
+    return 0;
+}
+
+int32_t
+metadisp_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+              fd_t *fd, dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    loc_t backend_loc = {
+        0,
+    };
+
+    if (build_backend_loc(loc->gfid, loc, &backend_loc)) {
+        goto unwind;
+    }
+
+    stub = fop_open_stub(frame, metadisp_open_resume, &backend_loc, flags, fd,
+                         xdata);
+    STACK_WIND_COOKIE(frame, metadisp_open_cbk, stub, METADATA_CHILD(this),
+                      METADATA_CHILD(this)->fops->open, loc, flags, fd, xdata);
+    return 0;
+unwind:
+    STACK_UNWIND_STRICT(open, frame, -1, EINVAL, NULL, NULL);
+    return 0;
+}
diff --git a/xlators/features/metadisp/src/metadisp-readdir.c b/xlators/features/metadisp/src/metadisp-readdir.c
new file mode 100644
index 00000000000..5f840b1e88f
--- /dev/null
+++ b/xlators/features/metadisp/src/metadisp-readdir.c
@@ -0,0 +1,65 @@
+#include "metadisp.h"
+
+/**
+ * With a change to the posix xlator, readdir and readdirp are shockingly
+ * simple.
+ *
+ * The issue with separating the backend data of the files
+ * with the metadata is that readdirs must now read from multiple sources
+ * to coalesce the directory entries.
+ *
+ * The way we do this is to tell the METADATA_CHILD that when it's
+ * running readdirp, each file entry should have a stat wound to
+ * 'stat-source-of-truth'.
+ *
+ * see metadisp_stat for how it handles winds _from_posix.
+ */
+
+int32_t
+metadisp_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                 off_t off, dict_t *xdata)
+{
+    METADISP_TRACE(".");
+    /*
+     * Always use readdirp, even if the original was readdir. Why? Because NFS.
+     * There are multiple translations between Gluster, UNIX, and NFS stat
+     * structures in that path. One of them uses the type etc. from the stat
+     * structure, which is only filled in by readdirp. If we use readdir, the
+     * entries do actually go all the way back to the client and are visible in
+     * getdents, but then the readdir throws them away because of the
+     * uninitialized type.
+     */
+    GF_UNUSED int32_t ret;
+    if (!xdata) {
+        xdata = dict_new();
+    }
+
+    // ret = dict_set_int32 (xdata, "list-xattr", 1);
+
+    // I'm my own source of truth!
+    ret = dict_set_static_ptr(xdata, "stat-source-of-truth", (void *)this);
+
+    STACK_WIND(frame, default_readdirp_cbk, METADATA_CHILD(this),
+               METADATA_CHILD(this)->fops->readdirp, fd, size, off, xdata);
+
+    return 0;
+}
+
+int32_t
+metadisp_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                  off_t off, dict_t *xdata)
+{
+    METADISP_TRACE(".");
+    if (!xdata) {
+        xdata = dict_new();
+    }
+    GF_UNUSED int32_t ret;
+    // ret = dict_set_int32 (xdata, "list-xattr", 1);
+
+    // I'm my own source of truth!
+    ret = dict_set_static_ptr(xdata, "stat-source-of-truth", (void *)this);
+
+    STACK_WIND(frame, default_readdirp_cbk, METADATA_CHILD(this),
+               METADATA_CHILD(this)->fops->readdirp, fd, size, off, xdata);
+    return 0;
+}
diff --git a/xlators/features/metadisp/src/metadisp-setattr.c b/xlators/features/metadisp/src/metadisp-setattr.c
new file mode 100644
index 00000000000..6991cf644f3
--- /dev/null
+++ b/xlators/features/metadisp/src/metadisp-setattr.c
@@ -0,0 +1,90 @@
+#include "metadisp.h"
+#include <glusterfs/call-stub.h>
+
+int32_t
+metadisp_backend_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                             int32_t op_ret, int32_t op_errno,
+                             struct iatt *statpre, struct iatt *statpost,
+                             dict_t *xdata)
+
+{
+    METADISP_TRACE("backend_setattr_cbk");
+    if (op_errno == ENOENT) {
+        op_errno = ENODATA;
+        op_ret = -1;
+    }
+    STACK_UNWIND_STRICT(setattr, frame, op_ret, op_errno, statpre, statpost,
+                        xdata);
+    return 0;
+}
+
+int32_t
+metadisp_backend_setattr_resume(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                                struct iatt *stbuf, int32_t valid,
+                                dict_t *xdata)
+
+{
+    METADISP_TRACE("backend_setattr_resume");
+    loc_t backend_loc = {
+        0,
+    };
+    if (build_backend_loc(loc->gfid, loc, &backend_loc)) {
+        goto unwind;
+    }
+
+    STACK_WIND(frame, metadisp_backend_setattr_cbk, DATA_CHILD(this),
+               DATA_CHILD(this)->fops->setattr, &backend_loc, stbuf, valid,
+               xdata);
+    return 0;
+
+unwind:
+    STACK_UNWIND_STRICT(setattr, frame, -1, EINVAL, NULL, NULL, NULL);
+    return 0;
+}
+
+int32_t
+metadisp_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *statpre,
+                     struct iatt *statpost, dict_t *xdata)
+{
+    METADISP_TRACE("%d %d", op_ret, op_errno);
+    call_stub_t *stub = NULL;
+    stub = cookie;
+
+    if (op_ret != 0) {
+        goto unwind;
+    }
+
+    if (!IA_ISREG(statpost->ia_type)) {
+        goto unwind;
+    } else if (!stub) {
+        op_errno = EINVAL;
+        goto unwind;
+    }
+
+    METADISP_TRACE("resuming stub");
+    call_resume(stub);
+    return 0;
+unwind:
+    METADISP_TRACE("unwinding %d %d", op_ret, op_errno);
+    STACK_UNWIND_STRICT(setattr, frame, op_ret, op_errno, statpre, statpost,
+                        xdata);
+    if (stub) {
+        call_stub_destroy(stub);
+    }
+    return 0;
+}
+
+int32_t
+metadisp_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                 struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    METADISP_TRACE("setattr");
+    call_stub_t *stub = NULL;
+    stub = fop_setattr_stub(frame, metadisp_backend_setattr_resume, loc, stbuf,
+                            valid, xdata);
+    STACK_WIND_COOKIE(frame, metadisp_setattr_cbk, stub, METADATA_CHILD(this),
+                      METADATA_CHILD(this)->fops->setattr, loc, stbuf, valid,
+                      xdata);
+    return 0;
+}
diff --git a/xlators/features/metadisp/src/metadisp-stat.c b/xlators/features/metadisp/src/metadisp-stat.c
new file mode 100644
index 00000000000..b06d0dbcddd
--- /dev/null
+++ b/xlators/features/metadisp/src/metadisp-stat.c
@@ -0,0 +1,124 @@
+#include "metadisp.h"
+#include <glusterfs/call-stub.h>
+
+/**
+ * The stat flow in METADISP is complicated because we must
+ * do ensure a few things:
+ *    1. stat, on the path within the metadata layer,
+ *       MUST get the backend FD of the data layer.
+ *        --- we wind to the metadata layer, then the data layer.
+ *
+ *    2. the metadata layer MUST be able to ask the data
+ *       layer for stat information.
+ *        --- this is 'syncop-internal-from-posix'
+ *
+ *    3. when the metadata exists BUT the data is missing,
+ *       we MUST mark the backend file as bad and heal it.
+ */
+
+int32_t
+metadisp_stat_backend_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                          dict_t *xdata)
+{
+    METADISP_TRACE("got backend stat results %d %d", op_ret, op_errno);
+    if (op_errno == ENOENT) {
+        STACK_UNWIND_STRICT(open, frame, -1, ENODATA, NULL, NULL);
+        return 0;
+    }
+    STACK_UNWIND_STRICT(stat, frame, op_ret, op_errno, buf, xdata);
+    return 0;
+}
+
+int32_t
+metadisp_stat_resume(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                     dict_t *xdata)
+{
+    METADISP_TRACE("winding stat to path %s", loc->path);
+    if (gf_uuid_is_null(loc->gfid)) {
+        METADISP_TRACE("bad object, sending EUCLEAN");
+        STACK_UNWIND_STRICT(open, frame, -1, EUCLEAN, NULL, NULL);
+        return 0;
+    }
+
+    STACK_WIND(frame, metadisp_stat_backend_cbk, SECOND_CHILD(this),
+               SECOND_CHILD(this)->fops->stat, loc, xdata);
+    return 0;
+}
+
+int32_t
+metadisp_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                  dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+
+    METADISP_TRACE("got stat results %d %d", op_ret, op_errno);
+
+    if (cookie) {
+        stub = cookie;
+    }
+
+    if (op_ret != 0) {
+        goto unwind;
+    }
+
+    // only use the stub for the files
+    if (!IA_ISREG(buf->ia_type)) {
+        goto unwind;
+    }
+
+    if (stub->poison) {
+        call_stub_destroy(stub);
+        stub = NULL;
+        return 0;
+    }
+
+    call_resume(stub);
+    return 0;
+
+unwind:
+    if (stub) {
+        call_stub_destroy(stub);
+    }
+    STACK_UNWIND_STRICT(stat, frame, op_ret, op_errno, buf, xdata);
+    return 0;
+}
+
+int32_t
+metadisp_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    int32_t ret = 0;
+    loc_t backend_loc = {
+        0,
+    };
+    METADISP_FILTER_ROOT(stat, loc, xdata);
+
+    if (build_backend_loc(loc->gfid, loc, &backend_loc)) {
+        goto unwind;
+    }
+
+    if (dict_get_int32(xdata, "syncop-internal-from-posix", &ret) == 0) {
+        // if we've just been sent a stat from posix, then we know
+        // that we must send down a stat for a file to the second child.
+        //
+        // that means we can skip the stat for the first child and just
+        // send to the data disk.
+        METADISP_TRACE("got syncop-internal-from-posix");
+        STACK_WIND(frame, default_stat_cbk, DATA_CHILD(this),
+                   DATA_CHILD(this)->fops->stat, &backend_loc, xdata);
+        return 0;
+    }
+
+    // we do not know if the request is for a file, folder, etc. wind
+    // to first child to find out.
+    stub = fop_stat_stub(frame, metadisp_stat_resume, &backend_loc, xdata);
+    METADISP_TRACE("winding stat to first child %s", loc->path);
+    STACK_WIND_COOKIE(frame, metadisp_stat_cbk, stub, METADATA_CHILD(this),
+                      METADATA_CHILD(this)->fops->stat, loc, xdata);
+    return 0;
+unwind:
+    STACK_UNWIND_STRICT(stat, frame, -1, EINVAL, NULL, NULL);
+    return 0;
+}
diff --git a/xlators/features/metadisp/src/metadisp-unlink.c b/xlators/features/metadisp/src/metadisp-unlink.c
new file mode 100644
index 00000000000..1f6a8eb35ce
--- /dev/null
+++ b/xlators/features/metadisp/src/metadisp-unlink.c
@@ -0,0 +1,160 @@
+
+#include "metadisp.h"
+#include <glusterfs/call-stub.h>
+
+/**
+ * The unlink flow in metadisp is complicated because we must
+ * do ensure that UNLINK causes both the metadata objects
+ * to get removed and the data objects to get removed.
+ */
+
+int32_t
+metadisp_unlink_resume(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                       int xflag, dict_t *xdata)
+{
+    METADISP_TRACE("winding backend unlink to path %s", loc->path);
+    STACK_WIND(frame, default_unlink_cbk, DATA_CHILD(this),
+               DATA_CHILD(this)->fops->unlink, loc, xflag, xdata);
+    return 0;
+}
+
+int32_t
+metadisp_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata)
+{
+    METADISP_TRACE(". %d %d", op_ret, op_errno);
+
+    int ret = 0;
+    call_stub_t *stub = NULL;
+    int nlink = 0;
+
+    if (cookie) {
+        stub = cookie;
+    }
+
+    if (op_ret != 0) {
+        goto unwind;
+    }
+
+    if (stub->poison) {
+        call_stub_destroy(stub);
+        stub = NULL;
+        return 0;
+    }
+
+    ret = dict_get_uint32(xdata, GF_RESPONSE_LINK_COUNT_XDATA, &nlink);
+    if (ret != 0) {
+        op_errno = EINVAL;
+        op_ret = -1;
+        goto unwind;
+    }
+    METADISP_TRACE("frontend hardlink count %d %d", ret, nlink);
+    if (nlink > 1) {
+        goto unwind;
+    }
+
+    call_resume(stub);
+    return 0;
+
+unwind:
+    if (stub) {
+        call_stub_destroy(stub);
+    }
+    STACK_UNWIND_STRICT(unlink, frame, op_ret, op_errno, preparent, postparent,
+                        xdata);
+    return 0;
+}
+
+int32_t
+metadisp_unlink_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno, inode_t *inode,
+                           struct iatt *buf, dict_t *xdata,
+                           struct iatt *postparent)
+{
+    call_stub_t *stub = NULL;
+
+    if (cookie) {
+        stub = cookie;
+    }
+
+    if (op_ret != 0) {
+        goto unwind;
+    }
+
+    // fail fast on empty gfid so we don't loop forever
+    if (gf_uuid_is_null(buf->ia_gfid)) {
+        op_ret = -1;
+        op_errno = ENODATA;
+        goto unwind;
+    }
+
+    // fill gfid since the stub is incomplete
+    memcpy(stub->args.loc.gfid, buf->ia_gfid, sizeof(uuid_t));
+    memcpy(stub->args.loc.pargfid, postparent->ia_gfid, sizeof(uuid_t));
+
+    if (stub->poison) {
+        call_stub_destroy(stub);
+        stub = NULL;
+        return 0;
+    }
+
+    call_resume(stub);
+    return 0;
+
+unwind:
+    if (stub) {
+        call_stub_destroy(stub);
+    }
+    STACK_UNWIND_STRICT(unlink, frame, op_ret, op_errno, NULL, NULL, NULL);
+    return 0;
+}
+
+int32_t
+metadisp_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+                dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    loc_t backend_loc = {
+        0,
+    };
+
+    if (gf_uuid_is_null(loc->gfid)) {
+        METADISP_TRACE("winding lookup for unlink to path %s", loc->path);
+
+        // loop back to ourselves after a lookup
+        stub = fop_unlink_stub(frame, metadisp_unlink, loc, xflag, xdata);
+        STACK_WIND_COOKIE(frame, metadisp_unlink_lookup_cbk, stub,
+                          METADATA_CHILD(this),
+                          METADATA_CHILD(this)->fops->lookup, loc, xdata);
+        return 0;
+    }
+
+    if (build_backend_loc(loc->gfid, loc, &backend_loc)) {
+        goto unwind;
+    }
+
+    //
+    // ensure we get the link count on the unlink response, so we can
+    // account for hardlinks before winding to the backend.
+    // NOTE:
+    //   multiple xlators use GF_REQUEST_LINK_COUNT_XDATA. confirmation
+    //   is needed to ensure that multiple requests will work in the same
+    //   xlator stack.
+    //
+    if (!xdata) {
+        xdata = dict_new();
+    }
+    dict_set_int32(xdata, GF_REQUEST_LINK_COUNT_XDATA, 1);
+
+    METADISP_TRACE("winding frontend unlink to path %s", loc->path);
+    stub = fop_unlink_stub(frame, metadisp_unlink_resume, &backend_loc, xflag,
+                           xdata);
+
+    STACK_WIND_COOKIE(frame, metadisp_unlink_cbk, stub, METADATA_CHILD(this),
+                      METADATA_CHILD(this)->fops->unlink, loc, xflag, xdata);
+    return 0;
+unwind:
+    STACK_UNWIND_STRICT(unlink, frame, -1, EINVAL, NULL, NULL, NULL);
+    return 0;
+}
diff --git a/xlators/features/metadisp/src/metadisp.c b/xlators/features/metadisp/src/metadisp.c
new file mode 100644
index 00000000000..3c8f150cebc
--- /dev/null
+++ b/xlators/features/metadisp/src/metadisp.c
@@ -0,0 +1,46 @@
+#include <glusterfs/call-stub.h>
+
+#include "metadisp.h"
+#include "metadisp-fops.h"
+
+int32_t
+init(xlator_t *this)
+{
+    if (!this->children) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "not configured with children. exiting");
+        return -1;
+    }
+
+    if (!this->parents) {
+        gf_log(this->name, GF_LOG_WARNING, "dangling volume. check volfile ");
+    }
+
+    return 0;
+}
+
+void
+fini(xlator_t *this)
+{
+    return;
+}
+
+/* defined in fops.c */
+struct xlator_fops fops;
+
+struct xlator_cbks cbks = {};
+
+struct volume_options options[] = {
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .op_version = {1},
+    .identifier = "metadisp",
+    .category = GF_EXPERIMENTAL,
+};
diff --git a/xlators/features/metadisp/src/metadisp.h b/xlators/features/metadisp/src/metadisp.h
new file mode 100644
index 00000000000..c8fd7a13c04
--- /dev/null
+++ b/xlators/features/metadisp/src/metadisp.h
@@ -0,0 +1,45 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#ifndef GF_METADISP_H_
+#define GF_METADISP_H_
+
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/dict.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#define METADATA_CHILD(_this) FIRST_CHILD(_this)
+#define DATA_CHILD(_this) SECOND_CHILD(_this)
+
+int32_t
+build_backend_loc(uuid_t gfid, loc_t *src_loc, loc_t *dst_loc);
+
+#define METADISP_TRACE(_args...) gf_log("metadisp", GF_LOG_INFO, _args)
+
+#define METADISP_FILTER_ROOT(_op, _args...)                                    \
+    if (strcmp(loc->path, "/") == 0) {                                         \
+        STACK_WIND(frame, default_##_op##_cbk, METADATA_CHILD(this),           \
+                   METADATA_CHILD(this)->fops->_op, _args);                    \
+        return 0;                                                              \
+    }
+
+#define METADISP_FILTER_ROOT_BY_GFID(_op, _gfid, _args...)                     \
+    if (__is_root_gfid(_gfid)) {                                               \
+        STACK_WIND(frame, default_##_op##_cbk, METADATA_CHILD(this),           \
+                   METADATA_CHILD(this)->fops->_op, _args);                    \
+        return 0;                                                              \
+    }
+
+#define RESOLVE_GFID_REQ(_dict, _dest, _lbl)                                   \
+    VALIDATE_OR_GOTO(dict_get_ptr(_dict, "gfid-req", (void **)&_dest) == 0,    \
+                     _lbl)
+
+#endif /* __TEMPLATE_H__ */
diff --git a/xlators/features/namespace/Makefile.am b/xlators/features/namespace/Makefile.am
new file mode 100644
index 00000000000..a985f42a877
--- /dev/null
+++ b/xlators/features/namespace/Makefile.am
@@ -0,0 +1,3 @@
+SUBDIRS = src
+
+CLEANFILES =
diff --git a/xlators/features/namespace/src/Makefile.am b/xlators/features/namespace/src/Makefile.am
new file mode 100644
index 00000000000..e355d42cf4e
--- /dev/null
+++ b/xlators/features/namespace/src/Makefile.am
@@ -0,0 +1,17 @@
+xlator_LTLIBRARIES = namespace.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+namespace_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+namespace_la_SOURCES = namespace.c
+namespace_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = namespace.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src \
+	-I$(top_srcdir)/xlators/lib/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/namespace/src/namespace.c b/xlators/features/namespace/src/namespace.c
new file mode 100644
index 00000000000..86c5ebee900
--- /dev/null
+++ b/xlators/features/namespace/src/namespace.c
@@ -0,0 +1,1344 @@
+/*
+ * Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+ * This file is part of GlusterFS.
+ *
+ * This file is licensed to you under your choice of the GNU Lesser
+ * General Public License, version 3 or any later version (LGPLv3 or
+ * later), or the GNU General Public License, version 2 (GPLv2), in all
+ * cases as published by the Free Software Foundation.
+ *
+ * xlators/features/namespace:
+ *      This translator tags each request with a namespace hash,
+ *      which then can be used in later translators to track and
+ *      throttle fops per namespace.
+ */
+
+#include <sys/types.h>
+
+#include <glusterfs/defaults.h>
+#include <glusterfs/hashfn.h>
+#include <glusterfs/logging.h>
+#include "namespace.h"
+
+/* Return codes for common path parsing functions. */
+enum _path_parse_result {
+    PATH_PARSE_RESULT_NO_PATH = 0,
+    PATH_PARSE_RESULT_FOUND = 1,
+    PATH_PARSE_RESULT_IS_GFID = 2,
+};
+
+typedef enum _path_parse_result path_parse_result_t;
+
+/* Clean up an ns_local struct. Wipe a loc (its inode is ref'd, so we're good.)
+ */
+static inline void
+ns_local_cleanup(ns_local_t *local)
+{
+    if (!local) {
+        return;
+    }
+
+    loc_wipe(&local->loc);
+    GF_FREE(local);
+}
+
+/* Create a new ns_local. We ref the inode, fake a new loc struct, and stash
+ * the stub given to us. */
+static inline ns_local_t *
+ns_local_new(call_stub_t *stub, inode_t *inode)
+{
+    ns_local_t *local = NULL;
+    loc_t loc = {
+        0,
+    };
+
+    if (!stub || !inode) {
+        goto out;
+    }
+
+    local = GF_CALLOC(1, sizeof(ns_local_t), 0);
+    if (local == NULL) {
+        goto out;
+    }
+
+    /* Set up a fake loc_t struct to give to the getxattr call. */
+    gf_uuid_copy(loc.gfid, inode->gfid);
+    loc.inode = inode_ref(inode);
+
+    /* If for some reason inode_ref() fails, then just give up. */
+    if (!loc.inode) {
+        GF_FREE(local);
+        goto out;
+    }
+
+    local->stub = stub;
+    local->loc = loc;
+
+out:
+    return local;
+}
+
+/* Try parsing a path string. If the path string is a GFID, then return
+ * with PATH_PARSE_RESULT_IS_GFID. If we have no namespace (i.e. '/') then
+ * return PATH_PARSE_RESULT_NO_PATH and set the hash to 1. Otherwise, hash the
+ * namespace and store it in the info struct. */
+static path_parse_result_t
+parse_path(ns_info_t *info, const char *path)
+{
+    int len = 0;
+    const char *ns_begin = path;
+    const char *ns_end = NULL;
+
+    if (!path || strlen(path) == 0) {
+        return PATH_PARSE_RESULT_NO_PATH;
+    }
+
+    if (path[0] == '<') {
+        return PATH_PARSE_RESULT_IS_GFID;
+    }
+
+    /* Right now we only want the top-level directory, so
+     * skip the initial '/' and read until the next '/'. */
+    while (*ns_begin == '/') {
+        ns_begin++;
+    }
+
+    /* ns_end will point to the next '/' or NULL if there is no delimiting
+     * '/' (i.e. "/directory" or the top level "/") */
+    ns_end = strchr(ns_begin, '/');
+    len = ns_end ? (ns_end - ns_begin) : strlen(ns_begin);
+
+    if (len != 0) {
+        info->hash = SuperFastHash(ns_begin, len);
+    } else {
+        /* If our substring is empty, then we can hash '/' instead.
+         * '/' is used in the namespace config for the top-level
+         * namespace. */
+        info->hash = SuperFastHash("/", 1);
+    }
+
+    info->found = _gf_true;
+    return PATH_PARSE_RESULT_FOUND;
+}
+
+/* Cache namespace info stored in the stack (info) into the inode. */
+static int
+ns_inode_ctx_put(inode_t *inode, xlator_t *this, ns_info_t *info)
+{
+    ns_info_t *cached_ns_info = NULL;
+    uint64_t ns_as_64 = 0;
+    int ret = -1;
+
+    if (!inode || !this) {
+        gf_log(this ? this->name : "namespace", GF_LOG_WARNING,
+               "Need a valid inode and xlator to cache ns_info.");
+        ret = -1;
+        goto out;
+    }
+
+    cached_ns_info = GF_CALLOC(1, sizeof(ns_info_t), 0);
+
+    /* If we've run out of memory, then return ENOMEM. */
+    if (cached_ns_info == NULL) {
+        gf_log(this->name, GF_LOG_WARNING, "No memory to cache ns_info.");
+        ret = -(ENOMEM);
+        goto out;
+    }
+
+    *cached_ns_info = *info;
+    ns_as_64 = (uint64_t)(uintptr_t)cached_ns_info;
+
+    ret = inode_ctx_put(inode, this, ns_as_64);
+
+    if (ret) {
+        goto out;
+    }
+
+    ret = 0;
+out:
+    if (ret && cached_ns_info) {
+        GF_FREE(cached_ns_info);
+    }
+
+    return ret;
+}
+
+/* Retrieve namespace info cached in the inode into the stack for use in later
+ * translators. */
+static int
+ns_inode_ctx_get(inode_t *inode, xlator_t *this, ns_info_t *info)
+{
+    ns_info_t *cached_ns_info = NULL;
+    uint64_t ns_as_64 = 0;
+    int ret = -1;
+
+    if (!inode) {
+        ret = -ENOENT;
+        goto out;
+    }
+
+    ret = inode_ctx_get(inode, this, &ns_as_64);
+
+    if (!ret) {
+        cached_ns_info = (ns_info_t *)(uintptr_t)ns_as_64;
+        *info = *cached_ns_info;
+    }
+
+out:
+    return ret;
+}
+
+/* This callback is the top of the unwind path of our attempt to get the path
+ * manually from the posix translator. We'll try to parse the path returned
+ * if it exists, then cache the hash if possible. Then just return to the
+ * default stub that we provide in the local, since there's nothing else to do
+ * once we've gotten the namespace hash. */
+int32_t
+get_path_resume_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *dict,
+                    dict_t *xdata)
+{
+    path_parse_result_t ret = PATH_PARSE_RESULT_NO_PATH;
+    call_frame_t *resume_frame = NULL;
+    ns_local_t *local = NULL;
+    call_stub_t *stub = NULL;
+    ns_info_t *info = NULL;
+    char *path = NULL;
+
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    local = frame->local;
+
+    GF_VALIDATE_OR_GOTO(this->name, local, out);
+    stub = local->stub;
+
+    GF_VALIDATE_OR_GOTO(this->name, stub, out);
+    /* Get the ns_info from the frame that we will eventually resume,
+     * not the frame that we're going to destroy (frame). */
+    resume_frame = stub->frame;
+
+    GF_VALIDATE_OR_GOTO(this->name, resume_frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, resume_frame->root, out);
+    info = &resume_frame->root->ns_info;
+
+    GF_VALIDATE_OR_GOTO(this->name, dict, out);
+
+    /* If we get a value back for the GET_ANCESTRY_PATH_KEY, then we
+     * try to access it and parse it like a path. */
+    if (!op_ret && !dict_get_str(dict, GET_ANCESTRY_PATH_KEY, &path)) {
+        gf_log(this->name, GF_LOG_DEBUG, "G>P %s retrieved path %s",
+               uuid_utoa(local->loc.gfid), path);
+        /* Now let's parse a path, finally. */
+        ret = parse_path(info, path);
+    }
+
+    if (ret == PATH_PARSE_RESULT_FOUND) {
+        /* If we finally found namespace, then stash it. */
+        ns_inode_ctx_put(local->loc.inode, this, info);
+
+        gf_log(this->name, GF_LOG_DEBUG, "G>P %s %10u namespace found %s",
+               uuid_utoa(local->loc.inode->gfid), info->hash, path);
+    } else if (ret == PATH_PARSE_RESULT_NO_PATH) {
+        gf_log(this->name, GF_LOG_WARNING, "G>P %s has no path",
+               uuid_utoa(local->loc.inode->gfid));
+    } else if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "G>P %s winding failed, still have gfid",
+               uuid_utoa(local->loc.inode->gfid));
+    }
+
+out:
+    /* Make sure to clean up local finally. */
+
+    if (frame) {
+        frame->local = NULL;
+        STACK_DESTROY(frame->root);
+    }
+
+    if (local) {
+        ns_local_cleanup(local);
+    }
+
+    if (stub) {
+        call_resume(stub);
+    }
+
+    return 0;
+}
+
+/* This function tries first to set a namespace based on the information that
+ * it can retrieve from an `loc_t`. This includes first looking for a cached
+ * namespace in the inode, then trying to parse the path string in the `loc_t`
+ * struct. If this fails, then it will try to call inode_path. */
+static path_parse_result_t
+set_ns_from_loc(const char *fn, call_frame_t *frame, xlator_t *this, loc_t *loc)
+{
+    path_parse_result_t ret = PATH_PARSE_RESULT_NO_PATH;
+    ns_private_t *priv = (ns_private_t *)this->private;
+    ns_info_t *info = &frame->root->ns_info;
+    char *path = NULL;
+
+    info->hash = 0;
+    info->found = _gf_false;
+
+    if (!priv->tag_namespaces) {
+        return ret;
+    }
+
+    /* This is our first pass at trying to get a path. Try getting
+     * from the inode context, then from the loc's path itself. */
+    if (!loc || !loc->path || !loc->inode) {
+        ret = PATH_PARSE_RESULT_NO_PATH;
+    } else if (!ns_inode_ctx_get(loc->inode, this, info)) {
+        ret = PATH_PARSE_RESULT_FOUND;
+    } else {
+        ret = parse_path(info, loc->path);
+        gf_log(this->name, GF_LOG_DEBUG, "%s: LOC retrieved path %s", fn,
+               loc->path);
+
+        if (ret == PATH_PARSE_RESULT_FOUND) {
+            ns_inode_ctx_put(loc->inode, this, info);
+        }
+    }
+
+    /* Keep trying by calling inode_path next, making sure to copy
+    the loc's gfid into its inode if necessary. */
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        if (gf_uuid_is_null(loc->inode->gfid)) {
+            gf_uuid_copy(loc->inode->gfid, loc->gfid);
+        }
+
+        if (inode_path(loc->inode, NULL, &path) >= 0 && path) {
+            ret = parse_path(info, loc->path);
+            gf_log(this->name, GF_LOG_DEBUG, "%s: LOC retrieved path %s", fn,
+                   path);
+
+            if (ret == PATH_PARSE_RESULT_FOUND) {
+                ns_inode_ctx_put(loc->inode, this, info);
+            }
+        }
+
+        if (path) {
+            GF_FREE(path);
+        }
+    }
+
+    /* Report our status, and if we have a GFID, we'll eventually try a
+     * GET_ANCESTRY_PATH_KEY wind when we return from this function. */
+    if (ret == PATH_PARSE_RESULT_FOUND) {
+        gf_log(this->name, GF_LOG_DEBUG,
+               "%s: LOC %s %10u namespace found for %s", fn,
+               uuid_utoa(loc->inode->gfid), info->hash, loc->path);
+    } else if (ret == PATH_PARSE_RESULT_NO_PATH) {
+        gf_log(this->name, GF_LOG_WARNING, "%s: LOC has no path", fn);
+    } else if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        /* Make sure to copy the inode's gfid for the eventual wind. */
+        if (gf_uuid_is_null(loc->inode->gfid)) {
+            gf_uuid_copy(loc->inode->gfid, loc->gfid);
+        }
+
+        gf_log(this->name, GF_LOG_DEBUG, "%s: LOC %s winding, looking for path",
+               fn, uuid_utoa(loc->inode->gfid));
+    }
+
+    return ret;
+}
+
+/* This function tries first to set a namespace based on the information that
+ * it can retrieve from an `fd_t`. This includes first looking for a cached
+ * namespace in the inode, then trying to call inode_path manually. */
+static path_parse_result_t
+set_ns_from_fd(const char *fn, call_frame_t *frame, xlator_t *this, fd_t *fd)
+{
+    path_parse_result_t ret = PATH_PARSE_RESULT_NO_PATH;
+    ns_private_t *priv = (ns_private_t *)this->private;
+    ns_info_t *info = &frame->root->ns_info;
+    char *path = NULL;
+
+    info->hash = 0;
+    info->found = _gf_false;
+
+    if (!priv->tag_namespaces) {
+        return ret;
+    }
+
+    /* This is our first pass at trying to get a path. Try getting
+     * from the inode context, then inode_path. */
+    if (!fd || !fd->inode) {
+        ret = PATH_PARSE_RESULT_NO_PATH;
+    } else if (!ns_inode_ctx_get(fd->inode, this, info)) {
+        ret = PATH_PARSE_RESULT_FOUND;
+    } else if (inode_path(fd->inode, NULL, &path) >= 0 && path) {
+        ret = parse_path(info, path);
+        gf_log(this->name, GF_LOG_DEBUG, "%s: FD  retrieved path %s", fn, path);
+
+        if (ret == PATH_PARSE_RESULT_FOUND) {
+            ns_inode_ctx_put(fd->inode, this, info);
+        }
+    }
+
+    if (path) {
+        GF_FREE(path);
+    }
+
+    /* Report our status, and if we have a GFID, we'll eventually try a
+     * GET_ANCESTRY_PATH_KEY wind when we return from this function. */
+    if (ret == PATH_PARSE_RESULT_FOUND) {
+        gf_log(this->name, GF_LOG_DEBUG, "%s: FD  %s %10u namespace found", fn,
+               uuid_utoa(fd->inode->gfid), info->hash);
+    } else if (ret == PATH_PARSE_RESULT_NO_PATH) {
+        gf_log(this->name, GF_LOG_WARNING, "%s: FD  has no path", fn);
+    } else if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        gf_log(this->name, GF_LOG_DEBUG, "%s: FD  %s winding, looking for path",
+               fn, uuid_utoa(fd->inode->gfid));
+    }
+
+    return ret;
+}
+
+/* This macro does the work of winding down a call of `getxattr` in the case
+ * that we have to retrieve the path manually. It assumes that there is a label
+ * called `wind` and the existence of several basic variables (frame, this),
+ * but otherwise is general enough for any fop (fd- or loc-based.) */
+#define GET_ANCESTRY_PATH_WIND(fop, inode, args...)                            \
+    do {                                                                       \
+        ns_info_t *info = &frame->root->ns_info;                               \
+        call_frame_t *new_frame = NULL;                                        \
+        ns_local_t *local = NULL;                                              \
+        call_stub_t *stub = NULL;                                              \
+                                                                               \
+        gf_log(this->name, GF_LOG_DEBUG, "    %s winding, looking for path",   \
+               uuid_utoa(inode->gfid));                                        \
+                                                                               \
+        new_frame = create_frame(this, this->ctx->pool);                       \
+        if (!new_frame) {                                                      \
+            gf_log(this->name, GF_LOG_ERROR,                                   \
+                   "Cannot allocate new call frame.");                         \
+            goto wind;                                                         \
+        }                                                                      \
+                                                                               \
+        stub = fop_##fop##_stub(frame, default_##fop, args);                   \
+        if (!stub) {                                                           \
+            gf_log(this->name, GF_LOG_ERROR,                                   \
+                   "Cannot allocate function stub.");                          \
+            goto wind;                                                         \
+        }                                                                      \
+                                                                               \
+        new_frame->root->uid = 0;                                              \
+        new_frame->root->gid = 0;                                              \
+        /* Put a phony "not found" NS info into this call. */                  \
+        new_frame->root->ns_info = *info;                                      \
+                                                                               \
+        local = ns_local_new(stub, inode);                                     \
+        if (!local) {                                                          \
+            gf_log(this->name, GF_LOG_ERROR,                                   \
+                   "Cannot allocate function local.");                         \
+            goto wind;                                                         \
+        }                                                                      \
+                                                                               \
+        new_frame->local = local;                                              \
+        /* After allocating a new frame, a call stub (to                       \
+         * resume our current fop), and a local variables                      \
+         * struct (for our loc to getxattr and our resume                      \
+         * stub), call getxattr and unwind to get_path_resume_cbk.             \
+         */                                                                    \
+        STACK_WIND(new_frame, get_path_resume_cbk, FIRST_CHILD(this),          \
+                   FIRST_CHILD(this)->fops->getxattr, &local->loc,             \
+                   GET_ANCESTRY_PATH_KEY, NULL);                               \
+    } while (0)
+
+int32_t
+ns_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflags,
+         dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this, loc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(rmdir, loc->inode, loc, xflags, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_rmdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rmdir, loc, xflags, xdata);
+    return 0;
+}
+
+int32_t
+ns_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflags,
+          dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this, loc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(unlink, loc->inode, loc, xflags, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_unlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->unlink, loc, xflags, xdata);
+    return 0;
+}
+
+int32_t
+ns_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+          dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this,
+                                              newloc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(rename, newloc->inode, oldloc, newloc, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_rename_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
+    return 0;
+}
+
+int32_t
+ns_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+        dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this,
+                                              newloc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(link, newloc->inode, oldloc, newloc, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_link_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->link, oldloc, newloc, xdata);
+    return 0;
+}
+
+int32_t
+ns_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+         mode_t umask, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this, loc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(mkdir, loc->inode, loc, mode, umask, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_mkdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mkdir, loc, mode, umask, xdata);
+    return 0;
+}
+
+int32_t
+ns_symlink(call_frame_t *frame, xlator_t *this, const char *linkname,
+           loc_t *loc, mode_t umask, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this, loc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(symlink, loc->inode, linkname, loc, umask,
+                               xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_symlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->symlink, linkname, loc, umask, xdata);
+    return 0;
+}
+
+int32_t
+ns_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+         dev_t dev, mode_t umask, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this, loc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(mknod, loc->inode, loc, mode, dev, umask, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_mknod_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mknod, loc, mode, dev, umask, xdata);
+    return 0;
+}
+
+int32_t
+ns_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+          mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this, loc);
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(create, loc->inode, loc, flags, mode, umask, fd,
+                               xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_create_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->create, loc, flags, mode, umask, fd,
+               xdata);
+    return 0;
+}
+
+int32_t
+ns_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
+            int32_t valid, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(fsetattr, fd->inode, fd, stbuf, valid, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_fsetattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetattr, fd, stbuf, valid, xdata);
+    return 0;
+}
+
+int32_t
+ns_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *stbuf,
+           int32_t valid, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this, loc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(setattr, loc->inode, loc, stbuf, valid, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_setattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setattr, loc, stbuf, valid, xdata);
+    return 0;
+}
+
+int32_t
+ns_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+                dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(fremovexattr, fd->inode, fd, name, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_fremovexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fremovexattr, fd, name, xdata);
+    return 0;
+}
+
+int32_t
+ns_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               const char *name, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this, loc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(removexattr, loc->inode, loc, name, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_removexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->removexattr, loc, name, xdata);
+    return 0;
+}
+
+int32_t
+ns_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+            int32_t flags, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this, loc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(setxattr, loc->inode, loc, dict, flags, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_setxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setxattr, loc, dict, flags, xdata);
+    return 0;
+}
+
+int32_t
+ns_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+             int32_t flags, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(fsetxattr, fd->inode, fd, dict, flags, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_fsetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
+    return 0;
+}
+
+int32_t
+ns_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+            dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this, loc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(truncate, loc->inode, loc, offset, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_truncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+    return 0;
+}
+
+int32_t
+ns_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(ftruncate, fd->inode, fd, offset, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_ftruncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+    return 0;
+}
+
+int32_t
+ns_writev(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
+          int32_t count, off_t offset, uint32_t flags, struct iobref *iobref,
+          dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(writev, fd->inode, fd, vector, count, offset,
+                               flags, iobref, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_writev_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->writev, fd, vector, count, offset,
+               flags, iobref, xdata);
+    return 0;
+}
+
+int32_t
+ns_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this, loc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(lookup, loc->inode, loc, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_lookup_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lookup, loc, xdata);
+    return 0;
+}
+
+int32_t
+ns_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this, loc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(stat, loc->inode, loc, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_stat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->stat, loc, xdata);
+    return 0;
+}
+
+int32_t
+ns_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(fstat, fd->inode, fd, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_fstat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fstat, fd, xdata);
+    return 0;
+}
+
+int32_t
+ns_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+            dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this, loc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(readlink, loc->inode, loc, size, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_readlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readlink, loc, size, xdata);
+    return 0;
+}
+
+int32_t
+ns_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+          dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this, loc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(access, loc->inode, loc, mask, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_access_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->access, loc, mask, xdata);
+    return 0;
+}
+
+int32_t
+ns_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+        fd_t *fd, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(open, fd->inode, loc, flags, fd, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_open_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->open, loc, flags, fd, xdata);
+    return 0;
+}
+
+int32_t
+ns_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+         off_t offset, uint32_t flags, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(readv, fd->inode, fd, size, offset, flags,
+                               xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_readv_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readv, fd, size, offset, flags, xdata);
+    return 0;
+}
+
+int32_t
+ns_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(flush, fd->inode, fd, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_flush_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->flush, fd, xdata);
+    return 0;
+}
+
+int32_t
+ns_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+         dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(fsync, fd->inode, fd, datasync, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_fsync_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsync, fd, datasync, xdata);
+    return 0;
+}
+
+int32_t
+ns_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+           dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this, loc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(opendir, loc->inode, loc, fd, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_opendir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->opendir, loc, fd, xdata);
+    return 0;
+}
+
+int32_t
+ns_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+            dict_t *xdata)
+
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(fsyncdir, fd->inode, fd, datasync, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_fsyncdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsyncdir, fd, datasync, xdata);
+    return 0;
+}
+
+int32_t
+ns_rchecksum(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             int32_t len, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(rchecksum, fd->inode, fd, offset, len, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_rchecksum_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rchecksum, fd, offset, len, xdata);
+    return 0;
+}
+
+int32_t
+ns_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this, loc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(statfs, loc->inode, loc, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_statfs_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->statfs, loc, xdata);
+    return 0;
+}
+
+int32_t
+ns_inodelk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+           int32_t cmd, struct gf_flock *flock, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this, loc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(inodelk, loc->inode, volume, loc, cmd, flock,
+                               xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_inodelk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->inodelk, volume, loc, cmd, flock,
+               xdata);
+    return 0;
+}
+
+int32_t
+ns_finodelk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+            int32_t cmd, struct gf_flock *flock, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(finodelk, fd->inode, volume, fd, cmd, flock,
+                               xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_finodelk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->finodelk, volume, fd, cmd, flock,
+               xdata);
+    return 0;
+}
+
+int32_t
+ns_entrylk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+           const char *basename, entrylk_cmd cmd, entrylk_type type,
+           dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this, loc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(entrylk, loc->inode, volume, loc, basename, cmd,
+                               type, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, volume, loc, basename, cmd,
+               type, xdata);
+    return 0;
+}
+
+int32_t
+ns_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+            const char *basename, entrylk_cmd cmd, entrylk_type type,
+            dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(fentrylk, fd->inode, volume, fd, basename, cmd,
+                               type, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_fentrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fentrylk, volume, fd, basename, cmd,
+               type, xdata);
+    return 0;
+}
+
+int32_t
+ns_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+             dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(fgetxattr, fd->inode, fd, name, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_fgetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fgetxattr, fd, name, xdata);
+    return 0;
+}
+
+int32_t
+ns_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, const char *name,
+            dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this, loc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(getxattr, loc->inode, loc, name, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_getxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->getxattr, loc, name, xdata);
+    return 0;
+}
+
+int32_t
+ns_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+      struct gf_flock *flock, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(lk, fd->inode, fd, cmd, flock, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_lk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lk, fd, cmd, flock, xdata);
+    return 0;
+}
+
+int32_t
+ns_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+           off_t offset, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(readdir, fd->inode, fd, size, offset, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_readdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdir, fd, size, offset, xdata);
+
+    return 0;
+}
+
+int32_t
+ns_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t offset, dict_t *dict)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(readdirp, fd->inode, fd, size, offset, dict);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_readdirp_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdirp, fd, size, offset, dict);
+    return 0;
+}
+
+int32_t
+ns_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+           gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_loc(__FUNCTION__, frame, this, loc);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(xattrop, loc->inode, loc, flags, dict, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_xattrop_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->xattrop, loc, flags, dict, xdata);
+
+    return 0;
+}
+
+int32_t
+ns_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+            gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(fxattrop, fd->inode, fd, flags, dict, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_fxattrop_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fxattrop, fd, flags, dict, xdata);
+
+    return 0;
+}
+
+int32_t
+ns_getspec(call_frame_t *frame, xlator_t *this, const char *key, int32_t flag)
+{
+    STACK_WIND(frame, default_getspec_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->getspec, key, flag);
+    return 0;
+}
+
+int32_t
+ns_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t keep_size,
+             off_t offset, size_t len, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(fallocate, fd->inode, fd, keep_size, offset, len,
+                               xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_fallocate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fallocate, fd, keep_size, offset, len,
+               xdata);
+    return 0;
+}
+
+int32_t
+ns_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+           size_t len, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(discard, fd->inode, fd, offset, len, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_discard_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->discard, fd, offset, len, xdata);
+    return 0;
+}
+
+int32_t
+ns_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+            off_t len, dict_t *xdata)
+{
+    path_parse_result_t ret = set_ns_from_fd(__FUNCTION__, frame, this, fd);
+
+    if (ret == PATH_PARSE_RESULT_IS_GFID) {
+        GET_ANCESTRY_PATH_WIND(zerofill, fd->inode, fd, offset, len, xdata);
+        return 0;
+    }
+wind:
+    STACK_WIND(frame, default_zerofill_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->zerofill, fd, offset, len, xdata);
+    return 0;
+}
+
+int
+ns_forget(xlator_t *this, inode_t *inode)
+{
+    uint64_t ns_as_64 = 0;
+    ns_info_t *info = NULL;
+
+    inode_ctx_del(inode, this, &ns_as_64);
+
+    if (!ns_as_64) {
+        return 0;
+    }
+
+    info = (ns_info_t *)(uintptr_t)ns_as_64;
+    GF_FREE(info);
+
+    return 0;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    int32_t ret = -1;
+    ns_private_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO(GF_NAMESPACE, this, out);
+
+    if (!this->children || this->children->next) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "translator needs a single subvolume.");
+        goto out;
+    }
+
+    if (!this->parents) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "dangling volume. please check volfile.");
+        goto out;
+    }
+
+    priv = GF_CALLOC(1, sizeof(ns_private_t), 0);
+
+    if (!priv) {
+        gf_log(this->name, GF_LOG_ERROR, "Can't allocate ns_priv structure.");
+        goto out;
+    }
+
+    GF_OPTION_INIT("tag-namespaces", priv->tag_namespaces, bool, out);
+
+    gf_log(this->name, GF_LOG_INFO, "Namespace xlator loaded");
+    this->private = priv;
+    ret = 0;
+
+out:
+    if (ret) {
+        GF_FREE(priv);
+    }
+
+    return ret;
+}
+
+void
+fini(xlator_t *this)
+{
+    GF_FREE(this->private);
+}
+
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    int ret = -1;
+    ns_private_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+    GF_VALIDATE_OR_GOTO(this->name, options, out);
+
+    priv = (ns_private_t *)this->private;
+
+    GF_OPTION_RECONF("tag-namespaces", priv->tag_namespaces, options, bool,
+                     out);
+
+    ret = 0;
+out:
+    return ret;
+}
+
+struct xlator_fops fops = {
+    .lookup = ns_lookup,
+    .stat = ns_stat,
+    .fstat = ns_fstat,
+    .truncate = ns_truncate,
+    .ftruncate = ns_ftruncate,
+    .access = ns_access,
+    .readlink = ns_readlink,
+    .mknod = ns_mknod,
+    .mkdir = ns_mkdir,
+    .unlink = ns_unlink,
+    .rmdir = ns_rmdir,
+    .symlink = ns_symlink,
+    .rename = ns_rename,
+    .link = ns_link,
+    .create = ns_create,
+    .open = ns_open,
+    .readv = ns_readv,
+    .writev = ns_writev,
+    .flush = ns_flush,
+    .fsync = ns_fsync,
+    .opendir = ns_opendir,
+    .readdir = ns_readdir,
+    .readdirp = ns_readdirp,
+    .fsyncdir = ns_fsyncdir,
+    .statfs = ns_statfs,
+    .setxattr = ns_setxattr,
+    .getxattr = ns_getxattr,
+    .fsetxattr = ns_fsetxattr,
+    .fgetxattr = ns_fgetxattr,
+    .removexattr = ns_removexattr,
+    .fremovexattr = ns_fremovexattr,
+    .lk = ns_lk,
+    .inodelk = ns_inodelk,
+    .finodelk = ns_finodelk,
+    .entrylk = ns_entrylk,
+    .fentrylk = ns_fentrylk,
+    .rchecksum = ns_rchecksum,
+    .xattrop = ns_xattrop,
+    .fxattrop = ns_fxattrop,
+    .setattr = ns_setattr,
+    .fsetattr = ns_fsetattr,
+    .getspec = ns_getspec,
+    .fallocate = ns_fallocate,
+    .discard = ns_discard,
+    .zerofill = ns_zerofill,
+};
+
+struct xlator_cbks cbks = {
+    .forget = ns_forget,
+};
+
+struct xlator_dumpops dumpops;
+
+struct volume_options options[] = {
+    {
+        .key = {"tag-namespaces"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "off",
+        .description = "This option enables this translator's functionality "
+                       "that tags every fop with a namespace hash for later "
+                       "throttling, stats collection, logging, etc.",
+        .op_version = {GD_OP_VERSION_4_1_0},
+        .tags = {"namespace"},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+    },
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .reconfigure = reconfigure,
+    .op_version = {GD_OP_VERSION_3_12_0},
+    .dumpops = &dumpops,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "namespace",
+    .category = GF_TECH_PREVIEW,
+};
diff --git a/xlators/features/namespace/src/namespace.h b/xlators/features/namespace/src/namespace.h
new file mode 100644
index 00000000000..3a9b84d6426
--- /dev/null
+++ b/xlators/features/namespace/src/namespace.h
@@ -0,0 +1,23 @@
+#ifndef __NAMESPACE_H__
+#define __NAMESPACE_H__
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/call-stub.h>
+
+#define GF_NAMESPACE "namespace"
+
+typedef struct {
+    gf_boolean_t tag_namespaces;
+} ns_private_t;
+
+typedef struct {
+    loc_t loc;         /* We store a "fake" loc_t for the getxattr wind. */
+    call_stub_t *stub; /* A stub back to the function we're resuming. */
+} ns_local_t;
+
+#endif /* __NAMESPACE_H__ */
diff --git a/xlators/features/path-convertor/Makefile.am b/xlators/features/path-convertor/Makefile.am
deleted file mode 100644
index d471a3f9243..00000000000
--- a/xlators/features/path-convertor/Makefile.am
+++ /dev/null
@@ -1,3 +0,0 @@
-SUBDIRS = src
-
-CLEANFILES = 
diff --git a/xlators/features/path-convertor/src/Makefile.am b/xlators/features/path-convertor/src/Makefile.am
deleted file mode 100644
index 58cfed0f983..00000000000
--- a/xlators/features/path-convertor/src/Makefile.am
+++ /dev/null
@@ -1,14 +0,0 @@
-
-xlator_LTLIBRARIES = path-converter.la
-xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/testing/features
-
-path_converter_la_LDFLAGS = -module -avoidversion 
-
-path_converter_la_SOURCES = path.c
-path_converter_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la 
-
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS)\
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
-
-CLEANFILES = 
-
diff --git a/xlators/features/path-convertor/src/path-mem-types.h b/xlators/features/path-convertor/src/path-mem-types.h
deleted file mode 100644
index adecb1f94a0..00000000000
--- a/xlators/features/path-convertor/src/path-mem-types.h
+++ /dev/null
@@ -1,32 +0,0 @@
-/*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef __PATH_MEM_TYPES_H__
-#define __PATH_MEM_TYPES_H__
-
-#include "mem-types.h"
-
-enum gf_path_mem_types_ {
-        gf_path_mt_path_private_t = gf_common_mt_end + 1,
-        gf_path_mt_char,
-        gf_path_mt_regex_t,
-        gf_path_mt_end
-};
-#endif
-
diff --git a/xlators/features/path-convertor/src/path.c b/xlators/features/path-convertor/src/path.c
deleted file mode 100644
index e4f52211353..00000000000
--- a/xlators/features/path-convertor/src/path.c
+++ /dev/null
@@ -1,1239 +0,0 @@
-/*
-  Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-/* TODO: add gf_log to all the cases returning errors */
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-/**
- * xlators/features/path-translator:
- *    This translator converts the path it gets into user specified targets.
- */
-
-#include <sys/types.h>
-#include <regex.h>
-#include <time.h>
-#include <errno.h>
-#include "glusterfs.h"
-#include "xlator.h"
-#include "path-mem-types.h"
-
-typedef struct path_private
-{
-	int32_t this_len;
-	int32_t start_off;
-	int32_t end_off;
-	char *this;
-	char *that;
-	char *path;
-	regex_t *preg;
-} path_private_t;
-
-static char *
-name_this_to_that (xlator_t *xl, const char *path, const char *name)
-{
-	path_private_t *priv = xl->private;
-	char priv_path[ZR_PATH_MAX] = {0,};
-	char *tmp_name = NULL;
-	int32_t path_len = strlen (path);
-	int32_t name_len = strlen (name) - ZR_FILE_CONTENT_STRLEN;
-	int32_t total_len = path_len + name_len;
-	int32_t i = 0, j = 0;
-
-	if (path_len >= priv->end_off)
-		return (char *)name;
-
-	if (priv->end_off && (total_len > priv->end_off)) {
-		j = priv->start_off;
-		tmp_name = GF_CALLOC (1, (total_len +
-                                          ZR_FILE_CONTENT_STRLEN),
-                                      gf_path_mt_char);
-		ERR_ABORT (tmp_name);
-
-		/* Get the complete path for the file first */
-		strcpy (tmp_name, path);
-		strcat (tmp_name, name + ZR_FILE_CONTENT_STRLEN);
-
-		strncpy (priv_path, tmp_name, priv->start_off);
-		for (i = priv->start_off; i < priv->end_off; i++) {
-			if (tmp_name[i] == '/')
-				continue;
-			priv_path[j++] = tmp_name[i];
-		}
-		memcpy ((priv_path + j), 
-			(tmp_name + priv->end_off), 
-			(total_len - priv->end_off));
-		priv_path[(total_len - (priv->end_off - j))] = '\0';
-
-		strcpy (tmp_name, ZR_FILE_CONTENT_STR);
-		strcat (tmp_name, priv_path);
-
-		return tmp_name;
-	}
-
-	return (char *)name;
-}
-
-/* This function should return 
- *  NULL - 
- *  converted path - if path match
- *  same path - if it doesn't match
- */
-static char *
-path_this_to_that (xlator_t *xl, const char *path)
-{
-	path_private_t *priv = xl->private;
-	char *priv_path = NULL;
-	int32_t path_len = strlen (path);
-	int32_t i = 0, j = 0;
-
-	if (priv->end_off && (path_len > priv->start_off)) {
-		priv_path = GF_CALLOC (1, path_len, gf_path_mt_char);
-		ERR_ABORT (priv_path);
-
-		if (priv->start_off && (path_len > priv->start_off))
-			memcpy (priv_path, path, priv->start_off);
-		if (path_len > priv->end_off) {
-			j = priv->start_off;
-			for (i = priv->start_off; i < priv->end_off; i++) {
-				if (path[i] == '/')
-					continue;
-				priv_path[j++] = path[i];
-			}
-			memcpy ((priv_path + j), 
-				(path + priv->end_off), 
-				(path_len - priv->end_off));
-			priv_path[(path_len - (priv->end_off - j))] = '\0';
-		}
-		return priv_path;
-	}
-	return (char *)path;
-}
-
-int32_t 
-path_create_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 fd_t *fd,
-		 inode_t *inode,
-		 struct iatt *buf,
-                 struct iatt *preparent,
-                 struct iatt *postparent)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, fd, inode, buf);
-	return 0;
-}
-
-int32_t 
-path_open_cbk (call_frame_t *frame,
-	       void *cookie,
-	       xlator_t *this,
-	       int32_t op_ret,
-	       int32_t op_errno,
-	       fd_t *fd)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, fd);
-	return 0;
-}
-
-int32_t 
-path_getdents_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-		   dir_entry_t *entries,
-		   int32_t count)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, entries, count);
-	return 0;
-}
-
-int32_t 
-path_readdir_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  gf_dirent_t *buf)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, buf);
-	return 0;
-}
-
-
-int32_t 
-path_readlink_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-		   const char *buf,
-                   struct iatt *sbuf)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, buf, sbuf);
-	return 0;
-}
-
-int32_t 
-path_lookup_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 inode_t *inode,
-		 struct iatt *buf,
-		 dict_t *xattr,
-                 struct iatt *postparent)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, inode, buf, xattr);
-	return 0;
-}
-
-
-int32_t 
-path_symlink_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  inode_t *inode,
-                  struct iatt *buf,
-                  struct iatt *preparent,
-                  struct iatt *postparent)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, inode, buf);
-	return 0;
-}
-
-int32_t 
-path_mknod_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		inode_t *inode,
-                struct iatt *buf,
-                struct iatt *preparent,
-                struct iatt *postparent)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, inode, buf);
-	return 0;
-}
-  
-
-int32_t 
-path_mkdir_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		inode_t *inode,
-                struct iatt *buf,
-                struct iatt *preparent,
-                struct iatt *postparent)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, inode, buf);
-	return 0;
-}
-  
-int32_t 
-path_link_cbk (call_frame_t *frame,
-	       void *cookie,
-	       xlator_t *this,
-	       int32_t op_ret,
-	       int32_t op_errno,
-	       inode_t *inode,
-               struct iatt *buf,
-               struct iatt *preparent,
-               struct iatt *postparent)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, inode, buf);
-	return 0;
-}
-
-int32_t 
-path_opendir_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  fd_t *fd)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, fd);
-	return 0;
-}
-
-
-int32_t
-path_rename_buf_cbk (call_frame_t *frame,
-		     void *cookie,
-		     xlator_t *this,
-		     int32_t op_ret,
-		     int32_t op_errno,
-		     struct iatt *buf,
-                     struct iatt *preoldparent,
-                     struct iatt *postoldparent,
-                     struct iatt *prenewparent,
-                     struct iatt *postnewparent)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, buf);
-	return 0;
-}
-
-
-
-int32_t
-path_common_buf_cbk (call_frame_t *frame,
-		     void *cookie,
-		     xlator_t *this,
-		     int32_t op_ret,
-		     int32_t op_errno,
-		     struct iatt *buf)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, buf);
-	return 0;
-}
-
-int32_t 
-path_common_dict_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno,
-		      dict_t *dict)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, dict);
-	return 0;
-}
-
-int32_t 
-path_common_remove_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno,struct iatt *preparent,
-                        struct iatt *postparent)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-int32_t
-path_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                   int32_t op_ret, int32_t op_errno,struct iatt *prebuf,
-                   struct iatt *postbuf)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, prebuf, postbuf);
-	return 0;
-}
-
-
-int32_t
-path_common_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno)
-{
-	STACK_UNWIND (frame, op_ret, op_errno);
-	return 0;
-}
-
-/* */
-int32_t 
-path_lookup (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     dict_t *xattr_req)
-{
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-	
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	STACK_WIND (frame, path_lookup_cbk,
-		    FIRST_CHILD(this), 
-		    FIRST_CHILD(this)->fops->lookup, 
-		    loc, xattr_req);
-
-	loc->path = loc_path;	
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	return 0;
-}
-
-int32_t 
-path_stat (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *loc)
-{
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-	
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	STACK_WIND (frame, 
-		    path_common_buf_cbk, 
-		    FIRST_CHILD(this), 
-		    FIRST_CHILD(this)->fops->stat, 
-		    loc);
-  
-	loc->path = loc_path;	
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	return 0;
-}
-
-int32_t 
-path_readlink (call_frame_t *frame,
-	       xlator_t *this,
-	       loc_t *loc,
-	       size_t size)
-{
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-	
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	STACK_WIND (frame, 
-		    path_readlink_cbk,
-		    FIRST_CHILD(this), 
-		    FIRST_CHILD(this)->fops->readlink, 
-		    loc, 
-		    size);
-  
-	loc->path = loc_path;	
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	return 0;
-}
-
-int32_t 
-path_mknod (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc,
-	    mode_t mode,
-	    dev_t dev)
-{
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-	
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	STACK_WIND (frame, 
-		    path_mknod_cbk,
-		    FIRST_CHILD(this), 
-		    FIRST_CHILD(this)->fops->mknod, 
-		    loc, 
-		    mode, 
-		    dev);
-
-	loc->path = loc_path;	
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	return 0;
-}
-
-int32_t 
-path_mkdir (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc,
-	    mode_t mode)
-{
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-	
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	STACK_WIND (frame, 
-		    path_mkdir_cbk,
-		    FIRST_CHILD(this), 
-		    FIRST_CHILD(this)->fops->mkdir, 
-		    loc, 
-		    mode);
-
-	loc->path = loc_path;	
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	return 0;
-}
-
-int32_t 
-path_unlink (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc)
-{
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-	
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	STACK_WIND (frame, 
-		    path_common_remove_cbk,
-		    FIRST_CHILD(this), 
-		    FIRST_CHILD(this)->fops->unlink, 
-		    loc);
-
-	loc->path = loc_path;	
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	return 0;
-}
-
-int32_t 
-path_rmdir (call_frame_t *frame,
-	    xlator_t *this,
-	    loc_t *loc)
-{
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-	
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	STACK_WIND (frame, 
-		    path_common_remove_cbk,
-		    FIRST_CHILD(this), 
-		    FIRST_CHILD(this)->fops->rmdir, 
-		    loc);
-  
-	loc->path = loc_path;	
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	return 0;
-}
-
-int32_t 
-path_symlink (call_frame_t *frame,
-	      xlator_t *this,
-	      const char *linkpath,
-	      loc_t *loc)
-{
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-	
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	STACK_WIND (frame, 
-		    path_symlink_cbk,
-		    FIRST_CHILD(this), 
-		    FIRST_CHILD(this)->fops->symlink, 
-		    linkpath,
-		    loc);
-  
-	loc->path = loc_path;	
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	return 0;
-}
-
-int32_t 
-path_rename (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *oldloc,
-	     loc_t *newloc)
-{  
-	char *oldloc_path = (char *)oldloc->path;
-	char *tmp_oldloc_path = NULL;
-
-	char *newloc_path = (char *)newloc->path;
-	char *tmp_newloc_path = NULL;
-	
-	if (!(tmp_oldloc_path = path_this_to_that (this, oldloc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	oldloc->path = tmp_oldloc_path;
-
-	if (!(tmp_newloc_path = path_this_to_that (this, newloc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	newloc->path = tmp_newloc_path;
-
-	STACK_WIND (frame, 
-		    path_rename_buf_cbk,
-		    FIRST_CHILD(this), 
-		    FIRST_CHILD(this)->fops->rename, 
-		    oldloc,
-		    newloc);
-  
-	oldloc->path = oldloc_path;	
-	if (tmp_oldloc_path != oldloc_path)
-		GF_FREE (tmp_oldloc_path);
-
-	newloc->path = newloc_path;	
-	if (tmp_newloc_path != newloc_path)
-		GF_FREE (tmp_newloc_path);
-
-	return 0;
-}
-
-int32_t 
-path_link (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *oldloc,
-	   loc_t *newloc)
-{
-	char *oldloc_path = (char *)oldloc->path;
-	char *tmp_oldloc_path = NULL;
-
-	char *newloc_path = (char *)newloc->path;
-	char *tmp_newloc_path = NULL;
-	
-	if (!(tmp_oldloc_path = path_this_to_that (this, oldloc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	oldloc->path = tmp_oldloc_path;
-
-	if (!(tmp_newloc_path = path_this_to_that (this, newloc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	newloc->path = tmp_newloc_path;
-
-	STACK_WIND (frame, 
-		    path_link_cbk, 
-		    FIRST_CHILD(this), 
-		    FIRST_CHILD(this)->fops->link, 
-		    oldloc, 
-		    newloc);
-
-	oldloc->path = oldloc_path;	
-	if (tmp_oldloc_path != oldloc_path)
-		GF_FREE (tmp_oldloc_path);
-
-	newloc->path = newloc_path;	
-	if (tmp_newloc_path != newloc_path)
-		GF_FREE (tmp_newloc_path);
-
-	return 0;
-}
-
-int32_t 
-path_setattr_cbk (call_frame_t *frame,
-                  void *cookie,
-                  xlator_t *this,
-                  int32_t op_ret,
-                  int32_t op_errno,
-                  struct iatt *preop,
-                  struct iatt *postop)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, preop, postop);
-	return 0;
-}
-
-int32_t
-path_setattr (call_frame_t *frame,
-              xlator_t *this,
-              loc_t *loc,
-              struct iatt *stbuf,
-              int32_t valid)
-{
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	STACK_WIND (frame,
-		    path_setattr_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->setattr,
-		    loc,
-		    stbuf, valid);
-
-	loc->path = loc_path;
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	return 0;
-}
-
-
-int32_t 
-path_truncate (call_frame_t *frame,
-	       xlator_t *this,
-	       loc_t *loc,
-	       off_t offset)
-{
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-	
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	STACK_WIND (frame, 
-		    path_truncate_cbk,
-		    FIRST_CHILD(this), 
-		    FIRST_CHILD(this)->fops->truncate, 
-		    loc, 
-		    offset);
-  
-	loc->path = loc_path;	
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	return 0;
-}
-
-
-int32_t 
-path_open (call_frame_t *frame,
-	   xlator_t *this,
-	   loc_t *loc,
-	   int32_t flags,
-	   fd_t *fd,
-           int32_t wbflags)
-{
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-	
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	STACK_WIND (frame, 
-		    path_open_cbk, 
-		    FIRST_CHILD(this), 
-		    FIRST_CHILD(this)->fops->open, 
-		    loc, 
-		    flags,
-		    fd,
-                    wbflags);
-
-	loc->path = loc_path;	
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	return 0;
-}
-
-int32_t 
-path_create (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     int32_t flags,
-	     mode_t mode,
-	     fd_t *fd)
-{
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-	
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	STACK_WIND (frame, 
-		    path_create_cbk, 
-		    FIRST_CHILD(this), 
-		    FIRST_CHILD(this)->fops->create, 
-		    loc, 
-		    flags,
-		    mode,
-		    fd);
-
-	loc->path = loc_path;	
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	return 0;
-}
-
-int32_t 
-path_setxattr (call_frame_t *frame,
-	       xlator_t *this,
-	       loc_t *loc,
-	       dict_t *dict,
-	       int32_t flags)
-{
-	char *tmp_name = NULL;
-	data_pair_t *trav = dict->members_list;
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-	
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	if (ZR_FILE_CONTENT_REQUEST(trav->key)) {
-		tmp_name = name_this_to_that (this, loc->path, trav->key);
-		if (tmp_name != trav->key) {
-			trav->key = tmp_name;
-		} else {
-			tmp_name = NULL;
-		}
-	}
-
-	STACK_WIND (frame, 
-		    path_common_cbk, 
-		    FIRST_CHILD(this), 
-		    FIRST_CHILD(this)->fops->setxattr, 
-		    loc, 
-		    dict,
-		    flags);
-
-	loc->path = loc_path;	
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	if (tmp_name)
-		GF_FREE (tmp_name);
-
-	return 0;
-}
-
-int32_t 
-path_getxattr (call_frame_t *frame,
-	       xlator_t *this,
-	       loc_t *loc,
-	       const char *name)
-{
-	char *tmp_name = (char *)name;
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-	
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	if (ZR_FILE_CONTENT_REQUEST(name)) {
-		tmp_name = name_this_to_that (this, loc->path, name);
-	}
-
-	STACK_WIND (frame, 
-		    path_common_dict_cbk, 
-		    FIRST_CHILD(this), 
-		    FIRST_CHILD(this)->fops->getxattr,
-		    loc, 
-		    tmp_name);
-
-	loc->path = loc_path;	
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	if (tmp_name != name)
-		GF_FREE (tmp_name);
-
-	return 0;
-}
-
-int32_t 
-path_removexattr (call_frame_t *frame,
-		  xlator_t *this,
-		  loc_t *loc,
-		  const char *name)
-{
-	char *tmp_name = (char *)name;
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-	
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	if (ZR_FILE_CONTENT_REQUEST(name)) {
-		tmp_name = name_this_to_that (this, loc->path, name);
-	}
-
-	STACK_WIND (frame, 
-		    path_common_cbk, 
-		    FIRST_CHILD(this), 
-		    FIRST_CHILD(this)->fops->removexattr, 
-		    loc, 
-		    tmp_name);
-
-	loc->path = loc_path;	
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	if (tmp_name != name)
-		GF_FREE (tmp_name);
-
-	return 0;
-}
-
-int32_t 
-path_opendir (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      fd_t *fd)
-{
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-	
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	STACK_WIND (frame, 
-		    path_opendir_cbk, 
-		    FIRST_CHILD(this), 
-		    FIRST_CHILD(this)->fops->opendir, 
-		    loc, 
-		    fd);
-
-	loc->path = loc_path;	
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	return 0;
-}
-
-int32_t 
-path_access (call_frame_t *frame,
-	     xlator_t *this,
-	     loc_t *loc,
-	     int32_t mask)
-{
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-	
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	STACK_WIND (frame, 
-		    path_common_cbk, 
-		    FIRST_CHILD(this), 
-		    FIRST_CHILD(this)->fops->access, 
-		    loc, 
-		    mask);
-
-	loc->path = loc_path;
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	return 0;
-}
-
-int32_t
-path_checksum_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-		   uint8_t *fchecksum,
-		   uint8_t *dchecksum)
-{
-	STACK_UNWIND (frame, op_ret, op_errno, fchecksum, dchecksum);
-	return 0;
-}
-
-int32_t
-path_checksum (call_frame_t *frame,
-	       xlator_t *this,
-	       loc_t *loc,
-	       int32_t flag)
-{
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-	
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	STACK_WIND (frame,
-		    path_checksum_cbk,
-		    FIRST_CHILD(this), 
-		    FIRST_CHILD(this)->fops->checksum, 
-		    loc, 
-		    flag);
-
-	loc->path = loc_path;	
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	return 0;
-}
-
-
-int32_t
-path_entrylk (call_frame_t *frame, xlator_t *this,
-	      const char *volume, loc_t *loc, const char *basename,
-	      entrylk_cmd cmd, entrylk_type type)
-{
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-	
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	STACK_WIND (frame, path_common_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->entrylk,
-		    volume, loc, basename, cmd, type);
-
-	loc->path = loc_path;	
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	return 0;
-}
-
-int32_t
-path_inodelk (call_frame_t *frame, xlator_t *this,
-              const char *volume, loc_t *loc, int32_t cmd, struct gf_flock *lock)
-{
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-	
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	STACK_WIND (frame,
-		    path_common_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->inodelk,
-		    volume, loc, cmd, lock);
-
-	loc->path = loc_path;	
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	return 0;
-}
-
-
-int32_t
-path_xattrop (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      gf_xattrop_flags_t flags,
-	      dict_t *dict)
-{
-	char *loc_path = (char *)loc->path;
-	char *tmp_path = NULL;
-	
-	if (!(tmp_path = path_this_to_that (this, loc->path))) {
-		STACK_UNWIND (frame, -1, ENOENT, NULL, NULL);
-		return 0;
-	}
-	loc->path = tmp_path;
-
-	STACK_WIND (frame,
-		    path_common_dict_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->xattrop,
-		    loc,
-		    flags,
-		    dict);
-
-	loc->path = loc_path;	
-	if (tmp_path != loc_path)
-		GF_FREE (tmp_path);
-
-	return 0;
-}
-
-int32_t
-mem_acct_init (xlator_t *this)
-{
-        int     ret = -1;
-
-        if (!this)
-                return ret;
-
-        ret = xlator_mem_acct_init (this, gf_path_mt_end + 1);
-        
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
-                        "failed");
-                return ret;
-        }
-
-        return ret;
-}
-
-int32_t 
-init (xlator_t *this)
-{
-	dict_t *options = this->options;
-	path_private_t *priv = NULL;
-
-	if (!this->children || this->children->next) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"path translator requires exactly one subvolume");
-		return -1;
-	}
-    
-	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"dangling volume. check volfile ");
-	}
-  
-	priv = GF_CALLOC (1, sizeof (*priv), gf_path_mt_path_private_t);
-	ERR_ABORT (priv);
-	if (dict_get (options, "start-offset")) {
-		priv->start_off = data_to_int32 (dict_get (options, 
-							   "start-offset"));
-	}
-	if (dict_get (options, "end-offset")) {
-		priv->end_off = data_to_int32 (dict_get (options, 
-							 "end-offset"));
-	}
-
-	if (dict_get (options, "regex")) {
-		int32_t ret = 0;
-		priv->preg = GF_CALLOC (1, sizeof (regex_t),
-                                        gf_path_mt_regex_t);
-		ERR_ABORT (priv->preg);
-		ret = regcomp (priv->preg, 
-			       data_to_str (dict_get (options, "regex")), 
-			       REG_EXTENDED);
-		if (ret) {
-			gf_log (this->name, GF_LOG_ERROR, 
-				"Failed to compile the 'option regex'");
-			GF_FREE (priv);
-			return -1;
-		}
-		if (dict_get (options, "replace-with")) {
-			priv->that = data_to_str (dict_get (options, 
-							    "replace-with"));
-		} else {
-			priv->that = "";
-		}
-	}
-
-	this->private = priv;
-	return 0;
-}
-
-void
-fini (xlator_t *this)
-{
-	return;
-}
-
-struct xlator_fops fops = {
-	.stat        = path_stat,
-	.readlink    = path_readlink,
-	.mknod       = path_mknod,
-	.mkdir       = path_mkdir,
-	.unlink      = path_unlink,
-	.rmdir       = path_rmdir,
-	.symlink     = path_symlink,
-	.rename      = path_rename,
-	.link        = path_link,
-	.truncate    = path_truncate,
-	.open        = path_open,
-	.setxattr    = path_setxattr,
-	.getxattr    = path_getxattr,
-	.removexattr = path_removexattr,
-	.opendir     = path_opendir,
-	.access      = path_access,
-	.create      = path_create,
-	.lookup      = path_lookup,
-	.checksum    = path_checksum,
-	.xattrop     = path_xattrop,
-	.entrylk     = path_entrylk,
-	.inodelk     = path_inodelk,
-        .setattr     = path_setattr,
-};
-
-struct xlator_cbks cbks = {
-};
-
-struct volume_options options[] = { 
-	{ .key  = {"start-offset"}, 
-	  .type = GF_OPTION_TYPE_INT, 
-	  .min  = 0, 
-	  .max  = 4095 
-	},
-	{ .key  = {"end-offset"}, 
-	  .type = GF_OPTION_TYPE_INT, 
-	  .min  = 1, 
-	  .max  = 4096 
-	},
-	{ .key  = {"replace-with"}, 
-	  .type = GF_OPTION_TYPE_ANY 
-	},
-	{ .key  = {NULL} },
-};
diff --git a/xlators/features/quiesce/src/Makefile.am b/xlators/features/quiesce/src/Makefile.am
index e8ab4cb2417..74ea999c045 100644
--- a/xlators/features/quiesce/src/Makefile.am
+++ b/xlators/features/quiesce/src/Makefile.am
@@ -1,14 +1,16 @@
 xlator_LTLIBRARIES = quiesce.la
 xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
 
-quiesce_la_LDFLAGS = -module -avoidversion
+quiesce_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
 
 quiesce_la_SOURCES = quiesce.c
 quiesce_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
 
-noinst_HEADERS = quiesce.h quiesce-mem-types.h
+noinst_HEADERS = quiesce.h quiesce-mem-types.h quiesce-messages.h
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS) \
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
 
 CLEANFILES =
diff --git a/xlators/features/quiesce/src/quiesce-mem-types.h b/xlators/features/quiesce/src/quiesce-mem-types.h
index f9154173a18..416456b13af 100644
--- a/xlators/features/quiesce/src/quiesce-mem-types.h
+++ b/xlators/features/quiesce/src/quiesce-mem-types.h
@@ -1,30 +1,21 @@
 /*
-   Copyright (c) 2010 Gluster, Inc. <http://www.gluster.com>
+   Copyright (c) 2010-2012 Red Hat, Inc. <http://www.redhat.com>
    This file is part of GlusterFS.
 
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
 */
 
-
 #ifndef __QUIESCE_MEM_TYPES_H__
 #define __QUIESCE_MEM_TYPES_H__
 
-#include "mem-types.h"
+#include <glusterfs/mem-types.h>
 
 enum gf_quiesce_mem_types_ {
-        gf_quiesce_mt_priv_t = gf_common_mt_end + 1,
-        gf_quiesce_mt_end
+    gf_quiesce_mt_priv_t = gf_common_mt_end + 1,
+    gf_quiesce_mt_failover_hosts,
+    gf_quiesce_mt_end
 };
 #endif
diff --git a/xlators/features/quiesce/src/quiesce-messages.h b/xlators/features/quiesce/src/quiesce-messages.h
new file mode 100644
index 00000000000..32ffd409807
--- /dev/null
+++ b/xlators/features/quiesce/src/quiesce-messages.h
@@ -0,0 +1,28 @@
+/*
+ *   Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com>
+ *   This file is part of GlusterFS.
+ *
+ *   This file is licensed to you under your choice of the GNU Lesser
+ *   General Public License, version 3 or any later version (LGPLv3 or
+ *   later), or the GNU General Public License, version 2 (GPLv2), in all
+ *   cases as published by the Free Software Foundation.
+ */
+
+#ifndef __QUIESCE_MESSAGES_H__
+#define __QUIESCE_MESSAGES_H__
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(QUIESCE, QUIESCE_MSG_INVAL_HOST, QUIESCE_MSG_FAILOVER_FAILED);
+
+#endif /* __NL_CACHE_MESSAGES_H__ */
diff --git a/xlators/features/quiesce/src/quiesce.c b/xlators/features/quiesce/src/quiesce.c
index cb52d0ee551..0e5eb60a16f 100644
--- a/xlators/features/quiesce/src/quiesce.c
+++ b/xlators/features/quiesce/src/quiesce.c
@@ -1,1567 +1,2704 @@
 /*
-  Copyright (c) 2010 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+   Copyright (c) 2010-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
 */
+#include "quiesce.h"
+#include <glusterfs/defaults.h>
+#include <glusterfs/call-stub.h>
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+/* TODO: */
+/* Think about 'writev/_*_lk/setattr/xattrop/' fops to do re-transmittion */
 
-#include "quiesce.h"
-#include "defaults.h"
-#include "call-stub.h"
+void
+gf_quiesce_timeout(void *data);
 
 /* Quiesce Specific Functions */
-call_stub_t *
-gf_quiesce_dequeue (quiesce_priv_t *priv)
+void
+gf_quiesce_local_wipe(xlator_t *this, quiesce_local_t *local)
 {
-        call_stub_t  *stub = NULL;
+    if (!local || !this || !this->private)
+        return;
 
-        if (list_empty (&priv->req))
-                return NULL;
+    if (local->loc.inode)
+        loc_wipe(&local->loc);
+    if (local->fd)
+        fd_unref(local->fd);
+    GF_FREE(local->name);
+    GF_FREE(local->volname);
+    if (local->dict)
+        dict_unref(local->dict);
+    if (local->iobref)
+        iobref_unref(local->iobref);
+    GF_FREE(local->vector);
+
+    mem_put(local);
+}
 
-        LOCK (&priv->lock);
-        {
-                stub = list_entry (priv->req.next, call_stub_t, list);
-                list_del_init (&stub->list);
-                priv->queue_size--;
-        }
-        UNLOCK (&priv->lock);
+void
+__gf_quiesce_start_timer(xlator_t *this, quiesce_priv_t *priv)
+{
+    struct timespec timeout = {
+        0,
+    };
+
+    if (!priv->timer) {
+        timeout.tv_sec = priv->timeout;
+        timeout.tv_nsec = 0;
 
-        return stub;
+        priv->timer = gf_timer_call_after(this->ctx, timeout,
+                                          gf_quiesce_timeout, (void *)this);
+        if (priv->timer == NULL) {
+            gf_log(this->name, GF_LOG_ERROR, "Cannot create timer");
+        }
+    }
 }
 
+static void
+__gf_quiesce_cleanup_failover_hosts(xlator_t *this, quiesce_priv_t *priv)
+{
+    quiesce_failover_hosts_t *tmp = NULL;
+    quiesce_failover_hosts_t *failover_host = NULL;
+
+    list_for_each_entry_safe(failover_host, tmp, &priv->failover_list, list)
+    {
+        GF_FREE(failover_host->addr);
+        list_del(&failover_host->list);
+        GF_FREE(failover_host);
+    }
+    return;
+}
 
 void
-gf_quiesce_enqueue (quiesce_priv_t *priv, call_stub_t *stub)
+gf_quiesce_populate_failover_hosts(xlator_t *this, quiesce_priv_t *priv,
+                                   const char *value)
 {
-        LOCK (&priv->lock);
-        {
-                list_add_tail (&stub->list, &priv->req);
-                priv->queue_size++;
-        }
-        UNLOCK (&priv->lock);
+    char *dup_val = NULL;
+    char *addr_tok = NULL;
+    char *save_ptr = NULL;
+    quiesce_failover_hosts_t *failover_host = NULL;
+
+    if (!value)
+        goto out;
+
+    dup_val = gf_strdup(value);
+    if (!dup_val)
+        goto out;
+
+    addr_tok = strtok_r(dup_val, ",", &save_ptr);
+    LOCK(&priv->lock);
+    {
+        if (!list_empty(&priv->failover_list))
+            __gf_quiesce_cleanup_failover_hosts(this, priv);
+
+        while (addr_tok) {
+            if (!valid_internet_address(addr_tok, _gf_true, _gf_false)) {
+                gf_msg(this->name, GF_LOG_INFO, 0, QUIESCE_MSG_INVAL_HOST,
+                       "Specified "
+                       "invalid internet address:%s",
+                       addr_tok);
+                continue;
+            }
+            failover_host = GF_CALLOC(1, sizeof(*failover_host),
+                                      gf_quiesce_mt_failover_hosts);
+            failover_host->addr = gf_strdup(addr_tok);
+            INIT_LIST_HEAD(&failover_host->list);
+            list_add(&failover_host->list, &priv->failover_list);
+            addr_tok = strtok_r(NULL, ",", &save_ptr);
+        }
+    }
+    UNLOCK(&priv->lock);
+    GF_FREE(dup_val);
+out:
+    return;
+}
 
-        return;
+int32_t
+gf_quiesce_failover_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    quiesce_priv_t *priv = NULL;
+
+    if (op_ret < 0) {
+        /* Failure here doesn't mean the failover to another host didn't
+         * succeed, we will know if failover succeeds or not by the
+         * CHILD_UP/CHILD_DOWN event. A failure here indicates something
+         * went wrong with the submission of failover command, hence
+         * just abort the failover attempts without retrying with other
+         * hosts.
+         */
+        gf_msg(this->name, GF_LOG_INFO, op_errno, QUIESCE_MSG_FAILOVER_FAILED,
+               "Initiating failover to host:%s failed:", (char *)cookie);
+    }
+
+    GF_FREE(cookie);
+    STACK_DESTROY(frame->root);
+
+    priv = this->private;
+    __gf_quiesce_start_timer(this, priv);
+
+    return 0;
 }
 
+int
+__gf_quiesce_perform_failover(xlator_t *this)
+{
+    int ret = 0;
+    call_frame_t *frame = NULL;
+    dict_t *dict = NULL;
+    quiesce_priv_t *priv = NULL;
+    quiesce_failover_hosts_t *failover_host = NULL;
+    quiesce_failover_hosts_t *host = NULL;
+
+    priv = this->private;
+
+    if (priv->pass_through) {
+        gf_msg_trace(this->name, 0,
+                     "child is up, hence not "
+                     "performing any failover");
+        goto out;
+    }
+
+    list_for_each_entry(failover_host, &priv->failover_list, list)
+    {
+        if (failover_host->tried == 0) {
+            host = failover_host;
+            failover_host->tried = 1;
+            break;
+        }
+    }
+    if (!host) {
+        /*TODO: Keep trying until any of the gfproxy comes back up.
+                Currently it tries failing over once for each host,
+                if it doesn't succeed then returns error to mount point
+           list_for_each_entry (failover_host,
+                        &priv->failover_list, list) {
+                failover_host->tried = 0;
+        }*/
+        gf_msg_debug(this->name, 0,
+                     "all the failover hosts have "
+                     "been tried and looks like didn't succeed");
+        ret = -1;
+        goto out;
+    }
+
+    frame = create_frame(this, this->ctx->pool);
+    if (!frame) {
+        gf_msg_debug(this->name, 0, "failed to create the frame");
+        ret = -1;
+        goto out;
+    }
+
+    dict = dict_new();
+
+    ret = dict_set_dynstr(dict, CLIENT_CMD_CONNECT, gf_strdup(host->addr));
+
+    gf_msg_trace(this->name, 0, "Initiating failover to:%s", host->addr);
+
+    STACK_WIND_COOKIE(frame, gf_quiesce_failover_cbk, NULL, FIRST_CHILD(this),
+                      FIRST_CHILD(this)->fops->setxattr, NULL, dict, 0, NULL);
+out:
+
+    if (dict)
+        dict_unref(dict);
+
+    return ret;
+}
+
+call_stub_t *
+gf_quiesce_dequeue(xlator_t *this)
+{
+    call_stub_t *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+
+    priv = this->private;
+
+    if (!priv || list_empty(&priv->req))
+        return NULL;
+
+    LOCK(&priv->lock);
+    {
+        stub = list_entry(priv->req.next, call_stub_t, list);
+        list_del_init(&stub->list);
+        priv->queue_size--;
+    }
+    UNLOCK(&priv->lock);
+
+    return stub;
+}
 
 void *
-gf_quiesce_dequeue_start (void *data)
+gf_quiesce_dequeue_start(void *data)
 {
-        xlator_t       *this = NULL;
-        quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    xlator_t *this = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
 
-        this = data;
-        priv = this->private;
-        THIS = this;
+    this = data;
+    priv = this->private;
+    THIS = this;
 
-        while (!list_empty (&priv->req)) {
-                stub = gf_quiesce_dequeue (priv);
-                if (stub) {
-                        call_resume (stub);
-                }
+    while (!list_empty(&priv->req)) {
+        stub = gf_quiesce_dequeue(this);
+        if (stub) {
+            call_resume(stub);
         }
+    }
 
-        return 0;
+    return 0;
 }
 
-
 void
-gf_quiesce_timeout (void *data)
+gf_quiesce_timeout(void *data)
 {
-        xlator_t       *this = NULL;
-        quiesce_priv_t *priv = NULL;
-        int             need_dequeue = 0;
+    xlator_t *this = NULL;
+    quiesce_priv_t *priv = NULL;
+    int ret = -1;
 
-        this = data;
-        priv = this->private;
-        THIS = this;
+    this = data;
+    priv = this->private;
+    THIS = this;
 
-        LOCK (&priv->lock);
-        {
-                priv->pass_through = _gf_true;
-                need_dequeue = (priv->queue_size)? 1:0;
+    LOCK(&priv->lock);
+    {
+        priv->timer = NULL;
+        if (priv->pass_through) {
+            UNLOCK(&priv->lock);
+            goto out;
         }
-        UNLOCK (&priv->lock);
+        ret = __gf_quiesce_perform_failover(THIS);
+    }
+    UNLOCK(&priv->lock);
 
-        gf_quiesce_dequeue_start (this);
+    if (ret < 0) {
+        priv->pass_through = _gf_true;
+        gf_quiesce_dequeue_start(this);
+    }
 
+out:
+    return;
+}
+
+void
+gf_quiesce_enqueue(xlator_t *this, call_stub_t *stub)
+{
+    quiesce_priv_t *priv = NULL;
+
+    priv = this->private;
+    if (!priv) {
+        gf_log_callingfn(this->name, GF_LOG_ERROR, "this->private == NULL");
         return;
+    }
+
+    LOCK(&priv->lock);
+    {
+        list_add_tail(&stub->list, &priv->req);
+        priv->queue_size++;
+        __gf_quiesce_start_timer(this, priv);
+    }
+    UNLOCK(&priv->lock);
+
+    return;
 }
 
-/* FOP */
+/* _CBK function section */
 
 int32_t
-quiesce_fgetxattr (call_frame_t *frame,
-                   xlator_t *this,
-                   fd_t *fd,
-                   const char *name)
+quiesce_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, inode_t *inode,
+                   struct iatt *buf, dict_t *dict, struct iatt *postparent)
 {
-        quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+    if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+        /* Re-transmit (by putting in the queue) */
+        stub = fop_lookup_stub(frame, default_lookup_resume, &local->loc,
+                               local->dict);
+        if (!stub) {
+            STACK_UNWIND_STRICT(lookup, frame, -1, ENOMEM, NULL, NULL, NULL,
+                                NULL);
+            goto out;
+        }
 
-        priv = this->private;
+        gf_quiesce_enqueue(this, stub);
+        goto out;
+    }
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_fgetxattr_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->fgetxattr,
-                            fd,
-                            name);
-                return 0;
-        }
+    STACK_UNWIND_STRICT(lookup, frame, op_ret, op_errno, inode, buf, dict,
+                        postparent);
+out:
+    gf_quiesce_local_wipe(this, local);
 
-        stub = fop_fgetxattr_stub (frame, default_fgetxattr_resume, fd, name);
+    return 0;
+}
+
+int32_t
+quiesce_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                 dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+    if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+        /* Re-transmit (by putting in the queue) */
+        stub = fop_stat_stub(frame, default_stat_resume, &local->loc, xdata);
         if (!stub) {
-                STACK_UNWIND_STRICT (fgetxattr, frame, -1, ENOMEM, NULL);
-                return 0;
+            STACK_UNWIND_STRICT(stat, frame, -1, ENOMEM, NULL, NULL);
+            goto out;
         }
 
-        gf_quiesce_enqueue (priv, stub);
+        gf_quiesce_enqueue(this, stub);
+        goto out;
+    }
 
-        return 0;
+    STACK_UNWIND_STRICT(stat, frame, op_ret, op_errno, buf, xdata);
+out:
+    gf_quiesce_local_wipe(this, local);
+
+    return 0;
 }
 
 int32_t
-quiesce_fsetxattr (call_frame_t *frame,
-                   xlator_t *this,
-                   fd_t *fd,
-                   dict_t *dict,
-                   int32_t flags)
+quiesce_access_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+    if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+        /* Re-transmit (by putting in the queue) */
+        stub = fop_access_stub(frame, default_access_resume, &local->loc,
+                               local->flag, xdata);
+        if (!stub) {
+            STACK_UNWIND_STRICT(access, frame, -1, ENOMEM, NULL);
+            goto out;
+        }
 
-        priv = this->private;
+        gf_quiesce_enqueue(this, stub);
+        goto out;
+    }
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_fsetxattr_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->fsetxattr,
-                            fd,
-                            dict,
-                            flags);
-	        return 0;
+    STACK_UNWIND_STRICT(access, frame, op_ret, op_errno, xdata);
+out:
+    gf_quiesce_local_wipe(this, local);
+
+    return 0;
+}
+
+int32_t
+quiesce_readlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, const char *path,
+                     struct iatt *buf, dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+    if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+        /* Re-transmit (by putting in the queue) */
+        stub = fop_readlink_stub(frame, default_readlink_resume, &local->loc,
+                                 local->size, xdata);
+        if (!stub) {
+            STACK_UNWIND_STRICT(readlink, frame, -1, ENOMEM, NULL, NULL, NULL);
+            goto out;
         }
 
-        stub = fop_fsetxattr_stub (frame, default_fsetxattr_resume,
-                                   fd, dict, flags);
+        gf_quiesce_enqueue(this, stub);
+        goto out;
+    }
+
+    STACK_UNWIND_STRICT(readlink, frame, op_ret, op_errno, path, buf, xdata);
+out:
+    gf_quiesce_local_wipe(this, local);
+
+    return 0;
+}
+
+int32_t
+quiesce_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+    if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+        /* Re-transmit (by putting in the queue) */
+        stub = fop_open_stub(frame, default_open_resume, &local->loc,
+                             local->flag, local->fd, xdata);
         if (!stub) {
-                STACK_UNWIND_STRICT (fsetxattr, frame, -1, ENOMEM);
-                return 0;
+            STACK_UNWIND_STRICT(open, frame, -1, ENOMEM, NULL, NULL);
+            goto out;
         }
 
-        gf_quiesce_enqueue (priv, stub);
+        gf_quiesce_enqueue(this, stub);
+        goto out;
+    }
 
-        return 0;
+    STACK_UNWIND_STRICT(open, frame, op_ret, op_errno, fd, xdata);
+out:
+    gf_quiesce_local_wipe(this, local);
+
+    return 0;
 }
 
 int32_t
-quiesce_setxattr (call_frame_t *frame,
-		  xlator_t *this,
-		  loc_t *loc,
-		  dict_t *dict,
-		  int32_t flags)
+quiesce_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct iovec *vector,
+                  int32_t count, struct iatt *stbuf, struct iobref *iobref,
+                  dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+    if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+        /* Re-transmit (by putting in the queue) */
+        stub = fop_readv_stub(frame, default_readv_resume, local->fd,
+                              local->size, local->offset, local->io_flag,
+                              xdata);
+        if (!stub) {
+            STACK_UNWIND_STRICT(readv, frame, -1, ENOMEM, NULL, 0, NULL, NULL,
+                                NULL);
+            goto out;
+        }
 
-        priv = this->private;
+        gf_quiesce_enqueue(this, stub);
+        goto out;
+    }
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_setxattr_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->setxattr,
-                            loc,
-                            dict,
-                            flags);
-	        return 0;
+    STACK_UNWIND_STRICT(readv, frame, op_ret, op_errno, vector, count, stbuf,
+                        iobref, xdata);
+out:
+    gf_quiesce_local_wipe(this, local);
+
+    return 0;
+}
+
+int32_t
+quiesce_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+    if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+        /* Re-transmit (by putting in the queue) */
+        stub = fop_flush_stub(frame, default_flush_resume, local->fd, xdata);
+        if (!stub) {
+            STACK_UNWIND_STRICT(flush, frame, -1, ENOMEM, NULL);
+            goto out;
         }
 
-        stub = fop_setxattr_stub (frame, default_setxattr_resume,
-                                  loc, dict, flags);
+        gf_quiesce_enqueue(this, stub);
+        goto out;
+    }
+
+    STACK_UNWIND_STRICT(flush, frame, op_ret, op_errno, xdata);
+out:
+    gf_quiesce_local_wipe(this, local);
+
+    return 0;
+}
+
+int32_t
+quiesce_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                  struct iatt *postbuf, dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+    if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+        /* Re-transmit (by putting in the queue) */
+        stub = fop_fsync_stub(frame, default_fsync_resume, local->fd,
+                              local->flag, xdata);
         if (!stub) {
-                STACK_UNWIND_STRICT (setxattr, frame, -1, ENOMEM);
-                return 0;
+            STACK_UNWIND_STRICT(fsync, frame, -1, ENOMEM, NULL, NULL, NULL);
+            goto out;
         }
 
-        gf_quiesce_enqueue (priv, stub);
+        gf_quiesce_enqueue(this, stub);
+        goto out;
+    }
 
-        return 0;
+    STACK_UNWIND_STRICT(fsync, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+out:
+    gf_quiesce_local_wipe(this, local);
+
+    return 0;
 }
 
 int32_t
-quiesce_statfs (call_frame_t *frame,
-		xlator_t *this,
-		loc_t *loc)
+quiesce_fstat_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                  dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+    if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+        /* Re-transmit (by putting in the queue) */
+        stub = fop_fstat_stub(frame, default_fstat_resume, local->fd, xdata);
+        if (!stub) {
+            STACK_UNWIND_STRICT(fstat, frame, -1, ENOMEM, NULL, NULL);
+            goto out;
+        }
 
-        priv = this->private;
+        gf_quiesce_enqueue(this, stub);
+        goto out;
+    }
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_statfs_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->statfs,
-                            loc);
-	        return 0;
+    STACK_UNWIND_STRICT(fstat, frame, op_ret, op_errno, buf, xdata);
+out:
+    gf_quiesce_local_wipe(this, local);
+
+    return 0;
+}
+
+int32_t
+quiesce_opendir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+    if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+        /* Re-transmit (by putting in the queue) */
+        stub = fop_opendir_stub(frame, default_opendir_resume, &local->loc,
+                                local->fd, xdata);
+        if (!stub) {
+            STACK_UNWIND_STRICT(opendir, frame, -1, ENOMEM, NULL, NULL);
+            goto out;
         }
 
-        stub = fop_statfs_stub (frame, default_statfs_resume, loc);
+        gf_quiesce_enqueue(this, stub);
+        goto out;
+    }
+
+    STACK_UNWIND_STRICT(opendir, frame, op_ret, op_errno, fd, xdata);
+out:
+    gf_quiesce_local_wipe(this, local);
+
+    return 0;
+}
+
+int32_t
+quiesce_fsyncdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+    if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+        /* Re-transmit (by putting in the queue) */
+        stub = fop_fsyncdir_stub(frame, default_fsyncdir_resume, local->fd,
+                                 local->flag, xdata);
         if (!stub) {
-                STACK_UNWIND_STRICT (statfs, frame, -1, ENOMEM, NULL);
-                return 0;
+            STACK_UNWIND_STRICT(fsyncdir, frame, -1, ENOMEM, NULL);
+            goto out;
         }
 
-        gf_quiesce_enqueue (priv, stub);
+        gf_quiesce_enqueue(this, stub);
+        goto out;
+    }
 
-        return 0;
+    STACK_UNWIND_STRICT(fsyncdir, frame, op_ret, op_errno, xdata);
+out:
+    gf_quiesce_local_wipe(this, local);
+
+    return 0;
 }
 
 int32_t
-quiesce_fsyncdir (call_frame_t *frame,
-		  xlator_t *this,
-		  fd_t *fd,
-		  int32_t flags)
+quiesce_statfs_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct statvfs *buf,
+                   dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+    if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+        /* Re-transmit (by putting in the queue) */
+        stub = fop_statfs_stub(frame, default_statfs_resume, &local->loc,
+                               xdata);
+        if (!stub) {
+            STACK_UNWIND_STRICT(statfs, frame, -1, ENOMEM, NULL, NULL);
+            goto out;
+        }
 
-        priv = this->private;
+        gf_quiesce_enqueue(this, stub);
+        goto out;
+    }
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_fsyncdir_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->fsyncdir,
-                            fd,
-                            flags);
-	        return 0;
+    STACK_UNWIND_STRICT(statfs, frame, op_ret, op_errno, buf, xdata);
+out:
+    gf_quiesce_local_wipe(this, local);
+
+    return 0;
+}
+
+int32_t
+quiesce_fgetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *dict,
+                      dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+    if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+        /* Re-transmit (by putting in the queue) */
+        stub = fop_fgetxattr_stub(frame, default_fgetxattr_resume, local->fd,
+                                  local->name, xdata);
+        if (!stub) {
+            STACK_UNWIND_STRICT(fgetxattr, frame, -1, ENOMEM, NULL, NULL);
+            goto out;
         }
 
-        stub = fop_fsyncdir_stub (frame, default_fsyncdir_resume, fd, flags);
+        gf_quiesce_enqueue(this, stub);
+        goto out;
+    }
+
+    STACK_UNWIND_STRICT(fgetxattr, frame, op_ret, op_errno, dict, xdata);
+out:
+    gf_quiesce_local_wipe(this, local);
+
+    return 0;
+}
+
+int32_t
+quiesce_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, dict_t *dict,
+                     dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+    if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+        /* Re-transmit (by putting in the queue) */
+        stub = fop_getxattr_stub(frame, default_getxattr_resume, &local->loc,
+                                 local->name, xdata);
         if (!stub) {
-                STACK_UNWIND_STRICT (fsyncdir, frame, -1, ENOMEM);
-                return 0;
+            STACK_UNWIND_STRICT(getxattr, frame, -1, ENOMEM, NULL, NULL);
+            goto out;
         }
 
-        gf_quiesce_enqueue (priv, stub);
+        gf_quiesce_enqueue(this, stub);
+        goto out;
+    }
 
-        return 0;
+    STACK_UNWIND_STRICT(getxattr, frame, op_ret, op_errno, dict, xdata);
+out:
+    gf_quiesce_local_wipe(this, local);
+
+    return 0;
 }
 
 int32_t
-quiesce_opendir (call_frame_t *frame,
-		 xlator_t *this,
-		 loc_t *loc, fd_t *fd)
+quiesce_rchecksum_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, uint32_t weak_checksum,
+                      uint8_t *strong_checksum, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+    if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+        /* Re-transmit (by putting in the queue) */
+        stub = fop_rchecksum_stub(frame, default_rchecksum_resume, local->fd,
+                                  local->offset, local->flag, xdata);
+        if (!stub) {
+            STACK_UNWIND_STRICT(rchecksum, frame, -1, ENOMEM, 0, NULL, NULL);
+            goto out;
+        }
 
-        priv = this->private;
+        gf_quiesce_enqueue(this, stub);
+        goto out;
+    }
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_opendir_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->opendir,
-                            loc, fd);
-	        return 0;
+    STACK_UNWIND_STRICT(rchecksum, frame, op_ret, op_errno, weak_checksum,
+                        strong_checksum, xdata);
+out:
+    gf_quiesce_local_wipe(this, local);
+
+    return 0;
+}
+
+int32_t
+quiesce_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, gf_dirent_t *entries,
+                    dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+    if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+        /* Re-transmit (by putting in the queue) */
+        stub = fop_readdir_stub(frame, default_readdir_resume, local->fd,
+                                local->size, local->offset, xdata);
+        if (!stub) {
+            STACK_UNWIND_STRICT(readdir, frame, -1, ENOMEM, NULL, NULL);
+            goto out;
         }
 
-        stub = fop_opendir_stub (frame, default_opendir_resume, loc, fd);
+        gf_quiesce_enqueue(this, stub);
+        goto out;
+    }
+
+    STACK_UNWIND_STRICT(readdir, frame, op_ret, op_errno, entries, xdata);
+out:
+    gf_quiesce_local_wipe(this, local);
+
+    return 0;
+}
+
+int32_t
+quiesce_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, gf_dirent_t *entries,
+                     dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+    if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+        /* Re-transmit (by putting in the queue) */
+        stub = fop_readdirp_stub(frame, default_readdirp_resume, local->fd,
+                                 local->size, local->offset, local->dict);
         if (!stub) {
-                STACK_UNWIND_STRICT (opendir, frame, -1, ENOMEM, NULL);
-                return 0;
+            STACK_UNWIND_STRICT(readdirp, frame, -1, ENOMEM, NULL, NULL);
+            goto out;
         }
 
-        gf_quiesce_enqueue (priv, stub);
+        gf_quiesce_enqueue(this, stub);
+        goto out;
+    }
 
-        return 0;
+    STACK_UNWIND_STRICT(readdirp, frame, op_ret, op_errno, entries, xdata);
+out:
+    gf_quiesce_local_wipe(this, local);
+
+    return 0;
 }
 
+#if 0
+
 int32_t
-quiesce_fstat (call_frame_t *frame,
-	       xlator_t *this,
-	       fd_t *fd)
+quiesce_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                    struct iatt *postbuf, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
+        quiesce_priv_t *priv = NULL;
         call_stub_t    *stub = NULL;
+        quiesce_local_t *local = NULL;
 
         priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_fstat_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->fstat,
-                            fd);
-	        return 0;
-        }
+        local = frame->local;
+        frame->local = NULL;
+        if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+                /* Re-transmit (by putting in the queue) */
+                stub = fop_writev_stub (frame, default_writev_resume,
+                                        local->fd, local->vector, local->flag,
+                                        local->offset, local->io_flags,
+                                        local->iobref, xdata);
+                if (!stub) {
+                        STACK_UNWIND_STRICT (writev, frame, -1, ENOMEM,
+                                             NULL, NULL, NULL);
+                        goto out;
+                }
 
-        stub = fop_fstat_stub (frame, default_fstat_resume, fd);
-        if (!stub) {
-                STACK_UNWIND_STRICT (fstat, frame, -1, ENOMEM, NULL);
-                return 0;
+                gf_quiesce_enqueue (this, stub);
+                goto out;
         }
 
-        gf_quiesce_enqueue (priv, stub);
+        STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+out:
+        gf_quiesce_local_wipe (this, local);
 
         return 0;
 }
 
 int32_t
-quiesce_fsync (call_frame_t *frame,
-	       xlator_t *this,
-	       fd_t *fd,
-	       int32_t flags)
+quiesce_xattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
+        quiesce_priv_t *priv = NULL;
         call_stub_t    *stub = NULL;
+        quiesce_local_t *local = NULL;
 
         priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_fsync_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->fsync,
-                            fd,
-                            flags);
-	        return 0;
-        }
+        local = frame->local;
+        frame->local = NULL;
+        if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+                /* Re-transmit (by putting in the queue) */
+                stub = fop_xattrop_stub (frame, default_xattrop_resume,
+                                         &local->loc, local->xattrop_flags,
+                                         local->dict, xdata);
+                if (!stub) {
+                        STACK_UNWIND_STRICT (xattrop, frame, -1, ENOMEM,
+                                             NULL, NULL);
+                        goto out;
+                }
 
-        stub = fop_fsync_stub (frame, default_fsync_resume, fd, flags);
-        if (!stub) {
-                STACK_UNWIND_STRICT (fsync, frame, -1, ENOMEM, NULL, NULL);
-                return 0;
+                gf_quiesce_enqueue (this, stub);
+                goto out;
         }
 
-        gf_quiesce_enqueue (priv, stub);
+        STACK_UNWIND_STRICT (xattrop, frame, op_ret, op_errno, dict, xdata);
+out:
+        gf_quiesce_local_wipe (this, local);
 
         return 0;
 }
 
 int32_t
-quiesce_flush (call_frame_t *frame,
-	       xlator_t *this,
-	       fd_t *fd)
+quiesce_fxattrop_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
+        quiesce_priv_t *priv = NULL;
         call_stub_t    *stub = NULL;
+        quiesce_local_t *local = NULL;
 
         priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_flush_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->flush,
-                            fd);
-	        return 0;
-        }
+        local = frame->local;
+        frame->local = NULL;
+        if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+                /* Re-transmit (by putting in the queue) */
+                stub = fop_fxattrop_stub (frame, default_fxattrop_resume,
+                                          local->fd, local->xattrop_flags,
+                                          local->dict, xdata);
+                if (!stub) {
+                        STACK_UNWIND_STRICT (fxattrop, frame, -1, ENOMEM,
+                                             NULL, NULL);
+                        goto out;
+                }
 
-        stub = fop_flush_stub (frame, default_flush_resume, fd);
-        if (!stub) {
-                STACK_UNWIND_STRICT (flush, frame, -1, ENOMEM);
-                return 0;
+                gf_quiesce_enqueue (this, stub);
+                goto out;
         }
 
-        gf_quiesce_enqueue (priv, stub);
+        STACK_UNWIND_STRICT (fxattrop, frame, op_ret, op_errno, dict, xdata);
+out:
+        gf_quiesce_local_wipe (this, local);
 
         return 0;
 }
 
 int32_t
-quiesce_writev (call_frame_t *frame,
-		xlator_t *this,
-		fd_t *fd,
-		struct iovec *vector,
-		int32_t count,
-		off_t off,
-                struct iobref *iobref)
+quiesce_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct gf_flock *lock, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
+        quiesce_priv_t *priv = NULL;
         call_stub_t    *stub = NULL;
+        quiesce_local_t *local = NULL;
 
         priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_writev_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->writev,
-                            fd,
-                            vector,
-                            count,
-                            off,
-                            iobref);
-	        return 0;
-        }
+        local = frame->local;
+        frame->local = NULL;
+        if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+                /* Re-transmit (by putting in the queue) */
+                stub = fop_lk_stub (frame, default_lk_resume,
+                                    local->fd, local->flag, &local->flock, xdata);
+                if (!stub) {
+                        STACK_UNWIND_STRICT (lk, frame, -1, ENOMEM,
+                                             NULL, NULL);
+                        goto out;
+                }
 
-        stub = fop_writev_stub (frame, default_writev_resume,
-                                fd, vector, count, off, iobref);
-        if (!stub) {
-                STACK_UNWIND_STRICT (writev, frame, -1, ENOMEM, NULL, NULL);
-                return 0;
+                gf_quiesce_enqueue (this, stub);
+                goto out;
         }
 
-        gf_quiesce_enqueue (priv, stub);
+        STACK_UNWIND_STRICT (lk, frame, op_ret, op_errno, lock, xdata);
+out:
+        gf_quiesce_local_wipe (this, local);
 
         return 0;
 }
 
 int32_t
-quiesce_readv (call_frame_t *frame,
-	       xlator_t *this,
-	       fd_t *fd,
-	       size_t size,
-	       off_t offset)
+quiesce_inodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
+        quiesce_priv_t *priv = NULL;
         call_stub_t    *stub = NULL;
+        quiesce_local_t *local = NULL;
 
         priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_readv_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->readv,
-                            fd,
-                            size,
-                            offset);
-	        return 0;
-        }
+        local = frame->local;
+        frame->local = NULL;
+        if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+                /* Re-transmit (by putting in the queue) */
+                stub = fop_inodelk_stub (frame, default_inodelk_resume,
+                                         local->volname, &local->loc,
+                                         local->flag, &local->flock, xdata);
+                if (!stub) {
+                        STACK_UNWIND_STRICT (inodelk, frame, -1, ENOMEM, NULL);
+                        goto out;
+                }
 
-        stub = fop_readv_stub (frame, default_readv_resume, fd, size, offset);
-        if (!stub) {
-                STACK_UNWIND_STRICT (readv, frame, -1, ENOMEM,
-                                     NULL, 0, NULL, NULL);
-                return 0;
+                gf_quiesce_enqueue (this, stub);
+                goto out;
         }
 
-        gf_quiesce_enqueue (priv, stub);
+        STACK_UNWIND_STRICT (inodelk, frame, op_ret, op_errno, xdata);
+out:
+        gf_quiesce_local_wipe (this, local);
 
         return 0;
 }
 
 
 int32_t
-quiesce_open (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc,
-	      int32_t flags, fd_t *fd,
-              int32_t wbflags)
+quiesce_finodelk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
+        quiesce_priv_t *priv = NULL;
         call_stub_t    *stub = NULL;
+        quiesce_local_t *local = NULL;
 
         priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_open_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->open,
-                            loc, flags, fd, wbflags);
-	        return 0;
-        }
+        local = frame->local;
+        frame->local = NULL;
+        if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+                /* Re-transmit (by putting in the queue) */
+                stub = fop_finodelk_stub (frame, default_finodelk_resume,
+                                         local->volname, local->fd,
+                                         local->flag, &local->flock, xdata);
+                if (!stub) {
+                        STACK_UNWIND_STRICT (finodelk, frame, -1, ENOMEM, NULL);
+                        goto out;
+                }
 
-        stub = fop_open_stub (frame, default_open_resume, loc,
-                              flags, fd, wbflags);
-        if (!stub) {
-                STACK_UNWIND_STRICT (open, frame, -1, ENOMEM, NULL);
-                return 0;
+                gf_quiesce_enqueue (this, stub);
+                goto out;
         }
 
-        gf_quiesce_enqueue (priv, stub);
+        STACK_UNWIND_STRICT (finodelk, frame, op_ret, op_errno, xdata);
+out:
+        gf_quiesce_local_wipe (this, local);
 
         return 0;
 }
 
 int32_t
-quiesce_create (call_frame_t *frame, xlator_t *this,
-		loc_t *loc, int32_t flags, mode_t mode,
-                fd_t *fd, dict_t *params)
+quiesce_entrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
+        quiesce_priv_t *priv = NULL;
         call_stub_t    *stub = NULL;
+        quiesce_local_t *local = NULL;
 
         priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame, default_create_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->create,
-                            loc, flags, mode, fd, params);
-	        return 0;
-        }
+        local = frame->local;
+        frame->local = NULL;
+        if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+                /* Re-transmit (by putting in the queue) */
+                stub = fop_entrylk_stub (frame, default_entrylk_resume,
+                                         local->volname, &local->loc,
+                                         local->name, local->cmd, local->type, xdata);
+                if (!stub) {
+                        STACK_UNWIND_STRICT (entrylk, frame, -1, ENOMEM, NULL);
+                        goto out;
+                }
 
-        stub = fop_create_stub (frame, default_create_resume,
-                                loc, flags, mode, fd, params);
-        if (!stub) {
-                STACK_UNWIND_STRICT (create, frame, -1, ENOMEM,
-                                     NULL, NULL, NULL, NULL, NULL);
-                return 0;
+                gf_quiesce_enqueue (this, stub);
+                goto out;
         }
 
-        gf_quiesce_enqueue (priv, stub);
+        STACK_UNWIND_STRICT (entrylk, frame, op_ret, op_errno, xdata);
+out:
+        gf_quiesce_local_wipe (this, local);
 
         return 0;
 }
 
 int32_t
-quiesce_link (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *oldloc,
-	      loc_t *newloc)
+quiesce_fentrylk_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
+        quiesce_priv_t *priv = NULL;
         call_stub_t    *stub = NULL;
+        quiesce_local_t *local = NULL;
 
         priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_link_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->link,
-                            oldloc, newloc);
-	        return 0;
-        }
+        local = frame->local;
+        frame->local = NULL;
+        if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+                /* Re-transmit (by putting in the queue) */
+                stub = fop_fentrylk_stub (frame, default_fentrylk_resume,
+                                          local->volname, local->fd,
+                                          local->name, local->cmd, local->type, xdata);
+                if (!stub) {
+                        STACK_UNWIND_STRICT (fentrylk, frame, -1, ENOMEM, NULL);
+                        goto out;
+                }
 
-        stub = fop_link_stub (frame, default_link_resume, oldloc, newloc);
-        if (!stub) {
-                STACK_UNWIND_STRICT (link, frame, -1, ENOMEM,
-                                     NULL, NULL, NULL, NULL);
-                return 0;
+                gf_quiesce_enqueue (this, stub);
+                goto out;
         }
 
-        gf_quiesce_enqueue (priv, stub);
+        STACK_UNWIND_STRICT (fentrylk, frame, op_ret, op_errno, xdata);
+out:
+        gf_quiesce_local_wipe (this, local);
 
         return 0;
 }
 
 int32_t
-quiesce_rename (call_frame_t *frame,
-		xlator_t *this,
-		loc_t *oldloc,
-		loc_t *newloc)
+quiesce_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *statpre,
+                     struct iatt *statpost, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
+        quiesce_priv_t *priv = NULL;
         call_stub_t    *stub = NULL;
+        quiesce_local_t *local = NULL;
 
         priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_rename_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->rename,
-                            oldloc, newloc);
-	        return 0;
-        }
+        local = frame->local;
+        frame->local = NULL;
+        if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+                /* Re-transmit (by putting in the queue) */
+                stub = fop_setattr_stub (frame, default_setattr_resume,
+                                         &local->loc, &local->stbuf, local->flag, xdata);
+                if (!stub) {
+                        STACK_UNWIND_STRICT (setattr, frame, -1, ENOMEM,
+                                             NULL, NULL, NULL);
+                        goto out;
+                }
 
-        stub = fop_rename_stub (frame, default_rename_resume, oldloc, newloc);
-        if (!stub) {
-                STACK_UNWIND_STRICT (rename, frame, -1, ENOMEM,
-                                     NULL, NULL, NULL, NULL, NULL);
-                return 0;
+                gf_quiesce_enqueue (this, stub);
+                goto out;
         }
 
-        gf_quiesce_enqueue (priv, stub);
+        STACK_UNWIND_STRICT (setattr, frame, op_ret, op_errno, statpre,
+                             statpost, xdata);
+out:
+        gf_quiesce_local_wipe (this, local);
 
         return 0;
 }
 
-
-int
-quiesce_symlink (call_frame_t *frame, xlator_t *this,
-		 const char *linkpath, loc_t *loc, dict_t *params)
+int32_t
+quiesce_fsetattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *statpre,
+                      struct iatt *statpost, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
+        quiesce_priv_t *priv = NULL;
         call_stub_t    *stub = NULL;
+        quiesce_local_t *local = NULL;
 
         priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame, default_symlink_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->symlink,
-                            linkpath, loc, params);
-	        return 0;
-        }
+        local = frame->local;
+        frame->local = NULL;
+
+        if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+                /* Re-transmit (by putting in the queue) */
+                stub = fop_fsetattr_stub (frame, default_fsetattr_resume,
+                                          local->fd, &local->stbuf, local->flag, xdata);
+                if (!stub) {
+                        STACK_UNWIND_STRICT (fsetattr, frame, -1, ENOMEM,
+                                             NULL, NULL, NULL);
+                        goto out;
+                }
 
-        stub = fop_symlink_stub (frame, default_symlink_resume,
-                                 linkpath, loc, params);
-        if (!stub) {
-                STACK_UNWIND_STRICT (symlink, frame, -1, ENOMEM,
-                                     NULL, NULL, NULL, NULL);
-                return 0;
+                gf_quiesce_enqueue (this, stub);
+                goto out;
         }
 
-        gf_quiesce_enqueue (priv, stub);
+        STACK_UNWIND_STRICT (fsetattr, frame, op_ret, op_errno, statpre,
+                             statpost, xdata);
+out:
+        gf_quiesce_local_wipe (this, local);
 
         return 0;
 }
 
+#endif /* if 0 */
 
-int
-quiesce_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags)
-{
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+/* FOP */
 
-        priv = this->private;
+/* No retransmittion */
 
-        if (priv->pass_through) {
-                STACK_WIND (frame, default_rmdir_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->rmdir,
-                            loc, flags);
-	        return 0;
-        }
+int32_t
+quiesce_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                    const char *name, dict_t *xdata)
+{
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
 
-        stub = fop_rmdir_stub (frame, default_rmdir_resume, loc, flags);
-        if (!stub) {
-                STACK_UNWIND_STRICT (rmdir, frame, -1, ENOMEM, NULL, NULL);
-                return 0;
-        }
+    priv = this->private;
 
-        gf_quiesce_enqueue (priv, stub);
+    if (priv->pass_through) {
+        STACK_WIND(frame, default_removexattr_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->removexattr, loc, name, xdata);
+        return 0;
+    }
 
+    stub = fop_removexattr_stub(frame, default_removexattr_resume, loc, name,
+                                xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(removexattr, frame, -1, ENOMEM, NULL);
         return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
 int32_t
-quiesce_unlink (call_frame_t *frame,
-		xlator_t *this,
-		loc_t *loc)
+quiesce_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                     const char *name, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_unlink_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->unlink,
-                            loc);
-	        return 0;
-        }
+    if (priv->pass_through) {
+        STACK_WIND(frame, default_fremovexattr_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fremovexattr, fd, name, xdata);
+        return 0;
+    }
 
-        stub = fop_unlink_stub (frame, default_unlink_resume, loc);
-        if (!stub) {
-                STACK_UNWIND_STRICT (unlink, frame, -1, ENOMEM, NULL, NULL);
-                return 0;
-        }
+    stub = fop_fremovexattr_stub(frame, default_fremovexattr_resume, fd, name,
+                                 xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(fremovexattr, frame, -1, ENOMEM, NULL);
+        return 0;
+    }
 
-        gf_quiesce_enqueue (priv, stub);
+    gf_quiesce_enqueue(this, stub);
 
-        return 0;
+    return 0;
 }
 
-int
-quiesce_mkdir (call_frame_t *frame, xlator_t *this,
-	       loc_t *loc, mode_t mode, dict_t *params)
+int32_t
+quiesce_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+                 dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame, default_mkdir_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->mkdir,
-                            loc, mode, params);
-	        return 0;
-        }
+    if (priv->pass_through) {
+        STACK_WIND(frame, default_truncate_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+        return 0;
+    }
 
-        stub = fop_mkdir_stub (frame, default_mkdir_resume,
-                               loc, mode, params);
-        if (!stub) {
-                STACK_UNWIND_STRICT (mkdir, frame, -1, ENOMEM,
-                                     NULL, NULL, NULL, NULL);
-                return 0;
-        }
+    stub = fop_truncate_stub(frame, default_truncate_resume, loc, offset,
+                             xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(truncate, frame, -1, ENOMEM, NULL, NULL, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
 
-        gf_quiesce_enqueue (priv, stub);
+    return 0;
+}
+
+int32_t
+quiesce_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+                  int32_t flags, dict_t *xdata)
+{
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+
+    priv = this->private;
 
+    if (priv->pass_through) {
+        STACK_WIND(frame, default_fsetxattr_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
         return 0;
+    }
+
+    stub = fop_fsetxattr_stub(frame, default_fsetxattr_resume, fd, dict, flags,
+                              xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(fsetxattr, frame, -1, ENOMEM, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
+int32_t
+quiesce_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+                 int32_t flags, dict_t *xdata)
+{
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+
+    priv = this->private;
 
-int
-quiesce_mknod (call_frame_t *frame, xlator_t *this,
-	       loc_t *loc, mode_t mode, dev_t rdev, dict_t *parms)
+    if (priv->pass_through) {
+        STACK_WIND(frame, default_setxattr_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->setxattr, loc, dict, flags, xdata);
+        return 0;
+    }
+
+    stub = fop_setxattr_stub(frame, default_setxattr_resume, loc, dict, flags,
+                             xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(setxattr, frame, -1, ENOMEM, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
+}
+
+int32_t
+quiesce_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+               mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame, default_mknod_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->mknod,
-                            loc, mode, rdev, parms);
-	        return 0;
-        }
+    if (priv->pass_through) {
+        /* Don't send O_APPEND below, as write() re-transmittions can
+           fail with O_APPEND */
+        STACK_WIND(frame, default_create_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->create, loc, (flags & ~O_APPEND),
+                   mode, umask, fd, xdata);
+        return 0;
+    }
 
-        stub = fop_mknod_stub (frame, default_mknod_resume,
-                               loc, mode, rdev, parms);
-        if (!stub) {
-                STACK_UNWIND_STRICT (mknod, frame, -1, ENOMEM,
-                                     NULL, NULL, NULL, NULL);
-                return 0;
-        }
+    stub = fop_create_stub(frame, default_create_resume, loc,
+                           (flags & ~O_APPEND), mode, umask, fd, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(create, frame, -1, ENOMEM, NULL, NULL, NULL, NULL,
+                            NULL, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
+}
+
+int32_t
+quiesce_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+             dict_t *xdata)
+{
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+
+    priv = this->private;
 
-        gf_quiesce_enqueue (priv, stub);
+    if (priv->pass_through) {
+        STACK_WIND(frame, default_link_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->link, oldloc, newloc, xdata);
+        return 0;
+    }
 
+    stub = fop_link_stub(frame, default_link_resume, oldloc, newloc, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(link, frame, -1, ENOMEM, NULL, NULL, NULL, NULL,
+                            NULL);
         return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
 int32_t
-quiesce_readlink (call_frame_t *frame,
-		  xlator_t *this,
-		  loc_t *loc,
-		  size_t size)
+quiesce_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc,
+               loc_t *newloc, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_readlink_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->readlink,
-                            loc,
-                            size);
-	        return 0;
-        }
+    if (priv->pass_through) {
+        STACK_WIND(frame, default_rename_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
+        return 0;
+    }
 
-        stub = fop_readlink_stub (frame, default_readlink_resume, loc, size);
-        if (!stub) {
-                STACK_UNWIND_STRICT (readlink, frame, -1, ENOMEM, NULL, NULL);
-                return 0;
-        }
+    stub = fop_rename_stub(frame, default_rename_resume, oldloc, newloc, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(rename, frame, -1, ENOMEM, NULL, NULL, NULL, NULL,
+                            NULL, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
+}
 
-        gf_quiesce_enqueue (priv, stub);
+int
+quiesce_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+                loc_t *loc, mode_t umask, dict_t *xdata)
+{
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+
+    priv = this->private;
+
+    if (priv->pass_through) {
+        STACK_WIND(frame, default_symlink_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->symlink, linkpath, loc, umask,
+                   xdata);
+        return 0;
+    }
 
+    stub = fop_symlink_stub(frame, default_symlink_resume, linkpath, loc, umask,
+                            xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(symlink, frame, -1, ENOMEM, NULL, NULL, NULL, NULL,
+                            NULL);
         return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
+int
+quiesce_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+              dict_t *xdata)
+{
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+
+    priv = this->private;
+
+    if (priv->pass_through) {
+        STACK_WIND(frame, default_rmdir_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->rmdir, loc, flags, xdata);
+        return 0;
+    }
+
+    stub = fop_rmdir_stub(frame, default_rmdir_resume, loc, flags, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(rmdir, frame, -1, ENOMEM, NULL, NULL, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
+}
 
 int32_t
-quiesce_access (call_frame_t *frame,
-		xlator_t *this,
-		loc_t *loc,
-		int32_t mask)
+quiesce_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+               dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_access_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->access,
-                            loc,
-                            mask);
-	        return 0;
-        }
+    if (priv->pass_through) {
+        STACK_WIND(frame, default_unlink_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->unlink, loc, xflag, xdata);
+        return 0;
+    }
 
-        stub = fop_access_stub (frame, default_access_resume, loc, mask);
-        if (!stub) {
-                STACK_UNWIND_STRICT (access, frame, -1, ENOMEM);
-                return 0;
-        }
+    stub = fop_unlink_stub(frame, default_unlink_resume, loc, xflag, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(unlink, frame, -1, ENOMEM, NULL, NULL, NULL);
+        return 0;
+    }
 
-        gf_quiesce_enqueue (priv, stub);
+    gf_quiesce_enqueue(this, stub);
 
-        return 0;
+    return 0;
 }
 
-int32_t
-quiesce_ftruncate (call_frame_t *frame,
-		   xlator_t *this,
-		   fd_t *fd,
-		   off_t offset)
+int
+quiesce_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+              mode_t umask, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_ftruncate_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->ftruncate,
-                            fd,
-                            offset);
-	        return 0;
-        }
+    if (priv->pass_through) {
+        STACK_WIND(frame, default_mkdir_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->mkdir, loc, mode, umask, xdata);
+        return 0;
+    }
 
-        stub = fop_ftruncate_stub (frame, default_ftruncate_resume, fd, offset);
-        if (!stub) {
-                STACK_UNWIND_STRICT (ftruncate, frame, -1, ENOMEM, NULL, NULL);
-                return 0;
-        }
+    stub = fop_mkdir_stub(frame, default_mkdir_resume, loc, mode, umask, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(mkdir, frame, -1, ENOMEM, NULL, NULL, NULL, NULL,
+                            NULL);
+        return 0;
+    }
 
-        gf_quiesce_enqueue (priv, stub);
+    gf_quiesce_enqueue(this, stub);
 
+    return 0;
+}
+
+int
+quiesce_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+              dev_t rdev, mode_t umask, dict_t *xdata)
+{
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+
+    priv = this->private;
+
+    if (priv->pass_through) {
+        STACK_WIND(frame, default_mknod_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->mknod, loc, mode, rdev, umask,
+                   xdata);
         return 0;
+    }
+
+    stub = fop_mknod_stub(frame, default_mknod_resume, loc, mode, rdev, umask,
+                          xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(mknod, frame, -1, ENOMEM, NULL, NULL, NULL, NULL,
+                            NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
 int32_t
-quiesce_getxattr (call_frame_t *frame,
-		  xlator_t *this,
-		  loc_t *loc,
-		  const char *name)
+quiesce_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                  dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_getxattr_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->getxattr,
-                            loc,
-                            name);
-	        return 0;
-        }
+    if (priv->pass_through) {
+        STACK_WIND(frame, default_ftruncate_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+        return 0;
+    }
 
-        stub = fop_getxattr_stub (frame, default_getxattr_resume, loc, name);
-        if (!stub) {
-                STACK_UNWIND_STRICT (getxattr, frame, -1, ENOMEM, NULL);
-                return 0;
-        }
+    stub = fop_ftruncate_stub(frame, default_ftruncate_resume, fd, offset,
+                              xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(ftruncate, frame, -1, ENOMEM, NULL, NULL, NULL);
+        return 0;
+    }
 
-        gf_quiesce_enqueue (priv, stub);
+    gf_quiesce_enqueue(this, stub);
 
-        return 0;
+    return 0;
 }
 
+/* Re-transmittion */
 
 int32_t
-quiesce_xattrop (call_frame_t *frame,
-		 xlator_t *this,
-		 loc_t *loc,
-		 gf_xattrop_flags_t flags,
-		 dict_t *dict)
+quiesce_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+                 dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_xattrop_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->xattrop,
-                            loc,
-                            flags,
-                            dict);
-	        return 0;
-        }
+    if (priv && priv->pass_through) {
+        local = mem_get0(priv->local_pool);
+        loc_dup(loc, &local->loc);
+        local->size = size;
+        frame->local = local;
 
-        stub = fop_xattrop_stub (frame, default_xattrop_resume,
-                                 loc, flags, dict);
-        if (!stub) {
-                STACK_UNWIND_STRICT (xattrop, frame, -1, ENOMEM, NULL);
-                return 0;
-        }
-
-        gf_quiesce_enqueue (priv, stub);
+        STACK_WIND(frame, quiesce_readlink_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->readlink, loc, size, xdata);
+        return 0;
+    }
 
+    stub = fop_readlink_stub(frame, default_readlink_resume, loc, size, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(readlink, frame, -1, ENOMEM, NULL, NULL, NULL);
         return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
 int32_t
-quiesce_fxattrop (call_frame_t *frame,
-		  xlator_t *this,
-		  fd_t *fd,
-		  gf_xattrop_flags_t flags,
-		  dict_t *dict)
+quiesce_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+               dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_fxattrop_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->fxattrop,
-                            fd,
-                            flags,
-                            dict);
-	        return 0;
-        }
-
-        stub = fop_fxattrop_stub (frame, default_fxattrop_resume,
-                                  fd, flags, dict);
-        if (!stub) {
-                STACK_UNWIND_STRICT (fxattrop, frame, -1, ENOMEM, NULL);
-                return 0;
-        }
+    if (priv && priv->pass_through) {
+        local = mem_get0(priv->local_pool);
+        loc_dup(loc, &local->loc);
+        local->flag = mask;
+        frame->local = local;
 
-        gf_quiesce_enqueue (priv, stub);
+        STACK_WIND(frame, quiesce_access_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->access, loc, mask, xdata);
+        return 0;
+    }
 
+    stub = fop_access_stub(frame, default_access_resume, loc, mask, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(access, frame, -1, ENOMEM, NULL);
         return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
 int32_t
-quiesce_removexattr (call_frame_t *frame,
-		     xlator_t *this,
-		     loc_t *loc,
-		     const char *name)
+quiesce_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                  const char *name, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_removexattr_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->removexattr,
-                            loc,
-                            name);
-	        return 0;
-        }
+    if (priv && priv->pass_through) {
+        local = mem_get0(priv->local_pool);
+        local->fd = fd_ref(fd);
+        if (name)
+            local->name = gf_strdup(name);
 
-        stub = fop_removexattr_stub (frame, default_removexattr_resume,
-                                     loc, name);
-        if (!stub) {
-                STACK_UNWIND_STRICT (removexattr, frame, -1, ENOMEM);
-                return 0;
-        }
+        frame->local = local;
 
-        gf_quiesce_enqueue (priv, stub);
+        STACK_WIND(frame, quiesce_fgetxattr_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fgetxattr, fd, name, xdata);
+        return 0;
+    }
 
+    stub = fop_fgetxattr_stub(frame, default_fgetxattr_resume, fd, name, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(fgetxattr, frame, -1, ENOMEM, NULL, NULL);
         return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
 int32_t
-quiesce_lk (call_frame_t *frame,
-	    xlator_t *this,
-	    fd_t *fd,
-	    int32_t cmd,
-	    struct gf_flock *lock)
+quiesce_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_lk_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->lk,
-                            fd,
-                            cmd,
-                            lock);
-	        return 0;
-        }
+    if (priv && priv->pass_through) {
+        local = mem_get0(priv->local_pool);
+        loc_dup(loc, &local->loc);
+        frame->local = local;
 
-        stub = fop_lk_stub (frame, default_lk_resume, fd, cmd, lock);
-        if (!stub) {
-                STACK_UNWIND_STRICT (lk, frame, -1, ENOMEM, NULL);
-                return 0;
-        }
-
-        gf_quiesce_enqueue (priv, stub);
+        STACK_WIND(frame, quiesce_statfs_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->statfs, loc, xdata);
+        return 0;
+    }
 
+    stub = fop_statfs_stub(frame, default_statfs_resume, loc, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(statfs, frame, -1, ENOMEM, NULL, NULL);
         return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
+int32_t
+quiesce_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+                 dict_t *xdata)
+{
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    priv = this->private;
+
+    if (priv && priv->pass_through) {
+        local = mem_get0(priv->local_pool);
+        local->fd = fd_ref(fd);
+        local->flag = flags;
+        frame->local = local;
+
+        STACK_WIND(frame, quiesce_fsyncdir_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fsyncdir, fd, flags, xdata);
+        return 0;
+    }
+
+    stub = fop_fsyncdir_stub(frame, default_fsyncdir_resume, fd, flags, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(fsyncdir, frame, -1, ENOMEM, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
+}
 
 int32_t
-quiesce_inodelk (call_frame_t *frame, xlator_t *this,
-		 const char *volume, loc_t *loc, int32_t cmd,
-                 struct gf_flock *lock)
+quiesce_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+                dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_inodelk_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->inodelk,
-                            volume, loc, cmd, lock);
-	        return 0;
-        }
+    if (priv && priv->pass_through) {
+        local = mem_get0(priv->local_pool);
+        loc_dup(loc, &local->loc);
+        local->fd = fd_ref(fd);
+        frame->local = local;
 
-        stub = fop_inodelk_stub (frame, default_inodelk_resume,
-                                 volume, loc, cmd, lock);
-        if (!stub) {
-                STACK_UNWIND_STRICT (inodelk, frame, -1, ENOMEM);
-                return 0;
-        }
+        STACK_WIND(frame, quiesce_opendir_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->opendir, loc, fd, xdata);
+        return 0;
+    }
+
+    stub = fop_opendir_stub(frame, default_opendir_resume, loc, fd, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(opendir, frame, -1, ENOMEM, NULL, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
+}
+
+int32_t
+quiesce_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    priv = this->private;
 
-        gf_quiesce_enqueue (priv, stub);
+    if (priv && priv->pass_through) {
+        local = mem_get0(priv->local_pool);
+        local->fd = fd_ref(fd);
+        frame->local = local;
 
+        STACK_WIND(frame, quiesce_fstat_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fstat, fd, xdata);
         return 0;
+    }
+
+    stub = fop_fstat_stub(frame, default_fstat_resume, fd, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(fstat, frame, -1, ENOMEM, NULL, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
 int32_t
-quiesce_finodelk (call_frame_t *frame, xlator_t *this,
-		  const char *volume, fd_t *fd, int32_t cmd, struct gf_flock *lock)
+quiesce_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+              dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_finodelk_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->finodelk,
-                            volume, fd, cmd, lock);
-	        return 0;
-        }
+    if (priv && priv->pass_through) {
+        local = mem_get0(priv->local_pool);
+        local->fd = fd_ref(fd);
+        local->flag = flags;
+        frame->local = local;
 
-        stub = fop_finodelk_stub (frame, default_finodelk_resume,
-                                  volume, fd, cmd, lock);
-        if (!stub) {
-                STACK_UNWIND_STRICT (finodelk, frame, -1, ENOMEM);
-                return 0;
-        }
+        STACK_WIND(frame, quiesce_fsync_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fsync, fd, flags, xdata);
+        return 0;
+    }
+
+    stub = fop_fsync_stub(frame, default_fsync_resume, fd, flags, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(fsync, frame, -1, ENOMEM, NULL, NULL, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
+}
+
+int32_t
+quiesce_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    priv = this->private;
 
-        gf_quiesce_enqueue (priv, stub);
+    if (priv && priv->pass_through) {
+        local = mem_get0(priv->local_pool);
+        local->fd = fd_ref(fd);
+        frame->local = local;
 
+        STACK_WIND(frame, quiesce_flush_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->flush, fd, xdata);
         return 0;
+    }
+
+    stub = fop_flush_stub(frame, default_flush_resume, fd, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(flush, frame, -1, ENOMEM, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
 int32_t
-quiesce_entrylk (call_frame_t *frame, xlator_t *this,
-		 const char *volume, loc_t *loc, const char *basename,
-		 entrylk_cmd cmd, entrylk_type type)
+quiesce_writev(call_frame_t *frame, xlator_t *this, fd_t *fd,
+               struct iovec *vector, int32_t count, off_t off, uint32_t flags,
+               struct iobref *iobref, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame, default_entrylk_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->entrylk,
-                            volume, loc, basename, cmd, type);
-	        return 0;
-        }
+    if (priv && priv->pass_through) {
+        STACK_WIND(frame, default_writev_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->writev, fd, vector, count, off,
+                   flags, iobref, xdata);
+        return 0;
+    }
 
-        stub = fop_entrylk_stub (frame, default_entrylk_resume,
-                                 volume, loc, basename, cmd, type);
-        if (!stub) {
-                STACK_UNWIND_STRICT (entrylk, frame, -1, ENOMEM);
-                return 0;
-        }
+    stub = fop_writev_stub(frame, default_writev_resume, fd, vector, count, off,
+                           flags, iobref, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(writev, frame, -1, ENOMEM, NULL, NULL, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
 
-        gf_quiesce_enqueue (priv, stub);
+    return 0;
+}
 
+int32_t
+quiesce_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+              off_t offset, uint32_t flags, dict_t *xdata)
+{
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    priv = this->private;
+
+    if (priv && priv->pass_through) {
+        local = mem_get0(priv->local_pool);
+        local->fd = fd_ref(fd);
+        local->size = size;
+        local->offset = offset;
+        local->io_flag = flags;
+        frame->local = local;
+
+        STACK_WIND(frame, quiesce_readv_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->readv, fd, size, offset, flags,
+                   xdata);
         return 0;
+    }
+
+    stub = fop_readv_stub(frame, default_readv_resume, fd, size, offset, flags,
+                          xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(readv, frame, -1, ENOMEM, NULL, 0, NULL, NULL,
+                            NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
 int32_t
-quiesce_fentrylk (call_frame_t *frame, xlator_t *this,
-		  const char *volume, fd_t *fd, const char *basename,
-		  entrylk_cmd cmd, entrylk_type type)
+quiesce_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+             fd_t *fd, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame, default_fentrylk_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->fentrylk,
-                            volume, fd, basename, cmd, type);
-	        return 0;
-        }
+    if (priv && priv->pass_through) {
+        local = mem_get0(priv->local_pool);
+        loc_dup(loc, &local->loc);
+        local->fd = fd_ref(fd);
 
-        stub = fop_fentrylk_stub (frame, default_fentrylk_resume,
-                                  volume, fd, basename, cmd, type);
-        if (!stub) {
-                STACK_UNWIND_STRICT (fentrylk, frame, -1, ENOMEM);
-                return 0;
-        }
+        /* Don't send O_APPEND below, as write() re-transmittions can
+           fail with O_APPEND */
+        local->flag = (flags & ~O_APPEND);
+        frame->local = local;
 
-        gf_quiesce_enqueue (priv, stub);
+        STACK_WIND(frame, quiesce_open_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->open, loc, (flags & ~O_APPEND), fd,
+                   xdata);
+        return 0;
+    }
 
+    stub = fop_open_stub(frame, default_open_resume, loc, (flags & ~O_APPEND),
+                         fd, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(open, frame, -1, ENOMEM, NULL, NULL);
         return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
 int32_t
-quiesce_rchecksum (call_frame_t *frame,
-                   xlator_t *this,
-                   fd_t *fd, off_t offset,
-                   int32_t len)
+quiesce_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                 const char *name, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_rchecksum_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->rchecksum,
-                            fd, offset, len);
-	        return 0;
-        }
+    if (priv && priv->pass_through) {
+        local = mem_get0(priv->local_pool);
+        loc_dup(loc, &local->loc);
+        if (name)
+            local->name = gf_strdup(name);
 
-        stub = fop_rchecksum_stub (frame, default_rchecksum_resume,
-                                   fd, offset, len);
-        if (!stub) {
-                STACK_UNWIND_STRICT (rchecksum, frame, -1, ENOMEM, 0, NULL);
-                return 0;
-        }
+        frame->local = local;
 
-        gf_quiesce_enqueue (priv, stub);
+        STACK_WIND(frame, quiesce_getxattr_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->getxattr, loc, name, xdata);
+        return 0;
+    }
 
+    stub = fop_getxattr_stub(frame, default_getxattr_resume, loc, name, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(getxattr, frame, -1, ENOMEM, NULL, NULL);
         return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
+int32_t
+quiesce_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+
+    priv = this->private;
+
+    if (priv && priv->pass_through) {
+        STACK_WIND(frame, default_xattrop_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->xattrop, loc, flags, dict, xdata);
+        return 0;
+    }
+
+    stub = fop_xattrop_stub(frame, default_xattrop_resume, loc, flags, dict,
+                            xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(xattrop, frame, -1, ENOMEM, NULL, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
+}
 
 int32_t
-quiesce_readdir (call_frame_t *frame,
-		 xlator_t *this,
-		 fd_t *fd,
-		 size_t size,
-		 off_t off)
+quiesce_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                 gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_readdir_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->readdir,
-                            fd, size, off);
-	        return 0;
-        }
+    if (priv && priv->pass_through) {
+        STACK_WIND(frame, default_fxattrop_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fxattrop, fd, flags, dict, xdata);
+        return 0;
+    }
 
-        stub = fop_readdir_stub (frame, default_readdir_resume, fd, size, off);
-        if (!stub) {
-                STACK_UNWIND_STRICT (readdir, frame, -1, ENOMEM, NULL);
-                return 0;
-        }
+    stub = fop_fxattrop_stub(frame, default_fxattrop_resume, fd, flags, dict,
+                             xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(fxattrop, frame, -1, ENOMEM, NULL, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
+}
+
+int32_t
+quiesce_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+           struct gf_flock *lock, dict_t *xdata)
+{
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+
+    priv = this->private;
 
-        gf_quiesce_enqueue (priv, stub);
+    if (priv && priv->pass_through) {
+        STACK_WIND(frame, default_lk_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->lk, fd, cmd, lock, xdata);
+        return 0;
+    }
 
+    stub = fop_lk_stub(frame, default_lk_resume, fd, cmd, lock, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(lk, frame, -1, ENOMEM, NULL, NULL);
         return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
+int32_t
+quiesce_inodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+                loc_t *loc, int32_t cmd, struct gf_flock *lock, dict_t *xdata)
+{
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+
+    priv = this->private;
+
+    if (priv && priv->pass_through) {
+        STACK_WIND(frame, default_inodelk_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->inodelk, volume, loc, cmd, lock,
+                   xdata);
+        return 0;
+    }
+
+    stub = fop_inodelk_stub(frame, default_inodelk_resume, volume, loc, cmd,
+                            lock, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(inodelk, frame, -1, ENOMEM, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
+}
 
 int32_t
-quiesce_readdirp (call_frame_t *frame,
-		  xlator_t *this,
-		  fd_t *fd,
-		  size_t size,
-		  off_t off)
+quiesce_finodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+                 fd_t *fd, int32_t cmd, struct gf_flock *lock, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_readdirp_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->readdirp,
-                            fd, size, off);
-	        return 0;
-        }
+    if (priv && priv->pass_through) {
+        STACK_WIND(frame, default_finodelk_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->finodelk, volume, fd, cmd, lock,
+                   xdata);
+        return 0;
+    }
 
-        stub = fop_readdirp_stub (frame, default_readdirp_resume, fd, size, off);
-        if (!stub) {
-                STACK_UNWIND_STRICT (readdirp, frame, -1, ENOMEM, NULL);
-                return 0;
-        }
+    stub = fop_finodelk_stub(frame, default_finodelk_resume, volume, fd, cmd,
+                             lock, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(finodelk, frame, -1, ENOMEM, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
 
-        gf_quiesce_enqueue (priv, stub);
+    return 0;
+}
+
+int32_t
+quiesce_entrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+                loc_t *loc, const char *basename, entrylk_cmd cmd,
+                entrylk_type type, dict_t *xdata)
+{
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+
+    priv = this->private;
 
+    if (priv && priv->pass_through) {
+        STACK_WIND(frame, default_entrylk_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->entrylk, volume, loc, basename, cmd,
+                   type, xdata);
         return 0;
+    }
+
+    stub = fop_entrylk_stub(frame, default_entrylk_resume, volume, loc,
+                            basename, cmd, type, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(entrylk, frame, -1, ENOMEM, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
 int32_t
-quiesce_setattr (call_frame_t *frame,
-                 xlator_t *this,
-                 loc_t *loc,
-                 struct iatt *stbuf,
-                 int32_t valid)
+quiesce_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+                 fd_t *fd, const char *basename, entrylk_cmd cmd,
+                 entrylk_type type, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_setattr_cbk,
-                            FIRST_CHILD (this),
-                            FIRST_CHILD (this)->fops->setattr,
-                            loc, stbuf, valid);
-	        return 0;
-        }
+    if (priv && priv->pass_through) {
+        STACK_WIND(frame, default_fentrylk_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fentrylk, volume, fd, basename, cmd,
+                   type, xdata);
+        return 0;
+    }
 
-        stub = fop_setattr_stub (frame, default_setattr_resume,
-                                   loc, stbuf, valid);
-        if (!stub) {
-                STACK_UNWIND_STRICT (setattr, frame, -1, ENOMEM, NULL, NULL);
-                return 0;
-        }
+    stub = fop_fentrylk_stub(frame, default_fentrylk_resume, volume, fd,
+                             basename, cmd, type, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(fentrylk, frame, -1, ENOMEM, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
+}
+
+int32_t
+quiesce_rchecksum(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                  int32_t len, dict_t *xdata)
+{
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    priv = this->private;
+
+    if (priv && priv->pass_through) {
+        local = mem_get0(priv->local_pool);
+        local->fd = fd_ref(fd);
+        local->offset = offset;
+        local->flag = len;
+        frame->local = local;
 
-        gf_quiesce_enqueue (priv, stub);
+        STACK_WIND(frame, quiesce_rchecksum_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->rchecksum, fd, offset, len, xdata);
+        return 0;
+    }
 
+    stub = fop_rchecksum_stub(frame, default_rchecksum_resume, fd, offset, len,
+                              xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(rchecksum, frame, -1, ENOMEM, 0, NULL, NULL);
         return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
 int32_t
-quiesce_truncate (call_frame_t *frame,
-		  xlator_t *this,
-		  loc_t *loc,
-		  off_t offset)
+quiesce_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                off_t off, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_truncate_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->truncate,
-                            loc,
-                            offset);
-	        return 0;
-        }
+    if (priv && priv->pass_through) {
+        local = mem_get0(priv->local_pool);
+        local->fd = fd_ref(fd);
+        local->size = size;
+        local->offset = off;
+        frame->local = local;
 
-        stub = fop_truncate_stub (frame, default_truncate_resume, loc, offset);
-        if (!stub) {
-                STACK_UNWIND_STRICT (truncate, frame, -1, ENOMEM, NULL, NULL);
-                return 0;
-        }
+        STACK_WIND(frame, quiesce_readdir_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->readdir, fd, size, off, xdata);
+        return 0;
+    }
+
+    stub = fop_readdir_stub(frame, default_readdir_resume, fd, size, off,
+                            xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(readdir, frame, -1, ENOMEM, NULL, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
 
-        gf_quiesce_enqueue (priv, stub);
+    return 0;
+}
 
+int32_t
+quiesce_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                 off_t off, dict_t *dict)
+{
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    priv = this->private;
+
+    if (priv && priv->pass_through) {
+        local = mem_get0(priv->local_pool);
+        local->fd = fd_ref(fd);
+        local->size = size;
+        local->offset = off;
+        local->dict = dict_ref(dict);
+        frame->local = local;
+
+        STACK_WIND(frame, quiesce_readdirp_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->readdirp, fd, size, off, dict);
         return 0;
+    }
+
+    stub = fop_readdirp_stub(frame, default_readdirp_resume, fd, size, off,
+                             dict);
+    if (!stub) {
+        STACK_UNWIND_STRICT(readdirp, frame, -1, ENOMEM, NULL, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
 int32_t
-quiesce_stat (call_frame_t *frame,
-	      xlator_t *this,
-	      loc_t *loc)
+quiesce_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                struct iatt *stbuf, int32_t valid, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_stat_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->stat,
-                            loc);
-	        return 0;
-        }
+    if (priv && priv->pass_through) {
+        STACK_WIND(frame, default_setattr_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->setattr, loc, stbuf, valid, xdata);
+        return 0;
+    }
 
-        stub = fop_stat_stub (frame, default_stat_resume, loc);
-        if (!stub) {
-                STACK_UNWIND_STRICT (stat, frame, -1, ENOMEM, NULL);
-                return 0;
-        }
+    stub = fop_setattr_stub(frame, default_setattr_resume, loc, stbuf, valid,
+                            xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(setattr, frame, -1, ENOMEM, NULL, NULL, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
+}
+
+int32_t
+quiesce_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    priv = this->private;
 
-        gf_quiesce_enqueue (priv, stub);
+    if (priv && priv->pass_through) {
+        local = mem_get0(priv->local_pool);
+        loc_dup(loc, &local->loc);
+        frame->local = local;
 
+        STACK_WIND(frame, quiesce_stat_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->stat, loc, xdata);
         return 0;
+    }
+
+    stub = fop_stat_stub(frame, default_stat_resume, loc, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(stat, frame, -1, ENOMEM, NULL, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
 int32_t
-quiesce_lookup (call_frame_t *frame,
-		xlator_t *this,
-		loc_t *loc,
-		dict_t *xattr_req)
+quiesce_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               dict_t *xattr_req)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_lookup_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->lookup,
-                            loc,
-                            xattr_req);
-	        return 0;
-        }
+    if (priv && priv->pass_through) {
+        local = mem_get0(priv->local_pool);
+        loc_dup(loc, &local->loc);
+        local->dict = dict_ref(xattr_req);
+        frame->local = local;
 
-        stub = fop_lookup_stub (frame, default_lookup_resume, loc, xattr_req);
-        if (!stub) {
-                STACK_UNWIND_STRICT (lookup, frame, -1, ENOMEM,
-                                     NULL, NULL, NULL, NULL);
-                return 0;
-        }
+        STACK_WIND(frame, quiesce_lookup_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->lookup, loc, xattr_req);
+        return 0;
+    }
+
+    stub = fop_lookup_stub(frame, default_lookup_resume, loc, xattr_req);
+    if (!stub) {
+        STACK_UNWIND_STRICT(lookup, frame, -1, ENOMEM, NULL, NULL, NULL, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
+}
+
+int32_t
+quiesce_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                 struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+
+    priv = this->private;
 
-        gf_quiesce_enqueue (priv, stub);
+    if (priv && priv->pass_through) {
+        STACK_WIND(frame, default_fsetattr_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fsetattr, fd, stbuf, valid, xdata);
+        return 0;
+    }
 
+    stub = fop_fsetattr_stub(frame, default_fsetattr_resume, fd, stbuf, valid,
+                             xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(fsetattr, frame, -1, ENOMEM, NULL, NULL, NULL);
         return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
 int32_t
-quiesce_fsetattr (call_frame_t *frame,
-                  xlator_t *this,
-                  fd_t *fd,
-                  struct iatt *stbuf,
-                  int32_t valid)
+quiesce_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+                  off_t offset, size_t len, dict_t *xdata)
 {
-	quiesce_priv_t *priv = NULL;
-        call_stub_t    *stub = NULL;
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
 
-        priv = this->private;
+    priv = this->private;
 
-        if (priv->pass_through) {
-                STACK_WIND (frame,
-                            default_fsetattr_cbk,
-                            FIRST_CHILD (this),
-                            FIRST_CHILD (this)->fops->fsetattr,
-                            fd, stbuf, valid);
-	        return 0;
-        }
+    if (priv && priv->pass_through) {
+        STACK_WIND(frame, default_fallocate_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fallocate, fd, mode, offset, len,
+                   xdata);
+        return 0;
+    }
 
-        stub = fop_fsetattr_stub (frame, default_fsetattr_resume,
-                                  fd, stbuf, valid);
+    stub = fop_fallocate_stub(frame, default_fallocate_resume, fd, mode, offset,
+                              len, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(fallocate, frame, -1, ENOMEM, NULL, NULL, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
+}
+
+int
+quiesce_seek_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, off_t offset, dict_t *xdata)
+{
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    local = frame->local;
+    frame->local = NULL;
+    if ((op_ret == -1) && (op_errno == ENOTCONN)) {
+        /* Re-transmit (by putting in the queue) */
+        stub = fop_seek_stub(frame, default_seek_resume, local->fd,
+                             local->offset, local->what, xdata);
         if (!stub) {
-                STACK_UNWIND_STRICT (fsetattr, frame, -1, ENOMEM, NULL, NULL);
-                return 0;
+            STACK_UNWIND_STRICT(seek, frame, -1, ENOMEM, 0, NULL);
+            goto out;
         }
 
-        gf_quiesce_enqueue (priv, stub);
+        gf_quiesce_enqueue(this, stub);
+        goto out;
+    }
+
+    STACK_UNWIND_STRICT(seek, frame, op_ret, op_errno, offset, xdata);
+out:
+    gf_quiesce_local_wipe(this, local);
+
+    return 0;
+}
+
+int
+quiesce_seek(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             gf_seek_what_t what, dict_t *xdata)
+{
+    quiesce_priv_t *priv = NULL;
+    call_stub_t *stub = NULL;
+    quiesce_local_t *local = NULL;
+
+    priv = this->private;
+
+    if (priv && priv->pass_through) {
+        local = mem_get0(priv->local_pool);
+        local->fd = fd_ref(fd);
+        local->offset = offset;
+        local->what = what;
+
+        frame->local = local;
 
+        STACK_WIND(frame, quiesce_seek_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->seek, fd, offset, what, xdata);
         return 0;
+    }
+
+    stub = fop_seek_stub(frame, default_seek_resume, fd, offset, what, xdata);
+    if (!stub) {
+        STACK_UNWIND_STRICT(seek, frame, -1, ENOMEM, 0, NULL);
+        return 0;
+    }
+
+    gf_quiesce_enqueue(this, stub);
+
+    return 0;
 }
 
 int32_t
-mem_acct_init (xlator_t *this)
+mem_acct_init(xlator_t *this)
 {
-        int     ret = -1;
+    int ret = -1;
 
-        ret = xlator_mem_acct_init (this, gf_quiesce_mt_end + 1);
+    ret = xlator_mem_acct_init(this, gf_quiesce_mt_end + 1);
 
-        return ret;
+    return ret;
 }
 
 int
-init (xlator_t *this)
+reconfigure(xlator_t *this, dict_t *options)
 {
-        int ret = -1;
-        quiesce_priv_t *priv = NULL;
+    int32_t ret = -1;
+    quiesce_priv_t *priv = NULL;
 
-	if (!this->children || this->children->next) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"'quiesce' not configured with exactly one child");
-                goto out;
-	}
+    priv = this->private;
 
-	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"dangling volume. check volfile ");
-	}
+    GF_OPTION_RECONF("timeout", priv->timeout, options, time, out);
+    GF_OPTION_RECONF("failover-hosts", priv->failover_hosts, options, str, out);
+    gf_quiesce_populate_failover_hosts(this, priv, priv->failover_hosts);
 
-        priv = GF_CALLOC (1, sizeof (*priv), gf_quiesce_mt_priv_t);
-        if (!priv)
-                goto out;
+    ret = 0;
+out:
+    return ret;
+}
+
+int
+init(xlator_t *this)
+{
+    int ret = -1;
+    quiesce_priv_t *priv = NULL;
+
+    if (!this->children || this->children->next) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "'quiesce' not configured with exactly one child");
+        goto out;
+    }
+
+    if (!this->parents) {
+        gf_log(this->name, GF_LOG_WARNING, "dangling volume. check volfile ");
+    }
+
+    priv = GF_CALLOC(1, sizeof(*priv), gf_quiesce_mt_priv_t);
+    if (!priv)
+        goto out;
+
+    INIT_LIST_HEAD(&priv->failover_list);
 
-        LOCK_INIT (&priv->lock);
-        priv->pass_through = _gf_false;
+    GF_OPTION_INIT("timeout", priv->timeout, time, out);
+    GF_OPTION_INIT("failover-hosts", priv->failover_hosts, str, out);
+    gf_quiesce_populate_failover_hosts(this, priv, priv->failover_hosts);
 
-        INIT_LIST_HEAD (&priv->req);
+    priv->local_pool = mem_pool_new(quiesce_local_t,
+                                    GF_FOPS_EXPECTED_IN_PARALLEL);
 
-        this->private = priv;
-        ret = 0;
+    LOCK_INIT(&priv->lock);
+    priv->pass_through = _gf_false;
+
+    INIT_LIST_HEAD(&priv->req);
+
+    this->private = priv;
+    ret = 0;
 out:
-        return ret;
+    return ret;
 }
 
 void
-fini (xlator_t *this)
+fini(xlator_t *this)
 {
-        quiesce_priv_t *priv = NULL;
+    quiesce_priv_t *priv = NULL;
 
-        priv = this->private;
-        if (!priv)
-                goto out;
-        this->private = NULL;
+    priv = this->private;
+    if (!priv)
+        goto out;
+    this->private = NULL;
 
-        LOCK_DESTROY (&priv->lock);
-        GF_FREE (priv);
+    mem_pool_destroy(priv->local_pool);
+    priv->local_pool = NULL;
+    LOCK_DESTROY(&priv->lock);
+    GF_FREE(priv);
 out:
-        return;
+    return;
 }
 
 int
-notify (xlator_t *this, int event, void *data, ...)
+notify(xlator_t *this, int event, void *data, ...)
 {
-        int             ret     = 0;
-        quiesce_priv_t *priv    = NULL;
-        struct timeval  timeout = {0,};
-
-        priv = this->private;
-        if (!priv)
-                goto out;
-
-        switch (event) {
-        case GF_EVENT_CHILD_UP:
-        {
-                ret = pthread_create (&priv->thr, NULL, gf_quiesce_dequeue_start,
-                                      this);
-                if (ret) {
-                        gf_log (this->name, GF_LOG_ERROR,
-                                "failed to create the quiesce-dequeue thread");
-                }
-
-                LOCK (&priv->lock);
-                {
-                        priv->pass_through = _gf_true;
-                }
-                UNLOCK (&priv->lock);
-                break;
+    int ret = 0;
+    quiesce_priv_t *priv = NULL;
+
+    priv = this->private;
+    if (!priv)
+        goto out;
+
+    switch (event) {
+        case GF_EVENT_CHILD_UP: {
+            ret = gf_thread_create(&priv->thr, NULL, gf_quiesce_dequeue_start,
+                                   this, "quiesce");
+            if (ret) {
+                gf_log(this->name, GF_LOG_ERROR,
+                       "failed to create the quiesce-dequeue thread");
+            }
+
+            LOCK(&priv->lock);
+            {
+                priv->pass_through = _gf_true;
+            }
+            UNLOCK(&priv->lock);
+            break;
         }
         case GF_EVENT_CHILD_DOWN:
-                LOCK (&priv->lock);
-                {
-                        priv->pass_through = _gf_false;
-                }
-                UNLOCK (&priv->lock);
-
-                if (!priv->timer) {
-                        timeout.tv_sec = 20;
-                        timeout.tv_usec = 0;
-
-                        gf_timer_call_cancel (this->ctx, priv->timer);
-                        priv->timer = gf_timer_call_after (this->ctx,
-                                                           timeout,
-                                                           gf_quiesce_timeout,
-                                                           (void *) this);
-
-                        if (priv->timer == NULL) {
-                                gf_log (this->name, GF_LOG_ERROR,
-                                        "Cannot create timer");
-                        }
-                }
-                break;
+            LOCK(&priv->lock);
+            {
+                priv->pass_through = _gf_false;
+                __gf_quiesce_start_timer(this, priv);
+            }
+            UNLOCK(&priv->lock);
+            break;
         default:
-                break;
-        }
+            break;
+    }
 
-        ret = default_notify (this, event, data);
+    ret = default_notify(this, event, data);
 out:
-        return ret;
+    return ret;
 }
 
-
 struct xlator_fops fops = {
-	.lookup      = quiesce_lookup,
-	.mknod       = quiesce_mknod,
-	.create      = quiesce_create,
-	.stat        = quiesce_stat,
-	.fstat       = quiesce_fstat,
-	.truncate    = quiesce_truncate,
-	.ftruncate   = quiesce_ftruncate,
-	.access      = quiesce_access,
-	.readlink    = quiesce_readlink,
-	.setxattr    = quiesce_setxattr,
-	.getxattr    = quiesce_getxattr,
-	.removexattr = quiesce_removexattr,
-	.open        = quiesce_open,
-	.readv       = quiesce_readv,
-	.writev      = quiesce_writev,
-	.flush       = quiesce_flush,
-	.fsync       = quiesce_fsync,
-	.statfs      = quiesce_statfs,
-	.lk          = quiesce_lk,
-	.opendir     = quiesce_opendir,
-	.readdir     = quiesce_readdir,
-	.readdirp    = quiesce_readdirp,
-	.fsyncdir    = quiesce_fsyncdir,
-	.symlink     = quiesce_symlink,
-	.unlink      = quiesce_unlink,
-	.link        = quiesce_link,
-	.mkdir       = quiesce_mkdir,
-	.rmdir       = quiesce_rmdir,
-	.rename      = quiesce_rename,
-	.inodelk     = quiesce_inodelk,
-	.finodelk    = quiesce_finodelk,
-	.entrylk     = quiesce_entrylk,
-	.fentrylk    = quiesce_fentrylk,
-	.xattrop     = quiesce_xattrop,
-	.fxattrop    = quiesce_fxattrop,
-        .setattr     = quiesce_setattr,
-        .fsetattr    = quiesce_fsetattr,
+    /* write/modifying fops */
+    .mknod = quiesce_mknod,
+    .create = quiesce_create,
+    .truncate = quiesce_truncate,
+    .ftruncate = quiesce_ftruncate,
+    .setxattr = quiesce_setxattr,
+    .fsetxattr = quiesce_fsetxattr,
+    .removexattr = quiesce_removexattr,
+    .fremovexattr = quiesce_fremovexattr,
+    .symlink = quiesce_symlink,
+    .unlink = quiesce_unlink,
+    .link = quiesce_link,
+    .mkdir = quiesce_mkdir,
+    .rmdir = quiesce_rmdir,
+    .rename = quiesce_rename,
+    .fallocate = quiesce_fallocate,
+
+    /* The below calls are known to change state, hence
+       re-transmittion is not advised */
+    .lk = quiesce_lk,
+    .inodelk = quiesce_inodelk,
+    .finodelk = quiesce_finodelk,
+    .entrylk = quiesce_entrylk,
+    .fentrylk = quiesce_fentrylk,
+    .xattrop = quiesce_xattrop,
+    .fxattrop = quiesce_fxattrop,
+    .setattr = quiesce_setattr,
+    .fsetattr = quiesce_fsetattr,
+
+    /* Special case, re-transmittion is not harmful *
+     * as offset is properly sent from above layers */
+    /* TODO: not re-transmitted as of now */
+    .writev = quiesce_writev,
+
+    /* re-transmittable fops */
+    .lookup = quiesce_lookup,
+    .stat = quiesce_stat,
+    .fstat = quiesce_fstat,
+    .access = quiesce_access,
+    .readlink = quiesce_readlink,
+    .getxattr = quiesce_getxattr,
+    .fgetxattr = quiesce_fgetxattr,
+    .open = quiesce_open,
+    .readv = quiesce_readv,
+    .flush = quiesce_flush,
+    .fsync = quiesce_fsync,
+    .statfs = quiesce_statfs,
+    .opendir = quiesce_opendir,
+    .readdir = quiesce_readdir,
+    .readdirp = quiesce_readdirp,
+    .fsyncdir = quiesce_fsyncdir,
+    .seek = quiesce_seek,
 };
 
-struct xlator_dumpops dumpops = {
-};
+struct xlator_dumpops dumpops;
 
+struct xlator_cbks cbks;
 
-struct xlator_cbks cbks = {
+struct volume_options options[] = {
+    {
+        .key = {"timeout"},
+        .type = GF_OPTION_TYPE_TIME,
+        .default_value = "45",
+        .description =
+            "After 'timeout' seconds since the time 'quiesce' "
+            "option was set to \"!pass-through\", acknowledgements to file "
+            "operations are no longer quiesced and previously "
+            "quiesced acknowledgements are sent to the application",
+        .op_version = {GD_OP_VERSION_4_0_0},
+        .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+    },
+    {.key = {"failover-hosts"},
+     .type = GF_OPTION_TYPE_INTERNET_ADDRESS_LIST,
+     .op_version = {GD_OP_VERSION_4_0_0},
+     .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .description = "It is a comma separated list of hostname/IP "
+                    "addresses. It Specifies the list of hosts where "
+                    "the gfproxy daemons are running, to which the "
+                    "the thin clients can failover to."},
+    {.key = {NULL}},
 };
 
-
-struct volume_options options[] = {
-	{ .key  = {NULL} },
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {GD_OP_VERSION_3_12_0},
+    .dumpops = &dumpops,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "quiesce",
+    .category = GF_TECH_PREVIEW,
 };
diff --git a/xlators/features/quiesce/src/quiesce.h b/xlators/features/quiesce/src/quiesce.h
index 2e1a560e2e9..6ab2af40a56 100644
--- a/xlators/features/quiesce/src/quiesce.h
+++ b/xlators/features/quiesce/src/quiesce.h
@@ -1,37 +1,65 @@
 /*
-   Copyright (c) 2010 Gluster, Inc. <http://www.gluster.com>
+   Copyright (c) 2010-2012 Red Hat, Inc. <http://www.redhat.com>
    This file is part of GlusterFS.
 
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
 */
 
-
 #ifndef __QUIESCE_H__
 #define __QUIESCE_H__
 
 #include "quiesce-mem-types.h"
-#include "xlator.h"
-#include "timer.h"
+#include "quiesce-messages.h"
+#include <glusterfs/xlator.h>
+#include <glusterfs/timer.h>
+
+#define GF_FOPS_EXPECTED_IN_PARALLEL 512
 
 typedef struct {
-        gf_timer_t       *timer;
-        gf_boolean_t      pass_through;
-        gf_lock_t         lock;
-        struct list_head  req;
-        int               queue_size;
-        pthread_t         thr;
+    struct list_head list;
+    char *addr;
+    gf_boolean_t tried; /* indicates attempted connecting */
+} quiesce_failover_hosts_t;
+
+typedef struct {
+    gf_timer_t *timer;
+    gf_boolean_t pass_through;
+    gf_lock_t lock;
+    struct list_head req;
+    int queue_size;
+    pthread_t thr;
+    struct mem_pool *local_pool;
+    uint32_t timeout;
+    char *failover_hosts;
+    struct list_head failover_list;
 } quiesce_priv_t;
 
+typedef struct {
+    fd_t *fd;
+    char *name;
+    char *volname;
+    loc_t loc;
+    off_t size;
+    off_t offset;
+    mode_t mode;
+    int32_t flag;
+    struct iatt stbuf;
+    struct iovec *vector;
+    struct iobref *iobref;
+    dict_t *dict;
+    struct gf_flock flock;
+    entrylk_cmd cmd;
+    entrylk_type type;
+    gf_xattrop_flags_t xattrop_flags;
+    int32_t wbflags;
+    uint32_t io_flag;
+    /* for fallocate */
+    size_t len;
+    /* for lseek */
+    gf_seek_what_t what;
+} quiesce_local_t;
+
 #endif
diff --git a/xlators/features/quota/src/Makefile.am b/xlators/features/quota/src/Makefile.am
index fe373c8b515..1c2dcef0ca3 100644
--- a/xlators/features/quota/src/Makefile.am
+++ b/xlators/features/quota/src/Makefile.am
@@ -1,15 +1,29 @@
-xlator_LTLIBRARIES = quota.la
+if WITH_SERVER
+xlator_LTLIBRARIES = quota.la quotad.la
+endif
 xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
 
-quota_la_LDFLAGS = -module -avoidversion
+quota_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+quotad_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
 
-quota_la_SOURCES = quota.c
-quota_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+quota_la_SOURCES = quota.c quota-enforcer-client.c
+quota_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la \
+                  $(top_builddir)/rpc/xdr/src/libgfxdr.la \
+                  $(top_builddir)/rpc/rpc-lib/src/libgfrpc.la
 
-noinst_HEADERS = quota-mem-types.h
+quotad_la_SOURCES = quotad.c quotad-helpers.c quotad-aggregator.c
+quotad_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la \
+                   $(top_builddir)/rpc/xdr/src/libgfxdr.la \
+                   $(top_builddir)/rpc/rpc-lib/src/libgfrpc.la
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS) \
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
+noinst_HEADERS = quota-mem-types.h quota.h quotad-aggregator.h \
+	quotad-helpers.h quota-messages.h
 
-CLEANFILES = 
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src/ -I$(top_builddir)/rpc/xdr/src/ \
+	-I$(top_srcdir)/rpc/rpc-lib/src \
+	-I$(top_srcdir)/xlators/cluster/dht/src
 
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/quota/src/quota-enforcer-client.c b/xlators/features/quota/src/quota-enforcer-client.c
new file mode 100644
index 00000000000..480d64ade27
--- /dev/null
+++ b/xlators/features/quota/src/quota-enforcer-client.c
@@ -0,0 +1,503 @@
+/*
+   Copyright (c) 2010-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include <stdio.h>
+#include <string.h>
+#include <netinet/in.h>
+#include <sys/socket.h>
+#include <sys/types.h>
+#include <sys/resource.h>
+#include <sys/file.h>
+#include <netdb.h>
+#include <signal.h>
+#include <libgen.h>
+
+#include <sys/utsname.h>
+
+#include <stdint.h>
+#include <pthread.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <time.h>
+#include <semaphore.h>
+#include <errno.h>
+
+#ifdef HAVE_MALLOC_H
+#include <malloc.h>
+#endif
+
+#include "quota.h"
+#include "quota-messages.h"
+
+extern struct rpc_clnt_program quota_enforcer_clnt;
+
+int32_t
+quota_validate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, inode_t *inode,
+                   struct iatt *buf, dict_t *xdata, struct iatt *postparent);
+
+int
+quota_enforcer_submit_request(void *req, call_frame_t *frame,
+                              rpc_clnt_prog_t *prog, int procnum,
+                              struct iobref *iobref, xlator_t *this,
+                              fop_cbk_fn_t cbkfn, xdrproc_t xdrproc)
+{
+    int ret = -1;
+    int count = 0;
+    struct iovec iov = {
+        0,
+    };
+    struct iobuf *iobuf = NULL;
+    char new_iobref = 0;
+    ssize_t xdr_size = 0;
+    quota_priv_t *priv = NULL;
+
+    GF_ASSERT(this);
+
+    priv = this->private;
+
+    if (req) {
+        xdr_size = xdr_sizeof(xdrproc, req);
+        iobuf = iobuf_get2(this->ctx->iobuf_pool, xdr_size);
+        if (!iobuf) {
+            goto out;
+        }
+
+        if (!iobref) {
+            iobref = iobref_new();
+            if (!iobref) {
+                goto out;
+            }
+
+            new_iobref = 1;
+        }
+
+        iobref_add(iobref, iobuf);
+
+        iov.iov_base = iobuf->ptr;
+        iov.iov_len = iobuf_size(iobuf);
+
+        /* Create the xdr payload */
+        ret = xdr_serialize_generic(iov, req, xdrproc);
+        if (ret == -1) {
+            goto out;
+        }
+        iov.iov_len = ret;
+        count = 1;
+    }
+
+    /* Send the msg */
+    ret = rpc_clnt_submit(priv->rpc_clnt, prog, procnum, cbkfn, &iov, count,
+                          NULL, 0, iobref, frame, NULL, 0, NULL, 0, NULL);
+    ret = 0;
+
+out:
+    if (new_iobref)
+        iobref_unref(iobref);
+    if (iobuf)
+        iobuf_unref(iobuf);
+
+    return ret;
+}
+
+int
+quota_enforcer_lookup_cbk(struct rpc_req *req, struct iovec *iov, int count,
+                          void *myframe)
+{
+    quota_local_t *local = NULL;
+    call_frame_t *frame = NULL;
+    int ret = 0;
+    gfs3_lookup_rsp rsp = {
+        0,
+    };
+    struct iatt stbuf = {
+        0,
+    };
+    struct iatt postparent = {
+        0,
+    };
+    int op_errno = EINVAL;
+    dict_t *xdata = NULL;
+    inode_t *inode = NULL;
+    xlator_t *this = NULL;
+    quota_priv_t *priv = NULL;
+    struct timespec retry_delay = {
+        0,
+    };
+    gf_timer_t *timer = NULL;
+
+    this = THIS;
+
+    frame = myframe;
+    local = frame->local;
+    inode = local->validate_loc.inode;
+    priv = this->private;
+
+    if (-1 == req->rpc_status) {
+        rsp.op_ret = -1;
+        op_errno = ENOTCONN;
+        goto out;
+    }
+
+    ret = xdr_to_generic(*iov, &rsp, (xdrproc_t)xdr_gfs3_lookup_rsp);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, Q_MSG_XDR_DECODING_FAILED,
+               "XDR decoding failed");
+        rsp.op_ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    op_errno = gf_error_to_errno(rsp.op_errno);
+    gf_stat_to_iatt(&rsp.postparent, &postparent);
+
+    if (rsp.op_ret == -1)
+        goto out;
+
+    rsp.op_ret = -1;
+    gf_stat_to_iatt(&rsp.stat, &stbuf);
+
+    GF_PROTOCOL_DICT_UNSERIALIZE(frame->this, xdata, (rsp.xdata.xdata_val),
+                                 (rsp.xdata.xdata_len), rsp.op_ret, op_errno,
+                                 out);
+
+    if ((!gf_uuid_is_null(inode->gfid)) &&
+        (gf_uuid_compare(stbuf.ia_gfid, inode->gfid) != 0)) {
+        gf_msg_debug(frame->this->name, ESTALE, "gfid changed for %s",
+                     local->validate_loc.path);
+        rsp.op_ret = -1;
+        op_errno = ESTALE;
+        goto out;
+    }
+
+    rsp.op_ret = 0;
+
+out:
+    rsp.op_errno = op_errno;
+
+    /* We need to retry connecting to quotad on ENOTCONN error.
+     * Suppose if there are two volumes vol1 and vol2,
+     * and quota is enabled and limit is set on vol1.
+     * Now if IO is happening on vol1 and quota is enabled/disabled
+     * on vol2, quotad gets restarted and client will receive
+     * ENOTCONN in the IO path of vol1
+     */
+    if (rsp.op_ret == -1 && rsp.op_errno == ENOTCONN) {
+        if (local->quotad_conn_retry >= 12) {
+            priv->quotad_conn_status = 1;
+            gf_log(this->name, GF_LOG_WARNING,
+                   "failed to connect "
+                   "to quotad after retry count %d)",
+                   local->quotad_conn_retry);
+        } else {
+            local->quotad_conn_retry++;
+        }
+
+        if (priv->quotad_conn_status == 0) {
+            /* retry connecting after 5secs for 12 retries
+             * (up to 60sec).
+             */
+            gf_log(this->name, GF_LOG_DEBUG,
+                   "retry connecting to "
+                   "quotad (retry count %d)",
+                   local->quotad_conn_retry);
+
+            retry_delay.tv_sec = 5;
+            retry_delay.tv_nsec = 0;
+            timer = gf_timer_call_after(this->ctx, retry_delay,
+                                        _quota_enforcer_lookup, (void *)frame);
+            if (timer == NULL) {
+                gf_log(this->name, GF_LOG_WARNING,
+                       "failed to "
+                       "set quota_enforcer_lookup with timer");
+            } else {
+                goto clean;
+            }
+        }
+    } else {
+        priv->quotad_conn_status = 0;
+    }
+
+    if (rsp.op_ret == -1) {
+        /* any error other than ENOENT */
+        if (rsp.op_errno != ENOENT)
+            gf_msg(
+                this->name, GF_LOG_WARNING, rsp.op_errno, Q_MSG_LOOKUP_FAILED,
+                "Getting cluster-wide size of directory failed "
+                "(path: %s gfid:%s)",
+                local->validate_loc.path, loc_gfid_utoa(&local->validate_loc));
+        else
+            gf_msg_trace(this->name, ENOENT, "not found on remote node");
+
+    } else if (local->quotad_conn_retry) {
+        gf_log(this->name, GF_LOG_DEBUG,
+               "connected to quotad after "
+               "retry count %d",
+               local->quotad_conn_retry);
+    }
+
+    local->validate_cbk(frame, NULL, this, rsp.op_ret, rsp.op_errno, inode,
+                        &stbuf, xdata, &postparent);
+
+clean:
+    if (xdata)
+        dict_unref(xdata);
+
+    free(rsp.xdata.xdata_val);
+
+    return 0;
+}
+
+void
+_quota_enforcer_lookup(void *data)
+{
+    quota_local_t *local = NULL;
+    gfs3_lookup_req req = {
+        {
+            0,
+        },
+    };
+    int ret = 0;
+    int op_errno = ESTALE;
+    quota_priv_t *priv = NULL;
+    call_frame_t *frame = NULL;
+    loc_t *loc = NULL;
+    xlator_t *this = NULL;
+    char *dir_path = NULL;
+
+    frame = data;
+    local = frame->local;
+    this = local->this;
+    loc = &local->validate_loc;
+
+    priv = this->private;
+
+    if (!(loc && loc->inode))
+        goto unwind;
+
+    if (!gf_uuid_is_null(loc->inode->gfid))
+        memcpy(req.gfid, loc->inode->gfid, 16);
+    else
+        memcpy(req.gfid, loc->gfid, 16);
+
+    if (local->validate_xdata) {
+        GF_PROTOCOL_DICT_SERIALIZE(this, local->validate_xdata,
+                                   (&req.xdata.xdata_val), req.xdata.xdata_len,
+                                   op_errno, unwind);
+    }
+
+    if (loc->name)
+        req.bname = (char *)loc->name;
+    else
+        req.bname = "";
+
+    if (loc->path)
+        dir_path = (char *)loc->path;
+    else
+        dir_path = "";
+
+    ret = quota_enforcer_submit_request(
+        &req, frame, priv->quota_enforcer, GF_AGGREGATOR_LOOKUP, NULL, this,
+        quota_enforcer_lookup_cbk, (xdrproc_t)xdr_gfs3_lookup_req);
+
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, Q_MSG_RPC_SUBMIT_FAILED,
+               "Couldn't send the request to "
+               "fetch cluster wide size of directory (path:%s gfid:%s)",
+               dir_path, req.gfid);
+    }
+
+    GF_FREE(req.xdata.xdata_val);
+
+    return;
+
+unwind:
+    local->validate_cbk(frame, NULL, this, -1, op_errno, NULL, NULL, NULL,
+                        NULL);
+
+    GF_FREE(req.xdata.xdata_val);
+
+    return;
+}
+
+int
+quota_enforcer_lookup(call_frame_t *frame, xlator_t *this, dict_t *xdata,
+                      fop_lookup_cbk_t validate_cbk)
+{
+    quota_local_t *local = NULL;
+
+    if (!frame || !this)
+        goto unwind;
+
+    local = frame->local;
+    local->this = this;
+    local->validate_cbk = validate_cbk;
+    local->validate_xdata = dict_ref(xdata);
+
+    _quota_enforcer_lookup(frame);
+
+    return 0;
+
+unwind:
+    validate_cbk(frame, NULL, this, -1, ESTALE, NULL, NULL, NULL, NULL);
+
+    return 0;
+}
+
+int
+quota_enforcer_notify(struct rpc_clnt *rpc, void *mydata,
+                      rpc_clnt_event_t event, void *data)
+{
+    xlator_t *this = NULL;
+    int ret = 0;
+    quota_priv_t *priv = NULL;
+
+    this = mydata;
+    priv = this->private;
+    switch (event) {
+        case RPC_CLNT_CONNECT: {
+            pthread_mutex_lock(&priv->conn_mutex);
+            {
+                priv->conn_status = _gf_true;
+            }
+            pthread_mutex_unlock(&priv->conn_mutex);
+            gf_msg_trace(this->name, 0, "got RPC_CLNT_CONNECT");
+            break;
+        }
+
+        case RPC_CLNT_DISCONNECT: {
+            pthread_mutex_lock(&priv->conn_mutex);
+            {
+                priv->conn_status = _gf_false;
+                pthread_cond_signal(&priv->conn_cond);
+            }
+            pthread_mutex_unlock(&priv->conn_mutex);
+            gf_msg_trace(this->name, 0, "got RPC_CLNT_DISCONNECT");
+            break;
+        }
+
+        default:
+            gf_msg_trace(this->name, 0, "got some other RPC event %d", event);
+            ret = 0;
+            break;
+    }
+
+    return ret;
+}
+
+int
+quota_enforcer_blocking_connect(rpc_clnt_t *rpc)
+{
+    dict_t *options = NULL;
+    int ret = -1;
+
+    options = dict_new();
+    if (options == NULL)
+        goto out;
+
+    ret = dict_set_sizen_str_sizen(options, "non-blocking-io", "no");
+    if (ret)
+        goto out;
+
+    rpc->conn.trans->reconfigure(rpc->conn.trans, options);
+
+    rpc_clnt_start(rpc);
+
+    ret = dict_set_sizen_str_sizen(options, "non-blocking-io", "yes");
+    if (ret)
+        goto out;
+
+    rpc->conn.trans->reconfigure(rpc->conn.trans, options);
+
+    ret = 0;
+out:
+    if (options)
+        dict_unref(options);
+
+    return ret;
+}
+
+// Returns a started rpc_clnt. Creates a new rpc_clnt if quota_priv doesn't have
+// one already
+struct rpc_clnt *
+quota_enforcer_init(xlator_t *this, dict_t *options)
+{
+    struct rpc_clnt *rpc = NULL;
+    quota_priv_t *priv = NULL;
+    int ret = -1;
+
+    priv = this->private;
+
+    LOCK(&priv->lock);
+    {
+        if (priv->rpc_clnt) {
+            ret = 0;
+            rpc = priv->rpc_clnt;
+        }
+    }
+    UNLOCK(&priv->lock);
+
+    if (rpc)
+        goto out;
+
+    priv->quota_enforcer = &quota_enforcer_clnt;
+
+    ret = dict_set_sizen_str_sizen(options, "transport.address-family", "unix");
+    if (ret)
+        goto out;
+
+    ret = dict_set_sizen_str_sizen(options, "transport-type", "socket");
+    if (ret)
+        goto out;
+
+    ret = dict_set_sizen_str_sizen(options, "transport.socket.connect-path",
+                                   "/var/run/gluster/quotad.socket");
+    if (ret)
+        goto out;
+
+    rpc = rpc_clnt_new(options, this, this->name, 16);
+    if (!rpc) {
+        ret = -1;
+        goto out;
+    }
+
+    ret = rpc_clnt_register_notify(rpc, quota_enforcer_notify, this);
+    if (ret) {
+        gf_msg("quota", GF_LOG_ERROR, 0, Q_MSG_RPCCLNT_REGISTER_NOTIFY_FAILED,
+               "failed to register notify");
+        goto out;
+    }
+
+    ret = quota_enforcer_blocking_connect(rpc);
+    if (ret)
+        goto out;
+
+    ret = 0;
+out:
+    if (ret) {
+        if (rpc)
+            rpc_clnt_unref(rpc);
+        rpc = NULL;
+    }
+
+    return rpc;
+}
+
+struct rpc_clnt_procedure quota_enforcer_actors[GF_AGGREGATOR_MAXVALUE] = {
+    [GF_AGGREGATOR_NULL] = {"NULL", NULL},
+    [GF_AGGREGATOR_LOOKUP] = {"LOOKUP", NULL},
+};
+
+struct rpc_clnt_program quota_enforcer_clnt = {
+    .progname = "Quota enforcer",
+    .prognum = GLUSTER_AGGREGATOR_PROGRAM,
+    .progver = GLUSTER_AGGREGATOR_VERSION,
+    .numproc = GF_AGGREGATOR_MAXVALUE,
+    .proctable = quota_enforcer_actors,
+};
diff --git a/xlators/features/quota/src/quota-mem-types.h b/xlators/features/quota/src/quota-mem-types.h
index 315b5b8298f..782a7de96bb 100644
--- a/xlators/features/quota/src/quota-mem-types.h
+++ b/xlators/features/quota/src/quota-mem-types.h
@@ -1,31 +1,30 @@
 /*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
    This file is part of GlusterFS.
 
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
 */
-
 #ifndef __QUOTA_MEM_TYPES_H__
 #define __QUOTA_MEM_TYPES_H__
 
-#include "mem-types.h"
+#include <glusterfs/mem-types.h>
 
 enum gf_quota_mem_types_ {
-        gf_quota_mt_quota_local = gf_common_mt_end + 1,
-        gf_quota_mt_quota_priv,
-        gf_quota_mt_end
+    /* Those are used by QUOTA_ALLOC_OR_GOTO macro */
+    gf_quota_mt_quota_priv_t = gf_common_mt_end + 1,
+    gf_quota_mt_quota_inode_ctx_t,
+    gf_quota_mt_loc_t,
+    gf_quota_mt_char,
+    gf_quota_mt_int64_t,
+    gf_quota_mt_int32_t,
+    gf_quota_mt_limits_t,
+    gf_quota_mt_quota_dentry_t,
+    gf_quota_mt_quota_limits_level_t,
+    gf_quota_mt_qd_vols_conf_t,
+    gf_quota_mt_aggregator_state_t,
+    gf_quota_mt_end
 };
 #endif
-
diff --git a/xlators/features/quota/src/quota-messages.h b/xlators/features/quota/src/quota-messages.h
new file mode 100644
index 00000000000..d434ed75e76
--- /dev/null
+++ b/xlators/features/quota/src/quota-messages.h
@@ -0,0 +1,39 @@
+/*
+  Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _QUOTA_MESSAGES_H_
+#define _QUOTA_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(QUOTA, Q_MSG_ENFORCEMENT_FAILED, Q_MSG_ENOMEM, Q_MSG_PARENT_NULL,
+           Q_MSG_CROSSED_SOFT_LIMIT, Q_MSG_QUOTA_ENFORCER_RPC_INIT_FAILED,
+           Q_MSG_REMOTE_OPERATION_FAILED, Q_MSG_FAILED_TO_SEND_FOP,
+           Q_MSG_INVALID_VOLFILE, Q_MSG_INODE_PARENT_NOT_FOUND,
+           Q_MSG_XDR_DECODE_ERROR, Q_MSG_DICT_UNSERIALIZE_FAIL,
+           Q_MSG_DICT_SERIALIZE_FAIL, Q_MSG_RPCSVC_INIT_FAILED,
+           Q_MSG_RPCSVC_LISTENER_CREATION_FAILED, Q_MSG_RPCSVC_REGISTER_FAILED,
+           Q_MSG_XDR_DECODING_FAILED, Q_MSG_RPCCLNT_REGISTER_NOTIFY_FAILED,
+           Q_MSG_ANCESTRY_BUILD_FAILED, Q_MSG_SIZE_KEY_MISSING,
+           Q_MSG_INODE_CTX_GET_FAILED, Q_MSG_INODE_CTX_SET_FAILED,
+           Q_MSG_LOOKUP_FAILED, Q_MSG_RPC_SUBMIT_FAILED,
+           Q_MSG_ENFORCEMENT_SKIPPED, Q_MSG_INTERNAL_FOP_KEY_MISSING);
+
+#endif /* !_QUOTA_MESSAGES_H_ */
diff --git a/xlators/features/quota/src/quota.c b/xlators/features/quota/src/quota.c
index b889a0c2a12..18df9ae6d19 100644
--- a/xlators/features/quota/src/quota.c
+++ b/xlators/features/quota/src/quota.c
@@ -1,1207 +1,5336 @@
 /*
-  Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
 */
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include <sys/time.h>
-
-#include "xlator.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "quota-mem-types.h"
-
-#ifndef MAX_IOVEC
-#define MAX_IOVEC 16
-#endif
-
-struct quota_local {
-	struct iatt    stbuf;
-	inode_t       *inode;
-	char          *path;
-	fd_t          *fd;
-	off_t          offset;
-	int32_t        count;
-	struct iovec   vector[MAX_IOVEC];
-	struct iobref *iobref;
-	loc_t          loc;
-        int            flags;
-};
+#include "quota.h"
+#include <glusterfs/statedump.h>
+#include "quota-messages.h"
+#include <glusterfs/events.h>
+
+struct volume_options options[];
 
+static int32_t
+__quota_init_inode_ctx(inode_t *inode, xlator_t *this,
+                       quota_inode_ctx_t **context)
+{
+    int32_t ret = -1;
+    quota_inode_ctx_t *ctx = NULL;
 
-struct quota_priv {
-	char       only_first_time;          /* Used to make sure a call is done only one time */
-	gf_lock_t  lock;                     /* Used while updating variables */
+    if (inode == NULL) {
+        goto out;
+    }
 
-	uint64_t   disk_usage_limit;         /* Used for Disk usage quota */
-	uint64_t   current_disk_usage;       /* Keep the current usage value */
+    QUOTA_ALLOC_OR_GOTO(ctx, quota_inode_ctx_t, out);
 
-	uint32_t   min_free_disk_limit;        /* user specified limit, in %*/
-	uint32_t   current_free_disk;          /* current free disk space available, in % */
-	uint32_t   refresh_interval;           /* interval in seconds */
-	uint32_t   min_disk_last_updated_time; /* used for interval calculation */	
+    LOCK_INIT(&ctx->lock);
 
-	loc_t      root_loc;		     /* Store '/' loc_t to make xattr calls */
-};
+    if (context != NULL) {
+        *context = ctx;
+    }
+
+    INIT_LIST_HEAD(&ctx->parents);
+
+    ret = __inode_ctx_put(inode, this, (uint64_t)(long)ctx);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, Q_MSG_INODE_CTX_SET_FAILED,
+               "cannot set quota context "
+               "in inode (gfid:%s)",
+               uuid_utoa(inode->gfid));
+        GF_FREE(ctx);
+    }
+out:
+    return ret;
+}
+
+static int32_t
+quota_inode_ctx_get(inode_t *inode, xlator_t *this, quota_inode_ctx_t **ctx,
+                    char create_if_absent)
+{
+    int32_t ret = 0;
+    uint64_t ctx_int;
+
+    LOCK(&inode->lock);
+    {
+        ret = __inode_ctx_get(inode, this, &ctx_int);
+
+        if ((ret == 0) && (ctx != NULL)) {
+            *ctx = (quota_inode_ctx_t *)(unsigned long)ctx_int;
+        } else if (create_if_absent) {
+            ret = __quota_init_inode_ctx(inode, this, ctx);
+        }
+    }
+    UNLOCK(&inode->lock);
+
+    return ret;
+}
+
+int
+quota_loc_fill(loc_t *loc, inode_t *inode, inode_t *parent, char *path)
+{
+    int ret = -1;
+
+    if (!loc || (inode == NULL))
+        return ret;
+
+    if (inode) {
+        loc->inode = inode_ref(inode);
+        gf_uuid_copy(loc->gfid, inode->gfid);
+    }
 
+    if (parent) {
+        loc->parent = inode_ref(parent);
+    }
+
+    if (path != NULL) {
+        loc->path = gf_strdup(path);
+
+        loc->name = strrchr(loc->path, '/');
+        if (loc->name) {
+            loc->name++;
+        }
+    }
+
+    ret = 0;
+
+    return ret;
+}
 
 int
-quota_statvfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		   int32_t op_ret, int32_t op_errno, struct statvfs *stbuf)
+quota_inode_loc_fill(inode_t *inode, loc_t *loc)
 {
-	struct quota_priv *priv = this->private;
-	
-	if (op_ret >= 0) {
-		priv->current_free_disk =
-			(stbuf->f_bavail * 100) / stbuf->f_blocks;
-	}
+    char *resolvedpath = NULL;
+    inode_t *parent = NULL;
+    int ret = -1;
+    xlator_t *this = NULL;
 
-	STACK_DESTROY (frame->root);
-	return 0;
+    if ((!inode) || (!loc)) {
+        return ret;
+    }
+
+    this = THIS;
+
+    if ((inode) && __is_root_gfid(inode->gfid)) {
+        loc->parent = NULL;
+        goto ignore_parent;
+    }
+
+    parent = inode_parent(inode, 0, NULL);
+    if (!parent) {
+        gf_msg_debug(this->name, 0,
+                     "cannot find parent for "
+                     "inode (gfid:%s)",
+                     uuid_utoa(inode->gfid));
+    }
+
+ignore_parent:
+    ret = inode_path(inode, NULL, &resolvedpath);
+    if (ret < 0) {
+        gf_msg_debug(this->name, 0,
+                     "cannot construct path for "
+                     "inode (gfid:%s)",
+                     uuid_utoa(inode->gfid));
+    }
+
+    ret = quota_loc_fill(loc, inode, parent, resolvedpath);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "cannot fill loc");
+        goto err;
+    }
+
+err:
+    if (parent) {
+        inode_unref(parent);
+    }
+
+    GF_FREE(resolvedpath);
+
+    return ret;
 }
 
+int32_t
+quota_local_cleanup(quota_local_t *local)
+{
+    if (local == NULL) {
+        goto out;
+    }
 
-void
-gf_quota_usage_subtract (xlator_t *this, size_t size)
+    loc_wipe(&local->loc);
+    loc_wipe(&local->newloc);
+    loc_wipe(&local->oldloc);
+    loc_wipe(&local->validate_loc);
+
+    inode_unref(local->inode);
+
+    if (local->xdata)
+        dict_unref(local->xdata);
+
+    if (local->validate_xdata)
+        dict_unref(local->validate_xdata);
+
+    if (local->stub)
+        call_stub_destroy(local->stub);
+
+    LOCK_DESTROY(&local->lock);
+
+    mem_put(local);
+out:
+    return 0;
+}
+
+static quota_local_t *
+quota_local_new()
 {
-	struct quota_priv *priv = NULL;
+    quota_local_t *local = NULL;
+    local = mem_get0(THIS->local_pool);
+    if (local == NULL)
+        goto out;
 
-	priv = this->private;
+    LOCK_INIT(&local->lock);
+    local->space_available = -1;
 
-	LOCK (&priv->lock);
-	{
-		if (priv->current_disk_usage < size)
-			priv->current_disk_usage = 0;
-		else
-			priv->current_disk_usage -= size;
-	}
-	UNLOCK (&priv->lock);
+out:
+    return local;
 }
 
+quota_dentry_t *
+__quota_dentry_new(quota_inode_ctx_t *ctx, char *name, uuid_t par)
+{
+    quota_dentry_t *dentry = NULL;
+    GF_UNUSED int32_t ret = 0;
+
+    QUOTA_ALLOC_OR_GOTO(dentry, quota_dentry_t, err);
+
+    INIT_LIST_HEAD(&dentry->next);
+
+    dentry->name = gf_strdup(name);
+    if (dentry->name == NULL) {
+        GF_FREE(dentry);
+        dentry = NULL;
+        goto err;
+    }
+
+    gf_uuid_copy(dentry->par, par);
+
+    if (ctx != NULL)
+        list_add_tail(&dentry->next, &ctx->parents);
+
+err:
+    return dentry;
+}
 
 void
-gf_quota_usage_add (xlator_t *this, size_t size)
+__quota_dentry_free(quota_dentry_t *dentry)
 {
-	struct quota_priv *priv = this->private;
+    if (dentry == NULL) {
+        goto out;
+    }
+
+    list_del_init(&dentry->next);
 
-	LOCK (&priv->lock);
-	{
-		priv->current_disk_usage += size;
-	}
-	UNLOCK (&priv->lock);
+    GF_FREE(dentry->name);
+    GF_FREE(dentry);
+out:
+    return;
 }
 
+void
+__quota_dentry_del(quota_inode_ctx_t *ctx, const char *name, uuid_t par)
+{
+    quota_dentry_t *dentry = NULL;
+    quota_dentry_t *tmp = NULL;
+
+    list_for_each_entry_safe(dentry, tmp, &ctx->parents, next)
+    {
+        if ((strcmp(dentry->name, name) == 0) &&
+            (gf_uuid_compare(dentry->par, par) == 0)) {
+            __quota_dentry_free(dentry);
+            break;
+        }
+    }
+}
 
-void 
-gf_quota_update_current_free_disk (xlator_t *this)
+void
+quota_dentry_del(quota_inode_ctx_t *ctx, const char *name, uuid_t par)
 {
-	call_frame_t *frame = NULL;
-	call_pool_t   *pool = NULL;
+    LOCK(&ctx->lock);
+    {
+        __quota_dentry_del(ctx, name, par);
+    }
+    UNLOCK(&ctx->lock);
+}
 
-	struct quota_priv *priv = NULL;
+static inode_t *
+__quota_inode_parent(inode_t *inode, uuid_t pargfid, const char *name)
+{
+    inode_t *parent = NULL;
 
-	pool  = this->ctx->pool;
-	frame = create_frame (this, pool);
-  
-	priv = this->private;
+    parent = inode_parent(inode, pargfid, name);
+    inode_unref(inode);
+    return parent;
+}
+
+static inode_t *
+quota_inode_parent(inode_t *inode, uuid_t pargfid, const char *name)
+{
+    inode_t *parent = NULL;
 
-	STACK_WIND (frame, quota_statvfs_cbk,
-		    this->children->xlator,
-		    this->children->xlator->fops->statfs, &(priv->root_loc));
+    parent = __quota_inode_parent(inode, pargfid, name);
+    if (!parent)
+        gf_msg_callingfn(THIS->name, GF_LOG_ERROR, 0, Q_MSG_PARENT_NULL,
+                         "Failed to find "
+                         "ancestor for inode (%s)",
+                         uuid_utoa(inode->gfid));
 
-	return ;
+    return parent;
 }
 
+int32_t
+quota_inode_depth(inode_t *inode)
+{
+    int depth = 0;
+    inode_t *cur_inode = NULL;
 
-int
-gf_quota_check_free_disk (xlator_t *this) 
+    cur_inode = inode_ref(inode);
+    while (cur_inode && !__is_root_gfid(cur_inode->gfid)) {
+        depth++;
+        cur_inode = quota_inode_parent(cur_inode, 0, NULL);
+        if (!cur_inode)
+            depth = -1;
+    }
+
+    if (cur_inode)
+        inode_unref(cur_inode);
+
+    return depth;
+}
+
+int32_t
+quota_find_common_ancestor(inode_t *inode1, inode_t *inode2,
+                           uuid_t *common_ancestor)
 {
-        struct quota_priv * priv = NULL;
-	struct timeval tv = {0, 0};
+    int32_t depth1 = 0;
+    int32_t depth2 = 0;
+    int32_t ret = -1;
+    inode_t *cur_inode1 = NULL;
+    inode_t *cur_inode2 = NULL;
+
+    depth1 = quota_inode_depth(inode1);
+    if (depth1 < 0)
+        goto out;
+
+    depth2 = quota_inode_depth(inode2);
+    if (depth2 < 0)
+        goto out;
+
+    cur_inode1 = inode_ref(inode1);
+    cur_inode2 = inode_ref(inode2);
+
+    while (cur_inode1 && depth1 > depth2) {
+        cur_inode1 = quota_inode_parent(cur_inode1, 0, NULL);
+        depth1--;
+    }
+
+    while (cur_inode2 && depth2 > depth1) {
+        cur_inode2 = quota_inode_parent(cur_inode2, 0, NULL);
+        depth2--;
+    }
+
+    while (depth1 && cur_inode1 && cur_inode2 && cur_inode1 != cur_inode2) {
+        cur_inode1 = quota_inode_parent(cur_inode1, 0, NULL);
+        cur_inode2 = quota_inode_parent(cur_inode2, 0, NULL);
+        depth1--;
+    }
+
+    if (cur_inode1 && cur_inode2) {
+        gf_uuid_copy(*common_ancestor, cur_inode1->gfid);
+        ret = 0;
+    }
+out:
+    if (cur_inode1)
+        inode_unref(cur_inode1);
+
+    if (cur_inode2)
+        inode_unref(cur_inode2);
+
+    return ret;
+}
 
-	priv = this->private;
-	if (priv->min_free_disk_limit) {
-		gettimeofday (&tv, NULL);
-		if (tv.tv_sec > (priv->refresh_interval + 
-				 priv->min_disk_last_updated_time)) {
-			priv->min_disk_last_updated_time = tv.tv_sec;
-			gf_quota_update_current_free_disk (this);
-		}
-		if (priv->current_free_disk <= priv->min_free_disk_limit)
-			return -1;
-	}
+void
+check_ancestory_continue(struct list_head *parents, inode_t *inode,
+                         int32_t op_ret, int32_t op_errno, void *data)
+{
+    call_frame_t *frame = NULL;
+    quota_local_t *local = NULL;
+    uint32_t link_count = 0;
+
+    frame = data;
+    local = frame->local;
+
+    if (parents && list_empty(parents)) {
+        gf_msg(THIS->name, GF_LOG_WARNING, EIO, Q_MSG_ANCESTRY_BUILD_FAILED,
+               "Couldn't build ancestry for inode (gfid:%s). "
+               "Without knowing ancestors till root, quota "
+               "cannot be enforced. "
+               "Hence, failing fop with EIO",
+               uuid_utoa(inode->gfid));
+        op_errno = EIO;
+        op_ret = -1;
+    }
+
+    LOCK(&local->lock);
+    {
+        link_count = --local->link_count;
+        if (op_ret < 0) {
+            local->op_ret = op_ret;
+            local->op_errno = op_errno;
+        }
+    }
+    UNLOCK(&local->lock);
 
-	return 0;
+    if (link_count == 0)
+        local->fop_continue_cbk(frame);
 }
 
+void
+check_ancestory(call_frame_t *frame, inode_t *inode)
+{
+    inode_t *cur_inode = NULL;
+    inode_t *parent = NULL;
+
+    cur_inode = inode_ref(inode);
+    while (cur_inode && !__is_root_gfid(cur_inode->gfid)) {
+        parent = inode_parent(cur_inode, 0, NULL);
+        if (!parent) {
+            quota_build_ancestry(cur_inode, check_ancestory_continue, frame);
+            inode_unref(cur_inode);
+            return;
+        }
+        inode_unref(cur_inode);
+        cur_inode = parent;
+    }
+
+    if (cur_inode) {
+        inode_unref(cur_inode);
+        check_ancestory_continue(NULL, NULL, 0, 0, frame);
+    } else {
+        check_ancestory_continue(NULL, NULL, -1, ESTALE, frame);
+    }
+}
 
-int
-quota_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		    int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                    struct iatt *postbuf)
+void
+check_ancestory_2_cbk(struct list_head *parents, inode_t *inode, int32_t op_ret,
+                      int32_t op_errno, void *data)
 {
-	struct quota_priv *priv = this->private;
-	struct quota_local *local = NULL;
+    inode_t *this_inode = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+
+    this_inode = data;
+
+    if (op_ret < 0)
+        goto out;
 
-	local = frame->local;
+    if (parents == NULL || list_empty(parents)) {
+        gf_msg(THIS->name, GF_LOG_WARNING, 0, Q_MSG_ENFORCEMENT_FAILED,
+               "Couldn't build ancestry for inode (gfid:%s). "
+               "Without knowing ancestors till root, quota "
+               "cannot be enforced.",
+               uuid_utoa(this_inode->gfid));
+        goto out;
+    }
 
-	if ((op_ret >= 0) && priv->disk_usage_limit) {
-		gf_quota_usage_subtract (this, (local->stbuf.ia_blocks -
-						postbuf->ia_blocks) * 512);
-		loc_wipe (&local->loc);
-	}
+    quota_inode_ctx_get(this_inode, THIS, &ctx, 0);
+    if (ctx)
+        ctx->ancestry_built = _gf_true;
 
-	STACK_UNWIND_STRICT (truncate, frame, op_ret, op_errno,
-                             prebuf, postbuf);
-	return 0;
+out:
+    inode_unref(this_inode);
 }
 
+void
+check_ancestory_2(xlator_t *this, quota_local_t *local, inode_t *inode)
+{
+    inode_t *cur_inode = NULL;
+    inode_t *parent = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    char *name = NULL;
+    uuid_t pgfid = {0};
+
+    name = (char *)local->loc.name;
+    if (local->loc.parent) {
+        gf_uuid_copy(pgfid, local->loc.parent->gfid);
+    }
+
+    cur_inode = inode_ref(inode);
+    while (cur_inode && !__is_root_gfid(cur_inode->gfid)) {
+        quota_inode_ctx_get(cur_inode, this, &ctx, 0);
+        /* build ancestry is required only on the first lookup,
+         * so stop crawling when the inode_ctx is set for an inode
+         */
+        if (ctx && ctx->ancestry_built)
+            goto setctx;
+
+        parent = inode_parent(cur_inode, pgfid, name);
+        if (!parent) {
+            quota_build_ancestry(cur_inode, check_ancestory_2_cbk,
+                                 inode_ref(inode));
+            goto out;
+        }
 
-int
-quota_truncate_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			 int32_t op_ret, int32_t op_errno, struct iatt *buf)
+        if (name != NULL) {
+            name = NULL;
+            gf_uuid_clear(pgfid);
+        }
+
+        inode_unref(cur_inode);
+        cur_inode = parent;
+    }
+
+setctx:
+    if (cur_inode && cur_inode != inode) {
+        quota_inode_ctx_get(inode, this, &ctx, 0);
+        if (ctx)
+            ctx->ancestry_built = _gf_true;
+    }
+out:
+    if (cur_inode)
+        inode_unref(cur_inode);
+}
+
+static void
+quota_link_count_decrement(call_frame_t *frame)
 {
-	struct quota_local *local = NULL;
+    call_frame_t *tmpframe = NULL;
+    quota_local_t *local = NULL;
+    call_stub_t *stub = NULL;
+    int link_count = -1;
+
+    local = frame->local;
+    if (local && local->par_frame) {
+        local = local->par_frame->local;
+        tmpframe = frame;
+    }
+
+    if (local == NULL)
+        goto out;
+
+    LOCK(&local->lock);
+    {
+        link_count = --local->link_count;
+        if (link_count == 0) {
+            stub = local->stub;
+            local->stub = NULL;
+        }
+    }
+    UNLOCK(&local->lock);
+
+    if (stub != NULL) {
+        call_resume(stub);
+    }
 
-	local = frame->local;
+out:
+    if (tmpframe) {
+        local = tmpframe->local;
+        tmpframe->local = NULL;
 
-	if (op_ret >= 0) {
-		local->stbuf = *buf;
-	}
+        STACK_DESTROY(frame->root);
+        if (local)
+            quota_local_cleanup(local);
+    }
 
-	STACK_WIND (frame, quota_truncate_cbk,
-		    FIRST_CHILD (this), FIRST_CHILD (this)->fops->truncate,
-		    &local->loc, local->offset);
-	return 0;
+    return;
 }
 
+static void
+quota_handle_validate_error(call_frame_t *frame, int32_t op_ret,
+                            int32_t op_errno)
+{
+    quota_local_t *local;
 
-int
-quota_truncate (call_frame_t *frame, xlator_t *this,
-		loc_t *loc, off_t offset)
+    local = frame->local;
+    if (local && local->par_frame)
+        local = local->par_frame->local;
+
+    if (local == NULL)
+        goto out;
+
+    if (op_ret < 0) {
+        LOCK(&local->lock);
+        {
+            local->op_ret = op_ret;
+            local->op_errno = op_errno;
+        }
+        UNLOCK(&local->lock);
+    }
+    /* we abort checking limits on this path to root */
+    quota_link_count_decrement(frame);
+out:
+    return;
+}
+
+int32_t
+quota_validate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, inode_t *inode,
+                   struct iatt *buf, dict_t *xdata, struct iatt *postparent)
 {
-	struct quota_local *local = NULL;
-	struct quota_priv  *priv = NULL;
+    quota_local_t *local = NULL;
+    int32_t ret = 0;
+    quota_inode_ctx_t *ctx = NULL;
+    uint64_t value = 0;
+    quota_meta_t size = {
+        0,
+    };
+
+    local = frame->local;
+
+    if (op_ret < 0) {
+        goto unwind;
+    }
+
+    GF_ASSERT(local);
+    GF_ASSERT(frame);
+    GF_VALIDATE_OR_GOTO_WITH_ERROR("quota", this, unwind, op_errno, EINVAL);
+    GF_VALIDATE_OR_GOTO_WITH_ERROR(this->name, xdata, unwind, op_errno, EINVAL);
+
+    ret = inode_ctx_get(local->validate_loc.inode, this, &value);
+
+    ctx = (quota_inode_ctx_t *)(unsigned long)value;
+    if ((ret == -1) || (ctx == NULL)) {
+        gf_msg(this->name, GF_LOG_WARNING, EINVAL, Q_MSG_INODE_CTX_GET_FAILED,
+               "quota context is"
+               " not present in  inode (gfid:%s)",
+               uuid_utoa(local->validate_loc.inode->gfid));
+        op_errno = EINVAL;
+        goto unwind;
+    }
+
+    ret = quota_dict_get_meta(xdata, QUOTA_SIZE_KEY, SLEN(QUOTA_SIZE_KEY),
+                              &size);
+    if (ret == -1) {
+        gf_msg(this->name, GF_LOG_WARNING, EINVAL, Q_MSG_SIZE_KEY_MISSING,
+               "quota size key not present "
+               "in dict");
+        op_errno = EINVAL;
+    }
+
+    local->just_validated = 1; /* so that we don't go into infinite
+                                * loop of validation and checking
+                                * limit when timeout is zero.
+                                */
+    LOCK(&ctx->lock);
+    {
+        ctx->size = size.size;
+        ctx->validate_time = gf_time();
+        ctx->file_count = size.file_count;
+        ctx->dir_count = size.dir_count;
+    }
+    UNLOCK(&ctx->lock);
+
+    quota_check_limit(frame, local->validate_loc.inode, this);
+    return 0;
 
-	priv = this->private;
+unwind:
+    quota_handle_validate_error(frame, op_ret, op_errno);
+    return 0;
+}
 
-	if (priv->disk_usage_limit) {
-		local = GF_CALLOC (1, sizeof (struct quota_local),
-                                   gf_quota_mt_quota_local);
-		frame->local  = local;
+static inline gf_boolean_t
+quota_timeout(time_t t, uint32_t timeout)
+{
+    return (gf_time() - t) >= timeout;
+}
 
-		loc_copy (&local->loc, loc);
-		local->offset = offset;
+/* Return: 1 if new entry added
+ *         0 no entry added
+ *        -1 on errors
+ */
+static int32_t
+quota_add_parent(struct list_head *list, char *name, uuid_t pgfid)
+{
+    quota_dentry_t *entry = NULL;
+    gf_boolean_t found = _gf_false;
+    int ret = 0;
+
+    if (!list_empty(list)) {
+        list_for_each_entry(entry, list, next)
+        {
+            if (gf_uuid_compare(pgfid, entry->par) == 0) {
+                found = _gf_true;
+                goto out;
+            }
+        }
+    }
+
+    entry = __quota_dentry_new(NULL, name, pgfid);
+    if (entry)
+        list_add_tail(&entry->next, list);
+    else
+        ret = -1;
+
+out:
+    if (found)
+        return 0;
+    else if (ret == 0)
+        return 1;
+    else
+        return -1;
+}
 
-		STACK_WIND (frame, quota_truncate_stat_cbk,
-			    FIRST_CHILD(this),
-			    FIRST_CHILD(this)->fops->stat, loc);
-		return 0;
-	}
+/* This function iterates the parent list in inode
+ * context and add unique parent to the list
+ * Returns number of dentry added to the list, or -1 on errors
+ */
+static int32_t
+quota_add_parents_from_ctx(quota_inode_ctx_t *ctx, struct list_head *list)
+{
+    int ret = 0;
+    quota_dentry_t *dentry = NULL;
+    int32_t count = 0;
+
+    if (ctx == NULL || list == NULL)
+        goto out;
+
+    LOCK(&ctx->lock);
+    {
+        list_for_each_entry(dentry, &ctx->parents, next)
+        {
+            ret = quota_add_parent(list, dentry->name, dentry->par);
+            if (ret == 1)
+                count++;
+            else if (ret == -1)
+                break;
+        }
+    }
+    UNLOCK(&ctx->lock);
 
-	STACK_WIND (frame, quota_truncate_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->truncate,
-		    loc, offset);
-	return 0;
+out:
+    return (ret == -1) ? -1 : count;
 }
 
+int32_t
+quota_build_ancestry_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, gf_dirent_t *entries,
+                         dict_t *xdata)
+{
+    inode_t *parent = NULL;
+    inode_t *tmp_parent = NULL;
+    inode_t *linked_inode = NULL;
+    inode_t *tmp_inode = NULL;
+    gf_dirent_t *entry = NULL;
+    loc_t loc = {
+        0,
+    };
+    quota_dentry_t *dentry = NULL;
+    quota_dentry_t *tmp = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    struct list_head parents;
+    quota_local_t *local = NULL;
+    int ret;
+
+    INIT_LIST_HEAD(&parents);
+
+    local = frame->local;
+    frame->local = NULL;
+
+    if (op_ret < 0)
+        goto err;
+
+    if ((op_ret > 0) && (entries != NULL)) {
+        list_for_each_entry(entry, &entries->list, list)
+        {
+            if (__is_root_gfid(entry->inode->gfid)) {
+                /* The list contains a sub-list for each
+                 * possible path to the target inode. Each
+                 * sub-list starts with the root entry of the
+                 * tree and is followed by the child entries
+                 * for a particular path to the target entry.
+                 * The root entry is an implied sub-list
+                 * delimiter, as it denotes we have started
+                 * processing a new path. Reset the parent
+                 * pointer and continue
+                 */
+
+                tmp_parent = NULL;
+            } else {
+                /* For a non-root entry, link this inode */
+                linked_inode = inode_link(entry->inode, tmp_parent,
+                                          entry->d_name, &entry->d_stat);
+                if (linked_inode) {
+                    tmp_inode = entry->inode;
+                    entry->inode = linked_inode;
+                    inode_unref(tmp_inode);
+                } else {
+                    gf_msg(this->name, GF_LOG_WARNING, EINVAL,
+                           Q_MSG_PARENT_NULL, "inode link failed");
+                    op_errno = EINVAL;
+                    goto err;
+                }
+            }
+
+            gf_uuid_copy(loc.gfid, entry->d_stat.ia_gfid);
+
+            loc.inode = inode_ref(entry->inode);
+            loc.parent = inode_ref(tmp_parent);
+            loc.name = entry->d_name;
+
+            quota_fill_inodectx(this, entry->inode, entry->dict, &loc,
+                                &entry->d_stat, &op_errno);
+
+            /* For non-directory, posix_get_ancestry_non_directory
+             * returns all hard-links that are represented by nodes
+             * adjacent to each other in the dentry-list.
+             * (Unlike the directory case where adjacent nodes
+             * either have a parent/child relationship or belong to
+             * different paths).
+             */
+            if (entry->inode->ia_type == IA_IFDIR)
+                tmp_parent = entry->inode;
+
+            loc_wipe(&loc);
+        }
+    }
+
+    parent = inode_parent(local->loc.inode, 0, NULL);
+    if (parent == NULL) {
+        gf_msg(this->name, GF_LOG_WARNING, EINVAL, Q_MSG_PARENT_NULL,
+               "parent is NULL");
+        op_errno = EINVAL;
+        goto err;
+    }
+
+    quota_inode_ctx_get(local->loc.inode, this, &ctx, 0);
+
+    ret = quota_add_parents_from_ctx(ctx, &parents);
+    if (ret == -1) {
+        op_errno = errno;
+        goto err;
+    }
+
+    if (list_empty(&parents)) {
+        /* we built ancestry for a directory */
+        list_for_each_entry(entry, &entries->list, list)
+        {
+            if (entry->inode == local->loc.inode)
+                break;
+        }
+
+        /* Getting assertion here, need to investigate
+           comment for now
+           GF_ASSERT (&entry->list != &entries->list);
+        */
+
+        ret = quota_add_parent(&parents, entry->d_name, parent->gfid);
+        if (ret == -1) {
+            op_errno = errno;
+            goto err;
+        }
+    }
+
+    local->ancestry_cbk(&parents, local->loc.inode, 0, 0, local->ancestry_data);
+    goto cleanup;
+
+err:
+    local->ancestry_cbk(NULL, NULL, -1, op_errno, local->ancestry_data);
+
+cleanup:
+    STACK_DESTROY(frame->root);
+    quota_local_cleanup(local);
+
+    if (parent != NULL) {
+        inode_unref(parent);
+        parent = NULL;
+    }
+
+    if (!list_empty(&parents)) {
+        list_for_each_entry_safe(dentry, tmp, &parents, next)
+        {
+            __quota_dentry_free(dentry);
+        }
+    }
+
+    return 0;
+}
 
 int
-quota_ftruncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		     int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                     struct iatt *postbuf)
+quota_build_ancestry(inode_t *inode, quota_ancestry_built_t ancestry_cbk,
+                     void *data)
 {
-	struct quota_priv  *priv = NULL;
-	struct quota_local *local = NULL;
+    fd_t *fd = NULL;
+    quota_local_t *local = NULL;
+    call_frame_t *new_frame = NULL;
+    int op_errno = ENOMEM;
+    int op_ret = -1;
+    xlator_t *this = NULL;
+    dict_t *xdata_req = NULL;
+
+    this = THIS;
+
+    xdata_req = dict_new();
+    if (xdata_req == NULL)
+        goto err;
+
+    fd = fd_anonymous(inode);
+    if (fd == NULL)
+        goto err;
+
+    new_frame = create_frame(this, this->ctx->pool);
+    if (new_frame == NULL)
+        goto err;
+
+    local = quota_local_new();
+    if (local == NULL)
+        goto err;
+
+    new_frame->root->uid = new_frame->root->gid = 0;
+    new_frame->local = local;
+    local->ancestry_cbk = ancestry_cbk;
+    local->ancestry_data = data;
+    local->loc.inode = inode_ref(inode);
+
+    op_ret = dict_set_int8(xdata_req, QUOTA_LIMIT_KEY, 1);
+    if (op_ret < 0) {
+        op_errno = -op_ret;
+        goto err;
+    }
+
+    op_ret = dict_set_int8(xdata_req, QUOTA_LIMIT_OBJECTS_KEY, 1);
+    if (op_ret < 0) {
+        op_errno = -op_ret;
+        goto err;
+    }
+
+    op_ret = dict_set_int8(xdata_req, GET_ANCESTRY_DENTRY_KEY, 1);
+    if (op_ret < 0) {
+        op_errno = -op_ret;
+        goto err;
+    }
+
+    /* This would ask posix layer to construct dentry chain till root
+     * We don't need to do a opendir, we can use the anonymous fd
+     * here for  the readidrp.
+     * avoiding opendir also reduces the window size where another FOP
+     * can be executed before completion of build ancestry
+     */
+    STACK_WIND(new_frame, quota_build_ancestry_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdirp, fd, 0, 0, xdata_req);
+
+    op_ret = 0;
+
+err:
+    if (fd)
+        fd_unref(fd);
+
+    if (xdata_req)
+        dict_unref(xdata_req);
+
+    if (op_ret < 0) {
+        ancestry_cbk(NULL, NULL, -1, op_errno, data);
+
+        if (new_frame) {
+            local = new_frame->local;
+            new_frame->local = NULL;
+            STACK_DESTROY(new_frame->root);
+        }
 
-	local = frame->local;
-	priv = this->private;
+        if (local)
+            quota_local_cleanup(local);
+    }
 
-	if ((op_ret >= 0) && priv->disk_usage_limit) {
-		gf_quota_usage_subtract (this, (local->stbuf.ia_blocks -
-						postbuf->ia_blocks) * 512);
-		fd_unref (local->fd);
-	}
+    return 0;
+}
 
-	STACK_UNWIND_STRICT (ftruncate, frame, op_ret, op_errno,
-                             prebuf, postbuf);
-	return 0;
+int
+quota_validate(call_frame_t *frame, inode_t *inode, xlator_t *this,
+               fop_lookup_cbk_t cbk_fn)
+{
+    quota_local_t *local = NULL;
+    int ret = 0;
+    dict_t *xdata = NULL;
+    quota_priv_t *priv = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    LOCK(&local->lock);
+    {
+        loc_wipe(&local->validate_loc);
+
+        ret = quota_inode_loc_fill(inode, &local->validate_loc);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENFORCEMENT_FAILED,
+                   "cannot fill loc for inode (gfid:%s), hence "
+                   "aborting quota-checks and continuing with fop",
+                   uuid_utoa(inode->gfid));
+        }
+    }
+    UNLOCK(&local->lock);
+
+    if (ret < 0) {
+        ret = -ENOMEM;
+        goto err;
+    }
+
+    xdata = dict_new();
+    if (xdata == NULL) {
+        ret = -ENOMEM;
+        goto err;
+    }
+
+    ret = dict_set_int8(xdata, QUOTA_SIZE_KEY, 1);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "dict set failed");
+        ret = -ENOMEM;
+        goto err;
+    }
+
+    ret = dict_set_str(xdata, "volume-uuid", priv->volume_uuid);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "dict set failed");
+        ret = -ENOMEM;
+        goto err;
+    }
+
+    ret = quota_enforcer_lookup(frame, this, xdata, cbk_fn);
+    if (ret < 0) {
+        ret = -ENOTCONN;
+        goto err;
+    }
+
+    ret = 0;
+err:
+    if (xdata)
+        dict_unref(xdata);
+
+    return ret;
 }
 
+void
+quota_check_limit_continuation(struct list_head *parents, inode_t *inode,
+                               int32_t op_ret, int32_t op_errno, void *data)
+{
+    call_frame_t *frame = NULL;
+    xlator_t *this = NULL;
+    quota_local_t *local = NULL;
+    quota_local_t *par_local = NULL;
+    quota_dentry_t *entry = NULL;
+    inode_t *parent = NULL;
+    int parent_count = 0;
+
+    frame = data;
+    local = frame->local;
+    this = THIS;
+
+    if (local->par_frame)
+        par_local = local->par_frame->local;
+    else
+        par_local = local;
+
+    if ((op_ret < 0) || list_empty(parents)) {
+        if (op_ret >= 0) {
+            gf_msg(this->name, GF_LOG_WARNING, EIO, Q_MSG_ANCESTRY_BUILD_FAILED,
+                   "Couldn't build ancestry for inode (gfid:%s). "
+                   "Without knowing ancestors till root, quota"
+                   "cannot be enforced. "
+                   "Hence, failing fop with EIO",
+                   uuid_utoa(inode->gfid));
+            op_errno = EIO;
+        }
 
-int
-quota_ftruncate_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			   int32_t op_ret, int32_t op_errno, struct iatt *buf)
+        quota_handle_validate_error(frame, -1, op_errno);
+        goto out;
+    }
+
+    list_for_each_entry(entry, parents, next) { parent_count++; }
+
+    LOCK(&par_local->lock);
+    {
+        par_local->link_count += (parent_count - 1);
+    }
+    UNLOCK(&par_local->lock);
+
+    if (local->par_frame) {
+        list_for_each_entry(entry, parents, next)
+        {
+            parent = inode_find(inode->table, entry->par);
+            quota_check_limit(frame, parent, this);
+            inode_unref(parent);
+        }
+    } else {
+        list_for_each_entry(entry, parents, next)
+        {
+            parent = do_quota_check_limit(frame, inode, this, entry, _gf_true);
+            if (parent)
+                inode_unref(parent);
+            else
+                quota_link_count_decrement(frame);
+        }
+    }
+
+out:
+    return;
+}
+
+int32_t
+quota_check_object_limit(call_frame_t *frame, quota_inode_ctx_t *ctx,
+                         quota_priv_t *priv, inode_t *_inode, xlator_t *this,
+                         int32_t *op_errno, int just_validated,
+                         quota_local_t *local, gf_boolean_t *skip_check)
 {
-	struct quota_local *local = NULL;
+    int32_t ret = -1;
+    uint32_t timeout = 0;
+    char need_validate = 0;
+    gf_boolean_t hard_limit_exceeded = 0;
+    int64_t object_aggr_count = 0;
+
+    GF_ASSERT(frame);
+    GF_ASSERT(priv);
+    GF_ASSERT(_inode);
+    GF_ASSERT(this);
+    GF_ASSERT(local);
+
+    if (ctx != NULL && (ctx->object_hard_lim > 0 || ctx->object_soft_lim)) {
+        LOCK(&ctx->lock);
+        {
+            timeout = priv->soft_timeout;
+
+            object_aggr_count = ctx->file_count + ctx->dir_count + 1;
+            if (((ctx->object_soft_lim >= 0) &&
+                 (object_aggr_count) > ctx->object_soft_lim)) {
+                timeout = priv->hard_timeout;
+            }
+
+            if (!just_validated && quota_timeout(ctx->validate_time, timeout)) {
+                need_validate = 1;
+            } else if ((object_aggr_count) > ctx->object_hard_lim) {
+                hard_limit_exceeded = 1;
+            }
+        }
+        UNLOCK(&ctx->lock);
+
+        if (need_validate && *skip_check != _gf_true) {
+            *skip_check = _gf_true;
+            ret = quota_validate(frame, _inode, this, quota_validate_cbk);
+            if (ret < 0) {
+                *op_errno = -ret;
+                *skip_check = _gf_false;
+            }
+            goto out;
+        }
+
+        if (hard_limit_exceeded) {
+            local->op_ret = -1;
+            local->op_errno = EDQUOT;
+            *op_errno = EDQUOT;
+            goto out;
+        }
 
-	local = frame->local;
+        /*We log usage only if quota limit is configured on
+           that inode
+        */
+        quota_log_usage(this, ctx, _inode, 0);
+    }
 
-	if (op_ret >= 0) {
-		local->stbuf = *buf;
-	}
+    ret = 0;
 
-	STACK_WIND (frame, quota_ftruncate_cbk,
-		    FIRST_CHILD (this), FIRST_CHILD (this)->fops->ftruncate,
-		    local->fd, local->offset);
-	return 0;
+out:
+    return ret;
 }
 
+int32_t
+quota_check_size_limit(call_frame_t *frame, quota_inode_ctx_t *ctx,
+                       quota_priv_t *priv, inode_t *_inode, xlator_t *this,
+                       int32_t *op_errno, int just_validated, int64_t delta,
+                       quota_local_t *local, gf_boolean_t *skip_check)
+{
+    int32_t ret = -1;
+    uint32_t timeout = 0;
+    char need_validate = 0;
+    gf_boolean_t hard_limit_exceeded = 0;
+    int64_t space_available = 0;
+    int64_t wouldbe_size = 0;
+
+    GF_ASSERT(frame);
+    GF_ASSERT(priv);
+    GF_ASSERT(_inode);
+    GF_ASSERT(this);
+    GF_ASSERT(local);
+
+    if (ctx != NULL && (ctx->hard_lim > 0 || ctx->soft_lim > 0)) {
+        wouldbe_size = ctx->size + delta;
+
+        LOCK(&ctx->lock);
+        {
+            timeout = priv->soft_timeout;
+
+            if ((ctx->soft_lim >= 0) && (wouldbe_size > ctx->soft_lim)) {
+                timeout = priv->hard_timeout;
+            }
+
+            if (!just_validated && quota_timeout(ctx->validate_time, timeout)) {
+                need_validate = 1;
+            } else if (wouldbe_size >= ctx->hard_lim) {
+                hard_limit_exceeded = 1;
+            }
+        }
+        UNLOCK(&ctx->lock);
+
+        if (need_validate && *skip_check != _gf_true) {
+            *skip_check = _gf_true;
+            ret = quota_validate(frame, _inode, this, quota_validate_cbk);
+            if (ret < 0) {
+                *op_errno = -ret;
+                *skip_check = _gf_false;
+            }
+            goto out;
+        }
+
+        if (hard_limit_exceeded) {
+            local->op_ret = -1;
+            local->op_errno = EDQUOT;
 
-int
-quota_ftruncate (call_frame_t *frame, xlator_t *this,
-		 fd_t *fd, off_t offset)
+            space_available = ctx->hard_lim - ctx->size;
+
+            if (space_available < 0)
+                space_available = 0;
+
+            if ((local->space_available < 0) ||
+                (local->space_available > space_available)) {
+                local->space_available = space_available;
+            }
+
+            if (space_available == 0) {
+                *op_errno = EDQUOT;
+                goto out;
+            }
+        }
+
+        /* We log usage only if quota limit is configured on
+           that inode. */
+        quota_log_usage(this, ctx, _inode, delta);
+    }
+
+    ret = 0;
+out:
+    return ret;
+}
+
+int32_t
+quota_check_limit(call_frame_t *frame, inode_t *inode, xlator_t *this)
 {
-	struct quota_local *local = NULL;
-	struct quota_priv  *priv = NULL;
+    int32_t ret = -1, op_errno = EINVAL;
+    inode_t *_inode = NULL, *parent = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    quota_priv_t *priv = NULL;
+    quota_local_t *local = NULL;
+    quota_local_t *par_local = NULL;
+    char just_validated = 0;
+    int64_t delta = 0;
+    int8_t object_delta = 0;
+    uint64_t value = 0;
+    gf_boolean_t skip_check = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("quota", this, err);
+    GF_VALIDATE_OR_GOTO(this->name, frame, err);
+    GF_VALIDATE_OR_GOTO(this->name, inode, err);
+
+    local = frame->local;
+    GF_VALIDATE_OR_GOTO(this->name, local, err);
+
+    if (local->par_frame) {
+        par_local = local->par_frame->local;
+        GF_VALIDATE_OR_GOTO(this->name, par_local, err);
+    } else {
+        par_local = local;
+    }
+
+    delta = par_local->delta;
+    object_delta = par_local->object_delta;
+
+    GF_VALIDATE_OR_GOTO(this->name, par_local->stub, err);
+    /* Allow all the trusted clients
+     * Don't block the gluster internal processes like rebalance, gsyncd,
+     * self heal etc from the disk quotas.
+     *
+     * Method: Allow all the clients with PID negative. This is by the
+     * assumption that any kernel assigned pid doesn't have the negative
+     * number.
+     */
+    if (0 > frame->root->pid) {
+        ret = 0;
+        quota_link_count_decrement(frame);
+        goto done;
+    }
+
+    priv = this->private;
+
+    inode_ctx_get(inode, this, &value);
+    ctx = (quota_inode_ctx_t *)(unsigned long)value;
+
+    _inode = inode_ref(inode);
+
+    LOCK(&local->lock);
+    {
+        just_validated = local->just_validated;
+        local->just_validated = 0;
+    }
+    UNLOCK(&local->lock);
+
+    do {
+        /* In a rename operation, enforce should be stopped at common
+           ancestor */
+        if (!gf_uuid_is_null(par_local->common_ancestor) &&
+            !gf_uuid_compare(_inode->gfid, par_local->common_ancestor)) {
+            quota_link_count_decrement(frame);
+            break;
+        }
 
+        if (object_delta <= 0)
+            goto skip_check_object_limit;
+
+        ret = quota_check_object_limit(frame, ctx, priv, _inode, this,
+                                       &op_errno, just_validated, par_local,
+                                       &skip_check);
+        if (skip_check == _gf_true)
+            goto done;
+
+        if (ret) {
+            if (op_errno != EDQUOT)
+                gf_msg(this->name, GF_LOG_ERROR, 0, Q_MSG_ENFORCEMENT_FAILED,
+                       "Failed to "
+                       "check quota object limit");
+            goto err;
+        }
 
-	priv = this->private;
+    skip_check_object_limit:
+        ret = quota_check_size_limit(frame, ctx, priv, _inode, this, &op_errno,
+                                     just_validated, delta, par_local,
+                                     &skip_check);
+        if (skip_check == _gf_true)
+            goto done;
+
+        if (ret) {
+            if (op_errno != EDQUOT)
+                gf_msg(this->name, GF_LOG_ERROR, 0, Q_MSG_ENFORCEMENT_FAILED,
+                       "Failed to "
+                       "check quota size limit");
+            goto err;
+        }
+
+        if (__is_root_gfid(_inode->gfid)) {
+            quota_link_count_decrement(frame);
+            break;
+        }
+
+        parent = inode_parent(_inode, 0, NULL);
+        if (parent == NULL) {
+            ret = quota_build_ancestry(_inode, quota_check_limit_continuation,
+                                       frame);
+            if (ret < 0) {
+                op_errno = -ret;
+                goto err;
+            }
+
+            break;
+        }
 
-	if (priv->disk_usage_limit) {
-		local = GF_CALLOC (1, sizeof (struct quota_local),
-                                   gf_quota_mt_quota_local);
-		frame->local  = local;
+        inode_unref(_inode);
+        _inode = parent;
+        just_validated = 0;
 
-		local->fd = fd_ref (fd);
-		local->offset = offset;
+        value = 0;
+        inode_ctx_get(_inode, this, &value);
+        ctx = (quota_inode_ctx_t *)(unsigned long)value;
+    } while (1);
 
-		STACK_WIND (frame, quota_ftruncate_fstat_cbk,
-			    FIRST_CHILD(this),
-			    FIRST_CHILD(this)->fops->fstat, fd);
-		return 0;
-	}
+done:
+    if (_inode != NULL) {
+        inode_unref(_inode);
+        _inode = NULL;
+    }
+    return 0;
 
-	STACK_WIND (frame, quota_ftruncate_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->ftruncate,
-		    fd, offset);
-	return 0;
+err:
+    quota_handle_validate_error(frame, -1, op_errno);
+
+    inode_unref(_inode);
+    return 0;
 }
 
+inode_t *
+do_quota_check_limit(call_frame_t *frame, inode_t *inode, xlator_t *this,
+                     quota_dentry_t *dentry, gf_boolean_t force)
+{
+    int32_t ret = -1;
+    inode_t *parent = NULL;
+    call_frame_t *new_frame = NULL;
+    quota_local_t *new_local = NULL;
+
+    parent = inode_parent(inode, dentry->par, dentry->name);
+    if (parent == NULL) {
+        if (force)
+            parent = inode_find(inode->table, dentry->par);
+        else
+            goto out;
+    }
+    if (parent == NULL)
+        goto out;
+
+    new_frame = copy_frame(frame);
+    if (new_frame == NULL)
+        goto out;
+
+    new_local = quota_local_new();
+    if (new_local == NULL)
+        goto out;
+
+    new_frame->local = new_local;
+    new_local->par_frame = frame;
+
+    quota_check_limit(new_frame, parent, this);
+
+    ret = 0;
+out:
+    if (ret < 0) {
+        if (parent) {
+            /* Caller should decrement link_count, in case parent is
+             * NULL
+             */
+            quota_handle_validate_error(frame, -1, ENOMEM);
+        }
+
+        if (new_frame) {
+            new_frame->local = NULL;
+            STACK_DESTROY(new_frame->root);
+        }
+    }
+
+    return parent;
+}
+
+static int
+quota_get_limits(xlator_t *this, dict_t *dict, int64_t *hard_lim,
+                 int64_t *soft_lim, int64_t *object_hard_limit,
+                 int64_t *object_soft_limit)
+{
+    quota_limits_t *limit = NULL;
+    quota_limits_t *object_limit = NULL;
+    quota_priv_t *priv = NULL;
+    int64_t soft_lim_percent = 0;
+    int64_t *ptr = NULL;
+    int ret = 0;
+
+    if ((this == NULL) || (dict == NULL) || (hard_lim == NULL) ||
+        (soft_lim == NULL))
+        goto out;
+
+    priv = this->private;
+
+    ret = dict_get_bin(dict, QUOTA_LIMIT_KEY, (void **)&ptr);
+    limit = (quota_limits_t *)ptr;
+
+    if (limit) {
+        *hard_lim = ntoh64(limit->hl);
+        soft_lim_percent = ntoh64(limit->sl);
+    }
+
+    if (soft_lim_percent < 0) {
+        soft_lim_percent = priv->default_soft_lim;
+    }
+
+    if ((*hard_lim > 0) && (soft_lim_percent > 0)) {
+        *soft_lim = (soft_lim_percent * (*hard_lim)) / 100;
+    }
+
+    ret = dict_get_bin(dict, QUOTA_LIMIT_OBJECTS_KEY, (void **)&ptr);
+    if (ret)
+        return 0;
+    object_limit = (quota_limits_t *)ptr;
+
+    if (object_limit) {
+        *object_hard_limit = ntoh64(object_limit->hl);
+        soft_lim_percent = ntoh64(object_limit->sl);
+    }
+
+    if (soft_lim_percent < 0) {
+        soft_lim_percent = priv->default_soft_lim;
+    }
+
+    if ((*object_hard_limit > 0) && (soft_lim_percent > 0)) {
+        *object_soft_limit = (soft_lim_percent * (*object_hard_limit)) / 100;
+    }
+
+out:
+    return 0;
+}
 
 int
-quota_mknod_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		 int32_t op_ret, int32_t op_errno,
-                 inode_t *inode, struct iatt *buf, struct iatt *preparent,
-                 struct iatt *postparent)
+quota_fill_inodectx(xlator_t *this, inode_t *inode, dict_t *dict, loc_t *loc,
+                    struct iatt *buf, int32_t *op_errno)
 {
-	struct quota_priv *priv = NULL;
+    int32_t ret = -1;
+    char found = 0;
+    quota_inode_ctx_t *ctx = NULL;
+    quota_dentry_t *dentry = NULL;
+    uint64_t value = 0;
+    int64_t hard_lim = 0;
+    int64_t soft_lim = 0;
+    int64_t object_hard_limit = 0;
+    int64_t object_soft_limit = 0;
+
+    quota_get_limits(this, dict, &hard_lim, &soft_lim, &object_hard_limit,
+                     &object_soft_limit);
+
+    inode_ctx_get(inode, this, &value);
+    ctx = (quota_inode_ctx_t *)(unsigned long)value;
+
+    if ((((ctx == NULL) || (ctx->hard_lim == hard_lim)) && (hard_lim < 0) &&
+         !QUOTA_REG_OR_LNK_FILE(buf->ia_type))) {
+        ret = 0;
+        goto out;
+    }
+
+    ret = quota_inode_ctx_get(inode, this, &ctx, 1);
+    if ((ret == -1) || (ctx == NULL)) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_INODE_CTX_GET_FAILED,
+               "cannot create quota "
+               "context in inode(gfid:%s)",
+               uuid_utoa(inode->gfid));
+        ret = -1;
+        *op_errno = ENOMEM;
+        goto out;
+    }
+
+    LOCK(&ctx->lock);
+    {
+        ctx->hard_lim = hard_lim;
+        ctx->soft_lim = soft_lim;
+        ctx->object_hard_lim = object_hard_limit;
+        ctx->object_soft_lim = object_soft_limit;
+
+        ctx->buf = *buf;
+
+        if (!QUOTA_REG_OR_LNK_FILE(buf->ia_type)) {
+            goto unlock;
+        }
 
-	priv = this->private;
+        /* do nothing if it is a nameless lookup */
+        if (loc->name == NULL || !loc->parent)
+            goto unlock;
+
+        list_for_each_entry(dentry, &ctx->parents, next)
+        {
+            if ((strcmp(dentry->name, loc->name) == 0) &&
+                (gf_uuid_compare(loc->parent->gfid, dentry->par) == 0)) {
+                found = 1;
+                break;
+            }
+        }
 
-	if ((op_ret >= 0) && priv->disk_usage_limit) {
-		gf_quota_usage_add (this, buf->ia_blocks * 512);
-	}
+        if (!found) {
+            dentry = __quota_dentry_new(ctx, (char *)loc->name,
+                                        loc->parent->gfid);
+            if (dentry == NULL) {
+                /*
+                  gf_msg (this->name, GF_LOG_WARNING, ENOMEM,
+                          Q_MSG_ENOMEM,
+                          "cannot create a new dentry (par:%"
+-                                          PRId64", name:%s) for inode(ino:%"
+-                                          PRId64", gfid:%s)",
+-                                          uuid_utoa (local->loc.inode->gfid));
+                */
+                ret = -1;
+                *op_errno = ENOMEM;
+                goto unlock;
+            }
+        }
+    }
+unlock:
+    UNLOCK(&ctx->lock);
 
-	STACK_UNWIND_STRICT (mknod, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent);
-	return 0;
+out:
+    return ret;
 }
 
+/*
+ * return _gf_true if enforcement is needed and _gf_false otherwise
+ */
+gf_boolean_t
+should_quota_enforce(xlator_t *this, dict_t *dict, glusterfs_fop_t fop)
+{
+    int ret = 0;
+
+    ret = dict_check_flag(dict, GF_INTERNAL_CTX_KEY, GF_DHT_HEAL_DIR);
+
+    if (fop == GF_FOP_MKDIR && ret == DICT_FLAG_SET) {
+        return _gf_false;
+    } else if (ret == -ENOENT) {
+        gf_msg(this->name, GF_LOG_DEBUG, EINVAL, Q_MSG_INTERNAL_FOP_KEY_MISSING,
+               "No internal fop context present");
+        goto out;
+    }
+out:
+    return _gf_true;
+}
 
-int
-quota_mknod (call_frame_t *frame, xlator_t *this,
-	     loc_t *loc, mode_t mode, dev_t rdev, dict_t *params)
+int32_t
+quota_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, inode_t *inode,
+                 struct iatt *buf, dict_t *dict, struct iatt *postparent)
 {
-	struct quota_priv *priv = NULL;
+    quota_local_t *local = NULL;
+    inode_t *this_inode = NULL;
 
-	priv = this->private;
+    local = frame->local;
+    frame->local = NULL;
 
-	if (gf_quota_check_free_disk (this) == -1) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"min-free-disk limit (%u) crossed, current available is %u",
-			priv->min_free_disk_limit, priv->current_free_disk);
-		STACK_UNWIND_STRICT (mknod, frame, -1, ENOSPC, NULL, NULL,
-                                     NULL, NULL);
-		return 0;
-	}
+    if (op_ret >= 0 && inode) {
+        this_inode = inode_ref(inode);
 
-        if (priv->current_disk_usage > priv->disk_usage_limit) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"Disk usage limit (%"PRIu64") crossed, current usage is %"PRIu64"",
-			priv->disk_usage_limit, priv->current_disk_usage);
-		STACK_UNWIND_STRICT (mknod, frame, -1, ENOSPC, NULL, NULL,
-                                     NULL, NULL);
-		return 0;
-        }
+        op_ret = quota_fill_inodectx(this, inode, dict, &local->loc, buf,
+                                     &op_errno);
+        if (op_ret < 0)
+            op_errno = ENOMEM;
+    }
 
-	STACK_WIND (frame, quota_mknod_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->mknod,
-		    loc, mode, rdev, params);
-	return 0;
-}
+    QUOTA_STACK_UNWIND(lookup, frame, op_ret, op_errno, inode, buf, dict,
+                       postparent);
 
+    if (op_ret < 0 || this_inode == NULL || gf_uuid_is_null(this_inode->gfid))
+        goto out;
 
-int
-quota_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		 int32_t op_ret, int32_t op_errno, inode_t *inode,
-                 struct iatt *buf, struct iatt *preparent,
-                 struct iatt *postparent)
+    check_ancestory_2(this, local, this_inode);
+
+out:
+    if (this_inode)
+        inode_unref(this_inode);
+
+    quota_local_cleanup(local);
+
+    return 0;
+}
+
+int32_t
+quota_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
 {
-	struct quota_priv *priv = NULL;
+    quota_priv_t *priv = NULL;
+    int32_t ret = -1;
+    quota_local_t *local = NULL;
+
+    priv = this->private;
+
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+
+    xattr_req = xattr_req ? dict_ref(xattr_req) : dict_new();
+    if (!xattr_req)
+        goto err;
+
+    local = quota_local_new();
+    if (local == NULL) {
+        goto err;
+    }
+
+    frame->local = local;
+    loc_copy(&local->loc, loc);
+
+    ret = dict_set_int8(xattr_req, QUOTA_LIMIT_KEY, 1);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "dict set of key for "
+               "hard-limit failed");
+        goto err;
+    }
+
+    ret = dict_set_int8(xattr_req, QUOTA_LIMIT_OBJECTS_KEY, 1);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "dict set of key for quota object limit failed");
+        goto err;
+    }
 
-	priv = this->private;
+    STACK_WIND(frame, quota_lookup_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lookup, loc, xattr_req);
 
-	if ((op_ret >= 0) && priv->disk_usage_limit) {
-		gf_quota_usage_subtract (this, buf->ia_blocks * 512);
-	}
+    ret = 0;
 
-	STACK_UNWIND_STRICT (mkdir, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent);
-	return 0;
+err:
+    if (xattr_req)
+        dict_unref(xattr_req);
+
+    if (ret < 0) {
+        QUOTA_STACK_UNWIND(lookup, frame, -1, ENOMEM, NULL, NULL, NULL, NULL);
+    }
+
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->lookup,
+                    loc, xattr_req);
+    return 0;
 }
 
+int32_t
+quota_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                 struct iatt *postbuf, dict_t *xdata)
+{
+    int32_t ret = 0;
+    uint64_t ctx_int = 0;
+    quota_inode_ctx_t *ctx = NULL;
+    quota_local_t *local = NULL;
+
+    local = frame->local;
+
+    if ((op_ret < 0) || (local == NULL) || (postbuf == NULL)) {
+        goto out;
+    }
+
+    ret = inode_ctx_get(local->loc.inode, this, &ctx_int);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, Q_MSG_INODE_CTX_GET_FAILED,
+               "%s: failed to get the "
+               "context",
+               local->loc.path);
+        goto out;
+    }
+
+    ctx = (quota_inode_ctx_t *)(unsigned long)ctx_int;
+
+    if (ctx == NULL) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, Q_MSG_INODE_CTX_GET_FAILED,
+               "quota context not set in %s (gfid:%s)", local->loc.path,
+               uuid_utoa(local->loc.inode->gfid));
+        goto out;
+    }
+
+    LOCK(&ctx->lock);
+    {
+        ctx->buf = *postbuf;
+    }
+    UNLOCK(&ctx->lock);
+
+out:
+    QUOTA_STACK_UNWIND(writev, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+
+    return 0;
+}
 
-int
-quota_mkdir (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
-             dict_t *params)
+static int gf_quota_enforcer_log;
+
+int32_t
+quota_writev_helper(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                    struct iovec *vector, int32_t count, off_t off,
+                    uint32_t flags, struct iobref *iobref, dict_t *xdata)
 {
-	struct quota_priv *priv = NULL;
+    quota_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+    struct iovec *new_vector = NULL;
+    int32_t new_count = 0;
+
+    local = frame->local;
+
+    GF_VALIDATE_OR_GOTO("quota", local, unwind);
+
+    if (local->op_ret == -1) {
+        op_errno = local->op_errno;
+
+        if ((op_errno == EDQUOT) && (local->space_available > 0)) {
+            new_count = iov_subset(vector, count, 0, local->space_available,
+                                   &new_vector, 0);
+            if (new_count < 0) {
+                local->op_ret = -1;
+                local->op_errno = ENOMEM;
+                goto unwind;
+            }
+
+            vector = new_vector;
+            count = new_count;
+        } else if (op_errno == ENOENT || op_errno == ESTALE) {
+            /* We may get ENOENT/ESTALE in case of below scenario
+             *     fd = open file.txt
+             *     unlink file.txt
+             *     write on fd
+             * Here build_ancestry can fail as the file is removed.
+             * For now ignore ENOENT/ESTALE with writes on active fd
+             * We need to re-visit this code once we understand
+             * how other file-system behave in this scenario
+             */
+            gf_msg_debug(this->name, 0,
+                         "quota enforcer failed "
+                         "with ENOENT/ESTALE on %s, cannot check "
+                         "quota limits and allowing writes",
+                         uuid_utoa(fd->inode->gfid));
+        } else if ((op_errno == EINVAL) &&
+                   !inode_parent(local->loc.inode, 0, NULL)) {
+            /* We may get INVAL with parent == NULL,
+             * in case of below scenario
+             *     1. enable quota
+             *     2. glusterfsd stop/start
+             *     3. nameless lookup
+             *     4. write on fd
+             * Here build_ancestry can fail as the file's pgfid
+             * is't exist.
+             * For now ignore EINVAL with writes on active fd
+             * untils the pgfid is created at name lookup
+             */
+            GF_LOG_OCCASIONALLY(gf_quota_enforcer_log, this->name,
+                                GF_LOG_CRITICAL,
+                                "Quota cannot be enforced as "
+                                "parent is not available and writes are being "
+                                "allowed without checking whether they are "
+                                "within quota limits. This can happen if Quota "
+                                "crawl is not complete. If crawl has been "
+                                "completed, please file a bug.");
+        } else {
+            goto unwind;
+        }
+    }
+
+    STACK_WIND(frame, quota_writev_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->writev, fd, vector, count, off, flags,
+               iobref, xdata);
 
-	priv = this->private;
+    if (new_vector != NULL)
+        GF_FREE(new_vector);
 
-	if (gf_quota_check_free_disk (this) == -1) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"min-free-disk limit (%u) crossed, current available is %u",
-			priv->min_free_disk_limit, priv->current_free_disk);
-		STACK_UNWIND_STRICT (mkdir, frame, -1, ENOSPC, NULL, NULL,
-                                     NULL, NULL);
-		return 0;
-		
-	}
+    return 0;
+
+unwind:
+    QUOTA_STACK_UNWIND(writev, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
+}
+
+int32_t
+quota_writev(call_frame_t *frame, xlator_t *this, fd_t *fd,
+             struct iovec *vector, int32_t count, off_t off, uint32_t flags,
+             struct iobref *iobref, dict_t *xdata)
+{
+    quota_priv_t *priv = NULL;
+    int32_t op_errno = EINVAL;
+    int32_t parents = 0;
+    int32_t fail_count = 0;
+    uint64_t size = 0;
+    quota_local_t *local = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    quota_dentry_t *dentry = NULL, *tmp = NULL;
+    call_stub_t *stub = NULL;
+    struct list_head head;
+    inode_t *par_inode = NULL;
+
+    priv = this->private;
+
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+
+    INIT_LIST_HEAD(&head);
+
+    GF_ASSERT(frame);
+    GF_VALIDATE_OR_GOTO("quota", this, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, fd, unwind);
+
+    local = quota_local_new();
+    if (local == NULL) {
+        goto unwind;
+    }
+
+    frame->local = local;
+    local->loc.inode = inode_ref(fd->inode);
+
+    (void)quota_inode_ctx_get(fd->inode, this, &ctx, 0);
+    if (ctx == NULL) {
+        gf_msg_debug(this->name, 0,
+                     "quota context is NULL on inode"
+                     " (%s). If quota is not enabled recently and "
+                     "crawler has finished crawling, its an error",
+                     uuid_utoa(fd->inode->gfid));
+    }
+
+    stub = fop_writev_stub(frame, quota_writev_helper, fd, vector, count, off,
+                           flags, iobref, xdata);
+    if (stub == NULL) {
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, priv, unwind);
+
+    parents = quota_add_parents_from_ctx(ctx, &head);
+    if (parents == -1) {
+        op_errno = errno;
+        goto unwind;
+    }
+
+    size = iov_length(vector, count);
+
+    LOCK(&local->lock);
+    {
+        local->delta = size;
+        local->object_delta = 0;
+        local->link_count = (parents != 0) ? parents : 1;
+        local->stub = stub;
+    }
+    UNLOCK(&local->lock);
+
+    if (parents == 0) {
+        /* nameless lookup on this inode, allow quota to reconstruct
+         * ancestry as part of check_limit.
+         */
+        quota_check_limit(frame, fd->inode, this);
+    } else {
+        list_for_each_entry_safe(dentry, tmp, &head, next)
+        {
+            par_inode = do_quota_check_limit(frame, fd->inode, this, dentry,
+                                             _gf_false);
+            if (par_inode == NULL) {
+                if (ctx) {
+                    /* remove stale entry from inode ctx */
+                    quota_dentry_del(ctx, dentry->name, dentry->par);
+                    parents--;
+                    fail_count++;
+                }
+            } else {
+                inode_unref(par_inode);
+            }
+            __quota_dentry_free(dentry);
+        }
+
+        if (parents == 0) {
+            LOCK(&local->lock);
+            {
+                local->link_count++;
+            }
+            UNLOCK(&local->lock);
+            quota_check_limit(frame, fd->inode, this);
+        }
 
-        if (priv->current_disk_usage > priv->disk_usage_limit) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"Disk usage limit (%"PRIu64") crossed, current usage is %"PRIu64"",
-			priv->disk_usage_limit, priv->current_disk_usage);
-		STACK_UNWIND_STRICT (mkdir, frame, -1, ENOSPC, NULL, NULL,
-                                     NULL, NULL);
-		return 0;
+        while (fail_count != 0) {
+            quota_link_count_decrement(frame);
+            fail_count--;
         }
+    }
 
-	STACK_WIND (frame, quota_mkdir_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->mkdir,
-		    loc, mode, params);
+    return 0;
 
-	return 0;
+unwind:
+    QUOTA_STACK_UNWIND(writev, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->writev,
+                    fd, vector, count, off, flags, iobref, xdata);
+    return 0;
 }
 
+int32_t
+quota_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, inode_t *inode,
+                struct iatt *buf, struct iatt *preparent,
+                struct iatt *postparent, dict_t *xdata)
+{
+    QUOTA_STACK_UNWIND(mkdir, frame, op_ret, op_errno, inode, buf, preparent,
+                       postparent, xdata);
+    return 0;
+}
 
-int
-quota_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		  int32_t op_ret, int32_t op_errno, struct iatt *preparent,
-                  struct iatt *postparent)
+int32_t
+quota_mkdir_helper(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+                   mode_t umask, dict_t *xdata)
 {
-	struct quota_local *local = NULL;
+    quota_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+
+    local = frame->local;
+
+    GF_VALIDATE_OR_GOTO("quota", local, unwind);
+
+    op_errno = local->op_errno;
 
-	local = frame->local;
+    if (local->op_ret == -1) {
+        goto unwind;
+    }
 
-	if (local) {
-		if (op_ret >= 0) {
-			gf_quota_usage_subtract (this,
-						 local->stbuf.ia_blocks * 512);
-		}
-		loc_wipe (&local->loc);
-	}
+    STACK_WIND(frame, quota_mkdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mkdir, loc, mode, umask, xdata);
 
-	STACK_UNWIND_STRICT (unlink, frame, op_ret, op_errno, preparent, postparent);
-	return 0;
+    return 0;
+
+unwind:
+    QUOTA_STACK_UNWIND(mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                       NULL);
+    return 0;
 }
 
+int32_t
+quota_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+            mode_t umask, dict_t *xdata)
+{
+    quota_priv_t *priv = NULL;
+    int32_t ret = 0, op_errno = 0;
+    quota_local_t *local = NULL;
+    call_stub_t *stub = NULL;
+
+    priv = this->private;
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+
+    if (!should_quota_enforce(this, xdata, GF_FOP_MKDIR)) {
+        gf_msg(this->name, GF_LOG_DEBUG, 0, Q_MSG_ENFORCEMENT_SKIPPED,
+               "Enforcement has been skipped(internal fop).");
+        goto off;
+    }
+
+    local = quota_local_new();
+    if (local == NULL) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    frame->local = local;
+
+    ret = loc_copy(&local->loc, loc);
+    if (ret) {
+        op_errno = ENOMEM;
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "loc_copy failed");
+        goto err;
+    }
+
+    stub = fop_mkdir_stub(frame, quota_mkdir_helper, loc, mode, umask, xdata);
+    if (stub == NULL) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    LOCK(&local->lock);
+    {
+        local->stub = stub;
+        local->delta = 0;
+        local->object_delta = 1;
+        local->link_count = 1;
+    }
+    UNLOCK(&local->lock);
+
+    quota_check_limit(frame, loc->parent, this);
+    return 0;
+
+err:
+    QUOTA_STACK_UNWIND(mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                       NULL);
+
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->mkdir,
+                    loc, mode, umask, xdata);
+
+    return 0;
+}
 
-int
-quota_unlink_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno, struct iatt *buf)
+int32_t
+quota_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, fd_t *fd, inode_t *inode,
+                 struct iatt *buf, struct iatt *preparent,
+                 struct iatt *postparent, dict_t *xdata)
+{
+    int32_t ret = -1;
+    quota_local_t *local = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    quota_dentry_t *dentry = NULL;
+
+    local = frame->local;
+    if (op_ret < 0) {
+        goto unwind;
+    }
+
+    ret = quota_inode_ctx_get(inode, this, &ctx, 1);
+    if ((ret == -1) || (ctx == NULL)) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_INODE_CTX_GET_FAILED,
+               "cannot create quota "
+               "context in inode(gfid:%s)",
+               uuid_utoa(inode->gfid));
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+
+    LOCK(&ctx->lock);
+    {
+        ctx->buf = *buf;
+
+        dentry = __quota_dentry_new(ctx, (char *)local->loc.name,
+                                    local->loc.parent->gfid);
+        if (dentry == NULL) {
+            gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+                   "cannot create a new dentry "
+                   "(name:%s) for inode(gfid:%s)",
+                   local->loc.name, uuid_utoa(local->loc.inode->gfid));
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto unlock;
+        }
+    }
+unlock:
+    UNLOCK(&ctx->lock);
+
+unwind:
+    QUOTA_STACK_UNWIND(create, frame, op_ret, op_errno, fd, inode, buf,
+                       preparent, postparent, xdata);
+    return 0;
+}
+
+int32_t
+quota_create_helper(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                    int32_t flags, mode_t mode, mode_t umask, fd_t *fd,
+                    dict_t *xdata)
 {
-	struct quota_local *local = NULL;
+    quota_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+
+    local = frame->local;
 
-	local = frame->local;
+    GF_VALIDATE_OR_GOTO("quota", local, unwind);
 
-	if (op_ret >= 0) {
-		if (buf->ia_nlink == 1) {
-			local->stbuf = *buf;
-		}
-	}
+    if (local->op_ret == -1) {
+        op_errno = local->op_errno;
+        goto unwind;
+    }
 
-	STACK_WIND (frame, quota_unlink_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->unlink,
-		    &local->loc);
+    STACK_WIND(frame, quota_create_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->create, loc, flags, mode, umask, fd,
+               xdata);
+    return 0;
 
-	return 0;
+unwind:
+    QUOTA_STACK_UNWIND(create, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                       NULL, NULL);
+    return 0;
 }
 
+int32_t
+quota_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+             mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+    quota_priv_t *priv = NULL;
+    int32_t ret = -1;
+    quota_local_t *local = NULL;
+    int32_t op_errno = 0;
+    call_stub_t *stub = NULL;
+
+    priv = this->private;
+
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+    QUOTA_WIND_FOR_INTERNAL_FOP(xdata, off);
+
+    local = quota_local_new();
+    if (local == NULL) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    frame->local = local;
+
+    ret = loc_copy(&local->loc, loc);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "loc_copy failed");
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    stub = fop_create_stub(frame, quota_create_helper, loc, flags, mode, umask,
+                           fd, xdata);
+    if (stub == NULL) {
+        goto err;
+    }
+
+    LOCK(&local->lock);
+    {
+        local->link_count = 1;
+        local->stub = stub;
+        local->delta = 0;
+        local->object_delta = 1;
+    }
+    UNLOCK(&local->lock);
+
+    quota_check_limit(frame, loc->parent, this);
+    return 0;
+err:
+    QUOTA_STACK_UNWIND(create, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                       NULL, NULL);
+
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->create,
+                    loc, flags, mode, umask, fd, xdata);
+    return 0;
+}
 
-int
-quota_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc)
+int32_t
+quota_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                 struct iatt *postparent, dict_t *xdata)
 {
-	struct quota_local *local = NULL;
-	struct quota_priv  *priv = NULL;
+    quota_local_t *local = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    uint64_t value = 0;
+
+    if (op_ret < 0) {
+        goto out;
+    }
 
-	priv = this->private;
+    local = (quota_local_t *)frame->local;
 
-	if (priv->disk_usage_limit) {
-		local = GF_CALLOC (1, sizeof (struct quota_local),
-                                   gf_quota_mt_quota_local);
-		frame->local = local;
+    inode_ctx_get(local->loc.inode, this, &value);
+    ctx = (quota_inode_ctx_t *)(unsigned long)value;
 
-		loc_copy (&local->loc, loc);
+    if (ctx == NULL) {
+        gf_msg(this->name, GF_LOG_INFO, EINVAL, Q_MSG_INODE_CTX_GET_FAILED,
+               "quota context not set inode (gfid:%s)",
+               uuid_utoa(local->loc.gfid));
+        goto out;
+    }
 
-		STACK_WIND (frame,
-			    quota_unlink_stat_cbk,
-			    FIRST_CHILD(this),
-			    FIRST_CHILD(this)->fops->stat,
-			    loc);
-		return 0;
-	}
+    quota_dentry_del(ctx, local->loc.name, local->loc.parent->gfid);
 
-	STACK_WIND (frame, quota_unlink_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->unlink,
-		    loc);
-	return 0;
+out:
+    QUOTA_STACK_UNWIND(unlink, frame, op_ret, op_errno, preparent, postparent,
+                       xdata);
+    return 0;
 }
 
+int32_t
+quota_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+             dict_t *xdata)
+{
+    quota_priv_t *priv = NULL;
+    int32_t ret = -1;
+    quota_local_t *local = NULL;
 
-int
-quota_rmdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		 int32_t op_ret, int32_t op_errno, struct iatt *preparent,
-                 struct iatt *postparent)
+    priv = this->private;
+
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+
+    local = quota_local_new();
+    if (local == NULL) {
+        goto err;
+    }
+
+    frame->local = local;
+
+    ret = loc_copy(&local->loc, loc);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "loc_copy failed");
+        goto err;
+    }
+
+    STACK_WIND(frame, quota_unlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->unlink, loc, xflag, xdata);
+
+    ret = 0;
+
+err:
+    if (ret == -1) {
+        QUOTA_STACK_UNWIND(unlink, frame, -1, 0, NULL, NULL, NULL);
+    }
+
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->unlink,
+                    loc, xflag, xdata);
+    return 0;
+}
+
+int32_t
+quota_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, inode_t *inode,
+               struct iatt *buf, struct iatt *preparent,
+               struct iatt *postparent, dict_t *xdata)
 {
-	struct quota_local *local = NULL;
+    int32_t ret = -1;
+    quota_local_t *local = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    quota_dentry_t *dentry = NULL;
+    char found = 0;
+
+    if (op_ret < 0) {
+        goto out;
+    }
+
+    local = (quota_local_t *)frame->local;
+
+    ret = quota_inode_ctx_get(inode, this, &ctx, 0);
+    if ((ret == -1) || (ctx == NULL)) {
+        gf_msg_debug(this->name, 0,
+                     "quota context is NULL on inode"
+                     " (%s). If quota is not enabled recently and "
+                     "crawler has finished crawling, its an error",
+                     uuid_utoa(inode->gfid));
+        goto out;
+    }
+
+    LOCK(&ctx->lock);
+    {
+        list_for_each_entry(dentry, &ctx->parents, next)
+        {
+            if ((strcmp(dentry->name, local->loc.name) == 0) &&
+                (gf_uuid_compare(local->loc.parent->gfid, dentry->par) == 0)) {
+                found = 1;
+
+                gf_msg_debug(this->name, 0,
+                             "new entry being"
+                             " linked (name:%s) for inode "
+                             "(gfid:%s) is already present "
+                             "in inode-dentry-list",
+                             dentry->name, uuid_utoa(local->loc.inode->gfid));
+                break;
+            }
+        }
+
+        if (!found) {
+            dentry = __quota_dentry_new(ctx, (char *)local->loc.name,
+                                        local->loc.parent->gfid);
+            if (dentry == NULL) {
+                gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+                       "cannot create a new dentry (name:%s)"
+                       "for inode(gfid:%s)",
+                       local->loc.name, uuid_utoa(local->loc.inode->gfid));
+                op_ret = -1;
+                op_errno = ENOMEM;
+                goto unlock;
+            }
+        }
 
-	local = frame->local;
+        ctx->buf = *buf;
+    }
+unlock:
+    UNLOCK(&ctx->lock);
 
-	if (local) {
-		if (op_ret >= 0) {
-			gf_quota_usage_subtract (this, local->stbuf.ia_blocks * 512);
-		}
-		loc_wipe (&local->loc);
-	}
+out:
+    QUOTA_STACK_UNWIND(link, frame, op_ret, op_errno, inode, buf, preparent,
+                       postparent, xdata);
 
-	STACK_UNWIND_STRICT (rmdir, frame, op_ret, op_errno, preparent, postparent);
-	return 0;
+    return 0;
 }
 
+int32_t
+quota_link_helper(call_frame_t *frame, xlator_t *this, loc_t *oldloc,
+                  loc_t *newloc, dict_t *xdata)
+{
+    quota_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
 
-int
-quota_rmdir_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		      int32_t op_ret, int32_t op_errno, struct iatt *buf)
+    local = frame->local;
+
+    GF_VALIDATE_OR_GOTO("quota", local, unwind);
+
+    op_errno = local->op_errno;
+
+    if (local->op_ret == -1) {
+        goto unwind;
+    }
+
+    STACK_WIND(frame, quota_link_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->link, oldloc, newloc, xdata);
+    return 0;
+
+unwind:
+    QUOTA_STACK_UNWIND(link, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+    return 0;
+}
+
+void
+quota_link_continue(call_frame_t *frame)
 {
-	struct quota_local *local = NULL;
+    int32_t ret = -1;
+    int32_t op_errno = EIO;
+    quota_local_t *local = NULL;
+    uuid_t common_ancestor = {0};
+    xlator_t *this = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    inode_t *src_parent = NULL;
+    inode_t *dst_parent = NULL;
+
+    local = frame->local;
+    this = THIS;
+
+    if (local->op_ret < 0) {
+        op_errno = local->op_errno;
+        goto err;
+    }
+
+    if (local->xdata && dict_get(local->xdata, GLUSTERFS_INTERNAL_FOP_KEY)) {
+        /* Treat link as rename, crawl upwards only till common ancestor
+         */
+        ret = quota_find_common_ancestor(
+            local->oldloc.inode, local->newloc.parent, &common_ancestor);
+        if (ret < 0 || gf_uuid_is_null(common_ancestor)) {
+            gf_msg(this->name, GF_LOG_ERROR, ESTALE,
+                   Q_MSG_ANCESTRY_BUILD_FAILED,
+                   "failed to get "
+                   "common_ancestor for %s and %s",
+                   local->oldloc.path, local->newloc.path);
+            op_errno = ESTALE;
+            goto err;
+        }
+    } else {
+        /* Treat link as a new file.
+         * TODO: Currently marker accounts twice for the links created
+         * across directories.
+         * This needs re-visit if marker accounts only once
+         * for the links created across directories
+         */
+        if (local->oldloc.parent)
+            src_parent = inode_ref(local->oldloc.parent);
+        else
+            src_parent = inode_parent(local->oldloc.inode, 0, NULL);
+        dst_parent = local->newloc.parent;
+
+        /* No need to check quota limit if src and dst parents are same
+         */
+        if (src_parent == dst_parent ||
+            gf_uuid_compare(src_parent->gfid, dst_parent->gfid) == 0) {
+            inode_unref(src_parent);
+            goto wind;
+        }
 
-	local = frame->local;
+        inode_unref(src_parent);
+    }
+
+    quota_inode_ctx_get(local->oldloc.inode, this, &ctx, 0);
+    if (ctx == NULL) {
+        gf_msg_debug(this->name, 0,
+                     "quota context is NULL on inode"
+                     " (%s). If quota is not enabled recently and "
+                     "crawler has finished crawling, its an error",
+                     uuid_utoa(local->oldloc.inode->gfid));
+    }
+
+    LOCK(&local->lock);
+    {
+        local->link_count = 1;
+        local->delta = (ctx != NULL) ? ctx->buf.ia_blocks * 512 : 0;
+        local->object_delta = 1;
+        gf_uuid_copy(local->common_ancestor, common_ancestor);
+    }
+    UNLOCK(&local->lock);
+
+    quota_check_limit(frame, local->newloc.parent, this);
+    return;
+
+err:
+    QUOTA_STACK_UNWIND(link, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+    return;
+
+wind:
+    STACK_WIND(frame, quota_link_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->link, &(local->oldloc),
+               &(local->newloc), local->xdata);
+    return;
+}
 
-	if (op_ret >= 0) {
-		local->stbuf = *buf;
-	}
+int32_t
+quota_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+           dict_t *xdata)
+{
+    quota_priv_t *priv = NULL;
+    int32_t ret = -1;
+    int32_t op_errno = ENOMEM;
+    quota_local_t *local = NULL;
+    call_stub_t *stub = NULL;
+
+    priv = this->private;
+
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+
+    local = quota_local_new();
+    if (local == NULL) {
+        goto err;
+    }
+
+    frame->local = (void *)local;
+
+    if (xdata)
+        local->xdata = dict_ref(xdata);
+
+    ret = loc_copy(&local->loc, newloc);
+    if (ret == -1) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "loc_copy failed");
+        goto err;
+    }
+
+    ret = loc_copy(&local->oldloc, oldloc);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "loc_copy failed");
+        goto err;
+    }
+
+    ret = loc_copy(&local->newloc, newloc);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "loc_copy failed");
+        goto err;
+    }
+
+    /* No need to check quota limit if src and dst parents are same */
+    if (oldloc->parent && newloc->parent &&
+        !gf_uuid_compare(oldloc->parent->gfid, newloc->parent->gfid)) {
+        gf_msg_debug(this->name, GF_LOG_DEBUG,
+                     "link %s -> %s are "
+                     "in the same directory, so skip check limit",
+                     oldloc->path, newloc->path);
+        goto wind;
+    }
+
+    stub = fop_link_stub(frame, quota_link_helper, oldloc, newloc, xdata);
+    if (stub == NULL) {
+        goto err;
+    }
+
+    LOCK(&local->lock);
+    {
+        local->link_count = 2;
+        local->fop_continue_cbk = quota_link_continue;
+        local->stub = stub;
+    }
+    UNLOCK(&local->lock);
+
+    check_ancestory(frame, newloc->parent);
+
+    /* source parent can be NULL, so do check_ancestry on a file */
+    if (oldloc->parent)
+        check_ancestory(frame, oldloc->parent);
+    else
+        check_ancestory(frame, oldloc->inode);
+
+    return 0;
+
+err:
+    QUOTA_STACK_UNWIND(link, frame, -1, op_errno, NULL, NULL, NULL, NULL, NULL);
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->link,
+                    oldloc, newloc, xdata);
+    return 0;
+
+wind:
+    STACK_WIND(frame, quota_link_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->link, oldloc, newloc, xdata);
+    return 0;
+}
 
-	STACK_WIND (frame, quota_rmdir_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->rmdir,
-		    &local->loc, local->flags);
+int32_t
+quota_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                 struct iatt *preoldparent, struct iatt *postoldparent,
+                 struct iatt *prenewparent, struct iatt *postnewparent,
+                 dict_t *xdata)
+{
+    int32_t ret = -1;
+    quota_local_t *local = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    quota_dentry_t *old_dentry = NULL, *dentry = NULL;
+    char new_dentry_found = 0;
+
+    if (op_ret < 0) {
+        goto out;
+    }
+
+    local = frame->local;
+
+    GF_VALIDATE_OR_GOTO("quota", local, out);
+
+    if (!QUOTA_REG_OR_LNK_FILE(local->oldloc.inode->ia_type))
+        goto out;
+
+    ret = quota_inode_ctx_get(local->oldloc.inode, this, &ctx, 0);
+    if ((ret == -1) || (ctx == NULL)) {
+        gf_msg_debug(this->name, 0,
+                     "quota context is NULL on inode"
+                     " (%s). If quota is not enabled recently and "
+                     "crawler has finished crawling, its an error",
+                     uuid_utoa(local->oldloc.inode->gfid));
+
+        goto out;
+    }
+
+    LOCK(&ctx->lock);
+    {
+        list_for_each_entry(dentry, &ctx->parents, next)
+        {
+            if ((strcmp(dentry->name, local->oldloc.name) == 0) &&
+                (gf_uuid_compare(local->oldloc.parent->gfid, dentry->par) ==
+                 0)) {
+                old_dentry = dentry;
+            } else if ((strcmp(dentry->name, local->newloc.name) == 0) &&
+                       (gf_uuid_compare(local->newloc.parent->gfid,
+                                        dentry->par) == 0)) {
+                new_dentry_found = 1;
+                gf_msg_debug(this->name, 0,
+                             "new entry being "
+                             "linked (name:%s) for inode (gfid:%s) "
+                             "is in inode-dentry-list",
+                             dentry->name,
+                             uuid_utoa(local->oldloc.inode->gfid));
+            }
+
+            if (old_dentry && new_dentry_found)
+                break;
+        }
 
-	return 0;
+        if (old_dentry != NULL) {
+            __quota_dentry_free(old_dentry);
+        } else {
+            gf_msg_debug(this->name, 0,
+                         "dentry corresponding"
+                         "the path just renamed (name:%s) is not"
+                         " present",
+                         local->oldloc.name);
+        }
+
+        if (!new_dentry_found) {
+            dentry = __quota_dentry_new(ctx, (char *)local->newloc.name,
+                                        local->newloc.parent->gfid);
+            if (dentry == NULL) {
+                gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+                       "cannot create a new dentry (name:%s) "
+                       "for inode(gfid:%s)",
+                       local->newloc.name,
+                       uuid_utoa(local->newloc.inode->gfid));
+                op_ret = -1;
+                op_errno = ENOMEM;
+                goto unlock;
+            }
+        }
+
+        ctx->buf = *buf;
+    }
+unlock:
+    UNLOCK(&ctx->lock);
+
+out:
+    QUOTA_STACK_UNWIND(rename, frame, op_ret, op_errno, buf, preoldparent,
+                       postoldparent, prenewparent, postnewparent, xdata);
+
+    return 0;
 }
 
+int32_t
+quota_rename_helper(call_frame_t *frame, xlator_t *this, loc_t *oldloc,
+                    loc_t *newloc, dict_t *xdata)
+{
+    quota_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
 
-int
-quota_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags)
+    local = frame->local;
+
+    GF_VALIDATE_OR_GOTO("quota", local, unwind);
+
+    op_errno = local->op_errno;
+
+    if (local->op_ret == -1) {
+        goto unwind;
+    }
+
+    STACK_WIND(frame, quota_rename_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
+
+    return 0;
+
+unwind:
+    QUOTA_STACK_UNWIND(rename, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                       NULL, NULL);
+    return 0;
+}
+
+static int32_t
+quota_rename_get_size_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, inode_t *inode,
+                          struct iatt *buf, dict_t *xdata,
+                          struct iatt *postparent)
+{
+    quota_local_t *local = NULL;
+    int32_t ret = 0;
+    int64_t *size = 0;
+
+    GF_ASSERT(frame);
+    GF_VALIDATE_OR_GOTO_WITH_ERROR("quota", this, out, op_errno, EINVAL);
+    GF_VALIDATE_OR_GOTO_WITH_ERROR(this->name, xdata, out, op_errno, EINVAL);
+    local = frame->local;
+    GF_ASSERT(local);
+    local->link_count = 1;
+
+    if (op_ret < 0)
+        goto out;
+
+    ret = dict_get_bin(xdata, QUOTA_SIZE_KEY, (void **)&size);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, EINVAL, Q_MSG_SIZE_KEY_MISSING,
+               "size key not present in dict");
+        op_errno = EINVAL;
+        goto out;
+    }
+    local->delta = ntoh64(*size);
+    local->object_delta = 1;
+    quota_check_limit(frame, local->newloc.parent, this);
+    return 0;
+
+out:
+    quota_handle_validate_error(frame, -1, op_errno);
+    return 0;
+}
+
+void
+quota_rename_continue(call_frame_t *frame)
 {
-	struct quota_local *local = NULL;
-	struct quota_priv  *priv = NULL;
+    int32_t ret = -1;
+    int32_t op_errno = EIO;
+    quota_local_t *local = NULL;
+    uuid_t common_ancestor = {0};
+    xlator_t *this = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+
+    local = frame->local;
+    this = THIS;
+
+    if (local->op_ret < 0) {
+        op_errno = local->op_errno;
+        goto err;
+    }
+
+    ret = quota_find_common_ancestor(local->oldloc.parent, local->newloc.parent,
+                                     &common_ancestor);
+    if (ret < 0 || gf_uuid_is_null(common_ancestor)) {
+        gf_msg(this->name, GF_LOG_ERROR, ESTALE, Q_MSG_ANCESTRY_BUILD_FAILED,
+               "failed to get "
+               "common_ancestor for %s and %s",
+               local->oldloc.path, local->newloc.path);
+        op_errno = ESTALE;
+        goto err;
+    }
+
+    LOCK(&local->lock);
+    {
+        local->link_count = 1;
+        gf_uuid_copy(local->common_ancestor, common_ancestor);
+    }
+    UNLOCK(&local->lock);
+
+    if (QUOTA_REG_OR_LNK_FILE(local->oldloc.inode->ia_type)) {
+        ret = quota_inode_ctx_get(local->oldloc.inode, this, &ctx, 0);
+        if (ctx == NULL) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, Q_MSG_INODE_CTX_GET_FAILED,
+                   "quota context not set in inode (gfid:%s), "
+                   "considering file size as zero while enforcing "
+                   "quota on new ancestry",
+                   uuid_utoa(local->oldloc.inode->gfid));
+
+            local->delta = 0;
+            local->object_delta = 1;
+        } else {
+            /* FIXME: We need to account for the size occupied by
+             * this inode on the target directory. To avoid double
+             * accounting, we need to modify enforcer to perform
+             * quota_check_limit only up till the least common
+             * ancestor directory inode*/
+
+            /* FIXME: The following code assumes that regular files
+             * and link files are present, in their entirety, in a
+             * single brick. This *assumption is invalid in the
+             * case of stripe.*/
+
+            local->delta = ctx->buf.ia_blocks * 512;
+            local->object_delta = 1;
+        }
 
-	priv = this->private;
+    } else if (IA_ISDIR(local->oldloc.inode->ia_type)) {
+        ret = quota_validate(frame, local->oldloc.inode, this,
+                             quota_rename_get_size_cbk);
+        if (ret) {
+            op_errno = -ret;
+            goto err;
+        }
 
-	if (priv->disk_usage_limit) {
-		local = GF_CALLOC (1, sizeof (struct quota_local),
-                                   gf_quota_mt_quota_local);
-		frame->local = local;
+        return;
+    }
 
-		loc_copy (&local->loc, loc);
-                local->flags = flags;
+    quota_check_limit(frame, local->newloc.parent, this);
+    return;
 
-		STACK_WIND (frame, quota_rmdir_stat_cbk,
-			    FIRST_CHILD(this),
-			    FIRST_CHILD(this)->fops->stat, loc);
-		return 0;
-	}
+err:
+    QUOTA_STACK_UNWIND(rename, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                       NULL, NULL);
+    return;
+}
 
-	STACK_WIND (frame, quota_rmdir_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->rmdir,
-		    loc, flags);
-	return 0;
+int32_t
+quota_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+             dict_t *xdata)
+{
+    quota_priv_t *priv = NULL;
+    int32_t ret = -1;
+    int32_t op_errno = ENOMEM;
+    quota_local_t *local = NULL;
+    call_stub_t *stub = NULL;
+
+    priv = this->private;
+
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+
+    local = quota_local_new();
+    if (local == NULL) {
+        goto err;
+    }
+
+    frame->local = local;
+
+    ret = loc_copy(&local->oldloc, oldloc);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "loc_copy failed");
+        goto err;
+    }
+
+    ret = loc_copy(&local->newloc, newloc);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "loc_copy failed");
+        goto err;
+    }
+
+    /* No need to check quota limit if src and dst parents are same */
+    if (oldloc->parent && newloc->parent &&
+        !gf_uuid_compare(oldloc->parent->gfid, newloc->parent->gfid)) {
+        gf_msg_debug(this->name, 0,
+                     "rename %s -> %s are "
+                     "in the same directory, so skip check limit",
+                     oldloc->path, newloc->path);
+        goto wind;
+    }
+
+    stub = fop_rename_stub(frame, quota_rename_helper, oldloc, newloc, xdata);
+    if (stub == NULL) {
+        goto err;
+    }
+
+    LOCK(&local->lock);
+    {
+        /* link_count here tell how many check_ancestry should be done
+         * before continuing the FOP
+         */
+        local->link_count = 2;
+        local->stub = stub;
+        local->fop_continue_cbk = quota_rename_continue;
+    }
+    UNLOCK(&local->lock);
+
+    check_ancestory(frame, newloc->parent);
+    check_ancestory(frame, oldloc->parent);
+    return 0;
+
+err:
+    QUOTA_STACK_UNWIND(rename, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                       NULL, NULL);
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->rename,
+                    oldloc, newloc, xdata);
+    return 0;
+
+wind:
+    STACK_WIND(frame, quota_rename_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
+    return 0;
 }
 
+int32_t
+quota_symlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, inode_t *inode,
+                  struct iatt *buf, struct iatt *preparent,
+                  struct iatt *postparent, dict_t *xdata)
+{
+    quota_local_t *local = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    quota_dentry_t *dentry = NULL;
+    int32_t ret = -1;
+
+    if (op_ret < 0) {
+        goto out;
+    }
+
+    local = frame->local;
+
+    ret = quota_inode_ctx_get(local->loc.inode, this, &ctx, 1);
+    if ((ret == -1) || (ctx == NULL)) {
+        gf_msg_debug(this->name, 0,
+                     "quota context is NULL on inode"
+                     " (%s). If quota is not enabled recently and "
+                     "crawler has finished crawling, its an error",
+                     uuid_utoa(local->loc.inode->gfid));
+
+        goto out;
+    }
+
+    LOCK(&ctx->lock);
+    {
+        ctx->buf = *buf;
+
+        dentry = __quota_dentry_new(ctx, (char *)local->loc.name,
+                                    local->loc.parent->gfid);
+        if (dentry == NULL) {
+            gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+                   "cannot create "
+                   "a new dentry (name:%s) for inode(gfid:%s)",
+                   local->loc.name, uuid_utoa(local->loc.inode->gfid));
+            op_ret = -1;
+            op_errno = ENOMEM;
+        }
+    }
+    UNLOCK(&ctx->lock);
+
+out:
+    QUOTA_STACK_UNWIND(symlink, frame, op_ret, op_errno, inode, buf, preparent,
+                       postparent, xdata);
+
+    return 0;
+}
 
 int
-quota_symlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		   int32_t op_ret, int32_t op_errno, inode_t *inode,
-                   struct iatt *buf, struct iatt *preparent,
-                   struct iatt *postparent)
+quota_symlink_helper(call_frame_t *frame, xlator_t *this, const char *linkpath,
+                     loc_t *loc, mode_t umask, dict_t *xdata)
 {
-	struct quota_priv *priv = NULL;
+    quota_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
 
-	priv = this->private;
+    local = frame->local;
 
-	if ((op_ret >= 0) && priv->disk_usage_limit) {
-		gf_quota_usage_add (this, buf->ia_blocks * 512);
-	}
+    GF_VALIDATE_OR_GOTO("quota", local, unwind);
 
-	STACK_UNWIND_STRICT (symlink, frame, op_ret, op_errno, inode, buf,
-                             preparent, postparent);
-	return 0;
-}
+    if (local->op_ret == -1) {
+        op_errno = local->op_errno;
+        goto unwind;
+    }
 
+    STACK_WIND(frame, quota_symlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->symlink, linkpath, loc, umask, xdata);
+    return 0;
+
+unwind:
+    QUOTA_STACK_UNWIND(symlink, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                       NULL);
+    return 0;
+}
 
 int
-quota_symlink (call_frame_t *frame, xlator_t *this,
-	       const char *linkpath, loc_t *loc, dict_t *params)
+quota_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+              loc_t *loc, mode_t umask, dict_t *xdata)
+{
+    quota_priv_t *priv = NULL;
+    int32_t ret = -1;
+    int32_t op_errno = ENOMEM;
+    quota_local_t *local = NULL;
+    call_stub_t *stub = NULL;
+
+    priv = this->private;
+
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+
+    local = quota_local_new();
+    if (local == NULL) {
+        goto err;
+    }
+
+    frame->local = local;
+
+    ret = loc_copy(&local->loc, loc);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "loc_copy failed");
+        goto err;
+    }
+
+    stub = fop_symlink_stub(frame, quota_symlink_helper, linkpath, loc, umask,
+                            xdata);
+    if (stub == NULL) {
+        goto err;
+    }
+
+    LOCK(&local->lock);
+    {
+        local->stub = stub;
+        local->delta = strlen(linkpath);
+        local->object_delta = 1;
+        local->link_count = 1;
+    }
+    UNLOCK(&local->lock);
+
+    quota_check_limit(frame, loc->parent, this);
+    return 0;
+
+err:
+    QUOTA_STACK_UNWIND(symlink, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                       NULL);
+
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->symlink,
+                    linkpath, loc, umask, xdata);
+    return 0;
+}
+
+int32_t
+quota_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                   struct iatt *postbuf, dict_t *xdata)
 {
-	struct quota_priv *priv = NULL;
+    quota_local_t *local = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+
+    if (op_ret < 0) {
+        goto out;
+    }
+
+    local = frame->local;
+
+    GF_VALIDATE_OR_GOTO("quota", local, out);
+
+    quota_inode_ctx_get(local->loc.inode, this, &ctx, 0);
+    if (ctx == NULL) {
+        gf_msg_debug(this->name, 0,
+                     "quota context is NULL on inode"
+                     " (%s). If quota is not enabled recently and "
+                     "crawler has finished crawling, its an error",
+                     uuid_utoa(local->loc.inode->gfid));
+        goto out;
+    }
+
+    LOCK(&ctx->lock);
+    {
+        ctx->buf = *postbuf;
+    }
+    UNLOCK(&ctx->lock);
+
+out:
+    QUOTA_STACK_UNWIND(truncate, frame, op_ret, op_errno, prebuf, postbuf,
+                       xdata);
+    return 0;
+}
+
+int32_t
+quota_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+               dict_t *xdata)
+{
+    quota_priv_t *priv = NULL;
+    int32_t ret = -1;
+    quota_local_t *local = NULL;
+
+    priv = this->private;
+
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+
+    local = quota_local_new();
+    if (local == NULL) {
+        goto err;
+    }
+
+    frame->local = local;
+
+    ret = loc_copy(&local->loc, loc);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "loc_copy failed");
+        goto err;
+    }
+
+    STACK_WIND(frame, quota_truncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+
+    return 0;
+
+err:
+    QUOTA_STACK_UNWIND(truncate, frame, -1, ENOMEM, NULL, NULL, NULL);
+
+    return 0;
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->truncate,
+                    loc, offset, xdata);
+    return 0;
+}
+
+int32_t
+quota_ftruncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                    struct iatt *postbuf, dict_t *xdata)
+{
+    quota_local_t *local = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+
+    if (op_ret < 0) {
+        goto out;
+    }
+
+    local = frame->local;
+
+    GF_VALIDATE_OR_GOTO("quota", local, out);
+
+    quota_inode_ctx_get(local->loc.inode, this, &ctx, 0);
+    if (ctx == NULL) {
+        gf_msg_debug(this->name, 0,
+                     "quota context is NULL on inode"
+                     " (%s). If quota is not enabled recently and "
+                     "crawler has finished crawling, its an error",
+                     uuid_utoa(local->loc.inode->gfid));
+        goto out;
+    }
+
+    LOCK(&ctx->lock);
+    {
+        ctx->buf = *postbuf;
+    }
+    UNLOCK(&ctx->lock);
+
+out:
+    QUOTA_STACK_UNWIND(ftruncate, frame, op_ret, op_errno, prebuf, postbuf,
+                       xdata);
+    return 0;
+}
+
+int32_t
+quota_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                dict_t *xdata)
+{
+    quota_priv_t *priv = NULL;
+    quota_local_t *local = NULL;
+
+    priv = this->private;
+
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+
+    local = quota_local_new();
+    if (local == NULL)
+        goto err;
+
+    frame->local = local;
+
+    local->loc.inode = inode_ref(fd->inode);
+
+    STACK_WIND(frame, quota_ftruncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+
+    return 0;
+err:
+    QUOTA_STACK_UNWIND(ftruncate, frame, -1, ENOMEM, NULL, NULL, NULL);
+
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+    return 0;
+}
 
-	priv = this->private;
+static int32_t
+quota_send_dir_limit_to_cli(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                            const char *name, const int namelen)
+{
+    int32_t ret = 0;
+    int dir_limit_len = 0;
+    char dir_limit[64] = {
+        0,
+    };
+    dict_t *dict = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    uint64_t value = 0;
+    quota_priv_t *priv = NULL;
+
+    priv = this->private;
+    if (!priv->is_quota_on) {
+        dir_limit_len = snprintf(dir_limit, sizeof(dir_limit),
+                                 "Quota is disabled please turn on");
+        goto dict_set;
+    }
+
+    ret = inode_ctx_get(inode, this, &value);
+    if (ret < 0)
+        goto out;
+
+    ctx = (quota_inode_ctx_t *)(unsigned long)value;
+    dir_limit_len = snprintf(dir_limit, sizeof(dir_limit),
+                             "%" PRId64 ",%" PRId64, ctx->size, ctx->hard_lim);
+
+dict_set:
+    dict = dict_new();
+    if (dict == NULL) {
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_set_nstrn(dict, (char *)name, namelen, dir_limit, dir_limit_len);
+    if (ret < 0)
+        goto out;
+
+    gf_msg_debug(this->name, 0, "str = %s", dir_limit);
+
+    QUOTA_STACK_UNWIND(getxattr, frame, 0, 0, dict, NULL);
+
+    ret = 0;
+
+out:
+    if (dict)
+        dict_unref(dict);
+    return ret;
+}
 
-	if (gf_quota_check_free_disk (this) == -1) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"min-free-disk limit (%u) crossed, current available is %u",
-			priv->min_free_disk_limit, priv->current_free_disk);
-		STACK_UNWIND_STRICT (symlink, frame, -1, ENOSPC, NULL, NULL,
-                                     NULL, NULL);
-		return 0;
-		
-	}
-        if (priv->current_disk_usage > priv->disk_usage_limit) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"Disk usage limit (%"PRIu64") crossed, current usage is %"PRIu64"",
-			priv->disk_usage_limit, priv->current_disk_usage);
-		STACK_UNWIND_STRICT (symlink, frame, -1, ENOSPC, NULL, NULL,
-                                     NULL, NULL);
-		return 0;
+int32_t
+quota_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+                dict_t *xdata)
+{
+    int32_t ret = 0;
+
+    if (name && strcasecmp(name, "trusted.limit.list") == 0) {
+        ret = quota_send_dir_limit_to_cli(frame, this, fd->inode,
+                                          "trusted.limit.list",
+                                          SLEN("trusted.limit.list"));
+        if (ret == 0) {
+            return 0;
         }
+    }
 
-	STACK_WIND (frame, quota_symlink_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->symlink,
-		    linkpath, loc, params);
-	return 0;
+    STACK_WIND(frame, default_fgetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fgetxattr, fd, name, xdata);
+    return 0;
 }
 
+int32_t
+quota_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               const char *name, dict_t *xdata)
+{
+    int32_t ret = 0;
+
+    if ((name != NULL) && strcasecmp(name, "trusted.limit.list") == 0) {
+        ret = quota_send_dir_limit_to_cli(frame, this, loc->inode,
+                                          "trusted.limit.list",
+                                          SLEN("trusted.limit.list"));
+        if (ret == 0)
+            return 0;
+    }
+
+    STACK_WIND(frame, default_getxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->getxattr, loc, name, xdata);
+    return 0;
+}
 
-int
-quota_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		  int32_t op_ret, int32_t op_errno,
-		  fd_t *fd, inode_t *inode, struct iatt *buf,
-                  struct iatt *preparent, struct iatt *postparent)
+int32_t
+quota_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, struct iatt *buf,
+               dict_t *xdata)
 {
-	struct quota_priv *priv = this->private;
+    quota_local_t *local = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+
+    if (op_ret < 0) {
+        goto out;
+    }
 
-	if ((op_ret >= 0) && priv->disk_usage_limit) {
-		gf_quota_usage_add (this, buf->ia_blocks * 512);
+    local = frame->local;
 
-		fd_ctx_set (fd, this, 1);
-	}
+    GF_VALIDATE_OR_GOTO("quota", local, out);
 
-	STACK_UNWIND_STRICT (create, frame, op_ret, op_errno, fd, inode, buf,
-                             preparent, postparent);
-	return 0;
+    quota_inode_ctx_get(local->loc.inode, this, &ctx, 0);
+    if (ctx == NULL) {
+        if (!IA_ISDIR(buf->ia_type)) {
+            gf_msg_debug(this->name, 0,
+                         "quota context is NULL on inode"
+                         " (%s). If quota is not enabled recently and "
+                         "crawler has finished crawling, its an error",
+                         uuid_utoa(local->loc.inode->gfid));
+        }
+
+        goto out;
+    }
+
+    if (buf) {
+        LOCK(&ctx->lock);
+        ctx->buf = *buf;
+        UNLOCK(&ctx->lock);
+    }
+
+out:
+    QUOTA_STACK_UNWIND(stat, frame, op_ret, op_errno, buf, xdata);
+    return 0;
 }
 
+int32_t
+quota_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    quota_priv_t *priv = NULL;
+    quota_local_t *local = NULL;
+    int32_t ret = -1;
+
+    priv = this->private;
 
-int
-quota_create (call_frame_t *frame, xlator_t *this,
-	      loc_t *loc, int32_t flags, mode_t mode, fd_t *fd, dict_t *params)
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+
+    local = quota_local_new();
+    if (local == NULL) {
+        goto unwind;
+    }
+
+    frame->local = local;
+    ret = loc_copy(&local->loc, loc);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "loc_copy failed");
+        goto unwind;
+    }
+
+    STACK_WIND(frame, quota_stat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->stat, loc, xdata);
+    return 0;
+
+unwind:
+    QUOTA_STACK_UNWIND(stat, frame, -1, ENOMEM, NULL, NULL);
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->stat,
+                    loc, xdata);
+    return 0;
+}
+
+int32_t
+quota_fstat_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                dict_t *xdata)
 {
-	struct quota_priv *priv = NULL;
+    quota_local_t *local = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+
+    if (op_ret < 0) {
+        goto out;
+    }
 
-	priv = this->private;
+    local = frame->local;
 
-	if (gf_quota_check_free_disk (this) == -1) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"min-free-disk limit (%u) crossed, current available is %u",
-			priv->min_free_disk_limit, priv->current_free_disk);
-		STACK_UNWIND_STRICT (create, frame, -1, ENOSPC, NULL, NULL, NULL,
-                                     NULL, NULL);
-		return 0;
-		
-	}
-        if (priv->current_disk_usage > priv->disk_usage_limit) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"Disk usage limit (%"PRIu64") crossed, current usage is %"PRIu64"",
-			priv->disk_usage_limit, priv->current_disk_usage);
-		STACK_UNWIND_STRICT (create, frame, -1, ENOSPC, NULL, NULL, NULL,
-                                     NULL, NULL);
-		return 0;
+    GF_VALIDATE_OR_GOTO("quota", local, out);
+
+    quota_inode_ctx_get(local->loc.inode, this, &ctx, 0);
+    if (ctx == NULL) {
+        if (!IA_ISDIR(buf->ia_type)) {
+            gf_msg_debug(this->name, 0,
+                         "quota context is NULL on inode"
+                         " (%s). If quota is not enabled recently and "
+                         "crawler has finished crawling, its an error",
+                         uuid_utoa(local->loc.inode->gfid));
         }
 
-	STACK_WIND (frame, quota_create_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->create,
-		    loc, flags, mode, fd, params);
-	return 0;
+        goto out;
+    }
+
+    if (buf) {
+        LOCK(&ctx->lock);
+        ctx->buf = *buf;
+        UNLOCK(&ctx->lock);
+    }
+
+out:
+    QUOTA_STACK_UNWIND(fstat, frame, op_ret, op_errno, buf, xdata);
+    return 0;
 }
 
+int32_t
+quota_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    quota_priv_t *priv = NULL;
+    quota_local_t *local = NULL;
 
-int
-quota_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		int32_t op_ret, int32_t op_errno, fd_t *fd)
+    priv = this->private;
+
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+
+    local = quota_local_new();
+    if (local == NULL) {
+        goto unwind;
+    }
+
+    frame->local = local;
+
+    local->loc.inode = inode_ref(fd->inode);
+
+    STACK_WIND(frame, quota_fstat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fstat, fd, xdata);
+    return 0;
+
+unwind:
+    QUOTA_STACK_UNWIND(fstat, frame, -1, ENOMEM, NULL, NULL);
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->fstat,
+                    fd, xdata);
+    return 0;
+}
+
+int32_t
+quota_readlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, const char *path,
+                   struct iatt *buf, dict_t *xdata)
 {
+    quota_local_t *local = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+
+    if (op_ret < 0) {
+        goto out;
+    }
+
+    local = frame->local;
+
+    GF_VALIDATE_OR_GOTO("quota", local, out);
+
+    quota_inode_ctx_get(local->loc.inode, this, &ctx, 0);
+    if (ctx == NULL) {
+        gf_msg_debug(this->name, 0,
+                     "quota context is NULL on inode"
+                     " (%s). If quota is not enabled recently and "
+                     "crawler has finished crawling, its an error",
+                     uuid_utoa(local->loc.inode->gfid));
+        goto out;
+    }
+
+    LOCK(&ctx->lock);
+    {
+        ctx->buf = *buf;
+    }
+    UNLOCK(&ctx->lock);
+
+out:
+    QUOTA_STACK_UNWIND(readlink, frame, op_ret, op_errno, path, buf, xdata);
+    return 0;
+}
+
+int32_t
+quota_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+               dict_t *xdata)
+{
+    quota_priv_t *priv = NULL;
+    quota_local_t *local = NULL;
+    int32_t ret = -1;
+
+    priv = this->private;
+
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
 
-	if (op_ret >= 0)
-		fd_ctx_set (fd, this, 1);
+    local = quota_local_new();
+    if (local == NULL) {
+        goto unwind;
+    }
 
-	STACK_UNWIND_STRICT (open, frame, op_ret, op_errno, fd);
-	return 0;
+    frame->local = local;
+
+    ret = loc_copy(&local->loc, loc);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "loc_copy failed");
+        goto unwind;
+    }
+
+    STACK_WIND(frame, quota_readlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readlink, loc, size, xdata);
+    return 0;
+
+unwind:
+    QUOTA_STACK_UNWIND(readlink, frame, -1, ENOMEM, NULL, NULL, NULL);
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->readlink,
+                    loc, size, xdata);
+    return 0;
 }
 
+int32_t
+quota_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iovec *vector,
+                int32_t count, struct iatt *buf, struct iobref *iobref,
+                dict_t *xdata)
+{
+    quota_local_t *local = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+
+    if (op_ret < 0) {
+        goto out;
+    }
+
+    local = frame->local;
+
+    GF_VALIDATE_OR_GOTO("quota", local, out);
+
+    quota_inode_ctx_get(local->loc.inode, this, &ctx, 0);
+    if (ctx == NULL) {
+        gf_msg_debug(this->name, 0,
+                     "quota context is NULL on inode"
+                     " (%s). If quota is not enabled recently and "
+                     "crawler has finished crawling, its an error",
+                     uuid_utoa(local->loc.inode->gfid));
+        goto out;
+    }
+
+    LOCK(&ctx->lock);
+    {
+        ctx->buf = *buf;
+    }
+    UNLOCK(&ctx->lock);
+
+out:
+    QUOTA_STACK_UNWIND(readv, frame, op_ret, op_errno, vector, count, buf,
+                       iobref, xdata);
+    return 0;
+}
 
-int
-quota_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
-            fd_t *fd, int32_t wbflags)
+int32_t
+quota_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t offset, uint32_t flags, dict_t *xdata)
 {
-	STACK_WIND (frame, quota_open_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->open,
-		    loc, flags, fd, wbflags);
-	return 0;
+    quota_priv_t *priv = NULL;
+    quota_local_t *local = NULL;
+
+    priv = this->private;
+
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+
+    local = quota_local_new();
+    if (local == NULL) {
+        goto unwind;
+    }
+
+    frame->local = local;
+
+    local->loc.inode = inode_ref(fd->inode);
+
+    STACK_WIND(frame, quota_readv_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readv, fd, size, offset, flags, xdata);
+    return 0;
+
+unwind:
+    QUOTA_STACK_UNWIND(readv, frame, -1, ENOMEM, NULL, -1, NULL, NULL, NULL);
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->readv,
+                    fd, size, offset, flags, xdata);
+    return 0;
 }
 
+int32_t
+quota_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                struct iatt *postbuf, dict_t *xdata)
+{
+    quota_local_t *local = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+
+    if (op_ret < 0) {
+        goto out;
+    }
+
+    local = frame->local;
+
+    GF_VALIDATE_OR_GOTO("quota", local, out);
+
+    quota_inode_ctx_get(local->loc.inode, this, &ctx, 0);
+    if (ctx == NULL) {
+        gf_msg_debug(this->name, 0,
+                     "quota context is NULL on inode"
+                     " (%s). If quota is not enabled recently and "
+                     "crawler has finished crawling, its an error",
+                     uuid_utoa(local->loc.inode->gfid));
+        goto out;
+    }
+
+    LOCK(&ctx->lock);
+    {
+        ctx->buf = *postbuf;
+    }
+    UNLOCK(&ctx->lock);
+
+out:
+    QUOTA_STACK_UNWIND(fsync, frame, op_ret, op_errno, prebuf, postbuf, xdata);
+    return 0;
+}
 
-int
-quota_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		  int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
-                  struct iatt *postbuf)
+int32_t
+quota_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+            dict_t *xdata)
 {
-	struct quota_priv *priv = NULL;
-	struct quota_local *local = NULL;
+    quota_priv_t *priv = NULL;
+    quota_local_t *local = NULL;
+
+    priv = this->private;
 
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
 
-	priv = this->private;
-	local = frame->local;
+    local = quota_local_new();
+    if (local == NULL) {
+        goto unwind;
+    }
 
-	if (priv->disk_usage_limit) {
-		if (op_ret >= 0) { 
-			gf_quota_usage_add (this, (postbuf->ia_blocks -
-						   prebuf->ia_blocks) * 512);
-		}
-		fd_unref (local->fd);
-		iobref_unref (local->iobref);
-	}
+    local->loc.inode = inode_ref(fd->inode);
 
-	STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, prebuf, postbuf);
-	return 0;
+    frame->local = local;
+
+    STACK_WIND(frame, quota_fsync_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsync, fd, flags, xdata);
+    return 0;
+
+unwind:
+    QUOTA_STACK_UNWIND(fsync, frame, -1, ENOMEM, NULL, NULL, NULL);
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->fsync,
+                    fd, flags, xdata);
+    return 0;
 }
 
+int32_t
+quota_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, struct iatt *statpre,
+                  struct iatt *statpost, dict_t *xdata)
+{
+    quota_local_t *local = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+
+    if (op_ret < 0) {
+        goto out;
+    }
 
-int
-quota_writev_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-			int32_t op_ret,	int32_t op_errno, struct iatt *buf)
+    local = frame->local;
+
+    GF_VALIDATE_OR_GOTO("quota", local, out);
+
+    quota_inode_ctx_get(local->loc.inode, this, &ctx, 0);
+    if (ctx == NULL) {
+        if (!IA_ISDIR(statpost->ia_type)) {
+            gf_msg_debug(this->name, 0,
+                         "quota context is NULL on inode"
+                         " (%s). If quota is not enabled recently and "
+                         "crawler has finished crawling, its an error",
+                         uuid_utoa(local->loc.inode->gfid));
+        }
+
+        goto out;
+    }
+
+    if (statpost) {
+        LOCK(&ctx->lock);
+        ctx->buf = *statpost;
+        UNLOCK(&ctx->lock);
+    }
+
+out:
+    QUOTA_STACK_UNWIND(setattr, frame, op_ret, op_errno, statpre, statpost,
+                       xdata);
+    return 0;
+}
+
+int32_t
+quota_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+              struct iatt *stbuf, int32_t valid, dict_t *xdata)
 {
-	struct quota_local *local = NULL;
-	struct quota_priv  *priv = NULL;
-	int                 iovlen = 0;
+    quota_priv_t *priv = NULL;
+    quota_local_t *local = NULL;
+    int32_t ret = -1;
+
+    priv = this->private;
+
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
 
+    local = quota_local_new();
+    if (local == NULL) {
+        goto unwind;
+    }
 
-	local = frame->local;
-	priv = this->private;
+    frame->local = local;
 
-	if (op_ret >= 0) {
-		if (priv->current_disk_usage > priv->disk_usage_limit) {
-			iovlen = iov_length (local->vector, local->count);
+    ret = loc_copy(&local->loc, loc);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "loc_copy failed");
+        goto unwind;
+    }
 
-			if (iovlen > (buf->ia_blksize - (buf->ia_size % buf->ia_blksize))) {
-				fd_unref (local->fd);
-				iobref_unref (local->iobref);
-				STACK_UNWIND_STRICT (writev, frame, -1, ENOSPC,
-                                                     NULL, NULL);
-				return 0;
-			}
-		}
-		local->stbuf = *buf;
-	}
-	
-	STACK_WIND (frame, quota_writev_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->writev,
-		    local->fd, local->vector, local->count, local->offset,
-                    local->iobref);
+    STACK_WIND(frame, quota_setattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setattr, loc, stbuf, valid, xdata);
+    return 0;
+
+unwind:
+    QUOTA_STACK_UNWIND(setattr, frame, -1, ENOMEM, NULL, NULL, NULL);
+    return 0;
 
-	return 0;
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->setattr,
+                    loc, stbuf, valid, xdata);
+    return 0;
 }
 
+int32_t
+quota_fsetattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iatt *statpre,
+                   struct iatt *statpost, dict_t *xdata)
+{
+    quota_local_t *local = NULL;
+    quota_inode_ctx_t *ctx = NULL;
 
-int
-quota_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
-	      struct iovec *vector, int32_t count, off_t off,
-              struct iobref *iobref)
-{
-	struct quota_local *local = NULL;
-	struct quota_priv  *priv  = NULL;
-        int                 i     = 0;
-
-	priv = this->private;
-
-	if (gf_quota_check_free_disk (this) == -1) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"min-free-disk limit (%u) crossed, current available is %u",
-			priv->min_free_disk_limit, priv->current_free_disk);
-		STACK_UNWIND_STRICT (writev, frame, -1, ENOSPC,
-                                     NULL, NULL);
-		return 0;
-	}
-
-	if (priv->disk_usage_limit) {
-		local = GF_CALLOC (1, sizeof (struct quota_local),
-                                   gf_quota_mt_quota_local);
-		local->fd     = fd_ref (fd);
-		local->iobref = iobref_ref (iobref);
-                for (i = 0; i < count; i++) {
-                        local->vector[i].iov_base = vector[i].iov_base;
-                        local->vector[i].iov_len = vector[i].iov_len;
-                }
+    if (op_ret < 0) {
+        goto out;
+    }
 
-                local->count = count;
-		local->offset = off;
-		frame->local  = local;
+    local = frame->local;
 
-		STACK_WIND (frame, quota_writev_fstat_cbk,
-			    FIRST_CHILD(this),
-			    FIRST_CHILD(this)->fops->fstat, fd);
-		return 0;
-	}
+    GF_VALIDATE_OR_GOTO("quota", local, out);
 
-	STACK_WIND (frame, quota_writev_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->writev,
-		    fd, vector, count, off, iobref);
-	return 0;
+    quota_inode_ctx_get(local->loc.inode, this, &ctx, 0);
+    if (ctx == NULL) {
+        if (!IA_ISDIR(statpost->ia_type)) {
+            gf_msg_debug(this->name, 0,
+                         "quota context is NULL on inode"
+                         " (%s). If quota is not enabled recently and "
+                         "crawler has finished crawling, its an error",
+                         uuid_utoa(local->loc.inode->gfid));
+        }
+
+        goto out;
+    }
+
+    LOCK(&ctx->lock);
+    {
+        ctx->buf = *statpost;
+    }
+    UNLOCK(&ctx->lock);
+
+out:
+    QUOTA_STACK_UNWIND(fsetattr, frame, op_ret, op_errno, statpre, statpost,
+                       xdata);
+    return 0;
 }
 
+int32_t
+quota_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+               struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    quota_priv_t *priv = NULL;
+    quota_local_t *local = NULL;
+
+    priv = this->private;
 
-int
-quota_removexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		       int32_t op_ret, int32_t op_errno)
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+
+    local = quota_local_new();
+    if (local == NULL) {
+        goto unwind;
+    }
+
+    frame->local = local;
+
+    local->loc.inode = inode_ref(fd->inode);
+
+    STACK_WIND(frame, quota_fsetattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetattr, fd, stbuf, valid, xdata);
+    return 0;
+
+unwind:
+    QUOTA_STACK_UNWIND(fsetattr, frame, -1, ENOMEM, NULL, NULL, NULL);
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->fsetattr,
+                    fd, stbuf, valid, xdata);
+    return 0;
+}
+
+int32_t
+quota_mknod_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, inode_t *inode,
+                struct iatt *buf, struct iatt *preparent,
+                struct iatt *postparent, dict_t *xdata)
 {
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_CRITICAL, 
-			"failed to remove the disk-usage value: %s",
-			strerror (op_errno));
-	} 
-	
-	STACK_DESTROY (frame->root);
-	return 0;
+    int32_t ret = -1;
+    quota_local_t *local = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    quota_dentry_t *dentry = NULL;
+
+    local = frame->local;
+    if (op_ret < 0) {
+        goto unwind;
+    }
+
+    ret = quota_inode_ctx_get(inode, this, &ctx, 1);
+    if ((ret == -1) || (ctx == NULL)) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, Q_MSG_INODE_CTX_GET_FAILED,
+               "cannot create quota context in "
+               "inode(gfid:%s)",
+               uuid_utoa(inode->gfid));
+        op_ret = -1;
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+
+    LOCK(&ctx->lock);
+    {
+        ctx->buf = *buf;
+
+        dentry = __quota_dentry_new(ctx, (char *)local->loc.name,
+                                    local->loc.parent->gfid);
+        if (dentry == NULL) {
+            gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+                   "cannot create a new dentry "
+                   "(name:%s) for inode(gfid:%s)",
+                   local->loc.name, uuid_utoa(local->loc.inode->gfid));
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto unlock;
+        }
+    }
+unlock:
+    UNLOCK(&ctx->lock);
+
+unwind:
+    QUOTA_STACK_UNWIND(mknod, frame, op_ret, op_errno, inode, buf, preparent,
+                       postparent, xdata);
+    return 0;
 }
 
+int
+quota_mknod_helper(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+                   dev_t rdev, mode_t umask, dict_t *xdata)
+{
+    quota_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+
+    local = frame->local;
+
+    GF_VALIDATE_OR_GOTO("quota", local, unwind);
+
+    if (local->op_ret == -1) {
+        op_errno = local->op_errno;
+        goto unwind;
+    }
+
+    STACK_WIND(frame, quota_mknod_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mknod, loc, mode, rdev, umask, xdata);
+
+    return 0;
+
+unwind:
+    QUOTA_STACK_UNWIND(mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                       NULL);
+    return 0;
+}
 
 int
-quota_setxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		    int32_t op_ret, int32_t op_errno)
+quota_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+            dev_t rdev, mode_t umask, dict_t *xdata)
 {
-        dict_t            *dict = NULL;
+    quota_priv_t *priv = NULL;
+    int32_t ret = -1;
+    quota_local_t *local = NULL;
+    call_stub_t *stub = NULL;
+
+    priv = this->private;
+
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+    QUOTA_WIND_FOR_INTERNAL_FOP(xdata, off);
+
+    local = quota_local_new();
+    if (local == NULL) {
+        goto err;
+    }
+
+    frame->local = local;
+
+    ret = loc_copy(&local->loc, loc);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "loc_copy failed");
+        goto err;
+    }
+
+    stub = fop_mknod_stub(frame, quota_mknod_helper, loc, mode, rdev, umask,
+                          xdata);
+    if (stub == NULL) {
+        goto err;
+    }
+
+    LOCK(&local->lock);
+    {
+        local->link_count = 1;
+        local->stub = stub;
+        local->delta = 0;
+        local->object_delta = 1;
+    }
+    UNLOCK(&local->lock);
+
+    quota_check_limit(frame, loc->parent, this);
+    return 0;
+
+err:
+    QUOTA_STACK_UNWIND(mknod, frame, -1, ENOMEM, NULL, NULL, NULL, NULL, NULL);
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->mknod,
+                    loc, mode, rdev, umask, xdata);
+    return 0;
+}
 
-	if (op_ret == -1) {
-		gf_log (this->name, GF_LOG_CRITICAL, 
-			"failed to set the disk-usage value: %s",
-			strerror (op_errno));
-	} 
+int
+quota_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int op_ret, int op_errno, dict_t *xdata)
+{
+    quota_local_t *local = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    int ret = 0;
+
+    if (op_ret < 0) {
+        goto out;
+    }
+
+    local = frame->local;
+    if (!local)
+        goto out;
+
+    ret = quota_inode_ctx_get(local->loc.inode, this, &ctx, 1);
+    if ((ret < 0) || (ctx == NULL)) {
+        op_errno = -1;
+        goto out;
+    }
+
+    LOCK(&ctx->lock);
+    {
+        ctx->hard_lim = local->limit.hl;
+        ctx->soft_lim = local->limit.sl;
+        ctx->object_hard_lim = local->object_limit.hl;
+        ctx->object_soft_lim = local->object_limit.sl;
+    }
+    UNLOCK(&ctx->lock);
+
+out:
+    QUOTA_STACK_UNWIND(setxattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
-        if (cookie) {
-                dict = (dict_t *) cookie;
-                dict_unref (dict);
+int
+quota_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+               int flags, dict_t *xdata)
+{
+    quota_priv_t *priv = NULL;
+    int op_errno = EINVAL;
+    int op_ret = -1;
+    int64_t hard_lim = -1;
+    int64_t soft_lim = -1;
+    int64_t object_hard_limit = -1;
+    int64_t object_soft_limit = -1;
+    quota_local_t *local = NULL;
+    gf_boolean_t internal_fop = _gf_false;
+
+    priv = this->private;
+
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+
+    if (xdata && dict_get_sizen(xdata, GLUSTERFS_INTERNAL_FOP_KEY))
+        internal_fop = _gf_true;
+
+    if (frame->root->pid >= 0 && internal_fop == _gf_false) {
+        GF_IF_INTERNAL_XATTR_GOTO("trusted.glusterfs.quota*", dict, op_errno,
+                                  err);
+        GF_IF_INTERNAL_XATTR_GOTO("trusted.pgfid*", dict, op_errno, err);
+    }
+
+    quota_get_limits(this, dict, &hard_lim, &soft_lim, &object_hard_limit,
+                     &object_soft_limit);
+
+    if (hard_lim > 0 || object_hard_limit > 0) {
+        local = quota_local_new();
+        if (local == NULL) {
+            op_errno = ENOMEM;
+            goto err;
         }
+        frame->local = local;
+        loc_copy(&local->loc, loc);
+    }
+
+    if (hard_lim > 0) {
+        local->limit.hl = hard_lim;
+        local->limit.sl = soft_lim;
+    }
+
+    if (object_hard_limit > 0) {
+        local->object_limit.hl = object_hard_limit;
+        local->object_limit.sl = object_soft_limit;
+    }
+
+    STACK_WIND(frame, quota_setxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setxattr, loc, dict, flags, xdata);
+    return 0;
+err:
+    QUOTA_STACK_UNWIND(setxattr, frame, op_ret, op_errno, NULL);
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->setxattr,
+                    loc, dict, flags, xdata);
+    return 0;
+}
 
-	STACK_DESTROY (frame->root);
-	return 0;
+int
+quota_fsetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int op_ret, int op_errno, dict_t *xdata)
+{
+    quota_inode_ctx_t *ctx = NULL;
+    quota_local_t *local = NULL;
+
+    if (op_ret < 0)
+        goto out;
+
+    local = frame->local;
+    if (!local)
+        goto out;
+
+    op_ret = quota_inode_ctx_get(local->loc.inode, this, &ctx, 1);
+    if ((op_ret < 0) || (ctx == NULL)) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    LOCK(&ctx->lock);
+    {
+        ctx->hard_lim = local->limit.hl;
+        ctx->soft_lim = local->limit.sl;
+        ctx->object_hard_lim = local->object_limit.hl;
+        ctx->object_soft_lim = local->object_limit.sl;
+    }
+    UNLOCK(&ctx->lock);
+
+out:
+    QUOTA_STACK_UNWIND(fsetxattr, frame, op_ret, op_errno, xdata);
+    return 0;
 }
 
+int
+quota_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+                int flags, dict_t *xdata)
+{
+    quota_priv_t *priv = NULL;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    quota_local_t *local = NULL;
+    int64_t hard_lim = -1;
+    int64_t soft_lim = -1;
+    int64_t object_hard_limit = -1;
+    int64_t object_soft_limit = -1;
+
+    priv = this->private;
+
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    if (0 <= frame->root->pid) {
+        GF_IF_INTERNAL_XATTR_GOTO("trusted.glusterfs.quota*", dict, op_errno,
+                                  err);
+        GF_IF_INTERNAL_XATTR_GOTO("trusted.pgfid*", dict, op_errno, err);
+    }
+
+    quota_get_limits(this, dict, &hard_lim, &soft_lim, &object_hard_limit,
+                     &object_soft_limit);
+
+    if (hard_lim > 0 || object_hard_limit > 0) {
+        local = quota_local_new();
+        if (local == NULL) {
+            op_errno = ENOMEM;
+            goto err;
+        }
+        frame->local = local;
+        local->loc.inode = inode_ref(fd->inode);
+    }
+
+    if (hard_lim > 0) {
+        local->limit.hl = hard_lim;
+        local->limit.sl = soft_lim;
+    }
+
+    if (object_hard_limit > 0) {
+        local->object_limit.hl = object_hard_limit;
+        local->object_limit.sl = object_soft_limit;
+    }
+
+    STACK_WIND(frame, quota_fsetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
+    return 0;
+err:
+    QUOTA_STACK_UNWIND(fsetxattr, frame, op_ret, op_errno, NULL);
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
+    return 0;
+}
+
+int
+quota_removexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    QUOTA_STACK_UNWIND(removexattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
 int
-quota_statfs_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		  int32_t op_ret, int32_t op_errno, struct statvfs *statvfs)
+quota_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                  const char *name, dict_t *xdata)
 {
-	struct quota_priv *priv = NULL;
-	uint64_t           f_blocks = 0;
-	int64_t            f_bfree = 0;
-	uint64_t           f_bused = 0;
+    quota_priv_t *priv = NULL;
+    int32_t op_errno = EINVAL;
 
+    priv = this->private;
 
-	priv = this->private;
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
 
-	if (op_ret != 0)
-		goto unwind;
+    VALIDATE_OR_GOTO(this, err);
 
-	f_blocks = priv->disk_usage_limit / statvfs->f_frsize;
-	f_bused = priv->current_disk_usage / statvfs->f_frsize;
+    /* all quota xattrs can be cleaned up by doing setxattr on special key.
+     * Hence its ok that we don't allow removexattr on quota keys here.
+     */
+    if (frame->root->pid >= 0) {
+        GF_IF_NATIVE_XATTR_GOTO("trusted.glusterfs.quota*", name, op_errno,
+                                err);
+        GF_IF_NATIVE_XATTR_GOTO("trusted.pgfid*", name, op_errno, err);
+    }
 
-	if (f_blocks && (f_blocks < statvfs->f_blocks))
-		statvfs->f_blocks = f_blocks;
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(loc, err);
 
-	f_bfree = (statvfs->f_blocks - f_bused);
+    STACK_WIND(frame, quota_removexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->removexattr, loc, name, xdata);
+    return 0;
 
-	if (f_bfree >= 0)
-		statvfs->f_bfree = statvfs->f_bavail = f_bfree;
-	else
-		statvfs->f_bfree = statvfs->f_bavail = 0;
+err:
+    QUOTA_STACK_UNWIND(removexattr, frame, -1, op_errno, NULL);
+    return 0;
 
-unwind:
-	STACK_UNWIND_STRICT (statfs, frame, op_ret, op_errno, statvfs);
-	return 0;
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->removexattr, loc, name, xdata);
+    return 0;
 }
 
-
 int
-quota_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc)
+quota_fremovexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *xdata)
 {
-	STACK_WIND (frame, quota_statfs_cbk,
-		    FIRST_CHILD (this), FIRST_CHILD (this)->fops->statfs, loc);
+    QUOTA_STACK_UNWIND(fremovexattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
 
-	return 0;
+int
+quota_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                   const char *name, dict_t *xdata)
+{
+    quota_priv_t *priv = NULL;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+
+    priv = this->private;
+
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(fd, err);
+
+    if (frame->root->pid >= 0) {
+        GF_IF_NATIVE_XATTR_GOTO("trusted.glusterfs.quota*", name, op_errno,
+                                err);
+        GF_IF_NATIVE_XATTR_GOTO("trusted.pgfid*", name, op_errno, err);
+    }
+    STACK_WIND(frame, quota_fremovexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fremovexattr, fd, name, xdata);
+    return 0;
+err:
+    QUOTA_STACK_UNWIND(fremovexattr, frame, op_ret, op_errno, NULL);
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->fremovexattr, fd, name, xdata);
+    return 0;
 }
 
+int32_t
+quota_statfs_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct statvfs *buf,
+                 dict_t *xdata)
+{
+    inode_t *inode = NULL;
+    uint64_t value = 0;
+    int64_t usage = -1;
+    int64_t avail = -1;
+    int64_t blocks = 0;
+    quota_inode_ctx_t *ctx = NULL;
+    int ret = 0;
+
+    inode = cookie;
+
+    /* This fop will fail mostly in case of client disconnect,
+     * which is already logged. Hence, not logging here */
+    if (op_ret == -1)
+        goto unwind;
+    /*
+     * We should never get here unless quota_statfs (below) sent us a
+     * cookie, and it would only do so if the value was non-NULL.  This
+     * check is therefore just routine defensive coding.
+     */
+
+    GF_VALIDATE_OR_GOTO("quota", inode, unwind);
+
+    inode_ctx_get(inode, this, &value);
+    ctx = (quota_inode_ctx_t *)(unsigned long)value;
+    if (!ctx || ctx->hard_lim <= 0)
+        goto unwind;
+
+    { /* statfs is adjusted in this code block */
+        usage = (ctx->size) / buf->f_bsize;
+
+        blocks = ctx->hard_lim / buf->f_bsize;
+        buf->f_blocks = blocks;
+
+        avail = buf->f_blocks - usage;
+        avail = max(avail, 0);
+
+        buf->f_bfree = avail;
+        /*
+         * We have to assume that the total assigned quota
+         * won't cause us to dip into the reserved space,
+         * because dealing with the overcommitted cases is
+         * just too hairy (especially when different bricks
+         * might be using different reserved percentages and
+         * such).
+         */
+        buf->f_bavail = buf->f_bfree;
+    }
+
+    xdata = xdata ? dict_ref(xdata) : dict_new();
+    if (!xdata)
+        goto unwind;
+
+    ret = dict_set_int8(xdata, "quota-deem-statfs", 1);
+    if (-1 == ret)
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, Q_MSG_ENOMEM,
+               "Dict set failed, deem-statfs option may "
+               "have no effect");
 
-int
-quota_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-		    int32_t op_ret, int32_t op_errno, dict_t *value)
+unwind:
+    QUOTA_STACK_UNWIND(statfs, frame, op_ret, op_errno, buf, xdata);
+
+    if (xdata)
+        dict_unref(xdata);
+
+    return 0;
+}
+
+int32_t
+quota_statfs_helper(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                    dict_t *xdata)
 {
-	data_t *data = NULL;
-	struct quota_priv *priv = this->private;
-	
-	if (op_ret >= 0) {
-		data = dict_get (value, "trusted.glusterfs-quota-du");
-		if (data) {
-			LOCK (&priv->lock);
-			{
-				priv->current_disk_usage = data_to_uint64 (data);
-			}
-			UNLOCK (&priv->lock);
+    quota_local_t *local = frame->local;
+    int op_errno = EINVAL;
 
-			return 0;
-		}
-	} 
+    GF_VALIDATE_OR_GOTO("quota", local, err);
 
-	STACK_DESTROY (frame->root);
+    if (-1 == local->op_ret) {
+        op_errno = local->op_errno;
+        goto err;
+    }
 
-	return 0;
+    STACK_WIND_COOKIE(frame, quota_statfs_cbk, local->inode, FIRST_CHILD(this),
+                      FIRST_CHILD(this)->fops->statfs, loc, xdata);
+    return 0;
+err:
+    QUOTA_STACK_UNWIND(statfs, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
 }
 
+int32_t
+quota_statfs_validate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, inode_t *inode,
+                          struct iatt *buf, dict_t *xdata,
+                          struct iatt *postparent)
+{
+    quota_local_t *local = NULL;
+    int32_t ret = 0;
+    quota_inode_ctx_t *ctx = NULL;
+    uint64_t value = 0;
+    quota_meta_t size = {
+        0,
+    };
+
+    local = frame->local;
+
+    if (op_ret < 0)
+        goto resume;
+
+    GF_ASSERT(local);
+    GF_ASSERT(frame);
+    GF_VALIDATE_OR_GOTO_WITH_ERROR("quota", this, resume, op_errno, EINVAL);
+    GF_VALIDATE_OR_GOTO_WITH_ERROR(this->name, xdata, resume, op_errno, EINVAL);
+
+    ret = inode_ctx_get(local->validate_loc.inode, this, &value);
+
+    ctx = (quota_inode_ctx_t *)(unsigned long)value;
+    if ((ret == -1) || (ctx == NULL)) {
+        gf_msg(this->name, GF_LOG_WARNING, EINVAL, Q_MSG_INODE_CTX_GET_FAILED,
+               "quota context is not present in inode (gfid:%s)",
+               uuid_utoa(local->validate_loc.inode->gfid));
+        op_errno = EINVAL;
+        goto resume;
+    }
+
+    ret = quota_dict_get_meta(xdata, QUOTA_SIZE_KEY, SLEN(QUOTA_SIZE_KEY),
+                              &size);
+    if (ret == -1) {
+        gf_msg(this->name, GF_LOG_WARNING, EINVAL, Q_MSG_SIZE_KEY_MISSING,
+               "size key not present in "
+               "dict");
+        op_errno = EINVAL;
+    }
+
+    LOCK(&ctx->lock);
+    {
+        ctx->size = size.size;
+        ctx->validate_time = gf_time();
+        ctx->file_count = size.file_count;
+        ctx->dir_count = size.dir_count;
+    }
+    UNLOCK(&ctx->lock);
+
+resume:
+    local->op_errno = op_errno;
+    quota_link_count_decrement(frame);
+    return 0;
+}
 
 void
-gf_quota_get_disk_usage (xlator_t *this)
+quota_get_limit_dir_continuation(struct list_head *parents, inode_t *inode,
+                                 int32_t op_ret, int32_t op_errno, void *data)
 {
-	call_frame_t *frame = NULL;
-	call_pool_t  *pool = NULL;
+    call_frame_t *frame = NULL;
+    xlator_t *this = NULL;
+    quota_dentry_t *entry = NULL;
+    inode_t *parent = NULL;
+
+    frame = data;
+    this = THIS;
+
+    if ((op_ret < 0) || list_empty(parents)) {
+        if (op_ret >= 0) {
+            gf_msg(this->name, GF_LOG_WARNING, EIO, Q_MSG_ANCESTRY_BUILD_FAILED,
+                   "Couldn't build ancestry for inode (gfid:%s). "
+                   "Without knowing ancestors till root, quota "
+                   "cannot be enforced. "
+                   "Hence, failing fop with EIO",
+                   uuid_utoa(inode->gfid));
+            op_errno = EIO;
+        }
+
+        quota_handle_validate_error(frame, -1, op_errno);
+        goto out;
+    }
 
-	struct quota_priv *priv = NULL;
+    entry = list_entry(parents, quota_dentry_t, next);
+    parent = inode_find(inode->table, entry->par);
 
-	pool = this->ctx->pool;
-	frame = create_frame (this, pool);
-	priv = this->private;
+    quota_get_limit_dir(frame, parent, this);
 
-	STACK_WIND (frame, quota_getxattr_cbk,
-		    this->children->xlator,
-		    this->children->xlator->fops->getxattr,
-		    &(priv->root_loc),
-		    "trusted.glusterfs-quota-du");
-	return ;
+    inode_unref(parent);
+out:
+    return;
 }
 
+void
+quota_statfs_continue(call_frame_t *frame, xlator_t *this, inode_t *inode)
+{
+    quota_local_t *local = frame->local;
+    int ret = -1;
+
+    LOCK(&local->lock);
+    {
+        local->inode = inode_ref(inode);
+    }
+    UNLOCK(&local->lock);
+
+    ret = quota_validate(frame, local->inode, this, quota_statfs_validate_cbk);
+    if (0 > ret)
+        quota_handle_validate_error(frame, -1, -ret);
+}
 
 void
-gf_quota_cache_sync (xlator_t *this)
+quota_get_limit_dir(call_frame_t *frame, inode_t *cur_inode, xlator_t *this)
 {
-	struct quota_priv *priv = NULL;
-	call_frame_t      *frame = NULL;
-	dict_t            *dict = get_new_dict ();
+    inode_t *inode = NULL;
+    inode_t *parent = NULL;
+    uint64_t value = 0;
+    quota_inode_ctx_t *ctx = NULL;
+    quota_local_t *local = frame->local;
+
+    if (!cur_inode)
+        goto out;
+
+    inode = inode_ref(cur_inode);
+    while (inode) {
+        value = 0;
+        inode_ctx_get(inode, this, &value);
+
+        if (value) {
+            ctx = (quota_inode_ctx_t *)(unsigned long)value;
+            if (ctx->hard_lim > 0)
+                break;
+        }
 
+        if (__is_root_gfid(inode->gfid))
+            goto off;
 
+        parent = inode_parent(inode, 0, NULL);
+        if (!parent) {
+            (void)quota_build_ancestry(inode, quota_get_limit_dir_continuation,
+                                       frame);
+            goto out;
+        }
 
-	priv = this->private;
+        inode_unref(inode);
+        inode = parent;
+    }
 
-	frame = create_frame (this, this->ctx->pool);
-	dict_set (dict, "trusted.glusterfs-quota-du", 
-		  data_from_uint64 (priv->current_disk_usage));
+    quota_statfs_continue(frame, this, inode);
+    inode_unref(inode);
+    return;
 
-        dict_ref (dict);
+off:
+    gf_msg_debug(this->name, 0, "No limit set on the inode or it's parents.");
 
-	STACK_WIND_COOKIE (frame, quota_setxattr_cbk,
-                           (void *) (dict_t *) dict,
-                           this->children->xlator,
-                           this->children->xlator->fops->setxattr,
-                           &(priv->root_loc), dict, 0);
+    QUOTA_STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                          FIRST_CHILD(this)->fops->statfs, &local->loc,
+                          local->xdata);
+out:
+    inode_unref(inode);
+
+    return;
 }
 
+int32_t
+quota_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    int op_errno = 0;
+    int ret = -1;
+    int8_t ignore_deem_statfs = 0;
+    quota_priv_t *priv = NULL;
+    quota_local_t *local = NULL;
+    call_stub_t *stub = NULL;
+
+    priv = this->private;
+    GF_ASSERT(loc);
+
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+
+    ret = dict_get_int8(xdata, GF_INTERNAL_IGNORE_DEEM_STATFS,
+                        &ignore_deem_statfs);
+    ret = 0;
+
+    if (ignore_deem_statfs)
+        goto off;
+
+    if (priv->consider_statfs && loc->inode) {
+        local = quota_local_new();
+        if (!local) {
+            op_errno = ENOMEM;
+            goto err;
+        }
+        frame->local = local;
+
+        ret = loc_copy(&local->loc, loc);
+        if (-1 == ret) {
+            op_errno = ENOMEM;
+            goto err;
+        }
+
+        if (xdata)
+            local->xdata = dict_ref(xdata);
+
+        stub = fop_statfs_stub(frame, quota_statfs_helper, &local->loc,
+                               local->xdata);
+        if (!stub) {
+            op_errno = ENOMEM;
+            goto err;
+        }
+
+        LOCK(&local->lock);
+        {
+            local->link_count = 1;
+            local->stub = stub;
+        }
+        UNLOCK(&local->lock);
+
+        quota_get_limit_dir(frame, loc->inode, this);
+
+        return 0;
+    }
+
+    /*
+     * We have to make sure that we never get to quota_statfs_cbk
+     * with a cookie that points to something other than an inode,
+     * which is exactly what would happen with STACK_UNWIND using
+     * that as a callback.  Therefore, use default_statfs_cbk in
+     * this case instead.
+     *
+     * Also if the option deem-statfs is not set to "on" don't
+     * bother calculating quota limit on / in statfs_cbk.
+     */
+    if (priv->consider_statfs)
+        gf_log(this->name, GF_LOG_ERROR,
+               "Missing inode, can't adjust for quota");
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->statfs,
+                    loc, xdata);
+    return 0;
+
+err:
+    QUOTA_STACK_UNWIND(statfs, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
+
+int
+quota_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int op_ret, int op_errno, gf_dirent_t *entries,
+                   dict_t *xdata)
+{
+    gf_dirent_t *entry = NULL;
+    quota_local_t *local = NULL;
+    loc_t loc = {
+        0,
+    };
+
+    if (op_ret <= 0)
+        goto unwind;
+
+    local = frame->local;
+
+    list_for_each_entry(entry, &entries->list, list)
+    {
+        if ((strcmp(entry->d_name, ".") == 0) ||
+            (strcmp(entry->d_name, "..") == 0) || entry->inode == NULL)
+            continue;
+
+        gf_uuid_copy(loc.gfid, entry->d_stat.ia_gfid);
+        loc.inode = inode_ref(entry->inode);
+        loc.parent = inode_ref(local->loc.inode);
+        gf_uuid_copy(loc.pargfid, loc.parent->gfid);
+        loc.name = entry->d_name;
+
+        quota_fill_inodectx(this, entry->inode, entry->dict, &loc,
+                            &entry->d_stat, &op_errno);
+
+        loc_wipe(&loc);
+    }
+
+unwind:
+    QUOTA_STACK_UNWIND(readdirp, frame, op_ret, op_errno, entries, xdata);
+
+    return 0;
+}
 
 int
-quota_release (xlator_t *this, fd_t *fd)
+quota_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+               off_t offset, dict_t *dict)
 {
-	gf_quota_cache_sync (this);
+    quota_priv_t *priv = NULL;
+    int ret = 0;
+    gf_boolean_t new_dict = _gf_false;
+    quota_local_t *local = NULL;
+
+    priv = this->private;
+
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+
+    local = quota_local_new();
+
+    if (local == NULL) {
+        goto err;
+    }
+
+    frame->local = local;
 
-	return 0;
+    local->loc.inode = inode_ref(fd->inode);
+
+    if (dict == NULL) {
+        dict = dict_new();
+        new_dict = _gf_true;
+    }
+
+    if (dict) {
+        ret = dict_set_int8(dict, QUOTA_LIMIT_KEY, 1);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+                   "dict set of key for hard-limit");
+            goto err;
+        }
+    }
+
+    if (dict) {
+        ret = dict_set_int8(dict, QUOTA_LIMIT_OBJECTS_KEY, 1);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+                   "dict set of key for hard-limit "
+                   "failed");
+            goto err;
+        }
+    }
+
+    STACK_WIND(frame, quota_readdirp_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdirp, fd, size, offset, dict);
+
+    if (new_dict) {
+        dict_unref(dict);
+    }
+
+    return 0;
+err:
+    STACK_UNWIND_STRICT(readdirp, frame, -1, EINVAL, NULL, NULL);
+
+    if (new_dict) {
+        dict_unref(dict);
+    }
+
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->readdirp,
+                    fd, size, offset, dict);
+    return 0;
 }
 
+int32_t
+quota_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                    struct iatt *postbuf, dict_t *xdata)
+{
+    int32_t ret = 0;
+    uint64_t ctx_int = 0;
+    quota_inode_ctx_t *ctx = NULL;
+    quota_local_t *local = NULL;
+
+    local = frame->local;
+
+    if ((op_ret < 0) || (local == NULL)) {
+        goto out;
+    }
+
+    ret = inode_ctx_get(local->loc.inode, this, &ctx_int);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, Q_MSG_INODE_CTX_GET_FAILED,
+               "%s: failed to get the context", local->loc.path);
+        goto out;
+    }
+
+    ctx = (quota_inode_ctx_t *)(unsigned long)ctx_int;
+
+    if (ctx == NULL) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, Q_MSG_INODE_CTX_GET_FAILED,
+               "quota context not set in %s (gfid:%s)", local->loc.path,
+               uuid_utoa(local->loc.inode->gfid));
+        goto out;
+    }
+
+    LOCK(&ctx->lock);
+    {
+        ctx->buf = *postbuf;
+    }
+    UNLOCK(&ctx->lock);
+
+out:
+    QUOTA_STACK_UNWIND(fallocate, frame, op_ret, op_errno, prebuf, postbuf,
+                       xdata);
+
+    return 0;
+}
 
-/* notify */
 int32_t
-notify (xlator_t *this,
-	int32_t event,
-	void *data,
-	...)
+quota_fallocate_helper(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                       int32_t mode, off_t offset, size_t len, dict_t *xdata)
 {
-	default_notify (this, event, data);
-	return 0;
+    quota_local_t *local = NULL;
+    int32_t op_errno = EINVAL;
+
+    local = frame->local;
+
+    GF_VALIDATE_OR_GOTO("quota", local, unwind);
+
+    if (local->op_ret == -1) {
+        op_errno = local->op_errno;
+        if (op_errno == ENOENT || op_errno == ESTALE) {
+            /* We may get ENOENT/ESTALE in case of below scenario
+             *     fd = open file.txt
+             *     unlink file.txt
+             *     fallocate on fd
+             * Here build_ancestry can fail as the file is removed.
+             * For now ignore ENOENT/ESTALE on active fd
+             * We need to re-visit this code once we understand
+             * how other file-system behave in this scenario
+             */
+            gf_msg_debug(this->name, 0,
+                         "quota enforcer failed "
+                         "with ENOENT/ESTALE on %s, cannot check "
+                         "quota limits and allowing fallocate",
+                         uuid_utoa(fd->inode->gfid));
+        } else {
+            goto unwind;
+        }
+    }
+
+    STACK_WIND(frame, quota_fallocate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fallocate, fd, mode, offset, len,
+               xdata);
+    return 0;
+
+unwind:
+    QUOTA_STACK_UNWIND(fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
 }
 
 int32_t
-quota_lookup_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    inode_t *inode,
-		    struct iatt *buf,
-                    dict_t *dict,
-                    struct iatt *postparent)
-{
-	STACK_UNWIND_STRICT (
-                     lookup, 
-                     frame,
-		      op_ret,
-		      op_errno,
-		      inode,
-		      buf,
-                      dict,
-                      postparent);
-	return 0;
+quota_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+                off_t offset, size_t len, dict_t *xdata)
+{
+    int32_t op_errno = EINVAL;
+    int32_t parents = 0;
+    int32_t fail_count = 0;
+    quota_local_t *local = NULL;
+    quota_inode_ctx_t *ctx = NULL;
+    quota_priv_t *priv = NULL;
+    quota_dentry_t *dentry = NULL;
+    quota_dentry_t *tmp = NULL;
+    call_stub_t *stub = NULL;
+    struct list_head head = {
+        0,
+    };
+    inode_t *par_inode = NULL;
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, priv, unwind);
+
+    WIND_IF_QUOTAOFF(priv->is_quota_on, off);
+
+    INIT_LIST_HEAD(&head);
+
+    GF_ASSERT(frame);
+    GF_VALIDATE_OR_GOTO("quota", this, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, fd, unwind);
+
+    local = quota_local_new();
+    if (local == NULL) {
+        goto unwind;
+    }
+
+    frame->local = local;
+    local->loc.inode = inode_ref(fd->inode);
+
+    (void)quota_inode_ctx_get(fd->inode, this, &ctx, 0);
+    if (ctx == NULL) {
+        gf_msg_debug(this->name, 0,
+                     "quota context is NULL on inode"
+                     " (%s). If quota is not enabled recently and "
+                     "crawler has finished crawling, its an error",
+                     uuid_utoa(local->loc.inode->gfid));
+    }
+
+    stub = fop_fallocate_stub(frame, quota_fallocate_helper, fd, mode, offset,
+                              len, xdata);
+    if (stub == NULL) {
+        op_errno = ENOMEM;
+        goto unwind;
+    }
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, priv, unwind);
+
+    parents = quota_add_parents_from_ctx(ctx, &head);
+    if (parents == -1) {
+        op_errno = errno;
+        goto unwind;
+    }
+
+    /*
+     * Note that by using len as the delta we're assuming the range from
+     * offset to offset+len has not already been allocated. This can result
+     * in ENOSPC errors attempting to allocate an already allocated range.
+     */
+    local->delta = len;
+    local->object_delta = 0;
+    local->stub = stub;
+    local->link_count = parents;
+
+    if (parents == 0) {
+        local->link_count = 1;
+        quota_check_limit(frame, fd->inode, this);
+    } else {
+        list_for_each_entry_safe(dentry, tmp, &head, next)
+        {
+            par_inode = do_quota_check_limit(frame, fd->inode, this, dentry,
+                                             _gf_false);
+            if (par_inode == NULL) {
+                /* remove stale entry from inode_ctx */
+                quota_dentry_del(ctx, dentry->name, dentry->par);
+                parents--;
+                fail_count++;
+            } else {
+                inode_unref(par_inode);
+            }
+            __quota_dentry_free(dentry);
+        }
+
+        if (parents == 0) {
+            LOCK(&local->lock);
+            {
+                local->link_count++;
+            }
+            UNLOCK(&local->lock);
+            quota_check_limit(frame, fd->inode, this);
+        }
+
+        while (fail_count != 0) {
+            quota_link_count_decrement(frame);
+            fail_count--;
+        }
+    }
+
+    return 0;
+
+unwind:
+    QUOTA_STACK_UNWIND(fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+    return 0;
+
+off:
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->fallocate, fd, mode, offset, len,
+                    xdata);
+    return 0;
+}
+
+void
+quota_log_helper(char **usage_str, int64_t cur_size, inode_t *inode,
+                 char **path, time_t *cur_time)
+{
+    xlator_t *this = THIS;
+
+    if (!usage_str || !inode || !path || !cur_time) {
+        gf_log(this->name, GF_LOG_ERROR, "Received null argument");
+        return;
+    }
+
+    *usage_str = gf_uint64_2human_readable(cur_size);
+    if (!(*usage_str))
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, Q_MSG_ENOMEM,
+               "integer to string conversion failed Reason"
+               ":\"Cannot allocate memory\"");
+
+    inode_path(inode, NULL, path);
+    if (!(*path))
+        *path = uuid_utoa(inode->gfid);
+
+    *cur_time = gf_time();
+}
+
+/* Logs if
+ *  i.   Usage crossed soft limit
+ *  ii.  Usage above soft limit and alert-time elapsed
+ */
+void
+quota_log_usage(xlator_t *this, quota_inode_ctx_t *ctx, inode_t *inode,
+                int64_t delta)
+{
+    time_t cur_time = 0;
+    char *usage_str = NULL;
+    char *path = NULL;
+    int64_t cur_size = 0;
+    quota_priv_t *priv = NULL;
+
+    priv = this->private;
+    cur_size = ctx->size + delta;
+
+    if ((ctx->soft_lim <= 0) || cur_size < ctx->soft_lim)
+        return;
+
+    /* Usage crossed/reached soft limit */
+    if (DID_REACH_LIMIT(ctx->soft_lim, ctx->size, cur_size)) {
+        quota_log_helper(&usage_str, cur_size, inode, &path, &cur_time);
+
+        gf_msg(this->name, GF_LOG_ALERT, 0, Q_MSG_CROSSED_SOFT_LIMIT,
+               "Usage crossed soft limit: "
+               "%s used by %s",
+               usage_str, path);
+
+        gf_event(EVENT_QUOTA_CROSSED_SOFT_LIMIT,
+                 "Usage=%s;volume=%s;"
+                 "path=%s",
+                 usage_str, priv->volume_uuid, path);
+
+        ctx->prev_log_time = cur_time;
+
+    }
+    /* Usage is above soft limit */
+    else if (cur_size > ctx->soft_lim &&
+             quota_timeout(ctx->prev_log_time, priv->log_timeout)) {
+        quota_log_helper(&usage_str, cur_size, inode, &path, &cur_time);
+
+        gf_msg(this->name, GF_LOG_ALERT, 0, Q_MSG_CROSSED_SOFT_LIMIT,
+               "Usage is above soft limit: %s used by %s", usage_str, path);
+
+        gf_event(EVENT_QUOTA_CROSSED_SOFT_LIMIT,
+                 "Usage=%s;volume=%s;"
+                 "path=%s",
+                 usage_str, priv->volume_uuid, path);
+
+        ctx->prev_log_time = cur_time;
+    }
+
+    if (path)
+        GF_FREE(path);
+
+    if (usage_str)
+        GF_FREE(usage_str);
 }
 
 int32_t
-quota_lookup (call_frame_t *frame,
-		xlator_t *this,
-		loc_t *loc,
-		dict_t *xattr_req)
+mem_acct_init(xlator_t *this)
 {
-	struct quota_priv *priv = NULL;
+    int ret = -1;
 
-	priv = this->private;
+    if (!this)
+        return ret;
 
-	if (priv->only_first_time) {
-		if (strcmp (loc->path, "/") == 0) {
-			loc_copy(&(priv->root_loc), loc); 
-			priv->only_first_time = 0;
-			if (priv->disk_usage_limit)
-				gf_quota_get_disk_usage (this);
-		}
-	}
+    ret = xlator_mem_acct_init(this, gf_quota_mt_end + 1);
 
-	STACK_WIND (frame,
-		    quota_lookup_cbk,
-		    FIRST_CHILD(this),
-		    FIRST_CHILD(this)->fops->lookup,
-		    loc,
-		    xattr_req);
-	return 0;
+    if (ret != 0) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "Memory accounting init failed");
+        return ret;
+    }
+
+    return ret;
 }
 
 int32_t
-mem_acct_init (xlator_t *this)
+quota_forget(xlator_t *this, inode_t *inode)
 {
-        int     ret = -1;
+    int32_t ret = 0;
+    uint64_t ctx_int = 0;
+    quota_inode_ctx_t *ctx = NULL;
+    quota_dentry_t *dentry = NULL, *tmp;
+
+    ret = inode_ctx_del(inode, this, &ctx_int);
+
+    if (ret < 0) {
+        return 0;
+    }
 
-        if (!this)
-                return ret;
+    ctx = (quota_inode_ctx_t *)(long)ctx_int;
 
-        ret = xlator_mem_acct_init (this, gf_quota_mt_end + 1);
-        
-        if (ret != 0) {
-                gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
-                        "failed");
-                return ret;
+    LOCK(&ctx->lock);
+    {
+        list_for_each_entry_safe(dentry, tmp, &ctx->parents, next)
+        {
+            __quota_dentry_free(dentry);
         }
+    }
+    UNLOCK(&ctx->lock);
 
-        return ret;
+    LOCK_DESTROY(&ctx->lock);
+
+    GF_FREE(ctx);
+
+    return 0;
 }
 
 int
-reconfigure (xlator_t *this, dict_t *options)
-{
-
-	struct quota_priv *_private = NULL;
-	uint64_t	   disk_usage_limit;	
-	uint32_t   	   min_free_disk_limit;
-	data_t 		  *data = NULL;
-	int		   ret = 0;
-	
-	_private = this->private;
-
-	data = dict_get (options, "disk-usage-limit");
-	if (data) {
-		if (gf_string2bytesize (data->data, &disk_usage_limit) != 0) {
-                        gf_log (this->name, GF_LOG_ERROR, 
-                                "Reconfigure: Invalid number '%s' "
-				"for disk-usage limit", data->data);
-			//return -1;
-			ret = -1;
-			goto out;
-                }
-		_private->disk_usage_limit = disk_usage_limit;
-		gf_log (this->name, GF_LOG_TRACE,
-                        "Reconfiguring disk-usage-limit %"PRIu64"",
-			disk_usage_limit);
-
-	}
-	
-    
-        data = dict_get (options, "min-free-disk-limit");
-        if (data) {
-		if (gf_string2percent (data->data, &min_free_disk_limit) != 0){
-                        gf_log (this->name, GF_LOG_ERROR, 
-                                "Reconfigure : Invalid percent '%s'  for"
-				" min-free-disk-limit", data->data);
-			ret = -1;
-			goto out;
-		}
-
-		_private->min_free_disk_limit = min_free_disk_limit;
-		gf_log (this->name, GF_LOG_TRACE,
-                        "Reconfiguring min-free-disk-limit %d \%",
-			min_free_disk_limit);
-		
-        }
-out:	
-	return ret;
-
-}
-
-int32_t 
-init (xlator_t *this)
-{
-	int     ret  = 0;
-        data_t *data = NULL;
-	struct quota_priv *_private = NULL;
-
-	if (!this->children || this->children->next) {
-		gf_log (this->name, GF_LOG_ERROR, 
-			"FATAL: quota should have exactly one child");
-		return -1;
-	}
-	
-	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"dangling volume. check volfile ");
-	}
-
-	_private = GF_CALLOC (1, sizeof (struct quota_priv),
-                              gf_quota_mt_quota_priv);
-        _private->disk_usage_limit = 0;
-        data = dict_get (this->options, "disk-usage-limit");
-        if (data) {
-		if (gf_string2bytesize (data->data, &_private->disk_usage_limit) != 0) {
-                        gf_log (this->name, GF_LOG_ERROR, 
-                                "invalid number '%s' for disk-usage limit", data->data);
-			ret = -1;
-			goto out;
+notify(xlator_t *this, int event, void *data, ...)
+{
+    quota_priv_t *priv = NULL;
+    int ret = 0;
+    rpc_clnt_t *rpc = NULL;
+    gf_boolean_t conn_status = _gf_true;
+    xlator_t *victim = data;
+
+    priv = this->private;
+    if (!priv || !priv->is_quota_on)
+        goto out;
+
+    if (event == GF_EVENT_PARENT_DOWN) {
+        rpc = priv->rpc_clnt;
+        if (rpc) {
+            rpc_clnt_disable(rpc);
+            pthread_mutex_lock(&priv->conn_mutex);
+            {
+                conn_status = priv->conn_status;
+                while (conn_status) {
+                    (void)pthread_cond_wait(&priv->conn_cond,
+                                            &priv->conn_mutex);
+                    conn_status = priv->conn_status;
                 }
+            }
+            pthread_mutex_unlock(&priv->conn_mutex);
+            gf_log(this->name, GF_LOG_INFO,
+                   "Notify GF_EVENT_PARENT_DOWN for brick %s", victim->name);
+        }
+    }
 
-		LOCK_INIT (&_private->lock);
-		_private->current_disk_usage = 0;
-	}
-	
-        _private->min_free_disk_limit = 0;
-        data = dict_get (this->options, "min-free-disk-limit");
-        if (data) {
-		if (gf_string2percent (data->data, &_private->min_free_disk_limit) != 0) {
-                        gf_log (this->name, GF_LOG_ERROR, 
-                                "invalid percent '%s' for min-free-disk limit", data->data);
-			ret = -1;
-			goto out;
-                }
-		_private->refresh_interval = 20; /* 20seconds is default */
-		data = dict_get (this->options, "refresh-interval");
-		if (data) {
-			if (gf_string2time (data->data, 
-					    &_private->refresh_interval)!= 0) {
-				gf_log (this->name, GF_LOG_ERROR, 
-					"invalid time '%s' for refresh "
-					"interval", data->data);
-				ret = -1;
-				goto out;
-			}
-		}
+out:
+    ret = default_notify(this, event, data);
+    return ret;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    int32_t ret = -1;
+    quota_priv_t *priv = NULL;
+    rpc_clnt_t *rpc = NULL;
+
+    if ((this->children == NULL) || this->children->next) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, Q_MSG_INVALID_VOLFILE,
+               "FATAL: quota (%s) not configured with "
+               "exactly one child",
+               this->name);
+        return -1;
+    }
+
+    if (this->parents == NULL) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, Q_MSG_INVALID_VOLFILE,
+               "dangling volume. check volfile");
+    }
+
+    QUOTA_ALLOC_OR_GOTO(priv, quota_priv_t, err);
+
+    LOCK_INIT(&priv->lock);
+
+    this->private = priv;
+
+    GF_OPTION_INIT("deem-statfs", priv->consider_statfs, bool, err);
+    GF_OPTION_INIT("server-quota", priv->is_quota_on, bool, err);
+    GF_OPTION_INIT("default-soft-limit", priv->default_soft_lim, percent, err);
+    GF_OPTION_INIT("soft-timeout", priv->soft_timeout, time, err);
+    GF_OPTION_INIT("hard-timeout", priv->hard_timeout, time, err);
+    GF_OPTION_INIT("alert-time", priv->log_timeout, time, err);
+    GF_OPTION_INIT("volume-uuid", priv->volume_uuid, str, err);
+
+    this->local_pool = mem_pool_new(quota_local_t, 64);
+    if (!this->local_pool) {
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, Q_MSG_ENOMEM,
+               "failed to create local_t's memory pool");
+        goto err;
+    }
+
+    pthread_mutex_init(&priv->conn_mutex, NULL);
+    pthread_cond_init(&priv->conn_cond, NULL);
+    priv->conn_status = _gf_false;
+
+    if (priv->is_quota_on) {
+        rpc = quota_enforcer_init(this, this->options);
+        if (rpc == NULL) {
+            ret = -1;
+            gf_msg(this->name, GF_LOG_WARNING, 0,
+                   Q_MSG_QUOTA_ENFORCER_RPC_INIT_FAILED,
+                   "quota enforcer rpc init failed");
+            goto err;
+        }
+
+        LOCK(&priv->lock);
+        {
+            priv->rpc_clnt = rpc;
         }
+        UNLOCK(&priv->lock);
+    }
 
-	_private->only_first_time = 1;
-        this->private = (void *)_private;
-	ret = 0;
- out:
-	return ret;
+    ret = 0;
+err:
+    return ret;
 }
 
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    int32_t ret = -1;
+    quota_priv_t *priv = NULL;
+    gf_boolean_t quota_on = _gf_false;
+    rpc_clnt_t *rpc = NULL;
+
+    priv = this->private;
+
+    GF_OPTION_RECONF("deem-statfs", priv->consider_statfs, options, bool, out);
+    GF_OPTION_RECONF("server-quota", quota_on, options, bool, out);
+    GF_OPTION_RECONF("default-soft-limit", priv->default_soft_lim, options,
+                     percent, out);
+    GF_OPTION_RECONF("alert-time", priv->log_timeout, options, time, out);
+    GF_OPTION_RECONF("soft-timeout", priv->soft_timeout, options, time, out);
+    GF_OPTION_RECONF("hard-timeout", priv->hard_timeout, options, time, out);
+
+    if (quota_on) {
+        priv->rpc_clnt = quota_enforcer_init(this, this->options);
+        if (priv->rpc_clnt == NULL) {
+            ret = -1;
+            gf_msg(this->name, GF_LOG_WARNING, 0,
+                   Q_MSG_QUOTA_ENFORCER_RPC_INIT_FAILED,
+                   "quota enforcer rpc init failed");
+            goto out;
+        }
+
+    } else {
+        LOCK(&priv->lock);
+        {
+            rpc = priv->rpc_clnt;
+            priv->rpc_clnt = NULL;
+        }
+        UNLOCK(&priv->lock);
+
+        if (rpc != NULL) {
+            // Quotad is shutdown when there is no started volume
+            // which has quota enabled. So, we should disable the
+            // enforcer client when quota is disabled on a volume,
+            // to avoid spurious reconnect attempts to a service
+            // (quotad), that is known to be down.
+            rpc_clnt_unref(rpc);
+        }
+    }
 
+    priv->is_quota_on = quota_on;
 
-void 
-fini (xlator_t *this)
+    ret = 0;
+out:
+    return ret;
+}
+
+int32_t
+quota_priv_dump(xlator_t *this)
 {
-	struct quota_priv *_private = this->private;
+    quota_priv_t *priv = NULL;
+    int32_t ret = -1;
+
+    GF_ASSERT(this);
+
+    priv = this->private;
+    if (!priv)
+        goto out;
+
+    gf_proc_dump_add_section("xlators.features.quota.priv");
+
+    ret = TRY_LOCK(&priv->lock);
+    if (ret)
+        goto out;
+    else {
+        gf_proc_dump_write("soft-timeout", "%u", priv->soft_timeout);
+        gf_proc_dump_write("hard-timeout", "%u", priv->hard_timeout);
+        gf_proc_dump_write("alert-time", "%u", priv->log_timeout);
+        gf_proc_dump_write("quota-on", "%d", priv->is_quota_on);
+        gf_proc_dump_write("statfs", "%d", priv->consider_statfs);
+        gf_proc_dump_write("volume-uuid", "%s", priv->volume_uuid);
+        gf_proc_dump_write("validation-count", "%" PRIu64,
+                           priv->validation_count);
+    }
+    UNLOCK(&priv->lock);
+
+out:
+    return 0;
+}
 
-	if (_private) {
-		gf_quota_cache_sync (this);
-		this->private = NULL;
-	}
-	
-	return ;
+void
+fini(xlator_t *this)
+{
+    quota_priv_t *priv = NULL;
+    rpc_clnt_t *rpc = NULL;
+
+    priv = this->private;
+    if (!priv)
+        return;
+    rpc = priv->rpc_clnt;
+    priv->rpc_clnt = NULL;
+    if (rpc) {
+        rpc_clnt_connection_cleanup(&rpc->conn);
+        rpc_clnt_unref(rpc);
+    }
+
+    this->private = NULL;
+    LOCK_DESTROY(&priv->lock);
+    pthread_mutex_destroy(&priv->conn_mutex);
+    pthread_cond_destroy(&priv->conn_cond);
+
+    GF_FREE(priv);
+    if (this->local_pool) {
+        mem_pool_destroy(this->local_pool);
+        this->local_pool = NULL;
+    }
+    return;
 }
 
 struct xlator_fops fops = {
-	.create      = quota_create,
-	.open        = quota_open,
-	.lookup	     = quota_lookup,
-	.truncate    = quota_truncate,
-	.ftruncate   = quota_ftruncate,
-	.writev      = quota_writev,
-	.unlink      = quota_unlink,
-	.rmdir       = quota_rmdir,
-	.mknod       = quota_mknod,
-	.mkdir       = quota_mkdir,
-	.symlink     = quota_symlink,
-	.statfs      = quota_statfs,
+    .statfs = quota_statfs,
+    .lookup = quota_lookup,
+    .writev = quota_writev,
+    .create = quota_create,
+    .mkdir = quota_mkdir,
+    .truncate = quota_truncate,
+    .ftruncate = quota_ftruncate,
+    .unlink = quota_unlink,
+    .symlink = quota_symlink,
+    .link = quota_link,
+    .rename = quota_rename,
+    .getxattr = quota_getxattr,
+    .fgetxattr = quota_fgetxattr,
+    .stat = quota_stat,
+    .fstat = quota_fstat,
+    .readlink = quota_readlink,
+    .readv = quota_readv,
+    .fsync = quota_fsync,
+    .setattr = quota_setattr,
+    .fsetattr = quota_fsetattr,
+    .mknod = quota_mknod,
+    .setxattr = quota_setxattr,
+    .fsetxattr = quota_fsetxattr,
+    .removexattr = quota_removexattr,
+    .fremovexattr = quota_fremovexattr,
+    .readdirp = quota_readdirp,
+    .fallocate = quota_fallocate,
 };
 
-struct xlator_cbks cbks = {
-	.release     = quota_release
-};
+struct xlator_cbks cbks = {.forget = quota_forget};
 
+struct xlator_dumpops dumpops = {
+    .priv = quota_priv_dump,
+};
 struct volume_options options[] = {
-	{ .key  = {"min-free-disk-limit"}, 
-	  .type = GF_OPTION_TYPE_PERCENT
-	},
-	{ .key  = {"refresh-interval"}, 
-	  .type = GF_OPTION_TYPE_TIME
-	},
-	{ .key  = {"disk-usage-limit"}, 
-	  .type = GF_OPTION_TYPE_SIZET 
-	},
-	{ .key = {NULL} },
+    {
+        .key = {"enable"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "off",
+        .description = "enable is the volume option that can be used "
+                       "to turn on quota.",
+        .op_version = {1},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .level = OPT_STATUS_BASIC,
+        .tags = {},
+    },
+    {
+        .key = {"deem-statfs"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "on",
+        .description = "If set to on, it takes quota limits into"
+                       " consideration while estimating fs size. (df command)"
+                       " (Default is on).",
+        .op_version = {2},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {},
+    },
+    {
+        .key = {"server-quota"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "off",
+        .description = "Skip the quota enforcement if the feature is"
+                       " not turned on. This is not a user exposed option.",
+        .flags = OPT_FLAG_NONE,
+    },
+    {
+        .key = {"default-soft-limit"},
+        .type = GF_OPTION_TYPE_PERCENT,
+        .default_value = "80%",
+        .op_version = {3},
+        .description = "Soft limit is expressed as a proportion of hard limit."
+                       " Default-soft-limit is the proportion used when the "
+                       " user does not supply any soft limit value.",
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {},
+    },
+    {
+        .key = {"soft-timeout"},
+        .type = GF_OPTION_TYPE_TIME,
+        .min = 0,
+        .max = 1800,
+        .default_value = "60",
+        .description = "quota caches the directory sizes on client. "
+                       "soft-timeout indicates the timeout for the validity of"
+                       " cache before soft-limit has been crossed.",
+        .op_version = {3},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {},
+    },
+    {
+        .key = {"hard-timeout"},
+        .type = GF_OPTION_TYPE_TIME,
+        .min = 0,
+        .max = 60,
+        .default_value = "5",
+        .description = "quota caches the directory sizes on client. "
+                       "hard-timeout indicates the timeout for the validity of"
+                       " cache after soft-limit has been crossed.",
+        .op_version = {3},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {},
+    },
+    {.key = {"volume-uuid"},
+     .type = GF_OPTION_TYPE_STR,
+     .default_value = "{{ volume.id }}",
+     .description = "uuid of the volume this brick is part of."},
+    {
+        .key = {"alert-time"},
+        .type = GF_OPTION_TYPE_TIME,
+        .min = 0,
+        .max = 7 * 86400,
+        .default_value = "86400",
+        .op_version = {3},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .description = "Frequency of limit breach messages in log.",
+        .tags = {},
+    },
+    {.key = {NULL}}};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "quota",
+    .category = GF_MAINTAINED,
 };
diff --git a/xlators/features/quota/src/quota.h b/xlators/features/quota/src/quota.h
new file mode 100644
index 00000000000..0395d78c9ef
--- /dev/null
+++ b/xlators/features/quota/src/quota.h
@@ -0,0 +1,266 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#ifndef _QUOTA_H
+#define _QUOTA_H
+
+#include <glusterfs/call-stub.h>
+#include "quota-mem-types.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/dict.h>
+#include <glusterfs/gf-event.h>
+#include "rpcsvc.h"
+#include "rpc-clnt.h"
+#include <glusterfs/byte-order.h>
+#include "glusterfs3-xdr.h"
+#include "glusterfs3.h"
+#include "xdr-generic.h"
+#include <glusterfs/compat-errno.h>
+#include "protocol-common.h"
+#include <glusterfs/quota-common-utils.h>
+#include "quota-messages.h"
+
+#define DIRTY "dirty"
+#define SIZE "size"
+#define CONTRIBUTION "contri"
+#define VAL_LENGTH 8
+#define READDIR_BUF 4096
+
+#ifndef UUID_CANONICAL_FORM_LEN
+#define UUID_CANONICAL_FORM_LEN 36
+#endif
+
+#define WIND_IF_QUOTAOFF(is_quota_on, label)                                   \
+    if (!is_quota_on)                                                          \
+        goto label;
+
+#define QUOTA_WIND_FOR_INTERNAL_FOP(xdata, label)                              \
+    do {                                                                       \
+        if (xdata && dict_get_sizen(xdata, GLUSTERFS_INTERNAL_FOP_KEY))        \
+            goto label;                                                        \
+    } while (0)
+
+#define DID_REACH_LIMIT(lim, prev_size, cur_size)                              \
+    ((cur_size) >= (lim) && (prev_size) < (lim))
+
+#define QUOTA_SAFE_INCREMENT(lock, var)                                        \
+    do {                                                                       \
+        LOCK(lock);                                                            \
+        var++;                                                                 \
+        UNLOCK(lock);                                                          \
+    } while (0)
+
+#define QUOTA_SAFE_DECREMENT(lock, var)                                        \
+    do {                                                                       \
+        LOCK(lock);                                                            \
+        var--;                                                                 \
+        UNLOCK(lock);                                                          \
+    } while (0)
+
+#define QUOTA_ALLOC_OR_GOTO(var, type, label)                                  \
+    do {                                                                       \
+        var = GF_CALLOC(sizeof(type), 1, gf_quota_mt_##type);                  \
+        if (!var) {                                                            \
+            gf_msg("", GF_LOG_ERROR, ENOMEM, Q_MSG_ENOMEM, "out of memory");   \
+            ret = -1;                                                          \
+            goto label;                                                        \
+        }                                                                      \
+    } while (0);
+
+#define QUOTA_STACK_WIND_TAIL(frame, params...)                                \
+    do {                                                                       \
+        quota_local_t *_local = NULL;                                          \
+                                                                               \
+        if (frame) {                                                           \
+            _local = frame->local;                                             \
+            frame->local = NULL;                                               \
+        }                                                                      \
+                                                                               \
+        STACK_WIND_TAIL(frame, params);                                        \
+                                                                               \
+        if (_local)                                                            \
+            quota_local_cleanup(_local);                                       \
+    } while (0)
+
+#define QUOTA_STACK_UNWIND(fop, frame, params...)                              \
+    do {                                                                       \
+        quota_local_t *_local = NULL;                                          \
+        if (frame) {                                                           \
+            _local = frame->local;                                             \
+            frame->local = NULL;                                               \
+        }                                                                      \
+        STACK_UNWIND_STRICT(fop, frame, params);                               \
+        quota_local_cleanup(_local);                                           \
+    } while (0)
+
+#define QUOTA_FREE_CONTRIBUTION_NODE(_contribution)                            \
+    do {                                                                       \
+        list_del(&_contribution->contri_list);                                 \
+        GF_FREE(_contribution);                                                \
+    } while (0)
+
+#define GET_CONTRI_KEY(var, _vol_name, _gfid, _ret)                            \
+    do {                                                                       \
+        char _gfid_unparsed[40];                                               \
+        if (_gfid != NULL) {                                                   \
+            gf_uuid_unparse(_gfid, _gfid_unparsed);                            \
+            _ret = gf_asprintf(var, QUOTA_XATTR_PREFIX "%s.%s." CONTRIBUTION,  \
+                               _vol_name, _gfid_unparsed);                     \
+        } else {                                                               \
+            _ret = gf_asprintf(var, QUOTA_XATTR_PREFIX "%s.." CONTRIBUTION,    \
+                               _vol_name);                                     \
+        }                                                                      \
+    } while (0)
+
+#define GET_CONTRI_KEY_OR_GOTO(var, _vol_name, _gfid, label)                   \
+    do {                                                                       \
+        GET_CONTRI_KEY(var, _vol_name, _gfid, ret);                            \
+        if (ret == -1)                                                         \
+            goto label;                                                        \
+    } while (0)
+
+#define GET_DIRTY_KEY_OR_GOTO(var, _vol_name, label)                           \
+    do {                                                                       \
+        ret = gf_asprintf(var, QUOTA_XATTR_PREFIX "%s." DIRTY, _vol_name);     \
+        if (ret == -1)                                                         \
+            goto label;                                                        \
+    } while (0)
+
+#define QUOTA_REG_OR_LNK_FILE(ia_type) (IA_ISREG(ia_type) || IA_ISLNK(ia_type))
+
+struct quota_dentry {
+    char *name;
+    uuid_t par;
+    struct list_head next;
+};
+typedef struct quota_dentry quota_dentry_t;
+
+struct quota_inode_ctx {
+    int64_t size;
+    int64_t hard_lim;
+    int64_t soft_lim;
+    int64_t file_count;
+    int64_t dir_count;
+    int64_t object_hard_lim;
+    int64_t object_soft_lim;
+    struct iatt buf;
+    struct list_head parents;
+    time_t validate_time;
+    time_t prev_log_time;
+    gf_boolean_t ancestry_built;
+    gf_lock_t lock;
+};
+typedef struct quota_inode_ctx quota_inode_ctx_t;
+
+typedef void (*quota_ancestry_built_t)(struct list_head *parents,
+                                       inode_t *inode, int32_t op_ret,
+                                       int32_t op_errno, void *data);
+
+typedef void (*quota_fop_continue_t)(call_frame_t *frame);
+
+struct quota_local {
+    gf_lock_t lock;
+    uint32_t link_count;
+    loc_t loc;
+    loc_t oldloc;
+    loc_t newloc;
+    loc_t validate_loc;
+    int64_t delta;
+    int8_t object_delta;
+    int32_t op_ret;
+    int32_t op_errno;
+    int64_t size;
+    char just_validated;
+    fop_lookup_cbk_t validate_cbk;
+    quota_fop_continue_t fop_continue_cbk;
+    inode_t *inode;
+    uuid_t common_ancestor; /* Used by quota_rename */
+    call_stub_t *stub;
+    struct iobref *iobref;
+    quota_limits_t limit;
+    quota_limits_t object_limit;
+    int64_t space_available;
+    quota_ancestry_built_t ancestry_cbk;
+    void *ancestry_data;
+    dict_t *xdata;
+    dict_t *validate_xdata;
+    int32_t quotad_conn_retry;
+    xlator_t *this;
+    call_frame_t *par_frame;
+};
+typedef struct quota_local quota_local_t;
+
+struct quota_priv {
+    /* FIXME: consider time_t for timeouts. */
+    uint32_t soft_timeout;
+    uint32_t hard_timeout;
+    uint32_t log_timeout;
+    double default_soft_lim;
+    gf_boolean_t is_quota_on;
+    gf_boolean_t consider_statfs;
+    gf_lock_t lock;
+    rpc_clnt_prog_t *quota_enforcer;
+    struct rpcsvc_program *quotad_aggregator;
+    struct rpc_clnt *rpc_clnt;
+    rpcsvc_t *rpcsvc;
+    inode_table_t *itable;
+    char *volume_uuid;
+    uint64_t validation_count;
+    int32_t quotad_conn_status;
+    pthread_mutex_t conn_mutex;
+    pthread_cond_t conn_cond;
+    gf_boolean_t conn_status;
+};
+typedef struct quota_priv quota_priv_t;
+
+int
+quota_enforcer_lookup(call_frame_t *frame, xlator_t *this, dict_t *xdata,
+                      fop_lookup_cbk_t cbk);
+
+void
+_quota_enforcer_lookup(void *data);
+
+struct rpc_clnt *
+quota_enforcer_init(xlator_t *this, dict_t *options);
+
+void
+quota_log_usage(xlator_t *this, quota_inode_ctx_t *ctx, inode_t *inode,
+                int64_t delta);
+
+int
+quota_build_ancestry(inode_t *inode, quota_ancestry_built_t ancestry_cbk,
+                     void *data);
+
+void
+quota_get_limit_dir(call_frame_t *frame, inode_t *cur_inode, xlator_t *this);
+
+int32_t
+quota_check_limit(call_frame_t *frame, inode_t *inode, xlator_t *this);
+
+inode_t *
+do_quota_check_limit(call_frame_t *frame, inode_t *inode, xlator_t *this,
+                     quota_dentry_t *dentry, gf_boolean_t force);
+int
+quota_fill_inodectx(xlator_t *this, inode_t *inode, dict_t *dict, loc_t *loc,
+                    struct iatt *buf, int32_t *op_errno);
+
+int32_t
+quota_check_size_limit(call_frame_t *frame, quota_inode_ctx_t *ctx,
+                       quota_priv_t *priv, inode_t *_inode, xlator_t *this,
+                       int32_t *op_errno, int just_validated, int64_t delta,
+                       quota_local_t *local, gf_boolean_t *skip_check);
+
+int32_t
+quota_check_object_limit(call_frame_t *frame, quota_inode_ctx_t *ctx,
+                         quota_priv_t *priv, inode_t *_inode, xlator_t *this,
+                         int32_t *op_errno, int just_validated,
+                         quota_local_t *local, gf_boolean_t *skip_check);
+#endif
diff --git a/xlators/features/quota/src/quotad-aggregator.c b/xlators/features/quota/src/quotad-aggregator.c
new file mode 100644
index 00000000000..75d47867b5b
--- /dev/null
+++ b/xlators/features/quota/src/quotad-aggregator.c
@@ -0,0 +1,494 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include "cli1-xdr.h"
+#include "quota.h"
+#include "quotad-helpers.h"
+#include "quotad-aggregator.h"
+
+static char *qd_ext_xattrs[] = {
+    QUOTA_SIZE_KEY,
+    QUOTA_LIMIT_KEY,
+    QUOTA_LIMIT_OBJECTS_KEY,
+    NULL,
+};
+
+static struct rpcsvc_program quotad_aggregator_prog;
+
+struct iobuf *
+quotad_serialize_reply(rpcsvc_request_t *req, void *arg, struct iovec *outmsg,
+                       xdrproc_t xdrproc)
+{
+    struct iobuf *iob = NULL;
+    ssize_t retlen = 0;
+    ssize_t xdr_size = 0;
+
+    GF_VALIDATE_OR_GOTO("server", req, ret);
+
+    /* First, get the io buffer into which the reply in arg will
+     * be serialized.
+     */
+    if (arg && xdrproc) {
+        xdr_size = xdr_sizeof(xdrproc, arg);
+        iob = iobuf_get2(req->svc->ctx->iobuf_pool, xdr_size);
+        if (!iob) {
+            gf_log_callingfn(THIS->name, GF_LOG_ERROR, "Failed to get iobuf");
+            goto ret;
+        };
+
+        iobuf_to_iovec(iob, outmsg);
+        /* Use the given serializer to translate the given C structure
+         * in arg to XDR format which will be written into the buffer
+         * in outmsg.
+         */
+        /* retlen is used to received the error since size_t is unsigned and we
+         * need -1 for error notification during encoding.
+         */
+
+        retlen = xdr_serialize_generic(*outmsg, arg, xdrproc);
+        if (retlen == -1) {
+            /* Failed to Encode 'GlusterFS' msg in RPC is not exactly
+               failure of RPC return values.. Client should get
+               notified about this, so there are no missing frames */
+            gf_log_callingfn("", GF_LOG_ERROR, "Failed to encode message");
+            req->rpc_err = GARBAGE_ARGS;
+            retlen = 0;
+        }
+    }
+    outmsg->iov_len = retlen;
+ret:
+    return iob;
+}
+
+int
+quotad_aggregator_submit_reply(call_frame_t *frame, rpcsvc_request_t *req,
+                               void *arg, struct iovec *payload,
+                               int payloadcount, struct iobref *iobref,
+                               xdrproc_t xdrproc)
+{
+    struct iobuf *iob = NULL;
+    int ret = -1;
+    struct iovec rsp = {
+        0,
+    };
+    quotad_aggregator_state_t *state = NULL;
+    char new_iobref = 0;
+
+    GF_VALIDATE_OR_GOTO("server", req, ret);
+
+    if (frame) {
+        state = frame->root->state;
+        frame->local = NULL;
+    }
+
+    if (!iobref) {
+        iobref = iobref_new();
+        if (!iobref) {
+            goto ret;
+        }
+
+        new_iobref = 1;
+    }
+
+    iob = quotad_serialize_reply(req, arg, &rsp, xdrproc);
+    if (!iob) {
+        gf_msg("", GF_LOG_ERROR, 0, Q_MSG_DICT_SERIALIZE_FAIL,
+               "Failed to serialize reply");
+        goto ret;
+    }
+
+    iobref_add(iobref, iob);
+
+    ret = rpcsvc_submit_generic(req, &rsp, 1, payload, payloadcount, iobref);
+
+    iobuf_unref(iob);
+
+    ret = 0;
+ret:
+    if (state) {
+        quotad_aggregator_free_state(state);
+    }
+
+    if (frame)
+        STACK_DESTROY(frame->root);
+
+    if (new_iobref) {
+        iobref_unref(iobref);
+    }
+
+    return ret;
+}
+
+int
+quotad_aggregator_getlimit_cbk(xlator_t *this, call_frame_t *frame,
+                               void *lookup_rsp)
+{
+    gfs3_lookup_rsp *rsp = lookup_rsp;
+    gf_cli_rsp cli_rsp = {
+        0,
+    };
+    dict_t *xdata = NULL;
+    quotad_aggregator_state_t *state = NULL;
+    int ret = -1;
+    int type = 0;
+
+    if (!rsp || (rsp->op_ret == -1))
+        goto reply;
+
+    GF_PROTOCOL_DICT_UNSERIALIZE(frame->this, xdata, (rsp->xdata.xdata_val),
+                                 (rsp->xdata.xdata_len), rsp->op_ret,
+                                 rsp->op_errno, out);
+
+    if (xdata) {
+        state = frame->root->state;
+        ret = dict_get_int32n(state->req_xdata, "type", SLEN("type"), &type);
+        if (ret < 0)
+            goto out;
+
+        ret = dict_set_int32_sizen(xdata, "type", type);
+        if (ret < 0)
+            goto out;
+    }
+
+    ret = 0;
+out:
+    rsp->op_ret = ret;
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, Q_MSG_DICT_UNSERIALIZE_FAIL,
+               "failed to unserialize "
+               "nameless lookup rsp");
+        goto reply;
+    }
+    cli_rsp.op_ret = rsp->op_ret;
+    cli_rsp.op_errno = rsp->op_errno;
+    cli_rsp.op_errstr = "";
+    if (xdata) {
+        GF_PROTOCOL_DICT_SERIALIZE(frame->this, xdata, (&cli_rsp.dict.dict_val),
+                                   (cli_rsp.dict.dict_len), cli_rsp.op_errno,
+                                   reply);
+    }
+
+reply:
+    quotad_aggregator_submit_reply(frame, (frame) ? frame->local : NULL,
+                                   (void *)&cli_rsp, NULL, 0, NULL,
+                                   (xdrproc_t)xdr_gf_cli_rsp);
+
+    dict_unref(xdata);
+    GF_FREE(cli_rsp.dict.dict_val);
+    return 0;
+}
+
+int
+quotad_aggregator_getlimit(rpcsvc_request_t *req)
+{
+    call_frame_t *frame = NULL;
+    gf_cli_req cli_req = {
+        {0},
+    };
+    gf_cli_rsp cli_rsp = {0};
+    quotad_aggregator_state_t *state = NULL;
+    xlator_t *this = NULL;
+    dict_t *dict = NULL;
+    int ret = -1, op_errno = 0;
+    char *gfid_str = NULL;
+    uuid_t gfid = {0};
+    char *volume_uuid = NULL;
+
+    GF_VALIDATE_OR_GOTO("quotad-aggregator", req, err);
+
+    this = THIS;
+
+    cli_req.dict.dict_val = alloca(req->msg[0].iov_len);
+
+    ret = xdr_to_generic(req->msg[0], &cli_req, (xdrproc_t)xdr_gf_cli_req);
+    if (ret < 0) {
+        // failed to decode msg;
+        gf_msg("this->name", GF_LOG_ERROR, 0, Q_MSG_XDR_DECODE_ERROR,
+               "xdr decoding error");
+        req->rpc_err = GARBAGE_ARGS;
+        goto err;
+    }
+
+    if (cli_req.dict.dict_len) {
+        dict = dict_new();
+        ret = dict_unserialize(cli_req.dict.dict_val, cli_req.dict.dict_len,
+                               &dict);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, Q_MSG_DICT_UNSERIALIZE_FAIL,
+                   "Failed to unserialize req-buffer to "
+                   "dictionary");
+            goto err;
+        }
+    }
+
+    ret = dict_get_strn(dict, "gfid", SLEN("gfid"), &gfid_str);
+    if (ret) {
+        goto err;
+    }
+
+    ret = dict_get_strn(dict, "volume-uuid", SLEN("volume-uuid"), &volume_uuid);
+    if (ret) {
+        goto err;
+    }
+
+    gf_uuid_parse((const char *)gfid_str, gfid);
+
+    frame = quotad_aggregator_get_frame_from_req(req);
+    if (frame == NULL) {
+        cli_rsp.op_errno = ENOMEM;
+        goto errx;
+    }
+    state = frame->root->state;
+    state->req_xdata = dict;
+    state->xdata = dict_new();
+    dict = NULL;
+
+    ret = dict_set_int32_sizen(state->xdata, QUOTA_LIMIT_KEY, 42);
+    if (ret)
+        goto err;
+
+    ret = dict_set_int32_sizen(state->xdata, QUOTA_LIMIT_OBJECTS_KEY, 42);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, Q_MSG_ENOMEM,
+               "Failed to set QUOTA_LIMIT_OBJECTS_KEY");
+        goto err;
+    }
+
+    ret = dict_set_int32_sizen(state->xdata, QUOTA_SIZE_KEY, 42);
+    if (ret)
+        goto err;
+
+    ret = dict_set_int32_sizen(state->xdata, GET_ANCESTRY_PATH_KEY, 42);
+    if (ret)
+        goto err;
+
+    ret = qd_nameless_lookup(this, frame, (char *)gfid, state->xdata,
+                             volume_uuid, quotad_aggregator_getlimit_cbk);
+    if (ret) {
+        cli_rsp.op_errno = ret;
+        goto errx;
+    }
+
+    return ret;
+
+err:
+    cli_rsp.op_errno = op_errno;
+errx:
+    cli_rsp.op_ret = -1;
+    cli_rsp.op_errstr = "";
+
+    quotad_aggregator_getlimit_cbk(this, frame, &cli_rsp);
+    if (dict)
+        dict_unref(dict);
+    return ret;
+}
+
+int
+quotad_aggregator_lookup_cbk(xlator_t *this, call_frame_t *frame, void *rsp)
+{
+    quotad_aggregator_submit_reply(frame, frame ? frame->local : NULL, rsp,
+                                   NULL, 0, NULL,
+                                   (xdrproc_t)xdr_gfs3_lookup_rsp);
+
+    return 0;
+}
+
+int
+quotad_aggregator_lookup(rpcsvc_request_t *req)
+{
+    call_frame_t *frame = NULL;
+    gfs3_lookup_req args = {
+        {
+            0,
+        },
+    };
+    int i = 0, ret = -1, op_errno = 0;
+    gfs3_lookup_rsp rsp = {
+        0,
+    };
+    quotad_aggregator_state_t *state = NULL;
+    xlator_t *this = NULL;
+    dict_t *dict = NULL;
+    char *volume_uuid = NULL;
+
+    GF_VALIDATE_OR_GOTO("quotad-aggregator", req, err);
+
+    this = THIS;
+
+    args.bname = alloca(req->msg[0].iov_len);
+    args.xdata.xdata_val = alloca(req->msg[0].iov_len);
+
+    ret = xdr_to_generic(req->msg[0], &args, (xdrproc_t)xdr_gfs3_lookup_req);
+    if (ret < 0) {
+        rsp.op_errno = EINVAL;
+        goto err;
+    }
+
+    frame = quotad_aggregator_get_frame_from_req(req);
+    if (frame == NULL) {
+        rsp.op_errno = ENOMEM;
+        goto err;
+    }
+
+    state = frame->root->state;
+
+    GF_PROTOCOL_DICT_UNSERIALIZE(this, dict, (args.xdata.xdata_val),
+                                 (args.xdata.xdata_len), ret, op_errno, err);
+
+    ret = dict_get_str(dict, "volume-uuid", &volume_uuid);
+    if (ret) {
+        goto err;
+    }
+
+    state->xdata = dict_new();
+
+    for (i = 0; qd_ext_xattrs[i]; i++) {
+        if (dict_get(dict, qd_ext_xattrs[i])) {
+            ret = dict_set_uint32(state->xdata, qd_ext_xattrs[i], 1);
+            if (ret < 0)
+                goto err;
+        }
+    }
+
+    ret = qd_nameless_lookup(this, frame, args.gfid, state->xdata, volume_uuid,
+                             quotad_aggregator_lookup_cbk);
+    if (ret) {
+        rsp.op_errno = ret;
+        goto err;
+    }
+
+    if (dict)
+        dict_unref(dict);
+
+    return ret;
+
+err:
+    rsp.op_ret = -1;
+    rsp.op_errno = op_errno;
+
+    quotad_aggregator_lookup_cbk(this, frame, &rsp);
+    if (dict)
+        dict_unref(dict);
+
+    return ret;
+}
+
+int
+quotad_aggregator_rpc_notify(rpcsvc_t *rpc, void *xl, rpcsvc_event_t event,
+                             void *data)
+{
+    if (!xl || !data) {
+        gf_log_callingfn("server", GF_LOG_WARNING,
+                         "Calling rpc_notify without initializing");
+        goto out;
+    }
+
+    switch (event) {
+        case RPCSVC_EVENT_ACCEPT:
+            break;
+
+        case RPCSVC_EVENT_DISCONNECT:
+            break;
+
+        default:
+            break;
+    }
+
+out:
+    return 0;
+}
+
+int
+quotad_aggregator_init(xlator_t *this)
+{
+    quota_priv_t *priv = NULL;
+    int ret = -1;
+
+    priv = this->private;
+
+    if (priv->rpcsvc) {
+        /* Listener already created */
+        return 0;
+    }
+
+    ret = dict_set_nstrn(this->options, "transport.address-family",
+                         SLEN("transport.address-family"), "unix",
+                         SLEN("unix"));
+    if (ret)
+        goto out;
+
+    ret = dict_set_nstrn(this->options, "transport-type",
+                         SLEN("transport-type"), "socket", SLEN("socket"));
+    if (ret)
+        goto out;
+
+    ret = dict_set_nstrn(this->options, "transport.socket.listen-path",
+                         SLEN("transport.socket.listen-path"),
+                         "/var/run/gluster/quotad.socket",
+                         SLEN("/var/run/gluster/quotad.socket"));
+    if (ret)
+        goto out;
+
+    /* RPC related */
+    priv->rpcsvc = rpcsvc_init(this, this->ctx, this->options, 0);
+    if (priv->rpcsvc == NULL) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, Q_MSG_RPCSVC_INIT_FAILED,
+               "creation of rpcsvc failed");
+        ret = -1;
+        goto out;
+    }
+
+    ret = rpcsvc_create_listeners(priv->rpcsvc, this->options, this->name);
+    if (ret < 1) {
+        gf_msg(this->name, GF_LOG_WARNING, 0,
+               Q_MSG_RPCSVC_LISTENER_CREATION_FAILED,
+               "creation of listener failed");
+        ret = -1;
+        goto out;
+    }
+
+    priv->quotad_aggregator = &quotad_aggregator_prog;
+    quotad_aggregator_prog.options = this->options;
+
+    ret = rpcsvc_program_register(priv->rpcsvc, &quotad_aggregator_prog,
+                                  _gf_false);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, Q_MSG_RPCSVC_REGISTER_FAILED,
+               "registration of program (name:%s, prognum:%d, "
+               "progver:%d) failed",
+               quotad_aggregator_prog.progname, quotad_aggregator_prog.prognum,
+               quotad_aggregator_prog.progver);
+        goto out;
+    }
+
+    ret = 0;
+out:
+    if (ret && priv->rpcsvc) {
+        GF_FREE(priv->rpcsvc);
+        priv->rpcsvc = NULL;
+    }
+
+    return ret;
+}
+
+static rpcsvc_actor_t quotad_aggregator_actors[GF_AGGREGATOR_MAXVALUE] = {
+    [GF_AGGREGATOR_NULL] = {"NULL", NULL, NULL, GF_AGGREGATOR_NULL, DRC_NA, 0},
+    [GF_AGGREGATOR_LOOKUP] = {"LOOKUP", quotad_aggregator_lookup, NULL,
+                              GF_AGGREGATOR_NULL, DRC_NA, 0},
+    [GF_AGGREGATOR_GETLIMIT] = {"GETLIMIT", quotad_aggregator_getlimit, NULL,
+                                GF_AGGREGATOR_GETLIMIT, DRC_NA, 0},
+};
+
+static struct rpcsvc_program quotad_aggregator_prog = {
+    .progname = "GlusterFS 3.3",
+    .prognum = GLUSTER_AGGREGATOR_PROGRAM,
+    .progver = GLUSTER_AGGREGATOR_VERSION,
+    .numactors = GF_AGGREGATOR_MAXVALUE,
+    .actors = quotad_aggregator_actors};
diff --git a/xlators/features/quota/src/quotad-aggregator.h b/xlators/features/quota/src/quotad-aggregator.h
new file mode 100644
index 00000000000..706592c7d50
--- /dev/null
+++ b/xlators/features/quota/src/quotad-aggregator.h
@@ -0,0 +1,38 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef _QUOTAD_AGGREGATOR_H
+#define _QUOTAD_AGGREGATOR_H
+
+#include "quota.h"
+#include <glusterfs/stack.h>
+#include "glusterfs3-xdr.h"
+#include <glusterfs/inode.h>
+
+typedef struct {
+    void *pool;
+    xlator_t *this;
+    xlator_t *active_subvol;
+    inode_table_t *itable;
+    loc_t loc;
+    dict_t *xdata;
+    dict_t *req_xdata;
+} quotad_aggregator_state_t;
+
+typedef int (*quotad_aggregator_lookup_cbk_t)(xlator_t *this,
+                                              call_frame_t *frame, void *rsp);
+int
+qd_nameless_lookup(xlator_t *this, call_frame_t *frame, char *gfid,
+                   dict_t *xdata, char *volume_uuid,
+                   quotad_aggregator_lookup_cbk_t lookup_cbk);
+int
+quotad_aggregator_init(xlator_t *this);
+
+#endif
diff --git a/xlators/features/quota/src/quotad-helpers.c b/xlators/features/quota/src/quotad-helpers.c
new file mode 100644
index 00000000000..51ff1d7e98d
--- /dev/null
+++ b/xlators/features/quota/src/quotad-helpers.c
@@ -0,0 +1,107 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include "quotad-helpers.h"
+
+quotad_aggregator_state_t *
+get_quotad_aggregator_state(xlator_t *this, rpcsvc_request_t *req)
+{
+    quotad_aggregator_state_t *state = NULL;
+    xlator_t *active_subvol = NULL;
+    quota_priv_t *priv = NULL;
+
+    state = (void *)GF_CALLOC(1, sizeof(*state),
+                              gf_quota_mt_aggregator_state_t);
+    if (!state)
+        return NULL;
+
+    state->this = THIS;
+    priv = this->private;
+
+    LOCK(&priv->lock);
+    {
+        active_subvol = state->active_subvol = FIRST_CHILD(this);
+    }
+    UNLOCK(&priv->lock);
+
+    if (active_subvol->itable == NULL)
+        active_subvol->itable = inode_table_new(4096, active_subvol);
+
+    state->itable = active_subvol->itable;
+
+    state->pool = this->ctx->pool;
+
+    return state;
+}
+
+void
+quotad_aggregator_free_state(quotad_aggregator_state_t *state)
+{
+    if (state->xdata)
+        dict_unref(state->xdata);
+
+    if (state->req_xdata)
+        dict_unref(state->req_xdata);
+
+    GF_FREE(state);
+}
+
+call_frame_t *
+quotad_aggregator_alloc_frame(rpcsvc_request_t *req)
+{
+    call_frame_t *frame = NULL;
+    quotad_aggregator_state_t *state = NULL;
+    xlator_t *this = NULL;
+
+    GF_VALIDATE_OR_GOTO("server", req, out);
+    GF_VALIDATE_OR_GOTO("server", req->trans, out);
+    GF_VALIDATE_OR_GOTO("server", req->svc, out);
+    GF_VALIDATE_OR_GOTO("server", req->svc->ctx, out);
+
+    this = req->svc->xl;
+
+    frame = create_frame(this, req->svc->ctx->pool);
+    if (!frame)
+        goto out;
+
+    state = get_quotad_aggregator_state(this, req);
+    if (!state)
+        goto out;
+
+    frame->root->state = state;
+
+    frame->this = this;
+out:
+    return frame;
+}
+
+call_frame_t *
+quotad_aggregator_get_frame_from_req(rpcsvc_request_t *req)
+{
+    call_frame_t *frame = NULL;
+
+    GF_VALIDATE_OR_GOTO("server", req, out);
+
+    frame = quotad_aggregator_alloc_frame(req);
+    if (!frame)
+        goto out;
+
+    frame->root->op = req->procnum;
+
+    frame->root->uid = req->uid;
+    frame->root->gid = req->gid;
+    frame->root->pid = req->pid;
+
+    frame->root->lk_owner = req->lk_owner;
+
+    frame->local = req;
+out:
+    return frame;
+}
diff --git a/xlators/features/quota/src/quotad-helpers.h b/xlators/features/quota/src/quotad-helpers.h
new file mode 100644
index 00000000000..bcb39fe845e
--- /dev/null
+++ b/xlators/features/quota/src/quotad-helpers.h
@@ -0,0 +1,24 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef QUOTAD_HELPERS_H
+#define QUOTAD_HELPERS_H
+
+#include "rpcsvc.h"
+#include "quota.h"
+#include "quotad-aggregator.h"
+
+void
+quotad_aggregator_free_state(quotad_aggregator_state_t *state);
+
+call_frame_t *
+quotad_aggregator_get_frame_from_req(rpcsvc_request_t *req);
+
+#endif
diff --git a/xlators/features/quota/src/quotad.c b/xlators/features/quota/src/quotad.c
new file mode 100644
index 00000000000..643f25c9c2a
--- /dev/null
+++ b/xlators/features/quota/src/quotad.c
@@ -0,0 +1,245 @@
+/*
+   Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include "quota.h"
+#include "quotad-aggregator.h"
+
+int
+qd_notify(xlator_t *this, int32_t event, void *data, ...)
+{
+    switch (event) {
+        case GF_EVENT_PARENT_UP:
+            quotad_aggregator_init(this);
+    }
+
+    default_notify(this, event, data);
+    return 0;
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    if (!this)
+        return ret;
+
+    ret = xlator_mem_acct_init(this, gf_quota_mt_end + 1);
+
+    if (0 != ret) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "Memory accounting "
+               "init failed");
+        return ret;
+    }
+
+    return ret;
+}
+
+int32_t
+qd_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, inode_t *inode, struct iatt *buf, dict_t *xdata,
+              struct iatt *postparent)
+{
+    quotad_aggregator_lookup_cbk_t lookup_cbk = NULL;
+    gfs3_lookup_rsp rsp = {
+        0,
+    };
+
+    lookup_cbk = cookie;
+
+    rsp.op_ret = op_ret;
+    rsp.op_errno = op_errno;
+
+    gf_stat_from_iatt(&rsp.postparent, postparent);
+
+    GF_PROTOCOL_DICT_SERIALIZE(this, xdata, (&rsp.xdata.xdata_val),
+                               rsp.xdata.xdata_len, rsp.op_errno, out);
+
+    gf_stat_from_iatt(&rsp.stat, buf);
+
+out:
+    lookup_cbk(this, frame, &rsp);
+
+    GF_FREE(rsp.xdata.xdata_val);
+
+    inode_unref(inode);
+
+    return 0;
+}
+
+xlator_t *
+qd_find_subvol(xlator_t *this, char *volume_uuid)
+{
+    xlator_list_t *child = NULL;
+    xlator_t *subvol = NULL;
+    char key[1024];
+    int keylen = 0;
+    char *optstr = NULL;
+
+    if (!this || !volume_uuid)
+        goto out;
+
+    for (child = this->children; child; child = child->next) {
+        keylen = snprintf(key, sizeof(key), "%s.volume-id",
+                          child->xlator->name);
+        if (dict_get_strn(this->options, key, keylen, &optstr) < 0)
+            continue;
+
+        if (strcmp(optstr, volume_uuid) == 0) {
+            subvol = child->xlator;
+            break;
+        }
+    }
+
+out:
+    return subvol;
+}
+
+int
+qd_nameless_lookup(xlator_t *this, call_frame_t *frame, char *gfid,
+                   dict_t *xdata, char *volume_uuid,
+                   quotad_aggregator_lookup_cbk_t lookup_cbk)
+{
+    gfs3_lookup_rsp rsp = {
+        0,
+    };
+    int op_errno = 0, ret = -1;
+    loc_t loc = {
+        0,
+    };
+    quotad_aggregator_state_t *state = NULL;
+    xlator_t *subvol = NULL;
+
+    state = frame->root->state;
+
+    frame->root->op = GF_FOP_LOOKUP;
+
+    loc.inode = inode_new(state->itable);
+    if (loc.inode == NULL) {
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    memcpy(loc.gfid, gfid, 16);
+
+    ret = dict_set_int8(xdata, QUOTA_READ_ONLY_KEY, 1);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, Q_MSG_ENOMEM,
+               "dict set failed");
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    subvol = qd_find_subvol(this, volume_uuid);
+    if (subvol == NULL) {
+        op_errno = EINVAL;
+        goto out;
+    }
+
+    STACK_WIND_COOKIE(frame, qd_lookup_cbk, lookup_cbk, subvol,
+                      subvol->fops->lookup, &loc, xdata);
+    return 0;
+
+out:
+    rsp.op_ret = -1;
+    rsp.op_errno = op_errno;
+
+    lookup_cbk(this, frame, &rsp);
+
+    inode_unref(loc.inode);
+    return 0;
+}
+
+int
+qd_reconfigure(xlator_t *this, dict_t *options)
+{
+    /* As of now quotad is restarted upon alteration of volfile */
+    return 0;
+}
+
+void
+qd_fini(xlator_t *this)
+{
+    quota_priv_t *priv = NULL;
+
+    if (this == NULL || this->private == NULL)
+        goto out;
+
+    priv = this->private;
+
+    if (priv->rpcsvc) {
+        GF_FREE(priv->rpcsvc);
+        priv->rpcsvc = NULL;
+    }
+
+    GF_FREE(priv);
+
+out:
+    return;
+}
+
+int32_t
+qd_init(xlator_t *this)
+{
+    int32_t ret = -1;
+    quota_priv_t *priv = NULL;
+
+    if (NULL == this->children) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "FATAL: quota (%s) not configured for min of 1 child",
+               this->name);
+        ret = -1;
+        goto err;
+    }
+
+    QUOTA_ALLOC_OR_GOTO(priv, quota_priv_t, err);
+    LOCK_INIT(&priv->lock);
+
+    this->private = priv;
+
+    ret = 0;
+err:
+    if (ret) {
+        GF_FREE(priv);
+    }
+    return ret;
+}
+
+struct xlator_fops fops = {};
+
+struct xlator_cbks cbks = {};
+
+struct volume_options options[] = {
+    {.key = {"transport-type"},
+     .value = {"rpc", "rpc-over-rdma", "tcp", "socket", "ib-verbs", "unix",
+               "ib-sdp", "tcp/server", "ib-verbs/server", "rdma",
+               "rdma*([ \t]),*([ \t])socket", "rdma*([ \t]),*([ \t])tcp",
+               "tcp*([ \t]),*([ \t])rdma", "socket*([ \t]),*([ \t])rdma"},
+     .type = GF_OPTION_TYPE_STR},
+    {
+        .key = {"transport.*"},
+        .type = GF_OPTION_TYPE_ANY,
+    },
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = qd_init,
+    .fini = qd_fini,
+    .reconfigure = qd_reconfigure,
+    .notify = qd_notify,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1},
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "quotad",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/features/read-only/src/Makefile.am b/xlators/features/read-only/src/Makefile.am
index 15f49966ff7..e4a2017ef0d 100644
--- a/xlators/features/read-only/src/Makefile.am
+++ b/xlators/features/read-only/src/Makefile.am
@@ -1,13 +1,23 @@
-xlator_LTLIBRARIES = read-only.la
+xlator_LTLIBRARIES = read-only.la worm.la
+
 xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
 
-read_only_la_LDFLAGS = -module -avoidversion
+noinst_HEADERS = read-only.h read-only-mem-types.h read-only-common.h worm-helper.h
+
+read_only_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
 
-read_only_la_SOURCES = read-only.c
+read_only_la_SOURCES = read-only.c read-only-common.c
 read_only_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la 
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS) \
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
+worm_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+worm_la_SOURCES = read-only-common.c worm-helper.c worm.c
+worm_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
 
 CLEANFILES = 
 
diff --git a/xlators/features/read-only/src/read-only-common.c b/xlators/features/read-only/src/read-only-common.c
new file mode 100644
index 00000000000..9640e7e3eee
--- /dev/null
+++ b/xlators/features/read-only/src/read-only-common.c
@@ -0,0 +1,406 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include "read-only.h"
+#include "read-only-mem-types.h"
+#include <glusterfs/defaults.h>
+
+gf_boolean_t
+is_readonly_or_worm_enabled(call_frame_t *frame, xlator_t *this)
+{
+    read_only_priv_t *priv = NULL;
+    gf_boolean_t readonly_or_worm_enabled = _gf_false;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    readonly_or_worm_enabled = priv->readonly_or_worm_enabled;
+
+    if (frame->root->pid < GF_CLIENT_PID_MAX)
+        readonly_or_worm_enabled = _gf_false;
+
+    return readonly_or_worm_enabled;
+}
+
+static int
+_check_key_is_zero_filled(dict_t *d, char *k, data_t *v, void *tmp)
+{
+    if (mem_0filled((const char *)v->data, v->len)) {
+        /* -1 means, no more iterations, treat as 'break' */
+        return -1;
+    }
+    return 0;
+}
+
+int32_t
+ro_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+           gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+    gf_boolean_t allzero = _gf_false;
+    int ret = 0;
+
+    ret = dict_foreach(dict, _check_key_is_zero_filled, NULL);
+    if (ret == 0)
+        allzero = _gf_true;
+
+    if (is_readonly_or_worm_enabled(frame, this) && !allzero)
+        STACK_UNWIND_STRICT(xattrop, frame, -1, EROFS, NULL, xdata);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->xattrop, loc, flags, dict,
+                        xdata);
+    return 0;
+}
+
+int32_t
+ro_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+            gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+    gf_boolean_t allzero = _gf_false;
+    int ret = 0;
+
+    ret = dict_foreach(dict, _check_key_is_zero_filled, NULL);
+    if (ret == 0)
+        allzero = _gf_true;
+
+    if (is_readonly_or_worm_enabled(frame, this) && !allzero)
+        STACK_UNWIND_STRICT(fxattrop, frame, -1, EROFS, NULL, xdata);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->fxattrop, fd, flags, dict,
+                        xdata);
+
+    return 0;
+}
+
+int32_t
+ro_entrylk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+           const char *basename, entrylk_cmd cmd, entrylk_type type,
+           dict_t *xdata)
+{
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->entrylk,
+                    volume, loc, basename, cmd, type, xdata);
+
+    return 0;
+}
+
+int32_t
+ro_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+            const char *basename, entrylk_cmd cmd, entrylk_type type,
+            dict_t *xdata)
+{
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->fentrylk,
+                    volume, fd, basename, cmd, type, xdata);
+
+    return 0;
+}
+
+int32_t
+ro_inodelk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+           int32_t cmd, struct gf_flock *lock, dict_t *xdata)
+{
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->inodelk,
+                    volume, loc, cmd, lock, xdata);
+
+    return 0;
+}
+
+int32_t
+ro_finodelk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+            int32_t cmd, struct gf_flock *lock, dict_t *xdata)
+{
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->finodelk,
+                    volume, fd, cmd, lock, xdata);
+
+    return 0;
+}
+
+int32_t
+ro_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int cmd,
+      struct gf_flock *flock, dict_t *xdata)
+{
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->lk, fd,
+                    cmd, flock, xdata);
+
+    return 0;
+}
+
+int32_t
+ro_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *stbuf,
+           int32_t valid, dict_t *xdata)
+{
+    if (is_readonly_or_worm_enabled(frame, this))
+        STACK_UNWIND_STRICT(setattr, frame, -1, EROFS, NULL, NULL, xdata);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->setattr, loc, stbuf, valid,
+                        xdata);
+
+    return 0;
+}
+
+int32_t
+ro_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
+            int32_t valid, dict_t *xdata)
+{
+    if (is_readonly_or_worm_enabled(frame, this))
+        STACK_UNWIND_STRICT(fsetattr, frame, -1, EROFS, NULL, NULL, xdata);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->fsetattr, fd, stbuf, valid,
+                        xdata);
+
+    return 0;
+}
+
+int32_t
+ro_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+            dict_t *xdata)
+{
+    if (is_readonly_or_worm_enabled(frame, this))
+        STACK_UNWIND_STRICT(truncate, frame, -1, EROFS, NULL, NULL, xdata);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+
+    return 0;
+}
+
+int32_t
+ro_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             dict_t *xdata)
+{
+    if (is_readonly_or_worm_enabled(frame, this))
+        STACK_UNWIND_STRICT(ftruncate, frame, -1, EROFS, NULL, NULL, xdata);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+
+    return 0;
+}
+
+int32_t
+ro_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+             off_t offset, size_t len, dict_t *xdata)
+{
+    if (is_readonly_or_worm_enabled(frame, this))
+        STACK_UNWIND_STRICT(fallocate, frame, -1, EROFS, NULL, NULL, xdata);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->fallocate, fd, mode, offset,
+                        len, xdata);
+    return 0;
+}
+
+int
+ro_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+         dev_t rdev, mode_t umask, dict_t *xdata)
+{
+    if (is_readonly_or_worm_enabled(frame, this))
+        STACK_UNWIND_STRICT(mknod, frame, -1, EROFS, NULL, NULL, NULL, NULL,
+                            xdata);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->mknod, loc, mode, rdev, umask,
+                        xdata);
+
+    return 0;
+}
+
+int
+ro_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+         mode_t umask, dict_t *xdata)
+{
+    if (is_readonly_or_worm_enabled(frame, this))
+        STACK_UNWIND_STRICT(mkdir, frame, -1, EROFS, NULL, NULL, NULL, NULL,
+                            xdata);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->mkdir, loc, mode, umask,
+                        xdata);
+
+    return 0;
+}
+
+int32_t
+ro_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+          dict_t *xdata)
+{
+    if (is_readonly_or_worm_enabled(frame, this))
+        STACK_UNWIND_STRICT(unlink, frame, -1, EROFS, NULL, NULL, xdata);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->unlink, loc, xflag, xdata);
+
+    return 0;
+}
+
+int
+ro_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+         dict_t *xdata)
+{
+    if (is_readonly_or_worm_enabled(frame, this))
+        STACK_UNWIND_STRICT(rmdir, frame, -1, EROFS, NULL, NULL, xdata);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->rmdir, loc, flags, xdata);
+
+    return 0;
+}
+
+int
+ro_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+           loc_t *loc, mode_t umask, dict_t *xdata)
+{
+    if (is_readonly_or_worm_enabled(frame, this))
+        STACK_UNWIND_STRICT(symlink, frame, -1, EROFS, NULL, NULL, NULL, NULL,
+                            xdata);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->symlink, linkpath, loc, umask,
+                        xdata);
+
+    return 0;
+}
+
+int32_t
+ro_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+          dict_t *xdata)
+{
+    if (is_readonly_or_worm_enabled(frame, this))
+        STACK_UNWIND_STRICT(rename, frame, -1, EROFS, NULL, NULL, NULL, NULL,
+                            NULL, xdata);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
+
+    return 0;
+}
+
+int32_t
+ro_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+        dict_t *xdata)
+{
+    if (is_readonly_or_worm_enabled(frame, this))
+        STACK_UNWIND_STRICT(link, frame, -1, EROFS, NULL, NULL, NULL, NULL,
+                            xdata);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->link,
+                        oldloc, newloc, xdata);
+
+    return 0;
+}
+
+int32_t
+ro_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+          mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+    if (is_readonly_or_worm_enabled(frame, this))
+        STACK_UNWIND_STRICT(create, frame, -1, EROFS, NULL, NULL, NULL, NULL,
+                            NULL, xdata);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->create, loc, flags, mode,
+                        umask, fd, xdata);
+
+    return 0;
+}
+
+static int32_t
+ro_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, fd_t *fd, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(open, frame, op_ret, op_errno, fd, xdata);
+    return 0;
+}
+
+int32_t
+ro_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+        fd_t *fd, dict_t *xdata)
+{
+    if (is_readonly_or_worm_enabled(frame, this) &&
+        (((flags & O_ACCMODE) == O_WRONLY) ||
+         ((flags & O_ACCMODE) == O_RDWR))) {
+        STACK_UNWIND_STRICT(open, frame, -1, EROFS, NULL, xdata);
+        return 0;
+    }
+
+    STACK_WIND(frame, ro_open_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->open, loc, flags, fd, xdata);
+    return 0;
+}
+
+int32_t
+ro_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+             int32_t flags, dict_t *xdata)
+{
+    if (is_readonly_or_worm_enabled(frame, this))
+        STACK_UNWIND_STRICT(fsetxattr, frame, -1, EROFS, xdata);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags,
+                        xdata);
+
+    return 0;
+}
+
+int32_t
+ro_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+            dict_t *xdata)
+{
+    if (is_readonly_or_worm_enabled(frame, this))
+        STACK_UNWIND_STRICT(fsyncdir, frame, -1, EROFS, xdata);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->fsyncdir, fd, flags, xdata);
+
+    return 0;
+}
+
+int32_t
+ro_writev(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
+          int32_t count, off_t off, uint32_t flags, struct iobref *iobref,
+          dict_t *xdata)
+{
+    if (is_readonly_or_worm_enabled(frame, this))
+        STACK_UNWIND_STRICT(writev, frame, -1, EROFS, NULL, NULL, xdata);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->writev, fd, vector, count, off,
+                        flags, iobref, xdata);
+
+    return 0;
+}
+
+int32_t
+ro_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+            int32_t flags, dict_t *xdata)
+{
+    if (is_readonly_or_worm_enabled(frame, this))
+        STACK_UNWIND_STRICT(setxattr, frame, -1, EROFS, xdata);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->setxattr, loc, dict, flags,
+                        xdata);
+
+    return 0;
+}
+
+int32_t
+ro_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               const char *name, dict_t *xdata)
+{
+    if (is_readonly_or_worm_enabled(frame, this))
+        STACK_UNWIND_STRICT(removexattr, frame, -1, EROFS, xdata);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->removexattr, loc, name, xdata);
+
+    return 0;
+}
diff --git a/xlators/features/read-only/src/read-only-common.h b/xlators/features/read-only/src/read-only-common.h
new file mode 100644
index 00000000000..5561961ffa2
--- /dev/null
+++ b/xlators/features/read-only/src/read-only-common.h
@@ -0,0 +1,121 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+gf_boolean_t
+is_readonly_or_worm_enabled(call_frame_t *frame, xlator_t *this);
+
+int32_t
+ro_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+           gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata);
+
+int32_t
+ro_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+            gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata);
+
+int32_t
+ro_entrylk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+           const char *basename, entrylk_cmd cmd, entrylk_type type,
+           dict_t *xdata);
+
+int32_t
+ro_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+            const char *basename, entrylk_cmd cmd, entrylk_type type,
+            dict_t *xdata);
+
+int32_t
+ro_inodelk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+           int32_t cmd, struct gf_flock *lock, dict_t *xdata);
+
+int32_t
+ro_finodelk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+            int32_t cmd, struct gf_flock *lock, dict_t *xdata);
+
+int32_t
+ro_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int cmd,
+      struct gf_flock *flock, dict_t *xdata);
+
+int32_t
+ro_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *stbuf,
+           int32_t valid, dict_t *xdata);
+
+int32_t
+ro_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
+            int32_t valid, dict_t *xdata);
+
+int32_t
+ro_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+            dict_t *xdata);
+
+int32_t
+ro_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             dict_t *xdata);
+
+int
+ro_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+         dev_t rdev, mode_t umask, dict_t *xdata);
+
+int
+ro_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+         mode_t umask, dict_t *xdata);
+
+int32_t
+ro_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+          dict_t *xdata);
+
+int
+ro_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+         dict_t *xdata);
+
+int
+ro_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+           loc_t *loc, mode_t umask, dict_t *xdata);
+
+int32_t
+ro_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+          dict_t *xdata);
+
+int32_t
+ro_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+        dict_t *xdata);
+
+int32_t
+ro_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+          mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata);
+
+int32_t
+ro_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+        fd_t *fd, dict_t *xdata);
+
+int32_t
+ro_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+             int32_t flags, dict_t *xdata);
+
+int32_t
+ro_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+            dict_t *xdata);
+
+int32_t
+ro_writev(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
+          int32_t count, off_t off, uint32_t flags, struct iobref *iobref,
+          dict_t *xdata);
+
+int32_t
+ro_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+            int32_t flags, dict_t *xdata);
+
+int32_t
+ro_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               const char *name, dict_t *xdata);
+
+int32_t
+ro_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+             off_t offset, size_t len, dict_t *xdata);
diff --git a/xlators/features/read-only/src/read-only-mem-types.h b/xlators/features/read-only/src/read-only-mem-types.h
new file mode 100644
index 00000000000..c67d6c02cd0
--- /dev/null
+++ b/xlators/features/read-only/src/read-only-mem-types.h
@@ -0,0 +1,20 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __READONLY_MEM_TYPES_H__
+#define __READONLY_MEM_TYPES_H__
+
+#include <glusterfs/mem-types.h>
+
+enum gf_read_only_mem_types_ {
+    gf_read_only_mt_priv_t = gf_common_mt_end + 1,
+    gf_read_only_mt_end
+};
+#endif
diff --git a/xlators/features/read-only/src/read-only.c b/xlators/features/read-only/src/read-only.c
index 13e01c8ce4d..48654998e63 100644
--- a/xlators/features/read-only/src/read-only.c
+++ b/xlators/features/read-only/src/read-only.c
@@ -1,306 +1,144 @@
 /*
-  Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
 */
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "xlator.h"
-#include "defaults.h"
+#include "read-only-common.h"
+#include "read-only-mem-types.h"
+#include "read-only.h"
 
 int32_t
-ro_xattrop (call_frame_t *frame, xlator_t *this, loc_t *loc,
-            gf_xattrop_flags_t flags, dict_t *dict)
+mem_acct_init(xlator_t *this)
 {
-        STACK_UNWIND_STRICT (xattrop, frame, -1, EROFS, NULL);
-        return 0;
-}
+    int ret = -1;
 
-int32_t
-ro_fxattrop (call_frame_t *frame, xlator_t *this,
-             fd_t *fd, gf_xattrop_flags_t flags, dict_t *dict)
-{
-        STACK_UNWIND_STRICT (fxattrop, frame, -1, EROFS, NULL);
-        return 0;
-}
+    ret = xlator_mem_acct_init(this, gf_read_only_mt_end + 1);
+    if (ret)
+        gf_log(this->name, GF_LOG_ERROR,
+               "Memory accounting "
+               "initialization failed.");
 
-int32_t
-ro_entrylk (call_frame_t *frame, xlator_t *this, const char *volume,
-            loc_t *loc, const char *basename, entrylk_cmd cmd,
-            entrylk_type type)
-{
-        STACK_UNWIND_STRICT (entrylk, frame, -1, EROFS);
-        return 0;
+    return ret;
 }
 
 int32_t
-ro_fentrylk (call_frame_t *frame, xlator_t *this, const char *volume,
-             fd_t *fd, const char *basename, entrylk_cmd cmd, entrylk_type type)
+init(xlator_t *this)
 {
-        STACK_UNWIND_STRICT (fentrylk, frame, -1, EROFS);
-        return 0;
-}
+    int ret = -1;
+    read_only_priv_t *priv = NULL;
 
-int32_t
-ro_inodelk (call_frame_t *frame, xlator_t *this, const char *volume,
-            loc_t *loc, int32_t cmd, struct gf_flock *lock)
-{
-        STACK_UNWIND_STRICT (inodelk, frame, -1, EROFS);
-        return 0;
-}
+    if (!this->children || this->children->next) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "translator not configured with exactly one child");
+        return -1;
+    }
 
-int32_t
-ro_finodelk (call_frame_t *frame, xlator_t *this, const char *volume,
-             fd_t *fd, int32_t cmd, struct gf_flock *lock)
-{
-        STACK_UNWIND_STRICT (finodelk, frame, -1, EROFS);
-        return 0;
-}
+    if (!this->parents) {
+        gf_log(this->name, GF_LOG_WARNING, "dangling volume. check volfile ");
+    }
 
-int32_t
-ro_lk (call_frame_t *frame, xlator_t *this, fd_t *fd, int cmd,
-       struct gf_flock *flock)
-{
-        STACK_UNWIND_STRICT (lk, frame, -1, EROFS, NULL);
-        return 0;
-}
+    priv = GF_CALLOC(1, sizeof(*priv), gf_read_only_mt_priv_t);
+    if (!priv)
+        goto out;
 
-int32_t
-ro_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-            struct iatt *stbuf, int32_t valid)
-{
-        STACK_UNWIND_STRICT (setattr, frame, -1, EROFS, NULL, NULL);
-	return 0;
-}
-
-int32_t
-ro_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
-             struct iatt *stbuf, int32_t valid)
-{
-        STACK_UNWIND_STRICT (fsetattr, frame, -1, EROFS, NULL, NULL);
-        return 0;
-}
+    this->private = priv;
 
+    GF_OPTION_INIT("read-only", priv->readonly_or_worm_enabled, bool, out);
 
-int32_t
-ro_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset)
-{
-        STACK_UNWIND_STRICT (truncate, frame, -1, EROFS, NULL, NULL);
-	return 0;
-}
-
-int32_t
-ro_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset)
-{
-        STACK_UNWIND_STRICT (ftruncate, frame, -1, EROFS, NULL, NULL);
-	return 0;
-}
-
-int
-ro_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
-          dev_t rdev, dict_t *params)
-{
-        STACK_UNWIND_STRICT (mknod, frame, -1, EROFS, NULL, NULL, NULL, NULL);
-        return 0;
-}
-
-
-int
-ro_mkdir (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
-          dict_t *params)
-{
-        STACK_UNWIND_STRICT (mkdir, frame, -1, EROFS, NULL, NULL, NULL, NULL);
-        return 0;
+    ret = 0;
+out:
+    return ret;
 }
 
-int32_t
-ro_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc)
-{
-        STACK_UNWIND_STRICT (unlink, frame, -1, EROFS, NULL, NULL);
-        return 0;
-}
-
-
 int
-ro_rmdir (call_frame_t *frame, xlator_t *this, loc_t *loc, int flags)
+reconfigure(xlator_t *this, dict_t *options)
 {
-        STACK_UNWIND_STRICT (rmdir, frame, -1, EROFS, NULL, NULL);
-        return 0;
-}
+    read_only_priv_t *priv = NULL;
+    int ret = -1;
+    gf_boolean_t readonly_or_worm_enabled = _gf_false;
 
+    priv = this->private;
+    GF_ASSERT(priv);
 
-int
-ro_symlink (call_frame_t *frame, xlator_t *this, const char *linkpath,
-            loc_t *loc, dict_t *params)
-{
-        STACK_UNWIND_STRICT (symlink, frame, -1, EROFS, NULL, NULL, NULL, NULL);
-        return 0;
+    GF_OPTION_RECONF("read-only", readonly_or_worm_enabled, options, bool, out);
+    priv->readonly_or_worm_enabled = readonly_or_worm_enabled;
+    ret = 0;
+out:
+    gf_log(this->name, GF_LOG_DEBUG, "returning %d", ret);
+    return ret;
 }
 
-
-
-int32_t
-ro_rename (call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc)
-{
-        STACK_UNWIND_STRICT (rename, frame, -1, EROFS, NULL, NULL, NULL, NULL,
-                             NULL);
-        return 0;
-}
-
-
-int32_t
-ro_link (call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc)
-{
-        STACK_UNWIND_STRICT (link, frame, -1, EROFS, NULL, NULL, NULL, NULL);
-        return 0;
-}
-
-int32_t
-ro_create (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
-           mode_t mode, fd_t *fd, dict_t *params)
-{
-        STACK_UNWIND_STRICT (create, frame, -1, EROFS, NULL, NULL, NULL,
-                             NULL, NULL);
-        return 0;
-}
-
-
-static int32_t
-ro_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
-             int32_t op_errno, fd_t *fd)
-{
-	STACK_UNWIND_STRICT (open, frame, op_ret, op_errno, fd);
-	return 0;
-}
-
-int32_t
-ro_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
-         fd_t *fd, int32_t wbflags)
-{
-        if (((flags & O_ACCMODE) == O_WRONLY) ||
-              ((flags & O_ACCMODE) == O_RDWR)) {
-                STACK_UNWIND_STRICT (open, frame, -1, EROFS, NULL);
-                return 0;
-	}
-
-	STACK_WIND (frame, ro_open_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->open, loc, flags, fd, wbflags);
-	return 0;
-}
-
-int32_t
-ro_fsetxattr (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
-              int32_t flags)
-{
-        STACK_UNWIND_STRICT (fsetxattr, frame, -1, EROFS);
-	return 0;
-}
-
-int32_t
-ro_fsyncdir (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags)
-{
-        STACK_UNWIND_STRICT (fsyncdir, frame, -1, EROFS);
-	return 0;
-}
-
-int32_t
-ro_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
-           int32_t count, off_t off, struct iobref *iobref)
-{
-        STACK_UNWIND_STRICT (writev, frame, -1, EROFS, NULL, NULL);
-        return 0;
-}
-
-
-int32_t
-ro_setxattr (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
-             int32_t flags)
-{
-        STACK_UNWIND_STRICT (setxattr, frame, -1, EROFS);
-        return 0;
-}
-
-int32_t
-ro_removexattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                const char *name)
+void
+fini(xlator_t *this)
 {
-        STACK_UNWIND_STRICT (removexattr, frame, -1, EROFS);
-        return 0;
-}
+    read_only_priv_t *priv = NULL;
 
-int32_t
-init (xlator_t *this)
-{
-	if (!this->children || this->children->next) {
-		gf_log (this->name, GF_LOG_ERROR,
-			"translator not configured with exactly one child");
-		return -1;
-	}
+    priv = this->private;
+    if (!priv)
+        return;
 
-	if (!this->parents) {
-		gf_log (this->name, GF_LOG_WARNING,
-			"dangling volume. check volfile ");
-	}
+    this->private = NULL;
+    GF_FREE(priv);
 
-	return 0;
+    return;
 }
 
-
-void
-fini (xlator_t *this)
-{
-	return;
-}
-
-
 struct xlator_fops fops = {
-        .mknod       = ro_mknod,
-        .mkdir       = ro_mkdir,
-        .unlink      = ro_unlink,
-        .rmdir       = ro_rmdir,
-        .symlink     = ro_symlink,
-        .rename      = ro_rename,
-        .link        = ro_link,
-        .truncate    = ro_truncate,
-        .open        = ro_open,
-        .writev      = ro_writev,
-        .setxattr    = ro_setxattr,
-        .fsetxattr   = ro_fsetxattr,
-        .removexattr = ro_removexattr,
-        .fsyncdir    = ro_fsyncdir,
-        .ftruncate   = ro_ftruncate,
-        .create      = ro_create,
-        .setattr     = ro_setattr,
-        .fsetattr    = ro_fsetattr,
-        .xattrop     = ro_xattrop,
-        .fxattrop    = ro_fxattrop,
-        .inodelk     = ro_inodelk,
-        .finodelk    = ro_finodelk,
-        .entrylk     = ro_entrylk,
-        .fentrylk    = ro_fentrylk,
-        .lk          = ro_lk,
+    .mknod = ro_mknod,
+    .mkdir = ro_mkdir,
+    .unlink = ro_unlink,
+    .rmdir = ro_rmdir,
+    .symlink = ro_symlink,
+    .rename = ro_rename,
+    .link = ro_link,
+    .truncate = ro_truncate,
+    .open = ro_open,
+    .writev = ro_writev,
+    .setxattr = ro_setxattr,
+    .fsetxattr = ro_fsetxattr,
+    .removexattr = ro_removexattr,
+    .fsyncdir = ro_fsyncdir,
+    .ftruncate = ro_ftruncate,
+    .create = ro_create,
+    .setattr = ro_setattr,
+    .fsetattr = ro_fsetattr,
+    .xattrop = ro_xattrop,
+    .fxattrop = ro_fxattrop,
+    .inodelk = ro_inodelk,
+    .finodelk = ro_finodelk,
+    .entrylk = ro_entrylk,
+    .fentrylk = ro_fentrylk,
+    .lk = ro_lk,
+    .fallocate = ro_fallocate,
 };
 
-struct xlator_cbks cbks = {
-};
+struct xlator_cbks cbks = {};
 
 struct volume_options options[] = {
-	{ .key = {NULL} },
+    {.key = {"read-only"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     /*.validate_fn = validate_boolean,*/
+     .op_version = {1},
+     .flags = OPT_FLAG_SETTABLE,
+     .description = "When \"on\", makes a volume read-only. It is turned "
+                    "\"off\" by default."},
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "read-only",
+    .category = GF_TECH_PREVIEW,
 };
diff --git a/xlators/features/read-only/src/read-only.h b/xlators/features/read-only/src/read-only.h
new file mode 100644
index 00000000000..aced5d3c577
--- /dev/null
+++ b/xlators/features/read-only/src/read-only.h
@@ -0,0 +1,37 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __READONLY_H__
+#define __READONLY_H__
+
+#include <stdint.h>               // for uint64_t, uint8_t
+#include <sys/time.h>             // for time_t
+#include "glusterfs/glusterfs.h"  // for gf_boolean_t
+
+typedef struct {
+    uint8_t worm : 1;
+    uint8_t retain : 1;
+    uint8_t legal_hold : 1;
+    uint8_t ret_mode : 1;
+    int64_t ret_period;
+    int64_t auto_commit_period;
+} worm_reten_state_t;
+
+typedef struct {
+    gf_boolean_t readonly_or_worm_enabled;
+    gf_boolean_t worm_file;
+    gf_boolean_t worm_files_deletable;
+    int64_t reten_period;
+    int64_t com_period;
+    int reten_mode;
+    time_t start_time;
+} read_only_priv_t;
+
+#endif
diff --git a/xlators/features/read-only/src/worm-helper.c b/xlators/features/read-only/src/worm-helper.c
new file mode 100644
index 00000000000..df45f2a940b
--- /dev/null
+++ b/xlators/features/read-only/src/worm-helper.c
@@ -0,0 +1,395 @@
+/*
+   Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include "read-only-mem-types.h"
+#include "read-only.h"
+#include <glusterfs/xlator.h>
+#include <glusterfs/syncop.h>
+#include "worm-helper.h"
+
+/*Function to check whether file is read-only.
+ * The input *stbuf contains the attributes of the file, which is used to check
+ * the write protection bits for all the users of the file.
+ * Return true if all the write bits are disabled,false otherwise*/
+gf_boolean_t
+gf_worm_write_disabled(struct iatt *stbuf)
+{
+    gf_boolean_t ret = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("worm", stbuf, out);
+
+    if (stbuf->ia_prot.owner.write == 0 && stbuf->ia_prot.group.write == 0 &&
+        stbuf->ia_prot.other.write == 0)
+        ret = _gf_true;
+out:
+    return ret;
+}
+
+int32_t
+worm_init_state(xlator_t *this, gf_boolean_t fop_with_fd, void *file_ptr)
+{
+    int ret = -1;
+    uint64_t start_time = 0;
+    dict_t *dict = NULL;
+
+    GF_VALIDATE_OR_GOTO("worm", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, file_ptr, out);
+
+    start_time = gf_time();
+    dict = dict_new();
+    if (!dict) {
+        gf_log(this->name, GF_LOG_ERROR, "Error creating the dict");
+        goto out;
+    }
+    ret = dict_set_uint64(dict, "trusted.start_time", start_time);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR, "Error in setting the dict");
+        goto out;
+    }
+    if (fop_with_fd)
+        ret = syncop_fsetxattr(this, (fd_t *)file_ptr, dict, 0, NULL, NULL);
+    else
+        ret = syncop_setxattr(this, (loc_t *)file_ptr, dict, 0, NULL, NULL);
+out:
+    if (dict)
+        dict_unref(dict);
+    return ret;
+}
+
+/*Function to set the retention state for a file.
+ * It loads the WORM/Retention state into the retention_state pointer.*/
+int32_t
+worm_set_state(xlator_t *this, gf_boolean_t fop_with_fd, void *file_ptr,
+               worm_reten_state_t *retention_state, struct iatt *stbuf)
+{
+    read_only_priv_t *priv = NULL;
+    struct iatt stpre = {
+        0,
+    };
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("worm", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, file_ptr, out);
+    GF_VALIDATE_OR_GOTO(this->name, retention_state, out);
+    GF_VALIDATE_OR_GOTO(this->name, stbuf, out);
+
+    priv = this->private;
+    GF_ASSERT(priv);
+    retention_state->worm = 1;
+    retention_state->retain = 1;
+    retention_state->legal_hold = 0;
+    retention_state->ret_mode = priv->reten_mode;
+    retention_state->ret_period = priv->reten_period;
+    retention_state->auto_commit_period = priv->com_period;
+    if (fop_with_fd)
+        ret = syncop_fstat(this, (fd_t *)file_ptr, &stpre, NULL, NULL);
+    else
+        ret = syncop_stat(this, (loc_t *)file_ptr, &stpre, NULL, NULL);
+    if (ret)
+        goto out;
+    stbuf->ia_mtime = stpre.ia_mtime;
+    stbuf->ia_atime = gf_time() + retention_state->ret_period;
+
+    if (fop_with_fd)
+        ret = syncop_fsetattr(this, (fd_t *)file_ptr, stbuf, GF_SET_ATTR_ATIME,
+                              NULL, NULL, NULL, NULL);
+    else
+        ret = syncop_setattr(this, (loc_t *)file_ptr, stbuf, GF_SET_ATTR_ATIME,
+                             NULL, NULL, NULL, NULL);
+    if (ret)
+        goto out;
+
+    ret = gf_worm_set_xattr(this, retention_state, fop_with_fd, file_ptr);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR, "Error setting xattr");
+        goto out;
+    }
+    ret = 0;
+out:
+    return ret;
+}
+
+/*This function gets the state of the WORM/Retention xattr and loads it in the
+ * dict pointer.*/
+int32_t
+worm_get_state(xlator_t *this, gf_boolean_t fop_with_fd, void *file_ptr,
+               worm_reten_state_t *reten_state)
+{
+    dict_t *dict = NULL;
+    char *val = NULL;
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("worm", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, file_ptr, out);
+    GF_VALIDATE_OR_GOTO(this->name, reten_state, out);
+
+    if (fop_with_fd)
+        ret = syncop_fgetxattr(this, (fd_t *)file_ptr, &dict,
+                               "trusted.reten_state", NULL, NULL);
+    else
+        ret = syncop_getxattr(this, (loc_t *)file_ptr, &dict,
+                              "trusted.reten_state", NULL, NULL);
+    if (ret < 0 || !dict) {
+        ret = -1;
+        goto out;
+    }
+    ret = dict_get_str(dict, "trusted.reten_state", &val);
+    if (ret) {
+        ret = -2;
+        gf_log(this->name, GF_LOG_ERROR, "Empty val");
+    }
+    gf_worm_deserialize_state(val, reten_state);
+out:
+    if (dict)
+        dict_unref(dict);
+    return ret;
+}
+
+/*Function to lookup the current state of the WORM/Retention profile.
+ * Based on the retain value and the access time of the file, the transition
+ * from WORM/Retention to WORM is made.*/
+void
+gf_worm_state_lookup(xlator_t *this, gf_boolean_t fop_with_fd, void *file_ptr,
+                     worm_reten_state_t *reten_state, struct iatt *stbuf)
+{
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("worm", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, file_ptr, out);
+    GF_VALIDATE_OR_GOTO(this->name, reten_state, out);
+    GF_VALIDATE_OR_GOTO(this->name, stbuf, out);
+
+    stbuf->ia_atime -= reten_state->ret_period;
+    reten_state->retain = 0;
+    reten_state->ret_period = 0;
+    reten_state->auto_commit_period = 0;
+    ret = gf_worm_set_xattr(this, reten_state, fop_with_fd, file_ptr);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR, "Error setting xattr");
+        goto out;
+    }
+
+    if (fop_with_fd)
+        ret = syncop_fsetattr(this, (fd_t *)file_ptr, stbuf, GF_SET_ATTR_ATIME,
+                              NULL, NULL, NULL, NULL);
+    else
+        ret = syncop_setattr(this, (loc_t *)file_ptr, stbuf, GF_SET_ATTR_ATIME,
+                             NULL, NULL, NULL, NULL);
+    if (ret)
+        goto out;
+    gf_log(this->name, GF_LOG_INFO, "Retention state reset");
+out:
+    return;
+}
+
+/*This function serializes and stores the WORM/Retention state of a file in an
+ * uint64_t variable by setting the bits using the bitwise operations.*/
+void
+gf_worm_serialize_state(worm_reten_state_t *reten_state, char *val)
+{
+    uint32_t state = 0;
+
+    GF_VALIDATE_OR_GOTO("worm", reten_state, out);
+    GF_VALIDATE_OR_GOTO("worm", val, out);
+
+    state |= reten_state->worm << 0;
+    state |= reten_state->retain << 1;
+    state |= reten_state->legal_hold << 2;
+    state |= reten_state->ret_mode << 3;
+    sprintf(val, "%d/%" PRIu64 "/%" PRIu64, state, reten_state->ret_period,
+            reten_state->auto_commit_period);
+
+out:
+    return;
+}
+
+/*This function deserializes the data stored in the xattr of the file and loads
+ * the value to the reten_state structure.*/
+void
+gf_worm_deserialize_state(char *val, worm_reten_state_t *reten_state)
+{
+    char *token = NULL;
+    uint32_t state = 0;
+
+    GF_VALIDATE_OR_GOTO("worm", val, out);
+    GF_VALIDATE_OR_GOTO("worm", reten_state, out);
+
+    token = strtok(val, "/");
+    state = atoi(token);
+    reten_state->worm = (state >> 0) & 1;
+    reten_state->retain = (state >> 1) & 1;
+    reten_state->legal_hold = (state >> 2) & 1;
+    reten_state->ret_mode = (state >> 3) & 1;
+    token = strtok(NULL, "/");
+    reten_state->ret_period = atoi(token);
+    token = strtok(NULL, "/");
+    reten_state->auto_commit_period = atoi(token);
+
+out:
+    return;
+}
+
+/*Function to set the xattr for a file.
+ * If the xattr is already present then it will replace that.*/
+int32_t
+gf_worm_set_xattr(xlator_t *this, worm_reten_state_t *reten_state,
+                  gf_boolean_t fop_with_fd, void *file_ptr)
+{
+    char val[100] = "";
+    int ret = -1;
+    dict_t *dict = NULL;
+
+    GF_VALIDATE_OR_GOTO("worm", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, reten_state, out);
+    GF_VALIDATE_OR_GOTO(this->name, file_ptr, out);
+
+    gf_worm_serialize_state(reten_state, val);
+    dict = dict_new();
+    if (!dict) {
+        gf_log(this->name, GF_LOG_ERROR, "Error creating the dict");
+        goto out;
+    }
+    ret = dict_set_str(dict, "trusted.reten_state", val);
+    if (ret) {
+        gf_log(this->name, GF_LOG_ERROR, "Error in setting the dict");
+        goto out;
+    }
+    if (fop_with_fd)
+        ret = syncop_fsetxattr(this, (fd_t *)file_ptr, dict, 0, NULL, NULL);
+    else
+        ret = syncop_setxattr(this, (loc_t *)file_ptr, dict, 0, NULL, NULL);
+out:
+    if (dict)
+        dict_unref(dict);
+    return ret;
+}
+
+/*This function checks whether a file's timeout is happened for the state
+ * transition and if yes, then it will do the transition from the current state
+ * to the appropriate state. It also decides whether to continue or to block
+ * the FOP.
+ * Return:
+ * 0 : If the FOP should continue i.e., if the file is not in the WORM-Retained
+ *     state or if the FOP is unlink and the file is not in the Retained state.
+ * 1: If the FOP sholud block i.e., if the file is in WORM-Retained/WORM state.
+ * 2: Blocks the FOP if any operation fails while doing the state transition or
+ *    fails to get the state of the file.*/
+int
+gf_worm_state_transition(xlator_t *this, gf_boolean_t fop_with_fd,
+                         void *file_ptr, glusterfs_fop_t op)
+{
+    int op_errno = EROFS;
+    int ret = -1;
+    time_t now = 0;
+    uint64_t com_period = 0;
+    uint64_t start_time = 0;
+    dict_t *dict = NULL;
+    worm_reten_state_t reten_state = {
+        0,
+    };
+    read_only_priv_t *priv = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    if (fop_with_fd)
+        ret = syncop_fgetxattr(this, (fd_t *)file_ptr, &dict,
+                               "trusted.start_time", NULL, NULL);
+    else
+        ret = syncop_getxattr(this, (loc_t *)file_ptr, &dict,
+                              "trusted.start_time", NULL, NULL);
+    if (ret < 0 || !dict) {
+        op_errno = ret;
+        gf_msg(this->name, GF_LOG_ERROR, -ret, 0, "Error getting xattr");
+        goto out;
+    }
+    ret = dict_get_uint64(dict, "trusted.start_time", &start_time);
+    if (ret) {
+        op_errno = ret;
+        gf_msg(this->name, GF_LOG_ERROR, -ret, 0, "Error getting start time");
+        goto out;
+    }
+
+    com_period = priv->com_period;
+    if (fop_with_fd)
+        ret = syncop_fstat(this, (fd_t *)file_ptr, &stbuf, NULL, NULL);
+    else
+        ret = syncop_stat(this, (loc_t *)file_ptr, &stbuf, NULL, NULL);
+    if (ret) {
+        op_errno = ret;
+        gf_msg(this->name, GF_LOG_ERROR, -ret, 0, "Error getting file stat");
+        goto out;
+    }
+
+    ret = worm_get_state(this, fop_with_fd, file_ptr, &reten_state);
+    if (ret == -2) {
+        op_errno = ret;
+        gf_msg(this->name, GF_LOG_ERROR, -ret, 0,
+               "Error getting worm/retention state");
+        goto out;
+    }
+
+    now = gf_time();
+
+    if (ret == -1 && (now - start_time) >= com_period) {
+        if ((now - stbuf.ia_mtime) >= com_period) {
+            ret = worm_set_state(this, fop_with_fd, file_ptr, &reten_state,
+                                 &stbuf);
+            if (ret) {
+                op_errno = ret;
+                gf_msg(this->name, GF_LOG_ERROR, -ret, 0,
+                       "Error setting worm/retention state");
+                goto out;
+            }
+            goto out;
+        } else {
+            op_errno = 0;
+            goto out;
+        }
+    } else if (ret == -1 && (now - start_time) < com_period) {
+        op_errno = 0;
+        goto out;
+    } else if (reten_state.retain && ((now >= stbuf.ia_atime))) {
+        gf_worm_state_lookup(this, fop_with_fd, file_ptr, &reten_state, &stbuf);
+    }
+    if (reten_state.worm && !reten_state.retain && priv->worm_files_deletable &&
+        op == GF_FOP_UNLINK) {
+        op_errno = 0;
+        goto out;
+    }
+
+out:
+    if (dict)
+        dict_unref(dict);
+    return op_errno;
+}
+
+/*Function to check whether a file is independently WORMed (i.e., file level
+ * WORM is set on the file). */
+int32_t
+is_wormfile(xlator_t *this, gf_boolean_t fop_with_fd, void *file_ptr)
+{
+    int ret = -1;
+    dict_t *dict = NULL;
+
+    if (fop_with_fd)
+        ret = syncop_fgetxattr(this, (fd_t *)file_ptr, &dict,
+                               "trusted.worm_file", NULL, NULL);
+    else
+        ret = syncop_getxattr(this, (loc_t *)file_ptr, &dict,
+                              "trusted.worm_file", NULL, NULL);
+    if (dict) {
+        ret = 0;
+        dict_unref(dict);
+    }
+    return ret;
+}
diff --git a/xlators/features/read-only/src/worm-helper.h b/xlators/features/read-only/src/worm-helper.h
new file mode 100644
index 00000000000..b42f8d2b40c
--- /dev/null
+++ b/xlators/features/read-only/src/worm-helper.h
@@ -0,0 +1,44 @@
+/*
+   Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+gf_boolean_t
+gf_worm_write_disabled(struct iatt *stbuf);
+
+int32_t
+worm_init_state(xlator_t *this, gf_boolean_t fop_with_fd, void *file_ptr);
+
+int32_t
+worm_set_state(xlator_t *this, gf_boolean_t fop_with_fd, void *file_ptr,
+               worm_reten_state_t *retention_state, struct iatt *stbuf);
+
+int32_t
+worm_get_state(xlator_t *this, gf_boolean_t fop_with_fd, void *file_ptr,
+               worm_reten_state_t *reten_state);
+
+void
+gf_worm_state_lookup(xlator_t *this, gf_boolean_t fop_with_fd, void *file_ptr,
+                     worm_reten_state_t *reten_state, struct iatt *stbuf);
+
+void
+gf_worm_serialize_state(worm_reten_state_t *reten_state, char *val);
+
+void
+gf_worm_deserialize_state(char *val, worm_reten_state_t *reten_state);
+
+int32_t
+gf_worm_set_xattr(xlator_t *this, worm_reten_state_t *reten_state,
+                  gf_boolean_t fop_with_fd, void *file_ptr);
+
+int
+gf_worm_state_transition(xlator_t *this, gf_boolean_t fop_with_fd,
+                         void *file_ptr, glusterfs_fop_t op);
+
+int32_t
+is_wormfile(xlator_t *this, gf_boolean_t fop_with_fd, void *file_ptr);
diff --git a/xlators/features/read-only/src/worm.c b/xlators/features/read-only/src/worm.c
new file mode 100644
index 00000000000..1cc5526d5cd
--- /dev/null
+++ b/xlators/features/read-only/src/worm.c
@@ -0,0 +1,722 @@
+/*
+   Copyright (c) 2008-2012, 2016 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+#include "read-only-common.h"
+#include "read-only-mem-types.h"
+#include "read-only.h"
+#include <glusterfs/syncop.h>
+#include "worm-helper.h"
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    ret = xlator_mem_acct_init(this, gf_read_only_mt_end + 1);
+    if (ret)
+        gf_log(this->name, GF_LOG_ERROR,
+               "Memory accounting "
+               "initialization failed.");
+
+    return ret;
+}
+
+static int32_t
+worm_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+          fd_t *fd, dict_t *xdata)
+{
+    if (is_readonly_or_worm_enabled(frame, this) &&
+        (flags & (O_WRONLY | O_RDWR | O_APPEND | O_TRUNC))) {
+        STACK_UNWIND_STRICT(open, frame, -1, EROFS, NULL, NULL);
+        return 0;
+    }
+
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->open,
+                    loc, flags, fd, xdata);
+    return 0;
+}
+
+static int32_t
+worm_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+          dict_t *xdata)
+{
+    int op_errno = EROFS;
+    read_only_priv_t *priv = NULL;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+    if (is_readonly_or_worm_enabled(frame, this))
+        goto out;
+    if (!priv->worm_file || (frame->root->pid < 0)) {
+        op_errno = 0;
+        goto out;
+    }
+
+    gf_uuid_copy(oldloc->gfid, oldloc->inode->gfid);
+    if (is_wormfile(this, _gf_false, oldloc)) {
+        op_errno = 0;
+        goto out;
+    }
+    op_errno = gf_worm_state_transition(this, _gf_false, oldloc, GF_FOP_LINK);
+
+out:
+    if (op_errno) {
+        if (op_errno < 0)
+            op_errno = EROFS;
+        STACK_UNWIND_STRICT(link, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                            NULL);
+    } else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->link,
+                        oldloc, newloc, xdata);
+    return 0;
+}
+
+static int32_t
+worm_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+            dict_t *xdata)
+{
+    int op_errno = EROFS;
+    read_only_priv_t *priv = NULL;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+    if (is_readonly_or_worm_enabled(frame, this)) {
+        goto out;
+    }
+    if (!priv->worm_file || (frame->root->pid < 0)) {
+        op_errno = 0;
+        goto out;
+    }
+
+    gf_uuid_copy(loc->gfid, loc->inode->gfid);
+    if (is_wormfile(this, _gf_false, loc)) {
+        op_errno = 0;
+        goto out;
+    }
+    op_errno = gf_worm_state_transition(this, _gf_false, loc, GF_FOP_UNLINK);
+out:
+    if (op_errno) {
+        if (op_errno < 0)
+            op_errno = EROFS;
+        STACK_UNWIND_STRICT(unlink, frame, -1, op_errno, NULL, NULL, NULL);
+    } else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->unlink, loc, flags, xdata);
+    return 0;
+}
+
+static int32_t
+worm_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+            dict_t *xdata)
+{
+    int op_errno = EROFS;
+    read_only_priv_t *priv = NULL;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+    if (is_readonly_or_worm_enabled(frame, this))
+        goto out;
+    if (!priv->worm_file || (frame->root->pid < 0)) {
+        op_errno = 0;
+        goto out;
+    }
+
+    gf_uuid_copy(oldloc->gfid, oldloc->inode->gfid);
+    if (is_wormfile(this, _gf_false, oldloc)) {
+        op_errno = 0;
+        goto check_newloc;
+    }
+    op_errno = gf_worm_state_transition(this, _gf_false, oldloc, GF_FOP_RENAME);
+
+    if (op_errno == 0) {
+    check_newloc:
+        if (newloc->inode != NULL) {
+            gf_uuid_copy(newloc->gfid, newloc->inode->gfid);
+            if (is_wormfile(this, _gf_false, newloc)) {
+                op_errno = 0;
+                goto out;
+            }
+            op_errno = gf_worm_state_transition(this, _gf_false, newloc,
+                                                GF_FOP_RENAME);
+        }
+    }
+
+out:
+    if (op_errno) {
+        if (op_errno < 0)
+            op_errno = EROFS;
+        STACK_UNWIND_STRICT(rename, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                            NULL, NULL);
+    } else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
+    return 0;
+}
+
+static int32_t
+worm_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+              dict_t *xdata)
+{
+    int op_errno = EROFS;
+    read_only_priv_t *priv = NULL;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+    if (is_readonly_or_worm_enabled(frame, this))
+        goto out;
+    if (!priv->worm_file || (frame->root->pid < 0)) {
+        op_errno = 0;
+        goto out;
+    }
+
+    if (is_wormfile(this, _gf_false, loc)) {
+        op_errno = 0;
+        goto out;
+    }
+    op_errno = gf_worm_state_transition(this, _gf_false, loc, GF_FOP_TRUNCATE);
+
+out:
+    if (op_errno) {
+        if (op_errno < 0)
+            op_errno = EROFS;
+        STACK_UNWIND_STRICT(truncate, frame, -1, op_errno, NULL, NULL, NULL);
+    } else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+    return 0;
+}
+
+static int32_t
+worm_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+               dict_t *xdata)
+{
+    int op_errno = EROFS;
+    read_only_priv_t *priv = NULL;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+    if (is_readonly_or_worm_enabled(frame, this))
+        goto out;
+    if (!priv->worm_file || (frame->root->pid < 0)) {
+        op_errno = 0;
+        goto out;
+    }
+
+    if (is_wormfile(this, _gf_true, fd)) {
+        op_errno = 0;
+        goto out;
+    }
+    op_errno = gf_worm_state_transition(this, _gf_true, fd, GF_FOP_FTRUNCATE);
+
+out:
+    if (op_errno) {
+        if (op_errno < 0)
+            op_errno = EROFS;
+        STACK_UNWIND_STRICT(ftruncate, frame, -1, op_errno, NULL, NULL, NULL);
+    } else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+    return 0;
+}
+
+static int32_t
+worm_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+             struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    gf_boolean_t rd_only = _gf_false;
+    worm_reten_state_t reten_state = {
+        0,
+    };
+    struct iatt stpre = {
+        0,
+    };
+    read_only_priv_t *priv = NULL;
+    int op_errno = EROFS;
+    int ret = -1;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+    if (!priv->worm_file) {
+        op_errno = 0;
+        goto out;
+    }
+
+    if (is_wormfile(this, _gf_false, loc)) {
+        op_errno = 0;
+        goto out;
+    }
+    if (valid & GF_SET_ATTR_MODE) {
+        rd_only = gf_worm_write_disabled(stbuf);
+        if (!rd_only) {
+            op_errno = 0;
+            goto out;
+        }
+
+        ret = worm_set_state(this, _gf_false, loc, &reten_state, stbuf);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR, "Error setting worm state");
+            goto out;
+        }
+    } else if (valid & GF_SET_ATTR_ATIME) {
+        ret = worm_get_state(this, _gf_false, loc, &reten_state);
+        if (ret) {
+            op_errno = 0;
+            goto out;
+        }
+        if (reten_state.retain) {
+            ret = syncop_stat(this, loc, &stpre, NULL, NULL);
+            if (ret)
+                goto out;
+            if (reten_state.ret_mode == 0) {
+                if (stbuf->ia_atime < stpre.ia_mtime) {
+                    gf_log(this->name, GF_LOG_ERROR,
+                           "Cannot set atime less than "
+                           "the mtime for a WORM-Retained "
+                           "file");
+                    goto out;
+                }
+            } else {
+                if (stbuf->ia_atime < stpre.ia_atime) {
+                    gf_log(this->name, GF_LOG_ERROR,
+                           "Cannot decrease the atime of a"
+                           " WORM-Retained file in "
+                           "Enterprise mode");
+                    goto out;
+                }
+            }
+            reten_state.ret_period = reten_state.ret_period + stbuf->ia_atime -
+                                     stpre.ia_atime;
+            ret = gf_worm_set_xattr(this, &reten_state, _gf_false, loc);
+            if (ret) {
+                goto out;
+            }
+            stbuf->ia_mtime = stpre.ia_mtime;
+        }
+    }
+    op_errno = 0;
+
+out:
+    if (op_errno)
+        STACK_UNWIND_STRICT(setattr, frame, -1, EROFS, NULL, NULL, NULL);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->setattr, loc, stbuf, valid,
+                        xdata);
+    return 0;
+}
+
+static int32_t
+worm_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
+              int32_t valid, dict_t *xdata)
+{
+    gf_boolean_t rd_only = _gf_false;
+    worm_reten_state_t reten_state = {
+        0,
+    };
+    struct iatt stpre = {
+        0,
+    };
+    read_only_priv_t *priv = NULL;
+    int op_errno = EROFS;
+    int ret = -1;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+    if (!priv->worm_file) {
+        op_errno = 0;
+        goto out;
+    }
+
+    if (is_wormfile(this, _gf_true, fd)) {
+        op_errno = 0;
+        goto out;
+    }
+    if (valid & GF_SET_ATTR_MODE) {
+        rd_only = gf_worm_write_disabled(stbuf);
+        if (!rd_only) {
+            op_errno = 0;
+            goto out;
+        }
+
+        ret = worm_set_state(this, _gf_true, fd, &reten_state, stbuf);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR, "Error setting worm state");
+            goto out;
+        }
+    } else if (valid & GF_SET_ATTR_ATIME) {
+        ret = worm_get_state(this, _gf_true, fd, &reten_state);
+        if (ret) {
+            op_errno = 0;
+            goto out;
+        }
+        if (reten_state.retain) {
+            ret = syncop_fstat(this, fd, &stpre, NULL, NULL);
+            if (ret)
+                goto out;
+            if (reten_state.ret_mode == 0) {
+                if (stbuf->ia_atime < stpre.ia_mtime) {
+                    gf_log(this->name, GF_LOG_ERROR,
+                           "Cannot set atime less than "
+                           "the mtime for a WORM-Retained "
+                           "file");
+                    goto out;
+                }
+            } else {
+                if (stbuf->ia_atime < stpre.ia_atime) {
+                    gf_log(this->name, GF_LOG_ERROR,
+                           "Cannot decrease the atime of a"
+                           " WORM-Retained file in "
+                           "Enterprise mode");
+                    goto out;
+                }
+            }
+            reten_state.ret_period = reten_state.ret_period + stbuf->ia_atime -
+                                     stpre.ia_atime;
+            ret = gf_worm_set_xattr(this, &reten_state, _gf_true, fd);
+            if (ret) {
+                goto out;
+            }
+
+            stbuf->ia_mtime = stpre.ia_mtime;
+        }
+    }
+    op_errno = 0;
+
+out:
+    if (op_errno)
+        STACK_UNWIND_STRICT(fsetattr, frame, -1, op_errno, NULL, NULL, NULL);
+    else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->fsetattr, fd, stbuf, valid,
+                        xdata);
+    return 0;
+}
+
+static int32_t
+worm_writev(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
+            int32_t count, off_t offset, uint32_t flags, struct iobref *iobref,
+            dict_t *xdata)
+{
+    read_only_priv_t *priv = NULL;
+    int op_errno = EROFS;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+    if (!priv->worm_file || (frame->root->pid < 0)) {
+        op_errno = 0;
+        goto out;
+    }
+    if (is_wormfile(this, _gf_true, fd)) {
+        op_errno = 0;
+        goto out;
+    }
+    op_errno = gf_worm_state_transition(this, _gf_true, fd, GF_FOP_WRITE);
+
+out:
+    if (op_errno) {
+        if (op_errno < 0)
+            op_errno = EROFS;
+        STACK_UNWIND_STRICT(writev, frame, -1, op_errno, NULL, NULL, NULL);
+    } else
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->writev, fd, vector, count,
+                        offset, flags, iobref, xdata);
+    return 0;
+}
+
+static int32_t
+worm_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, fd_t *fd, inode_t *inode,
+                struct iatt *buf, struct iatt *preparent,
+                struct iatt *postparent, dict_t *xdata)
+{
+    int ret = 0;
+    read_only_priv_t *priv = NULL;
+    // In case of an error exit because fd can be NULL and this would
+    // cause an segfault when performing fsetxattr . We explicitly
+    // unwind to avoid future problems
+    if (op_ret < 0) {
+        goto out;
+    }
+
+    priv = this->private;
+    GF_ASSERT(priv);
+    if (priv->worm_file) {
+        ret = fd_ctx_set(fd, this, 1);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "Failed to set the fd ctx "
+                   "for gfid:%s . Worm feature may not work for the gfid",
+                   uuid_utoa(inode->gfid));
+        }
+        ret = worm_init_state(this, _gf_true, fd);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR, "Error initializing state");
+        }
+    }
+
+out:
+    STACK_UNWIND_STRICT(create, frame, op_ret, op_errno, fd, inode, buf,
+                        preparent, postparent, xdata);
+    return ret;
+}
+
+static int32_t
+worm_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+            mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+    STACK_WIND(frame, worm_create_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->create, loc, flags, mode, umask, fd,
+               xdata);
+    return 0;
+}
+
+static void
+set_reten_mode(read_only_priv_t *priv, char *reten_mode)
+{
+    if (strcmp(reten_mode, "relax") == 0)
+        priv->reten_mode = 0;
+    else
+        priv->reten_mode = 1;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    int ret = -1;
+    read_only_priv_t *priv = NULL;
+    char *reten_mode = NULL;
+
+    if (!this->children || this->children->next) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "translator not configured with exactly one child");
+        return -1;
+    }
+
+    if (!this->parents) {
+        gf_log(this->name, GF_LOG_WARNING, "dangling volume. check volfile ");
+    }
+
+    this->local_pool = mem_pool_new(read_only_priv_t, 64);
+    if (!this->local_pool) {
+        ret = -1;
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to create read_only_priv_t's memory pool");
+        goto out;
+    }
+
+    priv = mem_get0(this->local_pool);
+    if (!priv) {
+        gf_log(this->name, GF_LOG_ERROR, "Error allocating priv");
+        goto out;
+    }
+
+    this->private = priv;
+
+    GF_OPTION_INIT("worm", priv->readonly_or_worm_enabled, bool, out);
+    GF_OPTION_INIT("worm-file-level", priv->worm_file, bool, out);
+    GF_OPTION_INIT("default-retention-period", priv->reten_period, int64, out);
+    GF_OPTION_INIT("auto-commit-period", priv->com_period, int64, out);
+    GF_OPTION_INIT("retention-mode", reten_mode, str, out);
+    set_reten_mode(priv, reten_mode);
+    GF_OPTION_INIT("worm-files-deletable", priv->worm_files_deletable, bool,
+                   out);
+
+    ret = 0;
+out:
+    return ret;
+}
+
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    read_only_priv_t *priv = NULL;
+    char *reten_mode = NULL;
+    int ret = -1;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    GF_OPTION_RECONF("worm", priv->readonly_or_worm_enabled, options, bool,
+                     out);
+    GF_OPTION_RECONF("worm-file-level", priv->worm_file, options, bool, out);
+    GF_OPTION_RECONF("default-retention-period", priv->reten_period, options,
+                     int64, out);
+    GF_OPTION_RECONF("retention-mode", reten_mode, options, str, out);
+    set_reten_mode(priv, reten_mode);
+    GF_OPTION_RECONF("auto-commit-period", priv->com_period, options, int64,
+                     out);
+    GF_OPTION_RECONF("worm-files-deletable", priv->worm_files_deletable,
+                     options, bool, out);
+    ret = 0;
+out:
+    gf_log(this->name, GF_LOG_DEBUG, "returning %d", ret);
+    return ret;
+}
+
+void
+fini(xlator_t *this)
+{
+    read_only_priv_t *priv = NULL;
+
+    priv = this->private;
+    if (!priv)
+        goto out;
+    mem_put(priv);
+    this->private = NULL;
+    mem_pool_destroy(this->local_pool);
+    this->local_pool = NULL;
+out:
+    return;
+}
+
+struct xlator_fops fops = {
+    .open = worm_open,
+    .writev = worm_writev,
+    .setattr = worm_setattr,
+    .fsetattr = worm_fsetattr,
+    .rename = worm_rename,
+    .link = worm_link,
+    .unlink = worm_unlink,
+    .truncate = worm_truncate,
+    .ftruncate = worm_ftruncate,
+    .create = worm_create,
+
+    .rmdir = ro_rmdir,
+    .removexattr = ro_removexattr,
+    .fsyncdir = ro_fsyncdir,
+    .xattrop = ro_xattrop,
+    .inodelk = ro_inodelk,
+    .finodelk = ro_finodelk,
+    .entrylk = ro_entrylk,
+    .fentrylk = ro_fentrylk,
+    .lk = ro_lk,
+};
+
+int32_t
+worm_release(xlator_t *this, fd_t *fd)
+{
+    dict_t *dict = NULL;
+    int ret = -1;
+    dict = dict_new();
+    uint64_t value = 0;
+    loc_t loc = {
+        0,
+    };
+    read_only_priv_t *priv = NULL;
+    priv = this->private;
+
+    if (priv->worm_file) {
+        if (!dict) {
+            gf_log(this->name, GF_LOG_ERROR, "Error creating the dict");
+            goto out;
+        }
+
+        ret = fd_ctx_get(fd, this, &value);
+        if (ret) {
+            gf_log(this->name, GF_LOG_DEBUG, "Failed to get the fd ctx");
+        }
+        if (!value) {
+            goto out;
+        }
+
+        ret = dict_set_int8(dict, "trusted.worm_file", 1);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "Error in setting "
+                   "the dict");
+            goto out;
+        }
+
+        loc.inode = inode_ref(fd->inode);
+        gf_uuid_copy(loc.gfid, fd->inode->gfid);
+        ret = syncop_setxattr(this, &loc, dict, 0, NULL, NULL);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR, "Error setting xattr");
+            goto out;
+        }
+
+        gf_worm_state_transition(this, _gf_false, &loc, GF_FOP_WRITE);
+    }
+
+out:
+    loc_wipe(&loc);
+    if (dict)
+        dict_unref(dict);
+    return 0;
+}
+
+struct xlator_cbks cbks = {
+    .release = worm_release,
+};
+
+struct volume_options options[] = {
+    {.key = {"worm"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     /*.validate_fn = validate_boolean,*/
+     .op_version = {2},
+     .flags = OPT_FLAG_SETTABLE,
+     .description = "When \"on\", makes a volume get write once read many "
+                    " feature. It is turned \"off\" by default."},
+    {.key = {"worm-file-level"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "off",
+     /*.validate_fn = validate_boolean,*/
+     .op_version = {GD_OP_VERSION_3_8_0},
+     .flags = OPT_FLAG_SETTABLE,
+     .description = "When \"on\", activates the file level worm. "
+                    "It is turned \"off\" by default."},
+    {.key = {"worm-files-deletable"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     /*.validate_fn = validate_boolean,*/
+     .op_version = {GD_OP_VERSION_3_13_0},
+     .flags = OPT_FLAG_SETTABLE,
+     .description = "When \"off\", doesn't allow the Worm files"
+                    "to be deleted. It is turned \"on\" by default."},
+    {.key = {"default-retention-period"},
+     .type = GF_OPTION_TYPE_TIME,
+     .default_value = "120",
+     /*.validate_fn = validate_worm_period,*/
+     .op_version = {GD_OP_VERSION_3_8_0},
+     .flags = OPT_FLAG_SETTABLE,
+     .description = "The default retention period for the files."},
+    {.key = {"retention-mode"},
+     .type = GF_OPTION_TYPE_STR,
+     .default_value = "relax",
+     /*.validate_fn = validate_reten_mode,*/
+     .op_version = {GD_OP_VERSION_3_8_0},
+     .flags = OPT_FLAG_SETTABLE,
+     .description = "The mode of retention (relax/enterprise). "
+                    "It is relax by default."},
+    {.key = {"auto-commit-period"},
+     .type = GF_OPTION_TYPE_TIME,
+     .default_value = "180",
+     /*.validate_fn = validate_worm_period,*/
+     .op_version = {GD_OP_VERSION_3_8_0},
+     .flags = OPT_FLAG_SETTABLE,
+     .description = "Auto commit period for the files."},
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "worm",
+    .category = GF_TECH_PREVIEW,
+};
diff --git a/xlators/features/sdfs/Makefile.am b/xlators/features/sdfs/Makefile.am
new file mode 100644
index 00000000000..a985f42a877
--- /dev/null
+++ b/xlators/features/sdfs/Makefile.am
@@ -0,0 +1,3 @@
+SUBDIRS = src
+
+CLEANFILES =
diff --git a/xlators/features/sdfs/src/Makefile.am b/xlators/features/sdfs/src/Makefile.am
new file mode 100644
index 00000000000..6118d46ad22
--- /dev/null
+++ b/xlators/features/sdfs/src/Makefile.am
@@ -0,0 +1,19 @@
+if WITH_SERVER
+xlator_LTLIBRARIES = sdfs.la
+endif
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+sdfs_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+sdfs_la_SOURCES = sdfs.c
+sdfs_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = sdfs.h sdfs-messages.h $(top_builddir)/xlators/lib/src/libxlator.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+                                 -I$(top_srcdir)/xlators/lib/src \
+        -I$(top_srcdir)/rpc/xdr/src/ -I$(top_builddir)/rpc/xdr/src/
+
+AM_CFLAGS = -Wall -fno-strict-aliasing $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/sdfs/src/sdfs-messages.h b/xlators/features/sdfs/src/sdfs-messages.h
new file mode 100644
index 00000000000..3053efa8935
--- /dev/null
+++ b/xlators/features/sdfs/src/sdfs-messages.h
@@ -0,0 +1,67 @@
+/*
+ Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com>
+ This file is part of GlusterFS.
+
+ This file is licensed to you under your choice of the GNU Lesser
+ General Public License, version 3 or any later version (LGPLv3 or
+ later), or the GNU General Public License, version 2 (GPLv2), in all
+ cases as published by the Free Software Foundation.
+ */
+
+#ifndef _DFS_MESSAGES_H_
+#define _DFS_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* file bit-rot-bitd-messages.h
+ * brief SDFS log-message IDs and their descriptions
+ */
+
+/* NOTE: Rules for message additions
+ * 1) Each instance of a message is _better_ left with a unique message ID, even
+ *    if the message format is the same. Reasoning is that, if the message
+ *    format needs to change in one instance, the other instances are not
+ *    impacted or the new change does not change the ID of the instance being
+ *    modified.
+ * 2) Addition of a message,
+ *       - Should increment the GLFS_NUM_MESSAGES
+ *       - Append to the list of messages defined, towards the end
+ *       - Retain macro naming as glfs_msg_X (for redability across developers)
+ * NOTE: Rules for message format modifications
+ * 3) Check acorss the code if the message ID macro in question is reused
+ *    anywhere. If reused then then the modifications should ensure correctness
+ *    everywhere, or needs a new message ID as (1) above was not adhered to. If
+ *    not used anywhere, proceed with the required modification.
+ * NOTE: Rules for message deletion
+ * 4) Check (3) and if used anywhere else, then cannot be deleted. If not used
+ *    anywhere, then can be deleted, but will leave a hole by design, as
+ *    addition rules specify modification to the end of the list and not filling
+ *    holes.
+ */
+
+#define GLFS_SDFS_BASE GLFS_MSGID_COMP_SDFS
+#define GLFS_SDFS_NUM_MESSAGES 2
+#define GLFS_MSGID_END (GLFS_SDFS_BASE + GLFS_SDFS_NUM_MESSAGES + 1)
+/* Messaged with message IDs */
+#define glfs_msg_start_x GLFS_DFS_BASE, "Invalid: Start of messages"
+/*------------*/
+
+#define SDFS_MSG_ENTRYLK_ERROR (GLFS_SDFS_BASE + 1)
+/*!
+ * @messageid
+ * @diagnosis
+ * @recommendedaction
+ *
+ */
+
+#define SDFS_MSG_MKDIR_ERROR (GLFS_SDFS_BASE + 2)
+/*!
+ * @messageid
+ * @diagnosis
+ * @recommendedaction
+ *
+ */
+/*------------*/
+
+#define glfs_msg_end_x GLFS_MSGID_END, "Invalid: End of messages"
+#endif /* !_SDFS_MESSAGES_H_ */
diff --git a/xlators/features/sdfs/src/sdfs.c b/xlators/features/sdfs/src/sdfs.c
new file mode 100644
index 00000000000..aaf13f0852e
--- /dev/null
+++ b/xlators/features/sdfs/src/sdfs.c
@@ -0,0 +1,1479 @@
+/*
+   Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include <libgen.h>
+#include "sdfs.h"
+
+static int
+sdfs_frame_return(call_frame_t *frame)
+{
+    sdfs_local_t *local = NULL;
+
+    if (!frame)
+        return -1;
+
+    local = frame->local;
+
+    return GF_ATOMIC_DEC(local->call_cnt);
+}
+
+static void
+sdfs_lock_free(sdfs_entry_lock_t *entrylk)
+{
+    if (entrylk == NULL)
+        goto out;
+
+    loc_wipe(&entrylk->parent_loc);
+    GF_FREE(entrylk->basename);
+
+out:
+    return;
+}
+
+static void
+sdfs_lock_array_free(sdfs_lock_t *lock)
+{
+    sdfs_entry_lock_t *entrylk = NULL;
+    int i = 0;
+
+    if (lock == NULL)
+        goto out;
+
+    for (i = 0; i < lock->lock_count; i++) {
+        entrylk = &lock->entrylk[i];
+        sdfs_lock_free(entrylk);
+    }
+
+out:
+    return;
+}
+
+static void
+sdfs_local_cleanup(sdfs_local_t *local)
+{
+    if (!local)
+        return;
+
+    loc_wipe(&local->loc);
+    loc_wipe(&local->parent_loc);
+
+    if (local->stub) {
+        call_stub_destroy(local->stub);
+        local->stub = NULL;
+    }
+
+    sdfs_lock_array_free(local->lock);
+    GF_FREE(local->lock);
+
+    mem_put(local);
+}
+
+static int
+sdfs_build_parent_loc(loc_t *parent, loc_t *child)
+{
+    int ret = -1;
+    char *path = NULL;
+
+    if (!child->parent) {
+        goto out;
+    }
+    parent->inode = inode_ref(child->parent);
+    path = gf_strdup(child->path);
+    if (!path) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    parent->path = dirname(path);
+    if (!parent->path) {
+        goto out;
+    }
+
+    gf_uuid_copy(parent->gfid, child->pargfid);
+    return 0;
+
+out:
+    GF_FREE(path);
+    return ret;
+}
+
+static sdfs_local_t *
+sdfs_local_init(call_frame_t *frame, xlator_t *this)
+{
+    sdfs_local_t *local = NULL;
+
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto out;
+
+    frame->local = local;
+out:
+    return local;
+}
+
+static int
+sdfs_get_new_frame_common(call_frame_t *frame, call_frame_t **new_frame)
+{
+    int ret = -1;
+    sdfs_local_t *local = NULL;
+    client_t *client = NULL;
+
+    *new_frame = copy_frame(frame);
+    if (!*new_frame) {
+        goto err;
+    }
+
+    client = frame->root->client;
+    gf_client_ref(client);
+    (*new_frame)->root->client = client;
+
+    local = sdfs_local_init(*new_frame, THIS);
+    if (!local) {
+        goto err;
+    }
+
+    local->main_frame = frame;
+    /*Set unique lk-owner for the fop*/
+    set_lk_owner_from_ptr(&(*new_frame)->root->lk_owner, (*new_frame)->root);
+
+    ret = 0;
+err:
+    if ((ret == -1) && (*new_frame)) {
+        SDFS_STACK_DESTROY((*new_frame));
+        *new_frame = NULL;
+    }
+
+    return ret;
+}
+
+static int
+sdfs_get_new_frame(call_frame_t *frame, loc_t *loc, call_frame_t **new_frame)
+{
+    int ret = -1;
+    sdfs_local_t *local = NULL;
+
+    ret = sdfs_get_new_frame_common(frame, new_frame);
+    if (ret < 0) {
+        goto err;
+    }
+
+    local = (*new_frame)->local;
+
+    ret = sdfs_build_parent_loc(&local->parent_loc, loc);
+    if (ret) {
+        goto err;
+    }
+
+    ret = loc_copy(&local->loc, loc);
+    if (ret == -1) {
+        goto err;
+    }
+
+    ret = 0;
+err:
+    if (ret && (*new_frame)) {
+        SDFS_STACK_DESTROY((*new_frame));
+        *new_frame = NULL;
+        ret = -1;
+    }
+
+    return ret;
+}
+
+static int
+sdfs_get_new_frame_readdirp(call_frame_t *frame, fd_t *fd,
+                            call_frame_t **new_frame)
+{
+    int ret = -1;
+    sdfs_local_t *local = NULL;
+
+    ret = sdfs_get_new_frame_common(frame, new_frame);
+    if (ret < 0) {
+        goto err;
+    }
+
+    local = (*new_frame)->local;
+    local->parent_loc.inode = inode_ref(fd->inode);
+    gf_uuid_copy(local->parent_loc.gfid, fd->inode->gfid);
+
+    ret = 0;
+err:
+    return ret;
+}
+
+int
+sdfs_entrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    call_stub_t *stub = NULL;
+
+    local = frame->local;
+
+    local->op_ret = op_ret;
+    local->op_errno = op_errno;
+
+    if (local->stub) {
+        stub = local->stub;
+        local->stub = NULL;
+        call_resume(stub);
+    } else {
+        if (op_ret < 0)
+            gf_msg(this->name, GF_LOG_ERROR, 0, SDFS_MSG_ENTRYLK_ERROR,
+                   "Unlocking entry lock failed for %s", local->loc.name);
+
+        SDFS_STACK_DESTROY(frame);
+    }
+
+    return 0;
+}
+
+int
+sdfs_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, inode_t *inode,
+               struct iatt *stbuf, struct iatt *preparent,
+               struct iatt *postparent, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+
+    local = frame->local;
+
+    STACK_UNWIND_STRICT(mkdir, local->main_frame, op_ret, op_errno, inode,
+                        stbuf, preparent, postparent, xdata);
+
+    local->main_frame = NULL;
+    STACK_WIND(frame, sdfs_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, this->name, &local->parent_loc,
+               local->loc.name, ENTRYLK_UNLOCK, ENTRYLK_WRLCK, xdata);
+    return 0;
+}
+
+int
+sdfs_mkdir_helper(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+                  mode_t umask, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+    int op_errno = -1;
+
+    local = frame->local;
+
+    gf_uuid_unparse(loc->pargfid, gfid);
+
+    if (local->op_ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SDFS_MSG_ENTRYLK_ERROR,
+               "Acquiring entry lock failed for directory %s "
+               "with parent gfid %s",
+               local->loc.name, gfid);
+        op_errno = local->op_errno;
+        goto err;
+    }
+
+    STACK_WIND(frame, sdfs_mkdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mkdir, loc, mode, umask, xdata);
+
+    return 0;
+err:
+    STACK_UNWIND_STRICT(mkdir, local->main_frame, -1, op_errno, NULL, NULL,
+                        NULL, NULL, NULL);
+
+    local->main_frame = NULL;
+    SDFS_STACK_DESTROY(frame);
+    return 0;
+}
+
+int
+sdfs_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+           mode_t umask, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    call_frame_t *new_frame = NULL;
+    call_stub_t *stub = NULL;
+    int op_errno = 0;
+
+    if (-1 == sdfs_get_new_frame(frame, loc, &new_frame)) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    stub = fop_mkdir_stub(new_frame, sdfs_mkdir_helper, loc, mode, umask,
+                          xdata);
+    if (!stub) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local = new_frame->local;
+    local->stub = stub;
+
+    STACK_WIND(new_frame, sdfs_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, this->name, &local->parent_loc,
+               local->loc.name, ENTRYLK_LOCK, ENTRYLK_WRLCK, xdata);
+
+    return 0;
+err:
+    STACK_UNWIND_STRICT(mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                        NULL);
+
+    if (new_frame)
+        SDFS_STACK_DESTROY(new_frame);
+
+    return 0;
+}
+
+int
+sdfs_rmdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+               struct iatt *postparent, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+
+    local = frame->local;
+
+    STACK_UNWIND_STRICT(rmdir, local->main_frame, op_ret, op_errno, preparent,
+                        postparent, xdata);
+
+    local->main_frame = NULL;
+    STACK_WIND(frame, sdfs_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, this->name, &local->parent_loc,
+               local->loc.name, ENTRYLK_UNLOCK, ENTRYLK_WRLCK, xdata);
+    return 0;
+}
+
+int
+sdfs_rmdir_helper(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+                  dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+
+    gf_uuid_unparse(loc->pargfid, gfid);
+
+    if (local->op_ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SDFS_MSG_ENTRYLK_ERROR,
+               "Acquiring entry lock failed for directory %s "
+               "with parent gfid %s",
+               local->loc.name, gfid);
+        goto err;
+    }
+
+    STACK_WIND(frame, sdfs_rmdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rmdir, loc, flags, xdata);
+
+    return 0;
+err:
+    STACK_UNWIND_STRICT(rmdir, local->main_frame, -1, local->op_errno, NULL,
+                        NULL, NULL);
+
+    local->main_frame = NULL;
+    SDFS_STACK_DESTROY(frame);
+    return 0;
+}
+
+int
+sdfs_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+           dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    call_frame_t *new_frame = NULL;
+    call_stub_t *stub = NULL;
+    int op_errno = 0;
+
+    if (-1 == sdfs_get_new_frame(frame, loc, &new_frame)) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    stub = fop_rmdir_stub(new_frame, sdfs_rmdir_helper, loc, flags, xdata);
+    if (!stub) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local = new_frame->local;
+    local->stub = stub;
+
+    STACK_WIND(new_frame, sdfs_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, this->name, &local->parent_loc,
+               local->loc.name, ENTRYLK_LOCK, ENTRYLK_WRLCK, xdata);
+
+    return 0;
+err:
+    STACK_UNWIND_STRICT(rmdir, frame, -1, op_errno, NULL, NULL, NULL);
+
+    if (new_frame)
+        SDFS_STACK_DESTROY(new_frame);
+
+    return 0;
+}
+
+int
+sdfs_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, fd_t *fd, inode_t *inode,
+                struct iatt *stbuf, struct iatt *preparent,
+                struct iatt *postparent, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+
+    local = frame->local;
+
+    STACK_UNWIND_STRICT(create, local->main_frame, op_ret, op_errno, fd, inode,
+                        stbuf, preparent, postparent, xdata);
+
+    local->main_frame = NULL;
+    STACK_WIND(frame, sdfs_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, this->name, &local->parent_loc,
+               local->loc.name, ENTRYLK_UNLOCK, ENTRYLK_WRLCK, xdata);
+    return 0;
+}
+
+int
+sdfs_create_helper(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   int32_t flags, mode_t mode, mode_t umask, fd_t *fd,
+                   dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+
+    gf_uuid_unparse(loc->pargfid, gfid);
+
+    if (local->op_ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SDFS_MSG_ENTRYLK_ERROR,
+               "Acquiring entry lock failed for directory %s "
+               "with parent gfid %s",
+               local->loc.name, gfid);
+        goto err;
+    }
+
+    STACK_WIND(frame, sdfs_create_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->create, loc, flags, mode, umask, fd,
+               xdata);
+
+    return 0;
+err:
+    STACK_UNWIND_STRICT(create, local->main_frame, -1, local->op_errno, NULL,
+                        NULL, NULL, NULL, NULL, NULL);
+
+    local->main_frame = NULL;
+    SDFS_STACK_DESTROY(frame);
+    return 0;
+}
+
+int
+sdfs_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+            mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    call_frame_t *new_frame = NULL;
+    call_stub_t *stub = NULL;
+    int op_errno = 0;
+
+    if (-1 == sdfs_get_new_frame(frame, loc, &new_frame)) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    stub = fop_create_stub(new_frame, sdfs_create_helper, loc, flags, mode,
+                           umask, fd, xdata);
+    if (!stub) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local = new_frame->local;
+    local->stub = stub;
+
+    STACK_WIND(new_frame, sdfs_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, this->name, &local->parent_loc,
+               local->loc.name, ENTRYLK_LOCK, ENTRYLK_WRLCK, xdata);
+
+    return 0;
+err:
+    STACK_UNWIND_STRICT(create, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                        NULL, NULL);
+
+    if (new_frame)
+        SDFS_STACK_DESTROY(new_frame);
+
+    return 0;
+}
+
+int
+sdfs_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                struct iatt *postparent, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+
+    local = frame->local;
+
+    STACK_UNWIND_STRICT(unlink, local->main_frame, op_ret, op_errno, preparent,
+                        postparent, xdata);
+
+    local->main_frame = NULL;
+    STACK_WIND(frame, sdfs_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, this->name, &local->parent_loc,
+               local->loc.name, ENTRYLK_UNLOCK, ENTRYLK_WRLCK, xdata);
+    return 0;
+}
+
+int
+sdfs_unlink_helper(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+                   dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+
+    gf_uuid_unparse(loc->pargfid, gfid);
+
+    if (local->op_ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SDFS_MSG_ENTRYLK_ERROR,
+               "Acquiring entry lock failed for directory %s "
+               "with parent gfid %s",
+               local->loc.name, gfid);
+        goto err;
+    }
+
+    STACK_WIND(frame, sdfs_unlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->unlink, loc, flags, xdata);
+
+    return 0;
+err:
+    STACK_UNWIND_STRICT(unlink, local->main_frame, -1, local->op_errno, NULL,
+                        NULL, NULL);
+
+    local->main_frame = NULL;
+    SDFS_STACK_DESTROY(frame);
+    return 0;
+}
+
+int
+sdfs_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+            dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    call_frame_t *new_frame = NULL;
+    call_stub_t *stub = NULL;
+    int op_errno = 0;
+
+    if (-1 == sdfs_get_new_frame(frame, loc, &new_frame)) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    stub = fop_unlink_stub(new_frame, sdfs_unlink_helper, loc, flags, xdata);
+    if (!stub) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local = new_frame->local;
+    local->stub = stub;
+
+    STACK_WIND(new_frame, sdfs_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, this->name, &local->parent_loc,
+               local->loc.name, ENTRYLK_LOCK, ENTRYLK_WRLCK, xdata);
+
+    return 0;
+err:
+    STACK_UNWIND_STRICT(unlink, frame, -1, op_errno, NULL, NULL, NULL);
+
+    if (new_frame)
+        SDFS_STACK_DESTROY(new_frame);
+
+    return 0;
+}
+
+int
+sdfs_symlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, inode_t *inode,
+                 struct iatt *stbuf, struct iatt *preparent,
+                 struct iatt *postparent, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+
+    local = frame->local;
+
+    STACK_UNWIND_STRICT(link, local->main_frame, op_ret, op_errno, inode, stbuf,
+                        preparent, postparent, xdata);
+
+    local->main_frame = NULL;
+    STACK_WIND(frame, sdfs_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, this->name, &local->parent_loc,
+               local->loc.name, ENTRYLK_UNLOCK, ENTRYLK_WRLCK, xdata);
+    return 0;
+}
+
+int
+sdfs_symlink_helper(call_frame_t *frame, xlator_t *this, const char *linkname,
+                    loc_t *loc, mode_t umask, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+
+    gf_uuid_unparse(loc->pargfid, gfid);
+
+    if (local->op_ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SDFS_MSG_ENTRYLK_ERROR,
+               "Acquiring entry lock failed for directory %s "
+               "with parent gfid %s",
+               local->loc.name, gfid);
+        goto err;
+    }
+
+    STACK_WIND(frame, sdfs_symlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->symlink, linkname, loc, umask, xdata);
+
+    return 0;
+err:
+    STACK_UNWIND_STRICT(link, local->main_frame, -1, local->op_errno, NULL,
+                        NULL, NULL, NULL, NULL);
+
+    local->main_frame = NULL;
+    SDFS_STACK_DESTROY(frame);
+    return 0;
+}
+
+int
+sdfs_symlink(call_frame_t *frame, xlator_t *this, const char *linkname,
+             loc_t *loc, mode_t umask, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    call_frame_t *new_frame = NULL;
+    call_stub_t *stub = NULL;
+    int op_errno = 0;
+
+    if (-1 == sdfs_get_new_frame(frame, loc, &new_frame)) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    stub = fop_symlink_stub(new_frame, sdfs_symlink_helper, linkname, loc,
+                            umask, xdata);
+    if (!stub) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local = new_frame->local;
+    local->stub = stub;
+
+    STACK_WIND(new_frame, sdfs_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, this->name, &local->parent_loc,
+               local->loc.name, ENTRYLK_LOCK, ENTRYLK_WRLCK, xdata);
+
+    return 0;
+err:
+    STACK_UNWIND_STRICT(link, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                        NULL);
+
+    if (new_frame)
+        SDFS_STACK_DESTROY(new_frame);
+
+    return 0;
+}
+
+int
+sdfs_common_entrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    int this_call_cnt = 0;
+    int lk_index = 0;
+    sdfs_lock_t *locks = NULL;
+    call_stub_t *stub = NULL;
+
+    local = frame->local;
+    locks = local->lock;
+    lk_index = (long)cookie;
+
+    if (op_ret < 0) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+    } else {
+        locks->entrylk->locked[lk_index] = _gf_true;
+    }
+
+    this_call_cnt = sdfs_frame_return(frame);
+    if (this_call_cnt > 0) {
+        gf_log(this->name, GF_LOG_DEBUG,
+               "As there are more callcnt (%d) returning without WIND",
+               this_call_cnt);
+        return 0;
+    }
+
+    if (local->stub) {
+        stub = local->stub;
+        local->stub = NULL;
+        call_resume(stub);
+    } else {
+        if (local->op_ret < 0)
+            gf_msg(this->name, GF_LOG_ERROR, 0, SDFS_MSG_ENTRYLK_ERROR,
+                   "unlocking entry lock failed ");
+        SDFS_STACK_DESTROY(frame);
+    }
+
+    return 0;
+}
+
+int
+sdfs_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, inode_t *inode, struct iatt *stbuf,
+              struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    sdfs_lock_t *lock = NULL;
+    int i = 0;
+    int lock_count = 0;
+
+    local = frame->local;
+    lock = local->lock;
+
+    STACK_UNWIND_STRICT(link, local->main_frame, op_ret, op_errno, inode, stbuf,
+                        preparent, postparent, xdata);
+
+    local->main_frame = NULL;
+    lock_count = lock->lock_count;
+    for (i = 0; i < lock_count; i++) {
+        STACK_WIND_COOKIE(frame, sdfs_common_entrylk_cbk, (void *)(long)i,
+                          FIRST_CHILD(this), FIRST_CHILD(this)->fops->entrylk,
+                          this->name, &lock->entrylk[i].parent_loc,
+                          lock->entrylk[i].basename, ENTRYLK_UNLOCK,
+                          ENTRYLK_WRLCK, xdata);
+    }
+
+    return 0;
+}
+
+int
+sdfs_link_helper(call_frame_t *frame, xlator_t *this, loc_t *oldloc,
+                 loc_t *newloc, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    sdfs_lock_t *locks = NULL;
+    gf_boolean_t stack_destroy = _gf_true;
+    int lock_count = 0;
+    int i = 0;
+
+    local = frame->local;
+    locks = local->lock;
+
+    if (local->op_ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SDFS_MSG_ENTRYLK_ERROR,
+               "Acquiring entry lock failed");
+        goto err;
+    }
+
+    STACK_WIND(frame, sdfs_link_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->link, oldloc, newloc, xdata);
+
+    return 0;
+err:
+    STACK_UNWIND_STRICT(link, local->main_frame, -1, local->op_errno, NULL,
+                        NULL, NULL, NULL, NULL);
+
+    local->main_frame = NULL;
+    for (i = 0; i < locks->lock_count && locks->entrylk->locked[i]; i++) {
+        lock_count++;
+    }
+    GF_ATOMIC_INIT(local->call_cnt, lock_count);
+
+    for (i = 0; i < lock_count; i++) {
+        if (!locks->entrylk->locked[i]) {
+            lock_count++;
+            continue;
+        }
+
+        stack_destroy = _gf_false;
+        STACK_WIND(frame, sdfs_common_entrylk_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->entrylk, this->name,
+                   &locks->entrylk[i].parent_loc, locks->entrylk[i].basename,
+                   ENTRYLK_UNLOCK, ENTRYLK_WRLCK, xdata);
+    }
+
+    if (stack_destroy)
+        SDFS_STACK_DESTROY(frame);
+
+    return 0;
+}
+
+static int
+sdfs_init_entry_lock(sdfs_entry_lock_t *lock, loc_t *loc)
+{
+    int ret = 0;
+
+    ret = sdfs_build_parent_loc(&lock->parent_loc, loc);
+    if (ret)
+        return -1;
+
+    lock->basename = gf_strdup(loc->name);
+    if (!lock->basename)
+        return -1;
+
+    return 0;
+}
+
+int
+sdfs_entry_lock_cmp(const void *l1, const void *l2)
+{
+    const sdfs_entry_lock_t *r1 = l1;
+    const sdfs_entry_lock_t *r2 = l2;
+    int ret = 0;
+    uuid_t gfid1 = {0};
+    uuid_t gfid2 = {0};
+
+    loc_gfid((loc_t *)&r1->parent_loc, gfid1);
+    loc_gfid((loc_t *)&r2->parent_loc, gfid2);
+    ret = gf_uuid_compare(gfid1, gfid2);
+    /*Entrylks with NULL basename are the 'smallest'*/
+    if (ret == 0) {
+        if (!r1->basename)
+            return -1;
+        if (!r2->basename)
+            return 1;
+        ret = strcmp(r1->basename, r2->basename);
+    }
+
+    if (ret <= 0)
+        return -1;
+    else
+        return 1;
+}
+
+int
+sdfs_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+          dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    call_frame_t *new_frame = NULL;
+    call_stub_t *stub = NULL;
+    sdfs_lock_t *lock = NULL;
+    client_t *client = NULL;
+    int ret = 0;
+    int op_errno = ENOMEM;
+
+    new_frame = copy_frame(frame);
+    if (!new_frame) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+    /*Set unique lk-owner for the fop*/
+    set_lk_owner_from_ptr(&new_frame->root->lk_owner, new_frame->root);
+
+    gf_client_ref(client);
+    new_frame->root->client = client;
+    local = sdfs_local_init(new_frame, this);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->main_frame = frame;
+
+    lock = GF_CALLOC(1, sizeof(*lock), gf_common_mt_char);
+    if (!lock)
+        goto err;
+
+    local->lock = lock;
+
+    ret = sdfs_init_entry_lock(&lock->entrylk[0], newloc);
+    if (ret)
+        goto err;
+
+    ++lock->lock_count;
+
+    local->lock = lock;
+    GF_ATOMIC_INIT(local->call_cnt, lock->lock_count);
+
+    ret = loc_copy(&local->loc, newloc);
+    if (ret == -1) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    stub = fop_link_stub(new_frame, sdfs_link_helper, oldloc, newloc, xdata);
+    if (!stub) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->stub = stub;
+
+    STACK_WIND_COOKIE(new_frame, sdfs_common_entrylk_cbk, 0, FIRST_CHILD(this),
+                      FIRST_CHILD(this)->fops->entrylk, this->name,
+                      &lock->entrylk[0].parent_loc, lock->entrylk[0].basename,
+                      ENTRYLK_LOCK, ENTRYLK_WRLCK, xdata);
+
+    return 0;
+err:
+
+    STACK_UNWIND_STRICT(link, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                        NULL);
+
+    if (new_frame)
+        SDFS_STACK_DESTROY(new_frame);
+
+    return 0;
+}
+
+int
+sdfs_mknod_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, inode_t *inode,
+               struct iatt *stbuf, struct iatt *preparent,
+               struct iatt *postparent, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+
+    local = frame->local;
+
+    STACK_UNWIND_STRICT(mknod, local->main_frame, op_ret, op_errno, inode,
+                        stbuf, preparent, postparent, xdata);
+
+    local->main_frame = NULL;
+    STACK_WIND(frame, sdfs_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, this->name, &local->parent_loc,
+               local->loc.name, ENTRYLK_UNLOCK, ENTRYLK_WRLCK, xdata);
+    return 0;
+}
+
+int
+sdfs_mknod_helper(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+                  dev_t rdev, mode_t umask, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+
+    gf_uuid_unparse(loc->pargfid, gfid);
+
+    if (local->op_ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SDFS_MSG_ENTRYLK_ERROR,
+               "Acquiring entry lock failed for directory %s "
+               "with parent gfid %s",
+               local->loc.name, gfid);
+        goto err;
+    }
+
+    STACK_WIND(frame, sdfs_mknod_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mknod, loc, mode, rdev, umask, xdata);
+
+    return 0;
+err:
+    STACK_UNWIND_STRICT(mknod, local->main_frame, -1, local->op_errno, NULL,
+                        NULL, NULL, NULL, NULL);
+
+    local->main_frame = NULL;
+    SDFS_STACK_DESTROY(frame);
+    return 0;
+}
+
+int
+sdfs_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+           dev_t rdev, mode_t umask, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    call_frame_t *new_frame = NULL;
+    call_stub_t *stub = NULL;
+    int op_errno = 0;
+
+    if (-1 == sdfs_get_new_frame(frame, loc, &new_frame)) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    stub = fop_mknod_stub(new_frame, sdfs_mknod_helper, loc, mode, rdev, umask,
+                          xdata);
+    if (!stub) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local = new_frame->local;
+    local->stub = stub;
+
+    STACK_WIND(new_frame, sdfs_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, this->name, &local->parent_loc,
+               local->loc.name, ENTRYLK_LOCK, ENTRYLK_WRLCK, xdata);
+
+    return 0;
+err:
+    STACK_UNWIND_STRICT(mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                        NULL);
+
+    if (new_frame)
+        SDFS_STACK_DESTROY(new_frame);
+
+    return 0;
+}
+
+int
+sdfs_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *stbuf,
+                struct iatt *preoldparent, struct iatt *postoldparent,
+                struct iatt *prenewparent, struct iatt *postnewparent,
+                dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    sdfs_lock_t *lock = NULL;
+    int i = 0;
+    int call_cnt = 0;
+
+    local = frame->local;
+    lock = local->lock;
+    GF_ATOMIC_INIT(local->call_cnt, lock->lock_count);
+
+    STACK_UNWIND_STRICT(rename, local->main_frame, op_ret, op_errno, stbuf,
+                        preoldparent, postoldparent, prenewparent,
+                        postnewparent, xdata);
+
+    local->main_frame = NULL;
+    call_cnt = GF_ATOMIC_GET(local->call_cnt);
+
+    for (i = 0; i < call_cnt; i++) {
+        STACK_WIND_COOKIE(frame, sdfs_common_entrylk_cbk, (void *)(long)i,
+                          FIRST_CHILD(this), FIRST_CHILD(this)->fops->entrylk,
+                          this->name, &lock->entrylk[i].parent_loc,
+                          lock->entrylk[i].basename, ENTRYLK_UNLOCK,
+                          ENTRYLK_WRLCK, xdata);
+    }
+
+    return 0;
+}
+
+int
+sdfs_rename_helper(call_frame_t *frame, xlator_t *this, loc_t *oldloc,
+                   loc_t *newloc, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    sdfs_lock_t *lock = NULL;
+    gf_boolean_t stack_destroy = _gf_true;
+    int lock_count = 0;
+    int i = 0;
+
+    local = frame->local;
+    lock = local->lock;
+
+    if (local->op_ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SDFS_MSG_ENTRYLK_ERROR,
+               "Acquiring entry lock failed ");
+        goto err;
+    }
+
+    STACK_WIND(frame, sdfs_rename_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
+
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(rename, local->main_frame, -1, local->op_errno, NULL,
+                        NULL, NULL, NULL, NULL, NULL);
+
+    local->main_frame = NULL;
+    for (i = 0; i < lock->lock_count && lock->entrylk->locked[i]; i++) {
+        lock_count++;
+    }
+    GF_ATOMIC_INIT(local->call_cnt, lock_count);
+
+    for (i = 0; i < lock_count; i++) {
+        if (!lock->entrylk->locked[i]) {
+            lock_count++;
+            continue;
+        }
+        stack_destroy = _gf_false;
+        STACK_WIND(frame, sdfs_common_entrylk_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->entrylk, this->name,
+                   &lock->entrylk[i].parent_loc, lock->entrylk[i].basename,
+                   ENTRYLK_UNLOCK, ENTRYLK_WRLCK, xdata);
+    }
+
+    if (stack_destroy)
+        SDFS_STACK_DESTROY(frame);
+
+    return 0;
+}
+
+int
+sdfs_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+            dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    sdfs_lock_t *lock = NULL;
+    call_frame_t *new_frame = NULL;
+    call_stub_t *stub = NULL;
+    client_t *client = NULL;
+    int ret = 0;
+    int op_errno = ENOMEM;
+    int i = 0;
+    int call_cnt = 0;
+
+    new_frame = copy_frame(frame);
+    if (!new_frame) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+    /*Set unique lk-owner for the fop*/
+    set_lk_owner_from_ptr(&new_frame->root->lk_owner, new_frame->root);
+
+    gf_client_ref(client);
+    new_frame->root->client = client;
+    local = sdfs_local_init(new_frame, this);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->main_frame = frame;
+
+    lock = GF_CALLOC(1, sizeof(*lock), gf_common_mt_char);
+    if (!lock)
+        goto err;
+
+    local->lock = lock;
+
+    ret = sdfs_init_entry_lock(&lock->entrylk[0], oldloc);
+    if (ret)
+        goto err;
+    lock->entrylk->locked[0] = _gf_false;
+
+    ++lock->lock_count;
+
+    ret = sdfs_init_entry_lock(&lock->entrylk[1], newloc);
+    if (ret)
+        goto err;
+    lock->entrylk->locked[1] = _gf_false;
+
+    ++lock->lock_count;
+
+    qsort(lock->entrylk, lock->lock_count, sizeof(*lock->entrylk),
+          sdfs_entry_lock_cmp);
+
+    local->lock = lock;
+    GF_ATOMIC_INIT(local->call_cnt, lock->lock_count);
+
+    stub = fop_rename_stub(new_frame, sdfs_rename_helper, oldloc, newloc,
+                           xdata);
+    if (!stub) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local->stub = stub;
+    call_cnt = GF_ATOMIC_GET(local->call_cnt);
+    for (i = 0; i < call_cnt; i++) {
+        STACK_WIND_COOKIE(new_frame, sdfs_common_entrylk_cbk, (void *)(long)i,
+                          FIRST_CHILD(this), FIRST_CHILD(this)->fops->entrylk,
+                          this->name, &lock->entrylk[i].parent_loc,
+                          lock->entrylk[i].basename, ENTRYLK_LOCK,
+                          ENTRYLK_WRLCK, xdata);
+    }
+
+    return 0;
+err:
+
+    STACK_UNWIND_STRICT(rename, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                        NULL, NULL);
+
+    if (new_frame)
+        SDFS_STACK_DESTROY(new_frame);
+
+    return 0;
+}
+
+int
+sdfs_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, inode_t *inode,
+                struct iatt *stbuf, dict_t *xdata, struct iatt *postparent)
+{
+    sdfs_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (!local->loc.parent) {
+        sdfs_local_cleanup(local);
+        frame->local = NULL;
+        STACK_UNWIND_STRICT(lookup, frame, op_ret, op_errno, inode, stbuf,
+                            xdata, postparent);
+        return 0;
+    }
+
+    STACK_UNWIND_STRICT(lookup, local->main_frame, op_ret, op_errno, inode,
+                        stbuf, xdata, postparent);
+
+    local->main_frame = NULL;
+    STACK_WIND(frame, sdfs_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, this->name, &local->parent_loc,
+               local->loc.name, ENTRYLK_UNLOCK, ENTRYLK_RDLCK, xdata);
+    return 0;
+}
+
+int
+sdfs_lookup_helper(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+
+    gf_uuid_unparse(loc->pargfid, gfid);
+
+    if (local->op_ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SDFS_MSG_ENTRYLK_ERROR,
+               "Acquiring entry lock failed for directory %s "
+               "with parent gfid %s",
+               local->loc.name, gfid);
+        goto err;
+    }
+
+    STACK_WIND(frame, sdfs_lookup_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lookup, loc, xdata);
+
+    return 0;
+err:
+    STACK_UNWIND_STRICT(lookup, local->main_frame, -1, local->op_errno, NULL,
+                        NULL, NULL, NULL);
+    local->main_frame = NULL;
+
+    SDFS_STACK_DESTROY(frame);
+    return 0;
+}
+
+int
+sdfs_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    call_frame_t *new_frame = NULL;
+    call_stub_t *stub = NULL;
+    int op_errno = 0;
+
+    if (!loc->parent) {
+        local = sdfs_local_init(frame, this);
+        if (!local) {
+            op_errno = ENOMEM;
+            goto err;
+        }
+
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->lookup, loc, xdata);
+        return 0;
+    }
+
+    if (-1 == sdfs_get_new_frame(frame, loc, &new_frame)) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    stub = fop_lookup_stub(new_frame, sdfs_lookup_helper, loc, xdata);
+    if (!stub) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local = new_frame->local;
+    local->stub = stub;
+
+    STACK_WIND(new_frame, sdfs_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, this->name, &local->parent_loc,
+               local->loc.name, ENTRYLK_LOCK, ENTRYLK_RDLCK, xdata);
+
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+
+    if (new_frame)
+        SDFS_STACK_DESTROY(new_frame);
+
+    return 0;
+}
+
+int32_t
+sdfs_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, gf_dirent_t *entries,
+                  dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+
+    local = frame->local;
+    STACK_UNWIND_STRICT(readdirp, local->main_frame, op_ret, op_errno, entries,
+                        xdata);
+
+    local->main_frame = NULL;
+    STACK_WIND(frame, sdfs_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, this->name, &local->parent_loc,
+               NULL, ENTRYLK_UNLOCK, ENTRYLK_RDLCK, xdata);
+    return 0;
+}
+
+int32_t
+sdfs_readdirp_helper(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                     off_t off, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    char gfid[GF_UUID_BUF_SIZE] = {0};
+
+    local = frame->local;
+
+    gf_uuid_unparse(fd->inode->gfid, gfid);
+
+    if (local->op_ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SDFS_MSG_ENTRYLK_ERROR,
+               "Acquiring entry lock failed for directory %s "
+               "with parent gfid %s",
+               local->loc.name, gfid);
+        goto err;
+    }
+
+    STACK_WIND(frame, sdfs_readdirp_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdirp, fd, size, off, xdata);
+
+    return 0;
+err:
+    STACK_UNWIND_STRICT(readdirp, local->main_frame, -1, local->op_errno, NULL,
+                        NULL);
+
+    local->main_frame = NULL;
+
+    SDFS_STACK_DESTROY(frame);
+    return 0;
+}
+
+int32_t
+sdfs_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+              off_t off, dict_t *xdata)
+{
+    sdfs_local_t *local = NULL;
+    call_frame_t *new_frame = NULL;
+    call_stub_t *stub = NULL;
+    int op_errno = 0;
+
+    if (-1 == sdfs_get_new_frame_readdirp(frame, fd, &new_frame)) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    stub = fop_readdirp_stub(new_frame, sdfs_readdirp_helper, fd, size, off,
+                             xdata);
+    if (!stub) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    local = new_frame->local;
+    local->stub = stub;
+
+    STACK_WIND(new_frame, sdfs_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, this->name, &local->parent_loc,
+               NULL, ENTRYLK_LOCK, ENTRYLK_RDLCK, xdata);
+
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(readdirp, frame, -1, op_errno, NULL, NULL);
+
+    if (new_frame)
+        SDFS_STACK_DESTROY(new_frame);
+
+    return 0;
+}
+
+int
+init(xlator_t *this)
+{
+    int ret = -1;
+
+    if (!this->children || this->children->next) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "'dentry-fop-serializer' not configured with exactly one child");
+        goto out;
+    }
+
+    if (!this->parents) {
+        gf_log(this->name, GF_LOG_WARNING, "dangling volume. check volfile ");
+    }
+
+    this->local_pool = mem_pool_new(sdfs_local_t, 512);
+    if (!this->local_pool) {
+        goto out;
+    }
+
+    GF_OPTION_INIT("pass-through", this->pass_through, bool, out);
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    int ret = -1;
+
+    GF_OPTION_RECONF("pass-through", this->pass_through, options, bool, out);
+
+    ret = 0;
+out:
+    return ret;
+}
+
+void
+fini(xlator_t *this)
+{
+    mem_pool_destroy(this->local_pool);
+    this->local_pool = NULL;
+    return;
+}
+
+struct xlator_fops fops = {
+    .mkdir = sdfs_mkdir,
+    .rmdir = sdfs_rmdir,
+    .create = sdfs_create,
+    .unlink = sdfs_unlink,
+    .symlink = sdfs_symlink,
+    .link = sdfs_link,
+    .mknod = sdfs_mknod,
+    .rename = sdfs_rename,
+    .lookup = sdfs_lookup,
+    .readdirp = sdfs_readdirp,
+};
+
+struct xlator_cbks cbks;
+
+struct volume_options options[] = {
+    {.key = {"pass-through"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "true",
+     .op_version = {GD_OP_VERSION_4_1_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC | OPT_FLAG_CLIENT_OPT,
+     .tags = {"sdfs"},
+     .description = "Enable/Disable dentry serialize functionality"},
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .reconfigure = reconfigure,
+    .op_version = {GD_OP_VERSION_4_0_0},
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "sdfs",
+    .category = GF_TECH_PREVIEW,
+};
diff --git a/xlators/features/sdfs/src/sdfs.h b/xlators/features/sdfs/src/sdfs.h
new file mode 100644
index 00000000000..dded5a2d7fc
--- /dev/null
+++ b/xlators/features/sdfs/src/sdfs.h
@@ -0,0 +1,49 @@
+/*
+   Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/call-stub.h>
+#include "sdfs-messages.h"
+#include <glusterfs/atomic.h>
+
+#define SDFS_LOCK_COUNT_MAX 2
+
+typedef struct {
+    loc_t parent_loc;
+    char *basename;
+    int locked[SDFS_LOCK_COUNT_MAX];
+} sdfs_entry_lock_t;
+
+typedef struct {
+    sdfs_entry_lock_t entrylk[SDFS_LOCK_COUNT_MAX];
+    int lock_count;
+} sdfs_lock_t;
+
+struct sdfs_local {
+    call_frame_t *main_frame;
+    loc_t loc;
+    loc_t parent_loc;
+    call_stub_t *stub;
+    sdfs_lock_t *lock;
+    int op_ret;
+    int op_errno;
+    gf_atomic_t call_cnt;
+};
+typedef struct sdfs_local sdfs_local_t;
+
+#define SDFS_STACK_DESTROY(frame)                                              \
+    do {                                                                       \
+        sdfs_local_t *__local = NULL;                                          \
+        __local = frame->local;                                                \
+        frame->local = NULL;                                                   \
+        gf_client_unref(frame->root->client);                                  \
+        STACK_DESTROY(frame->root);                                            \
+        sdfs_local_cleanup(__local);                                           \
+    } while (0)
diff --git a/xlators/features/selinux/Makefile.am b/xlators/features/selinux/Makefile.am
new file mode 100644
index 00000000000..a985f42a877
--- /dev/null
+++ b/xlators/features/selinux/Makefile.am
@@ -0,0 +1,3 @@
+SUBDIRS = src
+
+CLEANFILES =
diff --git a/xlators/features/selinux/src/Makefile.am b/xlators/features/selinux/src/Makefile.am
new file mode 100644
index 00000000000..4f1e5e149b3
--- /dev/null
+++ b/xlators/features/selinux/src/Makefile.am
@@ -0,0 +1,20 @@
+if WITH_SERVER
+xlator_LTLIBRARIES = selinux.la
+endif
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+selinux_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+selinux_la_SOURCES = selinux.c
+
+selinux_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = selinux.h selinux-messages.h selinux-mem-types.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
+
diff --git a/xlators/features/selinux/src/selinux-mem-types.h b/xlators/features/selinux/src/selinux-mem-types.h
new file mode 100644
index 00000000000..553e59e5a9d
--- /dev/null
+++ b/xlators/features/selinux/src/selinux-mem-types.h
@@ -0,0 +1,19 @@
+/*
+   Copyright (c) 2017 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#ifndef __SELINUX_MEM_TYPES_H__
+#define __SELINUX_MEM_TYPES_H__
+
+#include <glusterfs/mem-types.h>
+
+enum gf_selinux_mem_types_ {
+    gf_selinux_mt_selinux_priv_t = gf_common_mt_end + 1,
+    gf_selinux_mt_end
+};
+#endif
diff --git a/xlators/features/selinux/src/selinux-messages.h b/xlators/features/selinux/src/selinux-messages.h
new file mode 100644
index 00000000000..f49a54f956c
--- /dev/null
+++ b/xlators/features/selinux/src/selinux-messages.h
@@ -0,0 +1,30 @@
+/*
+  Copyright (c) 2017 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _SELINUX_MESSAGES_H__
+#define _SELINUX_MESSAGES_H__
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(SL, SL_MSG_INVALID_VOLFILE, SL_MSG_ENOMEM,
+           SL_MSG_MEM_ACCT_INIT_FAILED, SL_MSG_SELINUX_GLUSTER_XATTR_MISSING,
+           SL_MSG_SELINUX_XATTR_MISSING);
+
+#endif /*_SELINUX_MESSAGES_H */
diff --git a/xlators/features/selinux/src/selinux.c b/xlators/features/selinux/src/selinux.c
new file mode 100644
index 00000000000..9b1b4b55e1a
--- /dev/null
+++ b/xlators/features/selinux/src/selinux.c
@@ -0,0 +1,323 @@
+/*
+   Copyright (c) 2017 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+
+#include "selinux.h"
+#include "selinux-messages.h"
+#include "selinux-mem-types.h"
+#include <glusterfs/compat-errno.h>
+
+static int
+selinux_fgetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, dict_t *dict, dict_t *xdata)
+{
+    int ret = 0;
+    char *name = cookie;
+
+    if (op_errno == 0 && dict && name &&
+        (!strcmp(name, SELINUX_GLUSTER_XATTR))) {
+        ret = dict_rename_key(dict, SELINUX_GLUSTER_XATTR, SELINUX_XATTR);
+        if (ret < 0)
+            gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                   SL_MSG_SELINUX_GLUSTER_XATTR_MISSING,
+                   "getxattr failed for %s", SELINUX_XATTR);
+    }
+
+    STACK_UNWIND_STRICT(fgetxattr, frame, op_ret, op_errno, dict, xdata);
+    return ret;
+}
+
+static int
+selinux_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                  const char *name, dict_t *xdata)
+{
+    selinux_priv_t *priv = NULL;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    char *xattr_name = (char *)name;
+
+    priv = this->private;
+
+    GF_VALIDATE_OR_GOTO("selinux", priv, err);
+
+    /* name can be NULL for listxattr calls */
+    if (!priv->selinux_enabled || !name)
+        goto off;
+
+    if (strcmp(name, SELINUX_XATTR) == 0)
+        xattr_name = SELINUX_GLUSTER_XATTR;
+
+off:
+    STACK_WIND_COOKIE(frame, selinux_fgetxattr_cbk, xattr_name,
+                      FIRST_CHILD(this), FIRST_CHILD(this)->fops->fgetxattr, fd,
+                      xattr_name, xdata);
+    return 0;
+err:
+    STACK_UNWIND_STRICT(fgetxattr, frame, op_ret, op_errno, NULL, xdata);
+
+    return 0;
+}
+
+static int
+selinux_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, dict_t *dict, dict_t *xdata)
+{
+    int ret = 0;
+    char *name = cookie;
+
+    if (op_errno == 0 && dict && name &&
+        (!strcmp(name, SELINUX_GLUSTER_XATTR))) {
+        ret = dict_rename_key(dict, SELINUX_GLUSTER_XATTR, SELINUX_XATTR);
+        if (ret < 0)
+            gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                   SL_MSG_SELINUX_GLUSTER_XATTR_MISSING,
+                   "getxattr failed for %s", SELINUX_XATTR);
+    }
+
+    STACK_UNWIND_STRICT(getxattr, frame, op_ret, op_errno, dict, xdata);
+
+    return 0;
+}
+
+static int
+selinux_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                 const char *name, dict_t *xdata)
+{
+    selinux_priv_t *priv = NULL;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    char *xattr_name = (char *)name;
+
+    priv = this->private;
+
+    GF_VALIDATE_OR_GOTO("selinux", priv, err);
+
+    /* name can be NULL for listxattr calls */
+    if (!priv->selinux_enabled || !name)
+        goto off;
+
+    if (strcmp(name, SELINUX_XATTR) == 0)
+        xattr_name = SELINUX_GLUSTER_XATTR;
+
+off:
+    STACK_WIND_COOKIE(frame, selinux_getxattr_cbk, xattr_name,
+                      FIRST_CHILD(this), FIRST_CHILD(this)->fops->getxattr, loc,
+                      xattr_name, xdata);
+    return 0;
+err:
+    STACK_UNWIND_STRICT(getxattr, frame, op_ret, op_errno, NULL, xdata);
+    return 0;
+}
+
+static int
+selinux_fsetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int op_ret, int op_errno, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(fsetxattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+static int
+selinux_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+                  int flags, dict_t *xdata)
+{
+    selinux_priv_t *priv = NULL;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    int32_t ret = -1;
+
+    priv = this->private;
+
+    GF_VALIDATE_OR_GOTO("selinux", priv, err);
+
+    if (!priv->selinux_enabled && !dict)
+        goto off;
+
+    ret = dict_rename_key(dict, SELINUX_XATTR, SELINUX_GLUSTER_XATTR);
+    if (ret < 0 && ret != -ENODATA)
+        goto err;
+
+off:
+    STACK_WIND(frame, selinux_fsetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
+
+    return 0;
+err:
+    STACK_UNWIND_STRICT(fsetxattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+static int
+selinux_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int op_ret, int op_errno, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(setxattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+static int
+selinux_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+                 int flags, dict_t *xdata)
+{
+    selinux_priv_t *priv = NULL;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    int32_t ret = -1;
+
+    priv = this->private;
+
+    GF_VALIDATE_OR_GOTO("selinux", priv, err);
+
+    if (!priv->selinux_enabled && !dict)
+        goto off;
+
+    ret = dict_rename_key(dict, SELINUX_XATTR, SELINUX_GLUSTER_XATTR);
+    if (ret < 0 && ret != -ENODATA)
+        goto err;
+
+off:
+    STACK_WIND(frame, selinux_setxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setxattr, loc, dict, flags, xdata);
+    return 0;
+err:
+    STACK_UNWIND_STRICT(setxattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("selinux", this, out);
+
+    ret = xlator_mem_acct_init(this, gf_selinux_mt_end + 1);
+
+    if (ret != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SL_MSG_MEM_ACCT_INIT_FAILED,
+               "Memory accounting init failed");
+        return ret;
+    }
+out:
+    return ret;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    int32_t ret = -1;
+    selinux_priv_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("selinux", this, out);
+
+    if (!this->children || this->children->next) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, SL_MSG_INVALID_VOLFILE,
+               "Error: SELinux (%s) not configured with exactly one "
+               "child",
+               this->name);
+        return -1;
+    }
+
+    if (this->parents == NULL) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, SL_MSG_INVALID_VOLFILE,
+               "Dangling volume. Please check the volfile");
+    }
+
+    priv = GF_CALLOC(1, sizeof(*priv), gf_selinux_mt_selinux_priv_t);
+    if (!priv) {
+        gf_log(this->name, GF_LOG_ERROR, "out of memory");
+        goto out;
+    }
+
+    GF_OPTION_INIT("selinux", priv->selinux_enabled, bool, out);
+
+    this->local_pool = mem_pool_new(selinux_priv_t, 64);
+    if (!this->local_pool) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, SL_MSG_ENOMEM,
+               "Failed to create local_t's memory pool");
+        goto out;
+    }
+
+    this->private = (void *)priv;
+    ret = 0;
+out:
+    if (ret) {
+        GF_FREE(priv);
+        mem_pool_destroy(this->local_pool);
+        this->local_pool = NULL;
+    }
+    return ret;
+}
+
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    int32_t ret = -1;
+    selinux_priv_t *priv = NULL;
+
+    priv = this->private;
+
+    GF_OPTION_RECONF("selinux", priv->selinux_enabled, options, bool, out);
+
+    ret = 0;
+out:
+    return ret;
+}
+
+void
+fini(xlator_t *this)
+{
+    selinux_priv_t *priv = NULL;
+
+    priv = this->private;
+    GF_FREE(priv);
+
+    mem_pool_destroy(this->local_pool);
+    this->local_pool = NULL;
+
+    return;
+}
+
+struct xlator_fops fops = {
+    .getxattr = selinux_getxattr,
+    .fgetxattr = selinux_fgetxattr,
+    .setxattr = selinux_setxattr,
+    .fsetxattr = selinux_fsetxattr,
+};
+
+struct xlator_cbks cbks = {};
+
+struct volume_options options[] = {
+    {
+        .key = {"selinux"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "on",
+        .description = "Enable/disable selinux translator",
+        .op_version = {GD_OP_VERSION_3_11_0},
+        .flags = OPT_FLAG_SETTABLE,
+        .tags = {"security", "linux"},
+    },
+    {
+        .key = {NULL},
+    }};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "selinux",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/features/selinux/src/selinux.h b/xlators/features/selinux/src/selinux.h
new file mode 100644
index 00000000000..1bbdad3bb36
--- /dev/null
+++ b/xlators/features/selinux/src/selinux.h
@@ -0,0 +1,24 @@
+/*
+   Copyright (c) 2017 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#ifndef __SELINUX_H__
+#define __SELINUX_H__
+
+#include <glusterfs/common-utils.h>
+
+#define SELINUX_XATTR "security.selinux"
+#define SELINUX_GLUSTER_XATTR "trusted.glusterfs.selinux"
+
+struct selinux_priv {
+    gf_boolean_t selinux_enabled;
+};
+
+typedef struct selinux_priv selinux_priv_t;
+
+#endif
diff --git a/xlators/features/shard/Makefile.am b/xlators/features/shard/Makefile.am
new file mode 100644
index 00000000000..a985f42a877
--- /dev/null
+++ b/xlators/features/shard/Makefile.am
@@ -0,0 +1,3 @@
+SUBDIRS = src
+
+CLEANFILES =
diff --git a/xlators/features/shard/src/Makefile.am b/xlators/features/shard/src/Makefile.am
new file mode 100644
index 00000000000..bf5700d4bcc
--- /dev/null
+++ b/xlators/features/shard/src/Makefile.am
@@ -0,0 +1,17 @@
+xlator_LTLIBRARIES = shard.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+shard_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+shard_la_SOURCES = shard.c
+
+shard_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = shard.h shard-mem-types.h shard-messages.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/shard/src/shard-mem-types.h b/xlators/features/shard/src/shard-mem-types.h
new file mode 100644
index 00000000000..1fe7e2e2798
--- /dev/null
+++ b/xlators/features/shard/src/shard-mem-types.h
@@ -0,0 +1,24 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#ifndef __SHARD_MEM_TYPES_H__
+#define __SHARD_MEM_TYPES_H__
+
+#include <glusterfs/mem-types.h>
+
+enum gf_shard_mem_types_ {
+    gf_shard_mt_priv_t = gf_common_mt_end + 1,
+    gf_shard_mt_inode_list,
+    gf_shard_mt_inode_ctx_t,
+    gf_shard_mt_iovec,
+    gf_shard_mt_int64_t,
+    gf_shard_mt_uint64_t,
+    gf_shard_mt_end
+};
+#endif
diff --git a/xlators/features/shard/src/shard-messages.h b/xlators/features/shard/src/shard-messages.h
new file mode 100644
index 00000000000..2d0867eb136
--- /dev/null
+++ b/xlators/features/shard/src/shard-messages.h
@@ -0,0 +1,39 @@
+/*
+ Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+ This file is part of GlusterFS.
+
+ This file is licensed to you under your choice of the GNU Lesser
+ General Public License, version 3 or any later version (LGPLv3 or
+ later), or the GNU General Public License, version 2 (GPLv2), in all
+ cases as published by the Free Software Foundation.
+ */
+
+#ifndef _SHARD_MESSAGES_H_
+#define _SHARD_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(SHARD, SHARD_MSG_BASE_FILE_LOOKUP_FAILED, SHARD_MSG_DICT_OP_FAILED,
+           SHARD_MSG_DOT_SHARD_NODIR, SHARD_MSG_FD_CTX_SET_FAILED,
+           SHARD_MSG_INODE_CTX_GET_FAILED, SHARD_MSG_INODE_CTX_SET_FAILED,
+           SHARD_MSG_INODE_PATH_FAILED, SHARD_MSG_INTERNAL_XATTR_MISSING,
+           SHARD_MSG_INVALID_VOLFILE, SHARD_MSG_LOOKUP_SHARD_FAILED,
+           SHARD_MSG_MEM_ACCT_INIT_FAILED, SHARD_MSG_NULL_THIS,
+           SHARD_MSG_SIZE_SET_FAILED, SHARD_MSG_STAT_FAILED,
+           SHARD_MSG_TRUNCATE_LAST_SHARD_FAILED,
+           SHARD_MSG_UPDATE_FILE_SIZE_FAILED, SHARD_MSG_FOP_NOT_SUPPORTED,
+           SHARD_MSG_INVALID_FOP, SHARD_MSG_MEMALLOC_FAILED,
+           SHARD_MSG_FOP_FAILED, SHARD_MSG_SHARDS_DELETION_FAILED,
+           SHARD_MSG_SHARD_DELETION_COMPLETED);
+
+#endif /* !_SHARD_MESSAGES_H_ */
diff --git a/xlators/features/shard/src/shard.c b/xlators/features/shard/src/shard.c
new file mode 100644
index 00000000000..e5f93063943
--- /dev/null
+++ b/xlators/features/shard/src/shard.c
@@ -0,0 +1,7382 @@
+/*
+  Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <unistd.h>
+
+#include "shard.h"
+#include "shard-mem-types.h"
+#include <glusterfs/byte-order.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/statedump.h>
+
+static gf_boolean_t
+__is_shard_dir(uuid_t gfid)
+{
+    shard_priv_t *priv = THIS->private;
+
+    if (gf_uuid_compare(gfid, priv->dot_shard_gfid) == 0)
+        return _gf_true;
+
+    return _gf_false;
+}
+
+static gf_boolean_t
+__is_gsyncd_on_shard_dir(call_frame_t *frame, loc_t *loc)
+{
+    if (frame->root->pid == GF_CLIENT_PID_GSYNCD &&
+        (__is_shard_dir(loc->pargfid) ||
+         (loc->parent && __is_shard_dir(loc->parent->gfid))))
+        return _gf_true;
+
+    return _gf_false;
+}
+
+void
+shard_make_block_bname(int block_num, uuid_t gfid, char *buf, size_t len)
+{
+    char gfid_str[GF_UUID_BUF_SIZE] = {
+        0,
+    };
+
+    gf_uuid_unparse(gfid, gfid_str);
+    snprintf(buf, len, "%s.%d", gfid_str, block_num);
+}
+
+void
+shard_make_block_abspath(int block_num, uuid_t gfid, char *filepath, size_t len)
+{
+    char gfid_str[GF_UUID_BUF_SIZE] = {
+        0,
+    };
+
+    gf_uuid_unparse(gfid, gfid_str);
+    snprintf(filepath, len, "/%s/%s.%d", GF_SHARD_DIR, gfid_str, block_num);
+}
+
+int
+__shard_inode_ctx_get(inode_t *inode, xlator_t *this, shard_inode_ctx_t **ctx)
+{
+    int ret = -1;
+    uint64_t ctx_uint = 0;
+    shard_inode_ctx_t *ctx_p = NULL;
+
+    ret = __inode_ctx_get(inode, this, &ctx_uint);
+    if (ret == 0) {
+        *ctx = (shard_inode_ctx_t *)(uintptr_t)ctx_uint;
+        return ret;
+    }
+
+    ctx_p = GF_CALLOC(1, sizeof(*ctx_p), gf_shard_mt_inode_ctx_t);
+    if (!ctx_p)
+        return ret;
+
+    INIT_LIST_HEAD(&ctx_p->ilist);
+    INIT_LIST_HEAD(&ctx_p->to_fsync_list);
+
+    ctx_uint = (uint64_t)(uintptr_t)ctx_p;
+    ret = __inode_ctx_set(inode, this, &ctx_uint);
+    if (ret < 0) {
+        GF_FREE(ctx_p);
+        return ret;
+    }
+
+    *ctx = ctx_p;
+
+    return ret;
+}
+
+int
+shard_inode_ctx_get(inode_t *inode, xlator_t *this, shard_inode_ctx_t **ctx)
+{
+    int ret = 0;
+
+    LOCK(&inode->lock);
+    {
+        ret = __shard_inode_ctx_get(inode, this, ctx);
+    }
+    UNLOCK(&inode->lock);
+
+    return ret;
+}
+
+int
+__shard_inode_ctx_set(inode_t *inode, xlator_t *this, struct iatt *stbuf,
+                      uint64_t block_size, int32_t valid)
+{
+    int ret = -1;
+    shard_inode_ctx_t *ctx = NULL;
+
+    ret = __shard_inode_ctx_get(inode, this, &ctx);
+    if (ret)
+        return ret;
+
+    if (valid & SHARD_MASK_BLOCK_SIZE)
+        ctx->block_size = block_size;
+
+    if (valid & SHARD_MASK_PROT)
+        ctx->stat.ia_prot = stbuf->ia_prot;
+
+    if (valid & SHARD_MASK_NLINK)
+        ctx->stat.ia_nlink = stbuf->ia_nlink;
+
+    if (valid & SHARD_MASK_UID)
+        ctx->stat.ia_uid = stbuf->ia_uid;
+
+    if (valid & SHARD_MASK_GID)
+        ctx->stat.ia_gid = stbuf->ia_gid;
+
+    if (valid & SHARD_MASK_SIZE)
+        ctx->stat.ia_size = stbuf->ia_size;
+
+    if (valid & SHARD_MASK_BLOCKS)
+        ctx->stat.ia_blocks = stbuf->ia_blocks;
+
+    if (valid & SHARD_MASK_TIMES) {
+        SHARD_TIME_UPDATE(ctx->stat.ia_mtime, ctx->stat.ia_mtime_nsec,
+                          stbuf->ia_mtime, stbuf->ia_mtime_nsec);
+        SHARD_TIME_UPDATE(ctx->stat.ia_ctime, ctx->stat.ia_ctime_nsec,
+                          stbuf->ia_ctime, stbuf->ia_ctime_nsec);
+        SHARD_TIME_UPDATE(ctx->stat.ia_atime, ctx->stat.ia_atime_nsec,
+                          stbuf->ia_atime, stbuf->ia_atime_nsec);
+    }
+
+    if (valid & SHARD_MASK_OTHERS) {
+        ctx->stat.ia_ino = stbuf->ia_ino;
+        gf_uuid_copy(ctx->stat.ia_gfid, stbuf->ia_gfid);
+        ctx->stat.ia_dev = stbuf->ia_dev;
+        ctx->stat.ia_type = stbuf->ia_type;
+        ctx->stat.ia_rdev = stbuf->ia_rdev;
+        ctx->stat.ia_blksize = stbuf->ia_blksize;
+    }
+
+    if (valid & SHARD_MASK_REFRESH_RESET)
+        ctx->refresh = _gf_false;
+
+    return 0;
+}
+
+int
+shard_inode_ctx_set(inode_t *inode, xlator_t *this, struct iatt *stbuf,
+                    uint64_t block_size, int32_t valid)
+{
+    int ret = -1;
+
+    LOCK(&inode->lock);
+    {
+        ret = __shard_inode_ctx_set(inode, this, stbuf, block_size, valid);
+    }
+    UNLOCK(&inode->lock);
+
+    return ret;
+}
+
+int
+__shard_inode_ctx_set_refresh_flag(inode_t *inode, xlator_t *this)
+{
+    int ret = -1;
+    shard_inode_ctx_t *ctx = NULL;
+
+    ret = __shard_inode_ctx_get(inode, this, &ctx);
+    if (ret)
+        return ret;
+
+    ctx->refresh = _gf_true;
+
+    return 0;
+}
+int
+shard_inode_ctx_set_refresh_flag(inode_t *inode, xlator_t *this)
+{
+    int ret = -1;
+
+    LOCK(&inode->lock);
+    {
+        ret = __shard_inode_ctx_set_refresh_flag(inode, this);
+    }
+    UNLOCK(&inode->lock);
+
+    return ret;
+}
+
+int
+__shard_inode_ctx_mark_dir_refreshed(inode_t *inode, xlator_t *this)
+{
+    int ret = -1;
+    shard_inode_ctx_t *ctx = NULL;
+
+    ret = __shard_inode_ctx_get(inode, this, &ctx);
+    if (ret)
+        return ret;
+
+    ctx->refreshed = _gf_true;
+    return 0;
+}
+
+int
+shard_inode_ctx_mark_dir_refreshed(inode_t *inode, xlator_t *this)
+{
+    int ret = -1;
+
+    LOCK(&inode->lock);
+    {
+        ret = __shard_inode_ctx_mark_dir_refreshed(inode, this);
+    }
+    UNLOCK(&inode->lock);
+
+    return ret;
+}
+
+int
+__shard_inode_ctx_add_to_fsync_list(inode_t *base_inode, xlator_t *this,
+                                    inode_t *shard_inode)
+{
+    int ret = -1;
+    shard_inode_ctx_t *base_ictx = NULL;
+    shard_inode_ctx_t *shard_ictx = NULL;
+
+    ret = __shard_inode_ctx_get(base_inode, this, &base_ictx);
+    if (ret)
+        return ret;
+
+    ret = __shard_inode_ctx_get(shard_inode, this, &shard_ictx);
+    if (ret)
+        return ret;
+
+    if (shard_ictx->fsync_needed) {
+        shard_ictx->fsync_needed++;
+        return 1;
+    }
+
+    list_add_tail(&shard_ictx->to_fsync_list, &base_ictx->to_fsync_list);
+    shard_ictx->inode = shard_inode;
+    shard_ictx->fsync_needed++;
+    base_ictx->fsync_count++;
+    shard_ictx->base_inode = base_inode;
+
+    return 0;
+}
+
+int
+shard_inode_ctx_add_to_fsync_list(inode_t *base_inode, xlator_t *this,
+                                  inode_t *shard_inode)
+{
+    int ret = -1;
+
+    /* This ref acts as a refkeepr on the base inode. We
+     * need to keep this inode alive as it holds the head
+     * of the to_fsync_list.
+     */
+    inode_ref(base_inode);
+    inode_ref(shard_inode);
+
+    LOCK(&base_inode->lock);
+    LOCK(&shard_inode->lock);
+    {
+        ret = __shard_inode_ctx_add_to_fsync_list(base_inode, this,
+                                                  shard_inode);
+    }
+    UNLOCK(&shard_inode->lock);
+    UNLOCK(&base_inode->lock);
+
+    /* Unref the base inode corresponding to the ref above, if the shard is
+     * found to be already part of the fsync list.
+     */
+    if (ret != 0) {
+        inode_unref(base_inode);
+        inode_unref(shard_inode);
+    }
+    return ret;
+}
+
+gf_boolean_t
+__shard_inode_ctx_needs_lookup(inode_t *inode, xlator_t *this)
+{
+    int ret = -1;
+    shard_inode_ctx_t *ctx = NULL;
+
+    ret = __shard_inode_ctx_get(inode, this, &ctx);
+    /* If inode ctx get fails, better to err on the side of caution and
+     * try again? Unless the failure is due to mem-allocation.
+     */
+    if (ret)
+        return _gf_true;
+
+    return !ctx->refreshed;
+}
+
+gf_boolean_t
+shard_inode_ctx_needs_lookup(inode_t *inode, xlator_t *this)
+{
+    gf_boolean_t flag = _gf_false;
+
+    LOCK(&inode->lock);
+    {
+        flag = __shard_inode_ctx_needs_lookup(inode, this);
+    }
+    UNLOCK(&inode->lock);
+
+    return flag;
+}
+int
+__shard_inode_ctx_invalidate(inode_t *inode, xlator_t *this, struct iatt *stbuf)
+{
+    int ret = -1;
+    shard_inode_ctx_t *ctx = NULL;
+
+    ret = __shard_inode_ctx_get(inode, this, &ctx);
+    if (ret)
+        return ret;
+
+    if ((stbuf->ia_size != ctx->stat.ia_size) ||
+        (stbuf->ia_blocks != ctx->stat.ia_blocks))
+        ctx->refresh = _gf_true;
+
+    return 0;
+}
+
+int
+shard_inode_ctx_invalidate(inode_t *inode, xlator_t *this, struct iatt *stbuf)
+{
+    int ret = -1;
+
+    LOCK(&inode->lock);
+    {
+        ret = __shard_inode_ctx_invalidate(inode, this, stbuf);
+    }
+    UNLOCK(&inode->lock);
+
+    return ret;
+}
+
+int
+__shard_inode_ctx_get_block_size(inode_t *inode, xlator_t *this,
+                                 uint64_t *block_size)
+{
+    int ret = -1;
+    uint64_t ctx_uint = 0;
+    shard_inode_ctx_t *ctx = NULL;
+
+    ret = __inode_ctx_get(inode, this, &ctx_uint);
+    if (ret < 0)
+        return ret;
+
+    ctx = (shard_inode_ctx_t *)(uintptr_t)ctx_uint;
+
+    *block_size = ctx->block_size;
+
+    return 0;
+}
+
+int
+shard_inode_ctx_get_block_size(inode_t *inode, xlator_t *this,
+                               uint64_t *block_size)
+{
+    int ret = -1;
+
+    LOCK(&inode->lock);
+    {
+        ret = __shard_inode_ctx_get_block_size(inode, this, block_size);
+    }
+    UNLOCK(&inode->lock);
+
+    return ret;
+}
+
+int
+__shard_inode_ctx_get_fsync_count(inode_t *inode, xlator_t *this,
+                                  int *fsync_count)
+{
+    int ret = -1;
+    uint64_t ctx_uint = 0;
+    shard_inode_ctx_t *ctx = NULL;
+
+    ret = __inode_ctx_get(inode, this, &ctx_uint);
+    if (ret < 0)
+        return ret;
+
+    ctx = (shard_inode_ctx_t *)(uintptr_t)ctx_uint;
+
+    *fsync_count = ctx->fsync_needed;
+
+    return 0;
+}
+
+int
+shard_inode_ctx_get_fsync_count(inode_t *inode, xlator_t *this,
+                                int *fsync_count)
+{
+    int ret = -1;
+
+    LOCK(&inode->lock);
+    {
+        ret = __shard_inode_ctx_get_fsync_count(inode, this, fsync_count);
+    }
+    UNLOCK(&inode->lock);
+
+    return ret;
+}
+int
+__shard_inode_ctx_get_all(inode_t *inode, xlator_t *this,
+                          shard_inode_ctx_t *ctx_out)
+{
+    int ret = -1;
+    uint64_t ctx_uint = 0;
+    shard_inode_ctx_t *ctx = NULL;
+
+    ret = __inode_ctx_get(inode, this, &ctx_uint);
+    if (ret < 0)
+        return ret;
+
+    ctx = (shard_inode_ctx_t *)(uintptr_t)ctx_uint;
+
+    memcpy(ctx_out, ctx, sizeof(shard_inode_ctx_t));
+    return 0;
+}
+
+int
+shard_inode_ctx_get_all(inode_t *inode, xlator_t *this,
+                        shard_inode_ctx_t *ctx_out)
+{
+    int ret = -1;
+
+    LOCK(&inode->lock);
+    {
+        ret = __shard_inode_ctx_get_all(inode, this, ctx_out);
+    }
+    UNLOCK(&inode->lock);
+
+    return ret;
+}
+
+int
+__shard_inode_ctx_fill_iatt_from_cache(inode_t *inode, xlator_t *this,
+                                       struct iatt *buf,
+                                       gf_boolean_t *need_refresh)
+{
+    int ret = -1;
+    uint64_t ctx_uint = 0;
+    shard_inode_ctx_t *ctx = NULL;
+
+    ret = __inode_ctx_get(inode, this, &ctx_uint);
+    if (ret < 0)
+        return ret;
+
+    ctx = (shard_inode_ctx_t *)(uintptr_t)ctx_uint;
+
+    if (ctx->refresh == _gf_false)
+        *buf = ctx->stat;
+    else
+        *need_refresh = _gf_true;
+
+    return 0;
+}
+
+int
+shard_inode_ctx_fill_iatt_from_cache(inode_t *inode, xlator_t *this,
+                                     struct iatt *buf,
+                                     gf_boolean_t *need_refresh)
+{
+    int ret = -1;
+
+    LOCK(&inode->lock);
+    {
+        ret = __shard_inode_ctx_fill_iatt_from_cache(inode, this, buf,
+                                                     need_refresh);
+    }
+    UNLOCK(&inode->lock);
+
+    return ret;
+}
+
+void
+shard_local_wipe(shard_local_t *local)
+{
+    int i = 0;
+    int count = 0;
+
+    count = local->num_blocks;
+
+    syncbarrier_destroy(&local->barrier);
+    loc_wipe(&local->loc);
+    loc_wipe(&local->dot_shard_loc);
+    loc_wipe(&local->dot_shard_rm_loc);
+    loc_wipe(&local->loc2);
+    loc_wipe(&local->tmp_loc);
+    loc_wipe(&local->int_inodelk.loc);
+    loc_wipe(&local->int_entrylk.loc);
+    loc_wipe(&local->newloc);
+
+    if (local->name)
+        GF_FREE(local->name);
+
+    if (local->int_entrylk.basename)
+        GF_FREE(local->int_entrylk.basename);
+    if (local->fd)
+        fd_unref(local->fd);
+
+    if (local->xattr_req)
+        dict_unref(local->xattr_req);
+    if (local->xattr_rsp)
+        dict_unref(local->xattr_rsp);
+
+    for (i = 0; i < count; i++) {
+        if (!local->inode_list)
+            break;
+
+        if (local->inode_list[i])
+            inode_unref(local->inode_list[i]);
+    }
+
+    GF_FREE(local->inode_list);
+
+    GF_FREE(local->vector);
+    if (local->iobref)
+        iobref_unref(local->iobref);
+    if (local->list_inited)
+        gf_dirent_free(&local->entries_head);
+    if (local->inodelk_frame)
+        SHARD_STACK_DESTROY(local->inodelk_frame);
+    if (local->entrylk_frame)
+        SHARD_STACK_DESTROY(local->entrylk_frame);
+}
+
+int
+shard_modify_size_and_block_count(struct iatt *stbuf, dict_t *dict)
+{
+    int ret = -1;
+    void *size_attr = NULL;
+    uint64_t size_array[4];
+
+    ret = dict_get_ptr(dict, GF_XATTR_SHARD_FILE_SIZE, &size_attr);
+    if (ret) {
+        gf_msg_callingfn(THIS->name, GF_LOG_ERROR, 0,
+                         SHARD_MSG_INTERNAL_XATTR_MISSING,
+                         "Failed to "
+                         "get " GF_XATTR_SHARD_FILE_SIZE " for %s",
+                         uuid_utoa(stbuf->ia_gfid));
+        return ret;
+    }
+
+    memcpy(size_array, size_attr, sizeof(size_array));
+
+    stbuf->ia_size = ntoh64(size_array[0]);
+    stbuf->ia_blocks = ntoh64(size_array[2]);
+
+    return 0;
+}
+
+int
+shard_call_count_return(call_frame_t *frame)
+{
+    int call_count = 0;
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    LOCK(&frame->lock);
+    {
+        call_count = --local->call_count;
+    }
+    UNLOCK(&frame->lock);
+
+    return call_count;
+}
+
+static char *
+shard_internal_dir_string(shard_internal_dir_type_t type)
+{
+    char *str = NULL;
+
+    switch (type) {
+        case SHARD_INTERNAL_DIR_DOT_SHARD:
+            str = GF_SHARD_DIR;
+            break;
+        case SHARD_INTERNAL_DIR_DOT_SHARD_REMOVE_ME:
+            str = GF_SHARD_REMOVE_ME_DIR;
+            break;
+        default:
+            break;
+    }
+    return str;
+}
+
+static int
+shard_init_internal_dir_loc(xlator_t *this, shard_local_t *local,
+                            shard_internal_dir_type_t type)
+{
+    int ret = -1;
+    char *bname = NULL;
+    inode_t *parent = NULL;
+    loc_t *internal_dir_loc = NULL;
+    shard_priv_t *priv = NULL;
+
+    priv = this->private;
+    if (!local)
+        return -1;
+
+    switch (type) {
+        case SHARD_INTERNAL_DIR_DOT_SHARD:
+            internal_dir_loc = &local->dot_shard_loc;
+            bname = GF_SHARD_DIR;
+            parent = inode_ref(this->itable->root);
+            break;
+        case SHARD_INTERNAL_DIR_DOT_SHARD_REMOVE_ME:
+            internal_dir_loc = &local->dot_shard_rm_loc;
+            bname = GF_SHARD_REMOVE_ME_DIR;
+            parent = inode_ref(priv->dot_shard_inode);
+            break;
+        default:
+            break;
+    }
+
+    internal_dir_loc->inode = inode_new(this->itable);
+    internal_dir_loc->parent = parent;
+    ret = inode_path(internal_dir_loc->parent, bname,
+                     (char **)&internal_dir_loc->path);
+    if (ret < 0 || !(internal_dir_loc->inode)) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_PATH_FAILED,
+               "Inode path failed on %s", bname);
+        goto out;
+    }
+
+    internal_dir_loc->name = strrchr(internal_dir_loc->path, '/');
+    if (internal_dir_loc->name)
+        internal_dir_loc->name++;
+
+    ret = 0;
+out:
+    return ret;
+}
+
+inode_t *
+__shard_update_shards_inode_list(inode_t *linked_inode, xlator_t *this,
+                                 inode_t *base_inode, int block_num,
+                                 uuid_t gfid)
+{
+    char block_bname[256] = {
+        0,
+    };
+    inode_t *lru_inode = NULL;
+    shard_priv_t *priv = NULL;
+    shard_inode_ctx_t *ctx = NULL;
+    shard_inode_ctx_t *lru_inode_ctx = NULL;
+    shard_inode_ctx_t *lru_base_inode_ctx = NULL;
+    inode_t *fsync_inode = NULL;
+    inode_t *lru_base_inode = NULL;
+    gf_boolean_t do_fsync = _gf_false;
+
+    priv = this->private;
+
+    shard_inode_ctx_get(linked_inode, this, &ctx);
+
+    if (list_empty(&ctx->ilist)) {
+        if (priv->inode_count + 1 <= priv->lru_limit) {
+            /* If this inode was linked here for the first time (indicated
+             * by empty list), and if there is still space in the priv list,
+             * add this ctx to the tail of the list.
+             */
+            /* For as long as an inode is in lru list, we try to
+             * keep it alive by holding a ref on it.
+             */
+            inode_ref(linked_inode);
+            if (base_inode)
+                gf_uuid_copy(ctx->base_gfid, base_inode->gfid);
+            else
+                gf_uuid_copy(ctx->base_gfid, gfid);
+            ctx->block_num = block_num;
+            list_add_tail(&ctx->ilist, &priv->ilist_head);
+            priv->inode_count++;
+            ctx->base_inode = inode_ref(base_inode);
+        } else {
+            /*If on the other hand there is no available slot for this inode
+             * in the list, delete the lru inode from the head of the list,
+             * unlink it. And in its place add this new inode into the list.
+             */
+            lru_inode_ctx = list_first_entry(&priv->ilist_head,
+                                             shard_inode_ctx_t, ilist);
+            GF_ASSERT(lru_inode_ctx->block_num > 0);
+            lru_base_inode = lru_inode_ctx->base_inode;
+            list_del_init(&lru_inode_ctx->ilist);
+            lru_inode = inode_find(linked_inode->table,
+                                   lru_inode_ctx->stat.ia_gfid);
+            /* If the lru inode was part of the pending-fsync list,
+             * the base inode needs to be unref'd, the lru inode
+             * deleted from fsync list and fsync'd in a new frame,
+             * and then unlinked in memory and forgotten.
+             */
+            if (!lru_base_inode)
+                goto after_fsync_check;
+            LOCK(&lru_base_inode->lock);
+            LOCK(&lru_inode->lock);
+            {
+                if (!list_empty(&lru_inode_ctx->to_fsync_list)) {
+                    list_del_init(&lru_inode_ctx->to_fsync_list);
+                    lru_inode_ctx->fsync_needed = 0;
+                    do_fsync = _gf_true;
+                    __shard_inode_ctx_get(lru_base_inode, this,
+                                          &lru_base_inode_ctx);
+                    lru_base_inode_ctx->fsync_count--;
+                }
+            }
+            UNLOCK(&lru_inode->lock);
+            UNLOCK(&lru_base_inode->lock);
+
+        after_fsync_check:
+            if (!do_fsync) {
+                shard_make_block_bname(lru_inode_ctx->block_num,
+                                       lru_inode_ctx->base_gfid, block_bname,
+                                       sizeof(block_bname));
+                /* The following unref corresponds to the ref held at
+                 * the time the shard was added to the lru list.
+                 */
+                inode_unref(lru_inode);
+                inode_unlink(lru_inode, priv->dot_shard_inode, block_bname);
+                inode_forget(lru_inode, 0);
+            } else {
+                /* The following unref corresponds to the ref
+                 * held when the shard was added to fsync list.
+                 */
+                inode_unref(lru_inode);
+                fsync_inode = lru_inode;
+                if (lru_base_inode)
+                    inode_unref(lru_base_inode);
+            }
+            /* The following unref corresponds to the ref
+             * held by inode_find() above.
+             */
+            inode_unref(lru_inode);
+
+            /* The following unref corresponds to the ref held on the base shard
+             * at the time of adding shard inode to lru list
+             */
+            if (lru_base_inode)
+                inode_unref(lru_base_inode);
+
+            /* For as long as an inode is in lru list, we try to
+             * keep it alive by holding a ref on it.
+             */
+            inode_ref(linked_inode);
+            if (base_inode)
+                gf_uuid_copy(ctx->base_gfid, base_inode->gfid);
+            else
+                gf_uuid_copy(ctx->base_gfid, gfid);
+            ctx->block_num = block_num;
+            ctx->base_inode = inode_ref(base_inode);
+            list_add_tail(&ctx->ilist, &priv->ilist_head);
+        }
+    } else {
+        /* If this is not the first time this inode is being operated on, move
+         * it to the most recently used end of the list.
+         */
+        list_move_tail(&ctx->ilist, &priv->ilist_head);
+    }
+    return fsync_inode;
+}
+
+int
+shard_common_failure_unwind(glusterfs_fop_t fop, call_frame_t *frame,
+                            int32_t op_ret, int32_t op_errno)
+{
+    switch (fop) {
+        case GF_FOP_LOOKUP:
+            SHARD_STACK_UNWIND(lookup, frame, op_ret, op_errno, NULL, NULL,
+                               NULL, NULL);
+            break;
+        case GF_FOP_STAT:
+            SHARD_STACK_UNWIND(stat, frame, op_ret, op_errno, NULL, NULL);
+            break;
+        case GF_FOP_FSTAT:
+            SHARD_STACK_UNWIND(fstat, frame, op_ret, op_errno, NULL, NULL);
+            break;
+        case GF_FOP_TRUNCATE:
+            SHARD_STACK_UNWIND(truncate, frame, op_ret, op_errno, NULL, NULL,
+                               NULL);
+            break;
+        case GF_FOP_FTRUNCATE:
+            SHARD_STACK_UNWIND(ftruncate, frame, op_ret, op_errno, NULL, NULL,
+                               NULL);
+            break;
+        case GF_FOP_MKNOD:
+            SHARD_STACK_UNWIND(mknod, frame, op_ret, op_errno, NULL, NULL, NULL,
+                               NULL, NULL);
+            break;
+        case GF_FOP_LINK:
+            SHARD_STACK_UNWIND(link, frame, op_ret, op_errno, NULL, NULL, NULL,
+                               NULL, NULL);
+            break;
+        case GF_FOP_CREATE:
+            SHARD_STACK_UNWIND(create, frame, op_ret, op_errno, NULL, NULL,
+                               NULL, NULL, NULL, NULL);
+            break;
+        case GF_FOP_UNLINK:
+            SHARD_STACK_UNWIND(unlink, frame, op_ret, op_errno, NULL, NULL,
+                               NULL);
+            break;
+        case GF_FOP_RENAME:
+            SHARD_STACK_UNWIND(rename, frame, op_ret, op_errno, NULL, NULL,
+                               NULL, NULL, NULL, NULL);
+            break;
+        case GF_FOP_WRITE:
+            SHARD_STACK_UNWIND(writev, frame, op_ret, op_errno, NULL, NULL,
+                               NULL);
+            break;
+        case GF_FOP_FALLOCATE:
+            SHARD_STACK_UNWIND(fallocate, frame, op_ret, op_errno, NULL, NULL,
+                               NULL);
+            break;
+        case GF_FOP_ZEROFILL:
+            SHARD_STACK_UNWIND(zerofill, frame, op_ret, op_errno, NULL, NULL,
+                               NULL);
+            break;
+        case GF_FOP_DISCARD:
+            SHARD_STACK_UNWIND(discard, frame, op_ret, op_errno, NULL, NULL,
+                               NULL);
+            break;
+        case GF_FOP_READ:
+            SHARD_STACK_UNWIND(readv, frame, op_ret, op_errno, NULL, -1, NULL,
+                               NULL, NULL);
+            break;
+        case GF_FOP_FSYNC:
+            SHARD_STACK_UNWIND(fsync, frame, op_ret, op_errno, NULL, NULL,
+                               NULL);
+            break;
+        case GF_FOP_REMOVEXATTR:
+            SHARD_STACK_UNWIND(removexattr, frame, op_ret, op_errno, NULL);
+            break;
+        case GF_FOP_FREMOVEXATTR:
+            SHARD_STACK_UNWIND(fremovexattr, frame, op_ret, op_errno, NULL);
+            break;
+        case GF_FOP_FGETXATTR:
+            SHARD_STACK_UNWIND(fgetxattr, frame, op_ret, op_errno, NULL, NULL);
+            break;
+        case GF_FOP_GETXATTR:
+            SHARD_STACK_UNWIND(getxattr, frame, op_ret, op_errno, NULL, NULL);
+            break;
+        case GF_FOP_FSETXATTR:
+            SHARD_STACK_UNWIND(fsetxattr, frame, op_ret, op_errno, NULL);
+            break;
+        case GF_FOP_SETXATTR:
+            SHARD_STACK_UNWIND(setxattr, frame, op_ret, op_errno, NULL);
+            break;
+        case GF_FOP_SETATTR:
+            SHARD_STACK_UNWIND(setattr, frame, op_ret, op_errno, NULL, NULL,
+                               NULL);
+            break;
+        case GF_FOP_FSETATTR:
+            SHARD_STACK_UNWIND(fsetattr, frame, op_ret, op_errno, NULL, NULL,
+                               NULL);
+            break;
+        case GF_FOP_SEEK:
+            SHARD_STACK_UNWIND(seek, frame, op_ret, op_errno, 0, NULL);
+            break;
+        default:
+            gf_msg(THIS->name, GF_LOG_WARNING, 0, SHARD_MSG_INVALID_FOP,
+                   "Invalid fop id = %d", fop);
+            break;
+    }
+    return 0;
+}
+
+int
+shard_common_inode_write_success_unwind(glusterfs_fop_t fop,
+                                        call_frame_t *frame, int32_t op_ret)
+{
+    shard_local_t *local = frame->local;
+
+    /* the below 3 variables are required because, in SHARD_STACK_UNWIND()
+       macro, there is a check for local being null. So many static analyzers
+       backtrace the code with assumption of possible (local == NULL) case,
+       and complains for below lines. By handling it like below, we overcome
+       the warnings */
+
+    struct iatt *prebuf = ((local) ? &local->prebuf : NULL);
+    struct iatt *postbuf = ((local) ? &local->postbuf : NULL);
+    dict_t *xattr_rsp = ((local) ? local->xattr_rsp : NULL);
+
+    switch (fop) {
+        case GF_FOP_WRITE:
+            SHARD_STACK_UNWIND(writev, frame, op_ret, 0, prebuf, postbuf,
+                               xattr_rsp);
+            break;
+        case GF_FOP_FALLOCATE:
+            SHARD_STACK_UNWIND(fallocate, frame, op_ret, 0, prebuf, postbuf,
+                               xattr_rsp);
+            break;
+        case GF_FOP_ZEROFILL:
+            SHARD_STACK_UNWIND(zerofill, frame, op_ret, 0, prebuf, postbuf,
+                               xattr_rsp);
+            break;
+        case GF_FOP_DISCARD:
+            SHARD_STACK_UNWIND(discard, frame, op_ret, 0, prebuf, postbuf,
+                               xattr_rsp);
+            break;
+        default:
+            gf_msg(THIS->name, GF_LOG_WARNING, 0, SHARD_MSG_INVALID_FOP,
+                   "Invalid fop id = %d", fop);
+            break;
+    }
+    return 0;
+}
+
+int
+shard_evicted_inode_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                              int32_t op_ret, int32_t op_errno,
+                              struct iatt *prebuf, struct iatt *postbuf,
+                              dict_t *xdata)
+{
+    char block_bname[256] = {
+        0,
+    };
+    fd_t *anon_fd = cookie;
+    inode_t *shard_inode = NULL;
+    shard_inode_ctx_t *ctx = NULL;
+    shard_priv_t *priv = NULL;
+
+    priv = this->private;
+
+    if (anon_fd == NULL || op_ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, op_errno, SHARD_MSG_MEMALLOC_FAILED,
+               "fsync failed on shard");
+        goto out;
+    }
+    shard_inode = anon_fd->inode;
+
+    LOCK(&priv->lock);
+    LOCK(&shard_inode->lock);
+    {
+        __shard_inode_ctx_get(shard_inode, this, &ctx);
+        if ((list_empty(&ctx->to_fsync_list)) && (list_empty(&ctx->ilist))) {
+            shard_make_block_bname(ctx->block_num, shard_inode->gfid,
+                                   block_bname, sizeof(block_bname));
+            inode_unlink(shard_inode, priv->dot_shard_inode, block_bname);
+            /* The following unref corresponds to the ref held by
+             * inode_link() at the time the shard was created or
+             * looked up
+             */
+            inode_unref(shard_inode);
+            inode_forget(shard_inode, 0);
+        }
+    }
+    UNLOCK(&shard_inode->lock);
+    UNLOCK(&priv->lock);
+
+out:
+    if (anon_fd)
+        fd_unref(anon_fd);
+    STACK_DESTROY(frame->root);
+    return 0;
+}
+
+int
+shard_initiate_evicted_inode_fsync(xlator_t *this, inode_t *inode)
+{
+    fd_t *anon_fd = NULL;
+    call_frame_t *fsync_frame = NULL;
+
+    fsync_frame = create_frame(this, this->ctx->pool);
+    if (!fsync_frame) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, SHARD_MSG_MEMALLOC_FAILED,
+               "Failed to create new frame "
+               "to fsync shard");
+        return -1;
+    }
+
+    anon_fd = fd_anonymous(inode);
+    if (!anon_fd) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, SHARD_MSG_MEMALLOC_FAILED,
+               "Failed to create anon fd to"
+               " fsync shard");
+        STACK_DESTROY(fsync_frame->root);
+        return -1;
+    }
+
+    STACK_WIND_COOKIE(fsync_frame, shard_evicted_inode_fsync_cbk, anon_fd,
+                      FIRST_CHILD(this), FIRST_CHILD(this)->fops->fsync,
+                      anon_fd, 1, NULL);
+    return 0;
+}
+
+int
+shard_common_inode_write_post_lookup_shards_handler(call_frame_t *frame,
+                                                    xlator_t *this);
+
+int
+shard_common_resolve_shards(call_frame_t *frame, xlator_t *this,
+                            shard_post_resolve_fop_handler_t post_res_handler)
+{
+    int i = -1;
+    uint32_t shard_idx_iter = 0;
+    char path[PATH_MAX] = {
+        0,
+    };
+    uuid_t gfid = {
+        0,
+    };
+    inode_t *inode = NULL;
+    inode_t *res_inode = NULL;
+    inode_t *fsync_inode = NULL;
+    shard_priv_t *priv = NULL;
+    shard_local_t *local = NULL;
+    uint64_t resolve_count = 0;
+
+    priv = this->private;
+    local = frame->local;
+    local->call_count = 0;
+    shard_idx_iter = local->first_block;
+    res_inode = local->resolver_base_inode;
+
+    if ((local->op_ret < 0) || (local->resolve_not))
+        goto out;
+
+    /* If this prealloc FOP is for fresh file creation, then the size of the
+     * file will be 0. Then there will be no shards associated with this file.
+     * So we can skip the lookup process for the shards which do not exists
+     * and directly issue mknod to crete shards.
+     *
+     * In case the prealloc fop is to extend the preallocated file to bigger
+     * size then just lookup and populate inodes of existing shards and
+     * update the create count
+     */
+    if (local->fop == GF_FOP_FALLOCATE) {
+        if (!local->prebuf.ia_size) {
+            local->inode_list[0] = inode_ref(res_inode);
+            local->create_count = local->last_block;
+            shard_common_inode_write_post_lookup_shards_handler(frame, this);
+            return 0;
+        }
+        if (local->prebuf.ia_size < local->total_size)
+            local->create_count = local->last_block -
+                                  ((local->prebuf.ia_size - 1) /
+                                   local->block_size);
+    }
+
+    resolve_count = local->last_block - local->create_count;
+
+    if (res_inode)
+        gf_uuid_copy(gfid, res_inode->gfid);
+    else
+        gf_uuid_copy(gfid, local->base_gfid);
+
+    while (shard_idx_iter <= resolve_count) {
+        i++;
+        if (shard_idx_iter == 0) {
+            local->inode_list[i] = inode_ref(res_inode);
+            shard_idx_iter++;
+            continue;
+        }
+
+        shard_make_block_abspath(shard_idx_iter, gfid, path, sizeof(path));
+
+        inode = NULL;
+        inode = inode_resolve(this->itable, path);
+        if (inode) {
+            gf_msg_debug(this->name, 0,
+                         "Shard %d already "
+                         "present. gfid=%s. Saving inode for future.",
+                         shard_idx_iter, uuid_utoa(inode->gfid));
+            local->inode_list[i] = inode;
+            /* Let the ref on the inodes that are already present
+             * in inode table still be held so that they don't get
+             * forgotten by the time the fop reaches the actual
+             * write stage.
+             */
+            LOCK(&priv->lock);
+            {
+                fsync_inode = __shard_update_shards_inode_list(
+                    inode, this, res_inode, shard_idx_iter, gfid);
+            }
+            UNLOCK(&priv->lock);
+            shard_idx_iter++;
+            if (fsync_inode)
+                shard_initiate_evicted_inode_fsync(this, fsync_inode);
+            continue;
+        } else {
+            local->call_count++;
+            shard_idx_iter++;
+        }
+    }
+out:
+    post_res_handler(frame, this);
+    return 0;
+}
+
+int
+shard_update_file_size_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno, dict_t *dict,
+                           dict_t *xdata)
+{
+    inode_t *inode = NULL;
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if ((local->fd) && (local->fd->inode))
+        inode = local->fd->inode;
+    else if (local->loc.inode)
+        inode = local->loc.inode;
+
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno,
+               SHARD_MSG_UPDATE_FILE_SIZE_FAILED,
+               "Update to file size"
+               " xattr failed on %s",
+               uuid_utoa(inode->gfid));
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    if (shard_modify_size_and_block_count(&local->postbuf, dict)) {
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        goto err;
+    }
+err:
+    local->post_update_size_handler(frame, this);
+    return 0;
+}
+
+int
+shard_set_size_attrs(int64_t size, int64_t block_count, int64_t **size_attr_p)
+{
+    int ret = -1;
+    int64_t *size_attr = NULL;
+
+    if (!size_attr_p)
+        goto out;
+
+    size_attr = GF_CALLOC(4, sizeof(int64_t), gf_shard_mt_int64_t);
+    if (!size_attr)
+        goto out;
+
+    size_attr[0] = hton64(size);
+    /* As sharding evolves, it _may_ be necessary to embed more pieces of
+     * information within the same xattr. So allocating slots for them in
+     * advance. For now, only bytes 0-63 and 128-191 which would make up the
+     * current size and block count respectively of the file are valid.
+     */
+    size_attr[2] = hton64(block_count);
+
+    *size_attr_p = size_attr;
+
+    ret = 0;
+out:
+    return ret;
+}
+
+int
+shard_update_file_size(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                       loc_t *loc, shard_post_update_size_fop_handler_t handler)
+{
+    int ret = -1;
+    int64_t *size_attr = NULL;
+    int64_t delta_blocks = 0;
+    inode_t *inode = NULL;
+    shard_local_t *local = NULL;
+    dict_t *xattr_req = NULL;
+
+    local = frame->local;
+    local->post_update_size_handler = handler;
+
+    xattr_req = dict_new();
+    if (!xattr_req) {
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        goto out;
+    }
+
+    if (fd)
+        inode = fd->inode;
+    else
+        inode = loc->inode;
+
+    /* If both size and block count have not changed, then skip the xattrop.
+     */
+    delta_blocks = GF_ATOMIC_GET(local->delta_blocks);
+    if ((local->delta_size + local->hole_size == 0) && (delta_blocks == 0)) {
+        goto out;
+    }
+
+    ret = shard_set_size_attrs(local->delta_size + local->hole_size,
+                               delta_blocks, &size_attr);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_SIZE_SET_FAILED,
+               "Failed to set size attrs for %s", uuid_utoa(inode->gfid));
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        goto out;
+    }
+
+    ret = dict_set_bin(xattr_req, GF_XATTR_SHARD_FILE_SIZE, size_attr, 8 * 4);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_DICT_OP_FAILED,
+               "Failed to set key %s into dict. gfid=%s",
+               GF_XATTR_SHARD_FILE_SIZE, uuid_utoa(inode->gfid));
+        GF_FREE(size_attr);
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        goto out;
+    }
+
+    if (fd)
+        STACK_WIND(frame, shard_update_file_size_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fxattrop, fd,
+                   GF_XATTROP_ADD_ARRAY64, xattr_req, NULL);
+    else
+        STACK_WIND(frame, shard_update_file_size_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->xattrop, loc,
+                   GF_XATTROP_ADD_ARRAY64, xattr_req, NULL);
+
+    dict_unref(xattr_req);
+    return 0;
+
+out:
+    if (xattr_req)
+        dict_unref(xattr_req);
+    handler(frame, this);
+    return 0;
+}
+
+static inode_t *
+shard_link_internal_dir_inode(shard_local_t *local, inode_t *inode,
+                              struct iatt *buf, shard_internal_dir_type_t type)
+{
+    inode_t *linked_inode = NULL;
+    shard_priv_t *priv = NULL;
+    char *bname = NULL;
+    inode_t **priv_inode = NULL;
+    inode_t *parent = NULL;
+
+    priv = THIS->private;
+
+    switch (type) {
+        case SHARD_INTERNAL_DIR_DOT_SHARD:
+            bname = GF_SHARD_DIR;
+            priv_inode = &priv->dot_shard_inode;
+            parent = inode->table->root;
+            break;
+        case SHARD_INTERNAL_DIR_DOT_SHARD_REMOVE_ME:
+            bname = GF_SHARD_REMOVE_ME_DIR;
+            priv_inode = &priv->dot_shard_rm_inode;
+            parent = priv->dot_shard_inode;
+            break;
+        default:
+            break;
+    }
+
+    linked_inode = inode_link(inode, parent, bname, buf);
+    inode_lookup(linked_inode);
+    *priv_inode = linked_inode;
+    return linked_inode;
+}
+
+int
+shard_refresh_internal_dir_cbk(call_frame_t *frame, void *cookie,
+                               xlator_t *this, int32_t op_ret, int32_t op_errno,
+                               inode_t *inode, struct iatt *buf, dict_t *xdata,
+                               struct iatt *postparent)
+{
+    shard_local_t *local = NULL;
+    inode_t *linked_inode = NULL;
+    shard_internal_dir_type_t type = (shard_internal_dir_type_t)cookie;
+
+    local = frame->local;
+
+    if (op_ret) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        goto out;
+    }
+
+    /* To-Do: Fix refcount increment per call to
+     * shard_link_internal_dir_inode().
+     */
+    linked_inode = shard_link_internal_dir_inode(local, inode, buf, type);
+    shard_inode_ctx_mark_dir_refreshed(linked_inode, this);
+out:
+    shard_common_resolve_shards(frame, this, local->post_res_handler);
+    return 0;
+}
+
+int
+shard_refresh_internal_dir(call_frame_t *frame, xlator_t *this,
+                           shard_internal_dir_type_t type)
+{
+    loc_t loc = {
+        0,
+    };
+    inode_t *inode = NULL;
+    shard_priv_t *priv = NULL;
+    shard_local_t *local = NULL;
+    uuid_t gfid = {
+        0,
+    };
+
+    local = frame->local;
+    priv = this->private;
+
+    switch (type) {
+        case SHARD_INTERNAL_DIR_DOT_SHARD:
+            gf_uuid_copy(gfid, priv->dot_shard_gfid);
+            break;
+        case SHARD_INTERNAL_DIR_DOT_SHARD_REMOVE_ME:
+            gf_uuid_copy(gfid, priv->dot_shard_rm_gfid);
+            break;
+        default:
+            break;
+    }
+
+    inode = inode_find(this->itable, gfid);
+
+    if (!shard_inode_ctx_needs_lookup(inode, this)) {
+        local->op_ret = 0;
+        goto out;
+    }
+
+    /* Plain assignment because the ref is already taken above through
+     * call to inode_find()
+     */
+    loc.inode = inode;
+    gf_uuid_copy(loc.gfid, gfid);
+
+    STACK_WIND_COOKIE(frame, shard_refresh_internal_dir_cbk, (void *)(long)type,
+                      FIRST_CHILD(this), FIRST_CHILD(this)->fops->lookup, &loc,
+                      NULL);
+    loc_wipe(&loc);
+
+    return 0;
+
+out:
+    shard_common_resolve_shards(frame, this, local->post_res_handler);
+    return 0;
+}
+
+int
+shard_lookup_internal_dir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                              int32_t op_ret, int32_t op_errno, inode_t *inode,
+                              struct iatt *buf, dict_t *xdata,
+                              struct iatt *postparent)
+{
+    inode_t *link_inode = NULL;
+    shard_local_t *local = NULL;
+    shard_internal_dir_type_t type = (shard_internal_dir_type_t)cookie;
+
+    local = frame->local;
+
+    if (op_ret) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        goto unwind;
+    }
+
+    if (!IA_ISDIR(buf->ia_type)) {
+        gf_msg(this->name, GF_LOG_CRITICAL, 0, SHARD_MSG_DOT_SHARD_NODIR,
+               "%s already exists and "
+               "is not a directory. Please remove it from all bricks "
+               "and try again",
+               shard_internal_dir_string(type));
+        local->op_ret = -1;
+        local->op_errno = EIO;
+        goto unwind;
+    }
+
+    link_inode = shard_link_internal_dir_inode(local, inode, buf, type);
+    if (link_inode != inode) {
+        shard_refresh_internal_dir(frame, this, type);
+    } else {
+        shard_inode_ctx_mark_dir_refreshed(link_inode, this);
+        shard_common_resolve_shards(frame, this, local->post_res_handler);
+    }
+    return 0;
+
+unwind:
+    local->post_res_handler(frame, this);
+    return 0;
+}
+
+int
+shard_lookup_internal_dir(call_frame_t *frame, xlator_t *this,
+                          shard_post_resolve_fop_handler_t post_res_handler,
+                          shard_internal_dir_type_t type)
+{
+    int ret = -1;
+    dict_t *xattr_req = NULL;
+    shard_priv_t *priv = NULL;
+    shard_local_t *local = NULL;
+    uuid_t *gfid = NULL;
+    loc_t *loc = NULL;
+    gf_boolean_t free_gfid = _gf_true;
+
+    local = frame->local;
+    priv = this->private;
+    local->post_res_handler = post_res_handler;
+
+    gfid = GF_MALLOC(sizeof(uuid_t), gf_common_mt_uuid_t);
+    if (!gfid)
+        goto err;
+
+    xattr_req = dict_new();
+    if (!xattr_req) {
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        goto err;
+    }
+
+    switch (type) {
+        case SHARD_INTERNAL_DIR_DOT_SHARD:
+            gf_uuid_copy(*gfid, priv->dot_shard_gfid);
+            loc = &local->dot_shard_loc;
+            break;
+        case SHARD_INTERNAL_DIR_DOT_SHARD_REMOVE_ME:
+            gf_uuid_copy(*gfid, priv->dot_shard_rm_gfid);
+            loc = &local->dot_shard_rm_loc;
+            break;
+        default:
+            bzero(*gfid, sizeof(uuid_t));
+            break;
+    }
+
+    ret = dict_set_gfuuid(xattr_req, "gfid-req", *gfid, false);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_DICT_OP_FAILED,
+               "Failed to set gfid of %s into dict",
+               shard_internal_dir_string(type));
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        goto err;
+    } else {
+        free_gfid = _gf_false;
+    }
+
+    STACK_WIND_COOKIE(frame, shard_lookup_internal_dir_cbk, (void *)(long)type,
+                      FIRST_CHILD(this), FIRST_CHILD(this)->fops->lookup, loc,
+                      xattr_req);
+
+    dict_unref(xattr_req);
+    return 0;
+
+err:
+    if (xattr_req)
+        dict_unref(xattr_req);
+    if (free_gfid)
+        GF_FREE(gfid);
+    post_res_handler(frame, this);
+    return 0;
+}
+
+static void
+shard_inode_ctx_update(inode_t *inode, xlator_t *this, dict_t *xdata,
+                       struct iatt *buf)
+{
+    int ret = 0;
+    uint64_t size = 0;
+    void *bsize = NULL;
+
+    if (shard_inode_ctx_get_block_size(inode, this, &size)) {
+        /* Fresh lookup */
+        ret = dict_get_ptr(xdata, GF_XATTR_SHARD_BLOCK_SIZE, &bsize);
+        if (!ret)
+            size = ntoh64(*((uint64_t *)bsize));
+        /* If the file is sharded, set its block size, otherwise just
+         * set 0.
+         */
+
+        shard_inode_ctx_set(inode, this, buf, size, SHARD_MASK_BLOCK_SIZE);
+    }
+    /* If the file is sharded, also set the remaining attributes,
+     * except for ia_size and ia_blocks.
+     */
+    if (size) {
+        shard_inode_ctx_set(inode, this, buf, 0, SHARD_LOOKUP_MASK);
+        (void)shard_inode_ctx_invalidate(inode, this, buf);
+    }
+}
+
+int
+shard_delete_shards(void *opaque);
+
+int
+shard_delete_shards_cbk(int ret, call_frame_t *frame, void *data);
+
+int
+shard_start_background_deletion(xlator_t *this)
+{
+    int ret = 0;
+    gf_boolean_t i_cleanup = _gf_true;
+    shard_priv_t *priv = NULL;
+    call_frame_t *cleanup_frame = NULL;
+
+    priv = this->private;
+
+    LOCK(&priv->lock);
+    {
+        switch (priv->bg_del_state) {
+            case SHARD_BG_DELETION_NONE:
+                i_cleanup = _gf_true;
+                priv->bg_del_state = SHARD_BG_DELETION_LAUNCHING;
+                break;
+            case SHARD_BG_DELETION_LAUNCHING:
+                i_cleanup = _gf_false;
+                break;
+            case SHARD_BG_DELETION_IN_PROGRESS:
+                priv->bg_del_state = SHARD_BG_DELETION_LAUNCHING;
+                i_cleanup = _gf_false;
+                break;
+            default:
+                break;
+        }
+    }
+    UNLOCK(&priv->lock);
+    if (!i_cleanup)
+        return 0;
+
+    cleanup_frame = create_frame(this, this->ctx->pool);
+    if (!cleanup_frame) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, SHARD_MSG_MEMALLOC_FAILED,
+               "Failed to create "
+               "new frame to delete shards");
+        ret = -ENOMEM;
+        goto err;
+    }
+
+    set_lk_owner_from_ptr(&cleanup_frame->root->lk_owner, cleanup_frame->root);
+
+    ret = synctask_new(this->ctx->env, shard_delete_shards,
+                       shard_delete_shards_cbk, cleanup_frame, cleanup_frame);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, errno,
+               SHARD_MSG_SHARDS_DELETION_FAILED,
+               "failed to create task to do background "
+               "cleanup of shards");
+        STACK_DESTROY(cleanup_frame->root);
+        goto err;
+    }
+    return 0;
+
+err:
+    LOCK(&priv->lock);
+    {
+        priv->bg_del_state = SHARD_BG_DELETION_NONE;
+    }
+    UNLOCK(&priv->lock);
+    return ret;
+}
+
+int
+shard_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, inode_t *inode,
+                 struct iatt *buf, dict_t *xdata, struct iatt *postparent)
+{
+    int ret = -1;
+    shard_priv_t *priv = NULL;
+    gf_boolean_t i_start_cleanup = _gf_false;
+
+    priv = this->private;
+
+    if (op_ret < 0)
+        goto unwind;
+
+    if (IA_ISDIR(buf->ia_type))
+        goto unwind;
+
+    /* Also, if the file is sharded, get the file size and block cnt xattr,
+     * and store them in the stbuf appropriately.
+     */
+
+    if (dict_get(xdata, GF_XATTR_SHARD_FILE_SIZE) &&
+        frame->root->pid != GF_CLIENT_PID_GSYNCD)
+        shard_modify_size_and_block_count(buf, xdata);
+
+    /* If this was a fresh lookup, there are two possibilities:
+     * 1) If the file is sharded (indicated by the presence of block size
+     *    xattr), store this block size, along with rdev and mode in its
+     *    inode ctx.
+     * 2) If the file is not sharded, store size along with rdev and mode
+     *    (which are anyway don't cares) in inode ctx. Since @ctx_tmp is
+     *    already initialised to all zeroes, nothing more needs to be done.
+     */
+
+    (void)shard_inode_ctx_update(inode, this, xdata, buf);
+
+    LOCK(&priv->lock);
+    {
+        if (priv->first_lookup_done == _gf_false) {
+            priv->first_lookup_done = _gf_true;
+            i_start_cleanup = _gf_true;
+        }
+    }
+    UNLOCK(&priv->lock);
+
+    if (!i_start_cleanup)
+        goto unwind;
+
+    ret = shard_start_background_deletion(this);
+    if (ret < 0) {
+        LOCK(&priv->lock);
+        {
+            priv->first_lookup_done = _gf_false;
+        }
+        UNLOCK(&priv->lock);
+    }
+
+unwind:
+    SHARD_STACK_UNWIND(lookup, frame, op_ret, op_errno, inode, buf, xdata,
+                       postparent);
+    return 0;
+}
+
+int
+shard_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
+{
+    int ret = -1;
+    int32_t op_errno = ENOMEM;
+    uint64_t block_size = 0;
+    shard_local_t *local = NULL;
+
+    this->itable = loc->inode->table;
+    if ((frame->root->pid != GF_CLIENT_PID_GSYNCD) &&
+        (frame->root->pid != GF_CLIENT_PID_GLFS_HEAL)) {
+        SHARD_ENTRY_FOP_CHECK(loc, op_errno, err);
+    }
+
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto err;
+
+    frame->local = local;
+
+    loc_copy(&local->loc, loc);
+
+    local->xattr_req = xattr_req ? dict_ref(xattr_req) : dict_new();
+    if (!local->xattr_req)
+        goto err;
+
+    if (shard_inode_ctx_get_block_size(loc->inode, this, &block_size)) {
+        ret = dict_set_uint64(local->xattr_req, GF_XATTR_SHARD_BLOCK_SIZE, 0);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, SHARD_MSG_DICT_OP_FAILED,
+                   "Failed to set dict"
+                   " value: key:%s for path %s",
+                   GF_XATTR_SHARD_BLOCK_SIZE, loc->path);
+            goto err;
+        }
+    }
+
+    if (frame->root->pid != GF_CLIENT_PID_GSYNCD) {
+        ret = dict_set_uint64(local->xattr_req, GF_XATTR_SHARD_FILE_SIZE,
+                              8 * 4);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, SHARD_MSG_DICT_OP_FAILED,
+                   "Failed to set dict value: key:%s for path %s.",
+                   GF_XATTR_SHARD_FILE_SIZE, loc->path);
+            goto err;
+        }
+    }
+
+    if ((xattr_req) && (dict_get(xattr_req, GF_CONTENT_KEY)))
+        dict_del(xattr_req, GF_CONTENT_KEY);
+
+    STACK_WIND(frame, shard_lookup_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lookup, loc, local->xattr_req);
+    return 0;
+err:
+    shard_common_failure_unwind(GF_FOP_LOOKUP, frame, -1, op_errno);
+    return 0;
+}
+
+int
+shard_set_iattr_invoke_post_handler(call_frame_t *frame, xlator_t *this,
+                                    inode_t *inode, int32_t op_ret,
+                                    int32_t op_errno, struct iatt *buf,
+                                    dict_t *xdata)
+{
+    int ret = -1;
+    int32_t mask = SHARD_INODE_WRITE_MASK;
+    shard_local_t *local = frame->local;
+    shard_inode_ctx_t ctx = {
+        0,
+    };
+
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno,
+               SHARD_MSG_BASE_FILE_LOOKUP_FAILED,
+               "Lookup on base file"
+               " failed : %s",
+               uuid_utoa(inode->gfid));
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        goto unwind;
+    }
+
+    local->prebuf = *buf;
+    if (shard_modify_size_and_block_count(&local->prebuf, xdata)) {
+        local->op_ret = -1;
+        local->op_errno = EINVAL;
+        goto unwind;
+    }
+
+    if (shard_inode_ctx_get_all(inode, this, &ctx))
+        mask = SHARD_ALL_MASK;
+
+    ret = shard_inode_ctx_set(inode, this, &local->prebuf, 0,
+                              (mask | SHARD_MASK_REFRESH_RESET));
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, SHARD_MSG_INODE_CTX_SET_FAILED, 0,
+               "Failed to set inode"
+               " write params into inode ctx for %s",
+               uuid_utoa(buf->ia_gfid));
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        goto unwind;
+    }
+
+unwind:
+    local->handler(frame, this);
+    return 0;
+}
+
+int
+shard_fstat_base_file_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                          dict_t *xdata)
+{
+    shard_local_t *local = frame->local;
+
+    shard_set_iattr_invoke_post_handler(frame, this, local->fd->inode, op_ret,
+                                        op_errno, buf, xdata);
+    return 0;
+}
+
+int
+shard_lookup_base_file_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno, inode_t *inode,
+                           struct iatt *buf, dict_t *xdata,
+                           struct iatt *postparent)
+{
+    /* In case of op_ret < 0, inode passed to this function will be NULL
+       ex: in case of op_errno = ENOENT. So refer prefilled inode data
+       which is part of local.
+       Note: Reassigning/overriding the inode passed to this cbk with inode
+       which is part of *struct shard_local_t* won't cause any issue as
+       both inodes have same reference/address as of the inode passed */
+    inode = ((shard_local_t *)frame->local)->loc.inode;
+
+    shard_set_iattr_invoke_post_handler(frame, this, inode, op_ret, op_errno,
+                                        buf, xdata);
+    return 0;
+}
+
+/* This function decides whether to make file based lookup or
+ * fd based lookup (fstat) depending on the 3rd and 4th arg.
+ * If fd != NULL and loc == NULL then call is for fstat
+ * If fd == NULL and loc != NULL then call is for file based
+ * lookup. Please pass args based on the requirement.
+ */
+int
+shard_refresh_base_file(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                        fd_t *fd, shard_post_fop_handler_t handler)
+{
+    int ret = -1;
+    inode_t *inode = NULL;
+    shard_local_t *local = NULL;
+    dict_t *xattr_req = NULL;
+    gf_boolean_t need_refresh = _gf_false;
+
+    local = frame->local;
+    local->handler = handler;
+    inode = fd ? fd->inode : loc->inode;
+
+    ret = shard_inode_ctx_fill_iatt_from_cache(inode, this, &local->prebuf,
+                                               &need_refresh);
+    /* By this time, inode ctx should have been created either in create,
+     * mknod, readdirp or lookup. If not it is a bug!
+     */
+    if ((ret == 0) && (need_refresh == _gf_false)) {
+        gf_msg_debug(this->name, 0,
+                     "Skipping lookup on base file: %s"
+                     "Serving prebuf off the inode ctx cache",
+                     uuid_utoa(inode->gfid));
+        goto out;
+    }
+
+    xattr_req = dict_new();
+    if (!xattr_req) {
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        goto out;
+    }
+
+    SHARD_MD_READ_FOP_INIT_REQ_DICT(this, xattr_req, inode->gfid, local, out);
+
+    if (fd)
+        STACK_WIND(frame, shard_fstat_base_file_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fstat, fd, xattr_req);
+    else
+        STACK_WIND(frame, shard_lookup_base_file_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->lookup, loc, xattr_req);
+
+    dict_unref(xattr_req);
+    return 0;
+
+out:
+    if (xattr_req)
+        dict_unref(xattr_req);
+    handler(frame, this);
+    return 0;
+}
+
+int
+shard_post_fstat_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->op_ret >= 0)
+        shard_inode_ctx_set(local->fd->inode, this, &local->prebuf, 0,
+                            SHARD_LOOKUP_MASK);
+
+    SHARD_STACK_UNWIND(fstat, frame, local->op_ret, local->op_errno,
+                       &local->prebuf, local->xattr_rsp);
+    return 0;
+}
+
+int
+shard_post_stat_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->op_ret >= 0)
+        shard_inode_ctx_set(local->loc.inode, this, &local->prebuf, 0,
+                            SHARD_LOOKUP_MASK);
+
+    SHARD_STACK_UNWIND(stat, frame, local->op_ret, local->op_errno,
+                       &local->prebuf, local->xattr_rsp);
+    return 0;
+}
+
+int
+shard_common_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                      dict_t *xdata)
+{
+    inode_t *inode = NULL;
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, SHARD_MSG_STAT_FAILED,
+               "stat failed: %s",
+               local->fd ? uuid_utoa(local->fd->inode->gfid)
+                         : uuid_utoa((local->loc.inode)->gfid));
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        goto unwind;
+    }
+
+    local->prebuf = *buf;
+    if (shard_modify_size_and_block_count(&local->prebuf, xdata)) {
+        local->op_ret = -1;
+        local->op_errno = EINVAL;
+        goto unwind;
+    }
+    local->xattr_rsp = dict_ref(xdata);
+
+    if (local->loc.inode)
+        inode = local->loc.inode;
+    else
+        inode = local->fd->inode;
+
+    shard_inode_ctx_invalidate(inode, this, &local->prebuf);
+
+unwind:
+    local->handler(frame, this);
+    return 0;
+}
+
+int
+shard_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    int ret = -1;
+    uint64_t block_size = 0;
+    shard_local_t *local = NULL;
+
+    if ((IA_ISDIR(loc->inode->ia_type)) || (IA_ISLNK(loc->inode->ia_type))) {
+        STACK_WIND(frame, default_stat_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->stat, loc, xdata);
+        return 0;
+    }
+
+    ret = shard_inode_ctx_get_block_size(loc->inode, this, &block_size);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_CTX_GET_FAILED,
+               "Failed to get block "
+               "size from inode ctx of %s",
+               uuid_utoa(loc->inode->gfid));
+        goto err;
+    }
+
+    if (!block_size || frame->root->pid == GF_CLIENT_PID_GSYNCD) {
+        STACK_WIND(frame, default_stat_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->stat, loc, xdata);
+        return 0;
+    }
+
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto err;
+
+    frame->local = local;
+
+    local->handler = shard_post_stat_handler;
+    loc_copy(&local->loc, loc);
+    local->xattr_req = (xdata) ? dict_ref(xdata) : dict_new();
+    if (!local->xattr_req)
+        goto err;
+
+    SHARD_MD_READ_FOP_INIT_REQ_DICT(this, local->xattr_req, local->loc.gfid,
+                                    local, err);
+
+    STACK_WIND(frame, shard_common_stat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->stat, loc, local->xattr_req);
+    return 0;
+err:
+    shard_common_failure_unwind(GF_FOP_STAT, frame, -1, ENOMEM);
+    return 0;
+}
+
+int
+shard_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    int ret = -1;
+    uint64_t block_size = 0;
+    shard_local_t *local = NULL;
+
+    if ((IA_ISDIR(fd->inode->ia_type)) || (IA_ISLNK(fd->inode->ia_type))) {
+        STACK_WIND(frame, default_fstat_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fstat, fd, xdata);
+        return 0;
+    }
+
+    ret = shard_inode_ctx_get_block_size(fd->inode, this, &block_size);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_CTX_GET_FAILED,
+               "Failed to get block "
+               "size from inode ctx of %s",
+               uuid_utoa(fd->inode->gfid));
+        goto err;
+    }
+
+    if (!block_size || frame->root->pid == GF_CLIENT_PID_GSYNCD) {
+        STACK_WIND(frame, default_fstat_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fstat, fd, xdata);
+        return 0;
+    }
+
+    if (!this->itable)
+        this->itable = fd->inode->table;
+
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto err;
+
+    frame->local = local;
+
+    local->handler = shard_post_fstat_handler;
+    local->fd = fd_ref(fd);
+    local->xattr_req = (xdata) ? dict_ref(xdata) : dict_new();
+    if (!local->xattr_req)
+        goto err;
+
+    SHARD_MD_READ_FOP_INIT_REQ_DICT(this, local->xattr_req, fd->inode->gfid,
+                                    local, err);
+
+    STACK_WIND(frame, shard_common_stat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fstat, fd, local->xattr_req);
+    return 0;
+err:
+    shard_common_failure_unwind(GF_FOP_FSTAT, frame, -1, ENOMEM);
+    return 0;
+}
+
+int
+shard_post_update_size_truncate_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->fop == GF_FOP_TRUNCATE)
+        SHARD_STACK_UNWIND(truncate, frame, local->op_ret, local->op_errno,
+                           &local->prebuf, &local->postbuf, NULL);
+    else
+        SHARD_STACK_UNWIND(ftruncate, frame, local->op_ret, local->op_errno,
+                           &local->prebuf, &local->postbuf, NULL);
+    return 0;
+}
+
+int
+shard_truncate_last_shard_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                              int32_t op_ret, int32_t op_errno,
+                              struct iatt *prebuf, struct iatt *postbuf,
+                              dict_t *xdata)
+{
+    inode_t *inode = NULL;
+    int64_t delta_blocks = 0;
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    SHARD_UNSET_ROOT_FS_ID(frame, local);
+
+    inode = (local->fop == GF_FOP_TRUNCATE) ? local->loc.inode
+                                            : local->fd->inode;
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno,
+               SHARD_MSG_TRUNCATE_LAST_SHARD_FAILED,
+               "truncate on last"
+               " shard failed : %s",
+               uuid_utoa(inode->gfid));
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    local->postbuf.ia_size = local->offset;
+    /* Let the delta be negative. We want xattrop to do subtraction */
+    local->delta_size = local->postbuf.ia_size - local->prebuf.ia_size;
+    delta_blocks = GF_ATOMIC_ADD(local->delta_blocks,
+                                 postbuf->ia_blocks - prebuf->ia_blocks);
+    GF_ASSERT(delta_blocks <= 0);
+    local->postbuf.ia_blocks += delta_blocks;
+    local->hole_size = 0;
+
+    shard_inode_ctx_set(inode, this, &local->postbuf, 0, SHARD_MASK_TIMES);
+    shard_update_file_size(frame, this, NULL, &local->loc,
+                           shard_post_update_size_truncate_handler);
+    return 0;
+err:
+    shard_common_failure_unwind(local->fop, frame, local->op_ret,
+                                local->op_errno);
+    return 0;
+}
+
+int
+shard_truncate_last_shard(call_frame_t *frame, xlator_t *this, inode_t *inode)
+{
+    size_t last_shard_size_after = 0;
+    loc_t loc = {
+        0,
+    };
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    /* A NULL inode could be due to the fact that the last shard which
+     * needs to be truncated does not exist due to it lying in a hole
+     * region. So the only thing left to do in that case would be an
+     * update to file size xattr.
+     */
+    if (!inode) {
+        gf_msg_debug(this->name, 0,
+                     "Last shard to be truncated absent in backend: %" PRIu64
+                     " of gfid %s. Directly proceeding to update file size",
+                     local->first_block, uuid_utoa(local->loc.inode->gfid));
+        shard_update_file_size(frame, this, NULL, &local->loc,
+                               shard_post_update_size_truncate_handler);
+        return 0;
+    }
+
+    SHARD_SET_ROOT_FS_ID(frame, local);
+
+    loc.inode = inode_ref(inode);
+    gf_uuid_copy(loc.gfid, inode->gfid);
+
+    last_shard_size_after = (local->offset % local->block_size);
+
+    STACK_WIND(frame, shard_truncate_last_shard_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->truncate, &loc, last_shard_size_after,
+               NULL);
+    loc_wipe(&loc);
+    return 0;
+}
+
+void
+shard_unlink_block_inode(shard_local_t *local, int shard_block_num);
+
+int
+shard_truncate_htol_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno,
+                        struct iatt *preparent, struct iatt *postparent,
+                        dict_t *xdata)
+{
+    int ret = 0;
+    int call_count = 0;
+    int shard_block_num = (long)cookie;
+    uint64_t block_count = 0;
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (op_ret < 0) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        goto done;
+    }
+    ret = dict_get_uint64(xdata, GF_GET_FILE_BLOCK_COUNT, &block_count);
+    if (!ret) {
+        GF_ATOMIC_SUB(local->delta_blocks, block_count);
+    } else {
+        /* dict_get failed possibly due to a heterogeneous cluster? */
+        gf_msg(this->name, GF_LOG_WARNING, 0, SHARD_MSG_DICT_OP_FAILED,
+               "Failed to get key %s from dict during truncate of gfid %s",
+               GF_GET_FILE_BLOCK_COUNT,
+               uuid_utoa(local->resolver_base_inode->gfid));
+    }
+
+    shard_unlink_block_inode(local, shard_block_num);
+done:
+    call_count = shard_call_count_return(frame);
+    if (call_count == 0) {
+        SHARD_UNSET_ROOT_FS_ID(frame, local);
+        shard_truncate_last_shard(frame, this, local->inode_list[0]);
+    }
+    return 0;
+}
+
+int
+shard_truncate_htol(call_frame_t *frame, xlator_t *this, inode_t *inode)
+{
+    int i = 1;
+    int ret = -1;
+    int call_count = 0;
+    uint32_t cur_block = 0;
+    uint32_t last_block = 0;
+    char path[PATH_MAX] = {
+        0,
+    };
+    char *bname = NULL;
+    loc_t loc = {
+        0,
+    };
+    gf_boolean_t wind_failed = _gf_false;
+    shard_local_t *local = NULL;
+    shard_priv_t *priv = NULL;
+    dict_t *xdata_req = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    cur_block = local->first_block + 1;
+    last_block = local->last_block;
+
+    /* Determine call count */
+    for (i = 1; i < local->num_blocks; i++) {
+        if (!local->inode_list[i])
+            continue;
+        call_count++;
+    }
+
+    if (!call_count) {
+        /* Call count = 0 implies that all of the shards that need to be
+         * unlinked do not exist. So shard xlator would now proceed to
+         * do the final truncate + size updates.
+         */
+        gf_msg_debug(this->name, 0,
+                     "Shards to be unlinked as part of "
+                     "truncate absent in backend: %s. Directly "
+                     "proceeding to update file size",
+                     uuid_utoa(inode->gfid));
+        local->postbuf.ia_size = local->offset;
+        local->postbuf.ia_blocks = local->prebuf.ia_blocks;
+        local->delta_size = local->postbuf.ia_size - local->prebuf.ia_size;
+        GF_ATOMIC_INIT(local->delta_blocks, 0);
+        local->hole_size = 0;
+        shard_update_file_size(frame, this, local->fd, &local->loc,
+                               shard_post_update_size_truncate_handler);
+        return 0;
+    }
+
+    local->call_count = call_count;
+    i = 1;
+    xdata_req = dict_new();
+    if (!xdata_req) {
+        shard_common_failure_unwind(local->fop, frame, -1, ENOMEM);
+        return 0;
+    }
+    ret = dict_set_uint64(xdata_req, GF_GET_FILE_BLOCK_COUNT, 8 * 8);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, SHARD_MSG_DICT_OP_FAILED,
+               "Failed to set key %s into dict during truncate of %s",
+               GF_GET_FILE_BLOCK_COUNT,
+               uuid_utoa(local->resolver_base_inode->gfid));
+        dict_unref(xdata_req);
+        shard_common_failure_unwind(local->fop, frame, -1, ENOMEM);
+        return 0;
+    }
+
+    SHARD_SET_ROOT_FS_ID(frame, local);
+    while (cur_block <= last_block) {
+        if (!local->inode_list[i]) {
+            cur_block++;
+            i++;
+            continue;
+        }
+        if (wind_failed) {
+            shard_truncate_htol_cbk(frame, (void *)(long)cur_block, this, -1,
+                                    ENOMEM, NULL, NULL, NULL);
+            goto next;
+        }
+
+        shard_make_block_abspath(cur_block, inode->gfid, path, sizeof(path));
+        bname = strrchr(path, '/') + 1;
+        loc.parent = inode_ref(priv->dot_shard_inode);
+        ret = inode_path(loc.parent, bname, (char **)&(loc.path));
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_PATH_FAILED,
+                   "Inode path failed"
+                   " on %s. Base file gfid = %s",
+                   bname, uuid_utoa(inode->gfid));
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            loc_wipe(&loc);
+            wind_failed = _gf_true;
+            shard_truncate_htol_cbk(frame, (void *)(long)cur_block, this, -1,
+                                    ENOMEM, NULL, NULL, NULL);
+            goto next;
+        }
+        loc.name = strrchr(loc.path, '/');
+        if (loc.name)
+            loc.name++;
+        loc.inode = inode_ref(local->inode_list[i]);
+
+        STACK_WIND_COOKIE(frame, shard_truncate_htol_cbk,
+                          (void *)(long)cur_block, FIRST_CHILD(this),
+                          FIRST_CHILD(this)->fops->unlink, &loc, 0, xdata_req);
+        loc_wipe(&loc);
+    next:
+        i++;
+        cur_block++;
+        if (!--call_count)
+            break;
+    }
+    dict_unref(xdata_req);
+    return 0;
+}
+
+int
+shard_truncate_do(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->num_blocks == 1) {
+        /* This means that there are no shards to be unlinked.
+         * The fop boils down to truncating the last shard, updating
+         * the size and unwinding.
+         */
+        shard_truncate_last_shard(frame, this, local->inode_list[0]);
+        return 0;
+    } else {
+        shard_truncate_htol(frame, this, local->loc.inode);
+    }
+    return 0;
+}
+
+int
+shard_post_lookup_shards_truncate_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->op_ret < 0) {
+        shard_common_failure_unwind(local->fop, frame, local->op_ret,
+                                    local->op_errno);
+        return 0;
+    }
+
+    shard_truncate_do(frame, this);
+    return 0;
+}
+
+void
+shard_link_block_inode(shard_local_t *local, int block_num, inode_t *inode,
+                       struct iatt *buf)
+{
+    int list_index = 0;
+    char block_bname[256] = {
+        0,
+    };
+    uuid_t gfid = {
+        0,
+    };
+    inode_t *linked_inode = NULL;
+    xlator_t *this = NULL;
+    inode_t *fsync_inode = NULL;
+    shard_priv_t *priv = NULL;
+    inode_t *base_inode = NULL;
+
+    this = THIS;
+    priv = this->private;
+    if (local->loc.inode) {
+        gf_uuid_copy(gfid, local->loc.inode->gfid);
+        base_inode = local->loc.inode;
+    } else if (local->resolver_base_inode) {
+        gf_uuid_copy(gfid, local->resolver_base_inode->gfid);
+        base_inode = local->resolver_base_inode;
+    } else {
+        gf_uuid_copy(gfid, local->base_gfid);
+    }
+
+    shard_make_block_bname(block_num, gfid, block_bname, sizeof(block_bname));
+
+    shard_inode_ctx_set(inode, this, buf, 0, SHARD_LOOKUP_MASK);
+    linked_inode = inode_link(inode, priv->dot_shard_inode, block_bname, buf);
+    inode_lookup(linked_inode);
+    list_index = block_num - local->first_block;
+    local->inode_list[list_index] = linked_inode;
+
+    LOCK(&priv->lock);
+    {
+        fsync_inode = __shard_update_shards_inode_list(
+            linked_inode, this, base_inode, block_num, gfid);
+    }
+    UNLOCK(&priv->lock);
+    if (fsync_inode)
+        shard_initiate_evicted_inode_fsync(this, fsync_inode);
+}
+
+int
+shard_common_lookup_shards_cbk(call_frame_t *frame, void *cookie,
+                               xlator_t *this, int32_t op_ret, int32_t op_errno,
+                               inode_t *inode, struct iatt *buf, dict_t *xdata,
+                               struct iatt *postparent)
+{
+    int call_count = 0;
+    int shard_block_num = (long)cookie;
+    uuid_t gfid = {
+        0,
+    };
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+    if (local->resolver_base_inode)
+        gf_uuid_copy(gfid, local->resolver_base_inode->gfid);
+    else
+        gf_uuid_copy(gfid, local->base_gfid);
+
+    if (op_ret < 0) {
+        /* Ignore absence of shards in the backend in truncate fop. */
+        switch (local->fop) {
+            case GF_FOP_TRUNCATE:
+            case GF_FOP_FTRUNCATE:
+            case GF_FOP_RENAME:
+            case GF_FOP_UNLINK:
+                if (op_errno == ENOENT)
+                    goto done;
+                break;
+            case GF_FOP_WRITE:
+            case GF_FOP_READ:
+            case GF_FOP_ZEROFILL:
+            case GF_FOP_DISCARD:
+            case GF_FOP_FALLOCATE:
+                if ((!local->first_lookup_done) && (op_errno == ENOENT)) {
+                    LOCK(&frame->lock);
+                    {
+                        local->create_count++;
+                    }
+                    UNLOCK(&frame->lock);
+                    goto done;
+                }
+                break;
+            default:
+                break;
+        }
+
+        /* else */
+        gf_msg(this->name, GF_LOG_ERROR, op_errno,
+               SHARD_MSG_LOOKUP_SHARD_FAILED,
+               "Lookup on shard %d "
+               "failed. Base file gfid = %s",
+               shard_block_num, uuid_utoa(gfid));
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        goto done;
+    }
+
+    shard_link_block_inode(local, shard_block_num, inode, buf);
+
+done:
+    if (local->lookup_shards_barriered) {
+        syncbarrier_wake(&local->barrier);
+        return 0;
+    } else {
+        call_count = shard_call_count_return(frame);
+        if (call_count == 0) {
+            if (!local->first_lookup_done)
+                local->first_lookup_done = _gf_true;
+            local->pls_fop_handler(frame, this);
+        }
+    }
+    return 0;
+}
+
+dict_t *
+shard_create_gfid_dict(dict_t *dict)
+{
+    int ret = 0;
+    dict_t *new = NULL;
+    unsigned char *gfid = NULL;
+
+    new = dict_copy_with_ref(dict, NULL);
+    if (!new)
+        return NULL;
+
+    gfid = GF_MALLOC(sizeof(uuid_t), gf_common_mt_char);
+    if (!gfid) {
+        ret = -1;
+        goto out;
+    }
+
+    gf_uuid_generate(gfid);
+
+    ret = dict_set_gfuuid(new, "gfid-req", gfid, false);
+
+out:
+    if (ret) {
+        dict_unref(new);
+        new = NULL;
+        GF_FREE(gfid);
+    }
+
+    return new;
+}
+
+int
+shard_common_lookup_shards(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                           shard_post_lookup_shards_fop_handler_t handler)
+{
+    int i = 0;
+    int ret = 0;
+    int count = 0;
+    int call_count = 0;
+    int32_t shard_idx_iter = 0;
+    int lookup_count = 0;
+    char path[PATH_MAX] = {
+        0,
+    };
+    char *bname = NULL;
+    uuid_t gfid = {
+        0,
+    };
+    loc_t loc = {
+        0,
+    };
+    shard_local_t *local = NULL;
+    shard_priv_t *priv = NULL;
+    gf_boolean_t wind_failed = _gf_false;
+    dict_t *xattr_req = NULL;
+
+    priv = this->private;
+    local = frame->local;
+    count = call_count = local->call_count;
+    shard_idx_iter = local->first_block;
+    lookup_count = local->last_block - local->create_count;
+    local->pls_fop_handler = handler;
+    if (local->lookup_shards_barriered)
+        local->barrier.waitfor = local->call_count;
+
+    if (inode)
+        gf_uuid_copy(gfid, inode->gfid);
+    else
+        gf_uuid_copy(gfid, local->base_gfid);
+
+    while (shard_idx_iter <= lookup_count) {
+        if (local->inode_list[i]) {
+            i++;
+            shard_idx_iter++;
+            continue;
+        }
+
+        if (wind_failed) {
+            shard_common_lookup_shards_cbk(frame, (void *)(long)shard_idx_iter,
+                                           this, -1, ENOMEM, NULL, NULL, NULL,
+                                           NULL);
+            goto next;
+        }
+
+        shard_make_block_abspath(shard_idx_iter, gfid, path, sizeof(path));
+
+        bname = strrchr(path, '/') + 1;
+        loc.inode = inode_new(this->itable);
+        loc.parent = inode_ref(priv->dot_shard_inode);
+        gf_uuid_copy(loc.pargfid, priv->dot_shard_gfid);
+        ret = inode_path(loc.parent, bname, (char **)&(loc.path));
+        if (ret < 0 || !(loc.inode)) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_PATH_FAILED,
+                   "Inode path failed"
+                   " on %s, base file gfid = %s",
+                   bname, uuid_utoa(gfid));
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            loc_wipe(&loc);
+            wind_failed = _gf_true;
+            shard_common_lookup_shards_cbk(frame, (void *)(long)shard_idx_iter,
+                                           this, -1, ENOMEM, NULL, NULL, NULL,
+                                           NULL);
+            goto next;
+        }
+
+        loc.name = strrchr(loc.path, '/');
+        if (loc.name)
+            loc.name++;
+
+        xattr_req = shard_create_gfid_dict(local->xattr_req);
+        if (!xattr_req) {
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            wind_failed = _gf_true;
+            loc_wipe(&loc);
+            shard_common_lookup_shards_cbk(frame, (void *)(long)shard_idx_iter,
+                                           this, -1, ENOMEM, NULL, NULL, NULL,
+                                           NULL);
+            goto next;
+        }
+
+        STACK_WIND_COOKIE(frame, shard_common_lookup_shards_cbk,
+                          (void *)(long)shard_idx_iter, FIRST_CHILD(this),
+                          FIRST_CHILD(this)->fops->lookup, &loc, xattr_req);
+        loc_wipe(&loc);
+        dict_unref(xattr_req);
+    next:
+        shard_idx_iter++;
+        i++;
+
+        if (!--call_count)
+            break;
+    }
+    if (local->lookup_shards_barriered) {
+        syncbarrier_wait(&local->barrier, count);
+        local->pls_fop_handler(frame, this);
+    }
+    return 0;
+}
+
+int
+shard_post_resolve_truncate_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->op_ret < 0) {
+        if (local->op_errno == ENOENT) {
+            /* If lookup on /.shard fails with ENOENT, it means that
+             * the file was 0-byte in size but truncated sometime in
+             * the past to a higher size which is reflected in the
+             * size xattr, and now being truncated to a lower size.
+             * In this case, the only thing that needs to be done is
+             * to update the size xattr of the file and unwind.
+             */
+            local->first_block = local->last_block = 0;
+            local->num_blocks = 1;
+            local->call_count = 0;
+            local->op_ret = 0;
+            local->postbuf.ia_size = local->offset;
+            shard_update_file_size(frame, this, local->fd, &local->loc,
+                                   shard_post_update_size_truncate_handler);
+            return 0;
+        } else {
+            shard_common_failure_unwind(local->fop, frame, local->op_ret,
+                                        local->op_errno);
+            return 0;
+        }
+    }
+
+    if (!local->call_count)
+        shard_truncate_do(frame, this);
+    else
+        shard_common_lookup_shards(frame, this, local->loc.inode,
+                                   shard_post_lookup_shards_truncate_handler);
+
+    return 0;
+}
+
+int
+shard_truncate_begin(call_frame_t *frame, xlator_t *this)
+{
+    int ret = 0;
+    shard_local_t *local = NULL;
+    shard_priv_t *priv = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    /* First participant block here is the lowest numbered block that would
+     * hold the last byte of the file post successful truncation.
+     * Last participant block is the block that contains the last byte in
+     * the current state of the file.
+     * If (first block == last_block):
+     *         then that means that the file only needs truncation of the
+     *         first (or last since both are same) block.
+     * Else
+     *         if (new_size % block_size == 0)
+     *                 then that means there is no truncate to be done with
+     *                 only shards from first_block + 1 through the last
+     *                 block needing to be unlinked.
+     *         else
+     *                 both truncate of the first block and unlink of the
+     *                 remaining shards until end of file is required.
+     */
+    local->first_block = (local->offset == 0)
+                             ? 0
+                             : get_lowest_block(local->offset - 1,
+                                                local->block_size);
+    local->last_block = get_highest_block(0, local->prebuf.ia_size,
+                                          local->block_size);
+
+    local->num_blocks = local->last_block - local->first_block + 1;
+    GF_ASSERT(local->num_blocks > 0);
+    local->resolver_base_inode = (local->fop == GF_FOP_TRUNCATE)
+                                     ? local->loc.inode
+                                     : local->fd->inode;
+
+    if ((local->first_block == 0) && (local->num_blocks == 1)) {
+        if (local->fop == GF_FOP_TRUNCATE)
+            STACK_WIND(frame, shard_truncate_last_shard_cbk, FIRST_CHILD(this),
+                       FIRST_CHILD(this)->fops->truncate, &local->loc,
+                       local->offset, local->xattr_req);
+        else
+            STACK_WIND(frame, shard_truncate_last_shard_cbk, FIRST_CHILD(this),
+                       FIRST_CHILD(this)->fops->ftruncate, local->fd,
+                       local->offset, local->xattr_req);
+        return 0;
+    }
+
+    local->inode_list = GF_CALLOC(local->num_blocks, sizeof(inode_t *),
+                                  gf_shard_mt_inode_list);
+    if (!local->inode_list)
+        goto err;
+
+    local->dot_shard_loc.inode = inode_find(this->itable, priv->dot_shard_gfid);
+    if (!local->dot_shard_loc.inode) {
+        ret = shard_init_internal_dir_loc(this, local,
+                                          SHARD_INTERNAL_DIR_DOT_SHARD);
+        if (ret)
+            goto err;
+        shard_lookup_internal_dir(frame, this,
+                                  shard_post_resolve_truncate_handler,
+                                  SHARD_INTERNAL_DIR_DOT_SHARD);
+    } else {
+        local->post_res_handler = shard_post_resolve_truncate_handler;
+        shard_refresh_internal_dir(frame, this, SHARD_INTERNAL_DIR_DOT_SHARD);
+    }
+    return 0;
+
+err:
+    shard_common_failure_unwind(local->fop, frame, -1, ENOMEM);
+    return 0;
+}
+
+int
+shard_post_lookup_truncate_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+    struct iatt tmp_stbuf = {
+        0,
+    };
+
+    local = frame->local;
+
+    if (local->op_ret < 0) {
+        shard_common_failure_unwind(local->fop, frame, local->op_ret,
+                                    local->op_errno);
+        return 0;
+    }
+
+    local->postbuf = tmp_stbuf = local->prebuf;
+
+    if (local->prebuf.ia_size == local->offset) {
+        /* If the file size is same as requested size, unwind the call
+         * immediately.
+         */
+        if (local->fop == GF_FOP_TRUNCATE)
+            SHARD_STACK_UNWIND(truncate, frame, 0, 0, &local->prebuf,
+                               &local->postbuf, NULL);
+        else
+            SHARD_STACK_UNWIND(ftruncate, frame, 0, 0, &local->prebuf,
+                               &local->postbuf, NULL);
+    } else if (local->offset > local->prebuf.ia_size) {
+        /* If the truncate is from a lower to a higher size, set the
+         * new size xattr and unwind.
+         */
+        local->hole_size = local->offset - local->prebuf.ia_size;
+        local->delta_size = 0;
+        GF_ATOMIC_INIT(local->delta_blocks, 0);
+        local->postbuf.ia_size = local->offset;
+        tmp_stbuf.ia_size = local->offset;
+        shard_inode_ctx_set(local->loc.inode, this, &tmp_stbuf, 0,
+                            SHARD_INODE_WRITE_MASK);
+        shard_update_file_size(frame, this, NULL, &local->loc,
+                               shard_post_update_size_truncate_handler);
+    } else {
+        /* ... else
+         * i.   unlink all shards that need to be unlinked.
+         * ii.  truncate the last of the shards.
+         * iii. update the new size using setxattr.
+         * and unwind the fop.
+         */
+        local->hole_size = 0;
+        local->delta_size = (local->offset - local->prebuf.ia_size);
+        GF_ATOMIC_INIT(local->delta_blocks, 0);
+        tmp_stbuf.ia_size = local->offset;
+        shard_inode_ctx_set(local->loc.inode, this, &tmp_stbuf, 0,
+                            SHARD_INODE_WRITE_MASK);
+        shard_truncate_begin(frame, this);
+    }
+    return 0;
+}
+
+/* TO-DO:
+ * Fix updates to size and block count with racing write(s) and truncate(s).
+ */
+
+int
+shard_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+               dict_t *xdata)
+{
+    int ret = -1;
+    uint64_t block_size = 0;
+    shard_local_t *local = NULL;
+
+    ret = shard_inode_ctx_get_block_size(loc->inode, this, &block_size);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_CTX_GET_FAILED,
+               "Failed to get block "
+               "size from inode ctx of %s",
+               uuid_utoa(loc->inode->gfid));
+        goto err;
+    }
+
+    if (!block_size || frame->root->pid == GF_CLIENT_PID_GSYNCD) {
+        STACK_WIND(frame, default_truncate_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+        return 0;
+    }
+
+    if (!this->itable)
+        this->itable = loc->inode->table;
+
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto err;
+
+    frame->local = local;
+
+    ret = syncbarrier_init(&local->barrier);
+    if (ret)
+        goto err;
+    loc_copy(&local->loc, loc);
+    local->offset = offset;
+    local->block_size = block_size;
+    local->fop = GF_FOP_TRUNCATE;
+    local->xattr_req = (xdata) ? dict_ref(xdata) : dict_new();
+    if (!local->xattr_req)
+        goto err;
+    local->resolver_base_inode = loc->inode;
+    GF_ATOMIC_INIT(local->delta_blocks, 0);
+
+    shard_refresh_base_file(frame, this, &local->loc, NULL,
+                            shard_post_lookup_truncate_handler);
+    return 0;
+
+err:
+    shard_common_failure_unwind(GF_FOP_TRUNCATE, frame, -1, ENOMEM);
+    return 0;
+}
+
+int
+shard_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                dict_t *xdata)
+{
+    int ret = -1;
+    uint64_t block_size = 0;
+    shard_local_t *local = NULL;
+
+    ret = shard_inode_ctx_get_block_size(fd->inode, this, &block_size);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_CTX_GET_FAILED,
+               "Failed to get block "
+               "size from inode ctx of %s",
+               uuid_utoa(fd->inode->gfid));
+        goto err;
+    }
+
+    if (!block_size || frame->root->pid == GF_CLIENT_PID_GSYNCD) {
+        STACK_WIND(frame, default_ftruncate_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+        return 0;
+    }
+
+    if (!this->itable)
+        this->itable = fd->inode->table;
+
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto err;
+
+    frame->local = local;
+    ret = syncbarrier_init(&local->barrier);
+    if (ret)
+        goto err;
+    local->fd = fd_ref(fd);
+    local->offset = offset;
+    local->block_size = block_size;
+    local->xattr_req = (xdata) ? dict_ref(xdata) : dict_new();
+    if (!local->xattr_req)
+        goto err;
+    local->fop = GF_FOP_FTRUNCATE;
+
+    local->loc.inode = inode_ref(fd->inode);
+    gf_uuid_copy(local->loc.gfid, fd->inode->gfid);
+    local->resolver_base_inode = fd->inode;
+    GF_ATOMIC_INIT(local->delta_blocks, 0);
+
+    shard_refresh_base_file(frame, this, NULL, fd,
+                            shard_post_lookup_truncate_handler);
+    return 0;
+err:
+    shard_common_failure_unwind(GF_FOP_FTRUNCATE, frame, -1, ENOMEM);
+    return 0;
+}
+
+int
+shard_mknod_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, inode_t *inode,
+                struct iatt *buf, struct iatt *preparent,
+                struct iatt *postparent, dict_t *xdata)
+{
+    int ret = -1;
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (op_ret == -1)
+        goto unwind;
+
+    ret = shard_inode_ctx_set(inode, this, buf, local->block_size,
+                              SHARD_ALL_MASK);
+    if (ret)
+        gf_msg(this->name, GF_LOG_WARNING, 0, SHARD_MSG_INODE_CTX_SET_FAILED,
+               "Failed to set inode "
+               "ctx for %s",
+               uuid_utoa(inode->gfid));
+
+unwind:
+    SHARD_STACK_UNWIND(mknod, frame, op_ret, op_errno, inode, buf, preparent,
+                       postparent, xdata);
+
+    return 0;
+}
+
+int
+shard_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+            dev_t rdev, mode_t umask, dict_t *xdata)
+{
+    shard_priv_t *priv = NULL;
+    shard_local_t *local = NULL;
+
+    priv = this->private;
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto err;
+
+    frame->local = local;
+    local->block_size = priv->block_size;
+    if (!__is_gsyncd_on_shard_dir(frame, loc)) {
+        SHARD_INODE_CREATE_INIT(this, local->block_size, xdata, loc, 0, 0, err);
+    }
+
+    STACK_WIND(frame, shard_mknod_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mknod, loc, mode, rdev, umask, xdata);
+    return 0;
+err:
+    shard_common_failure_unwind(GF_FOP_MKNOD, frame, -1, ENOMEM);
+    return 0;
+}
+
+int32_t
+shard_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, inode_t *inode,
+               struct iatt *buf, struct iatt *preparent,
+               struct iatt *postparent, dict_t *xdata)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+    if (op_ret < 0)
+        goto err;
+
+    shard_inode_ctx_set(inode, this, buf, 0,
+                        SHARD_MASK_NLINK | SHARD_MASK_TIMES);
+    buf->ia_size = local->prebuf.ia_size;
+    buf->ia_blocks = local->prebuf.ia_blocks;
+
+    SHARD_STACK_UNWIND(link, frame, op_ret, op_errno, inode, buf, preparent,
+                       postparent, xdata);
+    return 0;
+err:
+    shard_common_failure_unwind(GF_FOP_LINK, frame, op_ret, op_errno);
+    return 0;
+}
+
+int
+shard_post_lookup_link_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->op_ret < 0) {
+        SHARD_STACK_UNWIND(link, frame, local->op_ret, local->op_errno, NULL,
+                           NULL, NULL, NULL, NULL);
+        return 0;
+    }
+
+    STACK_WIND(frame, shard_link_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->link, &local->loc, &local->loc2,
+               local->xattr_req);
+    return 0;
+}
+
+int32_t
+shard_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+           dict_t *xdata)
+{
+    int ret = -1;
+    uint64_t block_size = 0;
+    shard_local_t *local = NULL;
+
+    ret = shard_inode_ctx_get_block_size(oldloc->inode, this, &block_size);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_CTX_GET_FAILED,
+               "Failed to get block "
+               "size from inode ctx of %s",
+               uuid_utoa(oldloc->inode->gfid));
+        goto err;
+    }
+
+    if (!block_size) {
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->link,
+                        oldloc, newloc, xdata);
+        return 0;
+    }
+
+    if (!this->itable)
+        this->itable = oldloc->inode->table;
+
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto err;
+
+    frame->local = local;
+
+    loc_copy(&local->loc, oldloc);
+    loc_copy(&local->loc2, newloc);
+    local->xattr_req = (xdata) ? dict_ref(xdata) : dict_new();
+    if (!local->xattr_req)
+        goto err;
+
+    shard_refresh_base_file(frame, this, &local->loc, NULL,
+                            shard_post_lookup_link_handler);
+    return 0;
+err:
+    shard_common_failure_unwind(GF_FOP_LINK, frame, -1, ENOMEM);
+    return 0;
+}
+
+int
+shard_unlink_shards_do(call_frame_t *frame, xlator_t *this, inode_t *inode);
+
+int
+shard_post_lookup_shards_unlink_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+    uuid_t gfid = {
+        0,
+    };
+
+    local = frame->local;
+
+    if (local->resolver_base_inode)
+        gf_uuid_copy(gfid, local->resolver_base_inode->gfid);
+    else
+        gf_uuid_copy(gfid, local->base_gfid);
+
+    if ((local->op_ret < 0) && (local->op_errno != ENOENT)) {
+        gf_msg(this->name, GF_LOG_ERROR, local->op_errno, SHARD_MSG_FOP_FAILED,
+               "failed to delete shards of %s", uuid_utoa(gfid));
+        return 0;
+    }
+    local->op_ret = 0;
+    local->op_errno = 0;
+
+    shard_unlink_shards_do(frame, this, local->resolver_base_inode);
+    return 0;
+}
+
+int
+shard_post_resolve_unlink_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+    local->lookup_shards_barriered = _gf_true;
+
+    if (!local->call_count)
+        shard_unlink_shards_do(frame, this, local->resolver_base_inode);
+    else
+        shard_common_lookup_shards(frame, this, local->resolver_base_inode,
+                                   shard_post_lookup_shards_unlink_handler);
+    return 0;
+}
+
+void
+shard_unlink_block_inode(shard_local_t *local, int shard_block_num)
+{
+    char block_bname[256] = {
+        0,
+    };
+    uuid_t gfid = {
+        0,
+    };
+    inode_t *inode = NULL;
+    inode_t *base_inode = NULL;
+    xlator_t *this = NULL;
+    shard_priv_t *priv = NULL;
+    shard_inode_ctx_t *ctx = NULL;
+    shard_inode_ctx_t *base_ictx = NULL;
+    int unref_base_inode = 0;
+    int unref_shard_inode = 0;
+
+    this = THIS;
+    priv = this->private;
+
+    inode = local->inode_list[shard_block_num - local->first_block];
+    shard_inode_ctx_get(inode, this, &ctx);
+    base_inode = ctx->base_inode;
+    if (base_inode)
+        gf_uuid_copy(gfid, base_inode->gfid);
+    else
+        gf_uuid_copy(gfid, ctx->base_gfid);
+    shard_make_block_bname(shard_block_num, gfid, block_bname,
+                           sizeof(block_bname));
+
+    LOCK(&priv->lock);
+    if (base_inode)
+        LOCK(&base_inode->lock);
+    LOCK(&inode->lock);
+    {
+        __shard_inode_ctx_get(inode, this, &ctx);
+        if (!list_empty(&ctx->ilist)) {
+            list_del_init(&ctx->ilist);
+            priv->inode_count--;
+            unref_base_inode++;
+            unref_shard_inode++;
+            GF_ASSERT(priv->inode_count >= 0);
+        }
+        if (ctx->fsync_needed) {
+            unref_base_inode++;
+            unref_shard_inode++;
+            list_del_init(&ctx->to_fsync_list);
+            if (base_inode) {
+                __shard_inode_ctx_get(base_inode, this, &base_ictx);
+                base_ictx->fsync_count--;
+            }
+        }
+    }
+    UNLOCK(&inode->lock);
+    if (base_inode)
+        UNLOCK(&base_inode->lock);
+
+    inode_unlink(inode, priv->dot_shard_inode, block_bname);
+    inode_ref_reduce_by_n(inode, unref_shard_inode);
+    inode_forget(inode, 0);
+
+    if (base_inode && unref_base_inode)
+        inode_ref_reduce_by_n(base_inode, unref_base_inode);
+    UNLOCK(&priv->lock);
+}
+
+int
+shard_rename_cbk(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    SHARD_STACK_UNWIND(rename, frame, local->op_ret, local->op_errno,
+                       &local->prebuf, &local->preoldparent,
+                       &local->postoldparent, &local->prenewparent,
+                       &local->postnewparent, local->xattr_rsp);
+    return 0;
+}
+
+int32_t
+shard_unlink_cbk(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = frame->local;
+
+    SHARD_STACK_UNWIND(unlink, frame, local->op_ret, local->op_errno,
+                       &local->preoldparent, &local->postoldparent,
+                       local->xattr_rsp);
+    return 0;
+}
+
+int
+shard_unlink_shards_do_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno,
+                           struct iatt *preparent, struct iatt *postparent,
+                           dict_t *xdata)
+{
+    int shard_block_num = (long)cookie;
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (op_ret < 0) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        goto done;
+    }
+
+    shard_unlink_block_inode(local, shard_block_num);
+done:
+    syncbarrier_wake(&local->barrier);
+    return 0;
+}
+
+int
+shard_unlink_shards_do(call_frame_t *frame, xlator_t *this, inode_t *inode)
+{
+    int i = 0;
+    int ret = -1;
+    int count = 0;
+    uint32_t cur_block = 0;
+    uint32_t cur_block_idx = 0; /*this is idx into inode_list[] array */
+    char *bname = NULL;
+    char path[PATH_MAX] = {
+        0,
+    };
+    uuid_t gfid = {
+        0,
+    };
+    loc_t loc = {
+        0,
+    };
+    gf_boolean_t wind_failed = _gf_false;
+    shard_local_t *local = NULL;
+    shard_priv_t *priv = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    if (inode)
+        gf_uuid_copy(gfid, inode->gfid);
+    else
+        gf_uuid_copy(gfid, local->base_gfid);
+
+    for (i = 0; i < local->num_blocks; i++) {
+        if (!local->inode_list[i])
+            continue;
+        count++;
+    }
+
+    if (!count) {
+        /* callcount = 0 implies that all of the shards that need to be
+         * unlinked are non-existent (in other words the file is full of
+         * holes).
+         */
+        gf_msg_debug(this->name, 0,
+                     "All shards that need to be "
+                     "unlinked are non-existent: %s",
+                     uuid_utoa(gfid));
+        return 0;
+    }
+
+    SHARD_SET_ROOT_FS_ID(frame, local);
+    local->barrier.waitfor = count;
+    cur_block = cur_block_idx + local->first_block;
+
+    while (cur_block_idx < local->num_blocks) {
+        if (!local->inode_list[cur_block_idx])
+            goto next;
+
+        if (wind_failed) {
+            shard_unlink_shards_do_cbk(frame, (void *)(long)cur_block, this, -1,
+                                       ENOMEM, NULL, NULL, NULL);
+            goto next;
+        }
+
+        shard_make_block_abspath(cur_block, gfid, path, sizeof(path));
+        bname = strrchr(path, '/') + 1;
+        loc.parent = inode_ref(priv->dot_shard_inode);
+        ret = inode_path(loc.parent, bname, (char **)&(loc.path));
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_PATH_FAILED,
+                   "Inode path failed"
+                   " on %s, base file gfid = %s",
+                   bname, uuid_utoa(gfid));
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            loc_wipe(&loc);
+            wind_failed = _gf_true;
+            shard_unlink_shards_do_cbk(frame, (void *)(long)cur_block, this, -1,
+                                       ENOMEM, NULL, NULL, NULL);
+            goto next;
+        }
+
+        loc.name = strrchr(loc.path, '/');
+        if (loc.name)
+            loc.name++;
+        loc.inode = inode_ref(local->inode_list[cur_block_idx]);
+
+        STACK_WIND_COOKIE(frame, shard_unlink_shards_do_cbk,
+                          (void *)(long)cur_block, FIRST_CHILD(this),
+                          FIRST_CHILD(this)->fops->unlink, &loc, local->xflag,
+                          local->xattr_req);
+        loc_wipe(&loc);
+    next:
+        cur_block++;
+        cur_block_idx++;
+    }
+    syncbarrier_wait(&local->barrier, count);
+    SHARD_UNSET_ROOT_FS_ID(frame, local);
+    return 0;
+}
+
+int
+shard_regulated_shards_deletion(call_frame_t *cleanup_frame, xlator_t *this,
+                                int now, int first_block, gf_dirent_t *entry)
+{
+    int i = 0;
+    int ret = 0;
+    shard_local_t *local = NULL;
+    uuid_t gfid = {
+        0,
+    };
+
+    local = cleanup_frame->local;
+
+    local->inode_list = GF_CALLOC(now, sizeof(inode_t *),
+                                  gf_shard_mt_inode_list);
+    if (!local->inode_list)
+        return -ENOMEM;
+
+    local->first_block = first_block;
+    local->last_block = first_block + now - 1;
+    local->num_blocks = now;
+    gf_uuid_parse(entry->d_name, gfid);
+    gf_uuid_copy(local->base_gfid, gfid);
+    local->resolver_base_inode = inode_find(this->itable, gfid);
+    local->call_count = 0;
+    ret = syncbarrier_init(&local->barrier);
+    if (ret) {
+        GF_FREE(local->inode_list);
+        local->inode_list = NULL;
+        inode_unref(local->resolver_base_inode);
+        local->resolver_base_inode = NULL;
+        return -errno;
+    }
+    shard_common_resolve_shards(cleanup_frame, this,
+                                shard_post_resolve_unlink_handler);
+
+    for (i = 0; i < local->num_blocks; i++) {
+        if (local->inode_list[i])
+            inode_unref(local->inode_list[i]);
+    }
+    GF_FREE(local->inode_list);
+    local->inode_list = NULL;
+    if (local->op_ret)
+        ret = -local->op_errno;
+    syncbarrier_destroy(&local->barrier);
+    inode_unref(local->resolver_base_inode);
+    local->resolver_base_inode = NULL;
+    STACK_RESET(cleanup_frame->root);
+    return ret;
+}
+
+int
+__shard_delete_shards_of_entry(call_frame_t *cleanup_frame, xlator_t *this,
+                               gf_dirent_t *entry, inode_t *inode)
+{
+    int ret = 0;
+    int shard_count = 0;
+    int first_block = 0;
+    int now = 0;
+    uint64_t size = 0;
+    uint64_t block_size = 0;
+    uint64_t size_array[4] = {
+        0,
+    };
+    void *bsize = NULL;
+    void *size_attr = NULL;
+    dict_t *xattr_rsp = NULL;
+    loc_t loc = {
+        0,
+    };
+    shard_local_t *local = NULL;
+    shard_priv_t *priv = NULL;
+
+    priv = this->private;
+    local = cleanup_frame->local;
+    ret = dict_reset(local->xattr_req);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, SHARD_MSG_DICT_OP_FAILED,
+               "Failed to reset dict");
+        ret = -ENOMEM;
+        goto err;
+    }
+
+    ret = dict_set_uint64(local->xattr_req, GF_XATTR_SHARD_BLOCK_SIZE, 0);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, SHARD_MSG_DICT_OP_FAILED,
+               "Failed to set dict value: key:%s", GF_XATTR_SHARD_BLOCK_SIZE);
+        ret = -ENOMEM;
+        goto err;
+    }
+
+    ret = dict_set_uint64(local->xattr_req, GF_XATTR_SHARD_FILE_SIZE, 8 * 4);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, SHARD_MSG_DICT_OP_FAILED,
+               "Failed to set dict value: key:%s", GF_XATTR_SHARD_FILE_SIZE);
+        ret = -ENOMEM;
+        goto err;
+    }
+
+    loc.inode = inode_ref(inode);
+    loc.parent = inode_ref(priv->dot_shard_rm_inode);
+    ret = inode_path(loc.parent, entry->d_name, (char **)&(loc.path));
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_PATH_FAILED,
+               "Inode path  failed on %s", entry->d_name);
+        ret = -ENOMEM;
+        goto err;
+    }
+
+    loc.name = strrchr(loc.path, '/');
+    if (loc.name)
+        loc.name++;
+    ret = syncop_lookup(FIRST_CHILD(this), &loc, NULL, NULL, local->xattr_req,
+                        &xattr_rsp);
+    if (ret)
+        goto err;
+
+    ret = dict_get_ptr(xattr_rsp, GF_XATTR_SHARD_BLOCK_SIZE, &bsize);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_DICT_OP_FAILED,
+               "Failed to get dict value: key:%s", GF_XATTR_SHARD_BLOCK_SIZE);
+        goto err;
+    }
+    block_size = ntoh64(*((uint64_t *)bsize));
+
+    ret = dict_get_ptr(xattr_rsp, GF_XATTR_SHARD_FILE_SIZE, &size_attr);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_DICT_OP_FAILED,
+               "Failed to get dict value: key:%s", GF_XATTR_SHARD_FILE_SIZE);
+        goto err;
+    }
+
+    memcpy(size_array, size_attr, sizeof(size_array));
+    size = ntoh64(size_array[0]);
+
+    shard_count = (size / block_size) - 1;
+    if (shard_count < 0) {
+        gf_msg_debug(this->name, 0,
+                     "Size of %s hasn't grown beyond "
+                     "its shard-block-size. Nothing to delete. "
+                     "Returning",
+                     entry->d_name);
+        /* File size < shard-block-size, so nothing to delete */
+        ret = 0;
+        goto delete_marker;
+    }
+    if ((size % block_size) > 0)
+        shard_count++;
+
+    if (shard_count == 0) {
+        gf_msg_debug(this->name, 0,
+                     "Size of %s is exactly equal to "
+                     "its shard-block-size. Nothing to delete. "
+                     "Returning",
+                     entry->d_name);
+        ret = 0;
+        goto delete_marker;
+    }
+    gf_msg_debug(this->name, 0,
+                 "base file = %s, "
+                 "shard-block-size=%" PRIu64 ", file-size=%" PRIu64
+                 ", "
+                 "shard_count=%d",
+                 entry->d_name, block_size, size, shard_count);
+
+    /* Perform a gfid-based lookup to see if gfid corresponding to marker
+     * file's base name exists.
+     */
+    loc_wipe(&loc);
+    loc.inode = inode_new(this->itable);
+    if (!loc.inode) {
+        ret = -ENOMEM;
+        goto err;
+    }
+    gf_uuid_parse(entry->d_name, loc.gfid);
+    ret = syncop_lookup(FIRST_CHILD(this), &loc, NULL, NULL, NULL, NULL);
+    if (!ret) {
+        gf_msg_debug(this->name, 0,
+                     "Base shard corresponding to gfid "
+                     "%s is present. Skipping shard deletion. "
+                     "Returning",
+                     entry->d_name);
+        ret = 0;
+        goto delete_marker;
+    }
+
+    first_block = 1;
+
+    while (shard_count) {
+        if (shard_count < local->deletion_rate) {
+            now = shard_count;
+            shard_count = 0;
+        } else {
+            now = local->deletion_rate;
+            shard_count -= local->deletion_rate;
+        }
+
+        gf_msg_debug(this->name, 0,
+                     "deleting %d shards starting from "
+                     "block %d of gfid %s",
+                     now, first_block, entry->d_name);
+        ret = shard_regulated_shards_deletion(cleanup_frame, this, now,
+                                              first_block, entry);
+        if (ret)
+            goto err;
+        first_block += now;
+    }
+
+delete_marker:
+    loc_wipe(&loc);
+    loc.inode = inode_ref(inode);
+    loc.parent = inode_ref(priv->dot_shard_rm_inode);
+    ret = inode_path(loc.parent, entry->d_name, (char **)&(loc.path));
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_PATH_FAILED,
+               "Inode path  failed on %s", entry->d_name);
+        ret = -ENOMEM;
+        goto err;
+    }
+    loc.name = strrchr(loc.path, '/');
+    if (loc.name)
+        loc.name++;
+    ret = syncop_unlink(FIRST_CHILD(this), &loc, NULL, NULL);
+    if (ret)
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_SHARDS_DELETION_FAILED,
+               "Failed to delete %s "
+               "from /%s",
+               entry->d_name, GF_SHARD_REMOVE_ME_DIR);
+err:
+    if (xattr_rsp)
+        dict_unref(xattr_rsp);
+    loc_wipe(&loc);
+    return ret;
+}
+
+int
+shard_delete_shards_of_entry(call_frame_t *cleanup_frame, xlator_t *this,
+                             gf_dirent_t *entry, inode_t *inode)
+{
+    int ret = -1;
+    loc_t loc = {
+        0,
+    };
+    shard_priv_t *priv = NULL;
+
+    priv = this->private;
+    loc.inode = inode_ref(priv->dot_shard_rm_inode);
+
+    ret = syncop_entrylk(FIRST_CHILD(this), this->name, &loc, entry->d_name,
+                         ENTRYLK_LOCK_NB, ENTRYLK_WRLCK, NULL, NULL);
+    if (ret < 0) {
+        if (ret == -EAGAIN) {
+            ret = 0;
+        }
+        goto out;
+    }
+    {
+        ret = __shard_delete_shards_of_entry(cleanup_frame, this, entry, inode);
+    }
+    syncop_entrylk(FIRST_CHILD(this), this->name, &loc, entry->d_name,
+                   ENTRYLK_UNLOCK, ENTRYLK_WRLCK, NULL, NULL);
+out:
+    loc_wipe(&loc);
+    return ret;
+}
+
+int
+shard_delete_shards_cbk(int ret, call_frame_t *frame, void *data)
+{
+    SHARD_STACK_DESTROY(frame);
+    return 0;
+}
+
+int
+shard_resolve_internal_dir(xlator_t *this, shard_local_t *local,
+                           shard_internal_dir_type_t type)
+{
+    int ret = 0;
+    char *bname = NULL;
+    loc_t *loc = NULL;
+    shard_priv_t *priv = NULL;
+    uuid_t gfid = {
+        0,
+    };
+    struct iatt stbuf = {
+        0,
+    };
+
+    priv = this->private;
+
+    switch (type) {
+        case SHARD_INTERNAL_DIR_DOT_SHARD:
+            loc = &local->dot_shard_loc;
+            gf_uuid_copy(gfid, priv->dot_shard_gfid);
+            bname = GF_SHARD_DIR;
+            break;
+        case SHARD_INTERNAL_DIR_DOT_SHARD_REMOVE_ME:
+            loc = &local->dot_shard_rm_loc;
+            gf_uuid_copy(gfid, priv->dot_shard_rm_gfid);
+            bname = GF_SHARD_REMOVE_ME_DIR;
+            break;
+        default:
+            break;
+    }
+
+    loc->inode = inode_find(this->itable, gfid);
+    if (!loc->inode) {
+        ret = shard_init_internal_dir_loc(this, local, type);
+        if (ret)
+            goto err;
+        ret = dict_reset(local->xattr_req);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, SHARD_MSG_DICT_OP_FAILED,
+                   "Failed to reset "
+                   "dict");
+            ret = -ENOMEM;
+            goto err;
+        }
+        ret = dict_set_gfuuid(local->xattr_req, "gfid-req", gfid, true);
+        ret = syncop_lookup(FIRST_CHILD(this), loc, &stbuf, NULL,
+                            local->xattr_req, NULL);
+        if (ret < 0) {
+            if (ret != -ENOENT)
+                gf_msg(this->name, GF_LOG_ERROR, -ret,
+                       SHARD_MSG_SHARDS_DELETION_FAILED,
+                       "Lookup on %s failed, exiting", bname);
+            goto err;
+        } else {
+            shard_link_internal_dir_inode(local, loc->inode, &stbuf, type);
+        }
+    }
+    ret = 0;
+err:
+    return ret;
+}
+
+int
+shard_lookup_marker_entry(xlator_t *this, shard_local_t *local,
+                          gf_dirent_t *entry)
+{
+    int ret = 0;
+    loc_t loc = {
+        0,
+    };
+
+    loc.inode = inode_new(this->itable);
+    if (!loc.inode) {
+        ret = -ENOMEM;
+        goto err;
+    }
+    loc.parent = inode_ref(local->fd->inode);
+
+    ret = inode_path(loc.parent, entry->d_name, (char **)&(loc.path));
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_PATH_FAILED,
+               "Inode path failed on %s", entry->d_name);
+        ret = -ENOMEM;
+        goto err;
+    }
+
+    loc.name = strrchr(loc.path, '/');
+    if (loc.name)
+        loc.name++;
+
+    ret = syncop_lookup(FIRST_CHILD(this), &loc, NULL, NULL, NULL, NULL);
+    if (ret < 0) {
+        goto err;
+    }
+    entry->inode = inode_ref(loc.inode);
+    ret = 0;
+err:
+    loc_wipe(&loc);
+    return ret;
+}
+
+int
+shard_delete_shards(void *opaque)
+{
+    int ret = 0;
+    off_t offset = 0;
+    loc_t loc = {
+        0,
+    };
+    inode_t *link_inode = NULL;
+    xlator_t *this = NULL;
+    shard_priv_t *priv = NULL;
+    shard_local_t *local = NULL;
+    gf_dirent_t entries;
+    gf_dirent_t *entry = NULL;
+    call_frame_t *cleanup_frame = NULL;
+    gf_boolean_t done = _gf_false;
+
+    this = THIS;
+    priv = this->private;
+    INIT_LIST_HEAD(&entries.list);
+
+    cleanup_frame = opaque;
+
+    local = mem_get0(this->local_pool);
+    if (!local) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, SHARD_MSG_MEMALLOC_FAILED,
+               "Failed to create local to "
+               "delete shards");
+        ret = -ENOMEM;
+        goto err;
+    }
+    cleanup_frame->local = local;
+    local->fop = GF_FOP_UNLINK;
+
+    local->xattr_req = dict_new();
+    if (!local->xattr_req) {
+        ret = -ENOMEM;
+        goto err;
+    }
+    local->deletion_rate = priv->deletion_rate;
+
+    ret = shard_resolve_internal_dir(this, local, SHARD_INTERNAL_DIR_DOT_SHARD);
+    if (ret == -ENOENT) {
+        gf_msg_debug(this->name, 0,
+                     ".shard absent. Nothing to"
+                     " delete. Exiting");
+        ret = 0;
+        goto err;
+    } else if (ret < 0) {
+        goto err;
+    }
+
+    ret = shard_resolve_internal_dir(this, local,
+                                     SHARD_INTERNAL_DIR_DOT_SHARD_REMOVE_ME);
+    if (ret == -ENOENT) {
+        gf_msg_debug(this->name, 0,
+                     ".remove_me absent. "
+                     "Nothing to delete. Exiting");
+        ret = 0;
+        goto err;
+    } else if (ret < 0) {
+        goto err;
+    }
+
+    local->fd = fd_anonymous(local->dot_shard_rm_loc.inode);
+    if (!local->fd) {
+        ret = -ENOMEM;
+        goto err;
+    }
+
+    for (;;) {
+        offset = 0;
+        LOCK(&priv->lock);
+        {
+            if (priv->bg_del_state == SHARD_BG_DELETION_LAUNCHING) {
+                priv->bg_del_state = SHARD_BG_DELETION_IN_PROGRESS;
+            } else if (priv->bg_del_state == SHARD_BG_DELETION_IN_PROGRESS) {
+                priv->bg_del_state = SHARD_BG_DELETION_NONE;
+                done = _gf_true;
+            }
+        }
+        UNLOCK(&priv->lock);
+        if (done)
+            break;
+        while (
+            (ret = syncop_readdirp(FIRST_CHILD(this), local->fd, 131072, offset,
+                                   &entries, local->xattr_req, NULL))) {
+            if (ret > 0)
+                ret = 0;
+            list_for_each_entry(entry, &entries.list, list)
+            {
+                offset = entry->d_off;
+
+                if (!strcmp(entry->d_name, ".") || !strcmp(entry->d_name, ".."))
+                    continue;
+
+                if (!entry->inode) {
+                    ret = shard_lookup_marker_entry(this, local, entry);
+                    if (ret < 0)
+                        continue;
+                }
+                link_inode = inode_link(entry->inode, local->fd->inode,
+                                        entry->d_name, &entry->d_stat);
+
+                gf_msg_debug(this->name, 0,
+                             "Initiating deletion of "
+                             "shards of gfid %s",
+                             entry->d_name);
+                ret = shard_delete_shards_of_entry(cleanup_frame, this, entry,
+                                                   link_inode);
+                inode_unlink(link_inode, local->fd->inode, entry->d_name);
+                inode_unref(link_inode);
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_ERROR, -ret,
+                           SHARD_MSG_SHARDS_DELETION_FAILED,
+                           "Failed to clean up shards of gfid %s",
+                           entry->d_name);
+                    continue;
+                }
+                gf_msg(this->name, GF_LOG_INFO, 0,
+                       SHARD_MSG_SHARD_DELETION_COMPLETED,
+                       "Deleted "
+                       "shards of gfid=%s from backend",
+                       entry->d_name);
+            }
+            gf_dirent_free(&entries);
+            if (ret)
+                break;
+        }
+    }
+    ret = 0;
+    loc_wipe(&loc);
+    return ret;
+
+err:
+    LOCK(&priv->lock);
+    {
+        priv->bg_del_state = SHARD_BG_DELETION_NONE;
+    }
+    UNLOCK(&priv->lock);
+    loc_wipe(&loc);
+    return ret;
+}
+
+int
+shard_unlock_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    if (op_ret)
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, SHARD_MSG_FOP_FAILED,
+               "Unlock failed. Please check brick logs for "
+               "more details");
+    SHARD_STACK_DESTROY(frame);
+    return 0;
+}
+
+int
+shard_unlock_inodelk(call_frame_t *frame, xlator_t *this)
+{
+    loc_t *loc = NULL;
+    call_frame_t *lk_frame = NULL;
+    shard_local_t *local = NULL;
+    shard_local_t *lk_local = NULL;
+    shard_inodelk_t *lock = NULL;
+
+    local = frame->local;
+    lk_frame = local->inodelk_frame;
+    lk_local = lk_frame->local;
+    local->inodelk_frame = NULL;
+    loc = &local->int_inodelk.loc;
+    lock = &lk_local->int_inodelk;
+    lock->flock.l_type = F_UNLCK;
+
+    STACK_WIND(lk_frame, shard_unlock_inodelk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->inodelk, lock->domain, loc, F_SETLK,
+               &lock->flock, NULL);
+    local->int_inodelk.acquired_lock = _gf_false;
+    return 0;
+}
+
+int
+shard_rename_src_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                     struct iatt *preoldparent, struct iatt *postoldparent,
+                     struct iatt *prenewparent, struct iatt *postnewparent,
+                     dict_t *xdata);
+int
+shard_rename_src_base_file(call_frame_t *frame, xlator_t *this)
+{
+    int ret = 0;
+    loc_t *dst_loc = NULL;
+    loc_t tmp_loc = {
+        0,
+    };
+    shard_local_t *local = frame->local;
+
+    if (local->dst_block_size) {
+        tmp_loc.parent = inode_ref(local->loc2.parent);
+        ret = inode_path(tmp_loc.parent, local->loc2.name,
+                         (char **)&tmp_loc.path);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_PATH_FAILED,
+                   "Inode path failed"
+                   " on pargfid=%s bname=%s",
+                   uuid_utoa(tmp_loc.parent->gfid), local->loc2.name);
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            goto err;
+        }
+
+        tmp_loc.name = strrchr(tmp_loc.path, '/');
+        if (tmp_loc.name)
+            tmp_loc.name++;
+        dst_loc = &tmp_loc;
+    } else {
+        dst_loc = &local->loc2;
+    }
+
+    /* To-Do: Request open-fd count on dst base file */
+    STACK_WIND(frame, shard_rename_src_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rename, &local->loc, dst_loc,
+               local->xattr_req);
+    loc_wipe(&tmp_loc);
+    return 0;
+err:
+    loc_wipe(&tmp_loc);
+    shard_common_failure_unwind(local->fop, frame, local->op_ret,
+                                local->op_errno);
+    return 0;
+}
+
+int
+shard_unlink_base_file(call_frame_t *frame, xlator_t *this);
+
+int
+shard_set_size_attrs_on_marker_file_cbk(call_frame_t *frame, void *cookie,
+                                        xlator_t *this, int32_t op_ret,
+                                        int32_t op_errno, dict_t *dict,
+                                        dict_t *xdata)
+{
+    shard_priv_t *priv = NULL;
+    shard_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, SHARD_MSG_FOP_FAILED,
+               "Xattrop on marker file failed "
+               "while performing %s; entry gfid=%s",
+               gf_fop_string(local->fop), local->newloc.name);
+        goto err;
+    }
+
+    inode_unlink(local->newloc.inode, priv->dot_shard_rm_inode,
+                 local->newloc.name);
+
+    if (local->fop == GF_FOP_UNLINK)
+        shard_unlink_base_file(frame, this);
+    else if (local->fop == GF_FOP_RENAME)
+        shard_rename_src_base_file(frame, this);
+    return 0;
+err:
+    shard_common_failure_unwind(local->fop, frame, op_ret, op_errno);
+    return 0;
+}
+
+int
+shard_set_size_attrs_on_marker_file(call_frame_t *frame, xlator_t *this)
+{
+    int op_errno = ENOMEM;
+    uint64_t bs = 0;
+    dict_t *xdata = NULL;
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+    xdata = dict_new();
+    if (!xdata)
+        goto err;
+
+    if (local->fop == GF_FOP_UNLINK)
+        bs = local->block_size;
+    else if (local->fop == GF_FOP_RENAME)
+        bs = local->dst_block_size;
+    SHARD_INODE_CREATE_INIT(this, bs, xdata, &local->newloc,
+                            local->prebuf.ia_size, 0, err);
+    STACK_WIND(frame, shard_set_size_attrs_on_marker_file_cbk,
+               FIRST_CHILD(this), FIRST_CHILD(this)->fops->xattrop,
+               &local->newloc, GF_XATTROP_GET_AND_SET, xdata, NULL);
+    dict_unref(xdata);
+    return 0;
+err:
+    if (xdata)
+        dict_unref(xdata);
+    shard_common_failure_unwind(local->fop, frame, -1, op_errno);
+    return 0;
+}
+
+int
+shard_lookup_marker_file_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                             int32_t op_ret, int32_t op_errno, inode_t *inode,
+                             struct iatt *buf, dict_t *xdata,
+                             struct iatt *postparent)
+{
+    inode_t *linked_inode = NULL;
+    shard_priv_t *priv = NULL;
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    if (op_ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, SHARD_MSG_FOP_FAILED,
+               "Lookup on marker file failed "
+               "while performing %s; entry gfid=%s",
+               gf_fop_string(local->fop), local->newloc.name);
+        goto err;
+    }
+
+    linked_inode = inode_link(inode, priv->dot_shard_rm_inode,
+                              local->newloc.name, buf);
+    inode_unref(local->newloc.inode);
+    local->newloc.inode = linked_inode;
+    shard_set_size_attrs_on_marker_file(frame, this);
+    return 0;
+err:
+    shard_common_failure_unwind(local->fop, frame, op_ret, op_errno);
+    return 0;
+}
+
+int
+shard_lookup_marker_file(call_frame_t *frame, xlator_t *this)
+{
+    int op_errno = ENOMEM;
+    dict_t *xattr_req = NULL;
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    xattr_req = shard_create_gfid_dict(local->xattr_req);
+    if (!xattr_req)
+        goto err;
+
+    STACK_WIND(frame, shard_lookup_marker_file_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lookup, &local->newloc, xattr_req);
+    dict_unref(xattr_req);
+    return 0;
+err:
+    shard_common_failure_unwind(local->fop, frame, -1, op_errno);
+    return 0;
+}
+
+int
+shard_create_marker_file_under_remove_me_cbk(
+    call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+    int32_t op_errno, inode_t *inode, struct iatt *buf, struct iatt *preparent,
+    struct iatt *postparent, dict_t *xdata)
+{
+    inode_t *linked_inode = NULL;
+    shard_priv_t *priv = NULL;
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+    priv = this->private;
+
+    SHARD_UNSET_ROOT_FS_ID(frame, local);
+    if (op_ret < 0) {
+        if ((op_errno != EEXIST) && (op_errno != ENODATA)) {
+            local->op_ret = op_ret;
+            local->op_errno = op_errno;
+            gf_msg(this->name, GF_LOG_ERROR, op_errno, SHARD_MSG_FOP_FAILED,
+                   "Marker file creation "
+                   "failed while performing %s; entry gfid=%s",
+                   gf_fop_string(local->fop), local->newloc.name);
+            goto err;
+        } else {
+            shard_lookup_marker_file(frame, this);
+            return 0;
+        }
+    }
+
+    linked_inode = inode_link(inode, priv->dot_shard_rm_inode,
+                              local->newloc.name, buf);
+    inode_unref(local->newloc.inode);
+    local->newloc.inode = linked_inode;
+
+    if (local->fop == GF_FOP_UNLINK)
+        shard_unlink_base_file(frame, this);
+    else if (local->fop == GF_FOP_RENAME)
+        shard_rename_src_base_file(frame, this);
+    return 0;
+err:
+    shard_common_failure_unwind(local->fop, frame, -1, local->op_errno);
+    return 0;
+}
+
+int
+shard_create_marker_file_under_remove_me(call_frame_t *frame, xlator_t *this,
+                                         loc_t *loc)
+{
+    int ret = 0;
+    int op_errno = ENOMEM;
+    uint64_t bs = 0;
+    char g1[64] = {
+        0,
+    };
+    char g2[64] = {
+        0,
+    };
+    dict_t *xattr_req = NULL;
+    shard_priv_t *priv = NULL;
+    shard_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    SHARD_SET_ROOT_FS_ID(frame, local);
+
+    xattr_req = shard_create_gfid_dict(local->xattr_req);
+    if (!xattr_req)
+        goto err;
+
+    local->newloc.inode = inode_new(this->itable);
+    local->newloc.parent = inode_ref(priv->dot_shard_rm_inode);
+    ret = inode_path(local->newloc.parent, uuid_utoa(loc->inode->gfid),
+                     (char **)&local->newloc.path);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_PATH_FAILED,
+               "Inode path failed on "
+               "pargfid=%s bname=%s",
+               uuid_utoa_r(priv->dot_shard_rm_gfid, g1),
+               uuid_utoa_r(loc->inode->gfid, g2));
+        goto err;
+    }
+    local->newloc.name = strrchr(local->newloc.path, '/');
+    if (local->newloc.name)
+        local->newloc.name++;
+
+    if (local->fop == GF_FOP_UNLINK)
+        bs = local->block_size;
+    else if (local->fop == GF_FOP_RENAME)
+        bs = local->dst_block_size;
+
+    SHARD_INODE_CREATE_INIT(this, bs, xattr_req, &local->newloc,
+                            local->prebuf.ia_size, 0, err);
+
+    STACK_WIND(frame, shard_create_marker_file_under_remove_me_cbk,
+               FIRST_CHILD(this), FIRST_CHILD(this)->fops->mknod,
+               &local->newloc, 0, 0, 0644, xattr_req);
+    dict_unref(xattr_req);
+    return 0;
+
+err:
+    if (xattr_req)
+        dict_unref(xattr_req);
+    shard_create_marker_file_under_remove_me_cbk(frame, 0, this, -1, op_errno,
+                                                 NULL, NULL, NULL, NULL, NULL);
+    return 0;
+}
+
+int
+shard_unlock_entrylk(call_frame_t *frame, xlator_t *this);
+
+int
+shard_unlink_base_file_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno,
+                           struct iatt *preparent, struct iatt *postparent,
+                           dict_t *xdata)
+{
+    int ret = 0;
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (op_ret < 0) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+    } else {
+        shard_inode_ctx_set_refresh_flag(local->int_inodelk.loc.inode, this);
+        local->preoldparent = *preparent;
+        local->postoldparent = *postparent;
+        if (xdata)
+            local->xattr_rsp = dict_ref(xdata);
+        if (local->cleanup_required)
+            shard_start_background_deletion(this);
+    }
+
+    if (local->entrylk_frame) {
+        ret = shard_unlock_entrylk(frame, this);
+        if (ret < 0) {
+            local->op_ret = -1;
+            local->op_errno = -ret;
+        }
+    }
+
+    ret = shard_unlock_inodelk(frame, this);
+    if (ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = -ret;
+    }
+
+    shard_unlink_cbk(frame, this);
+    return 0;
+}
+
+int
+shard_unlink_base_file(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = frame->local;
+
+    /* To-Do: Request open-fd count on base file */
+    STACK_WIND(frame, shard_unlink_base_file_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->unlink, &local->loc, local->xflag,
+               local->xattr_req);
+    return 0;
+}
+
+int
+shard_unlock_entrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    if (op_ret)
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, SHARD_MSG_FOP_FAILED,
+               "Unlock failed. Please check brick logs for "
+               "more details");
+    SHARD_STACK_DESTROY(frame);
+    return 0;
+}
+
+int
+shard_unlock_entrylk(call_frame_t *frame, xlator_t *this)
+{
+    loc_t *loc = NULL;
+    call_frame_t *lk_frame = NULL;
+    shard_local_t *local = NULL;
+    shard_local_t *lk_local = NULL;
+    shard_entrylk_t *lock = NULL;
+
+    local = frame->local;
+    lk_frame = local->entrylk_frame;
+    lk_local = lk_frame->local;
+    local->entrylk_frame = NULL;
+    lock = &lk_local->int_entrylk;
+    loc = &lock->loc;
+
+    STACK_WIND(lk_frame, shard_unlock_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, this->name, loc,
+               lk_local->int_entrylk.basename, ENTRYLK_UNLOCK, ENTRYLK_WRLCK,
+               NULL);
+    local->int_entrylk.acquired_lock = _gf_false;
+    return 0;
+}
+
+int
+shard_post_entrylk_fop_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    switch (local->fop) {
+        case GF_FOP_UNLINK:
+        case GF_FOP_RENAME:
+            shard_create_marker_file_under_remove_me(frame, this,
+                                                     &local->int_inodelk.loc);
+            break;
+        default:
+            gf_msg(this->name, GF_LOG_WARNING, 0, SHARD_MSG_INVALID_FOP,
+                   "post-entrylk handler not defined. This case should not"
+                   " be hit");
+            break;
+    }
+    return 0;
+}
+
+int
+shard_acquire_entrylk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    call_frame_t *main_frame = NULL;
+    shard_local_t *local = NULL;
+    shard_local_t *main_local = NULL;
+
+    local = frame->local;
+    main_frame = local->main_frame;
+    main_local = main_frame->local;
+
+    if (local->op_ret < 0) {
+        shard_common_failure_unwind(main_local->fop, main_frame, op_ret,
+                                    op_errno);
+        return 0;
+    }
+    main_local->int_entrylk.acquired_lock = _gf_true;
+    shard_post_entrylk_fop_handler(main_frame, this);
+    return 0;
+}
+
+int
+shard_acquire_entrylk(call_frame_t *frame, xlator_t *this, inode_t *inode,
+                      uuid_t gfid)
+{
+    char gfid_str[GF_UUID_BUF_SIZE] = {
+        0,
+    };
+    shard_local_t *local = NULL;
+    shard_local_t *entrylk_local = NULL;
+    shard_entrylk_t *int_entrylk = NULL;
+    call_frame_t *entrylk_frame = NULL;
+
+    local = frame->local;
+    entrylk_frame = create_frame(this, this->ctx->pool);
+    if (!entrylk_frame) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, SHARD_MSG_MEMALLOC_FAILED,
+               "Failed to create new frame "
+               "to lock marker file");
+        goto err;
+    }
+
+    entrylk_local = mem_get0(this->local_pool);
+    if (!entrylk_local) {
+        STACK_DESTROY(entrylk_frame->root);
+        goto err;
+    }
+
+    entrylk_frame->local = entrylk_local;
+    entrylk_local->main_frame = frame;
+    int_entrylk = &entrylk_local->int_entrylk;
+
+    int_entrylk->loc.inode = inode_ref(inode);
+    set_lk_owner_from_ptr(&entrylk_frame->root->lk_owner, entrylk_frame->root);
+    local->entrylk_frame = entrylk_frame;
+    gf_uuid_unparse(gfid, gfid_str);
+    int_entrylk->basename = gf_strdup(gfid_str);
+
+    STACK_WIND(entrylk_frame, shard_acquire_entrylk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->entrylk, this->name, &int_entrylk->loc,
+               int_entrylk->basename, ENTRYLK_LOCK, ENTRYLK_WRLCK, NULL);
+    return 0;
+err:
+    shard_common_failure_unwind(local->fop, frame, -1, ENOMEM);
+    return 0;
+}
+
+int
+shard_post_lookup_base_shard_rm_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+    shard_priv_t *priv = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    if (local->op_ret < 0) {
+        shard_common_failure_unwind(local->fop, frame, -1, local->op_errno);
+        return 0;
+    }
+
+    if (local->prebuf.ia_nlink > 1) {
+        gf_msg_debug(this->name, 0,
+                     "link count on %s > 1:%d, "
+                     "performing rename()/unlink()",
+                     local->int_inodelk.loc.path, local->prebuf.ia_nlink);
+        if (local->fop == GF_FOP_RENAME)
+            shard_rename_src_base_file(frame, this);
+        else if (local->fop == GF_FOP_UNLINK)
+            shard_unlink_base_file(frame, this);
+    } else {
+        gf_msg_debug(this->name, 0,
+                     "link count on %s = 1, creating "
+                     "file under .remove_me",
+                     local->int_inodelk.loc.path);
+        local->cleanup_required = _gf_true;
+        shard_acquire_entrylk(frame, this, priv->dot_shard_rm_inode,
+                              local->prebuf.ia_gfid);
+    }
+    return 0;
+}
+
+int
+shard_post_inodelk_fop_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    switch (local->fop) {
+        case GF_FOP_UNLINK:
+        case GF_FOP_RENAME:
+            shard_refresh_base_file(frame, this, &local->int_inodelk.loc, NULL,
+                                    shard_post_lookup_base_shard_rm_handler);
+            break;
+        default:
+            gf_msg(this->name, GF_LOG_WARNING, 0, SHARD_MSG_INVALID_FOP,
+                   "post-inodelk handler not defined. This case should not"
+                   " be hit");
+            break;
+    }
+    return 0;
+}
+
+int
+shard_acquire_inodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    call_frame_t *main_frame = NULL;
+    shard_local_t *local = NULL;
+    shard_local_t *main_local = NULL;
+
+    local = frame->local;
+    main_frame = local->main_frame;
+    main_local = main_frame->local;
+
+    if (local->op_ret < 0) {
+        shard_common_failure_unwind(main_local->fop, main_frame, op_ret,
+                                    op_errno);
+        return 0;
+    }
+    main_local->int_inodelk.acquired_lock = _gf_true;
+    shard_post_inodelk_fop_handler(main_frame, this);
+    return 0;
+}
+
+int
+shard_acquire_inodelk(call_frame_t *frame, xlator_t *this, loc_t *loc)
+{
+    call_frame_t *lk_frame = NULL;
+    shard_local_t *local = NULL;
+    shard_local_t *lk_local = NULL;
+    shard_inodelk_t *int_inodelk = NULL;
+
+    local = frame->local;
+    lk_frame = create_frame(this, this->ctx->pool);
+    if (!lk_frame) {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM, SHARD_MSG_MEMALLOC_FAILED,
+               "Failed to create new frame "
+               "to lock base shard");
+        goto err;
+    }
+    lk_local = mem_get0(this->local_pool);
+    if (!lk_local) {
+        STACK_DESTROY(lk_frame->root);
+        goto err;
+    }
+
+    lk_frame->local = lk_local;
+    lk_local->main_frame = frame;
+    int_inodelk = &lk_local->int_inodelk;
+
+    int_inodelk->flock.l_len = 0;
+    int_inodelk->flock.l_start = 0;
+    int_inodelk->domain = this->name;
+    int_inodelk->flock.l_type = F_WRLCK;
+    loc_copy(&local->int_inodelk.loc, loc);
+    set_lk_owner_from_ptr(&lk_frame->root->lk_owner, lk_frame->root);
+    local->inodelk_frame = lk_frame;
+
+    STACK_WIND(lk_frame, shard_acquire_inodelk_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->inodelk, int_inodelk->domain,
+               &local->int_inodelk.loc, F_SETLKW, &int_inodelk->flock, NULL);
+    return 0;
+err:
+    shard_common_failure_unwind(local->fop, frame, -1, ENOMEM);
+    return 0;
+}
+
+int
+shard_post_mkdir_rm_handler(call_frame_t *frame, xlator_t *this)
+{
+    loc_t *loc = NULL;
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->op_ret < 0) {
+        shard_common_failure_unwind(local->fop, frame, -1, local->op_errno);
+        return 0;
+    }
+    if (local->fop == GF_FOP_UNLINK)
+        loc = &local->loc;
+    else if (local->fop == GF_FOP_RENAME)
+        loc = &local->loc2;
+    shard_acquire_inodelk(frame, this, loc);
+    return 0;
+}
+
+int
+shard_mkdir_internal_dir(call_frame_t *frame, xlator_t *this,
+                         shard_post_resolve_fop_handler_t handler,
+                         shard_internal_dir_type_t type);
+int
+shard_pre_mkdir_rm_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->op_ret < 0) {
+        shard_common_failure_unwind(local->fop, frame, -1, local->op_errno);
+        return 0;
+    }
+    shard_mkdir_internal_dir(frame, this, shard_post_mkdir_rm_handler,
+                             SHARD_INTERNAL_DIR_DOT_SHARD_REMOVE_ME);
+    return 0;
+}
+
+void
+shard_begin_rm_resolution(call_frame_t *frame, xlator_t *this)
+{
+    shard_priv_t *priv = NULL;
+    shard_local_t *local = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    local->dot_shard_rm_loc.inode = inode_find(this->itable,
+                                               priv->dot_shard_rm_gfid);
+    if (!local->dot_shard_rm_loc.inode) {
+        local->dot_shard_loc.inode = inode_find(this->itable,
+                                                priv->dot_shard_gfid);
+        if (!local->dot_shard_loc.inode) {
+            shard_mkdir_internal_dir(frame, this, shard_pre_mkdir_rm_handler,
+                                     SHARD_INTERNAL_DIR_DOT_SHARD);
+        } else {
+            local->post_res_handler = shard_pre_mkdir_rm_handler;
+            shard_refresh_internal_dir(frame, this,
+                                       SHARD_INTERNAL_DIR_DOT_SHARD);
+        }
+    } else {
+        local->post_res_handler = shard_post_mkdir_rm_handler;
+        shard_refresh_internal_dir(frame, this,
+                                   SHARD_INTERNAL_DIR_DOT_SHARD_REMOVE_ME);
+    }
+}
+
+int
+shard_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+             dict_t *xdata)
+{
+    int ret = -1;
+    uint64_t block_size = 0;
+    shard_local_t *local = NULL;
+
+    ret = shard_inode_ctx_get_block_size(loc->inode, this, &block_size);
+    if ((ret) && (!IA_ISLNK(loc->inode->ia_type))) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_CTX_GET_FAILED,
+               "Failed to get block "
+               "size from inode ctx of %s",
+               uuid_utoa(loc->inode->gfid));
+        goto err;
+    }
+
+    if (!block_size || frame->root->pid == GF_CLIENT_PID_GSYNCD) {
+        STACK_WIND(frame, default_unlink_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->unlink, loc, xflag, xdata);
+        return 0;
+    }
+
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto err;
+
+    frame->local = local;
+
+    loc_copy(&local->loc, loc);
+    local->xflag = xflag;
+    local->xattr_req = (xdata) ? dict_ref(xdata) : dict_new();
+    local->block_size = block_size;
+    local->resolver_base_inode = loc->inode;
+    local->fop = GF_FOP_UNLINK;
+    if (!this->itable)
+        this->itable = (local->loc.inode)->table;
+
+    local->resolve_not = _gf_true;
+    shard_begin_rm_resolution(frame, this);
+    return 0;
+err:
+    shard_common_failure_unwind(GF_FOP_UNLINK, frame, -1, ENOMEM);
+    return 0;
+}
+
+int
+shard_post_rename_lookup_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_rename_cbk(frame, this);
+    return 0;
+}
+
+int
+shard_rename_src_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                     struct iatt *preoldparent, struct iatt *postoldparent,
+                     struct iatt *prenewparent, struct iatt *postnewparent,
+                     dict_t *xdata)
+{
+    int ret = 0;
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (op_ret < 0) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        goto err;
+    }
+    /* Set ctx->refresh to TRUE to force a lookup on disk when
+     * shard_lookup_base_file() is called next to refresh the hard link
+     * count in ctx. Note that this is applicable only to the case where
+     * the rename dst is already existent and sharded.
+     */
+    if ((local->dst_block_size) && (!local->cleanup_required))
+        shard_inode_ctx_set_refresh_flag(local->int_inodelk.loc.inode, this);
+
+    local->prebuf = *buf;
+    local->preoldparent = *preoldparent;
+    local->postoldparent = *postoldparent;
+    local->prenewparent = *prenewparent;
+    local->postnewparent = *postnewparent;
+    if (xdata)
+        local->xattr_rsp = dict_ref(xdata);
+
+    if (local->dst_block_size) {
+        if (local->entrylk_frame) {
+            ret = shard_unlock_entrylk(frame, this);
+            if (ret < 0) {
+                local->op_ret = -1;
+                local->op_errno = -ret;
+            }
+        }
+
+        ret = shard_unlock_inodelk(frame, this);
+        if (ret < 0) {
+            local->op_ret = -1;
+            local->op_errno = -ret;
+            goto err;
+        }
+        if (local->cleanup_required)
+            shard_start_background_deletion(this);
+    }
+
+    /* Now the base file of src, if sharded, is looked up to gather ia_size
+     * and ia_blocks.*/
+    if (local->block_size) {
+        local->tmp_loc.inode = inode_new(this->itable);
+        gf_uuid_copy(local->tmp_loc.gfid, (local->loc.inode)->gfid);
+        shard_refresh_base_file(frame, this, &local->tmp_loc, NULL,
+                                shard_post_rename_lookup_handler);
+    } else {
+        shard_rename_cbk(frame, this);
+    }
+    return 0;
+err:
+    shard_common_failure_unwind(local->fop, frame, local->op_ret,
+                                local->op_errno);
+    return 0;
+}
+
+int
+shard_post_lookup_dst_base_file_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->op_ret < 0) {
+        shard_common_failure_unwind(local->fop, frame, local->op_ret,
+                                    local->op_errno);
+        return 0;
+    }
+
+    /* Save dst base file attributes into postbuf so the information is not
+     * lost when it is overwritten after lookup on base file of src in
+     * shard_lookup_base_file_cbk().
+     */
+    local->postbuf = local->prebuf;
+    shard_rename_src_base_file(frame, this);
+    return 0;
+}
+
+int
+shard_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+             dict_t *xdata)
+{
+    int ret = -1;
+    uint64_t block_size = 0;
+    uint64_t dst_block_size = 0;
+    shard_local_t *local = NULL;
+
+    if (IA_ISDIR(oldloc->inode->ia_type)) {
+        STACK_WIND(frame, default_rename_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
+        return 0;
+    }
+
+    ret = shard_inode_ctx_get_block_size(oldloc->inode, this, &block_size);
+    if ((ret) && (!IA_ISLNK(oldloc->inode->ia_type))) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_CTX_GET_FAILED,
+               "Failed to get block "
+               "size from inode ctx of %s",
+               uuid_utoa(oldloc->inode->gfid));
+        goto err;
+    }
+
+    if (newloc->inode)
+        ret = shard_inode_ctx_get_block_size(newloc->inode, this,
+                                             &dst_block_size);
+
+    /* The following stack_wind covers the case where:
+     * a. the src file is not sharded and dst doesn't exist, OR
+     * b. the src and dst both exist but are not sharded.
+     */
+    if (((!block_size) && (!dst_block_size)) ||
+        frame->root->pid == GF_CLIENT_PID_GSYNCD) {
+        STACK_WIND(frame, default_rename_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
+        return 0;
+    }
+
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto err;
+
+    frame->local = local;
+    loc_copy(&local->loc, oldloc);
+    loc_copy(&local->loc2, newloc);
+    local->resolver_base_inode = newloc->inode;
+    local->fop = GF_FOP_RENAME;
+    local->xattr_req = (xdata) ? dict_ref(xdata) : dict_new();
+    if (!local->xattr_req)
+        goto err;
+
+    local->block_size = block_size;
+    local->dst_block_size = dst_block_size;
+    if (!this->itable)
+        this->itable = (local->loc.inode)->table;
+    local->resolve_not = _gf_true;
+
+    /* The following if-block covers the case where the dst file exists
+     * and is sharded.
+     */
+    if (local->dst_block_size) {
+        shard_begin_rm_resolution(frame, this);
+    } else {
+        /* The following block covers the case where the dst either doesn't
+         * exist or is NOT sharded but the src is sharded. In this case, shard
+         * xlator would go ahead and rename src to dst. Once done, it would also
+         * lookup the base shard of src to get the ia_size and ia_blocks xattr
+         * values.
+         */
+        shard_rename_src_base_file(frame, this);
+    }
+    return 0;
+
+err:
+    shard_common_failure_unwind(GF_FOP_RENAME, frame, -1, ENOMEM);
+    return 0;
+}
+
+int
+shard_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, fd_t *fd, inode_t *inode,
+                 struct iatt *stbuf, struct iatt *preparent,
+                 struct iatt *postparent, dict_t *xdata)
+{
+    int ret = -1;
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (op_ret == -1)
+        goto unwind;
+
+    ret = shard_inode_ctx_set(inode, this, stbuf, local->block_size,
+                              SHARD_ALL_MASK);
+    if (ret)
+        gf_msg(this->name, GF_LOG_WARNING, 0, SHARD_MSG_INODE_CTX_SET_FAILED,
+               "Failed to set inode "
+               "ctx for %s",
+               uuid_utoa(inode->gfid));
+
+unwind:
+    SHARD_STACK_UNWIND(create, frame, op_ret, op_errno, fd, inode, stbuf,
+                       preparent, postparent, xdata);
+    return 0;
+}
+
+int
+shard_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+             mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+    shard_priv_t *priv = NULL;
+    shard_local_t *local = NULL;
+
+    priv = this->private;
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto err;
+
+    frame->local = local;
+    local->block_size = priv->block_size;
+
+    if (!__is_gsyncd_on_shard_dir(frame, loc)) {
+        SHARD_INODE_CREATE_INIT(this, local->block_size, xdata, loc, 0, 0, err);
+    }
+
+    STACK_WIND(frame, shard_create_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->create, loc, flags, mode, umask, fd,
+               xdata);
+    return 0;
+err:
+    shard_common_failure_unwind(GF_FOP_CREATE, frame, -1, ENOMEM);
+    return 0;
+}
+
+int
+shard_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
+{
+    /* To-Do: Handle open with O_TRUNC under locks */
+    SHARD_STACK_UNWIND(open, frame, op_ret, op_errno, fd, xdata);
+    return 0;
+}
+
+int
+shard_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+           fd_t *fd, dict_t *xdata)
+{
+    STACK_WIND(frame, shard_open_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->open, loc, flags, fd, xdata);
+    return 0;
+}
+
+int
+shard_readv_do_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, struct iovec *vector,
+                   int32_t count, struct iatt *stbuf, struct iobref *iobref,
+                   dict_t *xdata)
+{
+    int i = 0;
+    int call_count = 0;
+    void *address = NULL;
+    uint64_t block_num = 0;
+    off_t off = 0;
+    struct iovec vec = {
+        0,
+    };
+    shard_local_t *local = NULL;
+    fd_t *anon_fd = cookie;
+    shard_inode_ctx_t *ctx = NULL;
+
+    local = frame->local;
+
+    /* If shard has already seen a failure here before, there is no point
+     * in aggregating subsequent reads, so just go to out.
+     */
+    if (local->op_ret < 0)
+        goto out;
+
+    if (op_ret < 0) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        goto out;
+    }
+
+    if (local->op_ret >= 0)
+        local->op_ret += op_ret;
+
+    shard_inode_ctx_get(anon_fd->inode, this, &ctx);
+    block_num = ctx->block_num;
+
+    if (block_num == local->first_block) {
+        address = local->iobuf->ptr;
+    } else {
+        /* else
+         * address to start writing to = beginning of buffer +
+         *                    number of bytes until end of first block +
+         *                    + block_size times number of blocks
+         *                    between the current block and the first
+         */
+        address = (char *)local->iobuf->ptr +
+                  (local->block_size - (local->offset % local->block_size)) +
+                  ((block_num - local->first_block - 1) * local->block_size);
+    }
+
+    for (i = 0; i < count; i++) {
+        address = (char *)address + off;
+        memcpy(address, vector[i].iov_base, vector[i].iov_len);
+        off += vector[i].iov_len;
+    }
+
+out:
+    if (anon_fd)
+        fd_unref(anon_fd);
+    call_count = shard_call_count_return(frame);
+    if (call_count == 0) {
+        SHARD_UNSET_ROOT_FS_ID(frame, local);
+        if (local->op_ret < 0) {
+            shard_common_failure_unwind(GF_FOP_READ, frame, local->op_ret,
+                                        local->op_errno);
+        } else {
+            if (xdata)
+                local->xattr_rsp = dict_ref(xdata);
+            vec.iov_base = local->iobuf->ptr;
+            if (local->offset + local->req_size > local->prebuf.ia_size)
+                local->total_size = local->prebuf.ia_size - local->offset;
+            vec.iov_len = local->total_size;
+            local->op_ret = local->total_size;
+            SHARD_STACK_UNWIND(readv, frame, local->op_ret, local->op_errno,
+                               &vec, 1, &local->prebuf, local->iobref,
+                               local->xattr_rsp);
+            return 0;
+        }
+    }
+
+    return 0;
+}
+
+int
+shard_readv_do(call_frame_t *frame, xlator_t *this)
+{
+    int i = 0;
+    int call_count = 0;
+    int last_block = 0;
+    int cur_block = 0;
+    off_t orig_offset = 0;
+    off_t shard_offset = 0;
+    size_t read_size = 0;
+    size_t remaining_size = 0;
+    fd_t *fd = NULL;
+    fd_t *anon_fd = NULL;
+    shard_local_t *local = NULL;
+    gf_boolean_t wind_failed = _gf_false;
+
+    local = frame->local;
+    fd = local->fd;
+
+    orig_offset = local->offset;
+    cur_block = local->first_block;
+    last_block = local->last_block;
+    remaining_size = local->total_size;
+    local->call_count = call_count = local->num_blocks;
+
+    SHARD_SET_ROOT_FS_ID(frame, local);
+
+    if (fd->flags & O_DIRECT)
+        local->flags = O_DIRECT;
+
+    while (cur_block <= last_block) {
+        if (wind_failed) {
+            shard_readv_do_cbk(frame, (void *)(long)0, this, -1, ENOMEM, NULL,
+                               0, NULL, NULL, NULL);
+            goto next;
+        }
+
+        shard_offset = orig_offset % local->block_size;
+        read_size = local->block_size - shard_offset;
+        if (read_size > remaining_size)
+            read_size = remaining_size;
+
+        remaining_size -= read_size;
+
+        if (cur_block == 0) {
+            anon_fd = fd_ref(fd);
+        } else {
+            anon_fd = fd_anonymous(local->inode_list[i]);
+            if (!anon_fd) {
+                local->op_ret = -1;
+                local->op_errno = ENOMEM;
+                wind_failed = _gf_true;
+                shard_readv_do_cbk(frame, (void *)(long)anon_fd, this, -1,
+                                   ENOMEM, NULL, 0, NULL, NULL, NULL);
+                goto next;
+            }
+        }
+
+        STACK_WIND_COOKIE(frame, shard_readv_do_cbk, anon_fd, FIRST_CHILD(this),
+                          FIRST_CHILD(this)->fops->readv, anon_fd, read_size,
+                          shard_offset, local->flags, local->xattr_req);
+
+        orig_offset += read_size;
+    next:
+        cur_block++;
+        i++;
+        call_count--;
+    }
+    return 0;
+}
+
+int
+shard_common_mknod_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, inode_t *inode,
+                       struct iatt *buf, struct iatt *preparent,
+                       struct iatt *postparent, dict_t *xdata)
+{
+    int shard_block_num = (long)cookie;
+    int call_count = 0;
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (op_ret < 0) {
+        if (op_errno == EEXIST) {
+            LOCK(&frame->lock);
+            {
+                local->eexist_count++;
+            }
+            UNLOCK(&frame->lock);
+        } else {
+            local->op_ret = op_ret;
+            local->op_errno = op_errno;
+        }
+        gf_msg_debug(this->name, 0,
+                     "mknod of shard %d "
+                     "failed: %s",
+                     shard_block_num, strerror(op_errno));
+        goto done;
+    }
+
+    shard_link_block_inode(local, shard_block_num, inode, buf);
+
+done:
+    call_count = shard_call_count_return(frame);
+    if (call_count == 0) {
+        SHARD_UNSET_ROOT_FS_ID(frame, local);
+        local->create_count = 0;
+        local->post_mknod_handler(frame, this);
+    }
+
+    return 0;
+}
+
+int
+shard_common_resume_mknod(call_frame_t *frame, xlator_t *this,
+                          shard_post_mknod_fop_handler_t post_mknod_handler)
+{
+    int i = 0;
+    int shard_idx_iter = 0;
+    int last_block = 0;
+    int ret = 0;
+    int call_count = 0;
+    char path[PATH_MAX] = {
+        0,
+    };
+    mode_t mode = 0;
+    char *bname = NULL;
+    shard_priv_t *priv = NULL;
+    shard_inode_ctx_t ctx_tmp = {
+        0,
+    };
+    shard_local_t *local = NULL;
+    gf_boolean_t wind_failed = _gf_false;
+    fd_t *fd = NULL;
+    loc_t loc = {
+        0,
+    };
+    dict_t *xattr_req = NULL;
+
+    local = frame->local;
+    priv = this->private;
+    fd = local->fd;
+    shard_idx_iter = local->first_block;
+    last_block = local->last_block;
+    call_count = local->call_count = local->create_count;
+    local->post_mknod_handler = post_mknod_handler;
+
+    SHARD_SET_ROOT_FS_ID(frame, local);
+
+    ret = shard_inode_ctx_get_all(fd->inode, this, &ctx_tmp);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_CTX_GET_FAILED,
+               "Failed to get inode "
+               "ctx for %s",
+               uuid_utoa(fd->inode->gfid));
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        goto err;
+    }
+    mode = st_mode_from_ia(ctx_tmp.stat.ia_prot, ctx_tmp.stat.ia_type);
+
+    while (shard_idx_iter <= last_block) {
+        if (local->inode_list[i]) {
+            shard_idx_iter++;
+            i++;
+            continue;
+        }
+
+        if (wind_failed) {
+            shard_common_mknod_cbk(frame, (void *)(long)shard_idx_iter, this,
+                                   -1, ENOMEM, NULL, NULL, NULL, NULL, NULL);
+            goto next;
+        }
+
+        shard_make_block_abspath(shard_idx_iter, fd->inode->gfid, path,
+                                 sizeof(path));
+
+        xattr_req = shard_create_gfid_dict(local->xattr_req);
+        if (!xattr_req) {
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            wind_failed = _gf_true;
+            shard_common_mknod_cbk(frame, (void *)(long)shard_idx_iter, this,
+                                   -1, ENOMEM, NULL, NULL, NULL, NULL, NULL);
+            goto next;
+        }
+
+        bname = strrchr(path, '/') + 1;
+        loc.inode = inode_new(this->itable);
+        loc.parent = inode_ref(priv->dot_shard_inode);
+        ret = inode_path(loc.parent, bname, (char **)&(loc.path));
+        if (ret < 0 || !(loc.inode)) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_PATH_FAILED,
+                   "Inode path failed"
+                   "on %s, base file gfid = %s",
+                   bname, uuid_utoa(fd->inode->gfid));
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            wind_failed = _gf_true;
+            loc_wipe(&loc);
+            dict_unref(xattr_req);
+            shard_common_mknod_cbk(frame, (void *)(long)shard_idx_iter, this,
+                                   -1, ENOMEM, NULL, NULL, NULL, NULL, NULL);
+            goto next;
+        }
+
+        loc.name = strrchr(loc.path, '/');
+        if (loc.name)
+            loc.name++;
+
+        STACK_WIND_COOKIE(frame, shard_common_mknod_cbk,
+                          (void *)(long)shard_idx_iter, FIRST_CHILD(this),
+                          FIRST_CHILD(this)->fops->mknod, &loc, mode,
+                          ctx_tmp.stat.ia_rdev, 0, xattr_req);
+        loc_wipe(&loc);
+        dict_unref(xattr_req);
+
+    next:
+        shard_idx_iter++;
+        i++;
+        if (!--call_count)
+            break;
+    }
+
+    return 0;
+err:
+    /*
+     * This block is for handling failure in shard_inode_ctx_get_all().
+     * Failures in the while-loop are handled within the loop.
+     */
+    SHARD_UNSET_ROOT_FS_ID(frame, local);
+    post_mknod_handler(frame, this);
+    return 0;
+}
+
+int
+shard_post_mknod_readv_handler(call_frame_t *frame, xlator_t *this);
+
+int
+shard_post_lookup_shards_readv_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->op_ret < 0) {
+        shard_common_failure_unwind(GF_FOP_READ, frame, local->op_ret,
+                                    local->op_errno);
+        return 0;
+    }
+
+    if (local->create_count) {
+        shard_common_resume_mknod(frame, this, shard_post_mknod_readv_handler);
+    } else {
+        shard_readv_do(frame, this);
+    }
+
+    return 0;
+}
+
+int
+shard_post_mknod_readv_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->op_ret < 0) {
+        shard_common_failure_unwind(GF_FOP_READ, frame, local->op_ret,
+                                    local->op_errno);
+        return 0;
+    }
+
+    if (!local->eexist_count) {
+        shard_readv_do(frame, this);
+    } else {
+        local->call_count = local->eexist_count;
+        shard_common_lookup_shards(frame, this, local->loc.inode,
+                                   shard_post_lookup_shards_readv_handler);
+    }
+    return 0;
+}
+
+int
+shard_post_resolve_readv_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->op_ret < 0) {
+        if (local->op_errno != ENOENT) {
+            shard_common_failure_unwind(GF_FOP_READ, frame, local->op_ret,
+                                        local->op_errno);
+            return 0;
+        } else {
+            struct iovec vec = {
+                0,
+            };
+
+            vec.iov_base = local->iobuf->ptr;
+            vec.iov_len = local->total_size;
+            local->op_ret = local->total_size;
+            SHARD_STACK_UNWIND(readv, frame, local->op_ret, 0, &vec, 1,
+                               &local->prebuf, local->iobref, NULL);
+            return 0;
+        }
+    }
+
+    if (local->call_count) {
+        shard_common_lookup_shards(frame, this, local->resolver_base_inode,
+                                   shard_post_lookup_shards_readv_handler);
+    } else {
+        shard_readv_do(frame, this);
+    }
+
+    return 0;
+}
+
+int
+shard_post_lookup_readv_handler(call_frame_t *frame, xlator_t *this)
+{
+    int ret = 0;
+    struct iobuf *iobuf = NULL;
+    shard_local_t *local = NULL;
+    shard_priv_t *priv = NULL;
+
+    priv = this->private;
+    local = frame->local;
+
+    if (local->op_ret < 0) {
+        shard_common_failure_unwind(GF_FOP_READ, frame, local->op_ret,
+                                    local->op_errno);
+        return 0;
+    }
+
+    if (local->offset >= local->prebuf.ia_size) {
+        /* If the read is being performed past the end of the file,
+         * unwind the FOP with 0 bytes read as status.
+         */
+        struct iovec vec = {
+            0,
+        };
+
+        iobuf = iobuf_get2(this->ctx->iobuf_pool, local->req_size);
+        if (!iobuf)
+            goto err;
+
+        vec.iov_base = iobuf->ptr;
+        vec.iov_len = 0;
+        local->iobref = iobref_new();
+        iobref_add(local->iobref, iobuf);
+        iobuf_unref(iobuf);
+
+        SHARD_STACK_UNWIND(readv, frame, 0, 0, &vec, 1, &local->prebuf,
+                           local->iobref, NULL);
+        return 0;
+    }
+
+    local->first_block = get_lowest_block(local->offset, local->block_size);
+
+    local->total_size = local->req_size;
+
+    local->last_block = get_highest_block(local->offset, local->total_size,
+                                          local->block_size);
+
+    local->num_blocks = local->last_block - local->first_block + 1;
+    GF_ASSERT(local->num_blocks > 0);
+    local->resolver_base_inode = local->loc.inode;
+
+    local->inode_list = GF_CALLOC(local->num_blocks, sizeof(inode_t *),
+                                  gf_shard_mt_inode_list);
+    if (!local->inode_list)
+        goto err;
+
+    iobuf = iobuf_get2(this->ctx->iobuf_pool, local->total_size);
+    if (!iobuf)
+        goto err;
+
+    local->iobref = iobref_new();
+    if (!local->iobref) {
+        iobuf_unref(iobuf);
+        goto err;
+    }
+
+    if (iobref_add(local->iobref, iobuf) != 0) {
+        iobuf_unref(iobuf);
+        goto err;
+    }
+
+    memset(iobuf->ptr, 0, local->total_size);
+    iobuf_unref(iobuf);
+    local->iobuf = iobuf;
+
+    local->dot_shard_loc.inode = inode_find(this->itable, priv->dot_shard_gfid);
+    if (!local->dot_shard_loc.inode) {
+        ret = shard_init_internal_dir_loc(this, local,
+                                          SHARD_INTERNAL_DIR_DOT_SHARD);
+        if (ret)
+            goto err;
+        shard_lookup_internal_dir(frame, this, shard_post_resolve_readv_handler,
+                                  SHARD_INTERNAL_DIR_DOT_SHARD);
+    } else {
+        local->post_res_handler = shard_post_resolve_readv_handler;
+        shard_refresh_internal_dir(frame, this, SHARD_INTERNAL_DIR_DOT_SHARD);
+    }
+    return 0;
+err:
+    shard_common_failure_unwind(GF_FOP_READ, frame, -1, ENOMEM);
+    return 0;
+}
+
+int
+shard_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t offset, uint32_t flags, dict_t *xdata)
+{
+    int ret = 0;
+    uint64_t block_size = 0;
+    shard_local_t *local = NULL;
+
+    ret = shard_inode_ctx_get_block_size(fd->inode, this, &block_size);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_CTX_GET_FAILED,
+               "Failed to get block "
+               "size for %s from its inode ctx",
+               uuid_utoa(fd->inode->gfid));
+        goto err;
+    }
+
+    if (!block_size || frame->root->pid == GF_CLIENT_PID_GSYNCD) {
+        /* block_size = 0 means that the file was created before
+         * sharding was enabled on the volume.
+         */
+        STACK_WIND(frame, default_readv_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->readv, fd, size, offset, flags,
+                   xdata);
+        return 0;
+    }
+
+    if (!this->itable)
+        this->itable = fd->inode->table;
+
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto err;
+
+    frame->local = local;
+
+    ret = syncbarrier_init(&local->barrier);
+    if (ret)
+        goto err;
+    local->fd = fd_ref(fd);
+    local->block_size = block_size;
+    local->offset = offset;
+    local->req_size = size;
+    local->flags = flags;
+    local->fop = GF_FOP_READ;
+    local->xattr_req = (xdata) ? dict_ref(xdata) : dict_new();
+    if (!local->xattr_req)
+        goto err;
+
+    local->loc.inode = inode_ref(fd->inode);
+    gf_uuid_copy(local->loc.gfid, fd->inode->gfid);
+
+    shard_refresh_base_file(frame, this, NULL, fd,
+                            shard_post_lookup_readv_handler);
+    return 0;
+err:
+    shard_common_failure_unwind(GF_FOP_READ, frame, -1, ENOMEM);
+    return 0;
+}
+
+int
+shard_common_inode_write_post_update_size_handler(call_frame_t *frame,
+                                                  xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->op_ret < 0) {
+        shard_common_failure_unwind(local->fop, frame, local->op_ret,
+                                    local->op_errno);
+    } else {
+        shard_common_inode_write_success_unwind(local->fop, frame,
+                                                local->written_size);
+    }
+    return 0;
+}
+
+static gf_boolean_t
+shard_is_appending_write(shard_local_t *local)
+{
+    if (local->fop != GF_FOP_WRITE)
+        return _gf_false;
+    if (local->flags & O_APPEND)
+        return _gf_true;
+    if (local->fd->flags & O_APPEND)
+        return _gf_true;
+    return _gf_false;
+}
+
+int
+__shard_get_delta_size_from_inode_ctx(shard_local_t *local, inode_t *inode,
+                                      xlator_t *this)
+{
+    int ret = -1;
+    uint64_t ctx_uint = 0;
+    shard_inode_ctx_t *ctx = NULL;
+
+    ret = __inode_ctx_get(inode, this, &ctx_uint);
+    if (ret < 0)
+        return ret;
+
+    ctx = (shard_inode_ctx_t *)(uintptr_t)ctx_uint;
+
+    if (shard_is_appending_write(local)) {
+        local->delta_size = local->total_size;
+    } else if (local->offset + local->total_size > ctx->stat.ia_size) {
+        local->delta_size = (local->offset + local->total_size) -
+                            ctx->stat.ia_size;
+    } else {
+        local->delta_size = 0;
+    }
+    ctx->stat.ia_size += (local->delta_size);
+    local->postbuf = ctx->stat;
+
+    return 0;
+}
+
+int
+shard_get_delta_size_from_inode_ctx(shard_local_t *local, inode_t *inode,
+                                    xlator_t *this)
+{
+    int ret = -1;
+
+    LOCK(&inode->lock);
+    {
+        ret = __shard_get_delta_size_from_inode_ctx(local, inode, this);
+    }
+    UNLOCK(&inode->lock);
+
+    return ret;
+}
+
+int
+shard_common_inode_write_do_cbk(call_frame_t *frame, void *cookie,
+                                xlator_t *this, int32_t op_ret,
+                                int32_t op_errno, struct iatt *pre,
+                                struct iatt *post, dict_t *xdata)
+{
+    int call_count = 0;
+    fd_t *anon_fd = cookie;
+    shard_local_t *local = NULL;
+    glusterfs_fop_t fop = 0;
+
+    local = frame->local;
+    fop = local->fop;
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret < 0) {
+            local->op_ret = op_ret;
+            local->op_errno = op_errno;
+        } else {
+            local->written_size += op_ret;
+            GF_ATOMIC_ADD(local->delta_blocks,
+                          post->ia_blocks - pre->ia_blocks);
+            local->delta_size += (post->ia_size - pre->ia_size);
+            shard_inode_ctx_set(local->fd->inode, this, post, 0,
+                                SHARD_MASK_TIMES);
+            if (local->fd->inode != anon_fd->inode)
+                shard_inode_ctx_add_to_fsync_list(local->fd->inode, this,
+                                                  anon_fd->inode);
+        }
+    }
+    UNLOCK(&frame->lock);
+
+    if (anon_fd)
+        fd_unref(anon_fd);
+
+    call_count = shard_call_count_return(frame);
+    if (call_count == 0) {
+        SHARD_UNSET_ROOT_FS_ID(frame, local);
+        if (local->op_ret < 0) {
+            shard_common_failure_unwind(fop, frame, local->op_ret,
+                                        local->op_errno);
+        } else {
+            shard_get_delta_size_from_inode_ctx(local, local->fd->inode, this);
+            local->hole_size = 0;
+            if (xdata)
+                local->xattr_rsp = dict_ref(xdata);
+            shard_update_file_size(
+                frame, this, local->fd, NULL,
+                shard_common_inode_write_post_update_size_handler);
+        }
+    }
+
+    return 0;
+}
+
+int
+shard_common_inode_write_wind(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                              struct iovec *vec, int count, off_t shard_offset,
+                              size_t size)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    switch (local->fop) {
+        case GF_FOP_WRITE:
+            STACK_WIND_COOKIE(
+                frame, shard_common_inode_write_do_cbk, fd, FIRST_CHILD(this),
+                FIRST_CHILD(this)->fops->writev, fd, vec, count, shard_offset,
+                local->flags, local->iobref, local->xattr_req);
+            break;
+        case GF_FOP_FALLOCATE:
+            STACK_WIND_COOKIE(
+                frame, shard_common_inode_write_do_cbk, fd, FIRST_CHILD(this),
+                FIRST_CHILD(this)->fops->fallocate, fd, local->flags,
+                shard_offset, size, local->xattr_req);
+            break;
+        case GF_FOP_ZEROFILL:
+            STACK_WIND_COOKIE(frame, shard_common_inode_write_do_cbk, fd,
+                              FIRST_CHILD(this),
+                              FIRST_CHILD(this)->fops->zerofill, fd,
+                              shard_offset, size, local->xattr_req);
+            break;
+        case GF_FOP_DISCARD:
+            STACK_WIND_COOKIE(frame, shard_common_inode_write_do_cbk, fd,
+                              FIRST_CHILD(this),
+                              FIRST_CHILD(this)->fops->discard, fd,
+                              shard_offset, size, local->xattr_req);
+            break;
+        default:
+            gf_msg(this->name, GF_LOG_WARNING, 0, SHARD_MSG_INVALID_FOP,
+                   "Invalid fop id = %d", local->fop);
+            break;
+    }
+    return 0;
+}
+
+int
+shard_common_inode_write_do(call_frame_t *frame, xlator_t *this)
+{
+    int i = 0;
+    int count = 0;
+    int call_count = 0;
+    int last_block = 0;
+    uint32_t cur_block = 0;
+    fd_t *fd = NULL;
+    fd_t *anon_fd = NULL;
+    shard_local_t *local = NULL;
+    struct iovec *vec = NULL;
+    gf_boolean_t wind_failed = _gf_false;
+    gf_boolean_t odirect = _gf_false;
+    off_t orig_offset = 0;
+    off_t shard_offset = 0;
+    off_t vec_offset = 0;
+    size_t remaining_size = 0;
+    size_t shard_write_size = 0;
+
+    local = frame->local;
+    fd = local->fd;
+
+    orig_offset = local->offset;
+    remaining_size = local->total_size;
+    cur_block = local->first_block;
+    local->call_count = call_count = local->num_blocks;
+    last_block = local->last_block;
+
+    SHARD_SET_ROOT_FS_ID(frame, local);
+
+    if (dict_set_uint32(local->xattr_req, GLUSTERFS_WRITE_UPDATE_ATOMIC, 4)) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_DICT_OP_FAILED,
+               "Failed to set " GLUSTERFS_WRITE_UPDATE_ATOMIC
+               " into "
+               "dict: %s",
+               uuid_utoa(fd->inode->gfid));
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        local->call_count = 1;
+        shard_common_inode_write_do_cbk(frame, (void *)(long)0, this, -1,
+                                        ENOMEM, NULL, NULL, NULL);
+        return 0;
+    }
+
+    if ((fd->flags & O_DIRECT) && (local->fop == GF_FOP_WRITE))
+        odirect = _gf_true;
+
+    while (cur_block <= last_block) {
+        if (wind_failed) {
+            shard_common_inode_write_do_cbk(frame, (void *)(long)0, this, -1,
+                                            ENOMEM, NULL, NULL, NULL);
+            goto next;
+        }
+
+        shard_offset = orig_offset % local->block_size;
+        shard_write_size = local->block_size - shard_offset;
+        if (shard_write_size > remaining_size)
+            shard_write_size = remaining_size;
+
+        remaining_size -= shard_write_size;
+
+        if (local->fop == GF_FOP_WRITE) {
+            vec = NULL;
+            count = iov_subset(local->vector, local->count, vec_offset,
+                               shard_write_size, &vec, 0);
+            if (count < 0) {
+                local->op_ret = -1;
+                local->op_errno = ENOMEM;
+                wind_failed = _gf_true;
+                shard_common_inode_write_do_cbk(frame, (void *)(long)0, this,
+                                                -1, ENOMEM, NULL, NULL, NULL);
+                goto next;
+            }
+        }
+
+        if (cur_block == 0) {
+            anon_fd = fd_ref(fd);
+        } else {
+            anon_fd = fd_anonymous(local->inode_list[i]);
+            if (!anon_fd) {
+                local->op_ret = -1;
+                local->op_errno = ENOMEM;
+                wind_failed = _gf_true;
+                GF_FREE(vec);
+                shard_common_inode_write_do_cbk(frame, (void *)(long)anon_fd,
+                                                this, -1, ENOMEM, NULL, NULL,
+                                                NULL);
+                goto next;
+            }
+
+            if (local->fop == GF_FOP_WRITE) {
+                if (odirect)
+                    local->flags = O_DIRECT;
+                else
+                    local->flags = GF_ANON_FD_FLAGS;
+            }
+        }
+
+        shard_common_inode_write_wind(frame, this, anon_fd, vec, count,
+                                      shard_offset, shard_write_size);
+        if (vec)
+            vec_offset += shard_write_size;
+        orig_offset += shard_write_size;
+        GF_FREE(vec);
+        vec = NULL;
+    next:
+        cur_block++;
+        i++;
+        call_count--;
+    }
+    return 0;
+}
+
+int
+shard_common_inode_write_post_mknod_handler(call_frame_t *frame,
+                                            xlator_t *this);
+
+int
+shard_common_inode_write_post_lookup_shards_handler(call_frame_t *frame,
+                                                    xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->op_ret < 0) {
+        shard_common_failure_unwind(local->fop, frame, local->op_ret,
+                                    local->op_errno);
+        return 0;
+    }
+
+    if (local->create_count) {
+        shard_common_resume_mknod(frame, this,
+                                  shard_common_inode_write_post_mknod_handler);
+    } else {
+        shard_common_inode_write_do(frame, this);
+    }
+
+    return 0;
+}
+
+int
+shard_common_inode_write_post_mknod_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->op_ret < 0) {
+        shard_common_failure_unwind(local->fop, frame, local->op_ret,
+                                    local->op_errno);
+        return 0;
+    }
+
+    if (!local->eexist_count) {
+        shard_common_inode_write_do(frame, this);
+    } else {
+        local->call_count = local->eexist_count;
+        shard_common_lookup_shards(
+            frame, this, local->loc.inode,
+            shard_common_inode_write_post_lookup_shards_handler);
+    }
+
+    return 0;
+}
+
+int
+shard_common_inode_write_post_resolve_handler(call_frame_t *frame,
+                                              xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->op_ret < 0) {
+        shard_common_failure_unwind(local->fop, frame, local->op_ret,
+                                    local->op_errno);
+        return 0;
+    }
+
+    if (local->call_count) {
+        shard_common_lookup_shards(
+            frame, this, local->resolver_base_inode,
+            shard_common_inode_write_post_lookup_shards_handler);
+    } else if (local->create_count) {
+        shard_common_inode_write_post_lookup_shards_handler(frame, this);
+    } else {
+        shard_common_inode_write_do(frame, this);
+    }
+
+    return 0;
+}
+
+int
+shard_common_inode_write_post_lookup_handler(call_frame_t *frame,
+                                             xlator_t *this)
+{
+    shard_local_t *local = frame->local;
+    shard_priv_t *priv = this->private;
+
+    if (local->op_ret < 0) {
+        shard_common_failure_unwind(local->fop, frame, local->op_ret,
+                                    local->op_errno);
+        return 0;
+    }
+
+    local->postbuf = local->prebuf;
+
+    /*Adjust offset to EOF so that correct shard is chosen for append*/
+    if (shard_is_appending_write(local))
+        local->offset = local->prebuf.ia_size;
+
+    local->first_block = get_lowest_block(local->offset, local->block_size);
+    local->last_block = get_highest_block(local->offset, local->total_size,
+                                          local->block_size);
+    local->num_blocks = local->last_block - local->first_block + 1;
+    GF_ASSERT(local->num_blocks > 0);
+    local->inode_list = GF_CALLOC(local->num_blocks, sizeof(inode_t *),
+                                  gf_shard_mt_inode_list);
+    if (!local->inode_list) {
+        shard_common_failure_unwind(local->fop, frame, -1, ENOMEM);
+        return 0;
+    }
+
+    gf_msg_trace(this->name, 0,
+                 "%s: gfid=%s first_block=%" PRIu64
+                 " "
+                 "last_block=%" PRIu64 " num_blocks=%" PRIu64 " offset=%" PRId64
+                 " total_size=%zu flags=%" PRId32 "",
+                 gf_fop_list[local->fop],
+                 uuid_utoa(local->resolver_base_inode->gfid),
+                 local->first_block, local->last_block, local->num_blocks,
+                 local->offset, local->total_size, local->flags);
+
+    local->dot_shard_loc.inode = inode_find(this->itable, priv->dot_shard_gfid);
+
+    if (!local->dot_shard_loc.inode) {
+        /*change handler*/
+        shard_mkdir_internal_dir(frame, this,
+                                 shard_common_inode_write_post_resolve_handler,
+                                 SHARD_INTERNAL_DIR_DOT_SHARD);
+    } else {
+        /*change handler*/
+        local->post_res_handler = shard_common_inode_write_post_resolve_handler;
+        shard_refresh_internal_dir(frame, this, SHARD_INTERNAL_DIR_DOT_SHARD);
+    }
+    return 0;
+}
+
+int
+shard_mkdir_internal_dir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                             int32_t op_ret, int32_t op_errno, inode_t *inode,
+                             struct iatt *buf, struct iatt *preparent,
+                             struct iatt *postparent, dict_t *xdata)
+{
+    inode_t *link_inode = NULL;
+    shard_local_t *local = NULL;
+    shard_internal_dir_type_t type = (shard_internal_dir_type_t)cookie;
+
+    local = frame->local;
+
+    SHARD_UNSET_ROOT_FS_ID(frame, local);
+
+    if (op_ret == -1) {
+        if (op_errno != EEXIST) {
+            local->op_ret = op_ret;
+            local->op_errno = op_errno;
+            goto unwind;
+        } else {
+            gf_msg_debug(this->name, 0,
+                         "mkdir on %s failed "
+                         "with EEXIST. Attempting lookup now",
+                         shard_internal_dir_string(type));
+            shard_lookup_internal_dir(frame, this, local->post_res_handler,
+                                      type);
+            return 0;
+        }
+    }
+
+    link_inode = shard_link_internal_dir_inode(local, inode, buf, type);
+    if (link_inode != inode) {
+        shard_refresh_internal_dir(frame, this, type);
+    } else {
+        shard_inode_ctx_mark_dir_refreshed(link_inode, this);
+        shard_common_resolve_shards(frame, this, local->post_res_handler);
+    }
+    return 0;
+unwind:
+    shard_common_resolve_shards(frame, this, local->post_res_handler);
+    return 0;
+}
+
+int
+shard_mkdir_internal_dir(call_frame_t *frame, xlator_t *this,
+                         shard_post_resolve_fop_handler_t handler,
+                         shard_internal_dir_type_t type)
+{
+    int ret = -1;
+    shard_local_t *local = NULL;
+    shard_priv_t *priv = NULL;
+    dict_t *xattr_req = NULL;
+    uuid_t *gfid = NULL;
+    loc_t *loc = NULL;
+    gf_boolean_t free_gfid = _gf_true;
+
+    local = frame->local;
+    priv = this->private;
+
+    local->post_res_handler = handler;
+    gfid = GF_MALLOC(sizeof(uuid_t), gf_common_mt_uuid_t);
+    if (!gfid)
+        goto err;
+
+    switch (type) {
+        case SHARD_INTERNAL_DIR_DOT_SHARD:
+            gf_uuid_copy(*gfid, priv->dot_shard_gfid);
+            loc = &local->dot_shard_loc;
+            break;
+        case SHARD_INTERNAL_DIR_DOT_SHARD_REMOVE_ME:
+            gf_uuid_copy(*gfid, priv->dot_shard_rm_gfid);
+            loc = &local->dot_shard_rm_loc;
+            break;
+        default:
+            bzero(*gfid, sizeof(uuid_t));
+            break;
+    }
+
+    xattr_req = dict_new();
+    if (!xattr_req)
+        goto err;
+
+    ret = shard_init_internal_dir_loc(this, local, type);
+    if (ret)
+        goto err;
+
+    ret = dict_set_gfuuid(xattr_req, "gfid-req", *gfid, false);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_DICT_OP_FAILED,
+               "Failed to set gfid-req for %s",
+               shard_internal_dir_string(type));
+        goto err;
+    } else {
+        free_gfid = _gf_false;
+    }
+
+    SHARD_SET_ROOT_FS_ID(frame, local);
+
+    STACK_WIND_COOKIE(frame, shard_mkdir_internal_dir_cbk, (void *)(long)type,
+                      FIRST_CHILD(this), FIRST_CHILD(this)->fops->mkdir, loc,
+                      0755, 0, xattr_req);
+    dict_unref(xattr_req);
+    return 0;
+
+err:
+    if (xattr_req)
+        dict_unref(xattr_req);
+    local->op_ret = -1;
+    local->op_errno = ENOMEM;
+    if (free_gfid)
+        GF_FREE(gfid);
+    handler(frame, this);
+    return 0;
+}
+
+int
+shard_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    /* To-Do: Wind flush on all shards of the file */
+    SHARD_STACK_UNWIND(flush, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+int
+shard_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    STACK_WIND(frame, shard_flush_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->flush, fd, xdata);
+    return 0;
+}
+
+int
+__shard_get_timestamps_from_inode_ctx(shard_local_t *local, inode_t *inode,
+                                      xlator_t *this)
+{
+    int ret = -1;
+    uint64_t ctx_uint = 0;
+    shard_inode_ctx_t *ctx = NULL;
+
+    ret = __inode_ctx_get(inode, this, &ctx_uint);
+    if (ret < 0)
+        return ret;
+
+    ctx = (shard_inode_ctx_t *)(uintptr_t)ctx_uint;
+
+    local->postbuf.ia_ctime = ctx->stat.ia_ctime;
+    local->postbuf.ia_ctime_nsec = ctx->stat.ia_ctime_nsec;
+    local->postbuf.ia_atime = ctx->stat.ia_atime;
+    local->postbuf.ia_atime_nsec = ctx->stat.ia_atime_nsec;
+    local->postbuf.ia_mtime = ctx->stat.ia_mtime;
+    local->postbuf.ia_mtime_nsec = ctx->stat.ia_mtime_nsec;
+
+    return 0;
+}
+
+int
+shard_get_timestamps_from_inode_ctx(shard_local_t *local, inode_t *inode,
+                                    xlator_t *this)
+{
+    int ret = 0;
+
+    LOCK(&inode->lock);
+    {
+        ret = __shard_get_timestamps_from_inode_ctx(local, inode, this);
+    }
+    UNLOCK(&inode->lock);
+
+    return ret;
+}
+
+int
+shard_fsync_shards_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                       struct iatt *postbuf, dict_t *xdata)
+{
+    int call_count = 0;
+    uint64_t fsync_count = 0;
+    fd_t *anon_fd = cookie;
+    shard_local_t *local = NULL;
+    shard_inode_ctx_t *ctx = NULL;
+    shard_inode_ctx_t *base_ictx = NULL;
+    inode_t *base_inode = NULL;
+    gf_boolean_t unref_shard_inode = _gf_false;
+
+    local = frame->local;
+    base_inode = local->fd->inode;
+
+    if (local->op_ret < 0)
+        goto out;
+
+    LOCK(&frame->lock);
+    {
+        if (op_ret < 0) {
+            local->op_ret = op_ret;
+            local->op_errno = op_errno;
+            UNLOCK(&frame->lock);
+            goto out;
+        }
+        shard_inode_ctx_set(local->fd->inode, this, postbuf, 0,
+                            SHARD_MASK_TIMES);
+    }
+    UNLOCK(&frame->lock);
+    fd_ctx_get(anon_fd, this, &fsync_count);
+out:
+    if (anon_fd && (base_inode != anon_fd->inode)) {
+        LOCK(&base_inode->lock);
+        LOCK(&anon_fd->inode->lock);
+        {
+            __shard_inode_ctx_get(anon_fd->inode, this, &ctx);
+            __shard_inode_ctx_get(base_inode, this, &base_ictx);
+            if (op_ret == 0)
+                ctx->fsync_needed -= fsync_count;
+            GF_ASSERT(ctx->fsync_needed >= 0);
+            if (ctx->fsync_needed != 0) {
+                list_add_tail(&ctx->to_fsync_list, &base_ictx->to_fsync_list);
+                base_ictx->fsync_count++;
+            } else {
+                unref_shard_inode = _gf_true;
+            }
+        }
+        UNLOCK(&anon_fd->inode->lock);
+        UNLOCK(&base_inode->lock);
+    }
+
+    if (unref_shard_inode)
+        inode_unref(anon_fd->inode);
+    if (anon_fd)
+        fd_unref(anon_fd);
+
+    call_count = shard_call_count_return(frame);
+    if (call_count != 0)
+        return 0;
+
+    if (local->op_ret < 0) {
+        shard_common_failure_unwind(GF_FOP_FSYNC, frame, local->op_ret,
+                                    local->op_errno);
+    } else {
+        shard_get_timestamps_from_inode_ctx(local, base_inode, this);
+        SHARD_STACK_UNWIND(fsync, frame, local->op_ret, local->op_errno,
+                           &local->prebuf, &local->postbuf, local->xattr_rsp);
+    }
+    return 0;
+}
+
+int
+shard_post_lookup_fsync_handler(call_frame_t *frame, xlator_t *this)
+{
+    int ret = 0;
+    int call_count = 0;
+    int fsync_count = 0;
+    fd_t *anon_fd = NULL;
+    inode_t *base_inode = NULL;
+    shard_local_t *local = NULL;
+    shard_inode_ctx_t *ctx = NULL;
+    shard_inode_ctx_t *iter = NULL;
+    struct list_head copy = {
+        0,
+    };
+    shard_inode_ctx_t *tmp = NULL;
+
+    local = frame->local;
+    base_inode = local->fd->inode;
+    local->postbuf = local->prebuf;
+    INIT_LIST_HEAD(&copy);
+
+    if (local->op_ret < 0) {
+        shard_common_failure_unwind(GF_FOP_FSYNC, frame, local->op_ret,
+                                    local->op_errno);
+        return 0;
+    }
+
+    LOCK(&base_inode->lock);
+    {
+        __shard_inode_ctx_get(base_inode, this, &ctx);
+        list_splice_init(&ctx->to_fsync_list, &copy);
+        call_count = ctx->fsync_count;
+        ctx->fsync_count = 0;
+    }
+    UNLOCK(&base_inode->lock);
+
+    local->call_count = ++call_count;
+
+    /* Send fsync() on the base shard first */
+    anon_fd = fd_ref(local->fd);
+    STACK_WIND_COOKIE(frame, shard_fsync_shards_cbk, anon_fd, FIRST_CHILD(this),
+                      FIRST_CHILD(this)->fops->fsync, anon_fd, local->datasync,
+                      local->xattr_req);
+    call_count--;
+    anon_fd = NULL;
+
+    list_for_each_entry_safe(iter, tmp, &copy, to_fsync_list)
+    {
+        list_del_init(&iter->to_fsync_list);
+        fsync_count = 0;
+        shard_inode_ctx_get_fsync_count(iter->inode, this, &fsync_count);
+        GF_ASSERT(fsync_count > 0);
+        anon_fd = fd_anonymous(iter->inode);
+        if (!anon_fd) {
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            gf_msg(this->name, GF_LOG_WARNING, ENOMEM,
+                   SHARD_MSG_MEMALLOC_FAILED,
+                   "Failed to create "
+                   "anon fd to fsync shard");
+            shard_fsync_shards_cbk(frame, (void *)(long)anon_fd, this, -1,
+                                   ENOMEM, NULL, NULL, NULL);
+            continue;
+        }
+
+        ret = fd_ctx_set(anon_fd, this, fsync_count);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_FD_CTX_SET_FAILED,
+                   "Failed to set fd "
+                   "ctx for shard inode gfid=%s",
+                   uuid_utoa(iter->inode->gfid));
+            local->op_ret = -1;
+            local->op_errno = ENOMEM;
+            shard_fsync_shards_cbk(frame, (void *)(long)anon_fd, this, -1,
+                                   ENOMEM, NULL, NULL, NULL);
+            continue;
+        }
+        STACK_WIND_COOKIE(frame, shard_fsync_shards_cbk, anon_fd,
+                          FIRST_CHILD(this), FIRST_CHILD(this)->fops->fsync,
+                          anon_fd, local->datasync, local->xattr_req);
+        call_count--;
+    }
+
+    return 0;
+}
+
+int
+shard_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+            dict_t *xdata)
+{
+    int ret = 0;
+    uint64_t block_size = 0;
+    shard_local_t *local = NULL;
+
+    ret = shard_inode_ctx_get_block_size(fd->inode, this, &block_size);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_CTX_GET_FAILED,
+               "Failed to get block "
+               "size for %s from its inode ctx",
+               uuid_utoa(fd->inode->gfid));
+        goto err;
+    }
+
+    if (!block_size || frame->root->pid == GF_CLIENT_PID_GSYNCD) {
+        STACK_WIND(frame, default_fsync_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fsync, fd, datasync, xdata);
+        return 0;
+    }
+
+    if (!this->itable)
+        this->itable = fd->inode->table;
+
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto err;
+
+    frame->local = local;
+
+    local->fd = fd_ref(fd);
+    local->fop = GF_FOP_FSYNC;
+    local->datasync = datasync;
+    local->xattr_req = (xdata) ? dict_ref(xdata) : dict_new();
+    if (!local->xattr_req)
+        goto err;
+
+    local->loc.inode = inode_ref(fd->inode);
+    gf_uuid_copy(local->loc.gfid, fd->inode->gfid);
+
+    shard_refresh_base_file(frame, this, NULL, fd,
+                            shard_post_lookup_fsync_handler);
+    return 0;
+err:
+    shard_common_failure_unwind(GF_FOP_FSYNC, frame, -1, ENOMEM);
+    return 0;
+}
+
+int
+shard_readdir_past_dot_shard_cbk(call_frame_t *frame, void *cookie,
+                                 xlator_t *this, int32_t op_ret,
+                                 int32_t op_errno, gf_dirent_t *orig_entries,
+                                 dict_t *xdata)
+{
+    gf_dirent_t *entry = NULL;
+    gf_dirent_t *tmp = NULL;
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (op_ret < 0)
+        goto unwind;
+
+    list_for_each_entry_safe(entry, tmp, (&orig_entries->list), list)
+    {
+        list_del_init(&entry->list);
+        list_add_tail(&entry->list, &local->entries_head.list);
+
+        if (!entry->dict)
+            continue;
+
+        if (IA_ISDIR(entry->d_stat.ia_type))
+            continue;
+
+        if (dict_get(entry->dict, GF_XATTR_SHARD_FILE_SIZE))
+            shard_modify_size_and_block_count(&entry->d_stat, entry->dict);
+        if (!entry->inode)
+            continue;
+
+        shard_inode_ctx_update(entry->inode, this, entry->dict, &entry->d_stat);
+    }
+    local->op_ret += op_ret;
+
+unwind:
+    if (local->fop == GF_FOP_READDIR)
+        SHARD_STACK_UNWIND(readdir, frame, local->op_ret, local->op_errno,
+                           &local->entries_head, xdata);
+    else
+        SHARD_STACK_UNWIND(readdirp, frame, op_ret, op_errno,
+                           &local->entries_head, xdata);
+    return 0;
+}
+
+int32_t
+shard_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, gf_dirent_t *orig_entries,
+                  dict_t *xdata)
+{
+    fd_t *fd = NULL;
+    gf_dirent_t *entry = NULL;
+    gf_dirent_t *tmp = NULL;
+    shard_local_t *local = NULL;
+    gf_boolean_t last_entry = _gf_false;
+
+    local = frame->local;
+    fd = local->fd;
+
+    if (op_ret < 0)
+        goto unwind;
+
+    list_for_each_entry_safe(entry, tmp, (&orig_entries->list), list)
+    {
+        if (last_entry)
+            last_entry = _gf_false;
+
+        if (__is_root_gfid(fd->inode->gfid) &&
+            !(strcmp(entry->d_name, GF_SHARD_DIR))) {
+            local->offset = entry->d_off;
+            op_ret--;
+            last_entry = _gf_true;
+            continue;
+        }
+
+        list_del_init(&entry->list);
+        list_add_tail(&entry->list, &local->entries_head.list);
+
+        if (!entry->dict)
+            continue;
+
+        if (IA_ISDIR(entry->d_stat.ia_type))
+            continue;
+
+        if (dict_get(entry->dict, GF_XATTR_SHARD_FILE_SIZE) &&
+            frame->root->pid != GF_CLIENT_PID_GSYNCD)
+            shard_modify_size_and_block_count(&entry->d_stat, entry->dict);
+
+        if (!entry->inode)
+            continue;
+
+        shard_inode_ctx_update(entry->inode, this, entry->dict, &entry->d_stat);
+    }
+
+    local->op_ret = op_ret;
+
+    if (last_entry) {
+        if (local->fop == GF_FOP_READDIR)
+            STACK_WIND(frame, shard_readdir_past_dot_shard_cbk,
+                       FIRST_CHILD(this), FIRST_CHILD(this)->fops->readdir,
+                       local->fd, local->readdir_size, local->offset,
+                       local->xattr_req);
+        else
+            STACK_WIND(frame, shard_readdir_past_dot_shard_cbk,
+                       FIRST_CHILD(this), FIRST_CHILD(this)->fops->readdirp,
+                       local->fd, local->readdir_size, local->offset,
+                       local->xattr_req);
+        return 0;
+    }
+
+unwind:
+    if (local->fop == GF_FOP_READDIR)
+        SHARD_STACK_UNWIND(readdir, frame, op_ret, op_errno,
+                           &local->entries_head, xdata);
+    else
+        SHARD_STACK_UNWIND(readdirp, frame, op_ret, op_errno,
+                           &local->entries_head, xdata);
+    return 0;
+}
+
+int
+shard_readdir_do(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                 off_t offset, int whichop, dict_t *xdata)
+{
+    int ret = 0;
+    shard_local_t *local = NULL;
+
+    local = mem_get0(this->local_pool);
+    if (!local) {
+        goto err;
+    }
+
+    frame->local = local;
+
+    local->fd = fd_ref(fd);
+    local->fop = whichop;
+    local->readdir_size = size;
+    INIT_LIST_HEAD(&local->entries_head.list);
+    local->list_inited = _gf_true;
+
+    if (whichop == GF_FOP_READDIR) {
+        STACK_WIND(frame, shard_readdir_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->readdir, fd, size, offset, xdata);
+    } else {
+        local->xattr_req = (xdata) ? dict_ref(xdata) : dict_new();
+        SHARD_MD_READ_FOP_INIT_REQ_DICT(this, local->xattr_req, fd->inode->gfid,
+                                        local, err);
+        ret = dict_set_uint64(local->xattr_req, GF_XATTR_SHARD_BLOCK_SIZE, 0);
+        if (ret) {
+            gf_log(this->name, GF_LOG_WARNING,
+                   "Failed to set "
+                   "dict value: key:%s, directory gfid=%s",
+                   GF_XATTR_SHARD_BLOCK_SIZE, uuid_utoa(fd->inode->gfid));
+            goto err;
+        }
+
+        STACK_WIND(frame, shard_readdir_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->readdirp, fd, size, offset,
+                   local->xattr_req);
+    }
+
+    return 0;
+
+err:
+    STACK_UNWIND_STRICT(readdir, frame, -1, ENOMEM, NULL, NULL);
+    return 0;
+}
+
+int32_t
+shard_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+              off_t offset, dict_t *xdata)
+{
+    shard_readdir_do(frame, this, fd, size, offset, GF_FOP_READDIR, xdata);
+    return 0;
+}
+
+int32_t
+shard_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+               off_t offset, dict_t *xdata)
+{
+    shard_readdir_do(frame, this, fd, size, offset, GF_FOP_READDIRP, xdata);
+    return 0;
+}
+
+int32_t
+shard_modify_and_set_iatt_in_dict(dict_t *xdata, shard_local_t *local,
+                                  char *key)
+{
+    int ret = 0;
+    struct iatt *tmpbuf = NULL;
+    struct iatt *stbuf = NULL;
+    data_t *data = NULL;
+
+    if (!xdata)
+        return 0;
+
+    data = dict_get(xdata, key);
+    if (!data)
+        return 0;
+
+    tmpbuf = data_to_iatt(data, key);
+    stbuf = GF_MALLOC(sizeof(struct iatt), gf_common_mt_char);
+    if (stbuf == NULL) {
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        goto err;
+    }
+    *stbuf = *tmpbuf;
+    stbuf->ia_size = local->prebuf.ia_size;
+    stbuf->ia_blocks = local->prebuf.ia_blocks;
+    ret = dict_set_iatt(xdata, key, stbuf, false);
+    if (ret < 0) {
+        local->op_ret = -1;
+        local->op_errno = ENOMEM;
+        goto err;
+    }
+    return 0;
+
+err:
+    GF_FREE(stbuf);
+    return -1;
+}
+
+int32_t
+shard_common_remove_xattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                              int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    int ret = -1;
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (op_ret < 0) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    ret = shard_modify_and_set_iatt_in_dict(xdata, local, GF_PRESTAT);
+    if (ret < 0)
+        goto err;
+
+    ret = shard_modify_and_set_iatt_in_dict(xdata, local, GF_POSTSTAT);
+    if (ret < 0)
+        goto err;
+
+    if (local->fd)
+        SHARD_STACK_UNWIND(fremovexattr, frame, local->op_ret, local->op_errno,
+                           xdata);
+    else
+        SHARD_STACK_UNWIND(removexattr, frame, local->op_ret, local->op_errno,
+                           xdata);
+    return 0;
+
+err:
+    shard_common_failure_unwind(local->fop, frame, local->op_ret,
+                                local->op_errno);
+    return 0;
+}
+
+int32_t
+shard_post_lookup_remove_xattr_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->op_ret < 0) {
+        shard_common_failure_unwind(local->fop, frame, local->op_ret,
+                                    local->op_errno);
+        return 0;
+    }
+
+    if (local->fd)
+        STACK_WIND(frame, shard_common_remove_xattr_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fremovexattr, local->fd,
+                   local->name, local->xattr_req);
+    else
+        STACK_WIND(frame, shard_common_remove_xattr_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->removexattr, &local->loc,
+                   local->name, local->xattr_req);
+    return 0;
+}
+
+int32_t
+shard_common_remove_xattr(call_frame_t *frame, xlator_t *this,
+                          glusterfs_fop_t fop, loc_t *loc, fd_t *fd,
+                          const char *name, dict_t *xdata)
+{
+    int ret = -1;
+    int op_errno = ENOMEM;
+    uint64_t block_size = 0;
+    shard_local_t *local = NULL;
+    inode_t *inode = loc ? loc->inode : fd->inode;
+
+    if ((IA_ISDIR(inode->ia_type)) || (IA_ISLNK(inode->ia_type))) {
+        if (loc)
+            STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->removexattr, loc, name,
+                            xdata);
+        else
+            STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->fremovexattr, fd, name,
+                            xdata);
+        return 0;
+    }
+
+    /* If shard's special xattrs are attempted to be removed,
+     * fail the fop with EPERM (except if the client is gsyncd).
+     */
+    if (frame->root->pid != GF_CLIENT_PID_GSYNCD) {
+        GF_IF_NATIVE_XATTR_GOTO(SHARD_XATTR_PREFIX "*", name, op_errno, err);
+    }
+
+    /* Repeat the same check for bulk-removexattr */
+    if (xdata && (frame->root->pid != GF_CLIENT_PID_GSYNCD)) {
+        dict_del(xdata, GF_XATTR_SHARD_BLOCK_SIZE);
+        dict_del(xdata, GF_XATTR_SHARD_FILE_SIZE);
+    }
+
+    ret = shard_inode_ctx_get_block_size(inode, this, &block_size);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_CTX_GET_FAILED,
+               "Failed to get block size from inode ctx of %s",
+               uuid_utoa(inode->gfid));
+        goto err;
+    }
+
+    if (!block_size || frame->root->pid == GF_CLIENT_PID_GSYNCD) {
+        if (loc)
+            STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->removexattr, loc, name,
+                            xdata);
+        else
+            STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->fremovexattr, fd, name,
+                            xdata);
+        return 0;
+    }
+
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto err;
+
+    frame->local = local;
+    local->fop = fop;
+    if (loc) {
+        if (loc_copy(&local->loc, loc) != 0)
+            goto err;
+    }
+
+    if (fd) {
+        local->fd = fd_ref(fd);
+        local->loc.inode = inode_ref(fd->inode);
+        gf_uuid_copy(local->loc.gfid, fd->inode->gfid);
+    }
+
+    if (name) {
+        local->name = gf_strdup(name);
+        if (!local->name)
+            goto err;
+    }
+
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
+    shard_refresh_base_file(frame, this, loc, fd,
+                            shard_post_lookup_remove_xattr_handler);
+    return 0;
+err:
+    shard_common_failure_unwind(fop, frame, -1, op_errno);
+    return 0;
+}
+
+int32_t
+shard_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                  const char *name, dict_t *xdata)
+{
+    shard_common_remove_xattr(frame, this, GF_FOP_REMOVEXATTR, loc, NULL, name,
+                              xdata);
+    return 0;
+}
+
+int32_t
+shard_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                   const char *name, dict_t *xdata)
+{
+    shard_common_remove_xattr(frame, this, GF_FOP_FREMOVEXATTR, NULL, fd, name,
+                              xdata);
+    return 0;
+}
+
+int32_t
+shard_fgetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *dict,
+                    dict_t *xdata)
+{
+    if (op_ret < 0)
+        goto unwind;
+
+    if (dict && (frame->root->pid != GF_CLIENT_PID_GSYNCD)) {
+        dict_del(dict, GF_XATTR_SHARD_BLOCK_SIZE);
+        dict_del(dict, GF_XATTR_SHARD_FILE_SIZE);
+    }
+
+unwind:
+    SHARD_STACK_UNWIND(fgetxattr, frame, op_ret, op_errno, dict, xdata);
+    return 0;
+}
+
+int32_t
+shard_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+                dict_t *xdata)
+{
+    int op_errno = EINVAL;
+
+    if ((frame->root->pid != GF_CLIENT_PID_GSYNCD) && (name) &&
+        (!strncmp(name, SHARD_XATTR_PREFIX, SLEN(SHARD_XATTR_PREFIX)))) {
+        op_errno = ENODATA;
+        goto out;
+    }
+
+    STACK_WIND(frame, shard_fgetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fgetxattr, fd, name, xdata);
+    return 0;
+out:
+    shard_common_failure_unwind(GF_FOP_FGETXATTR, frame, -1, op_errno);
+    return 0;
+}
+
+int32_t
+shard_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *dict,
+                   dict_t *xdata)
+{
+    if (op_ret < 0)
+        goto unwind;
+
+    if (dict && (frame->root->pid != GF_CLIENT_PID_GSYNCD)) {
+        dict_del(dict, GF_XATTR_SHARD_BLOCK_SIZE);
+        dict_del(dict, GF_XATTR_SHARD_FILE_SIZE);
+    }
+
+unwind:
+    SHARD_STACK_UNWIND(getxattr, frame, op_ret, op_errno, dict, xdata);
+    return 0;
+}
+
+int32_t
+shard_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               const char *name, dict_t *xdata)
+{
+    int op_errno = EINVAL;
+
+    if ((frame->root->pid != GF_CLIENT_PID_GSYNCD) && (name) &&
+        (!strncmp(name, SHARD_XATTR_PREFIX, sizeof(SHARD_XATTR_PREFIX) - 1))) {
+        op_errno = ENODATA;
+        goto out;
+    }
+
+    STACK_WIND(frame, shard_getxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->getxattr, loc, name, xdata);
+    return 0;
+out:
+    shard_common_failure_unwind(GF_FOP_GETXATTR, frame, -1, op_errno);
+    return 0;
+}
+
+int32_t
+shard_common_set_xattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    int ret = -1;
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (op_ret < 0) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        goto err;
+    }
+
+    ret = shard_modify_and_set_iatt_in_dict(xdata, local, GF_PRESTAT);
+    if (ret < 0)
+        goto err;
+
+    ret = shard_modify_and_set_iatt_in_dict(xdata, local, GF_POSTSTAT);
+    if (ret < 0)
+        goto err;
+
+    if (local->fd)
+        SHARD_STACK_UNWIND(fsetxattr, frame, local->op_ret, local->op_errno,
+                           xdata);
+    else
+        SHARD_STACK_UNWIND(setxattr, frame, local->op_ret, local->op_errno,
+                           xdata);
+    return 0;
+
+err:
+    shard_common_failure_unwind(local->fop, frame, local->op_ret,
+                                local->op_errno);
+    return 0;
+}
+
+int32_t
+shard_post_lookup_set_xattr_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->op_ret < 0) {
+        shard_common_failure_unwind(local->fop, frame, local->op_ret,
+                                    local->op_errno);
+        return 0;
+    }
+
+    if (local->fd)
+        STACK_WIND(frame, shard_common_set_xattr_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fsetxattr, local->fd,
+                   local->xattr_req, local->flags, local->xattr_rsp);
+    else
+        STACK_WIND(frame, shard_common_set_xattr_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->setxattr, &local->loc,
+                   local->xattr_req, local->flags, local->xattr_rsp);
+    return 0;
+}
+
+int32_t
+shard_common_set_xattr(call_frame_t *frame, xlator_t *this, glusterfs_fop_t fop,
+                       loc_t *loc, fd_t *fd, dict_t *dict, int32_t flags,
+                       dict_t *xdata)
+{
+    int ret = -1;
+    int op_errno = ENOMEM;
+    uint64_t block_size = 0;
+    shard_local_t *local = NULL;
+    inode_t *inode = loc ? loc->inode : fd->inode;
+
+    if ((IA_ISDIR(inode->ia_type)) || (IA_ISLNK(inode->ia_type))) {
+        if (loc)
+            STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->setxattr, loc, dict, flags,
+                            xdata);
+        else
+            STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags,
+                            xdata);
+        return 0;
+    }
+
+    /* Sharded or not, if shard's special xattrs are attempted to be set,
+     * fail the fop with EPERM (except if the client is gsyncd.
+     */
+    if (frame->root->pid != GF_CLIENT_PID_GSYNCD) {
+        GF_IF_INTERNAL_XATTR_GOTO(SHARD_XATTR_PREFIX "*", dict, op_errno, err);
+    }
+
+    ret = shard_inode_ctx_get_block_size(inode, this, &block_size);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_CTX_GET_FAILED,
+               "Failed to get block size from inode ctx of %s",
+               uuid_utoa(inode->gfid));
+        goto err;
+    }
+
+    if (!block_size || frame->root->pid == GF_CLIENT_PID_GSYNCD) {
+        if (loc)
+            STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->setxattr, loc, dict, flags,
+                            xdata);
+        else
+            STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                            FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags,
+                            xdata);
+        return 0;
+    }
+
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto err;
+
+    frame->local = local;
+    local->fop = fop;
+    if (loc) {
+        if (loc_copy(&local->loc, loc) != 0)
+            goto err;
+    }
+
+    if (fd) {
+        local->fd = fd_ref(fd);
+        local->loc.inode = inode_ref(fd->inode);
+        gf_uuid_copy(local->loc.gfid, fd->inode->gfid);
+    }
+    local->flags = flags;
+    /* Reusing local->xattr_req and local->xattr_rsp to store the setxattr dict
+     * and the xdata dict
+     */
+    if (dict)
+        local->xattr_req = dict_ref(dict);
+    if (xdata)
+        local->xattr_rsp = dict_ref(xdata);
+
+    shard_refresh_base_file(frame, this, loc, fd,
+                            shard_post_lookup_set_xattr_handler);
+    return 0;
+err:
+    shard_common_failure_unwind(fop, frame, -1, op_errno);
+    return 0;
+}
+
+int32_t
+shard_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+                int32_t flags, dict_t *xdata)
+{
+    shard_common_set_xattr(frame, this, GF_FOP_FSETXATTR, NULL, fd, dict, flags,
+                           xdata);
+    return 0;
+}
+
+int32_t
+shard_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+               int32_t flags, dict_t *xdata)
+{
+    shard_common_set_xattr(frame, this, GF_FOP_SETXATTR, loc, NULL, dict, flags,
+                           xdata);
+    return 0;
+}
+
+int
+shard_post_setattr_handler(call_frame_t *frame, xlator_t *this)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (local->fop == GF_FOP_SETATTR) {
+        if (local->op_ret >= 0)
+            shard_inode_ctx_set(local->loc.inode, this, &local->postbuf, 0,
+                                SHARD_LOOKUP_MASK);
+        SHARD_STACK_UNWIND(setattr, frame, local->op_ret, local->op_errno,
+                           &local->prebuf, &local->postbuf, local->xattr_rsp);
+    } else if (local->fop == GF_FOP_FSETATTR) {
+        if (local->op_ret >= 0)
+            shard_inode_ctx_set(local->fd->inode, this, &local->postbuf, 0,
+                                SHARD_LOOKUP_MASK);
+        SHARD_STACK_UNWIND(fsetattr, frame, local->op_ret, local->op_errno,
+                           &local->prebuf, &local->postbuf, local->xattr_rsp);
+    }
+
+    return 0;
+}
+
+int
+shard_common_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                         struct iatt *postbuf, dict_t *xdata)
+{
+    shard_local_t *local = NULL;
+
+    local = frame->local;
+
+    if (op_ret < 0) {
+        local->op_ret = op_ret;
+        local->op_errno = op_errno;
+        goto unwind;
+    }
+
+    local->prebuf = *prebuf;
+    if (shard_modify_size_and_block_count(&local->prebuf, xdata)) {
+        local->op_ret = -1;
+        local->op_errno = EINVAL;
+        goto unwind;
+    }
+    if (xdata)
+        local->xattr_rsp = dict_ref(xdata);
+    local->postbuf = *postbuf;
+    local->postbuf.ia_size = local->prebuf.ia_size;
+    local->postbuf.ia_blocks = local->prebuf.ia_blocks;
+
+unwind:
+    local->handler(frame, this);
+    return 0;
+}
+
+int
+shard_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+              struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    int ret = -1;
+    uint64_t block_size = 0;
+    shard_local_t *local = NULL;
+
+    if ((IA_ISDIR(loc->inode->ia_type)) || (IA_ISLNK(loc->inode->ia_type))) {
+        STACK_WIND(frame, default_setattr_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->setattr, loc, stbuf, valid, xdata);
+        return 0;
+    }
+
+    ret = shard_inode_ctx_get_block_size(loc->inode, this, &block_size);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_CTX_GET_FAILED,
+               "Failed to get block size from inode ctx of %s",
+               uuid_utoa(loc->inode->gfid));
+        goto err;
+    }
+
+    if (!block_size || frame->root->pid == GF_CLIENT_PID_GSYNCD) {
+        STACK_WIND(frame, default_setattr_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->setattr, loc, stbuf, valid, xdata);
+        return 0;
+    }
+
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto err;
+
+    frame->local = local;
+
+    local->handler = shard_post_setattr_handler;
+    local->xattr_req = (xdata) ? dict_ref(xdata) : dict_new();
+    if (!local->xattr_req)
+        goto err;
+    local->fop = GF_FOP_SETATTR;
+    loc_copy(&local->loc, loc);
+
+    SHARD_MD_READ_FOP_INIT_REQ_DICT(this, local->xattr_req, local->loc.gfid,
+                                    local, err);
+
+    STACK_WIND(frame, shard_common_setattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setattr, loc, stbuf, valid,
+               local->xattr_req);
+    return 0;
+err:
+    shard_common_failure_unwind(GF_FOP_SETATTR, frame, -1, ENOMEM);
+    return 0;
+}
+
+int
+shard_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+               struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    int ret = -1;
+    uint64_t block_size = 0;
+    shard_local_t *local = NULL;
+
+    if ((IA_ISDIR(fd->inode->ia_type)) || (IA_ISLNK(fd->inode->ia_type))) {
+        STACK_WIND(frame, default_fsetattr_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fsetattr, fd, stbuf, valid, xdata);
+        return 0;
+    }
+
+    ret = shard_inode_ctx_get_block_size(fd->inode, this, &block_size);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_CTX_GET_FAILED,
+               "Failed to get block size from inode ctx of %s",
+               uuid_utoa(fd->inode->gfid));
+        goto err;
+    }
+
+    if (!block_size || frame->root->pid == GF_CLIENT_PID_GSYNCD) {
+        STACK_WIND(frame, default_fsetattr_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fsetattr, fd, stbuf, valid, xdata);
+        return 0;
+    }
+
+    if (!this->itable)
+        this->itable = fd->inode->table;
+
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto err;
+
+    frame->local = local;
+
+    local->handler = shard_post_setattr_handler;
+    local->xattr_req = (xdata) ? dict_ref(xdata) : dict_new();
+    if (!local->xattr_req)
+        goto err;
+    local->fop = GF_FOP_FSETATTR;
+    local->fd = fd_ref(fd);
+
+    SHARD_MD_READ_FOP_INIT_REQ_DICT(this, local->xattr_req, fd->inode->gfid,
+                                    local, err);
+
+    STACK_WIND(frame, shard_common_setattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetattr, fd, stbuf, valid,
+               local->xattr_req);
+    return 0;
+err:
+    shard_common_failure_unwind(GF_FOP_FSETATTR, frame, -1, ENOMEM);
+    return 0;
+}
+
+int
+shard_common_inode_write_begin(call_frame_t *frame, xlator_t *this,
+                               glusterfs_fop_t fop, fd_t *fd,
+                               struct iovec *vector, int32_t count,
+                               off_t offset, uint32_t flags, size_t len,
+                               struct iobref *iobref, dict_t *xdata)
+{
+    int ret = 0;
+    int i = 0;
+    uint64_t block_size = 0;
+    shard_local_t *local = NULL;
+
+    ret = shard_inode_ctx_get_block_size(fd->inode, this, &block_size);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INODE_CTX_GET_FAILED,
+               "Failed to get block "
+               "size for %s from its inode ctx",
+               uuid_utoa(fd->inode->gfid));
+        goto out;
+    }
+
+    if (!block_size || frame->root->pid == GF_CLIENT_PID_GSYNCD) {
+        /* block_size = 0 means that the file was created before
+         * sharding was enabled on the volume.
+         */
+        switch (fop) {
+            case GF_FOP_WRITE:
+                STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                                FIRST_CHILD(this)->fops->writev, fd, vector,
+                                count, offset, flags, iobref, xdata);
+                break;
+            case GF_FOP_FALLOCATE:
+                STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                                FIRST_CHILD(this)->fops->fallocate, fd, flags,
+                                offset, len, xdata);
+                break;
+            case GF_FOP_ZEROFILL:
+                STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                                FIRST_CHILD(this)->fops->zerofill, fd, offset,
+                                len, xdata);
+                break;
+            case GF_FOP_DISCARD:
+                STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                                FIRST_CHILD(this)->fops->discard, fd, offset,
+                                len, xdata);
+                break;
+            default:
+                gf_msg(this->name, GF_LOG_WARNING, 0, SHARD_MSG_INVALID_FOP,
+                       "Invalid fop id = %d", fop);
+                break;
+        }
+        return 0;
+    }
+
+    if (!this->itable)
+        this->itable = fd->inode->table;
+
+    local = mem_get0(this->local_pool);
+    if (!local)
+        goto out;
+
+    frame->local = local;
+
+    ret = syncbarrier_init(&local->barrier);
+    if (ret)
+        goto out;
+    local->xattr_req = (xdata) ? dict_ref(xdata) : dict_new();
+    if (!local->xattr_req)
+        goto out;
+
+    if (vector) {
+        local->vector = iov_dup(vector, count);
+        if (!local->vector)
+            goto out;
+        for (i = 0; i < count; i++)
+            local->total_size += vector[i].iov_len;
+        local->count = count;
+    } else {
+        local->total_size = len;
+    }
+
+    local->fop = fop;
+    local->offset = offset;
+    local->flags = flags;
+    if (iobref)
+        local->iobref = iobref_ref(iobref);
+    local->fd = fd_ref(fd);
+    local->block_size = block_size;
+    local->resolver_base_inode = local->fd->inode;
+    GF_ATOMIC_INIT(local->delta_blocks, 0);
+
+    local->loc.inode = inode_ref(fd->inode);
+    gf_uuid_copy(local->loc.gfid, fd->inode->gfid);
+
+    shard_refresh_base_file(frame, this, NULL, fd,
+                            shard_common_inode_write_post_lookup_handler);
+    return 0;
+out:
+    shard_common_failure_unwind(fop, frame, -1, ENOMEM);
+    return 0;
+}
+
+int
+shard_writev(call_frame_t *frame, xlator_t *this, fd_t *fd,
+             struct iovec *vector, int32_t count, off_t offset, uint32_t flags,
+             struct iobref *iobref, dict_t *xdata)
+{
+    shard_common_inode_write_begin(frame, this, GF_FOP_WRITE, fd, vector, count,
+                                   offset, flags, 0, iobref, xdata);
+    return 0;
+}
+
+int
+shard_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                int32_t keep_size, off_t offset, size_t len, dict_t *xdata)
+{
+    if ((keep_size != 0) && (keep_size != FALLOC_FL_ZERO_RANGE) &&
+        (keep_size != (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)))
+        goto out;
+
+    shard_common_inode_write_begin(frame, this, GF_FOP_FALLOCATE, fd, NULL, 0,
+                                   offset, keep_size, len, NULL, xdata);
+    return 0;
+out:
+    shard_common_failure_unwind(GF_FOP_FALLOCATE, frame, -1, ENOTSUP);
+    return 0;
+}
+
+int
+shard_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+               off_t len, dict_t *xdata)
+{
+    shard_common_inode_write_begin(frame, this, GF_FOP_ZEROFILL, fd, NULL, 0,
+                                   offset, 0, len, NULL, xdata);
+    return 0;
+}
+
+int
+shard_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+              size_t len, dict_t *xdata)
+{
+    shard_common_inode_write_begin(frame, this, GF_FOP_DISCARD, fd, NULL, 0,
+                                   offset, 0, len, NULL, xdata);
+    return 0;
+}
+
+int32_t
+shard_seek(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+           gf_seek_what_t what, dict_t *xdata)
+{
+    /* TBD */
+    gf_msg(this->name, GF_LOG_INFO, ENOTSUP, SHARD_MSG_FOP_NOT_SUPPORTED,
+           "seek called on %s.", uuid_utoa(fd->inode->gfid));
+    shard_common_failure_unwind(GF_FOP_SEEK, frame, -1, ENOTSUP);
+    return 0;
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    if (!this)
+        return ret;
+
+    ret = xlator_mem_acct_init(this, gf_shard_mt_end + 1);
+
+    if (ret != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_MEM_ACCT_INIT_FAILED,
+               "Memory accounting init"
+               "failed");
+        return ret;
+    }
+
+    return ret;
+}
+
+int
+init(xlator_t *this)
+{
+    int ret = -1;
+    shard_priv_t *priv = NULL;
+
+    if (!this) {
+        gf_msg("shard", GF_LOG_ERROR, 0, SHARD_MSG_NULL_THIS,
+               "this is NULL. init() failed");
+        return -1;
+    }
+
+    if (!this->parents) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INVALID_VOLFILE,
+               "Dangling volume. Check volfile");
+        goto out;
+    }
+
+    if (!this->children || this->children->next) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SHARD_MSG_INVALID_VOLFILE,
+               "shard not configured with exactly one sub-volume. "
+               "Check volfile");
+        goto out;
+    }
+
+    priv = GF_CALLOC(1, sizeof(shard_priv_t), gf_shard_mt_priv_t);
+    if (!priv)
+        goto out;
+
+    GF_OPTION_INIT("shard-block-size", priv->block_size, size_uint64, out);
+
+    GF_OPTION_INIT("shard-deletion-rate", priv->deletion_rate, uint32, out);
+
+    GF_OPTION_INIT("shard-lru-limit", priv->lru_limit, uint64, out);
+
+    this->local_pool = mem_pool_new(shard_local_t, 128);
+    if (!this->local_pool) {
+        ret = -1;
+        goto out;
+    }
+    gf_uuid_parse(SHARD_ROOT_GFID, priv->dot_shard_gfid);
+    gf_uuid_parse(DOT_SHARD_REMOVE_ME_GFID, priv->dot_shard_rm_gfid);
+
+    this->private = priv;
+    LOCK_INIT(&priv->lock);
+    INIT_LIST_HEAD(&priv->ilist_head);
+    ret = 0;
+out:
+    if (ret) {
+        GF_FREE(priv);
+        mem_pool_destroy(this->local_pool);
+    }
+
+    return ret;
+}
+
+void
+fini(xlator_t *this)
+{
+    shard_priv_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("shard", this, out);
+
+    /*Itable was not created by shard, hence setting to NULL.*/
+    this->itable = NULL;
+
+    mem_pool_destroy(this->local_pool);
+    this->local_pool = NULL;
+
+    priv = this->private;
+    if (!priv)
+        goto out;
+
+    this->private = NULL;
+    LOCK_DESTROY(&priv->lock);
+    GF_FREE(priv);
+
+out:
+    return;
+}
+
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    int ret = -1;
+    shard_priv_t *priv = NULL;
+
+    priv = this->private;
+
+    GF_OPTION_RECONF("shard-block-size", priv->block_size, options, size, out);
+
+    GF_OPTION_RECONF("shard-deletion-rate", priv->deletion_rate, options,
+                     uint32, out);
+    ret = 0;
+
+out:
+    return ret;
+}
+
+int
+shard_forget(xlator_t *this, inode_t *inode)
+{
+    uint64_t ctx_uint = 0;
+    shard_inode_ctx_t *ctx = NULL;
+    shard_priv_t *priv = NULL;
+
+    priv = this->private;
+    if (!priv)
+        return 0;
+
+    inode_ctx_del(inode, this, &ctx_uint);
+    if (!ctx_uint)
+        return 0;
+
+    ctx = (shard_inode_ctx_t *)(uintptr_t)ctx_uint;
+
+    /* When LRU limit reaches inode will be forcefully removed from the
+     * table, inode needs to be removed from LRU of shard as well.
+     */
+    if (!list_empty(&ctx->ilist)) {
+        LOCK(&priv->lock);
+        {
+            list_del_init(&ctx->ilist);
+            priv->inode_count--;
+        }
+        UNLOCK(&priv->lock);
+    }
+    GF_FREE(ctx);
+
+    return 0;
+}
+
+int
+shard_release(xlator_t *this, fd_t *fd)
+{
+    /* TBD */
+    return 0;
+}
+
+int
+shard_priv_dump(xlator_t *this)
+{
+    shard_priv_t *priv = NULL;
+    char key_prefix[GF_DUMP_MAX_BUF_LEN] = {
+        0,
+    };
+    char *str = NULL;
+
+    priv = this->private;
+
+    snprintf(key_prefix, GF_DUMP_MAX_BUF_LEN, "%s.%s", this->type, this->name);
+    gf_proc_dump_add_section("%s", key_prefix);
+    str = gf_uint64_2human_readable(priv->block_size);
+    gf_proc_dump_write("shard-block-size", "%s", str);
+    gf_proc_dump_write("inode-count", "%d", priv->inode_count);
+    gf_proc_dump_write("ilist_head", "%p", &priv->ilist_head);
+    gf_proc_dump_write("lru-max-limit", "%" PRIu64, priv->lru_limit);
+
+    GF_FREE(str);
+
+    return 0;
+}
+
+int
+shard_releasedir(xlator_t *this, fd_t *fd)
+{
+    return 0;
+}
+
+struct xlator_fops fops = {
+    .lookup = shard_lookup,
+    .open = shard_open,
+    .flush = shard_flush,
+    .fsync = shard_fsync,
+    .stat = shard_stat,
+    .fstat = shard_fstat,
+    .getxattr = shard_getxattr,
+    .fgetxattr = shard_fgetxattr,
+    .readv = shard_readv,
+    .writev = shard_writev,
+    .truncate = shard_truncate,
+    .ftruncate = shard_ftruncate,
+    .setxattr = shard_setxattr,
+    .fsetxattr = shard_fsetxattr,
+    .setattr = shard_setattr,
+    .fsetattr = shard_fsetattr,
+    .removexattr = shard_removexattr,
+    .fremovexattr = shard_fremovexattr,
+    .fallocate = shard_fallocate,
+    .discard = shard_discard,
+    .zerofill = shard_zerofill,
+    .readdir = shard_readdir,
+    .readdirp = shard_readdirp,
+    .create = shard_create,
+    .mknod = shard_mknod,
+    .link = shard_link,
+    .unlink = shard_unlink,
+    .rename = shard_rename,
+    .seek = shard_seek,
+};
+
+struct xlator_cbks cbks = {
+    .forget = shard_forget,
+    .release = shard_release,
+    .releasedir = shard_releasedir,
+};
+
+struct xlator_dumpops dumpops = {
+    .priv = shard_priv_dump,
+};
+
+struct volume_options options[] = {
+    {
+        .key = {"shard"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "off",
+        .description = "enable/disable shard",
+        .op_version = {GD_OP_VERSION_6_0},
+        .flags = OPT_FLAG_SETTABLE,
+    },
+    {
+        .key = {"shard-block-size"},
+        .type = GF_OPTION_TYPE_SIZET,
+        .op_version = {GD_OP_VERSION_3_7_0},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+        .tags = {"shard"},
+        .default_value = "64MB",
+        .min = SHARD_MIN_BLOCK_SIZE,
+        .max = SHARD_MAX_BLOCK_SIZE,
+        .description = "The size unit used to break a file into multiple "
+                       "chunks",
+    },
+    {
+        .key = {"shard-deletion-rate"},
+        .type = GF_OPTION_TYPE_INT,
+        .op_version = {GD_OP_VERSION_5_0},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+        .tags = {"shard"},
+        .default_value = "100",
+        .min = 100,
+        .max = INT_MAX,
+        .description = "The number of shards to send deletes on at a time",
+    },
+    {
+        .key = {"shard-lru-limit"},
+        .type = GF_OPTION_TYPE_INT,
+        .op_version = {GD_OP_VERSION_5_0},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT,
+        .tags = {"shard"},
+        .default_value = "16384",
+        .min = 20,
+        .max = INT_MAX,
+        .description = "The number of resolved shard inodes to keep in "
+                       "memory. A higher number means shards that are "
+                       "resolved will remain in memory longer, avoiding "
+                       "frequent lookups on them when they participate in "
+                       "file operations. The option also has a bearing on "
+                       "amount of memory consumed by these inodes and their "
+                       "internal metadata",
+    },
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .dumpops = &dumpops,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "shard",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/features/shard/src/shard.h b/xlators/features/shard/src/shard.h
new file mode 100644
index 00000000000..4fe181b64d5
--- /dev/null
+++ b/xlators/features/shard/src/shard.h
@@ -0,0 +1,348 @@
+/*
+  Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __SHARD_H__
+#define __SHARD_H__
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/compat-errno.h>
+#include "shard-messages.h"
+#include <glusterfs/syncop.h>
+
+#define GF_SHARD_DIR ".shard"
+#define GF_SHARD_REMOVE_ME_DIR ".remove_me"
+#define SHARD_MIN_BLOCK_SIZE (4 * GF_UNIT_MB)
+#define SHARD_MAX_BLOCK_SIZE (4 * GF_UNIT_TB)
+#define SHARD_XATTR_PREFIX "trusted.glusterfs.shard."
+#define GF_XATTR_SHARD_BLOCK_SIZE "trusted.glusterfs.shard.block-size"
+/**
+ *  Bit masks for the valid flag, which is used while updating ctx
+ **/
+#define SHARD_MASK_BLOCK_SIZE (1 << 0)
+#define SHARD_MASK_PROT (1 << 1)
+#define SHARD_MASK_NLINK (1 << 2)
+#define SHARD_MASK_UID (1 << 3)
+#define SHARD_MASK_GID (1 << 4)
+#define SHARD_MASK_SIZE (1 << 6)
+#define SHARD_MASK_BLOCKS (1 << 7)
+#define SHARD_MASK_TIMES (1 << 8)
+#define SHARD_MASK_OTHERS (1 << 9)
+#define SHARD_MASK_REFRESH_RESET (1 << 10)
+
+#define SHARD_INODE_WRITE_MASK                                                 \
+    (SHARD_MASK_SIZE | SHARD_MASK_BLOCKS | SHARD_MASK_TIMES)
+
+#define SHARD_LOOKUP_MASK                                                      \
+    (SHARD_MASK_PROT | SHARD_MASK_NLINK | SHARD_MASK_UID | SHARD_MASK_GID |    \
+     SHARD_MASK_TIMES | SHARD_MASK_OTHERS)
+
+#define SHARD_ALL_MASK                                                         \
+    (SHARD_MASK_BLOCK_SIZE | SHARD_MASK_PROT | SHARD_MASK_NLINK |              \
+     SHARD_MASK_UID | SHARD_MASK_GID | SHARD_MASK_SIZE | SHARD_MASK_BLOCKS |   \
+     SHARD_MASK_TIMES | SHARD_MASK_OTHERS)
+
+#define get_lowest_block(off, shard_size) ((off) / (shard_size))
+#define get_highest_block(off, len, shard_size)                                \
+    (((((off) + (len)) == 0) ? 0 : ((off) + (len)-1)) / (shard_size))
+
+int
+shard_unlock_inodelk(call_frame_t *frame, xlator_t *this);
+
+int
+shard_unlock_entrylk(call_frame_t *frame, xlator_t *this);
+
+#define SHARD_ENTRY_FOP_CHECK(loc, op_errno, label)                            \
+    do {                                                                       \
+        if ((loc->name && !strcmp(GF_SHARD_DIR, loc->name)) &&                 \
+            (((loc->parent) && __is_root_gfid(loc->parent->gfid)) ||           \
+             __is_root_gfid(loc->pargfid))) {                                  \
+            op_errno = EPERM;                                                  \
+            goto label;                                                        \
+        }                                                                      \
+                                                                               \
+        if ((loc->parent && __is_shard_dir(loc->parent->gfid)) ||              \
+            __is_shard_dir(loc->pargfid)) {                                    \
+            op_errno = EPERM;                                                  \
+            goto label;                                                        \
+        }                                                                      \
+    } while (0)
+
+#define SHARD_INODE_OP_CHECK(gfid, err, label)                                 \
+    do {                                                                       \
+        if (__is_shard_dir(gfid)) {                                            \
+            err = EPERM;                                                       \
+            goto label;                                                        \
+        }                                                                      \
+    } while (0)
+
+#define SHARD_STACK_UNWIND(fop, frame, params...)                              \
+    do {                                                                       \
+        shard_local_t *__local = NULL;                                         \
+        if (frame) {                                                           \
+            __local = frame->local;                                            \
+            if (__local && __local->int_inodelk.acquired_lock)                 \
+                shard_unlock_inodelk(frame, frame->this);                      \
+            if (__local && __local->int_entrylk.acquired_lock)                 \
+                shard_unlock_entrylk(frame, frame->this);                      \
+            frame->local = NULL;                                               \
+        }                                                                      \
+        STACK_UNWIND_STRICT(fop, frame, params);                               \
+        if (__local) {                                                         \
+            shard_local_wipe(__local);                                         \
+            mem_put(__local);                                                  \
+        }                                                                      \
+    } while (0)
+
+#define SHARD_STACK_DESTROY(frame)                                             \
+    do {                                                                       \
+        shard_local_t *__local = NULL;                                         \
+        __local = frame->local;                                                \
+        frame->local = NULL;                                                   \
+        STACK_DESTROY(frame->root);                                            \
+        if (__local) {                                                         \
+            shard_local_wipe(__local);                                         \
+            mem_put(__local);                                                  \
+        }                                                                      \
+    } while (0);
+
+#define SHARD_INODE_CREATE_INIT(this, block_size, xattr_req, loc, size,        \
+                                block_count, label)                            \
+    do {                                                                       \
+        int __ret = -1;                                                        \
+        int64_t *__size_attr = NULL;                                           \
+        uint64_t *__bs = 0;                                                    \
+                                                                               \
+        __bs = GF_MALLOC(sizeof(uint64_t), gf_shard_mt_uint64_t);              \
+        if (!__bs)                                                             \
+            goto label;                                                        \
+        *__bs = hton64(block_size);                                            \
+        __ret = dict_set_bin(xattr_req, GF_XATTR_SHARD_BLOCK_SIZE, __bs,       \
+                             sizeof(*__bs));                                   \
+        if (__ret) {                                                           \
+            gf_msg(this->name, GF_LOG_WARNING, 0, SHARD_MSG_DICT_OP_FAILED,    \
+                   "Failed to set key: %s "                                    \
+                   "on path %s",                                               \
+                   GF_XATTR_SHARD_BLOCK_SIZE, (loc)->path);                    \
+            GF_FREE(__bs);                                                     \
+            goto label;                                                        \
+        }                                                                      \
+                                                                               \
+        __ret = shard_set_size_attrs(size, block_count, &__size_attr);         \
+        if (__ret)                                                             \
+            goto label;                                                        \
+                                                                               \
+        __ret = dict_set_bin(xattr_req, GF_XATTR_SHARD_FILE_SIZE, __size_attr, \
+                             8 * 4);                                           \
+        if (__ret) {                                                           \
+            gf_msg(this->name, GF_LOG_WARNING, 0, SHARD_MSG_DICT_OP_FAILED,    \
+                   "Failed to set key: %s "                                    \
+                   "on path %s",                                               \
+                   GF_XATTR_SHARD_FILE_SIZE, (loc)->path);                     \
+            GF_FREE(__size_attr);                                              \
+            goto label;                                                        \
+        }                                                                      \
+    } while (0)
+
+#define SHARD_MD_READ_FOP_INIT_REQ_DICT(this, dict, gfid, local, label)        \
+    do {                                                                       \
+        int __ret = -1;                                                        \
+                                                                               \
+        __ret = dict_set_uint64(dict, GF_XATTR_SHARD_FILE_SIZE, 8 * 4);        \
+        if (__ret) {                                                           \
+            local->op_ret = -1;                                                \
+            local->op_errno = ENOMEM;                                          \
+            gf_msg(this->name, GF_LOG_WARNING, 0, SHARD_MSG_DICT_OP_FAILED,    \
+                   "Failed to set dict value:"                                 \
+                   " key:%s for %s.",                                          \
+                   GF_XATTR_SHARD_FILE_SIZE, uuid_utoa(gfid));                 \
+            goto label;                                                        \
+        }                                                                      \
+    } while (0)
+
+#define SHARD_SET_ROOT_FS_ID(frame, local)                                     \
+    do {                                                                       \
+        if (!local->is_set_fsid) {                                             \
+            local->uid = frame->root->uid;                                     \
+            local->gid = frame->root->gid;                                     \
+            frame->root->uid = 0;                                              \
+            frame->root->gid = 0;                                              \
+            local->is_set_fsid = _gf_true;                                     \
+        }                                                                      \
+    } while (0)
+
+#define SHARD_UNSET_ROOT_FS_ID(frame, local)                                   \
+    do {                                                                       \
+        if (local->is_set_fsid) {                                              \
+            frame->root->uid = local->uid;                                     \
+            frame->root->gid = local->gid;                                     \
+            local->is_set_fsid = _gf_false;                                    \
+        }                                                                      \
+    } while (0)
+
+#define SHARD_TIME_UPDATE(ctx_sec, ctx_nsec, new_sec, new_nsec)                \
+    do {                                                                       \
+        if (ctx_sec == new_sec)                                                \
+            ctx_nsec = new_nsec = max(new_nsec, ctx_nsec);                     \
+        else if (ctx_sec > new_sec) {                                          \
+            new_sec = ctx_sec;                                                 \
+            new_nsec = ctx_nsec;                                               \
+        } else {                                                               \
+            ctx_sec = new_sec;                                                 \
+            ctx_nsec = new_nsec;                                               \
+        }                                                                      \
+    } while (0)
+
+typedef enum {
+    SHARD_BG_DELETION_NONE = 0,
+    SHARD_BG_DELETION_LAUNCHING,
+    SHARD_BG_DELETION_IN_PROGRESS,
+} shard_bg_deletion_state_t;
+
+/* rm = "remove me" */
+
+typedef struct shard_priv {
+    uint64_t block_size;
+    uuid_t dot_shard_gfid;
+    uuid_t dot_shard_rm_gfid;
+    inode_t *dot_shard_inode;
+    inode_t *dot_shard_rm_inode;
+    gf_lock_t lock;
+    int inode_count;
+    struct list_head ilist_head;
+    uint32_t deletion_rate;
+    shard_bg_deletion_state_t bg_del_state;
+    gf_boolean_t first_lookup_done;
+    uint64_t lru_limit;
+} shard_priv_t;
+
+typedef struct {
+    loc_t loc;
+    char *domain;
+    struct gf_flock flock;
+    gf_boolean_t acquired_lock;
+} shard_inodelk_t;
+
+typedef struct {
+    loc_t loc;
+    char *domain;
+    char *basename;
+    entrylk_cmd cmd;
+    entrylk_type type;
+    gf_boolean_t acquired_lock;
+} shard_entrylk_t;
+
+typedef int32_t (*shard_post_fop_handler_t)(call_frame_t *frame,
+                                            xlator_t *this);
+typedef int32_t (*shard_post_resolve_fop_handler_t)(call_frame_t *frame,
+                                                    xlator_t *this);
+typedef int32_t (*shard_post_lookup_shards_fop_handler_t)(call_frame_t *frame,
+                                                          xlator_t *this);
+
+typedef int32_t (*shard_post_mknod_fop_handler_t)(call_frame_t *frame,
+                                                  xlator_t *this);
+
+typedef int32_t (*shard_post_update_size_fop_handler_t)(call_frame_t *frame,
+                                                        xlator_t *this);
+
+typedef struct shard_local {
+    int op_ret;
+    int op_errno;
+    uint64_t first_block;
+    uint64_t last_block;
+    uint64_t num_blocks;
+    int call_count;
+    int eexist_count;
+    int create_count;
+    int xflag;
+    int count;
+    uint32_t flags;
+    uint32_t uid;
+    uint32_t gid;
+    uint64_t block_size;
+    uint64_t dst_block_size;
+    int32_t datasync;
+    off_t offset;
+    size_t total_size;
+    size_t written_size;
+    size_t hole_size;
+    size_t req_size;
+    size_t readdir_size;
+    int64_t delta_size;
+    gf_atomic_t delta_blocks;
+    loc_t loc;
+    loc_t dot_shard_loc;
+    loc_t dot_shard_rm_loc;
+    loc_t loc2;
+    loc_t tmp_loc;
+    fd_t *fd;
+    dict_t *xattr_req;
+    dict_t *xattr_rsp;
+    inode_t **inode_list;
+    glusterfs_fop_t fop;
+    struct iatt prebuf;
+    struct iatt postbuf;
+    struct iatt preoldparent;
+    struct iatt postoldparent;
+    struct iatt prenewparent;
+    struct iatt postnewparent;
+    struct iovec *vector;
+    struct iobref *iobref;
+    struct iobuf *iobuf;
+    gf_dirent_t entries_head;
+    gf_boolean_t is_set_fsid;
+    gf_boolean_t list_inited;
+    shard_post_fop_handler_t handler;
+    shard_post_lookup_shards_fop_handler_t pls_fop_handler;
+    shard_post_resolve_fop_handler_t post_res_handler;
+    shard_post_mknod_fop_handler_t post_mknod_handler;
+    shard_post_update_size_fop_handler_t post_update_size_handler;
+    shard_inodelk_t int_inodelk;
+    shard_entrylk_t int_entrylk;
+    inode_t *resolver_base_inode;
+    gf_boolean_t first_lookup_done;
+    syncbarrier_t barrier;
+    gf_boolean_t lookup_shards_barriered;
+    gf_boolean_t unlink_shards_barriered;
+    gf_boolean_t resolve_not;
+    loc_t newloc;
+    call_frame_t *main_frame;
+    call_frame_t *inodelk_frame;
+    call_frame_t *entrylk_frame;
+    uint32_t deletion_rate;
+    gf_boolean_t cleanup_required;
+    uuid_t base_gfid;
+    char *name;
+} shard_local_t;
+
+typedef struct shard_inode_ctx {
+    uint64_t block_size; /* The block size with which this inode is
+                            sharded */
+    struct iatt stat;
+    gf_boolean_t refresh;
+    /* The following members of inode ctx will be applicable only to the
+     * individual shards' ctx and never the base file ctx.
+     */
+    struct list_head ilist;
+    uuid_t base_gfid;
+    int block_num;
+    gf_boolean_t refreshed;
+    struct list_head to_fsync_list;
+    int fsync_needed;
+    inode_t *inode;
+    int fsync_count;
+    inode_t *base_inode;
+} shard_inode_ctx_t;
+
+typedef enum {
+    SHARD_INTERNAL_DIR_DOT_SHARD = 1,
+    SHARD_INTERNAL_DIR_DOT_SHARD_REMOVE_ME,
+} shard_internal_dir_type_t;
+
+#endif /* __SHARD_H__ */
diff --git a/xlators/performance/stat-prefetch/Makefile.am b/xlators/features/snapview-client/Makefile.am
index af437a64d6d..af437a64d6d 100644
--- a/xlators/performance/stat-prefetch/Makefile.am
+++ b/xlators/features/snapview-client/Makefile.am
diff --git a/xlators/features/snapview-client/src/Makefile.am b/xlators/features/snapview-client/src/Makefile.am
new file mode 100644
index 00000000000..fa08656c537
--- /dev/null
+++ b/xlators/features/snapview-client/src/Makefile.am
@@ -0,0 +1,16 @@
+xlator_LTLIBRARIES = snapview-client.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+snapview_client_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+snapview_client_la_SOURCES = snapview-client.c
+snapview_client_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = snapview-client.h snapview-client-mem-types.h snapview-client-messages.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/snapview-client/src/snapview-client-mem-types.h b/xlators/features/snapview-client/src/snapview-client-mem-types.h
new file mode 100644
index 00000000000..3c3ab555a55
--- /dev/null
+++ b/xlators/features/snapview-client/src/snapview-client-mem-types.h
@@ -0,0 +1,24 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef _SVC_MEM_TYPES_H
+#define _SVC_MEM_TYPES_H
+
+#include <glusterfs/mem-types.h>
+
+enum svc_mem_types {
+    gf_svc_mt_svc_private_t = gf_common_mt_end + 1,
+    gf_svc_mt_svc_local_t,
+    gf_svc_mt_svc_inode_t,
+    gf_svc_mt_svc_fd_t,
+    gf_svc_mt_end
+};
+
+#endif
diff --git a/xlators/features/snapview-client/src/snapview-client-messages.h b/xlators/features/snapview-client/src/snapview-client-messages.h
new file mode 100644
index 00000000000..c02fb154930
--- /dev/null
+++ b/xlators/features/snapview-client/src/snapview-client-messages.h
@@ -0,0 +1,71 @@
+/*
+ Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com>
+ This file is part of GlusterFS.
+
+ This file is licensed to you under your choice of the GNU Lesser
+ General Public License, version 3 or any later version (LGPLv3 or
+ later), or the GNU General Public License, version 2 (GPLv2), in all
+ cases as published by the Free Software Foundation.
+ */
+
+#ifndef _SNAPVIEW_CLIENT_MESSAGES_H_
+#define _SNAPVIEW_CLIENT_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(SNAPVIEW_CLIENT, SVC_MSG_NO_MEMORY, SVC_MSG_MEM_ACNT_FAILED,
+           SVC_MSG_SET_INODE_CONTEXT_FAILED, SVC_MSG_GET_INODE_CONTEXT_FAILED,
+           SVC_MSG_DELETE_INODE_CONTEXT_FAILED, SVC_MSG_SET_FD_CONTEXT_FAILED,
+           SVC_MSG_GET_FD_CONTEXT_FAILED, SVC_MSG_DICT_SET_FAILED,
+           SVC_MSG_SUBVOLUME_NULL, SVC_MSG_NO_CHILD_FOR_XLATOR,
+           SVC_MSG_XLATOR_CHILDREN_WRONG, SVC_MSG_NORMAL_GRAPH_LOOKUP_FAIL,
+           SVC_MSG_SNAPVIEW_GRAPH_LOOKUP_FAIL, SVC_MSG_OPENDIR_SPECIAL_DIR,
+           SVC_MSG_RENAME_SNAPSHOT_ENTRY, SVC_MSG_LINK_SNAPSHOT_ENTRY,
+           SVC_MSG_COPY_ENTRY_POINT_FAILED, SVC_MSG_ENTRY_POINT_SPECIAL_DIR,
+           SVC_MSG_STR_LEN, SVC_MSG_INVALID_ENTRY_POINT, SVC_MSG_NULL_PRIV,
+           SVC_MSG_PRIV_DESTROY_FAILED, SVC_MSG_ALLOC_FD_FAILED,
+           SVC_MSG_ALLOC_INODE_FAILED, SVC_MSG_NULL_SPECIAL_DIR,
+           SVC_MSG_MEM_POOL_GET_FAILED);
+
+#define SVC_MSG_ALLOC_FD_FAILED_STR "failed to allocate new fd context"
+#define SVC_MSG_SET_FD_CONTEXT_FAILED_STR "failed to set fd context"
+#define SVC_MSG_STR_LEN_STR                                                    \
+    "destination buffer size is less than the length of entry point name"
+#define SVC_MSG_NORMAL_GRAPH_LOOKUP_FAIL_STR "lookup failed on normal graph"
+#define SVC_MSG_SNAPVIEW_GRAPH_LOOKUP_FAIL_STR "lookup failed on snapview graph"
+#define SVC_MSG_SET_INODE_CONTEXT_FAILED_STR "failed to set inode context"
+#define SVC_MSG_NO_MEMORY_STR "failed to allocate memory"
+#define SVC_MSG_COPY_ENTRY_POINT_FAILED_STR                                    \
+    "failed to copy the entry point string"
+#define SVC_MSG_GET_FD_CONTEXT_FAILED_STR "fd context not found"
+#define SVC_MSG_GET_INODE_CONTEXT_FAILED_STR "failed to get inode context"
+#define SVC_MSG_ALLOC_INODE_FAILED_STR "failed to allocate new inode"
+#define SVC_MSG_DICT_SET_FAILED_STR "failed to set dict"
+#define SVC_MSG_RENAME_SNAPSHOT_ENTRY_STR                                      \
+    "rename happening on a entry residing in snapshot"
+#define SVC_MSG_DELETE_INODE_CONTEXT_FAILED_STR "failed to delete inode context"
+#define SVC_MSG_NULL_PRIV_STR "priv NULL"
+#define SVC_MSG_INVALID_ENTRY_POINT_STR "not a valid entry point"
+#define SVC_MSG_MEM_ACNT_FAILED_STR "Memory accouting init failed"
+#define SVC_MSG_NO_CHILD_FOR_XLATOR_STR "configured without any child"
+#define SVC_MSG_XLATOR_CHILDREN_WRONG_STR                                      \
+    "snap-view-client has got wrong subvolumes. It can have only 2"
+#define SVC_MSG_ENTRY_POINT_SPECIAL_DIR_STR                                    \
+    "entry point directory cannot be part of special directory"
+#define SVC_MSG_NULL_SPECIAL_DIR_STR "null special directory"
+#define SVC_MSG_MEM_POOL_GET_FAILED_STR                                        \
+    "could not get mem pool for frame->local"
+#define SVC_MSG_PRIV_DESTROY_FAILED_STR "failed to destroy private"
+#define SVC_MSG_LINK_SNAPSHOT_ENTRY_STR                                        \
+    "link happening on a entry residin gin snapshot"
+#endif /* !_SNAPVIEW_CLIENT_MESSAGES_H_ */
diff --git a/xlators/features/snapview-client/src/snapview-client.c b/xlators/features/snapview-client/src/snapview-client.c
new file mode 100644
index 00000000000..486c5179d5b
--- /dev/null
+++ b/xlators/features/snapview-client/src/snapview-client.c
@@ -0,0 +1,2791 @@
+/*
+  Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "snapview-client.h"
+#include <glusterfs/inode.h>
+#include <glusterfs/byte-order.h>
+
+static void
+svc_local_free(svc_local_t *local)
+{
+    if (local) {
+        loc_wipe(&local->loc);
+        if (local->fd)
+            fd_unref(local->fd);
+        if (local->xdata)
+            dict_unref(local->xdata);
+        mem_put(local);
+    }
+}
+
+static xlator_t *
+svc_get_subvolume(xlator_t *this, int inode_type)
+{
+    xlator_t *subvolume = NULL;
+
+    GF_VALIDATE_OR_GOTO("snapview-client", this, out);
+
+    if (inode_type == VIRTUAL_INODE)
+        subvolume = SECOND_CHILD(this);
+    else
+        subvolume = FIRST_CHILD(this);
+
+out:
+    return subvolume;
+}
+
+static int32_t
+__svc_inode_ctx_set(xlator_t *this, inode_t *inode, int inode_type)
+{
+    uint64_t value = 0;
+    int32_t ret = -1;
+
+    GF_VALIDATE_OR_GOTO("snapview-client", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+
+    value = inode_type;
+
+    ret = __inode_ctx_set(inode, this, &value);
+
+out:
+    return ret;
+}
+
+static int
+__svc_inode_ctx_get(xlator_t *this, inode_t *inode, int *inode_type)
+{
+    uint64_t value = 0;
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("snapview-client", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+
+    ret = __inode_ctx_get(inode, this, &value);
+    if (ret < 0)
+        goto out;
+
+    *inode_type = (int)(value);
+
+out:
+    return ret;
+}
+
+static int
+svc_inode_ctx_get(xlator_t *this, inode_t *inode, int *inode_type)
+{
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("snapview-client", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+
+    LOCK(&inode->lock);
+    {
+        ret = __svc_inode_ctx_get(this, inode, inode_type);
+    }
+    UNLOCK(&inode->lock);
+
+out:
+    return ret;
+}
+
+static int32_t
+svc_inode_ctx_set(xlator_t *this, inode_t *inode, int inode_type)
+{
+    int32_t ret = -1;
+
+    GF_VALIDATE_OR_GOTO("snapview-client", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+
+    LOCK(&inode->lock);
+    {
+        ret = __svc_inode_ctx_set(this, inode, inode_type);
+    }
+    UNLOCK(&inode->lock);
+
+out:
+    return ret;
+}
+
+static svc_fd_t *
+svc_fd_new(void)
+{
+    svc_fd_t *svc_fd = NULL;
+
+    svc_fd = GF_CALLOC(1, sizeof(*svc_fd), gf_svc_mt_svc_fd_t);
+
+    return svc_fd;
+}
+
+static svc_fd_t *
+__svc_fd_ctx_get(xlator_t *this, fd_t *fd)
+{
+    svc_fd_t *svc_fd = NULL;
+    uint64_t value = 0;
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("snapview-client", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    ret = __fd_ctx_get(fd, this, &value);
+    if (ret)
+        return NULL;
+
+    svc_fd = (svc_fd_t *)((long)value);
+
+out:
+    return svc_fd;
+}
+
+static svc_fd_t *
+svc_fd_ctx_get(xlator_t *this, fd_t *fd)
+{
+    svc_fd_t *svc_fd = NULL;
+
+    GF_VALIDATE_OR_GOTO("snapview-client", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    LOCK(&fd->lock);
+    {
+        svc_fd = __svc_fd_ctx_get(this, fd);
+    }
+    UNLOCK(&fd->lock);
+
+out:
+    return svc_fd;
+}
+
+static int
+__svc_fd_ctx_set(xlator_t *this, fd_t *fd, svc_fd_t *svc_fd)
+{
+    uint64_t value = 0;
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("snapview-client", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+    GF_VALIDATE_OR_GOTO(this->name, svc_fd, out);
+
+    value = (uint64_t)(long)svc_fd;
+
+    ret = __fd_ctx_set(fd, this, value);
+
+out:
+    return ret;
+}
+
+static svc_fd_t *
+__svc_fd_ctx_get_or_new(xlator_t *this, fd_t *fd)
+{
+    svc_fd_t *svc_fd = NULL;
+    int ret = -1;
+    inode_t *inode = NULL;
+
+    GF_VALIDATE_OR_GOTO("snapview-client", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    inode = fd->inode;
+    svc_fd = __svc_fd_ctx_get(this, fd);
+    if (svc_fd) {
+        ret = 0;
+        goto out;
+    }
+
+    svc_fd = svc_fd_new();
+    if (!svc_fd) {
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, SVC_MSG_ALLOC_FD_FAILED,
+                "gfid=%s", uuid_utoa(inode->gfid), NULL);
+        goto out;
+    }
+
+    ret = __svc_fd_ctx_set(this, fd, svc_fd);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_SET_FD_CONTEXT_FAILED,
+                "gfid=%s", uuid_utoa(inode->gfid), NULL);
+        ret = -1;
+    }
+
+out:
+    if (ret) {
+        GF_FREE(svc_fd);
+        svc_fd = NULL;
+    }
+
+    return svc_fd;
+}
+
+static svc_fd_t *
+svc_fd_ctx_get_or_new(xlator_t *this, fd_t *fd)
+{
+    svc_fd_t *svc_fd = NULL;
+
+    GF_VALIDATE_OR_GOTO("snapview-client", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    LOCK(&fd->lock);
+    {
+        svc_fd = __svc_fd_ctx_get_or_new(this, fd);
+    }
+    UNLOCK(&fd->lock);
+
+out:
+    return svc_fd;
+}
+
+/**
+ * @this: xlator
+ * @entry_point: pointer to the buffer provided by consumer
+ *
+ * This function is mainly for copying the entry point name
+ * (stored as string in priv->path) to a buffer point to by
+ * @entry_point within the lock. It is for the consumer to
+ * allocate the memory for the buffer.
+ *
+ * This function is called by all the functions (or fops)
+ * who need to use priv->path for avoiding the race.
+ * For example, either in lookup or in any other fop,
+ * while priv->path is being accessed, a reconfigure can
+ * happen to change priv->path. This ensures that, a lock
+ * is taken before accessing priv->path.
+ **/
+int
+gf_svc_get_entry_point(xlator_t *this, char *entry_point, size_t dest_size)
+{
+    int ret = -1;
+    svc_private_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("snapview-client", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, entry_point, out);
+
+    priv = this->private;
+
+    LOCK(&priv->lock);
+    {
+        if (dest_size <= strlen(priv->path)) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_STR_LEN,
+                    "dest-size=%zu", dest_size, "priv-path-len=%zu",
+                    strlen(priv->path), "path=%s", priv->path, NULL);
+        } else {
+            snprintf(entry_point, dest_size, "%s", priv->path);
+            ret = 0;
+        }
+    }
+    UNLOCK(&priv->lock);
+
+out:
+    return ret;
+}
+
+static int32_t
+gf_svc_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, inode_t *inode,
+                  struct iatt *buf, dict_t *xdata, struct iatt *postparent)
+{
+    svc_local_t *local = NULL;
+    xlator_t *subvolume = NULL;
+    gf_boolean_t do_unwind = _gf_true;
+    int inode_type = -1;
+    int ret = -1;
+
+    local = frame->local;
+    subvolume = local->subvolume;
+    if (!subvolume) {
+        gf_msg_callingfn(this->name, GF_LOG_ERROR, 0, SVC_MSG_SUBVOLUME_NULL,
+                         "path: %s gfid: %s ", local->loc.path,
+                         inode ? uuid_utoa(inode->gfid) : "");
+        GF_ASSERT(0);
+    }
+
+    /* There is a possibility that, the client process just came online
+       and does not have the inode on which the lookup came. In that case,
+       the fresh inode created from fuse for the lookup fop, won't have
+       the inode context set without which svc cannot decide where to
+       STACK_WIND to. So by default it decides to send the fop to the
+       regular subvolume (i.e first child of the xlator). If lookup fails
+       on the regular volume, then there is a possibility that the lookup
+       is happening on a virtual inode (i.e history data residing in snaps).
+       So if lookup fails with ENOENT and the inode context is not there,
+       then send the lookup to the 2nd child of svc.
+
+       If there are any changes in volfile/client-restarted then inode-ctx
+       is lost. In this case if nameless lookup fails with ESTALE,
+       then send the lookup to the 2nd child of svc.
+    */
+    if (op_ret) {
+        if (subvolume == FIRST_CHILD(this)) {
+            gf_smsg(this->name,
+                    (op_errno == ENOENT || op_errno == ESTALE) ? GF_LOG_DEBUG
+                                                               : GF_LOG_ERROR,
+                    op_errno, SVC_MSG_NORMAL_GRAPH_LOOKUP_FAIL, "error=%s",
+                    strerror(op_errno), NULL);
+        } else {
+            gf_smsg(this->name,
+                    (op_errno == ENOENT || op_errno == ESTALE) ? GF_LOG_DEBUG
+                                                               : GF_LOG_ERROR,
+                    op_errno, SVC_MSG_SNAPVIEW_GRAPH_LOOKUP_FAIL, "error=%s",
+                    strerror(op_errno), NULL);
+            goto out;
+        }
+
+        if ((op_errno == ENOENT || op_errno == ESTALE) &&
+            !gf_uuid_is_null(local->loc.gfid)) {
+            if (inode != NULL)
+                ret = svc_inode_ctx_get(this, inode, &inode_type);
+
+            if (ret < 0 || inode == NULL) {
+                gf_msg_debug(this->name, 0,
+                             "Lookup on normal graph failed. "
+                             " Sending lookup to snapview-server");
+                subvolume = SECOND_CHILD(this);
+                local->subvolume = subvolume;
+                STACK_WIND(frame, gf_svc_lookup_cbk, subvolume,
+                           subvolume->fops->lookup, &local->loc, xdata);
+                do_unwind = _gf_false;
+            }
+        }
+
+        goto out;
+    }
+
+    if (subvolume == FIRST_CHILD(this))
+        inode_type = NORMAL_INODE;
+    else
+        inode_type = VIRTUAL_INODE;
+
+    ret = svc_inode_ctx_set(this, inode, inode_type);
+    if (ret)
+        gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_SET_INODE_CONTEXT_FAILED,
+                "gfid=%s", uuid_utoa(inode->gfid), NULL);
+
+out:
+    if (do_unwind) {
+        SVC_STACK_UNWIND(lookup, frame, op_ret, op_errno, inode, buf, xdata,
+                         postparent);
+    }
+
+    return 0;
+}
+
+static int32_t
+gf_svc_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    int32_t ret = -1;
+    svc_local_t *local = NULL;
+    xlator_t *subvolume = NULL;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    inode_t *parent = NULL;
+    dict_t *new_xdata = NULL;
+    int inode_type = -1;
+    int parent_type = -1;
+    gf_boolean_t wind = _gf_false;
+    char entry_point[NAME_MAX + 1] = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    ret = svc_inode_ctx_get(this, loc->inode, &inode_type);
+    if (!__is_root_gfid(loc->gfid)) {
+        if (loc->parent) {
+            parent = inode_ref(loc->parent);
+            ret = svc_inode_ctx_get(this, loc->parent, &parent_type);
+        } else {
+            parent = inode_parent(loc->inode, loc->pargfid, NULL);
+            if (parent)
+                ret = svc_inode_ctx_get(this, parent, &parent_type);
+        }
+    }
+
+    local = mem_get0(this->local_pool);
+    if (!local) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno, SVC_MSG_NO_MEMORY, NULL);
+        goto out;
+    }
+
+    frame->local = local;
+    loc_copy(&local->loc, loc);
+
+    if (__is_root_gfid(loc->inode->gfid)) {
+        subvolume = FIRST_CHILD(this);
+        GF_ASSERT(subvolume);
+        local->subvolume = subvolume;
+        wind = _gf_true;
+        goto out;
+    }
+
+    /* nfs sends nameless lookups directly using the gfid. In that case
+       loc->name will be NULL. So check if loc->name is NULL. If so, then
+       try to get the subvolume using inode context. But if the inode has
+       not been looked up yet, then send the lookup call to the first
+       subvolume.
+    */
+
+    if (!loc->name) {
+        if (gf_uuid_is_null(loc->inode->gfid)) {
+            subvolume = FIRST_CHILD(this);
+            local->subvolume = subvolume;
+            wind = _gf_true;
+            goto out;
+        } else {
+            if (inode_type >= 0)
+                subvolume = svc_get_subvolume(this, inode_type);
+            else
+                subvolume = FIRST_CHILD(this);
+            local->subvolume = subvolume;
+            wind = _gf_true;
+            goto out;
+        }
+    }
+
+    if (gf_svc_get_entry_point(this, entry_point, sizeof(entry_point))) {
+        gf_smsg(this->name, GF_LOG_WARNING, op_errno,
+                SVC_MSG_COPY_ENTRY_POINT_FAILED, NULL);
+        goto out;
+    }
+
+    if (strcmp(loc->name, entry_point)) {
+        if (parent_type == VIRTUAL_INODE) {
+            subvolume = SECOND_CHILD(this);
+        } else {
+            /*
+             * Either parent type is normal graph, or the parent
+             * type is uncertain.
+             */
+            subvolume = FIRST_CHILD(this);
+        }
+        local->subvolume = subvolume;
+    } else {
+        subvolume = SECOND_CHILD(this);
+        local->subvolume = subvolume;
+        if (parent_type == NORMAL_INODE) {
+            /* Indication of whether the lookup is happening on the
+               entry point or not, to the snapview-server.
+            */
+            SVC_ENTRY_POINT_SET(this, xdata, op_ret, op_errno, new_xdata, ret,
+                                out);
+        }
+    }
+
+    wind = _gf_true;
+
+out:
+    if (wind)
+        STACK_WIND(frame, gf_svc_lookup_cbk, subvolume, subvolume->fops->lookup,
+                   loc, xdata);
+    else
+        SVC_STACK_UNWIND(lookup, frame, op_ret, op_errno, NULL, NULL, NULL,
+                         NULL);
+    if (new_xdata)
+        dict_unref(new_xdata);
+
+    if (parent)
+        inode_unref(parent);
+
+    return 0;
+}
+
+static int32_t
+gf_svc_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    xlator_t *subvolume = NULL;
+    int32_t ret = -1;
+    int inode_type = -1;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    gf_boolean_t wind = _gf_false;
+    svc_private_t *priv = NULL;
+    const char *path = NULL;
+    int path_len = -1;
+    int snap_len = -1;
+    loc_t root_loc = {
+        0,
+    };
+    loc_t *temp_loc = NULL;
+    char entry_point[NAME_MAX + 1] = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    priv = this->private;
+    SVC_GET_SUBVOL_FROM_CTX(this, op_ret, op_errno, inode_type, ret, loc->inode,
+                            subvolume, out);
+    path_len = strlen(loc->path);
+    snap_len = strlen(priv->path);
+    temp_loc = loc;
+
+    if (path_len >= snap_len && inode_type == VIRTUAL_INODE) {
+        path = &loc->path[path_len - snap_len];
+        if (gf_svc_get_entry_point(this, entry_point, sizeof(entry_point))) {
+            gf_smsg(this->name, GF_LOG_WARNING, op_errno,
+                    SVC_MSG_COPY_ENTRY_POINT_FAILED, NULL);
+            goto out;
+        }
+
+        if (!strcmp(path, entry_point)) {
+            /*
+             * statfs call for virtual snap directory.
+             * Sent the fops to parent volume by removing
+             * virtual directory from path
+             */
+            subvolume = FIRST_CHILD(this);
+            root_loc.path = gf_strdup("/");
+            gf_uuid_clear(root_loc.gfid);
+            root_loc.gfid[15] = 1;
+            root_loc.inode = inode_ref(loc->inode->table->root);
+            temp_loc = &root_loc;
+        }
+    }
+
+    STACK_WIND_TAIL(frame, subvolume, subvolume->fops->statfs, temp_loc, xdata);
+    if (temp_loc == &root_loc)
+        loc_wipe(temp_loc);
+
+    wind = _gf_true;
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(statfs, frame, op_ret, op_errno, NULL, NULL);
+    return 0;
+}
+
+static int32_t
+gf_svc_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                dict_t *xdata)
+{
+    /* TODO: FIX ME
+     * Consider a testcase:
+     * #mount -t nfs host1:/vol1 /mnt
+     * #ls /mnt
+     * #ls /mnt/.snaps (As expected this fails)
+     * #gluster volume set vol1 features.uss enable
+     * Now `ls /mnt/.snaps` should work, but fails with No such file or
+     * directory. This is because NFS client (gNFS) caches the list of files
+     * in a directory. This cache is updated if there are any changes in the
+     * directory attributes. So, one way to solve this problem is to change
+     * 'ctime' attribute when USS is enabled as below.
+     *
+     * if (op_ret == 0 && IA_ISDIR(buf->ia_type))
+     *     buf->ia_ctime_nsec++;
+     *
+     * But this is not the ideal solution as applications see the unexpected
+     * ctime change causing failures.
+     */
+
+    SVC_STACK_UNWIND(stat, frame, op_ret, op_errno, buf, xdata);
+    return 0;
+}
+
+/* should all the fops be handled like lookup is supposed to be
+   handled? i.e just based on inode type decide where the call should
+   be sent and in the call back update the contexts.
+*/
+static int32_t
+gf_svc_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    int32_t ret = -1;
+    int inode_type = -1;
+    xlator_t *subvolume = NULL;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    gf_boolean_t wind = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    SVC_GET_SUBVOL_FROM_CTX(this, op_ret, op_errno, inode_type, ret, loc->inode,
+                            subvolume, out);
+
+    STACK_WIND(frame, gf_svc_stat_cbk, subvolume, subvolume->fops->stat, loc,
+               xdata);
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(stat, frame, op_ret, op_errno, NULL, NULL);
+    return 0;
+}
+
+static int32_t
+gf_svc_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    int32_t ret = -1;
+    int inode_type = -1;
+    xlator_t *subvolume = NULL;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    gf_boolean_t wind = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd->inode, out);
+
+    SVC_GET_SUBVOL_FROM_CTX(this, op_ret, op_errno, inode_type, ret, fd->inode,
+                            subvolume, out);
+
+    STACK_WIND_TAIL(frame, subvolume, subvolume->fops->fstat, fd, xdata);
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(fstat, frame, op_ret, op_errno, NULL, NULL);
+
+    return ret;
+}
+
+static int32_t
+gf_svc_opendir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
+{
+    svc_fd_t *svc_fd = NULL;
+    svc_local_t *local = NULL;
+    svc_private_t *priv = NULL;
+    gf_boolean_t special_dir = _gf_false;
+    char path[PATH_MAX] = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("snapview-client", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+    if (op_ret)
+        goto out;
+
+    priv = this->private;
+    local = frame->local;
+
+    if (local->subvolume == FIRST_CHILD(this) && priv->special_dir &&
+        strcmp(priv->special_dir, "")) {
+        if (!__is_root_gfid(fd->inode->gfid))
+            snprintf(path, sizeof(path), "%s/.", priv->special_dir);
+        else
+            snprintf(path, sizeof(path), "/.");
+
+        if (!strcmp(local->loc.path, priv->special_dir) ||
+            !strcmp(local->loc.path, path)) {
+            gf_msg_debug(this->name, 0,
+                         "got opendir on special directory"
+                         " %s (gfid: %s)",
+                         path, uuid_utoa(fd->inode->gfid));
+            special_dir = _gf_true;
+        }
+    }
+
+    if (special_dir) {
+        svc_fd = svc_fd_ctx_get_or_new(this, fd);
+        if (!svc_fd) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_GET_FD_CONTEXT_FAILED,
+                    "gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+            goto out;
+        }
+
+        svc_fd->last_offset = -1;
+        svc_fd->special_dir = special_dir;
+    }
+
+out:
+    STACK_UNWIND_STRICT(opendir, frame, op_ret, op_errno, fd, xdata);
+
+    return 0;
+}
+
+/* If the inode represents a directory which is actually
+   present in a snapshot, then opendir on that directory
+   should be sent to the snap-view-server which opens
+   the directory in the corresponding graph.
+   In fact any opendir call on a virtual directory
+   should be sent to svs. Because if it fakes success
+   here, then later when readdir on that fd comes, there
+   will not be any corresponding fd opened on svs and
+   svc has to do things that open-behind is doing.
+*/
+static int32_t
+gf_svc_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+               dict_t *xdata)
+{
+    int32_t ret = -1;
+    int inode_type = -1;
+    xlator_t *subvolume = NULL;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    gf_boolean_t wind = _gf_false;
+    svc_local_t *local = NULL;
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    local = mem_get0(this->local_pool);
+    if (!local) {
+        op_errno = ENOMEM;
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno, SVC_MSG_NO_MEMORY,
+                "path=%s", loc->path, "gfid=%s", uuid_utoa(fd->inode->gfid),
+                NULL);
+        goto out;
+    }
+    loc_copy(&local->loc, loc);
+    frame->local = local;
+
+    SVC_GET_SUBVOL_FROM_CTX(this, op_ret, op_errno, inode_type, ret, loc->inode,
+                            subvolume, out);
+    local->subvolume = subvolume;
+
+    STACK_WIND(frame, gf_svc_opendir_cbk, subvolume, subvolume->fops->opendir,
+               loc, fd, xdata);
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(opendir, frame, op_ret, op_errno, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+gf_svc_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    int32_t ret = -1;
+    int inode_type = -1;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    gf_boolean_t wind = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    ret = svc_inode_ctx_get(this, loc->inode, &inode_type);
+    if (ret < 0) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                SVC_MSG_GET_INODE_CONTEXT_FAILED, "path=%s", loc->path,
+                "gfid= %s", uuid_utoa(loc->inode->gfid), NULL);
+        goto out;
+    }
+
+    if (inode_type == NORMAL_INODE) {
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->setattr, loc, stbuf, valid,
+                        xdata);
+    } else {
+        op_ret = -1;
+        op_errno = EROFS;
+        goto out;
+    }
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(setattr, frame, op_ret, op_errno, NULL, NULL, NULL);
+    return 0;
+}
+
+/* XXX: This function is currently not used. Remove "#if 0" when required */
+#if 0
+static int32_t
+gf_svc_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
+                 struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+        int32_t      ret        = -1;
+        int          inode_type = -1;
+        int          op_ret     = -1;
+        int          op_errno   = EINVAL;
+        gf_boolean_t wind       = _gf_false;
+
+        GF_VALIDATE_OR_GOTO ("svc", this, out);
+        GF_VALIDATE_OR_GOTO (this->name, frame, out);
+        GF_VALIDATE_OR_GOTO (this->name, fd, out);
+        GF_VALIDATE_OR_GOTO (this->name, fd->inode, out);
+
+        ret = svc_inode_ctx_get (this, fd->inode, &inode_type);
+        if (ret < 0) {
+                op_ret = -1;
+                op_errno = EINVAL;
+                gf_msg (this->name, GF_LOG_ERROR, op_errno,
+                        SVC_MSG_GET_INODE_CONTEXT_FAILED, "failed to "
+                        "get the inode context for %s",
+                        uuid_utoa (fd->inode->gfid));
+                goto out;
+        }
+
+        if (inode_type == NORMAL_INODE) {
+                STACK_WIND_TAIL (frame, FIRST_CHILD (this),
+                                 FIRST_CHILD (this)->fops->fsetattr, fd, stbuf,
+                                 valid, xdata);
+        } else {
+                op_ret = -1;
+                op_errno = EROFS;
+                goto out;
+        }
+
+        wind = _gf_true;
+
+out:
+        if (!wind)
+                SVC_STACK_UNWIND (fsetattr, frame, op_ret, op_errno,
+                                  NULL, NULL, NULL);
+        return 0;
+}
+#endif /* gf_svc_fsetattr() is not used */
+
+static int32_t
+gf_svc_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                const char *name, dict_t *xdata)
+{
+    int32_t ret = -1;
+    int inode_type = -1;
+    xlator_t *subvolume = NULL;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    gf_boolean_t wind = _gf_false;
+    svc_private_t *priv = NULL;
+    char attrname[PATH_MAX] = "";
+    char attrval[64] = "";
+    dict_t *dict = NULL;
+    char entry_point[NAME_MAX + 1] = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, priv, out);
+
+    /*
+     * Samba sends this special key for case insensitive
+     * filename check. This request comes with a parent
+     * path and with a special key GF_XATTR_GET_REAL_FILENAME_KEY.
+     * e.g. "glusterfs.get_real_filename:.snaps".
+     * If the name variable matches this key then we have
+     * to send back .snaps as the real filename.
+     */
+    if (!name)
+        goto stack_wind;
+
+    sscanf(name, "%[^:]:%[^@]", attrname, attrval);
+    strcat(attrname, ":");
+
+    if (!strcmp(attrname, GF_XATTR_GET_REAL_FILENAME_KEY)) {
+        if (gf_svc_get_entry_point(this, entry_point, sizeof(entry_point))) {
+            gf_smsg(this->name, GF_LOG_WARNING, op_errno,
+                    SVC_MSG_COPY_ENTRY_POINT_FAILED, NULL);
+            goto out;
+        }
+
+        if (!strcasecmp(attrval, entry_point)) {
+            dict = dict_new();
+            if (NULL == dict) {
+                op_errno = ENOMEM;
+                goto out;
+            }
+
+            ret = dict_set_dynstr_with_alloc(dict, (char *)name, entry_point);
+
+            if (ret) {
+                op_errno = ENOMEM;
+                goto out;
+            }
+
+            op_errno = 0;
+            op_ret = strlen(entry_point) + 1;
+            /* We should return from here */
+            goto out;
+        }
+    }
+stack_wind:
+    SVC_GET_SUBVOL_FROM_CTX(this, op_ret, op_errno, inode_type, ret, loc->inode,
+                            subvolume, out);
+
+    STACK_WIND_TAIL(frame, subvolume, subvolume->fops->getxattr, loc, name,
+                    xdata);
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(getxattr, frame, op_ret, op_errno, dict, NULL);
+
+    if (dict)
+        dict_unref(dict);
+
+    return 0;
+}
+
+/* XXX: This function is currently not used. Mark it '#if 0' when required */
+#if 0
+static int32_t
+gf_svc_fgetxattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
+                  const char *name, dict_t *xdata)
+{
+        int32_t       ret        = -1;
+        int           inode_type = -1;
+        xlator_t     *subvolume  = NULL;
+        gf_boolean_t  wind       = _gf_false;
+        int           op_ret     = -1;
+        int           op_errno   = EINVAL;
+
+        GF_VALIDATE_OR_GOTO ("svc", this, out);
+        GF_VALIDATE_OR_GOTO (this->name, frame, out);
+        GF_VALIDATE_OR_GOTO (this->name, fd, out);
+        GF_VALIDATE_OR_GOTO (this->name, fd->inode, out);
+
+        SVC_GET_SUBVOL_FROM_CTX (this, op_ret, op_errno, inode_type, ret,
+                                 fd->inode, subvolume, out);
+
+        STACK_WIND_TAIL (frame, subvolume,
+                         subvolume->fops->fgetxattr, fd, name, xdata);
+
+        wind = _gf_true;
+
+out:
+        if (!wind)
+                SVC_STACK_UNWIND (fgetxattr, frame, op_ret, op_errno,
+                                  NULL, NULL);
+        return 0;
+}
+#endif /* gf_svc_fgetxattr() is not used */
+
+static int32_t
+gf_svc_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+                int32_t flags, dict_t *xdata)
+{
+    int32_t ret = -1;
+    int inode_type = -1;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    gf_boolean_t wind = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    ret = svc_inode_ctx_get(this, loc->inode, &inode_type);
+    if (ret < 0) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                SVC_MSG_GET_INODE_CONTEXT_FAILED, "name=%s", loc->name,
+                "gfid=%s", uuid_utoa(loc->inode->gfid), NULL);
+        goto out;
+    }
+
+    if (inode_type == NORMAL_INODE) {
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->setxattr, loc, dict, flags,
+                        xdata);
+    } else {
+        op_ret = -1;
+        op_errno = EROFS;
+        goto out;
+    }
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(setxattr, frame, op_ret, op_errno, NULL);
+
+    return 0;
+}
+
+static int32_t
+gf_svc_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+                 int32_t flags, dict_t *xdata)
+{
+    int32_t ret = -1;
+    int inode_type = -1;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    gf_boolean_t wind = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd->inode, out);
+
+    ret = svc_inode_ctx_get(this, fd->inode, &inode_type);
+    if (ret < 0) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                SVC_MSG_GET_INODE_CONTEXT_FAILED, "gfid=%s",
+                uuid_utoa(fd->inode->gfid), NULL);
+        goto out;
+    }
+
+    if (inode_type == NORMAL_INODE) {
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags,
+                        xdata);
+    } else {
+        op_ret = -1;
+        op_errno = EROFS;
+        goto out;
+    }
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        STACK_UNWIND_STRICT(fsetxattr, frame, op_ret, op_errno, NULL);
+
+    return 0;
+}
+
+static int32_t
+gf_svc_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+             dict_t *xdata)
+{
+    int inode_type = -1;
+    int ret = -1;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    gf_boolean_t wind = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    ret = svc_inode_ctx_get(this, loc->inode, &inode_type);
+    if (ret < 0) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                SVC_MSG_GET_INODE_CONTEXT_FAILED, "name=%s", loc->name,
+                "gfid=%s", uuid_utoa(loc->inode->gfid), NULL);
+        goto out;
+    }
+
+    if (inode_type == NORMAL_INODE) {
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->rmdir, loc, flags, xdata);
+    } else {
+        op_ret = -1;
+        op_errno = EROFS;
+        goto out;
+    }
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(rmdir, frame, op_ret, op_errno, NULL, NULL, NULL);
+    return 0;
+}
+
+static int32_t
+gf_svc_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, inode_t *inode,
+                 struct iatt *buf, struct iatt *preparent,
+                 struct iatt *postparent, dict_t *xdata)
+{
+    int inode_type = -1;
+    int ret = -1;
+
+    if (op_ret < 0)
+        goto out;
+
+    inode_type = NORMAL_INODE;
+    ret = svc_inode_ctx_set(this, inode, inode_type);
+    if (ret)
+        gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_SET_INODE_CONTEXT_FAILED,
+                NULL);
+
+out:
+    SVC_STACK_UNWIND(mkdir, frame, op_ret, op_errno, inode, buf, preparent,
+                     postparent, xdata);
+    return 0;
+}
+
+static int32_t
+gf_svc_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+             mode_t umask, dict_t *xdata)
+{
+    int parent_type = -1;
+    int ret = -1;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    gf_boolean_t wind = _gf_false;
+    char entry_point[NAME_MAX + 1] = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    ret = svc_inode_ctx_get(this, loc->parent, &parent_type);
+    if (ret < 0) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                SVC_MSG_GET_INODE_CONTEXT_FAILED, "gfid=%s",
+                uuid_utoa(loc->parent->gfid), NULL);
+        goto out;
+    }
+
+    if (gf_svc_get_entry_point(this, entry_point, sizeof(entry_point))) {
+        gf_smsg(this->name, GF_LOG_WARNING, op_errno,
+                SVC_MSG_COPY_ENTRY_POINT_FAILED, NULL);
+        goto out;
+    }
+
+    if (strcmp(loc->name, entry_point) && parent_type == NORMAL_INODE) {
+        STACK_WIND(frame, gf_svc_mkdir_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->mkdir, loc, mode, umask, xdata);
+    } else {
+        op_ret = -1;
+        op_errno = EROFS;
+        goto out;
+    }
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(mkdir, frame, op_ret, op_errno, NULL, NULL, NULL, NULL,
+                         NULL);
+    return 0;
+}
+
+static int32_t
+gf_svc_mknod_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, inode_t *inode,
+                 struct iatt *buf, struct iatt *preparent,
+                 struct iatt *postparent, dict_t *xdata)
+{
+    int inode_type = -1;
+    int ret = -1;
+
+    if (op_ret < 0)
+        goto out;
+
+    inode_type = NORMAL_INODE;
+    ret = svc_inode_ctx_set(this, inode, inode_type);
+    if (ret)
+        gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_SET_INODE_CONTEXT_FAILED,
+                NULL);
+
+out:
+    SVC_STACK_UNWIND(mknod, frame, op_ret, op_errno, inode, buf, preparent,
+                     postparent, xdata);
+    return 0;
+}
+
+static int32_t
+gf_svc_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+             dev_t rdev, mode_t umask, dict_t *xdata)
+{
+    int parent_type = -1;
+    int ret = -1;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    gf_boolean_t wind = _gf_false;
+    char entry_point[NAME_MAX + 1] = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    ret = svc_inode_ctx_get(this, loc->parent, &parent_type);
+    if (ret < 0) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                SVC_MSG_GET_INODE_CONTEXT_FAILED, "gfid=%s",
+                uuid_utoa(loc->parent->gfid), NULL);
+        goto out;
+    }
+
+    if (gf_svc_get_entry_point(this, entry_point, sizeof(entry_point))) {
+        gf_smsg(this->name, GF_LOG_WARNING, op_errno,
+                SVC_MSG_COPY_ENTRY_POINT_FAILED, NULL);
+        goto out;
+    }
+
+    if (strcmp(loc->name, entry_point) && parent_type == NORMAL_INODE) {
+        STACK_WIND(frame, gf_svc_mknod_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->mknod, loc, mode, rdev, umask,
+                   xdata);
+    } else {
+        op_ret = -1;
+        op_errno = EROFS;
+        goto out;
+    }
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(mknod, frame, op_ret, op_errno, NULL, NULL, NULL, NULL,
+                         NULL);
+    return 0;
+}
+
+/* If the flags of the open call contain O_WRONLY or O_RDWR and the inode is
+   a virtual inode, then unwind the call back with EROFS. Otherwise simply
+   STACK_WIND the call to the first child of svc xlator.
+*/
+static int32_t
+gf_svc_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+            fd_t *fd, dict_t *xdata)
+{
+    xlator_t *subvolume = NULL;
+    int inode_type = -1;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    int ret = -1;
+    gf_boolean_t wind = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    /* Another way is to STACK_WIND to normal subvolume, if inode
+       type is not there in the context. If the file actually resides
+       in snapshots, then ENOENT would be returned. Needs more analysis.
+    */
+    SVC_GET_SUBVOL_FROM_CTX(this, op_ret, op_errno, inode_type, ret, loc->inode,
+                            subvolume, out);
+
+    if (((flags & O_ACCMODE) == O_WRONLY) || ((flags & O_ACCMODE) == O_RDWR)) {
+        if (subvolume != FIRST_CHILD(this)) {
+            op_ret = -1;
+            op_errno = EINVAL;
+            goto out;
+        }
+    }
+
+    STACK_WIND_TAIL(frame, subvolume, subvolume->fops->open, loc, flags, fd,
+                    xdata);
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(open, frame, op_ret, op_errno, NULL, NULL);
+    return 0;
+}
+
+static int32_t
+gf_svc_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                  int32_t op_ret, int32_t op_errno, fd_t *fd, inode_t *inode,
+                  struct iatt *stbuf, struct iatt *preparent,
+                  struct iatt *postparent, dict_t *xdata)
+{
+    int inode_type = -1;
+    int ret = -1;
+
+    if (op_ret < 0)
+        goto out;
+
+    inode_type = NORMAL_INODE;
+    ret = svc_inode_ctx_set(this, inode, inode_type);
+    if (ret)
+        gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_SET_INODE_CONTEXT_FAILED,
+                NULL);
+
+out:
+    SVC_STACK_UNWIND(create, frame, op_ret, op_errno, fd, inode, stbuf,
+                     preparent, postparent, xdata);
+
+    return 0;
+}
+
+static int32_t
+gf_svc_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+              mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
+{
+    int parent_type = -1;
+    int ret = -1;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    gf_boolean_t wind = _gf_false;
+    char entry_point[NAME_MAX + 1] = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    ret = svc_inode_ctx_get(this, loc->parent, &parent_type);
+    if (ret < 0) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                SVC_MSG_GET_INODE_CONTEXT_FAILED, "gfid=%s",
+                uuid_utoa(loc->parent->gfid), NULL);
+        goto out;
+    }
+
+    if (gf_svc_get_entry_point(this, entry_point, sizeof(entry_point))) {
+        gf_smsg(this->name, GF_LOG_WARNING, op_errno,
+                SVC_MSG_COPY_ENTRY_POINT_FAILED, NULL);
+        goto out;
+    }
+
+    if (strcmp(loc->name, entry_point) && parent_type == NORMAL_INODE) {
+        STACK_WIND(frame, gf_svc_create_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->create, loc, flags, mode, umask, fd,
+                   xdata);
+    } else {
+        op_ret = -1;
+        op_errno = EROFS;
+        goto out;
+    }
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(create, frame, op_ret, op_errno, NULL, NULL, NULL,
+                         NULL, NULL, NULL);
+    return 0;
+}
+
+static int32_t
+gf_svc_symlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, inode_t *inode,
+                   struct iatt *buf, struct iatt *preparent,
+                   struct iatt *postparent, dict_t *xdata)
+{
+    int inode_type = -1;
+    int ret = -1;
+
+    if (op_ret < 0)
+        goto out;
+
+    inode_type = NORMAL_INODE;
+    ret = svc_inode_ctx_set(this, inode, inode_type);
+    if (ret)
+        gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_SET_INODE_CONTEXT_FAILED,
+                NULL);
+
+out:
+    SVC_STACK_UNWIND(symlink, frame, op_ret, op_errno, inode, buf, preparent,
+                     postparent, xdata);
+
+    return 0;
+}
+
+static int32_t
+gf_svc_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+               loc_t *loc, mode_t umask, dict_t *xdata)
+{
+    int parent_type = -1;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    int ret = -1;
+    gf_boolean_t wind = _gf_false;
+    char entry_point[NAME_MAX + 1] = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    ret = svc_inode_ctx_get(this, loc->parent, &parent_type);
+    if (ret < 0) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                SVC_MSG_GET_INODE_CONTEXT_FAILED, "gfid=%s",
+                uuid_utoa(loc->parent->gfid), NULL);
+        goto out;
+    }
+
+    if (gf_svc_get_entry_point(this, entry_point, sizeof(entry_point))) {
+        gf_smsg(this->name, GF_LOG_WARNING, op_errno,
+                SVC_MSG_COPY_ENTRY_POINT_FAILED, NULL);
+        goto out;
+    }
+
+    if (strcmp(loc->name, entry_point) && parent_type == NORMAL_INODE) {
+        STACK_WIND(frame, gf_svc_symlink_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->symlink, linkpath, loc, umask,
+                   xdata);
+    } else {
+        op_ret = -1;
+        op_errno = EROFS;
+        goto out;
+    }
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(symlink, frame, op_ret, op_errno, NULL, NULL, NULL,
+                         NULL, NULL);
+    return 0;
+}
+
+static int32_t
+gf_svc_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+              dict_t *xdata)
+{
+    int inode_type = -1;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    int ret = -1;
+    gf_boolean_t wind = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    ret = svc_inode_ctx_get(this, loc->inode, &inode_type);
+    if (ret < 0) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                SVC_MSG_GET_INODE_CONTEXT_FAILED, "gfid=%s",
+                uuid_utoa(loc->parent->gfid), NULL);
+        goto out;
+    }
+
+    if (inode_type == NORMAL_INODE) {
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->unlink, loc, flags, xdata);
+    } else {
+        op_ret = -1;
+        op_errno = EROFS;
+        goto out;
+    }
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(unlink, frame, op_ret, op_errno, NULL, NULL, NULL);
+    return 0;
+}
+
+static int32_t
+gf_svc_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+             off_t offset, uint32_t flags, dict_t *xdata)
+{
+    int inode_type = -1;
+    xlator_t *subvolume = NULL;
+    int ret = -1;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    gf_boolean_t wind = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd->inode, out);
+
+    SVC_GET_SUBVOL_FROM_CTX(this, op_ret, op_errno, inode_type, ret, fd->inode,
+                            subvolume, out);
+
+    STACK_WIND_TAIL(frame, subvolume, subvolume->fops->readv, fd, size, offset,
+                    flags, xdata);
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(readv, frame, op_ret, op_errno, NULL, 0, NULL, NULL,
+                         NULL);
+    return 0;
+}
+
+static int32_t
+gf_svc_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+                dict_t *xdata)
+{
+    int inode_type = -1;
+    xlator_t *subvolume = NULL;
+    int ret = -1;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    gf_boolean_t wind = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    SVC_GET_SUBVOL_FROM_CTX(this, op_ret, op_errno, inode_type, ret, loc->inode,
+                            subvolume, out);
+
+    STACK_WIND_TAIL(frame, subvolume, subvolume->fops->readlink, loc, size,
+                    xdata);
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        STACK_UNWIND_STRICT(readlink, frame, op_ret, op_errno, NULL, NULL,
+                            NULL);
+    return 0;
+}
+
+static int32_t
+gf_svc_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+              dict_t *xdata)
+{
+    int ret = -1;
+    int inode_type = -1;
+    xlator_t *subvolume = NULL;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    gf_boolean_t wind = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    SVC_GET_SUBVOL_FROM_CTX(this, op_ret, op_errno, inode_type, ret, loc->inode,
+                            subvolume, out);
+
+    STACK_WIND_TAIL(frame, subvolume, subvolume->fops->access, loc, mask,
+                    xdata);
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(access, frame, op_ret, op_errno, NULL);
+
+    return 0;
+}
+
+int32_t
+gf_svc_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, gf_dirent_t *entries,
+                   dict_t *xdata)
+{
+    gf_dirent_t *entry = NULL;
+    gf_dirent_t *tmpentry = NULL;
+    svc_local_t *local = NULL;
+    char entry_point[NAME_MAX + 1] = {
+        0,
+    };
+
+    if (op_ret < 0)
+        goto out;
+
+    local = frame->local;
+
+    /* If .snaps pre-exists, then it should not be listed
+     * in the NORMAL INODE directory when USS is enabled,
+     * so filter the .snaps entry if exists.
+     * However it is OK to list .snaps in VIRTUAL world
+     */
+    if (local->subvolume != FIRST_CHILD(this))
+        goto out;
+
+    /*
+     * Better to goto out if getting the entry point
+     * fails. We might end up sending the directory
+     * entry for the snapview entry point in the readdir
+     * response. But, the intention is to avoid the race
+     * condition where priv->path is being changed in
+     * reconfigure while this is accessing it.
+     */
+    if (gf_svc_get_entry_point(this, entry_point, sizeof(entry_point))) {
+        gf_smsg(this->name, GF_LOG_WARNING, op_errno,
+                SVC_MSG_COPY_ENTRY_POINT_FAILED, NULL);
+        goto out;
+    }
+
+    list_for_each_entry_safe(entry, tmpentry, &entries->list, list)
+    {
+        if (strcmp(entry_point, entry->d_name) == 0)
+            gf_dirent_entry_free(entry);
+    }
+
+out:
+    SVC_STACK_UNWIND(readdir, frame, op_ret, op_errno, entries, xdata);
+    return 0;
+}
+
+static int32_t
+gf_svc_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+               off_t off, dict_t *xdata)
+{
+    int inode_type = -1;
+    xlator_t *subvolume = NULL;
+    svc_local_t *local = NULL;
+    int ret = -1;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    gf_boolean_t wind = _gf_false;
+    svc_fd_t *svc_fd = NULL;
+    gf_dirent_t entries;
+
+    INIT_LIST_HEAD(&entries);
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd->inode, out);
+
+    svc_fd = svc_fd_ctx_get_or_new(this, fd);
+    if (!svc_fd)
+        gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_GET_FD_CONTEXT_FAILED,
+                "gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+    else {
+        if (svc_fd->entry_point_handled && off == svc_fd->last_offset) {
+            op_ret = 0;
+            op_errno = ENOENT;
+            goto out;
+        }
+    }
+
+    SVC_GET_SUBVOL_FROM_CTX(this, op_ret, op_errno, inode_type, ret, fd->inode,
+                            subvolume, out);
+
+    local = mem_get0(this->local_pool);
+    if (!local) {
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno, SVC_MSG_NO_MEMORY,
+                "inode-gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        goto out;
+    }
+    local->subvolume = subvolume;
+    frame->local = local;
+
+    STACK_WIND(frame, gf_svc_readdir_cbk, subvolume, subvolume->fops->readdir,
+               fd, size, off, xdata);
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(readdir, frame, op_ret, op_errno, &entries, NULL);
+
+    gf_dirent_free(&entries);
+
+    return 0;
+}
+
+/*
+ * This lookup if mainly for supporting USS for windows.
+ * Since the dentry for the entry-point directory is not sent in
+ * the readdir response, from windows explorer, there is no way
+ * to access the snapshots. If the explicit path of the entry-point
+ * directory is mentioned in the address bar, then windows sends
+ * readdir on the parent directory and compares if the entry point
+ * directory's name is there in readdir response. If it is not there
+ * then access to snapshot world is denied. And windows users cannot
+ * access snapshots via samba.
+ * So, to handle this a new option called special-directory is created,
+ * which if set, snapview-client will send the entry-point's dentry
+ * in readdirp o/p for the special directory, so that it will be
+ * visible from windows explorer.
+ * But to send that virtual entry, the following mechanism is used.
+ * 1) Check if readdir from posix is over.
+ * 2) If so, then send a lookup on entry point directory to snap daemon
+ * (this is needed because in readdirp inodes are linked, so we need to
+ * maintain 1:1 mapping between inodes (gfids) from snapview server to
+ * snapview client).
+ * 3) Once successful lookup response received, send a new entry to
+ * windows.
+ */
+
+static int32_t
+gf_svc_readdirp_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                           int32_t op_ret, int32_t op_errno, inode_t *inode,
+                           struct iatt *buf, dict_t *xdata,
+                           struct iatt *postparent)
+{
+    gf_dirent_t entries;
+    gf_dirent_t *entry = NULL;
+    svc_fd_t *svc_fd = NULL;
+    svc_local_t *local = NULL;
+    int inode_type = -1;
+    int ret = -1;
+    char entry_point[NAME_MAX + 1] = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("snapview-client", this, out);
+
+    INIT_LIST_HEAD(&entries.list);
+
+    local = frame->local;
+
+    if (op_ret) {
+        if (op_errno == ESTALE && !local->revalidate) {
+            local->revalidate = 1;
+            ret = gf_svc_special_dir_revalidate_lookup(frame, this, xdata);
+
+            if (!ret)
+                return 0;
+        }
+        op_ret = 0;
+        op_errno = ENOENT;
+        goto out;
+    }
+
+    svc_fd = svc_fd_ctx_get(this, local->fd);
+    if (!svc_fd) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_GET_FD_CONTEXT_FAILED,
+                "gfid=%s", uuid_utoa(local->fd->inode->gfid), NULL);
+        op_ret = 0;
+        op_errno = ENOENT;
+        goto out;
+    }
+
+    if (gf_svc_get_entry_point(this, entry_point, sizeof(entry_point))) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, SVC_MSG_COPY_ENTRY_POINT_FAILED,
+                NULL);
+        op_ret = 0;
+        op_errno = ENOENT;
+        goto out;
+    }
+
+    entry = gf_dirent_for_name(entry_point);
+    if (!entry) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_NO_MEMORY,
+                "entry-point=%s", entry_point, NULL);
+        op_ret = 0;
+        op_errno = ENOMEM;
+        goto out;
+    }
+
+    entry->inode = inode_ref(inode);
+    entry->d_off = svc_fd->last_offset + 22;
+    entry->d_ino = buf->ia_ino;
+    entry->d_type = DT_DIR;
+    entry->d_stat = *buf;
+    inode_type = VIRTUAL_INODE;
+    ret = svc_inode_ctx_set(this, entry->inode, inode_type);
+    if (ret)
+        gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_SET_INODE_CONTEXT_FAILED,
+                "entry-name=%s", entry->d_name, NULL);
+
+    list_add_tail(&entry->list, &entries.list);
+    op_ret = 1;
+    svc_fd->last_offset = entry->d_off;
+    svc_fd->entry_point_handled = _gf_true;
+
+out:
+    SVC_STACK_UNWIND(readdirp, frame, op_ret, op_errno, &entries,
+                     local ? local->xdata : NULL);
+
+    gf_dirent_free(&entries);
+
+    return 0;
+}
+
+int
+gf_svc_special_dir_revalidate_lookup(call_frame_t *frame, xlator_t *this,
+                                     dict_t *xdata)
+{
+    svc_local_t *local = NULL;
+    loc_t *loc = NULL;
+    dict_t *tmp_xdata = NULL;
+    char *path = NULL;
+    int ret = -1;
+    char entry_point[NAME_MAX + 1] = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("snapview-client", this, out);
+
+    local = frame->local;
+    loc = &local->loc;
+
+    if (local->xdata) {
+        dict_unref(local->xdata);
+        local->xdata = NULL;
+    }
+
+    if (xdata)
+        local->xdata = dict_ref(xdata);
+
+    inode_unref(loc->inode);
+    loc->inode = inode_new(loc->parent->table);
+    if (!loc->inode) {
+        gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, SVC_MSG_ALLOC_INODE_FAILED,
+                NULL);
+        goto out;
+    }
+
+    if (gf_svc_get_entry_point(this, entry_point, sizeof(entry_point))) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, SVC_MSG_COPY_ENTRY_POINT_FAILED,
+                NULL);
+        goto out;
+    }
+
+    gf_uuid_copy(local->loc.gfid, loc->inode->gfid);
+    ret = inode_path(loc->parent, entry_point, &path);
+    if (ret < 0)
+        goto out;
+
+    if (loc->path)
+        GF_FREE((char *)loc->path);
+
+    loc->path = gf_strdup(path);
+    if (loc->path) {
+        if (!loc->name || (loc->name && !strcmp(loc->name, ""))) {
+            loc->name = strrchr(loc->path, '/');
+            if (loc->name)
+                loc->name++;
+        }
+    } else
+        loc->path = NULL;
+
+    tmp_xdata = dict_new();
+    if (!tmp_xdata) {
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_set_str(tmp_xdata, "entry-point", "true");
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_DICT_SET_FAILED, NULL);
+        goto out;
+    }
+
+    STACK_WIND(frame, gf_svc_readdirp_lookup_cbk, SECOND_CHILD(this),
+               SECOND_CHILD(this)->fops->lookup, loc, tmp_xdata);
+out:
+    if (tmp_xdata)
+        dict_unref(tmp_xdata);
+
+    GF_FREE(path);
+    return ret;
+}
+
+static gf_boolean_t
+gf_svc_readdir_on_special_dir(call_frame_t *frame, void *cookie, xlator_t *this,
+                              int32_t op_ret, int32_t op_errno,
+                              gf_dirent_t *entries, dict_t *xdata)
+{
+    svc_local_t *local = NULL;
+    svc_private_t *private = NULL;
+    inode_t *inode = NULL;
+    fd_t *fd = NULL;
+    char *path = NULL;
+    loc_t *loc = NULL;
+    dict_t *tmp_xdata = NULL;
+    int ret = -1;
+    gf_boolean_t unwind = _gf_true;
+    svc_fd_t *svc_fd = NULL;
+    char entry_point[NAME_MAX + 1] = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("snapview-client", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+
+   private
+    = this->private;
+    local = frame->local;
+
+    loc = &local->loc;
+    fd = local->fd;
+    svc_fd = svc_fd_ctx_get(this, fd);
+    if (!svc_fd) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_GET_FD_CONTEXT_FAILED,
+                "gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+        goto out;
+    }
+
+    /*
+     * check if its end of readdir operation from posix, if special_dir
+     * option is set, if readdir is done on special directory and if
+     * readdirp is from normal regular graph.
+     */
+
+    if (!private->show_entry_point)
+        goto out;
+
+    if (op_ret == 0 && op_errno == ENOENT && private->special_dir &&
+        strcmp(private->special_dir, "") && svc_fd->special_dir &&
+        local->subvolume == FIRST_CHILD(this)) {
+        if (gf_svc_get_entry_point(this, entry_point, sizeof(entry_point))) {
+            gf_smsg(this->name, GF_LOG_WARNING, 0,
+                    SVC_MSG_GET_FD_CONTEXT_FAILED, NULL);
+            goto out;
+        }
+
+        inode = inode_grep(fd->inode->table, fd->inode, entry_point);
+        if (!inode) {
+            inode = inode_new(fd->inode->table);
+            if (!inode) {
+                gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_ALLOC_INODE_FAILED,
+                        NULL);
+                goto out;
+            }
+        }
+
+        gf_uuid_copy(local->loc.pargfid, fd->inode->gfid);
+        gf_uuid_copy(local->loc.gfid, inode->gfid);
+        if (gf_uuid_is_null(inode->gfid))
+            ret = inode_path(fd->inode, entry_point, &path);
+        else
+            ret = inode_path(inode, NULL, &path);
+
+        if (ret < 0)
+            goto out;
+        loc->path = gf_strdup(path);
+        if (loc->path) {
+            if (!loc->name || (loc->name && !strcmp(loc->name, ""))) {
+                loc->name = strrchr(loc->path, '/');
+                if (loc->name)
+                    loc->name++;
+            }
+        }
+
+        loc->inode = inode;
+        loc->parent = inode_ref(fd->inode);
+        tmp_xdata = dict_new();
+        if (!tmp_xdata)
+            goto out;
+        ret = dict_set_str(tmp_xdata, "entry-point", "true");
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_DICT_SET_FAILED, NULL);
+            goto out;
+        }
+
+        local->cookie = cookie;
+        if (local->xdata) {
+            dict_unref(local->xdata);
+            local->xdata = NULL;
+        }
+        if (xdata)
+            local->xdata = dict_ref(xdata);
+
+        STACK_WIND(frame, gf_svc_readdirp_lookup_cbk, SECOND_CHILD(this),
+                   SECOND_CHILD(this)->fops->lookup, loc, tmp_xdata);
+        unwind = _gf_false;
+    }
+
+out:
+    if (tmp_xdata)
+        dict_unref(tmp_xdata);
+
+    GF_FREE(path);
+    return unwind;
+}
+
+static int32_t
+gf_svc_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, gf_dirent_t *entries,
+                    dict_t *xdata)
+{
+    gf_dirent_t *entry = NULL;
+    gf_dirent_t *tmpentry = NULL;
+    svc_local_t *local = NULL;
+    int inode_type = -1;
+    int ret = -1;
+    svc_fd_t *svc_fd = NULL;
+    gf_boolean_t unwind = _gf_true;
+    char entry_point[NAME_MAX + 1] = {
+        0,
+    };
+
+    if (op_ret < 0)
+        goto out;
+
+    GF_VALIDATE_OR_GOTO("snapview-client", this, out);
+
+    local = frame->local;
+
+    svc_fd = svc_fd_ctx_get(this, local->fd);
+    if (!svc_fd) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, SVC_MSG_GET_FD_CONTEXT_FAILED,
+                "gfid=%s", uuid_utoa(local->fd->inode->gfid), NULL);
+    }
+
+    if (local->subvolume == FIRST_CHILD(this))
+        inode_type = NORMAL_INODE;
+    else
+        inode_type = VIRTUAL_INODE;
+
+    /*
+     * Better to goto out and return whatever is there in the
+     * readdirp response (even if the readdir response contains
+     * a directory entry for the snapshot entry point). Otherwise
+     * if we ignore the error, then there is a chance of race
+     * condition where, priv->path is changed in reconfigure
+     */
+    if (gf_svc_get_entry_point(this, entry_point, sizeof(entry_point))) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, SVC_MSG_COPY_ENTRY_POINT_FAILED,
+                NULL);
+        goto out;
+    }
+
+    list_for_each_entry_safe(entry, tmpentry, &entries->list, list)
+    {
+        /* If .snaps pre-exists, then it should not be listed
+         * in the NORMAL INODE directory when USS is enabled,
+         * so filter the .snaps entry if exists.
+         * However it is OK to list .snaps in VIRTUAL world
+         */
+        if (inode_type == NORMAL_INODE && !strcmp(entry_point, entry->d_name)) {
+            gf_dirent_entry_free(entry);
+            continue;
+        }
+
+        if (!entry->inode)
+            continue;
+
+        ret = svc_inode_ctx_set(this, entry->inode, inode_type);
+        if (ret)
+            gf_smsg(this->name, GF_LOG_ERROR, 0,
+                    SVC_MSG_SET_INODE_CONTEXT_FAILED, NULL);
+        if (svc_fd)
+            svc_fd->last_offset = entry->d_off;
+    }
+
+    unwind = gf_svc_readdir_on_special_dir(frame, cookie, this, op_ret,
+                                           op_errno, entries, xdata);
+
+out:
+    if (unwind)
+        SVC_STACK_UNWIND(readdirp, frame, op_ret, op_errno, entries, xdata);
+
+    return 0;
+}
+
+static int32_t
+gf_svc_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                off_t off, dict_t *xdata)
+{
+    int inode_type = -1;
+    xlator_t *subvolume = NULL;
+    svc_local_t *local = NULL;
+    int ret = -1;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    gf_boolean_t wind = _gf_false;
+    svc_fd_t *svc_fd = NULL;
+    gf_dirent_t entries;
+
+    INIT_LIST_HEAD(&entries.list);
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd->inode, out);
+
+    local = mem_get0(this->local_pool);
+    if (!local) {
+        op_errno = ENOMEM;
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno, SVC_MSG_NO_MEMORY, NULL);
+        goto out;
+    }
+
+    /*
+     * This is mainly for samba shares (or windows clients). As part of
+     * readdirp on the directory used as samba share, the entry point
+     * directory would have been added at the end. So when a new readdirp
+     * request comes, we have to check if the entry point has been handled
+     * or not in readdirp. That information and the offset used for it
+     * is remembered in fd context. If it has been handled, then simply
+     * unwind indication end of readdir operation.
+     */
+    svc_fd = svc_fd_ctx_get_or_new(this, fd);
+    if (!svc_fd)
+        gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_GET_FD_CONTEXT_FAILED,
+                "gfid=%s", uuid_utoa(fd->inode->gfid), NULL);
+    else {
+        if (svc_fd->entry_point_handled && off == svc_fd->last_offset) {
+            op_ret = 0;
+            op_errno = ENOENT;
+            goto out;
+        }
+    }
+
+    SVC_GET_SUBVOL_FROM_CTX(this, op_ret, op_errno, inode_type, ret, fd->inode,
+                            subvolume, out);
+
+    local->subvolume = subvolume;
+    local->fd = fd_ref(fd);
+    frame->local = local;
+
+    STACK_WIND(frame, gf_svc_readdirp_cbk, subvolume, subvolume->fops->readdirp,
+               fd, size, off, xdata);
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(readdirp, frame, op_ret, op_errno, &entries, NULL);
+
+    gf_dirent_free(&entries);
+
+    return 0;
+}
+
+/* Renaming the entries from or to snapshots is not allowed as the snapshots
+   are read-only.
+*/
+static int32_t
+gf_svc_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+              dict_t *xdata)
+{
+    int src_inode_type = -1;
+    int dst_inode_type = -1;
+    int dst_parent_type = -1;
+    int32_t op_ret = -1;
+    int32_t op_errno = 0;
+    int32_t ret = -1;
+    gf_boolean_t wind = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, oldloc, out);
+    GF_VALIDATE_OR_GOTO(this->name, oldloc->inode, out);
+    GF_VALIDATE_OR_GOTO(this->name, newloc, out);
+
+    ret = svc_inode_ctx_get(this, oldloc->inode, &src_inode_type);
+    if (ret < 0) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                SVC_MSG_GET_INODE_CONTEXT_FAILED, "gfid=%s",
+                uuid_utoa(oldloc->inode->gfid), NULL);
+        goto out;
+    }
+
+    if (src_inode_type == VIRTUAL_INODE) {
+        op_ret = -1;
+        op_errno = EROFS;
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                SVC_MSG_RENAME_SNAPSHOT_ENTRY, "name=%s", oldloc->name, NULL);
+        goto out;
+    }
+
+    if (newloc->inode) {
+        ret = svc_inode_ctx_get(this, newloc->inode, &dst_inode_type);
+        if (!ret && dst_inode_type == VIRTUAL_INODE) {
+            op_ret = -1;
+            op_errno = EROFS;
+            gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                    SVC_MSG_RENAME_SNAPSHOT_ENTRY, "oldloc-name=%s",
+                    oldloc->name, "newloc-name=%s", newloc->name, NULL);
+            goto out;
+        }
+    }
+
+    if (dst_inode_type < 0) {
+        ret = svc_inode_ctx_get(this, newloc->parent, &dst_parent_type);
+        if (!ret && dst_parent_type == VIRTUAL_INODE) {
+            op_ret = -1;
+            op_errno = EROFS;
+            gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                    SVC_MSG_RENAME_SNAPSHOT_ENTRY, "oldloc-name=%s",
+                    oldloc->name, "newloc-name=%s", newloc->name, NULL);
+            goto out;
+        }
+    }
+
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->rename,
+                    oldloc, newloc, xdata);
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(rename, frame, op_ret, op_errno, NULL, NULL, NULL,
+                         NULL, NULL, NULL);
+    return 0;
+}
+
+/* Creating hardlinks for the files from the snapshot is not allowed as it
+   will be equivalent of creating hardlinks across different filesystems.
+   And so is vice versa.
+*/
+static int32_t
+gf_svc_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+            dict_t *xdata)
+{
+    int src_inode_type = -1;
+    int dst_parent_type = -1;
+    int32_t op_ret = -1;
+    int32_t op_errno = 0;
+    int32_t ret = -1;
+    gf_boolean_t wind = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, oldloc, out);
+    GF_VALIDATE_OR_GOTO(this->name, oldloc->inode, out);
+    GF_VALIDATE_OR_GOTO(this->name, newloc, out);
+
+    ret = svc_inode_ctx_get(this, oldloc->inode, &src_inode_type);
+    if (!ret && src_inode_type == VIRTUAL_INODE) {
+        op_ret = -1;
+        op_errno = EROFS;
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno, SVC_MSG_LINK_SNAPSHOT_ENTRY,
+                "oldloc-name=%s", oldloc->name, NULL);
+        goto out;
+    }
+
+    ret = svc_inode_ctx_get(this, newloc->parent, &dst_parent_type);
+    if (!ret && dst_parent_type == VIRTUAL_INODE) {
+        op_ret = -1;
+        op_errno = EROFS;
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno, SVC_MSG_LINK_SNAPSHOT_ENTRY,
+                "oldloc-name=%s", oldloc->name, "newloc-name=%s", newloc->name,
+                NULL);
+        goto out;
+    }
+
+    STACK_WIND_TAIL(frame, FIRST_CHILD(this), FIRST_CHILD(this)->fops->link,
+                    oldloc, newloc, xdata);
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(link, frame, op_ret, op_errno, NULL, NULL, NULL, NULL,
+                         NULL);
+    return 0;
+}
+
+static int32_t
+gf_svc_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   const char *name, dict_t *xdata)
+{
+    int ret = -1;
+    int inode_type = -1;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    gf_boolean_t wind = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    ret = svc_inode_ctx_get(this, loc->inode, &inode_type);
+    if (ret < 0) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                SVC_MSG_GET_INODE_CONTEXT_FAILED, "path=%s", loc->path,
+                "gfid=%s", uuid_utoa(loc->inode->gfid), NULL);
+        goto out;
+    }
+
+    if (inode_type == NORMAL_INODE) {
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->removexattr, loc, name, xdata);
+    } else {
+        op_ret = -1;
+        op_errno = EROFS;
+        goto out;
+    }
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(removexattr, frame, op_ret, op_errno, NULL);
+
+    return 0;
+}
+
+static int
+gf_svc_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int datasync,
+             dict_t *xdata)
+{
+    int inode_type = -1;
+    int ret = -1;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    gf_boolean_t wind = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd->inode, out);
+
+    ret = svc_inode_ctx_get(this, fd->inode, &inode_type);
+    if (ret < 0) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        gf_smsg(this->name, GF_LOG_ERROR, op_errno,
+                SVC_MSG_GET_INODE_CONTEXT_FAILED, "gfid=%s",
+                uuid_utoa(fd->inode->gfid), NULL);
+        goto out;
+    }
+
+    if (inode_type == NORMAL_INODE) {
+        STACK_WIND_TAIL(frame, FIRST_CHILD(this),
+                        FIRST_CHILD(this)->fops->fsync, fd, datasync, xdata);
+    } else {
+        op_ret = -1;
+        op_errno = EROFS;
+        goto out;
+    }
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(fsync, frame, op_ret, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+gf_svc_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    int32_t op_ret = -1;
+    int32_t op_errno = 0;
+    int ret = -1;
+    int inode_type = -1;
+    xlator_t *subvolume = NULL;
+    gf_boolean_t wind = _gf_false;
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd->inode, out);
+
+    SVC_GET_SUBVOL_FROM_CTX(this, op_ret, op_errno, inode_type, ret, fd->inode,
+                            subvolume, out);
+
+    STACK_WIND_TAIL(frame, subvolume, subvolume->fops->flush, fd, xdata);
+
+    wind = _gf_true;
+
+out:
+    if (!wind)
+        SVC_STACK_UNWIND(flush, frame, op_ret, op_errno, NULL);
+
+    return 0;
+}
+
+static int32_t
+gf_svc_releasedir(xlator_t *this, fd_t *fd)
+{
+    svc_fd_t *sfd = NULL;
+    uint64_t tmp_pfd = 0;
+    int ret = 0;
+
+    GF_VALIDATE_OR_GOTO("snapview-client", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    ret = fd_ctx_del(fd, this, &tmp_pfd);
+    if (ret < 0) {
+        gf_msg_debug(this->name, 0, "pfd from fd=%p is NULL", fd);
+        goto out;
+    }
+
+    GF_FREE(sfd);
+
+out:
+    return 0;
+}
+
+static int32_t
+gf_svc_forget(xlator_t *this, inode_t *inode)
+{
+    int ret = -1;
+    uint64_t value = 0;
+
+    GF_VALIDATE_OR_GOTO("svc", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+
+    ret = inode_ctx_del(inode, this, &value);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0,
+                SVC_MSG_DELETE_INODE_CONTEXT_FAILED, "gfid=%s",
+                uuid_utoa(inode->gfid), NULL);
+        goto out;
+    }
+
+out:
+    return 0;
+}
+
+static int
+gf_svc_priv_destroy(xlator_t *this, svc_private_t *priv)
+{
+    int ret = -1;
+
+    if (!priv) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, SVC_MSG_NULL_PRIV, NULL);
+        goto out;
+    }
+
+    GF_FREE(priv->path);
+    GF_FREE(priv->special_dir);
+
+    LOCK_DESTROY(&priv->lock);
+
+    GF_FREE(priv);
+
+    if (this->local_pool) {
+        mem_pool_destroy(this->local_pool);
+        this->local_pool = NULL;
+    }
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+/**
+ * ** NOTE **:
+ * =============
+ * The option "snapdir-entry-path" is NOT reconfigurable.
+ * That option as of now is only for the consumption of
+ * samba, where, it needs to tell glusterfs about the
+ * directory that is shared with windows client for the
+ * access. Now, in windows-explorer (GUI) interface, for
+ * the directory shared, the entry point to the snapshot
+ * world (snapshot-directory option) should be visible,
+ * atleast as a hidden entry. For that to happen, glusterfs
+ * has to send that entry in the readdir response coming on
+ * the directory used as the smb share. Therefore, samba,
+ * while initializing the gluster volume (via gfapi) sets
+ * the xlator option "snapdir-entry-path" to the directory
+ * which is to be shared with windows (check the file
+ * vfs_glusterfs.c from samba source code). So to avoid
+ * problems with smb access, not allowing snapdir-entry-path
+ * option to be configurable. That option is for those
+ * consumers who know what they are doing.
+ **/
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    svc_private_t *priv = NULL;
+    char *path = NULL;
+    gf_boolean_t show_entry_point = _gf_false;
+    char *tmp = NULL;
+
+    priv = this->private;
+
+    GF_OPTION_RECONF("snapshot-directory", path, options, str, out);
+    if (!path || (strlen(path) > NAME_MAX) || path[0] != '.') {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_INVALID_ENTRY_POINT,
+                "path=%s", path, NULL);
+        goto out;
+    }
+
+    GF_OPTION_RECONF("show-snapshot-directory", show_entry_point, options, bool,
+                     out);
+
+    /*
+     * The assumption now is that priv->path is an allocated memory (either
+     * in init or in a previous reconfigure).
+     * So, the intention here is to preserve the older contents of the option
+     * until the new option's value has been completely stored in the priv.
+     * So, do this.
+     *  - Store the pointer of priv->path in a temporary pointer.
+     *  - Allocate new memory for the new value of the option that is just
+     *    obtained from the above call to GF_OPTION_RECONF.
+     *  - If the above allocation fails, again set the pointer from priv
+     *    to the address stored in tmp. i.e. the previous value.
+     *  - If the allocation succeeds, then free the tmp pointer.
+     * WARNING: Before changing the allocation and freeing logic of
+     *          priv->path, always check the init function to see how
+     *          priv->path is set. Take decisions accordingly. As of now,
+     *          the assumption is that, the string elements of private
+     *          structure of snapview-client are allocated (either in
+     *          init or here in reconfugure).
+     */
+    LOCK(&priv->lock);
+    {
+        tmp = priv->path;
+        priv->path = NULL;
+        priv->path = gf_strdup(path);
+        if (!priv->path) {
+            gf_log(this->name, GF_LOG_ERROR,
+                   "failed to reconfigure snapshot-directory option to %s",
+                   path);
+            priv->path = tmp;
+        } else {
+            GF_FREE(tmp);
+            tmp = NULL;
+        }
+
+        priv->show_entry_point = show_entry_point;
+    }
+    UNLOCK(&priv->lock);
+
+out:
+    return 0;
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int32_t ret = -1;
+
+    if (!this)
+        return ret;
+
+    ret = xlator_mem_acct_init(this, gf_svc_mt_end + 1);
+
+    if (ret != 0) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, SVC_MSG_MEM_ACNT_FAILED, NULL);
+    }
+
+    return ret;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    svc_private_t *private = NULL;
+    int ret = -1;
+    int children = 0;
+    xlator_list_t *xl = NULL;
+    char *path = NULL;
+    char *special_dir = NULL;
+
+    if (!this->children) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_NO_CHILD_FOR_XLATOR, NULL);
+        goto out;
+    }
+
+    xl = this->children;
+    while (xl) {
+        children++;
+        xl = xl->next;
+    }
+
+    if (children != 2) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_XLATOR_CHILDREN_WRONG,
+                "subvol-num=%d", children, NULL);
+        goto out;
+    }
+
+    /* This can be the top of graph in certain cases */
+    if (!this->parents) {
+        gf_msg_debug(this->name, 0,
+                     "dangling volume. Check "
+                     "volfile");
+    }
+
+   private
+    = GF_CALLOC(1, sizeof(*private), gf_svc_mt_svc_private_t);
+    if (!private)
+        goto out;
+
+    LOCK_INIT(&private->lock);
+
+    GF_OPTION_INIT("snapshot-directory", path, str, out);
+    if (!path || (strlen(path) > NAME_MAX) || path[0] != '.') {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_INVALID_ENTRY_POINT,
+                "path=%s", path, NULL);
+        goto out;
+    }
+
+   private
+    ->path = gf_strdup(path);
+    if (!private->path) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_NO_MEMORY,
+                "entry-point-path=%s", path, NULL);
+        goto out;
+    }
+
+    GF_OPTION_INIT("snapdir-entry-path", special_dir, str, out);
+    if (!special_dir || strstr(special_dir, path)) {
+        if (special_dir)
+            gf_smsg(this->name, GF_LOG_ERROR, 0,
+                    SVC_MSG_ENTRY_POINT_SPECIAL_DIR, "path=%s", path,
+                    "special-dir=%s", special_dir);
+        else
+            gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_NULL_SPECIAL_DIR,
+                    NULL);
+        goto out;
+    }
+
+   private
+    ->special_dir = gf_strdup(special_dir);
+    if (!private->special_dir) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_NO_MEMORY,
+                "special-directory=%s", special_dir, NULL);
+        goto out;
+    }
+
+    GF_OPTION_INIT("show-snapshot-directory", private->show_entry_point, bool,
+                   out);
+
+    this->local_pool = mem_pool_new(svc_local_t, 128);
+    if (!this->local_pool) {
+        gf_smsg(this->name, GF_LOG_ERROR, 0, SVC_MSG_MEM_POOL_GET_FAILED, NULL);
+        goto out;
+    }
+
+    this->private = private;
+
+    ret = 0;
+
+out:
+    if (ret)
+        (void)gf_svc_priv_destroy(this, private);
+
+    return ret;
+}
+
+void
+fini(xlator_t *this)
+{
+    svc_private_t *priv = NULL;
+
+    if (!this)
+        return;
+
+    priv = this->private;
+    if (!priv)
+        return;
+
+    /*
+     * Just log the failure and go ahead to
+     * set this->priv to NULL.
+     */
+    if (gf_svc_priv_destroy(this, priv))
+        gf_smsg(this->name, GF_LOG_WARNING, 0, SVC_MSG_PRIV_DESTROY_FAILED,
+                NULL);
+
+    this->private = NULL;
+
+    return;
+}
+
+int
+notify(xlator_t *this, int event, void *data, ...)
+{
+    xlator_t *subvol = NULL;
+    int ret = 0;
+
+    subvol = data;
+
+    /* As there are two subvolumes in snapview-client, there is
+     * a possibility that the regular subvolume is still down and
+     * snapd subvolume come up first. So if we don't handle this situation
+     * CHILD_UP event will be propagated upwards to fuse when
+     * regular subvolume is still down.
+     * This can cause data unavailable for the application.
+     * So for now send notifications up only for regular subvolume.
+     *
+     * TODO: In future if required we may need to handle
+     * notifications from virtual subvolume
+     */
+    if (subvol != SECOND_CHILD(this))
+        ret = default_notify(this, event, data);
+
+    return ret;
+}
+
+struct xlator_fops fops = {
+    .lookup = gf_svc_lookup,
+    .opendir = gf_svc_opendir,
+    .stat = gf_svc_stat,
+    .fstat = gf_svc_fstat,
+    .statfs = gf_svc_statfs,
+    .rmdir = gf_svc_rmdir,
+    .rename = gf_svc_rename,
+    .mkdir = gf_svc_mkdir,
+    .open = gf_svc_open,
+    .unlink = gf_svc_unlink,
+    .setattr = gf_svc_setattr,
+    .getxattr = gf_svc_getxattr,
+    .setxattr = gf_svc_setxattr,
+    .fsetxattr = gf_svc_fsetxattr,
+    .readv = gf_svc_readv,
+    .readdir = gf_svc_readdir,
+    .readdirp = gf_svc_readdirp,
+    .create = gf_svc_create,
+    .readlink = gf_svc_readlink,
+    .mknod = gf_svc_mknod,
+    .symlink = gf_svc_symlink,
+    .flush = gf_svc_flush,
+    .link = gf_svc_link,
+    .access = gf_svc_access,
+    .removexattr = gf_svc_removexattr,
+    .fsync = gf_svc_fsync,
+};
+
+struct xlator_cbks cbks = {
+    .forget = gf_svc_forget,
+    .releasedir = gf_svc_releasedir,
+};
+
+struct volume_options options[] = {
+    {
+        .key = {"snapshot-directory"},
+        .type = GF_OPTION_TYPE_STR,
+        .default_value = ".snaps",
+    },
+    {
+        .key = {"snapdir-entry-path"},
+        .type = GF_OPTION_TYPE_STR,
+        .description = "An option to set the path of a directory on which "
+                       "when readdir comes, dentry for the snapshot-directory"
+                       " should be created and added in the readdir response",
+        .default_value = "",
+    },
+    {
+        .key = {"show-snapshot-directory"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .description = "If this option is set, and the option "
+                       "\"snapdir-entry-path\" is set (which is set by samba "
+                       "vfs plugin for glusterfs, then send the entry point "
+                       "when readdir comes on the snapdir-entry-path",
+        .default_value = "off",
+    },
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1},
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "snapview-client",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/features/snapview-client/src/snapview-client.h b/xlators/features/snapview-client/src/snapview-client.h
new file mode 100644
index 00000000000..166116a439d
--- /dev/null
+++ b/xlators/features/snapview-client/src/snapview-client.h
@@ -0,0 +1,101 @@
+/*
+  Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+#ifndef __SNAP_VIEW_CLIENT_H__
+#define __SNAP_VIEW_CLIENT_H__
+
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/dict.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+#include "snapview-client-mem-types.h"
+#include "snapview-client-messages.h"
+
+struct __svc_local {
+    loc_t loc;
+    xlator_t *subvolume;
+    fd_t *fd;
+    void *cookie;
+    dict_t *xdata;
+    uint16_t revalidate;
+};
+typedef struct __svc_local svc_local_t;
+
+#define SVC_STACK_UNWIND(fop, frame, params...)                                \
+    do {                                                                       \
+        svc_local_t *__local = NULL;                                           \
+        if (frame) {                                                           \
+            __local = frame->local;                                            \
+            frame->local = NULL;                                               \
+        }                                                                      \
+        STACK_UNWIND_STRICT(fop, frame, params);                               \
+        svc_local_free(__local);                                               \
+    } while (0)
+
+#define SVC_ENTRY_POINT_SET(this, xdata, op_ret, op_errno, new_xdata, ret,     \
+                            label)                                             \
+    do {                                                                       \
+        if (!xdata) {                                                          \
+            xdata = new_xdata = dict_new();                                    \
+            if (!new_xdata) {                                                  \
+                gf_log(this->name, GF_LOG_ERROR,                               \
+                       "failed to allocate new dict");                         \
+                op_ret = -1;                                                   \
+                op_errno = ENOMEM;                                             \
+                goto label;                                                    \
+            }                                                                  \
+        }                                                                      \
+        ret = dict_set_str(xdata, "entry-point", "true");                      \
+        if (ret) {                                                             \
+            gf_log(this->name, GF_LOG_ERROR, "failed to set dict");            \
+            op_ret = -1;                                                       \
+            op_errno = ENOMEM;                                                 \
+            goto label;                                                        \
+        }                                                                      \
+    } while (0);
+
+#define SVC_GET_SUBVOL_FROM_CTX(this, op_ret, op_errno, inode_type, ret,       \
+                                inode, subvolume, label)                       \
+    do {                                                                       \
+        ret = svc_inode_ctx_get(this, inode, &inode_type);                     \
+        if (ret < 0) {                                                         \
+            gf_log(this->name, GF_LOG_ERROR,                                   \
+                   "inode context not found for gfid %s",                      \
+                   uuid_utoa(inode->gfid));                                    \
+            op_ret = -1;                                                       \
+            op_errno = EINVAL;                                                 \
+            goto label;                                                        \
+        }                                                                      \
+                                                                               \
+        subvolume = svc_get_subvolume(this, inode_type);                       \
+    } while (0);
+
+struct svc_private {
+    char *path;
+    char *special_dir; /* needed for samba */
+    gf_boolean_t show_entry_point;
+    gf_lock_t lock; /* mainly to guard private->path */
+};
+typedef struct svc_private svc_private_t;
+
+struct svc_fd {
+    off_t last_offset;
+    gf_boolean_t entry_point_handled;
+    gf_boolean_t special_dir;
+};
+typedef struct svc_fd svc_fd_t;
+
+typedef enum { NORMAL_INODE = 1, VIRTUAL_INODE } inode_type_t;
+
+int
+gf_svc_special_dir_revalidate_lookup(call_frame_t *frame, xlator_t *this,
+                                     dict_t *xdata);
+
+#endif /* __SNAP_VIEW_CLIENT_H__ */
diff --git a/xlators/features/snapview-server/Makefile.am b/xlators/features/snapview-server/Makefile.am
new file mode 100644
index 00000000000..af437a64d6d
--- /dev/null
+++ b/xlators/features/snapview-server/Makefile.am
@@ -0,0 +1 @@
+SUBDIRS = src
diff --git a/xlators/features/snapview-server/src/Makefile.am b/xlators/features/snapview-server/src/Makefile.am
new file mode 100644
index 00000000000..2935f138a4c
--- /dev/null
+++ b/xlators/features/snapview-server/src/Makefile.am
@@ -0,0 +1,25 @@
+if WITH_SERVER
+xlator_LTLIBRARIES = snapview-server.la
+endif
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+snapview_server_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+snapview_server_la_SOURCES = snapview-server.c snapview-server-mgmt.c \
+	snapview-server-helpers.c
+
+snapview_server_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la \
+	$(top_builddir)/api/src/libgfapi.la \
+	$(RLLIBS) $(top_builddir)/rpc/xdr/src/libgfxdr.la \
+	$(top_builddir)/rpc/rpc-lib/src/libgfrpc.la
+
+noinst_HEADERS = snapview-server.h snapview-server-mem-types.h snapview-server-messages.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/api/src -I$(top_srcdir)/rpc/rpc-lib/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src \
+	-DDATADIR=\"$(localstatedir)\"
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/snapview-server/src/snapview-server-helpers.c b/xlators/features/snapview-server/src/snapview-server-helpers.c
new file mode 100644
index 00000000000..62c1ddac49c
--- /dev/null
+++ b/xlators/features/snapview-server/src/snapview-server-helpers.c
@@ -0,0 +1,715 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include "snapview-server.h"
+#include "snapview-server-mem-types.h"
+
+#include <glusterfs/xlator.h>
+#include "rpc-clnt.h"
+#include "xdr-generic.h"
+#include "protocol-common.h"
+#include <pthread.h>
+
+int
+__svs_inode_ctx_set(xlator_t *this, inode_t *inode, svs_inode_t *svs_inode)
+{
+    uint64_t value = 0;
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+    GF_VALIDATE_OR_GOTO(this->name, svs_inode, out);
+
+    value = (uint64_t)(long)svs_inode;
+
+    ret = __inode_ctx_set(inode, this, &value);
+
+out:
+    return ret;
+}
+
+svs_inode_t *
+__svs_inode_ctx_get(xlator_t *this, inode_t *inode)
+{
+    svs_inode_t *svs_inode = NULL;
+    uint64_t value = 0;
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+
+    ret = __inode_ctx_get(inode, this, &value);
+    if (ret)
+        goto out;
+
+    svs_inode = (svs_inode_t *)((long)value);
+
+out:
+    return svs_inode;
+}
+
+svs_inode_t *
+svs_inode_ctx_get(xlator_t *this, inode_t *inode)
+{
+    svs_inode_t *svs_inode = NULL;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+
+    LOCK(&inode->lock);
+    {
+        svs_inode = __svs_inode_ctx_get(this, inode);
+    }
+    UNLOCK(&inode->lock);
+
+out:
+    return svs_inode;
+}
+
+int32_t
+svs_inode_ctx_set(xlator_t *this, inode_t *inode, svs_inode_t *svs_inode)
+{
+    int32_t ret = -1;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+    GF_VALIDATE_OR_GOTO(this->name, svs_inode, out);
+
+    LOCK(&inode->lock);
+    {
+        ret = __svs_inode_ctx_set(this, inode, svs_inode);
+    }
+    UNLOCK(&inode->lock);
+
+out:
+    return ret;
+}
+
+svs_inode_t *
+svs_inode_new(void)
+{
+    svs_inode_t *svs_inode = NULL;
+
+    svs_inode = GF_CALLOC(1, sizeof(*svs_inode), gf_svs_mt_svs_inode_t);
+
+    return svs_inode;
+}
+
+svs_inode_t *
+svs_inode_ctx_get_or_new(xlator_t *this, inode_t *inode)
+{
+    svs_inode_t *svs_inode = NULL;
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+
+    LOCK(&inode->lock);
+    {
+        svs_inode = __svs_inode_ctx_get(this, inode);
+        if (!svs_inode) {
+            svs_inode = svs_inode_new();
+            if (svs_inode) {
+                ret = __svs_inode_ctx_set(this, inode, svs_inode);
+                if (ret) {
+                    GF_FREE(svs_inode);
+                    svs_inode = NULL;
+                }
+            }
+        }
+    }
+    UNLOCK(&inode->lock);
+
+out:
+    return svs_inode;
+}
+
+svs_fd_t *
+svs_fd_new(void)
+{
+    svs_fd_t *svs_fd = NULL;
+
+    svs_fd = GF_CALLOC(1, sizeof(*svs_fd), gf_svs_mt_svs_fd_t);
+
+    return svs_fd;
+}
+
+int
+__svs_fd_ctx_set(xlator_t *this, fd_t *fd, svs_fd_t *svs_fd)
+{
+    uint64_t value = 0;
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+    GF_VALIDATE_OR_GOTO(this->name, svs_fd, out);
+
+    value = (uint64_t)(long)svs_fd;
+
+    ret = __fd_ctx_set(fd, this, value);
+
+out:
+    return ret;
+}
+
+svs_fd_t *
+__svs_fd_ctx_get(xlator_t *this, fd_t *fd)
+{
+    svs_fd_t *svs_fd = NULL;
+    uint64_t value = 0;
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    ret = __fd_ctx_get(fd, this, &value);
+    if (ret)
+        return NULL;
+
+    svs_fd = (svs_fd_t *)((long)value);
+
+out:
+    return svs_fd;
+}
+
+svs_fd_t *
+svs_fd_ctx_get(xlator_t *this, fd_t *fd)
+{
+    svs_fd_t *svs_fd = NULL;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    LOCK(&fd->lock);
+    {
+        svs_fd = __svs_fd_ctx_get(this, fd);
+    }
+    UNLOCK(&fd->lock);
+
+out:
+    return svs_fd;
+}
+
+int32_t
+svs_fd_ctx_set(xlator_t *this, fd_t *fd, svs_fd_t *svs_fd)
+{
+    int32_t ret = -1;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+    GF_VALIDATE_OR_GOTO(this->name, svs_fd, out);
+
+    LOCK(&fd->lock);
+    {
+        ret = __svs_fd_ctx_set(this, fd, svs_fd);
+    }
+    UNLOCK(&fd->lock);
+
+out:
+    return ret;
+}
+
+svs_fd_t *
+__svs_fd_ctx_get_or_new(xlator_t *this, fd_t *fd)
+{
+    svs_fd_t *svs_fd = NULL;
+    int ret = -1;
+    glfs_t *fs = NULL;
+    glfs_object_t *object = NULL;
+    svs_inode_t *inode_ctx = NULL;
+    glfs_fd_t *glfd = NULL;
+    inode_t *inode = NULL;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    inode = fd->inode;
+    svs_fd = __svs_fd_ctx_get(this, fd);
+    if (svs_fd) {
+        ret = 0;
+        goto out;
+    }
+
+    svs_fd = svs_fd_new();
+    if (!svs_fd) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SVS_MSG_NEW_FD_CTX_FAILED,
+               "failed to allocate new fd "
+               "context for gfid %s",
+               uuid_utoa(inode->gfid));
+        goto out;
+    }
+
+    if (fd_is_anonymous(fd)) {
+        inode_ctx = svs_inode_ctx_get(this, inode);
+        if (!inode_ctx) {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   SVS_MSG_GET_INODE_CONTEXT_FAILED,
+                   "failed to get inode "
+                   "context for %s",
+                   uuid_utoa(inode->gfid));
+            goto out;
+        }
+
+        fs = inode_ctx->fs;
+        object = inode_ctx->object;
+
+        if (inode->ia_type == IA_IFDIR) {
+            glfd = glfs_h_opendir(fs, object);
+            if (!glfd) {
+                gf_msg(this->name, GF_LOG_ERROR, errno, SVS_MSG_OPENDIR_FAILED,
+                       "failed to "
+                       "open the directory %s",
+                       uuid_utoa(inode->gfid));
+                goto out;
+            }
+        }
+
+        if (inode->ia_type == IA_IFREG) {
+            glfd = glfs_h_open(fs, object, O_RDONLY | O_LARGEFILE);
+            if (!glfd) {
+                gf_msg(this->name, GF_LOG_ERROR, errno, SVS_MSG_OPEN_FAILED,
+                       "failed to "
+                       "open the file %s",
+                       uuid_utoa(inode->gfid));
+                goto out;
+            }
+        }
+
+        svs_fd->fd = glfd;
+    }
+
+    ret = __svs_fd_ctx_set(this, fd, svs_fd);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SVS_MSG_SET_FD_CONTEXT_FAILED,
+               "failed to set fd context "
+               "for gfid %s",
+               uuid_utoa(inode->gfid));
+        if (svs_fd->fd) {
+            if (inode->ia_type == IA_IFDIR) {
+                ret = glfs_closedir(svs_fd->fd);
+                if (ret)
+                    gf_msg(this->name, GF_LOG_ERROR, errno,
+                           SVS_MSG_CLOSEDIR_FAILED,
+                           "failed to close the fd for %s",
+                           uuid_utoa(inode->gfid));
+            }
+            if (inode->ia_type == IA_IFREG) {
+                ret = glfs_close(svs_fd->fd);
+                if (ret)
+                    gf_msg(this->name, GF_LOG_ERROR, 0, SVS_MSG_CLOSE_FAILED,
+                           "failed to close the fd for %s",
+                           uuid_utoa(inode->gfid));
+            }
+        }
+        ret = -1;
+    }
+
+out:
+    if (ret) {
+        GF_FREE(svs_fd);
+        svs_fd = NULL;
+    }
+
+    return svs_fd;
+}
+
+svs_fd_t *
+svs_fd_ctx_get_or_new(xlator_t *this, fd_t *fd)
+{
+    svs_fd_t *svs_fd = NULL;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    LOCK(&fd->lock);
+    {
+        svs_fd = __svs_fd_ctx_get_or_new(this, fd);
+    }
+    UNLOCK(&fd->lock);
+
+out:
+    return svs_fd;
+}
+
+int
+svs_uuid_generate(xlator_t *this, uuid_t gfid, char *snapname,
+                  uuid_t origin_gfid)
+{
+    char ino_string[NAME_MAX + 32] = "";
+    uuid_t tmp = {
+        0,
+    };
+    int ret = -1;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, snapname, out);
+
+    (void)snprintf(ino_string, sizeof(ino_string), "%s%s", snapname,
+                   uuid_utoa(origin_gfid));
+
+    if (gf_gfid_generate_from_xxh64(tmp, ino_string)) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, SVS_MSG_GFID_GEN_FAILED,
+               "failed to generate "
+               "gfid for object with actual gfid of %s "
+               "(snapname: %s, key: %s)",
+               uuid_utoa(origin_gfid), snapname, ino_string);
+        goto out;
+    }
+
+    gf_uuid_copy(gfid, tmp);
+
+    ret = 0;
+
+    gf_msg_debug(this->name, 0, "gfid generated is %s ", uuid_utoa(gfid));
+
+out:
+    return ret;
+}
+
+void
+svs_fill_ino_from_gfid(struct iatt *buf)
+{
+    xlator_t *this = NULL;
+
+    this = THIS;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, buf, out);
+
+    /* consider least significant 8 bytes of value out of gfid */
+    if (gf_uuid_is_null(buf->ia_gfid)) {
+        buf->ia_ino = -1;
+        goto out;
+    }
+
+    buf->ia_ino = gfid_to_ino(buf->ia_gfid);
+out:
+    return;
+}
+
+void
+svs_iatt_fill(uuid_t gfid, struct iatt *buf)
+{
+    struct timeval tv = {
+        0,
+    };
+    xlator_t *this = NULL;
+
+    this = THIS;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, buf, out);
+
+    buf->ia_type = IA_IFDIR;
+    buf->ia_uid = 0;
+    buf->ia_gid = 0;
+    buf->ia_size = 0;
+    buf->ia_nlink = 2;
+    buf->ia_blocks = 8;
+    buf->ia_size = 4096;
+
+    gf_uuid_copy(buf->ia_gfid, gfid);
+    svs_fill_ino_from_gfid(buf);
+
+    buf->ia_prot = ia_prot_from_st_mode(0755);
+
+    gettimeofday(&tv, 0);
+
+    buf->ia_mtime = buf->ia_atime = buf->ia_ctime = tv.tv_sec;
+    buf->ia_mtime_nsec = buf->ia_atime_nsec = buf->ia_ctime_nsec = (tv.tv_usec *
+                                                                    1000);
+
+out:
+    return;
+}
+
+/* priv->snaplist_lock should be held before calling this function */
+snap_dirent_t *
+__svs_get_snap_dirent(xlator_t *this, const char *name)
+{
+    svs_private_t *private = NULL;
+    int i = 0;
+    snap_dirent_t *dirents = NULL;
+    snap_dirent_t *tmp_dirent = NULL;
+    snap_dirent_t *dirent = NULL;
+
+   private
+    = this->private;
+
+    dirents = private->dirents;
+    if (!dirents) {
+        goto out;
+    }
+
+    tmp_dirent = dirents;
+    for (i = 0; i < private->num_snaps; i++) {
+        if (!strcmp(tmp_dirent->name, name)) {
+            dirent = tmp_dirent;
+            break;
+        }
+        tmp_dirent++;
+    }
+
+out:
+    return dirent;
+}
+
+glfs_t *
+__svs_initialise_snapshot_volume(xlator_t *this, const char *name,
+                                 int32_t *op_errno)
+{
+    svs_private_t *priv = NULL;
+    int32_t ret = -1;
+    int32_t local_errno = ESTALE;
+    snap_dirent_t *dirent = NULL;
+    char volname[PATH_MAX] = {
+        0,
+    };
+    glfs_t *fs = NULL;
+    int loglevel = GF_LOG_INFO;
+    char logfile[PATH_MAX] = {
+        0,
+    };
+    char *volfile_server = NULL;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+    GF_VALIDATE_OR_GOTO(this->name, name, out);
+
+    priv = this->private;
+
+    dirent = __svs_get_snap_dirent(this, name);
+    if (!dirent) {
+        gf_msg_debug(this->name, 0,
+                     "snap entry for "
+                     "name %s not found",
+                     name);
+        local_errno = ENOENT;
+        goto out;
+    }
+
+    if (dirent->fs) {
+        ret = 0;
+        fs = dirent->fs;
+        goto out;
+    }
+
+    snprintf(volname, sizeof(volname), "/snaps/%s/%s/%s", dirent->name,
+             dirent->snap_volname, dirent->snap_volname);
+
+    fs = glfs_new(volname);
+    if (!fs) {
+        local_errno = ENOMEM;
+        gf_msg(this->name, GF_LOG_ERROR, local_errno, SVS_MSG_GLFS_NEW_FAILED,
+               "glfs instance for snap volume %s "
+               "failed",
+               dirent->name);
+        goto out;
+    }
+
+    /*
+     * Before, localhost was used as the volfile server. But, with that
+     * method, accessing snapshots started giving ENOENT error if a
+     * specific bind address is mentioned in the glusterd volume file.
+     * Check the bug https://bugzilla.redhat.com/show_bug.cgi?id=1725211.
+     * So, the new method is tried below, where, snapview-server first
+     * uses the volfile server used by the snapd (obtained from the
+     * command line arguments saved in the global context of the process).
+     * If the volfile server in global context is NULL, then localhost
+     * is tried (like before).
+     */
+    if (this->ctx->cmd_args.volfile_server) {
+        volfile_server = gf_strdup(this->ctx->cmd_args.volfile_server);
+        if (!volfile_server) {
+            gf_msg(this->name, GF_LOG_WARNING, ENOMEM,
+                   SVS_MSG_VOLFILE_SERVER_GET_FAIL,
+                   "failed to copy volfile server %s. ",
+                   this->ctx->cmd_args.volfile_server);
+            ret = -1;
+            goto out;
+        }
+    } else {
+        gf_msg(this->name, GF_LOG_WARNING, ENOMEM,
+               SVS_MSG_VOLFILE_SERVER_GET_FAIL,
+               "volfile server is NULL in cmd args. "
+               "Trying with localhost");
+        volfile_server = gf_strdup("localhost");
+        if (!volfile_server) {
+            gf_msg(this->name, GF_LOG_WARNING, ENOMEM,
+                   SVS_MSG_VOLFILE_SERVER_GET_FAIL,
+                   "failed to copy volfile server localhost.");
+            ret = -1;
+            goto out;
+        }
+    }
+
+    ret = glfs_set_volfile_server(fs, "tcp", volfile_server, 24007);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, local_errno,
+               SVS_MSG_SET_VOLFILE_SERVR_FAILED,
+               "setting the "
+               "volfile server %s for snap volume %s "
+               "failed",
+               volfile_server, dirent->name);
+        goto out;
+    }
+
+    snprintf(logfile, sizeof(logfile),
+             DEFAULT_SVD_LOG_FILE_DIRECTORY "/snaps/%s/%s-%s.log",
+             priv->volname, name, dirent->uuid);
+
+    ret = glfs_set_logging(fs, logfile, loglevel);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, local_errno,
+               SVS_MSG_SET_LOGGING_FAILED,
+               "failed to set the "
+               "log file path");
+        goto out;
+    }
+
+    ret = glfs_init(fs);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, local_errno, SVS_MSG_GLFS_INIT_FAILED,
+               "initing the "
+               "fs for %s failed",
+               dirent->name);
+        goto out;
+    }
+
+    ret = 0;
+
+out:
+    if (ret) {
+        if (op_errno)
+            *op_errno = local_errno;
+
+        if (fs)
+            glfs_fini(fs);
+        fs = NULL;
+    }
+
+    if (fs) {
+        dirent->fs = fs;
+    }
+
+    GF_FREE(volfile_server);
+    return fs;
+}
+
+glfs_t *
+svs_initialise_snapshot_volume(xlator_t *this, const char *name,
+                               int32_t *op_errno)
+{
+    glfs_t *fs = NULL;
+    svs_private_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+    GF_VALIDATE_OR_GOTO(this->name, name, out);
+
+    priv = this->private;
+
+    LOCK(&priv->snaplist_lock);
+    {
+        fs = __svs_initialise_snapshot_volume(this, name, op_errno);
+    }
+    UNLOCK(&priv->snaplist_lock);
+
+out:
+
+    return fs;
+}
+
+snap_dirent_t *
+svs_get_latest_snap_entry(xlator_t *this)
+{
+    svs_private_t *priv = NULL;
+    snap_dirent_t *dirents = NULL;
+    snap_dirent_t *dirent = NULL;
+
+    GF_VALIDATE_OR_GOTO("svs", this, out);
+
+    priv = this->private;
+
+    LOCK(&priv->snaplist_lock);
+    {
+        dirents = priv->dirents;
+        if (!dirents) {
+            goto unlock;
+        }
+        if (priv->num_snaps)
+            dirent = &dirents[priv->num_snaps - 1];
+    }
+unlock:
+    UNLOCK(&priv->snaplist_lock);
+
+out:
+    return dirent;
+}
+
+glfs_t *
+svs_get_latest_snapshot(xlator_t *this)
+{
+    glfs_t *fs = NULL;
+    snap_dirent_t *dirent = NULL;
+    svs_private_t *priv = NULL;
+
+    GF_VALIDATE_OR_GOTO("svs", this, out);
+    priv = this->private;
+
+    dirent = svs_get_latest_snap_entry(this);
+
+    if (dirent) {
+        LOCK(&priv->snaplist_lock);
+        {
+            fs = dirent->fs;
+        }
+        UNLOCK(&priv->snaplist_lock);
+    }
+
+out:
+    return fs;
+}
+
+glfs_t *
+svs_inode_ctx_glfs_mapping(xlator_t *this, svs_inode_t *inode_ctx)
+{
+    glfs_t *fs = NULL;
+
+    GF_VALIDATE_OR_GOTO("svs", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode_ctx, out);
+
+    fs = inode_ctx->fs;
+
+    SVS_CHECK_VALID_SNAPSHOT_HANDLE(fs, this);
+
+out:
+    return fs;
+}
+
+glfs_t *
+svs_inode_glfs_mapping(xlator_t *this, inode_t *inode)
+{
+    svs_inode_t *inode_ctx = NULL;
+    glfs_t *fs = NULL;
+
+    inode_ctx = svs_inode_ctx_get(this, inode);
+    if (!inode_ctx) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SVS_MSG_GET_INODE_CONTEXT_FAILED,
+               "inode context not found for"
+               " the inode %s",
+               uuid_utoa(inode->gfid));
+        goto out;
+    }
+
+    fs = svs_inode_ctx_glfs_mapping(this, inode_ctx);
+
+out:
+    return fs;
+}
diff --git a/xlators/features/snapview-server/src/snapview-server-mem-types.h b/xlators/features/snapview-server/src/snapview-server-mem-types.h
new file mode 100644
index 00000000000..63456b85323
--- /dev/null
+++ b/xlators/features/snapview-server/src/snapview-server-mem-types.h
@@ -0,0 +1,25 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __SNAP_VIEW_MEM_TYPES_H
+#define __SNAP_VIEW_MEM_TYPES_H
+
+#include <glusterfs/mem-types.h>
+
+enum snapview_mem_types {
+    gf_svs_mt_priv_t = gf_common_mt_end + 1,
+    gf_svs_mt_svs_inode_t,
+    gf_svs_mt_dirents_t,
+    gf_svs_mt_svs_fd_t,
+    gf_svs_mt_snaplist_t,
+    gf_svs_mt_end
+};
+
+#endif
diff --git a/xlators/features/snapview-server/src/snapview-server-messages.h b/xlators/features/snapview-server/src/snapview-server-messages.h
new file mode 100644
index 00000000000..f634ab5d2b0
--- /dev/null
+++ b/xlators/features/snapview-server/src/snapview-server-messages.h
@@ -0,0 +1,54 @@
+/*
+ Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com>
+ This file is part of GlusterFS.
+
+ This file is licensed to you under your choice of the GNU Lesser
+ General Public License, version 3 or any later version (LGPLv3 or
+ later), or the GNU General Public License, version 2 (GPLv2), in all
+ cases as published by the Free Software Foundation.
+ */
+
+#ifndef _SNAPVIEW_SERVER_MESSAGES_H_
+#define _SNAPVIEW_SERVER_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(SNAPVIEW_SERVER, SVS_MSG_NO_MEMORY, SVS_MSG_MEM_ACNT_FAILED,
+           SVS_MSG_NULL_GFID, SVS_MSG_GET_LATEST_SNAP_FAILED,
+           SVS_MSG_INVALID_GLFS_CTX, SVS_MSG_LOCK_DESTROY_FAILED,
+           SVS_MSG_SNAPSHOT_LIST_CHANGED, SVS_MSG_MGMT_INIT_FAILED,
+           SVS_MSG_GET_SNAPSHOT_LIST_FAILED, SVS_MSG_GET_GLFS_H_OBJECT_FAILED,
+           SVS_MSG_PARENT_CTX_OR_NAME_NULL, SVS_MSG_SET_INODE_CONTEXT_FAILED,
+           SVS_MSG_GET_INODE_CONTEXT_FAILED, SVS_MSG_NEW_INODE_CTX_FAILED,
+           SVS_MSG_DELETE_INODE_CONTEXT_FAILED, SVS_MSG_SET_FD_CONTEXT_FAILED,
+           SVS_MSG_NEW_FD_CTX_FAILED, SVS_MSG_DELETE_FD_CTX_FAILED,
+           SVS_MSG_GETXATTR_FAILED, SVS_MSG_LISTXATTR_FAILED,
+           SVS_MSG_RELEASEDIR_FAILED, SVS_MSG_RELEASE_FAILED,
+           SVS_MSG_TELLDIR_FAILED, SVS_MSG_STAT_FAILED, SVS_MSG_STATFS_FAILED,
+           SVS_MSG_OPEN_FAILED, SVS_MSG_READ_FAILED, SVS_MSG_READLINK_FAILED,
+           SVS_MSG_ACCESS_FAILED, SVS_MSG_GET_FD_CONTEXT_FAILED,
+           SVS_MSG_DICT_SET_FAILED, SVS_MSG_OPENDIR_FAILED,
+           SVS_MSG_FS_INSTANCE_INVALID, SVS_MSG_SETFSUID_FAIL,
+           SVS_MSG_SETFSGID_FAIL, SVS_MSG_SETFSGRPS_FAIL,
+           SVS_MSG_BUILD_TRNSPRT_OPT_FAILED, SVS_MSG_RPC_INIT_FAILED,
+           SVS_MSG_REG_NOTIFY_FAILED, SVS_MSG_REG_CBK_PRGM_FAILED,
+           SVS_MSG_RPC_CLNT_START_FAILED, SVS_MSG_XDR_PAYLOAD_FAILED,
+           SVS_MSG_NULL_CTX, SVS_MSG_RPC_CALL_FAILED, SVS_MSG_XDR_DECODE_FAILED,
+           SVS_MSG_RSP_DICT_EMPTY, SVS_MSG_DICT_GET_FAILED,
+           SVS_MSG_SNAP_LIST_REFRESH_FAILED, SVS_MSG_RPC_REQ_FAILED,
+           SVS_MSG_CLOSEDIR_FAILED, SVS_MSG_CLOSE_FAILED,
+           SVS_MSG_GFID_GEN_FAILED, SVS_MSG_GLFS_NEW_FAILED,
+           SVS_MSG_SET_VOLFILE_SERVR_FAILED, SVS_MSG_SET_LOGGING_FAILED,
+           SVS_MSG_VOLFILE_SERVER_GET_FAIL, SVS_MSG_GLFS_INIT_FAILED);
+
+#endif /* !_SNAPVIEW_CLIENT_MESSAGES_H_ */
diff --git a/xlators/features/snapview-server/src/snapview-server-mgmt.c b/xlators/features/snapview-server/src/snapview-server-mgmt.c
new file mode 100644
index 00000000000..ecf31c3b880
--- /dev/null
+++ b/xlators/features/snapview-server/src/snapview-server-mgmt.c
@@ -0,0 +1,524 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include "snapview-server.h"
+#include "snapview-server-mem-types.h"
+#include <pthread.h>
+
+int
+mgmt_cbk_snap(struct rpc_clnt *rpc, void *mydata, void *data)
+{
+    xlator_t *this = NULL;
+
+    this = mydata;
+    GF_ASSERT(this);
+
+    gf_msg("mgmt", GF_LOG_INFO, 0, SVS_MSG_SNAPSHOT_LIST_CHANGED,
+           "list of snapshots changed");
+
+    svs_get_snapshot_list(this);
+    return 0;
+}
+
+static rpcclnt_cb_actor_t svs_cbk_actors[GF_CBK_MAXVALUE] = {
+    [GF_CBK_GET_SNAPS] = {"GETSNAPS", mgmt_cbk_snap, GF_CBK_GET_SNAPS},
+};
+
+static struct rpcclnt_cb_program svs_cbk_prog = {
+    .progname = "GlusterFS Callback",
+    .prognum = GLUSTER_CBK_PROGRAM,
+    .progver = GLUSTER_CBK_VERSION,
+    .actors = svs_cbk_actors,
+    .numactors = GF_CBK_MAXVALUE,
+};
+
+static char *clnt_handshake_procs[GF_HNDSK_MAXVALUE] = {
+    [GF_HNDSK_NULL] = "NULL",
+    [GF_HNDSK_EVENT_NOTIFY] = "EVENTNOTIFY",
+};
+
+static rpc_clnt_prog_t svs_clnt_handshake_prog = {
+    .progname = "GlusterFS Handshake",
+    .prognum = GLUSTER_HNDSK_PROGRAM,
+    .progver = GLUSTER_HNDSK_VERSION,
+    .procnames = clnt_handshake_procs,
+};
+
+static int
+svs_rpc_notify(struct rpc_clnt *rpc, void *mydata, rpc_clnt_event_t event,
+               void *data)
+{
+    xlator_t *this = NULL;
+    int ret = 0;
+
+    this = mydata;
+
+    switch (event) {
+        case RPC_CLNT_CONNECT:
+            ret = svs_get_snapshot_list(this);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, EINVAL,
+                       SVS_MSG_GET_SNAPSHOT_LIST_FAILED,
+                       "Error in refreshing the snaplist "
+                       "infrastructure");
+                ret = -1;
+            }
+            break;
+        default:
+            break;
+    }
+    return ret;
+}
+
+int
+svs_mgmt_init(xlator_t *this)
+{
+    int ret = -1;
+    svs_private_t *priv = NULL;
+    dict_t *options = NULL;
+    int port = GF_DEFAULT_BASE_PORT;
+    char *host = NULL;
+    cmd_args_t *cmd_args = NULL;
+    glusterfs_ctx_t *ctx = NULL;
+    xlator_cmdline_option_t *opt = NULL;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->ctx, out);
+
+    priv = this->private;
+
+    ctx = this->ctx;
+    cmd_args = &ctx->cmd_args;
+
+    host = "localhost";
+    if (cmd_args->volfile_server)
+        host = cmd_args->volfile_server;
+
+    options = dict_new();
+    if (!options)
+        goto out;
+
+    opt = find_xlator_option_in_cmd_args_t("address-family", cmd_args);
+    ret = rpc_transport_inet_options_build(options, host, port,
+                                           (opt != NULL ? opt->value : NULL));
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SVS_MSG_BUILD_TRNSPRT_OPT_FAILED,
+               "failed to build the "
+               "transport options");
+        goto out;
+    }
+
+    priv->rpc = rpc_clnt_new(options, this, this->name, 8);
+    if (!priv->rpc) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SVS_MSG_RPC_INIT_FAILED,
+               "failed to initialize RPC");
+        goto out;
+    }
+
+    ret = rpc_clnt_register_notify(priv->rpc, svs_rpc_notify, this);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, SVS_MSG_REG_NOTIFY_FAILED,
+               "failed to register notify function");
+        goto out;
+    }
+
+    ret = rpcclnt_cbk_program_register(priv->rpc, &svs_cbk_prog, this);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SVS_MSG_REG_CBK_PRGM_FAILED,
+               "failed to register callback program");
+        goto out;
+    }
+
+    ret = rpc_clnt_start(priv->rpc);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SVS_MSG_RPC_CLNT_START_FAILED,
+               "failed to start the rpc "
+               "client");
+        goto out;
+    }
+
+    ret = 0;
+
+    gf_msg_debug(this->name, 0, "svs mgmt init successful");
+
+out:
+    if (options)
+        dict_unref(options);
+    if (ret)
+        if (priv) {
+            rpc_clnt_connection_cleanup(&priv->rpc->conn);
+            rpc_clnt_unref(priv->rpc);
+            priv->rpc = NULL;
+        }
+
+    return ret;
+}
+
+int
+svs_mgmt_submit_request(void *req, call_frame_t *frame, glusterfs_ctx_t *ctx,
+                        rpc_clnt_prog_t *prog, int procnum, fop_cbk_fn_t cbkfn,
+                        xdrproc_t xdrproc)
+{
+    int ret = -1;
+    int count = 0;
+    struct iovec iov = {
+        0,
+    };
+    struct iobuf *iobuf = NULL;
+    struct iobref *iobref = NULL;
+    ssize_t xdr_size = 0;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", frame, out);
+    GF_VALIDATE_OR_GOTO("snapview-server", req, out);
+    GF_VALIDATE_OR_GOTO("snapview-server", ctx, out);
+    GF_VALIDATE_OR_GOTO("snapview-server", prog, out);
+
+    GF_ASSERT(frame->this);
+
+    iobref = iobref_new();
+    if (!iobref) {
+        gf_msg(frame->this->name, GF_LOG_WARNING, ENOMEM, SVS_MSG_NO_MEMORY,
+               "failed to allocate "
+               "new iobref");
+        goto out;
+    }
+
+    if (req) {
+        xdr_size = xdr_sizeof(xdrproc, req);
+
+        iobuf = iobuf_get2(ctx->iobuf_pool, xdr_size);
+        if (!iobuf) {
+            goto out;
+        }
+
+        iobref_add(iobref, iobuf);
+
+        iov.iov_base = iobuf->ptr;
+        iov.iov_len = iobuf_pagesize(iobuf);
+
+        /* Create the xdr payload */
+        ret = xdr_serialize_generic(iov, req, xdrproc);
+        if (ret == -1) {
+            gf_msg(frame->this->name, GF_LOG_WARNING, 0,
+                   SVS_MSG_XDR_PAYLOAD_FAILED, "Failed to create XDR payload");
+            goto out;
+        }
+        iov.iov_len = ret;
+        count = 1;
+    }
+
+    ret = rpc_clnt_submit(ctx->mgmt, prog, procnum, cbkfn, &iov, count, NULL, 0,
+                          iobref, frame, NULL, 0, NULL, 0, NULL);
+
+out:
+    if (iobref)
+        iobref_unref(iobref);
+
+    if (iobuf)
+        iobuf_unref(iobuf);
+    return ret;
+}
+
+int
+mgmt_get_snapinfo_cbk(struct rpc_req *req, struct iovec *iov, int count,
+                      void *myframe)
+{
+    gf_getsnap_name_uuid_rsp rsp = {
+        0,
+    };
+    call_frame_t *frame = NULL;
+    glusterfs_ctx_t *ctx = NULL;
+    int ret = -1;
+    dict_t *dict = NULL;
+    char key[32] = {0};
+    int len;
+    int snapcount = 0;
+    svs_private_t *priv = NULL;
+    xlator_t *this = NULL;
+    int i = 0;
+    int j = 0;
+    char *value = NULL;
+    snap_dirent_t *dirents = NULL;
+    snap_dirent_t *old_dirents = NULL;
+    int oldcount = 0;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", req, error_out);
+    GF_VALIDATE_OR_GOTO("snapview-server", myframe, error_out);
+    GF_VALIDATE_OR_GOTO("snapview-server", iov, error_out);
+
+    frame = myframe;
+    this = frame->this;
+    ctx = frame->this->ctx;
+    priv = this->private;
+
+    if (!ctx) {
+        errno = EINVAL;
+        gf_msg(frame->this->name, GF_LOG_ERROR, errno, SVS_MSG_NULL_CTX,
+               "NULL context");
+        goto out;
+    }
+
+    if (-1 == req->rpc_status) {
+        errno = EINVAL;
+        gf_msg(frame->this->name, GF_LOG_ERROR, errno, SVS_MSG_RPC_CALL_FAILED,
+               "RPC call is not successful");
+        goto out;
+    }
+
+    ret = xdr_to_generic(*iov, &rsp, (xdrproc_t)xdr_gf_getsnap_name_uuid_rsp);
+    if (ret < 0) {
+        gf_msg(frame->this->name, GF_LOG_ERROR, 0, SVS_MSG_XDR_DECODE_FAILED,
+               "Failed to decode xdr response, rsp.op_ret = %d", rsp.op_ret);
+        goto out;
+    }
+
+    if (rsp.op_ret == -1) {
+        errno = rsp.op_errno;
+        ret = -1;
+        goto out;
+    }
+
+    if (!rsp.dict.dict_len) {
+        ret = -1;
+        errno = EINVAL;
+        gf_msg(frame->this->name, GF_LOG_ERROR, errno, SVS_MSG_RSP_DICT_EMPTY,
+               "Response dict is not populated");
+        goto out;
+    }
+
+    dict = dict_new();
+    if (!dict) {
+        ret = -1;
+        errno = ENOMEM;
+        goto out;
+    }
+
+    ret = dict_unserialize(rsp.dict.dict_val, rsp.dict.dict_len, &dict);
+    if (ret) {
+        errno = EINVAL;
+        gf_msg(frame->this->name, GF_LOG_ERROR, errno,
+               LG_MSG_DICT_UNSERIAL_FAILED, "Failed to unserialize dictionary");
+        goto out;
+    }
+
+    ret = dict_get_int32(dict, "snap-count", (int32_t *)&snapcount);
+    if (ret) {
+        errno = EINVAL;
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, errno, SVS_MSG_DICT_GET_FAILED,
+               "Error retrieving snapcount");
+        goto out;
+    }
+
+    if (snapcount > 0) {
+        /* first time we are fetching snap list */
+        dirents = GF_CALLOC(snapcount, sizeof(snap_dirent_t),
+                            gf_svs_mt_dirents_t);
+        if (!dirents) {
+            errno = ENOMEM;
+            ret = -1;
+            gf_msg(frame->this->name, GF_LOG_ERROR, errno, SVS_MSG_NO_MEMORY,
+                   "Unable to allocate memory");
+            goto out;
+        }
+    }
+
+    for (i = 0; i < snapcount; i++) {
+        len = snprintf(key, sizeof(key), "snap-volname.%d", i + 1);
+        ret = dict_get_strn(dict, key, len, &value);
+        if (ret) {
+            errno = EINVAL;
+            ret = -1;
+            gf_msg(this->name, GF_LOG_ERROR, errno, SVS_MSG_DICT_GET_FAILED,
+                   "Error retrieving snap volname %d", i + 1);
+            goto out;
+        }
+
+        strncpy(dirents[i].snap_volname, value,
+                sizeof(dirents[i].snap_volname));
+
+        len = snprintf(key, sizeof(key), "snap-id.%d", i + 1);
+        ret = dict_get_strn(dict, key, len, &value);
+        if (ret) {
+            errno = EINVAL;
+            ret = -1;
+            gf_msg(this->name, GF_LOG_ERROR, errno, SVS_MSG_DICT_GET_FAILED,
+                   "Error retrieving snap uuid %d", i + 1);
+            goto out;
+        }
+        strncpy(dirents[i].uuid, value, sizeof(dirents[i].uuid));
+
+        len = snprintf(key, sizeof(key), "snapname.%d", i + 1);
+        ret = dict_get_strn(dict, key, len, &value);
+        if (ret) {
+            errno = EINVAL;
+            ret = -1;
+            gf_msg(this->name, GF_LOG_ERROR, errno, SVS_MSG_DICT_GET_FAILED,
+                   "Error retrieving snap name %d", i + 1);
+            goto out;
+        }
+        strncpy(dirents[i].name, value, sizeof(dirents[i].name));
+    }
+
+    /*
+     * Got the new snap list populated in dirents
+     * The new snap list is either a subset or a superset of
+     * the existing snaplist old_dirents which has priv->num_snaps
+     * number of entries.
+     *
+     * If subset, then clean up the fs for entries which are
+     * no longer relevant.
+     *
+     * For other overlapping entries set the fs for new dirents
+     * entries which have a fs assigned already in old_dirents
+     *
+     * We do this as we don't want to do new glfs_init()s repeatedly
+     * as the dirents entries for snapshot volumes get repatedly
+     * cleaned up and allocated. And if we don't then that will lead
+     * to memleaks
+     */
+
+    LOCK(&priv->snaplist_lock);
+    {
+        oldcount = priv->num_snaps;
+        old_dirents = priv->dirents;
+        for (i = 0; i < priv->num_snaps; i++) {
+            for (j = 0; j < snapcount; j++) {
+                if ((!strcmp(old_dirents[i].name, dirents[j].name)) &&
+                    (!strcmp(old_dirents[i].uuid, dirents[j].uuid))) {
+                    dirents[j].fs = old_dirents[i].fs;
+                    old_dirents[i].fs = NULL;
+                    break;
+                }
+            }
+        }
+
+        priv->dirents = dirents;
+        priv->num_snaps = snapcount;
+    }
+    UNLOCK(&priv->snaplist_lock);
+
+    if (old_dirents) {
+        for (i = 0; i < oldcount; i++) {
+            if (old_dirents[i].fs)
+                gf_msg_debug(this->name, 0,
+                             "calling glfs_fini on "
+                             "name: %s, snap_volname: %s, uuid: %s",
+                             old_dirents[i].name, old_dirents[i].snap_volname,
+                             old_dirents[i].uuid);
+            glfs_fini(old_dirents[i].fs);
+        }
+    }
+
+    GF_FREE(old_dirents);
+
+    ret = 0;
+
+out:
+    if (dict) {
+        dict_unref(dict);
+    }
+    free(rsp.dict.dict_val);
+    free(rsp.op_errstr);
+
+    if (ret && dirents) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, SVS_MSG_SNAP_LIST_REFRESH_FAILED,
+               "Could not update dirents with refreshed snap list");
+        GF_FREE(dirents);
+    }
+
+    if (myframe)
+        SVS_STACK_DESTROY(myframe);
+
+error_out:
+    return ret;
+}
+
+int
+svs_get_snapshot_list(xlator_t *this)
+{
+    gf_getsnap_name_uuid_req req = {{
+        0,
+    }};
+    int ret = -1;
+    dict_t *dict = NULL;
+    glusterfs_ctx_t *ctx = NULL;
+    call_frame_t *frame = NULL;
+    svs_private_t *priv = NULL;
+    gf_boolean_t frame_cleanup = _gf_true;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+
+    ctx = this->ctx;
+    if (!ctx) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SVS_MSG_NULL_CTX, "ctx is NULL");
+        goto out;
+    }
+
+    frame = create_frame(this, ctx->pool);
+    if (!frame) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, LG_MSG_FRAME_ERROR,
+               "Error allocating frame");
+        goto out;
+    }
+
+    priv = this->private;
+
+    dict = dict_new();
+    if (!dict) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, SVS_MSG_NO_MEMORY,
+               "Error allocating dictionary");
+        goto out;
+    }
+
+    ret = dict_set_str(dict, "volname", priv->volname);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SVS_MSG_DICT_SET_FAILED,
+               "Error setting volname in dict");
+        goto out;
+    }
+
+    ret = dict_allocate_and_serialize(dict, &req.dict.dict_val,
+                                      &req.dict.dict_len);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, LG_MSG_DICT_UNSERIAL_FAILED,
+               "Failed to serialize dictionary");
+        ret = -1;
+        goto out;
+    }
+
+    ret = svs_mgmt_submit_request(
+        &req, frame, ctx, &svs_clnt_handshake_prog, GF_HNDSK_GET_SNAPSHOT_INFO,
+        mgmt_get_snapinfo_cbk, (xdrproc_t)xdr_gf_getsnap_name_uuid_req);
+
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SVS_MSG_RPC_REQ_FAILED,
+               "Error sending snapshot names RPC request");
+    }
+
+    frame_cleanup = _gf_false;
+
+out:
+    if (dict) {
+        dict_unref(dict);
+    }
+    GF_FREE(req.dict.dict_val);
+
+    if (frame_cleanup && frame) {
+        /*
+         * Destroy the frame if we encountered an error
+         * Else we need to clean it up in
+         * mgmt_get_snapinfo_cbk
+         */
+        SVS_STACK_DESTROY(frame);
+    }
+
+    return ret;
+}
diff --git a/xlators/features/snapview-server/src/snapview-server.c b/xlators/features/snapview-server/src/snapview-server.c
new file mode 100644
index 00000000000..76cccae5914
--- /dev/null
+++ b/xlators/features/snapview-server/src/snapview-server.c
@@ -0,0 +1,2720 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include "snapview-server.h"
+#include "snapview-server-mem-types.h"
+#include <glusterfs/compat-errno.h>
+
+#include <glusterfs/xlator.h>
+#include "rpc-clnt.h"
+#include "xdr-generic.h"
+#include "protocol-common.h"
+#include <glusterfs/syscall.h>
+#include <pthread.h>
+
+#include "glfs-internal.h"
+
+int
+gf_setcredentials(uid_t *uid, gid_t *gid, uint16_t ngrps, uint32_t *groups)
+{
+    int ret = 0;
+
+    if (uid) {
+        ret = glfs_setfsuid(*uid);
+        if (ret != 0) {
+            gf_msg("snapview-server", GF_LOG_ERROR, 0, SVS_MSG_SETFSUID_FAIL,
+                   "failed to set uid "
+                   "%u in thread context",
+                   *uid);
+            return ret;
+        }
+    }
+    if (gid) {
+        ret = glfs_setfsgid(*gid);
+        if (ret != 0) {
+            gf_msg("snapview-server", GF_LOG_ERROR, 0, SVS_MSG_SETFSGID_FAIL,
+                   "failed to set gid "
+                   "%u in thread context",
+                   *gid);
+            return ret;
+        }
+    }
+
+    if (ngrps != 0 && groups) {
+        ret = glfs_setfsgroups(ngrps, groups);
+        if (ret != 0) {
+            gf_msg("snapview-server", GF_LOG_ERROR, 0, SVS_MSG_SETFSGRPS_FAIL,
+                   "failed to set "
+                   "groups in thread context");
+            return ret;
+        }
+    }
+    return 0;
+}
+
+int32_t
+svs_lookup_entry_point(xlator_t *this, loc_t *loc, inode_t *parent,
+                       struct iatt *buf, struct iatt *postparent,
+                       int32_t *op_errno)
+{
+    uuid_t gfid;
+    svs_inode_t *inode_ctx = NULL;
+    int op_ret = -1;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+    GF_VALIDATE_OR_GOTO(this->name, buf, out);
+    GF_VALIDATE_OR_GOTO(this->name, postparent, out);
+
+    if (gf_uuid_is_null(loc->inode->gfid)) {
+        gf_uuid_generate(gfid);
+        svs_iatt_fill(gfid, buf);
+
+        /* Here the inode context of the entry point directory
+           is filled with just the type of the inode and the gfid
+           of the parent from where the entry point was entered.
+           The glfs object and the fs instance will be NULL.
+        */
+        if (parent)
+            svs_iatt_fill(parent->gfid, postparent);
+        else {
+            svs_iatt_fill(buf->ia_gfid, postparent);
+        }
+
+        inode_ctx = svs_inode_ctx_get_or_new(this, loc->inode);
+        if (!inode_ctx) {
+            op_ret = -1;
+            *op_errno = ENOMEM;
+            gf_msg(this->name, GF_LOG_ERROR, *op_errno,
+                   SVS_MSG_NEW_INODE_CTX_FAILED,
+                   "failed to "
+                   "allocate inode context for entry point "
+                   "directory");
+            goto out;
+        }
+
+        gf_uuid_copy(inode_ctx->pargfid, loc->pargfid);
+        memcpy(&inode_ctx->buf, buf, sizeof(*buf));
+        inode_ctx->type = SNAP_VIEW_ENTRY_POINT_INODE;
+    } else {
+        inode_ctx = svs_inode_ctx_get(this, loc->inode);
+        if (inode_ctx) {
+            memcpy(buf, &inode_ctx->buf, sizeof(*buf));
+            svs_iatt_fill(inode_ctx->pargfid, postparent);
+        } else {
+            svs_iatt_fill(loc->inode->gfid, buf);
+            if (parent)
+                svs_iatt_fill(parent->gfid, postparent);
+            else {
+                svs_iatt_fill(loc->inode->gfid, postparent);
+            }
+        }
+    }
+
+    op_ret = 0;
+
+out:
+    return op_ret;
+}
+
+/* When lookup comes from client and the protocol/server tries to resolve
+   the pargfid via just sending the gfid as part of lookup, if the inode
+   for the parent gfid is not found. But since that gfid has not yet been
+   looked  up yet, inode will not be having inode context and parent is not
+   there (as it is the parent of the entry that is being resolved). So
+   without parent and inode context, svs cannot know which snapshot
+   to look into. In such cases, the amguity is handled by looking
+   into the latest snapshot. If the directory is there in the latest
+   snapshot, lookup is successful, otherwise it is a failure. So for
+   any directory created after taking the latest snapshot, entry into
+   snapshot world is denied. i.e you have to be part of snapshot world
+   to enter it. If the gfid is not found there, then unwind with
+   ESTALE
+   This gets executed mainly in the situation where the snapshot entry
+   point is entered from a non-root directory and that non-root directory's
+   inode (or gfid) is not yet looked up. And in each case when a gfid has to
+   be looked up (without any inode contex and parent context present), last
+   snapshot is referred and a random gfid is not generated.
+*/
+int32_t
+svs_lookup_gfid(xlator_t *this, loc_t *loc, struct iatt *buf,
+                struct iatt *postparent, int32_t *op_errno)
+{
+    int32_t op_ret = -1;
+    unsigned char handle_obj[GFAPI_HANDLE_LENGTH] = {
+        0,
+    };
+    glfs_t *fs = NULL;
+    glfs_object_t *object = NULL;
+    struct stat statbuf = {
+        0,
+    };
+    svs_inode_t *inode_ctx = NULL;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+    GF_VALIDATE_OR_GOTO(this->name, buf, out);
+    GF_VALIDATE_OR_GOTO(this->name, postparent, out);
+
+    if (gf_uuid_is_null(loc->gfid) && gf_uuid_is_null(loc->inode->gfid)) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SVS_MSG_NULL_GFID, "gfid is NULL");
+        goto out;
+    }
+
+    if (!gf_uuid_is_null(loc->inode->gfid))
+        memcpy(handle_obj, loc->inode->gfid, GFAPI_HANDLE_LENGTH);
+    else
+        memcpy(handle_obj, loc->gfid, GFAPI_HANDLE_LENGTH);
+
+    fs = svs_get_latest_snapshot(this);
+    if (!fs) {
+        op_ret = -1;
+        *op_errno = EINVAL;
+        gf_msg(this->name, GF_LOG_ERROR, *op_errno,
+               SVS_MSG_GET_LATEST_SNAP_FAILED,
+               "failed to get the latest "
+               "snapshot");
+        goto out;
+    }
+
+    object = glfs_h_create_from_handle(fs, handle_obj, GFAPI_HANDLE_LENGTH,
+                                       &statbuf);
+    if (!object) {
+        op_ret = -1;
+        *op_errno = ESTALE;
+        gf_msg(this->name, GF_LOG_ERROR, *op_errno,
+               SVS_MSG_GET_GLFS_H_OBJECT_FAILED,
+               "failed to do lookup and get "
+               "the handle on the snapshot %s (path: %s, gfid: %s)",
+               loc->name, loc->path, uuid_utoa(loc->gfid));
+        goto out;
+    }
+
+    inode_ctx = svs_inode_ctx_get_or_new(this, loc->inode);
+    if (!inode_ctx) {
+        op_ret = -1;
+        *op_errno = ENOMEM;
+        gf_msg(this->name, GF_LOG_ERROR, *op_errno,
+               SVS_MSG_NEW_INODE_CTX_FAILED,
+               "failed to allocate inode "
+               "context");
+        goto out;
+    }
+
+    iatt_from_stat(buf, &statbuf);
+    if (!gf_uuid_is_null(loc->gfid))
+        gf_uuid_copy(buf->ia_gfid, loc->gfid);
+    else
+        gf_uuid_copy(buf->ia_gfid, loc->inode->gfid);
+
+    inode_ctx->type = SNAP_VIEW_VIRTUAL_INODE;
+    inode_ctx->fs = fs;
+    inode_ctx->object = object;
+    memcpy(&inode_ctx->buf, buf, sizeof(*buf));
+    svs_iatt_fill(buf->ia_gfid, postparent);
+
+    op_ret = 0;
+
+out:
+    return op_ret;
+}
+
+/* If the parent is an entry point inode, then create the handle for the
+   snapshot on which lookup came. i.e in reality lookup came on
+   the directory from which the entry point directory was entered, but
+   lookup is into the past. So create the handle for it by doing
+   the name-less lookup on the gfid (which can be obtained from
+   parent's context
+*/
+int32_t
+svs_lookup_snapshot(xlator_t *this, loc_t *loc, struct iatt *buf,
+                    struct iatt *postparent, inode_t *parent,
+                    svs_inode_t *parent_ctx, int32_t *op_errno)
+{
+    int32_t op_ret = -1;
+    unsigned char handle_obj[GFAPI_HANDLE_LENGTH] = {
+        0,
+    };
+    glfs_t *fs = NULL;
+    glfs_object_t *object = NULL;
+    struct stat statbuf = {
+        0,
+    };
+    svs_inode_t *inode_ctx = NULL;
+    uuid_t gfid;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+    GF_VALIDATE_OR_GOTO(this->name, buf, out);
+    GF_VALIDATE_OR_GOTO(this->name, postparent, out);
+    GF_VALIDATE_OR_GOTO(this->name, parent_ctx, out);
+    GF_VALIDATE_OR_GOTO(this->name, parent, out);
+
+    fs = svs_initialise_snapshot_volume(this, loc->name, op_errno);
+    if (!fs) {
+        gf_msg_debug(this->name, 0,
+                     "failed to create "
+                     "the fs instance for snap %s",
+                     loc->name);
+        *op_errno = ENOENT;
+        op_ret = -1;
+        goto out;
+    }
+
+    memcpy(handle_obj, parent_ctx->pargfid, GFAPI_HANDLE_LENGTH);
+    object = glfs_h_create_from_handle(fs, handle_obj, GFAPI_HANDLE_LENGTH,
+                                       &statbuf);
+    if (!object) {
+        op_ret = -1;
+        *op_errno = errno;
+        /* Should this be in warning or error mode? */
+        gf_msg_debug(this->name, 0,
+                     "failed to do lookup and "
+                     "get the handle on the snapshot %s",
+                     loc->name);
+        goto out;
+    }
+
+    inode_ctx = svs_inode_ctx_get_or_new(this, loc->inode);
+    if (!inode_ctx) {
+        op_ret = -1;
+        *op_errno = ENOMEM;
+        gf_msg(this->name, GF_LOG_ERROR, *op_errno,
+               SVS_MSG_NEW_INODE_CTX_FAILED,
+               "failed to allocate "
+               "inode context");
+        goto out;
+    }
+
+    if (gf_uuid_is_null(loc->gfid) && gf_uuid_is_null(loc->inode->gfid))
+        gf_uuid_generate(gfid);
+    else {
+        if (!gf_uuid_is_null(loc->inode->gfid))
+            gf_uuid_copy(gfid, loc->inode->gfid);
+        else
+            gf_uuid_copy(gfid, loc->gfid);
+    }
+    iatt_from_stat(buf, &statbuf);
+    gf_uuid_copy(buf->ia_gfid, gfid);
+    svs_fill_ino_from_gfid(buf);
+    inode_ctx->type = SNAP_VIEW_SNAPSHOT_INODE;
+    inode_ctx->fs = fs;
+    inode_ctx->object = object;
+    memcpy(&inode_ctx->buf, buf, sizeof(*buf));
+    svs_iatt_fill(parent->gfid, postparent);
+
+    SVS_STRDUP(inode_ctx->snapname, loc->name);
+    if (!inode_ctx->snapname) {
+        op_ret = -1;
+        *op_errno = ENOMEM;
+        goto out;
+    }
+    op_ret = 0;
+
+out:
+    if (op_ret) {
+        if (object)
+            glfs_h_close(object);
+
+        if (inode_ctx)
+            inode_ctx->object = NULL;
+    }
+
+    return op_ret;
+}
+
+/* Both parent and entry are from snapshot world */
+int32_t
+svs_lookup_entry(xlator_t *this, loc_t *loc, struct iatt *buf,
+                 struct iatt *postparent, inode_t *parent,
+                 svs_inode_t *parent_ctx, int32_t *op_errno)
+{
+    int32_t op_ret = -1;
+    glfs_t *fs = NULL;
+    glfs_object_t *object = NULL;
+    struct stat statbuf = {
+        0,
+    };
+    svs_inode_t *inode_ctx = NULL;
+    glfs_object_t *parent_object = NULL;
+    uuid_t gfid = {
+        0,
+    };
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+    GF_VALIDATE_OR_GOTO(this->name, buf, out);
+    GF_VALIDATE_OR_GOTO(this->name, postparent, out);
+    GF_VALIDATE_OR_GOTO(this->name, parent_ctx, out);
+    GF_VALIDATE_OR_GOTO(this->name, parent, out);
+
+    parent_object = parent_ctx->object;
+    fs = parent_ctx->fs;
+
+    object = glfs_h_lookupat(fs, parent_object, loc->name, &statbuf, 0);
+    if (!object) {
+        /* should this be in WARNING or ERROR mode? */
+        gf_msg_debug(this->name, 0,
+                     "failed to do lookup and "
+                     "get the handle for entry %s (path: %s)",
+                     loc->name, loc->path);
+        op_ret = -1;
+        *op_errno = errno;
+        goto out;
+    }
+
+    if (gf_uuid_is_null(object->gfid)) {
+        /* should this be in WARNING or ERROR mode? */
+        gf_msg_debug(this->name, 0,
+                     "gfid from glfs handle is "
+                     "NULL for entry %s (path: %s)",
+                     loc->name, loc->path);
+        op_ret = -1;
+        *op_errno = errno;
+        goto out;
+    }
+
+    inode_ctx = svs_inode_ctx_get_or_new(this, loc->inode);
+    if (!inode_ctx) {
+        op_ret = -1;
+        *op_errno = ENOMEM;
+        gf_msg(this->name, GF_LOG_ERROR, *op_errno,
+               SVS_MSG_NEW_INODE_CTX_FAILED,
+               "failed to allocate "
+               "inode context");
+        goto out;
+    }
+
+    if (gf_uuid_is_null(loc->gfid) && gf_uuid_is_null(loc->inode->gfid)) {
+        if (svs_uuid_generate(this, gfid, parent_ctx->snapname, object->gfid)) {
+            /*
+             * should op_errno be something else such as
+             * EINVAL or ESTALE?
+             */
+            op_ret = -1;
+            *op_errno = EIO;
+            goto out;
+        }
+    } else {
+        if (!gf_uuid_is_null(loc->inode->gfid))
+            gf_uuid_copy(gfid, loc->inode->gfid);
+        else
+            gf_uuid_copy(gfid, loc->gfid);
+    }
+
+    iatt_from_stat(buf, &statbuf);
+    gf_uuid_copy(buf->ia_gfid, gfid);
+    svs_fill_ino_from_gfid(buf);
+    inode_ctx->type = SNAP_VIEW_VIRTUAL_INODE;
+    inode_ctx->fs = fs;
+    inode_ctx->object = object;
+    memcpy(&inode_ctx->buf, buf, sizeof(*buf));
+    svs_iatt_fill(parent->gfid, postparent);
+
+    if (IA_ISDIR(buf->ia_type)) {
+        SVS_STRDUP(inode_ctx->snapname, parent_ctx->snapname);
+        if (!inode_ctx->snapname) {
+            op_ret = -1;
+            *op_errno = ENOMEM;
+            goto out;
+        }
+    }
+
+    op_ret = 0;
+
+out:
+    if (op_ret) {
+        if (object)
+            glfs_h_close(object);
+
+        if (inode_ctx)
+            inode_ctx->object = NULL;
+    }
+
+    return op_ret;
+}
+
+/* inode context is there means lookup has come on an object which was
+   built either as part of lookup or as part of readdirp. But in readdirp
+   we would not have got the handle to access the object in the gfapi
+   world.
+   So if inode context contains glfs_t instance for the right
+   gfapi world and glfs_object_t handle for accessing it in the gfapi
+   world, then unwind with success as the snapshots as of now are
+   read-only.
+   If the above condition is not met, then send lookup call again to
+   the gfapi world. It can happen only if both parent context and
+   the name of the entry are present.
+
+   If parent is an entry point to snapshot world:
+   * parent is needed for getting the gfid on which lookup has to be done
+     (the gfid present in the inode is a virtual gfid) in the snapshot
+     world.
+   * name is required to get the right glfs_t instance on which lookup
+     has to be done
+
+   If parent is a directory from snapshot world:
+   * parent context is needed to get the glfs_t instance and to get the
+     handle to parent directory in the snapshot world.
+   * name is needed to do the lookup on the right entry in the snapshot
+     world
+*/
+int32_t
+svs_revalidate(xlator_t *this, loc_t *loc, inode_t *parent,
+               svs_inode_t *inode_ctx, svs_inode_t *parent_ctx,
+               struct iatt *buf, struct iatt *postparent, int32_t *op_errno)
+{
+    int32_t op_ret = -1;
+    int ret = -1;
+    char tmp_uuid[64] = {
+        0,
+    };
+    glfs_t *fs = NULL;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, buf, out);
+    GF_VALIDATE_OR_GOTO(this->name, postparent, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode_ctx, out);
+
+    if (inode_ctx->type == SNAP_VIEW_ENTRY_POINT_INODE) {
+        svs_iatt_fill(loc->inode->gfid, buf);
+        if (parent)
+            svs_iatt_fill(parent->gfid, postparent);
+        else
+            svs_iatt_fill(loc->inode->gfid, postparent);
+        op_ret = 0;
+        goto out;
+    } else {
+        /* Though fs and object are present in the inode context, its
+         * better to check if fs is valid or not before doing anything.
+         * Its for the protection from the following operations.
+         * 1) Create a file on the glusterfs mount point
+         * 2) Create a snapshot (say "snap1")
+         * 3) Access the contents of the snapshot
+         * 4) Delete the file from the mount point
+         * 5) Delete the snapshot "snap1"
+         * 6) Create a new snapshot "snap1"
+         *
+         * Now accessing the new snapshot "snap1" gives problems.
+         * Because the inode and dentry created for snap1 would not be
+         * deleted upon the deletion of the snapshot (as deletion of
+         * snapshot is a gluster cli operation, not a fop). So next time
+         * upon creation of a new snap with same name, the previous
+         * inode and dentry itself will be used. But the inode context
+         * contains old information about the glfs_t instance and the
+         * handle in the gfapi world. Thus the glfs_t instance should
+         * be checked before accessing. If its wrong, then right
+         * instance should be obtained by doing the lookup.
+         */
+        if (inode_ctx->fs && inode_ctx->object) {
+            fs = inode_ctx->fs;
+            SVS_CHECK_VALID_SNAPSHOT_HANDLE(fs, this);
+            if (fs) {
+                memcpy(buf, &inode_ctx->buf, sizeof(*buf));
+                if (parent)
+                    svs_iatt_fill(parent->gfid, postparent);
+                else
+                    svs_iatt_fill(buf->ia_gfid, postparent);
+                op_ret = 0;
+                goto out;
+            } else {
+                inode_ctx->fs = NULL;
+                inode_ctx->object = NULL;
+                ret = svs_get_handle(this, loc, inode_ctx, op_errno);
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_ERROR, *op_errno,
+                           SVS_MSG_GET_GLFS_H_OBJECT_FAILED,
+                           "failed to get the handle for "
+                           "%s (gfid %s)",
+                           loc->path, uuid_utoa_r(loc->inode->gfid, tmp_uuid));
+                    op_ret = -1;
+                    goto out;
+                }
+            }
+        }
+
+        /* To send the lookup to gfapi world, both the name of the
+           entry as well as the parent context is needed.
+        */
+        if (!loc->name || !parent_ctx) {
+            *op_errno = ESTALE;
+            gf_msg(this->name, GF_LOG_ERROR, *op_errno,
+                   SVS_MSG_PARENT_CTX_OR_NAME_NULL, "%s is NULL",
+                   loc->name ? "parent context" : "loc->name");
+            goto out;
+        }
+
+        if (parent_ctx->type == SNAP_VIEW_ENTRY_POINT_INODE)
+            op_ret = svs_lookup_snapshot(this, loc, buf, postparent, parent,
+                                         parent_ctx, op_errno);
+        else
+            op_ret = svs_lookup_entry(this, loc, buf, postparent, parent,
+                                      parent_ctx, op_errno);
+
+        goto out;
+    }
+
+out:
+    return op_ret;
+}
+
+int32_t
+svs_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    struct iatt buf = {
+        0,
+    };
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    struct iatt postparent = {
+        0,
+    };
+    svs_inode_t *inode_ctx = NULL;
+    svs_inode_t *parent_ctx = NULL;
+    int32_t ret = -1;
+    inode_t *parent = NULL;
+    gf_boolean_t entry_point_key = _gf_false;
+    gf_boolean_t entry_point = _gf_false;
+    call_stack_t *root = NULL;
+
+    GF_VALIDATE_OR_GOTO("svs", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    root = frame->root;
+    op_ret = gf_setcredentials(&root->uid, &root->gid, root->ngrps,
+                               root->groups);
+    if (op_ret != 0) {
+        goto out;
+    }
+
+    /* For lookups sent on inodes (i.e not parent inode + basename, but
+       direct inode itself which usually is a nameless lookup or revalidate
+       on the inode), loc->name will not be there. Get it from path if
+       it is there.
+       This is the difference between nameless lookup and revalidate lookup
+       on an inode:
+       nameless lookup: loc->path contains gfid and strrchr on it fails
+       revalidate lookup: loc->path contains the entry name of the inode
+                          and strrchr gives the name of the entry from path
+    */
+    if (loc->path) {
+        if (!loc->name || (loc->name && !strcmp(loc->name, ""))) {
+            loc->name = strrchr(loc->path, '/');
+            if (loc->name)
+                loc->name++;
+        }
+    }
+
+    if (loc->parent)
+        parent = inode_ref(loc->parent);
+    else {
+        parent = inode_find(loc->inode->table, loc->pargfid);
+        if (!parent)
+            parent = inode_parent(loc->inode, NULL, NULL);
+    }
+    if (parent)
+        parent_ctx = svs_inode_ctx_get(this, parent);
+
+    inode_ctx = svs_inode_ctx_get(this, loc->inode);
+
+    if (xdata && !inode_ctx) {
+        ret = dict_get_str_boolean(xdata, "entry-point", _gf_false);
+        if (ret == -1) {
+            gf_msg_debug(this->name, 0,
+                         "failed to get the "
+                         "entry point info");
+            entry_point_key = _gf_false;
+        } else {
+            entry_point_key = ret;
+        }
+
+        if (loc->name && strlen(loc->name)) {
+            /* lookup can come with the entry-point set in the dict
+             * for the parent directory of the entry-point as well.
+             * So consider entry_point only for named lookup
+             */
+            entry_point = entry_point_key;
+        }
+    }
+
+    if (inode_ctx && inode_ctx->type == SNAP_VIEW_ENTRY_POINT_INODE) {
+        /* entry-point may not be set in the dictonary.
+         * This can happen if snap-view client is restarted where
+         * inode-ctx not available and a nameless lookup has come
+         */
+        entry_point = _gf_true;
+    }
+
+    /* lookup is on the entry point to the snapshot world */
+    if (entry_point) {
+        op_ret = svs_lookup_entry_point(this, loc, parent, &buf, &postparent,
+                                        &op_errno);
+        goto out;
+    }
+
+    /* revalidate */
+    if (inode_ctx) {
+        op_ret = svs_revalidate(this, loc, parent, inode_ctx, parent_ctx, &buf,
+                                &postparent, &op_errno);
+        goto out;
+    }
+
+    /* This can happen when entry point directory is entered from non-root
+       directory. (ex: if /mnt/glusterfs is the mount point, then entry
+       point (say .snaps) is entered from /mnt/glusterfs/dir/.snaps). Also
+       it can happen when client sends a nameless lookup on just a gfid and
+       the server does not have the inode in the inode table.
+    */
+    if (!inode_ctx && !parent_ctx) {
+        if (gf_uuid_is_null(loc->gfid) && gf_uuid_is_null(loc->inode->gfid)) {
+            op_ret = -1;
+            op_errno = ESTALE;
+            gf_msg_debug(this->name, 0,
+                         "gfid is NULL. Either the lookup "
+                         "came on missing entry or the "
+                         "entry is stale");
+            goto out;
+        }
+
+        if (!entry_point_key) {
+            /* This can happen when there is no inode_ctx available.
+             * snapview-server might have restarted or
+             * graph change might have happened
+             */
+            op_ret = -1;
+            op_errno = ESTALE;
+            goto out;
+        }
+
+        /* lookup is on the parent directory of entry-point.
+         * this would have already looked up by snap-view client
+         * so return success
+         */
+        if (!gf_uuid_is_null(loc->gfid))
+            gf_uuid_copy(buf.ia_gfid, loc->gfid);
+        else
+            gf_uuid_copy(buf.ia_gfid, loc->inode->gfid);
+
+        svs_iatt_fill(buf.ia_gfid, &buf);
+        svs_iatt_fill(buf.ia_gfid, &postparent);
+
+        op_ret = 0;
+        goto out;
+    }
+
+    if (parent_ctx) {
+        if (parent_ctx->type == SNAP_VIEW_ENTRY_POINT_INODE)
+            op_ret = svs_lookup_snapshot(this, loc, &buf, &postparent, parent,
+                                         parent_ctx, &op_errno);
+        else
+            op_ret = svs_lookup_entry(this, loc, &buf, &postparent, parent,
+                                      parent_ctx, &op_errno);
+        goto out;
+    }
+
+out:
+    STACK_UNWIND_STRICT(lookup, frame, op_ret, op_errno,
+                        loc ? loc->inode : NULL, &buf, xdata, &postparent);
+
+    if (parent)
+        inode_unref(parent);
+
+    return 0;
+}
+
+int32_t
+svs_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+            dict_t *xdata)
+{
+    svs_inode_t *inode_ctx = NULL;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    svs_fd_t *svs_fd = NULL;
+    glfs_fd_t *glfd = NULL;
+    glfs_t *fs = NULL;
+    glfs_object_t *object = NULL;
+    call_stack_t *root = NULL;
+
+    GF_VALIDATE_OR_GOTO("snap-view-daemon", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    root = frame->root;
+    op_ret = gf_setcredentials(&root->uid, &root->gid, root->ngrps,
+                               root->groups);
+    if (op_ret != 0) {
+        goto out;
+    }
+
+    inode_ctx = svs_inode_ctx_get(this, loc->inode);
+    if (!inode_ctx) {
+        op_ret = -1;
+        op_errno = ESTALE;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno,
+               SVS_MSG_GET_INODE_CONTEXT_FAILED,
+               "inode context not found "
+               "for the inode %s",
+               uuid_utoa(loc->inode->gfid));
+        goto out;
+    }
+
+    /* Fake success is sent if the opendir is on the entry point directory
+       or the inode is SNAP_VIEW_ENTRY_POINT_INODE
+    */
+    if (inode_ctx->type == SNAP_VIEW_ENTRY_POINT_INODE) {
+        op_ret = 0;
+        op_errno = 0;
+        goto out;
+    } else {
+        SVS_GET_INODE_CTX_INFO(inode_ctx, fs, object, this, loc, op_ret,
+                               op_errno, out);
+
+        glfd = glfs_h_opendir(fs, object);
+        if (!glfd) {
+            op_ret = -1;
+            op_errno = errno;
+            gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_OPENDIR_FAILED,
+                   "opendir on %s failed "
+                   "(gfid: %s)",
+                   loc->name, uuid_utoa(loc->inode->gfid));
+            goto out;
+        }
+        svs_fd = svs_fd_ctx_get_or_new(this, fd);
+        if (!svs_fd) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                   SVS_MSG_NEW_FD_CTX_FAILED,
+                   "failed to allocate fd context "
+                   "for %s (gfid: %s)",
+                   loc->name, uuid_utoa(fd->inode->gfid));
+            glfs_closedir(glfd);
+            goto out;
+        }
+        svs_fd->fd = glfd;
+
+        op_ret = 0;
+        op_errno = 0;
+    }
+
+out:
+    STACK_UNWIND_STRICT(opendir, frame, op_ret, op_errno, fd, NULL);
+
+    return 0;
+}
+
+/*
+ * This function adds the xattr keys present in the list (@list) to the dict.
+ * But the list contains only the names of the xattrs (and no value, as
+ * the gfapi functions for the listxattr operations would return only the
+ * names of the xattrs in the buffer provided by the caller, though they had
+ * got the values of those xattrs from posix) as described in the man page of
+ * listxattr. But before unwinding snapview-server has to put those names
+ * back into the dict. But to get the values for those xattrs it has to do the
+ * getxattr operation on each xattr which might turn out to be a costly
+ * operation. So for each of the xattrs present in the list, a 0 byte value
+ * ("") is set into the dict before unwinding. Since ("") is also a valid xattr
+ * value(in a file system) we use an extra key in the same dictionary as an
+ * indicator to other xlators which want to cache the xattrs (as of now,
+ * md-cache which caches acl and selinux related xattrs) to not to cache the
+ * values of the xattrs present in the dict.
+ */
+int32_t
+svs_add_xattrs_to_dict(xlator_t *this, dict_t *dict, char *list, ssize_t size)
+{
+    char keybuffer[4096] = {
+        0,
+    };
+    size_t remaining_size = 0;
+    int32_t list_offset = 0;
+    int32_t ret = -1;
+
+    GF_VALIDATE_OR_GOTO("snapview-daemon", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, dict, out);
+    GF_VALIDATE_OR_GOTO(this->name, list, out);
+
+    remaining_size = size;
+    list_offset = 0;
+    while (remaining_size > 0) {
+        strncpy(keybuffer, list + list_offset, sizeof(keybuffer) - 1);
+#ifdef GF_DARWIN_HOST_OS
+        /* The protocol expect namespace for now */
+        char *newkey = NULL;
+        gf_add_prefix(XATTR_USER_PREFIX, keybuffer, &newkey);
+        strcpy(keybuffer, newkey);
+        GF_FREE(newkey);
+#endif
+        ret = dict_set_str(dict, keybuffer, "");
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, SVS_MSG_DICT_SET_FAILED,
+                   "dict set operation "
+                   "for the key %s failed.",
+                   keybuffer);
+            goto out;
+        }
+
+        remaining_size -= strlen(keybuffer) + 1;
+        list_offset += strlen(keybuffer) + 1;
+    } /* while (remaining_size > 0) */
+
+    /* Add an additional key to indicate that we don't need to cache these
+     * xattrs(with value "") */
+    ret = dict_set_str(dict, "glusterfs.skip-cache", "");
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SVS_MSG_DICT_SET_FAILED,
+               "dict set operation for the key glusterfs.skip-cache failed.");
+        goto out;
+    }
+
+    ret = 0;
+
+out:
+    return ret;
+}
+
+int32_t
+svs_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, const char *name,
+             dict_t *xdata)
+{
+    svs_inode_t *inode_ctx = NULL;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    glfs_t *fs = NULL;
+    glfs_object_t *object = NULL;
+    char *value = 0;
+    ssize_t size = 0;
+    dict_t *dict = NULL;
+    call_stack_t *root = NULL;
+
+    GF_VALIDATE_OR_GOTO("snap-view-daemon", this, out);
+    GF_VALIDATE_OR_GOTO("snap-view-daemon", frame, out);
+    GF_VALIDATE_OR_GOTO("snap-view-daemon", loc, out);
+    GF_VALIDATE_OR_GOTO("snap-view-daemon", loc->inode, out);
+
+    root = frame->root;
+    op_ret = gf_setcredentials(&root->uid, &root->gid, root->ngrps,
+                               root->groups);
+    if (op_ret != 0) {
+        goto out;
+    }
+
+    inode_ctx = svs_inode_ctx_get(this, loc->inode);
+    if (!inode_ctx) {
+        op_ret = -1;
+        op_errno = ESTALE;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno,
+               SVS_MSG_GET_INODE_CONTEXT_FAILED,
+               "inode context not found "
+               "for the inode %s",
+               uuid_utoa(loc->inode->gfid));
+        goto out;
+    }
+
+    /* ENODATA is sent if the getxattr is on entry point directory
+       or the inode is SNAP_VIEW_ENTRY_POINT_INODE. Entry point is
+       a virtual directory on which setxattr operations are not
+       allowed. If getxattr has to be faked as success, then a value
+       for the name of the xattr has to be sent which we don't have.
+    */
+    if (inode_ctx->type == SNAP_VIEW_ENTRY_POINT_INODE) {
+        op_ret = -1;
+        op_errno = ENODATA;
+        goto out;
+    } else {
+        SVS_GET_INODE_CTX_INFO(inode_ctx, fs, object, this, loc, op_ret,
+                               op_errno, out);
+
+        dict = dict_new();
+        if (!dict) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_NO_MEMORY,
+                   "failed to allocate dict");
+            goto out;
+        }
+
+        size = glfs_h_getxattrs(fs, object, name, NULL, 0);
+        if (size == -1) {
+            op_ret = -1;
+            op_errno = errno;
+            if (errno == ENODATA) {
+                gf_msg_debug(this->name, 0,
+                             "getxattr on "
+                             "%s failed (ket: %s) with %s",
+                             loc->path, name, strerror(errno));
+            } else {
+                gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                       SVS_MSG_GETXATTR_FAILED,
+                       "getxattr on %s failed (key: %s) with %s", loc->path,
+                       name, strerror(errno));
+            }
+            goto out;
+        }
+        value = GF_CALLOC(size + 1, sizeof(char), gf_common_mt_char);
+        if (!value) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_NO_MEMORY,
+                   "failed to allocate memory for getxattr "
+                   "on %s (key: %s)",
+                   loc->name, name);
+            goto out;
+        }
+
+        size = glfs_h_getxattrs(fs, object, name, value, size);
+        if (size == -1) {
+            op_ret = -1;
+            op_errno = errno;
+            gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_GETXATTR_FAILED,
+                   "failed to get the xattr %s for "
+                   "entry %s",
+                   name, loc->name);
+            goto out;
+        }
+        value[size] = '\0';
+
+        if (name) {
+            op_ret = dict_set_dynptr(dict, (char *)name, value, size);
+            if (op_ret < 0) {
+                op_errno = -op_ret;
+                gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                       SVS_MSG_DICT_SET_FAILED,
+                       "dict set operation for %s for "
+                       "the key %s failed.",
+                       loc->path, name);
+                GF_FREE(value);
+                value = NULL;
+                goto out;
+            }
+        } else {
+            op_ret = svs_add_xattrs_to_dict(this, dict, value, size);
+            if (op_ret == -1) {
+                op_errno = ENOMEM;
+                gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_NO_MEMORY,
+                       "failed to add xattrs from the list to "
+                       "dict for %s (gfid: %s)",
+                       loc->path, uuid_utoa(loc->inode->gfid));
+                goto out;
+            }
+            GF_FREE(value);
+            value = NULL;
+        }
+    }
+
+out:
+    if (op_ret && value)
+        GF_FREE(value);
+
+    STACK_UNWIND_STRICT(getxattr, frame, op_ret, op_errno, dict, NULL);
+
+    if (dict)
+        dict_unref(dict);
+
+    return 0;
+}
+
+int32_t
+svs_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+              dict_t *xdata)
+{
+    svs_inode_t *inode_ctx = NULL;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    char *value = 0;
+    ssize_t size = 0;
+    dict_t *dict = NULL;
+    svs_fd_t *sfd = NULL;
+    glfs_fd_t *glfd = NULL;
+
+    GF_VALIDATE_OR_GOTO("snap-view-daemon", this, out);
+    GF_VALIDATE_OR_GOTO("snap-view-daemon", frame, out);
+    GF_VALIDATE_OR_GOTO("snap-view-daemon", fd, out);
+    GF_VALIDATE_OR_GOTO("snap-view-daemon", fd->inode, out);
+
+    inode_ctx = svs_inode_ctx_get(this, fd->inode);
+    if (!inode_ctx) {
+        op_ret = -1;
+        op_errno = ESTALE;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno,
+               SVS_MSG_GET_INODE_CONTEXT_FAILED,
+               "inode context not found "
+               "for the inode %s",
+               uuid_utoa(fd->inode->gfid));
+        goto out;
+    }
+
+    if (!(svs_inode_ctx_glfs_mapping(this, inode_ctx))) {
+        op_ret = -1;
+        op_errno = EBADF;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_FS_INSTANCE_INVALID,
+               "glfs instance %p to which the inode %s "
+               "belongs to does not exist. The snapshot "
+               "corresponding to the instance might have"
+               "been deleted or deactivated",
+               inode_ctx->fs, uuid_utoa(fd->inode->gfid));
+        goto out;
+    }
+
+    sfd = svs_fd_ctx_get_or_new(this, fd);
+    if (!sfd) {
+        op_ret = -1;
+        op_errno = EBADFD;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno,
+               SVS_MSG_GET_FD_CONTEXT_FAILED,
+               "failed to get the fd "
+               "context for %s",
+               uuid_utoa(fd->inode->gfid));
+        goto out;
+    }
+
+    glfd = sfd->fd;
+    /* EINVAL is sent if the getxattr is on entry point directory
+       or the inode is SNAP_VIEW_ENTRY_POINT_INODE. Entry point is
+       a virtual directory on which setxattr operations are not
+       allowed. If getxattr has to be faked as success, then a value
+       for the name of the xattr has to be sent which we don't have.
+    */
+    if (inode_ctx->type == SNAP_VIEW_ENTRY_POINT_INODE) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        goto out;
+    } else {
+        dict = dict_new();
+        if (!dict) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_NO_MEMORY,
+                   "failed to allocate dict "
+                   "(gfid: %s, key: %s)",
+                   uuid_utoa(fd->inode->gfid), name);
+            goto out;
+        }
+
+        if (name) {
+            size = glfs_fgetxattr(glfd, name, NULL, 0);
+            if (size == -1) {
+                op_ret = -1;
+                op_errno = errno;
+                gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                       SVS_MSG_GETXATTR_FAILED,
+                       "getxattr on %s failed "
+                       "(key: %s)",
+                       uuid_utoa(fd->inode->gfid), name);
+                goto out;
+            }
+            value = GF_CALLOC(size + 1, sizeof(char), gf_common_mt_char);
+            if (!value) {
+                op_ret = -1;
+                op_errno = ENOMEM;
+                gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_NO_MEMORY,
+                       "failed to "
+                       "allocate memory for getxattr on %s "
+                       "(key: %s)",
+                       uuid_utoa(fd->inode->gfid), name);
+                goto out;
+            }
+
+            size = glfs_fgetxattr(glfd, name, value, size);
+            if (size == -1) {
+                op_ret = -1;
+                op_errno = errno;
+                gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                       SVS_MSG_GETXATTR_FAILED,
+                       "failed to get the xattr %s "
+                       "for inode %s",
+                       name, uuid_utoa(fd->inode->gfid));
+                goto out;
+            }
+            value[size] = '\0';
+
+            op_ret = dict_set_dynptr(dict, (char *)name, value, size);
+            if (op_ret < 0) {
+                op_errno = -op_ret;
+                gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                       SVS_MSG_DICT_SET_FAILED,
+                       "dict set operation for gfid %s "
+                       "for the key %s failed.",
+                       uuid_utoa(fd->inode->gfid), name);
+                goto out;
+            }
+        } else {
+            size = glfs_flistxattr(glfd, NULL, 0);
+            if (size == -1) {
+                op_errno = errno;
+                gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                       SVS_MSG_LISTXATTR_FAILED, "listxattr on %s failed",
+                       uuid_utoa(fd->inode->gfid));
+                goto out;
+            }
+
+            value = GF_CALLOC(size + 1, sizeof(char), gf_common_mt_char);
+            if (!value) {
+                op_ret = -1;
+                op_errno = ENOMEM;
+                gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_NO_MEMORY,
+                       "failed to "
+                       "allocate buffer for xattr "
+                       "list (%s)",
+                       uuid_utoa(fd->inode->gfid));
+                goto out;
+            }
+
+            size = glfs_flistxattr(glfd, value, size);
+            if (size == -1) {
+                op_ret = -1;
+                op_errno = errno;
+                gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                       SVS_MSG_LISTXATTR_FAILED, "listxattr on %s failed",
+                       uuid_utoa(fd->inode->gfid));
+                goto out;
+            }
+
+            op_ret = svs_add_xattrs_to_dict(this, dict, value, size);
+            if (op_ret == -1) {
+                op_errno = ENOMEM;
+                gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_NO_MEMORY,
+                       "failed to add xattrs from the list "
+                       "to dict (gfid: %s)",
+                       uuid_utoa(fd->inode->gfid));
+                goto out;
+            }
+            GF_FREE(value);
+        }
+
+        op_ret = 0;
+        op_errno = 0;
+    }
+
+out:
+    if (op_ret)
+        GF_FREE(value);
+
+    STACK_UNWIND_STRICT(fgetxattr, frame, op_ret, op_errno, dict, NULL);
+
+    if (dict)
+        dict_unref(dict);
+
+    return 0;
+}
+
+int32_t
+svs_releasedir(xlator_t *this, fd_t *fd)
+{
+    svs_fd_t *sfd = NULL;
+    uint64_t tmp_pfd = 0;
+    int ret = 0;
+    svs_inode_t *svs_inode = NULL;
+    glfs_t *fs = NULL;
+    inode_t *inode = NULL;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    ret = fd_ctx_del(fd, this, &tmp_pfd);
+    if (ret < 0) {
+        gf_msg_debug(this->name, 0, "pfd from fd=%p is NULL", fd);
+        goto out;
+    }
+
+    inode = fd->inode;
+
+    svs_inode = svs_inode_ctx_get(this, inode);
+    if (svs_inode) {
+        fs = svs_inode->fs; /* should inode->lock be held for this? */
+        SVS_CHECK_VALID_SNAPSHOT_HANDLE(fs, this);
+        if (fs) {
+            sfd = (svs_fd_t *)(long)tmp_pfd;
+            if (sfd->fd) {
+                ret = glfs_closedir(sfd->fd);
+                if (ret)
+                    gf_msg(this->name, GF_LOG_WARNING, errno,
+                           SVS_MSG_RELEASEDIR_FAILED,
+                           "failed to close the glfd for "
+                           "directory %s",
+                           uuid_utoa(fd->inode->gfid));
+            }
+        }
+    }
+
+    GF_FREE(sfd);
+
+out:
+    return 0;
+}
+
+int32_t
+svs_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    int32_t op_ret = -1;
+    int32_t op_errno = 0;
+    int ret = -1;
+    uint64_t value = 0;
+    svs_inode_t *inode_ctx = NULL;
+    call_stack_t *root = NULL;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    root = frame->root;
+    op_ret = gf_setcredentials(&root->uid, &root->gid, root->ngrps,
+                               root->groups);
+    if (op_ret != 0) {
+        goto out;
+    }
+
+    inode_ctx = svs_inode_ctx_get(this, fd->inode);
+    if (!inode_ctx) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno,
+               SVS_MSG_GET_INODE_CONTEXT_FAILED,
+               "inode context not found for"
+               " the inode %s",
+               uuid_utoa(fd->inode->gfid));
+        goto out;
+    }
+
+    ret = fd_ctx_get(fd, this, &value);
+    if (ret < 0 && inode_ctx->type != SNAP_VIEW_ENTRY_POINT_INODE) {
+        op_errno = EINVAL;
+        gf_msg(this->name, GF_LOG_WARNING, op_errno,
+               SVS_MSG_GET_FD_CONTEXT_FAILED, "pfd is NULL on fd=%p", fd);
+        goto out;
+    }
+
+    op_ret = 0;
+
+out:
+    STACK_UNWIND_STRICT(flush, frame, op_ret, op_errno, NULL);
+
+    return 0;
+}
+
+int32_t
+svs_release(xlator_t *this, fd_t *fd)
+{
+    svs_fd_t *sfd = NULL;
+    uint64_t tmp_pfd = 0;
+    int ret = 0;
+    inode_t *inode = NULL;
+    svs_inode_t *svs_inode = NULL;
+    glfs_t *fs = NULL;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+    ret = fd_ctx_del(fd, this, &tmp_pfd);
+    if (ret < 0) {
+        gf_msg_debug(this->name, 0, "pfd from fd=%p is NULL", fd);
+        goto out;
+    }
+
+    inode = fd->inode;
+
+    svs_inode = svs_inode_ctx_get(this, inode);
+    if (svs_inode) {
+        fs = svs_inode->fs; /* should inode->lock be held for this? */
+        SVS_CHECK_VALID_SNAPSHOT_HANDLE(fs, this);
+        if (fs) {
+            sfd = (svs_fd_t *)(long)tmp_pfd;
+            if (sfd->fd) {
+                ret = glfs_close(sfd->fd);
+                if (ret)
+                    gf_msg(this->name, GF_LOG_ERROR, errno,
+                           SVS_MSG_RELEASE_FAILED,
+                           "failed to close "
+                           "the glfd for %s",
+                           uuid_utoa(fd->inode->gfid));
+            }
+        }
+    }
+
+    GF_FREE(sfd);
+out:
+    return 0;
+}
+
+int32_t
+svs_forget(xlator_t *this, inode_t *inode)
+{
+    int ret = -1;
+    uint64_t value = 0;
+    svs_inode_t *inode_ctx = NULL;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+
+    ret = inode_ctx_del(inode, this, &value);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, SVS_MSG_DELETE_INODE_CONTEXT_FAILED,
+               "failed to delete the inode "
+               "context of %s",
+               uuid_utoa(inode->gfid));
+        goto out;
+    }
+
+    inode_ctx = (svs_inode_t *)(uintptr_t)value;
+    if (!inode_ctx)
+        goto out;
+
+    if (inode_ctx->snapname)
+        GF_FREE(inode_ctx->snapname);
+
+    /*
+     * glfs_h_close leads to unref and forgetting of the
+     * underlying inode in the gfapi world. i.e. the inode
+     * which inode_ctx->object points to.
+     * As of now the only possibility is, this forget came as a
+     * result of snapdaemon's inode table reaching the lru
+     * limit and receiving forget as a result of purging of
+     * extra inodes that exceeded the limit. But, care must
+     * be taken to ensure that, the gfapi instance to which
+     * the glfs_h_object belongs to is not deleted. Otherwise
+     * this might result in access of a freed pointer.
+     * This will still be helpful in reducing the memory
+     * footprint of snapdaemon when the fs instance itself is
+     * valid (i.e. present and not destroyed due to either snap
+     * deactivate or snap delete), but the lru limit is reached.
+     * The forget due to lru limit will make the underlying inode
+     * being unrefed and forgotten.
+     */
+    if (svs_inode_ctx_glfs_mapping(this, inode_ctx)) {
+        glfs_h_close(inode_ctx->object);
+        inode_ctx->object = NULL;
+    }
+    GF_FREE(inode_ctx);
+
+out:
+    return 0;
+}
+
+int
+svs_fill_readdir(xlator_t *this, gf_dirent_t *entries, size_t size, off_t off)
+{
+    gf_dirent_t *entry = NULL;
+    svs_private_t *priv = NULL;
+    int i = 0;
+    snap_dirent_t *dirents = NULL;
+    int this_size = 0;
+    int filled_size = 0;
+    int count = 0;
+
+    GF_VALIDATE_OR_GOTO("snap-view-daemon", this, out);
+    GF_VALIDATE_OR_GOTO("snap-view-daemon", entries, out);
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    /* create the dir entries */
+    LOCK(&priv->snaplist_lock);
+    {
+        dirents = priv->dirents;
+
+        for (i = off; i < priv->num_snaps;) {
+            this_size = sizeof(gf_dirent_t) + strlen(dirents[i].name) + 1;
+            if (this_size + filled_size > size)
+                goto unlock;
+
+            entry = gf_dirent_for_name(dirents[i].name);
+            if (!entry) {
+                gf_msg(this->name, GF_LOG_ERROR, ENOMEM, SVS_MSG_NO_MEMORY,
+                       "failed to allocate dentry for %s", dirents[i].name);
+                goto unlock;
+            }
+
+            entry->d_off = i + 1;
+            /*
+             * readdir on the entry-point directory to the snapshot
+             * world, will return elements in the list of the
+             * snapshots as the directory entries. Since the entries
+             * returned are virtual entries which does not exist
+             * physically on the disk, pseudo inode numbers are
+             * generated.
+             */
+            entry->d_ino = i + 2 * 42;
+            entry->d_type = DT_DIR;
+            list_add_tail(&entry->list, &entries->list);
+            ++i;
+            count++;
+            filled_size += this_size;
+        }
+    }
+unlock:
+    UNLOCK(&priv->snaplist_lock);
+
+out:
+    return count;
+}
+
+int32_t
+svs_glfs_readdir(xlator_t *this, glfs_fd_t *glfd, gf_dirent_t *entries,
+                 int32_t *op_errno, struct iatt *buf, gf_boolean_t readdirplus,
+                 size_t size)
+{
+    int filled_size = 0;
+    int this_size = 0;
+    int32_t ret = -1;
+    int32_t count = 0;
+    gf_dirent_t *entry = NULL;
+    struct dirent *dirents = NULL;
+    struct dirent de = {
+        0,
+    };
+    struct stat statbuf = {
+        0,
+    };
+    off_t in_case = -1;
+
+    GF_VALIDATE_OR_GOTO("svs", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, glfd, out);
+    GF_VALIDATE_OR_GOTO(this->name, entries, out);
+
+    while (filled_size < size) {
+        in_case = glfs_telldir(glfd);
+        if (in_case == -1) {
+            gf_msg(this->name, GF_LOG_ERROR, errno, SVS_MSG_TELLDIR_FAILED,
+                   "telldir failed");
+            break;
+        }
+
+        if (readdirplus)
+            ret = glfs_readdirplus_r(glfd, &statbuf, &de, &dirents);
+        else
+            ret = glfs_readdir_r(glfd, &de, &dirents);
+
+        if (ret == 0 && dirents != NULL) {
+            if (readdirplus)
+                this_size = max(sizeof(gf_dirent_t), sizeof(gfs3_dirplist)) +
+                            strlen(de.d_name) + 1;
+            else
+                this_size = sizeof(gf_dirent_t) + strlen(de.d_name) + 1;
+
+            if (this_size + filled_size > size) {
+                glfs_seekdir(glfd, in_case);
+                break;
+            }
+
+            entry = gf_dirent_for_name(de.d_name);
+            if (!entry) {
+                /*
+                 * Since gf_dirent_for_name can return
+                 * NULL only when it fails to allocate
+                 * memory for the directory entry,
+                 * SVS_MSG_NO_MEMORY is used as the
+                 * message-id.
+                 */
+                gf_msg(this->name, GF_LOG_ERROR, errno, SVS_MSG_NO_MEMORY,
+                       "could not create gf_dirent "
+                       "for entry %s: (%s)",
+                       entry->d_name, strerror(errno));
+                break;
+            }
+            entry->d_off = glfs_telldir(glfd);
+            entry->d_ino = de.d_ino;
+            entry->d_type = de.d_type;
+            if (readdirplus) {
+                iatt_from_stat(buf, &statbuf);
+                entry->d_stat = *buf;
+            }
+            list_add_tail(&entry->list, &entries->list);
+
+            filled_size += this_size;
+            count++;
+        } else if (ret == 0 && dirents == NULL) {
+            *op_errno = ENOENT;
+            break;
+        } else if (ret != 0) {
+            *op_errno = errno;
+            break;
+        }
+        dirents = NULL;
+    }
+
+out:
+    return count;
+}
+
+/* readdirp can be of 2 types.
+   1) It can come on entry point directory where the list of snapshots
+      is sent as dirents. In this case, the iatt structure is filled
+      on the fly if the inode is not found for the entry or the inode
+      context is NULL. Other wise if inode is found and inode context
+      is there the iatt structure saved in the context is used.
+   2) It can be on a directory in one of the snapshots. In this case,
+      the readdirp call would have sent us a iatt structure. So the same
+      structure is used with the exception that the gfid and the inode
+      numbers will be newly generated and filled in.
+*/
+void
+svs_readdirp_fill(xlator_t *this, inode_t *parent, svs_inode_t *parent_ctx,
+                  gf_dirent_t *entry)
+{
+    inode_t *inode = NULL;
+    uuid_t random_gfid = {
+        0,
+    };
+    struct iatt buf = {
+        0,
+    };
+    svs_inode_t *inode_ctx = NULL;
+
+    GF_VALIDATE_OR_GOTO("snapview-server", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, parent, out);
+    GF_VALIDATE_OR_GOTO(this->name, parent_ctx, out);
+    GF_VALIDATE_OR_GOTO(this->name, entry, out);
+
+    if (!strcmp(entry->d_name, ".") || !strcmp(entry->d_name, ".."))
+        goto out;
+
+    inode = inode_grep(parent->table, parent, entry->d_name);
+    if (inode) {
+        entry->inode = inode;
+        inode_ctx = svs_inode_ctx_get(this, inode);
+        if (!inode_ctx) {
+            gf_uuid_copy(buf.ia_gfid, inode->gfid);
+            svs_iatt_fill(inode->gfid, &buf);
+            buf.ia_type = inode->ia_type;
+        } else {
+            buf = inode_ctx->buf;
+        }
+
+        entry->d_ino = buf.ia_ino;
+
+        if (parent_ctx->type == SNAP_VIEW_ENTRY_POINT_INODE)
+            entry->d_stat = buf;
+        else {
+            entry->d_stat.ia_ino = buf.ia_ino;
+            gf_uuid_copy(entry->d_stat.ia_gfid, buf.ia_gfid);
+        }
+    } else {
+        if (parent_ctx->type == SNAP_VIEW_ENTRY_POINT_INODE) {
+            inode = inode_new(parent->table);
+            entry->inode = inode;
+
+            /* If inode context allocation fails, then do not send
+             * the inode for that particular entry as part of
+             * readdirp response. Fuse and protocol/server will link
+             * the inodes in readdirp only if the entry contains
+             * inode in it.
+             */
+            inode_ctx = svs_inode_ctx_get_or_new(this, inode);
+            if (!inode_ctx) {
+                gf_msg(this->name, GF_LOG_ERROR, ENOMEM, SVS_MSG_NO_MEMORY,
+                       "failed to allocate inode "
+                       "context for %s",
+                       entry->d_name);
+                inode_unref(entry->inode);
+                entry->inode = NULL;
+                goto out;
+            }
+
+            /* Generate virtual gfid for SNAPSHOT dir and
+             * update the statbuf
+             */
+            gf_uuid_generate(random_gfid);
+            gf_uuid_copy(buf.ia_gfid, random_gfid);
+            svs_fill_ino_from_gfid(&buf);
+            buf.ia_type = IA_IFDIR;
+            entry->d_ino = buf.ia_ino;
+            entry->d_stat = buf;
+            inode_ctx->buf = buf;
+            inode_ctx->type = SNAP_VIEW_SNAPSHOT_INODE;
+        } else {
+            /* For files under snapshot world do not set
+             * entry->inode and reset statbuf (except ia_ino),
+             * so that FUSE/Kernel will send an explicit lookup.
+             * entry->d_stat contains the statbuf information
+             * of original file, so for NFS not to cache this
+             * information and to send explicit lookup, it is
+             * required to reset the statbuf.
+             * Virtual gfid for these files will be generated in the
+             * first lookup.
+             */
+            buf.ia_ino = entry->d_ino;
+            entry->d_stat = buf;
+        }
+    }
+
+out:
+    return;
+}
+
+/* In readdirp, though new inode is created along with the generation of
+   new gfid, the inode context created will not contain the glfs_t instance
+   for the filesystem it belongs to and the handle for it in the gfapi
+   world. (handle is obtained only by doing the lookup call on the entry
+   and doing lookup on each entry received as part of readdir call is a
+   costly operation. So the fs and handle is NULL in the inode context
+   and is filled in when lookup comes on that object.
+*/
+int32_t
+svs_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+             off_t off, dict_t *dict)
+{
+    gf_dirent_t entries;
+    gf_dirent_t *entry = NULL;
+    struct iatt buf = {
+        0,
+    };
+    int count = 0;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    svs_inode_t *parent_ctx = NULL;
+    svs_fd_t *svs_fd = NULL;
+    call_stack_t *root = NULL;
+
+    GF_VALIDATE_OR_GOTO("snap-view-daemon", this, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, frame, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, fd, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, fd->inode, unwind);
+
+    INIT_LIST_HEAD(&entries.list);
+
+    root = frame->root;
+    op_ret = gf_setcredentials(&root->uid, &root->gid, root->ngrps,
+                               root->groups);
+    if (op_ret != 0) {
+        goto unwind;
+    }
+
+    parent_ctx = svs_inode_ctx_get(this, fd->inode);
+    if (!parent_ctx) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno,
+               SVS_MSG_GET_INODE_CONTEXT_FAILED,
+               "failed to get the inode "
+               "context for %s",
+               uuid_utoa(fd->inode->gfid));
+        goto unwind;
+    }
+
+    if (parent_ctx->type == SNAP_VIEW_ENTRY_POINT_INODE) {
+        LOCK(&fd->lock);
+        {
+            count = svs_fill_readdir(this, &entries, size, off);
+        }
+        UNLOCK(&fd->lock);
+
+        op_ret = count;
+
+        list_for_each_entry(entry, &entries.list, list)
+        {
+            svs_readdirp_fill(this, fd->inode, parent_ctx, entry);
+        }
+
+        goto unwind;
+    } else {
+        svs_fd = svs_fd_ctx_get_or_new(this, fd);
+        if (!svs_fd) {
+            op_ret = -1;
+            op_errno = EBADFD;
+            gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                   SVS_MSG_GET_FD_CONTEXT_FAILED,
+                   "failed to get the fd context "
+                   "for the inode %s",
+                   uuid_utoa(fd->inode->gfid));
+            goto unwind;
+        }
+
+        glfs_seekdir(svs_fd->fd, off);
+
+        LOCK(&fd->lock);
+        {
+            count = svs_glfs_readdir(this, svs_fd->fd, &entries, &op_errno,
+                                     &buf, _gf_true, size);
+        }
+        UNLOCK(&fd->lock);
+
+        op_ret = count;
+
+        list_for_each_entry(entry, &entries.list, list)
+        {
+            svs_readdirp_fill(this, fd->inode, parent_ctx, entry);
+        }
+
+        goto unwind;
+    }
+
+unwind:
+    STACK_UNWIND_STRICT(readdirp, frame, op_ret, op_errno, &entries, dict);
+
+    gf_dirent_free(&entries);
+
+    return 0;
+}
+
+int32_t
+svs_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t off, dict_t *xdata)
+{
+    gf_dirent_t entries = {
+        {
+            {
+                0,
+            },
+        },
+    };
+    int count = 0;
+    svs_inode_t *inode_ctx = NULL;
+    int op_errno = EINVAL;
+    int op_ret = -1;
+    svs_fd_t *svs_fd = NULL;
+    glfs_fd_t *glfd = NULL;
+
+    INIT_LIST_HEAD(&entries.list);
+
+    GF_VALIDATE_OR_GOTO("snap-view-server", this, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, frame, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, fd, unwind);
+    GF_VALIDATE_OR_GOTO(this->name, fd->inode, unwind);
+
+    inode_ctx = svs_inode_ctx_get(this, fd->inode);
+    if (!inode_ctx) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno,
+               SVS_MSG_GET_INODE_CONTEXT_FAILED,
+               "inode context not found in "
+               "the inode %s",
+               uuid_utoa(fd->inode->gfid));
+        goto unwind;
+    }
+
+    if (inode_ctx->type == SNAP_VIEW_ENTRY_POINT_INODE) {
+        LOCK(&fd->lock);
+        {
+            count = svs_fill_readdir(this, &entries, size, off);
+        }
+        UNLOCK(&fd->lock);
+    } else {
+        svs_fd = svs_fd_ctx_get_or_new(this, fd);
+        if (!svs_fd) {
+            op_ret = -1;
+            op_errno = EBADFD;
+            gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                   SVS_MSG_GET_FD_CONTEXT_FAILED,
+                   "failed to get the fd "
+                   "context for %s",
+                   uuid_utoa(fd->inode->gfid));
+            goto unwind;
+        }
+
+        glfd = svs_fd->fd;
+
+        LOCK(&fd->lock);
+        {
+            count = svs_glfs_readdir(this, glfd, &entries, &op_errno, NULL,
+                                     _gf_false, size);
+        }
+        UNLOCK(&fd->lock);
+    }
+
+    op_ret = count;
+
+unwind:
+    STACK_UNWIND_STRICT(readdir, frame, op_ret, op_errno, &entries, xdata);
+
+    gf_dirent_free(&entries);
+
+    return 0;
+}
+
+/*
+ * This function is mainly helpful for NFS. Till now NFS server was not linking
+ * the inodes in readdirp, which caused problems when below operations were
+ * performed.
+ *
+ * 1) ls -l in one of the snaopshots (snapview-server would generate gfids for
+ *    each entry on the fly and link the inodes associated with those entries)
+ * 2) NFS server upon getting readdirp reply would not link the inodes of the
+ *    entries. But it used to generate filehandles for each entry and associate
+ *    the gfid of that entry with the filehandle and send it as part of the
+ *    reply to nfs client.
+ * 3) NFS client would send the filehandle of one of those entries when some
+ *    activity is done on it.
+ * 4) NFS server would not be able to find the inode for the gfid present in the
+ *    filehandle (as the inode was not linked) and would go for hard resolution
+ *    by sending a lookup on the gfid by creating a new inode.
+ * 5) snapview-client will not able to identify whether the inode is a real
+ *    inode existing in the main volume or a virtual inode existing in the
+ *    snapshots as there would not be any inode context.
+ * 6) Since the gfid upon which lookup is sent is a virtual gfid which is not
+ *    present in the disk, lookup would fail and the application would get an
+ *    error.
+ *
+ * The above problem is fixed by the below commit which makes snapview server
+ * more compatible with nfs server (1dea949cb60c3814c9206df6ba8dddec8d471a94).
+ * But now because NFS server does inode linking in readdirp has introduced
+ * the below issue.
+ * In readdirp though snapview-server allocates inode contexts it does not
+ * actually perform lookup on each entry it obtained in readdirp (as doing
+ * a lookup via gfapi over the network for each entry would be costly).
+ *
+ * Till now it was not a problem with NFS server, as NFS was sending a lookup on
+ * the gfid it got from NFS client, for which it was not able to find the right
+ * inode. So snapview-server was able to get the fs instance (glfs_t) of the
+ * snapshot volume to which the entry belongs to, and the handle for the entry
+ * from the corresponding snapshot volume and fill those information in the
+ * inode context.
+ *
+ * But now, since NFS server is able to find the inode from the inode table for
+ * the gfid it got from the NFS client, it won't send lookup. Rather it directly
+ * sends the fop it received from the client. Now this causes problems for
+ * snapview-server. Because for each fop snapview-server assumes that lookup has
+ * been performed on that entry and the entry's inode context contains the
+ * pointers for the fs instance and the handle to the entry in that fs. When NFS
+ * server sends the fop and snapview-server finds that the fs instance and the
+ * handle within the inode context are NULL it unwinds with EINVAL.
+ *
+ * So to handle this, if fs instance or handle within the inode context are
+ * NULL, then do a lookup based on parent inode context's fs instance. And
+ * unwind the results obtained as part of lookup
+ */
+
+int32_t
+svs_get_handle(xlator_t *this, loc_t *loc, svs_inode_t *inode_ctx,
+               int32_t *op_errno)
+{
+    svs_inode_t *parent_ctx = NULL;
+    int ret = -1;
+    inode_t *parent = NULL;
+    struct iatt postparent = {
+        0,
+    };
+    struct iatt buf = {
+        0,
+    };
+    char uuid1[64];
+
+    GF_VALIDATE_OR_GOTO("snap-view-daemon", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    if (loc->path) {
+        if (!loc->name || (loc->name && !strcmp(loc->name, ""))) {
+            loc->name = strrchr(loc->path, '/');
+            if (loc->name)
+                loc->name++;
+        }
+    }
+
+    if (loc->parent)
+        parent = inode_ref(loc->parent);
+    else {
+        parent = inode_find(loc->inode->table, loc->pargfid);
+        if (!parent)
+            parent = inode_parent(loc->inode, NULL, NULL);
+    }
+
+    if (parent)
+        parent_ctx = svs_inode_ctx_get(this, parent);
+
+    if (!parent_ctx) {
+        *op_errno = EINVAL;
+        gf_msg(this->name, GF_LOG_WARNING, *op_errno,
+               SVS_MSG_GET_INODE_CONTEXT_FAILED,
+               "failed to get the parent "
+               "context for %s (%s)",
+               loc->path, uuid_utoa_r(loc->inode->gfid, uuid1));
+        goto out;
+    }
+
+    if (parent_ctx) {
+        if (parent_ctx->type == SNAP_VIEW_ENTRY_POINT_INODE)
+            ret = svs_lookup_snapshot(this, loc, &buf, &postparent, parent,
+                                      parent_ctx, op_errno);
+        else
+            ret = svs_lookup_entry(this, loc, &buf, &postparent, parent,
+                                   parent_ctx, op_errno);
+    }
+
+out:
+    if (parent)
+        inode_unref(parent);
+
+    return ret;
+}
+
+int32_t
+svs_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    struct iatt buf = {
+        0,
+    };
+    int32_t op_errno = EINVAL;
+    int32_t op_ret = -1;
+    svs_inode_t *inode_ctx = NULL;
+    glfs_t *fs = NULL;
+    glfs_object_t *object = NULL;
+    struct stat stat = {
+        0,
+    };
+    int ret = -1;
+    call_stack_t *root = NULL;
+
+    GF_VALIDATE_OR_GOTO("snap-view-daemon", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    root = frame->root;
+    op_ret = gf_setcredentials(&root->uid, &root->gid, root->ngrps,
+                               root->groups);
+    if (op_ret != 0) {
+        goto out;
+    }
+
+    /* Instead of doing the check of whether it is a entry point directory
+       or not by checking the name of the entry and then deciding what
+       to do, just check the inode context and decide what to be done.
+    */
+
+    inode_ctx = svs_inode_ctx_get(this, loc->inode);
+    if (!inode_ctx) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno,
+               SVS_MSG_GET_INODE_CONTEXT_FAILED,
+               "inode context not found for %s", uuid_utoa(loc->inode->gfid));
+        goto out;
+    }
+
+    if (inode_ctx->type == SNAP_VIEW_ENTRY_POINT_INODE) {
+        svs_iatt_fill(loc->inode->gfid, &buf);
+        op_ret = 0;
+    } else {
+        SVS_GET_INODE_CTX_INFO(inode_ctx, fs, object, this, loc, op_ret,
+                               op_errno, out);
+
+        ret = glfs_h_stat(fs, object, &stat);
+        if (ret) {
+            op_ret = -1;
+            op_errno = errno;
+            gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_STAT_FAILED,
+                   "glfs_h_stat on %s (gfid: %s) "
+                   "failed",
+                   loc->name, uuid_utoa(loc->inode->gfid));
+            goto out;
+        } else
+            gf_msg_debug(this->name, 0, "stat on %s (%s) successful", loc->path,
+                         uuid_utoa(loc->inode->gfid));
+
+        iatt_from_stat(&buf, &stat);
+        gf_uuid_copy(buf.ia_gfid, loc->inode->gfid);
+        svs_fill_ino_from_gfid(&buf);
+        op_ret = ret;
+    }
+
+out:
+    STACK_UNWIND_STRICT(stat, frame, op_ret, op_errno, &buf, xdata);
+    return 0;
+}
+
+int32_t
+svs_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    struct iatt buf = {
+        0,
+    };
+    int32_t op_errno = EINVAL;
+    int32_t op_ret = -1;
+    svs_inode_t *inode_ctx = NULL;
+    struct stat stat = {
+        0,
+    };
+    int ret = -1;
+    glfs_fd_t *glfd = NULL;
+    svs_fd_t *sfd = NULL;
+    call_stack_t *root = NULL;
+
+    GF_VALIDATE_OR_GOTO("snap-view-daemon", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd->inode, out);
+
+    /* Instead of doing the check of whether it is a entry point directory
+       or not by checking the name of the entry and then deciding what
+       to do, just check the inode context and decide what to be done.
+    */
+
+    root = frame->root;
+    op_ret = gf_setcredentials(&root->uid, &root->gid, root->ngrps,
+                               root->groups);
+    if (op_ret != 0) {
+        goto out;
+    }
+
+    inode_ctx = svs_inode_ctx_get(this, fd->inode);
+    if (!inode_ctx) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno,
+               SVS_MSG_GET_INODE_CONTEXT_FAILED,
+               "inode context not found for"
+               " the inode %s",
+               uuid_utoa(fd->inode->gfid));
+        goto out;
+    }
+
+    if (inode_ctx->type == SNAP_VIEW_ENTRY_POINT_INODE) {
+        svs_iatt_fill(fd->inode->gfid, &buf);
+        op_ret = 0;
+    } else {
+        if (!(svs_inode_ctx_glfs_mapping(this, inode_ctx))) {
+            op_ret = -1;
+            op_errno = EBADF;
+            gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                   SVS_MSG_FS_INSTANCE_INVALID,
+                   "glfs instance %p to which the inode %s "
+                   "belongs to does not exist. That snapshot "
+                   "corresponding to the fs instance "
+                   "might have been deleted or deactivated.",
+                   inode_ctx->fs, uuid_utoa(fd->inode->gfid));
+            goto out;
+        }
+
+        sfd = svs_fd_ctx_get_or_new(this, fd);
+        if (!sfd) {
+            op_ret = -1;
+            op_errno = EBADFD;
+            gf_msg(this->name, GF_LOG_ERROR, op_errno,
+                   SVS_MSG_GET_FD_CONTEXT_FAILED,
+                   "failed to get the fd context "
+                   "for %s",
+                   uuid_utoa(fd->inode->gfid));
+            goto out;
+        }
+
+        glfd = sfd->fd;
+        ret = glfs_fstat(glfd, &stat);
+        if (ret) {
+            op_ret = -1;
+            op_errno = errno;
+            gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_STAT_FAILED,
+                   "glfs_fstat on gfid: %s failed", uuid_utoa(fd->inode->gfid));
+            goto out;
+        }
+
+        iatt_from_stat(&buf, &stat);
+        gf_uuid_copy(buf.ia_gfid, fd->inode->gfid);
+        svs_fill_ino_from_gfid(&buf);
+        op_ret = ret;
+    }
+
+out:
+    STACK_UNWIND_STRICT(fstat, frame, op_ret, op_errno, &buf, xdata);
+    return 0;
+}
+
+int32_t
+svs_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    struct statvfs buf = {
+        0,
+    };
+    int32_t op_errno = EINVAL;
+    int32_t op_ret = -1;
+    svs_inode_t *inode_ctx = NULL;
+    glfs_t *fs = NULL;
+    glfs_object_t *object = NULL;
+    int ret = -1;
+    call_stack_t *root = NULL;
+
+    GF_VALIDATE_OR_GOTO("snap-view-daemon", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    root = frame->root;
+    op_ret = gf_setcredentials(&root->uid, &root->gid, root->ngrps,
+                               root->groups);
+    if (op_ret != 0) {
+        goto out;
+    }
+
+    /* Instead of doing the check of whether it is a entry point directory
+       or not by checking the name of the entry and then deciding what
+       to do, just check the inode context and decide what to be done.
+    */
+    inode_ctx = svs_inode_ctx_get(this, loc->inode);
+    if (!inode_ctx) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno,
+               SVS_MSG_GET_INODE_CONTEXT_FAILED,
+               "inode context not found for %s", uuid_utoa(loc->inode->gfid));
+        goto out;
+    }
+
+    SVS_GET_INODE_CTX_INFO(inode_ctx, fs, object, this, loc, op_ret, op_errno,
+                           out);
+
+    ret = glfs_h_statfs(fs, object, &buf);
+    if (ret) {
+        op_ret = -1;
+        op_errno = errno;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_STATFS_FAILED,
+               "glfs_h_statvfs on %s (gfid: %s) "
+               "failed",
+               loc->name, uuid_utoa(loc->inode->gfid));
+        goto out;
+    }
+    op_ret = ret;
+
+out:
+    STACK_UNWIND_STRICT(statfs, frame, op_ret, op_errno, &buf, xdata);
+    return 0;
+}
+
+int32_t
+svs_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+         fd_t *fd, dict_t *xdata)
+{
+    svs_inode_t *inode_ctx = NULL;
+    svs_fd_t *sfd = NULL;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    glfs_fd_t *glfd = NULL;
+    glfs_t *fs = NULL;
+    glfs_object_t *object = NULL;
+    call_stack_t *root = NULL;
+
+    GF_VALIDATE_OR_GOTO("snap-view-daemon", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    root = frame->root;
+
+    inode_ctx = svs_inode_ctx_get(this, loc->inode);
+    if (!inode_ctx) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno,
+               SVS_MSG_GET_INODE_CONTEXT_FAILED,
+               "inode context for %s (gfid: %s) "
+               "not found",
+               loc->name, uuid_utoa(loc->inode->gfid));
+        goto out;
+    }
+
+    if (inode_ctx->type == SNAP_VIEW_ENTRY_POINT_INODE)
+        GF_ASSERT(0);  // on entry point it should always be opendir
+
+    SVS_GET_INODE_CTX_INFO(inode_ctx, fs, object, this, loc, op_ret, op_errno,
+                           out);
+
+    op_ret = gf_setcredentials(&root->uid, &root->gid, root->ngrps,
+                               root->groups);
+    if (op_ret != 0) {
+        goto out;
+    }
+
+    glfd = glfs_h_open(fs, object, flags);
+    if (!glfd) {
+        op_ret = -1;
+        op_errno = errno;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_OPEN_FAILED,
+               "glfs_h_open on %s failed (gfid: %s)", loc->name,
+               uuid_utoa(loc->inode->gfid));
+        goto out;
+    }
+
+    sfd = svs_fd_ctx_get_or_new(this, fd);
+    if (!sfd) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_NO_MEMORY,
+               "failed to allocate fd context "
+               "for %s (gfid: %s)",
+               loc->name, uuid_utoa(loc->inode->gfid));
+        glfs_close(glfd);
+        goto out;
+    }
+    sfd->fd = glfd;
+
+    op_ret = 0;
+
+out:
+    STACK_UNWIND_STRICT(open, frame, op_ret, op_errno, fd, NULL);
+    return 0;
+}
+
+int32_t
+svs_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+          off_t offset, uint32_t flags, dict_t *xdata)
+{
+    int32_t op_ret = -1;
+    int32_t op_errno = 0;
+    svs_private_t *priv = NULL;
+    struct iobuf *iobuf = NULL;
+    struct iobref *iobref = NULL;
+    struct iovec vec = {
+        0,
+    };
+    svs_fd_t *sfd = NULL;
+    int ret = -1;
+    struct glfs_stat fstatbuf = {
+        0,
+    };
+    glfs_fd_t *glfd = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+    call_stack_t *root = NULL;
+
+    GF_VALIDATE_OR_GOTO("snap-view-daemon", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd, out);
+    GF_VALIDATE_OR_GOTO(this->name, fd->inode, out);
+
+    priv = this->private;
+    VALIDATE_OR_GOTO(priv, out);
+
+    root = frame->root;
+    op_ret = gf_setcredentials(&root->uid, &root->gid, root->ngrps,
+                               root->groups);
+    if (op_ret != 0) {
+        goto out;
+    }
+
+    if (!svs_inode_glfs_mapping(this, fd->inode)) {
+        op_ret = -1;
+        op_errno = EBADF; /* should this be some other error? */
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_FS_INSTANCE_INVALID,
+               "glfs instance to which the inode "
+               "%s receiving read request belongs, "
+               "does not exist anymore",
+               uuid_utoa(fd->inode->gfid));
+        goto out;
+    }
+
+    sfd = svs_fd_ctx_get_or_new(this, fd);
+    if (!sfd) {
+        op_ret = -1;
+        op_errno = EBADFD;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno,
+               SVS_MSG_GET_INODE_CONTEXT_FAILED,
+               "failed to get the fd "
+               "context for %s",
+               uuid_utoa(fd->inode->gfid));
+        goto out;
+    }
+
+    glfd = sfd->fd;
+
+    iobuf = iobuf_get2(this->ctx->iobuf_pool, size);
+    if (!iobuf) {
+        op_ret = -1;
+        op_errno = ENOMEM;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_NO_MEMORY,
+               "failed to "
+               "allocate iobuf while reading the "
+               "file with gfid %s",
+               uuid_utoa(fd->inode->gfid));
+        goto out;
+    }
+
+    ret = glfs_pread(glfd, iobuf->ptr, size, offset, 0, &fstatbuf);
+    if (ret < 0) {
+        op_ret = -1;
+        op_errno = errno;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_READ_FAILED,
+               "glfs_read failed on %s (%s)", uuid_utoa(fd->inode->gfid),
+               strerror(op_errno));
+        goto out;
+    }
+
+    vec.iov_base = iobuf->ptr;
+    vec.iov_len = ret;
+
+    iobref = iobref_new();
+
+    iobref_add(iobref, iobuf);
+    glfs_iatt_from_statx(&stbuf, &fstatbuf);
+    gf_uuid_copy(stbuf.ia_gfid, fd->inode->gfid);
+    svs_fill_ino_from_gfid(&stbuf);
+
+    /* Hack to notify higher layers of EOF. */
+    if (!stbuf.ia_size || (offset + vec.iov_len) >= stbuf.ia_size)
+        op_errno = ENOENT;
+
+    op_ret = vec.iov_len;
+
+out:
+
+    STACK_UNWIND_STRICT(readv, frame, op_ret, op_errno, &vec, 1, &stbuf, iobref,
+                        NULL);
+
+    if (iobref)
+        iobref_unref(iobref);
+    if (iobuf)
+        iobuf_unref(iobuf);
+
+    return 0;
+}
+
+int32_t
+svs_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+             dict_t *xdata)
+{
+    svs_inode_t *inode_ctx = NULL;
+    glfs_t *fs = NULL;
+    glfs_object_t *object = NULL;
+    int op_ret = -1;
+    int op_errno = EINVAL;
+    char *buf = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+    int ret = -1;
+    struct stat stat = {
+        0,
+    };
+    call_stack_t *root = NULL;
+
+    GF_VALIDATE_OR_GOTO("snap-view-daemon", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    root = frame->root;
+    op_ret = gf_setcredentials(&root->uid, &root->gid, root->ngrps,
+                               root->groups);
+    if (op_ret != 0) {
+        goto out;
+    }
+
+    inode_ctx = svs_inode_ctx_get(this, loc->inode);
+    if (!inode_ctx) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno,
+               SVS_MSG_GET_INODE_CONTEXT_FAILED,
+               "failed to get inode context "
+               "for %s (gfid: %s)",
+               loc->name, uuid_utoa(loc->inode->gfid));
+        goto out;
+    }
+
+    SVS_GET_INODE_CTX_INFO(inode_ctx, fs, object, this, loc, op_ret, op_errno,
+                           out);
+
+    ret = glfs_h_stat(fs, object, &stat);
+    if (ret) {
+        op_ret = -1;
+        op_errno = errno;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_STAT_FAILED,
+               "glfs_h_stat on %s (gfid: %s) "
+               "failed",
+               loc->name, uuid_utoa(loc->inode->gfid));
+        goto out;
+    }
+
+    iatt_from_stat(&stbuf, &stat);
+    gf_uuid_copy(stbuf.ia_gfid, loc->inode->gfid);
+    svs_fill_ino_from_gfid(&stbuf);
+
+    buf = alloca(size + 1);
+    op_ret = glfs_h_readlink(fs, object, buf, size);
+    if (op_ret == -1) {
+        op_errno = errno;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_READLINK_FAILED,
+               "readlink on %s failed (gfid: %s)", loc->name,
+               uuid_utoa(loc->inode->gfid));
+        goto out;
+    }
+
+    buf[op_ret] = 0;
+
+out:
+    STACK_UNWIND_STRICT(readlink, frame, op_ret, op_errno, buf, &stbuf, NULL);
+
+    return 0;
+}
+
+int32_t
+svs_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int mask,
+           dict_t *xdata)
+{
+    int ret = -1;
+    int32_t op_ret = -1;
+    int32_t op_errno = EINVAL;
+    glfs_t *fs = NULL;
+    glfs_object_t *object = NULL;
+    svs_inode_t *inode_ctx = NULL;
+    gf_boolean_t is_fuse_call = 0;
+    int mode = 0;
+    call_stack_t *root = NULL;
+
+    GF_VALIDATE_OR_GOTO("svs", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, this->private, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc, out);
+    GF_VALIDATE_OR_GOTO(this->name, loc->inode, out);
+
+    root = frame->root;
+    op_ret = gf_setcredentials(&root->uid, &root->gid, root->ngrps,
+                               root->groups);
+    if (op_ret != 0) {
+        goto out;
+    }
+
+    inode_ctx = svs_inode_ctx_get(this, loc->inode);
+    if (!inode_ctx) {
+        op_ret = -1;
+        op_errno = EINVAL;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno,
+               SVS_MSG_GET_INODE_CONTEXT_FAILED,
+               "inode context not found for %s", uuid_utoa(loc->inode->gfid));
+        goto out;
+    }
+
+    is_fuse_call = __is_fuse_call(frame);
+
+    /*
+     * For entry-point directory, set read and execute bits. But not write
+     * permissions.
+     */
+    if (inode_ctx->type == SNAP_VIEW_ENTRY_POINT_INODE) {
+        if (is_fuse_call) {
+            op_ret = 0;
+            op_errno = 0;
+        } else {
+            op_ret = 0;
+            mode |= POSIX_ACL_READ;
+            mode |= POSIX_ACL_EXECUTE;
+            op_errno = mode;
+        }
+        goto out;
+    }
+
+    SVS_GET_INODE_CTX_INFO(inode_ctx, fs, object, this, loc, op_ret, op_errno,
+                           out);
+
+    /* The actual posix_acl xlator does acl checks differently for
+       fuse and nfs. So set frame->root->pid as fspid of the syncop
+       if the call came from nfs
+    */
+    if (!is_fuse_call) {
+        syncopctx_setfspid(&frame->root->pid);
+        syncopctx_setfsuid(&frame->root->uid);
+        syncopctx_setfsgid(&frame->root->gid);
+        syncopctx_setfsgroups(frame->root->ngrps, frame->root->groups);
+    }
+
+    ret = glfs_h_access(fs, object, mask);
+    if (ret < 0) {
+        op_ret = -1;
+        op_errno = errno;
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, SVS_MSG_ACCESS_FAILED,
+               "failed to access %s (gfid: %s)", loc->path,
+               uuid_utoa(loc->inode->gfid));
+        goto out;
+    }
+
+    op_ret = 0;
+    op_errno = ret;
+
+out:
+
+    STACK_UNWIND_STRICT(access, frame, op_ret, op_errno, NULL);
+    return 0;
+}
+
+int32_t
+notify(xlator_t *this, int32_t event, void *data, ...)
+{
+    switch (event) {
+        case GF_EVENT_PARENT_UP: {
+            /* Tell the parent that snapview-server xlator is up */
+            default_notify(this, GF_EVENT_CHILD_UP, data);
+        } break;
+        default:
+            break;
+    }
+    return 0;
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    if (!this)
+        return ret;
+
+    ret = xlator_mem_acct_init(this, gf_svs_mt_end + 1);
+
+    if (ret != 0) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, SVS_MSG_MEM_ACNT_FAILED,
+               "Memory accounting"
+               " init failed");
+        return ret;
+    }
+
+    return ret;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    svs_private_t *priv = NULL;
+    int ret = -1;
+
+    /* This can be the top of graph in certain cases */
+    if (!this->parents) {
+        gf_msg_debug(this->name, 0, "dangling volume. check volfile ");
+    }
+
+    priv = GF_CALLOC(1, sizeof(*priv), gf_svs_mt_priv_t);
+    if (!priv) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, SVS_MSG_NO_MEMORY,
+               "failed to "
+               "allocate memory for this->private ");
+        goto out;
+    }
+
+    this->private = priv;
+
+    GF_OPTION_INIT("volname", priv->volname, str, out);
+    LOCK_INIT(&priv->snaplist_lock);
+
+    LOCK(&priv->snaplist_lock);
+    {
+        priv->num_snaps = 0;
+    }
+    UNLOCK(&priv->snaplist_lock);
+
+    /* What to do here upon failure? should init be failed or succeed? */
+    /* If succeeded, then dynamic management of snapshots will not */
+    /* happen.*/
+    ret = svs_mgmt_init(this);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, EINVAL, SVS_MSG_MGMT_INIT_FAILED,
+               "failed to initiate the "
+               "mgmt rpc callback for svs. Dymamic management of the"
+               "snapshots will not happen");
+        goto out;
+    }
+
+    /* get the list of snaps first to return to client xlator */
+    ret = svs_get_snapshot_list(this);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL,
+               SVS_MSG_GET_SNAPSHOT_LIST_FAILED,
+               "Error initializing snaplist infrastructure");
+        ret = -1;
+        goto out;
+    }
+
+    ret = 0;
+
+out:
+    if (ret && priv) {
+        LOCK_DESTROY(&priv->snaplist_lock);
+        GF_FREE(priv->dirents);
+        GF_FREE(priv);
+    }
+
+    return ret;
+}
+
+void
+fini(xlator_t *this)
+{
+    svs_private_t *priv = NULL;
+    glusterfs_ctx_t *ctx = NULL;
+    int ret = 0;
+
+    GF_ASSERT(this);
+    priv = this->private;
+    this->private = NULL;
+    ctx = this->ctx;
+    if (!ctx)
+        gf_msg(this->name, GF_LOG_ERROR, 0, SVS_MSG_INVALID_GLFS_CTX,
+               "Invalid ctx found");
+
+    if (priv) {
+        ret = LOCK_DESTROY(&priv->snaplist_lock);
+        if (ret != 0) {
+            gf_msg(this->name, GF_LOG_WARNING, errno,
+                   SVS_MSG_LOCK_DESTROY_FAILED,
+                   "Could not destroy mutex snaplist_lock");
+        }
+
+        if (priv->dirents) {
+            GF_FREE(priv->dirents);
+        }
+
+        if (priv->rpc) {
+            /* cleanup the saved-frames before last unref */
+            rpc_clnt_connection_cleanup(&priv->rpc->conn);
+            rpc_clnt_unref(priv->rpc);
+        }
+
+        GF_FREE(priv);
+    }
+
+    return;
+}
+
+struct xlator_fops fops = {
+    .lookup = svs_lookup,
+    .stat = svs_stat,
+    .statfs = svs_statfs,
+    .opendir = svs_opendir,
+    .readdirp = svs_readdirp,
+    .readdir = svs_readdir,
+    .open = svs_open,
+    .readv = svs_readv,
+    .flush = svs_flush,
+    .fstat = svs_fstat,
+    .getxattr = svs_getxattr,
+    .access = svs_access,
+    .readlink = svs_readlink,
+    /* entry fops */
+};
+
+struct xlator_cbks cbks = {
+    .release = svs_release,
+    .releasedir = svs_releasedir,
+    .forget = svs_forget,
+};
+
+struct volume_options options[] = {
+    {
+        .key = {"volname"},
+        .type = GF_OPTION_TYPE_STR,
+    },
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1},
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "snapview-server",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/features/snapview-server/src/snapview-server.h b/xlators/features/snapview-server/src/snapview-server.h
new file mode 100644
index 00000000000..6472422e715
--- /dev/null
+++ b/xlators/features/snapview-server/src/snapview-server.h
@@ -0,0 +1,255 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#ifndef __SNAP_VIEW_H__
+#define __SNAP_VIEW_H__
+
+#include <glusterfs/dict.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/mem-types.h>
+#include <glusterfs/call-stub.h>
+#include <glusterfs/byte-order.h>
+#include <glusterfs/iatt.h>
+#include <ctype.h>
+#include <sys/uio.h>
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/logging.h>
+#include "glfs.h"
+#include "glfs-handles.h"
+#include "glfs-internal.h"
+#include "glusterfs3-xdr.h"
+#include <glusterfs/glusterfs-acl.h>
+#include <glusterfs/syncop.h>
+#include <glusterfs/list.h>
+#include <glusterfs/timer.h>
+#include "rpc-clnt.h"
+#include "protocol-common.h"
+#include "xdr-generic.h"
+#include "snapview-server-messages.h"
+
+#define DEFAULT_SVD_LOG_FILE_DIRECTORY DATADIR "/log/glusterfs"
+
+#define SNAP_VIEW_MAX_GLFS_T 256
+#define SNAP_VIEW_MAX_GLFS_FDS 1024
+#define SNAP_VIEW_MAX_GLFS_OBJ_HANDLES 1024
+
+#define SVS_STACK_DESTROY(_frame)                                              \
+    do {                                                                       \
+        ((call_frame_t *)_frame)->local = NULL;                                \
+        STACK_DESTROY(((call_frame_t *)_frame)->root);                         \
+    } while (0)
+
+#define SVS_CHECK_VALID_SNAPSHOT_HANDLE(fs, this)                              \
+    do {                                                                       \
+        svs_private_t *_private = NULL;                                        \
+        _private = this->private;                                              \
+        int i = 0;                                                             \
+        gf_boolean_t found = _gf_false;                                        \
+        glfs_t *tmp_fs = NULL;                                                 \
+        LOCK(&_private->snaplist_lock);                                        \
+        {                                                                      \
+            for (i = 0; i < _private->num_snaps; i++) {                        \
+                tmp_fs = _private->dirents[i].fs;                              \
+                gf_log(this->name, GF_LOG_DEBUG,                               \
+                       "snap name: %s, snap volume: %s,"                       \
+                       "dirent->fs: %p",                                       \
+                       _private->dirents[i].name,                              \
+                       _private->dirents[i].snap_volname, tmp_fs);             \
+                if (tmp_fs && fs && (tmp_fs == fs)) {                          \
+                    found = _gf_true;                                          \
+                    gf_msg_debug(this->name, 0,                                \
+                                 "found the fs "                               \
+                                 "instance");                                  \
+                    break;                                                     \
+                }                                                              \
+            }                                                                  \
+        }                                                                      \
+        UNLOCK(&_private->snaplist_lock);                                      \
+                                                                               \
+        if (!found) {                                                          \
+            gf_log(this->name, GF_LOG_WARNING,                                 \
+                   "failed to"                                                 \
+                   " find the fs instance %p",                                 \
+                   fs);                                                        \
+            fs = NULL;                                                         \
+        }                                                                      \
+    } while (0)
+
+#define SVS_GET_INODE_CTX_INFO(inode_ctx, fs, object, this, loc, ret,          \
+                               op_errno, label)                                \
+    do {                                                                       \
+        fs = inode_ctx->fs;                                                    \
+        object = inode_ctx->object;                                            \
+        SVS_CHECK_VALID_SNAPSHOT_HANDLE(fs, this);                             \
+        if (!fs)                                                               \
+            object = NULL;                                                     \
+                                                                               \
+        if (!fs || !object) {                                                  \
+            int32_t tmp = -1;                                                  \
+            char tmp_uuid[64];                                                 \
+                                                                               \
+            tmp = svs_get_handle(this, loc, inode_ctx, &op_errno);             \
+            if (tmp) {                                                         \
+                gf_log(this->name, GF_LOG_ERROR,                               \
+                       "failed to get the handle for %s "                      \
+                       "(gfid: %s)",                                           \
+                       loc->path, uuid_utoa_r(loc->inode->gfid, tmp_uuid));    \
+                ret = -1;                                                      \
+                goto label;                                                    \
+            }                                                                  \
+                                                                               \
+            fs = inode_ctx->fs;                                                \
+            object = inode_ctx->object;                                        \
+        }                                                                      \
+    } while (0);
+
+#define SVS_STRDUP(dst, src)                                                   \
+    do {                                                                       \
+        if (dst && strcmp(src, dst)) {                                         \
+            GF_FREE(dst);                                                      \
+            dst = NULL;                                                        \
+        }                                                                      \
+                                                                               \
+        if (!dst)                                                              \
+            dst = gf_strdup(src);                                              \
+    } while (0)
+
+int
+svs_mgmt_submit_request(void *req, call_frame_t *frame, glusterfs_ctx_t *ctx,
+                        rpc_clnt_prog_t *prog, int procnum, fop_cbk_fn_t cbkfn,
+                        xdrproc_t xdrproc);
+
+int
+svs_get_snapshot_list(xlator_t *this);
+
+int
+mgmt_get_snapinfo_cbk(struct rpc_req *req, struct iovec *iov, int count,
+                      void *myframe);
+
+typedef enum {
+    SNAP_VIEW_ENTRY_POINT_INODE = 0,
+    SNAP_VIEW_SNAPSHOT_INODE,
+    SNAP_VIEW_VIRTUAL_INODE
+} inode_type_t;
+
+struct svs_inode {
+    glfs_t *fs;
+    glfs_object_t *object;
+    inode_type_t type;
+
+    /* used only for entry point directory where gfid of the directory
+       from where the entry point was entered is saved.
+    */
+    uuid_t pargfid;
+
+    /* This is used to generate gfid for all sub files/dirs under this
+     * snapshot
+     */
+    char *snapname;
+    struct iatt buf;
+};
+typedef struct svs_inode svs_inode_t;
+
+struct svs_fd {
+    glfs_fd_t *fd;
+};
+typedef struct svs_fd svs_fd_t;
+
+struct snap_dirent {
+    char name[NAME_MAX];
+    char uuid[UUID_CANONICAL_FORM_LEN + 1];
+    char snap_volname[NAME_MAX];
+    glfs_t *fs;
+};
+typedef struct snap_dirent snap_dirent_t;
+
+struct svs_private {
+    snap_dirent_t *dirents;
+    int num_snaps;
+    char *volname;
+    struct list_head snaplist;
+    gf_lock_t snaplist_lock;
+    struct rpc_clnt *rpc;
+};
+typedef struct svs_private svs_private_t;
+
+int
+__svs_inode_ctx_set(xlator_t *this, inode_t *inode, svs_inode_t *svs_inode);
+
+svs_inode_t *
+__svs_inode_ctx_get(xlator_t *this, inode_t *inode);
+
+svs_inode_t *
+svs_inode_ctx_get(xlator_t *this, inode_t *inode);
+
+int32_t
+svs_inode_ctx_set(xlator_t *this, inode_t *inode, svs_inode_t *svs_inode);
+
+svs_inode_t *
+svs_inode_ctx_get_or_new(xlator_t *this, inode_t *inode);
+
+int
+__svs_fd_ctx_set(xlator_t *this, fd_t *fd, svs_fd_t *svs_fd);
+
+svs_fd_t *
+__svs_fd_ctx_get(xlator_t *this, fd_t *fd);
+
+svs_fd_t *
+svs_fd_ctx_get(xlator_t *this, fd_t *fd);
+
+int32_t
+svs_fd_ctx_set(xlator_t *this, fd_t *fd, svs_fd_t *svs_fd);
+
+svs_fd_t *
+__svs_fd_ctx_get_or_new(xlator_t *this, fd_t *fd);
+
+svs_fd_t *
+svs_fd_ctx_get_or_new(xlator_t *this, fd_t *fd);
+
+int
+svs_uuid_generate(xlator_t *this, uuid_t gfid, char *snapname,
+                  uuid_t origin_gfid);
+
+void
+svs_fill_ino_from_gfid(struct iatt *buf);
+
+void
+svs_iatt_fill(uuid_t gfid, struct iatt *buf);
+
+snap_dirent_t *
+svs_get_latest_snap_entry(xlator_t *this);
+
+glfs_t *
+svs_get_latest_snapshot(xlator_t *this);
+
+glfs_t *
+svs_initialise_snapshot_volume(xlator_t *this, const char *name,
+                               int32_t *op_errno);
+
+glfs_t *
+__svs_initialise_snapshot_volume(xlator_t *this, const char *name,
+                                 int32_t *op_errno);
+
+snap_dirent_t *
+__svs_get_snap_dirent(xlator_t *this, const char *name);
+
+int
+svs_mgmt_init(xlator_t *this);
+
+int32_t
+svs_get_handle(xlator_t *this, loc_t *loc, svs_inode_t *inode_ctx,
+               int32_t *op_errno);
+
+glfs_t *
+svs_inode_glfs_mapping(xlator_t *this, inode_t *inode);
+
+glfs_t *
+svs_inode_ctx_glfs_mapping(xlator_t *this, svs_inode_t *inode_ctx);
+
+#endif /* __SNAP_VIEW_H__ */
diff --git a/xlators/features/thin-arbiter/Makefile.am b/xlators/features/thin-arbiter/Makefile.am
new file mode 100644
index 00000000000..a985f42a877
--- /dev/null
+++ b/xlators/features/thin-arbiter/Makefile.am
@@ -0,0 +1,3 @@
+SUBDIRS = src
+
+CLEANFILES =
diff --git a/xlators/features/thin-arbiter/src/Makefile.am b/xlators/features/thin-arbiter/src/Makefile.am
new file mode 100644
index 00000000000..a3c133e7798
--- /dev/null
+++ b/xlators/features/thin-arbiter/src/Makefile.am
@@ -0,0 +1,22 @@
+xlator_LTLIBRARIES = thin-arbiter.la
+
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+thin_arbiter_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+thin_arbiter_la_SOURCES = thin-arbiter.c \
+    $(top_builddir)/xlators/lib/src/libxlator.c
+
+thin_arbiter_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = thin-arbiter.h thin-arbiter-mem-types.h thin-arbiter-messages.h \
+    $(top_builddir)/xlators/lib/src/libxlator.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) \
+    -I$(top_srcdir)/libglusterfs/src -I$(top_srcdir)/xlators/lib/src \
+    -I$(top_srcdir)/rpc/rpc-lib/src \
+    -I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/thin-arbiter/src/thin-arbiter-mem-types.h b/xlators/features/thin-arbiter/src/thin-arbiter-mem-types.h
new file mode 100644
index 00000000000..69562d2febc
--- /dev/null
+++ b/xlators/features/thin-arbiter/src/thin-arbiter-mem-types.h
@@ -0,0 +1,19 @@
+/*
+  Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __THIN_ARBITER_MEM_TYPES_H__
+#define __THIN_ARBITER_MEM_TYPES_H__
+#include <glusterfs/mem-types.h>
+
+typedef enum gf_ta_mem_types_ {
+    gf_ta_mt_local_t = gf_common_mt_end + 1,
+    gf_ta_mt_char,
+    gf_ta_mt_end
+} gf_ta_mem_types_t;
+#endif
diff --git a/xlators/features/thin-arbiter/src/thin-arbiter-messages.h b/xlators/features/thin-arbiter/src/thin-arbiter-messages.h
new file mode 100644
index 00000000000..81d7491577a
--- /dev/null
+++ b/xlators/features/thin-arbiter/src/thin-arbiter-messages.h
@@ -0,0 +1,28 @@
+/*
+  Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _TA_MESSAGES_H_
+#define _TA_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(TA, TA_MSG_INVALID_FOP);
+
+#endif /* !_TA_MESSAGES_H_ */
diff --git a/xlators/features/thin-arbiter/src/thin-arbiter.c b/xlators/features/thin-arbiter/src/thin-arbiter.c
new file mode 100644
index 00000000000..ce3008636f1
--- /dev/null
+++ b/xlators/features/thin-arbiter/src/thin-arbiter.c
@@ -0,0 +1,661 @@
+/*
+  Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "thin-arbiter.h"
+#include "thin-arbiter-messages.h"
+#include "thin-arbiter-mem-types.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/byte-order.h>
+#include <glusterfs/common-utils.h>
+
+int
+ta_set_incoming_values(dict_t *dict, char *key, data_t *value, void *data)
+{
+    int32_t ret = 0;
+    ta_fop_t *fop = (ta_fop_t *)data;
+    int32_t *pending = NULL;
+
+    pending = GF_CALLOC(1, value->len, gf_ta_mt_char);
+    if (!pending) {
+        ret = -ENOMEM;
+        goto out;
+    }
+    ret = dict_set_bin(fop->brick_xattr, key, pending, value->len);
+out:
+    return ret;
+}
+
+int
+ta_get_incoming_and_brick_values(dict_t *dict, char *key, data_t *value,
+                                 void *data)
+{
+    ta_fop_t *fop = data;
+    char *source = NULL;
+    char *in_coming = NULL;
+    int32_t len = 0, ret = 0;
+
+    source = GF_CALLOC(1, value->len, gf_ta_mt_char);
+    if (!source) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    ret = dict_get_ptr_and_len(fop->dict, key, (void **)&in_coming, &len);
+
+    if (!in_coming || value->len != len) {
+        ret = -EINVAL;
+        goto out;
+    }
+
+    if (!memcmp(value->data, source, value->len) &&
+        (!memcmp(in_coming, source, len))) {
+        fop->on_disk[fop->idx] = 0;
+    } else {
+        fop->on_disk[fop->idx] = 1;
+    }
+
+    fop->idx++;
+out:
+    GF_FREE(source);
+    return ret;
+}
+
+void
+ta_release_fop(ta_fop_t *fop)
+{
+    if (!fop) {
+        return;
+    }
+    if (fop->fd) {
+        fd_unref(fop->fd);
+    }
+    loc_wipe(&fop->loc);
+    if (fop->dict) {
+        dict_unref(fop->dict);
+    }
+    if (fop->brick_xattr) {
+        dict_unref(fop->brick_xattr);
+    }
+
+    GF_FREE(fop);
+    return;
+}
+
+int32_t
+ta_set_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *dict,
+                   dict_t *xdata)
+{
+    TA_STACK_UNWIND(xattrop, frame, op_ret, op_errno, dict, xdata);
+    return 0;
+}
+
+/*
+case 1 - If brick value is 0 and incoming value is also 0, fine
+case 2 - If brick value is 0 and incoming value is non 0, fine
+case 3 - If brick value is non 0 and incoming value is also 0, fine
+case 4 - If brick value is non 0 and incoming value is non 0, fine
+case 5 - If incoming value is non zero on both brick, it is wrong
+case 6 - If incoming value is non zero but brick value for other
+brick is also non zero, wrong
+*/
+
+int32_t
+ta_verify_on_disk_source(ta_fop_t *fop, dict_t *dict)
+{
+    int ret = 0;
+
+    if (!fop) {
+        return -EINVAL;
+    }
+
+    ret = dict_foreach(dict, ta_get_incoming_and_brick_values, (void *)fop);
+    if (ret < 0) {
+        return ret;
+    }
+    if (fop->on_disk[0] && fop->on_disk[1]) {
+        return -EINVAL;
+    }
+    return 0;
+}
+
+int32_t
+ta_get_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *dict,
+                   dict_t *xdata)
+{
+    ta_fop_t *fop = NULL;
+    int ret = 0;
+
+    fop = frame->local;
+    if (op_ret) {
+        goto unwind;
+    }
+
+    ret = ta_verify_on_disk_source(fop, dict);
+    if (ret < 0) {
+        op_errno = -ret;
+        goto unwind;
+    }
+
+    if (fop->fd) {
+        STACK_WIND(frame, ta_set_xattrop_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->fxattrop, fop->fd,
+                   fop->xattrop_flags, fop->dict, NULL);
+    } else {
+        STACK_WIND(frame, ta_set_xattrop_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->xattrop, &fop->loc,
+                   fop->xattrop_flags, fop->dict, NULL);
+    }
+    return 0;
+
+unwind:
+
+    TA_STACK_UNWIND(xattrop, frame, -1, op_errno, NULL, NULL);
+    return -1;
+}
+
+ta_fop_t *
+ta_prepare_fop(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+               gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+    ta_fop_t *fop = NULL;
+    int ret = 0;
+
+    fop = GF_CALLOC(1, sizeof(*fop), gf_ta_mt_local_t);
+    if (!fop) {
+        goto out;
+    }
+
+    if (loc) {
+        loc_copy(&fop->loc, loc);
+    }
+
+    if (fd) {
+        fop->fd = fd_ref(fd);
+    }
+
+    fop->xattrop_flags = flags;
+    fop->idx = 0;
+
+    if (dict != NULL) {
+        fop->dict = dict_ref(dict);
+    }
+    fop->brick_xattr = dict_new();
+    if (fop->brick_xattr == NULL) {
+        goto out;
+    }
+    ret = dict_foreach(dict, ta_set_incoming_values, (void *)fop);
+    if (ret < 0) {
+        goto out;
+    }
+    frame->local = fop;
+    return fop;
+
+out:
+    ta_release_fop(fop);
+    return NULL;
+}
+
+int32_t
+ta_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+            gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+    int ret = 0;
+    ta_fop_t *fop = NULL;
+
+    fop = ta_prepare_fop(frame, this, NULL, fd, flags, dict, xdata);
+    if (!fop) {
+        ret = -ENOMEM;
+        goto unwind;
+    }
+
+    STACK_WIND(frame, ta_get_xattrop_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fxattrop, fd, flags, fop->brick_xattr,
+               xdata);
+    return 0;
+
+unwind:
+
+    TA_STACK_UNWIND(xattrop, frame, -1, -ret, NULL, NULL);
+    return 0;
+}
+
+int32_t
+ta_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+           gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+    int ret = 0;
+    ta_fop_t *fop = NULL;
+
+    fop = ta_prepare_fop(frame, this, loc, NULL, flags, dict, xdata);
+    if (!fop) {
+        ret = -ENOMEM;
+        goto unwind;
+    }
+
+    STACK_WIND(frame, ta_get_xattrop_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->xattrop, loc, flags, fop->brick_xattr,
+               xdata);
+    return 0;
+
+unwind:
+
+    TA_STACK_UNWIND(xattrop, frame, -1, -ret, NULL, NULL);
+    return 0;
+}
+
+int32_t
+ta_writev(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
+          int32_t count, off_t off, uint32_t flags, struct iobref *iobref,
+          dict_t *xdata)
+{
+    TA_FAILED_FOP(writev, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+             int32_t flags, dict_t *xdata)
+{
+    TA_FAILED_FOP(fsetxattr, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+            int32_t flags, dict_t *xdata)
+{
+    TA_FAILED_FOP(setxattr, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t keep_size,
+             off_t offset, size_t len, dict_t *xdata)
+{
+    TA_FAILED_FOP(fallocate, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+          dict_t *xdata)
+{
+    TA_FAILED_FOP(access, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+           size_t len, dict_t *xdata)
+{
+    TA_FAILED_FOP(discard, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_entrylk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+           const char *basename, entrylk_cmd cmd, entrylk_type type,
+           dict_t *xdata)
+{
+    TA_FAILED_FOP(entrylk, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+            const char *basename, entrylk_cmd cmd, entrylk_type type,
+            dict_t *xdata)
+{
+    TA_FAILED_FOP(fentrylk, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    TA_FAILED_FOP(flush, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+         dict_t *xdata)
+{
+    TA_FAILED_FOP(fsync, frame, EINVAL);
+    return 0;
+}
+int32_t
+ta_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+            dict_t *xdata)
+{
+    TA_FAILED_FOP(fsyncdir, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, const char *name,
+            dict_t *xdata)
+{
+    TA_FAILED_FOP(getxattr, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+             dict_t *xdata)
+{
+    TA_FAILED_FOP(fgetxattr, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+        dict_t *xdata)
+{
+    TA_FAILED_FOP(link, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+      struct gf_flock *flock, dict_t *xdata)
+{
+    TA_FAILED_FOP(lk, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+         mode_t umask, dict_t *xdata)
+{
+    TA_FAILED_FOP(mkdir, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+         dev_t rdev, mode_t umask, dict_t *xdata)
+{
+    TA_FAILED_FOP(mknod, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+        fd_t *fd, dict_t *xdata)
+{
+    TA_FAILED_FOP(open, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+           dict_t *xdata)
+{
+    TA_FAILED_FOP(opendir, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+           off_t offset, dict_t *xdata)
+{
+    TA_FAILED_FOP(readdir, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t offset, dict_t *xdata)
+{
+    TA_FAILED_FOP(readdirp, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+            dict_t *xdata)
+{
+    TA_FAILED_FOP(readlink, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+         off_t offset, uint32_t flags, dict_t *xdata)
+{
+    TA_FAILED_FOP(readv, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               const char *name, dict_t *xdata)
+{
+    TA_FAILED_FOP(removexattr, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+                dict_t *xdata)
+{
+    TA_FAILED_FOP(fremovexattr, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+          dict_t *xdata)
+{
+    TA_FAILED_FOP(rename, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflags,
+         dict_t *xdata)
+{
+    TA_FAILED_FOP(rmdir, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *stbuf,
+           int32_t valid, dict_t *xdata)
+{
+    TA_FAILED_FOP(setattr, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
+            int32_t valid, dict_t *xdata)
+{
+    TA_FAILED_FOP(fsetattr, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    TA_FAILED_FOP(stat, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    TA_FAILED_FOP(fstat, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    TA_FAILED_FOP(statfs, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_symlink(call_frame_t *frame, xlator_t *this, const char *linkname,
+           loc_t *loc, mode_t umask, dict_t *xdata)
+{
+    TA_FAILED_FOP(symlink, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+            dict_t *xdata)
+{
+    TA_FAILED_FOP(truncate, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             dict_t *xdata)
+{
+    TA_FAILED_FOP(ftruncate, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflags,
+          dict_t *xdata)
+{
+    TA_FAILED_FOP(unlink, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+            off_t len, dict_t *xdata)
+{
+    TA_FAILED_FOP(zerofill, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+ta_seek(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+        gf_seek_what_t what, dict_t *xdata)
+{
+    TA_FAILED_FOP(seek, frame, EINVAL);
+    return 0;
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    ret = xlator_mem_acct_init(this, gf_ta_mt_end + 1);
+    if (ret)
+        gf_log(this->name, GF_LOG_ERROR,
+               "Memory accounting "
+               "initialization failed.");
+    return ret;
+}
+
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    return 0;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    if (!this->children || this->children->next) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "'thin_arbiter' not configured with exactly one child");
+        return -1;
+    }
+
+    if (!this->parents) {
+        gf_log(this->name, GF_LOG_ERROR, "dangling volume. check volfile ");
+    }
+    return 0;
+}
+
+void
+fini(xlator_t *this)
+{
+    return;
+}
+
+struct xlator_fops fops = {
+    /*Passed fop*/
+    .xattrop = ta_xattrop,
+    .fxattrop = ta_fxattrop,
+    /*Failed fop*/
+    .writev = ta_writev,
+    .stat = ta_stat,
+    .fstat = ta_fstat,
+    .truncate = ta_truncate,
+    .ftruncate = ta_ftruncate,
+    .access = ta_access,
+    .readlink = ta_readlink,
+    .mknod = ta_mknod,
+    .mkdir = ta_mkdir,
+    .unlink = ta_unlink,
+    .rmdir = ta_rmdir,
+    .symlink = ta_symlink,
+    .rename = ta_rename,
+    .link = ta_link,
+    .open = ta_open,
+    .readv = ta_readv,
+    .flush = ta_flush,
+    .fsync = ta_fsync,
+    .opendir = ta_opendir,
+    .readdir = ta_readdir,
+    .readdirp = ta_readdirp,
+    .fsyncdir = ta_fsyncdir,
+    .statfs = ta_statfs,
+    .setxattr = ta_setxattr,
+    .getxattr = ta_getxattr,
+    .fsetxattr = ta_fsetxattr,
+    .fgetxattr = ta_fgetxattr,
+    .removexattr = ta_removexattr,
+    .fremovexattr = ta_fremovexattr,
+    .lk = ta_lk,
+    .entrylk = ta_entrylk,
+    .fentrylk = ta_fentrylk,
+    .setattr = ta_setattr,
+    .fsetattr = ta_fsetattr,
+    .fallocate = ta_fallocate,
+    .discard = ta_discard,
+    .zerofill = ta_zerofill,
+    .seek = ta_seek,
+};
+
+struct xlator_cbks cbks = {};
+
+struct volume_options options[] = {
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {GD_OP_VERSION_6_0},
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "thin-arbiter",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/features/thin-arbiter/src/thin-arbiter.h b/xlators/features/thin-arbiter/src/thin-arbiter.h
new file mode 100644
index 00000000000..e5f914b84bf
--- /dev/null
+++ b/xlators/features/thin-arbiter/src/thin-arbiter.h
@@ -0,0 +1,59 @@
+/*
+  Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _THIN_ARBITER_H
+#define _THIN_ARBITER_H
+
+#include <glusterfs/locking.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/list.h>
+
+#define THIN_ARBITER_SOURCE_XATTR "trusted.ta.source"
+#define THIN_ARBITER_SOURCE_SIZE 2
+
+#define TA_FAILED_FOP(fop, frame, op_errno)                                    \
+    do {                                                                       \
+        default_##fop##_failure_cbk(frame, op_errno);                          \
+    } while (0)
+
+#define TA_STACK_UNWIND(fop, frame, op_ret, op_errno, params...)               \
+    do {                                                                       \
+        ta_fop_t *__local = NULL;                                              \
+        int32_t __op_ret = 0;                                                  \
+        int32_t __op_errno = 0;                                                \
+                                                                               \
+        __local = frame->local;                                                \
+        __op_ret = op_ret;                                                     \
+        __op_errno = op_errno;                                                 \
+        if (__local) {                                                         \
+            ta_release_fop(__local);                                           \
+            frame->local = NULL;                                               \
+        }                                                                      \
+        STACK_UNWIND_STRICT(fop, frame, __op_ret, __op_errno, params);         \
+                                                                               \
+    } while (0)
+
+struct _ta_fop;
+typedef struct _ta_fop ta_fop_t;
+
+struct _ta_fop {
+    gf_xattrop_flags_t xattrop_flags;
+    loc_t loc;
+    fd_t *fd;
+    dict_t *dict;
+    dict_t *brick_xattr;
+    int32_t on_disk[2];
+    int32_t idx;
+};
+
+#endif /* _THIN_ARBITER_H */
diff --git a/xlators/features/trash/src/Makefile.am b/xlators/features/trash/src/Makefile.am
index 4671d06d309..8557e7171af 100644
--- a/xlators/features/trash/src/Makefile.am
+++ b/xlators/features/trash/src/Makefile.am
@@ -1,15 +1,19 @@
+if WITH_SERVER
 xlator_LTLIBRARIES = trash.la
-xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/testing/features
+endif
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
 
-trash_la_LDFLAGS = -module -avoidversion 
+trash_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
 
 trash_la_SOURCES = trash.c
 trash_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la 
 
 noinst_HEADERS = trash.h trash-mem-types.h
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS)\
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
 
 CLEANFILES = 
 
diff --git a/xlators/features/trash/src/trash-mem-types.h b/xlators/features/trash/src/trash-mem-types.h
index c0b1b916451..43353c8f095 100644
--- a/xlators/features/trash/src/trash-mem-types.h
+++ b/xlators/features/trash/src/trash-mem-types.h
@@ -1,33 +1,22 @@
 /*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
    This file is part of GlusterFS.
 
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
 */
-
 #ifndef __TRASH_MEM_TYPES_H__
 #define __TRASH_MEM_TYPES_H__
 
-#include "mem-types.h"
+#include <glusterfs/mem-types.h>
 
 enum gf_trash_mem_types_ {
-        gf_trash_mt_trash_local_t = gf_common_mt_end + 1,
-        gf_trash_mt_trash_private_t,
-        gf_trash_mt_char,
-        gf_trash_mt_trash_elim_pattern_t,
-        gf_trash_mt_end
+    gf_trash_mt_trash_private_t = gf_common_mt_end + 1,
+    gf_trash_mt_char,
+    gf_trash_mt_uuid,
+    gf_trash_mt_trash_elim_path,
+    gf_trash_mt_end
 };
 #endif
-
diff --git a/xlators/features/trash/src/trash.c b/xlators/features/trash/src/trash.c
index 087fe2b9903..7d09cba3e9c 100644
--- a/xlators/features/trash/src/trash.c
+++ b/xlators/features/trash/src/trash.c
@@ -1,1564 +1,2653 @@
 /*
-  Copyright (c) 2006-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+   Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
 #include "trash.h"
 #include "trash-mem-types.h"
+#include <glusterfs/syscall.h>
 
-int32_t
-trash_ftruncate_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int32_t op_ret, int32_t op_errno,
-                           struct iovec *vector, int32_t count,
-                           struct iatt *stbuf, struct iobref *iobuf);
+#define root_gfid                                                              \
+    (uuid_t) { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1 }
+#define trash_gfid                                                             \
+    (uuid_t) { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 5 }
+#define internal_op_gfid                                                       \
+    (uuid_t) { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 6 }
 
 int32_t
-trash_truncate_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int32_t op_ret, int32_t op_errno,
-                           struct iatt *prebuf, struct iatt *postbuf);
+trash_truncate_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                          struct iatt *postbuf, dict_t *xdata);
 
 int32_t
-trash_truncate_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t op_ret, int32_t op_errno, inode_t *inode,
-                          struct iatt *stbuf, struct iatt *preparent,
-                          struct iatt *postparent);
+trash_truncate_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, inode_t *inode,
+                         struct iatt *stbuf, struct iatt *preparent,
+                         struct iatt *postparent, dict_t *xdata);
 
 int32_t
-trash_unlink_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, struct iatt *buf,
-                         struct iatt *preoldparent, struct iatt *postoldparent,
-                         struct iatt *prenewparent, struct iatt *postnewparent);
+trash_unlink_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                        struct iatt *preoldparent, struct iatt *postoldparent,
+                        struct iatt *prenewparent, struct iatt *postnewparent,
+                        dict_t *xdata);
+/* Common routines used in this translator */
 
-void
-trash_local_wipe (trash_local_t *local)
+/**
+ * When a directory/file is created under trash directory, it should have
+ * the same permission as before. This function will fetch permission from
+ * the existing directory and returns the same
+ */
+mode_t
+get_permission(char *path)
 {
-        if (!local)
-                goto out;
-
-        loc_wipe (&local->loc);
-        loc_wipe (&local->newloc);
-
-        if (local->fd)
-                fd_unref (local->fd);
-
-        if (local->newfd)
-                fd_unref (local->newfd);
+    mode_t mode = 0755;
+    struct stat sbuf = {
+        0,
+    };
+    struct iatt ibuf = {
+        0,
+    };
+    int ret = 0;
+
+    ret = sys_stat(path, &sbuf);
+    if (!ret) {
+        iatt_from_stat(&ibuf, &sbuf);
+        mode = st_mode_from_ia(ibuf.ia_prot, ibuf.ia_type);
+    } else
+        gf_log("trash", GF_LOG_DEBUG,
+               "stat on %s failed"
+               " using default",
+               path);
+    return mode;
+}
 
-        GF_FREE (local);
+/**
+ * For normalization, trash directory name is stored inside priv structure as
+ * '/trash_directory/'. As a result the trailing and leading slashes are being
+ * striped out for additional usage.
+ */
+int
+extract_trash_directory(char *priv_value, const char **trash_directory)
+{
+    char *tmp = NULL;
+    int ret = 0;
+
+    GF_VALIDATE_OR_GOTO("trash", priv_value, out);
+
+    tmp = gf_strdup(priv_value + 1);
+    if (!tmp) {
+        ret = ENOMEM;
+        goto out;
+    }
+    if (tmp[strlen(tmp) - 1] == '/')
+        tmp[strlen(tmp) - 1] = '\0';
+    *trash_directory = gf_strdup(tmp);
+    if (!(*trash_directory)) {
+        ret = ENOMEM;
+        goto out;
+    }
 out:
-        return;
+    if (tmp)
+        GF_FREE(tmp);
+    return ret;
 }
 
-int32_t
-trash_common_unwind_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno,
-                         struct iatt *preparent, struct iatt *postparent)
+/**
+ * The trash directory path should be append at beginning of file path for
+ * delete or truncate operations. Normal trashing moves the contents to
+ * trash directory and trashing done by internal operations are moved to
+ * internal_op directory inside trash.
+ */
+void
+copy_trash_path(const char *priv_value, gf_boolean_t internal, char *path,
+                size_t path_size)
 {
-        TRASH_STACK_UNWIND (unlink, frame, op_ret, op_errno, preparent, postparent);
-        return 0;
+    char trash_path[PATH_MAX] = {
+        0,
+    };
+
+    strncpy(trash_path, priv_value, sizeof(trash_path));
+    trash_path[sizeof(trash_path) - 1] = 0;
+    if (internal)
+        strncat(trash_path, "internal_op/",
+                sizeof(trash_path) - strlen(trash_path) - 1);
+
+    strncpy(path, trash_path, path_size);
+    path[path_size - 1] = 0;
 }
 
-int32_t
-trash_unlink_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, inode_t *inode,
-                        struct iatt *stbuf, struct iatt *preparent,
-                        struct iatt *postparent)
+/**
+ * This function performs the reverse operation of copy_trash_path(). It gives
+ * out a pointer, whose starting value will be the path inside trash directory,
+ * similar to original path.
+ */
+void
+remove_trash_path(const char *path, gf_boolean_t internal, char **rem_path)
 {
-        trash_local_t *local       = NULL;
-        char          *tmp_str     = NULL;
-        char          *tmp_path    = NULL;
-        char          *tmp_dirname = NULL;
-        char          *dir_name    = NULL;
-        int32_t        count       = 0;
-        int32_t        loop_count  = 0;
-        int            i           = 0;
-        loc_t          tmp_loc     = {0,};
-
-        local   = frame->local;
-        tmp_str = gf_strdup (local->newpath);
-        if (!tmp_str) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto out;
-        }
-        loop_count = local->loop_count;
-
-        if ((op_ret == -1) &&  (op_errno == ENOENT)) {
-                tmp_dirname = strchr (tmp_str, '/');
-                while (tmp_dirname) {
-                        count = tmp_dirname - tmp_str;
-                        if (count == 0)
-                                count = 1;
-                        i++;
-                        if (i > loop_count)
-                                break;
-                        tmp_dirname = strchr (tmp_str + count + 1, '/');
-                }
-                tmp_path = memdup (local->newpath, count);
-                if (!tmp_path) {
-                        gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                        goto out;
-                }
-
-                tmp_loc.path = tmp_path;
-
-                /* TODO:create the directory with proper permissions */
-                STACK_WIND_COOKIE (frame, trash_unlink_mkdir_cbk, tmp_path,
-                                   this->children->xlator,
-                                   this->children->xlator->fops->mkdir,
-                                   &tmp_loc, 0755, NULL);
-
-                goto out;
-        }
-
-        if (op_ret == 0) {
-                dir_name = dirname (tmp_str);
-                if (strcmp((char*)cookie, dir_name) == 0) {
-                        tmp_loc.path = local->newpath;
-                        STACK_WIND (frame, trash_unlink_rename_cbk,
-                                    this->children->xlator,
-                                    this->children->xlator->fops->rename,
-                                    &local->loc, &tmp_loc);
-                        goto out;
-                }
-        }
-
-        LOCK (&frame->lock);
-        {
-                loop_count = ++local->loop_count;
-        }
-        UNLOCK (&frame->lock);
-        tmp_dirname = strchr (tmp_str, '/');
-        while (tmp_dirname) {
-                count = tmp_dirname - tmp_str;
-                if (count == 0)
-                        count = 1;
-                i++;
-                if ((i > loop_count) || (count > PATH_MAX))
-                        break;
-                tmp_dirname = strchr (tmp_str + count + 1, '/');
-        }
-        tmp_path = memdup (local->newpath, count);
-        if (!tmp_path) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                goto out;
-        }
-        tmp_loc.path = tmp_path;
-
-        STACK_WIND_COOKIE (frame, trash_unlink_mkdir_cbk, tmp_path,
-                           this->children->xlator,
-                           this->children->xlator->fops->mkdir,
-                           &tmp_loc, 0755, NULL);
-
-out:
-        GF_FREE (cookie);
-        if (tmp_str)
-                GF_FREE (tmp_str);
+    if (rem_path == NULL) {
+        return;
+    }
 
-        return 0;
+    *rem_path = strchr(path + 1, '/');
+    if (internal)
+        *rem_path = strchr(*rem_path + 1, '/');
 }
 
-int32_t
-trash_rename_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, inode_t *inode,
-                        struct iatt *stbuf, struct iatt *preparent,
-                        struct iatt *postparent);
-
-int32_t
-trash_unlink_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, struct iatt *buf,
-                         struct iatt *preoldparent, struct iatt *postoldparent,
-                         struct iatt *prenewparent, struct iatt *postnewparent)
+/**
+ * Checks whether the given path reside under the specified eliminate path
+ */
+int
+check_whether_eliminate_path(trash_elim_path *trav, const char *path)
 {
-        trash_local_t   *local      = NULL;
-        char            *tmp_str    = NULL;
-        char            *dir_name   = NULL;
-        char            *tmp_cookie = NULL;
-        loc_t            tmp_loc    = {0,};
-
-        local = frame->local;
-
-        if ((op_ret == -1) && (op_errno == ENOENT)) {
-                tmp_str = gf_strdup (local->newpath);
-                if (!tmp_str) {
-                        gf_log (this->name, GF_LOG_DEBUG, "out of memory");
-                }
-                dir_name = dirname (tmp_str);
-
-                tmp_loc.path = dir_name;
-
-                tmp_cookie = gf_strdup (dir_name);
-                if (!tmp_cookie) {
-                        gf_log (this->name, GF_LOG_DEBUG, "out of memory");
-                }
-                /* TODO: create the directory with proper permissions */
-                STACK_WIND_COOKIE (frame, trash_unlink_mkdir_cbk, tmp_cookie,
-                                   FIRST_CHILD(this),
-                                   FIRST_CHILD(this)->fops->mkdir,
-                                   &tmp_loc, 0755, NULL);
-
-                GF_FREE (tmp_str);
-
-                return 0;
-        }
-
-        if ((op_ret == -1) && (op_errno == ENOTDIR)) {
-
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "target(%s) exists, cannot keep the copy, deleting",
-                        local->newpath);
+    int match = 0;
 
-                STACK_WIND (frame, trash_common_unwind_cbk,
-                            this->children->xlator,
-                            this->children->xlator->fops->unlink, &local->loc);
-
-                return 0;
+    while (trav) {
+        if (strncmp(path, trav->path, strlen(trav->path)) == 0) {
+            match++;
+            break;
         }
-
-        if ((op_ret == -1) && (op_errno == EISDIR)) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "target(%s) exists as directory, cannot keep copy, "
-                        "deleting", local->newpath);
-
-                STACK_WIND (frame, trash_common_unwind_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->unlink, &local->loc);
-                return 0;
-        }
-
-        /* All other cases, unlink should return success */
-        TRASH_STACK_UNWIND (unlink, frame, 0, op_errno, &local->preparent,
-                            &local->postparent);
-
-        return 0;
+        trav = trav->next;
+    }
+    return match;
 }
 
-
-
-int32_t
-trash_common_unwind_buf_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                             int32_t op_ret, int32_t op_errno,
-                             struct iatt *prebuf, struct iatt *postbuf)
+/**
+ * Stores the eliminate path into internal eliminate path structure
+ */
+int
+store_eliminate_path(char *str, trash_elim_path **eliminate)
 {
-        TRASH_STACK_UNWIND (truncate, frame, op_ret, op_errno, prebuf, postbuf);
-        return 0;
+    trash_elim_path *trav = NULL;
+    char *component = NULL;
+    char elm_path[PATH_MAX] = {
+        0,
+    };
+    int ret = 0;
+    char *strtokptr = NULL;
+
+    if ((str == NULL) || (eliminate == NULL)) {
+        ret = EINVAL;
+        goto out;
+    }
+
+    component = strtok_r(str, ",", &strtokptr);
+    while (component) {
+        trav = GF_CALLOC(1, sizeof(*trav), gf_trash_mt_trash_elim_path);
+        if (!trav) {
+            ret = ENOMEM;
+            goto out;
+        }
+        if (component[0] == '/')
+            sprintf(elm_path, "%s", component);
+        else
+            sprintf(elm_path, "/%s", component);
+
+        if (component[strlen(component) - 1] != '/')
+            strncat(elm_path, "/", sizeof(elm_path) - strlen(elm_path) - 1);
+
+        trav->path = gf_strdup(elm_path);
+        if (!trav->path) {
+            ret = ENOMEM;
+            gf_log("trash", GF_LOG_DEBUG, "out of memory");
+            GF_FREE(trav);
+            goto out;
+        }
+        trav->next = *eliminate;
+        *eliminate = trav;
+        component = strtok_r(NULL, ",", &strtokptr);
+    }
+out:
+    return ret;
 }
 
-int
-trash_common_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, struct iatt *stbuf,
-                         struct iatt *preoldparent, struct iatt *postoldparent,
-                         struct iatt *prenewparent, struct iatt *postnewparent)
+/**
+ * Appends time stamp to given string
+ */
+void
+append_time_stamp(char *name, size_t name_size)
 {
-        TRASH_STACK_UNWIND (rename, frame, op_ret, op_errno, stbuf, preoldparent,
-                            postoldparent, prenewparent, postnewparent);
-        return 0;
+    int i;
+    char timestr[GF_TIMESTR_SIZE] = {
+        0,
+    };
+
+    gf_time_fmt(timestr, sizeof(timestr), gf_time(), gf_timefmt_F_HMS);
+
+    /* removing white spaces in timestamp */
+    for (i = 0; i < strlen(timestr); i++) {
+        if (timestr[i] == ' ')
+            timestr[i] = '_';
+    }
+    strncat(name, "_", name_size - strlen(name) - 1);
+    strncat(name, timestr, name_size - strlen(name) - 1);
 }
 
+/* *
+ * Check whether delete/rename operation is permitted on
+ * trash directory
+ */
 
-int32_t
-trash_unlink_stat_cbk (call_frame_t *frame,  void *cookie, xlator_t *this,
-                       int32_t op_ret, int32_t op_errno, struct iatt *buf)
+gf_boolean_t
+check_whether_op_permitted(trash_private_t *priv, loc_t *loc)
 {
-        trash_private_t *priv    = NULL;
-        trash_local_t   *local   = NULL;
-        loc_t            new_loc = {0,};
+    if ((priv->state && (gf_uuid_compare(loc->inode->gfid, trash_gfid) == 0)))
+        return _gf_false;
+    if (priv->internal &&
+        (gf_uuid_compare(loc->inode->gfid, internal_op_gfid) == 0))
+        return _gf_false;
 
-        priv  = this->private;
-        local = frame->local;
-
-        if (-1 == op_ret) {
-                gf_log (this->name, GF_LOG_DEBUG, "%s: %s",
-                        local->loc.path, strerror (op_errno));
-                goto fail;
-        }
-
-        if ((buf->ia_size == 0) ||
-            (buf->ia_size > priv->max_trash_file_size)) {
-                /* if the file is too big or zero, just unlink it */
-
-                if (buf->ia_size > priv->max_trash_file_size) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s: file size too big (%"PRId64") to "
-                                "move into trash directory",
-                                local->loc.path, buf->ia_size);
-                }
-
-                STACK_WIND (frame, trash_common_unwind_cbk,
-                            this->children->xlator,
-                            this->children->xlator->fops->unlink, &local->loc);
-                return 0;
-        }
-
-        new_loc.path = local->newpath;
-
-        STACK_WIND (frame, trash_unlink_rename_cbk,
-                    this->children->xlator,
-                    this->children->xlator->fops->rename,
-                    &local->loc, &new_loc);
+    return _gf_true;
+}
 
-        return 0;
+/**
+ * Wipe the memory used by trash location variable
+ */
+void
+trash_local_wipe(trash_local_t *local)
+{
+    if (!local)
+        goto out;
 
-fail:
-        TRASH_STACK_UNWIND (unlink, frame, op_ret, op_errno, buf,
-                            NULL);
+    loc_wipe(&local->loc);
+    loc_wipe(&local->newloc);
 
-        return 0;
+    if (local->fd)
+        fd_unref(local->fd);
+    if (local->newfd)
+        fd_unref(local->newfd);
 
+    mem_put(local);
+out:
+    return;
 }
 
-int32_t
-trash_rename_rename_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, struct iatt *buf,
-                         struct iatt *preoldparent, struct iatt *postoldparent,
-                         struct iatt *prenewparent, struct iatt *postnewparent)
+/**
+ * Wipe the memory used by eliminate path through a
+ * recursive call
+ */
+void
+wipe_eliminate_path(trash_elim_path **trav)
 {
-        trash_local_t *local    = NULL;
-        char          *tmp_str  = NULL;
-        char          *dir_name = NULL;
-        char          *tmp_path = NULL;
-        loc_t          tmp_loc  = {0,};
-
-        local = frame->local;
-        if ((op_ret == -1) && (op_errno == ENOENT)) {
-                tmp_str  = gf_strdup (local->newpath);
-                if (!tmp_str) {
-                        gf_log (this->name, GF_LOG_DEBUG, "out of memory");
-                }
-                dir_name = dirname (tmp_str);
-
-                /* check for the errno, if its ENOENT create directory and call
-                 * rename later
-                 */
-                tmp_path = gf_strdup (dir_name);
-                if (!tmp_path) {
-                        gf_log (this->name, GF_LOG_DEBUG, "out of memory");
-                }
-                tmp_loc.path = tmp_path;
-
-                /* TODO: create the directory with proper permissions */
-                STACK_WIND_COOKIE (frame, trash_rename_mkdir_cbk, tmp_path,
-                                   this->children->xlator,
-                                   this->children->xlator->fops->mkdir,
-                                   &tmp_loc, 0755, NULL);
-
-                GF_FREE (tmp_str);
-                return 0;
-        }
-
-        if ((op_ret == -1) && (op_errno == ENOTDIR)) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "target(%s) exists, cannot keep the dest entry(%s): "
-                        "renaming", local->newpath, local->origpath);
-        } else if ((op_ret == -1) && (op_errno == EISDIR)) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "target(%s) exists as a directory, cannot keep the "
-                        "copy (%s), renaming", local->newpath, local->origpath);
-        }
+    if (trav == NULL) {
+        return;
+    }
 
-        STACK_WIND (frame, trash_common_rename_cbk,
-                    this->children->xlator,
-                    this->children->xlator->fops->rename, &local->loc,
-                    &local->newloc);
+    if (*trav == NULL) {
+        return;
+    }
 
-        return 0;
+    wipe_eliminate_path(&(*trav)->next);
+    GF_FREE((*trav)->path);
+    GF_FREE(*trav);
+    *trav = NULL;
 }
 
-
+/**
+ * This is the call back of rename fop initated using STACK_WIND in
+ * reconfigure/notify function which is used to rename trash directory
+ * in the brick when it is required either in volume start or set.
+ * This frame  must destroyed from this function itself since it was
+ * created by trash xlator
+ */
 int32_t
-trash_rename_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                        int32_t op_ret, int32_t op_errno, inode_t *inode,
-                        struct iatt *stbuf, struct iatt *preparent,
-                        struct iatt *postparent)
+trash_dir_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                     struct iatt *preoldparent, struct iatt *postoldparent,
+                     struct iatt *prenewparent, struct iatt *postnewparent,
+                     dict_t *xdata)
 {
-        trash_local_t *local = NULL;
-        char          *tmp_str = NULL;
-        char          *tmp_path = NULL;
-        char          *tmp_dirname = NULL;
-        char          *dir_name = NULL;
-        int32_t        count = 0;
-        loc_t          tmp_loc = {0,};
-
-        local   = frame->local;
-        tmp_str = gf_strdup (local->newpath);
-        if (!tmp_str) {
-                gf_log (this->name, GF_LOG_DEBUG, "out of memory");
-                goto out;
-        }
+    trash_private_t *priv = NULL;
+    trash_local_t *local = NULL;
 
-        if ((op_ret == -1) && (op_errno == ENOENT)) {
-                tmp_dirname = strchr (tmp_str, '/');
-                while (tmp_dirname) {
-                        count = tmp_dirname - tmp_str;
-                        if (count == 0)
-                                count = 1;
+    priv = this->private;
 
-                        tmp_dirname = strchr (tmp_str + count + 1, '/');
+    local = frame->local;
 
-                        tmp_path = memdup (local->newpath, count);
-                        if (!tmp_path) {
-                                gf_log (this->name, GF_LOG_DEBUG, "out of memory");
-                        }
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "rename trash directory "
+               "failed: %s",
+               strerror(op_errno));
+        goto out;
+    }
 
-                        tmp_loc.path = tmp_path;
+    GF_FREE(priv->oldtrash_dir);
 
-                        /* TODO: create the directory with proper permissions */
-                        STACK_WIND_COOKIE (frame, trash_rename_mkdir_cbk,
-                                           tmp_path,  this->children->xlator,
-                                           this->children->xlator->fops->mkdir,
-                                           &tmp_loc, 0755, NULL);
-                }
+    priv->oldtrash_dir = gf_strdup(priv->newtrash_dir);
+    if (!priv->oldtrash_dir) {
+        op_ret = ENOMEM;
+        gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+    }
 
-                goto out;
-        }
-
-        dir_name = dirname (tmp_str);
-        if (strcmp ((char*)cookie, dir_name) == 0) {
-                tmp_loc.path = local->newpath;
+out:
+    frame->local = NULL;
+    STACK_DESTROY(frame->root);
+    trash_local_wipe(local);
+    return op_ret;
+}
 
-                STACK_WIND (frame, trash_rename_rename_cbk,
-                            this->children->xlator,
-                            this->children->xlator->fops->rename,
-                            &local->newloc, &tmp_loc);
-        }
+int
+rename_trash_directory(xlator_t *this)
+{
+    trash_private_t *priv = NULL;
+    int ret = 0;
+    loc_t loc = {
+        0,
+    };
+    loc_t old_loc = {
+        0,
+    };
+    call_frame_t *frame = NULL;
+    trash_local_t *local = NULL;
+
+    priv = this->private;
+
+    frame = create_frame(this, this->ctx->pool);
+    if (frame == NULL) {
+        gf_log(this->name, GF_LOG_ERROR, "failed to create frame");
+        ret = ENOMEM;
+        goto out;
+    }
+
+    local = mem_get0(this->local_pool);
+    if (!local) {
+        gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+        ret = ENOMEM;
+        goto out;
+    }
+    frame->local = local;
+
+    /* assign new location values to new_loc members */
+    gf_uuid_copy(loc.gfid, trash_gfid);
+    gf_uuid_copy(loc.pargfid, root_gfid);
+    ret = extract_trash_directory(priv->newtrash_dir, &loc.name);
+    if (ret) {
+        gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+        goto out;
+    }
+    loc.path = gf_strdup(priv->newtrash_dir);
+    if (!loc.path) {
+        ret = ENOMEM;
+        gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+        goto out;
+    }
+
+    /* assign old location values to old_loc members */
+    gf_uuid_copy(old_loc.gfid, trash_gfid);
+    gf_uuid_copy(old_loc.pargfid, root_gfid);
+    ret = extract_trash_directory(priv->oldtrash_dir, &old_loc.name);
+    if (ret) {
+        gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+        goto out;
+    }
+    old_loc.path = gf_strdup(priv->oldtrash_dir);
+    if (!old_loc.path) {
+        ret = ENOMEM;
+        gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+        goto out;
+    }
+
+    old_loc.inode = inode_ref(priv->trash_inode);
+    gf_uuid_copy(old_loc.inode->gfid, old_loc.gfid);
+
+    loc_copy(&local->loc, &old_loc);
+    loc_copy(&local->newloc, &loc);
+
+    STACK_WIND(frame, trash_dir_rename_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rename, &old_loc, &loc, NULL);
+    return 0;
 
 out:
-        GF_FREE (cookie); /* strdup (dir_name) was sent here :) */
-        if (tmp_str)
-                GF_FREE (tmp_str);
+    if (frame) {
+        frame->local = NULL;
+        STACK_DESTROY(frame->root);
+    }
 
-        return 0;
+    trash_local_wipe(local);
+
+    return ret;
 }
 
 int32_t
-trash_rename_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, inode_t *inode,
-                         struct iatt *buf, dict_t *xattr,
-                         struct iatt *postparent)
+trash_internal_op_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int32_t op_ret, int32_t op_errno, inode_t *inode,
+                            struct iatt *buf, struct iatt *preparent,
+                            struct iatt *postparent, dict_t *xdata)
 {
-        trash_private_t *priv = NULL;
-        trash_local_t   *local = NULL;
-        loc_t            tmp_loc = {0,};
-
-        local = frame->local;
-        priv  = this->private;
-
-        if (op_ret == -1) {
-                STACK_WIND (frame, trash_common_rename_cbk,
-                            this->children->xlator,
-                            this->children->xlator->fops->rename,
-                            &local->loc, &local->newloc);
-                return 0;
-        }
-        if ((buf->ia_size == 0) ||
-            (buf->ia_size > priv->max_trash_file_size)) {
-                /* if the file is too big or zero, just unlink it */
-
-                if (buf->ia_size > priv->max_trash_file_size) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s: file size too big (%"PRId64") to "
-                                "move into trash directory",
-                                local->newloc.path, buf->ia_size);
-                }
-
-                STACK_WIND (frame, trash_common_rename_cbk,
-                            this->children->xlator,
-                            this->children->xlator->fops->rename,
-                            &local->loc, &local->newloc);
-                return 0;
-        }
-
-        tmp_loc.path = local->newpath;
-
-        STACK_WIND (frame, trash_rename_rename_cbk,
-                    this->children->xlator,
-                    this->children->xlator->fops->rename,
-                    &local->newloc, &tmp_loc);
-
-        return 0;
+    trash_local_t *local = NULL;
+    local = frame->local;
+
+    if (op_ret != 0 && !(op_errno == EEXIST))
+        gf_log(this->name, GF_LOG_ERROR,
+               "mkdir failed for "
+               "internal op directory : %s",
+               strerror(op_errno));
+
+    frame->local = NULL;
+    STACK_DESTROY(frame->root);
+    trash_local_wipe(local);
+    return op_ret;
 }
 
+/**
+ * This is the call back of mkdir fop initated using STACK_WIND in
+ * notify/reconfigure function which is used to create trash directory
+ * in the brick when "trash" is on. The frame of the mkdir must
+ * destroyed from this function itself since it was created by trash xlator
+ */
 
 int32_t
-trash_rename (call_frame_t *frame, xlator_t *this, loc_t *oldloc,
-              loc_t *newloc)
+trash_dir_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, inode_t *inode,
+                    struct iatt *buf, struct iatt *preparent,
+                    struct iatt *postparent, dict_t *xdata)
 {
-        trash_elim_pattern_t  *trav  = NULL;
-        trash_private_t *priv = NULL;
-        trash_local_t   *local = NULL;
-        struct tm       *tm = NULL;
-        char             timestr[256] = {0,};
-        time_t           utime = 0;
-        int32_t          match = 0;
-
-        priv = this->private;
-        if (priv->eliminate) {
-                trav = priv->eliminate;
-                while (trav) {
-                        if (fnmatch(trav->pattern, newloc->name, 0) == 0) {
-                                match++;
-                                break;
-                        }
-                        trav = trav->next;
-                }
-        }
+    trash_private_t *priv = NULL;
+    trash_local_t *local = NULL;
 
-        if ((strncmp (oldloc->path, priv->trash_dir,
-                      strlen (priv->trash_dir)) == 0) || match) {
-                /* Trying to rename from the trash dir,
-                   do the actual rename */
-                STACK_WIND (frame, trash_common_rename_cbk,
-                            this->children->xlator,
-                            this->children->xlator->fops->rename,
-                            oldloc, newloc);
+    priv = this->private;
 
-                return 0;
-        }
+    local = frame->local;
 
-        local = GF_CALLOC (1, sizeof (trash_local_t), 
-                           gf_trash_mt_trash_local_t);
-        if (!local) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                TRASH_STACK_UNWIND (rename, frame, -1, ENOMEM,
-                                    NULL, NULL, NULL, NULL, NULL);
-                return 0;
+    if (op_ret == 0) {
+        priv->oldtrash_dir = gf_strdup(priv->newtrash_dir);
+        if (!priv->oldtrash_dir) {
+            gf_log(this->name, GF_LOG_ERROR, "out of memory");
+            op_ret = ENOMEM;
         }
+    } else if (op_ret != 0 && errno != EEXIST)
+        gf_log(this->name, GF_LOG_ERROR,
+               "mkdir failed for trash"
+               " directory : %s",
+               strerror(op_errno));
 
-        frame->local = local;
-        loc_copy (&local->loc, oldloc);
-
-        loc_copy (&local->newloc, newloc);
-
-        strcpy (local->origpath, newloc->path);
-        strcpy (local->newpath, priv->trash_dir);
-        strcat (local->newpath, newloc->path);
-
-        {
-                /* append timestamp to file name */
-                /* TODO: can we make it optional? */
-                utime = time (NULL);
-                tm    = localtime (&utime);
-                strftime (timestr, 256, ".%Y-%m-%d-%H%M%S", tm);
-                strcat (local->newpath, timestr);
-        }
+    frame->local = NULL;
+    STACK_DESTROY(frame->root);
+    trash_local_wipe(local);
+    return op_ret;
+}
 
-        /* Send a lookup call on newloc, to ensure we are not
-           overwriting */
-        STACK_WIND (frame, trash_rename_lookup_cbk,
-                    this->children->xlator,
-                    this->children->xlator->fops->lookup, newloc, 0);
+/**
+ * This getxattr calls returns existing trash directory path in
+ * the dictionary
+ */
+int32_t
+trash_dir_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *dict,
+                       dict_t *xdata)
+{
+    data_t *data = NULL;
+    trash_private_t *priv = NULL;
+    int ret = 0;
+    trash_local_t *local = NULL;
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO("trash", priv, out);
+
+    local = frame->local;
+
+    data = dict_get(dict, GET_ANCESTRY_PATH_KEY);
+    if (!data) {
+        goto out;
+    }
+    priv->oldtrash_dir = GF_MALLOC(PATH_MAX, gf_common_mt_char);
+    if (!priv->oldtrash_dir) {
+        gf_log(this->name, GF_LOG_ERROR, "out of memory");
+        ret = ENOMEM;
+        goto out;
+    }
+    /* appending '/' if it is not present */
+    sprintf(priv->oldtrash_dir, "%s%c", data->data,
+            data->data[strlen(data->data) - 1] != '/' ? '/' : '\0');
+    gf_log(this->name, GF_LOG_DEBUG,
+           "old trash directory path "
+           "is %s",
+           priv->oldtrash_dir);
+    if (strcmp(priv->newtrash_dir, priv->oldtrash_dir) != 0) {
+        /* When user set a new name for trash directory, trash
+         * xlator will perform a rename operation on old trash
+         * directory to the new one using a STACK_WIND from here.
+         * This option can be configured only when volume is in
+         * started state
+         */
+        ret = rename_trash_directory(this);
+    }
 
+out:
+    frame->local = NULL;
+    STACK_DESTROY(frame->root);
+    trash_local_wipe(local);
+    return ret;
+}
+/**
+ * This is a nameless look up for internal op directory
+ * The lookup is based on gfid, because internal op directory
+ * has fixed gfid.
+ */
+int32_t
+trash_internalop_dir_lookup_cbk(call_frame_t *frame, void *cookie,
+                                xlator_t *this, int32_t op_ret,
+                                int32_t op_errno, inode_t *inode,
+                                struct iatt *buf, dict_t *xdata,
+                                struct iatt *postparent)
+{
+    trash_private_t *priv = NULL;
+    int ret = 0;
+    uuid_t *gfid_ptr = NULL;
+    loc_t loc = {
+        0,
+    };
+    char internal_op_path[PATH_MAX] = {
+        0,
+    };
+    dict_t *dict = NULL;
+    trash_local_t *local = NULL;
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO("trash", priv, out);
+
+    local = frame->local;
+    if (op_ret != 0 && op_errno == ENOENT) {
+        loc_wipe(&local->loc);
+        gfid_ptr = GF_MALLOC(sizeof(uuid_t), gf_common_mt_uuid_t);
+        if (!gfid_ptr) {
+            ret = ENOMEM;
+            goto out;
+        }
+
+        gf_uuid_copy(*gfid_ptr, internal_op_gfid);
+
+        dict = dict_new();
+        if (!dict) {
+            ret = ENOMEM;
+            goto out;
+        }
+        ret = dict_set_gfuuid(dict, "gfid-req", *gfid_ptr, false);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR, "setting key gfid-req failed");
+            goto out;
+        }
+        gf_uuid_copy(loc.gfid, internal_op_gfid);
+        gf_uuid_copy(loc.pargfid, trash_gfid);
+
+        loc.inode = inode_new(priv->trash_itable);
+
+        /* The mkdir call for creating internal op directory */
+        loc.name = gf_strdup("internal_op");
+        if (!loc.name) {
+            gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+            ret = ENOMEM;
+            goto out;
+        }
+        sprintf(internal_op_path, "%s%s/", priv->newtrash_dir, loc.name);
+
+        loc.path = gf_strdup(internal_op_path);
+        if (!loc.path) {
+            gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+            ret = ENOMEM;
+            goto out;
+        }
+
+        loc_copy(&local->loc, &loc);
+        STACK_WIND(frame, trash_internal_op_mkdir_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->mkdir, &loc, 0755, 0022, dict);
         return 0;
+    }
+
+out:
+    if (ret && gfid_ptr)
+        GF_FREE(gfid_ptr);
+    if (dict)
+        dict_unref(dict);
+    frame->local = NULL;
+    STACK_DESTROY(frame->root);
+    trash_local_wipe(local);
+    return op_ret;
 }
 
+/**
+ * This is a nameless look up for old trash directory
+ * The lookup is based on gfid, because trash directory
+ * has fixed gfid.
+ */
 int32_t
-trash_unlink (call_frame_t *frame, xlator_t *this, loc_t *loc)
+trash_dir_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno, inode_t *inode,
+                     struct iatt *buf, dict_t *xdata, struct iatt *postparent)
 {
-        trash_elim_pattern_t  *trav  = NULL;
-        trash_private_t *priv = NULL;
-        trash_local_t   *local = NULL;
-        struct tm       *tm = NULL;
-        char             timestr[256] = {0,};
-        time_t           utime = 0;
-        int32_t          match = 0;
+    trash_private_t *priv = NULL;
+    loc_t loc = {
+        0,
+    };
+    int ret = 0;
+    uuid_t *gfid_ptr = NULL;
+    dict_t *dict = NULL;
+    trash_local_t *local = NULL;
 
-        priv = this->private;
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO("trash", priv, out);
 
-        if (priv->eliminate) {
-                trav = priv->eliminate;
-                while (trav) {
-                        if (fnmatch(trav->pattern, loc->name, 0) == 0) {
-                                match++;
-                                break;
-                        }
-                        trav = trav->next;
-                }
-        }
+    local = frame->local;
 
-        if ((strncmp (loc->path, priv->trash_dir,
-                      strlen (priv->trash_dir)) == 0) || (match)) {
-                if (match) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s: file matches eliminate pattern, "
-                                "not moved to trash", loc->name);
-                } else {
-                        /* unlink from the trash-dir, not keeping any copy */
-                        ;
-                }
-
-                STACK_WIND (frame, trash_common_unwind_cbk,
-                            this->children->xlator,
-                            this->children->xlator->fops->unlink, loc);
-                return 0;
-        }
+    loc_wipe(&local->loc);
+    if (op_ret == 0) {
+        gf_log(this->name, GF_LOG_DEBUG, "inode found with gfid %s",
+               uuid_utoa(buf->ia_gfid));
 
-        local = GF_CALLOC (1, sizeof (trash_local_t),
-                           gf_trash_mt_trash_local_t);
-        if (!local) {
-                gf_log (this->name, GF_LOG_DEBUG, "out of memory");
-                TRASH_STACK_UNWIND (unlink, frame, -1, ENOMEM, NULL, NULL);
-                return 0;
-        }
-        frame->local = local;
-        loc_copy (&local->loc, loc);
-
-        strcpy (local->origpath, loc->path);
-        strcpy (local->newpath, priv->trash_dir);
-        strcat (local->newpath, loc->path);
-
-        {
-                /* append timestamp to file name */
-                /* TODO: can we make it optional? */
-                utime = time (NULL);
-                tm    = localtime (&utime);
-                strftime (timestr, 256, ".%Y-%m-%d-%H%M%S", tm);
-                strcat (local->newpath, timestr);
-        }
+        gf_uuid_copy(loc.gfid, trash_gfid);
 
-        LOCK_INIT (&frame->lock);
+        /* Find trash inode using available information */
+        priv->trash_inode = inode_link(inode, NULL, NULL, buf);
 
-        STACK_WIND (frame, trash_unlink_stat_cbk,
-                    this->children->xlator,
-                    this->children->xlator->fops->stat, loc);
+        loc.inode = inode_ref(priv->trash_inode);
+        loc_copy(&local->loc, &loc);
 
+        /*Used to find path of old trash directory*/
+        STACK_WIND(frame, trash_dir_getxattr_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->getxattr, &loc,
+                   GET_ANCESTRY_PATH_KEY, xdata);
+        return 0;
+    }
+
+    /* If there is no old trash directory we set its value to new one,
+     * which is the valid condition for trash directory creation
+     */
+    else {
+        gf_log(this->name, GF_LOG_DEBUG,
+               "Creating trash "
+               "directory %s ",
+               priv->newtrash_dir);
+
+        gfid_ptr = GF_MALLOC(sizeof(uuid_t), gf_common_mt_uuid_t);
+        if (!gfid_ptr) {
+            ret = ENOMEM;
+            goto out;
+        }
+        gf_uuid_copy(*gfid_ptr, trash_gfid);
+
+        gf_uuid_copy(loc.gfid, trash_gfid);
+        gf_uuid_copy(loc.pargfid, root_gfid);
+        ret = extract_trash_directory(priv->newtrash_dir, &loc.name);
+        if (ret) {
+            gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+            goto out;
+        }
+        loc.path = gf_strdup(priv->newtrash_dir);
+        if (!loc.path) {
+            gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+            ret = ENOMEM;
+            goto out;
+        }
+
+        priv->trash_inode = inode_new(priv->trash_itable);
+        priv->trash_inode->ia_type = IA_IFDIR;
+        loc.inode = inode_ref(priv->trash_inode);
+        dict = dict_new();
+        if (!dict) {
+            ret = ENOMEM;
+            goto out;
+        }
+        /* Fixed gfid is set for trash directory with
+         * this function
+         */
+        ret = dict_set_gfuuid(dict, "gfid-req", *gfid_ptr, false);
+        if (ret) {
+            gf_log(this->name, GF_LOG_ERROR, "setting key gfid-req failed");
+            goto out;
+        }
+        loc_copy(&local->loc, &loc);
+
+        /* The mkdir call for creating trash directory */
+        STACK_WIND(frame, trash_dir_mkdir_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->mkdir, &loc, 0755, 0022, dict);
         return 0;
+    }
+out:
+    if (ret && gfid_ptr)
+        GF_FREE(gfid_ptr);
+    if (dict)
+        dict_unref(dict);
+    frame->local = NULL;
+    STACK_DESTROY(frame->root);
+    trash_local_wipe(local);
+    return ret;
 }
 
-int32_t
-trash_truncate_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int32_t op_ret, int32_t op_errno,
-                           struct iatt *preparent, struct iatt *postparent)
+int
+create_or_rename_trash_directory(xlator_t *this)
 {
-        /* use this Function when a failure occurs, and
-           delete the newly created file. */
-        trash_local_t *local = NULL;
-
-        local = frame->local;
-
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "deleting the newly created file: %s",
-                        strerror (op_errno));
-        }
+    trash_private_t *priv = NULL;
+    int ret = 0;
+    loc_t loc = {
+        0,
+    };
+    call_frame_t *frame = NULL;
+    trash_local_t *local = NULL;
+
+    priv = this->private;
+
+    frame = create_frame(this, this->ctx->pool);
+    if (frame == NULL) {
+        gf_log(this->name, GF_LOG_ERROR, "failed to create frame");
+        ret = ENOMEM;
+        goto out;
+    }
+
+    local = mem_get0(this->local_pool);
+    if (!local) {
+        gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+        ret = ENOMEM;
+        goto out;
+    }
+    frame->local = local;
+
+    loc.inode = inode_new(priv->trash_itable);
+    gf_uuid_copy(loc.gfid, trash_gfid);
+    loc_copy(&local->loc, &loc);
+    gf_log(this->name, GF_LOG_DEBUG,
+           "nameless lookup for"
+           "old trash directory");
+    STACK_WIND(frame, trash_dir_lookup_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lookup, &loc, NULL);
+out:
+    return ret;
+}
 
-        STACK_WIND (frame, trash_common_unwind_buf_cbk,
-                    FIRST_CHILD(this), FIRST_CHILD(this)->fops->truncate,
-                    &local->loc, local->fop_offset);
+int
+create_internalop_directory(xlator_t *this)
+{
+    trash_private_t *priv = NULL;
+    int ret = 0;
+    loc_t loc = {
+        0,
+    };
+    call_frame_t *frame = NULL;
+    trash_local_t *local = NULL;
+
+    priv = this->private;
+
+    frame = create_frame(this, this->ctx->pool);
+    if (frame == NULL) {
+        gf_log(this->name, GF_LOG_ERROR, "failed to create frame");
+        ret = ENOMEM;
+        goto out;
+    }
+
+    local = mem_get0(this->local_pool);
+    if (!local) {
+        gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+        ret = ENOMEM;
+        goto out;
+    }
+    frame->local = local;
+
+    gf_uuid_copy(loc.gfid, internal_op_gfid);
+    gf_uuid_copy(loc.pargfid, trash_gfid);
+    loc.inode = inode_new(priv->trash_itable);
+    loc.inode->ia_type = IA_IFDIR;
+
+    loc_copy(&local->loc, &loc);
+    STACK_WIND(frame, trash_internalop_dir_lookup_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lookup, &loc, NULL);
+out:
 
-        return 0;
+    return ret;
 }
 
 int32_t
-trash_truncate_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t op_ret, int32_t op_errno,
-                          struct iovec *vector, int32_t count,
-                          struct iatt *stbuf, struct iobref *iobuf)
+trash_common_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, inode_t *inode,
+                       struct iatt *buf, struct iatt *preparent,
+                       struct iatt *postparent, dict_t *xdata)
 {
-        trash_local_t *local = NULL;
+    STACK_UNWIND_STRICT(mkdir, frame, op_ret, op_errno, inode, buf, preparent,
+                        postparent, xdata);
+    return 0;
+}
 
-        local = frame->local;
+int32_t
+trash_common_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                        struct iatt *preoldparent, struct iatt *postoldparent,
+                        struct iatt *prenewparent, struct iatt *postnewparent,
+                        dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(rename, frame, op_ret, op_errno, buf, preoldparent,
+                        postoldparent, prenewparent, postnewparent, xdata);
+    return 0;
+}
 
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "readv on the existing file failed: %s",
-                        strerror (op_errno));
+int32_t
+trash_common_rmdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, struct iatt *preparent,
+                       struct iatt *postparent, dict_t *xdata)
+{
+    STACK_UNWIND_STRICT(rmdir, frame, op_ret, op_errno, preparent, postparent,
+                        xdata);
+    return 0;
+}
 
-                STACK_WIND (frame, trash_truncate_unlink_cbk,
-                            FIRST_CHILD(this), FIRST_CHILD(this)->fops->unlink,
-                            &local->newloc);
-                goto out;
-        }
+/**
+ * move backs from trash translator to unlink call
+ */
+int32_t
+trash_common_unwind_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno,
+                        struct iatt *preparent, struct iatt *postparent,
+                        dict_t *xdata)
+{
+    TRASH_STACK_UNWIND(unlink, frame, op_ret, op_errno, preparent, postparent,
+                       xdata);
+    return 0;
+}
 
-        local->fsize = stbuf->ia_size;
-        STACK_WIND (frame, trash_truncate_writev_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->writev,
-                    local->newfd, vector, count, local->cur_offset, iobuf);
+/**
+ * If the path is not present in the trash directory,it will recursively
+ * call this call-back and one by one directories will be created from
+ * the starting
+ */
+int32_t
+trash_unlink_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, inode_t *inode,
+                       struct iatt *stbuf, struct iatt *preparent,
+                       struct iatt *postparent, dict_t *xdata)
+{
+    trash_local_t *local = NULL;
+    char *tmp_str = NULL;
+    char *tmp_path = NULL;
+    char *tmp_dirname = NULL;
+    char *tmp_stat = NULL;
+    char real_path[PATH_MAX] = {
+        0,
+    };
+    char *dir_name = NULL;
+    size_t count = 0;
+    int32_t loop_count = 0;
+    int i = 0;
+    loc_t tmp_loc = {
+        0,
+    };
+    trash_private_t *priv = NULL;
+    int ret = 0;
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO("trash", priv, out);
+
+    local = frame->local;
+    GF_VALIDATE_OR_GOTO("trash", local, out);
+
+    TRASH_UNSET_PID(frame, local);
+
+    tmp_str = gf_strdup(local->newpath);
+    if (!tmp_str) {
+        gf_log(this->name, GF_LOG_ERROR, "out of memory");
+        ret = -1;
+        goto out;
+    }
+    loop_count = local->loop_count;
+
+    /* The directory is not present , need to create it */
+    if ((op_ret == -1) && (op_errno == ENOENT)) {
+        tmp_dirname = strchr(tmp_str, '/');
+        while (tmp_dirname) {
+            count = tmp_dirname - tmp_str;
+            if (count == 0)
+                count = 1;
+            i++;
+            if (i > loop_count)
+                break;
+            tmp_dirname = strchr(tmp_str + count + 1, '/');
+        }
+        tmp_path = gf_memdup(local->newpath, count + 1);
+        if (!tmp_path) {
+            gf_log(this->name, GF_LOG_ERROR, "out of memory");
+            ret = ENOMEM;
+            goto out;
+        }
+        tmp_path[count] = '\0';
+
+        loc_copy(&tmp_loc, &local->loc);
+        tmp_loc.path = gf_strdup(tmp_path);
+        if (!tmp_loc.path) {
+            gf_log(this->name, GF_LOG_ERROR, "out of memory");
+            ret = ENOMEM;
+            goto out;
+        }
+
+        /* Stores the the name of directory to be created */
+        tmp_loc.name = gf_strdup(strrchr(tmp_path, '/') + 1);
+        if (!tmp_loc.name) {
+            gf_log(this->name, GF_LOG_ERROR, "out of memory");
+            ret = ENOMEM;
+            goto out;
+        }
+        strncpy(real_path, priv->brick_path, sizeof(real_path));
+        real_path[sizeof(real_path) - 1] = 0;
+
+        remove_trash_path(tmp_path, (frame->root->pid < 0), &tmp_stat);
+        if (tmp_stat)
+            strncat(real_path, tmp_stat,
+                    sizeof(real_path) - strlen(real_path) - 1);
+
+        TRASH_SET_PID(frame, local);
+
+        STACK_WIND_COOKIE(frame, trash_unlink_mkdir_cbk, tmp_path,
+                          FIRST_CHILD(this), FIRST_CHILD(this)->fops->mkdir,
+                          &tmp_loc, get_permission(real_path), 0022, xdata);
+        loc_wipe(&tmp_loc);
+        goto out;
+    }
+
+    /* Given path is created , comparing to the required path */
+    if (op_ret == 0) {
+        dir_name = dirname(tmp_str);
+        if (strcmp((char *)cookie, dir_name) == 0) {
+            /* File path exists we can rename it*/
+            loc_copy(&tmp_loc, &local->loc);
+            tmp_loc.path = local->newpath;
+            STACK_WIND(frame, trash_unlink_rename_cbk, FIRST_CHILD(this),
+                       FIRST_CHILD(this)->fops->rename, &local->loc, &tmp_loc,
+                       xdata);
+            goto out;
+        }
+    }
+
+    if ((op_ret == -1) && (op_errno != EEXIST)) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Directory creation failed [%s]. "
+               "Therefore unlinking %s without moving to trash "
+               "directory",
+               strerror(op_errno), local->loc.name);
+        STACK_WIND(frame, trash_common_unwind_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->unlink, &local->loc, 0, xdata);
+        goto out;
+    }
+
+    LOCK(&frame->lock);
+    {
+        loop_count = ++local->loop_count;
+    }
+    UNLOCK(&frame->lock);
+
+    tmp_dirname = strchr(tmp_str, '/');
+
+    /* Path is not completed , need to create remaining path */
+    while (tmp_dirname) {
+        count = tmp_dirname - tmp_str;
+        if (count == 0)
+            count = 1;
+        i++;
+        if (i > loop_count)
+            break;
+        tmp_dirname = strchr(tmp_str + count + 1, '/');
+    }
+    tmp_path = gf_memdup(local->newpath, count + 1);
+    if (!tmp_path) {
+        gf_log(this->name, GF_LOG_ERROR, "out of memory");
+        ret = -1;
+        goto out;
+    }
+    tmp_path[count] = '\0';
+
+    loc_copy(&tmp_loc, &local->loc);
+    tmp_loc.path = gf_strdup(tmp_path);
+    if (!tmp_loc.path) {
+        gf_log(this->name, GF_LOG_ERROR, "out of memory");
+        ret = -1;
+        goto out;
+    }
+
+    /* Stores the the name of directory to be created */
+    tmp_loc.name = gf_strdup(strrchr(tmp_path, '/') + 1);
+    if (!tmp_loc.name) {
+        gf_log(this->name, GF_LOG_ERROR, "out of memory");
+        ret = -1;
+        goto out;
+    }
+
+    strncpy(real_path, priv->brick_path, sizeof(real_path));
+    real_path[sizeof(real_path) - 1] = 0;
+
+    remove_trash_path(tmp_path, (frame->root->pid < 0), &tmp_stat);
+    if (tmp_stat)
+        strncat(real_path, tmp_stat, sizeof(real_path) - strlen(real_path) - 1);
+
+    TRASH_SET_PID(frame, local);
+
+    STACK_WIND_COOKIE(frame, trash_unlink_mkdir_cbk, tmp_path,
+                      FIRST_CHILD(this), FIRST_CHILD(this)->fops->mkdir,
+                      &tmp_loc, get_permission(real_path), 0022, xdata);
 
 out:
-        return 0;
-
+    if (tmp_path)
+        GF_FREE(tmp_path);
+    if (tmp_str)
+        GF_FREE(tmp_str);
+    return ret;
 }
 
+/**
+ * The name of unlinking file should be renamed as starting
+ * from trash directory as mentioned in the mount point
+ */
 int32_t
-trash_truncate_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int32_t op_ret, int32_t op_errno,
-                           struct iatt *prebuf, struct iatt *postbuf)
+trash_unlink_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                        struct iatt *preoldparent, struct iatt *postoldparent,
+                        struct iatt *prenewparent, struct iatt *postnewparent,
+                        dict_t *xdata)
 {
-        trash_local_t *local = NULL;
-
-        local = frame->local;
-
-        if (op_ret == -1) {
-                /* Let truncate work, but previous copy is not preserved. */
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "writev on the existing file failed: %s",
-                        strerror (op_errno));
-
-                STACK_WIND (frame, trash_truncate_unlink_cbk, FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->unlink, &local->newloc);
-                goto out;
-        }
-
-        if (local->cur_offset < local->fsize) {
-                local->cur_offset += GF_BLOCK_READV_SIZE;
-                /* Loop back and Read the contents again. */
-                STACK_WIND (frame, trash_truncate_readv_cbk,
-                            FIRST_CHILD(this), FIRST_CHILD(this)->fops->readv,
-                            local->fd, (size_t)GF_BLOCK_READV_SIZE,
-                            local->cur_offset);
-                goto out;
-        }
-
-
-        /* OOFH.....Finally calling Truncate. */
-        STACK_WIND (frame, trash_common_unwind_buf_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->truncate, &local->loc,
-                    local->fop_offset);
-
+    trash_local_t *local = NULL;
+    trash_private_t *priv = NULL;
+    char *tmp_str = NULL;
+    char *dir_name = NULL;
+    char *tmp_cookie = NULL;
+    loc_t tmp_loc = {
+        0,
+    };
+    dict_t *new_xdata = NULL;
+    char *tmp_stat = NULL;
+    char real_path[PATH_MAX] = {
+        0,
+    };
+    int ret = 0;
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO("trash", priv, out);
+
+    local = frame->local;
+    GF_VALIDATE_OR_GOTO("trash", local, out);
+
+    if ((op_ret == -1) && (op_errno == ENOENT)) {
+        /* the file path does not exist we want to create path
+         * for the file
+         */
+        tmp_str = gf_strdup(local->newpath);
+        if (!tmp_str) {
+            gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+            ret = ENOMEM;
+            goto out;
+        }
+        dir_name = dirname(tmp_str); /* stores directory name */
+
+        loc_copy(&tmp_loc, &local->loc);
+        tmp_loc.path = gf_strdup(dir_name);
+        if (!tmp_loc.path) {
+            gf_log(this->name, GF_LOG_ERROR, "out of memory");
+            ret = ENOMEM;
+            goto out;
+        }
+
+        tmp_cookie = gf_strdup(dir_name);
+        if (!tmp_cookie) {
+            gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+            ret = ENOMEM;
+            goto out;
+        }
+        strncpy(real_path, priv->brick_path, sizeof(real_path));
+        real_path[sizeof(real_path) - 1] = 0;
+        remove_trash_path(tmp_str, (frame->root->pid < 0), &tmp_stat);
+        if (tmp_stat)
+            strncat(real_path, tmp_stat,
+                    sizeof(real_path) - strlen(real_path) - 1);
+
+        TRASH_SET_PID(frame, local);
+
+        /* create the directory with proper permissions */
+        STACK_WIND_COOKIE(frame, trash_unlink_mkdir_cbk, tmp_cookie,
+                          FIRST_CHILD(this), FIRST_CHILD(this)->fops->mkdir,
+                          &tmp_loc, get_permission(real_path), 0022, xdata);
+        loc_wipe(&tmp_loc);
+        goto out;
+    }
+
+    if ((op_ret == -1) && (op_errno == ENOTDIR)) {
+        /* if entry is already present in trash directory,
+         * new one is not copied*/
+        gf_log(this->name, GF_LOG_DEBUG,
+               "target(%s) exists, cannot keep the copy, deleting",
+               local->newpath);
+
+        STACK_WIND(frame, trash_common_unwind_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->unlink, &local->loc, 0, xdata);
+
+        goto out;
+    }
+
+    if ((op_ret == -1) && (op_errno == EISDIR)) {
+        /* if entry is directory,we remove directly */
+        gf_log(this->name, GF_LOG_DEBUG,
+               "target(%s) exists as directory, cannot keep copy, "
+               "deleting",
+               local->newpath);
+
+        STACK_WIND(frame, trash_common_unwind_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->unlink, &local->loc, 0, xdata);
+        goto out;
+    }
+
+    /**********************************************************************
+     *
+     * CTR Xlator message handling done here!
+     *
+     **********************************************************************/
+    /**
+     * If unlink is handled by trash translator, it should inform the
+     * CTR Xlator. And trash translator only handles the unlink for
+     * the last hardlink.
+     *
+     * Check if there is a GF_REQUEST_LINK_COUNT_XDATA from CTR Xlator
+     *
+     */
+
+    if (local->ctr_link_count_req) {
+        /* Sending back inode link count to ctr_unlink
+         * (changetimerecoder xlator) via
+         * "GF_RESPONSE_LINK_COUNT_XDATA" key using xdata.
+         * */
+        if (xdata) {
+            ret = dict_set_uint32(xdata, GF_RESPONSE_LINK_COUNT_XDATA, 1);
+            if (ret == -1) {
+                gf_log(this->name, GF_LOG_WARNING,
+                       "Failed to set"
+                       " GF_RESPONSE_LINK_COUNT_XDATA");
+            }
+        } else {
+            new_xdata = dict_new();
+            if (!new_xdata) {
+                gf_log(this->name, GF_LOG_WARNING,
+                       "Memory allocation failure while "
+                       "creating new_xdata");
+                goto ctr_out;
+            }
+            ret = dict_set_uint32(new_xdata, GF_RESPONSE_LINK_COUNT_XDATA, 1);
+            if (ret == -1) {
+                gf_log(this->name, GF_LOG_WARNING,
+                       "Failed to set"
+                       " GF_RESPONSE_LINK_COUNT_XDATA");
+            }
+        ctr_out:
+            TRASH_STACK_UNWIND(unlink, frame, 0, op_errno, preoldparent,
+                               postoldparent, new_xdata);
+            goto out;
+        }
+    }
+    /* All other cases, unlink should return success */
+    TRASH_STACK_UNWIND(unlink, frame, 0, op_errno, preoldparent, postoldparent,
+                       xdata);
 out:
-        return 0;
-}
 
+    if (tmp_str)
+        GF_FREE(tmp_str);
+    if (tmp_cookie)
+        GF_FREE(tmp_cookie);
+    if (new_xdata)
+        dict_unref(new_xdata);
 
+    return ret;
+}
 
+/**
+ * move backs from trash translator to truncate call
+ */
 int32_t
-trash_truncate_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, fd_t *fd)
+trash_common_unwind_buf_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                            int32_t op_ret, int32_t op_errno,
+                            struct iatt *prebuf, struct iatt *postbuf,
+                            dict_t *xdata)
 {
-        trash_local_t *local = NULL;
-
-        local = frame->local;
-
-        if (op_ret == -1) {
-                //Let truncate work, but previous copy is not preserved.
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "open on the existing file failed: %s",
-                        strerror (op_errno));
+    TRASH_STACK_UNWIND(truncate, frame, op_ret, op_errno, prebuf, postbuf,
+                       xdata);
+    return 0;
+}
 
-                STACK_WIND (frame, trash_truncate_unlink_cbk,
-                            FIRST_CHILD(this), FIRST_CHILD(this)->fops->unlink,
-                            &local->newloc);
-                goto out;
-        }
+int32_t
+trash_unlink_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                      int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                      dict_t *xdata)
+{
+    trash_private_t *priv = NULL;
+    trash_local_t *local = NULL;
+    loc_t new_loc = {
+        0,
+    };
+    int ret = 0;
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO("trash", priv, out);
+
+    local = frame->local;
+    GF_VALIDATE_OR_GOTO("trash", local, out);
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_DEBUG, "%s: %s", local->loc.path,
+               strerror(op_errno));
+        TRASH_STACK_UNWIND(unlink, frame, op_ret, op_errno, buf, NULL, xdata);
+        ret = -1;
+        goto out;
+    }
+
+    /* Only last hardlink will be moved to trash directory */
+    if (buf->ia_nlink > 1) {
+        STACK_WIND(frame, trash_common_unwind_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->unlink, &local->loc, 0, xdata);
+        goto out;
+    }
+
+    /* if the file is too big  just unlink it */
+    if (buf->ia_size > (priv->max_trash_file_size)) {
+        gf_log(this->name, GF_LOG_DEBUG,
+               "%s: file size too big (%" PRId64
+               ") to "
+               "move into trash directory",
+               local->loc.path, buf->ia_size);
+
+        STACK_WIND(frame, trash_common_unwind_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->unlink, &local->loc, 0, xdata);
+        goto out;
+    }
+
+    /* Copies new path for renaming */
+    loc_copy(&new_loc, &local->loc);
+    new_loc.path = gf_strdup(local->newpath);
+    if (!new_loc.path) {
+        gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+        ret = ENOMEM;
+        goto out;
+    }
+
+    STACK_WIND(frame, trash_unlink_rename_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rename, &local->loc, &new_loc, xdata);
 
-        local->cur_offset = local->fop_offset;
+out:
+    loc_wipe(&new_loc);
 
-        STACK_WIND (frame, trash_truncate_readv_cbk,
-                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->readv,
-                    local->fd, (size_t)GF_BLOCK_READV_SIZE, local->cur_offset);
+    return ret;
+}
 
+/**
+ * Unlink is called internally by rm system call and also
+ * by internal operations of gluster such as self-heal
+ */
+int32_t
+trash_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflags,
+             dict_t *xdata)
+{
+    trash_private_t *priv = NULL;
+    trash_local_t *local = NULL; /* files inside trash */
+    int32_t match = 0;
+    int32_t ctr_link_req = 0;
+    char *pathbuf = NULL;
+    int ret = 0;
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO("trash", priv, out);
+
+    /* If trash is not active or not enabled through cli, then
+     *  we bypass and wind back
+     */
+    if (!priv->state) {
+        STACK_WIND(frame, trash_common_unwind_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->unlink, loc, 0, xdata);
+        goto out;
+    }
+
+    /* The files removed by gluster internal operations such as self-heal,
+     * should moved to trash directory , but files by client should not
+     * moved
+     */
+    if ((frame->root->pid < 0) && !priv->internal) {
+        STACK_WIND(frame, trash_common_unwind_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->unlink, loc, 0, xdata);
+        goto out;
+    }
+    /* loc need some gfid which will be present in inode */
+    gf_uuid_copy(loc->gfid, loc->inode->gfid);
+
+    /* Checking for valid location */
+    if (gf_uuid_is_null(loc->gfid) && gf_uuid_is_null(loc->inode->gfid)) {
+        gf_log(this->name, GF_LOG_DEBUG, "Bad address");
+        STACK_WIND(frame, trash_common_unwind_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->unlink, loc, 0, xdata);
+        ret = EFAULT;
+        goto out;
+    }
+
+    /* This will be more accurate */
+    inode_path(loc->inode, NULL, &pathbuf);
+    /* Check whether the file is present under eliminate paths or
+     * inside trash directory. In both cases we don't need to move the
+     * file to trash directory. Instead delete it permanently
+     */
+    match = check_whether_eliminate_path(priv->eliminate, pathbuf);
+    if ((strncmp(pathbuf, priv->newtrash_dir, strlen(priv->newtrash_dir)) ==
+         0) ||
+        (match)) {
+        if (match) {
+            gf_log(this->name, GF_LOG_DEBUG,
+                   "%s is a file comes under an eliminate path, "
+                   "so it is not moved to trash",
+                   loc->name);
+        }
+
+        /* Trying to unlink from the trash-dir. So do the
+         * actual unlink without moving to trash-dir.
+         */
+        STACK_WIND(frame, trash_common_unwind_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->unlink, loc, 0, xdata);
+        goto out;
+    }
+
+    local = mem_get0(this->local_pool);
+    if (!local) {
+        gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+        TRASH_STACK_UNWIND(unlink, frame, -1, ENOMEM, NULL, NULL, xdata);
+        ret = ENOMEM;
+        goto out;
+    }
+    frame->local = local;
+    loc_copy(&local->loc, loc);
+
+    /* rename new location of file as starting from trash directory */
+    copy_trash_path(priv->newtrash_dir, (frame->root->pid < 0), local->newpath,
+                    sizeof(local->newpath));
+    strncat(local->newpath, pathbuf,
+            sizeof(local->newpath) - strlen(local->newpath) - 1);
+
+    /* append timestamp to file name so that we can avoid
+     * name collisions inside trash
+     */
+    append_time_stamp(local->newpath, sizeof(local->newpath));
+    if (strlen(local->newpath) > PATH_MAX) {
+        STACK_WIND(frame, trash_common_unwind_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->unlink, loc, 0, xdata);
+        goto out;
+    }
+
+    /* To know whether CTR xlator requested for the link count */
+    ret = dict_get_int32(xdata, GF_REQUEST_LINK_COUNT_XDATA, &ctr_link_req);
+    if (ret) {
+        local->ctr_link_count_req = _gf_false;
+        ret = 0;
+    } else
+        local->ctr_link_count_req = _gf_true;
+
+    LOCK_INIT(&frame->lock);
+
+    STACK_WIND(frame, trash_unlink_stat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->stat, loc, xdata);
 out:
-        return 0;
+    return ret;
 }
 
-
+/**
+ * Use this when a failure occurs, and delete the newly created file
+ */
 int32_t
-trash_truncate_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int32_t op_ret, int32_t op_errno, fd_t *fd,
-                           inode_t *inode, struct iatt *buf,
-                           struct iatt *preparent, struct iatt *postparent)
+trash_truncate_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno,
+                          struct iatt *preparent, struct iatt *postparent,
+                          dict_t *xdata)
 {
-        trash_local_t       *local    = NULL;
-        char                *tmp_str  = NULL;
-        char                *dir_name = NULL;
-        char                *tmp_path = NULL;
-        int32_t              flags    = 0;
-        loc_t                tmp_loc  = {0,};
-
-        local = frame->local;
-
-        if ((op_ret == -1) && (op_errno == ENOENT)) {
-                //Creating the directory structure here.
-                tmp_str = gf_strdup (local->newpath);
-                if (!tmp_str) {
-                        gf_log (this->name, GF_LOG_DEBUG, "out of memory");
-                }
-                dir_name = dirname (tmp_str);
-
-                tmp_path = gf_strdup (dir_name);
-                if (!tmp_path) {
-                        gf_log (this->name, GF_LOG_DEBUG, "out of memory");
-                }
-                tmp_loc.path = tmp_path;
-
-                /* TODO: create the directory with proper permissions */
-                STACK_WIND_COOKIE (frame, trash_truncate_mkdir_cbk,
-                                   tmp_path, FIRST_CHILD(this),
-                                   FIRST_CHILD(this)->fops->mkdir,
-                                   &tmp_loc, 0755, NULL);
-                GF_FREE (tmp_str);
-                goto out;
-        }
-
-        if (op_ret == -1) {
-                //Let truncate work, but previous copy is not preserved.
-                //Deleting the newly created copy.
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "creation of new file in trash-dir failed, "
-                        "when truncate was called: %s", strerror (op_errno));
-
-                STACK_WIND (frame, trash_common_unwind_buf_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->truncate, &local->loc,
-                            local->fop_offset);
-                goto out;
-        }
+    trash_local_t *local = NULL;
 
-        flags = O_RDONLY;
+    local = frame->local;
+    GF_VALIDATE_OR_GOTO("trash", local, out);
 
-        local->fd = fd_create (local->loc.inode, frame->root->pid);
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_DEBUG, "deleting the newly created file: %s",
+               strerror(op_errno));
+    }
 
-        STACK_WIND (frame, trash_truncate_open_cbk,  FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->open, &local->loc, flags,
-                    local->fd, 0);
+    STACK_WIND(frame, trash_common_unwind_buf_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->truncate, &local->loc,
+               local->fop_offset, xdata);
 out:
-        return 0;
+    return 0;
 }
 
+/**
+ * Read from source file
+ */
 int32_t
-trash_truncate_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                          int32_t op_ret, int32_t op_errno, inode_t *inode,
-                          struct iatt *stbuf, struct iatt *preparent,
-                          struct iatt *postparent)
+trash_truncate_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, struct iovec *vector,
+                         int32_t count, struct iatt *stbuf,
+                         struct iobref *iobuf, dict_t *xdata)
 {
-        trash_local_t       *local = NULL;
-        char                *tmp_str = NULL;
-        char                *tmp_path = NULL;
-        char                *tmp_dirname = NULL;
-        char                *dir_name = NULL;
-        int32_t              count = 0;
-        int32_t              flags = 0;
-        int32_t              loop_count = 0;
-        int                  i = 0;
-        loc_t                tmp_loc = {0,};
-
-        local   = frame->local;
-        if (!local)
-                goto out;
-
-        loop_count = local->loop_count;
-
-        tmp_str = gf_strdup (local->newpath);
-        if (!tmp_str) {
-                gf_log (this->name, GF_LOG_DEBUG, "out of memory");
-                goto out;
-        }
-
-        if ((op_ret == -1) && (op_errno == ENOENT)) {
-                tmp_dirname = strchr (tmp_str, '/');
-                while (tmp_dirname) {
-                        count = tmp_dirname - tmp_str;
-                        if (count == 0)
-                                count = 1;
-                        i++;
-                        if (i > loop_count)
-                                break;
-                        tmp_dirname = strchr (tmp_str + count + 1, '/');
-                }
-                tmp_path = memdup (local->newpath, count);
-                if (!tmp_path) {
-                        gf_log (this->name, GF_LOG_DEBUG, "out of memory");
-                }
-                tmp_loc.path = tmp_path;
-                STACK_WIND_COOKIE (frame, trash_truncate_mkdir_cbk,
-                                   tmp_path, this->children->xlator,
-                                   this->children->xlator->fops->mkdir,
-                                   &tmp_loc, 0755, NULL);
-
-                goto out;
-        }
+    trash_local_t *local = NULL;
 
-        if (op_ret == 0) {
-                dir_name = dirname (tmp_str);
-                if (strcmp ((char*)cookie, dir_name) == 0) {
-                        flags = O_CREAT|O_EXCL|O_WRONLY;
-                        ia_prot_t prot = {0, };
-
-                        //Call create again once directory structure is created.
-                        STACK_WIND (frame, trash_truncate_create_cbk,
-                                    FIRST_CHILD(this), FIRST_CHILD(this)->fops->create,
-                                    &local->newloc, flags,
-                                    st_mode_from_ia (prot, local->loc.inode->ia_type),
-                                    local->newfd, NULL);
-                        goto out;
-                }
-        }
+    local = frame->local;
+    GF_VALIDATE_OR_GOTO("trash", local, out);
 
-        LOCK (&frame->lock);
-        {
-                loop_count = ++local->loop_count;
-        }
-        UNLOCK (&frame->lock);
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_DEBUG,
+               "readv on the existing file failed: %s", strerror(op_errno));
 
-        tmp_dirname = strchr (tmp_str, '/');
-        while (tmp_dirname) {
-                count = tmp_dirname - tmp_str;
-                if (count == 0)
-                        count = 1;
-
-                i++;
-                if ((i > loop_count) || (count > PATH_MAX))
-                        break;
-                tmp_dirname = strchr (tmp_str + count + 1, '/');
-        }
-        tmp_path = memdup (local->newpath, count);
-        if (!tmp_path) {
-                gf_log (this->name, GF_LOG_DEBUG, "out of memory");
-        }
-        tmp_loc.path = tmp_path;
+        STACK_WIND(frame, trash_truncate_unlink_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->unlink, &local->newloc, 0, xdata);
+        goto out;
+    }
 
-        STACK_WIND_COOKIE (frame, trash_truncate_mkdir_cbk, tmp_path,
-                           this->children->xlator,
-                           this->children->xlator->fops->mkdir,
-                           &tmp_loc, 0755, NULL);
+    local->fsize = stbuf->ia_size;
+    STACK_WIND(frame, trash_truncate_writev_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->writev, local->newfd, vector, count,
+               local->cur_offset, 0, iobuf, xdata);
 
 out:
-        GF_FREE (cookie); /* strdup (dir_name) was sent here :) */
-        if (tmp_str)
-                GF_FREE (tmp_str);
-
-        return 0;
+    return 0;
 }
 
-
+/**
+ * Write to file created in trash directory
+ */
 int32_t
-trash_truncate_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                         int32_t op_ret, int32_t op_errno, struct iatt *buf)
+trash_truncate_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+                          struct iatt *postbuf, dict_t *xdata)
 {
-        trash_private_t     *priv  = NULL;
-        trash_local_t       *local = NULL;
-        struct tm           *tm = NULL;
-        char                 timestr[256] = {0,};
-        char                 loc_newname[PATH_MAX] = {0,};
-        time_t               utime = 0;
-        int32_t              flags = 0;
-
-        priv = this->private;
-        local = frame->local;
-
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "fstat on the file failed: %s",
-                        strerror (op_errno));
-
-                TRASH_STACK_UNWIND (truncate, frame, op_ret, op_errno, buf, NULL);
-                return 0;
-        }
-
-        if ((buf->ia_size == 0) || (buf->ia_size > priv->max_trash_file_size)) {
-                // If the file is too big, just unlink it.
-                if (buf->ia_size > priv->max_trash_file_size)
-                        gf_log (this->name, GF_LOG_DEBUG, "%s: file too big, "
-                                "not moving to trash", local->loc.path);
-
-                STACK_WIND (frame,  trash_common_unwind_buf_cbk,
-                            this->children->xlator,
-                            this->children->xlator->fops->truncate,
-                            &local->loc, local->fop_offset);
-                return 0;
-        }
-
-        strcpy (local->newpath, priv->trash_dir);
-        strcat (local->newpath, local->loc.path);
-
-        {
-                utime = time (NULL);
-                tm    = localtime (&utime);
-                strftime (timestr, 256, ".%Y-%m-%d-%H%M%S", tm);
-                strcat (local->newpath, timestr);
-        }
-        strcpy (loc_newname,local->loc.name);
-        strcat (loc_newname,timestr);
-
-        local->newloc.name = gf_strdup (loc_newname);
-        local->newloc.path = gf_strdup (local->newpath);
-        local->newloc.inode = inode_new (local->loc.inode->table);
-        local->newloc.ino   = local->newloc.inode->ino;
-        local->newfd = fd_create (local->newloc.inode, frame->root->pid);
+    trash_local_t *local = NULL;
+
+    local = frame->local;
+    GF_VALIDATE_OR_GOTO("trash", local, out);
+
+    if (op_ret == -1) {
+        /* Let truncate work, but previous copy is not preserved. */
+        gf_log(this->name, GF_LOG_DEBUG,
+               "writev on the existing file failed: %s", strerror(op_errno));
+
+        STACK_WIND(frame, trash_truncate_unlink_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->unlink, &local->newloc, 0, xdata);
+        goto out;
+    }
+
+    if (local->cur_offset < local->fsize) {
+        local->cur_offset += GF_BLOCK_READV_SIZE;
+        /* Loop back and Read the contents again. */
+        STACK_WIND(frame, trash_truncate_readv_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->readv, local->fd,
+                   (size_t)GF_BLOCK_READV_SIZE, local->cur_offset, 0, xdata);
+        goto out;
+    }
+
+    /* OOFH.....Finally calling Truncate. */
+    STACK_WIND(frame, trash_common_unwind_buf_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->truncate, &local->loc,
+               local->fop_offset, xdata);
 
-        flags = O_CREAT|O_EXCL|O_WRONLY;
-
-        STACK_WIND (frame, trash_truncate_create_cbk,
-                    FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->create,
-                    &local->newloc, flags,
-                    st_mode_from_ia (buf->ia_prot, local->loc.inode->ia_type),
-                    local->newfd, NULL);
-
-        return 0;
+out:
+    return 0;
 }
 
+/**
+ * The source file is opened for reading and writing
+ */
 int32_t
-trash_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc,
-                off_t offset)
+trash_truncate_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, fd_t *fd,
+                        dict_t *xdata)
 {
-        trash_elim_pattern_t *trav = NULL;
-        trash_private_t      *priv = NULL;
-        trash_local_t        *local = NULL;
-        int32_t               match = 0;
+    trash_local_t *local = NULL;
 
-        priv  = this->private;
-        if (priv->eliminate) {
-                trav = priv->eliminate;
-                while (trav) {
-                        if (fnmatch(trav->pattern, loc->name, 0) == 0) {
-                                match++;
-                                break;
-                        }
-                        trav = trav->next;
-                }
-        }
+    local = frame->local;
+    GF_VALIDATE_OR_GOTO("trash", local, out);
 
-        if ((strncmp (loc->path, priv->trash_dir,
-                      strlen (priv->trash_dir)) == 0) || (offset) || (match)) {
-                if (match) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "%s: file not moved to trash as per option "
-                                "'eliminate'", loc->path);
-                }
-
-                // Trying to truncate from the trash can dir,
-                //   do the actual truncate without moving to trash-dir.
-                STACK_WIND (frame, trash_common_unwind_buf_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->truncate, loc, offset);
-                goto out;
-        }
-
-        LOCK_INIT (&frame->lock);
-
-        local = GF_CALLOC (1, sizeof (trash_local_t), 
-                           gf_trash_mt_trash_local_t);
-        if (!local) {
-                gf_log (this->name, GF_LOG_DEBUG, "out of memory");
-                TRASH_STACK_UNWIND (truncate, frame, -1, ENOMEM, NULL, NULL);
-                return 0;
-        }
+    if (op_ret == -1) {
+        /* Let truncate work, but previous copy is not preserved. */
+        gf_log(this->name, GF_LOG_DEBUG, "open on the existing file failed: %s",
+               strerror(op_errno));
 
-        loc_copy (&local->loc, loc);
+        STACK_WIND(frame, trash_truncate_unlink_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->unlink, &local->newloc, 0, xdata);
+        goto out;
+    }
 
-        local->fop_offset = offset;
+    fd_bind(fd);
 
-        frame->local = local;
+    local->cur_offset = 0;
 
-        STACK_WIND (frame, trash_truncate_stat_cbk,
-                    this->children->xlator,
-                    this->children->xlator->fops->stat, loc);
+    STACK_WIND(frame, trash_truncate_readv_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readv, local->fd,
+               (size_t)GF_BLOCK_READV_SIZE, local->cur_offset, 0, xdata);
 
 out:
-        return 0;
+    return 0;
 }
 
+/**
+ * Creates new file descriptor for read and write operations,
+ * if the path is present in trash directory
+ */
 int32_t
-trash_ftruncate_unlink_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                            int32_t op_ret, int32_t op_errno,
-                            struct iatt *preparent, struct iatt *postparent)
+trash_truncate_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                          int32_t op_ret, int32_t op_errno, fd_t *fd,
+                          inode_t *inode, struct iatt *buf,
+                          struct iatt *preparent, struct iatt *postparent,
+                          dict_t *xdata)
 {
-        trash_local_t *local = NULL;
-
-        local = frame->local;
-
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "%s: failed to unlink new file: %s",
-                        local->newloc.path, strerror(op_errno));
-
+    trash_local_t *local = NULL;
+    char *tmp_str = NULL;
+    char *dir_name = NULL;
+    char *tmp_path = NULL;
+    int32_t flags = 0;
+    loc_t tmp_loc = {
+        0,
+    };
+    char *tmp_stat = NULL;
+    char real_path[PATH_MAX] = {
+        0,
+    };
+    trash_private_t *priv = NULL;
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO("trash", priv, out);
+
+    local = frame->local;
+    GF_VALIDATE_OR_GOTO("trash", local, out);
+
+    TRASH_UNSET_PID(frame, local);
+
+    /* Checks whether path is present in trash directory or not */
+
+    if ((op_ret == -1) && (op_errno == ENOENT)) {
+        /* Creating the directory structure here. */
+        tmp_str = gf_strdup(local->newpath);
+        if (!tmp_str) {
+            gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+            goto out;
         }
+        dir_name = dirname(tmp_str);
 
-        STACK_WIND (frame, trash_common_unwind_buf_cbk,
-                    FIRST_CHILD(this), FIRST_CHILD(this)->fops->ftruncate,
-                    local->fd, local->fop_offset);
+        tmp_path = gf_strdup(dir_name);
+        if (!tmp_path) {
+            gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+            goto out;
+        }
+        loc_copy(&tmp_loc, &local->newloc);
+        tmp_loc.path = gf_strdup(tmp_path);
+        if (!tmp_loc.path) {
+            gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+            goto out;
+        }
+        strncpy(real_path, priv->brick_path, sizeof(real_path));
+        real_path[sizeof(real_path) - 1] = 0;
+        remove_trash_path(tmp_path, (frame->root->pid < 0), &tmp_stat);
+        if (tmp_stat)
+            strncat(real_path, tmp_stat,
+                    sizeof(real_path) - strlen(real_path) - 1);
+
+        TRASH_SET_PID(frame, local);
+
+        /* create the directory with proper permissions */
+        STACK_WIND_COOKIE(frame, trash_truncate_mkdir_cbk, tmp_path,
+                          FIRST_CHILD(this), FIRST_CHILD(this)->fops->mkdir,
+                          &tmp_loc, get_permission(real_path), 0022, xdata);
+        loc_wipe(&tmp_loc);
+        goto out;
+    }
+
+    if (op_ret == -1) {
+        /* Let truncate work, but previous copy is not preserved.
+         * Deleting the newly created copy.
+         */
+        gf_log(this->name, GF_LOG_DEBUG,
+               "creation of new file in trash-dir failed, "
+               "when truncate was called: %s",
+               strerror(op_errno));
+
+        STACK_WIND(frame, trash_common_unwind_buf_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->truncate, &local->loc,
+                   local->fop_offset, xdata);
+        goto out;
+    }
+
+    fd_bind(fd);
+    flags = O_RDONLY;
+
+    /* fd which represents source file for reading and writing from it */
+
+    local->fd = fd_create(local->loc.inode, frame->root->pid);
+
+    STACK_WIND(frame, trash_truncate_open_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->open, &local->loc, flags, local->fd, 0);
+out:
+    if (tmp_str)
+        GF_FREE(tmp_str);
+    if (tmp_path)
+        GF_FREE(tmp_path);
 
-        return 0;
+    return 0;
 }
 
+/**
+ * If the path is not present in the trash directory,it will recursively call
+ * this call-back and one by one directories will be created from the
+ * beginning
+ */
 int32_t
-trash_ftruncate_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                            int32_t op_ret, int32_t op_errno,
-                            struct iatt *prebuf, struct iatt *postbuf)
+trash_truncate_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                         int32_t op_ret, int32_t op_errno, inode_t *inode,
+                         struct iatt *stbuf, struct iatt *preparent,
+                         struct iatt *postparent, dict_t *xdata)
 {
-        trash_local_t *local = NULL;
-
-        local = frame->local;
-
-        if (op_ret == -1) {
-                STACK_WIND (frame, trash_ftruncate_unlink_cbk,
-                            FIRST_CHILD(this), FIRST_CHILD(this)->fops->unlink,
-                            &local->newloc);
-                return 0;
-        }
-
-        if (local->cur_offset < local->fsize) {
-                local->cur_offset += GF_BLOCK_READV_SIZE;
-                STACK_WIND (frame, trash_ftruncate_readv_cbk,
-                            FIRST_CHILD(this), FIRST_CHILD(this)->fops->readv,
-                            local->fd, (size_t)GF_BLOCK_READV_SIZE,
-                            local->cur_offset);
-                return 0;
-        }
+    trash_local_t *local = NULL;
+    trash_private_t *priv = NULL;
+    char *tmp_str = NULL;
+    char *tmp_path = NULL;
+    char *tmp_dirname = NULL;
+    char *dir_name = NULL;
+    char *tmp_stat = NULL;
+    char real_path[PATH_MAX] = {
+        0,
+    };
+    size_t count = 0;
+    int32_t flags = 0;
+    int32_t loop_count = 0;
+    int i = 0;
+    loc_t tmp_loc = {
+        0,
+    };
+    int ret = 0;
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO("trash", priv, out);
+
+    local = frame->local;
+    GF_VALIDATE_OR_GOTO("trash", local, out);
+
+    loop_count = local->loop_count;
+
+    TRASH_UNSET_PID(frame, local);
+
+    tmp_str = gf_strdup(local->newpath);
+    if (!tmp_str) {
+        gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+        ret = ENOMEM;
+        goto out;
+    }
+
+    if ((op_ret == -1) && (op_errno == ENOENT)) {
+        tmp_dirname = strchr(tmp_str, '/');
+        while (tmp_dirname) {
+            count = tmp_dirname - tmp_str;
+            if (count == 0)
+                count = 1;
+            i++;
+            if (i > loop_count)
+                break;
+            tmp_dirname = strchr(tmp_str + count + 1, '/');
+        }
+        tmp_path = gf_memdup(local->newpath, count + 1);
+        if (!tmp_path) {
+            gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+            ret = ENOMEM;
+            goto out;
+        }
+        tmp_path[count] = '\0';
+
+        loc_copy(&tmp_loc, &local->newloc);
+        tmp_loc.path = gf_strdup(tmp_path);
+        if (!tmp_loc.path) {
+            gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+            ret = ENOMEM;
+            goto out;
+        }
+
+        /* Stores the the name of directory to be created */
+        tmp_loc.name = gf_strdup(strrchr(tmp_path, '/') + 1);
+        if (!tmp_loc.name) {
+            gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+            ret = ENOMEM;
+            goto out;
+        }
+        strncpy(real_path, priv->brick_path, sizeof(real_path));
+        real_path[sizeof(real_path) - 1] = 0;
+        remove_trash_path(tmp_path, (frame->root->pid < 0), &tmp_stat);
+        if (tmp_stat)
+            strncat(real_path, tmp_stat,
+                    sizeof(real_path) - strlen(real_path) - 1);
+
+        TRASH_SET_PID(frame, local);
+
+        STACK_WIND_COOKIE(frame, trash_truncate_mkdir_cbk, tmp_path,
+                          FIRST_CHILD(this), FIRST_CHILD(this)->fops->mkdir,
+                          &tmp_loc, get_permission(real_path), 0022, xdata);
+        loc_wipe(&tmp_loc);
+        goto out;
+    }
+
+    if (op_ret == 0) {
+        dir_name = dirname(tmp_str);
+        if (strcmp((char *)cookie, dir_name) == 0) {
+            flags = O_CREAT | O_EXCL | O_WRONLY;
+            strncpy(real_path, priv->brick_path, sizeof(real_path));
+            real_path[sizeof(real_path) - 1] = 0;
+            strncat(real_path, local->origpath,
+                    sizeof(real_path) - strlen(real_path) - 1);
+            /* Call create again once directory structure
+               is created. */
+
+            TRASH_SET_PID(frame, local);
+
+            STACK_WIND(frame, trash_truncate_create_cbk, FIRST_CHILD(this),
+                       FIRST_CHILD(this)->fops->create, &local->newloc, flags,
+                       get_permission(real_path), 0022, local->newfd, xdata);
+            goto out;
+        }
+    }
+
+    if ((op_ret == -1) && (op_errno != EEXIST)) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Directory creation failed [%s]. "
+               "Therefore truncating %s without moving the "
+               "original copy to trash directory",
+               strerror(op_errno), local->loc.name);
+        STACK_WIND(frame, trash_common_unwind_buf_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->truncate, &local->loc,
+                   local->fop_offset, xdata);
+        goto out;
+    }
+
+    LOCK(&frame->lock);
+    {
+        loop_count = ++local->loop_count;
+    }
+    UNLOCK(&frame->lock);
+
+    tmp_dirname = strchr(tmp_str, '/');
+    while (tmp_dirname) {
+        count = tmp_dirname - tmp_str;
+        if (count == 0)
+            count = 1;
+        i++;
+        if (i > loop_count)
+            break;
+        tmp_dirname = strchr(tmp_str + count + 1, '/');
+    }
+    tmp_path = gf_memdup(local->newpath, count + 1);
+    if (!tmp_path) {
+        gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+        ret = ENOMEM;
+        goto out;
+    }
+    tmp_path[count] = '\0';
+
+    loc_copy(&tmp_loc, &local->newloc);
+    tmp_loc.path = gf_strdup(tmp_path);
+    if (!tmp_loc.path) {
+        gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+        ret = ENOMEM;
+        goto out;
+    }
+
+    /* Stores the the name of directory to be created */
+    tmp_loc.name = gf_strdup(strrchr(tmp_path, '/') + 1);
+    if (!tmp_loc.name) {
+        gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+        goto out;
+    }
+
+    strncpy(real_path, priv->brick_path, sizeof(real_path));
+    real_path[sizeof(real_path) - 1] = 0;
+    remove_trash_path(tmp_path, (frame->root->pid < 0), &tmp_stat);
+    if (tmp_stat)
+        strncat(real_path, tmp_stat, sizeof(real_path) - strlen(real_path) - 1);
+
+    TRASH_SET_PID(frame, local);
+
+    STACK_WIND_COOKIE(frame, trash_truncate_mkdir_cbk, tmp_path,
+                      FIRST_CHILD(this), FIRST_CHILD(this)->fops->mkdir,
+                      &tmp_loc, get_permission(real_path), 0022, xdata);
 
-        STACK_WIND (frame, trash_common_unwind_buf_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->ftruncate, local->fd,
-                    local->fop_offset);
+out:
+    if (tmp_str)
+        GF_FREE(tmp_str);
+    if (tmp_path)
+        GF_FREE(tmp_path);
 
-        return 0;
+    return ret;
 }
 
-
 int32_t
-trash_ftruncate_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int32_t op_ret, int32_t op_errno,
-                           struct iovec *vector, int32_t count,
-                           struct iatt *stbuf, struct iobref *iobuf)
+trash_truncate_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int32_t op_ret, int32_t op_errno, struct iatt *buf,
+                        dict_t *xdata)
 {
-        trash_local_t *local = NULL;
+    trash_private_t *priv = NULL;
+    trash_local_t *local = NULL;
+    char loc_newname[PATH_MAX] = {
+        0,
+    };
+    int32_t flags = 0;
+    dentry_t *dir_entry = NULL;
+    inode_table_t *table = NULL;
+    int ret = 0;
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO("trash", priv, out);
+
+    local = frame->local;
+    GF_VALIDATE_OR_GOTO("trash", local, out);
+
+    table = local->loc.inode->table;
+
+    pthread_mutex_lock(&table->lock);
+    {
+        dir_entry = __dentry_search_arbit(local->loc.inode);
+    }
+    pthread_mutex_unlock(&table->lock);
+
+    if (op_ret == -1) {
+        gf_log(this->name, GF_LOG_DEBUG, "fstat on the file failed: %s",
+               strerror(op_errno));
+
+        TRASH_STACK_UNWIND(truncate, frame, op_ret, op_errno, buf, NULL, xdata);
+        goto out;
+    }
+
+    /* Only last hardlink will be moved to trash directory */
+    if (buf->ia_nlink > 1) {
+        STACK_WIND(frame, trash_common_unwind_buf_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->truncate, &local->loc,
+                   local->fop_offset, xdata);
+        goto out;
+    }
+
+    /**
+     * If the file is too big or if it is extended truncate,
+     * just don't move it to trash directory.
+     */
+    if (buf->ia_size > (priv->max_trash_file_size) ||
+        buf->ia_size <= local->fop_offset) {
+        gf_log(this->name, GF_LOG_DEBUG,
+               "%s: file is too large to move to trash", local->loc.path);
+
+        STACK_WIND(frame, trash_common_unwind_buf_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->truncate, &local->loc,
+                   local->fop_offset, xdata);
+        goto out;
+    }
+
+    /* Retrieves the name of file from path */
+    local->loc.name = gf_strdup(strrchr(local->loc.path, '/'));
+    if (!local->loc.name) {
+        gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+        goto out;
+    }
+
+    /* Stores new path for source file */
+    copy_trash_path(priv->newtrash_dir, (frame->root->pid < 0), local->newpath,
+                    sizeof(local->newpath));
+    strncat(local->newpath, local->loc.path,
+            sizeof(local->newpath) - strlen(local->newpath) - 1);
+
+    /* append timestamp to file name so that we can avoid
+       name collisions inside trash */
+    append_time_stamp(local->newpath, sizeof(local->newpath));
+    if (strlen(local->newpath) > PATH_MAX) {
+        STACK_WIND(frame, trash_common_unwind_buf_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->truncate, &local->loc,
+                   local->fop_offset, xdata);
+        goto out;
+    }
+
+    strncpy(loc_newname, local->loc.name, sizeof(loc_newname));
+    loc_newname[sizeof(loc_newname) - 1] = 0;
+    append_time_stamp(loc_newname, sizeof(loc_newname));
+    /* local->newloc represents old file(file inside trash),
+       where as local->loc represents truncated file. We need
+       to create new inode and fd for new file*/
+    local->newloc.name = gf_strdup(loc_newname);
+    if (!local->newloc.name) {
+        gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+        ret = ENOMEM;
+        goto out;
+    }
+    local->newloc.path = gf_strdup(local->newpath);
+    if (!local->newloc.path) {
+        gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+        ret = ENOMEM;
+        goto out;
+    }
+    local->newloc.inode = inode_new(local->loc.inode->table);
+    local->newfd = fd_create(local->newloc.inode, frame->root->pid);
+
+    /* Creating valid parent and pargfids for both files */
+
+    if (dir_entry == NULL) {
+        ret = EINVAL;
+        goto out;
+    }
+    local->loc.parent = inode_ref(dir_entry->parent);
+    gf_uuid_copy(local->loc.pargfid, dir_entry->parent->gfid);
+
+    local->newloc.parent = inode_ref(dir_entry->parent);
+    gf_uuid_copy(local->newloc.pargfid, dir_entry->parent->gfid);
+
+    flags = O_CREAT | O_EXCL | O_WRONLY;
+
+    TRASH_SET_PID(frame, local);
+
+    STACK_WIND(frame, trash_truncate_create_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->create, &local->newloc, flags,
+               st_mode_from_ia(buf->ia_prot, local->loc.inode->ia_type), 0022,
+               local->newfd, xdata);
 
-        local = frame->local;
-        local->fsize = stbuf->ia_size;
-
-        if (op_ret == -1) {
-                STACK_WIND (frame, trash_ftruncate_unlink_cbk,
-                            FIRST_CHILD(this), FIRST_CHILD(this)->fops->unlink,
-                            &local->newloc);
-                return 0;
-        }
+out:
+    return ret;
+}
 
-        STACK_WIND (frame, trash_ftruncate_writev_cbk,
-                    FIRST_CHILD(this), FIRST_CHILD(this)->fops->writev,
-                    local->newfd, vector, count, local->cur_offset, NULL);
+/**
+ * Truncate can be explicitly called or implicitly by some other applications
+ * like text editors etc..
+ */
+int32_t
+trash_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+               dict_t *xdata)
+{
+    trash_private_t *priv = NULL;
+    trash_local_t *local = NULL;
+    int32_t match = 0;
+    char *pathbuf = NULL;
+    int ret = 0;
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO("trash", priv, out);
+    /* If trash is not active or not enabled through cli, then
+     * we bypass and wind back
+     */
+    if (!priv->state) {
+        STACK_WIND(frame, trash_common_unwind_buf_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+        goto out;
+    }
+
+    /* The files removed by gluster operations such as self-heal,
+       should moved to trash directory, but files by client should
+       not moved */
+    if ((frame->root->pid < 0) && !priv->internal) {
+        STACK_WIND(frame, trash_common_unwind_buf_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+        goto out;
+    }
+    /* This will be more accurate */
+    inode_path(loc->inode, NULL, &pathbuf);
+
+    /* Checks whether file is in trash directory or eliminate path.
+     * In all such cases it does not move to trash directory,
+     * truncate will be performed
+     */
+    match = check_whether_eliminate_path(priv->eliminate, pathbuf);
+
+    if ((strncmp(pathbuf, priv->newtrash_dir, strlen(priv->newtrash_dir)) ==
+         0) ||
+        (match)) {
+        if (match) {
+            gf_log(this->name, GF_LOG_DEBUG,
+                   "%s: file not moved to trash as per option "
+                   "'eliminate path'",
+                   loc->path);
+        }
+
+        /* Trying to truncate from the trash-dir. So do the
+         * actual truncate without moving to trash-dir.
+         */
+        STACK_WIND(frame, trash_common_unwind_buf_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+        goto out;
+    }
+
+    LOCK_INIT(&frame->lock);
+
+    local = mem_get0(this->local_pool);
+    if (!local) {
+        gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+        TRASH_STACK_UNWIND(truncate, frame, -1, ENOMEM, NULL, NULL, xdata);
+        ret = ENOMEM;
+        goto out;
+    }
+
+    strncpy(local->origpath, pathbuf, sizeof(local->origpath));
+    local->origpath[sizeof(local->origpath) - 1] = 0;
+
+    loc_copy(&local->loc, loc);
+    local->loc.path = pathbuf;
+    local->fop_offset = offset;
+
+    frame->local = local;
+
+    STACK_WIND(frame, trash_truncate_stat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->stat, loc, xdata);
 
-        return 0;
+out:
+    return ret;
 }
 
+/**
+ * When we call truncate from terminal it comes to ftruncate of trash-xlator.
+ * Since truncate internally calls ftruncate and we receive fd of the file,
+ * other than that it also called by Rebalance operation
+ */
+int32_t
+trash_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+                dict_t *xdata)
+{
+    trash_private_t *priv = NULL;
+    trash_local_t *local = NULL; /* file inside trash */
+    char *pathbuf = NULL;        /* path of file from fd */
+    int32_t retval = 0;
+    int32_t match = 0;
+    int ret = 0;
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO("trash", priv, out);
+    /* If trash is not active or not enabled through cli, then
+     * we bypass and wind back
+     */
+    if (!priv->state) {
+        STACK_WIND(frame, trash_common_unwind_buf_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+        goto out;
+    }
+
+    /* The files removed by gluster operations such as self-heal,
+     * should moved to trash directory, but files by client
+     * should not moved
+     */
+    if ((frame->root->pid < 0) && !priv->internal) {
+        STACK_WIND(frame, trash_common_unwind_buf_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+        goto out;
+    }
+    /* This will be more accurate */
+    retval = inode_path(fd->inode, NULL, &pathbuf);
+
+    /* Checking  the eliminate path */
+
+    /* Checks whether file is trash directory or eliminate path or
+     * invalid fd. In all such cases it does not move to trash directory,
+     * ftruncate will be performed
+     */
+    match = check_whether_eliminate_path(priv->eliminate, pathbuf);
+    if ((strncmp(pathbuf, priv->newtrash_dir, strlen(priv->newtrash_dir)) ==
+         0) ||
+        match || !retval) {
+        if (match) {
+            gf_log(this->name, GF_LOG_DEBUG,
+                   "%s: file matches eliminate path, "
+                   "not moved to trash",
+                   pathbuf);
+        }
+
+        /* Trying to ftruncate from the trash-dir. So do the
+         * actual ftruncate without moving to trash-dir
+         */
+        STACK_WIND(frame, trash_common_unwind_buf_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+        goto out;
+    }
+
+    local = mem_get0(this->local_pool);
+    if (!local) {
+        gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+        TRASH_STACK_UNWIND(ftruncate, frame, -1, ENOMEM, NULL, NULL, xdata);
+        ret = -1;
+        goto out;
+    }
+
+    strncpy(local->origpath, pathbuf, sizeof(local->origpath));
+    local->origpath[sizeof(local->origpath) - 1] = 0;
+
+    /* To convert fd to location */
+    frame->local = local;
+
+    local->loc.path = pathbuf;
+    local->loc.inode = inode_ref(fd->inode);
+    gf_uuid_copy(local->loc.gfid, local->loc.inode->gfid);
+
+    local->fop_offset = offset;
+
+    /* Else remains same to truncate code, so from here flow goes
+     * to truncate_stat
+     */
+    STACK_WIND(frame, trash_truncate_stat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fstat, fd, xdata);
+out:
+    return ret;
+}
 
+/**
+ * The mkdir call is intercepted to avoid creation of
+ * trash directory in the mount by the user
+ */
 int32_t
-trash_ftruncate_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                            int32_t op_ret, int32_t op_errno, fd_t *fd,
-                            inode_t *inode, struct iatt *buf,
-                            struct iatt *preparent, struct iatt *postparent)
+trash_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+            mode_t umask, dict_t *xdata)
 {
-        trash_local_t *local = NULL;
-        char          *tmp_str = NULL;
-        char          *dir_name = NULL;
-        char          *tmp_path = NULL;
-        loc_t          tmp_loc = {0,};
-
-        local = frame->local;
-
-        if ((op_ret == -1) && (op_errno == ENOENT)) {
-                tmp_str = gf_strdup (local->newpath);
-                if (!tmp_str) {
-                        gf_log (this->name, GF_LOG_DEBUG, "out of memory");
-                }
-                dir_name = dirname (tmp_str);
-
-                tmp_path = gf_strdup (dir_name);
-                if (!tmp_path) {
-                        gf_log (this->name, GF_LOG_DEBUG, "out of memory");
-                }
-                tmp_loc.path = tmp_path;
-
-                /* TODO: create the directory with proper permissions */
-                STACK_WIND_COOKIE (frame, trash_truncate_mkdir_cbk,
-                                   tmp_path, FIRST_CHILD(this),
-                                   FIRST_CHILD(this)->fops->mkdir,
-                                   &tmp_loc, 0755, NULL);
-                GF_FREE (tmp_str);
-                return 0;
-        }
+    int32_t op_ret = 0;
+    int32_t op_errno = 0;
+    trash_private_t *priv = NULL;
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO("trash", priv, out);
+
+    if (!check_whether_op_permitted(priv, loc)) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "mkdir issued on %s, which is not permitted",
+               priv->newtrash_dir);
+        op_errno = EPERM;
+        op_ret = -1;
+
+        STACK_UNWIND_STRICT(mkdir, frame, op_ret, op_errno, NULL, NULL, NULL,
+                            NULL, xdata);
+    } else {
+        STACK_WIND(frame, trash_common_mkdir_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->mkdir, loc, mode, umask, xdata);
+    }
 
-        if (op_ret == -1) {
-                STACK_WIND (frame, trash_common_unwind_buf_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->ftruncate,
-                            local->fd, local->fop_offset);
-                return 0;
-        }
+out:
+    return 0;
+}
 
-        STACK_WIND (frame, trash_ftruncate_readv_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->readv, local->fd,
-                    (size_t)GF_BLOCK_READV_SIZE, local->cur_offset);
+/**
+ * The rename call is intercepted to avoid renaming
+ * of trash directory in the mount by the user
+ */
+int
+trash_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+             dict_t *xdata)
+{
+    int32_t op_ret = 0;
+    int32_t op_errno = 0;
+    trash_private_t *priv = NULL;
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO("trash", priv, out);
+
+    if (!check_whether_op_permitted(priv, oldloc)) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "rename issued on %s, which is not permitted",
+               priv->newtrash_dir);
+        op_errno = EPERM;
+        op_ret = -1;
+
+        STACK_UNWIND_STRICT(rename, frame, op_ret, op_errno, NULL, NULL, NULL,
+                            NULL, NULL, xdata);
+    } else {
+        STACK_WIND(frame, trash_common_rename_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
+    }
 
-        return 0;
+out:
+    return 0;
 }
 
-
+/**
+ * The rmdir call is intercepted to avoid deletion of
+ * trash directory in the mount by the user
+ */
 int32_t
-trash_ftruncate_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int32_t op_ret, int32_t op_errno, inode_t *inode,
-                           struct iatt *stbuf, struct iatt *preparent,
-                           struct iatt *postparent)
+trash_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+            dict_t *xdata)
 {
-        trash_local_t       *local = NULL;
-        char                *tmp_str = NULL;
-        char                *tmp_path = NULL;
-        char                *tmp_dirname = NULL;
-        char                *dir_name = NULL;
-        int32_t              count = 0;
-        int32_t              flags = 0;
-        int32_t              loop_count = 0;
-        int                  i = 0;
-        loc_t                tmp_loc = {0,};
-
-        local   = frame->local;
-        if (!local)
-                goto out;
+    int32_t op_ret = 0;
+    int32_t op_errno = 0;
+    trash_private_t *priv = NULL;
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO("trash", priv, out);
+
+    if (!check_whether_op_permitted(priv, loc)) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "rmdir issued on %s, which is not permitted",
+               priv->newtrash_dir);
+        op_errno = EPERM;
+        op_ret = -1;
+
+        STACK_UNWIND_STRICT(rmdir, frame, op_ret, op_errno, NULL, NULL, xdata);
+    } else {
+        STACK_WIND(frame, trash_common_rmdir_cbk, FIRST_CHILD(this),
+                   FIRST_CHILD(this)->fops->rmdir, loc, flags, xdata);
+    }
 
-        loop_count = local->loop_count;
+out:
+    return 0;
+}
 
-        tmp_str = gf_strdup (local->newpath);
-        if (!tmp_str) {
-                gf_log (this->name, GF_LOG_DEBUG, "out of memory");
+/**
+ * Volume set option is handled by the reconfigure function.
+ * Here we checks whether each option is set or not ,if it
+ * sets then corresponding modifciations will be made
+ */
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    uint64_t max_fsize = 0;
+    int ret = 0;
+    char *tmp = NULL;
+    char *tmp_str = NULL;
+    trash_private_t *priv = NULL;
+    char trash_dir[PATH_MAX] = {
+        0,
+    };
+
+    priv = this->private;
+
+    GF_VALIDATE_OR_GOTO("trash", priv, out);
+
+    GF_OPTION_RECONF("trash-internal-op", priv->internal, options, bool, out);
+    GF_OPTION_RECONF("trash-dir", tmp, options, str, out);
+
+    GF_OPTION_RECONF("trash", priv->state, options, bool, out);
+
+    if (priv->state) {
+        ret = create_or_rename_trash_directory(this);
+
+        if (tmp)
+            sprintf(trash_dir, "/%s/", tmp);
+        else
+            sprintf(trash_dir, "%s", priv->oldtrash_dir);
+
+        if (strcmp(priv->newtrash_dir, trash_dir) != 0) {
+            /* When user set a new name for trash directory, trash
+             * xlator will perform a rename operation on old trash
+             * directory to the new one using a STACK_WIND from here.
+             * This option can be configured only when volume is in
+             * started state
+             */
+
+            GF_FREE(priv->newtrash_dir);
+
+            priv->newtrash_dir = gf_strdup(trash_dir);
+            if (!priv->newtrash_dir) {
+                ret = ENOMEM;
+                gf_log(this->name, GF_LOG_DEBUG, "out of memory");
                 goto out;
-        }
-
-        if ((op_ret == -1) && (op_errno == ENOENT)) {
-                tmp_dirname = strchr (tmp_str, '/');
-                while (tmp_dirname) {
-                        count = tmp_dirname - tmp_str;
-                        if (count == 0)
-                                count = 1;
-                        i++;
-                        if (i > loop_count)
-                                break;
-                        tmp_dirname = strchr (tmp_str + count + 1, '/');
-                }
-                tmp_path = memdup (local->newpath, count);
-                if (!tmp_path) {
-                        gf_log (this->name, GF_LOG_DEBUG, "out of memory");
-                }
-                tmp_loc.path = tmp_path;
-                STACK_WIND_COOKIE (frame, trash_ftruncate_mkdir_cbk,
-                                   tmp_path, this->children->xlator,
-                                   this->children->xlator->fops->mkdir,
-                                   &tmp_loc, 0755, NULL);
-
+            }
+            gf_log(this->name, GF_LOG_DEBUG,
+                   "Renaming %s -> %s from reconfigure", priv->oldtrash_dir,
+                   priv->newtrash_dir);
+
+            if (!priv->newtrash_dir) {
+                gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+                ret = ENOMEM;
                 goto out;
+            }
+            ret = rename_trash_directory(this);
+        }
+
+        if (priv->internal) {
+            ret = create_internalop_directory(this);
+        }
+    }
+    tmp = NULL;
+
+    GF_OPTION_RECONF("trash-max-filesize", max_fsize, options, size_uint64,
+                     out);
+    if (max_fsize) {
+        priv->max_trash_file_size = max_fsize;
+        gf_log(this->name, GF_LOG_DEBUG, "%" GF_PRI_SIZET " max-size",
+               priv->max_trash_file_size);
+    }
+    GF_OPTION_RECONF("trash-eliminate-path", tmp, options, str, out);
+    if (!tmp) {
+        gf_log(this->name, GF_LOG_DEBUG,
+               "no option specified for 'eliminate', using NULL");
+    } else {
+        if (priv->eliminate)
+            wipe_eliminate_path(&priv->eliminate);
+
+        tmp_str = gf_strdup(tmp);
+        if (!tmp_str) {
+            gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+            ret = ENOMEM;
+            goto out;
         }
+        ret = store_eliminate_path(tmp_str, &priv->eliminate);
+    }
 
-        if (op_ret == 0) {
-                dir_name = dirname (tmp_str);
-                if (strcmp ((char*)cookie, dir_name) == 0) {
-                        ia_prot_t prot = {0, };
-                        flags = O_CREAT|O_EXCL|O_WRONLY;
-
-                        //Call create again once directory structure is created.
-                        STACK_WIND (frame, trash_ftruncate_create_cbk,
-                                    FIRST_CHILD(this),
-                                    FIRST_CHILD(this)->fops->create,
-                                    &local->newloc, flags,
-                                    st_mode_from_ia (prot, local->loc.inode->ia_type),
-                                    local->newfd, NULL);
-                        goto out;
-                }
-        }
+out:
 
-        LOCK (&frame->lock);
-        {
-                loop_count = ++local->loop_count;
-        }
-        UNLOCK (&frame->lock);
-        tmp_dirname = strchr (tmp_str, '/');
-        while (tmp_dirname) {
-                count = tmp_dirname - tmp_str;
-                if (count == 0)
-                        count = 1;
-
-                i++;
-                if ((i > loop_count) || (count > PATH_MAX))
-                        break;
-                tmp_dirname = strchr (tmp_str + count + 1, '/');
-        }
-        tmp_path = memdup (local->newpath, count);
-        if (!tmp_path) {
-                gf_log (this->name, GF_LOG_DEBUG, "out of memory");
-        }
-        tmp_loc.path = tmp_path;
+    return ret;
+}
 
-        STACK_WIND_COOKIE (frame, trash_ftruncate_mkdir_cbk, tmp_path,
-                           this->children->xlator,
-                           this->children->xlator->fops->mkdir,
-                           &tmp_loc, 0755, NULL);
+/**
+ * Notify is used to create the trash directory with fixed gfid
+ * using STACK_WIND only when posix xlator is up
+ */
+int
+notify(xlator_t *this, int event, void *data, ...)
+{
+    trash_private_t *priv = NULL;
+    int ret = 0;
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO("trash", priv, out);
+
+    /* Check whether posix is up not */
+    if (event == GF_EVENT_CHILD_UP) {
+        if (!priv->state) {
+            gf_log(this->name, GF_LOG_DEBUG, "trash xlator is off");
+            goto out;
+        }
+
+        /* Here there is two possibilities ,if trash directory already
+         * exist ,then we need to perform a rename operation on the
+         * old one. Otherwise, we need to create the trash directory
+         * For both, we need to pass location variable, gfid of parent
+         * and a frame for calling STACK_WIND.The location variable
+         * requires name,path,gfid and inode
+         */
+        if (!priv->oldtrash_dir)
+            ret = create_or_rename_trash_directory(this);
+        else if (strcmp(priv->newtrash_dir, priv->oldtrash_dir) != 0)
+            ret = rename_trash_directory(this);
+        if (ret)
+            goto out;
+
+        if (priv->internal)
+            (void)create_internalop_directory(this);
+    }
 
 out:
-        GF_FREE (cookie); /* strdup (dir_name) was sent here :) */
-        if (tmp_str)
-                GF_FREE (tmp_str);
-
-        return 0;
+    ret = default_notify(this, event, data);
+    if (ret)
+        gf_log(this->name, GF_LOG_INFO, "default notify event failed");
+    return ret;
 }
 
-
 int32_t
-trash_ftruncate_fstat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
-                           int32_t op_ret, int32_t op_errno, struct iatt *buf)
+mem_acct_init(xlator_t *this)
 {
-        trash_private_t *priv  = NULL;
-        trash_local_t   *local = NULL;
-
-        priv  = this->private;
-        local = frame->local;
-
-        if (op_ret == -1) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "%s: %s",local->newloc.path, strerror(op_errno));
-
-                TRASH_STACK_UNWIND (ftruncate, frame, -1, op_errno, buf, NULL);
-                return 0;
-        }
-        if ((buf->ia_size == 0) || (buf->ia_size > priv->max_trash_file_size))
-        {
-                STACK_WIND (frame, trash_common_unwind_buf_cbk,
-                            this->children->xlator,
-                            this->children->xlator->fops->ftruncate,
-                            local->fd, local->fop_offset);
-                return 0;
-        }
-
+    int ret = -1;
 
-        STACK_WIND (frame, trash_ftruncate_create_cbk, FIRST_CHILD(this),
-                    FIRST_CHILD(this)->fops->create, &local->newloc,
-                    ( O_CREAT | O_EXCL | O_WRONLY ),
-                    st_mode_from_ia (buf->ia_prot, local->loc.inode->ia_type),
-                    local->newfd, NULL);
+    GF_VALIDATE_OR_GOTO("trash", this, out);
 
-        return 0;
+    ret = xlator_mem_acct_init(this, gf_trash_mt_end + 1);
+    if (ret != 0) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "Memory accounting init"
+               "failed");
+        return ret;
+    }
+out:
+    return ret;
 }
 
+/**
+ * trash_init
+ */
 int32_t
-trash_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset)
+init(xlator_t *this)
 {
-        trash_elim_pattern_t  *trav = NULL;
-        trash_private_t       *priv = NULL;
-        trash_local_t         *local = NULL;
-        dentry_t              *dir_entry = NULL;
-        struct tm             *tm = NULL;
-        char                  *pathbuf = NULL;
-        inode_t               *newinode = NULL;
-        time_t                 utime = 0;
-        char                   timestr[256];
-        int32_t                retval = 0;
-        int32_t                match = 0;
-
-        priv = this->private;
-
-        dir_entry = __dentry_search_arbit (fd->inode);
-        retval = inode_path (fd->inode, NULL, &pathbuf);
-
-        if (priv->eliminate) {
-                trav = priv->eliminate;
-                while (trav) {
-                        if (fnmatch(trav->pattern, dir_entry->name, 0) == 0) {
-                                match++;
-                                break;
-                        }
-                        trav = trav->next;
-                }
-        }
-
-        if ((strncmp (pathbuf, priv->trash_dir,
-                      strlen (priv->trash_dir)) == 0) ||
-            (offset >= priv->max_trash_file_size) ||
-            (!retval) ||
-            match) {
-                STACK_WIND (frame, trash_common_unwind_buf_cbk,
-                            FIRST_CHILD(this),
-                            FIRST_CHILD(this)->fops->ftruncate,
-                            fd, offset);
-                return 0;
-        }
-
-        local = GF_CALLOC (1, sizeof (trash_local_t), 
-                           gf_trash_mt_trash_local_t);
-        if (!local) {
-                gf_log (this->name, GF_LOG_DEBUG, "out of memory");
-                TRASH_STACK_UNWIND (ftruncate, frame, -1, ENOMEM, NULL, NULL);
-                return 0;
-        }
-
-        utime = time (NULL);
-        tm    = localtime (&utime);
-        strftime (timestr, 256, ".%Y-%m-%d-%H%M%S", tm);
-
-        strcpy (local->newpath, priv->trash_dir);
-        strcat (local->newpath, pathbuf);
-        strcat (local->newpath, timestr);
-
-        local->fd = fd_ref (fd);
-        newinode = inode_new (fd->inode->table);
-        local->newfd = fd_create (newinode, frame->root->pid);
-        frame->local=local;
-
-        local->newloc.inode = newinode;
-        local->newloc.path = local->newpath;
-
-        local->loc.inode = inode_ref (fd->inode);
-        local->loc.ino   = fd->inode->ino;
-        local->loc.path  = pathbuf;
-
-        local->fop_offset = offset;
-        local->cur_offset = offset;
-
-        STACK_WIND (frame, trash_ftruncate_fstat_cbk, this->children->xlator,
-                    this->children->xlator->fops->fstat, fd);
+    trash_private_t *priv = NULL;
+    int ret = -1;
+    char *tmp = NULL;
+    char *tmp_str = NULL;
+    char trash_dir[PATH_MAX] = {
+        0,
+    };
+    uint64_t max_trash_file_size64 = 0;
+    data_t *data = NULL;
+
+    GF_VALIDATE_OR_GOTO("trash", this, out);
+
+    if (!this->children || this->children->next) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "not configured with exactly one child. exiting");
+        ret = -1;
+        goto out;
+    }
+
+    if (!this->parents) {
+        gf_log(this->name, GF_LOG_WARNING, "dangling volume. check volfile");
+    }
+
+    priv = GF_CALLOC(1, sizeof(*priv), gf_trash_mt_trash_private_t);
+    if (!priv) {
+        gf_log(this->name, GF_LOG_ERROR, "out of memory");
+        ret = ENOMEM;
+        goto out;
+    }
+
+    /* Trash priv data members are initialized through the following
+     * set of statements
+     */
+    GF_OPTION_INIT("trash", priv->state, bool, out);
+
+    GF_OPTION_INIT("trash-dir", tmp, str, out);
+
+    /* We store trash dir value as path for easier manipulation*/
+    if (!tmp) {
+        gf_log(this->name, GF_LOG_INFO,
+               "no option specified for 'trash-dir', "
+               "using \"/.trashcan/\"");
+        priv->newtrash_dir = gf_strdup("/.trashcan/");
+        if (!priv->newtrash_dir) {
+            ret = ENOMEM;
+            gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+            goto out;
+        }
+    } else {
+        sprintf(trash_dir, "/%s/", tmp);
+        priv->newtrash_dir = gf_strdup(trash_dir);
+        if (!priv->newtrash_dir) {
+            ret = ENOMEM;
+            gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+            goto out;
+        }
+    }
+    tmp = NULL;
+
+    GF_OPTION_INIT("trash-eliminate-path", tmp, str, out);
+    if (!tmp) {
+        gf_log(this->name, GF_LOG_INFO,
+               "no option specified for 'eliminate', using NULL");
+    } else {
+        tmp_str = gf_strdup(tmp);
+        if (!tmp_str) {
+            gf_log(this->name, GF_LOG_ERROR, "out of memory");
+            ret = ENOMEM;
+            goto out;
+        }
+        ret = store_eliminate_path(tmp_str, &priv->eliminate);
+    }
+    tmp = NULL;
+
+    GF_OPTION_INIT("trash-max-filesize", max_trash_file_size64, size_uint64,
+                   out);
+    if (!max_trash_file_size64) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "no option specified for 'max-trashable-file-size', "
+               "using default = %lld MB",
+               GF_DEFAULT_MAX_FILE_SIZE / GF_UNIT_MB);
+        priv->max_trash_file_size = GF_DEFAULT_MAX_FILE_SIZE;
+    } else {
+        priv->max_trash_file_size = max_trash_file_size64;
+        gf_log(this->name, GF_LOG_DEBUG, "%" GF_PRI_SIZET " max-size",
+               priv->max_trash_file_size);
+    }
+
+    GF_OPTION_INIT("trash-internal-op", priv->internal, bool, out);
+
+    this->local_pool = mem_pool_new(trash_local_t, 64);
+    if (!this->local_pool) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "failed to create local_t's memory pool");
+        ret = ENOMEM;
+        goto out;
+    }
+
+    /* For creating directories inside trash with proper permissions,
+     * we need to perform stat on that directories, for this we use
+     * brick path
+     */
+    data = dict_get(this->options, "brick-path");
+    if (!data) {
+        gf_log(this->name, GF_LOG_ERROR,
+               "no option specified for 'brick-path'");
+        ret = ENOMEM;
+        goto out;
+    }
+    priv->brick_path = gf_strdup(data->data);
+    if (!priv->brick_path) {
+        ret = ENOMEM;
+        gf_log(this->name, GF_LOG_DEBUG, "out of memory");
+        goto out;
+    }
+
+    priv->trash_itable = inode_table_new(0, this);
+    gf_log(this->name, GF_LOG_DEBUG, "brick path is%s", priv->brick_path);
+
+    this->private = (void *)priv;
+    ret = 0;
 
-        return 0;
+out:
+    if (tmp_str)
+        GF_FREE(tmp_str);
+    if (ret) {
+        if (priv) {
+            if (priv->newtrash_dir)
+                GF_FREE(priv->newtrash_dir);
+            if (priv->oldtrash_dir)
+                GF_FREE(priv->oldtrash_dir);
+            if (priv->brick_path)
+                GF_FREE(priv->brick_path);
+            if (priv->eliminate)
+                wipe_eliminate_path(&priv->eliminate);
+            GF_FREE(priv);
+        }
+        mem_pool_destroy(this->local_pool);
+        this->local_pool = NULL;
+    }
+    return ret;
 }
 
 /**
- * trash_init -
+ * trash_fini
  */
-int32_t
-init (xlator_t *this)
+void
+fini(xlator_t *this)
 {
-        data_t                *data  = NULL;
-        trash_private_t       *_priv = NULL;
-        trash_elim_pattern_t  *trav  = NULL;
-        char                  *tmp_str = NULL;
-        char                  *strtokptr = NULL;
-        char                  *component = NULL;
-        char                   trash_dir[PATH_MAX] = {0,};
-        uint64_t               max_trash_file_size64 = 0;
-
-        /* Create .trashcan directory in init */
-        if (!this->children || this->children->next) {
-                gf_log (this->name, GF_LOG_ERROR,
-                        "not configured with exactly one child. exiting");
-                return -1;
-        }
+    trash_private_t *priv = NULL;
+    inode_table_t *inode_table = NULL;
 
-        if (!this->parents) {
-                gf_log (this->name, GF_LOG_WARNING,
-                        "dangling volume. check volfile ");
+    GF_VALIDATE_OR_GOTO("trash", this, out);
+    priv = this->private;
+    if (priv) {
+        inode_table = priv->trash_itable;
+        if (priv->newtrash_dir) {
+            GF_FREE(priv->newtrash_dir);
+            priv->newtrash_dir = NULL;
         }
-
-        _priv = GF_CALLOC (1, sizeof (*_priv), gf_trash_mt_trash_private_t);
-        if (!_priv) {
-                gf_log (this->name, GF_LOG_ERROR, "out of memory");
-                return -1;
+        if (priv->oldtrash_dir) {
+            GF_FREE(priv->oldtrash_dir);
+            priv->oldtrash_dir = NULL;
         }
-
-        data = dict_get (this->options, "trash-dir");
-        if (!data) {
-                gf_log (this->name, GF_LOG_NORMAL,
-                        "no option specified for 'trash-dir', "
-                        "using \"/.trashcan/\"");
-                _priv->trash_dir = gf_strdup ("/.trashcan");
-        } else {
-                /* Need a path with '/' as the first char, if not
-                   given, append it */
-                if (data->data[0] == '/') {
-                        _priv->trash_dir = gf_strdup (data->data);
-                } else {
-                        /* TODO: Make sure there is no ".." in the path */
-                        strcpy (trash_dir, "/");
-                        strcat (trash_dir, data->data);
-                        _priv->trash_dir = gf_strdup (trash_dir);
-                }
+        if (priv->brick_path) {
+            GF_FREE(priv->brick_path);
+            priv->brick_path = NULL;
         }
-
-        data = dict_get (this->options, "eliminate-pattern");
-        if (!data) {
-                gf_log (this->name, GF_LOG_TRACE,
-                        "no option specified for 'eliminate', using NULL");
-        } else {
-                tmp_str = gf_strdup (data->data);
-                if (!tmp_str) {
-                        gf_log (this->name, GF_LOG_DEBUG, "out of memory");
-                }
-
-                /*  Match Filename to option specified in eliminate. */
-                component = strtok_r (tmp_str, "|", &strtokptr);
-                while (component) {
-                        trav = GF_CALLOC (1, sizeof (*trav),
-                                          gf_trash_mt_trash_elim_pattern_t);
-                        if (!trav) {
-                                gf_log (this->name, GF_LOG_DEBUG, "out of memory");
-                                break;
-                        }
-                        trav->pattern = component;
-                        trav->next = _priv->eliminate;
-                        _priv->eliminate = trav;
-
-                        component = strtok_r (NULL, "|", &strtokptr);
-                }
+        if (priv->eliminate) {
+            wipe_eliminate_path(&priv->eliminate);
+            priv->eliminate = NULL;
         }
-
-        /* TODO: do gf_string2sizet () */
-        data = dict_get (this->options, "max-trashable-file-size");
-        if (!data) {
-                gf_log (this->name, GF_LOG_DEBUG,
-                        "no option specified for 'max-trashable-file-size', "
-                        "using default = %lld MB",
-                        GF_DEFAULT_MAX_FILE_SIZE / GF_UNIT_MB);
-                _priv->max_trash_file_size = GF_DEFAULT_MAX_FILE_SIZE;
-        } else {
-                (void)gf_string2bytesize (data->data,
-                                          &max_trash_file_size64);
-                if( max_trash_file_size64 > GF_ALLOWED_MAX_FILE_SIZE ) {
-                        gf_log (this->name, GF_LOG_DEBUG,
-                                "Size specified for max-size(in MB) is too "
-                                "large so using 1GB as max-size (NOT IDEAL)");
-                        _priv->max_trash_file_size = GF_ALLOWED_MAX_FILE_SIZE;
-                } else
-                        _priv->max_trash_file_size = max_trash_file_size64;
-                gf_log (this->name, GF_LOG_DEBUG, "%"GF_PRI_SIZET" max-size",
-                        _priv->max_trash_file_size);
+        if (inode_table) {
+            inode_table_destroy(inode_table);
+            priv->trash_itable = NULL;
         }
+        GF_FREE(priv);
+    }
 
-        this->private = (void *)_priv;
-        return 0;
-}
-
-void
-fini (xlator_t *this)
-{
-        trash_private_t *priv = NULL;
-
-        priv = this->private;
-        if (priv)
-                GF_FREE (priv);
-
-        return;
+    if (this->local_pool) {
+        mem_pool_destroy(this->local_pool);
+        this->local_pool = NULL;
+    }
+    this->private = NULL;
+out:
+    return;
 }
 
 struct xlator_fops fops = {
-        .unlink    = trash_unlink,
-        .rename    = trash_rename,
-        .truncate  = trash_truncate,
-        .ftruncate = trash_ftruncate,
+    .unlink = trash_unlink,
+    .truncate = trash_truncate,
+    .ftruncate = trash_ftruncate,
+    .rmdir = trash_rmdir,
+    .mkdir = trash_mkdir,
+    .rename = trash_rename,
 };
 
-struct xlator_cbks cbks = {
-};
+struct xlator_cbks cbks = {};
 
 struct volume_options options[] = {
-        { .key  = { "trash-directory" },
-          .type = GF_OPTION_TYPE_PATH,
-        },
-        { .key  = { "eliminate-pattern" },
-          .type = GF_OPTION_TYPE_STR,
-        },
-        { .key  = { "max-trashable-file-size" },
-          .type = GF_OPTION_TYPE_SIZET,
-        },
-        { .key  = {NULL} },
+    {
+        .key = {"trash"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "off",
+        .description = "Enable/disable trash translator",
+        .op_version = {GD_OP_VERSION_3_7_0},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"backup"},
+    },
+    {
+        .key = {"trash-dir"},
+        .type = GF_OPTION_TYPE_STR,
+        .default_value = ".trashcan",
+        .description = "Directory for trash files",
+        .op_version = {GD_OP_VERSION_3_7_0},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"backup"},
+    },
+    {
+        .key = {"trash-eliminate-path"},
+        .type = GF_OPTION_TYPE_STR,
+        .description = "Eliminate paths to be excluded "
+                       "from trashing",
+        .op_version = {GD_OP_VERSION_3_7_0},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"backup"},
+    },
+    {
+        .key = {"trash-max-filesize"},
+        .type = GF_OPTION_TYPE_SIZET,
+        .default_value = "5MB",
+        .description = "Maximum size of file that can be "
+                       "moved to trash",
+        .op_version = {GD_OP_VERSION_3_7_0},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"backup"},
+    },
+    {
+        .key = {"trash-internal-op"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "off",
+        .description = "Enable/disable trash translator for "
+                       "internal operations",
+        .op_version = {GD_OP_VERSION_3_7_0},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"backup"},
+    },
+    {.key = {"brick-path"},
+     .type = GF_OPTION_TYPE_PATH,
+     .default_value = "{{ brick.path }}"},
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "trash",
+    .category = GF_TECH_PREVIEW,
 };
diff --git a/xlators/features/trash/src/trash.h b/xlators/features/trash/src/trash.h
index 65c0ed85712..6671617c2c6 100644
--- a/xlators/features/trash/src/trash.h
+++ b/xlators/features/trash/src/trash.h
@@ -1,89 +1,97 @@
 /*
-  Copyright (c) 2006-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
+   Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
 */
-
 #ifndef __TRASH_H__
 #define __TRASH_H__
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "logging.h"
-#include "dict.h"
-#include "xlator.h"
-#include "defaults.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/dict.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
 #include "inode.c"
 #include "fnmatch.h"
 
 #include <libgen.h>
 
 #ifndef GF_BLOCK_READV_SIZE
-#define GF_BLOCK_READV_SIZE      (128 * GF_UNIT_KB)
+#define GF_BLOCK_READV_SIZE (128 * GF_UNIT_KB)
 #endif
 
 #ifndef GF_DEFAULT_MAX_FILE_SIZE
 #define GF_DEFAULT_MAX_FILE_SIZE (200 * GF_UNIT_MB)
 #endif
 
-#ifndef GF_ALLOWED_MAX_FILE_SIZE
-#define GF_ALLOWED_MAX_FILE_SIZE (1 * GF_UNIT_GB)
-#endif
-
-
 struct trash_struct {
-        fd_t    *fd;         /* for the fd of existing file */
-        fd_t    *newfd;      /* for the newly created file */
-        loc_t    loc;        /* to store the location of the existing file */
-        loc_t    newloc;     /* to store the location for the new file */
-        size_t   fsize;      /* for keeping the size of existing file */
-        off_t    cur_offset; /* current offset for read and write ops */
-        off_t    fop_offset;
-        char     origpath[PATH_MAX];
-        char     newpath[PATH_MAX];
-        int32_t  loop_count;
-        struct iatt preparent;
-        struct iatt postparent;
+    fd_t *fd;         /* for the fd of existing file */
+    fd_t *newfd;      /* for the newly created file */
+    loc_t loc;        /* to store the location of the existing file */
+    loc_t newloc;     /* to store the location for the new file */
+    size_t fsize;     /* for keeping the size of existing file */
+    off_t cur_offset; /* current offset for read and write ops */
+    off_t fop_offset; /* original offset received with the fop */
+    pid_t pid;
+    char origpath[PATH_MAX];
+    char newpath[PATH_MAX];
+    int32_t loop_count;
+    gf_boolean_t is_set_pid;
+    struct iatt preparent;
+    struct iatt postparent;
+    gf_boolean_t ctr_link_count_req;
 };
 typedef struct trash_struct trash_local_t;
 
-struct _trash_elim_pattern;
-typedef struct _trash_elim_pattern {
-        struct _trash_elim_pattern *next;
-        char                       *pattern;
-} trash_elim_pattern_t;
+struct _trash_elim_path {
+    struct _trash_elim_path *next;
+    char *path;
+};
+typedef struct _trash_elim_path trash_elim_path;
 
 struct trash_priv {
-        char                 *trash_dir;
-        trash_elim_pattern_t *eliminate;
-        size_t                max_trash_file_size;
+    char *oldtrash_dir;
+    char *newtrash_dir;
+    char *brick_path;
+    trash_elim_path *eliminate;
+    size_t max_trash_file_size;
+    gf_boolean_t state;
+    gf_boolean_t internal;
+    inode_t *trash_inode;
+    inode_table_t *trash_itable;
 };
 typedef struct trash_priv trash_private_t;
 
-#define TRASH_STACK_UNWIND(op, frame, params ...) do {     \
-		trash_local_t *__local = NULL;         \
-		__local = frame->local;                \
-		frame->local = NULL;		       \
-		STACK_UNWIND_STRICT (op, frame, params);          \
-		trash_local_wipe (__local);	       \
-	} while (0)
-
+#define TRASH_SET_PID(frame, local)                                            \
+    do {                                                                       \
+        GF_ASSERT(!local->is_set_pid);                                         \
+        if (!local->is_set_pid) {                                              \
+            local->pid = frame->root->pid;                                     \
+            frame->root->pid = GF_SERVER_PID_TRASH;                            \
+            local->is_set_pid = _gf_true;                                      \
+        }                                                                      \
+    } while (0)
+
+#define TRASH_UNSET_PID(frame, local)                                          \
+    do {                                                                       \
+        GF_ASSERT(local->is_set_pid);                                          \
+        if (local->is_set_pid) {                                               \
+            frame->root->pid = local->pid;                                     \
+            local->is_set_pid = _gf_false;                                     \
+        }                                                                      \
+    } while (0)
+
+#define TRASH_STACK_UNWIND(op, frame, params...)                               \
+    do {                                                                       \
+        trash_local_t *__local = NULL;                                         \
+        __local = frame->local;                                                \
+        frame->local = NULL;                                                   \
+        STACK_UNWIND_STRICT(op, frame, params);                                \
+        trash_local_wipe(__local);                                             \
+    } while (0)
 
 #endif /* __TRASH_H__ */
diff --git a/xlators/features/upcall/Makefile.am b/xlators/features/upcall/Makefile.am
new file mode 100644
index 00000000000..a985f42a877
--- /dev/null
+++ b/xlators/features/upcall/Makefile.am
@@ -0,0 +1,3 @@
+SUBDIRS = src
+
+CLEANFILES =
diff --git a/xlators/features/upcall/src/Makefile.am b/xlators/features/upcall/src/Makefile.am
new file mode 100644
index 00000000000..72b7f55ae0a
--- /dev/null
+++ b/xlators/features/upcall/src/Makefile.am
@@ -0,0 +1,23 @@
+if WITH_SERVER
+xlator_LTLIBRARIES = upcall.la
+endif
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+upcall_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+
+upcall_la_SOURCES = upcall.c upcall-internal.c
+
+upcall_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la \
+	$(top_builddir)/rpc/rpc-lib/src/libgfrpc.la \
+	$(top_builddir)/rpc/xdr/src/libgfxdr.la
+
+noinst_HEADERS = upcall.h upcall-mem-types.h upcall-messages.h \
+	upcall-cache-invalidation.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/rpc-lib/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+
+AM_CFLAGS = -Wall -fno-strict-aliasing $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/features/upcall/src/upcall-cache-invalidation.h b/xlators/features/upcall/src/upcall-cache-invalidation.h
new file mode 100644
index 00000000000..db649b2c9a6
--- /dev/null
+++ b/xlators/features/upcall/src/upcall-cache-invalidation.h
@@ -0,0 +1,18 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __UPCALL_CACHE_INVALIDATION_H__
+#define __UPCALL_CACHE_INVALIDATION_H__
+
+/* The time period for which a client will be notified of cache_invalidation
+ * events post its last access */
+#define CACHE_INVALIDATION_TIMEOUT "60"
+
+#endif /* __UPCALL_CACHE_INVALIDATION_H__ */
diff --git a/xlators/features/upcall/src/upcall-internal.c b/xlators/features/upcall/src/upcall-internal.c
new file mode 100644
index 00000000000..c641bd6f432
--- /dev/null
+++ b/xlators/features/upcall/src/upcall-internal.c
@@ -0,0 +1,689 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <unistd.h>
+#include <fcntl.h>
+#include <limits.h>
+
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/common-utils.h>
+
+#include <glusterfs/statedump.h>
+#include <glusterfs/syncop.h>
+
+#include "upcall.h"
+#include "upcall-mem-types.h"
+#include "glusterfs3-xdr.h"
+#include "protocol-common.h"
+#include <glusterfs/defaults.h>
+
+/*
+ * Check if any of the upcall options are enabled:
+ *     - cache_invalidation
+ */
+gf_boolean_t
+is_upcall_enabled(xlator_t *this)
+{
+    upcall_private_t *priv = NULL;
+
+    if (this->private) {
+        priv = (upcall_private_t *)this->private;
+        return priv->cache_invalidation_enabled;
+    }
+
+    return _gf_false;
+}
+
+/*
+ * Get the cache_invalidation_timeout
+ */
+static int32_t
+get_cache_invalidation_timeout(xlator_t *this)
+{
+    upcall_private_t *priv = NULL;
+
+    if (this->private) {
+        priv = (upcall_private_t *)this->private;
+        return priv->cache_invalidation_timeout;
+    }
+
+    return 0;
+}
+
+static upcall_client_t *
+__add_upcall_client(call_frame_t *frame, client_t *client,
+                    upcall_inode_ctx_t *up_inode_ctx, time_t now)
+{
+    upcall_client_t *up_client_entry = GF_MALLOC(
+        sizeof(*up_client_entry), gf_upcall_mt_upcall_client_entry_t);
+    if (!up_client_entry) {
+        gf_msg("upcall", GF_LOG_WARNING, 0, UPCALL_MSG_NO_MEMORY,
+               "Memory allocation failed");
+        return NULL;
+    }
+    INIT_LIST_HEAD(&up_client_entry->client_list);
+    up_client_entry->client_uid = gf_strdup(client->client_uid);
+    up_client_entry->access_time = now;
+    up_client_entry->expire_time_attr = get_cache_invalidation_timeout(
+        frame->this);
+
+    list_add_tail(&up_client_entry->client_list, &up_inode_ctx->client_list);
+
+    gf_log(THIS->name, GF_LOG_DEBUG, "upcall_entry_t client added - %s",
+           up_client_entry->client_uid);
+
+    return up_client_entry;
+}
+
+static int
+__upcall_inode_ctx_set(inode_t *inode, xlator_t *this)
+{
+    upcall_inode_ctx_t *inode_ctx = NULL;
+    upcall_private_t *priv = NULL;
+    int ret = -1;
+    uint64_t ctx = 0;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    ret = __inode_ctx_get(inode, this, &ctx);
+
+    if (!ret)
+        goto out;
+
+    inode_ctx = GF_MALLOC(sizeof(upcall_inode_ctx_t),
+                          gf_upcall_mt_upcall_inode_ctx_t);
+
+    if (!inode_ctx) {
+        ret = -ENOMEM;
+        goto out;
+    }
+
+    pthread_mutex_init(&inode_ctx->client_list_lock, NULL);
+    INIT_LIST_HEAD(&inode_ctx->inode_ctx_list);
+    INIT_LIST_HEAD(&inode_ctx->client_list);
+    inode_ctx->destroy = 0;
+    gf_uuid_copy(inode_ctx->gfid, inode->gfid);
+
+    ctx = (long)inode_ctx;
+    ret = __inode_ctx_set(inode, this, &ctx);
+    if (ret) {
+        gf_log(this->name, GF_LOG_DEBUG, "failed to set inode ctx (%p)", inode);
+        GF_FREE(inode_ctx);
+        goto out;
+    }
+
+    /* add this inode_ctx to the global list */
+    LOCK(&priv->inode_ctx_lk);
+    {
+        list_add_tail(&inode_ctx->inode_ctx_list, &priv->inode_ctx_list);
+    }
+    UNLOCK(&priv->inode_ctx_lk);
+out:
+    return ret;
+}
+
+static upcall_inode_ctx_t *
+__upcall_inode_ctx_get(inode_t *inode, xlator_t *this)
+{
+    upcall_inode_ctx_t *inode_ctx = NULL;
+    uint64_t ctx = 0;
+    int ret = 0;
+
+    ret = __inode_ctx_get(inode, this, &ctx);
+
+    if (ret < 0) {
+        ret = __upcall_inode_ctx_set(inode, this);
+        if (ret < 0)
+            goto out;
+
+        ret = __inode_ctx_get(inode, this, &ctx);
+        if (ret < 0)
+            goto out;
+    }
+
+    inode_ctx = (upcall_inode_ctx_t *)(long)(ctx);
+
+out:
+    return inode_ctx;
+}
+
+upcall_inode_ctx_t *
+upcall_inode_ctx_get(inode_t *inode, xlator_t *this)
+{
+    upcall_inode_ctx_t *inode_ctx = NULL;
+
+    LOCK(&inode->lock);
+    {
+        inode_ctx = __upcall_inode_ctx_get(inode, this);
+    }
+    UNLOCK(&inode->lock);
+
+    return inode_ctx;
+}
+
+static int
+__upcall_cleanup_client_entry(upcall_client_t *up_client)
+{
+    list_del_init(&up_client->client_list);
+
+    GF_FREE(up_client->client_uid);
+    GF_FREE(up_client);
+
+    return 0;
+}
+
+static int
+upcall_cleanup_expired_clients(xlator_t *this, upcall_inode_ctx_t *up_inode_ctx,
+                               time_t now)
+{
+    upcall_client_t *up_client = NULL;
+    upcall_client_t *tmp = NULL;
+    int ret = -1;
+    time_t timeout = 0;
+    time_t t_expired = 0;
+
+    timeout = get_cache_invalidation_timeout(this);
+
+    pthread_mutex_lock(&up_inode_ctx->client_list_lock);
+    {
+        list_for_each_entry_safe(up_client, tmp, &up_inode_ctx->client_list,
+                                 client_list)
+        {
+            t_expired = now - up_client->access_time;
+
+            if (t_expired > (2 * timeout)) {
+                gf_log(THIS->name, GF_LOG_TRACE, "Cleaning up client_entry(%s)",
+                       up_client->client_uid);
+
+                ret = __upcall_cleanup_client_entry(up_client);
+
+                if (ret) {
+                    gf_msg("upcall", GF_LOG_WARNING, 0,
+                           UPCALL_MSG_INTERNAL_ERROR,
+                           "Client entry cleanup failed (%p)", up_client);
+                    goto out;
+                }
+            }
+        }
+    }
+    pthread_mutex_unlock(&up_inode_ctx->client_list_lock);
+
+    ret = 0;
+out:
+    return ret;
+}
+
+/*
+ * Free Upcall inode_ctx client list
+ */
+int
+__upcall_cleanup_inode_ctx_client_list(upcall_inode_ctx_t *inode_ctx)
+{
+    upcall_client_t *up_client = NULL;
+    upcall_client_t *tmp = NULL;
+
+    list_for_each_entry_safe(up_client, tmp, &inode_ctx->client_list,
+                             client_list)
+    {
+        __upcall_cleanup_client_entry(up_client);
+    }
+
+    return 0;
+}
+
+static void
+upcall_cache_forget(xlator_t *this, inode_t *inode,
+                    upcall_inode_ctx_t *up_inode_ctx);
+
+/*
+ * Free upcall_inode_ctx
+ */
+int
+upcall_cleanup_inode_ctx(xlator_t *this, inode_t *inode)
+{
+    uint64_t ctx = 0;
+    upcall_inode_ctx_t *inode_ctx = NULL;
+    int ret = 0;
+    upcall_private_t *priv = NULL;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    ret = inode_ctx_del(inode, this, &ctx);
+
+    if (ret < 0) {
+        gf_msg("upcall", GF_LOG_WARNING, 0, UPCALL_MSG_INTERNAL_ERROR,
+               "Failed to del upcall_inode_ctx (%p)", inode);
+        goto out;
+    }
+
+    inode_ctx = (upcall_inode_ctx_t *)(long)ctx;
+
+    if (inode_ctx) {
+        /* Invalidate all the upcall cache entries */
+        upcall_cache_forget(this, inode, inode_ctx);
+
+        /* do we really need lock? yes now reaper thread
+         * may also be trying to cleanup the client entries.
+         */
+        pthread_mutex_lock(&inode_ctx->client_list_lock);
+        {
+            if (!list_empty(&inode_ctx->client_list)) {
+                __upcall_cleanup_inode_ctx_client_list(inode_ctx);
+            }
+        }
+        pthread_mutex_unlock(&inode_ctx->client_list_lock);
+
+        /* Mark the inode_ctx to be destroyed */
+        inode_ctx->destroy = 1;
+        gf_msg_debug("upcall", 0, "set upcall_inode_ctx (%p) to destroy mode",
+                     inode_ctx);
+    }
+
+out:
+    return ret;
+}
+
+/*
+ * Traverse through the list of upcall_inode_ctx(s),
+ * cleanup the expired client entries and destroy the ctx
+ * which is no longer valid and has destroy bit set.
+ */
+void *
+upcall_reaper_thread(void *data)
+{
+    upcall_private_t *priv = NULL;
+    upcall_inode_ctx_t *inode_ctx = NULL;
+    upcall_inode_ctx_t *tmp = NULL;
+    xlator_t *this = NULL;
+    time_t timeout = 0;
+    time_t time_now;
+
+    this = (xlator_t *)data;
+    GF_ASSERT(this);
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    time_now = gf_time();
+    while (!priv->fini) {
+        list_for_each_entry_safe(inode_ctx, tmp, &priv->inode_ctx_list,
+                                 inode_ctx_list)
+        {
+            /* cleanup expired clients */
+            upcall_cleanup_expired_clients(this, inode_ctx, time_now);
+
+            if (!inode_ctx->destroy) {
+                continue;
+            }
+
+            /* client list would have been cleaned up*/
+            gf_msg_debug("upcall", 0, "Freeing upcall_inode_ctx (%p)",
+                         inode_ctx);
+            LOCK(&priv->inode_ctx_lk);
+            {
+                list_del_init(&inode_ctx->inode_ctx_list);
+                pthread_mutex_destroy(&inode_ctx->client_list_lock);
+            }
+            UNLOCK(&priv->inode_ctx_lk);
+            GF_FREE(inode_ctx);
+            inode_ctx = NULL;
+        }
+
+        /* don't do a very busy loop */
+        timeout = get_cache_invalidation_timeout(this);
+        sleep(timeout / 2);
+        time_now = gf_time();
+    }
+
+    return NULL;
+}
+
+/*
+ * Initialize upcall reaper thread.
+ */
+int
+upcall_reaper_thread_init(xlator_t *this)
+{
+    upcall_private_t *priv = NULL;
+    int ret = -1;
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    ret = gf_thread_create(&priv->reaper_thr, NULL, upcall_reaper_thread, this,
+                           "upreaper");
+
+    return ret;
+}
+
+int
+up_compare_afr_xattr(dict_t *d, char *k, data_t *v, void *tmp)
+{
+    dict_t *dict = tmp;
+
+    if (!strncmp(k, AFR_XATTR_PREFIX, SLEN(AFR_XATTR_PREFIX)) &&
+        (!is_data_equal(v, dict_get(dict, k))))
+        return -1;
+
+    return 0;
+}
+
+static void
+up_filter_afr_xattr(dict_t *xattrs, char *xattr, data_t *v)
+{
+    /* Filter the afr pending xattrs, with value 0. Ideally this should
+     * be executed only in case of xattrop and not in set and removexattr,
+     * butset and remove xattr fops do not come with keys AFR_XATTR_PREFIX
+     */
+    if (!strncmp(xattr, AFR_XATTR_PREFIX, SLEN(AFR_XATTR_PREFIX)) &&
+        (mem_0filled(v->data, v->len) == 0)) {
+        dict_del(xattrs, xattr);
+    }
+    return;
+}
+
+static gf_boolean_t
+up_key_is_regd_xattr(dict_t *regd_xattrs, char *regd_xattr, data_t *v,
+                     void *xattr)
+{
+    int ret = _gf_false;
+    char *key = xattr;
+
+    if (fnmatch(regd_xattr, key, 0) == 0)
+        ret = _gf_true;
+
+    return ret;
+}
+
+int
+up_filter_unregd_xattr(dict_t *xattrs, char *xattr, data_t *v,
+                       void *regd_xattrs)
+{
+    int ret = 0;
+
+    ret = dict_foreach_match(regd_xattrs, up_key_is_regd_xattr, xattr,
+                             dict_null_foreach_fn, NULL);
+    if (ret == 0) {
+        /* xattr was not found in the registered xattr, hence do not
+         * send notification for its change
+         */
+        dict_del(xattrs, xattr);
+        goto out;
+    }
+    up_filter_afr_xattr(xattrs, xattr, v);
+out:
+    return 0;
+}
+
+int
+up_filter_xattr(dict_t *xattr, dict_t *regd_xattrs)
+{
+    int ret = 0;
+
+    ret = dict_foreach(xattr, up_filter_unregd_xattr, regd_xattrs);
+
+    return ret;
+}
+
+static void
+upcall_client_cache_invalidate(xlator_t *this, uuid_t gfid,
+                               upcall_client_t *up_client_entry, uint32_t flags,
+                               struct iatt *stbuf, struct iatt *p_stbuf,
+                               struct iatt *oldp_stbuf, dict_t *xattr,
+                               time_t now);
+
+gf_boolean_t
+up_invalidate_needed(dict_t *xattrs)
+{
+    if (dict_key_count(xattrs) == 0) {
+        gf_msg_trace("upcall", 0,
+                     "None of xattrs requested for"
+                     " invalidation, were changed. Nothing to "
+                     "invalidate");
+        return _gf_false;
+    }
+
+    return _gf_true;
+}
+
+/*
+ * Given a client, first fetch upcall_entry_t from the inode_ctx client list.
+ * Later traverse through the client list of that upcall entry. If this client
+ * is not present in the list, create one client entry with this client info.
+ * Also check if there are other clients which need to be notified of this
+ * op. If yes send notify calls to them.
+ *
+ * Since sending notifications for cache_invalidation is a best effort,
+ * any errors during the process are logged and ignored.
+ */
+void
+upcall_cache_invalidate(call_frame_t *frame, xlator_t *this, client_t *client,
+                        inode_t *inode, uint32_t flags, struct iatt *stbuf,
+                        struct iatt *p_stbuf, struct iatt *oldp_stbuf,
+                        dict_t *xattr)
+{
+    upcall_client_t *up_client_entry = NULL;
+    upcall_client_t *tmp = NULL;
+    upcall_inode_ctx_t *up_inode_ctx = NULL;
+    gf_boolean_t found = _gf_false;
+    time_t time_now;
+    inode_t *linked_inode = NULL;
+
+    if (!is_upcall_enabled(this))
+        return;
+
+    /* server-side generated fops like quota/marker will not have any
+     * client associated with them. Ignore such fops.
+     */
+    if (!client) {
+        gf_msg_debug("upcall", 0, "Internal fop - client NULL");
+        return;
+    }
+
+    /* For nameless LOOKUPs, inode created shall always be
+     * invalid. Hence check if there is any already linked inode.
+     * If yes, update the inode_ctx of that valid inode
+     */
+    if (inode && (inode->ia_type == IA_INVAL) && stbuf) {
+        linked_inode = inode_find(inode->table, stbuf->ia_gfid);
+        if (linked_inode) {
+            gf_log("upcall", GF_LOG_DEBUG,
+                   "upcall_inode_ctx_get of linked inode (%p)", inode);
+            up_inode_ctx = upcall_inode_ctx_get(linked_inode, this);
+        }
+    }
+
+    if (inode && !up_inode_ctx)
+        up_inode_ctx = upcall_inode_ctx_get(inode, this);
+
+    if (!up_inode_ctx) {
+        gf_msg("upcall", GF_LOG_WARNING, 0, UPCALL_MSG_INTERNAL_ERROR,
+               "upcall_inode_ctx_get failed (%p)", inode);
+        return;
+    }
+
+    /* In case of LOOKUP, if first time, inode created shall be
+     * invalid till it gets linked to inode table. Read gfid from
+     * the stat returned in such cases.
+     */
+    if (gf_uuid_is_null(up_inode_ctx->gfid) && stbuf) {
+        /* That means inode must have been invalid when this inode_ctx
+         * is created. Copy the gfid value from stbuf instead.
+         */
+        gf_uuid_copy(up_inode_ctx->gfid, stbuf->ia_gfid);
+    }
+
+    if (gf_uuid_is_null(up_inode_ctx->gfid)) {
+        gf_msg_debug(this->name, 0,
+                     "up_inode_ctx->gfid and "
+                     "stbuf->ia_gfid is NULL, fop:%s",
+                     gf_fop_list[frame->root->op]);
+        goto out;
+    }
+
+    time_now = gf_time();
+    pthread_mutex_lock(&up_inode_ctx->client_list_lock);
+    {
+        list_for_each_entry_safe(up_client_entry, tmp,
+                                 &up_inode_ctx->client_list, client_list)
+        {
+            /* Do not send UPCALL event if same client. */
+            if (!strcmp(client->client_uid, up_client_entry->client_uid)) {
+                up_client_entry->access_time = time_now;
+                found = _gf_true;
+                continue;
+            }
+
+            /*
+             * Ignore sending notifications in case of only UP_ATIME
+             */
+            if (!(flags & ~(UP_ATIME))) {
+                if (found)
+                    break;
+                else /* we still need to find current client entry*/
+                    continue;
+            }
+
+            /* any other client */
+
+            /* XXX: Send notifications asynchrounously
+             * instead of in the I/O path - BZ 1200264
+             *  Also if the file is frequently accessed, set
+             *  expire_time_attr to 0.
+             */
+            upcall_client_cache_invalidate(
+                this, up_inode_ctx->gfid, up_client_entry, flags, stbuf,
+                p_stbuf, oldp_stbuf, xattr, time_now);
+        }
+
+        if (!found) {
+            up_client_entry = __add_upcall_client(frame, client, up_inode_ctx,
+                                                  time_now);
+        }
+    }
+    pthread_mutex_unlock(&up_inode_ctx->client_list_lock);
+out:
+    /* release the ref from inode_find */
+    if (linked_inode)
+        inode_unref(linked_inode);
+    return;
+}
+
+/*
+ * If the upcall_client_t has recently accessed the file (i.e, within
+ * priv->cache_invalidation_timeout), send a upcall notification.
+ */
+static void
+upcall_client_cache_invalidate(xlator_t *this, uuid_t gfid,
+                               upcall_client_t *up_client_entry, uint32_t flags,
+                               struct iatt *stbuf, struct iatt *p_stbuf,
+                               struct iatt *oldp_stbuf, dict_t *xattr,
+                               time_t now)
+{
+    struct gf_upcall up_req = {
+        0,
+    };
+    struct gf_upcall_cache_invalidation ca_req = {
+        0,
+    };
+    time_t timeout = 0;
+    int ret = -1;
+    time_t t_expired = now - up_client_entry->access_time;
+
+    GF_VALIDATE_OR_GOTO("upcall_client_cache_invalidate",
+                        !(gf_uuid_is_null(gfid)), out);
+    timeout = get_cache_invalidation_timeout(this);
+
+    if (t_expired < timeout) {
+        /* Send notify call */
+        up_req.client_uid = up_client_entry->client_uid;
+        gf_uuid_copy(up_req.gfid, gfid);
+
+        ca_req.flags = flags;
+        ca_req.expire_time_attr = up_client_entry->expire_time_attr;
+        if (stbuf)
+            ca_req.stat = *stbuf;
+        if (p_stbuf)
+            ca_req.p_stat = *p_stbuf;
+        if (oldp_stbuf)
+            ca_req.oldp_stat = *oldp_stbuf;
+        ca_req.dict = xattr;
+
+        up_req.data = &ca_req;
+        up_req.event_type = GF_UPCALL_CACHE_INVALIDATION;
+
+        gf_log(THIS->name, GF_LOG_TRACE,
+               "Cache invalidation notification sent to %s",
+               up_client_entry->client_uid);
+
+        /* Need to send inode flags */
+        ret = this->notify(this, GF_EVENT_UPCALL, &up_req);
+
+        /*
+         * notify may fail as the client could have been
+         * dis(re)connected. Cleanup the client entry.
+         */
+        if (ret < 0)
+            __upcall_cleanup_client_entry(up_client_entry);
+
+    } else {
+        gf_log(THIS->name, GF_LOG_TRACE,
+               "Cache invalidation notification NOT sent to %s",
+               up_client_entry->client_uid);
+
+        if (t_expired > (2 * timeout)) {
+            /* Cleanup the entry */
+            __upcall_cleanup_client_entry(up_client_entry);
+        }
+    }
+out:
+    return;
+}
+
+/*
+ * This is called during upcall_inode_ctx cleanup in case of 'inode_forget'.
+ * Send "UP_FORGET" to all the clients so that they invalidate their cache
+ * entry and do a fresh lookup next time when any I/O comes in.
+ */
+static void
+upcall_cache_forget(xlator_t *this, inode_t *inode,
+                    upcall_inode_ctx_t *up_inode_ctx)
+{
+    upcall_client_t *up_client_entry = NULL;
+    upcall_client_t *tmp = NULL;
+    uint32_t flags = UP_FORGET;
+    time_t time_now;
+
+    if (!up_inode_ctx) {
+        return;
+    }
+
+    time_now = gf_time();
+    pthread_mutex_lock(&up_inode_ctx->client_list_lock);
+    {
+        list_for_each_entry_safe(up_client_entry, tmp,
+                                 &up_inode_ctx->client_list, client_list)
+        {
+            /* Set the access time to gf_time()
+             * to send notify */
+            up_client_entry->access_time = time_now;
+
+            upcall_client_cache_invalidate(this, up_inode_ctx->gfid,
+                                           up_client_entry, flags, NULL, NULL,
+                                           NULL, NULL, time_now);
+        }
+    }
+    pthread_mutex_unlock(&up_inode_ctx->client_list_lock);
+}
diff --git a/xlators/features/upcall/src/upcall-mem-types.h b/xlators/features/upcall/src/upcall-mem-types.h
new file mode 100644
index 00000000000..f9883d9d72c
--- /dev/null
+++ b/xlators/features/upcall/src/upcall-mem-types.h
@@ -0,0 +1,23 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __UPCALL_MEM_TYPES_H__
+#define __UPCALL_MEM_TYPES_H__
+
+#include <glusterfs/mem-types.h>
+
+enum gf_upcall_mem_types_ {
+    gf_upcall_mt_conf_t = gf_common_mt_end + 1,
+    gf_upcall_mt_private_t,
+    gf_upcall_mt_upcall_inode_ctx_t,
+    gf_upcall_mt_upcall_client_entry_t,
+    gf_upcall_mt_end
+};
+#endif
diff --git a/xlators/features/upcall/src/upcall-messages.h b/xlators/features/upcall/src/upcall-messages.h
new file mode 100644
index 00000000000..4095a34c200
--- /dev/null
+++ b/xlators/features/upcall/src/upcall-messages.h
@@ -0,0 +1,29 @@
+/*
+ Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+ This file is part of GlusterFS.
+
+ This file is licensed to you under your choice of the GNU Lesser
+ General Public License, version 3 or any later version (LGPLv3 or
+ later), or the GNU General Public License, version 2 (GPLv2), in all
+ cases as published by the Free Software Foundation.
+ */
+
+#ifndef _UPCALL_MESSAGES_H_
+#define _UPCALL_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(UPCALL, UPCALL_MSG_NO_MEMORY, UPCALL_MSG_INTERNAL_ERROR,
+           UPCALL_MSG_NOTIFY_FAILED);
+
+#endif /* !_UPCALL_MESSAGES_H_ */
diff --git a/xlators/features/upcall/src/upcall.c b/xlators/features/upcall/src/upcall.c
new file mode 100644
index 00000000000..0795f58059d
--- /dev/null
+++ b/xlators/features/upcall/src/upcall.c
@@ -0,0 +1,2505 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <unistd.h>
+#include <fcntl.h>
+#include <limits.h>
+#include <pthread.h>
+
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/common-utils.h>
+
+#include <glusterfs/statedump.h>
+
+#include "upcall.h"
+#include "upcall-mem-types.h"
+#include "glusterfs3-xdr.h"
+#include "protocol-common.h"
+#include <glusterfs/defaults.h>
+
+static int32_t
+up_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, fd_t *fd, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, NULL,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(open, frame, op_ret, op_errno, fd, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+        fd_t *fd, dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, fd->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_open_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->open, loc, flags, fd, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(open, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+              dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = UP_WRITE_FLAGS;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, postbuf,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(writev, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+
+    return 0;
+}
+
+static int32_t
+up_writev(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
+          int count, off_t off, uint32_t flags, struct iobref *iobref,
+          dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, fd->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_writev_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->writev, fd, vector, count, off, flags,
+               iobref, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(writev, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+             int op_errno, struct iovec *vector, int count, struct iatt *stbuf,
+             struct iobref *iobref, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, stbuf,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(readv, frame, op_ret, op_errno, vector, count, stbuf,
+                        iobref, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+         off_t offset, uint32_t flags, dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, fd->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_readv_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readv, fd, size, offset, flags, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(readv, frame, -1, op_errno, NULL, 0, NULL, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_lk_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+          int32_t op_errno, struct gf_flock *lock, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, NULL,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(lk, frame, op_ret, op_errno, lock, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+      struct gf_flock *flock, dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, fd->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_lk_cbk, FIRST_CHILD(this), FIRST_CHILD(this)->fops->lk,
+               fd, cmd, flock, xdata);
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(lk, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                int op_errno, struct iatt *prebuf, struct iatt *postbuf,
+                dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = UP_WRITE_FLAGS;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, postbuf,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(truncate, frame, op_ret, op_errno, prebuf, postbuf,
+                        xdata);
+
+    return 0;
+}
+
+static int32_t
+up_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+            dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, loc->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_truncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(truncate, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+               int op_errno, struct iatt *statpre, struct iatt *statpost,
+               dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    /* XXX: setattr -> UP_SIZE or UP_OWN or UP_MODE or UP_TIMES
+     * or INODE_UPDATE (or UP_PERM esp in case of ACLs -> INODE_INVALIDATE)
+     * Need to check what attr is changed and accordingly pass UP_FLAGS.
+     * Bug1200271.
+     */
+    flags = UP_ATTR_FLAGS;
+    /* If mode bits have changed invalidate the xattrs, as posix-acl and
+     * others store permission related information in xattrs. With changing
+     * of permissions/mode, we need to make clients to forget all the
+     * xattrs related to permissions.
+     * TODO: Invalidate the xattr system.posix_acl_access alone.
+     */
+    if (is_same_mode(statpre->ia_prot, statpost->ia_prot) != 0)
+        flags |= UP_XATTR;
+
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, statpost,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(setattr, frame, op_ret, op_errno, statpre, statpost,
+                        xdata);
+
+    return 0;
+}
+
+static int32_t
+up_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *stbuf,
+           int32_t valid, dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, loc->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_setattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setattr, loc, stbuf, valid, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(setattr, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct iatt *stbuf, struct iatt *preoldparent,
+              struct iatt *postoldparent, struct iatt *prenewparent,
+              struct iatt *postnewparent, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = (UP_RENAME_FLAGS | UP_PARENT_DENTRY_FLAGS);
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, stbuf,
+                            postnewparent, postoldparent, NULL);
+
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->rename_oldloc.parent,
+                            flags, postoldparent, NULL, NULL, NULL);
+
+    if (local->rename_oldloc.parent == local->loc.parent)
+        goto out;
+
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->loc.parent, flags,
+                            postnewparent, NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(rename, frame, op_ret, op_errno, stbuf, preoldparent,
+                        postoldparent, prenewparent, postnewparent, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+          dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, newloc, NULL, oldloc->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+    /* copy oldloc */
+    loc_copy(&local->rename_oldloc, oldloc);
+out:
+    STACK_WIND(frame, up_rename_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rename, oldloc, newloc, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(rename, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                        NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, struct iatt *preparent, struct iatt *postparent,
+              dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = (UP_NLINK_FLAGS | UP_PARENT_DENTRY_FLAGS);
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, NULL,
+                            postparent, NULL, NULL);
+
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->loc.parent, flags,
+                            postparent, NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(unlink, frame, op_ret, op_errno, preparent, postparent,
+                        xdata);
+
+    return 0;
+}
+
+static int32_t
+up_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+          dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, loc, NULL, loc->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_unlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->unlink, loc, xflag, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(unlink, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+            int op_errno, inode_t *inode, struct iatt *stbuf,
+            struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = (UP_NLINK_FLAGS | UP_PARENT_DENTRY_FLAGS);
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, stbuf,
+                            postparent, NULL, NULL);
+
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->loc.parent, flags,
+                            postparent, NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(link, frame, op_ret, op_errno, inode, stbuf, preparent,
+                        postparent, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+        dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, newloc, NULL, oldloc->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_link_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->link, oldloc, newloc, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(link, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                        NULL);
+
+    return 0;
+}
+
+static int32_t
+up_rmdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+             int op_errno, struct iatt *preparent, struct iatt *postparent,
+             dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+
+    flags = (UP_NLINK_FLAGS | UP_PARENT_DENTRY_FLAGS);
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, NULL,
+                            postparent, NULL, NULL);
+
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->loc.parent, flags,
+                            postparent, NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(rmdir, frame, op_ret, op_errno, preparent, postparent,
+                        xdata);
+
+    return 0;
+}
+
+static int32_t
+up_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+         dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, loc, NULL, loc->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_rmdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->rmdir, loc, flags, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(rmdir, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+             int op_errno, inode_t *inode, struct iatt *stbuf,
+             struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+
+    /* invalidate parent's entry too */
+    flags = UP_TIMES;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags,
+                            postparent, NULL, NULL, NULL);
+
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->loc.inode, flags, stbuf,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(mkdir, frame, op_ret, op_errno, inode, stbuf, preparent,
+                        postparent, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+         mode_t umask, dict_t *params)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, loc, NULL, loc->parent, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_mkdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mkdir, loc, mode, umask, params);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(mkdir, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                        NULL);
+
+    return 0;
+}
+
+static int32_t
+up_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, fd_t *fd, inode_t *inode, struct iatt *stbuf,
+              struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+
+    /* As its a new file create, no need of sending notification
+     * However invalidate parent's entry and update that fact that the
+     * client has accessed the newly created entry */
+    flags = UP_TIMES;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags,
+                            postparent, NULL, NULL, NULL);
+
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->loc.inode, flags, stbuf,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(create, frame, op_ret, op_errno, fd, inode, stbuf,
+                        preparent, postparent, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+          mode_t mode, mode_t umask, fd_t *fd, dict_t *params)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, loc, NULL, loc->parent, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_create_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->create, loc, flags, mode, umask, fd,
+               params);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(create, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                        NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, inode_t *inode, struct iatt *stbuf, dict_t *xattr,
+              struct iatt *postparent)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, stbuf,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(lookup, frame, op_ret, op_errno, inode, stbuf, xattr,
+                        postparent);
+
+    return 0;
+}
+
+static int32_t
+up_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, loc->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_lookup_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lookup, loc, xattr_req);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+            int32_t op_errno, struct iatt *buf, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, buf, NULL,
+                            NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(stat, frame, op_ret, op_errno, buf, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, loc->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_stat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->stat, loc, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(stat, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, fd->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_stat_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fstat, fd, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(fstat, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+             dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, fd->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_truncate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(ftruncate, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_access_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+              int op_errno, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, NULL,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(access, frame, op_ret, op_errno, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+          dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, loc->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_access_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->access, loc, mask, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(access, frame, -1, op_errno, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_readlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                int op_errno, const char *path, struct iatt *stbuf,
+                dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, stbuf,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(readlink, frame, op_ret, op_errno, path, stbuf, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+            dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, loc->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_readlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readlink, loc, size, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(readlink, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_mknod_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+             int32_t op_errno, inode_t *inode, struct iatt *buf,
+             struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+
+    /* invalidate parent's entry too */
+    flags = UP_TIMES;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags,
+                            postparent, NULL, NULL, NULL);
+
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->loc.inode, flags, buf,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(mknod, frame, op_ret, op_errno, inode, buf, preparent,
+                        postparent, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+         dev_t rdev, mode_t umask, dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, loc, NULL, loc->parent, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_mknod_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->mknod, loc, mode, rdev, umask, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                        NULL);
+
+    return 0;
+}
+
+static int32_t
+up_symlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, inode_t *inode,
+               struct iatt *buf, struct iatt *preparent,
+               struct iatt *postparent, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+
+    /* invalidate parent's entry too */
+    flags = UP_TIMES;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags,
+                            postparent, NULL, NULL, NULL);
+
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->loc.inode, flags, buf,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(symlink, frame, op_ret, op_errno, inode, buf, preparent,
+                        postparent, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+           loc_t *loc, mode_t umask, dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, loc, NULL, loc->parent, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_symlink_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->symlink, linkpath, loc, umask, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(symlink, frame, -1, op_errno, NULL, NULL, NULL, NULL,
+                        NULL);
+
+    return 0;
+}
+
+static int32_t
+up_opendir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, NULL,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(opendir, frame, op_ret, op_errno, fd, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+           dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, loc->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_opendir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->opendir, loc, fd, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(opendir, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_statfs_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+              int32_t op_errno, struct statvfs *buf, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, NULL,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(statfs, frame, op_ret, op_errno, buf, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, loc->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_statfs_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->statfs, loc, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(statfs, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, gf_dirent_t *entries,
+               dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, NULL,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(readdir, frame, op_ret, op_errno, entries, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+           off_t off, dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, fd->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_readdir_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdir, fd, size, off, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(readdir, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, gf_dirent_t *entries,
+                dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+    gf_dirent_t *entry = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, NULL,
+                            NULL, NULL, NULL);
+
+    list_for_each_entry(entry, &entries->list, list)
+    {
+        if (entry->inode == NULL) {
+            continue;
+        }
+        upcall_cache_invalidate(frame, this, client, entry->inode, flags,
+                                &entry->d_stat, NULL, NULL, NULL);
+    }
+
+out:
+    UPCALL_STACK_UNWIND(readdirp, frame, op_ret, op_errno, entries, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+            off_t off, dict_t *dict)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, fd->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_readdirp_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->readdirp, fd, size, off, dict);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(readdirp, frame, -1, op_errno, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
+            int32_t valid, dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, fd->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_setattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetattr, fd, stbuf, valid, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(fsetattr, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, struct iatt *pre,
+                 struct iatt *post, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = UP_WRITE_FLAGS;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, post,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(fallocate, frame, op_ret, op_errno, pre, post, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t mode,
+             off_t offset, size_t len, dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, fd->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_fallocate_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fallocate, fd, mode, offset, len,
+               xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(fallocate, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, struct iatt *pre,
+               struct iatt *post, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = UP_WRITE_FLAGS;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, post,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(discard, frame, op_ret, op_errno, pre, post, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+           size_t len, dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, fd->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_discard_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->discard, fd, offset, len, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(discard, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_zerofill_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, struct iatt *pre,
+                struct iatt *post, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = UP_WRITE_FLAGS;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, post,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(zerofill, frame, op_ret, op_errno, pre, post, xdata);
+
+    return 0;
+}
+
+static int
+up_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+            off_t len, dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, fd->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_zerofill_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->zerofill, fd, offset, len, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(zerofill, frame, -1, op_errno, NULL, NULL, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_seek_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+            int op_errno, off_t offset, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, NULL,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(seek, frame, op_ret, op_errno, offset, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_seek(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+        gf_seek_what_t what, dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, fd->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_seek_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->seek, fd, offset, what, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(seek, frame, -1, op_errno, 0, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+    int ret = 0;
+    struct iatt stbuf = {
+        0,
+    };
+    upcall_private_t *priv = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, priv, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+
+    flags = UP_XATTR;
+
+    ret = up_filter_xattr(local->xattr, priv->xattrs);
+    if (ret < 0) {
+        op_ret = ret;
+        goto out;
+    }
+    if (!up_invalidate_needed(local->xattr))
+        goto out;
+
+    ret = dict_get_iatt(xdata, GF_POSTSTAT, &stbuf);
+    if (ret == 0)
+        flags |= UP_TIMES;
+
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, &stbuf,
+                            NULL, NULL, local->xattr);
+
+out:
+    UPCALL_STACK_UNWIND(setxattr, frame, op_ret, op_errno, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+            int32_t flags, dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, loc, NULL, loc->inode, dict);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_setxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->setxattr, loc, dict, flags, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(setxattr, frame, -1, op_errno, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_fsetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+    int ret = 0;
+    struct iatt stbuf = {
+        0,
+    };
+    upcall_private_t *priv = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, priv, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+
+    flags = UP_XATTR;
+
+    ret = up_filter_xattr(local->xattr, priv->xattrs);
+    if (ret < 0) {
+        op_ret = ret;
+        goto out;
+    }
+    if (!up_invalidate_needed(local->xattr))
+        goto out;
+
+    ret = dict_get_iatt(xdata, GF_POSTSTAT, &stbuf);
+    if (ret == 0)
+        flags |= UP_TIMES;
+
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, &stbuf,
+                            NULL, NULL, local->xattr);
+
+out:
+    UPCALL_STACK_UNWIND(fsetxattr, frame, op_ret, op_errno, xdata);
+
+    return 0;
+}
+
+static int32_t
+up_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+             int32_t flags, dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, fd, fd->inode, dict);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_fsetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
+
+    return 0;
+
+err:
+    UPCALL_STACK_UNWIND(fsetxattr, frame, -1, op_errno, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_fremovexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                    int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+    int ret = 0;
+    upcall_private_t *priv = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, priv, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = UP_XATTR_RM;
+
+    ret = up_filter_xattr(local->xattr, priv->xattrs);
+    if (ret < 0) {
+        op_ret = ret;
+        goto out;
+    }
+    if (!up_invalidate_needed(local->xattr))
+        goto out;
+
+    ret = dict_get_iatt(xdata, GF_POSTSTAT, &stbuf);
+    if (ret == 0)
+        flags |= UP_TIMES;
+
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, &stbuf,
+                            NULL, NULL, local->xattr);
+
+out:
+    UPCALL_STACK_UNWIND(fremovexattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+static int32_t
+up_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+                dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+    dict_t *xattr = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    xattr = dict_for_key_value(name, "", 1, _gf_true);
+    if (!xattr) {
+        goto err;
+    }
+
+    local = upcall_local_init(frame, this, NULL, fd, fd->inode, xattr);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    if (xattr)
+        dict_unref(xattr);
+
+    STACK_WIND(frame, up_fremovexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fremovexattr, fd, name, xdata);
+    return 0;
+
+err:
+    if (xattr)
+        dict_unref(xattr);
+
+    UPCALL_STACK_UNWIND(fremovexattr, frame, -1, op_errno, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_removexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+    struct iatt stbuf = {
+        0,
+    };
+    int ret = 0;
+    upcall_private_t *priv = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, priv, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+    flags = UP_XATTR_RM;
+
+    ret = up_filter_xattr(local->xattr, priv->xattrs);
+    if (ret < 0) {
+        op_ret = ret;
+        goto out;
+    }
+    if (!up_invalidate_needed(local->xattr))
+        goto out;
+
+    ret = dict_get_iatt(xdata, GF_POSTSTAT, &stbuf);
+    if (ret == 0)
+        flags |= UP_TIMES;
+
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, &stbuf,
+                            NULL, NULL, local->xattr);
+
+out:
+    UPCALL_STACK_UNWIND(removexattr, frame, op_ret, op_errno, xdata);
+    return 0;
+}
+
+static int32_t
+up_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+               const char *name, dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+    dict_t *xattr = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    xattr = dict_for_key_value(name, "", 1, _gf_true);
+    if (!xattr) {
+        goto err;
+    }
+
+    local = upcall_local_init(frame, this, loc, NULL, loc->inode, xattr);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    if (xattr)
+        dict_unref(xattr);
+
+    STACK_WIND(frame, up_removexattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->removexattr, loc, name, xdata);
+    return 0;
+
+err:
+    if (xattr)
+        dict_unref(xattr);
+
+    UPCALL_STACK_UNWIND(removexattr, frame, -1, op_errno, NULL);
+
+    return 0;
+}
+
+static int32_t
+up_fgetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, NULL,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(fgetxattr, frame, op_ret, op_errno, dict, xdata);
+    return 0;
+}
+
+static int32_t
+up_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+             dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, fd->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_fgetxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fgetxattr, fd, name, xdata);
+    return 0;
+err:
+    UPCALL_STACK_UNWIND(fgetxattr, frame, -1, op_errno, NULL, NULL);
+    return 0;
+}
+
+static int32_t
+up_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+    client_t *client = NULL;
+    uint32_t flags = 0;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+
+    flags = UP_UPDATE_CLIENT;
+    upcall_cache_invalidate(frame, this, client, local->inode, flags, NULL,
+                            NULL, NULL, NULL);
+
+out:
+    UPCALL_STACK_UNWIND(getxattr, frame, op_ret, op_errno, dict, xdata);
+    return 0;
+}
+
+static int32_t
+up_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, const char *name,
+            dict_t *xdata)
+{
+    int32_t op_errno = ENOMEM;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    local = upcall_local_init(frame, this, NULL, NULL, loc->inode, NULL);
+    if (!local) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_getxattr_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->getxattr, loc, name, xdata);
+    return 0;
+err:
+    UPCALL_STACK_UNWIND(getxattr, frame, -1, op_errno, NULL, NULL);
+    return 0;
+}
+
+/* The xattrops here mainly tracks changes in afr pending xattr.
+ *    1. xattrop doesn't carry info saying post op/pre op.
+ *    2. Pre xattrop will have 0 value for all pending xattrs,
+ *       the cbk of pre xattrop carries the on-disk xattr value.
+ *       Non zero on-disk xattr indicates pending healing.
+ *    3. Post xattrop will either have 0 or 1 as value of pending xattrs,
+ *       0 on success, 1 on failure. But the post xattrop cbk will have
+ *       0 or 1 or any higher value.
+ *       0 - if no healing required*
+ *       1 - if this is the first time pending xattr is being set.
+ *       n - if there is already a pending xattr set, it will increment
+ *       the on-disk value and send that in cbk.
+ * Our aim is to send an invalidation, only the first time a pending
+ * xattr was set on a file. Below are some of the exceptions in handling
+ * xattrop:
+ * - Do not filter unregistered xattrs in the cbk, but in the call path.
+ *   Else, we will be invalidating on every preop, if the file already has
+ *   pending xattr set. Filtering unregistered xattrs on the fop path
+ *   ensures we invalidate only in postop, every time a postop comes with
+ *   pending xattr value 1.
+ * - Consider a brick is down, and the postop sets pending xattrs as long
+ *   as the other brick is down. But we do not want to invalidate every time
+ *   a pending xattr is set, but we want to invalidate only the first time
+ *   a pending xattr is set on any file. Hence, to identify if its the first
+ *   time a pending xattr is set, we compare the value of pending xattrs that
+ *   came in postop and postop cbk, if its same then its the first time.
+ */
+static int32_t
+up_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+               int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+    client_t *client = NULL;
+    upcall_local_t *local = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    client = frame->root->client;
+    local = frame->local;
+
+    if ((op_ret < 0) || !local) {
+        goto out;
+    }
+
+    if (up_invalidate_needed(local->xattr)) {
+        if (dict_foreach(local->xattr, up_compare_afr_xattr, dict) < 0)
+            goto out;
+
+        upcall_cache_invalidate(frame, this, client, local->inode, UP_XATTR,
+                                NULL, NULL, NULL, local->xattr);
+    }
+out:
+    if (frame->root->op == GF_FOP_FXATTROP) {
+        UPCALL_STACK_UNWIND(fxattrop, frame, op_ret, op_errno, dict, xdata);
+    } else {
+        UPCALL_STACK_UNWIND(xattrop, frame, op_ret, op_errno, dict, xdata);
+    }
+    return 0;
+}
+
+static int32_t
+up_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+           gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+    int32_t op_errno = EINVAL;
+    upcall_local_t *local = NULL;
+    int ret = 0;
+    upcall_private_t *priv = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, priv, out);
+
+    local = upcall_local_init(frame, this, loc, NULL, loc->inode, xattr);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    ret = up_filter_xattr(local->xattr, priv->xattrs);
+    if (ret < 0) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_xattrop_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->xattrop, loc, optype, xattr, xdata);
+    return 0;
+err:
+    UPCALL_STACK_UNWIND(xattrop, frame, -1, op_errno, NULL, NULL);
+    return 0;
+}
+
+static int32_t
+up_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+            gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+    int32_t op_errno = EINVAL;
+    upcall_local_t *local = NULL;
+    int ret = 0;
+    upcall_private_t *priv = NULL;
+
+    EXIT_IF_UPCALL_OFF(this, out);
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, priv, out);
+
+    local = upcall_local_init(frame, this, NULL, fd, fd->inode, xattr);
+    if (!local) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    ret = up_filter_xattr(local->xattr, priv->xattrs);
+    if (ret < 0) {
+        goto err;
+    }
+
+out:
+    STACK_WIND(frame, up_xattrop_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->fxattrop, fd, optype, xattr, xdata);
+    return 0;
+err:
+    STACK_UNWIND_STRICT(fxattrop, frame, -1, op_errno, NULL, NULL);
+    return 0;
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    int ret = -1;
+
+    if (!this)
+        return ret;
+
+    ret = xlator_mem_acct_init(this, gf_upcall_mt_end + 1);
+
+    if (ret != 0) {
+        gf_msg("upcall", GF_LOG_WARNING, 0, UPCALL_MSG_NO_MEMORY,
+               "Memory allocation failed");
+        return ret;
+    }
+
+    return ret;
+}
+
+void
+upcall_local_wipe(xlator_t *this, upcall_local_t *local)
+{
+    if (local) {
+        inode_unref(local->inode);
+        if (local->xattr)
+            dict_unref(local->xattr);
+        loc_wipe(&local->rename_oldloc);
+        loc_wipe(&local->loc);
+        if (local->fd)
+            fd_unref(local->fd);
+        mem_put(local);
+    }
+}
+
+upcall_local_t *
+upcall_local_init(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+                  inode_t *inode, dict_t *xattr)
+{
+    upcall_local_t *local = NULL;
+
+    GF_VALIDATE_OR_GOTO("upcall", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, frame, out);
+    GF_VALIDATE_OR_GOTO(this->name, inode, out);
+
+    local = mem_get0(THIS->local_pool);
+
+    if (!local)
+        goto out;
+
+    local->inode = inode_ref(inode);
+    if (xattr)
+        local->xattr = dict_copy_with_ref(xattr, NULL);
+
+    if (loc)
+        loc_copy(&local->loc, loc);
+    if (fd)
+        local->fd = fd_ref(fd);
+
+    frame->local = local;
+
+out:
+    return local;
+}
+
+static int32_t
+update_xattrs(dict_t *dict, char *key, data_t *value, void *data)
+{
+    dict_t *xattrs = data;
+    int ret = 0;
+
+    ret = dict_set_int8(xattrs, key, 0);
+    return ret;
+}
+
+int32_t
+up_ipc(call_frame_t *frame, xlator_t *this, int32_t op, dict_t *xdata)
+{
+    upcall_private_t *priv = NULL;
+    int ret = 0;
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, priv, out);
+
+    if (op != GF_IPC_TARGET_UPCALL)
+        goto wind;
+
+    /* TODO: Bz-1371622 Along with the xattrs also store list of clients
+     * that are interested in notifications, so that the notification
+     * can be sent to the clients that have registered.
+     * Once this implemented there can be unregister of xattrs for
+     * notifications. Until then there is no unregister of xattrs*/
+    if (xdata && priv->xattrs) {
+        ret = dict_foreach(xdata, update_xattrs, priv->xattrs);
+    }
+
+out:
+    STACK_UNWIND_STRICT(ipc, frame, ret, 0, NULL);
+    return 0;
+
+wind:
+    STACK_WIND(frame, default_ipc_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->ipc, op, xdata);
+    return 0;
+}
+
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+    upcall_private_t *priv = NULL;
+    int ret = -1;
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, priv, out);
+
+    GF_OPTION_RECONF("cache-invalidation", priv->cache_invalidation_enabled,
+                     options, bool, out);
+    GF_OPTION_RECONF("cache-invalidation-timeout",
+                     priv->cache_invalidation_timeout, options, int32, out);
+
+    ret = 0;
+
+    if (priv->cache_invalidation_enabled && !priv->reaper_init_done) {
+        ret = upcall_reaper_thread_init(this);
+
+        if (ret) {
+            gf_msg("upcall", GF_LOG_WARNING, 0, UPCALL_MSG_INTERNAL_ERROR,
+                   "reaper_thread creation failed (%s)."
+                   " Disabling cache_invalidation",
+                   strerror(errno));
+        }
+        priv->reaper_init_done = _gf_true;
+    }
+
+out:
+    return ret;
+}
+
+int
+init(xlator_t *this)
+{
+    int ret = -1;
+    upcall_private_t *priv = NULL;
+
+    priv = GF_CALLOC(1, sizeof(*priv), gf_upcall_mt_private_t);
+    if (!priv)
+        goto out;
+
+    priv->xattrs = dict_new();
+    if (!priv->xattrs)
+        goto out;
+
+    GF_OPTION_INIT("cache-invalidation", priv->cache_invalidation_enabled, bool,
+                   out);
+    GF_OPTION_INIT("cache-invalidation-timeout",
+                   priv->cache_invalidation_timeout, int32, out);
+
+    LOCK_INIT(&priv->inode_ctx_lk);
+    INIT_LIST_HEAD(&priv->inode_ctx_list);
+
+    priv->fini = 0;
+    priv->reaper_init_done = _gf_false;
+
+    this->private = priv;
+    this->local_pool = mem_pool_new(upcall_local_t, 512);
+    ret = 0;
+
+    if (priv->cache_invalidation_enabled) {
+        ret = upcall_reaper_thread_init(this);
+
+        if (ret) {
+            gf_msg("upcall", GF_LOG_WARNING, 0, UPCALL_MSG_INTERNAL_ERROR,
+                   "reaper_thread creation failed (%s)."
+                   " Disabling cache_invalidation",
+                   strerror(errno));
+        }
+        priv->reaper_init_done = _gf_true;
+    }
+out:
+    if (ret && priv) {
+        if (priv->xattrs)
+            dict_unref(priv->xattrs);
+
+        GF_FREE(priv);
+    }
+
+    return ret;
+}
+
+void
+fini(xlator_t *this)
+{
+    upcall_private_t *priv = NULL;
+
+    priv = this->private;
+    if (!priv) {
+        return;
+    }
+    this->private = NULL;
+
+    priv->fini = 1;
+
+    if (priv->reaper_thr) {
+        gf_thread_cleanup_xint(priv->reaper_thr);
+        priv->reaper_thr = 0;
+        priv->reaper_init_done = _gf_false;
+    }
+
+    dict_unref(priv->xattrs);
+    LOCK_DESTROY(&priv->inode_ctx_lk);
+
+    /* Do we need to cleanup the inode_ctxs? IMO not required
+     * as inode_forget would have been done on all the inodes
+     * before calling xlator_fini */
+    GF_FREE(priv);
+
+    if (this->local_pool) {
+        mem_pool_destroy(this->local_pool);
+        this->local_pool = NULL;
+    }
+
+    return;
+}
+
+int
+upcall_forget(xlator_t *this, inode_t *inode)
+{
+    upcall_private_t *priv = this->private;
+
+    if (!priv)
+        goto out;
+
+    upcall_cleanup_inode_ctx(this, inode);
+out:
+    return 0;
+}
+
+int
+upcall_release(xlator_t *this, fd_t *fd)
+{
+    return 0;
+}
+
+int
+notify(xlator_t *this, int32_t event, void *data, ...)
+{
+    int ret = -1;
+    struct gf_upcall *up_req = NULL;
+
+    switch (event) {
+        case GF_EVENT_UPCALL: {
+            gf_log(this->name, GF_LOG_DEBUG, "Upcall Notify event = %d", event);
+
+            up_req = (struct gf_upcall *)data;
+
+            GF_VALIDATE_OR_GOTO(this->name, up_req, out);
+
+            ret = default_notify(this, event, up_req);
+
+            if (ret) {
+                gf_msg(this->name, GF_LOG_WARNING, 0, UPCALL_MSG_NOTIFY_FAILED,
+                       "Failed to notify cache invalidation"
+                       " to client(%s)",
+                       up_req->client_uid);
+                goto out;
+            }
+        } break;
+        default:
+            default_notify(this, event, data);
+            break;
+    }
+    ret = 0;
+
+out:
+    return ret;
+}
+
+struct xlator_fops fops = {
+    .ipc = up_ipc,
+    /* fops which change only "ATIME" do not result
+     * in any cache invalidation. Hence upcall
+     * notifications are not sent in this case.
+     * But however, we need to store/update the
+     * client info in the upcall state to be able
+     * to notify them in case of any changes done
+     * to the data.
+     *
+     * Below such fops do not trigger upcall
+     * notifications but will add/update
+     * clients info in the upcall inode ctx.*/
+    .lookup = up_lookup,
+    .open = up_open,
+    .statfs = up_statfs,
+    .opendir = up_opendir,
+    .readdir = up_readdir,
+    .readdirp = up_readdirp,
+    .stat = up_stat,
+    .fstat = up_fstat,
+    .access = up_access,
+    .readlink = up_readlink,
+    .readv = up_readv,
+    .lk = up_lk,
+    .seek = up_seek,
+
+    /* fops doing  write */
+    .truncate = up_truncate,
+    .ftruncate = up_ftruncate,
+    .writev = up_writev,
+    .zerofill = up_zerofill,
+    .fallocate = up_fallocate,
+    .discard = up_discard,
+
+    /* fops changing attributes */
+    .fsetattr = up_fsetattr,
+    .setattr = up_setattr,
+
+    /* fops affecting parent dirent */
+    .mknod = up_mknod,
+    .create = up_create,
+    .symlink = up_symlink,
+    .mkdir = up_mkdir,
+
+    /* fops affecting both file and parent
+     * cache entries */
+    .unlink = up_unlink,
+    .link = up_link,
+    .rmdir = up_rmdir,
+    .rename = up_rename,
+
+    .setxattr = up_setxattr,
+    .fsetxattr = up_fsetxattr,
+    .getxattr = up_getxattr,
+    .fgetxattr = up_fgetxattr,
+    .fremovexattr = up_fremovexattr,
+    .removexattr = up_removexattr,
+    .xattrop = up_xattrop,
+    .fxattrop = up_fxattrop,
+
+#ifdef NOT_SUPPORTED
+    /* internal lk fops */
+    .inodelk = up_inodelk,
+    .finodelk = up_finodelk,
+    .entrylk = up_entrylk,
+    .fentrylk = up_fentrylk,
+
+    /* Below fops follow 'WRITE' which
+     * would have already sent upcall
+     * notifications */
+    .flush = up_flush,
+    .fsync = up_fsync,
+    .fsyncdir = up_fsyncdir,
+#endif
+};
+
+struct xlator_cbks cbks = {
+    .forget = upcall_forget,
+    .release = upcall_release,
+};
+
+struct volume_options options[] = {
+    {
+        .key = {"cache-invalidation"},
+        .type = GF_OPTION_TYPE_BOOL,
+        .default_value = "off",
+        .description = "When \"on\", sends cache-invalidation"
+                       " notifications.",
+        .op_version = {GD_OP_VERSION_3_7_0},
+        .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+        .tags = {"cache", "cacheconsistency", "upcall"},
+    },
+    {.key = {"cache-invalidation-timeout"},
+     .type = GF_OPTION_TYPE_INT,
+     .default_value = CACHE_INVALIDATION_TIMEOUT,
+     .description = "After 'timeout' seconds since the time"
+                    " client accessed any file, cache-invalidation"
+                    " notifications are no longer sent to that client.",
+     .op_version = {GD_OP_VERSION_3_7_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+     .tags = {"cache", "cachetimeout", "upcall"}},
+    {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "upcall",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/features/upcall/src/upcall.h b/xlators/features/upcall/src/upcall.h
new file mode 100644
index 00000000000..aa535088ad7
--- /dev/null
+++ b/xlators/features/upcall/src/upcall.h
@@ -0,0 +1,131 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#ifndef __UPCALL_H__
+#define __UPCALL_H__
+
+#include <glusterfs/compat-errno.h>
+#include "upcall-mem-types.h"
+#include <glusterfs/client_t.h>
+#include "upcall-messages.h"
+#include "upcall-cache-invalidation.h"
+#include <glusterfs/upcall-utils.h>
+
+#define EXIT_IF_UPCALL_OFF(this, label)                                        \
+    do {                                                                       \
+        if (!is_upcall_enabled(this))                                          \
+            goto label;                                                        \
+    } while (0)
+
+#define UPCALL_STACK_UNWIND(fop, frame, params...)                             \
+    do {                                                                       \
+        upcall_local_t *__local = NULL;                                        \
+        xlator_t *__xl = NULL;                                                 \
+        if (frame) {                                                           \
+            __xl = frame->this;                                                \
+            __local = frame->local;                                            \
+            frame->local = NULL;                                               \
+        }                                                                      \
+        STACK_UNWIND_STRICT(fop, frame, params);                               \
+        upcall_local_wipe(__xl, __local);                                      \
+    } while (0)
+
+#define UPCALL_STACK_DESTROY(frame)                                            \
+    do {                                                                       \
+        upcall_local_t *__local = NULL;                                        \
+        xlator_t *__xl = NULL;                                                 \
+        __xl = frame->this;                                                    \
+        __local = frame->local;                                                \
+        frame->local = NULL;                                                   \
+        STACK_DESTROY(frame->root);                                            \
+        upcall_local_wipe(__xl, __local);                                      \
+    } while (0)
+
+struct _upcall_private {
+    gf_boolean_t cache_invalidation_enabled;
+    int32_t cache_invalidation_timeout;
+    struct list_head inode_ctx_list;
+    gf_lock_t inode_ctx_lk;
+    gf_boolean_t reaper_init_done;
+    pthread_t reaper_thr;
+    int32_t fini;
+    dict_t *xattrs; /* list of xattrs registered by clients
+                       for receiving invalidation */
+};
+typedef struct _upcall_private upcall_private_t;
+
+struct _upcall_client {
+    struct list_head client_list;
+    /* strdup to store client_uid, strdup. Free it explicitly */
+    char *client_uid;
+    time_t access_time; /* time last accessed */
+    /* the amount of time which client can cache this entry */
+    uint32_t expire_time_attr;
+};
+typedef struct _upcall_client upcall_client_t;
+
+/* Upcall entries are maintained in inode_ctx */
+struct _upcall_inode_ctx {
+    struct list_head inode_ctx_list;
+    struct list_head client_list;
+    pthread_mutex_t client_list_lock; /* mutex for clients list
+                                         of this upcall entry */
+    int destroy;
+    uuid_t gfid; /* gfid of the entry */
+};
+typedef struct _upcall_inode_ctx upcall_inode_ctx_t;
+
+struct upcall_local {
+    /* XXX: need to check if we can store
+     * pointers in 'local' which may get freed
+     * in future by other thread
+     */
+    inode_t *inode;
+    loc_t rename_oldloc;
+    loc_t loc; /* required for stat in *xattr_cbk */
+    fd_t *fd;  /* required for fstat in *xattr_cbk */
+    dict_t *xattr;
+};
+typedef struct upcall_local upcall_local_t;
+
+void
+upcall_local_wipe(xlator_t *this, upcall_local_t *local);
+upcall_local_t *
+upcall_local_init(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+                  inode_t *inode, dict_t *xattr);
+
+upcall_inode_ctx_t *
+upcall_inode_ctx_get(inode_t *inode, xlator_t *this);
+int
+upcall_cleanup_inode_ctx(xlator_t *this, inode_t *inode);
+
+void *
+upcall_reaper_thread(void *data);
+int
+upcall_reaper_thread_init(xlator_t *this);
+
+/* Xlator options */
+gf_boolean_t
+is_upcall_enabled(xlator_t *this);
+
+/* Cache invalidation specific */
+void
+upcall_cache_invalidate(call_frame_t *frame, xlator_t *this, client_t *client,
+                        inode_t *inode, uint32_t flags, struct iatt *stbuf,
+                        struct iatt *p_stbuf, struct iatt *oldp_stbuf,
+                        dict_t *xattr);
+int
+up_filter_xattr(dict_t *xattr, dict_t *regd_xattrs);
+
+int
+up_compare_afr_xattr(dict_t *d, char *k, data_t *v, void *tmp);
+
+gf_boolean_t
+up_invalidate_needed(dict_t *xattrs);
+#endif /* __UPCALL_H__ */
diff --git a/xlators/features/utime/Makefile.am b/xlators/features/utime/Makefile.am
new file mode 100644
index 00000000000..a985f42a877
--- /dev/null
+++ b/xlators/features/utime/Makefile.am
@@ -0,0 +1,3 @@
+SUBDIRS = src
+
+CLEANFILES =
diff --git a/xlators/features/utime/src/Makefile.am b/xlators/features/utime/src/Makefile.am
new file mode 100644
index 00000000000..7c3adbc2195
--- /dev/null
+++ b/xlators/features/utime/src/Makefile.am
@@ -0,0 +1,41 @@
+xlator_LTLIBRARIES = utime.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
+
+UTIME_SRC = $(top_srcdir)/xlators/features/utime/src
+
+utime_sources  = $(UTIME_SRC)/utime-helpers.c
+utime_sources += $(UTIME_SRC)/utime.c
+
+utime_la_SOURCES = $(utime_sources)
+nodist_utime_la_SOURCES = utime-autogen-fops.c utime-autogen-fops.h
+BUILT_SOURCES = utime-autogen-fops.h
+
+utime_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+utime_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS_utime  = $(UTIME_SRC)/utime-helpers.h
+noinst_HEADERS_utime += $(UTIME_SRC)/utime.h
+noinst_HEADERS_utime += $(UTIME_SRC)/utime-messages.h
+noinst_HEADERS_utime += $(UTIME_SRC)/utime-mem-types.h
+noinst_HEADERS = $(top_srcdir)/xlators/lib/src/libxlator.h
+noinst_HEADERS += $(noinst_HEADERS_utime)
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src \
+	-I$(top_srcdir)/xlators/lib/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+noinst_PYTHON	= utime-gen-fops-c.py utime-gen-fops-h.py
+EXTRA_DIST	= utime-autogen-fops-tmpl.c utime-autogen-fops-tmpl.h
+
+utime-autogen-fops.c: utime-gen-fops-c.py utime-autogen-fops-tmpl.c
+	$(PYTHON) $(UTIME_SRC)/utime-gen-fops-c.py $(UTIME_SRC)/utime-autogen-fops-tmpl.c > $@
+
+utime-autogen-fops.h: utime-gen-fops-h.py utime-autogen-fops-tmpl.h
+	$(PYTHON) $(UTIME_SRC)/utime-gen-fops-h.py $(UTIME_SRC)/utime-autogen-fops-tmpl.h > $@
+
+CLEANFILES = $(nodist_utime_la_SOURCES)
+
+uninstall-local:
+	rm -f $(DESTDIR)$(xlatordir)/utime.so
diff --git a/xlators/features/utime/src/utime-autogen-fops-tmpl.c b/xlators/features/utime/src/utime-autogen-fops-tmpl.c
new file mode 100644
index 00000000000..f2f35322926
--- /dev/null
+++ b/xlators/features/utime/src/utime-autogen-fops-tmpl.c
@@ -0,0 +1,28 @@
+/*
+  Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+/* File: utime-autogen-fops-tmpl.c
+ * This file contains the utime autogenerated FOPs. This is run through
+ * the code generator, generator.py to generate the required FOPs.
+ */
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/statedump.h>
+#include "utime-helpers.h"
+#include <glusterfs/timespec.h>
+
+#pragma generate
diff --git a/xlators/features/utime/src/utime-autogen-fops-tmpl.h b/xlators/features/utime/src/utime-autogen-fops-tmpl.h
new file mode 100644
index 00000000000..4e102ffed6c
--- /dev/null
+++ b/xlators/features/utime/src/utime-autogen-fops-tmpl.h
@@ -0,0 +1,22 @@
+/*
+  Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+/* File: utime-autogen-fops-tmpl.h
+ * This file contains the utime autogenerated FOPs declarations.
+ */
+
+#ifndef _UTIME_AUTOGEN_FOPS_H
+#define _UTIME_AUTOGEN_FOPS_H
+
+#include <glusterfs/xlator.h>
+
+#pragma generate
+
+#endif /* _UTIME_AUTOGEN_FOPS_H */
diff --git a/xlators/features/utime/src/utime-gen-fops-c.py b/xlators/features/utime/src/utime-gen-fops-c.py
new file mode 100755
index 00000000000..9fb3e1b8b1a
--- /dev/null
+++ b/xlators/features/utime/src/utime-gen-fops-c.py
@@ -0,0 +1,147 @@
+#!/usr/bin/python3
+
+from __future__ import print_function
+import os
+import sys
+
+curdir = os.path.dirname(sys.argv[0])
+gendir = os.path.join(curdir, '../../../../libglusterfs/src')
+sys.path.append(gendir)
+from generator import ops, fop_subs, cbk_subs, generate
+
+FOPS_COMMON_TEMPLATE = """
+int32_t
+gf_utime_@NAME@ (call_frame_t *frame, xlator_t *this,
+                @LONG_ARGS@)
+{
+        gl_timespec_get(&frame->root->ctime);
+
+        (void) utime_update_attribute_flags(frame, this, GF_FOP_@UPNAME@);
+        STACK_WIND (frame, gf_utime_@NAME@_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->@NAME@, @SHORT_ARGS@);
+        return 0;
+}
+"""
+
+FOPS_CBK_COMMON_TEMPLATE = """
+int32_t
+gf_utime_@NAME@_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                     int32_t op_ret, int32_t op_errno,
+                    @LONG_ARGS@)
+{
+        STACK_UNWIND_STRICT (@NAME@, frame, op_ret, op_errno, @SHORT_ARGS@);
+        return 0;
+}
+"""
+
+FOPS_READ_TEMPLATE = """
+int32_t
+gf_utime_@NAME@ (call_frame_t *frame, xlator_t *this,
+                @LONG_ARGS@)
+{
+        gl_timespec_get(&frame->root->ctime);
+
+        (void) utime_update_attribute_flags(frame, this, GF_FOP_READ);
+        STACK_WIND (frame, gf_utime_@NAME@_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->@NAME@, @SHORT_ARGS@);
+        return 0;
+}
+"""
+
+FOPS_WRITE_TEMPLATE = """
+int32_t
+gf_utime_@NAME@ (call_frame_t *frame, xlator_t *this,
+                @LONG_ARGS@)
+{
+        gl_timespec_get(&frame->root->ctime);
+
+        (void) utime_update_attribute_flags(frame, this, GF_FOP_WRITE);
+        STACK_WIND (frame, gf_utime_@NAME@_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->@NAME@, @SHORT_ARGS@);
+        return 0;
+}
+"""
+
+FOPS_COPY_FILE_RANGE_TEMPLATE = """
+int32_t
+gf_utime_@NAME@ (call_frame_t *frame, xlator_t *this,
+                @LONG_ARGS@)
+{
+        gl_timespec_get(&frame->root->ctime);
+
+        (void) utime_update_attribute_flags(frame, this, GF_FOP_COPY_FILE_RANGE);
+        STACK_WIND (frame, gf_utime_@NAME@_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->@NAME@, @SHORT_ARGS@);
+        return 0;
+}
+"""
+
+FOPS_SETATTR_TEMPLATE = """
+int32_t
+gf_utime_@NAME@ (call_frame_t *frame, xlator_t *this,
+             @LONG_ARGS@)
+{
+        gl_timespec_get(&frame->root->ctime);
+
+        if (!valid) {
+                frame->root->flags |= MDATA_CTIME;
+        }
+
+        if (valid & (GF_SET_ATTR_UID | GF_SET_ATTR_GID)) {
+                frame->root->flags |= MDATA_CTIME;
+        }
+
+        if (valid & GF_SET_ATTR_MODE) {
+                frame->root->flags |= MDATA_CTIME;
+        }
+
+        if (valid & (GF_SET_ATTR_ATIME | GF_SET_ATTR_MTIME)) {
+            if (valid & GF_ATTR_ATIME_NOW) {
+                frame->root->ctime.tv_sec = stbuf->ia_atime;
+                frame->root->ctime.tv_nsec = stbuf->ia_atime_nsec;
+            } else if (valid & GF_ATTR_MTIME_NOW) {
+                frame->root->ctime.tv_sec = stbuf->ia_mtime;
+                frame->root->ctime.tv_nsec = stbuf->ia_mtime_nsec;
+            }
+        }
+
+        STACK_WIND (frame, gf_utime_@NAME@_cbk, FIRST_CHILD(this),
+                    FIRST_CHILD(this)->fops->@NAME@, @SHORT_ARGS@);
+        return 0;
+}
+"""
+
+utime_ops = ['fallocate', 'zerofill', 'opendir', 'mknod', 'mkdir',
+             'unlink', 'rmdir', 'symlink', 'rename', 'link', 'truncate',
+             'ftruncate', 'create', 'open', 'removexattr', 'fremovexattr']
+
+utime_read_op = ['readv']
+utime_write_op = ['writev']
+utime_setattr_ops = ['setattr', 'fsetattr']
+utime_copy_file_range_ops = ['copy_file_range']
+
+def gen_defaults():
+    for name in ops:
+        if name in utime_ops:
+            print(generate(FOPS_CBK_COMMON_TEMPLATE, name, cbk_subs))
+            print(generate(FOPS_COMMON_TEMPLATE, name, fop_subs))
+        if name in utime_read_op:
+            print(generate(FOPS_CBK_COMMON_TEMPLATE, name, cbk_subs))
+            print(generate(FOPS_READ_TEMPLATE, name, fop_subs))
+        if name in utime_write_op:
+            print(generate(FOPS_CBK_COMMON_TEMPLATE, name, cbk_subs))
+            print(generate(FOPS_WRITE_TEMPLATE, name, fop_subs))
+        if name in utime_setattr_ops:
+            print(generate(FOPS_CBK_COMMON_TEMPLATE, name, cbk_subs))
+            print(generate(FOPS_SETATTR_TEMPLATE, name, fop_subs))
+        if name in utime_copy_file_range_ops:
+            print(generate(FOPS_CBK_COMMON_TEMPLATE, name, cbk_subs))
+            print(generate(FOPS_COPY_FILE_RANGE_TEMPLATE, name, fop_subs))
+
+for l in open(sys.argv[1], 'r').readlines():
+    if l.find('#pragma generate') != -1:
+        print("/* BEGIN GENERATED CODE - DO NOT MODIFY */")
+        gen_defaults()
+        print("/* END GENERATED CODE */")
+    else:
+        print(l[:-1])
diff --git a/xlators/features/utime/src/utime-gen-fops-h.py b/xlators/features/utime/src/utime-gen-fops-h.py
new file mode 100755
index 00000000000..e96274c229a
--- /dev/null
+++ b/xlators/features/utime/src/utime-gen-fops-h.py
@@ -0,0 +1,35 @@
+#!/usr/bin/python3
+
+from __future__ import print_function
+import os
+import sys
+
+curdir = os.path.dirname(sys.argv[0])
+gendir = os.path.join(curdir, '../../../../libglusterfs/src')
+sys.path.append(gendir)
+from generator import ops, fop_subs, generate
+
+OP_FOP_TEMPLATE = """
+int32_t
+gf_utime_@NAME@ (call_frame_t *frame, xlator_t *this,
+                @LONG_ARGS@);
+"""
+
+utime_ops = ['fallocate', 'zerofill', 'opendir', 'mknod', 'mkdir',
+             'unlink', 'rmdir', 'symlink', 'rename', 'link', 'truncate',
+             'ftruncate', 'create', 'open', 'removexattr', 'fremovexattr',
+             'readv', 'writev', 'setattr', 'fsetattr', 'copy_file_range']
+
+def gen_defaults():
+    for name, value in ops.items():
+        if name in utime_ops:
+            print(generate(OP_FOP_TEMPLATE, name, fop_subs))
+
+
+for l in open(sys.argv[1], 'r').readlines():
+    if l.find('#pragma generate') != -1:
+        print("/* BEGIN GENERATED CODE - DO NOT MODIFY */")
+        gen_defaults()
+        print("/* END GENERATED CODE */")
+    else:
+        print(l[:-1])
diff --git a/xlators/features/utime/src/utime-helpers.c b/xlators/features/utime/src/utime-helpers.c
new file mode 100644
index 00000000000..29d9ad93561
--- /dev/null
+++ b/xlators/features/utime/src/utime-helpers.c
@@ -0,0 +1,110 @@
+/*
+  Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "utime-helpers.h"
+#include "utime.h"
+
+void
+gl_timespec_get(struct timespec *ts)
+{
+#ifdef TIME_UTC
+    timespec_get(ts, TIME_UTC);
+#else
+    timespec_now_realtime(ts);
+#endif
+}
+
+void
+utime_update_attribute_flags(call_frame_t *frame, xlator_t *this,
+                             glusterfs_fop_t fop)
+{
+    utime_priv_t *utime_priv = NULL;
+
+    if (!frame || !this) {
+        goto out;
+    }
+
+    utime_priv = this->private;
+
+    switch (fop) {
+        case GF_FOP_SETXATTR:
+        case GF_FOP_FSETXATTR:
+            frame->root->flags |= MDATA_CTIME;
+            break;
+
+        case GF_FOP_FALLOCATE:
+        case GF_FOP_ZEROFILL:
+            frame->root->flags |= MDATA_MTIME;
+            frame->root->flags |= MDATA_ATIME;
+            break;
+
+        case GF_FOP_OPENDIR:
+        case GF_FOP_OPEN:
+        case GF_FOP_READ:
+            if (!utime_priv->noatime) {
+                frame->root->flags |= MDATA_ATIME;
+            }
+            break;
+        case GF_FOP_MKNOD:
+        case GF_FOP_MKDIR:
+        case GF_FOP_SYMLINK:
+        case GF_FOP_CREATE:
+            frame->root->flags |= MDATA_ATIME;
+            frame->root->flags |= MDATA_CTIME;
+            frame->root->flags |= MDATA_MTIME;
+            frame->root->flags |= MDATA_PAR_CTIME;
+            frame->root->flags |= MDATA_PAR_MTIME;
+            break;
+
+        case GF_FOP_UNLINK:
+        case GF_FOP_RMDIR:
+            frame->root->flags |= MDATA_CTIME;
+            frame->root->flags |= MDATA_PAR_CTIME;
+            frame->root->flags |= MDATA_PAR_MTIME;
+            break;
+
+        case GF_FOP_WRITE:
+            frame->root->flags |= MDATA_MTIME;
+            frame->root->flags |= MDATA_CTIME;
+            break;
+
+        case GF_FOP_LINK:
+        case GF_FOP_RENAME:
+            frame->root->flags |= MDATA_CTIME;
+            frame->root->flags |= MDATA_PAR_CTIME;
+            frame->root->flags |= MDATA_PAR_MTIME;
+            break;
+
+        case GF_FOP_TRUNCATE:
+        case GF_FOP_FTRUNCATE:
+            frame->root->flags |= MDATA_CTIME;
+            frame->root->flags |= MDATA_MTIME;
+            break;
+
+        case GF_FOP_REMOVEXATTR:
+        case GF_FOP_FREMOVEXATTR:
+            frame->root->flags |= MDATA_CTIME;
+            break;
+
+        case GF_FOP_COPY_FILE_RANGE:
+            /* Below 2 are for destination fd */
+            frame->root->flags |= MDATA_CTIME;
+            frame->root->flags |= MDATA_MTIME;
+            /* Below flag is for the source fd */
+            if (!utime_priv->noatime) {
+                frame->root->flags |= MDATA_ATIME;
+            }
+            break;
+        default:
+            frame->root->flags = 0;
+    }
+out:
+    return;
+}
diff --git a/xlators/features/utime/src/utime-helpers.h b/xlators/features/utime/src/utime-helpers.h
new file mode 100644
index 00000000000..2e32d4bece6
--- /dev/null
+++ b/xlators/features/utime/src/utime-helpers.h
@@ -0,0 +1,25 @@
+/*
+  Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _UTIME_HELPERS_H
+#define _UTIME_HELPERS_H
+
+#include <glusterfs/stack.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/timespec.h>
+#include <time.h>
+
+void
+gl_timespec_get(struct timespec *ts);
+void
+utime_update_attribute_flags(call_frame_t *frame, xlator_t *this,
+                             glusterfs_fop_t fop);
+
+#endif /* _UTIME_HELPERS_H */
diff --git a/xlators/features/utime/src/utime-mem-types.h b/xlators/features/utime/src/utime-mem-types.h
new file mode 100644
index 00000000000..ad1255f85f3
--- /dev/null
+++ b/xlators/features/utime/src/utime-mem-types.h
@@ -0,0 +1,21 @@
+/*
+  Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __UTIME_MEM_TYPES_H__
+#define __UTIME_MEM_TYPES_H__
+
+#include <glusterfs/mem-types.h>
+
+enum gf_utime_mem_types_ {
+    utime_mt_utime_t = gf_common_mt_end + 1,
+    utime_mt_end
+};
+
+#endif /* __UTIME_MEM_TYPES_H__ */
diff --git a/xlators/features/utime/src/utime-messages.h b/xlators/features/utime/src/utime-messages.h
new file mode 100644
index 00000000000..bd40265abaf
--- /dev/null
+++ b/xlators/features/utime/src/utime-messages.h
@@ -0,0 +1,29 @@
+/*
+  Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __UTIME_MESSAGES_H__
+#define __UTIME_MESSAGES_H__
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(UTIME, UTIME_MSG_NO_MEMORY, UTIME_MSG_SET_MDATA_FAILED,
+           UTIME_MSG_DICT_SET_FAILED);
+
+#endif /* __UTIME_MESSAGES_H__ */
diff --git a/xlators/features/utime/src/utime.c b/xlators/features/utime/src/utime.c
new file mode 100644
index 00000000000..2acc63e6a05
--- /dev/null
+++ b/xlators/features/utime/src/utime.c
@@ -0,0 +1,392 @@
+/*
+  Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "utime.h"
+#include "utime-helpers.h"
+#include "utime-messages.h"
+#include "utime-mem-types.h"
+#include <glusterfs/call-stub.h>
+
+int32_t
+gf_utime_invalidate(xlator_t *this, inode_t *inode)
+{
+    return 0;
+}
+
+int32_t
+gf_utime_forget(xlator_t *this, inode_t *inode)
+{
+    return 0;
+}
+
+int32_t
+gf_utime_client_destroy(xlator_t *this, client_t *client)
+{
+    return 0;
+}
+
+void
+gf_utime_ictxmerge(xlator_t *this, fd_t *fd, inode_t *inode,
+                   inode_t *linked_inode)
+{
+    return;
+}
+
+int32_t
+gf_utime_release(xlator_t *this, fd_t *fd)
+{
+    return 0;
+}
+
+int32_t
+gf_utime_releasedir(xlator_t *this, fd_t *fd)
+{
+    return 0;
+}
+
+int32_t
+gf_utime_client_disconnect(xlator_t *this, client_t *client)
+{
+    return 0;
+}
+
+int32_t
+gf_utime_fdctx_to_dict(xlator_t *this, fd_t *fd, dict_t *dict)
+{
+    return 0;
+}
+
+int32_t
+gf_utime_inode(xlator_t *this)
+{
+    return 0;
+}
+
+int32_t
+gf_utime_inode_to_dict(xlator_t *this, dict_t *dict)
+{
+    return 0;
+}
+
+int32_t
+gf_utime_history(xlator_t *this)
+{
+    return 0;
+}
+
+int32_t
+gf_utime_fd(xlator_t *this)
+{
+    return 0;
+}
+
+int32_t
+gf_utime_fd_to_dict(xlator_t *this, dict_t *dict)
+{
+    return 0;
+}
+
+int32_t
+gf_utime_fdctx(xlator_t *this, fd_t *fd)
+{
+    return 0;
+}
+
+int32_t
+gf_utime_inodectx(xlator_t *this, inode_t *ino)
+{
+    return 0;
+}
+
+int32_t
+gf_utime_inodectx_to_dict(xlator_t *this, inode_t *ino, dict_t *dict)
+{
+    return 0;
+}
+
+int32_t
+gf_utime_priv_to_dict(xlator_t *this, dict_t *dict, char *brickname)
+{
+    return 0;
+}
+
+int32_t
+gf_utime_priv(xlator_t *this)
+{
+    return 0;
+}
+
+int32_t
+mem_acct_init(xlator_t *this)
+{
+    if (xlator_mem_acct_init(this, utime_mt_end + 1) != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, UTIME_MSG_NO_MEMORY,
+               "Memory accounting initialization failed.");
+        return -1;
+    }
+    return 0;
+}
+
+int32_t
+gf_utime_set_mdata_setxattr_cbk(call_frame_t *frame, void *cookie,
+                                xlator_t *this, int op_ret, int op_errno,
+                                dict_t *xdata)
+{
+    call_stub_t *stub = frame->local;
+    /* Don't fail lookup if mdata setxattr fails */
+    if (op_ret) {
+        gf_msg(this->name, GF_LOG_ERROR, op_errno, UTIME_MSG_SET_MDATA_FAILED,
+               "dict set of key for set-ctime-mdata failed");
+    }
+    frame->local = NULL;
+    call_resume(stub);
+    STACK_DESTROY(frame->root);
+    return 0;
+}
+
+int32_t
+gf_utime_set_mdata_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                              int32_t op_ret, int32_t op_errno, inode_t *inode,
+                              struct iatt *stbuf, dict_t *xdata,
+                              struct iatt *postparent)
+{
+    dict_t *dict = NULL;
+    struct mdata_iatt *mdata = NULL;
+    int ret = 0;
+    loc_t loc = {
+        0,
+    };
+    call_frame_t *new_frame = NULL;
+
+    if (!op_ret && dict_get(xdata, GF_XATTR_MDATA_KEY) == NULL) {
+        dict = dict_new();
+        if (!dict) {
+            op_errno = ENOMEM;
+            goto err;
+        }
+        mdata = GF_MALLOC(sizeof(struct mdata_iatt), gf_common_mt_char);
+        if (mdata == NULL) {
+            op_errno = ENOMEM;
+            goto err;
+        }
+        iatt_to_mdata(mdata, stbuf);
+        ret = dict_set_mdata(dict, CTIME_MDATA_XDATA_KEY, mdata, _gf_false);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_WARNING, ENOMEM, UTIME_MSG_NO_MEMORY,
+                   "dict set of key for set-ctime-mdata failed");
+            goto err;
+        }
+        new_frame = copy_frame(frame);
+        if (!new_frame) {
+            op_errno = ENOMEM;
+            goto stub_err;
+        }
+
+        new_frame->local = fop_lookup_cbk_stub(frame, default_lookup_cbk,
+                                               op_ret, op_errno, inode, stbuf,
+                                               xdata, postparent);
+        if (!new_frame->local) {
+            gf_msg(this->name, GF_LOG_WARNING, ENOMEM, UTIME_MSG_NO_MEMORY,
+                   "lookup_cbk stub allocation failed");
+            op_errno = ENOMEM;
+            STACK_DESTROY(new_frame->root);
+            goto stub_err;
+        }
+
+        loc.inode = inode_ref(inode);
+        gf_uuid_copy(loc.gfid, stbuf->ia_gfid);
+
+        new_frame->root->uid = 0;
+        new_frame->root->gid = 0;
+        new_frame->root->pid = GF_CLIENT_PID_SET_UTIME;
+        STACK_WIND(new_frame, gf_utime_set_mdata_setxattr_cbk,
+                   FIRST_CHILD(this), FIRST_CHILD(this)->fops->setxattr, &loc,
+                   dict, 0, NULL);
+
+        dict_unref(dict);
+        inode_unref(loc.inode);
+        return 0;
+    }
+
+    STACK_UNWIND_STRICT(lookup, frame, op_ret, op_errno, inode, stbuf, xdata,
+                        postparent);
+    return 0;
+
+err:
+    if (mdata) {
+        GF_FREE(mdata);
+    }
+stub_err:
+    if (dict) {
+        dict_unref(dict);
+    }
+    STACK_UNWIND_STRICT(lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+    return 0;
+}
+
+int
+gf_utime_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+    int op_errno = EINVAL;
+    int ret = -1;
+
+    VALIDATE_OR_GOTO(frame, err);
+    VALIDATE_OR_GOTO(this, err);
+    VALIDATE_OR_GOTO(loc, err);
+    VALIDATE_OR_GOTO(loc->inode, err);
+
+    xdata = xdata ? dict_ref(xdata) : dict_new();
+    if (!xdata) {
+        op_errno = ENOMEM;
+        goto err;
+    }
+
+    ret = dict_set_int8(xdata, GF_XATTR_MDATA_KEY, 1);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, -ret, UTIME_MSG_DICT_SET_FAILED,
+               "%s: Unable to set dict value for %s", loc->path,
+               GF_XATTR_MDATA_KEY);
+        op_errno = -ret;
+        goto free_dict;
+    }
+
+    STACK_WIND(frame, gf_utime_set_mdata_lookup_cbk, FIRST_CHILD(this),
+               FIRST_CHILD(this)->fops->lookup, loc, xdata);
+    dict_unref(xdata);
+    return 0;
+
+free_dict:
+    dict_unref(xdata);
+err:
+    STACK_UNWIND_STRICT(lookup, frame, ret, op_errno, NULL, NULL, NULL, NULL);
+    return 0;
+}
+
+int32_t
+init(xlator_t *this)
+{
+    utime_priv_t *utime = NULL;
+
+    utime = GF_MALLOC(sizeof(*utime), utime_mt_utime_t);
+    if (utime == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, UTIME_MSG_NO_MEMORY,
+               "Failed to allocate private memory.");
+        return -1;
+    }
+    memset(utime, 0, sizeof(*utime));
+
+    this->private = utime;
+    GF_OPTION_INIT("noatime", utime->noatime, bool, err);
+
+    return 0;
+err:
+    return -1;
+}
+
+void
+fini(xlator_t *this)
+{
+    utime_priv_t *utime = NULL;
+
+    utime = this->private;
+    GF_FREE(utime);
+    return;
+}
+
+int32_t
+reconfigure(xlator_t *this, dict_t *options)
+{
+    utime_priv_t *utime = this->private;
+
+    GF_OPTION_RECONF("noatime", utime->noatime, options, bool, err);
+
+    return 0;
+err:
+    return -1;
+}
+
+int
+notify(xlator_t *this, int event, void *data, ...)
+{
+    return default_notify(this, event, data);
+}
+
+struct xlator_fops fops = {
+    .rename = gf_utime_rename,
+    .mknod = gf_utime_mknod,
+    .readv = gf_utime_readv,
+    .fremovexattr = gf_utime_fremovexattr,
+    .open = gf_utime_open,
+    .create = gf_utime_create,
+    .mkdir = gf_utime_mkdir,
+    .writev = gf_utime_writev,
+    .rmdir = gf_utime_rmdir,
+    .fallocate = gf_utime_fallocate,
+    .truncate = gf_utime_truncate,
+    .symlink = gf_utime_symlink,
+    .zerofill = gf_utime_zerofill,
+    .link = gf_utime_link,
+    .ftruncate = gf_utime_ftruncate,
+    .unlink = gf_utime_unlink,
+    .setattr = gf_utime_setattr,
+    .fsetattr = gf_utime_fsetattr,
+    .opendir = gf_utime_opendir,
+    .removexattr = gf_utime_removexattr,
+    .lookup = gf_utime_lookup,
+};
+struct xlator_cbks cbks = {
+    .invalidate = gf_utime_invalidate,
+    .forget = gf_utime_forget,
+    .client_destroy = gf_utime_client_destroy,
+    .ictxmerge = gf_utime_ictxmerge,
+    .release = gf_utime_release,
+    .releasedir = gf_utime_releasedir,
+    .client_disconnect = gf_utime_client_disconnect,
+};
+struct xlator_dumpops dumpops = {
+    .fdctx_to_dict = gf_utime_fdctx_to_dict,
+    .inode = gf_utime_inode,
+    .inode_to_dict = gf_utime_inode_to_dict,
+    .history = gf_utime_history,
+    .fd = gf_utime_fd,
+    .fd_to_dict = gf_utime_fd_to_dict,
+    .fdctx = gf_utime_fdctx,
+    .inodectx = gf_utime_inodectx,
+    .inodectx_to_dict = gf_utime_inodectx_to_dict,
+    .priv_to_dict = gf_utime_priv_to_dict,
+    .priv = gf_utime_priv,
+};
+
+struct volume_options options[] = {
+    {.key = {"noatime"},
+     .type = GF_OPTION_TYPE_BOOL,
+     .default_value = "on",
+     .op_version = {GD_OP_VERSION_5_0},
+     .flags = OPT_FLAG_SETTABLE | OPT_FLAG_CLIENT_OPT | OPT_FLAG_DOC,
+     .tags = {"ctime"},
+     .description = "Enable/Disable atime updation when ctime feature is "
+                    "enabled. When noatime is on, atime is not updated with "
+                    "ctime feature enabled and vice versa."},
+    {.key = {NULL}}};
+
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .notify = notify,
+    .reconfigure = reconfigure,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {GD_OP_VERSION_5_0},
+    .dumpops = &dumpops,
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "utime",
+    .category = GF_MAINTAINED,
+};
diff --git a/xlators/features/utime/src/utime.h b/xlators/features/utime/src/utime.h
new file mode 100644
index 00000000000..ba55eec00de
--- /dev/null
+++ b/xlators/features/utime/src/utime.h
@@ -0,0 +1,23 @@
+/*
+  Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef __UTIME_H__
+#define __UTIME_H__
+
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+#include "utime-autogen-fops.h"
+
+typedef struct utime_priv {
+    gf_boolean_t noatime;
+} utime_priv_t;
+
+#endif /* __UTIME_H__ */
diff --git a/xlators/lib/src/libxlator.c b/xlators/lib/src/libxlator.c
new file mode 100644
index 00000000000..8075fa0c29f
--- /dev/null
+++ b/xlators/lib/src/libxlator.c
@@ -0,0 +1,490 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include "libxlator.h"
+
+int marker_xtime_default_gauge[] = {
+    [MCNT_FOUND] = 1,     [MCNT_NOTFOUND] = -1, [MCNT_ENODATA] = -1,
+    [MCNT_ENOTCONN] = -1, [MCNT_ENOENT] = -1,   [MCNT_EOTHER] = -1,
+};
+
+int marker_uuid_default_gauge[] = {
+    [MCNT_FOUND] = 1,    [MCNT_NOTFOUND] = 0, [MCNT_ENODATA] = 0,
+    [MCNT_ENOTCONN] = 0, [MCNT_ENOENT] = 0,   [MCNT_EOTHER] = 0,
+};
+
+static int marker_idx_errno_map[] = {
+    [MCNT_FOUND] = EINVAL,    [MCNT_NOTFOUND] = EINVAL,
+    [MCNT_ENOENT] = ENOENT,   [MCNT_ENOTCONN] = ENOTCONN,
+    [MCNT_ENODATA] = ENODATA, [MCNT_EOTHER] = EINVAL,
+    [MCNT_MAX] = 0,
+};
+
+/*Copy the contents of oldtimebuf to newtimbuf*/
+static void
+update_timebuf(uint32_t *oldtimbuf, uint32_t *newtimebuf)
+{
+    newtimebuf[0] = (oldtimbuf[0]);
+    newtimebuf[1] = (oldtimbuf[1]);
+}
+
+/* Convert Timebuf in network order to host order */
+static void
+get_hosttime(uint32_t *oldtimbuf, uint32_t *newtimebuf)
+{
+    newtimebuf[0] = ntohl(oldtimbuf[0]);
+    newtimebuf[1] = ntohl(oldtimbuf[1]);
+}
+
+/* Match the Incoming trusted.glusterfs.<uuid>.xtime against volume uuid */
+int
+match_uuid_local(const char *name, char *uuid)
+{
+    if (!uuid || !*uuid)
+        return -1;
+
+    name = strtail((char *)name, MARKER_XATTR_PREFIX);
+    if (!name || name++ [0] != '.')
+        return -1;
+
+    name = strtail((char *)name, uuid);
+    if (!name || strcmp(name, ".xtime") != 0)
+        return -1;
+
+    return 0;
+}
+
+static void
+marker_local_incr_errcount(xl_marker_local_t *local, int op_errno)
+{
+    marker_result_idx_t i = -1;
+
+    if (!local)
+        return;
+
+    switch (op_errno) {
+        case ENODATA:
+            i = MCNT_ENODATA;
+            break;
+        case ENOENT:
+            i = MCNT_ENOENT;
+            break;
+        case ENOTCONN:
+            i = MCNT_ENOTCONN;
+            break;
+        default:
+            i = MCNT_EOTHER;
+            break;
+    }
+
+    local->count[i]++;
+}
+
+static int
+evaluate_marker_results(int *gauge, int *count)
+{
+    int i = 0;
+    int op_errno = 0;
+    gf_boolean_t sane = _gf_true;
+
+    /* check if the policy of the gauge is violated;
+     * if yes, try to get the best errno, ie. look
+     * for the first position where there is a more
+     * specific kind of vioilation than the generic EINVAL
+     */
+    for (i = 0; i < MCNT_MAX; i++) {
+        if (sane) {
+            if ((gauge[i] > 0 && count[i] < gauge[i]) ||
+                (gauge[i] < 0 && count[i] >= -gauge[i])) {
+                sane = _gf_false;
+                /* generic action: adopt corresponding errno */
+                op_errno = marker_idx_errno_map[i];
+            }
+        } else {
+            /* already insane; trying to get a more informative
+             * errno by checking subsequent counters
+             */
+            if (count[i] > 0)
+                op_errno = marker_idx_errno_map[i];
+        }
+        if (op_errno && op_errno != EINVAL)
+            break;
+    }
+
+    return op_errno;
+}
+
+static void
+cluster_marker_unwind(call_frame_t *frame, char *key, void *value, size_t size,
+                      dict_t *dict)
+{
+    xl_marker_local_t *local = frame->local;
+    int ret = 0;
+    int32_t op_ret = 0;
+    int32_t op_errno = 0;
+    gf_boolean_t unref = _gf_false;
+
+    frame->local = local->xl_local;
+
+    if (local->count[MCNT_FOUND]) {
+        if (!dict) {
+            dict = dict_new();
+            if (dict) {
+                unref = _gf_true;
+            } else {
+                op_ret = -1;
+                op_errno = ENOMEM;
+                goto out;
+            }
+        }
+
+        ret = dict_set_static_bin(dict, key, value, size);
+        if (ret) {
+            op_ret = -1;
+            op_errno = ENOMEM;
+            goto out;
+        }
+    }
+
+    op_errno = evaluate_marker_results(local->gauge, local->count);
+    if (op_errno)
+        op_ret = -1;
+
+out:
+    if (local->xl_specf_unwind) {
+        local->xl_specf_unwind(frame, op_ret, op_errno, dict, NULL);
+    } else {
+        STACK_UNWIND_STRICT(getxattr, frame, op_ret, op_errno, dict, NULL);
+    }
+
+    GF_FREE(local);
+    if (unref)
+        dict_unref(dict);
+}
+
+/* Aggregate all the <volid>.xtime attrs of the cluster and send the max*/
+int32_t
+cluster_markerxtime_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, dict_t *dict, dict_t *xdata)
+
+{
+    int32_t callcnt = 0;
+    uint32_t *net_timebuf = NULL;
+    uint32_t host_timebuf[2] = {
+        0,
+    };
+    char marker_xattr[128] = {0};
+    xl_marker_local_t *local = NULL;
+
+    local = frame->local;
+
+    snprintf(marker_xattr, sizeof(marker_xattr), "%s.%s.%s",
+             MARKER_XATTR_PREFIX, local->vol_uuid, XTIME);
+
+    LOCK(&frame->lock);
+    {
+        callcnt = --local->call_count;
+
+        if (op_ret) {
+            marker_local_incr_errcount(local, op_errno);
+            goto unlock;
+        }
+
+        if (dict_get_ptr(dict, marker_xattr, (void **)&net_timebuf)) {
+            local->count[MCNT_NOTFOUND]++;
+            UNLOCK(&frame->lock);
+            gf_log(this->name, GF_LOG_WARNING,
+                   "Unable to get <uuid>.xtime attr");
+            goto post_unlock;
+        }
+
+        if (local->count[MCNT_FOUND]) {
+            get_hosttime(net_timebuf, host_timebuf);
+            if ((host_timebuf[0] > local->host_timebuf[0]) ||
+                (host_timebuf[0] == local->host_timebuf[0] &&
+                 host_timebuf[1] >= local->host_timebuf[1])) {
+                update_timebuf(net_timebuf, local->net_timebuf);
+                update_timebuf(host_timebuf, local->host_timebuf);
+            }
+
+        } else {
+            get_hosttime(net_timebuf, local->host_timebuf);
+            update_timebuf(net_timebuf, local->net_timebuf);
+            local->count[MCNT_FOUND]++;
+        }
+    }
+unlock:
+    UNLOCK(&frame->lock);
+post_unlock:
+    if (callcnt == 0)
+        cluster_marker_unwind(frame, marker_xattr, local->net_timebuf, 8, dict);
+
+    return 0;
+}
+
+int32_t
+cluster_markeruuid_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, dict_t *dict, dict_t *xdata)
+{
+    int32_t callcnt = 0;
+    struct volume_mark *volmark = NULL;
+    xl_marker_local_t *local = NULL;
+    int32_t ret = -1;
+    char *vol_uuid = NULL;
+
+    local = frame->local;
+
+    LOCK(&frame->lock);
+    {
+        callcnt = --local->call_count;
+        vol_uuid = local->vol_uuid;
+
+        if (op_ret) {
+            marker_local_incr_errcount(local, op_errno);
+            goto unlock;
+        }
+
+        ret = dict_get_bin(dict, GF_XATTR_MARKER_KEY, (void *)&volmark);
+        if (ret)
+            goto unlock;
+
+        if (local->count[MCNT_FOUND]) {
+            if ((local->volmark->major != volmark->major) ||
+                (local->volmark->minor != volmark->minor)) {
+                op_ret = -1;
+                op_errno = EINVAL;
+                goto unlock;
+            }
+
+            if (local->retval) {
+                goto unlock;
+            } else if (volmark->retval) {
+                GF_FREE(local->volmark);
+                local->volmark = gf_memdup(volmark, sizeof(*volmark));
+                local->retval = volmark->retval;
+            } else if ((volmark->sec > local->volmark->sec) ||
+                       ((volmark->sec == local->volmark->sec) &&
+                        (volmark->usec >= local->volmark->usec))) {
+                GF_FREE(local->volmark);
+                local->volmark = gf_memdup(volmark, sizeof(*volmark));
+            }
+
+        } else {
+            local->volmark = gf_memdup(volmark, sizeof(*volmark));
+            VALIDATE_OR_GOTO(local->volmark, unlock);
+            gf_uuid_unparse(volmark->uuid, vol_uuid);
+            if (volmark->retval)
+                local->retval = volmark->retval;
+            local->count[MCNT_FOUND]++;
+        }
+    }
+unlock:
+    UNLOCK(&frame->lock);
+
+    if (callcnt == 0)
+        cluster_marker_unwind(frame, GF_XATTR_MARKER_KEY, local->volmark,
+                              sizeof(*local->volmark), dict);
+
+    return 0;
+}
+
+int
+gf_get_min_stime(xlator_t *this, dict_t *dst, char *key, data_t *value)
+{
+    int ret = -1;
+    uint32_t *net_timebuf = NULL;
+    uint32_t *value_timebuf = NULL;
+    uint32_t host_timebuf[2] = {
+        0,
+    };
+    uint32_t host_value_timebuf[2] = {
+        0,
+    };
+
+    /* stime should be minimum of all the other nodes */
+    ret = dict_get_bin(dst, key, (void **)&net_timebuf);
+    if (ret < 0) {
+        net_timebuf = GF_CALLOC(1, sizeof(int64_t), gf_common_mt_char);
+        if (!net_timebuf)
+            goto out;
+
+        ret = dict_set_bin(dst, key, net_timebuf, sizeof(int64_t));
+        if (ret < 0) {
+            gf_log(this->name, GF_LOG_WARNING, "key=%s: dict set failed", key);
+            goto error;
+        }
+    }
+
+    value_timebuf = data_to_bin(value);
+    if (!value_timebuf) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "key=%s: getting value of stime failed", key);
+        ret = -1;
+        goto out;
+    }
+
+    get_hosttime(value_timebuf, host_value_timebuf);
+    get_hosttime(net_timebuf, host_timebuf);
+
+    /* can't use 'min()' macro here as we need to compare two fields
+       in the array, selectively */
+    if ((host_value_timebuf[0] < host_timebuf[0]) ||
+        ((host_value_timebuf[0] == host_timebuf[0]) &&
+         (host_value_timebuf[1] < host_timebuf[1]))) {
+        update_timebuf(value_timebuf, net_timebuf);
+    }
+
+    ret = 0;
+out:
+    return ret;
+error:
+    /* To be used only when net_timebuf is not set in the dict */
+    if (net_timebuf)
+        GF_FREE(net_timebuf);
+
+    return ret;
+}
+
+int
+gf_get_max_stime(xlator_t *this, dict_t *dst, char *key, data_t *value)
+{
+    int ret = -ENOMEM;
+    uint32_t *net_timebuf = NULL;
+    uint32_t *value_timebuf = NULL;
+    uint32_t host_timebuf[2] = {
+        0,
+    };
+    uint32_t host_value_timebuf[2] = {
+        0,
+    };
+
+    /* stime should be maximum of all the other nodes */
+    ret = dict_get_bin(dst, key, (void **)&net_timebuf);
+    if (ret < 0) {
+        net_timebuf = GF_CALLOC(1, sizeof(int64_t), gf_common_mt_char);
+        if (!net_timebuf)
+            goto out;
+
+        ret = dict_set_bin(dst, key, net_timebuf, sizeof(int64_t));
+        if (ret < 0) {
+            gf_log(this->name, GF_LOG_WARNING, "key=%s: dict set failed", key);
+            goto error;
+        }
+    }
+
+    value_timebuf = data_to_bin(value);
+    if (!value_timebuf) {
+        gf_log(this->name, GF_LOG_WARNING,
+               "key=%s: getting value of stime failed", key);
+        ret = -EINVAL;
+        goto out;
+    }
+
+    get_hosttime(value_timebuf, host_value_timebuf);
+    get_hosttime(net_timebuf, host_timebuf);
+
+    /* can't use 'max()' macro here as we need to compare two fields
+       in the array, selectively */
+    if ((host_value_timebuf[0] > host_timebuf[0]) ||
+        ((host_value_timebuf[0] == host_timebuf[0]) &&
+         (host_value_timebuf[1] > host_timebuf[1]))) {
+        update_timebuf(value_timebuf, net_timebuf);
+    }
+
+    ret = 0;
+out:
+    return ret;
+error:
+    /* To be used only when net_timebuf is not set in the dict */
+    if (net_timebuf)
+        GF_FREE(net_timebuf);
+
+    return ret;
+}
+
+static int
+_get_children_count(xlator_t *xl)
+{
+    int i = 0;
+    xlator_list_t *trav = NULL;
+    for (i = 0, trav = xl->children; trav; trav = trav->next, i++) {
+        /*'i' will have the value */
+    }
+
+    return i;
+}
+
+int
+cluster_handle_marker_getxattr(call_frame_t *frame, loc_t *loc,
+                               const char *name, char *vol_uuid,
+                               xlator_specf_unwind_t unwind,
+                               int (*populate_args)(call_frame_t *frame,
+                                                    int type, int *gauge,
+                                                    xlator_t **subvols))
+{
+    xlator_t *this = frame->this;
+    xlator_t **subvols = NULL;
+    int num_subvols = 0;
+    int type = 0;
+    int i = 0;
+    int gauge[MCNT_MAX] = {0};
+    xl_marker_local_t *local = NULL;
+
+    if (GF_CLIENT_PID_GSYNCD != frame->root->pid)
+        return -EINVAL;
+
+    if (name == NULL)
+        return -EINVAL;
+
+    if (strcmp(GF_XATTR_MARKER_KEY, name) == 0) {
+        type = MARKER_UUID_TYPE;
+        memcpy(gauge, marker_uuid_default_gauge, sizeof(gauge));
+    } else if (match_uuid_local(name, vol_uuid) == 0) {
+        type = MARKER_XTIME_TYPE;
+        memcpy(gauge, marker_xtime_default_gauge, sizeof(gauge));
+    } else {
+        return -EINVAL;
+    }
+
+    num_subvols = _get_children_count(this);
+    subvols = alloca(num_subvols * sizeof(*subvols));
+    num_subvols = populate_args(frame, type, gauge, subvols);
+
+    local = GF_CALLOC(sizeof(struct marker_str), 1,
+                      gf_common_mt_libxl_marker_local);
+
+    if (!local)
+        goto fail;
+
+    local->xl_local = frame->local;
+    local->call_count = num_subvols;
+    local->xl_specf_unwind = unwind;
+    local->vol_uuid = vol_uuid;
+    memcpy(local->gauge, gauge, sizeof(local->gauge));
+
+    frame->local = local;
+
+    for (i = 0; i < num_subvols; i++) {
+        if (MARKER_UUID_TYPE == type)
+            STACK_WIND(frame, cluster_markeruuid_cbk, subvols[i],
+                       subvols[i]->fops->getxattr, loc, name, NULL);
+        else if (MARKER_XTIME_TYPE == type)
+            STACK_WIND(frame, cluster_markerxtime_cbk, subvols[i],
+                       subvols[i]->fops->getxattr, loc, name, NULL);
+    }
+
+    return 0;
+fail:
+    if (unwind)
+        unwind(frame, -1, ENOMEM, NULL, NULL);
+    else
+        default_getxattr_failure_cbk(frame, ENOMEM);
+    return 0;
+}
diff --git a/xlators/lib/src/libxlator.h b/xlators/lib/src/libxlator.h
new file mode 100644
index 00000000000..81da4060d55
--- /dev/null
+++ b/xlators/lib/src/libxlator.h
@@ -0,0 +1,147 @@
+/*
+   Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#ifndef _LIBXLATOR_H
+#define _LIBXLATOR_H
+
+#include <glusterfs/defaults.h>
+
+#include <stdint.h>             // for int32_t
+#include "glusterfs/dict.h"     // for dict_t, data_t
+#include "glusterfs/globals.h"  // for xlator_t, loc_t
+#include "glusterfs/stack.h"    // for call_frame_t
+#include <glusterfs/compat.h>
+#include <glusterfs/compat-errno.h>
+
+#define MARKER_XATTR_PREFIX "trusted.glusterfs"
+#define XTIME "xtime"
+#define VOLUME_MARK "volume-mark"
+#define GF_XATTR_MARKER_KEY MARKER_XATTR_PREFIX "." VOLUME_MARK
+#define UUID_SIZE 36
+#define MARKER_UUID_TYPE 1
+#define MARKER_XTIME_TYPE 2
+
+typedef int32_t (*xlator_specf_unwind_t)(call_frame_t *frame, int op_ret,
+                                         int op_errno, dict_t *dict,
+                                         dict_t *xdata);
+
+struct volume_mark {
+    uint8_t major;
+    uint8_t minor;
+    uint8_t uuid[16];
+    uint8_t retval;
+    uint32_t sec;
+    uint32_t usec;
+} __attribute__((__packed__));
+
+/*
+ * The enumerated type here
+ * is used to index two kind
+ * of integer arrays:
+ * - gauges
+ * - counters
+
+ * A counter is used internally,
+ * in getxattr callbacks, to count
+ * the results, categorized as
+ * the enum names suggest. So values
+ * in the counter are always non-negative.
+
+ * Gauges are part of the API.
+ * The caller passes one to the
+ * top-level aggregator function,
+ * cluster_getmarkerattr(). The gauge
+ * defines an evaluation policy for the
+ * counter. That is, at the
+ * end of the aggregation process
+ * the gauge is matched against the
+ * counter, and the policy
+ * represented by the gauge decides
+ * whether to return with success or failure,
+ * and in latter case, what particular failure
+ * case (errno).
+
+ * The rules are the following: for some index i,
+ * - if gauge[i] == 0, no requirement is set
+ *   against counter[i];
+ * - if gauge[i] > 0, counter[i] >= gauge[i]
+ *   is required;
+ * - if gauge[i] < 0, counter[i] < |gauge[i]|
+ *   is required.
+
+ * If the requirement is not met, then i is mapped
+ * to the respective errno (MCNT_ENOENT -> ENOENT),
+ * or in lack of that, EINVAL.
+
+ * Cf. evaluate_marker_results() and marker_idx_errno_map[]
+ * in libxlator.c
+
+ * We provide two default gauges, one intended for xtime
+ * aggregation, other for volume mark aggregation. The
+ * policies they represent agree with the hard-coded
+ * one prior to gauges. Cf. marker_xtime_default_gauge
+ * and marker_uuid_default_gauge in libxlator.c
+ */
+
+typedef enum {
+    MCNT_FOUND,
+    MCNT_NOTFOUND,
+    MCNT_ENODATA,
+    MCNT_ENOTCONN,
+    MCNT_ENOENT,
+    MCNT_EOTHER,
+    MCNT_MAX
+} marker_result_idx_t;
+
+extern int marker_xtime_default_gauge[];
+extern int marker_uuid_default_gauge[];
+
+struct marker_str {
+    struct volume_mark *volmark;
+    data_t *data;
+
+    uint32_t host_timebuf[2];
+    uint32_t net_timebuf[2];
+    int32_t call_count;
+    int gauge[MCNT_MAX];
+    int count[MCNT_MAX];
+
+    xlator_specf_unwind_t xl_specf_unwind;
+    void *xl_local;
+    char *vol_uuid;
+    uint8_t retval;
+};
+
+typedef struct marker_str xl_marker_local_t;
+
+int32_t
+cluster_markerxtime_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                        int op_ret, int op_errno, dict_t *dict, dict_t *xdata);
+
+int32_t
+cluster_markeruuid_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int op_ret, int op_errno, dict_t *dict, dict_t *xdata);
+
+int
+cluster_handle_marker_getxattr(call_frame_t *frame, loc_t *loc,
+                               const char *name, char *vol_uuid,
+                               xlator_specf_unwind_t unwind,
+                               int (*populate_args)(call_frame_t *frame,
+                                                    int type, int *gauge,
+                                                    xlator_t **subvols));
+int
+match_uuid_local(const char *name, char *uuid);
+
+int
+gf_get_min_stime(xlator_t *this, dict_t *dst, char *key, data_t *value);
+
+int
+gf_get_max_stime(xlator_t *this, dict_t *dst, char *key, data_t *value);
+
+#endif /* !_LIBXLATOR_H */
diff --git a/xlators/meta/Makefile.am b/xlators/meta/Makefile.am
index e1c45f3051c..af437a64d6d 100644
--- a/xlators/meta/Makefile.am
+++ b/xlators/meta/Makefile.am
@@ -1 +1 @@
-SUBDIRS=src
\ No newline at end of file
+SUBDIRS = src
diff --git a/xlators/meta/src/Makefile.am b/xlators/meta/src/Makefile.am
index 385ff553f59..29b871d7984 100644
--- a/xlators/meta/src/Makefile.am
+++ b/xlators/meta/src/Makefile.am
@@ -1,10 +1,44 @@
-xlator_PROGRAMS = meta.so
-xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/
+xlator_LTLIBRARIES = meta.la
+xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator
 
-meta_so_SOURCES = meta.c tree.c misc.c view.c
-noinst_HEADERS = meta.h tree.h misc.h view.h
+meta_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall \
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles
+meta_la_SOURCES = meta.c meta-helpers.c meta-defaults.c \
+	root-dir.c \
+	graphs-dir.c \
+	frames-file.c \
+	graph-dir.c \
+	active-link.c \
+	xlator-dir.c \
+	top-link.c \
+	logging-dir.c \
+	logfile-link.c \
+	loglevel-file.c \
+	process_uuid-file.c \
+	volfile-file.c \
+	view-dir.c \
+	subvolumes-dir.c \
+	subvolume-link.c \
+	type-file.c \
+	version-file.c \
+	options-dir.c \
+	option-file.c \
+	cmdline-file.c \
+	name-file.c \
+	private-file.c \
+	history-file.c \
+	mallinfo-file.c \
+	meminfo-file.c \
+	measure-file.c \
+	profile-file.c
 
-CLEANFILES = 
+meta_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
+
+noinst_HEADERS = meta.h meta-hooks.h meta-mem-types.h
+
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
+
+AM_CFLAGS = -Wall $(GF_CFLAGS)
+
+CLEANFILES =
diff --git a/xlators/meta/src/active-link.c b/xlators/meta/src/active-link.c
new file mode 100644
index 00000000000..7ee780d89e9
--- /dev/null
+++ b/xlators/meta/src/active-link.c
@@ -0,0 +1,34 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+
+static int
+active_link_fill(xlator_t *this, inode_t *inode, strfd_t *strfd)
+{
+    strprintf(strfd, "%s", this->ctx->active->graph_uuid);
+
+    return 0;
+}
+
+struct meta_ops active_link_ops = {.link_fill = active_link_fill};
+
+int
+meta_active_link_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                      dict_t *xdata)
+{
+    meta_ops_set(loc->inode, this, &active_link_ops);
+
+    return 0;
+}
diff --git a/xlators/meta/src/cmdline-file.c b/xlators/meta/src/cmdline-file.c
new file mode 100644
index 00000000000..eb24e985af9
--- /dev/null
+++ b/xlators/meta/src/cmdline-file.c
@@ -0,0 +1,39 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include <glusterfs/strfd.h>
+#include <glusterfs/lkowner.h>
+
+static int
+cmdline_file_fill(xlator_t *this, inode_t *file, strfd_t *strfd)
+{
+    if (this->ctx->cmdlinestr)
+        strprintf(strfd, "{ \n  \"Cmdlinestr\": \"%s\"\n}",
+                  this->ctx->cmdlinestr);
+    return strfd->size;
+}
+
+static struct meta_ops cmdline_file_ops = {
+    .file_fill = cmdline_file_fill,
+};
+
+int
+meta_cmdline_file_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                       dict_t *xdata)
+{
+    meta_ops_set(loc->inode, this, &cmdline_file_ops);
+
+    return 0;
+}
diff --git a/xlators/meta/src/frames-file.c b/xlators/meta/src/frames-file.c
new file mode 100644
index 00000000000..9a13db9a934
--- /dev/null
+++ b/xlators/meta/src/frames-file.c
@@ -0,0 +1,107 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include <glusterfs/strfd.h>
+#include <glusterfs/lkowner.h>
+
+static int
+frames_file_fill(xlator_t *this, inode_t *file, strfd_t *strfd)
+{
+    struct call_pool *pool = NULL;
+    call_stack_t *stack = NULL;
+    call_frame_t *frame = NULL;
+    int i = 0;
+    int j = 1;
+
+    if (!this || !file || !strfd)
+        return -1;
+
+    pool = this->ctx->pool;
+
+    strprintf(strfd, "{ \n\t\"Stack\": [\n");
+
+    LOCK(&pool->lock);
+    {
+        list_for_each_entry(stack, &pool->all_frames, all_frames)
+        {
+            strprintf(strfd, "\t   {\n");
+            strprintf(strfd, "\t\t\"Number\": %d,\n", ++i);
+            strprintf(strfd, "\t\t\"Frame\": [\n");
+            j = 1;
+            list_for_each_entry(frame, &stack->myframes, frames)
+            {
+                strprintf(strfd, "\t\t   {\n");
+                strprintf(strfd, "\t\t\t\"Number\": %d,\n", j++);
+                strprintf(strfd, "\t\t\t\"Xlator\": \"%s\",\n",
+                          frame->this->name);
+                if (frame->begin.tv_sec)
+                    strprintf(strfd, "\t\t\t\"Creation_time\": %d.%09d,\n",
+                              (int)frame->begin.tv_sec,
+                              (int)frame->begin.tv_nsec);
+                strprintf(strfd, " \t\t\t\"Refcount\": %d,\n",
+                          frame->ref_count);
+                if (frame->parent)
+                    strprintf(strfd, "\t\t\t\"Parent\": \"%s\",\n",
+                              frame->parent->this->name);
+                if (frame->wind_from)
+                    strprintf(strfd, "\t\t\t\"Wind_from\": \"%s\",\n",
+                              frame->wind_from);
+                if (frame->wind_to)
+                    strprintf(strfd, "\t\t\t\"Wind_to\": \"%s\",\n",
+                              frame->wind_to);
+                if (frame->unwind_from)
+                    strprintf(strfd, "\t\t\t\"Unwind_from\": \"%s\",\n",
+                              frame->unwind_from);
+                if (frame->unwind_to)
+                    strprintf(strfd, "\t\t\t\"Unwind_to\": \"%s\",\n",
+                              frame->unwind_to);
+                strprintf(strfd, "\t\t\t\"Complete\": %d\n", frame->complete);
+                if (list_is_last(&frame->frames, &stack->myframes))
+                    strprintf(strfd, "\t\t   }\n");
+                else
+                    strprintf(strfd, "\t\t   },\n");
+            }
+            strprintf(strfd, "\t\t],\n");
+            strprintf(strfd, "\t\t\"Unique\": %" PRId64 ",\n", stack->unique);
+            strprintf(strfd, "\t\t\"Type\": \"%s\",\n", gf_fop_list[stack->op]);
+            strprintf(strfd, "\t\t\"UID\": %d,\n", stack->uid);
+            strprintf(strfd, "\t\t\"GID\": %d,\n", stack->gid);
+            strprintf(strfd, "\t\t\"LK_owner\": \"%s\"\n",
+                      lkowner_utoa(&stack->lk_owner));
+            if (i == (int)pool->cnt)
+                strprintf(strfd, "\t   }\n");
+            else
+                strprintf(strfd, "\t   },\n");
+        }
+        strprintf(strfd, "\t],\n");
+        strprintf(strfd, "\t\"Call_Count\": %d\n", (int)pool->cnt);
+        strprintf(strfd, "}");
+    }
+    UNLOCK(&pool->lock);
+
+    return strfd->size;
+}
+
+static struct meta_ops frames_file_ops = {
+    .file_fill = frames_file_fill,
+};
+
+int
+meta_frames_file_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                      dict_t *xdata)
+{
+    meta_ops_set(loc->inode, this, &frames_file_ops);
+    return 0;
+}
diff --git a/xlators/meta/src/graph-dir.c b/xlators/meta/src/graph-dir.c
new file mode 100644
index 00000000000..a8f4787880d
--- /dev/null
+++ b/xlators/meta/src/graph-dir.c
@@ -0,0 +1,98 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include "meta-hooks.h"
+
+static struct meta_dirent graph_dir_dirents[] = {
+    DOT_DOTDOT,
+
+    {
+        .name = "top",
+        .type = IA_IFLNK,
+        .hook = meta_top_link_hook,
+    },
+    {
+        .name = "volfile",
+        .type = IA_IFREG,
+        .hook = meta_volfile_file_hook,
+    },
+    {.name = NULL}};
+
+static int
+graph_dir_fill(xlator_t *this, inode_t *inode, struct meta_dirent **dp)
+{
+    struct meta_dirent *dirents = NULL;
+    glusterfs_graph_t *graph = NULL;
+    int i = 0;
+    int count = 0;
+    xlator_t *xl = NULL;
+
+    graph = meta_ctx_get(inode, this);
+
+    for (xl = graph->first; xl; xl = xl->next)
+        count++;
+
+    dirents = GF_MALLOC(sizeof(*dirents) * count, gf_meta_mt_dirents_t);
+    if (!dirents)
+        return -1;
+
+    i = 0;
+    for (xl = graph->first; xl; xl = xl->next) {
+        dirents[i].name = gf_strdup(xl->name);
+        dirents[i].type = IA_IFDIR;
+        dirents[i].hook = meta_xlator_dir_hook;
+        i++;
+    }
+
+    *dp = dirents;
+    return i;
+}
+
+struct meta_ops graph_dir_ops = {
+    .fixed_dirents = graph_dir_dirents,
+    .dir_fill = graph_dir_fill,
+};
+
+static glusterfs_graph_t *
+glusterfs_graph_lookup(xlator_t *this, const char *graph_uuid)
+{
+    glusterfs_graph_t *graph = NULL;
+    glusterfs_graph_t *tmp = NULL;
+
+    list_for_each_entry(tmp, &this->ctx->graphs, list)
+    {
+        if (strcmp(graph_uuid, tmp->graph_uuid) == 0) {
+            graph = tmp;
+            break;
+        }
+    }
+
+    return graph;
+}
+
+int
+meta_graph_dir_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                    dict_t *xdata)
+{
+    glusterfs_graph_t *graph = NULL;
+
+    graph = glusterfs_graph_lookup(this, loc->name);
+
+    meta_ops_set(loc->inode, this, &graph_dir_ops);
+
+    meta_ctx_set(loc->inode, this, (void *)graph);
+
+    return 0;
+}
diff --git a/xlators/meta/src/graphs-dir.c b/xlators/meta/src/graphs-dir.c
new file mode 100644
index 00000000000..a1ffbca7d5a
--- /dev/null
+++ b/xlators/meta/src/graphs-dir.c
@@ -0,0 +1,67 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include "meta-hooks.h"
+
+static struct meta_dirent graphs_dir_dirents[] = {
+    DOT_DOTDOT,
+
+    {
+        .name = "active",
+        .type = IA_IFLNK,
+        .hook = meta_active_link_hook,
+    },
+    {.name = NULL}};
+
+static int
+graphs_dir_fill(xlator_t *this, inode_t *dir, struct meta_dirent **dp)
+{
+    glusterfs_graph_t *graph = NULL;
+    int graphs_count = 0;
+    int i = 0;
+    struct meta_dirent *dirents = NULL;
+
+    list_for_each_entry(graph, &this->ctx->graphs, list) { graphs_count++; }
+
+    dirents = GF_CALLOC(sizeof(*dirents), graphs_count + 3,
+                        gf_meta_mt_dirents_t);
+    if (!dirents)
+        return -1;
+
+    i = 0;
+    list_for_each_entry(graph, &this->ctx->graphs, list)
+    {
+        dirents[i].name = gf_strdup(graph->graph_uuid);
+        dirents[i].type = IA_IFDIR;
+        dirents[i].hook = meta_graph_dir_hook;
+        i++;
+    }
+
+    *dp = dirents;
+
+    return i;
+}
+
+struct meta_ops graphs_dir_ops = {.fixed_dirents = graphs_dir_dirents,
+                                  .dir_fill = graphs_dir_fill};
+
+int
+meta_graphs_dir_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                     dict_t *xdata)
+{
+    meta_ops_set(loc->inode, this, &graphs_dir_ops);
+
+    return 0;
+}
diff --git a/xlators/meta/src/history-file.c b/xlators/meta/src/history-file.c
new file mode 100644
index 00000000000..7742a635fed
--- /dev/null
+++ b/xlators/meta/src/history-file.c
@@ -0,0 +1,44 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include <glusterfs/strfd.h>
+#include <glusterfs/statedump.h>
+
+static int
+history_file_fill(xlator_t *this, inode_t *file, strfd_t *strfd)
+{
+    xlator_t *xl = NULL;
+
+    xl = meta_ctx_get(file, this);
+
+    gf_proc_dump_xlator_history(xl, strfd);
+
+    return strfd->size;
+}
+
+static struct meta_ops history_file_ops = {
+    .file_fill = history_file_fill,
+};
+
+int
+meta_history_file_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                       dict_t *xdata)
+{
+    meta_ops_set(loc->inode, this, &history_file_ops);
+
+    meta_ctx_set(loc->inode, this, meta_ctx_get(loc->parent, this));
+
+    return 0;
+}
diff --git a/xlators/meta/src/logfile-link.c b/xlators/meta/src/logfile-link.c
new file mode 100644
index 00000000000..616a54518c0
--- /dev/null
+++ b/xlators/meta/src/logfile-link.c
@@ -0,0 +1,34 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+
+static int
+logfile_link_fill(xlator_t *this, inode_t *inode, strfd_t *strfd)
+{
+    strprintf(strfd, "%s", this->ctx->log.filename);
+
+    return 0;
+}
+
+struct meta_ops logfile_link_ops = {.link_fill = logfile_link_fill};
+
+int
+meta_logfile_link_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                       dict_t *xdata)
+{
+    meta_ops_set(loc->inode, this, &logfile_link_ops);
+
+    return 0;
+}
diff --git a/xlators/meta/src/logging-dir.c b/xlators/meta/src/logging-dir.c
new file mode 100644
index 00000000000..46e6f9e95dd
--- /dev/null
+++ b/xlators/meta/src/logging-dir.c
@@ -0,0 +1,44 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include "meta-hooks.h"
+
+static struct meta_dirent logging_dir_dirents[] = {
+    DOT_DOTDOT,
+
+    {
+        .name = "logfile",
+        .type = IA_IFLNK,
+        .hook = meta_logfile_link_hook,
+    },
+    {
+        .name = "loglevel",
+        .type = IA_IFREG,
+        .hook = meta_loglevel_file_hook,
+    },
+    {.name = NULL}};
+
+struct meta_ops logging_dir_ops = {
+    .fixed_dirents = logging_dir_dirents,
+};
+
+int
+meta_logging_dir_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                      dict_t *xdata)
+{
+    meta_ops_set(loc->inode, this, &logging_dir_ops);
+
+    return 0;
+}
diff --git a/xlators/meta/src/loglevel-file.c b/xlators/meta/src/loglevel-file.c
new file mode 100644
index 00000000000..eeeeeaa5907
--- /dev/null
+++ b/xlators/meta/src/loglevel-file.c
@@ -0,0 +1,50 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include <glusterfs/strfd.h>
+
+static int
+loglevel_file_fill(xlator_t *this, inode_t *file, strfd_t *strfd)
+{
+    strprintf(strfd, "%d\n", this->ctx->log.loglevel);
+
+    return strfd->size;
+}
+
+static int
+loglevel_file_write(xlator_t *this, fd_t *fd, struct iovec *iov, int count)
+{
+    long int level = -1;
+
+    level = strtol(iov[0].iov_base, NULL, 0);
+    if (level >= GF_LOG_NONE && level <= GF_LOG_TRACE)
+        gf_log_set_loglevel(this->ctx, level);
+
+    return iov_length(iov, count);
+}
+
+static struct meta_ops loglevel_file_ops = {
+    .file_fill = loglevel_file_fill,
+    .file_write = loglevel_file_write,
+};
+
+int
+meta_loglevel_file_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                        dict_t *xdata)
+{
+    meta_ops_set(loc->inode, this, &loglevel_file_ops);
+
+    return 0;
+}
diff --git a/xlators/meta/src/mallinfo-file.c b/xlators/meta/src/mallinfo-file.c
new file mode 100644
index 00000000000..b4396d72189
--- /dev/null
+++ b/xlators/meta/src/mallinfo-file.c
@@ -0,0 +1,36 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include <glusterfs/statedump.h>
+
+static int
+mallinfo_file_fill(xlator_t *this, inode_t *file, strfd_t *strfd)
+{
+    gf_proc_dump_mallinfo(strfd);
+    return strfd->size;
+}
+
+static struct meta_ops mallinfo_file_ops = {
+    .file_fill = mallinfo_file_fill,
+};
+
+int
+meta_mallinfo_file_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                        dict_t *xdata)
+{
+    meta_ops_set(loc->inode, this, &mallinfo_file_ops);
+
+    return 0;
+}
diff --git a/xlators/meta/src/measure-file.c b/xlators/meta/src/measure-file.c
new file mode 100644
index 00000000000..52e92e48590
--- /dev/null
+++ b/xlators/meta/src/measure-file.c
@@ -0,0 +1,49 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include <glusterfs/strfd.h>
+
+static int
+measure_file_fill(xlator_t *this, inode_t *file, strfd_t *strfd)
+{
+    strprintf(strfd, "%d\n", this->ctx->measure_latency);
+
+    return strfd->size;
+}
+
+static int
+measure_file_write(xlator_t *this, fd_t *fd, struct iovec *iov, int count)
+{
+    long int num = -1;
+
+    num = strtol(iov[0].iov_base, NULL, 0);
+    this->ctx->measure_latency = !!num;
+
+    return iov_length(iov, count);
+}
+
+static struct meta_ops measure_file_ops = {
+    .file_fill = measure_file_fill,
+    .file_write = measure_file_write,
+};
+
+int
+meta_measure_file_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                       dict_t *xdata)
+{
+    meta_ops_set(loc->inode, this, &measure_file_ops);
+
+    return 0;
+}
diff --git a/xlators/meta/src/meminfo-file.c b/xlators/meta/src/meminfo-file.c
new file mode 100644
index 00000000000..d889dfb2ae8
--- /dev/null
+++ b/xlators/meta/src/meminfo-file.c
@@ -0,0 +1,44 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include <glusterfs/strfd.h>
+#include <glusterfs/statedump.h>
+
+static int
+meminfo_file_fill(xlator_t *this, inode_t *file, strfd_t *strfd)
+{
+    xlator_t *xl = NULL;
+
+    xl = meta_ctx_get(file, this);
+
+    gf_proc_dump_xlator_meminfo(xl, strfd);
+
+    return strfd->size;
+}
+
+static struct meta_ops meminfo_file_ops = {
+    .file_fill = meminfo_file_fill,
+};
+
+int
+meta_meminfo_file_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                       dict_t *xdata)
+{
+    meta_ops_set(loc->inode, this, &meminfo_file_ops);
+
+    meta_ctx_set(loc->inode, this, meta_ctx_get(loc->parent, this));
+
+    return 0;
+}
diff --git a/xlators/meta/src/meta-defaults.c b/xlators/meta/src/meta-defaults.c
new file mode 100644
index 00000000000..91c328473f8
--- /dev/null
+++ b/xlators/meta/src/meta-defaults.c
@@ -0,0 +1,655 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+
+#include <glusterfs/compat-errno.h>
+
+int
+meta_default_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                       const char *name, dict_t *xdata)
+{
+    return default_fgetxattr_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                       dict_t *dict, int32_t flags, dict_t *xdata)
+{
+    return default_fsetxattr_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                      dict_t *dict, int32_t flags, dict_t *xdata)
+{
+    return default_setxattr_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                    dict_t *xdata)
+{
+    return default_statfs_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                      int32_t flags, dict_t *xdata)
+{
+    return default_fsyncdir_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+                     dict_t *xdata)
+{
+    META_STACK_UNWIND(opendir, frame, 0, 0, fd, xdata);
+    return 0;
+}
+
+int
+meta_default_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    struct iatt iatt = {};
+
+    meta_iatt_fill(&iatt, fd->inode, fd->inode->ia_type);
+
+    META_STACK_UNWIND(fstat, frame, 0, 0, &iatt, xdata);
+
+    return 0;
+}
+
+int
+meta_default_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+                   dict_t *xdata)
+{
+    return default_fsync_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
+{
+    META_STACK_UNWIND(flush, frame, 0, 0, xdata);
+    return 0;
+}
+
+int
+meta_default_writev(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                    struct iovec *vector, int32_t count, off_t off,
+                    uint32_t flags, struct iobref *iobref, dict_t *xdata)
+{
+    struct meta_ops *ops = NULL;
+    int ret = 0;
+    struct iatt dummy = {};
+
+    ops = meta_ops_get(fd->inode, this);
+    if (!ops)
+        goto err;
+
+    if (!ops->file_write)
+        goto err;
+
+    ret = ops->file_write(this, fd, vector, count);
+
+    META_STACK_UNWIND(writev, frame, (ret >= 0 ? ret : -1),
+                      (ret < 0 ? -ret : 0), &dummy, &dummy, xdata);
+    return 0;
+err:
+    return default_writev_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                   off_t offset, uint32_t flags, dict_t *xdata)
+{
+    meta_fd_t *meta_fd = NULL;
+    struct iovec iov = {};
+    struct iobuf *iobuf = NULL;
+    struct iobref *iobref = NULL;
+    off_t copy_offset = 0;
+    int copy_size = 0;
+    struct iatt iatt = {};
+
+    meta_fd = meta_fd_get(fd, this);
+    if (!meta_fd)
+        return default_readv_failure_cbk(frame, ENODATA);
+
+    if (!meta_fd->size)
+        meta_file_fill(this, fd);
+
+    iobuf = iobuf_get2(this->ctx->iobuf_pool, size);
+    if (!iobuf)
+        return default_readv_failure_cbk(frame, ENOMEM);
+
+    iobref = iobref_new();
+    if (!iobref) {
+        iobuf_unref(iobuf);
+        return default_readv_failure_cbk(frame, ENOMEM);
+    }
+
+    if (iobref_add(iobref, iobuf) != 0) {
+        iobref_unref(iobref);
+        iobuf_unref(iobuf);
+        return default_readv_failure_cbk(frame, ENOMEM);
+    }
+
+    iov.iov_base = iobuf_ptr(iobuf);
+
+    /* iobref would have taken a ref */
+    iobuf_unref(iobuf);
+
+    copy_offset = min(meta_fd->size, offset);
+    copy_size = min(size, (meta_fd->size - copy_offset));
+
+    if (copy_size)
+        memcpy(iov.iov_base, meta_fd->data + copy_offset, copy_size);
+    iov.iov_len = copy_size;
+
+    META_STACK_UNWIND(readv, frame, copy_size, 0, &iov, 1, &iatt, iobref, 0);
+
+    iobref_unref(iobref);
+
+    return 0;
+}
+
+int
+meta_default_open(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                  int32_t flags, fd_t *fd, dict_t *xdata)
+{
+    dict_t *xdata_rsp = NULL;
+
+    xdata_rsp = meta_direct_io_mode(xdata, frame);
+
+    META_STACK_UNWIND(open, frame, 0, 0, fd, xdata_rsp);
+
+    return 0;
+}
+
+int
+meta_default_create(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                    int32_t flags, mode_t mode, mode_t umask, fd_t *fd,
+                    dict_t *xdata)
+{
+    return default_create_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc,
+                  loc_t *newloc, dict_t *xdata)
+{
+    return default_link_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc,
+                    loc_t *newloc, dict_t *xdata)
+{
+    return default_rename_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_symlink(call_frame_t *frame, xlator_t *this, const char *linkpath,
+                     loc_t *loc, mode_t umask, dict_t *xdata)
+{
+    return default_symlink_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags,
+                   dict_t *xdata)
+{
+    return default_rmdir_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+                    dict_t *xdata)
+{
+    return default_unlink_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+                   mode_t umask, dict_t *xdata)
+{
+    return default_mkdir_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+                   dev_t rdev, mode_t umask, dict_t *xdata)
+{
+    return default_mknod_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                      size_t size, dict_t *xdata)
+{
+    struct meta_ops *ops = NULL;
+    strfd_t *strfd = NULL;
+    struct iatt iatt = {};
+    int len = -1;
+
+    ops = meta_ops_get(loc->inode, this);
+    if (!ops || !ops->link_fill) {
+        META_STACK_UNWIND(readlink, frame, -1, EPERM, 0, 0, 0);
+        return 0;
+    }
+
+    strfd = strfd_open();
+    if (!strfd) {
+        META_STACK_UNWIND(readlink, frame, -1, ENOMEM, 0, 0, 0);
+        return 0;
+    }
+
+    ops->link_fill(this, loc->inode, strfd);
+
+    meta_iatt_fill(&iatt, loc->inode, IA_IFLNK);
+
+    if (strfd->data) {
+        len = strlen(strfd->data);
+        META_STACK_UNWIND(readlink, frame, len, 0, strfd->data, &iatt, xdata);
+    } else
+        META_STACK_UNWIND(readlink, frame, -1, ENODATA, 0, 0, 0);
+
+    strfd_close(strfd);
+
+    return 0;
+}
+
+int
+meta_default_access(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                    int32_t mask, dict_t *xdata)
+{
+    return default_access_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                       off_t offset, dict_t *xdata)
+{
+    struct iatt iatt = {};
+
+    meta_iatt_fill(&iatt, fd->inode, IA_IFREG);
+
+    META_STACK_UNWIND(ftruncate, frame, 0, 0, &iatt, &iatt, xdata);
+
+    return 0;
+}
+
+int
+meta_default_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                      const char *name, dict_t *xdata)
+{
+    return default_getxattr_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                     gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+    return default_xattrop_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                      gf_xattrop_flags_t flags, dict_t *dict, dict_t *xdata)
+{
+    return default_fxattrop_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                         const char *name, dict_t *xdata)
+{
+    return default_removexattr_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                          const char *name, dict_t *xdata)
+{
+    return default_fremovexattr_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+                struct gf_flock *lock, dict_t *xdata)
+{
+    return default_lk_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_inodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+                     loc_t *loc, int32_t cmd, struct gf_flock *lock,
+                     dict_t *xdata)
+{
+    return default_inodelk_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_finodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+                      fd_t *fd, int32_t cmd, struct gf_flock *lock,
+                      dict_t *xdata)
+{
+    return default_finodelk_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_entrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+                     loc_t *loc, const char *basename, entrylk_cmd cmd,
+                     entrylk_type type, dict_t *xdata)
+{
+    return default_entrylk_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+                      fd_t *fd, const char *basename, entrylk_cmd cmd,
+                      entrylk_type type, dict_t *xdata)
+{
+    return default_fentrylk_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_rchecksum(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                       off_t offset, int32_t len, dict_t *xdata)
+{
+    return default_rchecksum_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+                     off_t off, dict_t *xdata)
+{
+    meta_fd_t *meta_fd = NULL;
+    int i = 0;
+    gf_dirent_t head;
+    gf_dirent_t *list = NULL;
+    int ret = 0;
+    int this_size = 0;
+    int filled_size = 0;
+    int fixed_size = 0;
+    int dyn_size = 0;
+    struct meta_dirent *fixed_dirents = NULL;
+    struct meta_dirent *dyn_dirents = NULL;
+    struct meta_dirent *dirents = NULL;
+    struct meta_dirent *end = NULL;
+    struct meta_ops *ops = NULL;
+
+    INIT_LIST_HEAD(&head.list);
+
+    ops = meta_ops_get(fd->inode, this);
+    if (!ops)
+        goto err;
+
+    meta_fd = meta_fd_get(fd, this);
+    if (!meta_fd)
+        goto err;
+
+    meta_dir_fill(this, fd);
+
+    fixed_dirents = ops->fixed_dirents;
+    fixed_size = fixed_dirents_len(fixed_dirents);
+
+    dyn_dirents = meta_fd->dirents;
+    dyn_size = meta_fd->size;
+
+    for (i = off; i < (fixed_size + dyn_size);) {
+        if (i >= fixed_size) {
+            dirents = dyn_dirents + (i - fixed_size);
+            end = dyn_dirents + dyn_size;
+        } else {
+            dirents = fixed_dirents + i;
+            end = fixed_dirents + fixed_size;
+        }
+
+        while (dirents < end) {
+            this_size = sizeof(gf_dirent_t) + strlen(dirents->name) + 1;
+            if (this_size + filled_size > size)
+                goto unwind;
+
+            list = gf_dirent_for_name(dirents->name);
+            if (!list)
+                break;
+
+            list->d_off = i + 1;
+            list->d_ino = i + 42;
+            switch (dirents->type) {
+                case IA_IFDIR:
+                    list->d_type = DT_DIR;
+                    break;
+                case IA_IFCHR:
+                    list->d_type = DT_CHR;
+                    break;
+                case IA_IFBLK:
+                    list->d_type = DT_BLK;
+                    break;
+                case IA_IFIFO:
+                    list->d_type = DT_FIFO;
+                    break;
+                case IA_IFLNK:
+                    list->d_type = DT_LNK;
+                    break;
+                case IA_IFREG:
+                    list->d_type = DT_REG;
+                    break;
+                case IA_IFSOCK:
+                    list->d_type = DT_SOCK;
+                    break;
+                case IA_INVAL:
+                    list->d_type = DT_UNKNOWN;
+                    break;
+            }
+
+            list_add_tail(&list->list, &head.list);
+            ret++;
+            i++;
+            dirents++;
+            filled_size += this_size;
+        }
+    }
+
+unwind:
+    META_STACK_UNWIND(readdir, frame, ret, 0, &head, xdata);
+
+    gf_dirent_free(&head);
+
+    return 0;
+err:
+    META_STACK_UNWIND(readdir, frame, -1, ENOMEM, 0, 0);
+    return 0;
+}
+
+int
+meta_default_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                      size_t size, off_t off, dict_t *xdata)
+{
+    return meta_default_readdir(frame, this, fd, size, off, xdata);
+}
+
+int
+meta_default_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                     struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    return default_setattr_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                      off_t offset, dict_t *xdata)
+{
+    struct iatt iatt = {};
+
+    meta_iatt_fill(&iatt, loc->inode, IA_IFREG);
+
+    META_STACK_UNWIND(truncate, frame, 0, 0, &iatt, &iatt, xdata);
+
+    return 0;
+}
+
+int
+meta_default_stat(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                  dict_t *xdata)
+{
+    struct iatt iatt = {};
+
+    meta_iatt_fill(&iatt, loc->inode, loc->inode->ia_type);
+
+    META_STACK_UNWIND(stat, frame, 0, 0, &iatt, xdata);
+
+    return 0;
+}
+
+int
+meta_default_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                    dict_t *xdata)
+{
+    struct meta_ops *ops = NULL;
+    struct meta_dirent *dirent = NULL;
+    struct meta_dirent *dp = NULL;
+    int i = 0;
+    int ret = 0;
+
+    if (!loc->name)
+        return meta_inode_discover(frame, this, loc, xdata);
+
+    ops = meta_ops_get(loc->parent, this);
+    if (!ops)
+        return default_lookup_failure_cbk(frame, EPERM);
+
+    for (dirent = ops->fixed_dirents; dirent && dirent->name; dirent++) {
+        if (strcmp(dirent->name, loc->name) == 0)
+            goto hook;
+    }
+
+    dirent = NULL;
+    if (ops->dir_fill)
+        ret = ops->dir_fill(this, loc->parent, &dp);
+
+    for (i = 0; i < ret; i++) {
+        if (strcmp(dp[i].name, loc->name) == 0) {
+            dirent = &dp[i];
+            goto hook;
+        }
+    }
+hook:
+    if (dirent && dirent->hook) {
+        struct iatt parent = {};
+        struct iatt iatt = {};
+
+        dirent->hook(frame, this, loc, xdata);
+
+        meta_iatt_fill(&iatt, loc->inode, dirent->type);
+
+        META_STACK_UNWIND(lookup, frame, 0, 0, loc->inode, &iatt, xdata,
+                          &parent);
+    } else {
+        META_STACK_UNWIND(lookup, frame, -1, ENOENT, 0, 0, 0, 0);
+    }
+
+    for (i = 0; i < ret; i++)
+        GF_FREE((void *)dp[i].name);
+    GF_FREE(dp);
+
+    return 0;
+}
+
+int
+meta_default_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                      struct iatt *stbuf, int32_t valid, dict_t *xdata)
+{
+    return default_fsetattr_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                       int32_t keep_size, off_t offset, size_t len,
+                       dict_t *xdata)
+{
+    return default_fallocate_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_discard(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                     off_t offset, size_t len, dict_t *xdata)
+{
+    return default_discard_failure_cbk(frame, EPERM);
+}
+
+int
+meta_default_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd,
+                      off_t offset, off_t len, dict_t *xdata)
+{
+    return default_zerofill_failure_cbk(frame, EPERM);
+}
+
+#define SET_META_DEFAULT_FOP(f, name)                                          \
+    do {                                                                       \
+        if (!f->name)                                                          \
+            f->name = meta_default_##name;                                     \
+    } while (0)
+
+struct xlator_fops *
+meta_defaults_init(struct xlator_fops *fops)
+{
+    SET_META_DEFAULT_FOP(fops, create);
+    SET_META_DEFAULT_FOP(fops, open);
+    SET_META_DEFAULT_FOP(fops, stat);
+    SET_META_DEFAULT_FOP(fops, readlink);
+    SET_META_DEFAULT_FOP(fops, mknod);
+    SET_META_DEFAULT_FOP(fops, mkdir);
+    SET_META_DEFAULT_FOP(fops, unlink);
+    SET_META_DEFAULT_FOP(fops, rmdir);
+    SET_META_DEFAULT_FOP(fops, symlink);
+    SET_META_DEFAULT_FOP(fops, rename);
+    SET_META_DEFAULT_FOP(fops, link);
+    SET_META_DEFAULT_FOP(fops, truncate);
+    SET_META_DEFAULT_FOP(fops, readv);
+    SET_META_DEFAULT_FOP(fops, writev);
+    SET_META_DEFAULT_FOP(fops, statfs);
+    SET_META_DEFAULT_FOP(fops, flush);
+    SET_META_DEFAULT_FOP(fops, fsync);
+    SET_META_DEFAULT_FOP(fops, setxattr);
+    SET_META_DEFAULT_FOP(fops, getxattr);
+    SET_META_DEFAULT_FOP(fops, fsetxattr);
+    SET_META_DEFAULT_FOP(fops, fgetxattr);
+    SET_META_DEFAULT_FOP(fops, removexattr);
+    SET_META_DEFAULT_FOP(fops, fremovexattr);
+    SET_META_DEFAULT_FOP(fops, opendir);
+    SET_META_DEFAULT_FOP(fops, readdir);
+    SET_META_DEFAULT_FOP(fops, readdirp);
+    SET_META_DEFAULT_FOP(fops, fsyncdir);
+    SET_META_DEFAULT_FOP(fops, access);
+    SET_META_DEFAULT_FOP(fops, ftruncate);
+    SET_META_DEFAULT_FOP(fops, fstat);
+    SET_META_DEFAULT_FOP(fops, lk);
+    SET_META_DEFAULT_FOP(fops, inodelk);
+    SET_META_DEFAULT_FOP(fops, finodelk);
+    SET_META_DEFAULT_FOP(fops, entrylk);
+    SET_META_DEFAULT_FOP(fops, fentrylk);
+    SET_META_DEFAULT_FOP(fops, lookup);
+    SET_META_DEFAULT_FOP(fops, rchecksum);
+    SET_META_DEFAULT_FOP(fops, xattrop);
+    SET_META_DEFAULT_FOP(fops, fxattrop);
+    SET_META_DEFAULT_FOP(fops, setattr);
+    SET_META_DEFAULT_FOP(fops, fsetattr);
+    SET_META_DEFAULT_FOP(fops, fallocate);
+    SET_META_DEFAULT_FOP(fops, discard);
+    SET_META_DEFAULT_FOP(fops, zerofill);
+
+    return fops;
+}
diff --git a/xlators/meta/src/meta-helpers.c b/xlators/meta/src/meta-helpers.c
new file mode 100644
index 00000000000..cb54f547468
--- /dev/null
+++ b/xlators/meta/src/meta-helpers.c
@@ -0,0 +1,332 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+
+meta_fd_t *
+meta_fd_get(fd_t *fd, xlator_t *this)
+{
+    uint64_t value = 0;
+    meta_fd_t *meta_fd = NULL;
+
+    LOCK(&fd->lock);
+    {
+        if (__fd_ctx_get(fd, this, &value) < 0) {
+            if (!value) {
+                meta_fd = GF_CALLOC(1, sizeof(*meta_fd), gf_meta_mt_fd_t);
+                if (!meta_fd)
+                    goto unlock;
+                value = (long)meta_fd;
+                __fd_ctx_set(fd, this, value);
+            }
+        } else {
+            meta_fd = (void *)(uintptr_t)value;
+        }
+    }
+unlock:
+    UNLOCK(&fd->lock);
+
+    return meta_fd;
+}
+
+int
+meta_fd_release(fd_t *fd, xlator_t *this)
+{
+    uint64_t value = 0;
+    meta_fd_t *meta_fd = NULL;
+    int i = 0;
+
+    fd_ctx_get(fd, this, &value);
+    meta_fd = (void *)(uintptr_t)value;
+
+    if (meta_fd && meta_fd->dirents) {
+        for (i = 0; i < meta_fd->size; i++)
+            GF_FREE((void *)meta_fd->dirents[i].name);
+        GF_FREE(meta_fd->dirents);
+    }
+
+    if (meta_fd) {
+        GF_FREE(meta_fd->data);
+        GF_FREE(meta_fd);
+    }
+    return 0;
+}
+
+struct meta_ops *
+meta_ops_get(inode_t *inode, xlator_t *this)
+{
+    struct meta_ops *ops = NULL;
+    uint64_t value = 0;
+
+    inode_ctx_get2(inode, this, NULL, &value);
+
+    ops = (void *)(uintptr_t)value;
+
+    return ops;
+}
+
+struct xlator_fops *
+meta_fops_get(inode_t *inode, xlator_t *this)
+{
+    struct meta_ops *ops = NULL;
+
+    ops = meta_ops_get(inode, this);
+    if (!ops)
+        return default_fops;
+
+    return &ops->fops;
+}
+
+int
+meta_ops_set(inode_t *inode, xlator_t *this, struct meta_ops *ops)
+{
+    uint64_t value = 0;
+    int ret = 0;
+
+    meta_defaults_init(&ops->fops);
+
+    value = (long)ops;
+
+    ret = inode_ctx_set2(inode, this, NULL, &value);
+
+    return ret;
+}
+
+void *
+meta_ctx_get(inode_t *inode, xlator_t *this)
+{
+    void *ctx = NULL;
+    uint64_t value = 0;
+
+    inode_ctx_get2(inode, this, &value, 0);
+
+    ctx = (void *)(uintptr_t)value;
+
+    return ctx;
+}
+
+int
+meta_ctx_set(inode_t *inode, xlator_t *this, void *ctx)
+{
+    uint64_t value = 0;
+    int ret = 0;
+
+    value = (long)ctx;
+
+    ret = inode_ctx_set2(inode, this, &value, 0);
+
+    return ret;
+}
+
+void
+meta_local_cleanup(meta_local_t *local, xlator_t *this)
+{
+    if (!local)
+        return;
+
+    if (local->xdata)
+        dict_unref(local->xdata);
+
+    GF_FREE(local);
+    return;
+}
+
+meta_local_t *
+meta_local(call_frame_t *frame)
+{
+    meta_local_t *local = NULL;
+
+    local = frame->local;
+    if (!local)
+        local = frame->local = GF_CALLOC(1, sizeof(*local), gf_meta_mt_local_t);
+    return local;
+}
+
+dict_t *
+meta_direct_io_mode(dict_t *xdata, call_frame_t *frame)
+{
+    meta_local_t *local = NULL;
+
+    if (!xdata) {
+        local = meta_local(frame);
+        if (!local)
+            return NULL;
+        xdata = local->xdata = dict_new();
+        if (!xdata)
+            return NULL;
+    }
+
+    if (dict_set_int8(xdata, "direct-io-mode", 1) != 0)
+        return NULL;
+
+    return xdata;
+}
+
+static void
+meta_uuid_copy(uuid_t dst, uuid_t src)
+{
+    gf_uuid_copy(dst, src);
+    if (gf_uuid_is_null(dst))
+        gf_uuid_generate(dst);
+}
+
+static void
+default_meta_iatt_fill(struct iatt *iatt, inode_t *inode, ia_type_t type,
+                       gf_boolean_t is_tunable)
+{
+    struct timeval tv = {};
+
+    iatt->ia_type = type;
+    switch (type) {
+        case IA_IFDIR:
+            iatt->ia_prot = ia_prot_from_st_mode(0555);
+            iatt->ia_nlink = 2;
+            break;
+        case IA_IFLNK:
+            iatt->ia_prot = ia_prot_from_st_mode(0777);
+            iatt->ia_nlink = 1;
+            break;
+        default:
+            iatt->ia_prot = ia_prot_from_st_mode(is_tunable ? 0644 : 0444);
+            iatt->ia_nlink = 1;
+            break;
+    }
+    iatt->ia_uid = 0;
+    iatt->ia_gid = 0;
+    iatt->ia_size = 0;
+
+    meta_uuid_copy(iatt->ia_gfid, inode->gfid);
+    iatt->ia_ino = gfid_to_ino(iatt->ia_gfid);
+
+    gettimeofday(&tv, 0);
+    iatt->ia_mtime = iatt->ia_ctime = iatt->ia_atime = tv.tv_sec;
+    iatt->ia_mtime_nsec = iatt->ia_ctime_nsec = iatt->ia_atime_nsec =
+        (tv.tv_usec * 1000);
+    return;
+}
+
+void
+meta_iatt_fill(struct iatt *iatt, inode_t *inode, ia_type_t type)
+{
+    struct meta_ops *ops = NULL;
+
+    ops = meta_ops_get(inode, THIS);
+    if (!ops)
+        return;
+
+    if (!ops->iatt_fill)
+        default_meta_iatt_fill(iatt, inode, type, !!ops->file_write);
+    else
+        ops->iatt_fill(THIS, inode, iatt);
+    return;
+}
+
+int
+meta_inode_discover(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                    dict_t *xdata)
+{
+    struct iatt iatt = {};
+    struct iatt postparent = {};
+
+    meta_iatt_fill(&iatt, loc->inode, loc->inode->ia_type);
+
+    META_STACK_UNWIND(lookup, frame, 0, 0, loc->inode, &iatt, xdata,
+                      &postparent);
+    return 0;
+}
+
+int
+meta_file_fill(xlator_t *this, fd_t *fd)
+{
+    meta_fd_t *meta_fd = NULL;
+    strfd_t *strfd = NULL;
+    struct meta_ops *ops = NULL;
+    int ret = 0;
+
+    meta_fd = meta_fd_get(fd, this);
+    if (!meta_fd)
+        return -1;
+
+    if (meta_fd->data)
+        return meta_fd->size;
+
+    strfd = strfd_open();
+    if (!strfd)
+        return -1;
+
+    ops = meta_ops_get(fd->inode, this);
+    if (!ops) {
+        strfd_close(strfd);
+        return -1;
+    }
+
+    if (ops->file_fill)
+        ret = ops->file_fill(this, fd->inode, strfd);
+
+    if (ret >= 0) {
+        meta_fd->data = strfd->data;
+        meta_fd->size = strfd->size;
+
+        strfd->data = NULL;
+    }
+
+    strfd_close(strfd);
+
+    return meta_fd->size;
+}
+
+int
+meta_dir_fill(xlator_t *this, fd_t *fd)
+{
+    meta_fd_t *meta_fd = NULL;
+    struct meta_ops *ops = NULL;
+    struct meta_dirent *dp = NULL;
+    int ret = 0;
+
+    meta_fd = meta_fd_get(fd, this);
+    if (!meta_fd)
+        return -1;
+
+    if (meta_fd->dirents)
+        return meta_fd->size;
+
+    ops = meta_ops_get(fd->inode, this);
+    if (!ops)
+        return -1;
+
+    if (ops->dir_fill)
+        ret = ops->dir_fill(this, fd->inode, &dp);
+
+    if (dp) {
+        meta_fd->dirents = dp;
+        meta_fd->size = ret;
+    }
+
+    return meta_fd->size;
+}
+
+int
+fixed_dirents_len(struct meta_dirent *dirents)
+{
+    int i = 0;
+    struct meta_dirent *dirent = NULL;
+
+    if (!dirents)
+        return 0;
+
+    for (dirent = dirents; dirent->name; dirent++)
+        i++;
+
+    return i;
+}
diff --git a/xlators/meta/src/meta-hooks.h b/xlators/meta/src/meta-hooks.h
new file mode 100644
index 00000000000..7208641398a
--- /dev/null
+++ b/xlators/meta/src/meta-hooks.h
@@ -0,0 +1,48 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef __META_HOOKS_H
+#define __META_HOOKS_H
+#include <glusterfs/xlator.h>
+
+#define DECLARE_HOOK(name)                                                     \
+    int meta_##name##_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,    \
+                           dict_t *xdata)
+
+DECLARE_HOOK(root_dir);
+DECLARE_HOOK(graphs_dir);
+DECLARE_HOOK(frames_file);
+DECLARE_HOOK(graph_dir);
+DECLARE_HOOK(active_link);
+DECLARE_HOOK(xlator_dir);
+DECLARE_HOOK(top_link);
+DECLARE_HOOK(logging_dir);
+DECLARE_HOOK(logfile_link);
+DECLARE_HOOK(loglevel_file);
+DECLARE_HOOK(process_uuid_file);
+DECLARE_HOOK(volfile_file);
+DECLARE_HOOK(view_dir);
+DECLARE_HOOK(subvolumes_dir);
+DECLARE_HOOK(subvolume_link);
+DECLARE_HOOK(type_file);
+DECLARE_HOOK(version_file);
+DECLARE_HOOK(options_dir);
+DECLARE_HOOK(option_file);
+DECLARE_HOOK(cmdline_file);
+DECLARE_HOOK(name_file);
+DECLARE_HOOK(private_file);
+DECLARE_HOOK(mallinfo_file);
+DECLARE_HOOK(history_file);
+DECLARE_HOOK(master_dir);
+DECLARE_HOOK(meminfo_file);
+DECLARE_HOOK(measure_file);
+DECLARE_HOOK(profile_file);
+
+#endif
diff --git a/xlators/meta/src/meta-mem-types.h b/xlators/meta/src/meta-mem-types.h
index 0285dcedec2..033c306682f 100644
--- a/xlators/meta/src/meta-mem-types.h
+++ b/xlators/meta/src/meta-mem-types.h
@@ -1,35 +1,25 @@
 /*
-   Copyright (c) 2008-2009 Gluster, Inc. <http://www.gluster.com>
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
    This file is part of GlusterFS.
 
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
 */
 
-
 #ifndef __META_MEM_TYPES_H__
 #define __META_MEM_TYPES_H__
 
-#include "mem-types.h"
+#include <glusterfs/mem-types.h>
 
 enum gf_meta_mem_types_ {
-        gf_meta_mt__open_local = gf_common_mt_end + 1,
-        gf_meta_mt_dir_entry_t,
-        gf_meta_mt_meta_dirent_t,
-        gf_meta_mt_meta_private_t,
-        gf_meta_mt_stat,
-        gf_meta_mt_end
+    gf_meta_mt_priv_t = gf_common_mt_end + 1,
+    gf_meta_mt_fd_t,
+    gf_meta_mt_fd_data_t,
+    gf_meta_mt_strfd_t,
+    gf_meta_mt_dirents_t,
+    gf_meta_mt_local_t,
+    gf_meta_mt_end
 };
 #endif
-
diff --git a/xlators/meta/src/meta.c b/xlators/meta/src/meta.c
index 0fcdf6e6897..e1b9a2b6581 100644
--- a/xlators/meta/src/meta.c
+++ b/xlators/meta/src/meta.c
@@ -1,1307 +1,276 @@
 /*
-   Copyright (c) 2006-2009 Gluster, Inc. <http://www.gluster.com>
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
    This file is part of GlusterFS.
 
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
 */
 
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <unistd.h>
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "dict.h"
-#include "xlator.h"
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
 
-#include "meta.h"
-#include "view.h"
 #include "meta-mem-types.h"
+#include "meta.h"
 
-int32_t
-meta_getattr_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  struct stat *buf)
-{
-  STACK_UNWIND (frame, op_ret, op_errno, buf);
-  return 0;
-}
-
-int32_t
-meta_getattr (call_frame_t *frame,
-	      xlator_t *this,
-	      const char *path)
-{
-  meta_private_t *priv = (meta_private_t *) this->private;
-  meta_dirent_t *root = priv->tree;
-  meta_dirent_t *file = lookup_meta_entry (root, path, NULL);
-  
-  if (file) {
-    if (file->fops && file->fops->getattr) {
-      STACK_WIND (frame, meta_getattr_cbk,
-		  this, file->fops->getattr, path);
-      return 0;
-    }
-    else {
-      STACK_UNWIND (frame, 0, 0, file->stbuf);
-      return 0;
-    }
-  }
-  else {
-    STACK_WIND (frame, meta_getattr_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->getattr,
-	      path);
-    return 0;
-  }
-}
-
-int32_t
-meta_chmod_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		struct stat *buf)
-{
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno,
-		buf);
-  return 0;
-}
-
-int32_t
-meta_chmod (call_frame_t *frame,
-	    xlator_t *this,
-	    const char *path,
-	    mode_t mode)
-{
-  STACK_WIND (frame,
-	      meta_chmod_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->chmod,
-	      path,
-	      mode);
-  return 0;
-}
-
-int32_t
-meta_chown_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		struct stat *buf)
-{
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno,
-		buf);
-  return 0;
-}
-
-int32_t
-meta_chown (call_frame_t *frame,
-	    xlator_t *this,
-	    const char *path,
-	    uid_t uid,
-	    gid_t gid)
-{
-  STACK_WIND (frame,	      
-	      meta_chown_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->chown,
-	      path,
-	      uid,
-	      gid);
-  return 0;
-}
-
-
-int32_t
-meta_truncate_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-		   struct stat *buf)
-{
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno,
-		buf);
-  return 0;
-}
-
-int32_t
-meta_truncate (call_frame_t *frame,
-	       xlator_t *this,
-	       const char *path,
-	       off_t offset)
-{
-  STACK_WIND (frame,
-	      meta_truncate_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->truncate,
-	      path,
-	      offset);
-  return 0;
-}
-
-
-int32_t
-meta_ftruncate_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    struct stat *buf)
-{
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno,
-		buf);
-  return 0;
-}
-
-int32_t
-meta_ftruncate (call_frame_t *frame,
-		xlator_t *this,
-		dict_t *fd,
-		off_t offset)
-{
-  STACK_WIND (frame,
-	      meta_ftruncate_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->ftruncate,
-	      fd,
-	      offset);
-  return 0;
-}
-
-
-int32_t
-meta_utimes_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 struct stat *buf)
-{
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno,
-		buf);
-  return 0;
-}
-
-int32_t
-meta_utimes (call_frame_t *frame,
-	     xlator_t *this,
-	     const char *path,
-	     struct timespec *buf)
-{
-  STACK_WIND (frame,
-	      meta_utimes_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->utimes,
-	      path,
-	      buf);
-  return 0;
-}
-
-
-int32_t
-meta_access_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno)
-{
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno);
-  return 0;
-}
-
-int32_t
-meta_access (call_frame_t *frame,
-	     xlator_t *this,
-	     const char *path,
-	     mode_t mode)
-{
-  STACK_WIND (frame,
-	      meta_access_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->access,
-	      path,
-	      mode);
-  return 0;
-}
-
-int32_t
-meta_readlink_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-		   char *dest)
-{
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno,
-		dest);
-  return 0;
-}
-
-int32_t
-meta_readlink (call_frame_t *frame,
-	       xlator_t *this,
-	       const char *path,
-	       size_t size)
-{
-  STACK_WIND (frame,
-	      meta_readlink_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->readlink,
-	      path,
-	      size);
-  return 0;
-}
-
-int32_t
-meta_mknod_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		struct stat *buf)
-{
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno,
-		buf);
-  return 0;
-}
-
-int32_t
-meta_mknod (call_frame_t *frame,
-	    xlator_t *this,
-	    const char *path,
-	    mode_t mode,
-	    dev_t dev)
-{
-  STACK_WIND (frame,
-	      meta_mknod_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->mknod,
-	      path,
-	      mode,
-	      dev);
-  return 0;
-}
-
-int32_t
-meta_mkdir_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		struct stat *buf)
-{
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno,
-		buf);
-  return 0;
-}
-
-int32_t
-meta_mkdir (call_frame_t *frame,
-	    xlator_t *this,
-	    const char *path,
-	    mode_t mode)
-{
-  STACK_WIND (frame,
-	      meta_mkdir_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->mkdir,
-	      path,
-	      mode);
-  return 0;
-}
-
-int32_t
-meta_unlink_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno)
-{
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno);
-  return 0;
-}
-
-int32_t
-meta_unlink (call_frame_t *frame,
-	     xlator_t *this,
-	     const char *path)
-{
-  STACK_WIND (frame,
-	      meta_unlink_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->unlink,
-	      path);
-  return 0;
-}
-
-int32_t
-meta_rmdir_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno)
-{
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno);
-  return 0;
-}
-
-int32_t
-meta_rmdir (call_frame_t *frame,
-	    xlator_t *this,
-	    const char *path)
-{
-  STACK_WIND (frame,
-	      meta_rmdir_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->rmdir,
-	      path);
-  return 0;
-}
-
-int32_t
-meta_symlink_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  struct stat *buf)
-{
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno,
-		buf);
-  return 0;
-}
-
-int32_t
-meta_symlink (call_frame_t *frame,
-	      xlator_t *this,
-	      const char *oldpath,
-	      const char *newpath)
-{
-  STACK_WIND (frame,
-	      meta_symlink_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->symlink,
-	      oldpath,
-	      newpath);
-  return 0;
-}
+#include "meta-hooks.h"
 
-int32_t
-meta_rename_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno)
+int
+meta_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno);
-  return 0;
-}
+    inode_t *inode = NULL;
 
-int32_t
-meta_rename (call_frame_t *frame,
-	     xlator_t *this,
-	     const char *oldpath,
-	     const char *newpath)
-{
-  STACK_WIND (frame,
-	      meta_rename_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->rename,
-	      oldpath,
-	      newpath);
-  return 0;
-}
+    if (META_HOOK(loc) || IS_META_ROOT_GFID(loc->gfid)) {
+        struct iatt iatt = {};
+        struct iatt parent = {};
 
-int32_t
-meta_link_cbk (call_frame_t *frame,
-	       void *cookie,
-	       xlator_t *this,
-	       int32_t op_ret,
-	       int32_t op_errno,
-	       struct stat *buf)
-{
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno,
-		buf);
-  return 0;
-}
+        meta_root_dir_hook(frame, this, loc, xdata);
 
-int32_t
-meta_link (call_frame_t *frame,
-	   xlator_t *this,
-	   const char *oldpath,
-	   const char *newpath)
-{
-  STACK_WIND (frame,
-	      meta_link_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->link,
-	      oldpath,
-	      newpath);
-  return 0;
-}
+        meta_iatt_fill(&iatt, loc->inode, IA_IFDIR);
+        gf_uuid_parse(META_ROOT_GFID, iatt.ia_gfid);
 
-struct _open_local {
-  const char *path;
-};
+        META_STACK_UNWIND(lookup, frame, 0, 0, loc->inode, &iatt, xdata,
+                          &parent);
+        return 0;
+    }
 
-int32_t
-meta_open_cbk (call_frame_t *frame, void *cookie,
-	       xlator_t *this, int32_t op_ret, int32_t op_errno,
-	       dict_t *ctx, struct stat *buf)
-{
-  struct _open_local *local = frame->local;
-  if (local)
-    dict_set (ctx, this->name, str_to_data (local->path));
-  STACK_UNWIND (frame, op_ret, op_errno, ctx, buf);
-  return 0;
-}
+    if (loc->parent)
+        inode = loc->parent;
+    else
+        inode = loc->inode;
 
-int32_t
-meta_open (call_frame_t *frame, xlator_t *this,
-	   const char *path, int32_t flags, mode_t mode)
-{
-  meta_private_t *priv = (meta_private_t *) this->private;
-  meta_dirent_t *root = priv->tree;
-  meta_dirent_t *file = lookup_meta_entry (root, path, NULL);
+    META_FOP(inode, lookup, frame, this, loc, xdata);
 
-  if (file) {
-    if (file->fops && file->fops->open) {
-            struct _open_local *local = GF_CALLOC (1, sizeof (struct _open_local), gf_meta_mt__open_local);
-      ERR_ABORT (local);
-      local->path = gf_strdup (path);
-      frame->local = local;
-      STACK_WIND (frame, meta_open_cbk,
-		  this, file->fops->open,
-		  path, flags, mode);
-      return 0;
-    }
-    else {
-      dict_t *ctx = get_new_dict ();
-      dict_ref (ctx);
-      dict_set (ctx, this->name, str_to_data (gf_strdup (path)));
-      STACK_UNWIND (frame, 0, 0, ctx, file->stbuf);
-      return 0;
-    }
-  }
-  else {  
-    STACK_WIND (frame, meta_open_cbk,
-		FIRST_CHILD(this), FIRST_CHILD(this)->fops->open,
-		path, flags, mode);
     return 0;
-  }
 }
 
-int32_t
-meta_create (call_frame_t *frame, xlator_t *this,
-	     const char *path, int32_t flags, mode_t mode)
+int
+meta_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+             dict_t *xdata)
 {
-  meta_private_t *priv = (meta_private_t *) this->private;
-  meta_dirent_t *root = priv->tree;
-  meta_dirent_t *file = lookup_meta_entry (root, path, NULL);
+    META_FOP(fd->inode, opendir, frame, this, loc, fd, xdata);
 
-  if (file) {
-    if (file->fops && file->fops->create) {
-            struct _open_local *local = GF_CALLOC (1, sizeof (struct _open_local), gf_meta_mt__open_local);
-      ERR_ABORT (local);
-      local->path = gf_strdup (path);
-      frame->local = local;
-      STACK_WIND (frame, meta_open_cbk,
-		  this, file->fops->create,
-		  path, flags, mode);
-      return 0;
-    }
-    else {
-      STACK_UNWIND (frame, -1, 0, NULL, NULL);
-      return 0;
-    }
-  }
-  else {
-    STACK_WIND (frame, meta_open_cbk,
-		FIRST_CHILD(this), FIRST_CHILD(this)->fops->create,
-		path, flags, mode);
     return 0;
-  }
 }
 
-int32_t
-meta_readv_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno,
-		struct iovec *vector,
-		int32_t count)
+int
+meta_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int flags, fd_t *fd,
+          dict_t *xdata)
 {
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno,
-		vector,
-		count);
-  return 0;
-}
-
-int32_t
-meta_readv (call_frame_t *frame,
-	    xlator_t *this,
-	    dict_t *fd,
-	    size_t size,
-	    off_t offset)
-{
-  meta_private_t *priv = (meta_private_t *) this->private;
-  meta_dirent_t *root = priv->tree;
-  data_t *path_data = dict_get (fd, this->name);
+    META_FOP(fd->inode, open, frame, this, loc, flags, fd, xdata);
 
-  if (path_data) {
-    const char *path = data_to_str (path_data);
-    meta_dirent_t *file = lookup_meta_entry (root, path, NULL);
-  
-    if (file && file->fops && file->fops->readv) {
-      STACK_WIND (frame, meta_readv_cbk, 
-		  this, file->fops->readv,
-		  fd, size, offset);
-      return 0;
-    }
-  }
-  else {
-    STACK_WIND (frame, meta_readv_cbk,
-		FIRST_CHILD(this), FIRST_CHILD(this)->fops->readv,
-		fd, size, offset);
     return 0;
-  }
 }
 
-int32_t
-meta_writev_cbk (call_frame_t *frame, void *cookie,
-		 xlator_t *this, int32_t op_ret,
-		 int32_t op_errno)
+int
+meta_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+           off_t offset, uint32_t flags, dict_t *xdata)
 {
-  STACK_UNWIND (frame, op_ret, op_errno);
-  return 0;
-}
+    META_FOP(fd->inode, readv, frame, this, fd, size, offset, flags, xdata);
 
-int32_t
-meta_writev (call_frame_t *frame, xlator_t *this,
-	     dict_t *fd, 
-	     struct iovec *vector, int32_t count, off_t offset)
-{
-  meta_private_t *priv = (meta_private_t *) this->private;
-  meta_dirent_t *root = priv->tree;
-  data_t *path_data = dict_get (fd, this->name);
-
-  if (path_data) {
-    const char *path = data_to_str (path_data);
-    meta_dirent_t *file = lookup_meta_entry (root, path, NULL);
-  
-    if (file && file->fops && file->fops->writev) {
-      STACK_WIND (frame, meta_writev_cbk, 
-		  this, file->fops->writev,
-		  fd, vector, count, offset);
-      return 0;
-    }
-  }
-  else {
-    STACK_WIND (frame, meta_readv_cbk,
-		FIRST_CHILD(this), FIRST_CHILD(this)->fops->writev,
-		fd, vector, count, offset);
     return 0;
-  }
 }
 
-int32_t
-meta_flush_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno)
+int
+meta_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno);
-  return 0;
-}
+    META_FOP(fd->inode, flush, frame, this, fd, xdata);
 
-int32_t
-meta_flush (call_frame_t *frame,
-	    xlator_t *this,
-	    dict_t *fd)
-{
-  meta_private_t *priv = (meta_private_t *) this->private;
-  meta_dirent_t *root = priv->tree;
-  data_t *path_data = dict_get (fd, this->name);
- 
-  if (path_data) {
-    const char *path = data_to_str (path_data);
-    meta_dirent_t *file = lookup_meta_entry (root, path, NULL);
-  
-    if (file) {
-      if (file->fops && file->fops->flush) {
-	STACK_WIND (frame, meta_flush_cbk,
-		    this, file->fops->flush,
-		    fd);
-	return 0;
-      }
-      else {
-	STACK_UNWIND (frame, 0, 0);
-	return 0;
-      }
-    }
-  }
-  else {
-    STACK_WIND (frame, meta_flush_cbk,
-		FIRST_CHILD(this), FIRST_CHILD(this)->fops->flush,
-		fd);
     return 0;
-  }
-}
-
-int32_t
-meta_release_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno)
-{
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno);
-  return 0;
 }
 
-int32_t
-meta_release (call_frame_t *frame,
-	      xlator_t *this,
-	      dict_t *fd)
+int
+meta_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
 {
-  meta_private_t *priv = (meta_private_t *) this->private;
-  meta_dirent_t *root = priv->tree;
-  data_t *path_data = dict_get (fd, this->name);
+    META_FOP(loc->inode, stat, frame, this, loc, xdata);
 
-  if (path_data) {
-    const char *path = data_to_str (path_data);
-    meta_dirent_t *file = lookup_meta_entry (root, path, NULL);
-  
-    if (file) {
-      dict_unref (fd);
-      STACK_UNWIND (frame, 0, 0);
-      return 0;
-    }
-  }
-  else {
-    STACK_WIND (frame, meta_release_cbk,
-		FIRST_CHILD(this), FIRST_CHILD(this)->fops->release,
-		fd);
     return 0;
-  }
 }
 
-int32_t
-meta_fsync_cbk (call_frame_t *frame,
-		void *cookie,
-		xlator_t *this,
-		int32_t op_ret,
-		int32_t op_errno)
+int
+meta_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
 {
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno);
-  return 0;
-}
+    META_FOP(fd->inode, fstat, frame, this, fd, xdata);
 
-int32_t
-meta_fsync (call_frame_t *frame,
-	    xlator_t *this,
-	    dict_t *fd,
-	    int32_t flags)
-{
-  STACK_WIND (frame,
-	      meta_fsync_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->fsync,
-	      fd,
-	      flags);
-  return 0;
+    return 0;
 }
 
-int32_t
-meta_fgetattr_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-		   struct stat *buf)
+int
+meta_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+             off_t offset, dict_t *xdata)
 {
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno,
-		buf);
-  return 0;
-}
+    META_FOP(fd->inode, readdir, frame, this, fd, size, offset, xdata);
 
-int32_t
-meta_fgetattr (call_frame_t *frame,
-	       xlator_t *this,
-	       dict_t *fd)
-{
-  STACK_WIND (frame,
-	      meta_fgetattr_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->fgetattr,
-	      fd);
-  return 0;
+    return 0;
 }
 
-int32_t
-meta_opendir_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  dict_t *fd)
+int
+meta_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+              off_t offset, dict_t *xdata)
 {
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno,
-		fd);
-  return 0;
-}
+    META_FOP(fd->inode, readdirp, frame, this, fd, size, offset, xdata);
 
-int32_t
-meta_opendir (call_frame_t *frame,
-	      xlator_t *this,
-	      const char *path)
-{
-  meta_private_t *priv = (meta_private_t *) this->private;
-  meta_dirent_t *root = priv->tree;
-  meta_dirent_t *dir = lookup_meta_entry (root, path, NULL);
-  
-  if (dir) {
-    dict_t *ctx = get_new_dict ();
-    dict_set (ctx, this->name, str_to_data (gf_strdup (path)));
-    STACK_UNWIND (frame, 0, 0, ctx);
-    return 0;
-  }
-  else {  
-    STACK_WIND (frame, meta_opendir_cbk,
-		FIRST_CHILD(this), FIRST_CHILD(this)->fops->opendir,
-		path);
     return 0;
-  }
 }
 
-int32_t
-meta_readdir_cbk (call_frame_t *frame,
-		  void *cookie,
-		  xlator_t *this,
-		  int32_t op_ret,
-		  int32_t op_errno,
-		  dir_entry_t *entries,
-		  int32_t count)
+int
+meta_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+              dict_t *xdata)
 {
-  meta_private_t *priv = (meta_private_t *)this->private;
-
-  if ((int) cookie == 1) {
-    dir_entry_t *dir = GF_CALLOC (1, sizeof (dir_entry_t),
-                                  gf_meta_mt_dir_entry_t);
-    ERR_ABORT (dir);
+    META_FOP(loc->inode, readlink, frame, this, loc, size, xdata);
 
-    dir->name = gf_strdup (".meta");
-    memcpy (&dir->buf, priv->tree->stbuf, sizeof (struct stat));
-    dir->next = entries->next;
-    entries->next = dir;
-
-    STACK_UNWIND (frame, op_ret, op_errno, entries, count+1);
     return 0;
-  }
-  
-  STACK_UNWIND (frame, op_ret, op_errno, entries, count);
-  return 0;
 }
 
-int32_t
-meta_readdir (call_frame_t *frame,
-	      xlator_t *this,
-	      const char *path)
+int
+meta_writev(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *iov,
+            int count, off_t offset, uint32_t flags, struct iobref *iobref,
+            dict_t *xdata)
 {
-  meta_private_t *priv = (meta_private_t *) this->private;
-  meta_dirent_t *root = priv->tree;
-
-  meta_dirent_t *dir = lookup_meta_entry (root, path, NULL);
-  if (dir) {
-    if (dir->fops && dir->fops->readdir) {
-      STACK_WIND (frame, meta_readdir_cbk, 
-		  this, dir->fops->readdir, path);
-      return 0;
-    }
-    else {
-      int count = 0;
-      dir = dir->children;
-      dir_entry_t *entries = NULL;
-
-      while (dir) {
-        dir_entry_t *d = GF_CALLOC (1, sizeof (dir_entry_t),
-                                    gf_meta_mt_dir_entry_t);
-	ERR_ABORT (d);
-	d->name = dir->name;
-	d->buf  = *dir->stbuf;
-	d->next = entries;
-	entries = d;
-	count++;
-	dir = dir->next;
-      }
-
-      dir_entry_t *header = GF_CALLOC (1, sizeof (dir_entry_t),
-                                       gf_meta_mt_dir_entry_t);
-      ERR_ABORT (header);
-      header->next = entries;
-      STACK_UNWIND (frame, 0, 0, header, count);
-      return 0;
-    }
-  }
-  else {
-    if (!strcmp (path, "/")) {
-      STACK_WIND_COOKIE (frame, meta_readdir_cbk, 
-		   (int) 1, /* cookie to tell _cbk to add .meta entry */
-		   FIRST_CHILD(this), FIRST_CHILD(this)->fops->readdir,
-		   path);
-    }
-    else {
-      STACK_WIND (frame, meta_readdir_cbk, 
-		  FIRST_CHILD(this), FIRST_CHILD(this)->fops->readdir,
-		  path);
-    }
-  }
-  return 0;
+    META_FOP(fd->inode, writev, frame, this, fd, iov, count, offset, flags,
+             iobref, xdata);
+    return 0;
 }
 
-int32_t
-meta_releasedir_cbk (call_frame_t *frame,
-		     void *cookie,
-		     xlator_t *this,
-		     int32_t op_ret,
-		     int32_t op_errno)
+int
+meta_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+              dict_t *xdata)
 {
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno);
-  return 0;
-}
+    META_FOP(loc->inode, truncate, frame, this, loc, offset, xdata);
 
-int32_t
-meta_releasedir (call_frame_t *frame,
-		 xlator_t *this,
-		 dict_t *fd)
-{
-  STACK_WIND (frame,
-	      meta_releasedir_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->releasedir,
-	      fd);
-  return 0;
+    return 0;
 }
 
-int32_t
-meta_fsyncdir_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno)
+int
+meta_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+               dict_t *xdata)
 {
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno);
-  return 0;
-}
+    META_FOP(fd->inode, ftruncate, frame, this, fd, offset, xdata);
 
-int32_t
-meta_fsyncdir (call_frame_t *frame,
-	       xlator_t *this,
-	       dict_t *fd,
-	       int32_t flags)
-{
-  STACK_WIND (frame,
-	      meta_fsyncdir_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->fsyncdir,
-	      fd,
-	      flags);
-  return 0;
+    return 0;
 }
 
 int32_t
-meta_statfs_cbk (call_frame_t *frame,
-		 void *cookie,
-		 xlator_t *this,
-		 int32_t op_ret,
-		 int32_t op_errno,
-		 struct statvfs *buf)
+meta_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+           dict_t *xdata)
 {
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno,
-		buf);
-  return 0;
-}
+    META_FOP(fd->inode, fsync, frame, this, fd, flags, xdata);
 
-int32_t
-meta_statfs (call_frame_t *frame,
-	     xlator_t *this,
-	     const char *path)
-{
-  STACK_WIND (frame,
-	      meta_statfs_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->statfs,
-	      path);
-  return 0;
+    return 0;
 }
 
 int32_t
-meta_setxattr_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno)
+meta_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t flags,
+              dict_t *xdata)
 {
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno);
-  return 0;
-}
+    META_FOP(fd->inode, fsyncdir, frame, this, fd, flags, xdata);
 
-int32_t
-meta_setxattr (call_frame_t *frame,
-	       xlator_t *this,
-	       const char *path,
-	       const char *name,
-	       const char *value,
-	       size_t size,
-	       int32_t flags)
-{
-  STACK_WIND (frame,
-	      meta_setxattr_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->setxattr,
-	      path,
-	      name,
-	      value,
-	      size,
-	      flags);
-  return 0;
+    return 0;
 }
 
-int32_t
-meta_getxattr_cbk (call_frame_t *frame,
-		   void *cookie,
-		   xlator_t *this,
-		   int32_t op_ret,
-		   int32_t op_errno,
-		   char *value)
+int
+meta_forget(xlator_t *this, inode_t *inode)
 {
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno,
-		value);
-  return 0;
+    return 0;
 }
 
-int32_t
-meta_getxattr (call_frame_t *frame,
-	       xlator_t *this,
-	       const char *path,
-	       const char *name,
-	       size_t size)
+int
+meta_release(xlator_t *this, fd_t *fd)
 {
-  STACK_WIND (frame,
-	      meta_getxattr_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->getxattr,
-	      path,
-	      name,
-	      size);
-  return 0;
+    return meta_fd_release(fd, this);
 }
 
-int32_t
-meta_listxattr_cbk (call_frame_t *frame,
-		    void *cookie,
-		    xlator_t *this,
-		    int32_t op_ret,
-		    int32_t op_errno,
-		    char *value)
+int
+meta_releasedir(xlator_t *this, fd_t *fd)
 {
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno,
-		value);
-  return 0;
+    return meta_fd_release(fd, this);
 }
 
-int32_t
-meta_listxattr (call_frame_t *frame,
-		xlator_t *this,
-		const char *path,
-		size_t size)
+int
+mem_acct_init(xlator_t *this)
 {
-  STACK_WIND (frame,
-	      meta_listxattr_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->listxattr,
-	      path,
-	      size);
-  return 0;
-}
+    int ret = -1;
 
-int32_t
-meta_removexattr_cbk (call_frame_t *frame,
-		      void *cookie,
-		      xlator_t *this,
-		      int32_t op_ret,
-		      int32_t op_errno)
-{
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno);
-  return 0;
-}
+    if (!this)
+        return ret;
 
-int32_t
-meta_removexattr (call_frame_t *frame,
-		  xlator_t *this,
-		  const char *path,
-		  const char *name)
-{
-  STACK_WIND (frame,
-	      meta_removexattr_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->removexattr,
-	      path,
-	      name);
-  return 0;
-}
+    ret = xlator_mem_acct_init(this, gf_meta_mt_end + 1);
 
-int32_t
-meta_lk_cbk (call_frame_t *frame,
-	     void *cookie,
-	     xlator_t *this,
-	     int32_t op_ret,
-	     int32_t op_errno,
-	     struct gf_flock *lock)
-{
-  STACK_UNWIND (frame,
-		op_ret,
-		op_errno,
-		lock);
-  return 0;
-}
+    if (ret != 0) {
+        gf_log(this->name, GF_LOG_ERROR, "Memory accounting init failed");
+        return ret;
+    }
 
-int32_t
-meta_lk (call_frame_t *frame,
-	 xlator_t *this,
-	 dict_t *file,
-	 int32_t cmd,
-	 struct gf_flock *lock)
-{
-  STACK_WIND (frame,
-	      meta_lk_cbk,
-	      FIRST_CHILD(this),
-	      FIRST_CHILD(this)->fops->lk,
-	      file,
-	      cmd,
-	      lock);
-  return 0;
+    return ret;
 }
 
-static void
-add_xlator_to_tree (meta_dirent_t *tree, xlator_t *this,
-		    const char *prefix)
+int
+init(xlator_t *this)
 {
-  char *dir;
-  gf_asprintf (&dir, "%s/%s", prefix, this->name);
-
-  char *children;
-  gf_asprintf (&children, "%s/%s", dir, "subvolumes");
-
-  char *type;
-  gf_asprintf (&type, "%s/%s", dir, "type");
-
-  char *view;
-  gf_asprintf (&view, "%s/%s", dir, "view");
-
-  insert_meta_entry (tree, dir, S_IFDIR, NULL, NULL);
-  insert_meta_entry (tree, children, S_IFDIR, NULL, NULL);
-  meta_dirent_t *v = insert_meta_entry (tree, view, S_IFDIR, NULL, 
-					&meta_xlator_view_fops);
-  v->view_xlator = this;
-  meta_dirent_t *t = insert_meta_entry (tree, type, S_IFREG, NULL, 
-					&meta_xlator_type_fops);
-  t->view_xlator = this;
-
-  xlator_list_t *trav = this->children;
-  while (trav) {
-    add_xlator_to_tree (tree, trav->xlator, children);
-    trav = trav->next;
-  }
-}
+    meta_priv_t *priv = NULL;
+    int ret = -1;
 
-static void
-build_meta_tree (xlator_t *this)
-{
-  meta_private_t *priv = (meta_private_t *) this->private;
-  priv->tree = GF_CALLOC (1, sizeof (meta_dirent_t),
-                          gf_meta_mt_meta_dirent_t);
-  ERR_ABORT (priv->tree);
-  priv->tree->name = gf_strdup (".meta");
-  priv->tree->stbuf = new_stbuf ();
-  priv->tree->stbuf->st_mode = S_IFDIR | S_IRUSR | S_IRGRP | S_IROTH |
-    S_IXUSR | S_IXGRP | S_IXOTH;
+    priv = GF_CALLOC(sizeof(*priv), 1, gf_meta_mt_priv_t);
+    if (!priv)
+        return ret;
 
-  insert_meta_entry (priv->tree, "/.meta/version", 
-		     S_IFREG, NULL, &meta_version_fops);
+    GF_OPTION_INIT("meta-dir-name", priv->meta_dir_name, str, out);
 
-  insert_meta_entry (priv->tree, "/.meta/xlators",
-		     S_IFDIR, NULL, NULL);
+    this->private = priv;
+    ret = 0;
+out:
+    if (ret)
+        GF_FREE(priv);
 
-  xlator_list_t *trav = this->children;
-  while (trav) {
-    add_xlator_to_tree (priv->tree, trav->xlator, "/.meta/xlators");
-    trav = trav->next;
-  }
+    return ret;
 }
 
-int32_t
-mem_acct_init (xlator_t *this)
+void
+fini(xlator_t *this)
 {
-        int     ret = -1;
-
-        if (!this)
-                return ret;
-
-        ret = xlator_mem_acct_init (this, gf_meta_mt_end + 1);
-        
-        if (ret != 0) {
-                gf_log(this->name, GF_LOG_ERROR, "Memory accounting init"
-                       "failed");
-                return ret;
-        }
-
-        return ret;
+    GF_FREE(this->private);
+    return;
 }
 
-int32_t
-init (xlator_t *this)
-{
-  if (this->parent != NULL) {
-    gf_log ("meta", GF_LOG_ERROR, "FATAL: meta should be the root of the xlator tree");
-    return -1;
-  }
-  
-  meta_private_t *priv = GF_CALLOC (1, sizeof (meta_private_t),
-                                    gf_meta_mt_meta_private_t);
-  ERR_ABORT (priv);
-  
-  data_t *directory = dict_get (this->options, "directory");
-  if (directory) {
-    priv->directory = gf_strdup (data_to_str (directory));
-  }
-  else {
-    priv->directory = ".meta";
-  }
-  
-  this->private = priv;
-  build_meta_tree (this);
+struct xlator_fops fops = {.lookup = meta_lookup,
+                           .opendir = meta_opendir,
+                           .open = meta_open,
+                           .readv = meta_readv,
+                           .flush = meta_flush,
+                           .stat = meta_stat,
+                           .fstat = meta_fstat,
+                           .readdir = meta_readdir,
+                           .readdirp = meta_readdirp,
+                           .readlink = meta_readlink,
+                           .writev = meta_writev,
+                           .truncate = meta_truncate,
+                           .ftruncate = meta_ftruncate,
+                           .fsync = meta_fsync,
+                           .fsyncdir = meta_fsyncdir};
 
-  return 0;
-}
+struct xlator_cbks cbks = {
+    .forget = meta_forget,
+    .release = meta_release,
+    .releasedir = meta_releasedir,
+};
 
-int32_t
-fini (xlator_t *this)
-{
-  return 0;
-}
+struct volume_options options[] = {
+    {.key = {"meta-dir-name"},
+     .type = GF_OPTION_TYPE_STR,
+     .default_value = DEFAULT_META_DIR_NAME,
+     .description = "Name of default meta directory."},
+    {.key = {NULL}},
+};
 
-struct xlator_fops fops = {
-  .getattr     = meta_getattr,
-  .readlink    = meta_readlink,
-  .mknod       = meta_mknod,
-  .mkdir       = meta_mkdir,
-  .unlink      = meta_unlink,
-  .rmdir       = meta_rmdir,
-  .symlink     = meta_symlink,
-  .rename      = meta_rename,
-  .link        = meta_link,
-  .chmod       = meta_chmod,
-  .chown       = meta_chown,
-  .truncate    = meta_truncate,
-  .utimes      = meta_utimes,
-  .open        = meta_open,
-  .readv       = meta_readv,
-  .writev      = meta_writev,
-  .statfs      = meta_statfs,
-  .flush       = meta_flush,
-  .release     = meta_release,
-  .fsync       = meta_fsync,
-  .setxattr    = meta_setxattr,
-  .getxattr    = meta_getxattr,
-  .listxattr   = meta_listxattr,
-  .removexattr = meta_removexattr,
-  .opendir     = meta_opendir,
-  .readdir     = meta_readdir,
-  .releasedir  = meta_releasedir,
-  .fsyncdir    = meta_fsyncdir,
-  .access      = meta_access,
-  .ftruncate   = meta_ftruncate,
-  .fgetattr    = meta_fgetattr,
-  .create      = meta_create,
-  .lk          = meta_lk,
+xlator_api_t xlator_api = {
+    .init = init,
+    .fini = fini,
+    .mem_acct_init = mem_acct_init,
+    .op_version = {1}, /* Present from the initial version */
+    .fops = &fops,
+    .cbks = &cbks,
+    .options = options,
+    .identifier = "meta",
+    .category = GF_TECH_PREVIEW,
 };
diff --git a/xlators/meta/src/meta.h b/xlators/meta/src/meta.h
index b5bf8a6d6e5..7f0cf28808a 100644
--- a/xlators/meta/src/meta.h
+++ b/xlators/meta/src/meta.h
@@ -1,48 +1,140 @@
 /*
-   Copyright (c) 2006-2009 Gluster, Inc. <http://www.gluster.com>
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
    This file is part of GlusterFS.
 
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
 */
-
 #ifndef __META_H__
 #define __META_H__
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-struct _meta_dirent {
-  const char *name;
-  int type;
-  struct _meta_dirent *children;
-  struct _meta_dirent *parent;
-  struct _meta_dirent *next;
-  struct stat *stbuf;
-  xlator_t *view_xlator;
-  struct xlator_fops *fops;
+#include <glusterfs/strfd.h>
+
+#define DEFAULT_META_DIR_NAME ".meta"
+
+#define META_ROOT_GFID "ba926388-bb9c-4eec-ad60-79dba4cc083a"
+
+#define IS_META_ROOT_GFID(g) (strcmp(uuid_utoa(g), META_ROOT_GFID) == 0)
+
+typedef int (*meta_hook_t)(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                           dict_t *xdata);
+
+typedef struct {
+    dict_t *xdata;
+} meta_local_t;
+
+typedef struct {
+    char *meta_dir_name;
+} meta_priv_t;
+
+struct meta_dirent {
+    const char *name;
+    ia_type_t type;
+    meta_hook_t hook;
+};
+
+#define DOT_DOTDOT                                                             \
+    {.name = ".", .type = IA_IFDIR}, { .name = "..", .type = IA_IFDIR }
+
+struct meta_ops {
+    struct meta_dirent *fixed_dirents;
+    int (*dir_fill)(xlator_t *this, inode_t *dir, struct meta_dirent **entries);
+    int (*file_fill)(xlator_t *this, inode_t *file, strfd_t *strfd);
+    int (*iatt_fill)(xlator_t *this, inode_t *inode, struct iatt *iatt);
+    int (*link_fill)(xlator_t *this, inode_t *inode, strfd_t *strfd);
+    int (*file_write)(xlator_t *this, fd_t *fd, struct iovec *iov, int count);
+    struct xlator_fops fops;
+    struct xlator_cbks cbks;
 };
-typedef struct _meta_dirent meta_dirent_t;
 
 typedef struct {
-  const char *directory;
-  meta_dirent_t *tree;
-} meta_private_t;
+    char *data;
+    struct meta_dirent *dirents;
+    size_t size;
+} meta_fd_t;
+
+#define COUNT(arr) (sizeof(arr) / sizeof(arr[0]))
+
+#define META_HOOK(loc)                                                         \
+    (__is_root_gfid(loc->pargfid) &&                                           \
+     !strcmp(loc->name, META_PRIV(THIS)->meta_dir_name))
+
+#define META_PRIV(t) ((meta_priv_t *)(t->private))
+
+#define META_STACK_UNWIND(fop, frame, params...)                               \
+    do {                                                                       \
+        meta_local_t *__local = NULL;                                          \
+        xlator_t *__this = NULL;                                               \
+        if (frame) {                                                           \
+            __local = frame->local;                                            \
+            __this = frame->this;                                              \
+            frame->local = NULL;                                               \
+        }                                                                      \
+        STACK_UNWIND_STRICT(fop, frame, params);                               \
+        if (__local) {                                                         \
+            meta_local_cleanup(__local, __this);                               \
+        }                                                                      \
+    } while (0)
+
+#define META_FOP(i, fop, fr, t, params...)                                     \
+    {                                                                          \
+        struct xlator_fops *_fops = NULL;                                      \
+                                                                               \
+        _fops = meta_fops_get(i, t);                                           \
+                                                                               \
+        _fops->fop(fr, t, params);                                             \
+    }                                                                          \
+    while (0)
+
+void
+meta_iatt_fill(struct iatt *iatt, inode_t *inode, ia_type_t type);
+
+int
+meta_inode_discover(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                    dict_t *xdata);
+
+int
+meta_ops_set(inode_t *inode, xlator_t *this, struct meta_ops *ops);
+
+struct xlator_fops *
+meta_fops_get(inode_t *inode, xlator_t *this);
+struct xlator_cbks *
+meta_cbks_get(inode_t *inode, xlator_t *this);
+struct meta_ops *
+meta_ops_get(inode_t *inode, xlator_t *this);
+
+int
+meta_ctx_set(inode_t *inode, xlator_t *this, void *ctx);
+
+void *
+meta_ctx_get(inode_t *inode, xlator_t *this);
+
+void
+meta_local_cleanup(meta_local_t *local, xlator_t *this);
+
+struct xlator_fops *
+meta_defaults_init(struct xlator_fops *fops);
+
+meta_fd_t *
+meta_fd_get(fd_t *fd, xlator_t *this);
+
+int
+meta_fd_release(fd_t *fd, xlator_t *this);
+
+dict_t *
+meta_direct_io_mode(dict_t *xdata, call_frame_t *frame);
+
+meta_local_t *
+meta_local(call_frame_t *frame);
+
+int
+meta_file_fill(xlator_t *this, fd_t *fd);
 
-#include "tree.h"
-#include "misc.h"
+int
+meta_dir_fill(xlator_t *this, fd_t *fd);
 
+int
+fixed_dirents_len(struct meta_dirent *dirents);
 #endif /* __META_H__ */
diff --git a/xlators/meta/src/misc.c b/xlators/meta/src/misc.c
deleted file mode 100644
index 76da599c4e8..00000000000
--- a/xlators/meta/src/misc.c
+++ /dev/null
@@ -1,67 +0,0 @@
-/*
-   Copyright (c) 2006-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-#include <unistd.h>
-#include <sys/uio.h>
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "xlator.h"
-#include "meta.h"
-
-#define min(x,y)   ((x) < (y) ? (x) : (y))
-
-/* /.meta/version */
-static const char *version_str = PACKAGE_NAME " " PACKAGE_VERSION "\n";
-
-int32_t
-meta_version_readv (call_frame_t *frame, xlator_t *this,
-		    dict_t *fd, size_t size, off_t offset)
-{
-  static int version_size;
-  version_size = strlen (version_str);
-  
-  struct iovec vec;
-  vec.iov_base = version_str + offset;
-  vec.iov_len  = min (version_size - offset, size);
-
-  STACK_UNWIND (frame, vec.iov_len, 0, &vec, 1);
-  return 0;
-}
-
-int32_t
-meta_version_getattr (call_frame_t *frame, 
-			      xlator_t *this,
-			      const char *path)
-{
-  meta_private_t *priv = (meta_private_t *) this->private;
-  meta_dirent_t *root = priv->tree;
-  meta_dirent_t *file = lookup_meta_entry (root, path, NULL);
-  file->stbuf->st_size = strlen (version_str);
-  STACK_UNWIND (frame, 0, 0, file->stbuf);
-}
-
-struct xlator_fops meta_version_fops = {
-  .readv   = meta_version_readv,
-  .getattr = meta_version_getattr
-};
-
diff --git a/xlators/meta/src/misc.h b/xlators/meta/src/misc.h
deleted file mode 100644
index fd7baced66e..00000000000
--- a/xlators/meta/src/misc.h
+++ /dev/null
@@ -1,31 +0,0 @@
-/*
-   Copyright (c) 2006-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef __MISC_H__
-#define __MISC_H__
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-
-struct xlator_fops meta_version_fops;
-
-#endif /* __MISC_H__ */
diff --git a/xlators/meta/src/name-file.c b/xlators/meta/src/name-file.c
new file mode 100644
index 00000000000..5874a24d78a
--- /dev/null
+++ b/xlators/meta/src/name-file.c
@@ -0,0 +1,44 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include <glusterfs/strfd.h>
+#include <glusterfs/lkowner.h>
+
+static int
+name_file_fill(xlator_t *this, inode_t *file, strfd_t *strfd)
+{
+    xlator_t *xl = NULL;
+
+    xl = meta_ctx_get(file, this);
+
+    strprintf(strfd, "%s\n", xl->name);
+
+    return strfd->size;
+}
+
+static struct meta_ops name_file_ops = {
+    .file_fill = name_file_fill,
+};
+
+int
+meta_name_file_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                    dict_t *xdata)
+{
+    meta_ops_set(loc->inode, this, &name_file_ops);
+
+    meta_ctx_set(loc->inode, this, meta_ctx_get(loc->parent, this));
+
+    return 0;
+}
diff --git a/xlators/meta/src/option-file.c b/xlators/meta/src/option-file.c
new file mode 100644
index 00000000000..ff55eca592f
--- /dev/null
+++ b/xlators/meta/src/option-file.c
@@ -0,0 +1,45 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include "meta-hooks.h"
+
+static int
+option_file_fill(xlator_t *this, inode_t *inode, strfd_t *strfd)
+{
+    data_t *data = NULL;
+
+    data = meta_ctx_get(inode, this);
+
+    strprintf(strfd, "%s\n", data_to_str(data));
+
+    return strfd->size;
+}
+
+static struct meta_ops option_file_ops = {.file_fill = option_file_fill};
+
+int
+meta_option_file_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                      dict_t *xdata)
+{
+    xlator_t *xl = NULL;
+
+    xl = meta_ctx_get(loc->parent, this);
+
+    meta_ctx_set(loc->inode, this, dict_get(xl->options, (char *)loc->name));
+
+    meta_ops_set(loc->inode, this, &option_file_ops);
+
+    return 0;
+}
diff --git a/xlators/meta/src/options-dir.c b/xlators/meta/src/options-dir.c
new file mode 100644
index 00000000000..d68a7eeaffc
--- /dev/null
+++ b/xlators/meta/src/options-dir.c
@@ -0,0 +1,65 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include "meta-hooks.h"
+
+static int
+dict_key_add(dict_t *dict, char *key, data_t *value, void *data)
+{
+    struct meta_dirent **direntp = data;
+
+    (*direntp)->name = gf_strdup(key);
+    (*direntp)->type = IA_IFREG;
+    (*direntp)->hook = meta_option_file_hook;
+
+    (*direntp)++;
+    return 0;
+}
+
+static int
+options_dir_fill(xlator_t *this, inode_t *inode, struct meta_dirent **dp)
+{
+    struct meta_dirent *dirent = NULL;
+    struct meta_dirent *direntp = NULL;
+    xlator_t *xl = NULL;
+
+    xl = meta_ctx_get(inode, this);
+
+    dirent = GF_CALLOC(sizeof(*dirent), xl->options->count,
+                       gf_meta_mt_dirents_t);
+    if (!dirent)
+        return -1;
+
+    direntp = dirent;
+
+    dict_foreach(xl->options, dict_key_add, &direntp);
+
+    *dp = dirent;
+
+    return xl->options->count;
+}
+
+static struct meta_ops options_dir_ops = {.dir_fill = options_dir_fill};
+
+int
+meta_options_dir_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                      dict_t *xdata)
+{
+    meta_ctx_set(loc->inode, this, meta_ctx_get(loc->parent, this));
+
+    meta_ops_set(loc->inode, this, &options_dir_ops);
+
+    return 0;
+}
diff --git a/xlators/meta/src/private-file.c b/xlators/meta/src/private-file.c
new file mode 100644
index 00000000000..23ec319456b
--- /dev/null
+++ b/xlators/meta/src/private-file.c
@@ -0,0 +1,44 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include <glusterfs/strfd.h>
+#include <glusterfs/statedump.h>
+
+static int
+private_file_fill(xlator_t *this, inode_t *file, strfd_t *strfd)
+{
+    xlator_t *xl = NULL;
+
+    xl = meta_ctx_get(file, this);
+
+    gf_proc_dump_xlator_private(xl, strfd);
+
+    return strfd->size;
+}
+
+static struct meta_ops private_file_ops = {
+    .file_fill = private_file_fill,
+};
+
+int
+meta_private_file_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                       dict_t *xdata)
+{
+    meta_ops_set(loc->inode, this, &private_file_ops);
+
+    meta_ctx_set(loc->inode, this, meta_ctx_get(loc->parent, this));
+
+    return 0;
+}
diff --git a/xlators/meta/src/process_uuid-file.c b/xlators/meta/src/process_uuid-file.c
new file mode 100644
index 00000000000..a24c1b57ab3
--- /dev/null
+++ b/xlators/meta/src/process_uuid-file.c
@@ -0,0 +1,37 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include <glusterfs/strfd.h>
+#include <glusterfs/lkowner.h>
+
+static int
+process_uuid_file_fill(xlator_t *this, inode_t *file, strfd_t *strfd)
+{
+    strprintf(strfd, "%s\n", this->ctx->process_uuid);
+    return strfd->size;
+}
+
+static struct meta_ops process_uuid_file_ops = {
+    .file_fill = process_uuid_file_fill,
+};
+
+int
+meta_process_uuid_file_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                            dict_t *xdata)
+{
+    meta_ops_set(loc->inode, this, &process_uuid_file_ops);
+
+    return 0;
+}
diff --git a/xlators/meta/src/profile-file.c b/xlators/meta/src/profile-file.c
new file mode 100644
index 00000000000..829dcb77451
--- /dev/null
+++ b/xlators/meta/src/profile-file.c
@@ -0,0 +1,44 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include <glusterfs/strfd.h>
+#include <glusterfs/statedump.h>
+
+static int
+profile_file_fill(xlator_t *this, inode_t *file, strfd_t *strfd)
+{
+    xlator_t *xl = NULL;
+
+    xl = meta_ctx_get(file, this);
+
+    gf_proc_dump_xlator_profile(xl, strfd);
+
+    return strfd->size;
+}
+
+static struct meta_ops profile_file_ops = {
+    .file_fill = profile_file_fill,
+};
+
+int
+meta_profile_file_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                       dict_t *xdata)
+{
+    meta_ops_set(loc->inode, this, &profile_file_ops);
+
+    meta_ctx_set(loc->inode, this, meta_ctx_get(loc->parent, this));
+
+    return 0;
+}
diff --git a/xlators/meta/src/root-dir.c b/xlators/meta/src/root-dir.c
new file mode 100644
index 00000000000..80292bd3dda
--- /dev/null
+++ b/xlators/meta/src/root-dir.c
@@ -0,0 +1,77 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include "meta-hooks.h"
+
+static struct meta_dirent root_dir_dirents[] = {
+    DOT_DOTDOT,
+
+    {
+        .name = "graphs",
+        .type = IA_IFDIR,
+        .hook = meta_graphs_dir_hook,
+    },
+    {
+        .name = "frames",
+        .type = IA_IFREG,
+        .hook = meta_frames_file_hook,
+    },
+    {
+        .name = "logging",
+        .type = IA_IFDIR,
+        .hook = meta_logging_dir_hook,
+    },
+    {
+        .name = "process_uuid",
+        .type = IA_IFREG,
+        .hook = meta_process_uuid_file_hook,
+    },
+    {
+        .name = "version",
+        .type = IA_IFREG,
+        .hook = meta_version_file_hook,
+    },
+    {
+        .name = "cmdline",
+        .type = IA_IFREG,
+        .hook = meta_cmdline_file_hook,
+    },
+    {
+        .name = "mallinfo",
+        .type = IA_IFREG,
+        .hook = meta_mallinfo_file_hook,
+    },
+    {
+        .name = "master",
+        .type = IA_IFDIR,
+        .hook = meta_master_dir_hook,
+    },
+    {
+        .name = "measure_latency",
+        .type = IA_IFREG,
+        .hook = meta_measure_file_hook,
+    },
+    {.name = NULL}};
+
+static struct meta_ops meta_root_dir_ops = {.fixed_dirents = root_dir_dirents};
+
+int
+meta_root_dir_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   dict_t *xdata)
+{
+    meta_ops_set(loc->inode, this, &meta_root_dir_ops);
+
+    return 0;
+}
diff --git a/xlators/meta/src/subvolume-link.c b/xlators/meta/src/subvolume-link.c
new file mode 100644
index 00000000000..5b1f752efd0
--- /dev/null
+++ b/xlators/meta/src/subvolume-link.c
@@ -0,0 +1,56 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+
+static int
+subvolume_link_fill(xlator_t *this, inode_t *inode, strfd_t *strfd)
+{
+    xlator_t *xl = NULL;
+
+    xl = meta_ctx_get(inode, this);
+
+    strprintf(strfd, "../../%s", xl->name);
+
+    return 0;
+}
+
+struct meta_ops subvolume_link_ops = {.link_fill = subvolume_link_fill};
+
+int
+meta_subvolume_link_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                         dict_t *xdata)
+{
+    int count = 0;
+    int i = 0;
+    xlator_t *xl = NULL;
+    xlator_list_t *subv = NULL;
+    xlator_t *subvol = NULL;
+
+    count = strtol(loc->name, 0, 0);
+    xl = meta_ctx_get(loc->parent, this);
+
+    for (subv = xl->children; subv; subv = subv->next) {
+        if (i == count) {
+            subvol = subv->xlator;
+            break;
+        }
+        i++;
+    }
+
+    meta_ctx_set(loc->inode, this, subvol);
+
+    meta_ops_set(loc->inode, this, &subvolume_link_ops);
+    return 0;
+}
diff --git a/xlators/meta/src/subvolumes-dir.c b/xlators/meta/src/subvolumes-dir.c
new file mode 100644
index 00000000000..3cb170ea1f4
--- /dev/null
+++ b/xlators/meta/src/subvolumes-dir.c
@@ -0,0 +1,62 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include "meta-hooks.h"
+
+static int
+subvolumes_dir_fill(xlator_t *this, inode_t *dir, struct meta_dirent **dp)
+{
+    struct meta_dirent *dirents = NULL;
+    xlator_t *xl = NULL;
+    xlator_list_t *subv = NULL;
+    int i = 0;
+    int count = 0;
+
+    xl = meta_ctx_get(dir, this);
+
+    for (subv = xl->children; subv; subv = subv->next)
+        count++;
+
+    dirents = GF_MALLOC(sizeof(*dirents) * count, gf_meta_mt_dirents_t);
+    if (!dirents)
+        return -1;
+
+    for (subv = xl->children; subv; subv = subv->next) {
+        char num[16] = {};
+        snprintf(num, 16, "%d", i);
+
+        dirents[i].name = gf_strdup(num);
+        dirents[i].type = IA_IFLNK;
+        dirents[i].hook = meta_subvolume_link_hook;
+        i++;
+    }
+
+    *dp = dirents;
+
+    return count;
+}
+
+static struct meta_ops subvolumes_dir_ops = {.dir_fill = subvolumes_dir_fill};
+
+int
+meta_subvolumes_dir_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                         dict_t *xdata)
+{
+    meta_ctx_set(loc->inode, this, meta_ctx_get(loc->parent, this));
+
+    meta_ops_set(loc->inode, this, &subvolumes_dir_ops);
+
+    return 0;
+}
diff --git a/xlators/meta/src/top-link.c b/xlators/meta/src/top-link.c
new file mode 100644
index 00000000000..33f0d407411
--- /dev/null
+++ b/xlators/meta/src/top-link.c
@@ -0,0 +1,40 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+
+static int
+top_link_fill(xlator_t *this, inode_t *inode, strfd_t *strfd)
+{
+    glusterfs_graph_t *graph = NULL;
+
+    graph = meta_ctx_get(inode, this);
+
+    strprintf(strfd, "%s", ((xlator_t *)graph->top)->name);
+
+    return 0;
+}
+
+struct meta_ops top_link_ops = {.link_fill = top_link_fill};
+
+int
+meta_top_link_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   dict_t *xdata)
+{
+    meta_ops_set(loc->inode, this, &top_link_ops);
+
+    meta_ctx_set(loc->inode, this, meta_ctx_get(loc->parent, this));
+
+    return 0;
+}
diff --git a/xlators/meta/src/tree.c b/xlators/meta/src/tree.c
deleted file mode 100644
index 62faad19ed8..00000000000
--- a/xlators/meta/src/tree.c
+++ /dev/null
@@ -1,179 +0,0 @@
-/*
-   Copyright (c) 2006-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include <sys/stat.h>
-#include <sys/types.h>
-#include <unistd.h>
-#include <string.h>
-
-#include "glusterfs.h"
-#include "xlator.h"
-
-#include "meta.h"
-#include "meta-mem-types.h"
-
-static int
-is_meta_path (const char *path)
-{
-  while (*path == '/')
-    path++;
-  if (!strncmp (path, ".meta", strlen (".meta")))
-    return 1;
-  return 0;
-}
-
-struct stat *
-new_stbuf (void)
-{
-  static int next_inode = 0;
-  struct stat *stbuf = GF_CALLOC (1, sizeof (struct stat), gf_meta_mt_stat);
-  
-  ERR_ABORT (stbuf);
-
-  stbuf->st_dev = 0;
-  stbuf->st_ino = next_inode++;
-  stbuf->st_mode = S_IRUSR | S_IRGRP | S_IROTH;
-  stbuf->st_nlink = 1;
-  stbuf->st_uid = 0;
-  stbuf->st_gid = 0;
-  stbuf->st_rdev = 0;
-  stbuf->st_size = 0;
-  stbuf->st_blksize = 0;
-  stbuf->st_blocks = 0;
-  stbuf->st_atime = time (NULL);
-  stbuf->st_atim.tv_nsec = 0;
-  stbuf->st_mtime = stbuf->st_atime;
-  stbuf->st_mtim.tv_nsec = 0;
-  stbuf->st_ctime = stbuf->st_ctime;
-  stbuf->st_ctim.tv_nsec = 0;
-
-  return stbuf;
-}
-
-/* find an entry among the siblings of an entry */
-static meta_dirent_t *
-find_entry (meta_dirent_t *node, const char *dir)
-{
-  meta_dirent_t *trav = node;
-  while (trav) {
-    if (!strcmp (trav->name, dir))
-      return trav;
-    trav = trav->next;
-  }
-  return NULL;
-}
-
-/*
- * Return the meta_dirent_t corresponding to the pathname.
- *
- * If pathname does not exist in the meta tree, try to return
- * its highest parent that does exist. The part of the
- * pathname that is left over is returned in the value-result
- * variable {remain}.
- * For example, for "/.meta/xlators/brick1/view/foo/bar/baz",
- * return the entry for "/.meta/xlators/brick1/view"
- * and set remain to "/bar/baz"
- */
-
-meta_dirent_t *
-lookup_meta_entry (meta_dirent_t *root, const char *path,
-		   char **remain)
-{
-  char *_path = gf_strdup (path);
-
-  if (!is_meta_path (path))
-    return NULL;
-
-  meta_dirent_t *trav = root;
-  char *dir = strtok (_path, "/");
-  dir = strtok (NULL, "/");
-
-  while (dir) {
-    meta_dirent_t *ntrav;
-    ntrav = find_entry (trav->children, dir);
-    if (!ntrav) {
-      /* we have reached bottom of the meta tree. 
-         Unknown dragons lie further below */
-      if (remain) {
-	char *piece = dir;
-	while (piece) {
-	  char *tmp = *remain;
-	  if (*remain)
-	    gf_asprintf (remain, "/%s/%s", *remain, piece);
-	  else
-	    gf_asprintf (remain, "/%s", piece);
-	  if (tmp) GF_FREE (tmp);
-	  piece = strtok (NULL, "/");
-	}
-      }
-      return trav;
-    }
-    dir = strtok (NULL, "/");
-    trav = ntrav;
-  }
-
-  GF_FREE (_path);
-  return trav;
-}
-
-meta_dirent_t *
-insert_meta_entry (meta_dirent_t *root, const char *path,
-		   int type, struct stat *stbuf, struct xlator_fops *fops)
-{
-  if (!is_meta_path (path))
-    return NULL;
-  char *slashpos = strrchr (path, '/');
-  char *dir = strndup (path, slashpos - path);
-  meta_dirent_t *parent = lookup_meta_entry (root, dir, NULL);
-  if (!dir)
-    return NULL;
-
-  meta_dirent_t *new = GF_CALLOC (1, sizeof (meta_dirent_t),
-                                  gf_meta_mt_meta_dirent_t);
-  ERR_ABORT (new);
-  new->name        = gf_strdup (slashpos+1);
-  new->type        = type;
-  new->parent      = parent;
-  new->next        = parent->children;
-  parent->children = new;
-  if (stbuf)
-    new->stbuf     = stbuf;
-  else 
-    new->stbuf     = new_stbuf ();
-
-  new->stbuf->st_mode |= type;
-  new->fops        = fops;
-  return new;
-}
-
-int main (void)
-{
-  meta_dirent_t *root = GF_CALLOC (1, sizeof (meta_dirent_t),
-                                   gf_meta_mt_meta_dirent_t);
-  ERR_ABORT (root);
-  root->name = gf_strdup (".meta");
-
-  insert_meta_entry (root, "/.meta/version", S_IFREG, NULL, NULL);
-  return 0;
-}
diff --git a/xlators/meta/src/tree.h b/xlators/meta/src/tree.h
deleted file mode 100644
index 977f880ac0d..00000000000
--- a/xlators/meta/src/tree.h
+++ /dev/null
@@ -1,35 +0,0 @@
-/*
-   Copyright (c) 2006-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef __TREE_H__
-#define __TREE_H__
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-meta_dirent_t *
-insert_meta_entry (meta_dirent_t *root, const char *path,
-		   int type, struct stat *stbuf, struct xlator_fops *fops);
-meta_dirent_t *
-lookup_meta_entry (meta_dirent_t *root, const char *path, 
-		   char **remain);
-
-#endif /* __TREE_H__ */
diff --git a/xlators/meta/src/type-file.c b/xlators/meta/src/type-file.c
new file mode 100644
index 00000000000..ece342a0b2a
--- /dev/null
+++ b/xlators/meta/src/type-file.c
@@ -0,0 +1,44 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include <glusterfs/strfd.h>
+#include <glusterfs/lkowner.h>
+
+static int
+type_file_fill(xlator_t *this, inode_t *file, strfd_t *strfd)
+{
+    xlator_t *xl = NULL;
+
+    xl = meta_ctx_get(file, this);
+
+    strprintf(strfd, "%s\n", xl->type);
+
+    return strfd->size;
+}
+
+static struct meta_ops type_file_ops = {
+    .file_fill = type_file_fill,
+};
+
+int
+meta_type_file_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                    dict_t *xdata)
+{
+    meta_ops_set(loc->inode, this, &type_file_ops);
+
+    meta_ctx_set(loc->inode, this, meta_ctx_get(loc->parent, this));
+
+    return 0;
+}
diff --git a/xlators/meta/src/version-file.c b/xlators/meta/src/version-file.c
new file mode 100644
index 00000000000..36276fb810a
--- /dev/null
+++ b/xlators/meta/src/version-file.c
@@ -0,0 +1,37 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include <glusterfs/strfd.h>
+#include <glusterfs/lkowner.h>
+
+static int
+version_file_fill(xlator_t *this, inode_t *file, strfd_t *strfd)
+{
+    strprintf(strfd, "{ \n  \"Package Version\": \"%s\"\n}", PACKAGE_VERSION);
+    return strfd->size;
+}
+
+static struct meta_ops version_file_ops = {
+    .file_fill = version_file_fill,
+};
+
+int
+meta_version_file_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                       dict_t *xdata)
+{
+    meta_ops_set(loc->inode, this, &version_file_ops);
+
+    return 0;
+}
diff --git a/xlators/meta/src/view-dir.c b/xlators/meta/src/view-dir.c
new file mode 100644
index 00000000000..30931061567
--- /dev/null
+++ b/xlators/meta/src/view-dir.c
@@ -0,0 +1,33 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include "meta-hooks.h"
+
+static struct meta_dirent view_dir_dirents[] = {DOT_DOTDOT,
+
+                                                {.name = NULL}};
+
+static struct meta_ops view_dir_ops = {.fixed_dirents = view_dir_dirents};
+
+int
+meta_view_dir_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                   dict_t *xdata)
+{
+    meta_ctx_set(loc->inode, this, meta_ctx_get(loc->parent, this));
+
+    meta_ops_set(loc->inode, this, &view_dir_ops);
+
+    return 0;
+}
diff --git a/xlators/meta/src/view.c b/xlators/meta/src/view.c
deleted file mode 100644
index 0fcacc0449a..00000000000
--- a/xlators/meta/src/view.c
+++ /dev/null
@@ -1,258 +0,0 @@
-/*
-   Copyright (c) 2006-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "xlator.h"
-
-#include "meta.h"
-
-/* 
- * This file contains fops for the files and directories in 
- * an xlator directory
- */
-
-/* /.meta/xlators/.../type */
-
-int32_t
-meta_xlator_type_readv (call_frame_t *frame, xlator_t *this,
-			dict_t *fd, size_t size, off_t offset)
-{
-  meta_private_t *priv = (meta_private_t *) this->private;
-  meta_dirent_t *root = priv->tree;
-  data_t *path_data = dict_get (fd, this->name);
-
-  if (path_data) {
-    const char *path = data_to_str (path_data);
-    meta_dirent_t *file = lookup_meta_entry (root, path, NULL);
-    xlator_t *view_xlator = file->view_xlator;
-
-    int type_size;
-    type_size = strlen (view_xlator->type);
-  
-    struct iovec vec;
-    vec.iov_base = view_xlator->type + offset;
-    vec.iov_len  = min (type_size - offset, size);
-
-    STACK_UNWIND (frame, vec.iov_len, 0, &vec, 1);
-    return 0;
-  }
-}
-
-int32_t
-meta_xlator_type_getattr (call_frame_t *frame, 
-			  xlator_t *this,
-			  const char *path)
-{
-  meta_private_t *priv = (meta_private_t *) this->private;
-  meta_dirent_t *root = priv->tree;
-
-  meta_dirent_t *file = lookup_meta_entry (root, path, NULL);
-  xlator_t *view_xlator = file->view_xlator;
-  file->stbuf->st_size = strlen (view_xlator->type);
-
-  STACK_UNWIND (frame, 0, 0, file->stbuf);
-  return 0;
-}
-
-struct xlator_fops meta_xlator_type_fops = {
-  .readv   = meta_xlator_type_readv,
-  .getattr = meta_xlator_type_getattr
-};
-
-/* 
- * fops for the "view" directory
- * {xlator}/view shows the filesystem as it appears
- * to {xlator}
- */
-
-static int32_t
-meta_xlator_view_getattr_cbk (call_frame_t *frame, void *cookie,
-			      xlator_t *this, int32_t op_ret, int32_t op_errno,
-			      struct stat *buf)
-{
-  STACK_UNWIND (frame, op_ret, op_errno, buf);
-  return 0;
-}
-
-int32_t
-meta_xlator_view_getattr (call_frame_t *frame, 
-			  xlator_t *this,
-			  const char *path)
-{
-  meta_private_t *priv = (meta_private_t *) this->private;
-  meta_dirent_t *root = priv->tree;
-  char *op_path = NULL;
-
-  meta_dirent_t *file = lookup_meta_entry (root, path, &op_path);
-
-  if (op_path) {
-    STACK_WIND (frame, meta_xlator_view_getattr_cbk, file->view_xlator,
-		file->view_xlator->fops->getattr,
-		op_path);
-  }
-  else {
-    STACK_UNWIND (frame, 0, 0, file->stbuf);
-  }
-
-  return 0;
-}
-
-static int32_t
-meta_xlator_view_readdir_cbk (call_frame_t *frame, void *cookie,
-			      xlator_t *this, int32_t op_ret, int32_t op_errno,
-			      dir_entry_t *entries, int32_t count)
-{
-  STACK_UNWIND (frame, op_ret, op_errno, entries, count);
-  return 0;
-}
-
-int32_t
-meta_xlator_view_readdir (call_frame_t *frame,
-			  xlator_t *this,
-			  const char *path)
-{
-  meta_private_t *priv = (meta_private_t *) this->private;
-  meta_dirent_t *root = priv->tree;
-  char *op_path = NULL;
-
-  meta_dirent_t *dir = lookup_meta_entry (root, path, &op_path);
-
-  STACK_WIND (frame, meta_xlator_view_readdir_cbk, 
-	      dir->view_xlator, dir->view_xlator->fops->readdir,
-	      op_path ? op_path : "/");
-  return 0;
-}
-
-static int32_t
-meta_xlator_view_open_cbk (call_frame_t *frame, void *cookie,
-			   xlator_t *this, 
-			   int32_t op_ret, int32_t op_errno,
-			   dict_t *ctx, struct stat *buf)
-{
-  STACK_UNWIND (frame, op_ret, op_errno, ctx, buf);
-  return 0;
-}
-
-int32_t
-meta_xlator_view_open (call_frame_t *frame, xlator_t *this,
-		       const char *path, int32_t flags, mode_t mode)
-{
-  meta_private_t *priv = (meta_private_t *) this->private;
-  meta_dirent_t *root = priv->tree;
-  char *op_path = NULL;
-
-  meta_dirent_t *file = lookup_meta_entry (root, path, &op_path);
-  STACK_WIND (frame, meta_xlator_view_open_cbk,
-	      file->view_xlator, file->view_xlator->fops->open,
-	      op_path, flags, mode);
-  return 0;
-}
-
-int32_t
-meta_xlator_view_create (call_frame_t *frame, xlator_t *this,
-			 const char *path, int32_t flags, mode_t mode)
-{
-  meta_private_t *priv = (meta_private_t *) this->private;
-  meta_dirent_t *root = priv->tree;
-  char *op_path = NULL;
-
-  meta_dirent_t *file = lookup_meta_entry (root, path, &op_path);
-  STACK_WIND (frame, meta_xlator_view_open_cbk,
-	      file->view_xlator, file->view_xlator->fops->create,
-	      op_path, flags, mode);
-  return 0;
-}
-
-static int32_t
-meta_xlator_view_readv_cbk (call_frame_t *frame, void *cookie,
-			    xlator_t *this, int32_t op_ret,
-			    int32_t op_errno, struct iovec *vector,
-			    int32_t count)
-{
-  STACK_UNWIND (frame, op_ret, op_errno, vector, count);
-  return 0;
-}
-
-int32_t
-meta_xlator_view_readv (call_frame_t *frame, xlator_t *this,
-			dict_t *fd, size_t size, off_t offset)
-{
-  meta_private_t *priv = (meta_private_t *) this->private;
-  meta_dirent_t *root = priv->tree;
-  data_t *path_data = dict_get (fd, this->name);
-
-  if (path_data) {
-    const char *path = data_to_str (path_data);
-    meta_dirent_t *file = lookup_meta_entry (root, path, NULL);
-
-    STACK_WIND (frame, meta_xlator_view_readv_cbk,
-		file->view_xlator, file->view_xlator->fops->readv,
-		fd, size, offset);
-    return 0;
-  }
-
-  STACK_UNWIND (frame, -1, EBADFD, NULL, 0);
-  return 0;
-}
-
-static int32_t
-meta_xlator_view_writev_cbk (call_frame_t *frame, void *cookie,
-			     xlator_t *this, int32_t op_ret,
-			     int32_t op_errno)
-{
-  STACK_UNWIND (frame, op_ret, op_errno);
-  return 0;
-}
-
-int32_t
-meta_xlator_view_writev (call_frame_t *frame, xlator_t *this,
-			 dict_t *fd, 
-			 struct iovec *vector, int32_t count, off_t offset)
-{
-  meta_private_t *priv = (meta_private_t *) this->private;
-  meta_dirent_t *root = priv->tree;
-  data_t *path_data = dict_get (fd, this->name);
-
-  if (path_data) {
-    const char *path = data_to_str (path_data);
-    meta_dirent_t *file = lookup_meta_entry (root, path, NULL);
-
-    STACK_WIND (frame, meta_xlator_view_writev_cbk,
-		file->view_xlator, file->view_xlator->fops->writev,
-		fd, vector, count, offset);
-    return 0;
-  }
-
-  STACK_UNWIND (frame, -1, EBADFD, NULL, 0);
-  return 0;
-}
-
-struct xlator_fops meta_xlator_view_fops = {
-  .getattr = meta_xlator_view_getattr,
-  .readdir = meta_xlator_view_readdir,
-  .open    = meta_xlator_view_open,
-  .create  = meta_xlator_view_create,
-  .readv   = meta_xlator_view_readv,
-  .writev  = meta_xlator_view_writev
-};
diff --git a/xlators/meta/src/view.h b/xlators/meta/src/view.h
deleted file mode 100644
index 0d0d909f111..00000000000
--- a/xlators/meta/src/view.h
+++ /dev/null
@@ -1,32 +0,0 @@
-/*
-   Copyright (c) 2006-2009 Gluster, Inc. <http://www.gluster.com>
-   This file is part of GlusterFS.
-
-   GlusterFS is free software; you can redistribute it and/or modify
-   it under the terms of the GNU Affero General Public License as published
-   by the Free Software Foundation; either version 3 of the License,
-   or (at your option) any later version.
-
-   GlusterFS is distributed in the hope that it will be useful, but
-   WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Affero General Public License for more details.
-
-   You should have received a copy of the GNU Affero General Public License
-   along with this program.  If not, see
-   <http://www.gnu.org/licenses/>.
-*/
-
-#ifndef __VIEW_H__
-#define __VIEW_H__
-
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-
-struct xlator_fops meta_xlator_type_fops;
-struct xlator_fops meta_xlator_view_fops;
-
-#endif /* __VIEW_H__ */
diff --git a/xlators/meta/src/volfile-file.c b/xlators/meta/src/volfile-file.c
new file mode 100644
index 00000000000..b2e2562ab8b
--- /dev/null
+++ b/xlators/meta/src/volfile-file.c
@@ -0,0 +1,79 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include <glusterfs/strfd.h>
+
+static int
+xldump_options(dict_t *this, char *key, data_t *value, void *strfd)
+{
+    strprintf(strfd, "    option %s %s\n", key, value->data);
+    return 0;
+}
+
+static void
+xldump_subvolumes(xlator_t *this, void *strfd)
+{
+    xlator_list_t *subv = NULL;
+
+    if (!this->children)
+        return;
+
+    strprintf(strfd, "    subvolumes");
+
+    for (subv = this->children; subv; subv = subv->next)
+        strprintf(strfd, " %s", subv->xlator->name);
+
+    strprintf(strfd, "\n");
+}
+
+static void
+xldump(xlator_t *each, void *strfd)
+{
+    strprintf(strfd, "volume %s\n", each->name);
+    strprintf(strfd, "    type %s\n", each->type);
+    dict_foreach(each->options, xldump_options, strfd);
+
+    xldump_subvolumes(each, strfd);
+
+    strprintf(strfd, "end-volume\n");
+    strprintf(strfd, "\n");
+}
+
+static int
+volfile_file_fill(xlator_t *this, inode_t *file, strfd_t *strfd)
+{
+    glusterfs_graph_t *graph = NULL;
+
+    graph = meta_ctx_get(file, this);
+
+    xlator_foreach_depth_first(graph->top, xldump, strfd);
+
+    return strfd->size;
+}
+
+static struct meta_ops volfile_file_ops = {
+    .file_fill = volfile_file_fill,
+};
+
+int
+meta_volfile_file_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                       dict_t *xdata)
+{
+    meta_ops_set(loc->inode, this, &volfile_file_ops);
+
+    meta_ctx_set(loc->inode, this, meta_ctx_get(loc->parent, this));
+
+    return 0;
+}
diff --git a/xlators/meta/src/xlator-dir.c b/xlators/meta/src/xlator-dir.c
new file mode 100644
index 00000000000..86189715790
--- /dev/null
+++ b/xlators/meta/src/xlator-dir.c
@@ -0,0 +1,97 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include <glusterfs/defaults.h>
+
+#include "meta-mem-types.h"
+#include "meta.h"
+#include "meta-hooks.h"
+
+static struct meta_dirent xlator_dir_dirents[] = {
+    DOT_DOTDOT,
+
+    {
+        .name = "view",
+        .type = IA_IFDIR,
+        .hook = meta_view_dir_hook,
+    },
+    {
+        .name = "type",
+        .type = IA_IFREG,
+        .hook = meta_type_file_hook,
+    },
+    {
+        .name = "name",
+        .type = IA_IFREG,
+        .hook = meta_name_file_hook,
+    },
+    {
+        .name = "subvolumes",
+        .type = IA_IFDIR,
+        .hook = meta_subvolumes_dir_hook,
+    },
+    {
+        .name = "options",
+        .type = IA_IFDIR,
+        .hook = meta_options_dir_hook,
+    },
+    {
+        .name = "private",
+        .type = IA_IFREG,
+        .hook = meta_private_file_hook,
+    },
+    {
+        .name = "history",
+        .type = IA_IFREG,
+        .hook = meta_history_file_hook,
+    },
+    {
+        .name = "meminfo",
+        .type = IA_IFREG,
+        .hook = meta_meminfo_file_hook,
+    },
+    {
+        .name = "profile",
+        .type = IA_IFREG,
+        .hook = meta_profile_file_hook,
+    },
+    {.name = NULL}};
+
+static struct meta_ops xlator_dir_ops = {.fixed_dirents = xlator_dir_dirents};
+
+int
+meta_xlator_dir_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                     dict_t *xdata)
+{
+    glusterfs_graph_t *graph = NULL;
+    xlator_t *xl = NULL;
+
+    graph = meta_ctx_get(loc->parent, this);
+
+    xl = xlator_search_by_name(graph->first, loc->name);
+
+    meta_ctx_set(loc->inode, this, xl);
+
+    meta_ops_set(loc->inode, this, &xlator_dir_ops);
+
+    return 0;
+}
+
+int
+meta_master_dir_hook(call_frame_t *frame, xlator_t *this, loc_t *loc,
+                     dict_t *xdata)
+{
+    meta_ctx_set(loc->inode, this, this->ctx->master);
+
+    meta_ops_set(loc->inode, this, &xlator_dir_ops);
+
+    return 0;
+}
diff --git a/xlators/mgmt/glusterd/src/Makefile.am b/xlators/mgmt/glusterd/src/Makefile.am
index 326d8e63e88..685beb42d27 100644
--- a/xlators/mgmt/glusterd/src/Makefile.am
+++ b/xlators/mgmt/glusterd/src/Makefile.am
@@ -1,22 +1,79 @@
+if WITH_SERVER
 xlator_LTLIBRARIES = glusterd.la
+endif
+
 xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/mgmt
-glusterd_la_LDFLAGS = -module -avoidversion
-glusterd_la_SOURCES = glusterd.c glusterd-handler.c glusterd-sm.c glusterd-op-sm.c \
-	glusterd-utils.c glusterd3_1-mops.c glusterd-store.c glusterd-handshake.c \
-	glusterd-pmap.c glusterd-volgen.c glusterd-rebalance.c
+glusterd_la_CPPFLAGS = $(AM_CPPFLAGS) \
+	-DFILTERDIR=\"$(libdir)/glusterfs/$(PACKAGE_VERSION)/filter\" \
+	-DXLATORDIR=\"$(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator\" \
+	-I$(top_srcdir)/libglusterd/src/
+
+glusterd_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
+glusterd_la_SOURCES = glusterd.c glusterd-handler.c glusterd-sm.c \
+	glusterd-op-sm.c glusterd-utils.c glusterd-rpc-ops.c \
+	glusterd-store.c glusterd-handshake.c glusterd-pmap.c \
+	glusterd-volgen.c glusterd-rebalance.c \
+	glusterd-quota.c glusterd-bitrot.c glusterd-geo-rep.c \
+	glusterd-replace-brick.c glusterd-log-ops.c \
+	glusterd-volume-ops.c glusterd-brick-ops.c glusterd-mountbroker.c \
+	glusterd-syncop.c glusterd-hooks.c glusterd-volume-set.c \
+	glusterd-locks.c glusterd-snapshot.c glusterd-mgmt-handler.c \
+	glusterd-mgmt.c glusterd-peer-utils.c glusterd-statedump.c \
+	glusterd-snapshot-utils.c glusterd-conn-mgmt.c \
+	glusterd-proc-mgmt.c glusterd-svc-mgmt.c \
+	glusterd-nfs-svc.c glusterd-quotad-svc.c glusterd-svc-helper.c \
+	glusterd-conn-helper.c glusterd-snapd-svc.c glusterd-snapd-svc-helper.c \
+	glusterd-bitd-svc.c glusterd-scrub-svc.c glusterd-server-quorum.c \
+	glusterd-reset-brick.c glusterd-shd-svc.c glusterd-shd-svc-helper.c \
+        glusterd-gfproxyd-svc.c glusterd-gfproxyd-svc-helper.c glusterd-ganesha.c \
+	$(CONTRIBDIR)/mount/mntent.c
+
+glusterd_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la \
+	$(top_builddir)/libglusterd/src/libglusterd.la \
+	$(top_builddir)/rpc/xdr/src/libgfxdr.la \
+	$(top_builddir)/rpc/rpc-lib/src/libgfrpc.la \
+	$(XML_LIBS) -lcrypto $(URCU_LIBS) $(URCU_CDS_LIBS) $(LIB_DL) $(GF_XLATOR_MGNT_LIBADD)
 
-glusterd_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la\
-		     $(top_builddir)/rpc/xdr/src/libgfxdr.la\
-		     $(top_builddir)/rpc/rpc-lib/src/libgfrpc.la
+noinst_HEADERS = glusterd.h glusterd-utils.h glusterd-op-sm.h \
+	glusterd-sm.h glusterd-store.h glusterd-mem-types.h \
+	glusterd-pmap.h glusterd-volgen.h glusterd-mountbroker.h \
+	glusterd-syncop.h glusterd-hooks.h glusterd-locks.h glusterd-quota.h \
+	glusterd-mgmt.h glusterd-messages.h glusterd-peer-utils.h \
+	glusterd-statedump.h glusterd-snapshot-utils.h glusterd-geo-rep.h \
+	glusterd-conn-mgmt.h glusterd-conn-helper.h glusterd-proc-mgmt.h \
+	glusterd-svc-mgmt.h glusterd-nfs-svc.h \
+	glusterd-quotad-svc.h glusterd-svc-helper.h glusterd-snapd-svc.h \
+	glusterd-snapd-svc-helper.h glusterd-rcu.h glusterd-bitd-svc.h \
+	glusterd-scrub-svc.h glusterd-server-quorum.h glusterd-errno.h \
+        glusterd-shd-svc.h glusterd-shd-svc-helper.h \
+        glusterd-gfproxyd-svc.h glusterd-gfproxyd-svc-helper.h \
+	$(CONTRIBDIR)/userspace-rcu/rculist-extra.h \
+	$(CONTRIBDIR)/mount/mntent_compat.h
 
-noinst_HEADERS = glusterd.h glusterd-utils.h glusterd-op-sm.h glusterd-sm.h \
-	glusterd-store.h glusterd-mem-types.h glusterd-pmap.h glusterd-volgen.h
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+	-I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src \
+	-I$(CONTRIBDIR)/rbtree -I$(top_srcdir)/rpc/rpc-lib/src \
+	-I$(CONTRIBDIR)/mount -I$(CONTRIBDIR)/userspace-rcu \
+	-DSBIN_DIR=\"$(sbindir)\" -DDATADIR=\"$(localstatedir)\" \
+	-DGSYNCD_PREFIX=\"$(GLUSTERFS_LIBEXECDIR)\" \
+	-DCONFDIR=\"$(localstatedir)/run/gluster/shared_storage/nfs-ganesha\" \
+	-DGANESHA_PREFIX=\"$(libexecdir)/ganesha\" \
+	-DSYNCDAEMON_COMPILE=$(SYNCDAEMON_COMPILE) \
+	-I$(top_srcdir)/libglusterd/src/
 
-AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS)\
-	-I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS)\
-	-I$(rpclibdir) -L$(xlatordir)/ -I$(CONTRIBDIR)/rbtree -I$(top_srcdir)/rpc/xdr/src\
-	-I$(top_srcdir)/rpc/rpc-lib/src -I$(CONTRIBDIR)/uuid -DGFS_PREFIX=\"$(prefix)\" \
-	-DDATADIR=\"$(localstatedir)\"
 
+AM_CFLAGS = -Wall $(GF_CFLAGS) $(URCU_CFLAGS) $(URCU_CDS_CFLAGS) $(XML_CFLAGS)
+
+AM_LDFLAGS = -L$(xlatordir) $(URCU_LIBS) $(URCU_CDS_LIBS)
 
 CLEANFILES =
+
+install-data-hook:
+if WITH_SERVER
+if GF_INSTALL_GLUSTERD_WORKDIR
+	$(mkdir_p) $(DESTDIR)$(GLUSTERD_WORKDIR)
+	(stat $(DESTDIR)$(sysconfdir)/glusterd && \
+	    mv $(DESTDIR)$(sysconfdir)/glusterd $(DESTDIR)$(GLUSTERD_WORKDIR)) || true;
+	(ln -sf $(DESTDIR)$(GLUSTERD_WORKDIR) $(sysconfdir)/glusterd) || true;
+endif
+endif
diff --git a/xlators/mgmt/glusterd/src/glusterd-bitd-svc.c b/xlators/mgmt/glusterd/src/glusterd-bitd-svc.c
new file mode 100644
index 00000000000..6adb799b18f
--- /dev/null
+++ b/xlators/mgmt/glusterd/src/glusterd-bitd-svc.c
@@ -0,0 +1,206 @@
+/*
+  Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/globals.h>
+#include <glusterfs/run.h>
+#include "glusterd.h"
+#include "glusterd-utils.h"
+#include "glusterd-volgen.h"
+#include "glusterd-bitd-svc.h"
+#include "glusterd-svc-helper.h"
+
+void
+glusterd_bitdsvc_build(glusterd_svc_t *svc)
+{
+    svc->manager = glusterd_bitdsvc_manager;
+    svc->start = glusterd_bitdsvc_start;
+    svc->stop = glusterd_bitdsvc_stop;
+}
+
+int
+glusterd_bitdsvc_init(glusterd_svc_t *svc)
+{
+    return glusterd_svc_init(svc, bitd_svc_name);
+}
+
+static int
+glusterd_bitdsvc_create_volfile()
+{
+    char filepath[PATH_MAX] = {
+        0,
+    };
+    int ret = -1;
+    glusterd_conf_t *conf = NULL;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    conf = this->private;
+    GF_ASSERT(conf);
+
+    glusterd_svc_build_volfile_path(bitd_svc_name, conf->workdir, filepath,
+                                    sizeof(filepath));
+
+    ret = glusterd_create_global_volfile(build_bitd_graph, filepath, NULL);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_VOLFILE_CREATE_FAIL,
+               "Failed to create volfile");
+        goto out;
+    }
+
+out:
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+
+    return ret;
+}
+
+int
+glusterd_bitdsvc_manager(glusterd_svc_t *svc, void *data, int flags)
+{
+    int ret = 0;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    if (!svc->inited) {
+        ret = glusterd_bitdsvc_init(svc);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_BITD_INIT_FAIL,
+                   "Failed to init "
+                   "bitd service");
+            goto out;
+        } else {
+            svc->inited = _gf_true;
+            gf_msg_debug(this->name, 0,
+                         "BitD service "
+                         "initialized");
+        }
+    }
+
+    if (glusterd_should_i_stop_bitd()) {
+        ret = svc->stop(svc, SIGTERM);
+    } else {
+        ret = glusterd_bitdsvc_create_volfile();
+        if (ret)
+            goto out;
+
+        ret = svc->stop(svc, SIGKILL);
+        if (ret)
+            goto out;
+
+        ret = svc->start(svc, flags);
+        if (ret)
+            goto out;
+
+        ret = glusterd_conn_connect(&(svc->conn));
+        if (ret)
+            goto out;
+    }
+
+out:
+    if (ret)
+        gf_event(EVENT_SVC_MANAGER_FAILED, "svc_name=%s", svc->name);
+
+    gf_msg_debug(THIS->name, 0, "Returning %d", ret);
+
+    return ret;
+}
+
+int
+glusterd_bitdsvc_start(glusterd_svc_t *svc, int flags)
+{
+    int ret = -1;
+    dict_t *cmdict = NULL;
+
+    cmdict = dict_new();
+    if (!cmdict)
+        goto error_return;
+
+    ret = dict_set_str(cmdict, "cmdarg0", "--global-timer-wheel");
+    if (ret)
+        goto dealloc_dict;
+
+    ret = glusterd_svc_start(svc, flags, cmdict);
+
+dealloc_dict:
+    dict_unref(cmdict);
+error_return:
+    return ret;
+}
+
+int
+glusterd_bitdsvc_stop(glusterd_svc_t *svc, int sig)
+{
+    return glusterd_svc_stop(svc, sig);
+}
+
+int
+glusterd_bitdsvc_reconfigure()
+{
+    int ret = -1;
+    xlator_t *this = NULL;
+    glusterd_conf_t *priv = NULL;
+    gf_boolean_t identical = _gf_false;
+
+    this = THIS;
+    GF_VALIDATE_OR_GOTO("glusterd", this, out);
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, priv, out);
+
+    if (glusterd_should_i_stop_bitd())
+        goto manager;
+    /*
+     * Check both OLD and NEW volfiles, if they are SAME by size
+     * and cksum i.e. "character-by-character". If YES, then
+     * NOTHING has been changed, just return.
+     */
+    ret = glusterd_svc_check_volfile_identical(priv->bitd_svc.name,
+                                               build_bitd_graph, &identical);
+    if (ret)
+        goto out;
+    if (identical) {
+        ret = 0;
+        goto out;
+    }
+
+    /*
+     * They are not identical. Find out if the topology is changed
+     * OR just the volume options. If just the options which got
+     * changed, then inform the xlator to reconfigure the options.
+     */
+    identical = _gf_false; /* RESET the FLAG */
+    ret = glusterd_svc_check_topology_identical(priv->bitd_svc.name,
+                                                build_bitd_graph, &identical);
+    if (ret)
+        goto out; /*not able to compare due to some corruption */
+
+    /* Topology is not changed, but just the options. But write the
+     * options to bitd volfile, so that bitd will be reconfigured.
+     */
+    if (identical) {
+        ret = glusterd_bitdsvc_create_volfile();
+        if (ret == 0) { /* Only if above PASSES */
+            ret = glusterd_fetchspec_notify(THIS);
+        }
+        goto out;
+    }
+
+manager:
+    /*
+     * bitd volfile's topology has been changed. bitd server needs
+     * to be RESTARTED to ACT on the changed volfile.
+     */
+    ret = priv->bitd_svc.manager(&(priv->bitd_svc), NULL, PROC_START_NO_WAIT);
+
+out:
+    gf_msg_debug(this ? this->name : "glusterd", 0, "Returning %d", ret);
+    return ret;
+}
diff --git a/xlators/mgmt/glusterd/src/glusterd-bitd-svc.h b/xlators/mgmt/glusterd/src/glusterd-bitd-svc.h
new file mode 100644
index 00000000000..1bff084a9a8
--- /dev/null
+++ b/xlators/mgmt/glusterd/src/glusterd-bitd-svc.h
@@ -0,0 +1,40 @@
+/*
+  Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _GLUSTERD_BITD_SVC_H_
+#define _GLUSTERD_BITD_SVC_H_
+
+#include "glusterd-svc-mgmt.h"
+
+#define bitd_svc_name "bitd"
+
+void
+glusterd_bitdsvc_build(glusterd_svc_t *svc);
+
+int
+glusterd_bitdsvc_init(glusterd_svc_t *svc);
+
+int
+glusterd_bitdsvc_manager(glusterd_svc_t *svc, void *data, int flags);
+
+int
+glusterd_bitdsvc_start(glusterd_svc_t *svc, int flags);
+
+int
+glusterd_bitdsvc_stop(glusterd_svc_t *svc, int sig);
+
+int
+glusterd_bitdsvc_reconfigure();
+
+void
+glusterd_bitdsvc_build_volfile_path(char *server, char *workdir, char *volfile,
+                                    size_t len);
+
+#endif
diff --git a/xlators/mgmt/glusterd/src/glusterd-bitrot.c b/xlators/mgmt/glusterd/src/glusterd-bitrot.c
new file mode 100644
index 00000000000..37429fe9214
--- /dev/null
+++ b/xlators/mgmt/glusterd/src/glusterd-bitrot.c
@@ -0,0 +1,822 @@
+/*
+   Copyright (c) 2011-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+ */
+
+#include <glusterfs/common-utils.h>
+#include "cli1-xdr.h"
+#include "xdr-generic.h"
+#include "glusterd.h"
+#include "glusterd-op-sm.h"
+#include "glusterd-store.h"
+#include "glusterd-utils.h"
+#include "glusterd-volgen.h"
+#include <glusterfs/run.h>
+#include <glusterfs/syscall.h>
+#include <glusterfs/byte-order.h>
+#include <glusterfs/compat-errno.h>
+#include "glusterd-scrub-svc.h"
+#include "glusterd-messages.h"
+
+#include <sys/wait.h>
+#include <dlfcn.h>
+
+const char *gd_bitrot_op_list[GF_BITROT_OPTION_TYPE_MAX] = {
+    [GF_BITROT_OPTION_TYPE_NONE] = "none",
+    [GF_BITROT_OPTION_TYPE_ENABLE] = "enable",
+    [GF_BITROT_OPTION_TYPE_DISABLE] = "disable",
+    [GF_BITROT_OPTION_TYPE_SCRUB_THROTTLE] = "scrub-throttle",
+    [GF_BITROT_OPTION_TYPE_SCRUB_FREQ] = "scrub-frequency",
+    [GF_BITROT_OPTION_TYPE_SCRUB] = "scrub",
+    [GF_BITROT_OPTION_TYPE_EXPIRY_TIME] = "expiry-time",
+    [GF_BITROT_OPTION_TYPE_SIGNER_THREADS] = "signer-threads",
+};
+
+int
+__glusterd_handle_bitrot(rpcsvc_request_t *req)
+{
+    int32_t ret = -1;
+    gf_cli_req cli_req = {{
+        0,
+    }};
+    dict_t *dict = NULL;
+    glusterd_op_t cli_op = GD_OP_BITROT;
+    char *volname = NULL;
+    char *scrub = NULL;
+    int32_t type = 0;
+    char msg[256] = {
+        0,
+    };
+    xlator_t *this = NULL;
+    glusterd_conf_t *conf = NULL;
+
+    GF_ASSERT(req);
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    conf = this->private;
+    GF_ASSERT(conf);
+
+    ret = xdr_to_generic(req->msg[0], &cli_req, (xdrproc_t)xdr_gf_cli_req);
+    if (ret < 0) {
+        req->rpc_err = GARBAGE_ARGS;
+        goto out;
+    }
+
+    if (cli_req.dict.dict_len) {
+        /* Unserialize the dictionary */
+        dict = dict_new();
+
+        ret = dict_unserialize(cli_req.dict.dict_val, cli_req.dict.dict_len,
+                               &dict);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_UNSERIALIZE_FAIL,
+                   "failed to "
+                   "unserialize req-buffer to dictionary");
+            snprintf(msg, sizeof(msg),
+                     "Unable to decode the "
+                     "command");
+            goto out;
+        } else {
+            dict->extra_stdfree = cli_req.dict.dict_val;
+        }
+    }
+
+    ret = dict_get_str(dict, "volname", &volname);
+    if (ret) {
+        snprintf(msg, sizeof(msg), "Unable to get volume name");
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to get volume name, "
+               "while handling bitrot command");
+        goto out;
+    }
+
+    ret = dict_get_int32(dict, "type", &type);
+    if (ret) {
+        snprintf(msg, sizeof(msg), "Unable to get type of command");
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to get type of cmd, "
+               "while handling bitrot command");
+        goto out;
+    }
+
+    if (conf->op_version < GD_OP_VERSION_3_7_0) {
+        snprintf(msg, sizeof(msg),
+                 "Cannot execute command. The "
+                 "cluster is operating at version %d. Bitrot command "
+                 "%s is unavailable in this version",
+                 conf->op_version, gd_bitrot_op_list[type]);
+        ret = -1;
+        goto out;
+    }
+
+    if (type == GF_BITROT_CMD_SCRUB_STATUS) {
+        /* Backward compatibility handling for scrub status command*/
+        if (conf->op_version < GD_OP_VERSION_3_7_7) {
+            snprintf(msg, sizeof(msg),
+                     "Cannot execute command. "
+                     "The cluster is operating at version %d. "
+                     "Bitrot scrub status command unavailable in "
+                     "this version",
+                     conf->op_version);
+            ret = -1;
+            goto out;
+        }
+
+        ret = dict_get_str(dict, "scrub-value", &scrub);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+                   "Failed to get scrub value.");
+            ret = -1;
+            goto out;
+        }
+
+        if (!strncmp(scrub, "status", SLEN("status"))) {
+            ret = glusterd_op_begin_synctask(req, GD_OP_SCRUB_STATUS, dict);
+            goto out;
+        }
+    }
+
+    if (type == GF_BITROT_CMD_SCRUB_ONDEMAND) {
+        /* Backward compatibility handling for scrub status command*/
+        if (conf->op_version < GD_OP_VERSION_3_9_0) {
+            snprintf(msg, sizeof(msg),
+                     "Cannot execute command. "
+                     "The cluster is operating at version %d. "
+                     "Bitrot scrub ondemand command unavailable in "
+                     "this version",
+                     conf->op_version);
+            ret = -1;
+            goto out;
+        }
+
+        ret = dict_get_str(dict, "scrub-value", &scrub);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+                   "Failed to get scrub value.");
+            ret = -1;
+            goto out;
+        }
+
+        if (!strncmp(scrub, "ondemand", SLEN("ondemand"))) {
+            ret = glusterd_op_begin_synctask(req, GD_OP_SCRUB_ONDEMAND, dict);
+            goto out;
+        }
+    }
+
+    ret = glusterd_op_begin_synctask(req, GD_OP_BITROT, dict);
+
+out:
+    if (ret) {
+        if (msg[0] == '\0')
+            snprintf(msg, sizeof(msg), "Bitrot operation failed");
+        ret = glusterd_op_send_cli_response(cli_op, ret, 0, req, dict, msg);
+    }
+
+    return ret;
+}
+
+int
+glusterd_handle_bitrot(rpcsvc_request_t *req)
+{
+    return glusterd_big_locked_handler(req, __glusterd_handle_bitrot);
+}
+
+static int
+glusterd_bitrot_scrub_throttle(glusterd_volinfo_t *volinfo, dict_t *dict,
+                               char *key, char **op_errstr)
+{
+    int32_t ret = -1;
+    char *scrub_throttle = NULL;
+    char *option = NULL;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    ret = dict_get_str(dict, "scrub-throttle-value", &scrub_throttle);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to fetch scrub-"
+               "throttle value");
+        goto out;
+    }
+
+    option = gf_strdup(scrub_throttle);
+    ret = dict_set_dynstr(volinfo->dict, key, option);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+               "Failed to set option %s", key);
+        goto out;
+    }
+
+    ret = glusterd_scrubsvc_reconfigure();
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SCRUBSVC_RECONF_FAIL,
+               "Failed to reconfigure scrub "
+               "services");
+        goto out;
+    }
+
+out:
+    return ret;
+}
+
+static int
+glusterd_bitrot_scrub_freq(glusterd_volinfo_t *volinfo, dict_t *dict, char *key,
+                           char **op_errstr)
+{
+    int32_t ret = -1;
+    char *scrub_freq = NULL;
+    xlator_t *this = NULL;
+    char *option = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    ret = dict_get_str(dict, "scrub-frequency-value", &scrub_freq);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to fetch scrub-"
+               "freq value");
+        goto out;
+    }
+
+    option = gf_strdup(scrub_freq);
+    ret = dict_set_dynstr(volinfo->dict, key, option);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+               "Failed to set option %s", key);
+        goto out;
+    }
+
+    ret = glusterd_scrubsvc_reconfigure();
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SCRUBSVC_RECONF_FAIL,
+               "Failed to reconfigure scrub "
+               "services");
+        goto out;
+    }
+
+out:
+    return ret;
+}
+
+static int
+glusterd_bitrot_scrub(glusterd_volinfo_t *volinfo, dict_t *dict, char *key,
+                      char **op_errstr)
+{
+    int32_t ret = -1;
+    char *scrub_value = NULL;
+    xlator_t *this = NULL;
+    char *option = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    ret = dict_get_str(dict, "scrub-value", &scrub_value);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+               "Unable to fetch scrub"
+               "value");
+        goto out;
+    }
+
+    if (!strcmp(scrub_value, "resume")) {
+        option = gf_strdup("Active");
+    } else {
+        option = gf_strdup(scrub_value);
+    }
+
+    ret = dict_set_dynstr(volinfo->dict, key, option);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+               "Failed to set option %s", key);
+        goto out;
+    }
+
+    ret = glusterd_scrubsvc_reconfigure();
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SCRUBSVC_RECONF_FAIL,
+               "Failed to reconfigure scrub "
+               "services");
+        goto out;
+    }
+
+out:
+    return ret;
+}
+
+static int
+glusterd_bitrot_expiry_time(glusterd_volinfo_t *volinfo, dict_t *dict,
+                            char *key, char **op_errstr)
+{
+    int32_t ret = -1;
+    uint32_t expiry_time = 0;
+    xlator_t *this = NULL;
+    char dkey[32] = {
+        0,
+    };
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    ret = dict_get_uint32(dict, "expiry-time", &expiry_time);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to get bitrot expiry"
+               " timer value.");
+        goto out;
+    }
+
+    snprintf(dkey, sizeof(dkey), "%d", expiry_time);
+
+    ret = dict_set_dynstr_with_alloc(volinfo->dict, key, dkey);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+               "Failed to set option %s", key);
+        goto out;
+    }
+
+    ret = glusterd_bitdsvc_reconfigure();
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_BITDSVC_RECONF_FAIL,
+               "Failed to reconfigure bitrot"
+               "services");
+        goto out;
+    }
+out:
+    return ret;
+}
+
+static gf_boolean_t
+is_bitd_configure_noop(xlator_t *this, glusterd_volinfo_t *volinfo)
+{
+    gf_boolean_t noop = _gf_true;
+    glusterd_brickinfo_t *brickinfo = NULL;
+
+    if (!glusterd_is_bitrot_enabled(volinfo))
+        goto out;
+    else if (volinfo->status != GLUSTERD_STATUS_STARTED)
+        goto out;
+    else {
+        cds_list_for_each_entry(brickinfo, &volinfo->bricks, brick_list)
+        {
+            if (!glusterd_is_local_brick(this, volinfo, brickinfo))
+                continue;
+            noop = _gf_false;
+            return noop;
+        }
+    }
+out:
+    return noop;
+}
+
+static int
+glusterd_bitrot_signer_threads(glusterd_volinfo_t *volinfo, dict_t *dict,
+                               char *key, char **op_errstr)
+{
+    int32_t ret = -1;
+    uint32_t signer_th_count = 0;
+    uint32_t existing_th_count = 0;
+    xlator_t *this = NULL;
+    glusterd_conf_t *priv = NULL;
+    char dkey[32] = {
+        0,
+    };
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, priv, out);
+
+    ret = dict_get_uint32(dict, "signer-threads", &signer_th_count);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to get bitrot signer thread count.");
+        goto out;
+    }
+
+    ret = dict_get_uint32(volinfo->dict, key, &existing_th_count);
+    if (ret == 0 && signer_th_count == existing_th_count) {
+        goto out;
+    }
+
+    snprintf(dkey, sizeof(dkey), "%d", signer_th_count);
+    ret = dict_set_dynstr_with_alloc(volinfo->dict, key, dkey);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+               "Failed to set option %s", key);
+        goto out;
+    }
+
+    if (!is_bitd_configure_noop(this, volinfo)) {
+        ret = priv->bitd_svc.manager(&(priv->bitd_svc), NULL,
+                                     PROC_START_NO_WAIT);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_BITDSVC_RECONF_FAIL,
+                   "Failed to reconfigure bitrot services");
+            goto out;
+        }
+    }
+out:
+    return ret;
+}
+
+static int
+glusterd_bitrot_enable(glusterd_volinfo_t *volinfo, char **op_errstr)
+{
+    int32_t ret = -1;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_VALIDATE_OR_GOTO(this->name, volinfo, out);
+    GF_VALIDATE_OR_GOTO(this->name, op_errstr, out);
+
+    if (glusterd_is_volume_started(volinfo) == 0) {
+        *op_errstr = gf_strdup(
+            "Volume is stopped, start volume "
+            "to enable bitrot.");
+        ret = -1;
+        goto out;
+    }
+
+    ret = glusterd_is_bitrot_enabled(volinfo);
+    if (ret) {
+        *op_errstr = gf_strdup("Bitrot is already enabled");
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_set_dynstr_with_alloc(volinfo->dict, VKEY_FEATURES_BITROT, "on");
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+               "dict set failed");
+        goto out;
+    }
+
+    /*Once bitrot is enable scrubber should be in Active state*/
+    ret = dict_set_dynstr_with_alloc(volinfo->dict, "features.scrub", "Active");
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+               "Failed to set option "
+               "features.scrub value");
+        goto out;
+    }
+
+    ret = 0;
+out:
+    if (ret && op_errstr && !*op_errstr)
+        gf_asprintf(op_errstr,
+                    "Enabling bitrot on volume %s has been "
+                    "unsuccessful",
+                    volinfo->volname);
+    return ret;
+}
+
+static int
+glusterd_bitrot_disable(glusterd_volinfo_t *volinfo, char **op_errstr)
+{
+    int32_t ret = -1;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_VALIDATE_OR_GOTO("glusterd", this, out);
+
+    GF_VALIDATE_OR_GOTO(this->name, volinfo, out);
+    GF_VALIDATE_OR_GOTO(this->name, op_errstr, out);
+
+    ret = dict_set_dynstr_with_alloc(volinfo->dict, VKEY_FEATURES_BITROT,
+                                     "off");
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+               "dict set failed");
+        goto out;
+    }
+
+    /*Once bitrot disabled scrubber should be Inactive state*/
+    ret = dict_set_dynstr_with_alloc(volinfo->dict, "features.scrub",
+                                     "Inactive");
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+               "Failed to set "
+               "features.scrub value");
+        goto out;
+    }
+
+    ret = 0;
+out:
+    if (ret && op_errstr && !*op_errstr)
+        gf_asprintf(op_errstr,
+                    "Disabling bitrot on volume %s has "
+                    "been unsuccessful",
+                    volinfo->volname);
+    return ret;
+}
+
+gf_boolean_t
+glusterd_should_i_stop_bitd()
+{
+    glusterd_conf_t *conf = THIS->private;
+    glusterd_volinfo_t *volinfo = NULL;
+    gf_boolean_t stopped = _gf_true;
+    glusterd_brickinfo_t *brickinfo = NULL;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    cds_list_for_each_entry(volinfo, &conf->volumes, vol_list)
+    {
+        if (!glusterd_is_bitrot_enabled(volinfo))
+            continue;
+        else if (volinfo->status != GLUSTERD_STATUS_STARTED)
+            continue;
+        else {
+            cds_list_for_each_entry(brickinfo, &volinfo->bricks, brick_list)
+            {
+                if (!glusterd_is_local_brick(this, volinfo, brickinfo))
+                    continue;
+                stopped = _gf_false;
+                return stopped;
+            }
+
+            /* Before stopping bitrot/scrubber daemon check
+             * other volume also whether respective volume
+             * host a brick from this node or not.*/
+            continue;
+        }
+    }
+
+    return stopped;
+}
+
+static int
+glusterd_manage_bitrot(int opcode)
+{
+    int ret = -1;
+    xlator_t *this = NULL;
+    glusterd_conf_t *priv = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    switch (opcode) {
+        case GF_BITROT_OPTION_TYPE_ENABLE:
+        case GF_BITROT_OPTION_TYPE_DISABLE:
+            ret = priv->bitd_svc.manager(&(priv->bitd_svc), NULL,
+                                         PROC_START_NO_WAIT);
+            if (ret)
+                break;
+            ret = priv->scrub_svc.manager(&(priv->scrub_svc), NULL,
+                                          PROC_START_NO_WAIT);
+            break;
+        default:
+            ret = 0;
+            break;
+    }
+
+    return ret;
+}
+
+int
+glusterd_op_bitrot(dict_t *dict, char **op_errstr, dict_t *rsp_dict)
+{
+    glusterd_volinfo_t *volinfo = NULL;
+    int32_t ret = -1;
+    char *volname = NULL;
+    int type = -1;
+    glusterd_conf_t *priv = NULL;
+    xlator_t *this = NULL;
+
+    GF_ASSERT(dict);
+    GF_ASSERT(op_errstr);
+
+    this = THIS;
+    GF_ASSERT(this);
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    ret = dict_get_str(dict, "volname", &volname);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to get volume name");
+        goto out;
+    }
+
+    ret = glusterd_volinfo_find(volname, &volinfo);
+    if (ret) {
+        gf_asprintf(op_errstr, FMTSTR_CHECK_VOL_EXISTS, volname);
+        goto out;
+    }
+
+    ret = dict_get_int32(dict, "type", &type);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to get type from "
+               "dict");
+        goto out;
+    }
+
+    switch (type) {
+        case GF_BITROT_OPTION_TYPE_ENABLE:
+            ret = glusterd_bitrot_enable(volinfo, op_errstr);
+            if (ret < 0)
+                goto out;
+            break;
+
+        case GF_BITROT_OPTION_TYPE_DISABLE:
+            ret = glusterd_bitrot_disable(volinfo, op_errstr);
+            if (ret < 0)
+                goto out;
+
+            break;
+
+        case GF_BITROT_OPTION_TYPE_SCRUB_THROTTLE:
+            ret = glusterd_bitrot_scrub_throttle(
+                volinfo, dict, "features.scrub-throttle", op_errstr);
+            if (ret)
+                goto out;
+            break;
+
+        case GF_BITROT_OPTION_TYPE_SCRUB_FREQ:
+            ret = glusterd_bitrot_scrub_freq(volinfo, dict,
+                                             "features.scrub-freq", op_errstr);
+            if (ret)
+                goto out;
+            break;
+
+        case GF_BITROT_OPTION_TYPE_SCRUB:
+            ret = glusterd_bitrot_scrub(volinfo, dict, "features.scrub",
+                                        op_errstr);
+            if (ret)
+                goto out;
+            break;
+
+        case GF_BITROT_OPTION_TYPE_EXPIRY_TIME:
+            ret = glusterd_bitrot_expiry_time(
+                volinfo, dict, "features.expiry-time", op_errstr);
+            if (ret)
+                goto out;
+            break;
+
+        case GF_BITROT_OPTION_TYPE_SIGNER_THREADS:
+            ret = glusterd_bitrot_signer_threads(
+                volinfo, dict, "features.signer-threads", op_errstr);
+            if (ret)
+                goto out;
+            break;
+
+        case GF_BITROT_CMD_SCRUB_STATUS:
+        case GF_BITROT_CMD_SCRUB_ONDEMAND:
+            break;
+
+        default:
+            gf_asprintf(op_errstr,
+                        "Bitrot command failed. Invalid "
+                        "opcode");
+            ret = -1;
+            goto out;
+    }
+
+    ret = glusterd_manage_bitrot(type);
+    if (ret)
+        goto out;
+
+    ret = glusterd_create_volfiles_and_notify_services(volinfo);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_VOLFILE_CREATE_FAIL,
+               "Unable to re-create "
+               "volfiles");
+        ret = -1;
+        goto out;
+    }
+
+    ret = glusterd_store_volinfo(volinfo, GLUSTERD_VOLINFO_VER_AC_INCREMENT);
+    if (ret) {
+        gf_msg_debug(this->name, 0,
+                     "Failed to store volinfo for "
+                     "bitrot");
+        goto out;
+    }
+
+out:
+    return ret;
+}
+
+int
+glusterd_op_stage_bitrot(dict_t *dict, char **op_errstr, dict_t *rsp_dict)
+{
+    int ret = 0;
+    char *volname = NULL;
+    char *scrub_cmd = NULL;
+    char *scrub_cmd_from_dict = NULL;
+    char msg[2048] = {
+        0,
+    };
+    int type = 0;
+    xlator_t *this = NULL;
+    glusterd_conf_t *priv = NULL;
+    glusterd_volinfo_t *volinfo = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    GF_ASSERT(dict);
+    GF_ASSERT(op_errstr);
+
+    ret = dict_get_str(dict, "volname", &volname);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to get volume name");
+        goto out;
+    }
+
+    ret = glusterd_volinfo_find(volname, &volinfo);
+    if (ret) {
+        gf_asprintf(op_errstr, FMTSTR_CHECK_VOL_EXISTS, volname);
+        goto out;
+    }
+
+    if (!glusterd_is_volume_started(volinfo)) {
+        *op_errstr = gf_strdup(
+            "Volume is stopped, start volume "
+            "before executing bit rot command.");
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_get_int32(dict, "type", &type);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to get type for "
+               "operation");
+
+        *op_errstr = gf_strdup(
+            "Staging stage failed for bitrot "
+            "operation.");
+        goto out;
+    }
+
+    if ((GF_BITROT_OPTION_TYPE_ENABLE != type) &&
+        (glusterd_is_bitrot_enabled(volinfo) == 0)) {
+        ret = -1;
+        gf_asprintf(op_errstr, "Bitrot is not enabled on volume %s", volname);
+        goto out;
+    }
+
+    if ((GF_BITROT_OPTION_TYPE_SCRUB == type)) {
+        ret = dict_get_str(volinfo->dict, "features.scrub",
+                           &scrub_cmd_from_dict);
+        if (!ret) {
+            ret = dict_get_str(dict, "scrub-value", &scrub_cmd);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+                       "Unable to "
+                       "get scrub-value");
+                *op_errstr = gf_strdup(
+                    "Staging failed for "
+                    "bitrot operation. "
+                    "Please check log file"
+                    " for more details.");
+                goto out;
+            }
+            /* If scrubber is resume then value of scrubber will be
+             * "Active" in the dictionary. */
+            if (!strcmp(scrub_cmd_from_dict, scrub_cmd) ||
+                (!strncmp("Active", scrub_cmd_from_dict, SLEN("Active")) &&
+                 !strncmp("resume", scrub_cmd, SLEN("resume")))) {
+                snprintf(msg, sizeof(msg),
+                         "Scrub is already"
+                         " %sd for volume %s",
+                         scrub_cmd, volinfo->volname);
+                *op_errstr = gf_strdup(msg);
+                ret = -1;
+                goto out;
+            }
+        }
+        ret = 0;
+    }
+
+out:
+    if (ret && op_errstr && *op_errstr)
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_OP_STAGE_BITROT_FAIL, "%s",
+               *op_errstr);
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+
+    return ret;
+}
diff --git a/xlators/mgmt/glusterd/src/glusterd-brick-ops.c b/xlators/mgmt/glusterd/src/glusterd-brick-ops.c
new file mode 100644
index 00000000000..e56cd0e6c74
--- /dev/null
+++ b/xlators/mgmt/glusterd/src/glusterd-brick-ops.c
@@ -0,0 +1,2796 @@
+/*
+   Copyright (c) 2011-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include <glusterfs/common-utils.h>
+#include "cli1-xdr.h"
+#include "xdr-generic.h"
+#include "glusterd.h"
+#include "glusterd-op-sm.h"
+#include "glusterd-geo-rep.h"
+#include "glusterd-store.h"
+#include "glusterd-mgmt.h"
+#include "glusterd-utils.h"
+#include "glusterd-volgen.h"
+#include "glusterd-svc-helper.h"
+#include "glusterd-messages.h"
+#include "glusterd-server-quorum.h"
+#include <glusterfs/run.h>
+#include <glusterfs/syscall.h>
+#include <sys/signal.h>
+
+/* misc */
+
+/* In this function, we decide, based on the 'count' of the brick,
+   where to add it in the current volume. 'count' tells us already
+   how many of the given bricks are added. other argument are self-
+   descriptive. */
+int
+add_brick_at_right_order(glusterd_brickinfo_t *brickinfo,
+                         glusterd_volinfo_t *volinfo, int count,
+                         int32_t stripe_cnt, int32_t replica_cnt)
+{
+    int idx = 0;
+    int i = 0;
+    int sub_cnt = 0;
+    glusterd_brickinfo_t *brick = NULL;
+
+    /* The complexity of the function is in deciding at which index
+       to add new brick. Even though it can be defined with a complex
+       single formula for all volume, it is separated out to make it
+       more readable */
+    if (stripe_cnt) {
+        /* common formula when 'stripe_count' is set */
+        /* idx = ((count / ((stripe_cnt * volinfo->replica_count) -
+           volinfo->dist_leaf_count)) * volinfo->dist_leaf_count) +
+           (count + volinfo->dist_leaf_count);
+        */
+
+        sub_cnt = volinfo->dist_leaf_count;
+
+        idx = ((count / ((stripe_cnt * volinfo->replica_count) - sub_cnt)) *
+               sub_cnt) +
+              (count + sub_cnt);
+
+        goto insert_brick;
+    }
+
+    /* replica count is set */
+    /* common formula when 'replica_count' is set */
+    /* idx = ((count / (replica_cnt - existing_replica_count)) *
+       existing_replica_count) +
+       (count + existing_replica_count);
+    */
+
+    sub_cnt = volinfo->replica_count;
+    idx = (count / (replica_cnt - sub_cnt) * sub_cnt) + (count + sub_cnt);
+
+insert_brick:
+    i = 0;
+    cds_list_for_each_entry(brick, &volinfo->bricks, brick_list)
+    {
+        i++;
+        if (i < idx)
+            continue;
+        gf_msg_debug(THIS->name, 0, "brick:%s index=%d, count=%d", brick->path,
+                     idx, count);
+
+        cds_list_add(&brickinfo->brick_list, &brick->brick_list);
+        break;
+    }
+
+    return 0;
+}
+
+static int
+gd_addbr_validate_replica_count(glusterd_volinfo_t *volinfo, int replica_count,
+                                int arbiter_count, int total_bricks, int *type,
+                                char *err_str, int err_len)
+{
+    int ret = -1;
+
+    /* replica count is set */
+    switch (volinfo->type) {
+        case GF_CLUSTER_TYPE_NONE:
+            if ((volinfo->brick_count * replica_count) == total_bricks) {
+                /* Change the volume type */
+                *type = GF_CLUSTER_TYPE_REPLICATE;
+                gf_msg(THIS->name, GF_LOG_INFO, 0,
+                       GD_MSG_VOL_TYPE_CHANGING_INFO,
+                       "Changing the type of volume %s from "
+                       "'distribute' to 'replica'",
+                       volinfo->volname);
+                ret = 0;
+                goto out;
+
+            } else {
+                snprintf(err_str, err_len,
+                         "Incorrect number of "
+                         "bricks (%d) supplied for replica count (%d).",
+                         (total_bricks - volinfo->brick_count), replica_count);
+                gf_msg(THIS->name, GF_LOG_ERROR, EINVAL, GD_MSG_INVALID_ENTRY,
+                       "%s", err_str);
+                goto out;
+            }
+            break;
+        case GF_CLUSTER_TYPE_REPLICATE:
+            if (replica_count < volinfo->replica_count) {
+                snprintf(err_str, err_len,
+                         "Incorrect replica count (%d) supplied. "
+                         "Volume already has (%d)",
+                         replica_count, volinfo->replica_count);
+                gf_msg(THIS->name, GF_LOG_ERROR, EINVAL, GD_MSG_INVALID_ENTRY,
+                       "%s", err_str);
+                goto out;
+            }
+            if (replica_count == volinfo->replica_count) {
+                if (arbiter_count && !volinfo->arbiter_count) {
+                    snprintf(err_str, err_len,
+                             "Cannot convert replica 3 volume "
+                             "to arbiter volume.");
+                    gf_msg(THIS->name, GF_LOG_ERROR, EINVAL,
+                           GD_MSG_INVALID_ENTRY, "%s", err_str);
+                    goto out;
+                }
+                if (!(total_bricks % volinfo->dist_leaf_count)) {
+                    ret = 1;
+                    goto out;
+                }
+            }
+            if (replica_count > volinfo->replica_count) {
+                /* We have to make sure before and after 'add-brick',
+                   the number or subvolumes for distribute will remain
+                   same, when replica count is given */
+                if ((total_bricks * volinfo->dist_leaf_count) ==
+                    (volinfo->brick_count *
+                     (replica_count * volinfo->stripe_count))) {
+                    /* Change the dist_leaf_count */
+                    gf_msg(THIS->name, GF_LOG_INFO, 0,
+                           GD_MSG_REPLICA_COUNT_CHANGE_INFO,
+                           "Changing the replica count of "
+                           "volume %s from %d to %d",
+                           volinfo->volname, volinfo->replica_count,
+                           replica_count);
+                    ret = 0;
+                    goto out;
+                }
+            }
+            break;
+        case GF_CLUSTER_TYPE_DISPERSE:
+            snprintf(err_str, err_len,
+                     "Volume %s cannot be converted "
+                     "from dispersed to replicated-"
+                     "dispersed",
+                     volinfo->volname);
+            gf_msg(THIS->name, GF_LOG_ERROR, EPERM, GD_MSG_OP_NOT_PERMITTED,
+                   "%s", err_str);
+            goto out;
+    }
+out:
+    return ret;
+}
+
+static int
+gd_rmbr_validate_replica_count(glusterd_volinfo_t *volinfo,
+                               int32_t replica_count, int32_t brick_count,
+                               char *err_str, size_t err_len)
+{
+    int ret = -1;
+    int replica_nodes = 0;
+    xlator_t *this = NULL;
+    this = THIS;
+    GF_ASSERT(this);
+
+    switch (volinfo->type) {
+        case GF_CLUSTER_TYPE_NONE:
+        case GF_CLUSTER_TYPE_DISPERSE:
+            snprintf(err_str, err_len,
+                     "replica count (%d) option given for non replicate "
+                     "volume %s",
+                     replica_count, volinfo->volname);
+            gf_smsg(this->name, GF_LOG_WARNING, EINVAL, GD_MSG_INVALID_ARGUMENT,
+                    err_str, NULL);
+            goto out;
+
+        case GF_CLUSTER_TYPE_REPLICATE:
+            /* in remove brick, you can only reduce the replica count */
+            if (replica_count > volinfo->replica_count) {
+                snprintf(err_str, err_len,
+                         "given replica count (%d) option is more "
+                         "than volume %s's replica count (%d)",
+                         replica_count, volinfo->volname,
+                         volinfo->replica_count);
+                gf_smsg(this->name, GF_LOG_WARNING, EINVAL,
+                        GD_MSG_INVALID_ARGUMENT, err_str, NULL);
+                goto out;
+            }
+            if (replica_count == volinfo->replica_count) {
+                /* This means the 'replica N' option on CLI was
+                   redundant. Check if the total number of bricks given
+                   for removal is same as 'dist_leaf_count' */
+                if (brick_count % volinfo->dist_leaf_count) {
+                    snprintf(err_str, err_len,
+                             "number of bricks provided (%d) is "
+                             "not valid. need at least %d "
+                             "(or %dxN)",
+                             brick_count, volinfo->dist_leaf_count,
+                             volinfo->dist_leaf_count);
+                    gf_smsg(this->name, GF_LOG_WARNING, EINVAL,
+                            GD_MSG_INVALID_ARGUMENT, err_str, NULL);
+                    goto out;
+                }
+                ret = 1;
+                goto out;
+            }
+
+            replica_nodes = ((volinfo->brick_count / volinfo->replica_count) *
+                             (volinfo->replica_count - replica_count));
+
+            if (brick_count % replica_nodes) {
+                snprintf(err_str, err_len,
+                         "need %d(xN) bricks for reducing replica "
+                         "count of the volume from %d to %d",
+                         replica_nodes, volinfo->replica_count, replica_count);
+                gf_smsg(this->name, GF_LOG_WARNING, EINVAL,
+                        GD_MSG_INVALID_ARGUMENT, err_str, NULL);
+                goto out;
+            }
+            break;
+    }
+
+    ret = 0;
+out:
+    return ret;
+}
+
+/* Handler functions */
+int
+__glusterd_handle_add_brick(rpcsvc_request_t *req)
+{
+    int32_t ret = -1;
+    gf_cli_req cli_req = {{
+        0,
+    }};
+    dict_t *dict = NULL;
+    char *bricks = NULL;
+    char *volname = NULL;
+    int brick_count = 0;
+    void *cli_rsp = NULL;
+    char err_str[2048] = "";
+    gf_cli_rsp rsp = {
+        0,
+    };
+    glusterd_volinfo_t *volinfo = NULL;
+    xlator_t *this = NULL;
+    int total_bricks = 0;
+    int32_t replica_count = 0;
+    int32_t arbiter_count = 0;
+    int32_t stripe_count = 0;
+    int type = 0;
+    glusterd_conf_t *conf = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(req);
+
+    conf = this->private;
+    GF_ASSERT(conf);
+
+    ret = xdr_to_generic(req->msg[0], &cli_req, (xdrproc_t)xdr_gf_cli_req);
+    if (ret < 0) {
+        // failed to decode msg;
+        req->rpc_err = GARBAGE_ARGS;
+        snprintf(err_str, sizeof(err_str), "Garbage args received");
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_GARBAGE_ARGS, NULL);
+        goto out;
+    }
+
+    gf_msg(this->name, GF_LOG_INFO, 0, GD_MSG_ADD_BRICK_REQ_RECVD,
+           "Received add brick req");
+
+    if (cli_req.dict.dict_len) {
+        /* Unserialize the dictionary */
+        dict = dict_new();
+
+        ret = dict_unserialize(cli_req.dict.dict_val, cli_req.dict.dict_len,
+                               &dict);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, errno,
+                   GD_MSG_DICT_UNSERIALIZE_FAIL,
+                   "failed to "
+                   "unserialize req-buffer to dictionary");
+            snprintf(err_str, sizeof(err_str),
+                     "Unable to decode "
+                     "the command");
+            goto out;
+        }
+    }
+
+    ret = dict_get_strn(dict, "volname", SLEN("volname"), &volname);
+
+    if (ret) {
+        snprintf(err_str, sizeof(err_str),
+                 "Unable to get volume "
+                 "name");
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED, "%s",
+               err_str);
+        goto out;
+    }
+
+    ret = glusterd_volinfo_find(volname, &volinfo);
+    if (ret) {
+        snprintf(err_str, sizeof(err_str),
+                 "Unable to get volinfo "
+                 "for volume name %s",
+                 volname);
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_VOLINFO_GET_FAIL, "%s",
+               err_str);
+        goto out;
+    }
+
+    ret = dict_get_int32n(dict, "count", SLEN("count"), &brick_count);
+    if (ret) {
+        snprintf(err_str, sizeof(err_str),
+                 "Unable to get volume "
+                 "brick count");
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED, "%s",
+               err_str);
+        goto out;
+    }
+
+    ret = dict_get_int32n(dict, "replica-count", SLEN("replica-count"),
+                          &replica_count);
+    if (!ret) {
+        gf_msg(this->name, GF_LOG_INFO, errno, GD_MSG_DICT_GET_SUCCESS,
+               "replica-count is %d", replica_count);
+    }
+
+    ret = dict_get_int32n(dict, "arbiter-count", SLEN("arbiter-count"),
+                          &arbiter_count);
+    if (!ret) {
+        gf_msg(this->name, GF_LOG_INFO, errno, GD_MSG_DICT_GET_SUCCESS,
+               "arbiter-count is %d", arbiter_count);
+    }
+
+    ret = dict_get_int32n(dict, "stripe-count", SLEN("stripe-count"),
+                          &stripe_count);
+    if (!ret) {
+        gf_msg(this->name, GF_LOG_INFO, errno, GD_MSG_DICT_GET_SUCCESS,
+               "stripe-count is %d", stripe_count);
+    }
+
+    if (!dict_getn(dict, "force", SLEN("force"))) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Failed to get flag");
+        goto out;
+    }
+
+    total_bricks = volinfo->brick_count + brick_count;
+
+    if (!stripe_count && !replica_count) {
+        if (volinfo->type == GF_CLUSTER_TYPE_NONE)
+            goto brick_val;
+
+        if ((volinfo->brick_count < volinfo->dist_leaf_count) &&
+            (total_bricks <= volinfo->dist_leaf_count))
+            goto brick_val;
+
+        if ((brick_count % volinfo->dist_leaf_count) != 0) {
+            snprintf(err_str, sizeof(err_str),
+                     "Incorrect number "
+                     "of bricks supplied %d with count %d",
+                     brick_count, volinfo->dist_leaf_count);
+            gf_msg(this->name, GF_LOG_ERROR, EINVAL, GD_MSG_VOL_NOT_REPLICA,
+                   "%s", err_str);
+            ret = -1;
+            goto out;
+        }
+        goto brick_val;
+        /* done with validation.. below section is if stripe|replica
+           count is given */
+    }
+
+    ret = gd_addbr_validate_replica_count(volinfo, replica_count, arbiter_count,
+                                          total_bricks, &type, err_str,
+                                          sizeof(err_str));
+    if (ret == -1) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_COUNT_VALIDATE_FAILED, "%s",
+               err_str);
+        goto out;
+    }
+
+    /* if replica count is same as earlier, set it back to 0 */
+    if (ret == 1)
+        replica_count = 0;
+
+    ret = dict_set_int32n(dict, "replica-count", SLEN("replica-count"),
+                          replica_count);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+               "failed to set the replica-count in dict");
+        goto out;
+    }
+
+brick_val:
+    ret = dict_get_strn(dict, "bricks", SLEN("bricks"), &bricks);
+    if (ret) {
+        snprintf(err_str, sizeof(err_str),
+                 "Unable to get volume "
+                 "bricks");
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED, "%s",
+               err_str);
+        goto out;
+    }
+
+    if (type != volinfo->type) {
+        ret = dict_set_int32n(dict, "type", SLEN("type"), type);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                   "failed to set the new type in dict");
+            goto out;
+        }
+    }
+
+    if (conf->op_version <= GD_OP_VERSION_3_7_5) {
+        gf_msg_debug(this->name, 0,
+                     "The cluster is operating at "
+                     "version less than or equal to %d. Falling back "
+                     "to syncop framework.",
+                     GD_OP_VERSION_3_7_5);
+        ret = glusterd_op_begin_synctask(req, GD_OP_ADD_BRICK, dict);
+    } else {
+        ret = glusterd_mgmt_v3_initiate_all_phases(req, GD_OP_ADD_BRICK, dict);
+    }
+
+out:
+    if (ret) {
+        rsp.op_ret = -1;
+        rsp.op_errno = 0;
+        if (err_str[0] == '\0')
+            snprintf(err_str, sizeof(err_str), "Operation failed");
+        rsp.op_errstr = err_str;
+        cli_rsp = &rsp;
+        glusterd_to_cli(req, cli_rsp, NULL, 0, NULL, (xdrproc_t)xdr_gf_cli_rsp,
+                        dict);
+        ret = 0;  // sent error to cli, prevent second reply
+    }
+
+    free(cli_req.dict.dict_val);  // its malloced by xdr
+
+    return ret;
+}
+
+int
+glusterd_handle_add_brick(rpcsvc_request_t *req)
+{
+    return glusterd_big_locked_handler(req, __glusterd_handle_add_brick);
+}
+
+static int
+subvol_matcher_init(int **subvols, int count)
+{
+    int ret = -1;
+
+    *subvols = GF_CALLOC(count, sizeof(int), gf_gld_mt_int);
+    if (*subvols)
+        ret = 0;
+
+    return ret;
+}
+
+static void
+subvol_matcher_update(int *subvols, glusterd_volinfo_t *volinfo,
+                      glusterd_brickinfo_t *brickinfo)
+{
+    glusterd_brickinfo_t *tmp = NULL;
+    int32_t sub_volume = 0;
+    int pos = 0;
+    if (subvols) {
+        cds_list_for_each_entry(tmp, &volinfo->bricks, brick_list)
+        {
+            if (strcmp(tmp->hostname, brickinfo->hostname) ||
+                strcmp(tmp->path, brickinfo->path)) {
+                pos++;
+                continue;
+            }
+            gf_msg_debug(THIS->name, 0, LOGSTR_FOUND_BRICK, brickinfo->hostname,
+                         brickinfo->path, volinfo->volname);
+            sub_volume = (pos / volinfo->dist_leaf_count);
+            subvols[sub_volume]++;
+            break;
+        }
+    }
+}
+
+static int
+subvol_matcher_verify(int *subvols, glusterd_volinfo_t *volinfo, char *err_str,
+                      size_t err_len, char *vol_type, int replica_count)
+{
+    int i = 0;
+    int ret = 0;
+    int count = volinfo->replica_count - replica_count;
+    xlator_t *this = THIS;
+    GF_ASSERT(this);
+
+    if (replica_count && subvols) {
+        for (i = 0; i < volinfo->subvol_count; i++) {
+            if (subvols[i] != count) {
+                ret = -1;
+                snprintf(err_str, err_len,
+                         "Remove exactly %d"
+                         " brick(s) from each subvolume.",
+                         count);
+                gf_smsg(this->name, GF_LOG_ERROR, errno,
+                        GD_MSG_BRICK_SUBVOL_VERIFY_FAIL, err_str, NULL);
+                break;
+            }
+        }
+        return ret;
+    }
+
+    do {
+        if (subvols && (subvols[i] % volinfo->dist_leaf_count == 0)) {
+            continue;
+        } else {
+            ret = -1;
+            snprintf(err_str, err_len, "Bricks not from same subvol for %s",
+                     vol_type);
+            gf_smsg(this->name, GF_LOG_ERROR, errno,
+                    GD_MSG_BRICK_SUBVOL_VERIFY_FAIL, err_str, NULL);
+            break;
+        }
+    } while (++i < volinfo->subvol_count);
+
+    return ret;
+}
+
+static void
+subvol_matcher_destroy(int *subvols)
+{
+    GF_FREE(subvols);
+}
+
+static int
+glusterd_remove_brick_validate_arbiters(glusterd_volinfo_t *volinfo,
+                                        int32_t count, int32_t replica_count,
+                                        glusterd_brickinfo_t **brickinfo_list,
+                                        char *err_str, size_t err_len)
+{
+    int i = 0;
+    int ret = 0;
+    glusterd_brickinfo_t *brickinfo = NULL;
+    glusterd_brickinfo_t *last = NULL;
+    char *arbiter_array = NULL;
+    xlator_t *this = NULL;
+    this = THIS;
+    GF_ASSERT(this);
+
+    if (volinfo->type != GF_CLUSTER_TYPE_REPLICATE)
+        goto out;
+
+    if (!replica_count || !volinfo->arbiter_count)
+        goto out;
+
+    if (replica_count == 2) {
+        /* If it is an arbiter to replica 2 conversion, only permit
+         *  removal of the arbiter brick.*/
+        for (i = 0; i < count; i++) {
+            brickinfo = brickinfo_list[i];
+            last = get_last_brick_of_brick_group(volinfo, brickinfo);
+            if (last != brickinfo) {
+                snprintf(err_str, err_len,
+                         "Remove arbiter "
+                         "brick(s) only when converting from "
+                         "arbiter to replica 2 subvolume.");
+                gf_smsg(this->name, GF_LOG_ERROR, errno,
+                        GD_MSG_REMOVE_ARBITER_BRICK, err_str, NULL);
+                ret = -1;
+                goto out;
+            }
+        }
+    } else if (replica_count == 1) {
+        /* If it is an arbiter to plain distribute conversion, in every
+         * replica subvol, the arbiter has to be one of the bricks that
+         * are removed. */
+        arbiter_array = GF_CALLOC(volinfo->subvol_count, sizeof(*arbiter_array),
+                                  gf_common_mt_char);
+        if (!arbiter_array)
+            return -1;
+        for (i = 0; i < count; i++) {
+            brickinfo = brickinfo_list[i];
+            last = get_last_brick_of_brick_group(volinfo, brickinfo);
+            if (last == brickinfo)
+                arbiter_array[brickinfo->group] = 1;
+        }
+        for (i = 0; i < volinfo->subvol_count; i++)
+            if (!arbiter_array[i]) {
+                snprintf(err_str, err_len,
+                         "Removed bricks "
+                         "must contain arbiter when converting"
+                         " to plain distribute.");
+                gf_smsg(this->name, GF_LOG_ERROR, errno,
+                        GD_MSG_REMOVE_ARBITER_BRICK, err_str, NULL);
+                ret = -1;
+                break;
+            }
+        GF_FREE(arbiter_array);
+    }
+
+out:
+    return ret;
+}
+
+int
+__glusterd_handle_remove_brick(rpcsvc_request_t *req)
+{
+    int32_t ret = -1;
+    gf_cli_req cli_req = {{
+        0,
+    }};
+    dict_t *dict = NULL;
+    int32_t count = 0;
+    char *brick = NULL;
+    char key[64] = "";
+    int keylen;
+    int i = 1;
+    glusterd_conf_t *conf = NULL;
+    glusterd_volinfo_t *volinfo = NULL;
+    glusterd_brickinfo_t *brickinfo = NULL;
+    glusterd_brickinfo_t **brickinfo_list = NULL;
+    int *subvols = NULL;
+    char err_str[2048] = "";
+    gf_cli_rsp rsp = {
+        0,
+    };
+    void *cli_rsp = NULL;
+    char vol_type[256] = "";
+    int32_t replica_count = 0;
+    char *volname = 0;
+    xlator_t *this = NULL;
+    int cmd = -1;
+
+    GF_ASSERT(req);
+    this = THIS;
+    GF_ASSERT(this);
+    conf = this->private;
+    GF_ASSERT(conf);
+
+    ret = xdr_to_generic(req->msg[0], &cli_req, (xdrproc_t)xdr_gf_cli_req);
+    if (ret < 0) {
+        // failed to decode msg;
+        req->rpc_err = GARBAGE_ARGS;
+        snprintf(err_str, sizeof(err_str), "Received garbage args");
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_GARBAGE_ARGS, NULL);
+        goto out;
+    }
+
+    gf_msg(this->name, GF_LOG_INFO, 0, GD_MSG_REM_BRICK_REQ_RECVD,
+           "Received rem brick req");
+
+    if (cli_req.dict.dict_len) {
+        /* Unserialize the dictionary */
+        dict = dict_new();
+
+        ret = dict_unserialize(cli_req.dict.dict_val, cli_req.dict.dict_len,
+                               &dict);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, errno,
+                   GD_MSG_DICT_UNSERIALIZE_FAIL,
+                   "failed to "
+                   "unserialize req-buffer to dictionary");
+            snprintf(err_str, sizeof(err_str),
+                     "Unable to decode "
+                     "the command");
+            goto out;
+        }
+    }
+
+    ret = dict_get_strn(dict, "volname", SLEN("volname"), &volname);
+    if (ret) {
+        snprintf(err_str, sizeof(err_str),
+                 "Unable to get volume "
+                 "name");
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED, "%s",
+               err_str);
+        goto out;
+    }
+
+    ret = dict_get_int32n(dict, "count", SLEN("count"), &count);
+    if (ret) {
+        snprintf(err_str, sizeof(err_str),
+                 "Unable to get brick "
+                 "count");
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED, "%s",
+               err_str);
+        goto out;
+    }
+
+    ret = glusterd_volinfo_find(volname, &volinfo);
+    if (ret) {
+        snprintf(err_str, sizeof(err_str), "Volume %s does not exist", volname);
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, GD_MSG_VOL_NOT_FOUND, "%s",
+               err_str);
+        goto out;
+    }
+
+    ret = dict_get_int32n(dict, "command", SLEN("command"), &cmd);
+    if (ret) {
+        snprintf(err_str, sizeof(err_str),
+                 "Unable to get cmd "
+                 "ccommand");
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED, "%s",
+               err_str);
+        goto out;
+    }
+
+    ret = dict_get_int32n(dict, "replica-count", SLEN("replica-count"),
+                          &replica_count);
+    if (!ret) {
+        gf_msg(this->name, GF_LOG_INFO, errno, GD_MSG_DICT_GET_FAILED,
+               "request to change replica-count to %d", replica_count);
+        ret = gd_rmbr_validate_replica_count(volinfo, replica_count, count,
+                                             err_str, sizeof(err_str));
+        if (ret < 0) {
+            /* logging and error msg are done in above function
+               itself */
+            goto out;
+        }
+        dict_deln(dict, "replica-count", SLEN("replica-count"));
+        if (ret) {
+            replica_count = 0;
+        } else {
+            ret = dict_set_int32n(dict, "replica-count", SLEN("replica-count"),
+                                  replica_count);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_WARNING, errno,
+                       GD_MSG_DICT_SET_FAILED,
+                       "failed to set the replica_count "
+                       "in dict");
+                goto out;
+            }
+        }
+    }
+
+    /* 'vol_type' is used for giving the meaning full error msg for user */
+    if (volinfo->type == GF_CLUSTER_TYPE_REPLICATE) {
+        strcpy(vol_type, "replica");
+    } else if (volinfo->type == GF_CLUSTER_TYPE_DISPERSE) {
+        strcpy(vol_type, "disperse");
+    } else {
+        strcpy(vol_type, "distribute");
+    }
+
+    if (!replica_count && (volinfo->type == GF_CLUSTER_TYPE_REPLICATE) &&
+        (volinfo->brick_count == volinfo->dist_leaf_count)) {
+        snprintf(err_str, sizeof(err_str),
+                 "Removing bricks from replicate configuration "
+                 "is not allowed without reducing replica count "
+                 "explicitly.");
+        gf_msg(this->name, GF_LOG_ERROR, EPERM, GD_MSG_OP_NOT_PERMITTED_AC_REQD,
+               "%s", err_str);
+        ret = -1;
+        goto out;
+    }
+
+    /* Do not allow remove-brick if the bricks given is less than
+       the replica count or stripe count */
+    if (!replica_count && (volinfo->type != GF_CLUSTER_TYPE_NONE)) {
+        if (volinfo->dist_leaf_count && (count % volinfo->dist_leaf_count)) {
+            snprintf(err_str, sizeof(err_str),
+                     "Remove brick "
+                     "incorrect brick count of %d for %s %d",
+                     count, vol_type, volinfo->dist_leaf_count);
+            gf_msg(this->name, GF_LOG_ERROR, EINVAL, GD_MSG_INVALID_ENTRY, "%s",
+                   err_str);
+            ret = -1;
+            goto out;
+        }
+    }
+
+    if ((volinfo->type != GF_CLUSTER_TYPE_NONE) &&
+        (volinfo->subvol_count > 1)) {
+        ret = subvol_matcher_init(&subvols, volinfo->subvol_count);
+        if (ret)
+            goto out;
+    }
+
+    brickinfo_list = GF_CALLOC(count, sizeof(*brickinfo_list),
+                               gf_common_mt_pointer);
+    if (!brickinfo_list) {
+        ret = -1;
+        goto out;
+    }
+
+    while (i <= count) {
+        keylen = snprintf(key, sizeof(key), "brick%d", i);
+        ret = dict_get_strn(dict, key, keylen, &brick);
+        if (ret) {
+            snprintf(err_str, sizeof(err_str), "Unable to get %s", key);
+            gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+                   "%s", err_str);
+            goto out;
+        }
+        gf_msg_debug(this->name, 0,
+                     "Remove brick count %d brick:"
+                     " %s",
+                     i, brick);
+
+        ret = glusterd_volume_brickinfo_get_by_brick(brick, volinfo, &brickinfo,
+                                                     _gf_false);
+
+        if (ret) {
+            snprintf(err_str, sizeof(err_str),
+                     "Incorrect brick "
+                     "%s for volume %s",
+                     brick, volname);
+            gf_msg(this->name, GF_LOG_ERROR, EINVAL, GD_MSG_BRICK_NOT_FOUND,
+                   "%s", err_str);
+            goto out;
+        }
+        brickinfo_list[i - 1] = brickinfo;
+
+        i++;
+        if ((volinfo->type == GF_CLUSTER_TYPE_NONE) ||
+            (volinfo->brick_count <= volinfo->dist_leaf_count))
+            continue;
+
+        subvol_matcher_update(subvols, volinfo, brickinfo);
+    }
+
+    if ((volinfo->type != GF_CLUSTER_TYPE_NONE) &&
+        (volinfo->subvol_count > 1)) {
+        ret = subvol_matcher_verify(subvols, volinfo, err_str, sizeof(err_str),
+                                    vol_type, replica_count);
+        if (ret)
+            goto out;
+    }
+
+    ret = glusterd_remove_brick_validate_arbiters(volinfo, count, replica_count,
+                                                  brickinfo_list, err_str,
+                                                  sizeof(err_str));
+    if (ret)
+        goto out;
+
+    if (conf->op_version < GD_OP_VERSION_8_0) {
+        gf_msg_debug(this->name, 0,
+                     "The cluster is operating at "
+                     "version less than %d. remove-brick operation"
+                     "falling back to syncop framework.",
+                     GD_OP_VERSION_8_0);
+        ret = glusterd_op_begin_synctask(req, GD_OP_REMOVE_BRICK, dict);
+    } else {
+        ret = glusterd_mgmt_v3_initiate_all_phases(req, GD_OP_REMOVE_BRICK,
+                                                   dict);
+    }
+
+out:
+    if (ret) {
+        rsp.op_ret = -1;
+        rsp.op_errno = 0;
+        if (err_str[0] == '\0')
+            snprintf(err_str, sizeof(err_str), "Operation failed");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GLUSTERD_OP_FAILED, "%s",
+               err_str);
+        rsp.op_errstr = err_str;
+        cli_rsp = &rsp;
+        glusterd_to_cli(req, cli_rsp, NULL, 0, NULL, (xdrproc_t)xdr_gf_cli_rsp,
+                        dict);
+
+        ret = 0;  // sent error to cli, prevent second reply
+    }
+
+    if (brickinfo_list)
+        GF_FREE(brickinfo_list);
+    subvol_matcher_destroy(subvols);
+    free(cli_req.dict.dict_val);  // its malloced by xdr
+
+    return ret;
+}
+
+int
+glusterd_handle_remove_brick(rpcsvc_request_t *req)
+{
+    return glusterd_big_locked_handler(req, __glusterd_handle_remove_brick);
+}
+
+static int
+_glusterd_restart_gsync_session(dict_t *this, char *key, data_t *value,
+                                void *data)
+{
+    char *slave = NULL;
+    char *slave_buf = NULL;
+    char *path_list = NULL;
+    char *slave_vol = NULL;
+    char *slave_host = NULL;
+    char *slave_url = NULL;
+    char *conf_path = NULL;
+    char **errmsg = NULL;
+    int ret = -1;
+    glusterd_gsync_status_temp_t *param = NULL;
+    gf_boolean_t is_running = _gf_false;
+
+    param = (glusterd_gsync_status_temp_t *)data;
+
+    GF_ASSERT(param);
+    GF_ASSERT(param->volinfo);
+
+    slave = strchr(value->data, ':');
+    if (slave) {
+        slave++;
+        slave_buf = gf_strdup(slave);
+        if (!slave_buf) {
+            gf_msg("glusterd", GF_LOG_ERROR, ENOMEM, GD_MSG_NO_MEMORY,
+                   "Failed to gf_strdup");
+            ret = -1;
+            goto out;
+        }
+    } else
+        return 0;
+
+    ret = dict_set_dynstrn(param->rsp_dict, "slave", SLEN("slave"), slave_buf);
+    if (ret) {
+        gf_msg("glusterd", GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+               "Unable to store slave");
+        if (slave_buf)
+            GF_FREE(slave_buf);
+        goto out;
+    }
+
+    ret = glusterd_get_slave_details_confpath(param->volinfo, param->rsp_dict,
+                                              &slave_url, &slave_host,
+                                              &slave_vol, &conf_path, errmsg);
+    if (ret) {
+        if (errmsg && *errmsg)
+            gf_msg("glusterd", GF_LOG_ERROR, 0,
+                   GD_MSG_SLAVE_CONFPATH_DETAILS_FETCH_FAIL, "%s", *errmsg);
+        else
+            gf_msg("glusterd", GF_LOG_ERROR, 0,
+                   GD_MSG_SLAVE_CONFPATH_DETAILS_FETCH_FAIL,
+                   "Unable to fetch slave or confpath details.");
+        goto out;
+    }
+
+    /* In cases that gsyncd is not running, we will not invoke it
+     * because of add-brick. */
+    ret = glusterd_check_gsync_running_local(param->volinfo->volname, slave,
+                                             conf_path, &is_running);
+    if (ret) {
+        gf_msg("glusterd", GF_LOG_ERROR, 0, GD_MSG_GSYNC_VALIDATION_FAIL,
+               "gsync running validation failed.");
+        goto out;
+    }
+    if (_gf_false == is_running) {
+        gf_msg_debug("glusterd", 0,
+                     "gsync session for %s and %s is"
+                     " not running on this node. Hence not restarting.",
+                     param->volinfo->volname, slave);
+        ret = 0;
+        goto out;
+    }
+
+    ret = glusterd_get_local_brickpaths(param->volinfo, &path_list);
+    if (!path_list) {
+        gf_msg_debug("glusterd", 0,
+                     "This node not being part of"
+                     " volume should not be running gsyncd. Hence"
+                     " no gsyncd process to restart.");
+        ret = 0;
+        goto out;
+    }
+
+    ret = glusterd_check_restart_gsync_session(
+        param->volinfo, slave, param->rsp_dict, path_list, conf_path, 0);
+    if (ret)
+        gf_msg("glusterd", GF_LOG_ERROR, 0, GD_MSG_GSYNC_RESTART_FAIL,
+               "Unable to restart gsync session.");
+
+out:
+    gf_msg_debug("glusterd", 0, "Returning %d.", ret);
+    return ret;
+}
+
+/* op-sm */
+
+int
+glusterd_op_perform_add_bricks(glusterd_volinfo_t *volinfo, int32_t count,
+                               char *bricks, dict_t *dict)
+{
+    char *brick = NULL;
+    int32_t i = 1;
+    char *brick_list = NULL;
+    char *free_ptr1 = NULL;
+    char *free_ptr2 = NULL;
+    char *saveptr = NULL;
+    int32_t ret = -1;
+    int32_t stripe_count = 0;
+    int32_t replica_count = 0;
+    int32_t arbiter_count = 0;
+    int32_t type = 0;
+    glusterd_brickinfo_t *brickinfo = NULL;
+    glusterd_gsync_status_temp_t param = {
+        0,
+    };
+    gf_boolean_t restart_needed = 0;
+    int brickid = 0;
+    char key[64] = "";
+    char *brick_mount_dir = NULL;
+    xlator_t *this = NULL;
+    glusterd_conf_t *conf = NULL;
+    gf_boolean_t is_valid_add_brick = _gf_false;
+    gf_boolean_t restart_shd = _gf_false;
+    struct statvfs brickstat = {
+        0,
+    };
+
+    this = THIS;
+    GF_ASSERT(this);
+    GF_ASSERT(volinfo);
+
+    conf = this->private;
+    GF_ASSERT(conf);
+
+    if (bricks) {
+        brick_list = gf_strdup(bricks);
+        free_ptr1 = brick_list;
+    }
+
+    if (count)
+        brick = strtok_r(brick_list + 1, " \n", &saveptr);
+
+    if (dict) {
+        ret = dict_get_int32n(dict, "stripe-count", SLEN("stripe-count"),
+                              &stripe_count);
+        if (!ret)
+            gf_msg(THIS->name, GF_LOG_INFO, errno, GD_MSG_DICT_GET_SUCCESS,
+                   "stripe-count is set %d", stripe_count);
+
+        ret = dict_get_int32n(dict, "replica-count", SLEN("replica-count"),
+                              &replica_count);
+        if (!ret)
+            gf_msg(THIS->name, GF_LOG_INFO, errno, GD_MSG_DICT_GET_SUCCESS,
+                   "replica-count is set %d", replica_count);
+        ret = dict_get_int32n(dict, "arbiter-count", SLEN("arbiter-count"),
+                              &arbiter_count);
+        if (!ret)
+            gf_msg(THIS->name, GF_LOG_INFO, errno, GD_MSG_DICT_GET_SUCCESS,
+                   "arbiter-count is set %d", arbiter_count);
+        ret = dict_get_int32n(dict, "type", SLEN("type"), &type);
+        if (!ret)
+            gf_msg(THIS->name, GF_LOG_INFO, errno, GD_MSG_DICT_GET_SUCCESS,
+                   "type is set %d, need to change it", type);
+    }
+
+    brickid = glusterd_get_next_available_brickid(volinfo);
+    if (brickid < 0)
+        goto out;
+    while (i <= count) {
+        ret = glusterd_brickinfo_new_from_brick(brick, &brickinfo, _gf_true,
+                                                NULL);
+        if (ret)
+            goto out;
+
+        GLUSTERD_ASSIGN_BRICKID_TO_BRICKINFO(brickinfo, volinfo, brickid++);
+
+        /* A bricks mount dir is required only by snapshots which were
+         * introduced in gluster-3.6.0
+         */
+        if (conf->op_version >= GD_OP_VERSION_3_6_0) {
+            brick_mount_dir = NULL;
+
+            snprintf(key, sizeof(key), "brick%d.mount_dir", i);
+            ret = dict_get_str(dict, key, &brick_mount_dir);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+                       "%s not present", key);
+                goto out;
+            }
+            strncpy(brickinfo->mount_dir, brick_mount_dir,
+                    SLEN(brickinfo->mount_dir));
+        }
+
+        ret = glusterd_resolve_brick(brickinfo);
+        if (ret)
+            goto out;
+
+        if (!gf_uuid_compare(brickinfo->uuid, MY_UUID)) {
+            ret = sys_statvfs(brickinfo->path, &brickstat);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_STATVFS_FAILED,
+                       "Failed to fetch disk utilization "
+                       "from the brick (%s:%s). Please check the health of "
+                       "the brick. Error code was %s",
+                       brickinfo->hostname, brickinfo->path, strerror(errno));
+
+                goto out;
+            }
+            brickinfo->statfs_fsid = brickstat.f_fsid;
+        }
+        if (stripe_count || replica_count) {
+            add_brick_at_right_order(brickinfo, volinfo, (i - 1), stripe_count,
+                                     replica_count);
+        } else {
+            cds_list_add_tail(&brickinfo->brick_list, &volinfo->bricks);
+        }
+        brick = strtok_r(NULL, " \n", &saveptr);
+        i++;
+        volinfo->brick_count++;
+    }
+
+    /* Gets changed only if the options are given in add-brick cli */
+    if (type)
+        volinfo->type = type;
+    /* performance.client-io-threads is turned on by default,
+     * however this has adverse effects on replicate volumes due to
+     * replication design issues, till that get addressed
+     * performance.client-io-threads option is turned off for all
+     * replicate volumes if not already explicitly enabled.
+     */
+    if (type && glusterd_is_volume_replicate(volinfo) &&
+        conf->op_version >= GD_OP_VERSION_3_12_2) {
+        ret = dict_set_nstrn(volinfo->dict, "performance.client-io-threads",
+                             SLEN("performance.client-io-threads"), "off",
+                             SLEN("off"));
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+                   "Failed to set "
+                   "performance.client-io-threads to off");
+            goto out;
+        }
+    }
+
+    if (replica_count) {
+        volinfo->replica_count = replica_count;
+    }
+    if (arbiter_count) {
+        volinfo->arbiter_count = arbiter_count;
+    }
+    if (stripe_count) {
+        volinfo->stripe_count = stripe_count;
+    }
+    volinfo->dist_leaf_count = glusterd_get_dist_leaf_count(volinfo);
+
+    /* backward compatibility */
+    volinfo->sub_count = ((volinfo->dist_leaf_count == 1)
+                              ? 0
+                              : volinfo->dist_leaf_count);
+
+    volinfo->subvol_count = (volinfo->brick_count / volinfo->dist_leaf_count);
+
+    ret = 0;
+    if (GLUSTERD_STATUS_STARTED != volinfo->status)
+        goto generate_volfiles;
+
+    ret = generate_brick_volfiles(volinfo);
+    if (ret)
+        goto out;
+
+    brick_list = gf_strdup(bricks);
+    free_ptr2 = brick_list;
+    i = 1;
+
+    if (count)
+        brick = strtok_r(brick_list + 1, " \n", &saveptr);
+
+    if (glusterd_is_volume_replicate(volinfo)) {
+        if (replica_count && conf->op_version >= GD_OP_VERSION_3_7_10) {
+            is_valid_add_brick = _gf_true;
+            if (volinfo->status == GLUSTERD_STATUS_STARTED) {
+                ret = volinfo->shd.svc.stop(&(volinfo->shd.svc), SIGTERM);
+                if (ret) {
+                    gf_msg("glusterd", GF_LOG_ERROR, 0,
+                           GD_MSG_GLUSTER_SERVICES_STOP_FAIL,
+                           "Failed to stop shd for %s.", volinfo->volname);
+                }
+                restart_shd = _gf_true;
+            }
+            ret = generate_dummy_client_volfiles(volinfo);
+            if (ret) {
+                gf_msg(THIS->name, GF_LOG_ERROR, 0, GD_MSG_VOLFILE_CREATE_FAIL,
+                       "Failed to create volfile.");
+                goto out;
+            }
+        }
+    }
+
+    while (i <= count) {
+        ret = glusterd_volume_brickinfo_get_by_brick(brick, volinfo, &brickinfo,
+                                                     _gf_true);
+        if (ret)
+            goto out;
+
+        if (gf_uuid_is_null(brickinfo->uuid)) {
+            ret = glusterd_resolve_brick(brickinfo);
+            if (ret) {
+                gf_msg("glusterd", GF_LOG_ERROR, 0, GD_MSG_RESOLVE_BRICK_FAIL,
+                       FMTSTR_RESOLVE_BRICK, brickinfo->hostname,
+                       brickinfo->path);
+                goto out;
+            }
+        }
+
+        /* if the volume is a replicate volume, do: */
+        if (is_valid_add_brick) {
+            if (!gf_uuid_compare(brickinfo->uuid, MY_UUID)) {
+                ret = glusterd_handle_replicate_brick_ops(volinfo, brickinfo,
+                                                          GD_OP_ADD_BRICK);
+                if (ret < 0)
+                    goto out;
+            }
+        }
+        ret = glusterd_brick_start(volinfo, brickinfo, _gf_true, _gf_false);
+        if (ret)
+            goto out;
+        i++;
+        brick = strtok_r(NULL, " \n", &saveptr);
+
+        /* Check if the brick is added in this node, and set
+         * the restart_needed flag. */
+        if ((!gf_uuid_compare(brickinfo->uuid, MY_UUID)) && !restart_needed) {
+            restart_needed = 1;
+            gf_msg_debug("glusterd", 0,
+                         "Restart gsyncd session, if it's already "
+                         "running.");
+        }
+    }
+
+    /* If the restart_needed flag is set, restart gsyncd sessions for that
+     * particular master with all the slaves. */
+    if (restart_needed) {
+        param.rsp_dict = dict;
+        param.volinfo = volinfo;
+        dict_foreach(volinfo->gsync_slaves, _glusterd_restart_gsync_session,
+                     &param);
+    }
+
+generate_volfiles:
+    if (conf->op_version <= GD_OP_VERSION_3_7_5) {
+        ret = glusterd_create_volfiles_and_notify_services(volinfo);
+    } else {
+        /*
+         * The cluster is operating at version greater than
+         * gluster-3.7.5. So no need to sent volfile fetch
+         * request in commit phase, the same will be done
+         * in post validate phase with v3 framework.
+         */
+    }
+
+out:
+    GF_FREE(free_ptr1);
+    GF_FREE(free_ptr2);
+    if (restart_shd) {
+        if (volinfo->shd.svc.manager(&(volinfo->shd.svc), volinfo,
+                                     PROC_START_NO_WAIT)) {
+            gf_msg("glusterd", GF_LOG_CRITICAL, 0,
+                   GD_MSG_GLUSTER_SERVICE_START_FAIL,
+                   "Failed to start shd for %s.", volinfo->volname);
+        }
+    }
+
+    gf_msg_debug("glusterd", 0, "Returning %d", ret);
+    return ret;
+}
+
+int
+glusterd_op_perform_remove_brick(glusterd_volinfo_t *volinfo, char *brick,
+                                 int force, int *need_migrate)
+{
+    glusterd_brickinfo_t *brickinfo = NULL;
+    int32_t ret = -1;
+    glusterd_conf_t *priv = NULL;
+
+    GF_ASSERT(volinfo);
+    GF_ASSERT(brick);
+
+    priv = THIS->private;
+    GF_ASSERT(priv);
+
+    ret = glusterd_volume_brickinfo_get_by_brick(brick, volinfo, &brickinfo,
+                                                 _gf_false);
+    if (ret)
+        goto out;
+
+    ret = glusterd_resolve_brick(brickinfo);
+    if (ret)
+        goto out;
+
+    glusterd_volinfo_reset_defrag_stats(volinfo);
+
+    if (!gf_uuid_compare(brickinfo->uuid, MY_UUID)) {
+        /* Only if the brick is in this glusterd, do the rebalance */
+        if (need_migrate)
+            *need_migrate = 1;
+    }
+
+    if (force) {
+        ret = glusterd_brick_stop(volinfo, brickinfo, _gf_true);
+        if (ret) {
+            gf_msg(THIS->name, GF_LOG_ERROR, 0, GD_MSG_BRICK_STOP_FAIL,
+                   "Unable to stop "
+                   "glusterfs, ret: %d",
+                   ret);
+        }
+        goto out;
+    }
+
+    brickinfo->decommissioned = 1;
+    ret = 0;
+out:
+    gf_msg_debug("glusterd", 0, "Returning %d", ret);
+    return ret;
+}
+
+int
+glusterd_op_stage_add_brick(dict_t *dict, char **op_errstr, dict_t *rsp_dict)
+{
+    int ret = 0;
+    char *volname = NULL;
+    int count = 0;
+    int replica_count = 0;
+    int arbiter_count = 0;
+    int i = 0;
+    int32_t local_brick_count = 0;
+    char *bricks = NULL;
+    char *brick_list = NULL;
+    char *saveptr = NULL;
+    char *free_ptr = NULL;
+    char *brick = NULL;
+    glusterd_brickinfo_t *brickinfo = NULL;
+    glusterd_volinfo_t *volinfo = NULL;
+    xlator_t *this = NULL;
+    char msg[4096] = "";
+    char key[64] = "";
+    gf_boolean_t brick_alloc = _gf_false;
+    char *all_bricks = NULL;
+    char *str_ret = NULL;
+    gf_boolean_t is_force = _gf_false;
+    glusterd_conf_t *conf = NULL;
+    int32_t len = 0;
+
+    this = THIS;
+    GF_ASSERT(this);
+    conf = this->private;
+    GF_ASSERT(conf);
+
+    ret = dict_get_strn(dict, "volname", SLEN("volname"), &volname);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to get volume name");
+        goto out;
+    }
+
+    ret = glusterd_volinfo_find(volname, &volinfo);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_VOL_NOT_FOUND,
+               "Unable to find volume: %s", volname);
+        goto out;
+    }
+
+    ret = glusterd_validate_volume_id(dict, volinfo);
+    if (ret)
+        goto out;
+
+    ret = dict_get_int32n(dict, "replica-count", SLEN("replica-count"),
+                          &replica_count);
+    if (ret) {
+        gf_msg_debug(this->name, 0, "Unable to get replica count");
+    }
+
+    if (replica_count > 0) {
+        ret = op_version_check(this, GD_OP_VER_PERSISTENT_AFR_XATTRS, msg,
+                               sizeof(msg));
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_OP_VERSION_MISMATCH,
+                   "%s", msg);
+            *op_errstr = gf_strdup(msg);
+            goto out;
+        }
+    }
+
+    glusterd_add_peers_to_auth_list(volname);
+
+    if (replica_count && glusterd_is_volume_replicate(volinfo)) {
+        /* Do not allow add-brick for stopped volumes when replica-count
+         * is being increased.
+         */
+        if (GLUSTERD_STATUS_STOPPED == volinfo->status &&
+            conf->op_version >= GD_OP_VERSION_3_7_10) {
+            ret = -1;
+            snprintf(msg, sizeof(msg),
+                     " Volume must not be in"
+                     " stopped state when replica-count needs to "
+                     " be increased.");
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_BRICK_ADD_FAIL, "%s",
+                   msg);
+            *op_errstr = gf_strdup(msg);
+            goto out;
+        }
+        /* op-version check for replica 2 to arbiter conversion. If we
+         * don't have this check, an older peer added as arbiter brick
+         * will not have the  arbiter xlator in its volfile. */
+        if ((replica_count == 3) && (conf->op_version < GD_OP_VERSION_3_8_0)) {
+            ret = dict_get_int32n(dict, "arbiter-count", SLEN("arbiter-count"),
+                                  &arbiter_count);
+            if (ret) {
+                gf_msg_debug(this->name, 0,
+                             "No arbiter count present in the dict");
+            } else if (arbiter_count == 1) {
+                ret = -1;
+                snprintf(msg, sizeof(msg),
+                         "Cluster op-version must "
+                         "be >= 30800 to add arbiter brick to a "
+                         "replica 2 volume.");
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_BRICK_ADD_FAIL, "%s",
+                       msg);
+                *op_errstr = gf_strdup(msg);
+                goto out;
+            }
+        }
+        /* Do not allow increasing replica count for arbiter volumes. */
+        if (volinfo->arbiter_count) {
+            ret = -1;
+            snprintf(msg, sizeof(msg),
+                     "Increasing replica count "
+                     "for arbiter volumes is not supported.");
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_BRICK_ADD_FAIL, "%s",
+                   msg);
+            *op_errstr = gf_strdup(msg);
+            goto out;
+        }
+    }
+
+    is_force = dict_get_str_boolean(dict, "force", _gf_false);
+
+    /* Check brick order if the volume type is replicate or disperse. If
+     * force at the end of command not given then check brick order.
+     * doing this check at the originator node is sufficient.
+     */
+
+    if (!is_force && is_origin_glusterd(dict)) {
+        ret = 0;
+        if (volinfo->type == GF_CLUSTER_TYPE_REPLICATE) {
+            gf_msg_debug(this->name, 0,
+                         "Replicate cluster type "
+                         "found. Checking brick order.");
+            if (replica_count)
+                ret = glusterd_check_brick_order(dict, msg, volinfo->type,
+                                                 &volname, &bricks, &count,
+                                                 replica_count);
+            else
+                ret = glusterd_check_brick_order(dict, msg, volinfo->type,
+                                                 &volname, &bricks, &count,
+                                                 volinfo->replica_count);
+        } else if (volinfo->type == GF_CLUSTER_TYPE_DISPERSE) {
+            gf_msg_debug(this->name, 0,
+                         "Disperse cluster type"
+                         " found. Checking brick order.");
+            ret = glusterd_check_brick_order(dict, msg, volinfo->type, &volname,
+                                             &bricks, &count,
+                                             volinfo->disperse_count);
+        }
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_BAD_BRKORDER,
+                   "Not adding brick because of "
+                   "bad brick order. %s",
+                   msg);
+            *op_errstr = gf_strdup(msg);
+            goto out;
+        }
+    }
+
+    if (volinfo->replica_count < replica_count && !is_force) {
+        cds_list_for_each_entry(brickinfo, &volinfo->bricks, brick_list)
+        {
+            if (gf_uuid_compare(brickinfo->uuid, MY_UUID))
+                continue;
+            if (brickinfo->status == GF_BRICK_STOPPED) {
+                ret = -1;
+                len = snprintf(msg, sizeof(msg),
+                               "Brick %s "
+                               "is down, changing replica "
+                               "count needs all the bricks "
+                               "to be up to avoid data loss",
+                               brickinfo->path);
+                if (len < 0) {
+                    strcpy(msg, "<error>");
+                }
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_BRICK_ADD_FAIL, "%s",
+                       msg);
+                *op_errstr = gf_strdup(msg);
+                goto out;
+            }
+        }
+    }
+
+    if (conf->op_version > GD_OP_VERSION_3_7_5 && is_origin_glusterd(dict)) {
+        ret = glusterd_validate_quorum(this, GD_OP_ADD_BRICK, dict, op_errstr);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_CRITICAL, 0, GD_MSG_SERVER_QUORUM_NOT_MET,
+                   "Server quorum not met. Rejecting operation.");
+            goto out;
+        }
+    } else {
+        /* Case 1: conf->op_version <= GD_OP_VERSION_3_7_5
+         *         in this case the add-brick is running
+         *         syncop framework that will do a quorum
+         *         check by default
+         * Case 2: We don't need to do quorum check on every
+         *         node, only originator glusterd need to
+         *         check for quorum
+         * So nothing need to be done in else
+         */
+    }
+
+    if (glusterd_is_defrag_on(volinfo)) {
+        snprintf(msg, sizeof(msg),
+                 "Volume name %s rebalance is in "
+                 "progress. Please retry after completion",
+                 volname);
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_OIP_RETRY_LATER, "%s", msg);
+        *op_errstr = gf_strdup(msg);
+        ret = -1;
+        goto out;
+    }
+
+    if (volinfo->snap_count > 0 || !cds_list_empty(&volinfo->snap_volumes)) {
+        snprintf(msg, sizeof(msg),
+                 "Volume %s  has %" PRIu64
+                 " snapshots. "
+                 "Changing the volume configuration will not effect snapshots."
+                 "But the snapshot brick mount should be intact to "
+                 "make them function.",
+                 volname, volinfo->snap_count);
+        gf_msg("glusterd", GF_LOG_WARNING, 0, GD_MSG_SNAP_WARN, "%s", msg);
+        msg[0] = '\0';
+    }
+
+    if (!count) {
+        ret = dict_get_int32n(dict, "count", SLEN("count"), &count);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+                   "Unable to get count");
+            goto out;
+        }
+    }
+
+    if (!bricks) {
+        ret = dict_get_strn(dict, "bricks", SLEN("bricks"), &bricks);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+                   "Unable to get bricks");
+            goto out;
+        }
+    }
+
+    if (bricks) {
+        brick_list = gf_strdup(bricks);
+        all_bricks = gf_strdup(bricks);
+        free_ptr = brick_list;
+    }
+
+    if (count)
+        brick = strtok_r(brick_list + 1, " \n", &saveptr);
+
+    while (i < count) {
+        if (!glusterd_store_is_valid_brickpath(volname, brick) ||
+            !glusterd_is_valid_volfpath(volname, brick)) {
+            snprintf(msg, sizeof(msg),
+                     "brick path %s is "
+                     "too long",
+                     brick);
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_BRKPATH_TOO_LONG, "%s",
+                   msg);
+            *op_errstr = gf_strdup(msg);
+
+            ret = -1;
+            goto out;
+        }
+
+        ret = glusterd_brickinfo_new_from_brick(brick, &brickinfo, _gf_true,
+                                                NULL);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_BRICK_NOT_FOUND,
+                   "Add-brick: Unable"
+                   " to get brickinfo");
+            goto out;
+        }
+        brick_alloc = _gf_true;
+
+        ret = glusterd_new_brick_validate(brick, brickinfo, msg, sizeof(msg),
+                                          NULL);
+        if (ret) {
+            *op_errstr = gf_strdup(msg);
+            ret = -1;
+            goto out;
+        }
+
+        if (!gf_uuid_compare(brickinfo->uuid, MY_UUID)) {
+            ret = glusterd_validate_and_create_brickpath(
+                brickinfo, volinfo->volume_id, volinfo->volname, op_errstr,
+                is_force, _gf_false);
+            if (ret)
+                goto out;
+
+            /* A bricks mount dir is required only by snapshots which were
+             * introduced in gluster-3.6.0
+             */
+            if (conf->op_version >= GD_OP_VERSION_3_6_0) {
+                ret = glusterd_get_brick_mount_dir(
+                    brickinfo->path, brickinfo->hostname, brickinfo->mount_dir);
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0,
+                           GD_MSG_BRICK_MOUNTDIR_GET_FAIL,
+                           "Failed to get brick mount_dir");
+                    goto out;
+                }
+
+                snprintf(key, sizeof(key), "brick%d.mount_dir", i + 1);
+                ret = dict_set_dynstr_with_alloc(rsp_dict, key,
+                                                 brickinfo->mount_dir);
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_ERROR, errno,
+                           GD_MSG_DICT_SET_FAILED, "Failed to set %s", key);
+                    goto out;
+                }
+            }
+
+            local_brick_count = i + 1;
+        }
+
+        glusterd_brickinfo_delete(brickinfo);
+        brick_alloc = _gf_false;
+        brickinfo = NULL;
+        brick = strtok_r(NULL, " \n", &saveptr);
+        i++;
+    }
+
+    ret = dict_set_int32n(rsp_dict, "brick_count", SLEN("brick_count"),
+                          local_brick_count);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+               "Failed to set local_brick_count");
+        goto out;
+    }
+
+out:
+    GF_FREE(free_ptr);
+    if (brick_alloc && brickinfo)
+        glusterd_brickinfo_delete(brickinfo);
+    GF_FREE(str_ret);
+    GF_FREE(all_bricks);
+
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+
+    return ret;
+}
+
+int
+glusterd_remove_brick_validate_bricks(gf1_op_commands cmd, int32_t brick_count,
+                                      dict_t *dict, glusterd_volinfo_t *volinfo,
+                                      char **errstr,
+                                      gf_cli_defrag_type cmd_defrag)
+{
+    char *brick = NULL;
+    char msg[2048] = "";
+    char key[64] = "";
+    int keylen;
+    glusterd_brickinfo_t *brickinfo = NULL;
+    glusterd_peerinfo_t *peerinfo = NULL;
+    int i = 0;
+    int ret = -1;
+    char pidfile[PATH_MAX + 1] = {
+        0,
+    };
+    glusterd_conf_t *priv = THIS->private;
+    int pid = -1;
+    xlator_t *this = THIS;
+    GF_ASSERT(this);
+
+    /* Check whether all the nodes of the bricks to be removed are
+     * up, if not fail the operation */
+    for (i = 1; i <= brick_count; i++) {
+        keylen = snprintf(key, sizeof(key), "brick%d", i);
+        ret = dict_get_strn(dict, key, keylen, &brick);
+        if (ret) {
+            snprintf(msg, sizeof(msg), "Unable to get %s", key);
+            gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+                    "key=%s", key, NULL);
+            *errstr = gf_strdup(msg);
+            goto out;
+        }
+
+        ret = glusterd_volume_brickinfo_get_by_brick(brick, volinfo, &brickinfo,
+                                                     _gf_false);
+        if (ret) {
+            snprintf(msg, sizeof(msg),
+                     "Incorrect brick "
+                     "%s for volume %s",
+                     brick, volinfo->volname);
+            gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_INCORRECT_BRICK,
+                    "Brick=%s, Volume=%s", brick, volinfo->volname, NULL);
+            *errstr = gf_strdup(msg);
+            goto out;
+        }
+        /* Do not allow commit if the bricks are not decommissioned
+         * if its a remove brick commit
+         */
+        if (!brickinfo->decommissioned && cmd == GF_OP_CMD_COMMIT) {
+            snprintf(msg, sizeof(msg),
+                     "Brick %s "
+                     "is not decommissioned. "
+                     "Use start or force option",
+                     brick);
+            gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_BRICK_NOT_DECOM,
+                    "Use 'start' or 'force' option, Brick=%s", brick, NULL);
+            *errstr = gf_strdup(msg);
+            ret = -1;
+            goto out;
+        }
+
+        if (glusterd_is_local_brick(THIS, volinfo, brickinfo)) {
+            switch (cmd) {
+                case GF_OP_CMD_START:
+                    goto check;
+                case GF_OP_CMD_NONE:
+                default:
+                    break;
+            }
+
+            switch (cmd_defrag) {
+                case GF_DEFRAG_CMD_NONE:
+                default:
+                    continue;
+            }
+        check:
+            if (brickinfo->status != GF_BRICK_STARTED) {
+                snprintf(msg, sizeof(msg),
+                         "Found stopped "
+                         "brick %s. Use force option to "
+                         "remove the offline brick",
+                         brick);
+                gf_smsg(
+                    this->name, GF_LOG_ERROR, errno, GD_MSG_BRICK_STOPPED,
+                    "Use 'force' option to remove the offline brick, Brick=%s",
+                    brick, NULL);
+                *errstr = gf_strdup(msg);
+                ret = -1;
+                goto out;
+            }
+            GLUSTERD_GET_BRICK_PIDFILE(pidfile, volinfo, brickinfo, priv);
+            if (!gf_is_service_running(pidfile, &pid)) {
+                snprintf(msg, sizeof(msg),
+                         "Found dead "
+                         "brick %s",
+                         brick);
+                gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_BRICK_DEAD,
+                        "Brick=%s", brick, NULL);
+                *errstr = gf_strdup(msg);
+                ret = -1;
+                goto out;
+            } else {
+                ret = 0;
+            }
+            continue;
+        }
+
+        RCU_READ_LOCK;
+        peerinfo = glusterd_peerinfo_find_by_uuid(brickinfo->uuid);
+        if (!peerinfo) {
+            RCU_READ_UNLOCK;
+            snprintf(msg, sizeof(msg),
+                     "Host node of the "
+                     "brick %s is not in cluster",
+                     brick);
+            gf_smsg(this->name, GF_LOG_ERROR, errno,
+                    GD_MSG_BRICK_HOST_NOT_FOUND, "Brick=%s", brick, NULL);
+            *errstr = gf_strdup(msg);
+            ret = -1;
+            goto out;
+        }
+        if (!peerinfo->connected) {
+            RCU_READ_UNLOCK;
+            snprintf(msg, sizeof(msg),
+                     "Host node of the "
+                     "brick %s is down",
+                     brick);
+            gf_smsg(this->name, GF_LOG_ERROR, 0, GD_MSG_BRICK_HOST_DOWN,
+                    "Brick=%s", brick, NULL);
+            *errstr = gf_strdup(msg);
+            ret = -1;
+            goto out;
+        }
+        RCU_READ_UNLOCK;
+    }
+
+out:
+    return ret;
+}
+
+int
+glusterd_op_stage_remove_brick(dict_t *dict, char **op_errstr)
+{
+    int ret = -1;
+    char *volname = NULL;
+    glusterd_volinfo_t *volinfo = NULL;
+    char *errstr = NULL;
+    int32_t brick_count = 0;
+    char msg[2048] = "";
+    int32_t flag = 0;
+    gf1_op_commands cmd = GF_OP_CMD_NONE;
+    char *task_id_str = NULL;
+    xlator_t *this = NULL;
+    gsync_status_param_t param = {
+        0,
+    };
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    ret = op_version_check(this, GD_OP_VER_PERSISTENT_AFR_XATTRS, msg,
+                           sizeof(msg));
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_OP_VERSION_MISMATCH, "%s",
+               msg);
+        *op_errstr = gf_strdup(msg);
+        goto out;
+    }
+
+    ret = dict_get_strn(dict, "volname", SLEN("volname"), &volname);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+               "Unable to get volume name");
+        goto out;
+    }
+
+    ret = glusterd_volinfo_find(volname, &volinfo);
+
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_VOL_NOT_FOUND,
+               "Volume %s does not exist", volname);
+        goto out;
+    }
+
+    ret = glusterd_validate_volume_id(dict, volinfo);
+    if (ret)
+        goto out;
+
+    ret = dict_get_int32n(dict, "command", SLEN("command"), &flag);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to get brick command");
+        goto out;
+    }
+    cmd = flag;
+
+    ret = dict_get_int32n(dict, "count", SLEN("count"), &brick_count);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to get brick count");
+        goto out;
+    }
+
+    ret = 0;
+    if (volinfo->brick_count == brick_count) {
+        errstr = gf_strdup(
+            "Deleting all the bricks of the "
+            "volume is not allowed");
+        gf_smsg(this->name, GF_LOG_ERROR, 0, GD_MSG_BRICK_DELETE, NULL);
+        ret = -1;
+        goto out;
+    }
+
+    ret = -1;
+    switch (cmd) {
+        case GF_OP_CMD_NONE:
+            errstr = gf_strdup("no remove-brick command issued");
+            gf_smsg(this->name, GF_LOG_ERROR, 0, GD_MSG_BRICK_NO_REMOVE_CMD,
+                    NULL);
+            goto out;
+
+        case GF_OP_CMD_STATUS:
+            ret = 0;
+            goto out;
+        case GF_OP_CMD_START: {
+            if ((volinfo->type == GF_CLUSTER_TYPE_REPLICATE) &&
+                dict_getn(dict, "replica-count", SLEN("replica-count"))) {
+                snprintf(msg, sizeof(msg),
+                         "Migration of data is not "
+                         "needed when reducing replica count. Use the"
+                         " 'force' option");
+                errstr = gf_strdup(msg);
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_USE_THE_FORCE, "%s",
+                       errstr);
+                goto out;
+            }
+
+            if (GLUSTERD_STATUS_STARTED != volinfo->status) {
+                snprintf(msg, sizeof(msg),
+                         "Volume %s needs "
+                         "to be started before remove-brick "
+                         "(you can use 'force' or 'commit' "
+                         "to override this behavior)",
+                         volinfo->volname);
+                errstr = gf_strdup(msg);
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_VOL_NOT_STARTED,
+                       "%s", errstr);
+                goto out;
+            }
+            if (!gd_is_remove_brick_committed(volinfo)) {
+                snprintf(msg, sizeof(msg),
+                         "An earlier remove-brick "
+                         "task exists for volume %s. Either commit it"
+                         " or stop it before starting a new task.",
+                         volinfo->volname);
+                errstr = gf_strdup(msg);
+                gf_msg(this->name, GF_LOG_ERROR, 0,
+                       GD_MSG_OLD_REMOVE_BRICK_EXISTS,
+                       "Earlier remove-brick"
+                       " task exists for volume %s.",
+                       volinfo->volname);
+                goto out;
+            }
+            if (glusterd_is_defrag_on(volinfo)) {
+                errstr = gf_strdup(
+                    "Rebalance is in progress. Please "
+                    "retry after completion");
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_OIP_RETRY_LATER,
+                       "%s", errstr);
+                goto out;
+            }
+
+            /* Check if the connected clients are all of version
+             * glusterfs-3.6 and higher. This is needed to prevent some data
+             * loss issues that could occur when older clients are connected
+             * when rebalance is run.
+             */
+            ret = glusterd_check_client_op_version_support(
+                volname, GD_OP_VERSION_3_6_0, NULL);
+            if (ret) {
+                ret = gf_asprintf(op_errstr,
+                                  "Volume %s has one or "
+                                  "more connected clients of a version"
+                                  " lower than GlusterFS-v3.6.0. "
+                                  "Starting remove-brick in this state "
+                                  "could lead to data loss.\nPlease "
+                                  "disconnect those clients before "
+                                  "attempting this command again.",
+                                  volname);
+                goto out;
+            }
+
+            if (volinfo->snap_count > 0 ||
+                !cds_list_empty(&volinfo->snap_volumes)) {
+                snprintf(msg, sizeof(msg),
+                         "Volume %s  has %" PRIu64
+                         " snapshots. "
+                         "Changing the volume configuration will not effect "
+                         "snapshots."
+                         "But the snapshot brick mount should be intact to "
+                         "make them function.",
+                         volname, volinfo->snap_count);
+                gf_msg("glusterd", GF_LOG_WARNING, 0, GD_MSG_SNAP_WARN, "%s",
+                       msg);
+                msg[0] = '\0';
+            }
+
+            ret = glusterd_remove_brick_validate_bricks(
+                cmd, brick_count, dict, volinfo, &errstr, GF_DEFRAG_CMD_NONE);
+            if (ret)
+                goto out;
+
+            if (is_origin_glusterd(dict)) {
+                ret = glusterd_generate_and_set_task_id(
+                    dict, GF_REMOVE_BRICK_TID_KEY,
+                    SLEN(GF_REMOVE_BRICK_TID_KEY));
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_TASKID_GEN_FAIL,
+                           "Failed to generate task-id");
+                    goto out;
+                }
+            } else {
+                ret = dict_get_strn(dict, GF_REMOVE_BRICK_TID_KEY,
+                                    SLEN(GF_REMOVE_BRICK_TID_KEY),
+                                    &task_id_str);
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_WARNING, errno,
+                           GD_MSG_DICT_GET_FAILED, "Missing remove-brick-id");
+                    ret = 0;
+                }
+            }
+            break;
+        }
+
+        case GF_OP_CMD_STOP:
+            ret = 0;
+            break;
+
+        case GF_OP_CMD_COMMIT:
+            if (volinfo->decommission_in_progress) {
+                errstr = gf_strdup(
+                    "use 'force' option as migration "
+                    "is in progress");
+                gf_smsg(this->name, GF_LOG_WARNING, 0, GD_MSG_MIGRATION_PROG,
+                        "Use 'force' option", NULL);
+                goto out;
+            }
+
+            if (volinfo->rebal.defrag_status == GF_DEFRAG_STATUS_FAILED) {
+                errstr = gf_strdup(
+                    "use 'force' option as migration "
+                    "has failed");
+                gf_smsg(this->name, GF_LOG_WARNING, 0, GD_MSG_MIGRATION_FAIL,
+                        "Use 'force' option", NULL);
+                goto out;
+            }
+
+            if (volinfo->rebal.defrag_status == GF_DEFRAG_STATUS_COMPLETE) {
+                if (volinfo->rebal.rebalance_failures > 0 ||
+                    volinfo->rebal.skipped_files > 0) {
+                    errstr = gf_strdup(
+                        "use 'force' option as migration "
+                        "of some files might have been skipped or "
+                        "has failed");
+                    gf_smsg(this->name, GF_LOG_WARNING, 0,
+                            GD_MSG_MIGRATION_FAIL,
+                            "Use 'force' option, some files might have been "
+                            "skipped",
+                            NULL);
+                    goto out;
+                }
+            }
+
+            ret = glusterd_remove_brick_validate_bricks(
+                cmd, brick_count, dict, volinfo, &errstr, GF_DEFRAG_CMD_NONE);
+            if (ret)
+                goto out;
+
+            /* If geo-rep is configured, for this volume, it should be
+             * stopped.
+             */
+            param.volinfo = volinfo;
+            ret = glusterd_check_geo_rep_running(&param, op_errstr);
+            if (ret || param.is_active) {
+                ret = -1;
+                goto out;
+            }
+
+            break;
+
+        case GF_OP_CMD_COMMIT_FORCE:
+        case GF_OP_CMD_DETACH_START:
+        case GF_OP_CMD_DETACH_COMMIT:
+        case GF_OP_CMD_DETACH_COMMIT_FORCE:
+        case GF_OP_CMD_STOP_DETACH_TIER:
+            break;
+    }
+    ret = 0;
+
+out:
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    if (ret && errstr) {
+        if (op_errstr)
+            *op_errstr = errstr;
+    }
+    if (!op_errstr && errstr)
+        GF_FREE(errstr);
+    return ret;
+}
+
+int
+glusterd_remove_brick_migrate_cbk(glusterd_volinfo_t *volinfo,
+                                  gf_defrag_status_t status)
+{
+    int ret = 0;
+
+#if 0 /* TODO: enable this behavior once cluster-wide awareness comes for      \
+         defrag cbk function */
+        glusterd_brickinfo_t *brickinfo = NULL;
+        glusterd_brickinfo_t *tmp = NULL;
+
+        switch (status) {
+        case GF_DEFRAG_STATUS_PAUSED:
+        case GF_DEFRAG_STATUS_FAILED:
+                /* No changes required in the volume file.
+                   everything should remain as is */
+                break;
+        case GF_DEFRAG_STATUS_STOPPED:
+                /* Fall back to the old volume file */
+                cds_list_for_each_entry_safe (brickinfo, tmp, &volinfo->bricks,
+                                              brick_list) {
+                        if (!brickinfo->decommissioned)
+                                continue;
+                        brickinfo->decommissioned = 0;
+                }
+                break;
+
+        case GF_DEFRAG_STATUS_COMPLETE:
+                /* Done with the task, you can remove the brick from the
+                   volume file */
+                cds_list_for_each_entry_safe (brickinfo, tmp, &volinfo->bricks,
+                                              brick_list) {
+                        if (!brickinfo->decommissioned)
+                                continue;
+                        gf_log (THIS->name, GF_LOG_INFO, "removing the brick %s",
+                                brickinfo->path);
+                        brickinfo->decommissioned = 0;
+                        if (GLUSTERD_STATUS_STARTED == volinfo->status) {
+                            /*TODO: use the 'atomic' flavour of brick_stop*/
+                                ret = glusterd_brick_stop (volinfo, brickinfo);
+                                if (ret) {
+                                        gf_log (THIS->name, GF_LOG_ERROR,
+                                                "Unable to stop glusterfs (%d)", ret);
+                                }
+                        }
+                        glusterd_delete_brick (volinfo, brickinfo);
+                }
+                break;
+
+        default:
+                GF_ASSERT (!"cbk function called with wrong status");
+                break;
+        }
+
+        ret = glusterd_create_volfiles_and_notify_services (volinfo);
+        if (ret)
+                gf_log (THIS->name, GF_LOG_ERROR,
+                        "Unable to write volume files (%d)", ret);
+
+        ret = glusterd_store_volinfo (volinfo, GLUSTERD_VOLINFO_VER_AC_INCREMENT);
+        if (ret)
+                gf_log (THIS->name, GF_LOG_ERROR,
+                        "Unable to store volume info (%d)", ret);
+
+
+        if (GLUSTERD_STATUS_STARTED == volinfo->status) {
+                ret = glusterd_check_generate_start_nfs ();
+                if (ret)
+                        gf_log (THIS->name, GF_LOG_ERROR,
+                                "Unable to start nfs process (%d)", ret);
+        }
+
+#endif
+
+    volinfo->decommission_in_progress = 0;
+    return ret;
+}
+
+int
+glusterd_op_add_brick(dict_t *dict, char **op_errstr)
+{
+    int ret = 0;
+    char *volname = NULL;
+    glusterd_conf_t *priv = NULL;
+    glusterd_volinfo_t *volinfo = NULL;
+    xlator_t *this = NULL;
+    char *bricks = NULL;
+    int32_t count = 0;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    ret = dict_get_strn(dict, "volname", SLEN("volname"), &volname);
+
+    if (ret) {
+        gf_msg("glusterd", GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to get volume name");
+        goto out;
+    }
+
+    ret = glusterd_volinfo_find(volname, &volinfo);
+
+    if (ret) {
+        gf_msg("glusterd", GF_LOG_ERROR, EINVAL, GD_MSG_VOL_NOT_FOUND,
+               "Unable to allocate memory");
+        goto out;
+    }
+
+    ret = dict_get_int32n(dict, "count", SLEN("count"), &count);
+    if (ret) {
+        gf_msg("glusterd", GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to get count");
+        goto out;
+    }
+
+    ret = dict_get_strn(dict, "bricks", SLEN("bricks"), &bricks);
+    if (ret) {
+        gf_msg("glusterd", GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to get bricks");
+        goto out;
+    }
+
+    ret = glusterd_op_perform_add_bricks(volinfo, count, bricks, dict);
+    if (ret) {
+        gf_msg("glusterd", GF_LOG_ERROR, 0, GD_MSG_BRICK_ADD_FAIL,
+               "Unable to add bricks");
+        goto out;
+    }
+    if (priv->op_version <= GD_OP_VERSION_3_7_5) {
+        ret = glusterd_store_volinfo(volinfo,
+                                     GLUSTERD_VOLINFO_VER_AC_INCREMENT);
+        if (ret)
+            goto out;
+    } else {
+        /*
+         * The cluster is operating at version greater than
+         * gluster-3.7.5. So no need to store volfiles
+         * in commit phase, the same will be done
+         * in post validate phase with v3 framework.
+         */
+    }
+
+    if (GLUSTERD_STATUS_STARTED == volinfo->status)
+        ret = glusterd_svcs_manager(volinfo);
+
+out:
+    return ret;
+}
+
+int
+glusterd_post_commit_add_brick(dict_t *dict, char **op_errstr)
+{
+    int ret = 0;
+    char *volname = NULL;
+
+    ret = dict_get_strn(dict, "volname", SLEN("volname"), &volname);
+
+    if (ret) {
+        gf_msg(THIS->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to get volume name");
+        goto out;
+    }
+    ret = glusterd_replace_old_auth_allow_list(volname);
+out:
+    return ret;
+}
+
+int
+glusterd_post_commit_replace_brick(dict_t *dict, char **op_errstr)
+{
+    int ret = 0;
+    char *volname = NULL;
+
+    ret = dict_get_strn(dict, "volname", SLEN("volname"), &volname);
+
+    if (ret) {
+        gf_msg(THIS->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to get volume name");
+        goto out;
+    }
+    ret = glusterd_replace_old_auth_allow_list(volname);
+out:
+    return ret;
+}
+
+int
+glusterd_set_rebalance_id_for_remove_brick(dict_t *req_dict, dict_t *rsp_dict)
+{
+    int ret = -1;
+    char *volname = NULL;
+    glusterd_volinfo_t *volinfo = NULL;
+    char msg[2048] = {0};
+    char *task_id_str = NULL;
+    xlator_t *this = NULL;
+    int32_t cmd = 0;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(rsp_dict);
+    GF_ASSERT(req_dict);
+
+    ret = dict_get_strn(rsp_dict, "volname", SLEN("volname"), &volname);
+    if (ret) {
+        gf_msg_debug(this->name, 0, "volname not found");
+        goto out;
+    }
+
+    ret = glusterd_volinfo_find(volname, &volinfo);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, GD_MSG_VOL_NOT_FOUND,
+               "Unable to allocate memory");
+        goto out;
+    }
+
+    ret = dict_get_int32n(rsp_dict, "command", SLEN("command"), &cmd);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to get command");
+        goto out;
+    }
+
+    /* remove brick task id is generted in glusterd_op_stage_remove_brick(),
+     * but rsp_dict is unavailable there. So copying it to rsp_dict from
+     * req_dict here. */
+
+    if (is_origin_glusterd(rsp_dict)) {
+        ret = dict_get_strn(req_dict, GF_REMOVE_BRICK_TID_KEY,
+                            SLEN(GF_REMOVE_BRICK_TID_KEY), &task_id_str);
+        if (ret) {
+            snprintf(msg, sizeof(msg), "Missing rebalance id for remove-brick");
+            gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_REBALANCE_ID_MISSING,
+                   "%s", msg);
+            ret = 0;
+        } else {
+            gf_uuid_parse(task_id_str, volinfo->rebal.rebalance_id);
+
+            ret = glusterd_copy_uuid_to_dict(volinfo->rebal.rebalance_id,
+                                             rsp_dict, GF_REMOVE_BRICK_TID_KEY,
+                                             SLEN(GF_REMOVE_BRICK_TID_KEY));
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, 0,
+                       GD_MSG_REMOVE_BRICK_ID_SET_FAIL,
+                       "Failed to set remove-brick-id");
+                goto out;
+            }
+        }
+    }
+    if (!gf_uuid_is_null(volinfo->rebal.rebalance_id) &&
+        GD_OP_REMOVE_BRICK == volinfo->rebal.op) {
+        ret = glusterd_copy_uuid_to_dict(volinfo->rebal.rebalance_id, rsp_dict,
+                                         GF_REMOVE_BRICK_TID_KEY,
+                                         SLEN(GF_REMOVE_BRICK_TID_KEY));
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+                   "Failed to set task-id for volume %s", volname);
+            goto out;
+        }
+    }
+out:
+    return ret;
+}
+int
+glusterd_op_remove_brick(dict_t *dict, char **op_errstr)
+{
+    int ret = -1;
+    char *volname = NULL;
+    glusterd_volinfo_t *volinfo = NULL;
+    char *brick = NULL;
+    int32_t count = 0;
+    int32_t i = 1;
+    char key[64] = "";
+    int keylen;
+    int32_t flag = 0;
+    int need_rebalance = 0;
+    int force = 0;
+    gf1_op_commands cmd = 0;
+    int32_t replica_count = 0;
+    char *task_id_str = NULL;
+    xlator_t *this = NULL;
+    dict_t *bricks_dict = NULL;
+    char *brick_tmpstr = NULL;
+    int start_remove = 0;
+    uint32_t commit_hash = 0;
+    int defrag_cmd = 0;
+    glusterd_conf_t *conf = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+    conf = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, conf, out);
+
+    ret = dict_get_strn(dict, "volname", SLEN("volname"), &volname);
+
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_BRICK_ADD_FAIL,
+               "Unable to get volume name");
+        goto out;
+    }
+
+    ret = glusterd_volinfo_find(volname, &volinfo);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, GD_MSG_VOL_NOT_FOUND,
+               "Unable to allocate memory");
+        goto out;
+    }
+
+    ret = dict_get_int32n(dict, "command", SLEN("command"), &flag);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to get command");
+        goto out;
+    }
+    cmd = flag;
+
+    if (GF_OP_CMD_START == cmd)
+        start_remove = 1;
+
+    /* Set task-id, if available, in ctx dict for operations other than
+     * start
+     */
+
+    if (is_origin_glusterd(dict) && (!start_remove)) {
+        if (!gf_uuid_is_null(volinfo->rebal.rebalance_id)) {
+            ret = glusterd_copy_uuid_to_dict(volinfo->rebal.rebalance_id, dict,
+                                             GF_REMOVE_BRICK_TID_KEY,
+                                             SLEN(GF_REMOVE_BRICK_TID_KEY));
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, 0,
+                       GD_MSG_REMOVE_BRICK_ID_SET_FAIL,
+                       "Failed to set remove-brick-id");
+                goto out;
+            }
+        }
+    }
+
+    /* Clear task-id, rebal.op and stored bricks on commmitting/stopping
+     * remove-brick */
+    if ((!start_remove) && (cmd != GF_OP_CMD_STATUS)) {
+        gf_uuid_clear(volinfo->rebal.rebalance_id);
+        volinfo->rebal.op = GD_OP_NONE;
+        dict_unref(volinfo->rebal.dict);
+        volinfo->rebal.dict = NULL;
+    }
+
+    ret = -1;
+    switch (cmd) {
+        case GF_OP_CMD_NONE:
+            goto out;
+
+        case GF_OP_CMD_STATUS:
+            ret = 0;
+            goto out;
+
+        case GF_OP_CMD_STOP:
+        case GF_OP_CMD_START:
+            /* Reset defrag status to 'NOT STARTED' whenever a
+             * remove-brick/rebalance command is issued to remove
+             * stale information from previous run.
+             * Update defrag_cmd as well or it will only be done
+             * for nodes on which the brick to be removed exists.
+             */
+            /* coverity[MIXED_ENUMS] */
+            volinfo->rebal.defrag_cmd = cmd;
+            volinfo->rebal.defrag_status = GF_DEFRAG_STATUS_NOT_STARTED;
+            ret = dict_get_strn(dict, GF_REMOVE_BRICK_TID_KEY,
+                                SLEN(GF_REMOVE_BRICK_TID_KEY), &task_id_str);
+            if (ret) {
+                gf_msg_debug(this->name, errno, "Missing remove-brick-id");
+                ret = 0;
+            } else {
+                gf_uuid_parse(task_id_str, volinfo->rebal.rebalance_id);
+                volinfo->rebal.op = GD_OP_REMOVE_BRICK;
+            }
+            force = 0;
+            break;
+
+        case GF_OP_CMD_COMMIT:
+            force = 1;
+            break;
+
+        case GF_OP_CMD_COMMIT_FORCE:
+
+            if (volinfo->decommission_in_progress) {
+                if (volinfo->rebal.defrag) {
+                    LOCK(&volinfo->rebal.defrag->lock);
+                    /* Fake 'rebalance-complete' so the graph change
+                       happens right away */
+                    volinfo->rebal.defrag_status = GF_DEFRAG_STATUS_COMPLETE;
+
+                    UNLOCK(&volinfo->rebal.defrag->lock);
+                }
+                /* Graph change happens in rebalance _cbk function,
+                   no need to do anything here */
+                /* TODO: '_cbk' function is not doing anything for now */
+            }
+
+            ret = 0;
+            force = 1;
+            break;
+        case GF_OP_CMD_DETACH_START:
+        case GF_OP_CMD_DETACH_COMMIT_FORCE:
+        case GF_OP_CMD_DETACH_COMMIT:
+        case GF_OP_CMD_STOP_DETACH_TIER:
+            break;
+    }
+
+    ret = dict_get_int32n(dict, "count", SLEN("count"), &count);
+    if (ret) {
+        gf_msg("glusterd", GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to get count");
+        goto out;
+    }
+    /* Save the list of bricks for later usage only on starting a
+     * remove-brick. Right now this is required for displaying the task
+     * parameters with task status in volume status.
+     */
+
+    if (start_remove) {
+        bricks_dict = dict_new();
+        if (!bricks_dict) {
+            ret = -1;
+            goto out;
+        }
+        ret = dict_set_int32n(bricks_dict, "count", SLEN("count"), count);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                   "Failed to save remove-brick count");
+            goto out;
+        }
+    }
+
+    while (i <= count) {
+        keylen = snprintf(key, sizeof(key), "brick%d", i);
+        ret = dict_get_strn(dict, key, keylen, &brick);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+                   "Unable to get %s", key);
+            goto out;
+        }
+
+        if (start_remove) {
+            brick_tmpstr = gf_strdup(brick);
+            if (!brick_tmpstr) {
+                ret = -1;
+                gf_msg(this->name, GF_LOG_ERROR, ENOMEM, GD_MSG_NO_MEMORY,
+                       "Failed to duplicate brick name");
+                goto out;
+            }
+            ret = dict_set_dynstrn(bricks_dict, key, keylen, brick_tmpstr);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                       "Failed to add brick to dict");
+                goto out;
+            }
+            brick_tmpstr = NULL;
+        }
+
+        ret = glusterd_op_perform_remove_brick(volinfo, brick, force,
+                                               &need_rebalance);
+        if (ret)
+            goto out;
+        i++;
+    }
+
+    if (start_remove)
+        volinfo->rebal.dict = dict_ref(bricks_dict);
+
+    ret = dict_get_int32n(dict, "replica-count", SLEN("replica-count"),
+                          &replica_count);
+    if (!ret) {
+        gf_msg(this->name, GF_LOG_INFO, errno, GD_MSG_DICT_GET_FAILED,
+               "changing replica count %d to %d on volume %s",
+               volinfo->replica_count, replica_count, volinfo->volname);
+        volinfo->replica_count = replica_count;
+        /* A reduction in replica count implies an arbiter volume
+         * earlier is now no longer one. */
+        if (volinfo->arbiter_count)
+            volinfo->arbiter_count = 0;
+        volinfo->sub_count = replica_count;
+        volinfo->dist_leaf_count = glusterd_get_dist_leaf_count(volinfo);
+
+        /*
+         * volinfo->type and sub_count have already been set for
+         * volumes undergoing a detach operation, they should not
+         * be modified here.
+         */
+        if (replica_count == 1) {
+            if (volinfo->type == GF_CLUSTER_TYPE_REPLICATE) {
+                volinfo->type = GF_CLUSTER_TYPE_NONE;
+                /* backward compatibility */
+                volinfo->sub_count = 0;
+            }
+        }
+    }
+    volinfo->subvol_count = (volinfo->brick_count / volinfo->dist_leaf_count);
+
+    if (!glusterd_is_volume_replicate(volinfo) &&
+        conf->op_version >= GD_OP_VERSION_3_12_2) {
+        ret = dict_set_nstrn(volinfo->dict, "performance.client-io-threads",
+                             SLEN("performance.client-io-threads"), "on",
+                             SLEN("on"));
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+                   "Failed to set "
+                   "performance.client-io-threads to on");
+            goto out;
+        }
+    }
+
+    ret = glusterd_create_volfiles_and_notify_services(volinfo);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_VOLFILE_CREATE_FAIL,
+               "failed to create volfiles");
+        goto out;
+    }
+
+    ret = glusterd_store_volinfo(volinfo, GLUSTERD_VOLINFO_VER_AC_INCREMENT);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_VOLINFO_STORE_FAIL,
+               "failed to store volinfo");
+        goto out;
+    }
+
+    if (start_remove && volinfo->status == GLUSTERD_STATUS_STARTED) {
+        ret = glusterd_svcs_reconfigure(volinfo);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_NFS_RECONF_FAIL,
+                   "Unable to reconfigure NFS-Server");
+            goto out;
+        }
+    }
+
+    /* Need to reset the defrag/rebalance status accordingly */
+    switch (volinfo->rebal.defrag_status) {
+        case GF_DEFRAG_STATUS_FAILED:
+        case GF_DEFRAG_STATUS_COMPLETE:
+            volinfo->rebal.defrag_status = 0;
+        /* FALLTHROUGH */
+        default:
+            break;
+    }
+    if (!force && need_rebalance) {
+        if (dict_get_uint32(dict, "commit-hash", &commit_hash) == 0) {
+            volinfo->rebal.commit_hash = commit_hash;
+        }
+        /* perform the rebalance operations */
+        defrag_cmd = GF_DEFRAG_CMD_START_FORCE;
+        /*
+         * We need to set this *before* we issue commands to the
+         * bricks, or else we might end up setting it after the bricks
+         * have responded.  If we fail to send the request(s) we'll
+         * clear it ourselves because nobody else will.
+         */
+        volinfo->decommission_in_progress = 1;
+        char err_str[4096] = "";
+        ret = glusterd_handle_defrag_start(
+            volinfo, err_str, sizeof(err_str), defrag_cmd,
+            glusterd_remove_brick_migrate_cbk, GD_OP_REMOVE_BRICK);
+
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REBALANCE_START_FAIL,
+                   "failed to start the rebalance");
+            /* TBD: shouldn't we do more than print a message? */
+            volinfo->decommission_in_progress = 0;
+            if (op_errstr)
+                *op_errstr = gf_strdup(err_str);
+        }
+    } else {
+        if (GLUSTERD_STATUS_STARTED == volinfo->status)
+            ret = glusterd_svcs_manager(volinfo);
+    }
+out:
+    GF_FREE(brick_tmpstr);
+    if (bricks_dict)
+        dict_unref(bricks_dict);
+    gf_msg_debug(this->name, 0, "returning %d ", ret);
+    return ret;
+}
+
+int
+glusterd_op_stage_barrier(dict_t *dict, char **op_errstr)
+{
+    int ret = -1;
+    xlator_t *this = NULL;
+    char *volname = NULL;
+    glusterd_volinfo_t *vol = NULL;
+    char *barrier_op = NULL;
+
+    GF_ASSERT(dict);
+    this = THIS;
+    GF_ASSERT(this);
+
+    ret = dict_get_strn(dict, "volname", SLEN("volname"), &volname);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Volname not present in "
+               "dict");
+        goto out;
+    }
+
+    ret = glusterd_volinfo_find(volname, &vol);
+    if (ret) {
+        gf_asprintf(op_errstr, "Volume %s does not exist", volname);
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_VOL_NOT_FOUND, "%s",
+               *op_errstr);
+        goto out;
+    }
+
+    if (!glusterd_is_volume_started(vol)) {
+        gf_asprintf(op_errstr, "Volume %s is not started", volname);
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_get_strn(dict, "barrier", SLEN("barrier"), &barrier_op);
+    if (ret == -1) {
+        gf_asprintf(op_errstr,
+                    "Barrier op for volume %s not present "
+                    "in dict",
+                    volname);
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED, "%s",
+               *op_errstr);
+        goto out;
+    }
+    ret = 0;
+out:
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+int
+glusterd_op_barrier(dict_t *dict, char **op_errstr)
+{
+    int ret = -1;
+    xlator_t *this = NULL;
+    char *volname = NULL;
+    glusterd_volinfo_t *vol = NULL;
+    char *barrier_op = NULL;
+
+    GF_ASSERT(dict);
+    this = THIS;
+    GF_ASSERT(this);
+
+    ret = dict_get_strn(dict, "volname", SLEN("volname"), &volname);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Volname not present in "
+               "dict");
+        goto out;
+    }
+
+    ret = glusterd_volinfo_find(volname, &vol);
+    if (ret) {
+        gf_asprintf(op_errstr, "Volume %s does not exist", volname);
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_VOL_NOT_FOUND, "%s",
+               *op_errstr);
+        goto out;
+    }
+
+    ret = dict_get_strn(dict, "barrier", SLEN("barrier"), &barrier_op);
+    if (ret) {
+        gf_asprintf(op_errstr,
+                    "Barrier op for volume %s not present "
+                    "in dict",
+                    volname);
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED, "%s",
+               *op_errstr);
+        goto out;
+    }
+
+    ret = dict_set_dynstr_with_alloc(vol->dict, "features.barrier", barrier_op);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+               "Failed to set barrier op in"
+               " volume option dict");
+        goto out;
+    }
+
+    gd_update_volume_op_versions(vol);
+    ret = glusterd_create_volfiles(vol);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_VOLFILE_CREATE_FAIL,
+               "Failed to create volfiles");
+        goto out;
+    }
+    ret = glusterd_store_volinfo(vol, GLUSTERD_VOLINFO_VER_AC_INCREMENT);
+
+out:
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+int
+glusterd_handle_add_tier_brick(rpcsvc_request_t *req)
+{
+    return 0;
+}
+
+int
+glusterd_handle_attach_tier(rpcsvc_request_t *req)
+{
+    return 0;
+}
+
+int
+glusterd_handle_detach_tier(rpcsvc_request_t *req)
+{
+    return 0;
+}
diff --git a/xlators/mgmt/glusterd/src/glusterd-conn-helper.c b/xlators/mgmt/glusterd/src/glusterd-conn-helper.c
new file mode 100644
index 00000000000..a7f54ec24b7
--- /dev/null
+++ b/xlators/mgmt/glusterd/src/glusterd-conn-helper.c
@@ -0,0 +1,21 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include "glusterd-conn-mgmt.h"
+#include "glusterd-svc-mgmt.h"
+
+#define _LGPL_SOURCE
+#include <urcu/rculist.h>
+
+glusterd_svc_t *
+glusterd_conn_get_svc_object(glusterd_conn_t *conn)
+{
+    return cds_list_entry(conn, glusterd_svc_t, conn);
+}
diff --git a/xlators/mgmt/glusterd/src/glusterd-conn-helper.h b/xlators/mgmt/glusterd/src/glusterd-conn-helper.h
new file mode 100644
index 00000000000..6f500309175
--- /dev/null
+++ b/xlators/mgmt/glusterd/src/glusterd-conn-helper.h
@@ -0,0 +1,21 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef _GLUSTERD_CONN_HELPER_H_
+#define _GLUSTERD_CONN_HELPER_H_
+
+#include "rpc-clnt.h"
+
+#include "glusterd-conn-mgmt.h"
+
+glusterd_svc_t *
+glusterd_conn_get_svc_object(glusterd_conn_t *conn);
+
+#endif
diff --git a/xlators/mgmt/glusterd/src/glusterd-conn-mgmt.c b/xlators/mgmt/glusterd/src/glusterd-conn-mgmt.c
new file mode 100644
index 00000000000..5c01f0c70b6
--- /dev/null
+++ b/xlators/mgmt/glusterd/src/glusterd-conn-mgmt.c
@@ -0,0 +1,191 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/xlator.h>
+#include "rpc-clnt.h"
+#include "glusterd.h"
+#include "glusterd-conn-mgmt.h"
+#include "glusterd-conn-helper.h"
+#include "glusterd-utils.h"
+#include "glusterd-messages.h"
+
+int
+glusterd_conn_init(glusterd_conn_t *conn, char *sockpath, int frame_timeout,
+                   glusterd_conn_notify_t notify)
+{
+    int ret = -1;
+    dict_t *options = NULL;
+    struct rpc_clnt *rpc = NULL;
+    xlator_t *this = THIS;
+    glusterd_svc_t *svc = NULL;
+
+    if (!this) {
+        gf_smsg(THIS->name, GF_LOG_ERROR, errno, GD_MSG_XLATOR_NOT_DEFINED,
+                NULL);
+        goto out;
+    }
+
+    options = dict_new();
+    if (!options) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_CREATE_FAIL, NULL);
+        goto out;
+    }
+
+    svc = glusterd_conn_get_svc_object(conn);
+    if (!svc) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SVC_GET_FAIL,
+               "Failed to get the service");
+        goto out;
+    }
+
+    ret = rpc_transport_unix_options_build(options, sockpath, frame_timeout);
+    if (ret)
+        goto out;
+
+    ret = dict_set_int32n(options, "transport.socket.ignore-enoent",
+                          SLEN("transport.socket.ignore-enoent"), 1);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                "Key=transport.socket.ignore-enoent", NULL);
+        goto out;
+    }
+
+    /* @options is free'd by rpc_transport when destroyed */
+    rpc = rpc_clnt_new(options, this, (char *)svc->name, 16);
+    if (!rpc) {
+        ret = -1;
+        goto out;
+    }
+
+    ret = rpc_clnt_register_notify(rpc, glusterd_conn_common_notify, conn);
+    if (ret)
+        goto out;
+
+    ret = snprintf(conn->sockpath, sizeof(conn->sockpath), "%s", sockpath);
+    if (ret < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_COPY_FAIL, NULL);
+        goto out;
+    } else
+        ret = 0;
+
+    conn->frame_timeout = frame_timeout;
+    conn->rpc = rpc;
+    conn->notify = notify;
+out:
+    if (options)
+        dict_unref(options);
+    if (ret) {
+        if (rpc) {
+            rpc_clnt_unref(rpc);
+            rpc = NULL;
+        }
+    }
+    return ret;
+}
+
+int
+glusterd_conn_term(glusterd_conn_t *conn)
+{
+    rpc_clnt_unref(conn->rpc);
+    return 0;
+}
+
+int
+glusterd_conn_connect(glusterd_conn_t *conn)
+{
+    return rpc_clnt_start(conn->rpc);
+}
+
+int
+glusterd_conn_disconnect(glusterd_conn_t *conn)
+{
+    rpc_clnt_disable(conn->rpc);
+
+    return 0;
+}
+
+int
+__glusterd_conn_common_notify(struct rpc_clnt *rpc, void *mydata,
+                              rpc_clnt_event_t event, void *data)
+{
+    glusterd_conn_t *conn = mydata;
+
+    /* Silently ignoring this error, exactly like the current
+     * implementation */
+    if (!conn)
+        return 0;
+
+    return conn->notify(conn, event);
+}
+
+int
+glusterd_conn_common_notify(struct rpc_clnt *rpc, void *mydata,
+                            rpc_clnt_event_t event, void *data)
+{
+    return glusterd_big_locked_notify(rpc, mydata, event, data,
+                                      __glusterd_conn_common_notify);
+}
+
+int32_t
+glusterd_conn_build_socket_filepath(char *rundir, uuid_t uuid, char *socketpath,
+                                    int len)
+{
+    char sockfilepath[PATH_MAX] = {
+        0,
+    };
+
+    snprintf(sockfilepath, sizeof(sockfilepath), "%s/run-%s", rundir,
+             uuid_utoa(uuid));
+
+    glusterd_set_socket_filepath(sockfilepath, socketpath, len);
+    return 0;
+}
+
+int
+__glusterd_muxsvc_conn_common_notify(struct rpc_clnt *rpc, void *mydata,
+                                     rpc_clnt_event_t event, void *data)
+{
+    glusterd_conf_t *conf = THIS->private;
+    glusterd_svc_proc_t *mux_proc = mydata;
+    int ret = -1;
+
+    /* Silently ignoring this error, exactly like the current
+     * implementation */
+    if (!mux_proc)
+        return 0;
+
+    if (event == RPC_CLNT_DESTROY) {
+        /*RPC_CLNT_DESTROY will only called after mux_proc detached from the
+         * list. So it is safe to call without lock. Processing
+         * RPC_CLNT_DESTROY under a lock will lead to deadlock.
+         */
+        if (mux_proc->data) {
+            glusterd_volinfo_unref(mux_proc->data);
+            mux_proc->data = NULL;
+        }
+        GF_FREE(mux_proc);
+        ret = 0;
+    } else {
+        pthread_mutex_lock(&conf->attach_lock);
+        {
+            ret = mux_proc->notify(mux_proc, event);
+        }
+        pthread_mutex_unlock(&conf->attach_lock);
+    }
+    return ret;
+}
+
+int
+glusterd_muxsvc_conn_common_notify(struct rpc_clnt *rpc, void *mydata,
+                                   rpc_clnt_event_t event, void *data)
+{
+    return glusterd_big_locked_notify(rpc, mydata, event, data,
+                                      __glusterd_muxsvc_conn_common_notify);
+}
diff --git a/xlators/mgmt/glusterd/src/glusterd-conn-mgmt.h b/xlators/mgmt/glusterd/src/glusterd-conn-mgmt.h
new file mode 100644
index 00000000000..1b225621ab1
--- /dev/null
+++ b/xlators/mgmt/glusterd/src/glusterd-conn-mgmt.h
@@ -0,0 +1,53 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef _GLUSTERD_CONN_MGMT_H_
+#define _GLUSTERD_CONN_MGMT_H_
+
+#include "rpc-clnt.h"
+
+typedef struct glusterd_conn_ glusterd_conn_t;
+
+typedef int (*glusterd_conn_notify_t)(glusterd_conn_t *conn,
+                                      rpc_clnt_event_t event);
+
+struct glusterd_conn_ {
+    struct rpc_clnt *rpc;
+    /* Existing daemons tend to specialize their respective
+     * notify implementations, so ... */
+    glusterd_conn_notify_t notify;
+    int frame_timeout;
+    char sockpath[PATH_MAX];
+};
+
+int
+glusterd_conn_init(glusterd_conn_t *conn, char *sockpath, int frame_timeout,
+                   glusterd_conn_notify_t notify);
+
+int
+glusterd_conn_term(glusterd_conn_t *conn);
+
+int
+glusterd_conn_connect(glusterd_conn_t *conn);
+
+int
+glusterd_conn_disconnect(glusterd_conn_t *conn);
+
+int
+glusterd_conn_common_notify(struct rpc_clnt *rpc, void *mydata,
+                            rpc_clnt_event_t event, void *data);
+int
+glusterd_muxsvc_conn_common_notify(struct rpc_clnt *rpc, void *mydata,
+                                   rpc_clnt_event_t event, void *data);
+
+int32_t
+glusterd_conn_build_socket_filepath(char *rundir, uuid_t uuid, char *socketpath,
+                                    int len);
+#endif
diff --git a/xlators/mgmt/glusterd/src/glusterd-errno.h b/xlators/mgmt/glusterd/src/glusterd-errno.h
new file mode 100644
index 00000000000..c74070e0e8d
--- /dev/null
+++ b/xlators/mgmt/glusterd/src/glusterd-errno.h
@@ -0,0 +1,33 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#ifndef _GLUSTERD_ERRNO_H
+#define _GLUSTERD_ERRNO_H
+
+enum glusterd_op_errno {
+    EG_INTRNL = 30800,    /* Internal Error                    */
+    EG_OPNOTSUP = 30801,  /* Gluster Op Not Supported          */
+    EG_ANOTRANS = 30802,  /* Another Transaction in Progress   */
+    EG_BRCKDWN = 30803,   /* One or more brick is down         */
+    EG_NODEDWN = 30804,   /* One or more node is down          */
+    EG_HRDLMT = 30805,    /* Hard Limit is reached             */
+    EG_NOVOL = 30806,     /* Volume does not exist             */
+    EG_NOSNAP = 30807,    /* Snap does not exist               */
+    EG_RBALRUN = 30808,   /* Rebalance is running              */
+    EG_VOLRUN = 30809,    /* Volume is running                 */
+    EG_VOLSTP = 30810,    /* Volume is not running             */
+    EG_VOLEXST = 30811,   /* Volume exists                     */
+    EG_SNAPEXST = 30812,  /* Snapshot exists                   */
+    EG_ISSNAP = 30813,    /* Volume is a snap volume           */
+    EG_GEOREPRUN = 30814, /* Geo-Replication is running        */
+    EG_NOTTHINP = 30815,  /* Bricks are not thinly provisioned */
+    EG_NOGANESHA = 30816, /* Global ganesha is not enabled   */
+};
+
+#endif
diff --git a/xlators/mgmt/glusterd/src/glusterd-ganesha.c b/xlators/mgmt/glusterd/src/glusterd-ganesha.c
new file mode 100644
index 00000000000..f08bd6cebee
--- /dev/null
+++ b/xlators/mgmt/glusterd/src/glusterd-ganesha.c
@@ -0,0 +1,927 @@
+/*
+   Copyright (c) 2015 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/common-utils.h>
+#include "glusterd.h"
+#include "glusterd-op-sm.h"
+#include "glusterd-store.h"
+#include "glusterd-utils.h"
+#include "glusterd-volgen.h"
+#include "glusterd-messages.h"
+#include <glusterfs/syscall.h>
+
+#include <ctype.h>
+
+int
+start_ganesha(char **op_errstr);
+
+typedef struct service_command {
+    char *binary;
+    char *service;
+    int (*action)(struct service_command *, char *);
+} service_command;
+
+/* parsing_ganesha_ha_conf will allocate the returned string
+ * to be freed (GF_FREE) by the caller
+ * return NULL if error or not found */
+static char *
+parsing_ganesha_ha_conf(const char *key)
+{
+#define MAX_LINE 1024
+    char scratch[MAX_LINE * 2] = {
+        0,
+    };
+    char *value = NULL, *pointer = NULL, *end_pointer = NULL;
+    FILE *fp;
+
+    fp = fopen(GANESHA_HA_CONF, "r");
+    if (fp == NULL) {
+        gf_msg(THIS->name, GF_LOG_ERROR, errno, GD_MSG_FILE_OP_FAILED,
+               "couldn't open the file %s", GANESHA_HA_CONF);
+        goto end_ret;
+    }
+    while ((pointer = fgets(scratch, MAX_LINE, fp)) != NULL) {
+        /* Read config file until we get matching "^[[:space:]]*key" */
+        if (*pointer == '#') {
+            continue;
+        }
+        while (isblank(*pointer)) {
+            pointer++;
+        }
+        if (strncmp(pointer, key, strlen(key))) {
+            continue;
+        }
+        pointer += strlen(key);
+        /* key found : if we fail to parse, we'll return an error
+         * rather than trying next one
+         * - supposition : conf file is bash compatible : no space
+         *   around the '=' */
+        if (*pointer != '=') {
+            gf_msg(THIS->name, GF_LOG_ERROR, errno,
+                   GD_MSG_GET_CONFIG_INFO_FAILED, "Parsing %s failed at key %s",
+                   GANESHA_HA_CONF, key);
+            goto end_close;
+        }
+        pointer++; /* jump the '=' */
+
+        if (*pointer == '"' || *pointer == '\'') {
+            /* dont get the quote */
+            pointer++;
+        }
+        end_pointer = pointer;
+        /* stop at the next closing quote or  blank/newline */
+        do {
+            end_pointer++;
+        } while (!(*end_pointer == '\'' || *end_pointer == '"' ||
+                   isspace(*end_pointer) || *end_pointer == '\0'));
+        *end_pointer = '\0';
+
+        /* got it. copy it and return */
+        value = gf_strdup(pointer);
+        break;
+    }
+
+end_close:
+    fclose(fp);
+end_ret:
+    return value;
+}
+
+static int
+sc_systemctl_action(struct service_command *sc, char *command)
+{
+    runner_t runner = {
+        0,
+    };
+
+    runinit(&runner);
+    runner_add_args(&runner, sc->binary, command, sc->service, NULL);
+    return runner_run(&runner);
+}
+
+static int
+sc_service_action(struct service_command *sc, char *command)
+{
+    runner_t runner = {
+        0,
+    };
+
+    runinit(&runner);
+    runner_add_args(&runner, sc->binary, sc->service, command, NULL);
+    return runner_run(&runner);
+}
+
+static int
+manage_service(char *action)
+{
+    int i = 0;
+    int ret = 0;
+    struct service_command sc_list[] = {{.binary = "/bin/systemctl",
+                                         .service = "nfs-ganesha",
+                                         .action = sc_systemctl_action},
+                                        {.binary = "/sbin/invoke-rc.d",
+                                         .service = "nfs-ganesha",
+                                         .action = sc_service_action},
+                                        {.binary = "/sbin/service",
+                                         .service = "nfs-ganesha",
+                                         .action = sc_service_action},
+                                        {.binary = NULL}};
+
+    while (sc_list[i].binary != NULL) {
+        ret = sys_access(sc_list[i].binary, X_OK);
+        if (ret == 0) {
+            gf_msg_debug(THIS->name, 0, "%s found.", sc_list[i].binary);
+            return sc_list[i].action(&sc_list[i], action);
+        }
+        i++;
+    }
+    gf_msg(THIS->name, GF_LOG_ERROR, 0, GD_MSG_UNRECOGNIZED_SVC_MNGR,
+           "Could not %s NFS-Ganesha.Service manager for distro"
+           " not recognized.",
+           action);
+    return ret;
+}
+
+/*
+ * Check if the cluster is a ganesha cluster or not *
+ */
+gf_boolean_t
+glusterd_is_ganesha_cluster()
+{
+    int ret = -1;
+    glusterd_conf_t *priv = NULL;
+    xlator_t *this = NULL;
+    gf_boolean_t ret_bool = _gf_false;
+
+    this = THIS;
+    GF_VALIDATE_OR_GOTO("ganesha", this, out);
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, priv, out);
+
+    ret = dict_get_str_boolean(priv->opts, GLUSTERD_STORE_KEY_GANESHA_GLOBAL,
+                               _gf_false);
+    if (ret == _gf_true) {
+        ret_bool = _gf_true;
+        gf_msg_debug(this->name, 0, "nfs-ganesha is enabled for the cluster");
+    } else
+        gf_msg_debug(this->name, 0, "nfs-ganesha is disabled for the cluster");
+
+out:
+    return ret_bool;
+}
+
+/* Check if ganesha.enable is set to 'on', that checks if
+ * a  particular volume is exported via NFS-Ganesha */
+gf_boolean_t
+glusterd_check_ganesha_export(glusterd_volinfo_t *volinfo)
+{
+    char *value = NULL;
+    gf_boolean_t is_exported = _gf_false;
+    int ret = 0;
+
+    ret = glusterd_volinfo_get(volinfo, "ganesha.enable", &value);
+    if ((ret == 0) && value) {
+        if (strcmp(value, "on") == 0) {
+            gf_msg_debug(THIS->name, 0,
+                         "ganesha.enable set"
+                         " to %s",
+                         value);
+            is_exported = _gf_true;
+        }
+    }
+    return is_exported;
+}
+
+/* *
+ * The below function is called as part of commit phase for volume set option
+ * "ganesha.enable". If the value is "on", it creates export configuration file
+ * and then export the volume via dbus command. Incase of "off", the volume
+ * will be already unexported during stage phase, so it will remove the conf
+ * file from shared storage
+ */
+int
+glusterd_check_ganesha_cmd(char *key, char *value, char **errstr, dict_t *dict)
+{
+    int ret = 0;
+    char *volname = NULL;
+
+    GF_ASSERT(key);
+    GF_ASSERT(value);
+    GF_ASSERT(dict);
+
+    if ((strcmp(key, "ganesha.enable") == 0)) {
+        if ((strcmp(value, "on")) && (strcmp(value, "off"))) {
+            gf_asprintf(errstr,
+                        "Invalid value"
+                        " for volume set command. Use on/off only.");
+            ret = -1;
+            goto out;
+        }
+        if (strcmp(value, "on") == 0) {
+            ret = glusterd_handle_ganesha_op(dict, errstr, key, value);
+
+        } else if (is_origin_glusterd(dict)) {
+            ret = dict_get_str(dict, "volname", &volname);
+            if (ret) {
+                gf_msg("glusterd-ganesha", GF_LOG_ERROR, errno,
+                       GD_MSG_DICT_GET_FAILED, "Unable to get volume name");
+                goto out;
+            }
+            ret = manage_export_config(volname, "off", errstr);
+        }
+    }
+out:
+    if (ret) {
+        gf_msg("glusterd-ganesha", GF_LOG_ERROR, 0,
+               GD_MSG_NFS_GNS_OP_HANDLE_FAIL,
+               "Handling NFS-Ganesha"
+               " op failed.");
+    }
+    return ret;
+}
+
+int
+glusterd_op_stage_set_ganesha(dict_t *dict, char **op_errstr)
+{
+    int ret = -1;
+    char *value = NULL;
+    char *str = NULL;
+    glusterd_conf_t *priv = NULL;
+    xlator_t *this = NULL;
+
+    GF_ASSERT(dict);
+    this = THIS;
+    GF_ASSERT(this);
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    ret = dict_get_str(dict, "value", &value);
+    if (value == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "value not present.");
+        goto out;
+    }
+    /* This dict_get will fail if the user had never set the key before */
+    /*Ignoring the ret value and proceeding */
+    ret = dict_get_str(priv->opts, GLUSTERD_STORE_KEY_GANESHA_GLOBAL, &str);
+    if (str ? strcmp(value, str) == 0 : strcmp(value, "disable") == 0) {
+        gf_asprintf(op_errstr, "nfs-ganesha is already %sd.", value);
+        ret = -1;
+        goto out;
+    }
+
+    if (strcmp(value, "enable") == 0) {
+        ret = start_ganesha(op_errstr);
+        if (ret) {
+            gf_msg(THIS->name, GF_LOG_ERROR, 0, GD_MSG_NFS_GNS_START_FAIL,
+                   "Could not start NFS-Ganesha");
+        }
+    } else {
+        ret = stop_ganesha(op_errstr);
+        if (ret)
+            gf_msg_debug(THIS->name, 0,
+                         "Could not stop "
+                         "NFS-Ganesha.");
+    }
+
+out:
+
+    if (ret) {
+        if (!(*op_errstr)) {
+            *op_errstr = gf_strdup("Error, Validation Failed");
+            gf_msg_debug(this->name, 0, "Error, Cannot Validate option :%s",
+                         GLUSTERD_STORE_KEY_GANESHA_GLOBAL);
+        } else {
+            gf_msg_debug(this->name, 0, "Error, Cannot Validate option");
+        }
+    }
+    return ret;
+}
+
+int
+glusterd_op_set_ganesha(dict_t *dict, char **errstr)
+{
+    int ret = 0;
+    xlator_t *this = NULL;
+    glusterd_conf_t *priv = NULL;
+    char *key = NULL;
+    char *value = NULL;
+    char *next_version = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+    GF_ASSERT(dict);
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    ret = dict_get_str(dict, "key", &key);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Couldn't get key in global option set");
+        goto out;
+    }
+
+    ret = dict_get_str(dict, "value", &value);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Couldn't get value in global option set");
+        goto out;
+    }
+
+    ret = glusterd_handle_ganesha_op(dict, errstr, key, value);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_NFS_GNS_SETUP_FAIL,
+               "Initial NFS-Ganesha set up failed");
+        ret = -1;
+        goto out;
+    }
+    ret = dict_set_dynstr_with_alloc(priv->opts,
+                                     GLUSTERD_STORE_KEY_GANESHA_GLOBAL, value);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, errno, GD_MSG_DICT_SET_FAILED,
+               "Failed to set"
+               " nfs-ganesha in dict.");
+        goto out;
+    }
+    ret = glusterd_get_next_global_opt_version_str(priv->opts, &next_version);
+    if (ret) {
+        gf_msg_debug(THIS->name, 0,
+                     "Could not fetch "
+                     " global op version");
+        goto out;
+    }
+    ret = dict_set_str(priv->opts, GLUSTERD_GLOBAL_OPT_VERSION, next_version);
+    if (ret)
+        goto out;
+
+    ret = glusterd_store_options(this, priv->opts);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_STORE_FAIL,
+               "Failed to store options");
+        goto out;
+    }
+
+out:
+    gf_msg_debug(this->name, 0, "returning %d", ret);
+    return ret;
+}
+
+/* Following function parse GANESHA_HA_CONF
+ * The sample file looks like below,
+ * HA_NAME="ganesha-ha-360"
+ * HA_VOL_NAME="ha-state"
+ * HA_CLUSTER_NODES="server1,server2"
+ * VIP_rhs_1="10.x.x.x"
+ * VIP_rhs_2="10.x.x.x." */
+
+/* Check if the localhost is listed as one of nfs-ganesha nodes */
+gf_boolean_t
+check_host_list(void)
+{
+    glusterd_conf_t *priv = NULL;
+    char *hostname, *hostlist;
+    gf_boolean_t ret = _gf_false;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    priv = THIS->private;
+    GF_ASSERT(priv);
+
+    hostlist = parsing_ganesha_ha_conf("HA_CLUSTER_NODES");
+    if (hostlist == NULL) {
+        gf_msg(this->name, GF_LOG_INFO, errno, GD_MSG_GET_CONFIG_INFO_FAILED,
+               "couldn't get HA_CLUSTER_NODES from file %s", GANESHA_HA_CONF);
+        return _gf_false;
+    }
+
+    /* Hostlist is a comma separated list now */
+    hostname = strtok(hostlist, ",");
+    while (hostname != NULL) {
+        ret = gf_is_local_addr(hostname);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_INFO, 0, GD_MSG_NFS_GNS_HOST_FOUND,
+                   "ganesha host found "
+                   "Hostname is %s",
+                   hostname);
+            break;
+        }
+        hostname = strtok(NULL, ",");
+    }
+
+    GF_FREE(hostlist);
+    return ret;
+}
+
+int
+gd_ganesha_send_dbus(char *volname, char *value)
+{
+    runner_t runner = {
+        0,
+    };
+    int ret = -1;
+    runinit(&runner);
+
+    GF_VALIDATE_OR_GOTO("glusterd-ganesha", volname, out);
+    GF_VALIDATE_OR_GOTO("glusterd-ganesha", value, out);
+
+    ret = 0;
+    if (check_host_list()) {
+        /* Check whether ganesha is running on this node */
+        if (manage_service("status")) {
+            gf_msg("glusterd-ganesha", GF_LOG_WARNING, 0,
+                   GD_MSG_GANESHA_NOT_RUNNING,
+                   "Export failed, NFS-Ganesha is not running");
+        } else {
+            runner_add_args(&runner, GANESHA_PREFIX "/dbus-send.sh", CONFDIR,
+                            value, volname, NULL);
+            ret = runner_run(&runner);
+        }
+    }
+out:
+    return ret;
+}
+
+int
+manage_export_config(char *volname, char *value, char **op_errstr)
+{
+    runner_t runner = {
+        0,
+    };
+    int ret = -1;
+
+    GF_ASSERT(volname);
+    runinit(&runner);
+    runner_add_args(&runner, GANESHA_PREFIX "/create-export-ganesha.sh",
+                    CONFDIR, value, volname, NULL);
+    ret = runner_run(&runner);
+
+    if (ret && op_errstr)
+        gf_asprintf(op_errstr,
+                    "Failed to create"
+                    " NFS-Ganesha export config file.");
+
+    return ret;
+}
+
+/* Exports and unexports a particular volume via NFS-Ganesha */
+int
+ganesha_manage_export(dict_t *dict, char *value,
+                      gf_boolean_t update_cache_invalidation, char **op_errstr)
+{
+    int ret = -1;
+    glusterd_volinfo_t *volinfo = NULL;
+    dict_t *vol_opts = NULL;
+    char *volname = NULL;
+    xlator_t *this = NULL;
+    glusterd_conf_t *priv = NULL;
+    gf_boolean_t option = _gf_false;
+
+    this = THIS;
+    GF_ASSERT(this);
+    priv = this->private;
+
+    GF_ASSERT(value);
+    GF_ASSERT(dict);
+    GF_ASSERT(priv);
+
+    ret = dict_get_str(dict, "volname", &volname);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_GET_FAILED,
+               "Unable to get volume name");
+        goto out;
+    }
+    ret = gf_string2boolean(value, &option);
+    if (ret == -1) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, GD_MSG_INVALID_ENTRY,
+               "invalid value.");
+        goto out;
+    }
+
+    ret = glusterd_volinfo_find(volname, &volinfo);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, GD_MSG_VOL_NOT_FOUND,
+               FMTSTR_CHECK_VOL_EXISTS, volname);
+        goto out;
+    }
+
+    ret = glusterd_check_ganesha_export(volinfo);
+    if (ret && option) {
+        gf_asprintf(op_errstr,
+                    "ganesha.enable "
+                    "is already 'on'.");
+        ret = -1;
+        goto out;
+
+    } else if (!option && !ret) {
+        gf_asprintf(op_errstr,
+                    "ganesha.enable "
+                    "is already 'off'.");
+        ret = -1;
+        goto out;
+    }
+
+    /* Check if global option is enabled, proceed only then */
+    ret = dict_get_str_boolean(priv->opts, GLUSTERD_STORE_KEY_GANESHA_GLOBAL,
+                               _gf_false);
+    if (ret == -1) {
+        gf_msg_debug(this->name, 0,
+                     "Failed to get "
+                     "global option dict.");
+        gf_asprintf(op_errstr,
+                    "The option "
+                    "nfs-ganesha should be "
+                    "enabled before setting ganesha.enable.");
+        goto out;
+    }
+    if (!ret) {
+        gf_asprintf(op_errstr,
+                    "The option "
+                    "nfs-ganesha should be "
+                    "enabled before setting ganesha.enable.");
+        ret = -1;
+        goto out;
+    }
+
+    /* *
+     * Create the export file from the node where ganesha.enable "on"
+     * is executed
+     * */
+    if (option && is_origin_glusterd(dict)) {
+        ret = manage_export_config(volname, "on", op_errstr);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_EXPORT_FILE_CREATE_FAIL,
+                   "Failed to create"
+                   "export file for NFS-Ganesha\n");
+            goto out;
+        }
+    }
+    ret = gd_ganesha_send_dbus(volname, value);
+    if (ret) {
+        gf_asprintf(op_errstr,
+                    "Dynamic export addition/deletion failed."
+                    " Please see log file for details");
+        goto out;
+    }
+    if (update_cache_invalidation) {
+        vol_opts = volinfo->dict;
+        ret = dict_set_dynstr_with_alloc(vol_opts,
+                                         "features.cache-invalidation", value);
+        if (ret)
+            gf_asprintf(op_errstr,
+                        "Cache-invalidation could not"
+                        " be set to %s.",
+                        value);
+        ret = glusterd_store_volinfo(volinfo,
+                                     GLUSTERD_VOLINFO_VER_AC_INCREMENT);
+        if (ret)
+            gf_asprintf(op_errstr, "failed to store volinfo for %s",
+                        volinfo->volname);
+    }
+out:
+    return ret;
+}
+
+int
+tear_down_cluster(gf_boolean_t run_teardown)
+{
+    int ret = 0;
+    runner_t runner = {
+        0,
+    };
+    struct stat st = {
+        0,
+    };
+    DIR *dir = NULL;
+    struct dirent *entry = NULL;
+    struct dirent scratch[2] = {
+        {
+            0,
+        },
+    };
+    char path[PATH_MAX] = {
+        0,
+    };
+
+    if (run_teardown) {
+        runinit(&runner);
+        runner_add_args(&runner, GANESHA_PREFIX "/ganesha-ha.sh", "teardown",
+                        CONFDIR, NULL);
+        ret = runner_run(&runner);
+        /* *
+         * Remove all the entries in CONFDIR expect ganesha.conf and
+         * ganesha-ha.conf
+         */
+        dir = sys_opendir(CONFDIR);
+        if (!dir) {
+            gf_msg_debug(THIS->name, 0,
+                         "Failed to open directory %s. "
+                         "Reason : %s",
+                         CONFDIR, strerror(errno));
+            ret = 0;
+            goto out;
+        }
+
+        while ((entry = sys_readdir(dir, scratch))) {
+            if (gf_irrelevant_entry(entry))
+                continue;
+            snprintf(path, PATH_MAX, "%s/%s", CONFDIR, entry->d_name);
+            ret = sys_lstat(path, &st);
+            if (ret == -1) {
+                gf_msg_debug(THIS->name, 0,
+                             "Failed to stat entry %s :"
+                             " %s",
+                             path, strerror(errno));
+                goto out;
+            }
+
+            if (strcmp(entry->d_name, "ganesha.conf") == 0 ||
+                strcmp(entry->d_name, "ganesha-ha.conf") == 0)
+                gf_msg_debug(THIS->name, 0,
+                             " %s is not required"
+                             " to remove",
+                             path);
+            else if (S_ISDIR(st.st_mode))
+                ret = recursive_rmdir(path);
+            else
+                ret = sys_unlink(path);
+
+            if (ret) {
+                gf_msg_debug(THIS->name, 0,
+                             " Failed to remove %s. "
+                             "Reason : %s",
+                             path, strerror(errno));
+            }
+
+            gf_msg_debug(THIS->name, 0, "%s %s",
+                         ret ? "Failed to remove" : "Removed", entry->d_name);
+        }
+
+        ret = sys_closedir(dir);
+        if (ret) {
+            gf_msg_debug(THIS->name, 0,
+                         "Failed to close dir %s. Reason :"
+                         " %s",
+                         CONFDIR, strerror(errno));
+        }
+        goto exit;
+    }
+
+out:
+    if (dir && sys_closedir(dir)) {
+        gf_msg_debug(THIS->name, 0,
+                     "Failed to close dir %s. Reason :"
+                     " %s",
+                     CONFDIR, strerror(errno));
+    }
+exit:
+    return ret;
+}
+
+int
+setup_cluster(gf_boolean_t run_setup)
+{
+    int ret = 0;
+    runner_t runner = {
+        0,
+    };
+
+    if (run_setup) {
+        runinit(&runner);
+        runner_add_args(&runner, GANESHA_PREFIX "/ganesha-ha.sh", "setup",
+                        CONFDIR, NULL);
+        ret = runner_run(&runner);
+    }
+    return ret;
+}
+
+static int
+teardown(gf_boolean_t run_teardown, char **op_errstr)
+{
+    runner_t runner = {
+        0,
+    };
+    int ret = 1;
+    glusterd_volinfo_t *volinfo = NULL;
+    glusterd_conf_t *priv = NULL;
+    dict_t *vol_opts = NULL;
+
+    priv = THIS->private;
+
+    ret = tear_down_cluster(run_teardown);
+    if (ret == -1) {
+        gf_asprintf(op_errstr,
+                    "Cleanup of NFS-Ganesha"
+                    " HA config failed.");
+        goto out;
+    }
+
+    runinit(&runner);
+    runner_add_args(&runner, GANESHA_PREFIX "/ganesha-ha.sh", "cleanup",
+                    CONFDIR, NULL);
+    ret = runner_run(&runner);
+    if (ret)
+        gf_msg_debug(THIS->name, 0,
+                     "Could not clean up"
+                     " NFS-Ganesha related config");
+
+    cds_list_for_each_entry(volinfo, &priv->volumes, vol_list)
+    {
+        vol_opts = volinfo->dict;
+        /* All the volumes exported via NFS-Ganesha will be
+        unexported, hence setting the appropriate keys */
+        ret = dict_set_str(vol_opts, "features.cache-invalidation", "off");
+        if (ret)
+            gf_msg(THIS->name, GF_LOG_WARNING, errno, GD_MSG_DICT_SET_FAILED,
+                   "Could not set features.cache-invalidation "
+                   "to off for %s",
+                   volinfo->volname);
+
+        ret = dict_set_str(vol_opts, "ganesha.enable", "off");
+        if (ret)
+            gf_msg(THIS->name, GF_LOG_WARNING, errno, GD_MSG_DICT_SET_FAILED,
+                   "Could not set ganesha.enable to off for %s",
+                   volinfo->volname);
+
+        ret = glusterd_store_volinfo(volinfo,
+                                     GLUSTERD_VOLINFO_VER_AC_INCREMENT);
+        if (ret)
+            gf_msg(THIS->name, GF_LOG_WARNING, 0, GD_MSG_VOLINFO_SET_FAIL,
+                   "failed to store volinfo for %s", volinfo->volname);
+    }
+out:
+    return ret;
+}
+
+int
+stop_ganesha(char **op_errstr)
+{
+    int ret = 0;
+    runner_t runner = {
+        0,
+    };
+
+    if (check_host_list()) {
+        runinit(&runner);
+        runner_add_args(&runner, GANESHA_PREFIX "/ganesha-ha.sh",
+                        "--setup-ganesha-conf-files", CONFDIR, "no", NULL);
+        ret = runner_run(&runner);
+        if (ret) {
+            gf_asprintf(op_errstr,
+                        "removal of symlink ganesha.conf "
+                        "in /etc/ganesha failed");
+        }
+        ret = manage_service("stop");
+        if (ret)
+            gf_asprintf(op_errstr,
+                        "NFS-Ganesha service could not"
+                        "be stopped.");
+    }
+    return ret;
+}
+
+int
+start_ganesha(char **op_errstr)
+{
+    int ret = -1;
+    glusterd_volinfo_t *volinfo = NULL;
+    glusterd_conf_t *priv = NULL;
+    runner_t runner = {
+        0,
+    };
+
+    priv = THIS->private;
+    GF_ASSERT(priv);
+
+    cds_list_for_each_entry(volinfo, &priv->volumes, vol_list)
+    {
+#ifdef BUILD_GNFS
+        /* Gluster-nfs has to be disabled across the trusted pool */
+        /* before attempting to start nfs-ganesha */
+        ret = dict_set_str_sizen(volinfo->dict, NFS_DISABLE_MAP_KEY, "on");
+        if (ret)
+            goto out;
+#endif
+        ret = glusterd_store_volinfo(volinfo,
+                                     GLUSTERD_VOLINFO_VER_AC_INCREMENT);
+        if (ret) {
+            *op_errstr = gf_strdup(
+                "Failed to store the "
+                "Volume information");
+            goto out;
+        }
+    }
+
+    /* If the nfs svc is not initialized it means that the service is not
+     * running, hence we can skip the process of stopping gluster-nfs
+     * service
+     */
+#ifdef BUILD_GNFS
+    if (priv->nfs_svc.inited) {
+        ret = priv->nfs_svc.stop(&(priv->nfs_svc), SIGKILL);
+        if (ret) {
+            ret = -1;
+            gf_asprintf(op_errstr,
+                        "Gluster-NFS service could"
+                        "not be stopped, exiting.");
+            goto out;
+        }
+    }
+#endif
+
+    if (check_host_list()) {
+        runinit(&runner);
+        runner_add_args(&runner, GANESHA_PREFIX "/ganesha-ha.sh",
+                        "--setup-ganesha-conf-files", CONFDIR, "yes", NULL);
+        ret = runner_run(&runner);
+        if (ret) {
+            gf_asprintf(op_errstr,
+                        "creation of symlink ganesha.conf "
+                        "in /etc/ganesha failed");
+            goto out;
+        }
+        ret = manage_service("start");
+        if (ret)
+            gf_asprintf(op_errstr,
+                        "NFS-Ganesha failed to start."
+                        "Please see log file for details");
+    }
+
+out:
+    return ret;
+}
+
+static int
+pre_setup(gf_boolean_t run_setup, char **op_errstr)
+{
+    int ret = 0;
+    if (run_setup) {
+        if (!check_host_list()) {
+            gf_asprintf(op_errstr,
+                        "Running nfs-ganesha setup command "
+                        "from node which is not part of ganesha cluster");
+            return -1;
+        }
+    }
+    ret = setup_cluster(run_setup);
+    if (ret == -1)
+        gf_asprintf(op_errstr,
+                    "Failed to set up HA "
+                    "config for NFS-Ganesha. "
+                    "Please check the log file for details");
+    return ret;
+}
+
+int
+glusterd_handle_ganesha_op(dict_t *dict, char **op_errstr, char *key,
+                           char *value)
+{
+    int32_t ret = -1;
+    gf_boolean_t option = _gf_false;
+
+    GF_ASSERT(dict);
+    GF_ASSERT(op_errstr);
+    GF_ASSERT(key);
+    GF_ASSERT(value);
+
+    if (strcmp(key, "ganesha.enable") == 0) {
+        ret = ganesha_manage_export(dict, value, _gf_true, op_errstr);
+        if (ret < 0)
+            goto out;
+    }
+
+    /* It is possible that the key might not be set */
+    ret = gf_string2boolean(value, &option);
+    if (ret == -1) {
+        gf_asprintf(op_errstr, "Invalid value in key-value pair.");
+        goto out;
+    }
+
+    if (strcmp(key, GLUSTERD_STORE_KEY_GANESHA_GLOBAL) == 0) {
+        /* *
+         * The set up/teardown of pcs cluster should be performed only
+         * once. This will done on the node in which the cli command
+         * 'gluster nfs-ganesha <enable/disable>' got executed. So that
+         * node should part of ganesha HA cluster
+         */
+        if (option) {
+            ret = pre_setup(is_origin_glusterd(dict), op_errstr);
+            if (ret < 0)
+                goto out;
+        } else {
+            ret = teardown(is_origin_glusterd(dict), op_errstr);
+            if (ret < 0)
+                goto out;
+        }
+    }
+
+out:
+    return ret;
+}
diff --git a/xlators/mgmt/glusterd/src/glusterd-geo-rep.c b/xlators/mgmt/glusterd/src/glusterd-geo-rep.c
new file mode 100644
index 00000000000..bf062c87060
--- /dev/null
+++ b/xlators/mgmt/glusterd/src/glusterd-geo-rep.c
@@ -0,0 +1,6782 @@
+/*
+   Copyright (c) 2011-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#include <glusterfs/common-utils.h>
+#include "cli1-xdr.h"
+#include "xdr-generic.h"
+#include "glusterd.h"
+#include "glusterd-op-sm.h"
+#include "glusterd-geo-rep.h"
+#include "glusterd-store.h"
+#include "glusterd-utils.h"
+#include "glusterd-volgen.h"
+#include "glusterd-svc-helper.h"
+#include <glusterfs/run.h>
+#include <glusterfs/syscall.h>
+#include "glusterd-messages.h"
+
+#include <signal.h>
+
+static int
+dict_get_param(dict_t *dict, char *key, char **param);
+
+struct gsync_config_opt_vals_ gsync_confopt_vals[] = {
+    {
+        .op_name = "change_detector",
+        .no_of_pos_vals = 2,
+        .case_sensitive = _gf_true,
+        .values = {"xsync", "changelog"},
+    },
+    {.op_name = "special_sync_mode",
+     .no_of_pos_vals = 2,
+     .case_sensitive = _gf_true,
+     .values = {"partial", "recover"}},
+    {.op_name = "log-level",
+     .no_of_pos_vals = 5,
+     .case_sensitive = _gf_false,
+     .values = {"critical", "error", "warning", "info", "debug"}},
+    {.op_name = "use-tarssh",
+     .no_of_pos_vals = 6,
+     .case_sensitive = _gf_false,
+     .values = {"true", "false", "0", "1", "yes", "no"}},
+    {.op_name = "ignore_deletes",
+     .no_of_pos_vals = 6,
+     .case_sensitive = _gf_false,
+     .values = {"true", "false", "0", "1", "yes", "no"}},
+    {.op_name = "use_meta_volume",
+     .no_of_pos_vals = 6,
+     .case_sensitive = _gf_false,
+     .values = {"true", "false", "0", "1", "yes", "no"}},
+    {.op_name = "use-meta-volume",
+     .no_of_pos_vals = 6,
+     .case_sensitive = _gf_false,
+     .values = {"true", "false", "0", "1", "yes", "no"}},
+    {
+        .op_name = NULL,
+    },
+};
+
+static char *gsync_reserved_opts[] = {"gluster-command",
+                                      "pid-file",
+                                      "state-file",
+                                      "session-owner",
+                                      "state-socket-unencoded",
+                                      "socketdir",
+                                      "local-id",
+                                      "local-path",
+                                      "slave-id",
+                                      NULL};
+
+static char *gsync_no_restart_opts[] = {"checkpoint", "log_rsync_performance",
+                                        "log-rsync-performance", NULL};
+
+void
+set_gsyncd_inet6_arg(runner_t *runner)
+{
+    xlator_t *this = NULL;
+    char *af;
+    int ret;
+
+    this = THIS;
+    ret = dict_get_str(this->options, "transport.address-family", &af);
+    if (ret == 0)
+        runner_argprintf(runner, "--%s", af);
+}
+
+int
+__glusterd_handle_sys_exec(rpcsvc_request_t *req)
+{
+    int32_t ret = 0;
+    dict_t *dict = NULL;
+    gf_cli_req cli_req = {
+        {0},
+    };
+    glusterd_op_t cli_op = GD_OP_SYS_EXEC;
+    glusterd_conf_t *priv = NULL;
+    char *host_uuid = NULL;
+    char err_str[64] = {
+        0,
+    };
+    xlator_t *this = NULL;
+
+    GF_ASSERT(req);
+
+    this = THIS;
+    GF_ASSERT(this);
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    ret = xdr_to_generic(req->msg[0], &cli_req, (xdrproc_t)xdr_gf_cli_req);
+    if (ret < 0) {
+        req->rpc_err = GARBAGE_ARGS;
+        snprintf(err_str, sizeof(err_str), "Garbage args received");
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_GARBAGE_ARGS, NULL);
+        goto out;
+    }
+
+    if (cli_req.dict.dict_len) {
+        dict = dict_new();
+        if (!dict) {
+            gf_smsg(THIS->name, GF_LOG_ERROR, errno, GD_MSG_DICT_CREATE_FAIL,
+                    NULL);
+            goto out;
+        }
+
+        ret = dict_unserialize(cli_req.dict.dict_val, cli_req.dict.dict_len,
+                               &dict);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_UNSERIALIZE_FAIL,
+                   "failed to "
+                   "unserialize req-buffer to dictionary");
+            snprintf(err_str, sizeof(err_str),
+                     "Unable to decode "
+                     "the command");
+            goto out;
+        } else {
+            dict->extra_stdfree = cli_req.dict.dict_val;
+        }
+
+        host_uuid = gf_strdup(uuid_utoa(MY_UUID));
+        if (host_uuid == NULL) {
+            snprintf(err_str, sizeof(err_str),
+                     "Failed to get "
+                     "the uuid of local glusterd");
+            gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_UUID_GET_FAIL,
+                    NULL);
+            ret = -1;
+            goto out;
+        }
+
+        ret = dict_set_dynstr(dict, "host-uuid", host_uuid);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                    "Key=host-uuid", NULL);
+            goto out;
+        }
+    }
+
+    ret = glusterd_op_begin_synctask(req, cli_op, dict);
+
+out:
+    if (ret) {
+        if (err_str[0] == '\0')
+            snprintf(err_str, sizeof(err_str), "Operation failed");
+        ret = glusterd_op_send_cli_response(cli_op, ret, 0, req, dict, err_str);
+    }
+    return ret;
+}
+
+int
+__glusterd_handle_copy_file(rpcsvc_request_t *req)
+{
+    int32_t ret = 0;
+    dict_t *dict = NULL;
+    gf_cli_req cli_req = {
+        {0},
+    };
+    glusterd_op_t cli_op = GD_OP_COPY_FILE;
+    glusterd_conf_t *priv = NULL;
+    char *host_uuid = NULL;
+    char err_str[64] = {
+        0,
+    };
+    xlator_t *this = NULL;
+
+    GF_ASSERT(req);
+
+    this = THIS;
+    GF_ASSERT(this);
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    ret = xdr_to_generic(req->msg[0], &cli_req, (xdrproc_t)xdr_gf_cli_req);
+    if (ret < 0) {
+        req->rpc_err = GARBAGE_ARGS;
+        snprintf(err_str, sizeof(err_str), "Garbage args received");
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_GARBAGE_ARGS, NULL);
+        goto out;
+    }
+
+    if (cli_req.dict.dict_len) {
+        dict = dict_new();
+        if (!dict) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_CREATE_FAIL,
+                    NULL);
+            goto out;
+        }
+
+        ret = dict_unserialize(cli_req.dict.dict_val, cli_req.dict.dict_len,
+                               &dict);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_UNSERIALIZE_FAIL,
+                   "failed to"
+                   "unserialize req-buffer to dictionary");
+            snprintf(err_str, sizeof(err_str),
+                     "Unable to decode "
+                     "the command");
+            goto out;
+        } else {
+            dict->extra_stdfree = cli_req.dict.dict_val;
+        }
+
+        host_uuid = gf_strdup(uuid_utoa(MY_UUID));
+        if (host_uuid == NULL) {
+            snprintf(err_str, sizeof(err_str),
+                     "Failed to get "
+                     "the uuid of local glusterd");
+            gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_UUID_GET_FAIL,
+                    NULL);
+            ret = -1;
+            goto out;
+        }
+
+        ret = dict_set_dynstr(dict, "host-uuid", host_uuid);
+        if (ret)
+            goto out;
+    }
+
+    ret = glusterd_op_begin_synctask(req, cli_op, dict);
+
+out:
+    if (ret) {
+        if (err_str[0] == '\0')
+            snprintf(err_str, sizeof(err_str), "Operation failed");
+        ret = glusterd_op_send_cli_response(cli_op, ret, 0, req, dict, err_str);
+    }
+    return ret;
+}
+
+int
+__glusterd_handle_gsync_set(rpcsvc_request_t *req)
+{
+    int32_t ret = 0;
+    dict_t *dict = NULL;
+    gf_cli_req cli_req = {
+        {0},
+    };
+    glusterd_op_t cli_op = GD_OP_GSYNC_SET;
+    char *master = NULL;
+    char *slave = NULL;
+    char operation[64] = {
+        0,
+    };
+    int type = 0;
+    glusterd_conf_t *priv = NULL;
+    char *host_uuid = NULL;
+    char err_str[64] = {
+        0,
+    };
+    xlator_t *this = NULL;
+
+    GF_ASSERT(req);
+
+    this = THIS;
+    GF_ASSERT(this);
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    ret = xdr_to_generic(req->msg[0], &cli_req, (xdrproc_t)xdr_gf_cli_req);
+    if (ret < 0) {
+        req->rpc_err = GARBAGE_ARGS;
+        snprintf(err_str, sizeof(err_str), "Garbage args received");
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_GARBAGE_ARGS, NULL);
+        goto out;
+    }
+
+    if (cli_req.dict.dict_len) {
+        dict = dict_new();
+        if (!dict) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_CREATE_FAIL,
+                    NULL);
+            goto out;
+        }
+
+        ret = dict_unserialize(cli_req.dict.dict_val, cli_req.dict.dict_len,
+                               &dict);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_UNSERIALIZE_FAIL,
+                   "failed to "
+                   "unserialize req-buffer to dictionary");
+            snprintf(err_str, sizeof(err_str),
+                     "Unable to decode "
+                     "the command");
+            goto out;
+        } else {
+            dict->extra_stdfree = cli_req.dict.dict_val;
+        }
+
+        host_uuid = gf_strdup(uuid_utoa(MY_UUID));
+        if (host_uuid == NULL) {
+            snprintf(err_str, sizeof(err_str),
+                     "Failed to get "
+                     "the uuid of local glusterd");
+            gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_UUID_GET_FAIL,
+                    NULL);
+            ret = -1;
+            goto out;
+        }
+        ret = dict_set_dynstr(dict, "host-uuid", host_uuid);
+        if (ret)
+            goto out;
+    }
+
+    ret = dict_get_str(dict, "master", &master);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_INFO, 0, GD_MSG_DICT_GET_FAILED,
+               "master not found, while handling " GEOREP " options");
+        master = "(No Master)";
+    }
+
+    ret = dict_get_str(dict, "slave", &slave);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_INFO, 0, GD_MSG_DICT_GET_FAILED,
+               "slave not found, while handling " GEOREP " options");
+        slave = "(No Slave)";
+    }
+
+    ret = dict_get_int32(dict, "type", &type);
+    if (ret < 0) {
+        snprintf(err_str, sizeof(err_str),
+                 "Command type not found "
+                 "while handling " GEOREP " options");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED, "%s",
+               err_str);
+        goto out;
+    }
+
+    switch (type) {
+        case GF_GSYNC_OPTION_TYPE_CREATE:
+            snprintf(operation, sizeof(operation), "create");
+            cli_op = GD_OP_GSYNC_CREATE;
+            break;
+
+        case GF_GSYNC_OPTION_TYPE_START:
+            snprintf(operation, sizeof(operation), "start");
+            break;
+
+        case GF_GSYNC_OPTION_TYPE_STOP:
+            snprintf(operation, sizeof(operation), "stop");
+            break;
+
+        case GF_GSYNC_OPTION_TYPE_PAUSE:
+            snprintf(operation, sizeof(operation), "pause");
+            break;
+
+        case GF_GSYNC_OPTION_TYPE_RESUME:
+            snprintf(operation, sizeof(operation), "resume");
+            break;
+
+        case GF_GSYNC_OPTION_TYPE_CONFIG:
+            snprintf(operation, sizeof(operation), "config");
+            break;
+
+        case GF_GSYNC_OPTION_TYPE_STATUS:
+            snprintf(operation, sizeof(operation), "status");
+            break;
+    }
+
+    ret = glusterd_op_begin_synctask(req, cli_op, dict);
+
+out:
+    if (ret) {
+        if (err_str[0] == '\0')
+            snprintf(err_str, sizeof(err_str), "Operation failed");
+        ret = glusterd_op_send_cli_response(cli_op, ret, 0, req, dict, err_str);
+    }
+    return ret;
+}
+
+int
+glusterd_handle_sys_exec(rpcsvc_request_t *req)
+{
+    return glusterd_big_locked_handler(req, __glusterd_handle_sys_exec);
+}
+
+int
+glusterd_handle_copy_file(rpcsvc_request_t *req)
+{
+    return glusterd_big_locked_handler(req, __glusterd_handle_copy_file);
+}
+
+int
+glusterd_handle_gsync_set(rpcsvc_request_t *req)
+{
+    return glusterd_big_locked_handler(req, __glusterd_handle_gsync_set);
+}
+
+/*****
+ *
+ * glusterd_urltransform* internal API
+ *
+ *****/
+
+static void
+glusterd_urltransform_init(runner_t *runner, const char *transname)
+{
+    runinit(runner);
+    runner_add_arg(runner, GSYNCD_PREFIX "/gsyncd");
+    set_gsyncd_inet6_arg(runner);
+    runner_argprintf(runner, "--%s-url", transname);
+}
+
+static void
+glusterd_urltransform_add(runner_t *runner, const char *url)
+{
+    runner_add_arg(runner, url);
+}
+
+/* Helper routine to terminate just before slave_voluuid */
+static int32_t
+parse_slave_url(char *slv_url, char **slave)
+{
+    char *tmp = NULL;
+    xlator_t *this = NULL;
+    int32_t ret = -1;
+
+    this = THIS;
+
+    /* slave format:
+     * master_node_uuid:ssh://slave_host::slave_vol:slave_voluuid */
+    *slave = strchr(slv_url, ':');
+    if (!(*slave)) {
+        goto out;
+    }
+    (*slave)++;
+
+    /* To terminate at : before slave volume uuid */
+    tmp = strstr(*slave, "::");
+    if (!tmp) {
+        goto out;
+    }
+    tmp += 2;
+    tmp = strchr(tmp, ':');
+    if (!tmp)
+        gf_msg_debug(this->name, 0, "old slave: %s!", *slave);
+    else
+        *tmp = '\0';
+
+    ret = 0;
+    gf_msg_debug(this->name, 0, "parsed slave: %s!", *slave);
+out:
+    return ret;
+}
+
+static int
+_glusterd_urltransform_add_iter(dict_t *dict, char *key, data_t *value,
+                                void *data)
+{
+    runner_t *runner = (runner_t *)data;
+    char slv_url[VOLINFO_SLAVE_URL_MAX] = {0};
+    char *slave = NULL;
+    xlator_t *this = NULL;
+    int32_t ret = -1;
+
+    this = THIS;
+    GF_VALIDATE_OR_GOTO("glusterd", this, out);
+
+    gf_msg_debug(this->name, 0, "value->data %s", value->data);
+
+    if (snprintf(slv_url, sizeof(slv_url), "%s", value->data) >=
+        sizeof(slv_url)) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVE_VOL_PARSE_FAIL,
+               "Error in copying slave: %s!", value->data);
+        goto out;
+    }
+
+    ret = parse_slave_url(slv_url, &slave);
+    if (ret == -1) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVE_VOL_PARSE_FAIL,
+               "Error in parsing slave: %s!", value->data);
+        goto out;
+    }
+
+    runner_add_arg(runner, slave);
+    ret = 0;
+out:
+    return ret;
+}
+
+static void
+glusterd_urltransform_free(char **linearr, unsigned n)
+{
+    int i = 0;
+
+    for (; i < n; i++)
+        GF_FREE(linearr[i]);
+
+    GF_FREE(linearr);
+}
+
+static int
+glusterd_urltransform(runner_t *runner, char ***linearrp)
+{
+    char **linearr = NULL;
+    char *line = NULL;
+    unsigned arr_len = 32;
+    unsigned arr_idx = 0;
+    gf_boolean_t error = _gf_false;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    linearr = GF_CALLOC(arr_len, sizeof(char *), gf_gld_mt_linearr);
+    if (!linearr) {
+        error = _gf_true;
+        goto out;
+    }
+
+    runner_redir(runner, STDOUT_FILENO, RUN_PIPE);
+    if (runner_start(runner) != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SPAWNING_CHILD_FAILED,
+               "spawning child failed");
+
+        error = _gf_true;
+        goto out;
+    }
+
+    arr_idx = 0;
+    for (;;) {
+        size_t len;
+        line = GF_MALLOC(1024, gf_gld_mt_linebuf);
+        if (!line) {
+            error = _gf_true;
+            goto out;
+        }
+
+        if (fgets(line, 1024, runner_chio(runner, STDOUT_FILENO)) == NULL) {
+            GF_FREE(line);
+            break;
+        }
+
+        len = strlen(line);
+        if (len == 0 || line[len - 1] != '\n') {
+            GF_FREE(line);
+            error = _gf_true;
+            goto out;
+        }
+        line[len - 1] = '\0';
+
+        if (arr_idx == arr_len) {
+            void *p = linearr;
+            arr_len <<= 1;
+            p = GF_REALLOC(linearr, arr_len);
+            if (!p) {
+                GF_FREE(line);
+                error = _gf_true;
+                goto out;
+            }
+            linearr = p;
+        }
+        linearr[arr_idx] = line;
+
+        arr_idx++;
+    }
+
+out:
+
+    /* XXX chpid field is not exported by run API
+     * but runner_end() does not abort the invoked
+     * process (ie. it might block in waitpid(2))
+     * so we resort to a manual kill a the private field
+     */
+    if (error && runner->chpid > 0)
+        kill(runner->chpid, SIGKILL);
+
+    if (runner_end(runner) != 0)
+        error = _gf_true;
+
+    if (error) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_READ_CHILD_DATA_FAILED,
+               "reading data from child failed");
+        glusterd_urltransform_free(linearr, arr_idx);
+        return -1;
+    }
+
+    *linearrp = linearr;
+    return arr_idx;
+}
+
+static int
+glusterd_urltransform_single(const char *url, const char *transname,
+                             char ***linearrp)
+{
+    runner_t runner = {
+        0,
+    };
+
+    glusterd_urltransform_init(&runner, transname);
+    glusterd_urltransform_add(&runner, url);
+    return glusterd_urltransform(&runner, linearrp);
+}
+
+struct dictidxmark {
+    unsigned isrch;
+    unsigned ithis;
+    char *ikey;
+};
+
+struct slave_vol_config {
+    char old_slvhost[_POSIX_HOST_NAME_MAX + 1];
+    char old_slvuser[LOGIN_NAME_MAX];
+    unsigned old_slvidx;
+    char slave_voluuid[UUID_CANONICAL_FORM_LEN + 1];
+};
+
+static int
+_dict_mark_atindex(dict_t *dict, char *key, data_t *value, void *data)
+{
+    struct dictidxmark *dim = data;
+
+    if (dim->isrch == dim->ithis)
+        dim->ikey = key;
+
+    dim->ithis++;
+    return 0;
+}
+
+static char *
+dict_get_by_index(dict_t *dict, unsigned i)
+{
+    struct dictidxmark dim = {
+        0,
+    };
+
+    dim.isrch = i;
+    dict_foreach(dict, _dict_mark_atindex, &dim);
+
+    return dim.ikey;
+}
+
+static int
+glusterd_get_slave(glusterd_volinfo_t *vol, const char *slaveurl,
+                   char **slavekey)
+{
+    runner_t runner = {
+        0,
+    };
+    int n = 0;
+    int i = 0;
+    char **linearr = NULL;
+    int32_t ret = 0;
+
+    glusterd_urltransform_init(&runner, "canonicalize");
+    ret = dict_foreach(vol->gsync_slaves, _glusterd_urltransform_add_iter,
+                       &runner);
+    if (ret < 0)
+        return -2;
+
+    glusterd_urltransform_add(&runner, slaveurl);
+
+    n = glusterd_urltransform(&runner, &linearr);
+    if (n == -1)
+        return -2;
+
+    for (i = 0; i < n - 1; i++) {
+        if (strcmp(linearr[i], linearr[n - 1]) == 0)
+            break;
+    }
+    glusterd_urltransform_free(linearr, n);
+
+    if (i < n - 1)
+        *slavekey = dict_get_by_index(vol->gsync_slaves, i);
+    else
+        i = -1;
+
+    return i;
+}
+
+static int
+glusterd_query_extutil_generic(char *resbuf, size_t blen, runner_t *runner,
+                               void *data,
+                               int (*fcbk)(char *resbuf, size_t blen, FILE *fp,
+                                           void *data))
+{
+    int ret = 0;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    runner_redir(runner, STDOUT_FILENO, RUN_PIPE);
+    if (runner_start(runner) != 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SPAWNING_CHILD_FAILED,
+               "spawning child failed");
+
+        return -1;
+    }
+
+    ret = fcbk(resbuf, blen, runner_chio(runner, STDOUT_FILENO), data);
+
+    ret |= runner_end(runner);
+    if (ret)
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_READ_CHILD_DATA_FAILED,
+               "reading data from child failed");
+
+    return ret ? -1 : 0;
+}
+
+static int
+_fcbk_singleline(char *resbuf, size_t blen, FILE *fp, void *data)
+{
+    char *ptr = NULL;
+
+    errno = 0;
+    ptr = fgets(resbuf, blen, fp);
+    if (ptr) {
+        size_t len = strlen(resbuf);
+        if (len && resbuf[len - 1] == '\n')
+            resbuf[len - 1] = '\0';  // strip off \n
+    }
+
+    return errno ? -1 : 0;
+}
+
+static int
+glusterd_query_extutil(char *resbuf, runner_t *runner)
+{
+    return glusterd_query_extutil_generic(resbuf, PATH_MAX, runner, NULL,
+                                          _fcbk_singleline);
+}
+
+static int
+glusterd_get_slave_voluuid(char *slave_host, char *slave_vol, char *vol_uuid)
+{
+    runner_t runner = {
+        0,
+    };
+    glusterd_conf_t *priv = NULL;
+    xlator_t *this = NULL;
+    int ret = -1;
+
+    this = THIS;
+    GF_VALIDATE_OR_GOTO("glusterd", this, out);
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, priv, out);
+
+    runinit(&runner);
+    runner_add_arg(&runner, GSYNCD_PREFIX "/gsyncd");
+    set_gsyncd_inet6_arg(&runner);
+    runner_add_arg(&runner, "--slavevoluuid-get");
+    runner_argprintf(&runner, "%s::%s", slave_host, slave_vol);
+
+    synclock_unlock(&priv->big_lock);
+    ret = glusterd_query_extutil(vol_uuid, &runner);
+    synclock_lock(&priv->big_lock);
+
+out:
+    return ret;
+}
+
+static int
+_fcbk_conftodict(char *resbuf, size_t blen, FILE *fp, void *data)
+{
+    char *ptr = NULL;
+    dict_t *dict = data;
+    char *v = NULL;
+
+    for (;;) {
+        errno = 0;
+        ptr = fgets(resbuf, blen - 2, fp);
+        if (!ptr)
+            break;
+        v = resbuf + strlen(resbuf) - 1;
+        while (isspace(*v))
+            /* strip trailing space */
+            *v-- = '\0';
+        if (v == resbuf)
+            /* skip empty line */
+            continue;
+        v = strchr(resbuf, ':');
+        if (!v)
+            return -1;
+        *v++ = '\0';
+        while (isspace(*v))
+            v++;
+        v = gf_strdup(v);
+        if (!v)
+            return -1;
+        if (dict_set_dynstr(dict, resbuf, v) != 0) {
+            GF_FREE(v);
+            return -1;
+        }
+    }
+
+    return errno ? -1 : 0;
+}
+
+static int
+glusterd_gsync_get_config(char *master, char *slave, char *conf_path,
+                          dict_t *dict)
+{
+    /* key + value, where value must be able to accommodate a path */
+    char resbuf[256 + PATH_MAX] = {
+        0,
+    };
+    runner_t runner = {
+        0,
+    };
+
+    runinit(&runner);
+    runner_add_args(&runner, GSYNCD_PREFIX "/gsyncd", "-c", NULL);
+    runner_argprintf(&runner, "%s", conf_path);
+    set_gsyncd_inet6_arg(&runner);
+    runner_argprintf(&runner, "--iprefix=%s", DATADIR);
+    runner_argprintf(&runner, ":%s", master);
+    runner_add_args(&runner, slave, "--config-get-all", NULL);
+
+    return glusterd_query_extutil_generic(resbuf, sizeof(resbuf), &runner, dict,
+                                          _fcbk_conftodict);
+}
+
+static int
+_fcbk_statustostruct(char *resbuf, size_t blen, FILE *fp, void *data)
+{
+    char *ptr = NULL;
+    char *v = NULL;
+    char *k = NULL;
+    gf_gsync_status_t *sts_val = NULL;
+    size_t len = 0;
+
+    sts_val = (gf_gsync_status_t *)data;
+
+    for (;;) {
+        errno = 0;
+        ptr = fgets(resbuf, blen - 2, fp);
+        if (!ptr)
+            break;
+
+        v = resbuf + strlen(resbuf) - 1;
+        while (isspace(*v))
+            /* strip trailing space */
+            *v-- = '\0';
+        if (v == resbuf)
+            /* skip empty line */
+            continue;
+        v = strchr(resbuf, ':');
+        if (!v)
+            return -1;
+        *v++ = '\0';
+        while (isspace(*v))
+            v++;
+        v = gf_strdup(v);
+        if (!v)
+            return -1;
+
+        k = gf_strdup(resbuf);
+        if (!k) {
+            GF_FREE(v);
+            return -1;
+        }
+
+        if (strcmp(k, "worker_status") == 0) {
+            len = min(strlen(v), (sizeof(sts_val->worker_status) - 1));
+            memcpy(sts_val->worker_status, v, len);
+            sts_val->worker_status[len] = '\0';
+        } else if (strcmp(k, "slave_node") == 0) {
+            len = min(strlen(v), (sizeof(sts_val->slave_node) - 1));
+            memcpy(sts_val->slave_node, v, len);
+            sts_val->slave_node[len] = '\0';
+        } else if (strcmp(k, "crawl_status") == 0) {
+            len = min(strlen(v), (sizeof(sts_val->crawl_status) - 1));
+            memcpy(sts_val->crawl_status, v, len);
+            sts_val->crawl_status[len] = '\0';
+        } else if (strcmp(k, "last_synced") == 0) {
+            len = min(strlen(v), (sizeof(sts_val->last_synced) - 1));
+            memcpy(sts_val->last_synced, v, len);
+            sts_val->last_synced[len] = '\0';
+        } else if (strcmp(k, "last_synced_utc") == 0) {
+            len = min(strlen(v), (sizeof(sts_val->last_synced_utc) - 1));
+            memcpy(sts_val->last_synced_utc, v, len);
+            sts_val->last_synced_utc[len] = '\0';
+        } else if (strcmp(k, "entry") == 0) {
+            len = min(strlen(v), (sizeof(sts_val->entry) - 1));
+            memcpy(sts_val->entry, v, len);
+            sts_val->entry[len] = '\0';
+        } else if (strcmp(k, "data") == 0) {
+            len = min(strlen(v), (sizeof(sts_val->data) - 1));
+            memcpy(sts_val->data, v, len);
+            sts_val->data[len] = '\0';
+        } else if (strcmp(k, "meta") == 0) {
+            len = min(strlen(v), (sizeof(sts_val->meta) - 1));
+            memcpy(sts_val->meta, v, len);
+            sts_val->meta[len] = '\0';
+        } else if (strcmp(k, "failures") == 0) {
+            len = min(strlen(v), (sizeof(sts_val->failures) - 1));
+            memcpy(sts_val->failures, v, len);
+            sts_val->failures[len] = '\0';
+        } else if (strcmp(k, "checkpoint_time") == 0) {
+            len = min(strlen(v), (sizeof(sts_val->checkpoint_time) - 1));
+            memcpy(sts_val->checkpoint_time, v, len);
+            sts_val->checkpoint_time[len] = '\0';
+        } else if (strcmp(k, "checkpoint_time_utc") == 0) {
+            len = min(strlen(v), (sizeof(sts_val->checkpoint_time_utc) - 1));
+            memcpy(sts_val->checkpoint_time_utc, v, len);
+            sts_val->checkpoint_time_utc[len] = '\0';
+        } else if (strcmp(k, "checkpoint_completed") == 0) {
+            len = min(strlen(v), (sizeof(sts_val->checkpoint_completed) - 1));
+            memcpy(sts_val->checkpoint_completed, v, len);
+            sts_val->checkpoint_completed[len] = '\0';
+        } else if (strcmp(k, "checkpoint_completion_time") == 0) {
+            len = min(strlen(v),
+                      (sizeof(sts_val->checkpoint_completion_time) - 1));
+            memcpy(sts_val->checkpoint_completion_time, v, len);
+            sts_val->checkpoint_completion_time[len] = '\0';
+        } else if (strcmp(k, "checkpoint_completion_time_utc") == 0) {
+            len = min(strlen(v),
+                      (sizeof(sts_val->checkpoint_completion_time_utc) - 1));
+            memcpy(sts_val->checkpoint_completion_time_utc, v, len);
+            sts_val->checkpoint_completion_time_utc[len] = '\0';
+        }
+        GF_FREE(v);
+        GF_FREE(k);
+    }
+
+    return errno ? -1 : 0;
+}
+
+static int
+glusterd_gsync_get_status(char *master, char *slave, char *conf_path,
+                          char *brick_path, gf_gsync_status_t *sts_val)
+{
+    /* key + value, where value must be able to accommodate a path */
+    char resbuf[256 + PATH_MAX] = {
+        0,
+    };
+    runner_t runner = {
+        0,
+    };
+
+    runinit(&runner);
+    runner_add_args(&runner, GSYNCD_PREFIX "/gsyncd", "-c", NULL);
+    runner_argprintf(&runner, "%s", conf_path);
+    set_gsyncd_inet6_arg(&runner);
+    runner_argprintf(&runner, "--iprefix=%s", DATADIR);
+    runner_argprintf(&runner, ":%s", master);
+    runner_add_args(&runner, slave, "--status-get", NULL);
+    runner_add_args(&runner, "--path", brick_path, NULL);
+
+    return glusterd_query_extutil_generic(resbuf, sizeof(resbuf), &runner,
+                                          sts_val, _fcbk_statustostruct);
+}
+
+static int
+glusterd_gsync_get_param_file(char *prmfile, const char *param, char *master,
+                              char *slave, char *conf_path)
+{
+    runner_t runner = {
+        0,
+    };
+
+    runinit(&runner);
+    runner_add_args(&runner, GSYNCD_PREFIX "/gsyncd", "-c", NULL);
+    runner_argprintf(&runner, "%s", conf_path);
+    set_gsyncd_inet6_arg(&runner);
+    runner_argprintf(&runner, "--iprefix=%s", DATADIR);
+    runner_argprintf(&runner, ":%s", master);
+    runner_add_args(&runner, slave, "--config-get", NULL);
+    runner_argprintf(&runner, "%s-file", param);
+
+    return glusterd_query_extutil(prmfile, &runner);
+}
+
+static int
+gsyncd_getpidfile(char *master, char *slave, char *pidfile, char *conf_path,
+                  gf_boolean_t *is_template_in_use)
+{
+    char temp_conf_path[PATH_MAX] = "";
+    char *working_conf_path = NULL;
+    glusterd_conf_t *priv = NULL;
+    int ret = -1;
+    struct stat stbuf = {
+        0,
+    };
+    xlator_t *this = NULL;
+    int32_t len = 0;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(this->private);
+    GF_ASSERT(conf_path);
+
+    priv = this->private;
+
+    GF_VALIDATE_OR_GOTO("gsync", master, out);
+    GF_VALIDATE_OR_GOTO("gsync", slave, out);
+
+    len = snprintf(temp_conf_path, sizeof(temp_conf_path),
+                   "%s/" GSYNC_CONF_TEMPLATE, priv->workdir);
+    if ((len < 0) || (len >= sizeof(temp_conf_path))) {
+        goto out;
+    }
+
+    ret = sys_lstat(conf_path, &stbuf);
+    if (!ret) {
+        gf_msg_debug(this->name, 0, "Using passed config template(%s).",
+                     conf_path);
+        working_conf_path = conf_path;
+    } else {
+        gf_msg(this->name, GF_LOG_WARNING, ENOENT, GD_MSG_FILE_OP_FAILED,
+               "Config file (%s) missing. Looking for template "
+               "config file (%s)",
+               conf_path, temp_conf_path);
+        ret = sys_lstat(temp_conf_path, &stbuf);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOENT, GD_MSG_FILE_OP_FAILED,
+                   "Template config file (%s) missing.", temp_conf_path);
+            goto out;
+        }
+        gf_msg(this->name, GF_LOG_INFO, 0, GD_MSG_DEFAULT_TEMP_CONFIG,
+               "Using default config template(%s).", temp_conf_path);
+        working_conf_path = temp_conf_path;
+        *is_template_in_use = _gf_true;
+    }
+
+fetch_data:
+
+    ret = glusterd_gsync_get_param_file(pidfile, "pid", master, slave,
+                                        working_conf_path);
+    if ((ret == -1) || strlen(pidfile) == 0) {
+        if (*is_template_in_use == _gf_false) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_PIDFILE_CREATE_FAILED,
+                   "failed to create the pidfile string. "
+                   "Trying default config template");
+            working_conf_path = temp_conf_path;
+            *is_template_in_use = _gf_true;
+            goto fetch_data;
+        } else {
+            ret = -2;
+            gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_PIDFILE_CREATE_FAILED,
+                   "failed to "
+                   "create the pidfile string from template "
+                   "config");
+            goto out;
+        }
+    }
+
+    gf_msg_debug(this->name, 0, "pidfile = %s", pidfile);
+
+    ret = open(pidfile, O_RDWR);
+out:
+    return ret;
+}
+
+static int
+gsync_status_byfd(int fd)
+{
+    GF_ASSERT(fd >= -1);
+
+    if (lockf(fd, F_TEST, 0) == -1 && (errno == EAGAIN || errno == EACCES))
+        /* gsyncd keeps the pidfile locked */
+        return 0;
+
+    return -1;
+}
+
+/* status: return 0 when gsync is running
+ * return -1 when not running
+ */
+int
+gsync_status(char *master, char *slave, char *conf_path, int *status,
+             gf_boolean_t *is_template_in_use)
+{
+    char pidfile[PATH_MAX] = {
+        0,
+    };
+    int fd = -1;
+
+    fd = gsyncd_getpidfile(master, slave, pidfile, conf_path,
+                           is_template_in_use);
+    if (fd == -2)
+        return -1;
+
+    *status = gsync_status_byfd(fd);
+
+    sys_close(fd);
+
+    return 0;
+}
+
+static int32_t
+glusterd_gsync_volinfo_dict_set(glusterd_volinfo_t *volinfo, char *key,
+                                char *value)
+{
+    int32_t ret = -1;
+    char *gsync_status = NULL;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    gsync_status = gf_strdup(value);
+    if (!gsync_status) {
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, GD_MSG_NO_MEMORY,
+               "Unable to allocate memory");
+        goto out;
+    }
+
+    ret = dict_set_dynstr(volinfo->dict, key, gsync_status);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+               "Unable to set dict");
+        goto out;
+    }
+
+    ret = 0;
+out:
+    return ret;
+}
+
+static int
+glusterd_verify_gsyncd_spawn(char *master, char *slave)
+{
+    int ret = 0;
+    runner_t runner = {
+        0,
+    };
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    runinit(&runner);
+    runner_add_args(&runner, GSYNCD_PREFIX "/gsyncd", "--verify", "spawning",
+                    NULL);
+    runner_argprintf(&runner, ":%s", master);
+    runner_add_args(&runner, slave, NULL);
+    runner_redir(&runner, STDOUT_FILENO, RUN_PIPE);
+    ret = runner_start(&runner);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SPAWNING_CHILD_FAILED,
+               "spawning child failed");
+        ret = -1;
+        goto out;
+    }
+
+    if (runner_end(&runner) != 0)
+        ret = -1;
+
+out:
+    gf_msg_debug(this->name, 0, "returning %d", ret);
+    return ret;
+}
+
+static int
+gsync_verify_config_options(dict_t *dict, char **op_errstr, char *volname)
+{
+    char **resopt = NULL;
+    int i = 0;
+    int ret = -1;
+    char *subop = NULL;
+    char *slave = NULL;
+    char *op_name = NULL;
+    char *op_value = NULL;
+    char *t = NULL;
+    char errmsg[PATH_MAX] = "";
+    gf_boolean_t banned = _gf_true;
+    gf_boolean_t op_match = _gf_true;
+    gf_boolean_t val_match = _gf_true;
+    struct gsync_config_opt_vals_ *conf_vals = NULL;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    if (dict_get_str(dict, "subop", &subop) != 0) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_DICT_GET_FAILED,
+               "missing subop");
+        *op_errstr = gf_strdup("Invalid config request");
+        return -1;
+    }
+
+    if (dict_get_str(dict, "slave", &slave) != 0) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_DICT_GET_FAILED,
+               GEOREP " CONFIG: no slave given");
+        *op_errstr = gf_strdup("Slave required");
+        return -1;
+    }
+
+    if (strcmp(subop, "get-all") == 0)
+        return 0;
+
+    if (dict_get_str(dict, "op_name", &op_name) != 0) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_DICT_GET_FAILED,
+               "option name missing");
+        *op_errstr = gf_strdup("Option name missing");
+        return -1;
+    }
+
+    if (runcmd(GSYNCD_PREFIX "/gsyncd", "--config-check", op_name, NULL)) {
+        ret = glusterd_verify_gsyncd_spawn(volname, slave);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GSYNCD_SPAWN_FAILED,
+                   "Unable to spawn "
+                   "gsyncd");
+            return 0;
+        }
+
+        gf_msg(this->name, GF_LOG_WARNING, EINVAL, GD_MSG_INVALID_ENTRY,
+               "Invalid option %s", op_name);
+        *op_errstr = gf_strdup("Invalid option");
+
+        return -1;
+    }
+
+    if (strcmp(subop, "get") == 0)
+        return 0;
+
+    t = strtail(subop, "set");
+    if (!t)
+        t = strtail(subop, "del");
+    if (!t || (t[0] && strcmp(t, "-glob") != 0)) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_SUBOP_NOT_FOUND,
+               "unknown subop %s", subop);
+        *op_errstr = gf_strdup("Invalid config request");
+        return -1;
+    }
+
+    if (strtail(subop, "set") &&
+        dict_get_str(dict, "op_value", &op_value) != 0) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_DICT_GET_FAILED,
+               "missing value for set");
+        *op_errstr = gf_strdup("missing value");
+    }
+
+    /* match option name against reserved options, modulo -/_
+     * difference
+     */
+    for (resopt = gsync_reserved_opts; *resopt; resopt++) {
+        banned = _gf_true;
+        for (i = 0; (*resopt)[i] && op_name[i]; i++) {
+            if ((*resopt)[i] == op_name[i] ||
+                ((*resopt)[i] == '-' && op_name[i] == '_'))
+                continue;
+            banned = _gf_false;
+        }
+
+        if (op_name[i] != '\0')
+            banned = _gf_false;
+
+        if (banned) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_RESERVED_OPTION,
+                   "Reserved option %s", op_name);
+            *op_errstr = gf_strdup("Reserved option");
+
+            return -1;
+            break;
+        }
+    }
+
+    /* Check options in gsync_confopt_vals for invalid values */
+    for (conf_vals = gsync_confopt_vals; conf_vals->op_name; conf_vals++) {
+        op_match = _gf_true;
+        for (i = 0; conf_vals->op_name[i] && op_name[i]; i++) {
+            if (conf_vals->op_name[i] == op_name[i] ||
+                (conf_vals->op_name[i] == '_' && op_name[i] == '-'))
+                continue;
+            op_match = _gf_false;
+        }
+
+        if (op_match) {
+            if (!op_value)
+                goto out;
+            val_match = _gf_false;
+            for (i = 0; i < conf_vals->no_of_pos_vals; i++) {
+                if (conf_vals->case_sensitive) {
+                    if (!strcmp(conf_vals->values[i], op_value))
+                        val_match = _gf_true;
+                } else {
+                    if (!strcasecmp(conf_vals->values[i], op_value))
+                        val_match = _gf_true;
+                }
+            }
+
+            if (!val_match) {
+                ret = snprintf(errmsg, sizeof(errmsg) - 1,
+                               "Invalid value(%s) for"
+                               " option %s",
+                               op_value, op_name);
+                errmsg[ret] = '\0';
+
+                gf_msg(this->name, GF_LOG_ERROR, EINVAL, GD_MSG_INVALID_ENTRY,
+                       "%s", errmsg);
+                *op_errstr = gf_strdup(errmsg);
+                return -1;
+            }
+        }
+    }
+out:
+    return 0;
+}
+
+static int
+glusterd_get_gsync_status_mst_slv(glusterd_volinfo_t *volinfo, char *slave,
+                                  char *conf_path, dict_t *rsp_dict,
+                                  char *node);
+
+static int
+_get_status_mst_slv(dict_t *dict, char *key, data_t *value, void *data)
+{
+    glusterd_gsync_status_temp_t *param = NULL;
+    char *slave = NULL;
+    char *slave_buf = NULL;
+    char *slave_url = NULL;
+    char *slave_vol = NULL;
+    char *slave_host = NULL;
+    char *errmsg = NULL;
+    char conf_path[PATH_MAX] = "";
+    int ret = -1;
+    glusterd_conf_t *priv = NULL;
+    xlator_t *this = NULL;
+    char slv_url[VOLINFO_SLAVE_URL_MAX] = {0};
+
+    this = THIS;
+    GF_VALIDATE_OR_GOTO("glusterd", this, out);
+
+    param = (glusterd_gsync_status_temp_t *)data;
+
+    GF_VALIDATE_OR_GOTO(this->name, param, out);
+    GF_VALIDATE_OR_GOTO(this->name, param->volinfo, out);
+
+    if (this)
+        priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, priv, out);
+
+    if (snprintf(slv_url, sizeof(slv_url), "%s", value->data) >=
+        sizeof(slv_url)) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVE_VOL_PARSE_FAIL,
+               "Error in copying slave: %s!", value->data);
+        goto out;
+    }
+
+    ret = parse_slave_url(slv_url, &slave);
+    if (ret == -1) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVE_VOL_PARSE_FAIL,
+               "Error in parsing slave: %s!", value->data);
+        goto out;
+    }
+
+    ret = glusterd_get_slave_info(slave, &slave_url, &slave_host, &slave_vol,
+                                  &errmsg);
+    if (ret) {
+        if (errmsg)
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVEINFO_FETCH_ERROR,
+                   "Unable to fetch slave details. Error: %s", errmsg);
+        else
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVEINFO_FETCH_ERROR,
+                   "Unable to fetch slave details.");
+        ret = -1;
+        goto out;
+    }
+
+    ret = snprintf(conf_path, sizeof(conf_path) - 1,
+                   "%s/" GEOREP "/%s_%s_%s/gsyncd.conf", priv->workdir,
+                   param->volinfo->volname, slave_host, slave_vol);
+    conf_path[ret] = '\0';
+
+    ret = glusterd_get_gsync_status_mst_slv(param->volinfo, slave, conf_path,
+                                            param->rsp_dict, param->node);
+out:
+
+    if (errmsg)
+        GF_FREE(errmsg);
+
+    if (slave_buf)
+        GF_FREE(slave_buf);
+
+    if (slave_vol)
+        GF_FREE(slave_vol);
+
+    if (slave_url)
+        GF_FREE(slave_url);
+
+    if (slave_host)
+        GF_FREE(slave_host);
+
+    gf_msg_debug(this ? this->name : "glusterd", 0, "Returning %d.", ret);
+    return ret;
+}
+
+static int
+_get_max_gsync_slave_num(dict_t *dict, char *key, data_t *value, void *data)
+{
+    int tmp_slvnum = 0;
+    int *slvnum = (int *)data;
+
+    sscanf(key, "slave%d", &tmp_slvnum);
+    if (tmp_slvnum > *slvnum)
+        *slvnum = tmp_slvnum;
+
+    return 0;
+}
+
+static int
+_get_slave_idx_slave_voluuid(dict_t *dict, char *key, data_t *value, void *data)
+{
+    char *slave_info = NULL;
+    xlator_t *this = NULL;
+    struct slave_vol_config *slave_cfg = NULL;
+    int i = 0;
+    int ret = -1;
+    unsigned tmp_slvnum = 0;
+
+    this = THIS;
+    GF_VALIDATE_OR_GOTO("glusterd", this, out);
+
+    slave_cfg = data;
+
+    if (value)
+        slave_info = value->data;
+
+    if (!(slave_info) || strlen(slave_info) == 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_INVALID_SLAVE,
+               "Invalid slave in dict");
+        ret = -2;
+        goto out;
+    }
+
+    /* slave format:
+     * master_node_uuid:ssh://slave_host::slave_vol:slave_voluuid */
+    while (i++ < 5) {
+        slave_info = strchr(slave_info, ':');
+        if (slave_info)
+            slave_info++;
+        else {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVE_VOL_PARSE_FAIL,
+                   "slave_info becomes NULL!");
+            ret = -2;
+            goto out;
+        }
+    }
+    if (strcmp(slave_info, slave_cfg->slave_voluuid) == 0) {
+        gf_msg_debug(this->name, 0,
+                     "Same slave volume "
+                     "already present %s",
+                     slave_cfg->slave_voluuid);
+        ret = -1;
+
+        sscanf(key, "slave%d", &tmp_slvnum);
+        slave_cfg->old_slvidx = tmp_slvnum;
+
+        gf_msg_debug(this->name, 0,
+                     "and "
+                     "its index is: %d",
+                     tmp_slvnum);
+        goto out;
+    }
+
+    ret = 0;
+out:
+    return ret;
+}
+
+static int
+glusterd_remove_slave_in_info(glusterd_volinfo_t *volinfo, char *slave,
+                              char **op_errstr)
+{
+    int zero_slave_entries = _gf_true;
+    int ret = 0;
+    char *slavekey = NULL;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(volinfo);
+    GF_ASSERT(slave);
+
+    do {
+        ret = glusterd_get_slave(volinfo, slave, &slavekey);
+        if (ret < 0 && zero_slave_entries) {
+            ret++;
+            goto out;
+        }
+        zero_slave_entries = _gf_false;
+        dict_del(volinfo->gsync_slaves, slavekey);
+    } while (ret >= 0);
+
+    ret = glusterd_store_volinfo(volinfo, GLUSTERD_VOLINFO_VER_AC_INCREMENT);
+    if (ret) {
+        *op_errstr = gf_strdup(
+            "Failed to store the Volume"
+            "information");
+        goto out;
+    }
+out:
+    gf_msg_debug(this->name, 0, "returning %d", ret);
+    return ret;
+}
+
+static int
+glusterd_gsync_get_uuid(char *slave, glusterd_volinfo_t *vol, uuid_t uuid)
+{
+    int ret = 0;
+    char *slavekey = NULL;
+    char *slaveentry = NULL;
+    char *t = NULL;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(vol);
+    GF_ASSERT(slave);
+
+    ret = glusterd_get_slave(vol, slave, &slavekey);
+    if (ret < 0) {
+        /* XXX colliding cases of failure and non-extant
+         * slave... now just doing this as callers of this
+         * function can make sense only of -1 and 0 as retvals;
+         * getting at the proper semanticals will involve
+         * fixing callers as well.
+         */
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_get_str(vol->gsync_slaves, slavekey, &slaveentry);
+    GF_ASSERT(ret == 0);
+
+    t = strchr(slaveentry, ':');
+    GF_ASSERT(t);
+    *t = '\0';
+    ret = gf_uuid_parse(slaveentry, uuid);
+    *t = ':';
+
+out:
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+static int
+update_slave_voluuid(dict_t *dict, char *key, data_t *value, void *data)
+{
+    char *slave = NULL;
+    char *slave_url = NULL;
+    char *slave_vol = NULL;
+    char *slave_host = NULL;
+    char *errmsg = NULL;
+    xlator_t *this = NULL;
+    int ret = -1;
+    char slv_url[VOLINFO_SLAVE_URL_MAX] = {0};
+    char slave_voluuid[GF_UUID_BUF_SIZE] = {0};
+    char *slave_info = NULL;
+    char *new_value = NULL;
+    char *same_key = NULL;
+    int cnt = 0;
+    gf_boolean_t *voluuid_updated = NULL;
+
+    this = THIS;
+
+    voluuid_updated = data;
+    slave_info = value->data;
+    gf_msg_debug(this->name, 0, "slave_info: %s!", slave_info);
+
+    /* old slave format:
+     * master_node_uuid:ssh://slave_host::slave_vol
+     * New slave format:
+     * master_node_uuid:ssh://slave_host::slave_vol:slave_voluuid */
+    while (slave_info) {
+        slave_info = strchr(slave_info, ':');
+        if (slave_info)
+            cnt++;
+        else
+            break;
+
+        slave_info++;
+    }
+
+    gf_msg_debug(this->name, 0, "cnt: %d", cnt);
+    /* check whether old slave format and update vol uuid if old format.
+     * With volume uuid, number of ':' is 5 and is 4 without.
+     */
+    if (cnt == 4) {
+        if (snprintf(slv_url, sizeof(slv_url), "%s", value->data) >=
+            sizeof(slv_url)) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVE_VOL_PARSE_FAIL,
+                   "Error in copying slave: %s!", value->data);
+            goto out;
+        }
+
+        ret = parse_slave_url(slv_url, &slave);
+        if (ret == -1) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVE_VOL_PARSE_FAIL,
+                   "Error in parsing slave: %s!", value->data);
+            goto out;
+        }
+
+        ret = glusterd_get_slave_info(slave, &slave_url, &slave_host,
+                                      &slave_vol, &errmsg);
+        if (ret) {
+            if (errmsg)
+                gf_msg(this->name, GF_LOG_ERROR, 0,
+                       GD_MSG_SLAVEINFO_FETCH_ERROR,
+                       "Unable to fetch slave details. Error: %s", errmsg);
+            else
+                gf_msg(this->name, GF_LOG_ERROR, 0,
+                       GD_MSG_SLAVEINFO_FETCH_ERROR,
+                       "Unable to fetch slave details.");
+            ret = -1;
+            goto out;
+        }
+
+        ret = glusterd_get_slave_voluuid(slave_host, slave_vol, slave_voluuid);
+        if ((ret) || (strlen(slave_voluuid) == 0)) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REMOTE_VOL_UUID_FAIL,
+                   "Unable to get remote volume uuid"
+                   "slavehost:%s slavevol:%s",
+                   slave_host, slave_vol);
+            /* Avoiding failure due to remote vol uuid fetch */
+            ret = 0;
+            goto out;
+        }
+        ret = gf_asprintf(&new_value, "%s:%s", value->data, slave_voluuid);
+        ret = gf_asprintf(&same_key, "%s", key);
+
+        /* delete old key and add new value */
+        dict_del(dict, key);
+
+        /* set new value for the same key*/
+        ret = dict_set_dynstr(dict, same_key, new_value);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REMOTE_VOL_UUID_FAIL,
+                   "Error in setting dict value"
+                   "new_value :%s",
+                   new_value);
+            goto out;
+        }
+        *voluuid_updated = _gf_true;
+    }
+
+    ret = 0;
+out:
+    if (errmsg)
+        GF_FREE(errmsg);
+
+    if (slave_url)
+        GF_FREE(slave_url);
+
+    if (slave_vol)
+        GF_FREE(slave_vol);
+
+    if (slave_host)
+        GF_FREE(slave_host);
+
+    gf_msg_debug(this->name, 0, "Returning %d.", ret);
+    return ret;
+}
+
+static int
+glusterd_update_slave_voluuid_slaveinfo(glusterd_volinfo_t *volinfo)
+{
+    int ret = -1;
+    xlator_t *this = NULL;
+    gf_boolean_t voluuid_updated = _gf_false;
+
+    this = THIS;
+    GF_VALIDATE_OR_GOTO("glusterd", this, out);
+    GF_VALIDATE_OR_GOTO(this->name, volinfo, out);
+
+    ret = dict_foreach(volinfo->gsync_slaves, update_slave_voluuid,
+                       &voluuid_updated);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REMOTE_VOL_UUID_FAIL,
+               "Error in updating"
+               "volinfo");
+        goto out;
+    }
+
+    if (_gf_true == voluuid_updated) {
+        ret = glusterd_store_volinfo(volinfo,
+                                     GLUSTERD_VOLINFO_VER_AC_INCREMENT);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_VOLINFO_STORE_FAIL,
+                   "Error in storing"
+                   "volinfo");
+            goto out;
+        }
+    }
+
+    ret = 0;
+out:
+    gf_msg_debug((this ? this->name : "glusterd"), 0, "Returning %d", ret);
+    return ret;
+}
+
+int
+glusterd_check_gsync_running_local(char *master, char *slave, char *conf_path,
+                                   gf_boolean_t *is_run)
+{
+    int ret = -1;
+    int ret_status = 0;
+    gf_boolean_t is_template_in_use = _gf_false;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(master);
+    GF_ASSERT(slave);
+    GF_ASSERT(is_run);
+
+    *is_run = _gf_false;
+    ret = gsync_status(master, slave, conf_path, &ret_status,
+                       &is_template_in_use);
+    if (ret == 0 && ret_status == 0)
+        *is_run = _gf_true;
+    else if (ret == -1) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_VALIDATE_FAILED,
+               GEOREP " validation failed");
+        goto out;
+    }
+    ret = 0;
+out:
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+static int
+glusterd_store_slave_in_info(glusterd_volinfo_t *volinfo, char *slave,
+                             char *host_uuid, char *slave_voluuid,
+                             char **op_errstr, gf_boolean_t is_force)
+{
+    int ret = 0;
+    int maxslv = 0;
+    char **linearr = NULL;
+    char *value = NULL;
+    char *slavekey = NULL;
+    char *slaveentry = NULL;
+    char key[32] = {
+        0,
+    };
+    int keylen;
+    char *t = NULL;
+    xlator_t *this = NULL;
+    struct slave_vol_config slave1 = {
+        {0},
+    };
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(volinfo);
+    GF_ASSERT(slave);
+    GF_ASSERT(host_uuid);
+    GF_VALIDATE_OR_GOTO(this->name, slave_voluuid, out);
+
+    ret = glusterd_get_slave(volinfo, slave, &slavekey);
+    switch (ret) {
+        case -2:
+            ret = -1;
+            goto out;
+        case -1:
+            break;
+        default:
+            if (!is_force)
+                GF_ASSERT(ret > 0);
+            ret = dict_get_str(volinfo->gsync_slaves, slavekey, &slaveentry);
+            GF_ASSERT(ret == 0);
+
+            /* same-name + same-uuid slave entries should have been filtered
+             * out in glusterd_op_verify_gsync_start_options(), so we can
+             * assert an uuid mismatch
+             */
+            t = strtail(slaveentry, host_uuid);
+            if (!is_force)
+                GF_ASSERT(!t || *t != ':');
+
+            if (is_force) {
+                gf_msg_debug(this->name, 0,
+                             GEOREP
+                             " has already "
+                             "been invoked for the %s (master) and "
+                             "%s (slave). Allowing without saving "
+                             "info again due to force command.",
+                             volinfo->volname, slave);
+                ret = 0;
+                goto out;
+            }
+
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_INVOKE_ERROR,
+                   GEOREP
+                   " has already been invoked for "
+                   "the %s (master) and %s (slave) from a different "
+                   "machine",
+                   volinfo->volname, slave);
+            *op_errstr = gf_strdup(GEOREP
+                                   " already running in "
+                                   "another machine");
+            ret = -1;
+            goto out;
+    }
+
+    ret = glusterd_urltransform_single(slave, "normalize", &linearr);
+    if (ret == -1)
+        goto out;
+
+    ret = gf_asprintf(&value, "%s:%s:%s", host_uuid, linearr[0], slave_voluuid);
+
+    glusterd_urltransform_free(linearr, 1);
+    if (ret == -1)
+        goto out;
+
+    /* Given the slave volume uuid, check and get any existing slave */
+    memcpy(slave1.slave_voluuid, slave_voluuid, UUID_CANONICAL_FORM_LEN);
+    ret = dict_foreach(volinfo->gsync_slaves, _get_slave_idx_slave_voluuid,
+                       &slave1);
+
+    if (ret == 0) { /* New slave */
+        dict_foreach(volinfo->gsync_slaves, _get_max_gsync_slave_num, &maxslv);
+        keylen = snprintf(key, sizeof(key), "slave%d", maxslv + 1);
+
+        ret = dict_set_dynstrn(volinfo->gsync_slaves, key, keylen, value);
+        if (ret) {
+            GF_FREE(value);
+            goto out;
+        }
+    } else if (ret == -1) { /* Existing slave */
+        keylen = snprintf(key, sizeof(key), "slave%d", slave1.old_slvidx);
+
+        gf_msg_debug(this->name, 0,
+                     "Replacing key:%s with new value"
+                     ":%s",
+                     key, value);
+
+        /* Add new slave's value, with the same slave index */
+        ret = dict_set_dynstrn(volinfo->gsync_slaves, key, keylen, value);
+        if (ret) {
+            GF_FREE(value);
+            goto out;
+        }
+    } else {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REMOTE_VOL_UUID_FAIL,
+               "_get_slave_idx_slave_voluuid failed!");
+        GF_FREE(value);
+        ret = -1;
+        goto out;
+    }
+
+    ret = glusterd_store_volinfo(volinfo, GLUSTERD_VOLINFO_VER_AC_INCREMENT);
+    if (ret) {
+        *op_errstr = gf_strdup(
+            "Failed to store the Volume "
+            "information");
+        goto out;
+    }
+    ret = 0;
+out:
+    return ret;
+}
+
+static int
+glusterd_op_verify_gsync_start_options(glusterd_volinfo_t *volinfo, char *slave,
+                                       char *conf_path, char *statefile,
+                                       char **op_errstr, gf_boolean_t is_force)
+{
+    int ret = -1;
+    int ret_status = 0;
+    gf_boolean_t is_template_in_use = _gf_false;
+    char msg[2048] = {0};
+    uuid_t uuid = {0};
+    xlator_t *this = NULL;
+    struct stat stbuf = {
+        0,
+    };
+    char statefiledir[PATH_MAX] = {
+        0,
+    };
+    char *statedir = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(volinfo);
+    GF_ASSERT(slave);
+    GF_ASSERT(op_errstr);
+    GF_ASSERT(conf_path);
+    GF_ASSERT(this && this->private);
+
+    if (GLUSTERD_STATUS_STARTED != volinfo->status) {
+        snprintf(msg, sizeof(msg),
+                 "Volume %s needs to be started "
+                 "before " GEOREP " start",
+                 volinfo->volname);
+        goto out;
+    }
+
+    /* check session directory as statefile may not present
+     * during upgrade */
+    if (snprintf(statefiledir, sizeof(statefiledir), "%s", statefile) >=
+        sizeof(statefiledir)) {
+        snprintf(msg, sizeof(msg), "statefiledir truncated");
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_FILE_OP_FAILED, "%s",
+               msg);
+        *op_errstr = gf_strdup(msg);
+        goto out;
+    }
+    statedir = dirname(statefiledir);
+
+    ret = sys_lstat(statedir, &stbuf);
+    if (ret) {
+        snprintf(msg, sizeof(msg),
+                 "Session between %s and %s has"
+                 " not been created. Please create session and retry.",
+                 volinfo->volname, slave);
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_FILE_OP_FAILED,
+               "%s statefile: %s", msg, statefile);
+        *op_errstr = gf_strdup(msg);
+        goto out;
+    }
+
+    /* Check if the gsync slave info is stored. If not
+     * session has not been created */
+    ret = glusterd_gsync_get_uuid(slave, volinfo, uuid);
+    if (ret) {
+        snprintf(msg, sizeof(msg),
+                 "Session between %s and %s has"
+                 " not been created. Please create session and retry.",
+                 volinfo->volname, slave);
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SESSION_CREATE_ERROR, "%s",
+               msg);
+        goto out;
+    }
+
+    /*Check if the gsync is already started in cmd. inited host
+     * If so initiate add it into the glusterd's priv*/
+    ret = gsync_status(volinfo->volname, slave, conf_path, &ret_status,
+                       &is_template_in_use);
+    if (ret == 0) {
+        if ((ret_status == 0) && !is_force) {
+            snprintf(msg, sizeof(msg),
+                     GEOREP
+                     " session between"
+                     " %s & %s already started",
+                     volinfo->volname, slave);
+            ret = -1;
+            goto out;
+        }
+    } else if (ret == -1) {
+        snprintf(msg, sizeof(msg),
+                 GEOREP
+                 " start option "
+                 "validation failed ");
+        goto out;
+    }
+
+    if (is_template_in_use == _gf_true) {
+        snprintf(msg, sizeof(msg),
+                 GEOREP
+                 " start "
+                 "failed : pid-file entry missing "
+                 "in config file.");
+        ret = -1;
+        goto out;
+    }
+
+    ret = glusterd_verify_gsyncd_spawn(volinfo->volname, slave);
+    if (ret && !is_force) {
+        snprintf(msg, sizeof(msg), "Unable to spawn gsyncd");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GSYNCD_SPAWN_FAILED, "%s",
+               msg);
+    }
+out:
+    if (ret && (msg[0] != '\0')) {
+        *op_errstr = gf_strdup(msg);
+    }
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+void
+glusterd_check_geo_rep_configured(glusterd_volinfo_t *volinfo,
+                                  gf_boolean_t *flag)
+{
+    GF_ASSERT(volinfo);
+    GF_ASSERT(flag);
+
+    if (volinfo->gsync_slaves->count)
+        *flag = _gf_true;
+    else
+        *flag = _gf_false;
+
+    return;
+}
+
+/*
+ * is_geo_rep_active:
+ *      This function reads the state_file and sets is_active to 1 if the
+ *      monitor status is neither "Stopped" or "Created"
+ *
+ * RETURN VALUE:
+ *       0: On successful read of state_file.
+ *      -1: error.
+ */
+
+static int
+is_geo_rep_active(glusterd_volinfo_t *volinfo, char *slave, char *conf_path,
+                  int *is_active)
+{
+    dict_t *confd = NULL;
+    char *statefile = NULL;
+    char *master = NULL;
+    char monitor_status[PATH_MAX] = "";
+    int ret = -1;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    master = volinfo->volname;
+
+    confd = dict_new();
+    if (!confd) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_CREATE_FAIL,
+               "Not able to create dict.");
+        goto out;
+    }
+
+    ret = glusterd_gsync_get_config(master, slave, conf_path, confd);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GET_CONFIG_INFO_FAILED,
+               "Unable to get configuration data "
+               "for %s(master), %s(slave)",
+               master, slave);
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_get_param(confd, "state_file", &statefile);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+               "Unable to get state_file's name "
+               "for %s(master), %s(slave). Please check gsync "
+               "config file.",
+               master, slave);
+        ret = -1;
+        goto out;
+    }
+
+    ret = glusterd_gsync_read_frm_status(statefile, monitor_status,
+                                         sizeof(monitor_status));
+    if (ret <= 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_STAT_FILE_READ_FAILED,
+               "Unable to read the status file for %s(master), "
+               "%s(slave)",
+               master, slave);
+        snprintf(monitor_status, sizeof(monitor_status), "defunct");
+    }
+
+    if ((!strcmp(monitor_status, "Stopped")) ||
+        (!strcmp(monitor_status, "Created"))) {
+        *is_active = 0;
+    } else {
+        *is_active = 1;
+    }
+    ret = 0;
+out:
+    if (confd)
+        dict_unref(confd);
+    return ret;
+}
+
+/*
+ * _get_slave_status:
+ *      Called for each slave in the volume from dict_foreach.
+ *      It calls is_geo_rep_active to get the monitor status.
+ *
+ * RETURN VALUE:
+ *      0: On successful read of state_file from is_geo_rep_active.
+ *         When it is found geo-rep is already active from previous calls.
+ *         When there is no slave.
+ *     -1: On error.
+ */
+
+int
+_get_slave_status(dict_t *dict, char *key, data_t *value, void *data)
+{
+    gsync_status_param_t *param = NULL;
+    char *slave = NULL;
+    char *slave_url = NULL;
+    char *slave_vol = NULL;
+    char *slave_host = NULL;
+    char *errmsg = NULL;
+    char conf_path[PATH_MAX] = "";
+    int ret = -1;
+    glusterd_conf_t *priv = NULL;
+    xlator_t *this = NULL;
+
+    param = (gsync_status_param_t *)data;
+
+    GF_ASSERT(param);
+    GF_ASSERT(param->volinfo);
+    if (param->is_active) {
+        ret = 0;
+        goto out;
+    }
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    priv = this->private;
+    if (priv == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GLUSTERD_PRIV_NOT_FOUND,
+               "priv of glusterd not present");
+        goto out;
+    }
+
+    slave = strchr(value->data, ':');
+    if (!slave) {
+        ret = 0;
+        goto out;
+    }
+    slave++;
+
+    ret = glusterd_get_slave_info(slave, &slave_url, &slave_host, &slave_vol,
+                                  &errmsg);
+    if (ret) {
+        if (errmsg)
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVEINFO_FETCH_ERROR,
+                   "Unable to fetch"
+                   " slave details. Error: %s",
+                   errmsg);
+        else
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVEINFO_FETCH_ERROR,
+                   "Unable to fetch slave details.");
+        ret = -1;
+        goto out;
+    }
+
+    ret = snprintf(conf_path, sizeof(conf_path) - 1,
+                   "%s/" GEOREP "/%s_%s_%s/gsyncd.conf", priv->workdir,
+                   param->volinfo->volname, slave_host, slave_vol);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_CONF_PATH_ASSIGN_FAILED,
+               "Unable to assign conf_path.");
+        ret = -1;
+        goto out;
+    }
+    conf_path[ret] = '\0';
+
+    ret = is_geo_rep_active(param->volinfo, slave, conf_path,
+                            &param->is_active);
+out:
+    if (errmsg)
+        GF_FREE(errmsg);
+
+    if (slave_vol)
+        GF_FREE(slave_vol);
+
+    if (slave_url)
+        GF_FREE(slave_url);
+    if (slave_host)
+        GF_FREE(slave_host);
+
+    return ret;
+}
+
+/* glusterd_check_geo_rep_running:
+ *          Checks if any geo-rep session is running for the volume.
+ *
+ *    RETURN VALUE:
+ *          Sets param.active to true if any geo-rep session is active.
+ *    This function sets op_errstr during some error and when any geo-rep
+ *    session is active. It is caller's responsibility to free op_errstr
+ *    in above cases.
+ */
+
+int
+glusterd_check_geo_rep_running(gsync_status_param_t *param, char **op_errstr)
+{
+    char msg[2048] = {
+        0,
+    };
+    gf_boolean_t enabled = _gf_false;
+    int ret = 0;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+    GF_ASSERT(param);
+    GF_ASSERT(param->volinfo);
+    GF_ASSERT(op_errstr);
+
+    glusterd_check_geo_rep_configured(param->volinfo, &enabled);
+
+    if (enabled) {
+        ret = dict_foreach(param->volinfo->gsync_slaves, _get_slave_status,
+                           param);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVEINFO_FETCH_ERROR,
+                   "_get_slave_satus failed");
+            snprintf(msg, sizeof(msg),
+                     GEOREP
+                     " Unable to"
+                     " get the status of active " GEOREP
+                     ""
+                     " session for the volume '%s'.\n"
+                     " Please check the log file for"
+                     " more info.",
+                     param->volinfo->volname);
+            *op_errstr = gf_strdup(msg);
+            ret = -1;
+            goto out;
+        }
+
+        if (param->is_active) {
+            snprintf(msg, sizeof(msg),
+                     GEOREP
+                     " sessions"
+                     " are active for the volume %s.\nStop"
+                     " " GEOREP
+                     " sessions involved in this"
+                     " volume. Use 'volume " GEOREP
+                     " status' command for more info.",
+                     param->volinfo->volname);
+            *op_errstr = gf_strdup(msg);
+            goto out;
+        }
+    }
+out:
+    return ret;
+}
+
+static int
+glusterd_op_verify_gsync_running(glusterd_volinfo_t *volinfo, char *slave,
+                                 char *conf_path, char **op_errstr)
+{
+    int pfd = -1;
+    int ret = -1;
+    char msg[2048] = {0};
+    char pidfile[PATH_MAX] = {
+        0,
+    };
+    gf_boolean_t is_template_in_use = _gf_false;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(THIS && THIS->private);
+    GF_ASSERT(volinfo);
+    GF_ASSERT(slave);
+    GF_ASSERT(conf_path);
+    GF_ASSERT(op_errstr);
+
+    if (GLUSTERD_STATUS_STARTED != volinfo->status) {
+        snprintf(msg, sizeof(msg),
+                 "Volume %s needs to be started "
+                 "before " GEOREP " start",
+                 volinfo->volname);
+        gf_smsg(this->name, GF_LOG_ERROR, 0, GD_MSG_GEO_REP_START_FAILED,
+                "Volume is not in a started state, Volname=%s",
+                volinfo->volname, NULL);
+
+        goto out;
+    }
+
+    pfd = gsyncd_getpidfile(volinfo->volname, slave, pidfile, conf_path,
+                            &is_template_in_use);
+    if (pfd == -2) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_VALIDATE_FAILED,
+               GEOREP " stop validation failed for %s & %s", volinfo->volname,
+               slave);
+        ret = -1;
+        goto out;
+    }
+    if (gsync_status_byfd(pfd) == -1) {
+        snprintf(msg, sizeof(msg),
+                 GEOREP
+                 " session b/w %s & %s is "
+                 "not running on this node.",
+                 volinfo->volname, slave);
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SESSION_INACTIVE, "%s", msg);
+        ret = -1;
+        /* monitor gsyncd already dead */
+        goto out;
+    }
+
+    if (is_template_in_use) {
+        snprintf(msg, sizeof(msg),
+                 "pid-file entry missing in "
+                 "the config file(%s).",
+                 conf_path);
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_PIDFILE_NOT_FOUND, "%s",
+               msg);
+        ret = -1;
+        goto out;
+    }
+
+    if (pfd < 0)
+        goto out;
+
+    ret = 0;
+out:
+    if (ret && (msg[0] != '\0')) {
+        *op_errstr = gf_strdup(msg);
+    }
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+static int
+glusterd_verify_gsync_status_opts(dict_t *dict, char **op_errstr)
+{
+    char *slave = NULL;
+    char *volname = NULL;
+    char errmsg[PATH_MAX] = {
+        0,
+    };
+    glusterd_volinfo_t *volinfo = NULL;
+    int ret = 0;
+    char *conf_path = NULL;
+    char *slave_url = NULL;
+    char *slave_host = NULL;
+    char *slave_vol = NULL;
+    glusterd_conf_t *priv = NULL;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    if (THIS)
+        priv = THIS->private;
+    if (priv == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GLUSTERD_PRIV_NOT_FOUND,
+               "priv of glusterd not present");
+        *op_errstr = gf_strdup("glusterd defunct");
+        goto out;
+    }
+
+    ret = dict_get_str(dict, "master", &volname);
+    if (ret < 0) {
+        ret = 0;
+        goto out;
+    }
+
+    ret = glusterd_volinfo_find(volname, &volinfo);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_VOL_NOT_FOUND,
+               "volume name does not exist");
+        snprintf(errmsg, sizeof(errmsg),
+                 "Volume name %s does not"
+                 " exist",
+                 volname);
+        *op_errstr = gf_strdup(errmsg);
+        goto out;
+    }
+
+    ret = dict_get_str(dict, "slave", &slave);
+    if (ret < 0) {
+        ret = 0;
+        goto out;
+    }
+
+    ret = glusterd_get_slave_details_confpath(volinfo, dict, &slave_url,
+                                              &slave_host, &slave_vol,
+                                              &conf_path, op_errstr);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVEINFO_FETCH_ERROR,
+               "Unable to fetch slave  or confpath details.");
+        ret = -1;
+        goto out;
+    }
+
+out:
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+int
+glusterd_op_gsync_args_get(dict_t *dict, char **op_errstr, char **master,
+                           char **slave, char **host_uuid)
+{
+    int ret = -1;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+    GF_ASSERT(dict);
+    GF_ASSERT(op_errstr);
+
+    if (master) {
+        ret = dict_get_str(dict, "master", master);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_DICT_GET_FAILED,
+                   "master not found");
+            *op_errstr = gf_strdup("master not found");
+            goto out;
+        }
+    }
+
+    if (slave) {
+        ret = dict_get_str(dict, "slave", slave);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_DICT_GET_FAILED,
+                   "slave not found");
+            *op_errstr = gf_strdup("slave not found");
+            goto out;
+        }
+    }
+
+    if (host_uuid) {
+        ret = dict_get_str(dict, "host-uuid", host_uuid);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_DICT_GET_FAILED,
+                   "host_uuid not found");
+            *op_errstr = gf_strdup("host_uuid not found");
+            goto out;
+        }
+    }
+
+    ret = 0;
+out:
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+int
+glusterd_op_stage_sys_exec(dict_t *dict, char **op_errstr)
+{
+    char errmsg[PATH_MAX] = "";
+    char *command = NULL;
+    char command_path[PATH_MAX] = "";
+    struct stat st = {
+        0,
+    };
+    int ret = -1;
+    glusterd_conf_t *conf = NULL;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+    conf = this->private;
+    GF_ASSERT(conf);
+
+    if (conf->op_version < 2) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_UNSUPPORTED_VERSION,
+               "Op Version not supported.");
+        snprintf(errmsg, sizeof(errmsg),
+                 "One or more nodes do not"
+                 " support the required op version.");
+        *op_errstr = gf_strdup(errmsg);
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_get_str(dict, "command", &command);
+    if (ret) {
+        strcpy(errmsg, "internal error");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+               "Unable to get command from dict");
+        goto out;
+    }
+
+    /* enforce local occurrence of the command */
+    if (strchr(command, '/')) {
+        strcpy(errmsg, "invalid command name");
+        ret = -1;
+        goto out;
+    }
+
+    sprintf(command_path, GSYNCD_PREFIX "/peer_%s", command);
+    /* check if it's executable */
+    ret = sys_access(command_path, X_OK);
+    if (!ret)
+        /* check if it's a regular file */
+        ret = sys_stat(command_path, &st);
+    if (!ret && !S_ISREG(st.st_mode))
+        ret = -1;
+
+out:
+    if (ret) {
+        if (errmsg[0] == '\0') {
+            if (command)
+                snprintf(errmsg, sizeof(errmsg),
+                         "gsync peer_%s command not found.", command);
+            else
+                snprintf(errmsg, sizeof(errmsg), "%s",
+                         "gsync peer command was not "
+                         "specified");
+        }
+        *op_errstr = gf_strdup(errmsg);
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_PEER_CMD_ERROR, "%s",
+               errmsg);
+    }
+
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+int
+glusterd_op_stage_copy_file(dict_t *dict, char **op_errstr)
+{
+    char abs_filename[PATH_MAX] = "";
+    char errmsg[PATH_MAX] = "";
+    char *filename = NULL;
+    char *host_uuid = NULL;
+    char uuid_str[64] = {0};
+    int ret = -1;
+    glusterd_conf_t *priv = NULL;
+    struct stat stbuf = {
+        0,
+    };
+    xlator_t *this = NULL;
+    char workdir[PATH_MAX] = {
+        0,
+    };
+    char realpath_filename[PATH_MAX] = {
+        0,
+    };
+    char realpath_workdir[PATH_MAX] = {
+        0,
+    };
+    int32_t len = 0;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    if (THIS)
+        priv = THIS->private;
+    if (priv == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GLUSTERD_PRIV_NOT_FOUND,
+               "priv of glusterd not present");
+        *op_errstr = gf_strdup("glusterd defunct");
+        goto out;
+    }
+
+    if (priv->op_version < 2) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_UNSUPPORTED_VERSION,
+               "Op Version not supported.");
+        snprintf(errmsg, sizeof(errmsg),
+                 "One or more nodes do not"
+                 " support the required op version.");
+        *op_errstr = gf_strdup(errmsg);
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_get_str(dict, "host-uuid", &host_uuid);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+               "Unable to fetch host-uuid from dict.");
+        goto out;
+    }
+
+    uuid_utoa_r(MY_UUID, uuid_str);
+    if (!strcmp(uuid_str, host_uuid)) {
+        ret = dict_get_str(dict, "source", &filename);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+                   "Unable to fetch filename from dict.");
+            *op_errstr = gf_strdup("command unsuccessful");
+            goto out;
+        }
+        len = snprintf(abs_filename, sizeof(abs_filename), "%s/%s",
+                       priv->workdir, filename);
+        if ((len < 0) || (len >= sizeof(abs_filename))) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, GD_MSG_COPY_FAIL, NULL);
+            ret = -1;
+            goto out;
+        }
+
+        if (!realpath(priv->workdir, realpath_workdir)) {
+            len = snprintf(errmsg, sizeof(errmsg),
+                           "Failed to "
+                           "get realpath of %s: %s",
+                           priv->workdir, strerror(errno));
+            if (len < 0) {
+                strcpy(errmsg, "<error>");
+            }
+            gf_smsg(this->name, GF_LOG_ERROR, 0, GD_MSG_REALPATH_GET_FAIL,
+                    "Realpath=%s, Reason=%s", priv->workdir, strerror(errno),
+                    NULL);
+            *op_errstr = gf_strdup(errmsg);
+            ret = -1;
+            goto out;
+        }
+
+        if (!realpath(abs_filename, realpath_filename)) {
+            snprintf(errmsg, sizeof(errmsg),
+                     "Failed to get "
+                     "realpath of %s: %s",
+                     filename, strerror(errno));
+            gf_smsg(this->name, GF_LOG_ERROR, 0, GD_MSG_REALPATH_GET_FAIL,
+                    "Filename=%s, Reason=%s", filename, strerror(errno), NULL);
+            *op_errstr = gf_strdup(errmsg);
+            ret = -1;
+            goto out;
+        }
+
+        /* Add Trailing slash to workdir, without slash strncmp
+           will succeed for /var/lib/glusterd_bad */
+        len = snprintf(workdir, sizeof(workdir), "%s/", realpath_workdir);
+        if ((len < 0) || (len >= sizeof(workdir))) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, GD_MSG_COPY_FAIL, NULL);
+            ret = -1;
+            goto out;
+        }
+
+        /* Protect against file copy outside $workdir */
+        if (strncmp(workdir, realpath_filename, strlen(workdir))) {
+            len = snprintf(errmsg, sizeof(errmsg),
+                           "Source file"
+                           " is outside of %s directory",
+                           priv->workdir);
+            if (len < 0) {
+                strcpy(errmsg, "<error>");
+            }
+            gf_smsg(this->name, GF_LOG_ERROR, 0, GD_MSG_SRC_FILE_ERROR, errmsg,
+                    NULL);
+            *op_errstr = gf_strdup(errmsg);
+            ret = -1;
+            goto out;
+        }
+
+        ret = sys_lstat(abs_filename, &stbuf);
+        if (ret) {
+            len = snprintf(errmsg, sizeof(errmsg),
+                           "Source file"
+                           " does not exist in %s",
+                           priv->workdir);
+            if (len < 0) {
+                strcpy(errmsg, "<error>");
+            }
+            gf_smsg(this->name, GF_LOG_ERROR, 0, GD_MSG_SRC_FILE_ERROR, errmsg,
+                    NULL);
+            *op_errstr = gf_strdup(errmsg);
+            goto out;
+        }
+
+        if (!S_ISREG(stbuf.st_mode)) {
+            snprintf(errmsg, sizeof(errmsg),
+                     "Source file"
+                     " is not a regular file.");
+            gf_smsg(this->name, GF_LOG_ERROR, 0, GD_MSG_SRC_FILE_ERROR, errmsg,
+                    NULL);
+            *op_errstr = gf_strdup(errmsg);
+            ret = -1;
+            goto out;
+        }
+    }
+
+    ret = 0;
+out:
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+int
+glusterd_get_statefile_name(glusterd_volinfo_t *volinfo, char *slave,
+                            char *conf_path, char **statefile,
+                            gf_boolean_t *is_template_in_use)
+{
+    char *master = NULL;
+    char *buf = NULL;
+    char *working_conf_path = NULL;
+    char temp_conf_path[PATH_MAX] = "";
+    dict_t *confd = NULL;
+    glusterd_conf_t *priv = NULL;
+    int ret = -1;
+    struct stat stbuf = {
+        0,
+    };
+    xlator_t *this = NULL;
+    int32_t len = 0;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(this->private);
+    GF_ASSERT(volinfo);
+    GF_ASSERT(conf_path);
+    GF_ASSERT(is_template_in_use);
+
+    master = volinfo->volname;
+
+    confd = dict_new();
+    if (!confd) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_CREATE_FAIL,
+               "Unable to create new dict");
+        goto out;
+    }
+
+    priv = THIS->private;
+
+    len = snprintf(temp_conf_path, sizeof(temp_conf_path),
+                   "%s/" GSYNC_CONF_TEMPLATE, priv->workdir);
+    if ((len < 0) || (len >= sizeof(temp_conf_path))) {
+        goto out;
+    }
+
+    ret = sys_lstat(conf_path, &stbuf);
+    if (!ret) {
+        gf_msg(this->name, GF_LOG_INFO, 0, GD_MSG_CONFIG_INFO,
+               "Using passed config template(%s).", conf_path);
+        working_conf_path = conf_path;
+    } else {
+        gf_msg(this->name, GF_LOG_WARNING, ENOENT, GD_MSG_FILE_OP_FAILED,
+               "Config file (%s) missing. Looking for template config"
+               " file (%s)",
+               conf_path, temp_conf_path);
+        ret = sys_lstat(temp_conf_path, &stbuf);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOENT, GD_MSG_FILE_OP_FAILED,
+                   "Template "
+                   "config file (%s) missing.",
+                   temp_conf_path);
+            goto out;
+        }
+        gf_msg(this->name, GF_LOG_INFO, 0, GD_MSG_DEFAULT_TEMP_CONFIG,
+               "Using default config template(%s).", temp_conf_path);
+        working_conf_path = temp_conf_path;
+        *is_template_in_use = _gf_true;
+    }
+
+fetch_data:
+    ret = glusterd_gsync_get_config(master, slave, working_conf_path, confd);
+    if (ret) {
+        if (*is_template_in_use == _gf_false) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GET_CONFIG_INFO_FAILED,
+                   "Unable to get configuration data "
+                   "for %s(master), %s(slave). "
+                   "Trying template config.",
+                   master, slave);
+            working_conf_path = temp_conf_path;
+            *is_template_in_use = _gf_true;
+            goto fetch_data;
+        } else {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GET_CONFIG_INFO_FAILED,
+                   "Unable to get configuration data "
+                   "for %s(master), %s(slave) from "
+                   "template config",
+                   master, slave);
+            goto out;
+        }
+    }
+
+    ret = dict_get_param(confd, "state_file", &buf);
+    if (ret) {
+        if (*is_template_in_use == _gf_false) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+                   "Unable to get state_file's name. "
+                   "Trying template config.");
+            working_conf_path = temp_conf_path;
+            *is_template_in_use = _gf_true;
+            goto fetch_data;
+        } else {
+            gf_msg(this->name, GF_LOG_ERROR, 0,
+                   GD_MSG_GET_STATEFILE_NAME_FAILED,
+                   "Unable to get state_file's "
+                   "name from template.");
+            goto out;
+        }
+    }
+
+    ret = 0;
+out:
+    if (buf) {
+        *statefile = gf_strdup(buf);
+        if (!*statefile)
+            ret = -1;
+    }
+
+    if (confd)
+        dict_unref(confd);
+
+    gf_msg_debug(this->name, 0, "Returning %d ", ret);
+    return ret;
+}
+
+int
+glusterd_create_status_file(char *master, char *slave, char *slave_host,
+                            char *slave_vol, char *status)
+{
+    int ret = -1;
+    runner_t runner = {
+        0,
+    };
+    glusterd_conf_t *priv = NULL;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    if (THIS)
+        priv = THIS->private;
+    if (priv == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GLUSTERD_PRIV_NOT_FOUND,
+               "priv of glusterd not present");
+        goto out;
+    }
+
+    if (!status) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_STATUS_NULL, "Status Empty");
+        goto out;
+    }
+    gf_msg_debug(this->name, 0, "slave = %s", slave);
+
+    runinit(&runner);
+    runner_add_args(&runner, GSYNCD_PREFIX "/gsyncd", "--create", status, "-c",
+                    NULL);
+    runner_argprintf(&runner, "%s/" GEOREP "/%s_%s_%s/gsyncd.conf",
+                     priv->workdir, master, slave_host, slave_vol);
+    runner_argprintf(&runner, "--iprefix=%s", DATADIR);
+    runner_argprintf(&runner, ":%s", master);
+    runner_add_args(&runner, slave, NULL);
+    synclock_unlock(&priv->big_lock);
+    ret = runner_run(&runner);
+    synclock_lock(&priv->big_lock);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_STATUSFILE_CREATE_FAILED,
+               "Creating status file failed.");
+        ret = -1;
+        goto out;
+    }
+
+    ret = 0;
+out:
+    gf_msg_debug(this->name, 0, "returning %d", ret);
+    return ret;
+}
+
+static int
+glusterd_verify_slave(char *volname, char *slave_url, char *slave_vol,
+                      int ssh_port, char **op_errstr,
+                      gf_boolean_t *is_force_blocker)
+{
+    int32_t ret = -1;
+    runner_t runner = {
+        0,
+    };
+    char log_file_path[PATH_MAX] = "";
+    char buf[PATH_MAX] = "";
+    char *tmp = NULL;
+    char *slave_url_buf = NULL;
+    char *save_ptr = NULL;
+    char *slave_user = NULL;
+    char *slave_ip = NULL;
+    glusterd_conf_t *priv = NULL;
+    xlator_t *this = NULL;
+    char *af = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+    priv = this->private;
+    GF_ASSERT(priv);
+    GF_ASSERT(volname);
+    GF_ASSERT(slave_url);
+    GF_ASSERT(slave_vol);
+
+    /* Fetch the slave_user and slave_ip from the slave_url.
+     * If the slave_user is not present. Use "root"
+     */
+    if (strstr(slave_url, "@")) {
+        slave_url_buf = gf_strdup(slave_url);
+        if (!slave_url_buf) {
+            gf_smsg(this->name, GF_LOG_ERROR, 0, GD_MSG_STRDUP_FAILED,
+                    "Slave_url=%s", slave_url, NULL);
+            goto out;
+        }
+
+        slave_user = strtok_r(slave_url_buf, "@", &save_ptr);
+        slave_ip = strtok_r(NULL, "@", &save_ptr);
+    } else {
+        slave_user = "root";
+        slave_ip = slave_url;
+    }
+
+    if (!slave_user || !slave_ip) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVE_URL_INVALID,
+               "Invalid slave url.");
+        goto out;
+    }
+
+    snprintf(log_file_path, sizeof(log_file_path), "%s/create_verify_log",
+             priv->logdir);
+
+    runinit(&runner);
+    runner_add_args(&runner, GSYNCD_PREFIX "/gverify.sh", NULL);
+    runner_argprintf(&runner, "%s", volname);
+    runner_argprintf(&runner, "%s", slave_user);
+    runner_argprintf(&runner, "%s", slave_ip);
+    runner_argprintf(&runner, "%s", slave_vol);
+    runner_argprintf(&runner, "%d", ssh_port);
+    runner_argprintf(&runner, "%s", log_file_path);
+    ret = dict_get_str(this->options, "transport.address-family", &af);
+    if (ret)
+        af = "-";
+
+    runner_argprintf(&runner, "%s", af);
+
+    gf_msg_debug(this->name, 0, "gverify Args = %s %s %s %s %s %s %s %s",
+                 runner.argv[0], runner.argv[1], runner.argv[2], runner.argv[3],
+                 runner.argv[4], runner.argv[5], runner.argv[6],
+                 runner.argv[7]);
+    runner_redir(&runner, STDOUT_FILENO, RUN_PIPE);
+    synclock_unlock(&priv->big_lock);
+    ret = runner_run(&runner);
+    synclock_lock(&priv->big_lock);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_INVALID_SLAVE,
+               "Not a valid slave");
+        ret = glusterd_gsync_read_frm_status(log_file_path, buf, sizeof(buf));
+        if (ret <= 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_READ_ERROR,
+                   "Unable to read from %s", log_file_path);
+            goto out;
+        }
+
+        /* Tokenize the error message from gverify.sh to figure out
+         * if the error is a force blocker or not. */
+        tmp = strtok_r(buf, "|", &save_ptr);
+        if (!tmp) {
+            ret = -1;
+            goto out;
+        }
+        if (!strcmp(tmp, "FORCE_BLOCKER"))
+            *is_force_blocker = 1;
+        else {
+            /* No FORCE_BLOCKER flag present so all that is
+             * present is the error message. */
+            *is_force_blocker = 0;
+            *op_errstr = gf_strdup(tmp);
+            ret = -1;
+            goto out;
+        }
+
+        /* Copy rest of the error message to op_errstr */
+        tmp = strtok_r(NULL, "|", &save_ptr);
+        if (tmp)
+            *op_errstr = gf_strdup(tmp);
+        ret = -1;
+        goto out;
+    }
+    ret = 0;
+out:
+    GF_FREE(slave_url_buf);
+    sys_unlink(log_file_path);
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+/** @slave_ip remains unmodified */
+int
+glusterd_geo_rep_parse_slave(char *slave_url, char **hostname, char **op_errstr)
+{
+    int ret = -1;
+    char *tmp = NULL;
+    char *save_ptr = NULL;
+    char *host = NULL;
+    char errmsg[PATH_MAX] = "";
+    char *saved_url = NULL;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(slave_url);
+    GF_ASSERT(*slave_url);
+
+    saved_url = gf_strdup(slave_url);
+    if (!saved_url)
+        goto out;
+
+    /* Checking if hostname has user specified */
+    host = strstr(saved_url, "@");
+    if (!host) { /* no user specified */
+        if (hostname) {
+            *hostname = gf_strdup(saved_url);
+            if (!*hostname)
+                goto out;
+        }
+
+        ret = 0;
+        goto out;
+    } else {
+        /* Moving the host past the '@' and checking if the
+         * actual hostname also has '@' */
+        host++;
+        if (strstr(host, "@")) {
+            gf_msg_debug(this->name, 0, "host = %s", host);
+            ret = snprintf(errmsg, sizeof(errmsg) - 1, "Invalid Hostname (%s).",
+                           host);
+            errmsg[ret] = '\0';
+            gf_msg(this->name, GF_LOG_ERROR, EINVAL, GD_MSG_INVALID_ENTRY, "%s",
+                   errmsg);
+            ret = -1;
+            if (op_errstr)
+                *op_errstr = gf_strdup(errmsg);
+            goto out;
+        }
+
+        ret = -1;
+
+        /**
+         * preliminary check for valid slave format.
+         */
+        tmp = strtok_r(saved_url, "@", &save_ptr);
+        tmp = strtok_r(NULL, "@", &save_ptr);
+        if (!tmp)
+            goto out;
+        if (hostname) {
+            *hostname = gf_strdup(tmp);
+            if (!*hostname)
+                goto out;
+        }
+    }
+
+    ret = 0;
+out:
+    GF_FREE(saved_url);
+    if (ret)
+        if (hostname)
+            GF_FREE(*hostname);
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+/* Return -1 only if there is a match in volume uuid */
+static int
+get_slavehost_from_voluuid(dict_t *dict, char *key, data_t *value, void *data)
+{
+    char *slave_info = NULL;
+    char *tmp = NULL;
+    char *slave_host = NULL;
+    xlator_t *this = NULL;
+    struct slave_vol_config *slave_vol = NULL;
+    int i = 0;
+    int ret = -1;
+
+    this = THIS;
+    GF_VALIDATE_OR_GOTO("glusterd", this, out);
+
+    slave_vol = data;
+    slave_info = value->data;
+
+    gf_msg_debug(this->name, 0, "slave_info:%s !", slave_info);
+
+    if (!(slave_info) || strlen(slave_info) == 0) {
+        /* no slaves present, peace  */
+        ret = 0;
+        goto out;
+    }
+
+    /* slave format:
+     * master_node_uuid:ssh://slave_host::slave_vol:slave_voluuid */
+    while (i++ < 5) {
+        slave_info = strchr(slave_info, ':');
+        if (slave_info)
+            slave_info++;
+        else
+            break;
+    }
+
+    if (!(slave_info) || strlen(slave_info) == 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVE_VOL_PARSE_FAIL,
+               "slave_info format is wrong!");
+        ret = -2;
+        goto out;
+    } else {
+        if (strcmp(slave_info, slave_vol->slave_voluuid) == 0) {
+            ret = -1;
+
+            /* get corresponding slave host for reference*/
+            slave_host = value->data;
+            slave_host = strstr(slave_host, "://");
+            if (slave_host) {
+                slave_host += 3;
+            } else {
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVE_VOL_PARSE_FAIL,
+                       "Invalid slave_host format!");
+                ret = -2;
+                goto out;
+            }
+            /* To go past username in non-root geo-rep session */
+            tmp = strchr(slave_host, '@');
+            if (tmp) {
+                if ((tmp - slave_host) >= LOGIN_NAME_MAX) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0,
+                           GD_MSG_SLAVE_VOL_PARSE_FAIL,
+                           "Invalid slave user length in %s", slave_host);
+                    ret = -2;
+                    goto out;
+                }
+                strncpy(slave_vol->old_slvuser, slave_host, (tmp - slave_host));
+                slave_vol->old_slvuser[(tmp - slave_host) + 1] = '\0';
+                slave_host = tmp + 1;
+            } else
+                strcpy(slave_vol->old_slvuser, "root");
+
+            tmp = strchr(slave_host, ':');
+            if (!tmp) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVE_VOL_PARSE_FAIL,
+                       "Invalid slave_host!");
+                ret = -2;
+                goto out;
+            }
+
+            strncpy(slave_vol->old_slvhost, slave_host, (tmp - slave_host));
+            slave_vol->old_slvhost[(tmp - slave_host) + 1] = '\0';
+
+            goto out;
+        }
+    }
+
+    ret = 0;
+out:
+    return ret;
+}
+
+/* Given slave host and slave volume, check whether slave volume uuid
+ * already present.
+ * If slave volume uuid is present, get corresponding slave host
+ * for reference */
+static int
+glusterd_get_slavehost_from_voluuid(glusterd_volinfo_t *volinfo,
+                                    char *slave_host, char *slave_vol,
+                                    struct slave_vol_config *slave1)
+{
+    int ret = -1;
+    xlator_t *this = NULL;
+
+    this = THIS;
+
+    GF_VALIDATE_OR_GOTO(this->name, volinfo, out);
+
+    ret = dict_foreach(volinfo->gsync_slaves, get_slavehost_from_voluuid,
+                       slave1);
+out:
+    return ret;
+}
+
+int
+glusterd_op_stage_gsync_create(dict_t *dict, char **op_errstr)
+{
+    char *down_peerstr = NULL;
+    char *slave = NULL;
+    char *volname = NULL;
+    char *host_uuid = NULL;
+    char *statefile = NULL;
+    char *slave_url = NULL;
+    char *slave_host = NULL;
+    char *slave_vol = NULL;
+    char *conf_path = NULL;
+    char errmsg[PATH_MAX] = "";
+    char common_pem_file[PATH_MAX] = "";
+    char hook_script[PATH_MAX] = "";
+    char uuid_str[64] = "";
+    int ret = -1;
+    int is_pem_push = -1;
+    int ssh_port = 22;
+    gf_boolean_t is_force = -1;
+    gf_boolean_t is_no_verify = -1;
+    gf_boolean_t is_force_blocker = -1;
+    gf_boolean_t is_template_in_use = _gf_false;
+    glusterd_conf_t *conf = NULL;
+    glusterd_volinfo_t *volinfo = NULL;
+    struct stat stbuf = {
+        0,
+    };
+    xlator_t *this = NULL;
+    struct slave_vol_config slave1 = {
+        {0},
+    };
+    char old_slave_url[SLAVE_URL_INFO_MAX] = {0};
+    char old_confpath[PATH_MAX] = {0};
+    gf_boolean_t is_running = _gf_false;
+    char *statedir = NULL;
+    char statefiledir[PATH_MAX] = {
+        0,
+    };
+    gf_boolean_t is_different_slavehost = _gf_false;
+    gf_boolean_t is_different_username = _gf_false;
+    char *slave_user = NULL;
+    char *save_ptr = NULL;
+    char *slave_url_buf = NULL;
+    int32_t len = 0;
+
+    this = THIS;
+    GF_ASSERT(this);
+    conf = this->private;
+    GF_ASSERT(conf);
+
+    ret = glusterd_op_gsync_args_get(dict, op_errstr, &volname, &slave,
+                                     &host_uuid);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_ARG_FETCH_ERROR,
+               "Unable to fetch arguments");
+        gf_msg_debug(this->name, 0, "Returning %d", ret);
+        return -1;
+    }
+
+    if (conf->op_version < 2) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_UNSUPPORTED_VERSION,
+               "Op Version not supported.");
+        snprintf(errmsg, sizeof(errmsg),
+                 "One or more nodes do not"
+                 " support the required op version.");
+        *op_errstr = gf_strdup(errmsg);
+        ret = -1;
+        goto out;
+    }
+
+    ret = glusterd_volinfo_find(volname, &volinfo);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_VOL_NOT_FOUND,
+               "volume name does not exist");
+        snprintf(errmsg, sizeof(errmsg),
+                 "Volume name %s does not"
+                 " exist",
+                 volname);
+        goto out;
+    }
+
+    ret = glusterd_get_slave_details_confpath(volinfo, dict, &slave_url,
+                                              &slave_host, &slave_vol,
+                                              &conf_path, op_errstr);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVEINFO_FETCH_ERROR,
+               "Unable to fetch slave or confpath details.");
+        ret = -1;
+        goto out;
+    }
+
+    is_force = dict_get_str_boolean(dict, "force", _gf_false);
+
+    uuid_utoa_r(MY_UUID, uuid_str);
+    if (!strcmp(uuid_str, host_uuid)) {
+        ret = glusterd_are_vol_all_peers_up(volinfo, &conf->peers,
+                                            &down_peerstr);
+        if ((ret == _gf_false) && !is_force) {
+            snprintf(errmsg, sizeof(errmsg),
+                     "Peer %s,"
+                     " which is a part of %s volume, is"
+                     " down. Please bring up the peer and"
+                     " retry.",
+                     down_peerstr, volinfo->volname);
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_PEER_DISCONNECTED, "%s",
+                   errmsg);
+            *op_errstr = gf_strdup(errmsg);
+            GF_FREE(down_peerstr);
+            down_peerstr = NULL;
+            gf_msg_debug(this->name, 0, "Returning %d", ret);
+            return -1;
+        } else if (ret == _gf_false) {
+            gf_msg(this->name, GF_LOG_INFO, 0, GD_MSG_PEER_DISCONNECTED,
+                   "Peer %s, which is a part of %s volume, is"
+                   " down. Force creating geo-rep session."
+                   " On bringing up the peer, re-run"
+                   " \"gluster system:: execute"
+                   " gsec_create\" and \"gluster volume"
+                   " geo-replication %s %s create push-pem"
+                   " force\"",
+                   down_peerstr, volinfo->volname, volinfo->volname, slave);
+            GF_FREE(down_peerstr);
+            down_peerstr = NULL;
+        }
+
+        ret = dict_get_int32(dict, "ssh_port", &ssh_port);
+        if (ret < 0 && ret != -ENOENT) {
+            snprintf(errmsg, sizeof(errmsg),
+                     "Fetching ssh_port failed while "
+                     "handling " GEOREP " options");
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED, "%s",
+                   errmsg);
+            goto out;
+        }
+
+        is_no_verify = dict_get_str_boolean(dict, "no_verify", _gf_false);
+
+        if (!is_no_verify) {
+            /* Checking if slave host is pingable, has proper passwordless
+             * ssh login setup, slave volume is created, slave vol is empty,
+             * and if it has enough memory and bypass in case of force if
+             * the error is not a force blocker */
+            ret = glusterd_verify_slave(volname, slave_url, slave_vol, ssh_port,
+                                        op_errstr, &is_force_blocker);
+            if (ret) {
+                if (is_force && !is_force_blocker) {
+                    gf_msg(this->name, GF_LOG_INFO, 0, GD_MSG_INVALID_SLAVE,
+                           "%s is not a valid slave "
+                           "volume. Error: %s. Force "
+                           "creating geo-rep"
+                           " session.",
+                           slave, *op_errstr);
+                } else {
+                    gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_INVALID_SLAVE,
+                           "%s is not a valid slave "
+                           "volume. Error: %s",
+                           slave, *op_errstr);
+                    ret = -1;
+
+                    goto out;
+                }
+            }
+        }
+
+        ret = dict_get_int32(dict, "push_pem", &is_pem_push);
+        if (!ret && is_pem_push) {
+            ret = snprintf(common_pem_file, sizeof(common_pem_file),
+                           "%s" GLUSTERD_COMMON_PEM_PUB_FILE, conf->workdir);
+            if ((ret < 0) || (ret >= sizeof(common_pem_file))) {
+                ret = -1;
+                goto out;
+            }
+
+            ret = snprintf(hook_script, sizeof(hook_script),
+                           "%s" GLUSTERD_CREATE_HOOK_SCRIPT, conf->workdir);
+            if ((ret < 0) || (ret >= sizeof(hook_script))) {
+                ret = -1;
+                goto out;
+            }
+
+            ret = sys_lstat(common_pem_file, &stbuf);
+            if (ret) {
+                len = snprintf(errmsg, sizeof(errmsg),
+                               "%s"
+                               " required for push-pem is"
+                               " not present. Please run"
+                               " \"gluster system:: execute"
+                               " gsec_create\"",
+                               common_pem_file);
+                if (len < 0) {
+                    strcpy(errmsg, "<error>");
+                }
+                gf_msg(this->name, GF_LOG_ERROR, ENOENT, GD_MSG_FILE_OP_FAILED,
+                       "%s", errmsg);
+                *op_errstr = gf_strdup(errmsg);
+                ret = -1;
+                goto out;
+            }
+
+            ret = sys_lstat(hook_script, &stbuf);
+            if (ret) {
+                len = snprintf(errmsg, sizeof(errmsg),
+                               "The hook-script (%s) "
+                               "required for push-pem is not "
+                               "present. Please install the "
+                               "hook-script and retry",
+                               hook_script);
+                if (len < 0) {
+                    strcpy(errmsg, "<error>");
+                }
+                gf_msg(this->name, GF_LOG_ERROR, ENOENT, GD_MSG_FILE_OP_FAILED,
+                       "%s", errmsg);
+                *op_errstr = gf_strdup(errmsg);
+                ret = -1;
+                goto out;
+            }
+
+            if (!S_ISREG(stbuf.st_mode)) {
+                len = snprintf(errmsg, sizeof(errmsg),
+                               "%s"
+                               " required for push-pem is"
+                               " not a regular file. Please"
+                               " run \"gluster system:: "
+                               "execute gsec_create\"",
+                               common_pem_file);
+                if (len < 0) {
+                    strcpy(errmsg, "<error>");
+                }
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REG_FILE_MISSING,
+                       "%s", errmsg);
+                ret = -1;
+                goto out;
+            }
+        }
+    }
+
+    ret = glusterd_get_statefile_name(volinfo, slave, conf_path, &statefile,
+                                      &is_template_in_use);
+    if (ret) {
+        if (!strstr(slave, "::"))
+            snprintf(errmsg, sizeof(errmsg), "%s is not a valid slave url.",
+                     slave);
+        else
+            snprintf(errmsg, sizeof(errmsg),
+                     "Please check gsync "
+                     "config file. Unable to get statefile's name");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_STATEFILE_NAME_NOT_FOUND,
+               "%s", errmsg);
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_set_str(dict, "statefile", statefile);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+               "Unable to store statefile path");
+        goto out;
+    }
+
+    if (snprintf(statefiledir, sizeof(statefiledir), "%s", statefile) >=
+        sizeof(statefiledir)) {
+        snprintf(errmsg, sizeof(errmsg), "Failed copying statefiledir");
+        goto out;
+    }
+    statedir = dirname(statefiledir);
+
+    ret = sys_lstat(statedir, &stbuf);
+    if (!ret && !is_force) {
+        snprintf(errmsg, sizeof(errmsg),
+                 "Session between %s"
+                 " and %s is already created.",
+                 volinfo->volname, slave);
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SESSION_ALREADY_EXIST, "%s",
+               errmsg);
+        ret = -1;
+        goto out;
+    } else if (!ret)
+        gf_msg(this->name, GF_LOG_INFO, 0, GD_MSG_FORCE_CREATE_SESSION,
+               "Session between %s and %s is already created. Force"
+               " creating again.",
+               volinfo->volname, slave);
+
+    ret = glusterd_get_slave_voluuid(slave_host, slave_vol,
+                                     slave1.slave_voluuid);
+    if ((ret) || (strlen(slave1.slave_voluuid) == 0)) {
+        snprintf(errmsg, sizeof(errmsg), "Unable to get remote volume uuid.");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REMOTE_VOL_UUID_FAIL, "%s",
+               errmsg);
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_set_dynstr_with_alloc(dict, "slave_voluuid",
+                                     slave1.slave_voluuid);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+               "Unable to set slave volume uuid in the dict");
+        goto out;
+    }
+
+    /* Check whether session is already created using slave volume uuid */
+    ret = glusterd_get_slavehost_from_voluuid(volinfo, slave_host, slave_vol,
+                                              &slave1);
+    if (ret == -1) {
+        if (!is_force) {
+            snprintf(errmsg, sizeof(errmsg),
+                     "Session between %s"
+                     " and %s:%s is already created! Cannot create "
+                     "with new slave:%s again!",
+                     volinfo->volname, slave1.old_slvhost, slave_vol,
+                     slave_host);
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_FORCE_CREATE_SESSION,
+                   "Session between"
+                   " %s and %s:%s is already created! "
+                   "Cannot create with new slave:%s again!",
+                   volinfo->volname, slave1.old_slvhost, slave_vol, slave_host);
+            goto out;
+        }
+
+        /* There is a remote possibility that slave_host can be NULL when
+           control reaches here. Add a check so we wouldn't crash in next
+           line */
+        if (!slave_host)
+            goto out;
+
+        /* Now, check whether session is already started.If so, warn!*/
+        is_different_slavehost = (strcmp(slave_host, slave1.old_slvhost) != 0)
+                                     ? _gf_true
+                                     : _gf_false;
+
+        if (strstr(slave_url, "@")) {
+            slave_url_buf = gf_strdup(slave_url);
+            if (!slave_url_buf) {
+                gf_msg(this->name, GF_LOG_ERROR, ENOMEM, GD_MSG_NO_MEMORY,
+                       "Unable to allocate memory");
+                ret = -1;
+                goto out;
+            }
+            slave_user = strtok_r(slave_url_buf, "@", &save_ptr);
+        } else
+            slave_user = "root";
+        is_different_username = (strcmp(slave_user, slave1.old_slvuser) != 0)
+                                    ? _gf_true
+                                    : _gf_false;
+
+        /* Do the check, only if different slave host/slave user */
+        if (is_different_slavehost || is_different_username) {
+            len = snprintf(old_confpath, sizeof(old_confpath),
+                           "%s/" GEOREP "/%s_%s_%s/gsyncd.conf", conf->workdir,
+                           volinfo->volname, slave1.old_slvhost, slave_vol);
+            if ((len < 0) || (len >= sizeof(old_confpath))) {
+                ret = -1;
+                goto out;
+            }
+
+            /* construct old slave url with (old) slave host */
+            len = snprintf(old_slave_url, sizeof(old_slave_url), "%s::%s",
+                           slave1.old_slvhost, slave_vol);
+            if ((len < 0) || (len >= sizeof(old_slave_url))) {
+                ret = -1;
+                goto out;
+            }
+
+            ret = glusterd_check_gsync_running_local(
+                volinfo->volname, old_slave_url, old_confpath, &is_running);
+            if (_gf_true == is_running) {
+                (void)snprintf(errmsg, sizeof(errmsg),
+                               "Geo"
+                               "-replication session between %s and %s"
+                               " is still active. Please stop the "
+                               "session and retry.",
+                               volinfo->volname, old_slave_url);
+                ret = -1;
+                goto out;
+            }
+        }
+
+        ret = dict_set_dynstr_with_alloc(dict, "old_slavehost",
+                                         slave1.old_slvhost);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+                   "Unable to set old_slavehost in the dict");
+            goto out;
+        }
+
+        ret = dict_set_int32(dict, "existing_session", _gf_true);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+                   "Unable to set existing_session in the dict");
+            goto out;
+        }
+    } else if (ret == -2) {
+        snprintf(errmsg, sizeof(errmsg),
+                 "get_slavehost_from_voluuid"
+                 " failed for %s::%s. Please check the glusterd logs.",
+                 slave_host, slave_vol);
+        gf_msg(this->name, GF_LOG_INFO, 0, GD_MSG_FORCE_CREATE_SESSION,
+               "get_slavehost_from_voluuid failed %s %s!!", slave_host,
+               slave_vol);
+        goto out;
+    }
+
+    ret = glusterd_verify_gsyncd_spawn(volinfo->volname, slave);
+    if (ret) {
+        snprintf(errmsg, sizeof(errmsg), "Unable to spawn gsyncd.");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GSYNCD_SPAWN_FAILED, "%s",
+               errmsg);
+        goto out;
+    }
+
+    ret = 0;
+out:
+
+    if (ret && errmsg[0] != '\0')
+        *op_errstr = gf_strdup(errmsg);
+
+    if (slave_url_buf)
+        GF_FREE(slave_url_buf);
+
+    return ret;
+}
+
+/* pre-condition check for geo-rep pause/resume.
+ * Return: 0 on success
+ *        -1 on any check failed.
+ */
+static int
+gd_pause_resume_validation(int type, glusterd_volinfo_t *volinfo, char *slave,
+                           char *statefile, char **op_errstr)
+{
+    int ret = 0;
+    char errmsg[PATH_MAX] = {
+        0,
+    };
+    char monitor_status[NAME_MAX] = {
+        0,
+    };
+
+    GF_ASSERT(volinfo);
+    GF_ASSERT(slave);
+    GF_ASSERT(statefile);
+    GF_ASSERT(op_errstr);
+
+    ret = glusterd_gsync_read_frm_status(statefile, monitor_status,
+                                         sizeof(monitor_status));
+    if (ret <= 0) {
+        snprintf(errmsg, sizeof(errmsg),
+                 "Pause check Failed:"
+                 " Geo-rep session is not setup");
+        ret = -1;
+        goto out;
+    }
+
+    if (type == GF_GSYNC_OPTION_TYPE_PAUSE &&
+        strstr(monitor_status, "Paused")) {
+        snprintf(errmsg, sizeof(errmsg),
+                 "Geo-replication"
+                 " session between %s and %s already Paused.",
+                 volinfo->volname, slave);
+        ret = -1;
+        goto out;
+    }
+    if (type == GF_GSYNC_OPTION_TYPE_RESUME &&
+        !strstr(monitor_status, "Paused")) {
+        snprintf(errmsg, sizeof(errmsg),
+                 "Geo-replication"
+                 " session between %s and %s is not Paused.",
+                 volinfo->volname, slave);
+        ret = -1;
+        goto out;
+    }
+    ret = 0;
+out:
+    if (ret && (errmsg[0] != '\0')) {
+        *op_errstr = gf_strdup(errmsg);
+    }
+    return ret;
+}
+
+int
+glusterd_op_stage_gsync_set(dict_t *dict, char **op_errstr)
+{
+    int ret = 0;
+    int type = 0;
+    char *volname = NULL;
+    char *slave = NULL;
+    char *slave_url = NULL;
+    char *slave_host = NULL;
+    char *slave_vol = NULL;
+    char *down_peerstr = NULL;
+    char *statefile = NULL;
+    char statefiledir[PATH_MAX] = {
+        0,
+    };
+    char *statedir = NULL;
+    char *path_list = NULL;
+    char *conf_path = NULL;
+    glusterd_volinfo_t *volinfo = NULL;
+    char errmsg[PATH_MAX] = {
+        0,
+    };
+    dict_t *ctx = NULL;
+    gf_boolean_t is_force = 0;
+    gf_boolean_t is_running = _gf_false;
+    gf_boolean_t is_template_in_use = _gf_false;
+    uuid_t uuid = {0};
+    char uuid_str[64] = {0};
+    char *host_uuid = NULL;
+    xlator_t *this = NULL;
+    glusterd_conf_t *conf = NULL;
+    struct stat stbuf = {
+        0,
+    };
+
+    this = THIS;
+    GF_ASSERT(this);
+    conf = this->private;
+    GF_ASSERT(conf);
+
+    ret = dict_get_int32(dict, "type", &type);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_DICT_GET_FAILED,
+               "command type not found");
+        *op_errstr = gf_strdup("command unsuccessful");
+        goto out;
+    }
+
+    if (type == GF_GSYNC_OPTION_TYPE_STATUS) {
+        ret = glusterd_verify_gsync_status_opts(dict, op_errstr);
+        goto out;
+    }
+
+    ret = glusterd_op_gsync_args_get(dict, op_errstr, &volname, &slave,
+                                     &host_uuid);
+    if (ret)
+        goto out;
+
+    uuid_utoa_r(MY_UUID, uuid_str);
+
+    if (conf->op_version < 2) {
+        snprintf(errmsg, sizeof(errmsg),
+                 "One or more nodes do not"
+                 " support the required op version.");
+        ret = -1;
+        goto out;
+    }
+
+    ret = glusterd_volinfo_find(volname, &volinfo);
+    if (ret) {
+        snprintf(errmsg, sizeof(errmsg),
+                 "Volume name %s does not"
+                 " exist",
+                 volname);
+        goto out;
+    }
+
+    ret = glusterd_get_slave_details_confpath(volinfo, dict, &slave_url,
+                                              &slave_host, &slave_vol,
+                                              &conf_path, op_errstr);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVEINFO_FETCH_ERROR,
+               "Unable to fetch slave or confpath details.");
+        ret = -1;
+        goto out;
+    }
+
+    is_force = dict_get_str_boolean(dict, "force", _gf_false);
+
+    ret = glusterd_get_statefile_name(volinfo, slave, conf_path, &statefile,
+                                      &is_template_in_use);
+    if (ret) {
+        if (!strstr(slave, "::")) {
+            snprintf(errmsg, sizeof(errmsg), "%s is not a valid slave url.",
+                     slave);
+            ret = -1;
+            goto out;
+        } else {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVE_URL_INVALID,
+                   "state_file entry missing in config file (%s)", conf_path);
+
+            if ((type == GF_GSYNC_OPTION_TYPE_STOP) && is_force) {
+                gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_STOP_FORCE,
+                       "Allowing stop "
+                       "force to bypass missing statefile "
+                       "entry in config file (%s), and "
+                       "template file",
+                       conf_path);
+                ret = 0;
+            } else
+                goto out;
+        }
+    } else {
+        ret = dict_set_str(dict, "statefile", statefile);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+                   "Unable to store statefile path");
+            goto out;
+        }
+    }
+
+    /* Allowing stop force to bypass the statefile check
+     * as this command acts as a fail safe method to stop geo-rep
+     * session. */
+    if (!((type == GF_GSYNC_OPTION_TYPE_STOP) && is_force)) {
+        /* check session directory as statefile may not present
+         * during upgrade */
+        if (snprintf(statefiledir, sizeof(statefiledir), "%s", statefile) >=
+            sizeof(statefiledir)) {
+            snprintf(errmsg, sizeof(errmsg), "Failed copying statefiledir");
+            ret = -1;
+            goto out;
+        }
+        statedir = dirname(statefiledir);
+
+        ret = sys_lstat(statedir, &stbuf);
+        if (ret) {
+            snprintf(errmsg, sizeof(errmsg),
+                     "Geo-replication"
+                     " session between %s and %s does not exist.",
+                     volinfo->volname, slave);
+            gf_msg(this->name, GF_LOG_ERROR, ENOENT, GD_MSG_FILE_OP_FAILED,
+                   "%s. statefile = %s", errmsg, statefile);
+            ret = -1;
+            goto out;
+        }
+    }
+
+    /* Check if all peers that are a part of the volume are up or not */
+    if ((type == GF_GSYNC_OPTION_TYPE_DELETE) ||
+        ((type == GF_GSYNC_OPTION_TYPE_STOP) && !is_force) ||
+        (type == GF_GSYNC_OPTION_TYPE_PAUSE) ||
+        (type == GF_GSYNC_OPTION_TYPE_RESUME)) {
+        if (!strcmp(uuid_str, host_uuid)) {
+            ret = glusterd_are_vol_all_peers_up(volinfo, &conf->peers,
+                                                &down_peerstr);
+            if (ret == _gf_false) {
+                snprintf(errmsg, sizeof(errmsg),
+                         "Peer %s,"
+                         " which is a part of %s volume, is"
+                         " down. Please bring up the peer and"
+                         " retry.",
+                         down_peerstr, volinfo->volname);
+                ret = -1;
+                GF_FREE(down_peerstr);
+                down_peerstr = NULL;
+                goto out;
+            }
+        }
+    }
+
+    switch (type) {
+        case GF_GSYNC_OPTION_TYPE_START:
+            if (is_template_in_use) {
+                snprintf(errmsg, sizeof(errmsg),
+                         "state-file entry "
+                         "missing in the config file(%s).",
+                         conf_path);
+                ret = -1;
+                goto out;
+            }
+
+            ret = glusterd_op_verify_gsync_start_options(
+                volinfo, slave, conf_path, statefile, op_errstr, is_force);
+            if (ret)
+                goto out;
+            ctx = glusterd_op_get_ctx();
+            if (ctx) {
+                /* gsyncd does a fuse mount to start
+                 * the geo-rep session */
+                if (!glusterd_is_fuse_available()) {
+                    gf_msg("glusterd", GF_LOG_ERROR, errno,
+                           GD_MSG_GEO_REP_START_FAILED,
+                           "Unable "
+                           "to open /dev/fuse (%s), "
+                           "geo-replication start failed",
+                           strerror(errno));
+                    snprintf(errmsg, sizeof(errmsg), "fuse unavailable");
+                    ret = -1;
+                    goto out;
+                }
+            }
+            break;
+
+        case GF_GSYNC_OPTION_TYPE_STOP:
+            if (!is_force) {
+                if (is_template_in_use) {
+                    snprintf(errmsg, sizeof(errmsg),
+                             "state-file entry missing in "
+                             "the config file(%s).",
+                             conf_path);
+                    ret = -1;
+                    goto out;
+                }
+
+                ret = glusterd_op_verify_gsync_running(volinfo, slave,
+                                                       conf_path, op_errstr);
+                if (ret) {
+                    ret = glusterd_get_local_brickpaths(volinfo, &path_list);
+                    if (!path_list && ret == -1)
+                        goto out;
+                }
+
+                /* Check for geo-rep session is active or not for
+                 * configured user.*/
+                ret = glusterd_gsync_get_uuid(slave, volinfo, uuid);
+                if (ret) {
+                    snprintf(errmsg, sizeof(errmsg),
+                             "Geo-replication session between %s "
+                             "and %s does not exist.",
+                             volinfo->volname, slave);
+                    ret = -1;
+                    goto out;
+                }
+            }
+            break;
+
+        case GF_GSYNC_OPTION_TYPE_PAUSE:
+        case GF_GSYNC_OPTION_TYPE_RESUME:
+            if (is_template_in_use) {
+                snprintf(errmsg, sizeof(errmsg),
+                         "state-file entry missing in "
+                         "the config file(%s).",
+                         conf_path);
+                ret = -1;
+                goto out;
+            }
+
+            ret = glusterd_op_verify_gsync_running(volinfo, slave, conf_path,
+                                                   op_errstr);
+            if (ret) {
+                ret = glusterd_get_local_brickpaths(volinfo, &path_list);
+                if (!path_list && ret == -1)
+                    goto out;
+            }
+
+            /* Check for geo-rep session is active or not
+             * for configured user.*/
+            ret = glusterd_gsync_get_uuid(slave, volinfo, uuid);
+            if (ret) {
+                snprintf(errmsg, sizeof(errmsg),
+                         "Geo-replication"
+                         " session between %s and %s does not exist.",
+                         volinfo->volname, slave);
+                ret = -1;
+                goto out;
+            }
+
+            if (!is_force) {
+                ret = gd_pause_resume_validation(type, volinfo, slave,
+                                                 statefile, op_errstr);
+                if (ret) {
+                    ret = glusterd_get_local_brickpaths(volinfo, &path_list);
+                    if (!path_list && ret == -1)
+                        goto out;
+                }
+            }
+            break;
+
+        case GF_GSYNC_OPTION_TYPE_CONFIG:
+            if (is_template_in_use) {
+                snprintf(errmsg, sizeof(errmsg),
+                         "state-file entry "
+                         "missing in the config file(%s).",
+                         conf_path);
+                ret = -1;
+                goto out;
+            }
+
+            ret = gsync_verify_config_options(dict, op_errstr, volname);
+            goto out;
+            break;
+
+        case GF_GSYNC_OPTION_TYPE_DELETE:
+            /* Check if the gsync session is still running
+             * If so ask the user to stop geo-replication first.*/
+            if (is_template_in_use) {
+                snprintf(errmsg, sizeof(errmsg),
+                         "state-file entry "
+                         "missing in the config file(%s).",
+                         conf_path);
+                ret = -1;
+                goto out;
+            }
+
+            ret = glusterd_gsync_get_uuid(slave, volinfo, uuid);
+            if (ret) {
+                snprintf(errmsg, sizeof(errmsg),
+                         "Geo-replication"
+                         " session between %s and %s does not exist.",
+                         volinfo->volname, slave);
+                ret = -1;
+                goto out;
+            } else {
+                ret = glusterd_check_gsync_running_local(
+                    volinfo->volname, slave, conf_path, &is_running);
+                if (_gf_true == is_running) {
+                    snprintf(errmsg, sizeof(errmsg),
+                             GEOREP
+                             " session between %s & %s is "
+                             "still active. Please stop the "
+                             "session and retry.",
+                             volinfo->volname, slave);
+                    ret = -1;
+                    goto out;
+                }
+            }
+
+            ret = glusterd_verify_gsyncd_spawn(volinfo->volname, slave);
+            if (ret) {
+                snprintf(errmsg, sizeof(errmsg), "Unable to spawn gsyncd");
+            }
+
+            break;
+    }
+
+out:
+
+    if (path_list)
+        GF_FREE(path_list);
+
+    if (ret && errmsg[0] != '\0') {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GSYNCD_ERROR, "%s", errmsg);
+        *op_errstr = gf_strdup(errmsg);
+    }
+
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+static int
+gd_pause_or_resume_gsync(dict_t *dict, char *master, char *slave,
+                         char *slave_host, char *slave_vol, char *conf_path,
+                         char **op_errstr, gf_boolean_t is_pause)
+{
+    int32_t ret = 0;
+    int pfd = -1;
+    long pid = 0;
+    char pidfile[PATH_MAX] = {
+        0,
+    };
+    char errmsg[PATH_MAX] = "";
+    char buf[4096] = {
+        0,
+    };
+    gf_boolean_t is_template_in_use = _gf_false;
+    char monitor_status[NAME_MAX] = {
+        0,
+    };
+    char *statefile = NULL;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+    GF_ASSERT(dict);
+    GF_ASSERT(master);
+    GF_ASSERT(slave);
+    GF_ASSERT(slave_host);
+    GF_ASSERT(slave_vol);
+    GF_ASSERT(conf_path);
+
+    pfd = gsyncd_getpidfile(master, slave, pidfile, conf_path,
+                            &is_template_in_use);
+    if (pfd == -2) {
+        snprintf(errmsg, sizeof(errmsg),
+                 "pid-file entry mising in config file and "
+                 "template config file.");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_PIDFILE_NOT_FOUND, "%s",
+               errmsg);
+        *op_errstr = gf_strdup(errmsg);
+        ret = -1;
+        goto out;
+    }
+
+    if (gsync_status_byfd(pfd) == -1) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GSYNCD_ERROR,
+               "gsyncd b/w %s & %s is not running", master, slave);
+        /* monitor gsyncd already dead */
+        goto out;
+    }
+
+    if (pfd < 0)
+        goto out;
+
+    /* Prepare to update status file*/
+    ret = dict_get_str(dict, "statefile", &statefile);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+               "Pause/Resume Failed: Unable to fetch statefile path");
+        goto out;
+    }
+    ret = glusterd_gsync_read_frm_status(statefile, monitor_status,
+                                         sizeof(monitor_status));
+    if (ret <= 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_STAT_FILE_READ_FAILED,
+               "Pause/Resume Failed: "
+               "Unable to read status file for %s(master)"
+               " %s(slave)",
+               master, slave);
+        goto out;
+    }
+
+    ret = sys_read(pfd, buf, sizeof(buf) - 1);
+    if (ret > 0) {
+        buf[ret] = '\0';
+        pid = strtol(buf, NULL, 10);
+        if (is_pause) {
+            ret = kill(-pid, SIGSTOP);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_PID_KILL_FAIL,
+                       "Failed"
+                       " to pause gsyncd. Error: %s",
+                       strerror(errno));
+                goto out;
+            }
+            /*On pause force, if status is already paused
+              do not update status again*/
+            if (strstr(monitor_status, "Paused"))
+                goto out;
+
+            ret = glusterd_create_status_file(master, slave, slave_host,
+                                              slave_vol, "Paused");
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, 0,
+                       GD_MSG_UPDATE_STATEFILE_FAILED,
+                       "Unable  to update state_file."
+                       " Error : %s",
+                       strerror(errno));
+                /* If status cannot be updated resume back */
+                if (kill(-pid, SIGCONT)) {
+                    snprintf(errmsg, sizeof(errmsg),
+                             "Pause successful but could "
+                             "not update status file. "
+                             "Please use 'resume force' to"
+                             " resume back and retry pause"
+                             " to reflect in status");
+                    gf_msg(this->name, GF_LOG_ERROR, errno,
+                           GD_MSG_PID_KILL_FAIL,
+                           "Resume back Failed. Error:"
+                           "%s",
+                           strerror(errno));
+                    *op_errstr = gf_strdup(errmsg);
+                }
+                goto out;
+            }
+        } else {
+            ret = glusterd_create_status_file(master, slave, slave_host,
+                                              slave_vol, "Started");
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, 0,
+                       GD_MSG_UPDATE_STATEFILE_FAILED,
+                       "Resume Failed: Unable to update "
+                       "state_file. Error : %s",
+                       strerror(errno));
+                goto out;
+            }
+            ret = kill(-pid, SIGCONT);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_PID_KILL_FAIL,
+                       "Resumed Failed: Unable to send"
+                       " SIGCONT. Error: %s",
+                       strerror(errno));
+                /* Process can't be resumed, update status
+                 * back to paused. */
+                ret = glusterd_create_status_file(master, slave, slave_host,
+                                                  slave_vol, monitor_status);
+                if (ret) {
+                    snprintf(errmsg, sizeof(errmsg),
+                             "Resume failed!!! Status "
+                             "inconsistent. Please use "
+                             "'resume force' to resume and"
+                             " reach consistent state");
+                    gf_msg(this->name, GF_LOG_ERROR, 0,
+                           GD_MSG_STATUS_UPDATE_FAILED,
+                           "Updating status back to paused"
+                           " Failed. Error: %s",
+                           strerror(errno));
+                    *op_errstr = gf_strdup(errmsg);
+                }
+                goto out;
+            }
+        }
+    }
+    ret = 0;
+
+out:
+    sys_close(pfd);
+    /* coverity[INTEGER_OVERFLOW] */
+    return ret;
+}
+
+static int
+stop_gsync(char *master, char *slave, char **msg, char *conf_path,
+           char **op_errstr, gf_boolean_t is_force)
+{
+    int32_t ret = 0;
+    int pfd = -1;
+    long pid = 0;
+    char pidfile[PATH_MAX] = {
+        0,
+    };
+    char errmsg[PATH_MAX] = "";
+    char buf[4096] = {
+        0,
+    };
+    int i = 0;
+    gf_boolean_t is_template_in_use = _gf_false;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(this->private);
+
+    pfd = gsyncd_getpidfile(master, slave, pidfile, conf_path,
+                            &is_template_in_use);
+    if (pfd == -2) {
+        snprintf(errmsg, sizeof(errmsg) - 1,
+                 "pid-file entry mising in config file and "
+                 "template config file.");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_PIDFILE_NOT_FOUND, "%s",
+               errmsg);
+        *op_errstr = gf_strdup(errmsg);
+        ret = -1;
+        goto out;
+    }
+    if (gsync_status_byfd(pfd) == -1 && !is_force) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GSYNCD_ERROR,
+               "gsyncd b/w %s & %s is not running", master, slave);
+        /* monitor gsyncd already dead */
+        goto out;
+    }
+
+    if (pfd < 0)
+        goto out;
+
+    ret = sys_read(pfd, buf, sizeof(buf) - 1);
+    if (ret > 0) {
+        buf[ret] = '\0';
+        pid = strtol(buf, NULL, 10);
+        ret = kill(-pid, SIGTERM);
+        if (ret && !is_force) {
+            gf_msg(this->name, GF_LOG_WARNING, errno, GD_MSG_PID_KILL_FAIL,
+                   "failed to kill gsyncd");
+            goto out;
+        }
+        for (i = 0; i < 20; i++) {
+            if (gsync_status_byfd(pfd) == -1) {
+                /* monitor gsyncd is dead but worker may
+                 * still be alive, give some more time
+                 * before SIGKILL (hack)
+                 */
+                gf_nanosleep(50000 * GF_US_IN_NS);
+                break;
+            }
+            gf_nanosleep(50000 * GF_US_IN_NS);
+        }
+        kill(-pid, SIGKILL);
+        sys_unlink(pidfile);
+    }
+    ret = 0;
+
+out:
+    sys_close(pfd);
+    /* coverity[INTEGER_OVERFLOW] */
+    return ret;
+}
+
+/*
+ * glusterd_gsync_op_already_set:
+ *      This function checks whether the op_value is same as in the
+ *      gsyncd.conf file.
+ *
+ * RETURN VALUE:
+ *      0 : op_value matches the conf file.
+ *      1 : op_value does not matches the conf file or op_param not
+ *          found in conf file.
+ *     -1 : error
+ */
+
+int
+glusterd_gsync_op_already_set(char *master, char *slave, char *conf_path,
+                              char *op_name, char *op_value)
+{
+    dict_t *confd = NULL;
+    char *op_val_buf = NULL;
+    int32_t op_val_conf = 0;
+    int32_t op_val_cli = 0;
+    int32_t ret = -1;
+    gf_boolean_t is_bool = _gf_true;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    confd = dict_new();
+    if (!confd) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_CREATE_FAIL,
+               "Not able to create dict.");
+        return -1;
+    }
+
+    ret = glusterd_gsync_get_config(master, slave, conf_path, confd);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GET_CONFIG_INFO_FAILED,
+               "Unable to get configuration data for %s(master), "
+               "%s(slave)",
+               master, slave);
+        goto out;
+    }
+
+    ret = dict_get_param(confd, op_name, &op_val_buf);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+               "Unable to get op_value for %s(master), %s(slave). "
+               "Please check gsync config file.",
+               master, slave);
+        ret = 1;
+        goto out;
+    }
+
+    gf_msg_debug(this->name, 0, "val_cli:%s  val_conf:%s", op_value,
+                 op_val_buf);
+
+    if (!strcmp(op_val_buf, "true") || !strcmp(op_val_buf, "1") ||
+        !strcmp(op_val_buf, "yes")) {
+        op_val_conf = 1;
+    } else if (!strcmp(op_val_buf, "false") || !strcmp(op_val_buf, "0") ||
+               !strcmp(op_val_buf, "no")) {
+        op_val_conf = 0;
+    } else {
+        is_bool = _gf_false;
+    }
+
+    if (is_bool) {
+        if (op_value && (!strcmp(op_value, "true") || !strcmp(op_value, "1") ||
+                         !strcmp(op_value, "yes"))) {
+            op_val_cli = 1;
+        } else {
+            op_val_cli = 0;
+        }
+
+        if (op_val_cli == op_val_conf) {
+            ret = 0;
+            goto out;
+        }
+    } else {
+        if (op_value && !strcmp(op_val_buf, op_value)) {
+            ret = 0;
+            goto out;
+        }
+    }
+
+    ret = 1;
+
+out:
+    dict_unref(confd);
+    return ret;
+}
+
+static int
+glusterd_gsync_configure(glusterd_volinfo_t *volinfo, char *slave,
+                         char *path_list, dict_t *dict, dict_t *resp_dict,
+                         char **op_errstr)
+{
+    int32_t ret = -1;
+    char *op_name = NULL;
+    char *op_value = NULL;
+    runner_t runner = {
+        0,
+    };
+    glusterd_conf_t *priv = NULL;
+    char *subop = NULL;
+    char *master = NULL;
+    char *conf_path = NULL;
+    char *slave_host = NULL;
+    char *slave_vol = NULL;
+    struct stat stbuf = {
+        0,
+    };
+    gf_boolean_t restart_required = _gf_true;
+    char **resopt = NULL;
+    gf_boolean_t op_already_set = _gf_false;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(slave);
+    GF_ASSERT(op_errstr);
+    GF_ASSERT(dict);
+    GF_ASSERT(resp_dict);
+
+    ret = dict_get_str(dict, "subop", &subop);
+    if (ret != 0)
+        goto out;
+
+    if (strcmp(subop, "get") == 0 || strcmp(subop, "get-all") == 0) {
+        /* deferred to cli */
+        gf_msg_debug(this->name, 0, "Returning 0");
+        return 0;
+    }
+
+    ret = dict_get_str(dict, "op_name", &op_name);
+    if (ret != 0)
+        goto out;
+
+    if (strtail(subop, "set")) {
+        ret = dict_get_str(dict, "op_value", &op_value);
+        if (ret != 0)
+            goto out;
+    }
+
+    priv = THIS->private;
+    if (priv == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GLUSTERD_PRIV_NOT_FOUND,
+               "priv of glusterd not present");
+        *op_errstr = gf_strdup("glusterd defunct");
+        goto out;
+    }
+
+    ret = dict_get_str(dict, "conf_path", &conf_path);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+               "Unable to fetch conf file path.");
+        goto out;
+    }
+
+    master = "";
+    runinit(&runner);
+    runner_add_args(&runner, GSYNCD_PREFIX "/gsyncd", "-c", NULL);
+    runner_argprintf(&runner, "%s", conf_path);
+    runner_argprintf(&runner, "--iprefix=%s", DATADIR);
+    if (volinfo) {
+        master = volinfo->volname;
+        runner_argprintf(&runner, ":%s", master);
+    }
+    runner_add_arg(&runner, slave);
+    runner_argprintf(&runner, "--config-%s", subop);
+    runner_add_arg(&runner, op_name);
+    if (op_value) {
+        runner_argprintf(&runner, "--value=%s", op_value);
+    }
+
+    if (strcmp(op_name, "checkpoint") != 0 && strtail(subop, "set")) {
+        ret = glusterd_gsync_op_already_set(master, slave, conf_path, op_name,
+                                            op_value);
+        if (ret == -1) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_GSYNCD_OP_SET_FAILED,
+                   "glusterd_gsync_op_already_set failed.");
+            gf_asprintf(op_errstr,
+                        GEOREP
+                        " config-%s failed for "
+                        "%s %s",
+                        subop, master, slave);
+            goto out;
+        }
+        if (ret == 0) {
+            gf_msg_debug(this->name, 0, "op_value is already set");
+            op_already_set = _gf_true;
+            goto out;
+        }
+    }
+
+    synclock_unlock(&priv->big_lock);
+    ret = runner_run(&runner);
+    synclock_lock(&priv->big_lock);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_GSYNCD_ERROR,
+               "gsyncd failed to %s %s option for "
+               "%s %s peers",
+               subop, op_name, master, slave);
+
+        gf_asprintf(op_errstr, GEOREP " config-%s failed for %s %s", subop,
+                    master, slave);
+
+        goto out;
+    }
+
+    if ((!strcmp(op_name, "state_file")) && (op_value)) {
+        ret = sys_lstat(op_value, &stbuf);
+        if (ret) {
+            ret = dict_get_str(dict, "slave_host", &slave_host);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+                       "Unable to fetch slave host.");
+                goto out;
+            }
+
+            ret = dict_get_str(dict, "slave_vol", &slave_vol);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+                       "Unable to fetch slave volume name.");
+                goto out;
+            }
+
+            ret = glusterd_create_status_file(volinfo->volname, slave,
+                                              slave_host, slave_vol,
+                                              "Switching Status "
+                                              "File");
+            if (ret || sys_lstat(op_value, &stbuf)) {
+                gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_FILE_OP_FAILED,
+                       "Unable to "
+                       "create %s. Error : %s",
+                       op_value, strerror(errno));
+                ret = -1;
+                goto out;
+            }
+        }
+    }
+
+    ret = 0;
+    gf_asprintf(op_errstr, "config-%s successful", subop);
+
+out:
+    if (!ret && volinfo && !op_already_set) {
+        for (resopt = gsync_no_restart_opts; *resopt; resopt++) {
+            restart_required = _gf_true;
+            if (!strcmp((*resopt), op_name)) {
+                restart_required = _gf_false;
+                break;
+            }
+        }
+
+        if (restart_required) {
+            ret = glusterd_check_restart_gsync_session(
+                volinfo, slave, resp_dict, path_list, conf_path, 0);
+            if (ret)
+                *op_errstr = gf_strdup("internal error");
+        }
+    }
+
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+int
+glusterd_gsync_read_frm_status(char *path, char *buf, size_t blen)
+{
+    int ret = 0;
+    int status_fd = -1;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(path);
+    GF_ASSERT(buf);
+    status_fd = open(path, O_RDONLY);
+    if (status_fd == -1) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_FILE_OP_FAILED,
+               "Unable to read gsyncd status file %s", path);
+        return -1;
+    }
+    ret = sys_read(status_fd, buf, blen - 1);
+    if (ret > 0) {
+        size_t len = strnlen(buf, ret);
+        /* Ensure there is a NUL byte and that it's not the first.  */
+        if (len == 0 || len == blen - 1) {
+            ret = -1;
+        } else {
+            char *p = buf + len - 1;
+            while (isspace(*p))
+                *p-- = '\0';
+        }
+    } else if (ret == 0)
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GSYNCD_ERROR,
+               "Status file of gsyncd is empty");
+    else /* ret < 0 */
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GSYNCD_ERROR,
+               "Status file of gsyncd is corrupt");
+
+    sys_close(status_fd);
+    return ret;
+}
+
+static int
+dict_get_param(dict_t *dict, char *key, char **param)
+{
+    char *dk = NULL;
+    char *s = NULL;
+    char x = '\0';
+    int ret = 0;
+
+    if (dict_get_str(dict, key, param) == 0)
+        return 0;
+
+    dk = gf_strdup(key);
+    if (!dk)
+        return -1;
+
+    s = strpbrk(dk, "-_");
+    if (!s) {
+        ret = -1;
+        goto out;
+    }
+    x = (*s == '-') ? '_' : '-';
+    *s++ = x;
+    while ((s = strpbrk(s, "-_")))
+        *s++ = x;
+
+    ret = dict_get_str(dict, dk, param);
+out:
+    GF_FREE(dk);
+    return ret;
+}
+
+int
+glusterd_fetch_values_from_config(char *master, char *slave, char *confpath,
+                                  dict_t *confd, char **statefile,
+                                  char **georep_session_wrkng_dir,
+                                  char **socketfile)
+{
+    int ret = 0;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    ret = glusterd_gsync_get_config(master, slave, confpath, confd);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GET_CONFIG_INFO_FAILED,
+               "Unable to get configuration data for %s(master), "
+               "%s(slave)",
+               master, slave);
+        goto out;
+    }
+
+    if (statefile) {
+        ret = dict_get_param(confd, "state_file", statefile);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+                   "Unable to get state_file's name "
+                   "for %s(master), %s(slave). "
+                   "Please check gsync config file.",
+                   master, slave);
+            goto out;
+        }
+    }
+
+    if (georep_session_wrkng_dir) {
+        ret = dict_get_param(confd, "georep_session_working_dir",
+                             georep_session_wrkng_dir);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+                   "Unable to get geo-rep session's "
+                   "working directory name for %s(master), "
+                   "%s(slave). Please check gsync config file.",
+                   master, slave);
+            goto out;
+        }
+    }
+
+    if (socketfile) {
+        ret = dict_get_param(confd, "state_socket_unencoded", socketfile);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+                   "Unable to get socket file's name "
+                   "for %s(master), %s(slave). "
+                   "Please check gsync config file.",
+                   master, slave);
+            goto out;
+        }
+    }
+
+    ret = 0;
+out:
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+int
+glusterd_read_status_file(glusterd_volinfo_t *volinfo, char *slave,
+                          char *conf_path, dict_t *dict, char *node)
+{
+    char temp_conf_path[PATH_MAX] = "";
+    char *working_conf_path = NULL;
+    char *georep_session_wrkng_dir = NULL;
+    char *master = NULL;
+    char sts_val_name[1024] = "";
+    char monitor_status[NAME_MAX] = "";
+    char *statefile = NULL;
+    char *socketfile = NULL;
+    dict_t *confd = NULL;
+    char *slavekey = NULL;
+    char *slaveentry = NULL;
+    char *slaveuser = NULL;
+    char *saveptr = NULL;
+    char *temp = NULL;
+    char *temp_inp = NULL;
+    char *brick_host_uuid = NULL;
+    int brick_host_uuid_length = 0;
+    int gsync_count = 0;
+    int ret = 0;
+    glusterd_brickinfo_t *brickinfo = NULL;
+    gf_gsync_status_t *sts_val = NULL;
+    gf_boolean_t is_template_in_use = _gf_false;
+    glusterd_conf_t *priv = NULL;
+    struct stat stbuf = {
+        0,
+    };
+    xlator_t *this = NULL;
+    int32_t len = 0;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(this->private);
+    GF_ASSERT(volinfo);
+    GF_ASSERT(conf_path);
+
+    master = volinfo->volname;
+
+    confd = dict_new();
+    if (!confd) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_CREATE_FAIL,
+               "Not able to create dict.");
+        return -1;
+    }
+
+    priv = THIS->private;
+
+    len = snprintf(temp_conf_path, sizeof(temp_conf_path),
+                   "%s/" GSYNC_CONF_TEMPLATE, priv->workdir);
+    if ((len < 0) || (len >= sizeof(temp_conf_path))) {
+        return -1;
+    }
+
+    ret = sys_lstat(conf_path, &stbuf);
+    if (!ret) {
+        gf_msg(this->name, GF_LOG_INFO, 0, GD_MSG_CONFIG_INFO,
+               "Using passed config template(%s).", conf_path);
+        working_conf_path = conf_path;
+    } else {
+        gf_msg(this->name, GF_LOG_WARNING, ENOENT, GD_MSG_FILE_OP_FAILED,
+               "Config file (%s) missing. Looking for template "
+               "config file (%s)",
+               conf_path, temp_conf_path);
+        ret = sys_lstat(temp_conf_path, &stbuf);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOENT, GD_MSG_FILE_OP_FAILED,
+                   "Template "
+                   "config file (%s) missing.",
+                   temp_conf_path);
+            goto out;
+        }
+        gf_msg(this->name, GF_LOG_INFO, 0, GD_MSG_DEFAULT_TEMP_CONFIG,
+               "Using default config template(%s).", temp_conf_path);
+        working_conf_path = temp_conf_path;
+        is_template_in_use = _gf_true;
+    }
+
+fetch_data:
+    ret = glusterd_fetch_values_from_config(
+        master, slave, working_conf_path, confd, &statefile,
+        &georep_session_wrkng_dir, &socketfile);
+    if (ret) {
+        if (is_template_in_use == _gf_false) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_FETCH_CONFIG_VAL_FAILED,
+                   "Unable to fetch config values "
+                   "for %s(master), %s(slave). "
+                   "Trying default config template",
+                   master, slave);
+            working_conf_path = temp_conf_path;
+            is_template_in_use = _gf_true;
+            goto fetch_data;
+        } else {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_FETCH_CONFIG_VAL_FAILED,
+                   "Unable to "
+                   "fetch config values for %s(master), "
+                   "%s(slave)",
+                   master, slave);
+            goto out;
+        }
+    }
+
+    ret = glusterd_gsync_read_frm_status(statefile, monitor_status,
+                                         sizeof(monitor_status));
+    if (ret <= 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_STAT_FILE_READ_FAILED,
+               "Unable to read the status file for %s(master), "
+               "%s(slave) statefile: %s",
+               master, slave, statefile);
+        snprintf(monitor_status, sizeof(monitor_status), "defunct");
+    }
+
+    ret = dict_get_int32(dict, "gsync-count", &gsync_count);
+    if (ret)
+        gsync_count = 0;
+
+    cds_list_for_each_entry(brickinfo, &volinfo->bricks, brick_list)
+    {
+        if (gf_uuid_compare(brickinfo->uuid, MY_UUID))
+            continue;
+
+        sts_val = GF_CALLOC(1, sizeof(gf_gsync_status_t),
+                            gf_common_mt_gsync_status_t);
+        if (!sts_val) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, GD_MSG_NO_MEMORY,
+                   "Out Of Memory");
+            goto out;
+        }
+
+        /* Slave Key */
+        ret = glusterd_get_slave(volinfo, slave, &slavekey);
+        if (ret < 0) {
+            GF_FREE(sts_val);
+            goto out;
+        }
+        memcpy(sts_val->slavekey, slavekey, strlen(slavekey));
+        sts_val->slavekey[strlen(slavekey)] = '\0';
+
+        /* Master Volume */
+        memcpy(sts_val->master, master, strlen(master));
+        sts_val->master[strlen(master)] = '\0';
+
+        /* Master Brick Node */
+        memcpy(sts_val->node, brickinfo->hostname, strlen(brickinfo->hostname));
+        sts_val->node[strlen(brickinfo->hostname)] = '\0';
+
+        /* Master Brick Path */
+        memcpy(sts_val->brick, brickinfo->path, strlen(brickinfo->path));
+        sts_val->brick[strlen(brickinfo->path)] = '\0';
+
+        /* Brick Host UUID */
+        brick_host_uuid = uuid_utoa(brickinfo->uuid);
+        brick_host_uuid_length = strlen(brick_host_uuid);
+        memcpy(sts_val->brick_host_uuid, brick_host_uuid,
+               brick_host_uuid_length);
+        sts_val->brick_host_uuid[brick_host_uuid_length] = '\0';
+
+        /* Slave */
+        memcpy(sts_val->slave, slave, strlen(slave));
+        sts_val->slave[strlen(slave)] = '\0';
+
+        snprintf(sts_val->slave_node, sizeof(sts_val->slave_node), "N/A");
+
+        snprintf(sts_val->worker_status, sizeof(sts_val->worker_status), "N/A");
+
+        snprintf(sts_val->crawl_status, sizeof(sts_val->crawl_status), "N/A");
+
+        snprintf(sts_val->last_synced, sizeof(sts_val->last_synced), "N/A");
+
+        snprintf(sts_val->last_synced_utc, sizeof(sts_val->last_synced_utc),
+                 "N/A");
+
+        snprintf(sts_val->entry, sizeof(sts_val->entry), "N/A");
+
+        snprintf(sts_val->data, sizeof(sts_val->data), "N/A");
+
+        snprintf(sts_val->meta, sizeof(sts_val->meta), "N/A");
+
+        snprintf(sts_val->failures, sizeof(sts_val->failures), "N/A");
+
+        snprintf(sts_val->checkpoint_time, sizeof(sts_val->checkpoint_time),
+                 "N/A");
+
+        snprintf(sts_val->checkpoint_time_utc,
+                 sizeof(sts_val->checkpoint_time_utc), "N/A");
+
+        snprintf(sts_val->checkpoint_completed,
+                 sizeof(sts_val->checkpoint_completed), "N/A");
+
+        snprintf(sts_val->checkpoint_completion_time,
+                 sizeof(sts_val->checkpoint_completion_time), "N/A");
+
+        snprintf(sts_val->checkpoint_completion_time_utc,
+                 sizeof(sts_val->checkpoint_completion_time_utc), "N/A");
+
+        /* Get all the other values from Gsyncd */
+        ret = glusterd_gsync_get_status(master, slave, conf_path,
+                                        brickinfo->path, sts_val);
+
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GET_STATUS_DATA_FAIL,
+                   "Unable to get status data "
+                   "for %s(master), %s(slave), %s(brick)",
+                   master, slave, brickinfo->path);
+            ret = -1;
+            goto out;
+        }
+
+        if (is_template_in_use) {
+            snprintf(sts_val->worker_status, sizeof(sts_val->worker_status),
+                     "Config Corrupted");
+        }
+
+        ret = dict_get_str(volinfo->gsync_slaves, slavekey, &slaveentry);
+        if (ret < 0) {
+            GF_FREE(sts_val);
+            goto out;
+        }
+
+        memcpy(sts_val->session_slave, slaveentry, strlen(slaveentry));
+        sts_val->session_slave[strlen(slaveentry)] = '\0';
+
+        temp_inp = gf_strdup(slaveentry);
+        if (!temp_inp)
+            goto out;
+
+        if (strstr(temp_inp, "@") == NULL) {
+            slaveuser = "root";
+        } else {
+            temp = strtok_r(temp_inp, "//", &saveptr);
+            temp = strtok_r(NULL, "/", &saveptr);
+            slaveuser = strtok_r(temp, "@", &saveptr);
+        }
+        memcpy(sts_val->slave_user, slaveuser, strlen(slaveuser));
+        sts_val->slave_user[strlen(slaveuser)] = '\0';
+
+        snprintf(sts_val_name, sizeof(sts_val_name), "status_value%d",
+                 gsync_count);
+        ret = dict_set_bin(dict, sts_val_name, sts_val,
+                           sizeof(gf_gsync_status_t));
+        if (ret) {
+            GF_FREE(sts_val);
+            goto out;
+        }
+
+        gsync_count++;
+        sts_val = NULL;
+    }
+
+    ret = dict_set_int32(dict, "gsync-count", gsync_count);
+    if (ret)
+        goto out;
+
+out:
+    GF_FREE(temp_inp);
+    dict_unref(confd);
+
+    return 0;
+}
+
+int
+glusterd_check_restart_gsync_session(glusterd_volinfo_t *volinfo, char *slave,
+                                     dict_t *resp_dict, char *path_list,
+                                     char *conf_path, gf_boolean_t is_force)
+{
+    int ret = 0;
+    glusterd_conf_t *priv = NULL;
+    char *status_msg = NULL;
+    gf_boolean_t is_running = _gf_false;
+    char *op_errstr = NULL;
+    char *key = NULL;
+    xlator_t *this = NULL;
+
+    GF_ASSERT(volinfo);
+    GF_ASSERT(slave);
+    this = THIS;
+    GF_ASSERT(this);
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    key = slave;
+
+    ret = glusterd_check_gsync_running_local(volinfo->volname, slave, conf_path,
+                                             &is_running);
+    if (!ret && (_gf_true != is_running))
+        /* gsynd not running, nothing to do */
+        goto out;
+
+    ret = stop_gsync(volinfo->volname, slave, &status_msg, conf_path,
+                     &op_errstr, is_force);
+    if (ret == 0 && status_msg)
+        ret = dict_set_str(resp_dict, "gsync-status", status_msg);
+    if (ret == 0) {
+        dict_del(volinfo->gsync_active_slaves, key);
+        ret = glusterd_start_gsync(volinfo, slave, path_list, conf_path,
+                                   uuid_utoa(MY_UUID), NULL, _gf_false);
+        if (!ret) {
+            /* Add slave to the dict indicating geo-rep session is
+             * running.*/
+            ret = dict_set_dynstr_with_alloc(volinfo->gsync_active_slaves, key,
+                                             "running");
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+                       "Unable to set"
+                       " key:%s value:running in dict. But "
+                       "the config succeeded.",
+                       key);
+                goto out;
+            }
+        }
+    }
+
+out:
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    if (op_errstr)
+        GF_FREE(op_errstr);
+    return ret;
+}
+
+static int32_t
+glusterd_marker_changelog_create_volfile(glusterd_volinfo_t *volinfo)
+{
+    int32_t ret = 0;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    ret = glusterd_create_volfiles_and_notify_services(volinfo);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_VOLFILE_CREATE_FAIL,
+               "Unable to create volfile for setting of marker "
+               "while '" GEOREP " start'");
+        ret = -1;
+        goto out;
+    }
+
+    ret = glusterd_store_volinfo(volinfo, GLUSTERD_VOLINFO_VER_AC_INCREMENT);
+    if (ret)
+        goto out;
+
+    if (GLUSTERD_STATUS_STARTED == volinfo->status) {
+        ret = glusterd_svcs_manager(volinfo);
+        goto out;
+    }
+    ret = 0;
+out:
+    return ret;
+}
+
+static int
+glusterd_set_gsync_knob(glusterd_volinfo_t *volinfo, char *key, int *vc)
+{
+    int ret = -1;
+    int conf_enabled = _gf_false;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(this->private);
+
+    conf_enabled = glusterd_volinfo_get_boolean(volinfo, key);
+    if (conf_enabled == -1) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GET_KEY_FAILED,
+               "failed to get key %s from volinfo", key);
+        goto out;
+    }
+
+    ret = 0;
+    if (conf_enabled == _gf_false) {
+        *vc = 1;
+        ret = glusterd_gsync_volinfo_dict_set(volinfo, key, "on");
+    }
+
+out:
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+static int
+glusterd_set_gsync_confs(glusterd_volinfo_t *volinfo)
+{
+    int ret = -1;
+    int volfile_changed = 0;
+
+    ret = glusterd_set_gsync_knob(volinfo, VKEY_MARKER_XTIME, &volfile_changed);
+    if (ret)
+        goto out;
+
+    /**
+     * enable ignore-pid-check blindly as it could be needed for
+     * cascading setups.
+     */
+    ret = glusterd_set_gsync_knob(volinfo, VKEY_MARKER_XTIME_FORCE,
+                                  &volfile_changed);
+    if (ret)
+        goto out;
+
+    ret = glusterd_set_gsync_knob(volinfo, VKEY_CHANGELOG, &volfile_changed);
+    if (ret)
+        goto out;
+
+    if (volfile_changed)
+        ret = glusterd_marker_changelog_create_volfile(volinfo);
+
+out:
+    return ret;
+}
+
+static int
+glusterd_get_gsync_status_mst_slv(glusterd_volinfo_t *volinfo, char *slave,
+                                  char *conf_path, dict_t *rsp_dict, char *node)
+{
+    char *statefile = NULL;
+    uuid_t uuid = {
+        0,
+    };
+    int ret = 0;
+    gf_boolean_t is_template_in_use = _gf_false;
+    struct stat stbuf = {
+        0,
+    };
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(volinfo);
+    GF_ASSERT(slave);
+    GF_ASSERT(this->private);
+
+    ret = glusterd_gsync_get_uuid(slave, volinfo, uuid);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_INFO, 0, GD_MSG_SESSION_INACTIVE,
+               "geo-replication status %s %s : session is not "
+               "active",
+               volinfo->volname, slave);
+
+        ret = glusterd_get_statefile_name(volinfo, slave, conf_path, &statefile,
+                                          &is_template_in_use);
+        if (ret) {
+            if (!strstr(slave, "::"))
+                gf_msg(this->name, GF_LOG_INFO, 0, GD_MSG_SLAVE_URL_INVALID,
+                       "%s is not a valid slave url.", slave);
+            else
+                gf_msg(this->name, GF_LOG_INFO, 0,
+                       GD_MSG_GET_STATEFILE_NAME_FAILED,
+                       "Unable to get statefile's name");
+            ret = 0;
+            goto out;
+        }
+
+        ret = sys_lstat(statefile, &stbuf);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_INFO, ENOENT, GD_MSG_FILE_OP_FAILED,
+                   "%s statefile not present.", statefile);
+            ret = 0;
+            goto out;
+        }
+    }
+
+    ret = glusterd_read_status_file(volinfo, slave, conf_path, rsp_dict, node);
+out:
+    if (statefile)
+        GF_FREE(statefile);
+
+    gf_msg_debug(this->name, 0, "Returning with %d", ret);
+    return ret;
+}
+
+int
+glusterd_get_gsync_status_mst(glusterd_volinfo_t *volinfo, dict_t *rsp_dict,
+                              char *node)
+{
+    glusterd_gsync_status_temp_t param = {
+        0,
+    };
+
+    GF_ASSERT(volinfo);
+
+    param.rsp_dict = rsp_dict;
+    param.volinfo = volinfo;
+    param.node = node;
+    dict_foreach(volinfo->gsync_slaves, _get_status_mst_slv, &param);
+
+    return 0;
+}
+
+static int
+glusterd_get_gsync_status_all(dict_t *rsp_dict, char *node)
+{
+    int32_t ret = 0;
+    glusterd_conf_t *priv = NULL;
+    glusterd_volinfo_t *volinfo = NULL;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    priv = this->private;
+
+    GF_ASSERT(priv);
+
+    cds_list_for_each_entry(volinfo, &priv->volumes, vol_list)
+    {
+        ret = glusterd_get_gsync_status_mst(volinfo, rsp_dict, node);
+        if (ret)
+            goto out;
+    }
+
+out:
+    gf_msg_debug(this->name, 0, "Returning with %d", ret);
+    return ret;
+}
+
+static int
+glusterd_get_gsync_status(dict_t *dict, char **op_errstr, dict_t *rsp_dict)
+{
+    char *slave = NULL;
+    char *volname = NULL;
+    char *conf_path = NULL;
+    char errmsg[PATH_MAX] = {
+        0,
+    };
+    glusterd_volinfo_t *volinfo = NULL;
+    int ret = 0;
+    char my_hostname[256] = {
+        0,
+    };
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    ret = gethostname(my_hostname, 256);
+    if (ret) {
+        /* stick to N/A */
+        (void)strcpy(my_hostname, "N/A");
+    }
+
+    ret = dict_get_str(dict, "master", &volname);
+    if (ret < 0) {
+        ret = glusterd_get_gsync_status_all(rsp_dict, my_hostname);
+        goto out;
+    }
+
+    ret = glusterd_volinfo_find(volname, &volinfo);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_VOL_NOT_FOUND,
+               "volume name does not exist");
+        snprintf(errmsg, sizeof(errmsg),
+                 "Volume name %s does not"
+                 " exist",
+                 volname);
+        *op_errstr = gf_strdup(errmsg);
+        goto out;
+    }
+
+    ret = dict_get_str(dict, "slave", &slave);
+    if (ret < 0) {
+        ret = glusterd_get_gsync_status_mst(volinfo, rsp_dict, my_hostname);
+        goto out;
+    }
+
+    ret = dict_get_str(dict, "conf_path", &conf_path);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+               "Unable to fetch conf file path.");
+        goto out;
+    }
+
+    ret = glusterd_get_gsync_status_mst_slv(volinfo, slave, conf_path, rsp_dict,
+                                            my_hostname);
+
+out:
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+static int
+glusterd_gsync_delete(glusterd_volinfo_t *volinfo, char *slave,
+                      char *slave_host, char *slave_vol, char *path_list,
+                      dict_t *dict, dict_t *resp_dict, char **op_errstr)
+{
+    int32_t ret = -1;
+    runner_t runner = {
+        0,
+    };
+    glusterd_conf_t *priv = NULL;
+    char *master = NULL;
+    char *gl_workdir = NULL;
+    char geo_rep_dir[PATH_MAX] = "";
+    char *conf_path = NULL;
+    xlator_t *this = NULL;
+    uint32_t reset_sync_time = _gf_false;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(slave);
+    GF_ASSERT(slave_host);
+    GF_ASSERT(slave_vol);
+    GF_ASSERT(op_errstr);
+    GF_ASSERT(dict);
+    GF_ASSERT(resp_dict);
+
+    if (THIS)
+        priv = THIS->private;
+    if (priv == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GLUSTERD_PRIV_NOT_FOUND,
+               "priv of glusterd not present");
+        *op_errstr = gf_strdup("glusterd defunct");
+        goto out;
+    }
+
+    ret = dict_get_str(dict, "conf_path", &conf_path);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+               "Unable to fetch conf file path.");
+        goto out;
+    }
+
+    gl_workdir = priv->workdir;
+    master = "";
+    runinit(&runner);
+    runner_add_args(&runner, GSYNCD_PREFIX "/gsyncd", "--delete", "-c", NULL);
+    runner_argprintf(&runner, "%s", conf_path);
+    runner_argprintf(&runner, "--iprefix=%s", DATADIR);
+
+    runner_argprintf(&runner, "--path-list=%s", path_list);
+
+    ret = dict_get_uint32(dict, "reset-sync-time", &reset_sync_time);
+    if (!ret && reset_sync_time) {
+        runner_add_args(&runner, "--reset-sync-time", NULL);
+    }
+
+    if (volinfo) {
+        master = volinfo->volname;
+        runner_argprintf(&runner, ":%s", master);
+    }
+    runner_add_arg(&runner, slave);
+    runner_redir(&runner, STDOUT_FILENO, RUN_PIPE);
+    synclock_unlock(&priv->big_lock);
+    ret = runner_run(&runner);
+    synclock_lock(&priv->big_lock);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SESSION_DEL_FAILED,
+               "gsyncd failed to delete session info for %s and "
+               "%s peers",
+               master, slave);
+
+        gf_asprintf(op_errstr,
+                    "gsyncd failed to "
+                    "delete session info for %s and %s peers",
+                    master, slave);
+
+        goto out;
+    }
+
+    ret = snprintf(geo_rep_dir, sizeof(geo_rep_dir) - 1,
+                   "%s/" GEOREP "/%s_%s_%s", gl_workdir, volinfo->volname,
+                   slave_host, slave_vol);
+    geo_rep_dir[ret] = '\0';
+
+    ret = sys_rmdir(geo_rep_dir);
+    if (ret) {
+        if (errno == ENOENT)
+            gf_msg_debug(this->name, 0, "Geo Rep Dir(%s) Not Present.",
+                         geo_rep_dir);
+        else {
+            gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DIR_OP_FAILED,
+                   "Unable to delete Geo Rep Dir(%s). Error: %s", geo_rep_dir,
+                   strerror(errno));
+            goto out;
+        }
+    }
+
+    ret = 0;
+
+    gf_asprintf(op_errstr, "delete successful");
+
+out:
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+int
+glusterd_op_sys_exec(dict_t *dict, char **op_errstr, dict_t *rsp_dict)
+{
+    char buf[PATH_MAX] = "";
+    char cmd_arg_name[PATH_MAX] = "";
+    char output_name[PATH_MAX] = "";
+    char errmsg[PATH_MAX] = "";
+    char *ptr = NULL;
+    char *bufp = NULL;
+    char *command = NULL;
+    char **cmd_args = NULL;
+    int ret = -1;
+    int i = -1;
+    int cmd_args_count = 0;
+    int output_count = 0;
+    glusterd_conf_t *priv = NULL;
+    runner_t runner = {
+        0,
+    };
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(dict);
+    GF_ASSERT(op_errstr);
+    GF_ASSERT(rsp_dict);
+
+    if (THIS)
+        priv = THIS->private;
+    if (priv == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GLUSTERD_PRIV_NOT_FOUND,
+               "priv of glusterd not present");
+        *op_errstr = gf_strdup("glusterd defunct");
+        goto out;
+    }
+
+    ret = dict_get_str(dict, "command", &command);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+               "Unable to get command from dict");
+        goto out;
+    }
+
+    ret = dict_get_int32(dict, "cmd_args_count", &cmd_args_count);
+    if (ret)
+        gf_msg(this->name, GF_LOG_INFO, 0, GD_MSG_DICT_GET_FAILED,
+               "No cmd_args_count");
+
+    if (cmd_args_count) {
+        cmd_args = GF_CALLOC(cmd_args_count, sizeof(char *), gf_common_mt_char);
+        if (!cmd_args) {
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, GD_MSG_NO_MEMORY,
+                   "Unable to calloc. Errno = %s", strerror(errno));
+            goto out;
+        }
+
+        for (i = 1; i <= cmd_args_count; i++) {
+            snprintf(cmd_arg_name, sizeof(cmd_arg_name), "cmd_arg_%d", i);
+            ret = dict_get_str(dict, cmd_arg_name, &cmd_args[i - 1]);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+                       "Unable to get"
+                       " %s in dict",
+                       cmd_arg_name);
+                goto out;
+            }
+        }
+    }
+
+    runinit(&runner);
+    runner_argprintf(&runner, GSYNCD_PREFIX "/peer_%s", command);
+    for (i = 0; i < cmd_args_count; i++)
+        runner_add_arg(&runner, cmd_args[i]);
+    runner_redir(&runner, STDOUT_FILENO, RUN_PIPE);
+    synclock_unlock(&priv->big_lock);
+    ret = runner_start(&runner);
+    if (ret == -1) {
+        snprintf(errmsg, sizeof(errmsg),
+                 "Unable to "
+                 "execute command. Error : %s",
+                 strerror(errno));
+        *op_errstr = gf_strdup(errmsg);
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_CMD_EXEC_FAIL, "%s", errmsg);
+        ret = -1;
+        synclock_lock(&priv->big_lock);
+        goto out;
+    }
+
+    do {
+        ptr = fgets(buf, sizeof(buf), runner_chio(&runner, STDOUT_FILENO));
+        if (ptr) {
+            ret = dict_get_int32(rsp_dict, "output_count", &output_count);
+            if (ret)
+                output_count = 1;
+            else
+                output_count++;
+            snprintf(output_name, sizeof(output_name), "output_%d",
+                     output_count);
+            if (buf[strlen(buf) - 1] == '\n')
+                buf[strlen(buf) - 1] = '\0';
+            bufp = gf_strdup(buf);
+            if (!bufp)
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_STRDUP_FAILED,
+                       "gf_strdup failed.");
+            ret = dict_set_dynstr(rsp_dict, output_name, bufp);
+            if (ret) {
+                GF_FREE(bufp);
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+                       "output set "
+                       "failed.");
+            }
+            ret = dict_set_int32(rsp_dict, "output_count", output_count);
+            if (ret)
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+                       "output_count "
+                       "set failed.");
+        }
+    } while (ptr);
+
+    ret = runner_end(&runner);
+    if (ret) {
+        snprintf(errmsg, sizeof(errmsg),
+                 "Unable to "
+                 "end. Error : %s",
+                 strerror(errno));
+        *op_errstr = gf_strdup(errmsg);
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_UNABLE_TO_END, "%s", errmsg);
+        ret = -1;
+        synclock_lock(&priv->big_lock);
+        goto out;
+    }
+    synclock_lock(&priv->big_lock);
+
+    ret = 0;
+out:
+    if (cmd_args) {
+        GF_FREE(cmd_args);
+        cmd_args = NULL;
+    }
+
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+int
+glusterd_op_copy_file(dict_t *dict, char **op_errstr)
+{
+    char abs_filename[PATH_MAX] = "";
+    char errmsg[PATH_MAX] = "";
+    char *filename = NULL;
+    char *host_uuid = NULL;
+    char uuid_str[64] = {0};
+    char *contents = NULL;
+    char buf[4096] = "";
+    int ret = -1;
+    int fd = -1;
+    int bytes_writen = 0;
+    int bytes_read = 0;
+    int contents_size = -1;
+    int file_mode = -1;
+    glusterd_conf_t *priv = NULL;
+    struct stat stbuf = {
+        0,
+    };
+    gf_boolean_t free_contents = _gf_true;
+    xlator_t *this = NULL;
+    int32_t len = 0;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    if (THIS)
+        priv = THIS->private;
+    if (priv == NULL) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GLUSTERD_PRIV_NOT_FOUND,
+               "priv of glusterd not present");
+        *op_errstr = gf_strdup("glusterd defunct");
+        goto out;
+    }
+
+    ret = dict_get_str(dict, "host-uuid", &host_uuid);
+    if (ret < 0)
+        goto out;
+
+    ret = dict_get_str(dict, "source", &filename);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+               "Unable to fetch filename from dict.");
+        *op_errstr = gf_strdup("command unsuccessful");
+        goto out;
+    }
+    len = snprintf(abs_filename, sizeof(abs_filename), "%s/%s", priv->workdir,
+                   filename);
+    if ((len < 0) || (len >= sizeof(abs_filename))) {
+        ret = -1;
+        goto out;
+    }
+
+    uuid_utoa_r(MY_UUID, uuid_str);
+    if (!strcmp(uuid_str, host_uuid)) {
+        ret = sys_lstat(abs_filename, &stbuf);
+        if (ret) {
+            len = snprintf(errmsg, sizeof(errmsg),
+                           "Source file "
+                           "does not exist in %s",
+                           priv->workdir);
+            if (len < 0) {
+                strcpy(errmsg, "<error>");
+            }
+            *op_errstr = gf_strdup(errmsg);
+            gf_msg(this->name, GF_LOG_ERROR, ENOENT, GD_MSG_FILE_OP_FAILED,
+                   "%s", errmsg);
+            goto out;
+        }
+
+        contents = GF_CALLOC(1, stbuf.st_size + 1, gf_common_mt_char);
+        if (!contents) {
+            snprintf(errmsg, sizeof(errmsg), "Unable to allocate memory");
+            *op_errstr = gf_strdup(errmsg);
+            gf_msg(this->name, GF_LOG_ERROR, ENOMEM, GD_MSG_NO_MEMORY, "%s",
+                   errmsg);
+            ret = -1;
+            goto out;
+        }
+
+        fd = open(abs_filename, O_RDONLY);
+        if (fd < 0) {
+            len = snprintf(errmsg, sizeof(errmsg), "Unable to open %s",
+                           abs_filename);
+            if (len < 0) {
+                strcpy(errmsg, "<error>");
+            }
+            *op_errstr = gf_strdup(errmsg);
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_FILE_OP_FAILED, "%s",
+                   errmsg);
+            ret = -1;
+            goto out;
+        }
+
+        do {
+            ret = sys_read(fd, buf, sizeof(buf) - 1);
+            if (ret > 0) {
+                buf[ret] = '\0';
+                memcpy(contents + bytes_read, buf, ret);
+                bytes_read += ret;
+            }
+        } while (ret > 0);
+
+        if (bytes_read != stbuf.st_size) {
+            len = snprintf(errmsg, sizeof(errmsg),
+                           "Unable to read all the data from %s", abs_filename);
+            if (len < 0) {
+                strcpy(errmsg, "<error>");
+            }
+            *op_errstr = gf_strdup(errmsg);
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_READ_ERROR, "%s",
+                   errmsg);
+            ret = -1;
+            goto out;
+        }
+
+        ret = dict_set_int32(dict, "contents_size", stbuf.st_size);
+        if (ret) {
+            snprintf(errmsg, sizeof(errmsg),
+                     "Unable to set"
+                     " contents size in dict.");
+            *op_errstr = gf_strdup(errmsg);
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED, "%s",
+                   errmsg);
+            goto out;
+        }
+
+        ret = dict_set_int32(dict, "file_mode", (int32_t)stbuf.st_mode);
+        if (ret) {
+            snprintf(errmsg, sizeof(errmsg),
+                     "Unable to set"
+                     " file mode in dict.");
+            *op_errstr = gf_strdup(errmsg);
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED, "%s",
+                   errmsg);
+            goto out;
+        }
+
+        ret = dict_set_bin(dict, "common_pem_contents", contents,
+                           stbuf.st_size);
+        if (ret) {
+            snprintf(errmsg, sizeof(errmsg),
+                     "Unable to set"
+                     " pem contents in dict.");
+            *op_errstr = gf_strdup(errmsg);
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED, "%s",
+                   errmsg);
+            goto out;
+        }
+        free_contents = _gf_false;
+    } else {
+        free_contents = _gf_false;
+        ret = dict_get_bin(dict, "common_pem_contents", (void **)&contents);
+        if (ret) {
+            snprintf(errmsg, sizeof(errmsg),
+                     "Unable to get"
+                     " pem contents in dict.");
+            *op_errstr = gf_strdup(errmsg);
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED, "%s",
+                   errmsg);
+            goto out;
+        }
+        ret = dict_get_int32(dict, "contents_size", &contents_size);
+        if (ret) {
+            snprintf(errmsg, sizeof(errmsg),
+                     "Unable to set"
+                     " contents size in dict.");
+            *op_errstr = gf_strdup(errmsg);
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED, "%s",
+                   errmsg);
+            goto out;
+        }
+
+        ret = dict_get_int32(dict, "file_mode", &file_mode);
+        if (ret) {
+            snprintf(errmsg, sizeof(errmsg),
+                     "Unable to get"
+                     " file mode in dict.");
+            *op_errstr = gf_strdup(errmsg);
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED, "%s",
+                   errmsg);
+            goto out;
+        }
+
+        fd = open(abs_filename, O_WRONLY | O_TRUNC | O_CREAT, 0600);
+        if (fd < 0) {
+            len = snprintf(errmsg, sizeof(errmsg), "Unable to open %s",
+                           abs_filename);
+            if (len < 0) {
+                strcpy(errmsg, "<error>");
+            }
+            *op_errstr = gf_strdup(errmsg);
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_FILE_OP_FAILED, "%s",
+                   errmsg);
+            ret = -1;
+            goto out;
+        }
+
+        bytes_writen = sys_write(fd, contents, contents_size);
+
+        if (bytes_writen != contents_size) {
+            len = snprintf(errmsg, sizeof(errmsg), "Failed to write to %s",
+                           abs_filename);
+            if (len < 0) {
+                strcpy(errmsg, "<error>");
+            }
+            *op_errstr = gf_strdup(errmsg);
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_FILE_OP_FAILED, "%s",
+                   errmsg);
+            ret = -1;
+            goto out;
+        }
+
+        sys_fchmod(fd, file_mode);
+    }
+
+    ret = 0;
+out:
+    if (fd != -1)
+        sys_close(fd);
+
+    if (free_contents)
+        GF_FREE(contents);
+
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+int
+glusterd_op_gsync_set(dict_t *dict, char **op_errstr, dict_t *rsp_dict)
+{
+    int32_t ret = -1;
+    int32_t type = -1;
+    char *host_uuid = NULL;
+    char *slave = NULL;
+    char *slave_url = NULL;
+    char *slave_vol = NULL;
+    char *slave_host = NULL;
+    char *volname = NULL;
+    char *path_list = NULL;
+    glusterd_volinfo_t *volinfo = NULL;
+    glusterd_conf_t *priv = NULL;
+    gf_boolean_t is_force = _gf_false;
+    char *status_msg = NULL;
+    gf_boolean_t is_running = _gf_false;
+    char *conf_path = NULL;
+    char *key = NULL;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+    priv = this->private;
+    GF_ASSERT(priv);
+    GF_ASSERT(dict);
+    GF_ASSERT(op_errstr);
+    GF_ASSERT(rsp_dict);
+
+    ret = dict_get_int32(dict, "type", &type);
+    if (ret < 0)
+        goto out;
+
+    ret = dict_get_str(dict, "host-uuid", &host_uuid);
+    if (ret < 0)
+        goto out;
+
+    if (type == GF_GSYNC_OPTION_TYPE_STATUS) {
+        ret = glusterd_get_gsync_status(dict, op_errstr, rsp_dict);
+        goto out;
+    }
+
+    ret = dict_get_str(dict, "slave", &slave);
+    if (ret < 0)
+        goto out;
+
+    key = slave;
+
+    ret = dict_get_str(dict, "slave_url", &slave_url);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+               "Unable to fetch slave url.");
+        goto out;
+    }
+
+    ret = dict_get_str(dict, "slave_host", &slave_host);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+               "Unable to fetch slave hostname.");
+        goto out;
+    }
+
+    ret = dict_get_str(dict, "slave_vol", &slave_vol);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+               "Unable to fetch slave volume name.");
+        goto out;
+    }
+
+    ret = dict_get_str(dict, "conf_path", &conf_path);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+               "Unable to fetch conf file path.");
+        goto out;
+    }
+
+    if (dict_get_str(dict, "master", &volname) == 0) {
+        ret = glusterd_volinfo_find(volname, &volinfo);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_DICT_GET_FAILED,
+                   "Volinfo for"
+                   " %s (master) not found",
+                   volname);
+            goto out;
+        }
+
+        ret = glusterd_get_local_brickpaths(volinfo, &path_list);
+        if (!path_list && ret == -1)
+            goto out;
+    }
+
+    if (type == GF_GSYNC_OPTION_TYPE_CONFIG) {
+        ret = glusterd_gsync_configure(volinfo, slave, path_list, dict,
+                                       rsp_dict, op_errstr);
+        if (!ret) {
+            ret = dict_set_str(rsp_dict, "conf_path", conf_path);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+                       "Unable to store conf_file_path.");
+                goto out;
+            }
+        }
+        goto out;
+    }
+
+    if (type == GF_GSYNC_OPTION_TYPE_DELETE) {
+        ret = glusterd_remove_slave_in_info(volinfo, slave, op_errstr);
+        if (ret && !is_force && path_list)
+            goto out;
+
+        ret = glusterd_gsync_delete(volinfo, slave, slave_host, slave_vol,
+                                    path_list, dict, rsp_dict, op_errstr);
+        goto out;
+    }
+
+    if (!volinfo) {
+        ret = -1;
+        goto out;
+    }
+
+    is_force = dict_get_str_boolean(dict, "force", _gf_false);
+
+    if (type == GF_GSYNC_OPTION_TYPE_START) {
+        /* Add slave to the dict indicating geo-rep session is running*/
+        ret = dict_set_dynstr_with_alloc(volinfo->gsync_active_slaves, key,
+                                         "running");
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+                   "Unable to set key:%s"
+                   " value:running in the dict",
+                   key);
+            goto out;
+        }
+
+        /* If slave volume uuid is not present in gsync_slaves
+         * update it*/
+        ret = glusterd_update_slave_voluuid_slaveinfo(volinfo);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REMOTE_VOL_UUID_FAIL,
+                   "Error in updating"
+                   " slave volume uuid for old slave info");
+            goto out;
+        }
+
+        ret = glusterd_start_gsync(volinfo, slave, path_list, conf_path,
+                                   host_uuid, op_errstr, _gf_false);
+
+        /* Delete added slave in the dict if start fails*/
+        if (ret)
+            dict_del(volinfo->gsync_active_slaves, key);
+    }
+
+    if (type == GF_GSYNC_OPTION_TYPE_STOP ||
+        type == GF_GSYNC_OPTION_TYPE_PAUSE ||
+        type == GF_GSYNC_OPTION_TYPE_RESUME) {
+        ret = glusterd_check_gsync_running_local(volinfo->volname, slave,
+                                                 conf_path, &is_running);
+        if (!ret && !is_force && path_list && (_gf_true != is_running)) {
+            gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_GSYNCD_OP_SET_FAILED,
+                   GEOREP
+                   " is not "
+                   "set up for %s(master) and %s(slave)",
+                   volname, slave);
+            *op_errstr = gf_strdup(GEOREP " is not set up");
+            goto out;
+        }
+
+        if (type == GF_GSYNC_OPTION_TYPE_PAUSE) {
+            ret = gd_pause_or_resume_gsync(dict, volname, slave, slave_host,
+                                           slave_vol, conf_path, op_errstr,
+                                           _gf_true);
+            if (ret)
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_PAUSE_FAILED,
+                       GEOREP " Pause Failed");
+            else
+                dict_del(volinfo->gsync_active_slaves, key);
+
+        } else if (type == GF_GSYNC_OPTION_TYPE_RESUME) {
+            /* Add slave to the dict indicating geo-rep session is
+             * running*/
+            ret = dict_set_dynstr_with_alloc(volinfo->gsync_active_slaves, key,
+                                             "running");
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+                       "Unable to set "
+                       "key:%s value:running in dict",
+                       key);
+                goto out;
+            }
+
+            ret = gd_pause_or_resume_gsync(dict, volname, slave, slave_host,
+                                           slave_vol, conf_path, op_errstr,
+                                           _gf_false);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_RESUME_FAILED,
+                       GEOREP " Resume Failed");
+                dict_del(volinfo->gsync_active_slaves, key);
+            }
+        } else {
+            ret = stop_gsync(volname, slave, &status_msg, conf_path, op_errstr,
+                             is_force);
+
+            if (ret == 0 && status_msg)
+                ret = dict_set_str(rsp_dict, "gsync-status", status_msg);
+            if (!ret) {
+                ret = glusterd_create_status_file(
+                    volinfo->volname, slave, slave_host, slave_vol, "Stopped");
+                if (ret) {
+                    gf_msg(this->name, GF_LOG_ERROR, 0,
+                           GD_MSG_UPDATE_STATEFILE_FAILED,
+                           "Unable to update state_file. "
+                           "Error : %s",
+                           strerror(errno));
+                }
+                dict_del(volinfo->gsync_active_slaves, key);
+            }
+        }
+    }
+
+out:
+    if (path_list) {
+        GF_FREE(path_list);
+        path_list = NULL;
+    }
+
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+int
+glusterd_get_slave_details_confpath(glusterd_volinfo_t *volinfo, dict_t *dict,
+                                    char **slave_url, char **slave_host,
+                                    char **slave_vol, char **conf_path,
+                                    char **op_errstr)
+{
+    int ret = -1;
+    char confpath[PATH_MAX] = "";
+    glusterd_conf_t *priv = NULL;
+    char *slave = NULL;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    ret = dict_get_str(dict, "slave", &slave);
+    if (ret || !slave) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED,
+               "Unable to fetch slave from dict");
+        ret = -1;
+        goto out;
+    }
+
+    ret = glusterd_get_slave_info(slave, slave_url, slave_host, slave_vol,
+                                  op_errstr);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVEINFO_FETCH_ERROR,
+               "Unable to fetch slave details.");
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_set_str(dict, "slave_url", *slave_url);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+               "Unable to store slave IP.");
+        goto out;
+    }
+
+    ret = dict_set_str(dict, "slave_host", *slave_host);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+               "Unable to store slave hostname");
+        goto out;
+    }
+
+    ret = dict_set_str(dict, "slave_vol", *slave_vol);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+               "Unable to store slave volume name.");
+        goto out;
+    }
+
+    ret = snprintf(confpath, sizeof(confpath) - 1,
+                   "%s/" GEOREP "/%s_%s_%s/gsyncd.conf", priv->workdir,
+                   volinfo->volname, *slave_host, *slave_vol);
+    confpath[ret] = '\0';
+    *conf_path = gf_strdup(confpath);
+    if (!(*conf_path)) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_STRDUP_FAILED,
+               "Unable to gf_strdup. Error: %s", strerror(errno));
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_set_str(dict, "conf_path", *conf_path);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+               "Unable to store conf_path");
+        goto out;
+    }
+
+out:
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+int
+glusterd_get_slave_info(char *slave, char **slave_url, char **hostname,
+                        char **slave_vol, char **op_errstr)
+{
+    char *tmp = NULL;
+    char *save_ptr = NULL;
+    char **linearr = NULL;
+    int32_t ret = -1;
+    char errmsg[PATH_MAX] = "";
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    ret = glusterd_urltransform_single(slave, "normalize", &linearr);
+    if ((ret == -1) || (linearr[0] == NULL)) {
+        ret = snprintf(errmsg, sizeof(errmsg) - 1, "Invalid Url: %s", slave);
+        errmsg[ret] = '\0';
+        *op_errstr = gf_strdup(errmsg);
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_NORMALIZE_URL_FAIL,
+               "Failed to normalize url");
+        goto out;
+    }
+
+    tmp = strtok_r(linearr[0], "/", &save_ptr);
+    tmp = strtok_r(NULL, "/", &save_ptr);
+    slave = NULL;
+    if (tmp != NULL) {
+        slave = strtok_r(tmp, ":", &save_ptr);
+    }
+    if (slave) {
+        ret = glusterd_geo_rep_parse_slave(slave, hostname, op_errstr);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVE_URL_INVALID,
+                   "Invalid slave url: %s", *op_errstr);
+            goto out;
+        }
+        gf_msg_debug(this->name, 0, "Hostname : %s", *hostname);
+
+        *slave_url = gf_strdup(slave);
+        if (!*slave_url) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_STRDUP_FAILED,
+                   "Failed to gf_strdup");
+            ret = -1;
+            goto out;
+        }
+        gf_msg_debug(this->name, 0, "Slave URL : %s", *slave_url);
+        ret = 0;
+    } else {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, GD_MSG_INVALID_ENTRY,
+               "Invalid slave name");
+        goto out;
+    }
+
+    slave = strtok_r(NULL, ":", &save_ptr);
+    if (slave) {
+        *slave_vol = gf_strdup(slave);
+        if (!*slave_vol) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_STRDUP_FAILED,
+                   "Failed to gf_strdup");
+            ret = -1;
+            GF_FREE(*slave_url);
+            goto out;
+        }
+        gf_msg_debug(this->name, 0, "Slave Vol : %s", *slave_vol);
+        ret = 0;
+    } else {
+        gf_msg(this->name, GF_LOG_ERROR, EINVAL, GD_MSG_INVALID_ENTRY,
+               "Invalid slave name");
+        goto out;
+    }
+
+out:
+    if (linearr)
+        glusterd_urltransform_free(linearr, 1);
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+static void
+runinit_gsyncd_setrx(runner_t *runner, char *conf_path)
+{
+    runinit(runner);
+    runner_add_args(runner, GSYNCD_PREFIX "/gsyncd", "-c", NULL);
+    runner_argprintf(runner, "%s", conf_path);
+    runner_add_arg(runner, "--config-set-rx");
+}
+
+static int
+glusterd_check_gsync_present(int *valid_state)
+{
+    char buff[PATH_MAX] = {
+        0,
+    };
+    runner_t runner = {
+        0,
+    };
+    char *ptr = NULL;
+    int ret = 0;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    runinit(&runner);
+    runner_add_args(&runner, GSYNCD_PREFIX "/gsyncd", "--version", NULL);
+    runner_redir(&runner, STDOUT_FILENO, RUN_PIPE);
+    ret = runner_start(&runner);
+    if (ret == -1) {
+        if (errno == ENOENT) {
+            gf_msg("glusterd", GF_LOG_INFO, ENOENT, GD_MSG_MODULE_NOT_INSTALLED,
+                   GEOREP
+                   " module "
+                   "not installed in the system");
+            *valid_state = 0;
+        } else {
+            gf_msg("glusterd", GF_LOG_ERROR, 0, GD_MSG_MODULE_ERROR,
+                   GEOREP " module not working as desired");
+            *valid_state = -1;
+        }
+        goto out;
+    }
+
+    ptr = fgets(buff, sizeof(buff), runner_chio(&runner, STDOUT_FILENO));
+    if (ptr) {
+        if (!strstr(buff, "gsyncd")) {
+            ret = -1;
+            gf_msg("glusterd", GF_LOG_ERROR, 0, GD_MSG_MODULE_ERROR,
+                   GEOREP " module not working as desired");
+            *valid_state = -1;
+            goto out;
+        }
+    } else {
+        ret = -1;
+        gf_msg("glusterd", GF_LOG_ERROR, 0, GD_MSG_MODULE_ERROR,
+               GEOREP " module not working as desired");
+        *valid_state = -1;
+        goto out;
+    }
+
+    ret = 0;
+out:
+
+    runner_end(&runner);
+
+    gf_msg_debug("glusterd", 0, "Returning %d", ret);
+    return ret;
+}
+
+static int
+create_conf_file(glusterd_conf_t *conf, char *conf_path)
+#define RUN_GSYNCD_CMD                                                         \
+    do {                                                                       \
+        ret = runner_run_reuse(&runner);                                       \
+        if (ret == -1) {                                                       \
+            runner_log(&runner, "glusterd", GF_LOG_ERROR, "command failed");   \
+            runner_end(&runner);                                               \
+            goto out;                                                          \
+        }                                                                      \
+        runner_end(&runner);                                                   \
+    } while (0)
+{
+    int ret = 0;
+    runner_t runner = {
+        0,
+    };
+    char georepdir[PATH_MAX] = {
+        0,
+    };
+    int valid_state = 0;
+
+    valid_state = -1;
+    ret = glusterd_check_gsync_present(&valid_state);
+    if (-1 == ret) {
+        ret = valid_state;
+        goto out;
+    }
+
+    ret = snprintf(georepdir, sizeof(georepdir) - 1, "%s/" GEOREP,
+                   conf->workdir);
+    georepdir[ret] = '\0';
+
+    /************
+     * master pre-configuration
+     ************/
+
+    /* remote-gsyncd */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_args(&runner, "remote-gsyncd", GSYNCD_PREFIX "/gsyncd", ".", ".",
+                    NULL);
+    RUN_GSYNCD_CMD;
+
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_args(&runner, "remote-gsyncd", "/nonexistent/gsyncd", ".",
+                    "^ssh:", NULL);
+    RUN_GSYNCD_CMD;
+
+    /* gluster-command-dir */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_args(&runner, "gluster-command-dir", SBIN_DIR "/", ".", ".",
+                    NULL);
+    RUN_GSYNCD_CMD;
+
+    /* gluster-params */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_args(&runner, "gluster-params", "aux-gfid-mount acl", ".", ".",
+                    NULL);
+    RUN_GSYNCD_CMD;
+
+    /* ssh-command */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_arg(&runner, "ssh-command");
+    runner_argprintf(&runner,
+                     "ssh -oPasswordAuthentication=no "
+                     "-oStrictHostKeyChecking=no "
+                     "-i %s/secret.pem",
+                     georepdir);
+    runner_add_args(&runner, ".", ".", NULL);
+    RUN_GSYNCD_CMD;
+
+    /* ssh-command tar */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_arg(&runner, "ssh-command-tar");
+    runner_argprintf(&runner,
+                     "ssh -oPasswordAuthentication=no "
+                     "-oStrictHostKeyChecking=no "
+                     "-i %s/tar_ssh.pem",
+                     georepdir);
+    runner_add_args(&runner, ".", ".", NULL);
+    RUN_GSYNCD_CMD;
+
+    /* pid-file */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_arg(&runner, "pid-file");
+    runner_argprintf(&runner,
+                     "%s/${mastervol}_${remotehost}_${slavevol}/monitor.pid",
+                     georepdir);
+    runner_add_args(&runner, ".", ".", NULL);
+    RUN_GSYNCD_CMD;
+
+    /* geo-rep-working-dir */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_arg(&runner, "georep-session-working-dir");
+    runner_argprintf(&runner, "%s/${mastervol}_${remotehost}_${slavevol}/",
+                     georepdir);
+    runner_add_args(&runner, ".", ".", NULL);
+    RUN_GSYNCD_CMD;
+
+    /* state-file */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_arg(&runner, "state-file");
+    runner_argprintf(&runner,
+                     "%s/${mastervol}_${remotehost}_${slavevol}/monitor.status",
+                     georepdir);
+    runner_add_args(&runner, ".", ".", NULL);
+    RUN_GSYNCD_CMD;
+
+    /* state-detail-file */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_arg(&runner, "state-detail-file");
+    runner_argprintf(
+        &runner,
+        "%s/${mastervol}_${remotehost}_${slavevol}/${eSlave}-detail.status",
+        georepdir);
+    runner_add_args(&runner, ".", ".", NULL);
+    RUN_GSYNCD_CMD;
+
+    /* state-socket */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_arg(&runner, "state-socket-unencoded");
+    runner_argprintf(
+        &runner, "%s/${mastervol}_${remotehost}_${slavevol}/${eSlave}.socket",
+        georepdir);
+    runner_add_args(&runner, ".", ".", NULL);
+    RUN_GSYNCD_CMD;
+
+    /* socketdir */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_args(&runner, "socketdir", GLUSTERD_SOCK_DIR, ".", ".", NULL);
+    RUN_GSYNCD_CMD;
+
+    /* log-file */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_arg(&runner, "log-file");
+    runner_argprintf(&runner, "%s/%s/${mastervol}/${eSlave}.log", conf->logdir,
+                     GEOREP);
+    runner_add_args(&runner, ".", ".", NULL);
+    RUN_GSYNCD_CMD;
+
+    /* changelog-log-file */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_arg(&runner, "changelog-log-file");
+    runner_argprintf(&runner,
+                     "%s/%s/${mastervol}/${eSlave}${local_id}-changes.log",
+                     conf->logdir, GEOREP);
+    runner_add_args(&runner, ".", ".", NULL);
+    RUN_GSYNCD_CMD;
+
+    /* gluster-log-file */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_arg(&runner, "gluster-log-file");
+    runner_argprintf(&runner,
+                     "%s/%s/${mastervol}/${eSlave}${local_id}.gluster.log",
+                     conf->logdir, GEOREP);
+    runner_add_args(&runner, ".", ".", NULL);
+    RUN_GSYNCD_CMD;
+
+    /* ignore-deletes */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_args(&runner, "ignore-deletes", "false", ".", ".", NULL);
+    RUN_GSYNCD_CMD;
+
+    /* special-sync-mode */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_args(&runner, "special-sync-mode", "partial", ".", ".", NULL);
+    RUN_GSYNCD_CMD;
+
+    /* change-detector == changelog */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_args(&runner, "change-detector", "changelog", ".", ".", NULL);
+    RUN_GSYNCD_CMD;
+
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_arg(&runner, "working-dir");
+    runner_argprintf(&runner, "%s/${mastervol}/${eSlave}",
+                     DEFAULT_GLUSTERFSD_MISC_DIRETORY);
+    runner_add_args(&runner, ".", ".", NULL);
+    RUN_GSYNCD_CMD;
+
+    /************
+     * slave pre-configuration
+     ************/
+
+    /* slave-gluster-command-dir */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_args(&runner, "slave-gluster-command-dir", SBIN_DIR "/", ".",
+                    NULL);
+    RUN_GSYNCD_CMD;
+
+    /* gluster-params */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_args(&runner, "gluster-params", "aux-gfid-mount acl", ".", NULL);
+    RUN_GSYNCD_CMD;
+
+    /* log-file */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_arg(&runner, "log-file");
+    runner_argprintf(&runner,
+                     "%s/%s-slaves/"
+                     "${session_owner}:${local_node}${local_id}.${slavevol}."
+                     "log",
+                     conf->logdir, GEOREP);
+    runner_add_args(&runner, ".", ".", NULL);
+    RUN_GSYNCD_CMD;
+
+    /* MountBroker log-file */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_arg(&runner, "log-file-mbr");
+    runner_argprintf(&runner,
+                     "%s/%s-slaves/mbr/"
+                     "${session_owner}:${local_node}${local_id}.${slavevol}."
+                     "log",
+                     conf->logdir, GEOREP);
+    runner_add_args(&runner, ".", ".", NULL);
+    RUN_GSYNCD_CMD;
+
+    /* gluster-log-file */
+    runinit_gsyncd_setrx(&runner, conf_path);
+    runner_add_arg(&runner, "gluster-log-file");
+    runner_argprintf(&runner,
+                     "%s/%s-slaves/"
+                     "${session_owner}:${local_node}${local_id}.${slavevol}."
+                     "gluster.log",
+                     conf->logdir, GEOREP);
+    runner_add_args(&runner, ".", ".", NULL);
+    RUN_GSYNCD_CMD;
+
+out:
+    return ret ? -1 : 0;
+}
+
+static int
+glusterd_create_essential_dir_files(glusterd_volinfo_t *volinfo, dict_t *dict,
+                                    char *slave, char *slave_host,
+                                    char *slave_vol, char **op_errstr)
+{
+    int ret = -1;
+    char *conf_path = NULL;
+    char *statefile = NULL;
+    char buf[PATH_MAX] = "";
+    char errmsg[PATH_MAX] = "";
+    glusterd_conf_t *conf = NULL;
+    struct stat stbuf = {
+        0,
+    };
+    xlator_t *this = NULL;
+    int32_t len = 0;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    conf = this->private;
+
+    ret = dict_get_str(dict, "conf_path", &conf_path);
+    if (ret) {
+        snprintf(errmsg, sizeof(errmsg), "Unable to fetch conf file path.");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED, "%s",
+               errmsg);
+        goto out;
+    }
+
+    ret = dict_get_str(dict, "statefile", &statefile);
+    if (ret) {
+        snprintf(errmsg, sizeof(errmsg), "Unable to fetch statefile path.");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED, "%s",
+               errmsg);
+        goto out;
+    }
+
+    ret = snprintf(buf, sizeof(buf), "%s/" GEOREP "/%s_%s_%s", conf->workdir,
+                   volinfo->volname, slave_host, slave_vol);
+    if ((ret < 0) || (ret >= sizeof(buf))) {
+        ret = -1;
+        goto out;
+    }
+    ret = mkdir_p(buf, 0755, _gf_true);
+    if (ret) {
+        len = snprintf(errmsg, sizeof(errmsg),
+                       "Unable to create %s"
+                       ". Error : %s",
+                       buf, strerror(errno));
+        if (len < 0) {
+            strcpy(errmsg, "<error>");
+        }
+        *op_errstr = gf_strdup(errmsg);
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DIR_OP_FAILED, "%s",
+               errmsg);
+        goto out;
+    }
+
+    ret = snprintf(buf, PATH_MAX, "%s/" GEOREP "/%s", conf->logdir,
+                   volinfo->volname);
+    if ((ret < 0) || (ret >= PATH_MAX)) {
+        ret = -1;
+        goto out;
+    }
+    ret = mkdir_p(buf, 0755, _gf_true);
+    if (ret) {
+        len = snprintf(errmsg, sizeof(errmsg),
+                       "Unable to create %s"
+                       ". Error : %s",
+                       buf, strerror(errno));
+        if (len < 0) {
+            strcpy(errmsg, "<error>");
+        }
+        *op_errstr = gf_strdup(errmsg);
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_DIR_OP_FAILED, "%s",
+               errmsg);
+        goto out;
+    }
+
+    ret = sys_lstat(conf_path, &stbuf);
+    if (!ret) {
+        gf_msg_debug(this->name, 0,
+                     "Session already running."
+                     " Not creating config file again.");
+    } else {
+        ret = create_conf_file(conf, conf_path);
+        if (ret || sys_lstat(conf_path, &stbuf)) {
+            snprintf(errmsg, sizeof(errmsg),
+                     "Failed to create"
+                     " config file(%s).",
+                     conf_path);
+            gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_FILE_OP_FAILED, "%s",
+                   errmsg);
+            goto out;
+        }
+    }
+
+    ret = sys_lstat(statefile, &stbuf);
+    if (!ret) {
+        gf_msg_debug(this->name, 0,
+                     "Session already running."
+                     " Not creating status file again.");
+        goto out;
+    } else {
+        ret = glusterd_create_status_file(volinfo->volname, slave, slave_host,
+                                          slave_vol, "Created");
+        if (ret || sys_lstat(statefile, &stbuf)) {
+            snprintf(errmsg, sizeof(errmsg),
+                     "Unable to create %s"
+                     ". Error : %s",
+                     statefile, strerror(errno));
+            *op_errstr = gf_strdup(errmsg);
+            gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_FILE_OP_FAILED, "%s",
+                   errmsg);
+            ret = -1;
+            goto out;
+        }
+    }
+
+out:
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
+
+int
+glusterd_op_gsync_create(dict_t *dict, char **op_errstr, dict_t *rsp_dict)
+{
+    char common_pem_file[PATH_MAX] = "";
+    char errmsg[PATH_MAX] = {
+        0,
+    };
+    char hooks_args[PATH_MAX] = "";
+    char uuid_str[64] = "";
+    char *host_uuid = NULL;
+    char *slave_url = NULL;
+    char *slave_url_buf = NULL;
+    char *slave_user = NULL;
+    char *slave_ip = NULL;
+    char *save_ptr = NULL;
+    char *slave_host = NULL;
+    char *slave_vol = NULL;
+    char *arg_buf = NULL;
+    char *volname = NULL;
+    char *slave = NULL;
+    int32_t ret = -1;
+    int32_t is_pem_push = -1;
+    int32_t ssh_port = 22;
+    gf_boolean_t is_force = -1;
+    glusterd_conf_t *conf = NULL;
+    glusterd_volinfo_t *volinfo = NULL;
+    xlator_t *this = NULL;
+    char old_working_dir[PATH_MAX] = {0};
+    char new_working_dir[PATH_MAX] = {0};
+    char *slave_voluuid = NULL;
+    char *old_slavehost = NULL;
+    gf_boolean_t is_existing_session = _gf_false;
+    int32_t len = 0;
+
+    this = THIS;
+    GF_ASSERT(this);
+    conf = this->private;
+    GF_ASSERT(conf);
+    GF_ASSERT(dict);
+    GF_ASSERT(op_errstr);
+
+    ret = glusterd_op_gsync_args_get(dict, op_errstr, &volname, &slave,
+                                     &host_uuid);
+    if (ret)
+        goto out;
+
+    len = snprintf(common_pem_file, sizeof(common_pem_file),
+                   "%s" GLUSTERD_COMMON_PEM_PUB_FILE, conf->workdir);
+    if ((len < 0) || (len >= sizeof(common_pem_file))) {
+        ret = -1;
+        goto out;
+    }
+
+    ret = glusterd_volinfo_find(volname, &volinfo);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_VOL_NOT_FOUND,
+               "Volinfo for %s (master) not found", volname);
+        goto out;
+    }
+
+    ret = dict_get_str(dict, "slave_vol", &slave_vol);
+    if (ret) {
+        snprintf(errmsg, sizeof(errmsg), "Unable to fetch slave volume name.");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED, "%s",
+               errmsg);
+        goto out;
+    }
+
+    ret = dict_get_str(dict, "slave_url", &slave_url);
+    if (ret) {
+        snprintf(errmsg, sizeof(errmsg), "Unable to fetch slave IP.");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED, "%s",
+               errmsg);
+        ret = -1;
+        goto out;
+    }
+
+    /* Fetch the slave_user and slave_ip from the slave_url.
+     * If the slave_user is not present. Use "root"
+     */
+    if (strstr(slave_url, "@")) {
+        slave_url_buf = gf_strdup(slave_url);
+        if (!slave_url_buf) {
+            ret = -1;
+            goto out;
+        }
+        slave_user = strtok_r(slave_url, "@", &save_ptr);
+        slave_ip = strtok_r(NULL, "@", &save_ptr);
+    } else {
+        slave_user = "root";
+        slave_ip = slave_url;
+    }
+
+    if (!slave_user || !slave_ip) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVE_URL_INVALID,
+               "Invalid slave url.");
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_get_str(dict, "slave_host", &slave_host);
+    if (ret) {
+        snprintf(errmsg, sizeof(errmsg), "Unable to fetch slave host");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED, "%s",
+               errmsg);
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_get_int32(dict, "ssh_port", &ssh_port);
+    if (ret < 0 && ret != -ENOENT) {
+        snprintf(errmsg, sizeof(errmsg), "Fetching ssh_port failed");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED, "%s",
+               errmsg);
+        ret = -1;
+        goto out;
+    }
+
+    is_force = dict_get_str_boolean(dict, "force", _gf_false);
+
+    uuid_utoa_r(MY_UUID, uuid_str);
+    if (!strcmp(uuid_str, host_uuid)) {
+        ret = dict_get_int32(dict, "push_pem", &is_pem_push);
+        if (!ret && is_pem_push) {
+            gf_msg_debug(this->name, 0,
+                         "Trying to setup"
+                         " pem files in slave");
+            is_pem_push = 1;
+        } else
+            is_pem_push = 0;
+
+        len = snprintf(hooks_args, sizeof(hooks_args),
+                       "is_push_pem=%d,pub_file=%s,slave_user=%s,"
+                       "slave_ip=%s,slave_vol=%s,ssh_port=%d",
+                       is_pem_push, common_pem_file, slave_user, slave_ip,
+                       slave_vol, ssh_port);
+        if ((len < 0) || (len >= sizeof(hooks_args))) {
+            ret = -1;
+            goto out;
+        }
+    } else
+        snprintf(hooks_args, sizeof(hooks_args),
+                 "This argument will stop the hooks script");
+
+    arg_buf = gf_strdup(hooks_args);
+    if (!arg_buf) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_STRDUP_FAILED,
+               "Failed to gf_strdup");
+        if (is_force) {
+            ret = 0;
+            goto create_essentials;
+        }
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_set_str(dict, "hooks_args", arg_buf);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+               "Failed to set hooks_args in dict.");
+        if (is_force) {
+            ret = 0;
+            goto create_essentials;
+        }
+        goto out;
+    }
+
+create_essentials:
+    /* Fetch slave volume uuid, to get stored in volume info. */
+    ret = dict_get_str(dict, "slave_voluuid", &slave_voluuid);
+    if (ret) {
+        snprintf(errmsg, sizeof(errmsg),
+                 "Unable to fetch slave volume uuid from dict");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED, "%s",
+               errmsg);
+        ret = -1;
+        goto out;
+    }
+
+    is_existing_session = dict_get_str_boolean(dict, "existing_session",
+                                               _gf_false);
+    if (is_existing_session) {
+        ret = dict_get_str(dict, "old_slavehost", &old_slavehost);
+        if (ret) {
+            snprintf(errmsg, sizeof(errmsg), "Unable to fetch old_slavehost");
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED, "%s",
+                   errmsg);
+            ret = -1;
+            goto out;
+        }
+
+        /* Rename existing geo-rep session with new Slave Host */
+        ret = snprintf(old_working_dir, sizeof(old_working_dir) - 1,
+                       "%s/" GEOREP "/%s_%s_%s", conf->workdir,
+                       volinfo->volname, old_slavehost, slave_vol);
+
+        ret = snprintf(new_working_dir, sizeof(new_working_dir) - 1,
+                       "%s/" GEOREP "/%s_%s_%s", conf->workdir,
+                       volinfo->volname, slave_host, slave_vol);
+
+        ret = sys_rename(old_working_dir, new_working_dir);
+        if (!ret) {
+            gf_msg(this->name, GF_LOG_INFO, 0, GD_MSG_FORCE_CREATE_SESSION,
+                   "rename of old working dir %s to "
+                   "new working dir %s is done! ",
+                   old_working_dir, new_working_dir);
+        } else {
+            if (errno == ENOENT) {
+                /* log error, but proceed with directory
+                 * creation below */
+                gf_msg_debug(this->name, 0,
+                             "old_working_dir(%s) "
+                             "not present.",
+                             old_working_dir);
+            } else {
+                len = snprintf(errmsg, sizeof(errmsg),
+                               "rename of old working dir %s "
+                               "to new working dir %s "
+                               "failed! Error: %s",
+                               old_working_dir, new_working_dir,
+                               strerror(errno));
+                if (len < 0) {
+                    strcpy(errmsg, "<error>");
+                }
+                gf_msg(this->name, GF_LOG_INFO, 0, GD_MSG_FORCE_CREATE_SESSION,
+                       "rename of old working dir %s to "
+                       "new working dir %s failed! Error: %s!",
+                       old_working_dir, new_working_dir, strerror(errno));
+
+                ret = -1;
+                goto out;
+            }
+        }
+    }
+
+    ret = glusterd_create_essential_dir_files(volinfo, dict, slave, slave_host,
+                                              slave_vol, op_errstr);
+    if (ret)
+        goto out;
+
+    ret = glusterd_store_slave_in_info(volinfo, slave, host_uuid, slave_voluuid,
+                                       op_errstr, is_force);
+    if (ret) {
+        snprintf(errmsg, sizeof(errmsg),
+                 "Unable to store"
+                 " slave info.");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SLAVEINFO_STORE_ERROR, "%s",
+               errmsg);
+        goto out;
+    }
+
+    /* Enable marker and changelog */
+    ret = glusterd_set_gsync_confs(volinfo);
+    if (ret != 0) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_MARKER_START_FAIL,
+               "marker/changelog"
+               " start failed");
+        snprintf(errmsg, sizeof(errmsg), "Index initialization failed");
+
+        ret = -1;
+        goto out;
+    }
+
+out:
+    if (ret && errmsg[0] != '\0') {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GSYNCD_ERROR, "%s", errmsg);
+        *op_errstr = gf_strdup(errmsg);
+    }
+
+    GF_FREE(slave_url_buf);
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
+}
diff --git a/xlators/mgmt/glusterd/src/glusterd-geo-rep.h b/xlators/mgmt/glusterd/src/glusterd-geo-rep.h
new file mode 100644
index 00000000000..7d1318f522c
--- /dev/null
+++ b/xlators/mgmt/glusterd/src/glusterd-geo-rep.h
@@ -0,0 +1,52 @@
+/*
+   Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+#ifndef _GLUSTERD_GEO_REP_H_
+#define _GLUSTERD_GEO_REP_H_
+
+#ifndef GSYNC_CONF_TEMPLATE
+#define GSYNC_CONF_TEMPLATE GEOREP "/gsyncd_template.conf"
+#endif
+
+/* <slave host>::<slave volume> */
+#define SLAVE_URL_INFO_MAX (_POSIX_HOST_NAME_MAX + GD_VOLUME_NAME_MAX + 3)
+
+/* slave info format:
+ * <master host uuid>:ssh://{<slave_user>@}<slave host>::<slave volume> \
+ * :<slave volume uuid> */
+#define VOLINFO_SLAVE_URL_MAX                                                  \
+    (LOGIN_NAME_MAX + (2 * GF_UUID_BUF_SIZE) + SLAVE_URL_INFO_MAX + 10)
+
+typedef struct glusterd_gsync_status_temp {
+    dict_t *rsp_dict;
+    glusterd_volinfo_t *volinfo;
+    char *node;
+} glusterd_gsync_status_temp_t;
+
+typedef struct gsync_status_param {
+    glusterd_volinfo_t *volinfo;
+    int is_active;
+} gsync_status_param_t;
+
+int
+gsync_status(char *master, char *slave, char *conf_path, int *status,
+             gf_boolean_t *is_template_in_use);
+
+void
+glusterd_check_geo_rep_configured(glusterd_volinfo_t *volinfo,
+                                  gf_boolean_t *flag);
+int
+_get_slave_status(dict_t *dict, char *key, data_t *value, void *data);
+int
+glusterd_check_geo_rep_running(gsync_status_param_t *param, char **op_errstr);
+
+int
+glusterd_get_gsync_status_mst(glusterd_volinfo_t *volinfo, dict_t *rsp_dict,
+                              char *node);
+#endif
diff --git a/xlators/mgmt/glusterd/src/glusterd-gfproxyd-svc-helper.c b/xlators/mgmt/glusterd/src/glusterd-gfproxyd-svc-helper.c
new file mode 100644
index 00000000000..319bfa140f3
--- /dev/null
+++ b/xlators/mgmt/glusterd/src/glusterd-gfproxyd-svc-helper.c
@@ -0,0 +1,235 @@
+/*
+   Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include "glusterd.h"
+#include "glusterd-utils.h"
+#include "glusterd-gfproxyd-svc-helper.h"
+#include "glusterd-messages.h"
+#include <glusterfs/syscall.h>
+#include "glusterd-volgen.h"
+
+void
+glusterd_svc_build_gfproxyd_rundir(glusterd_volinfo_t *volinfo, char *path,
+                                   int path_len)
+{
+    char workdir[PATH_MAX] = {
+        0,
+    };
+    glusterd_conf_t *priv = THIS->private;
+
+    GLUSTERD_GET_VOLUME_PID_DIR(workdir, volinfo, priv);
+
+    snprintf(path, path_len, "%s", workdir);
+}
+
+void
+glusterd_svc_build_gfproxyd_socket_filepath(glusterd_volinfo_t *volinfo,
+                                            char *path, int path_len)
+{
+    char sockfilepath[PATH_MAX] = {
+        0,
+    };
+    char rundir[PATH_MAX] = {
+        0,
+    };
+    int32_t len = 0;
+
+    glusterd_svc_build_gfproxyd_rundir(volinfo, rundir, sizeof(rundir));
+    len = snprintf(sockfilepath, sizeof(sockfilepath), "%s/run-%s", rundir,
+                   uuid_utoa(MY_UUID));
+    if ((len < 0) || (len >= sizeof(sockfilepath))) {
+        sockfilepath[0] = 0;
+    }
+
+    glusterd_set_socket_filepath(sockfilepath, path, path_len);
+}
+
+void
+glusterd_svc_build_gfproxyd_pidfile(glusterd_volinfo_t *volinfo, char *path,
+                                    int path_len)
+{
+    char rundir[PATH_MAX] = {
+        0,
+    };
+
+    glusterd_svc_build_gfproxyd_rundir(volinfo, rundir, sizeof(rundir));
+
+    snprintf(path, path_len, "%s/%s.gfproxyd.pid", rundir, volinfo->volname);
+}
+
+void
+glusterd_svc_build_gfproxyd_volfile_path(glusterd_volinfo_t *volinfo,
+                                         char *path, int path_len)
+{
+    char workdir[PATH_MAX] = {
+        0,
+    };
+    glusterd_conf_t *priv = THIS->private;
+
+    GLUSTERD_GET_VOLUME_DIR(workdir, volinfo, priv);
+
+    snprintf(path, path_len, "%s/%s.gfproxyd.vol", workdir, volinfo->volname);
+}
+
+void
+glusterd_svc_build_gfproxyd_logdir(char *logdir, char *volname, size_t len)
+{
+    glusterd_conf_t *conf = THIS->private;
+    snprintf(logdir, len, "%s/gfproxy/%s", conf->logdir, volname);
+}
+
+void
+glusterd_svc_build_gfproxyd_logfile(char *logfile, char *logdir, size_t len)
+{
+    snprintf(logfile, len, "%s/gfproxyd.log", logdir);
+}
+
+int
+glusterd_is_gfproxyd_enabled(glusterd_volinfo_t *volinfo)
+{
+    return glusterd_volinfo_get_boolean(volinfo, VKEY_CONFIG_GFPROXY);
+}
+
+static int
+glusterd_svc_get_gfproxyd_volfile(glusterd_volinfo_t *volinfo, char *svc_name,
+                                  char *orgvol, char **tmpvol, int path_len)
+{
+    int tmp_fd = -1;
+    int ret = -1;
+    int need_unlink = 0;
+
+    glusterd_svc_build_gfproxyd_volfile_path(volinfo, orgvol, path_len);
+
+    ret = gf_asprintf(tmpvol, "/tmp/g%s-XXXXXX", svc_name);
+    if (ret < 0) {
+        goto out;
+    }
+
+    /* coverity[SECURE_TEMP] mkstemp uses 0600 as the mode and is safe */
+    tmp_fd = mkstemp(*tmpvol);
+    if (tmp_fd < 0) {
+        gf_msg("glusterd", GF_LOG_WARNING, errno, GD_MSG_FILE_OP_FAILED,
+               "Unable to create temp file"
+               " %s:(%s)",
+               *tmpvol, strerror(errno));
+        ret = -1;
+        goto out;
+    }
+
+    need_unlink = 1;
+    ret = glusterd_build_gfproxyd_volfile(volinfo, *tmpvol);
+out:
+    if (need_unlink && ret < 0)
+        sys_unlink(*tmpvol);
+
+    if ((ret < 0) && (*tmpvol != NULL)) {
+        GF_FREE(*tmpvol);
+        *tmpvol = NULL;
+    }
+
+    if (tmp_fd >= 0)
+        sys_close(tmp_fd);
+
+    return ret;
+}
+
+int
+glusterd_svc_check_gfproxyd_volfile_identical(char *svc_name,
+                                              glusterd_volinfo_t *volinfo,
+                                              gf_boolean_t *identical)
+{
+    char orgvol[PATH_MAX] = {
+        0,
+    };
+    char *tmpvol = NULL;
+    int ret = -1;
+    int need_unlink = 0;
+
+    GF_VALIDATE_OR_GOTO("glusterd", identical, out);
+
+    ret = glusterd_svc_get_gfproxyd_volfile(volinfo, svc_name, orgvol, &tmpvol,
+                                            PATH_MAX);
+    if (ret)
+        goto out;
+
+    need_unlink = 1;
+    ret = glusterd_check_files_identical(orgvol, tmpvol, identical);
+    if (ret)
+        goto out;
+
+out:
+    if (need_unlink)
+        sys_unlink(tmpvol);
+
+    if (tmpvol != NULL)
+        GF_FREE(tmpvol);
+
+    return ret;
+}
+
+int
+glusterd_svc_check_gfproxyd_topology_identical(char *svc_name,
+                                               glusterd_volinfo_t *volinfo,
+                                               gf_boolean_t *identical)
+{
+    char orgvol[PATH_MAX] = {
+        0,
+    };
+    char *tmpvol = NULL;
+    int ret = -1;
+    int tmpclean = 0;
+
+    GF_VALIDATE_OR_GOTO("glusterd", identical, out);
+
+    ret = glusterd_svc_get_gfproxyd_volfile(volinfo, svc_name, orgvol, &tmpvol,
+                                            PATH_MAX);
+    if (ret)
+        goto out;
+
+    tmpclean = 1; /* SET the flag to unlink() tmpfile */
+
+    /* Compare the topology of volfiles */
+    ret = glusterd_check_topology_identical(orgvol, tmpvol, identical);
+out:
+    if (tmpclean)
+        sys_unlink(tmpvol);
+
+    if (tmpvol != NULL)
+        GF_FREE(tmpvol);
+
+    return ret;
+}
+
+glusterd_volinfo_t *
+glusterd_gfproxyd_volinfo_from_svc(glusterd_svc_t *svc)
+{
+    glusterd_volinfo_t *volinfo = NULL;
+    glusterd_gfproxydsvc_t *gfproxyd = NULL;
+
+    /* Get volinfo->gfproxyd from svc object */
+    gfproxyd = cds_list_entry(svc, glusterd_gfproxydsvc_t, svc);
+    if (!gfproxyd) {
+        gf_msg("glusterd", GF_LOG_ERROR, 0, GD_MSG_SNAPD_OBJ_GET_FAIL,
+               "Failed to get gfproxyd "
+               "object from gfproxyd service");
+        goto out;
+    }
+
+    /* Get volinfo from gfproxyd */
+    volinfo = cds_list_entry(gfproxyd, glusterd_volinfo_t, gfproxyd);
+    if (!volinfo) {
+        gf_msg("glusterd", GF_LOG_ERROR, 0, GD_MSG_VOLINFO_GET_FAIL,
+               "Failed to get volinfo from "
+               "from gfproxyd");
+        goto out;
+    }
+out:
+    return volinfo;
+}
diff --git a/xlators/mgmt/glusterd/src/glusterd-gfproxyd-svc-helper.h b/xlators/mgmt/glusterd/src/glusterd-gfproxyd-svc-helper.h
new file mode 100644
index 00000000000..3aca218a65d
--- /dev/null
+++ b/xlators/mgmt/glusterd/src/glusterd-gfproxyd-svc-helper.h
@@ -0,0 +1,51 @@
+/*
+   Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#ifndef _GLUSTERD_GFPROXYD_SVC_HELPER_H_
+#define _GLUSTERD_GFPROXYD_SVC_HELPER_H_
+
+#include "glusterd.h"
+
+void
+glusterd_svc_build_gfproxyd_rundir(glusterd_volinfo_t *volinfo, char *path,
+                                   int path_len);
+
+void
+glusterd_svc_build_gfproxyd_socket_filepath(glusterd_volinfo_t *volinfo,
+                                            char *path, int path_len);
+
+void
+glusterd_svc_build_gfproxyd_pidfile(glusterd_volinfo_t *volinfo, char *path,
+                                    int path_len);
+
+void
+glusterd_svc_build_gfproxyd_volfile_path(glusterd_volinfo_t *volinfo,
+                                         char *path, int path_len);
+
+void
+glusterd_svc_build_gfproxyd_logdir(char *logdir, char *volname, size_t len);
+
+void
+glusterd_svc_build_gfproxyd_logfile(char *logfile, char *logdir, size_t len);
+
+int
+glusterd_svc_check_gfproxyd_volfile_identical(char *svc_name,
+                                              glusterd_volinfo_t *volinfo,
+                                              gf_boolean_t *identical);
+int
+glusterd_svc_check_gfproxyd_topology_identical(char *svc_name,
+                                               glusterd_volinfo_t *volinfo,
+                                               gf_boolean_t *identical);
+int
+glusterd_is_gfproxyd_enabled(glusterd_volinfo_t *volinfo);
+
+glusterd_volinfo_t *
+glusterd_gfproxyd_volinfo_from_svc(glusterd_svc_t *svc);
+#endif
diff --git a/xlators/mgmt/glusterd/src/glusterd-gfproxyd-svc.c b/xlators/mgmt/glusterd/src/glusterd-gfproxyd-svc.c
new file mode 100644
index 00000000000..a0bfea41f0f
--- /dev/null
+++ b/xlators/mgmt/glusterd/src/glusterd-gfproxyd-svc.c
@@ -0,0 +1,478 @@
+/*
+   Copyright (c) 2014 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
+
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
+
+#include <glusterfs/globals.h>
+#include <glusterfs/run.h>
+#include "glusterd.h"
+#include "glusterd-utils.h"
+#include "glusterd-volgen.h"
+#include "glusterd-gfproxyd-svc.h"
+#include "glusterd-messages.h"
+#include "glusterd-svc-helper.h"
+#include "glusterd-svc-mgmt.h"
+#include "glusterd-gfproxyd-svc-helper.h"
+#include <glusterfs/syscall.h>
+
+void
+glusterd_gfproxydsvc_build(glusterd_svc_t *svc)
+{
+    svc->manager = glusterd_gfproxydsvc_manager;
+    svc->start = glusterd_gfproxydsvc_start;
+    svc->stop = glusterd_gfproxydsvc_stop;
+    svc->reconfigure = glusterd_gfproxydsvc_reconfigure;
+}
+
+int
+glusterd_gfproxydsvc_stop(glusterd_svc_t *svc, int sig)
+{
+    glusterd_volinfo_t *volinfo = NULL;
+    int ret = 0;
+
+    ret = glusterd_svc_stop(svc, sig);
+    if (ret)
+        goto out;
+
+    volinfo = glusterd_gfproxyd_volinfo_from_svc(svc);
+    volinfo->gfproxyd.port = 0;
+
+out:
+    return ret;
+}
+
+int
+glusterd_gfproxydsvc_init(glusterd_volinfo_t *volinfo)
+{
+    int ret = -1;
+    char rundir[PATH_MAX] = {
+        0,
+    };
+    char sockpath[PATH_MAX] = {
+        0,
+    };
+    char pidfile[PATH_MAX] = {
+        0,
+    };
+    char volfile[PATH_MAX] = {
+        0,
+    };
+    char logdir[PATH_MAX] = {
+        0,
+    };
+    char logfile[PATH_MAX] = {
+        0,
+    };
+    char volfileid[256] = {0};
+    glusterd_svc_t *svc = NULL;
+    glusterd_conf_t *priv = NULL;
+    glusterd_conn_notify_t notify = NULL;
+    xlator_t *this = NULL;
+    char *volfileserver = NULL;
+    int32_t len = 0;
+
+    this = THIS;
+    GF_VALIDATE_OR_GOTO("glusterd", this, out);
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, priv, out);
+
+    svc = &(volinfo->gfproxyd.svc);
+
+    ret = snprintf(svc->name, sizeof(svc->name), "%s", gfproxyd_svc_name);
+    if (ret < 0)
+        goto out;
+
+    notify = glusterd_svc_common_rpc_notify;
+
+    glusterd_svc_build_gfproxyd_rundir(volinfo, rundir, sizeof(rundir));
+    glusterd_svc_create_rundir(rundir);
+
+    /* Initialize the connection mgmt */
+    glusterd_svc_build_gfproxyd_socket_filepath(volinfo, sockpath,
+                                                sizeof(sockpath));
+    ret = glusterd_conn_init(&(svc->conn), sockpath, 600, notify);
+    if (ret)
+        goto out;
+
+    /* Initialize the process mgmt */
+    glusterd_svc_build_gfproxyd_pidfile(volinfo, pidfile, sizeof(pidfile));
+    glusterd_svc_build_gfproxyd_volfile_path(volinfo, volfile, sizeof(volfile));
+    glusterd_svc_build_gfproxyd_logdir(logdir, volinfo->volname,
+                                       sizeof(logdir));
+    ret = mkdir_p(logdir, 0755, _gf_true);
+    if ((ret == -1) && (EEXIST != errno)) {
+        gf_msg(this->name, GF_LOG_ERROR, errno, GD_MSG_CREATE_DIR_FAILED,
+               "Unable to create logdir %s", logdir);
+        goto out;
+    }
+    glusterd_svc_build_gfproxyd_logfile(logfile, logdir, sizeof(logfile));
+    len = snprintf(volfileid, sizeof(volfileid), "gfproxyd/%s",
+                   volinfo->volname);
+    if ((len < 0) || (len >= sizeof(volfileid))) {
+        ret = -1;
+        goto out;
+    }
+
+    if (dict_get_strn(this->options, "transport.socket.bind-address",
+                      SLEN("transport.socket.bind-address"),
+                      &volfileserver) != 0) {
+        volfileserver = "localhost";
+    }
+    ret = glusterd_proc_init(&(svc->proc), gfproxyd_svc_name, pidfile, logdir,
+                             logfile, volfile, volfileid, volfileserver);
+    if (ret)
+        goto out;
+
+out:
+    gf_msg_debug(this ? this->name : "glusterd", 0, "Returning %d", ret);
+    return ret;
+}
+
+static int
+glusterd_gfproxydsvc_create_volfile(glusterd_volinfo_t *volinfo)
+{
+    int ret = -1;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_VALIDATE_OR_GOTO("glusterd", this, out);
+
+    ret = glusterd_generate_gfproxyd_volfile(volinfo);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_VOLFILE_CREATE_FAIL,
+               "Failed to create volfile");
+        goto out;
+    }
+
+out:
+    gf_msg_debug(this ? this->name : "glusterd", 0, "Returning %d", ret);
+
+    return ret;
+}
+
+int
+glusterd_gfproxydsvc_manager(glusterd_svc_t *svc, void *data, int flags)
+{
+    int ret = -1;
+    glusterd_volinfo_t *volinfo = NULL;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_VALIDATE_OR_GOTO("glusterd", this, out);
+
+    volinfo = data;
+    GF_VALIDATE_OR_GOTO(this->name, data, out);
+
+    if (!svc->inited) {
+        ret = glusterd_gfproxydsvc_init(volinfo);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_FAILED_INIT_QUOTASVC,
+                   "Failed to init "
+                   "gfproxyd service");
+            goto out;
+        } else {
+            svc->inited = _gf_true;
+            gf_msg_debug(this->name, 0,
+                         "gfproxyd service "
+                         "initialized");
+        }
+    }
+
+    ret = glusterd_is_gfproxyd_enabled(volinfo);
+    if (ret == -1) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_VOLINFO_GET_FAIL,
+               "Failed to read volume "
+               "options");
+        goto out;
+    }
+
+    if (ret) {
+        if (!glusterd_is_volume_started(volinfo)) {
+            if (glusterd_proc_is_running(&svc->proc)) {
+                ret = svc->stop(svc, SIGTERM);
+                if (ret)
+                    gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SNAPD_STOP_FAIL,
+                           "Couldn't stop gfproxyd for "
+                           "volume: %s",
+                           volinfo->volname);
+            } else {
+                /* Since gfproxyd is not running set ret to 0 */
+                ret = 0;
+            }
+            goto out;
+        }
+
+        ret = glusterd_gfproxydsvc_create_volfile(volinfo);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SNAPD_CREATE_FAIL,
+                   "Couldn't create "
+                   "gfroxyd volfile for volume: %s",
+                   volinfo->volname);
+            goto out;
+        }
+        ret = svc->stop(svc, SIGTERM);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SNAPD_START_FAIL,
+                   "Couldn't stop "
+                   "gfproxyd for volume: %s",
+                   volinfo->volname);
+            goto out;
+        }
+
+        ret = svc->start(svc, flags);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SNAPD_START_FAIL,
+                   "Couldn't start "
+                   "gfproxyd for volume: %s",
+                   volinfo->volname);
+            goto out;
+        }
+
+        glusterd_volinfo_ref(volinfo);
+        ret = glusterd_conn_connect(&(svc->conn));
+        if (ret) {
+            glusterd_volinfo_unref(volinfo);
+            volinfo = NULL;
+            goto out;
+        }
+
+    } else if (glusterd_proc_is_running(&svc->proc)) {
+        ret = svc->stop(svc, SIGTERM);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SNAPD_STOP_FAIL,
+                   "Couldn't stop gfproxyd for volume: %s", volinfo->volname);
+            goto out;
+        }
+    }
+
+out:
+    if (ret) {
+        if (volinfo) {
+            gf_event(EVENT_SVC_MANAGER_FAILED, "volume=%s;svc_name=%s",
+                     volinfo->volname, svc->name);
+        }
+    }
+
+    gf_msg_debug("glusterd", 0, "Returning %d", ret);
+
+    return ret;
+}
+
+int
+glusterd_gfproxydsvc_start(glusterd_svc_t *svc, int flags)
+{
+    int ret = -1;
+    runner_t runner = {
+        0,
+    };
+    glusterd_conf_t *priv = NULL;
+    xlator_t *this = NULL;
+    char valgrind_logfile[PATH_MAX] = {0};
+    int gfproxyd_port = 0;
+    char msg[1024] = {
+        0,
+    };
+    char gfproxyd_id[PATH_MAX] = {
+        0,
+    };
+    glusterd_volinfo_t *volinfo = NULL;
+    char *localtime_logging = NULL;
+    int32_t len = 0;
+
+    this = THIS;
+    GF_VALIDATE_OR_GOTO("glusterd", this, out);
+
+    priv = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, priv, out);
+
+    volinfo = glusterd_gfproxyd_volinfo_from_svc(svc);
+    if (!volinfo)
+        goto out;
+
+    ret = sys_access(svc->proc.volfile, F_OK);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_DEBUG, 0, GD_MSG_VOLINFO_GET_FAIL,
+               "gfproxyd Volfile %s is not present", svc->proc.volfile);
+        ret = glusterd_gfproxydsvc_create_volfile(volinfo);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_VOLFILE_CREATE_FAIL,
+                   "Couldn't create "
+                   "gfproxyd volfile for volume: %s",
+                   volinfo->volname);
+            goto out;
+        }
+    }
+    runinit(&runner);
+
+    if (this->ctx->cmd_args.vgtool != _gf_none) {
+        len = snprintf(valgrind_logfile, PATH_MAX, "%s/valgrind-%s",
+                       svc->proc.logdir, svc->proc.logfile);
+        if ((len < 0) || (len >= PATH_MAX)) {
+            ret = -1;
+            goto out;
+        }
+
+        if (this->ctx->cmd_args.vgtool == _gf_memcheck)
+            runner_add_args(&runner, "valgrind", "--leak-check=full",
+                            "--trace-children=yes", "--track-origins=yes",
+                            NULL);
+        else
+            runner_add_args(&runner, "valgrind", "--tool=drd", NULL);
+
+        runner_argprintf(&runner, "--log-file=%s", valgrind_logfile);
+    }
+
+    snprintf(gfproxyd_id, sizeof(gfproxyd_id), "gfproxyd-%s", volinfo->volname);
+    runner_add_args(&runner, SBIN_DIR "/glusterfsd", "-s",
+                    svc->proc.volfileserver, "--volfile-id",
+                    svc->proc.volfileid, "-p", svc->proc.pidfile, "-l",
+                    svc->proc.logfile, "--brick-name", gfproxyd_id, "-S",
+                    svc->conn.sockpath, NULL);
+
+    if (volinfo->memory_accounting)
+        runner_add_arg(&runner, "--mem-accounting");
+    if (dict_get_strn(priv->opts, GLUSTERD_LOCALTIME_LOGGING_KEY,
+                      SLEN(GLUSTERD_LOCALTIME_LOGGING_KEY),
+                      &localtime_logging) == 0) {
+        if (strcmp(localtime_logging, "enable") == 0)
+            runner_add_arg(&runner, "--localtime-logging");
+    }
+
+    gfproxyd_port = pmap_assign_port(this, volinfo->gfproxyd.port, gfproxyd_id);
+    volinfo->gfproxyd.port = gfproxyd_port;
+
+    runner_add_arg(&runner, "--brick-port");
+    runner_argprintf(&runner, "%d", gfproxyd_port);
+    runner_add_arg(&runner, "--xlator-option");
+    runner_argprintf(&runner, "%s-server.listen-port=%d", volinfo->volname,
+                     gfproxyd_port);
+
+    snprintf(msg, sizeof(msg), "Starting the gfproxyd service for volume %s",
+             volinfo->volname);
+    runner_log(&runner, this->name, GF_LOG_DEBUG, msg);
+
+    if (flags == PROC_START_NO_WAIT) {
+        ret = runner_run_nowait(&runner);
+    } else {
+        synclock_unlock(&priv->big_lock);
+        {
+            ret = runner_run(&runner);
+        }
+        synclock_lock(&priv->big_lock);
+    }
+
+out:
+    return ret;
+}
+
+int
+glusterd_gfproxydsvc_restart()
+{
+    glusterd_volinfo_t *volinfo = NULL;
+    glusterd_volinfo_t *tmp = NULL;
+    int ret = -1;
+    xlator_t *this = THIS;
+    glusterd_conf_t *conf = NULL;
+    glusterd_svc_t *svc = NULL;
+
+    GF_VALIDATE_OR_GOTO("glusterd", this, out);
+
+    conf = this->private;
+    GF_VALIDATE_OR_GOTO(this->name, conf, out);
+
+    cds_list_for_each_entry_safe(volinfo, tmp, &conf->volumes, vol_list)
+    {
+        /* Start per volume gfproxyd svc */
+        if (volinfo->status == GLUSTERD_STATUS_STARTED) {
+            svc = &(volinfo->gfproxyd.svc);
+            ret = svc->manager(svc, volinfo, PROC_START_NO_WAIT);
+            if (ret) {
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_SNAPD_START_FAIL,
+                       "Couldn't resolve gfproxyd for "
+                       "vol: %s on restart",
+                       volinfo->volname);
+                gf_event(EVENT_SVC_MANAGER_FAILED, "volume=%s;svc_name=%s",
+                         volinfo->volname, svc->name);
+                goto out;
+            }
+        }
+    }
+out:
+    return ret;
+}
+
+int
+glusterd_gfproxydsvc_reconfigure(void *data)
+{
+    int ret = -1;
+    xlator_t *this = NULL;
+    gf_boolean_t identical = _gf_false;
+    glusterd_volinfo_t *volinfo = NULL;
+
+    volinfo = data;
+
+    this = THIS;
+    GF_VALIDATE_OR_GOTO("glusterd", this, out);
+
+    if (!volinfo->gfproxyd.svc.inited)
+        goto manager;
+
+    if (!glusterd_is_gfproxyd_enabled(volinfo))
+        goto manager;
+    else if (!glusterd_proc_is_running(&volinfo->gfproxyd.svc.proc))
+        goto manager;
+
+    /*
+     * Check both OLD and NEW volfiles, if they are SAME by size
+     * and cksum i.e. "character-by-character". If YES, then
+     * NOTHING has been changed, just return.
+     */
+    ret = glusterd_svc_check_gfproxyd_volfile_identical(
+        volinfo->gfproxyd.svc.name, volinfo, &identical);
+    if (ret)
+        goto out;
+
+    if (identical) {
+        ret = 0;
+        goto out;
+    }
+
+    /*
+     * They are not identical. Find out if the topology is changed
+     * OR just the volume options. If just the options which got
+     * changed, then inform the xlator to reconfigure the options.
+     */
+    identical = _gf_false; /* RESET the FLAG */
+    ret = glusterd_svc_check_gfproxyd_topology_identical(
+        volinfo->gfproxyd.svc.name, volinfo, &identical);
+    if (ret)
+        goto out;
+
+    /* Topology is not changed, but just the options. But write the
+     * options to gfproxyd volfile, so that gfproxyd will be reconfigured.
+     */
+    if (identical) {
+        ret = glusterd_gfproxydsvc_create_volfile(volinfo);
+        if (ret == 0) { /* Only if above PASSES */
+            ret = glusterd_fetchspec_notify(this);
+        }
+        goto out;
+    }
+manager:
+    /*
+     * gfproxyd volfile's topology has been changed. gfproxyd server needs
+     * to be RESTARTED to ACT on the changed volfile.
+     */
+    ret = volinfo->gfproxyd.svc.manager(&(volinfo->gfproxyd.svc), volinfo,
+                                        PROC_START_NO_WAIT);
+
+out:
+    gf_msg_debug("glusterd", 0, "Returning %d", ret);
+    return ret;
+}
diff --git a/xlators/mgmt/glusterd/src/glusterd-gfproxyd-svc.h b/xlators/mgmt/glusterd/src/glusterd-gfproxyd-svc.h
new file mode 100644
index 00000000000..d396b4015f3
--- /dev/null
+++ b/xlators/mgmt/glusterd/src/glusterd-gfproxyd-svc.h
@@ -0,0 +1,47 @@
+/*
+  Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _GLUSTERD_GFPROXYD_SVC_H_
+#define _GLUSTERD_GFPROXYD_SVC_H_
+
+#include "glusterd-svc-mgmt.h"
+
+#define gfproxyd_svc_name "gfproxyd"
+
+struct glusterd_gfproxydsvc_ {
+    glusterd_svc_t svc;
+    gf_store_handle_t *handle;
+    int port;
+};
+
+typedef struct glusterd_gfproxydsvc_ glusterd_gfproxydsvc_t;
+
+void
+glusterd_gfproxydsvc_build(glusterd_svc_t *svc);
+
+int
+glusterd_gfproxydsvc_manager(glusterd_svc_t *svc, void *data, int flags);
+
+int
+glusterd_gfproxydsvc_start(glusterd_svc_t *svc, int flags);
+
+int
+glusterd_gfproxydsvc_stop(glusterd_svc_t *svc, int sig);
+
+int
+glusterd_gfproxydsvc_reconfigure();
+
+void
+glusterd_gfproxydsvc_build_volfile_path(char *server, char *workdir,
+                                        char *volfile, size_t len);
+
+int
+glusterd_gfproxydsvc_restart();
+#endif
diff --git a/xlators/mgmt/glusterd/src/glusterd-handler.c b/xlators/mgmt/glusterd/src/glusterd-handler.c
index 448dd65348d..1b21c40596d 100644
--- a/xlators/mgmt/glusterd/src/glusterd-handler.c
+++ b/xlators/mgmt/glusterd/src/glusterd-handler.c
@@ -1,3187 +1,6713 @@
 /*
-  Copyright (c) 2006-2009 Gluster, Inc. <http://www.gluster.com>
-  This file is part of GlusterFS.
-
-  GlusterFS is free software; you can redistribute it and/or modify
-  it under the terms of the GNU Affero General Public License as published
-  by the Free Software Foundation; either version 3 of the License,
-  or (at your option) any later version.
-
-  GlusterFS is distributed in the hope that it will be useful, but
-  WITHOUT ANY WARRANTY; without even the implied warranty of
-  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-  Affero General Public License for more details.
-
-  You should have received a copy of the GNU Affero General Public License
-  along with this program.  If not, see
-  <http://www.gnu.org/licenses/>.
-*/
+   Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
+   This file is part of GlusterFS.
 
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
+   This file is licensed to you under your choice of the GNU Lesser
+   General Public License, version 3 or any later version (LGPLv3 or
+   later), or the GNU General Public License, version 2 (GPLv2), in all
+   cases as published by the Free Software Foundation.
+*/
 #include <inttypes.h>
 
-
-#include "globals.h"
-#include "glusterfs.h"
-#include "compat.h"
-#include "dict.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/dict.h>
 #include "protocol-common.h"
-#include "xlator.h"
-#include "logging.h"
-#include "timer.h"
-#include "defaults.h"
-#include "compat.h"
-#include "compat-errno.h"
-#include "statedump.h"
+#include <glusterfs/xlator.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/syscall.h>
+#include <glusterfs/timer.h>
+#include <glusterfs/defaults.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/statedump.h>
+#include <glusterfs/run.h>
 #include "glusterd-mem-types.h"
 #include "glusterd.h"
 #include "glusterd-sm.h"
 #include "glusterd-op-sm.h"
 #include "glusterd-utils.h"
+#include "glusterd-mgmt.h"
+#include "glusterd-server-quorum.h"
 #include "glusterd-store.h"
+#include "glusterd-locks.h"
+#include "glusterd-snapshot-utils.h"
+#include "glusterd-geo-rep.h"
 
-#include "glusterd1.h"
-#include "cli1.h"
-#include "rpc-clnt.h"
 #include "glusterd1-xdr.h"
+#include "cli1-xdr.h"
+#include "xdr-generic.h"
+#include "rpc-clnt.h"
 #include "glusterd-volgen.h"
+#include "glusterd-mountbroker.h"
+#include "glusterd-messages.h"
+#include "glusterd-errno.h"
 
 #include <sys/resource.h>
 #include <inttypes.h>
 
-#include "defaults.c"
-#include "common-utils.h"
-
-#define glusterd_start_volume(req, volname, flags) \
-        glusterd_volume_txn (req, volname, flags, GD_OP_START_VOLUME)
+#include <glusterfs/common-utils.h>
 
-#define glusterd_stop_volume(req, volname, flags) \
-        glusterd_volume_txn (req, volname, flags, GD_OP_STOP_VOLUME)
+#include "glusterd-syncop.h"
+#include "glusterd-messages.h"
 
-static int
-glusterd_handle_friend_req (rpcsvc_request_t *req, uuid_t  uuid,
-                            char *hostname, int port, dict_t *dict)
-{
-        int                             ret = -1;
-        glusterd_peerinfo_t             *peerinfo = NULL;
-        glusterd_friend_sm_event_t      *event = NULL;
-        glusterd_friend_req_ctx_t       *ctx = NULL;
-        glusterd_peerctx_args_t         args = {0};
-        char                            rhost[UNIX_PATH_MAX + 1] = {0};
-        uuid_t                          friend_uuid = {0};
-        char                            uuid_str[50] = {0,};
-
-        uuid_unparse (uuid, uuid_str);
-        uuid_parse (uuid_str, friend_uuid);
-        if (!port)
-                port = 6969; // TODO: use define values.
-
-        ret = glusterd_remote_hostname_get (req, rhost, sizeof (rhost));
-        ret = glusterd_friend_find (uuid, rhost, &peerinfo);
-
-        if (ret) {
-                gf_log ("glusterd", GF_LOG_NORMAL,
-                        " for host: %s (%d)", rhost, port);
-                args.mode = GD_MODE_SWITCH_ON;
-                args.req  = NULL;
-                ret = glusterd_friend_add (rhost, port,
-                                           GD_FRIEND_STATE_DEFAULT,
-                                           &friend_uuid, NULL, &peerinfo, 0,
-                                           &args);
+extern glusterd_op_info_t opinfo;
+static int volcount;
 
-        }
-
-        ret = glusterd_friend_sm_new_event
-                        (GD_FRIEND_EVENT_RCVD_FRIEND_REQ, &event);
-
-        if (ret) {
-                gf_log ("", GF_LOG_ERROR, "event generation failed: %d", ret);
-                return ret;
-        }
-
-        event->peerinfo = peerinfo;
+int
+glusterd_big_locked_notify(struct rpc_clnt *rpc, void *mydata,
+                           rpc_clnt_event_t event, void *data,
+                           rpc_clnt_notify_t notify_fn)
+{
+    glusterd_conf_t *priv = THIS->private;
+    int ret = -1;
 
-        ctx = GF_CALLOC (1, sizeof (*ctx), gf_gld_mt_friend_req_ctx_t);
+    synclock_lock(&priv->big_lock);
+    ret = notify_fn(rpc, mydata, event, data);
+    synclock_unlock(&priv->big_lock);
 
-        if (!ctx) {
-                gf_log ("", GF_LOG_ERROR, "Unable to allocate memory");
-                ret = -1;
-                goto out;
-        }
-
-        uuid_copy (ctx->uuid, uuid);
-        if (hostname)
-                ctx->hostname = gf_strdup (hostname);
-        ctx->req = req;
-        ctx->vols = dict;
+    return ret;
+}
 
-        event->ctx = ctx;
+int
+glusterd_big_locked_handler(rpcsvc_request_t *req, rpcsvc_actor actor_fn)
+{
+    glusterd_conf_t *priv = THIS->private;
+    int ret = -1;
 
-        ret = glusterd_friend_sm_inject_event (event);
+    synclock_lock(&priv->big_lock);
+    ret = actor_fn(req);
+    synclock_unlock(&priv->big_lock);
 
-        if (ret) {
-                gf_log ("glusterd", GF_LOG_ERROR, "Unable to inject event %d, "
-                        "ret = %d", event->event, ret);
-                goto out;
-        }
+    return ret;
+}
 
-        ret = 0;
+static int
+glusterd_handle_friend_req(rpcsvc_request_t *req, uuid_t uuid, char *hostname,
+                           int port, gd1_mgmt_friend_req *friend_req)
+{
+    int ret = -1;
+    glusterd_peerinfo_t *peerinfo = NULL;
+    glusterd_friend_sm_event_t *event = NULL;
+    glusterd_friend_req_ctx_t *ctx = NULL;
+    char rhost[UNIX_PATH_MAX + 1] = {0};
+    dict_t *dict = NULL;
+
+    if (!port)
+        port = GF_DEFAULT_BASE_PORT;
+
+    ret = glusterd_remote_hostname_get(req, rhost, sizeof(rhost));
+
+    ctx = GF_CALLOC(1, sizeof(*ctx), gf_gld_mt_friend_req_ctx_t);
+    dict = dict_new();
+
+    RCU_READ_LOCK;
+
+    peerinfo = glusterd_peerinfo_find(uuid, rhost);
+
+    if (peerinfo == NULL) {
+        gf_event(EVENT_PEER_REJECT, "peer=%s", hostname);
+        ret = glusterd_xfer_friend_add_resp(req, hostname, rhost, port, -1,
+                                            GF_PROBE_UNKNOWN_PEER);
+        if (friend_req->vols.vols_val) {
+            free(friend_req->vols.vols_val);
+            friend_req->vols.vols_val = NULL;
+        }
+        goto out;
+    }
+
+    ret = glusterd_friend_sm_new_event(GD_FRIEND_EVENT_RCVD_FRIEND_REQ, &event);
+
+    if (ret) {
+        gf_msg("glusterd", GF_LOG_ERROR, 0, GD_MSG_EVENT_NEW_GET_FAIL,
+               "event generation failed: %d", ret);
+        goto out;
+    }
+
+    event->peername = gf_strdup(peerinfo->hostname);
+    gf_uuid_copy(event->peerid, peerinfo->uuid);
+
+    if (!ctx) {
+        gf_msg("glusterd", GF_LOG_ERROR, ENOMEM, GD_MSG_NO_MEMORY,
+               "Unable to allocate memory");
+        ret = -1;
+        goto out;
+    }
+
+    gf_uuid_copy(ctx->uuid, uuid);
+    if (hostname)
+        ctx->hostname = gf_strdup(hostname);
+    ctx->req = req;
+
+    if (!dict) {
+        gf_smsg("glusterd", GF_LOG_ERROR, errno, GD_MSG_DICT_CREATE_FAIL, NULL);
+        ret = -1;
+        goto out;
+    }
+
+    ret = dict_unserialize(friend_req->vols.vols_val, friend_req->vols.vols_len,
+                           &dict);
+
+    if (ret) {
+        gf_smsg("glusterd", GF_LOG_ERROR, 0, GD_MSG_DICT_UNSERIALIZE_FAIL,
+                NULL);
+        goto out;
+    } else
+        dict->extra_stdfree = friend_req->vols.vols_val;
+
+    ctx->vols = dict;
+    event->ctx = ctx;
+
+    ret = glusterd_friend_sm_inject_event(event);
+    if (ret) {
+        gf_msg("glusterd", GF_LOG_ERROR, 0, GD_MSG_EVENT_INJECT_FAIL,
+               "Unable to inject event %d, "
+               "ret = %d",
+               event->event, ret);
+        goto out;
+    }
+
+    ret = 0;
+    if (peerinfo && (0 == peerinfo->connected))
+        ret = GLUSTERD_CONNECTION_AWAITED;
 
 out:
-        if (0 != ret) {
-                if (ctx && ctx->hostname)
-                        GF_FREE (ctx->hostname);
-                if (ctx && ctx->vols)
-                        dict_destroy (ctx->vols);
-                if (ctx)
-                        GF_FREE (ctx);
+    RCU_READ_UNLOCK;
+
+    if (ret && (ret != GLUSTERD_CONNECTION_AWAITED)) {
+        if (ctx && ctx->hostname)
+            GF_FREE(ctx->hostname);
+        GF_FREE(ctx);
+        if (dict) {
+            if ((!dict->extra_stdfree) && friend_req->vols.vols_val)
+                free(friend_req->vols.vols_val);
+            dict_unref(dict);
+        } else {
+            free(friend_req->vols.vols_val);
         }
+        if (event)
+            GF_FREE(event->peername);
+        GF_FREE(event);
+    }
 
-        if (0 == peerinfo->connected)
-                ret = GLUSTERD_CONNECTION_AWAITED;
-        return ret;
+    return ret;
 }
 
-
 static int
-glusterd_handle_unfriend_req (rpcsvc_request_t *req, uuid_t  uuid,
-                              char *hostname, int port)
+glusterd_handle_unfriend_req(rpcsvc_request_t *req, uuid_t uuid, char *hostname,
+                             int port)
 {
-        int                             ret = -1;
-        glusterd_peerinfo_t             *peerinfo = NULL;
-        glusterd_friend_sm_event_t      *event = NULL;
-        glusterd_friend_req_ctx_t       *ctx = NULL;
+    int ret = -1;
+    glusterd_peerinfo_t *peerinfo = NULL;
+    glusterd_friend_sm_event_t *event = NULL;
+    glusterd_friend_req_ctx_t *ctx = NULL;
 
-        if (!port)
-                port = 6969; //TODO: use define'd macro
+    if (!port)
+        port = GF_DEFAULT_BASE_PORT;
 
-        ret = glusterd_friend_find (uuid, hostname, &peerinfo);
+    ctx = GF_CALLOC(1, sizeof(*ctx), gf_gld_mt_friend_req_ctx_t);
 
-        if (ret) {
-                gf_log ("glusterd", GF_LOG_NORMAL,
-                         "Unable to find peer");
+    RCU_READ_LOCK;
 
-        }
+    peerinfo = glusterd_peerinfo_find(uuid, hostname);
 
-        ret = glusterd_friend_sm_new_event
-                        (GD_FRIEND_EVENT_RCVD_REMOVE_FRIEND, &event);
+    if (peerinfo == NULL) {
+        RCU_READ_UNLOCK;
+        gf_msg("glusterd", GF_LOG_CRITICAL, 0, GD_MSG_REQ_FROM_UNKNOWN_PEER,
+               "Received remove-friend from unknown peer %s", hostname);
+        ret = glusterd_xfer_friend_remove_resp(req, hostname, port);
+        goto out;
+    }
 
-        if (ret) {
-                gf_log ("", GF_LOG_ERROR, "event generation failed: %d", ret);
-                return ret;
-        }
+    ret = glusterd_friend_sm_new_event(GD_FRIEND_EVENT_RCVD_REMOVE_FRIEND,
+                                       &event);
 
-        event->peerinfo = peerinfo;
+    if (ret) {
+        RCU_READ_UNLOCK;
+        gf_msg("glusterd", GF_LOG_ERROR, 0, GD_MSG_EVENT_NEW_GET_FAIL,
+               "event generation failed: %d", ret);
+        goto out;
+    }
 
-        ctx = GF_CALLOC (1, sizeof (*ctx), gf_gld_mt_friend_req_ctx_t);
+    if (hostname)
+        event->peername = gf_strdup(hostname);
 
-        if (!ctx) {
-                gf_log ("", GF_LOG_ERROR, "Unable to allocate memory");
-                ret = -1;
-                goto out;
-        }
+    gf_uuid_copy(event->peerid, uuid);
 
-        uuid_copy (ctx->uuid, uuid);
-        if (hostname)
-                ctx->hostname = gf_strdup (hostname);
-        ctx->req = req;
+    if (!ctx) {
+        RCU_READ_UNLOCK;
+        ret = -1;
+        gf_msg("glusterd", GF_LOG_ERROR, ENOMEM, GD_MSG_NO_MEMORY,
+               "Unable to allocate memory");
+        goto out;
+    }
 
-        event->ctx = ctx;
+    gf_uuid_copy(ctx->uuid, uuid);
+    if (hostname)
+        ctx->hostname = gf_strdup(hostname);
+    ctx->req = req;
 
-        ret = glusterd_friend_sm_inject_event (event);
+    event->ctx = ctx;
 
-        if (ret) {
-                gf_log ("glusterd", GF_LOG_ERROR, "Unable to inject event %d, "
-                        "ret = %d", event->event, ret);
-                goto out;
-        }
+    ret = glusterd_friend_sm_inject_event(event);
 
-        ret = 0;
+    if (ret) {
+        RCU_READ_UNLOCK;
+        gf_msg("glusterd", GF_LOG_ERROR, 0, GD_MSG_EVENT_INJECT_FAIL,
+               "Unable to inject event %d, "
+               "ret = %d",
+               event->event, ret);
+        goto out;
+    }
+
+    RCU_READ_UNLOCK;
+
+    return 0;
 
 out:
-        if (0 != ret) {
-                if (ctx && ctx->hostname)
-                        GF_FREE (ctx->hostname);
-                if (ctx)
-                        GF_FREE (ctx);
-        }
 
-        return ret;
+    if (0 != ret) {
+        if (ctx && ctx->hostname)
+            GF_FREE(ctx->hostname);
+        GF_FREE(ctx);
+        if (event)
+            GF_FREE(event->peername);
+        GF_FREE(event);
+    }
+
+    return ret;
 }
 
+struct args_pack {
+    dict_t *dict;
+    int vol_count;
+    int opt_count;
+};
+
 static int
-glusterd_add_peer_detail_to_dict (glusterd_peerinfo_t   *peerinfo,
-                                  dict_t  *friends, int   count)
+_build_option_key(dict_t *d, char *k, data_t *v, void *tmp)
 {
+    char reconfig_key[256] = {
+        0,
+    };
+    int keylen;
+    struct args_pack *pack = NULL;
+    int ret = -1;
+    xlator_t *this = NULL;
+    glusterd_conf_t *priv = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    pack = tmp;
+    if (strcmp(k, GLUSTERD_GLOBAL_OPT_VERSION) == 0)
+        return 0;
 
-        int             ret = -1;
-        char            key[256] = {0, };
-
-        GF_ASSERT (peerinfo);
-        GF_ASSERT (friends);
-
-        snprintf (key, 256, "friend%d.uuid", count);
-        uuid_unparse (peerinfo->uuid, peerinfo->uuid_str);
-        ret = dict_set_str (friends, key, peerinfo->uuid_str);
-        if (ret)
-                goto out;
-
-        snprintf (key, 256, "friend%d.hostname", count);
-        ret = dict_set_str (friends, key, peerinfo->hostname);
-        if (ret)
-                goto out;
+    if (priv->op_version > GD_OP_VERSION_MIN) {
+        if ((strcmp(k, "features.limit-usage") == 0) ||
+            (strcmp(k, "features.soft-limit") == 0))
+            return 0;
+    }
+
+    /* snap-max-hard-limit and snap-max-soft-limit are system   *
+     * options set and managed by snapshot config option. Hence *
+     * they should not be displayed in gluster volume info.     *
+     */
+    if ((strcmp(k, "snap-max-hard-limit") == 0) ||
+        (strcmp(k, "snap-max-soft-limit") == 0))
+        return 0;
 
-        snprintf (key, 256, "friend%d.port", count);
-        ret = dict_set_int32 (friends, key, peerinfo->port);
-        if (ret)
-                goto out;
+    keylen = snprintf(reconfig_key, sizeof(reconfig_key), "volume%d.option.%s",
+                      pack->vol_count, k);
+    ret = dict_set_strn(pack->dict, reconfig_key, keylen, v->data);
+    if (0 == ret)
+        pack->opt_count++;
 
-        snprintf (key, 256, "friend%d.state", count);
-        ret = dict_set_str (friends, key,
-                    glusterd_friend_sm_state_name_get(peerinfo->state.state));
-        if (ret)
-                goto out;
+    return 0;
+}
 
-        snprintf (key, 256, "friend%d.connected", count);
-        ret = dict_set_int32 (friends, key, (int32_t)peerinfo->connected);
+int
+glusterd_add_arbiter_info_to_bricks(glusterd_volinfo_t *volinfo,
+                                    dict_t *volumes, int count)
+{
+    char key[64] = {
+        0,
+    };
+    int keylen;
+    int i = 0;
+    int ret = 0;
+
+    if (volinfo->replica_count == 1 || volinfo->arbiter_count != 1)
+        return 0;
+    for (i = 1; i <= volinfo->brick_count; i++) {
+        if (i % volinfo->replica_count != 0)
+            continue;
+        keylen = snprintf(key, sizeof(key), "volume%d.brick%d.isArbiter", count,
+                          i);
+        ret = dict_set_int32n(volumes, key, keylen, 1);
         if (ret)
-                goto out;
-
-out:
-        return ret;
+            return ret;
+    }
+    return 0;
 }
 
 int
-glusterd_add_volume_detail_to_dict (glusterd_volinfo_t *volinfo,
-                                    dict_t  *volumes, int   count)
+glusterd_add_volume_detail_to_dict(glusterd_volinfo_t *volinfo, dict_t *volumes,
+                                   int count)
 {
+    int ret = -1;
+    char key[64] = {
+        0,
+    };
+    int keylen;
+    glusterd_brickinfo_t *brickinfo = NULL;
+    glusterd_brickinfo_t *ta_brickinfo = NULL;
+    char *buf = NULL;
+    int i = 1;
+    dict_t *dict = NULL;
+    glusterd_conf_t *priv = NULL;
+    char *volume_id_str = NULL;
+    struct args_pack pack = {
+        0,
+    };
+    xlator_t *this = NULL;
+    int32_t len = 0;
+
+    char ta_brick[4096] = {
+        0,
+    };
+
+    GF_ASSERT(volinfo);
+    GF_ASSERT(volumes);
+
+    this = THIS;
+    priv = this->private;
+
+    GF_ASSERT(priv);
+
+    keylen = snprintf(key, sizeof(key), "volume%d.name", count);
+    ret = dict_set_strn(volumes, key, keylen, volinfo->volname);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                "Key=%s", key, NULL);
+        goto out;
+    }
+
+    keylen = snprintf(key, sizeof(key), "volume%d.type", count);
+    ret = dict_set_int32n(volumes, key, keylen, volinfo->type);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                "Key=%s", key, NULL);
+        goto out;
+    }
+
+    keylen = snprintf(key, sizeof(key), "volume%d.status", count);
+    ret = dict_set_int32n(volumes, key, keylen, volinfo->status);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                "Key=%s", key, NULL);
+        goto out;
+    }
+
+    keylen = snprintf(key, sizeof(key), "volume%d.brick_count", count);
+    ret = dict_set_int32n(volumes, key, keylen, volinfo->brick_count);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                "Key=%s", key, NULL);
+        goto out;
+    }
+
+    keylen = snprintf(key, sizeof(key), "volume%d.dist_count", count);
+    ret = dict_set_int32n(volumes, key, keylen, volinfo->dist_leaf_count);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                "Key=%s", key, NULL);
+        goto out;
+    }
+
+    keylen = snprintf(key, sizeof(key), "volume%d.stripe_count", count);
+    ret = dict_set_int32n(volumes, key, keylen, volinfo->stripe_count);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                "Key=%s", key, NULL);
+        goto out;
+    }
+
+    keylen = snprintf(key, sizeof(key), "volume%d.replica_count", count);
+    ret = dict_set_int32n(volumes, key, keylen, volinfo->replica_count);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                "Key=%s", key, NULL);
+        goto out;
+    }
+
+    keylen = snprintf(key, sizeof(key), "volume%d.disperse_count", count);
+    ret = dict_set_int32n(volumes, key, keylen, volinfo->disperse_count);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                "Key=%s", key, NULL);
+        goto out;
+    }
+
+    keylen = snprintf(key, sizeof(key), "volume%d.redundancy_count", count);
+    ret = dict_set_int32n(volumes, key, keylen, volinfo->redundancy_count);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                "Key=%s", key, NULL);
+        goto out;
+    }
+
+    keylen = snprintf(key, sizeof(key), "volume%d.arbiter_count", count);
+    ret = dict_set_int32n(volumes, key, keylen, volinfo->arbiter_count);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                "Key=%s", key, NULL);
+        goto out;
+    }
+
+    keylen = snprintf(key, sizeof(key), "volume%d.transport", count);
+    ret = dict_set_int32n(volumes, key, keylen, volinfo->transport_type);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                "Key=%s", key, NULL);
+        goto out;
+    }
+
+    keylen = snprintf(key, sizeof(key), "volume%d.thin_arbiter_count", count);
+    ret = dict_set_int32n(volumes, key, keylen, volinfo->thin_arbiter_count);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                "Key=%s", key, NULL);
+        goto out;
+    }
+
+    volume_id_str = gf_strdup(uuid_utoa(volinfo->volume_id));
+    if (!volume_id_str) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_STRDUP_FAILED, NULL);
+        goto out;
+    }
+
+    keylen = snprintf(key, sizeof(key), "volume%d.volume_id", count);
+    ret = dict_set_dynstrn(volumes, key, keylen, volume_id_str);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                "Key=%s", key, NULL);
+        goto out;
+    }
+
+    keylen = snprintf(key, sizeof(key), "volume%d.rebalance", count);
+    ret = dict_set_int32n(volumes, key, keylen, volinfo->rebal.defrag_cmd);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                "Key=%s", key, NULL);
+        goto out;
+    }
+
+    keylen = snprintf(key, sizeof(key), "volume%d.snap_count", count);
+    ret = dict_set_int32n(volumes, key, keylen, volinfo->snap_count);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                "Key=%s", key, NULL);
+        goto out;
+    }
+
+    cds_list_for_each_entry(brickinfo, &volinfo->bricks, brick_list)
+    {
+        char brick[1024] = {
+            0,
+        };
+        char brick_uuid[64] = {
+            0,
+        };
+        len = snprintf(brick, sizeof(brick), "%s:%s", brickinfo->hostname,
+                       brickinfo->path);
+        if ((len < 0) || (len >= sizeof(brick))) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_COPY_FAIL, NULL);
+            ret = -1;
+            goto out;
+        }
+        buf = gf_strdup(brick);
+        keylen = snprintf(key, sizeof(key), "volume%d.brick%d", count, i);
+        ret = dict_set_dynstrn(volumes, key, keylen, buf);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                    "Key=%s", key, NULL);
+            goto out;
+        }
+        keylen = snprintf(key, sizeof(key), "volume%d.brick%d.uuid", count, i);
+        snprintf(brick_uuid, sizeof(brick_uuid), "%s",
+                 uuid_utoa(brickinfo->uuid));
+        buf = gf_strdup(brick_uuid);
+        if (!buf) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_STRDUP_FAILED,
+                    "brick_uuid=%s", brick_uuid, NULL);
+            goto out;
+        }
+        ret = dict_set_dynstrn(volumes, key, keylen, buf);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                    "Key=%s", key, NULL);
+            goto out;
+        }
+
+        i++;
+    }
+    if (volinfo->thin_arbiter_count == 1) {
+        ta_brickinfo = list_first_entry(&volinfo->ta_bricks,
+                                        glusterd_brickinfo_t, brick_list);
+        len = snprintf(ta_brick, sizeof(ta_brick), "%s:%s",
+                       ta_brickinfo->hostname, ta_brickinfo->path);
+        if ((len < 0) || (len >= sizeof(ta_brick))) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_COPY_FAIL, NULL);
+            ret = -1;
+            goto out;
+        }
+        buf = gf_strdup(ta_brick);
+        keylen = snprintf(key, sizeof(key), "volume%d.thin_arbiter_brick",
+                          count);
+        ret = dict_set_dynstrn(volumes, key, keylen, buf);
+        if (ret) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                    "Key=%s", key, NULL);
+            goto out;
+        }
+    }
+
+    ret = glusterd_add_arbiter_info_to_bricks(volinfo, volumes, count);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                GD_MSG_ARBITER_BRICK_SET_INFO_FAIL, NULL);
+        goto out;
+    }
+
+    dict = volinfo->dict;
+    if (!dict) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_CREATE_FAIL, NULL);
+        ret = 0;
+        goto out;
+    }
 
-        int                     ret = -1;
-        char                    key[256] = {0, };
-        glusterd_brickinfo_t    *brickinfo = NULL;
-        char                    *buf = NULL;
-        int                     i = 1;
-
-        GF_ASSERT (volinfo);
-        GF_ASSERT (volumes);
-
-        snprintf (key, 256, "volume%d.name", count);
-        ret = dict_set_str (volumes, key, volinfo->volname);
-        if (ret)
-                goto out;
-
-        snprintf (key, 256, "volume%d.type", count);
-        ret = dict_set_int32 (volumes, key, volinfo->type);
-        if (ret)
-                goto out;
+    pack.dict = volumes;
+    pack.vol_count = count;
+    pack.opt_count = 0;
+    dict_foreach(dict, _build_option_key, (void *)&pack);
+    dict_foreach(priv->opts, _build_option_key, &pack);
 
-        snprintf (key, 256, "volume%d.status", count);
-        ret = dict_set_int32 (volumes, key, volinfo->status);
-        if (ret)
-                goto out;
+    keylen = snprintf(key, sizeof(key), "volume%d.opt_count", pack.vol_count);
+    ret = dict_set_int32n(volumes, key, keylen, pack.opt_count);
+out:
+    return ret;
+}
 
-        snprintf (key, 256, "volume%d.brick_count", count);
-        ret = dict_set_int32 (volumes, key, volinfo->brick_count);
-        if (ret)
+int32_t
+glusterd_op_txn_begin(rpcsvc_request_t *req, glusterd_op_t op, void *ctx,
+                      char *err_str, size_t err_len)
+{
+    int32_t ret = -1;
+    dict_t *dict = NULL;
+    xlator_t *this = NULL;
+    glusterd_conf_t *priv = NULL;
+    int32_t locked = 0;
+    char *tmp = NULL;
+    char *volname = NULL;
+    uuid_t *txn_id = NULL;
+    glusterd_op_info_t txn_op_info = {
+        {0},
+    };
+    glusterd_op_sm_event_type_t event_type = GD_OP_EVENT_NONE;
+    uint32_t op_errno = 0;
+    uint32_t timeout = 0;
+
+    GF_ASSERT(req);
+    GF_ASSERT((op > GD_OP_NONE) && (op < GD_OP_MAX));
+    GF_ASSERT(NULL != ctx);
+
+    this = THIS;
+    GF_ASSERT(this);
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    dict = ctx;
+
+    /* Generate a transaction-id for this operation and
+     * save it in the dict. This transaction id distinguishes
+     * each transaction, and helps separate opinfos in the
+     * op state machine. */
+    ret = glusterd_generate_txn_id(dict, &txn_id);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_TRANS_IDGEN_FAIL,
+               "Failed to generate transaction id");
+        goto out;
+    }
+
+    /* Save the MY_UUID as the originator_uuid. This originator_uuid
+     * will be used by is_origin_glusterd() to determine if a node
+     * is the originator node for a command. */
+    ret = glusterd_set_originator_uuid(dict);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_UUID_SET_FAIL,
+               "Failed to set originator_uuid.");
+        goto out;
+    }
+
+    /* Based on the op_version, acquire a cluster or mgmt_v3 lock */
+    if (priv->op_version < GD_OP_VERSION_3_6_0) {
+        ret = glusterd_lock(MY_UUID);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_GLUSTERD_LOCK_FAIL,
+                   "Unable to acquire lock on localhost, ret: %d", ret);
+            snprintf(err_str, err_len,
+                     "Another transaction is in progress. "
+                     "Please try again after some time.");
+            goto out;
+        }
+    } else {
+        /* If no volname is given as a part of the command, locks will
+         * not be held */
+        ret = dict_get_strn(dict, "volname", SLEN("volname"), &tmp);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_INFO, errno, GD_MSG_DICT_GET_FAILED,
+                   "No Volume name present. "
+                   "Locks not being held.");
+            goto local_locking_done;
+        } else {
+            /* Use a copy of volname, as cli response will be
+             * sent before the unlock, and the volname in the
+             * dict, might be removed */
+            volname = gf_strdup(tmp);
+            if (!volname)
                 goto out;
+        }
 
-        snprintf (key, 256, "volume%d.sub_count", count);
-        ret = dict_set_int32 (volumes, key, volinfo->sub_count);
-        if (ret)
-                goto out;
+        /* Cli will add timeout key to dict if the default timeout is
+         * other than 2 minutes. Here we use this value to check whether
+         * mgmt_v3_lock_timeout should be set to default value or we
+         * need to change the value according to timeout value
+         * i.e, timeout + 120 seconds. */
+        ret = dict_get_uint32(dict, "timeout", &timeout);
+        if (!ret)
+            priv->mgmt_v3_lock_timeout = timeout + 120;
 
-        snprintf (key, 256, "volume%d.transport", count);
-        ret = dict_set_int32 (volumes, key, volinfo->transport_type);
-        if (ret)
-                goto out;
+        ret = glusterd_mgmt_v3_lock(volname, MY_UUID, &op_errno, "vol");
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_MGMTV3_LOCK_GET_FAIL,
+                   "Unable to acquire lock for %s", volname);
+            snprintf(err_str, err_len,
+                     "Another transaction is in progress for %s. "
+                     "Please try again after some time.",
+                     volname);
+            goto out;
+        }
+    }
+
+    locked = 1;
+    gf_msg_debug(this->name, 0, "Acquired lock on localhost");
+
+local_locking_done:
+    /* If no volname is given as a part of the command, locks will
+     * not be held, hence sending stage event. */
+    if (volname || (priv->op_version < GD_OP_VERSION_3_6_0))
+        event_type = GD_OP_EVENT_START_LOCK;
+    else {
+        txn_op_info.state.state = GD_OP_STATE_LOCK_SENT;
+        event_type = GD_OP_EVENT_ALL_ACC;
+    }
+
+    /* Save opinfo for this transaction with the transaction id */
+    glusterd_txn_opinfo_init(&txn_op_info, NULL, &op, ctx, req);
+
+    ret = glusterd_set_txn_opinfo(txn_id, &txn_op_info);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_TRANS_OPINFO_SET_FAIL,
+               "Unable to set transaction's opinfo");
+        if (ctx)
+            dict_unref(ctx);
+        goto out;
+    }
+
+    ret = glusterd_op_sm_inject_event(event_type, txn_id, ctx);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_EVENT_INJECT_FAIL,
+               "Failed to acquire cluster"
+               " lock.");
+        goto out;
+    }
 
-        list_for_each_entry (brickinfo, &volinfo->bricks, brick_list) {
-                char    brick[1024] = {0,};
-                snprintf (key, 256, "volume%d.brick%d", count, i);
-                snprintf (brick, 1024, "%s:%s", brickinfo->hostname,
-                          brickinfo->path);
-                buf = gf_strdup (brick);
-                ret = dict_set_dynstr (volumes, key, buf);
-                if (ret)
-                        goto out;
-                i++;
-        }
 out:
-        return ret;
+    if (locked && ret) {
+        /* Based on the op-version, we release the
+         * cluster or mgmt_v3 lock */
+        if (priv->op_version < GD_OP_VERSION_3_6_0)
+            glusterd_unlock(MY_UUID);
+        else {
+            ret = glusterd_mgmt_v3_unlock(volname, MY_UUID, "vol");
+            if (ret)
+                gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_MGMTV3_UNLOCK_FAIL,
+                       "Unable to release lock for %s", volname);
+            ret = -1;
+        }
+    }
+
+    if (volname)
+        GF_FREE(volname);
+
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
+    return ret;
 }
 
 int
-glusterd_friend_find (uuid_t uuid, char *hostname,
-                      glusterd_peerinfo_t **peerinfo)
+__glusterd_handle_cluster_lock(rpcsvc_request_t *req)
 {
-        int     ret = -1;
+    dict_t *op_ctx = NULL;
+    int32_t ret = -1;
+    gd1_mgmt_cluster_lock_req lock_req = {
+        {0},
+    };
+    glusterd_op_lock_ctx_t *ctx = NULL;
+    glusterd_op_sm_event_type_t op = GD_OP_EVENT_LOCK;
+    glusterd_op_info_t txn_op_info = {
+        {0},
+    };
+    glusterd_conf_t *priv = NULL;
+    uuid_t *txn_id = NULL;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+    priv = this->private;
+    GF_ASSERT(priv);
+    GF_ASSERT(req);
+
+    txn_id = &priv->global_txn_id;
+
+    ret = xdr_to_generic(req->msg[0], &lock_req,
+                         (xdrproc_t)xdr_gd1_mgmt_cluster_lock_req);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REQ_DECODE_FAIL,
+               "Failed to decode lock "
+               "request received from peer");
+        req->rpc_err = GARBAGE_ARGS;
+        goto out;
+    }
+
+    gf_msg_debug(this->name, 0, "Received LOCK from uuid: %s",
+                 uuid_utoa(lock_req.uuid));
+
+    RCU_READ_LOCK;
+    ret = (glusterd_peerinfo_find_by_uuid(lock_req.uuid) == NULL);
+    RCU_READ_UNLOCK;
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_PEER_NOT_FOUND,
+               "%s doesn't "
+               "belong to the cluster. Ignoring request.",
+               uuid_utoa(lock_req.uuid));
+        ret = -1;
+        goto out;
+    }
+
+    ctx = GF_CALLOC(1, sizeof(*ctx), gf_gld_mt_op_lock_ctx_t);
+
+    if (!ctx) {
+        // respond here
+        return -1;
+    }
+
+    gf_uuid_copy(ctx->uuid, lock_req.uuid);
+    ctx->req = req;
+    ctx->dict = NULL;
+
+    op_ctx = dict_new();
+    if (!op_ctx) {
+        ret = -1;
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, GD_MSG_DICT_CREATE_FAIL,
+               "Unable to set new dict");
+        goto out;
+    }
+
+    glusterd_txn_opinfo_init(&txn_op_info, NULL, &op, op_ctx, req);
+
+    ret = glusterd_set_txn_opinfo(txn_id, &txn_op_info);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_TRANS_OPINFO_SET_FAIL,
+               "Unable to set transaction's opinfo");
+        dict_unref(txn_op_info.op_ctx);
+        goto out;
+    }
+
+    ret = glusterd_op_sm_inject_event(GD_OP_EVENT_LOCK, txn_id, ctx);
+    if (ret)
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_EVENT_INJECT_FAIL,
+               "Failed to inject event GD_OP_EVENT_LOCK");
 
-        if (uuid) {
-                ret = glusterd_friend_find_by_uuid (uuid, peerinfo);
+out:
+    gf_msg_debug(this->name, 0, "Returning %d", ret);
 
-                if (ret) {
-                        gf_log ("glusterd", GF_LOG_NORMAL,
-                                 "Unable to find peer by uuid");
-                } else {
-                        goto out;
-                }
+    glusterd_friend_sm();
+    glusterd_op_sm();
 
-        }
+    if (ret)
+        GF_FREE(ctx);
 
-        if (hostname) {
-                ret = glusterd_friend_find_by_hostname (hostname, peerinfo);
+    return ret;
+}
 
-                if (ret) {
-                        gf_log ("glusterd", GF_LOG_NORMAL,
-                                "Unable to find hostname: %s", hostname);
-                } else {
-                        goto out;
-                }
-        }
+int
+glusterd_handle_cluster_lock(rpcsvc_request_t *req)
+{
+    return glusterd_big_locked_handler(req, __glusterd_handle_cluster_lock);
+}
 
+static int
+glusterd_req_ctx_create(rpcsvc_request_t *rpc_req, int op, uuid_t uuid,
+                        char *buf_val, size_t buf_len,
+                        gf_gld_mem_types_t mem_type,
+                        glusterd_req_ctx_t **req_ctx_out)
+{
+    int ret = -1;
+    char str[50] = {
+        0,
+    };
+    glusterd_req_ctx_t *req_ctx = NULL;
+    dict_t *dict = NULL;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    gf_uuid_unparse(uuid, str);
+    gf_msg_debug(this->name, 0, "Received op from uuid %s", str);
+
+    dict = dict_new();
+    if (!dict) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_CREATE_FAIL, NULL);
+        goto out;
+    }
+
+    req_ctx = GF_CALLOC(1, sizeof(*req_ctx), mem_type);
+    if (!req_ctx) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_NO_MEMORY, NULL);
+        goto out;
+    }
+
+    gf_uuid_copy(req_ctx->uuid, uuid);
+    req_ctx->op = op;
+    ret = dict_unserialize(buf_val, buf_len, &dict);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_WARNING, 0, GD_MSG_DICT_UNSERIALIZE_FAIL,
+                NULL);
+        goto out;
+    }
+
+    req_ctx->dict = dict;
+    req_ctx->req = rpc_req;
+    *req_ctx_out = req_ctx;
+    ret = 0;
 out:
-        return ret;
+    if (ret) {
+        if (dict)
+            dict_unref(dict);
+        GF_FREE(req_ctx);
+    }
+    return ret;
 }
 
 int
-glusterd_handle_cluster_lock (rpcsvc_request_t *req)
+__glusterd_handle_stage_op(rpcsvc_request_t *req)
 {
-        gd1_mgmt_cluster_lock_req       lock_req = {{0},};
-        int32_t                         ret = -1;
-        char                            str[50] = {0,};
-        glusterd_op_lock_ctx_t          *ctx = NULL;
-
-        GF_ASSERT (req);
-
-        if (!gd_xdr_to_mgmt_cluster_lock_req (req->msg[0], &lock_req)) {
-                //failed to decode msg;
-                req->rpc_err = GARBAGE_ARGS;
-                goto out;
+    int32_t ret = -1;
+    glusterd_req_ctx_t *req_ctx = NULL;
+    gd1_mgmt_stage_op_req op_req = {
+        {0},
+    };
+    xlator_t *this = NULL;
+    uuid_t *txn_id = NULL;
+    glusterd_op_info_t txn_op_info = {
+        {0},
+    };
+    glusterd_op_sm_state_info_t state = {
+        0,
+    };
+    glusterd_conf_t *priv = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+    priv = this->private;
+    GF_ASSERT(priv);
+    GF_ASSERT(req);
+
+    txn_id = &priv->global_txn_id;
+
+    ret = xdr_to_generic(req->msg[0], &op_req,
+                         (xdrproc_t)xdr_gd1_mgmt_stage_op_req);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REQ_DECODE_FAIL,
+               "Failed to decode stage "
+               "request received from peer");
+        req->rpc_err = GARBAGE_ARGS;
+        goto out;
+    }
+
+    ret = glusterd_req_ctx_create(req, op_req.op, op_req.uuid,
+                                  op_req.buf.buf_val, op_req.buf.buf_len,
+                                  gf_gld_mt_op_stage_ctx_t, &req_ctx);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REQ_CTX_CREATE_FAIL,
+               "Failed to create req_ctx");
+        goto out;
+    }
+
+    ret = dict_get_bin(req_ctx->dict, "transaction_id", (void **)&txn_id);
+    gf_msg_debug(this->name, 0, "transaction ID = %s", uuid_utoa(*txn_id));
+
+    RCU_READ_LOCK;
+    ret = (glusterd_peerinfo_find_by_uuid(op_req.uuid) == NULL);
+    RCU_READ_UNLOCK;
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_PEER_NOT_FOUND,
+               "%s doesn't "
+               "belong to the cluster. Ignoring request.",
+               uuid_utoa(op_req.uuid));
+        ret = -1;
+        goto out;
+    }
+
+    /* In cases where there is no volname, the receivers won't have a
+     * transaction opinfo created, as for those operations, the locking
+     * phase where the transaction opinfos are created, won't be called.
+     * skip_locking will be true for all such transaction and we clear
+     * the txn_opinfo after the staging phase, except for geo-replication
+     * operations where we need to access txn_opinfo in the later phases also.
+     */
+    ret = glusterd_get_txn_opinfo(txn_id, &txn_op_info);
+    if (ret) {
+        gf_msg_debug(this->name, 0, "No transaction's opinfo set");
+
+        state.state = GD_OP_STATE_LOCKED;
+        glusterd_txn_opinfo_init(&txn_op_info, &state, &op_req.op,
+                                 req_ctx->dict, req);
+
+        if (req_ctx->op != GD_OP_GSYNC_SET)
+            txn_op_info.skip_locking = _gf_true;
+        ret = glusterd_set_txn_opinfo(txn_id, &txn_op_info);
+        if (ret) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_TRANS_OPINFO_SET_FAIL,
+                   "Unable to set transaction's opinfo");
+            dict_unref(req_ctx->dict);
+            goto out;
         }
-        uuid_unparse (lock_req.uuid, str);
-
-        gf_log ("glusterd", GF_LOG_NORMAL,
-                "Received LOCK from uuid: %s", str);
+    }
 
+    ret = glusterd_op_sm_inject_event(GD_OP_EVENT_STAGE_OP, txn_id, req_ctx);
+    if (ret)
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_EVENT_INJECT_FAIL,
+               "Failed to inject event GD_OP_EVENT_STAGE_OP");
 
-        ctx = GF_CALLOC (1, sizeof (*ctx), gf_gld_mt_op_lock_ctx_t);
-
-        if (!ctx) {
-                //respond here
-                return -1;
-        }
+out:
+    free(op_req.buf.buf_val);  // malloced by xdr
+    glusterd_friend_sm();
+    glusterd_op_sm();
+    return ret;
+}
 
-        uuid_copy (ctx->uuid, lock_req.uuid);
-        ctx->req = req;
+int
+glusterd_handle_stage_op(rpcsvc_request_t *req)
+{
+    return glusterd_big_locked_handler(req, __glusterd_handle_stage_op);
+}
 
-        ret = glusterd_op_sm_inject_event (GD_OP_EVENT_LOCK, ctx);
+int
+__glusterd_handle_commit_op(rpcsvc_request_t *req)
+{
+    int32_t ret = -1;
+    glusterd_req_ctx_t *req_ctx = NULL;
+    gd1_mgmt_commit_op_req op_req = {
+        {0},
+    };
+    xlator_t *this = NULL;
+    uuid_t *txn_id = NULL;
+    glusterd_conf_t *priv = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+    priv = this->private;
+    GF_ASSERT(priv);
+    GF_ASSERT(req);
+
+    txn_id = &priv->global_txn_id;
+
+    ret = xdr_to_generic(req->msg[0], &op_req,
+                         (xdrproc_t)xdr_gd1_mgmt_commit_op_req);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REQ_DECODE_FAIL,
+               "Failed to decode commit "
+               "request received from peer");
+        req->rpc_err = GARBAGE_ARGS;
+        goto out;
+    }
+
+    RCU_READ_LOCK;
+    ret = (glusterd_peerinfo_find_by_uuid(op_req.uuid) == NULL);
+    RCU_READ_UNLOCK;
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_PEER_NOT_FOUND,
+               "%s doesn't "
+               "belong to the cluster. Ignoring request.",
+               uuid_utoa(op_req.uuid));
+        ret = -1;
+        goto out;
+    }
+
+    // the structures should always be equal
+    GF_ASSERT(sizeof(gd1_mgmt_commit_op_req) == sizeof(gd1_mgmt_stage_op_req));
+    ret = glusterd_req_ctx_create(req, op_req.op, op_req.uuid,
+                                  op_req.buf.buf_val, op_req.buf.buf_len,
+                                  gf_gld_mt_op_commit_ctx_t, &req_ctx);
+    if (ret)
+        goto out;
+
+    ret = dict_get_bin(req_ctx->dict, "transaction_id", (void **)&txn_id);
+    gf_msg_debug(this->name, 0, "transaction ID = %s", uuid_utoa(*txn_id));
+
+    ret = glusterd_op_sm_inject_event(GD_OP_EVENT_COMMIT_OP, txn_id, req_ctx);
 
 out:
-        gf_log ("", GF_LOG_NORMAL, "Returning %d", ret);
+    free(op_req.buf.buf_val);  // malloced by xdr
+    glusterd_friend_sm();
+    glusterd_op_sm();
+    return ret;
+}
 
-        return ret;
+int
+glusterd_handle_commit_op(rpcsvc_request_t *req)
+{
+    return glusterd_big_locked_handler(req, __glusterd_handle_commit_op);
 }
 
 int
-glusterd_handle_stage_op (rpcsvc_request_t *req)
+__glusterd_handle_cli_probe(rpcsvc_request_t *req)
 {
-        int32_t                         ret = -1;
-        char                            str[50] = {0,};
-        gd1_mgmt_stage_op_req           stage_req = {{0,}};
-        glusterd_op_stage_ctx_t         *ctx = NULL;
+    int32_t ret = -1;
+    gf_cli_req cli_req = {
+        {
+            0,
+        },
+    };
+    glusterd_peerinfo_t *peerinfo = NULL;
+    gf_boolean_t run_fsm = _gf_true;
+    xlator_t *this = NULL;
+    char *bind_name = NULL;
+    dict_t *dict = NULL;
+    char *hostname = NULL;
+    int port = 0;
+    int op_errno = 0;
+
+    GF_ASSERT(req);
+    this = THIS;
+
+    ret = xdr_to_generic(req->msg[0], &cli_req, (xdrproc_t)xdr_gf_cli_req);
+    if (ret < 0) {
+        // failed to decode msg;
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REQ_DECODE_FAIL,
+               "xdr decoding error");
+        req->rpc_err = GARBAGE_ARGS;
+        goto out;
+    }
+
+    if (cli_req.dict.dict_len) {
+        dict = dict_new();
+
+        ret = dict_unserialize(cli_req.dict.dict_val, cli_req.dict.dict_len,
+                               &dict);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_UNSERIALIZE_FAIL,
+                   "Failed to "
+                   "unserialize req-buffer to dictionary");
+            goto out;
+        }
+    }
+
+    ret = dict_get_strn(dict, "hostname", SLEN("hostname"), &hostname);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_HOSTNAME_NOTFOUND_IN_DICT,
+               "Failed to get hostname");
+        goto out;
+    }
+
+    ret = dict_get_int32n(dict, "port", SLEN("port"), &port);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_PORT_NOTFOUND_IN_DICT,
+               "Failed to get port");
+        goto out;
+    }
+
+    if (glusterd_is_any_volume_in_server_quorum(this) &&
+        !does_gd_meet_server_quorum(this)) {
+        glusterd_xfer_cli_probe_resp(req, -1, GF_PROBE_QUORUM_NOT_MET, NULL,
+                                     hostname, port, dict);
+        gf_msg(this->name, GF_LOG_CRITICAL, 0, GD_MSG_SERVER_QUORUM_NOT_MET,
+               "Server quorum not met. Rejecting operation.");
+        ret = 0;
+        goto out;
+    }
+
+    gf_msg("glusterd", GF_LOG_INFO, 0, GD_MSG_CLI_REQ_RECVD,
+           "Received CLI probe req %s %d", hostname, port);
+
+    if (dict_get_strn(this->options, "transport.socket.bind-address",
+                      SLEN("transport.socket.bind-address"), &bind_name) == 0) {
+        gf_msg_debug("glusterd", 0,
+                     "only checking probe address vs. bind address");
+        ret = gf_is_same_address(bind_name, hostname);
+    } else {
+        ret = gf_is_local_addr(hostname);
+    }
+    if (ret) {
+        glusterd_xfer_cli_probe_resp(req, 0, GF_PROBE_LOCALHOST, NULL, hostname,
+                                     port, dict);
+        ret = 0;
+        goto out;
+    }
 
-        GF_ASSERT (req);
+    RCU_READ_LOCK;
 
-        if (!gd_xdr_to_mgmt_stage_op_req (req->msg[0], &stage_req)) {
-                //failed to decode msg;
-                req->rpc_err = GARBAGE_ARGS;
-                goto out;
-        }
+    peerinfo = glusterd_peerinfo_find_by_hostname(hostname);
+    ret = (peerinfo && gd_peer_has_address(peerinfo, hostname));
 
-        uuid_unparse (stage_req.uuid, str);
-        gf_log ("glusterd", GF_LOG_NORMAL,
-                "Received stage op from uuid: %s", str);
+    RCU_READ_UNLOCK;
 
-        ctx = GF_CALLOC (1, sizeof (*ctx), gf_gld_mt_op_stage_ctx_t);
+    if (ret) {
+        gf_msg_debug("glusterd", 0,
+                     "Probe host %s port %d "
+                     "already a peer",
+                     hostname, port);
+        glusterd_xfer_cli_probe_resp(req, 0, GF_PROBE_FRIEND, NULL, hostname,
+                                     port, dict);
+        ret = 0;
+        goto out;
+    }
 
-        if (!ctx) {
-                //respond here
-                return -1;
-        }
+    ret = glusterd_probe_begin(req, hostname, port, dict, &op_errno);
 
-        //CHANGE THIS
-        uuid_copy (ctx->stage_req.uuid, stage_req.uuid);
-        ctx->stage_req.op = stage_req.op;
-        ctx->stage_req.buf.buf_len = stage_req.buf.buf_len;
-        ctx->stage_req.buf.buf_val = GF_CALLOC (1, stage_req.buf.buf_len,
-                                                gf_gld_mt_string);
-        if (!ctx->stage_req.buf.buf_val)
-                goto out;
+    if (ret == GLUSTERD_CONNECTION_AWAITED) {
+        // fsm should be run after connection establishes
+        run_fsm = _gf_false;
+        ret = 0;
 
-        memcpy (ctx->stage_req.buf.buf_val, stage_req.buf.buf_val,
-                stage_req.buf.buf_len);
+    } else if (ret == -1) {
+        glusterd_xfer_cli_probe_resp(req, -1, op_errno, NULL, hostname, port,
+                                     dict);
+        goto out;
+    }
 
-        ctx->req   = req;
+out:
+    free(cli_req.dict.dict_val);
 
-        ret = glusterd_op_sm_inject_event (GD_OP_EVENT_STAGE_OP, ctx);
+    if (run_fsm) {
+        glusterd_friend_sm();
+        glusterd_op_sm();
+    }
 
-out:
-        if (stage_req.buf.buf_val)
-                free (stage_req.buf.buf_val);//malloced by xdr
-        return ret;
+    return ret;
 }
 
 int
-glusterd_handle_commit_op (rpcsvc_request_t *req)
+glusterd_handle_cli_probe(rpcsvc_request_t *req)
 {
-        int32_t                         ret = -1;
-        char                            str[50] = {0,};
-        gd1_mgmt_commit_op_req          commit_req = {{0},};
-        glusterd_op_commit_ctx_t        *ctx = NULL;
-
-        GF_ASSERT (req);
+    return glusterd_big_locked_handler(req, __glusterd_handle_cli_probe);
+}
 
-        if (!gd_xdr_to_mgmt_commit_op_req (req->msg[0], &commit_req)) {
-                //failed to decode msg;
-                req->rpc_err = GARBAGE_ARGS;
-                goto out;
+int
+__glusterd_handle_cli_deprobe(rpcsvc_request_t *req)
+{
+    int32_t ret = -1;
+    gf_cli_req cli_req = {
+        {
+            0,
+        },
+    };
+    uuid_t uuid = {0};
+    int op_errno = 0;
+    xlator_t *this = NULL;
+    glusterd_conf_t *priv = NULL;
+    dict_t *dict = NULL;
+    char *hostname = NULL;
+    int port = 0;
+    int flags = 0;
+    glusterd_volinfo_t *volinfo = NULL;
+    glusterd_volinfo_t *tmp = NULL;
+    glusterd_snap_t *snapinfo = NULL;
+    glusterd_snap_t *tmpsnap = NULL;
+    gf_boolean_t need_free = _gf_false;
+
+    this = THIS;
+    GF_ASSERT(this);
+    priv = this->private;
+    GF_ASSERT(priv);
+    GF_ASSERT(req);
+
+    ret = xdr_to_generic(req->msg[0], &cli_req, (xdrproc_t)xdr_gf_cli_req);
+    if (ret < 0) {
+        // failed to decode msg;
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REQ_DECODE_FAIL,
+               "Failed to decode "
+               "request received from cli");
+        req->rpc_err = GARBAGE_ARGS;
+        goto out;
+    }
+
+    if (cli_req.dict.dict_len) {
+        dict = dict_new();
+
+        if (dict) {
+            need_free = _gf_true;
+        } else {
+            ret = -1;
+            goto out;
         }
 
-        uuid_unparse (commit_req.uuid, str);
-
-        gf_log ("glusterd", GF_LOG_NORMAL,
-                "Received commit op from uuid: %s", str);
+        ret = dict_unserialize(cli_req.dict.dict_val, cli_req.dict.dict_len,
+                               &dict);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_UNSERIALIZE_FAIL,
+                   "Failed to "
+                   "unserialize req-buffer to dictionary");
+            goto out;
+        }
+    }
+
+    gf_msg("glusterd", GF_LOG_INFO, 0, GD_MSG_CLI_REQ_RECVD,
+           "Received CLI deprobe req");
+
+    ret = dict_get_strn(dict, "hostname", SLEN("hostname"), &hostname);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_HOSTNAME_NOTFOUND_IN_DICT,
+               "Failed to get hostname");
+        goto out;
+    }
+
+    ret = dict_get_int32n(dict, "port", SLEN("port"), &port);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_PORT_NOTFOUND_IN_DICT,
+               "Failed to get port");
+        goto out;
+    }
+    ret = dict_get_int32n(dict, "flags", SLEN("flags"), &flags);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_FLAGS_NOTFOUND_IN_DICT,
+               "Failed to get flags");
+        goto out;
+    }
+
+    ret = glusterd_hostname_to_uuid(hostname, uuid);
+    if (ret) {
+        op_errno = GF_DEPROBE_NOT_FRIEND;
+        goto out;
+    }
+
+    if (!gf_uuid_compare(uuid, MY_UUID)) {
+        op_errno = GF_DEPROBE_LOCALHOST;
+        ret = -1;
+        goto out;
+    }
+
+    if (!(flags & GF_CLI_FLAG_OP_FORCE)) {
+        /* Check if peers are connected, except peer being
+         * detached*/
+        if (!glusterd_chk_peers_connected_befriended(uuid)) {
+            ret = -1;
+            op_errno = GF_DEPROBE_FRIEND_DOWN;
+            goto out;
+        }
+    }
+
+    /* Check for if volumes exist with some bricks on the peer being
+     * detached. It's not a problem if a volume contains none or all
+     * of its bricks on the peer being detached
+     */
+    cds_list_for_each_entry_safe(volinfo, tmp, &priv->volumes, vol_list)
+    {
+        ret = glusterd_friend_contains_vol_bricks(volinfo, uuid);
+        if (ret == 1) {
+            op_errno = GF_DEPROBE_BRICK_EXIST;
+            goto out;
+        }
+    }
+
+    cds_list_for_each_entry_safe(snapinfo, tmpsnap, &priv->snapshots, snap_list)
+    {
+        ret = glusterd_friend_contains_snap_bricks(snapinfo, uuid);
+        if (ret == 1) {
+            op_errno = GF_DEPROBE_SNAP_BRICK_EXIST;
+            goto out;
+        }
+    }
+    if (!(flags & GF_CLI_FLAG_OP_FORCE)) {
+        if (glusterd_is_any_volume_in_server_quorum(this) &&
+            !does_gd_meet_server_quorum(this)) {
+            gf_msg(this->name, GF_LOG_CRITICAL, 0, GD_MSG_SERVER_QUORUM_NOT_MET,
+                   "Server quorum not met. Rejecting operation.");
+            ret = -1;
+            op_errno = GF_DEPROBE_QUORUM_NOT_MET;
+            goto out;
+        }
+    }
+
+    if (!gf_uuid_is_null(uuid)) {
+        ret = glusterd_deprobe_begin(req, hostname, port, uuid, dict,
+                                     &op_errno);
+    } else {
+        ret = glusterd_deprobe_begin(req, hostname, port, NULL, dict,
+                                     &op_errno);
+    }
+
+    need_free = _gf_false;
 
-        ctx = GF_CALLOC (1, sizeof (*ctx), gf_gld_mt_op_commit_ctx_t);
+out:
+    free(cli_req.dict.dict_val);
 
-        if (!ctx) {
-                //respond here
-                return -1;
+    if (ret) {
+        ret = glusterd_xfer_cli_deprobe_resp(req, ret, op_errno, NULL, hostname,
+                                             dict);
+        if (need_free) {
+            dict_unref(dict);
         }
+    }
 
-        ctx->req = req;
-        //CHANGE THIS
-        uuid_copy (ctx->stage_req.uuid, commit_req.uuid);
-        ctx->stage_req.op = commit_req.op;
-        ctx->stage_req.buf.buf_len = commit_req.buf.buf_len;
-        ctx->stage_req.buf.buf_val = GF_CALLOC (1, commit_req.buf.buf_len,
-                                                gf_gld_mt_string);
-        if (!ctx->stage_req.buf.buf_val)
-                goto out;
+    glusterd_friend_sm();
+    glusterd_op_sm();
 
-        memcpy (ctx->stage_req.buf.buf_val, commit_req.buf.buf_val,
-                commit_req.buf.buf_len);
-
-        ret = glusterd_op_sm_inject_event (GD_OP_EVENT_COMMIT_OP, ctx);
-
-out:
-        if (commit_req.buf.buf_val)
-                free (commit_req.buf.buf_val);//malloced by xdr
-        return ret;
+    return ret;
 }
 
 int
-glusterd_handle_cli_probe (rpcsvc_request_t *req)
+glusterd_handle_cli_deprobe(rpcsvc_request_t *req)
 {
-        int32_t                         ret = -1;
-        gf1_cli_probe_req               cli_req = {0,};
-        glusterd_peerinfo_t             *peerinfo = NULL;
-        GF_ASSERT (req);
-
-        if (!gf_xdr_to_cli_probe_req (req->msg[0], &cli_req)) {
-                //failed to decode msg;
-                gf_log ("", GF_LOG_ERROR, "xdr decoding error");
-                req->rpc_err = GARBAGE_ARGS;
-                goto out;
-        }
-
-        gf_cmd_log ("peer probe", " on host %s:%d", cli_req.hostname,
-                    cli_req.port);
-        gf_log ("glusterd", GF_LOG_NORMAL, "Received CLI probe req %s %d",
-                cli_req.hostname, cli_req.port);
-
-        if (!(ret = glusterd_is_local_addr(cli_req.hostname))) {
-                glusterd_xfer_cli_probe_resp (req, 0, GF_PROBE_LOCALHOST,
-                                              cli_req.hostname, cli_req.port);
-                goto out;
-        }
-        if (!(ret = glusterd_friend_find_by_hostname(cli_req.hostname,
-                                         &peerinfo))) {
-                if ((peerinfo->state.state != GD_FRIEND_STATE_REQ_RCVD)
-                    || (peerinfo->state.state != GD_FRIEND_STATE_DEFAULT)) {
-
-                        gf_log ("glusterd", GF_LOG_NORMAL, "Probe host %s port %d"
-                               "already a friend", cli_req.hostname, cli_req.port);
-                        glusterd_xfer_cli_probe_resp (req, 0, GF_PROBE_FRIEND,
-                                                      cli_req.hostname, cli_req.port);
-                        goto out;
-                }
-        }
-        ret = glusterd_probe_begin (req, cli_req.hostname, cli_req.port);
-
-        gf_cmd_log ("peer probe","on host %s:%d %s",cli_req.hostname, cli_req.port,
-                    (ret) ? "FAILED" : "SUCCESS");
-out:
-        if (cli_req.hostname)
-                free (cli_req.hostname);//its malloced by xdr
-        return ret;
+    return glusterd_big_locked_handler(req, __glusterd_handle_cli_deprobe);
 }
 
 int
-glusterd_handle_cli_deprobe (rpcsvc_request_t *req)
+__glusterd_handle_cli_list_friends(rpcsvc_request_t *req)
 {
-        int32_t                         ret = -1;
-        gf1_cli_probe_req               cli_req = {0,};
-        uuid_t                          uuid = {0};
-        int                             op_errno = 0;
-        xlator_t                        *this = NULL;
-        glusterd_conf_t                 *priv = NULL;
-
-        this = THIS;
-        GF_ASSERT (this);
-        priv = this->private;
-        GF_ASSERT (priv);
-        GF_ASSERT (req);
-
-        if (!gf_xdr_to_cli_probe_req (req->msg[0], &cli_req)) {
-                //failed to decode msg;
-                req->rpc_err = GARBAGE_ARGS;
-                goto out;
-        }
-
-        gf_log ("glusterd", GF_LOG_NORMAL, "Received CLI deprobe req");
-
-        ret = glusterd_hostname_to_uuid (cli_req.hostname, uuid);
-        if (ret) {
-                op_errno = GF_DEPROBE_NOT_FRIEND;
-                goto out;
+    int32_t ret = -1;
+    gf1_cli_peer_list_req cli_req = {
+        0,
+    };
+    dict_t *dict = NULL;
+
+    GF_ASSERT(req);
+
+    ret = xdr_to_generic(req->msg[0], &cli_req,
+                         (xdrproc_t)xdr_gf1_cli_peer_list_req);
+    if (ret < 0) {
+        // failed to decode msg;
+        gf_msg("glusterd", GF_LOG_ERROR, 0, GD_MSG_REQ_DECODE_FAIL,
+               "Failed to decode "
+               "request received from cli");
+        req->rpc_err = GARBAGE_ARGS;
+        goto out;
+    }
+
+    gf_msg("glusterd", GF_LOG_INFO, 0, GD_MSG_CLI_REQ_RECVD,
+           "Received cli list req");
+
+    if (cli_req.dict.dict_len) {
+        /* Unserialize the dictionary */
+        dict = dict_new();
+
+        ret = dict_unserialize(cli_req.dict.dict_val, cli_req.dict.dict_len,
+                               &dict);
+        if (ret < 0) {
+            gf_msg("glusterd", GF_LOG_ERROR, 0, GD_MSG_DICT_UNSERIALIZE_FAIL,
+                   "failed to "
+                   "unserialize req-buffer to dictionary");
+            goto out;
+        } else {
+            dict->extra_stdfree = cli_req.dict.dict_val;
         }
+    }
 
-        if (!uuid_compare (uuid, priv->uuid)) {
-                op_errno = GF_DEPROBE_LOCALHOST;
-                ret = -1;
-                goto out;
-        }
+    ret = glusterd_list_friends(req, dict, cli_req.flags);
 
-        ret = glusterd_all_volume_cond_check (glusterd_friend_brick_belongs,
-                                              -1, &uuid);
-        if (ret) {
-                op_errno = GF_DEPROBE_BRICK_EXIST;
-                goto out;
-        }
+out:
+    if (dict)
+        dict_unref(dict);
 
-        ret = glusterd_deprobe_begin (req, cli_req.hostname,
-                                      cli_req.port, uuid);
+    glusterd_friend_sm();
+    glusterd_op_sm();
 
-        gf_cmd_log ("peer deprobe", "on host %s:%d %s", cli_req.hostname,
-                    cli_req.port, (ret) ? "FAILED" : "SUCCESS");
-out:
-        if (ret) {
-                ret = glusterd_xfer_cli_deprobe_resp (req, ret, op_errno,
-                                                      cli_req.hostname);
-        }
-        if (cli_req.hostname)
-                free (cli_req.hostname);//malloced by xdr
-        return ret;
+    return ret;
 }
 
 int
-glusterd_handle_cli_list_friends (rpcsvc_request_t *req)
+glusterd_handle_cli_list_friends(rpcsvc_request_t *req)
 {
-        int32_t                         ret = -1;
-        gf1_cli_peer_list_req           cli_req = {0,};
-        dict_t                          *dict = NULL;
-
-        GF_ASSERT (req);
+    return glusterd_big_locked_handler(req, __glusterd_handle_cli_list_friends);
+}
 
-        if (!gf_xdr_to_cli_peer_list_req (req->msg[0], &cli_req)) {
-                //failed to decode msg;
-                req->rpc_err = GARBAGE_ARGS;
-                goto out;
+static int
+__glusterd_handle_cli_get_volume(rpcsvc_request_t *req)
+{
+    int32_t ret = -1;
+    gf_cli_req cli_req = {{
+        0,
+    }};
+    int32_t flags = 0;
+    dict_t *dict = NULL;
+    xlator_t *this = NULL;
+
+    GF_ASSERT(req);
+    this = THIS;
+
+    ret = xdr_to_generic(req->msg[0], &cli_req, (xdrproc_t)xdr_gf_cli_req);
+    if (ret < 0) {
+        // failed to decode msg;
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REQ_DECODE_FAIL,
+               "Failed to decode "
+               "request received from cli");
+        req->rpc_err = GARBAGE_ARGS;
+        goto out;
+    }
+
+    gf_msg(this->name, GF_LOG_DEBUG, 0, GD_MSG_GET_VOL_REQ_RCVD,
+           "Received get vol req");
+
+    if (cli_req.dict.dict_len) {
+        /* Unserialize the dictionary */
+        dict = dict_new();
+
+        ret = dict_unserialize(cli_req.dict.dict_val, cli_req.dict.dict_len,
+                               &dict);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_UNSERIALIZE_FAIL,
+                   "failed to "
+                   "unserialize req-buffer to dictionary");
+            goto out;
+        } else {
+            dict->extra_stdfree = cli_req.dict.dict_val;
         }
+    }
 
-        gf_log ("glusterd", GF_LOG_NORMAL, "Received cli list req");
-
-        if (cli_req.dict.dict_len) {
-                /* Unserialize the dictionary */
-                dict  = dict_new ();
+    ret = dict_get_int32n(dict, "flags", SLEN("flags"), &flags);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_FLAGS_NOTFOUND_IN_DICT,
+               "failed to get flags");
+        goto out;
+    }
+    ret = glusterd_get_volumes(req, dict, flags);
 
-                ret = dict_unserialize (cli_req.dict.dict_val,
-                                        cli_req.dict.dict_len,
-                                        &dict);
-                if (ret < 0) {
-                        gf_log ("glusterd", GF_LOG_ERROR,
-                                "failed to "
-                                "unserialize req-buffer to dictionary");
-                        goto out;
-                } else {
-                        dict->extra_stdfree = cli_req.dict.dict_val;
-                }
-        }
+out:
+    if (dict)
+        dict_unref(dict);
 
-        ret = glusterd_list_friends (req, dict, cli_req.flags);
+    glusterd_friend_sm();
+    glusterd_op_sm();
 
-out:
-        if (dict)
-                dict_unref (dict);
-        return ret;
+    return ret;
 }
 
 int
-glusterd_handle_cli_get_volume (rpcsvc_request_t *req)
+glusterd_handle_cli_get_volume(rpcsvc_request_t *req)
 {
-        int32_t                         ret = -1;
-        gf1_cli_get_vol_req             cli_req = {0,};
-        dict_t                          *dict = NULL;
-
-        GF_ASSERT (req);
-
-        if (!gf_xdr_to_cli_get_vol_req (req->msg[0], &cli_req)) {
-                //failed to decode msg;
-                req->rpc_err = GARBAGE_ARGS;
-                goto out;
-        }
+    return glusterd_big_locked_handler(req, __glusterd_handle_cli_get_volume);
+}
 
-        gf_log ("glusterd", GF_LOG_NORMAL, "Received get vol req");
+int
+__glusterd_handle_cli_uuid_reset(rpcsvc_request_t *req)
+{
+    int ret = -1;
+    dict_t *dict = NULL;
+    xlator_t *this = NULL;
+    glusterd_conf_t *priv = NULL;
+    uuid_t uuid = {0};
+    gf_cli_rsp rsp = {
+        0,
+    };
+    gf_cli_req cli_req = {{
+        0,
+    }};
+    char msg_str[128] = {
+        0,
+    };
+
+    GF_ASSERT(req);
+
+    this = THIS;
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    ret = xdr_to_generic(req->msg[0], &cli_req, (xdrproc_t)xdr_gf_cli_req);
+    if (ret < 0) {
+        // failed to decode msg;
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REQ_DECODE_FAIL,
+               "Failed to decode "
+               "request received from cli");
+        req->rpc_err = GARBAGE_ARGS;
+        goto out;
+    }
+
+    gf_msg_debug("glusterd", 0, "Received uuid reset req");
+
+    if (cli_req.dict.dict_len) {
+        /* Unserialize the dictionary */
+        dict = dict_new();
+
+        ret = dict_unserialize(cli_req.dict.dict_val, cli_req.dict.dict_len,
+                               &dict);
+        if (ret < 0) {
+            gf_msg("glusterd", GF_LOG_ERROR, 0, GD_MSG_DICT_UNSERIALIZE_FAIL,
+                   "failed to "
+                   "unserialize req-buffer to dictionary");
+            snprintf(msg_str, sizeof(msg_str),
+                     "Unable to decode "
+                     "the buffer");
+            goto out;
+        } else {
+            dict->extra_stdfree = cli_req.dict.dict_val;
+        }
+    }
+
+    /* In the above section if dict_unserialize is successful, ret is set
+     * to zero.
+     */
+    ret = -1;
+    // Do not allow peer reset if there are any volumes in the cluster
+    if (!cds_list_empty(&priv->volumes)) {
+        snprintf(msg_str, sizeof(msg_str),
+                 "volumes are already "
+                 "present in the cluster. Resetting uuid is not "
+                 "allowed");
+        gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_VOLS_ALREADY_PRESENT, "%s",
+               msg_str);
+        goto out;
+    }
+
+    // Do not allow peer reset if trusted storage pool is already formed
+    if (!cds_list_empty(&priv->peers)) {
+        snprintf(msg_str, sizeof(msg_str),
+                 "trusted storage pool "
+                 "has been already formed. Please detach this peer "
+                 "from the pool and reset its uuid.");
+        gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_TSP_ALREADY_FORMED, "%s",
+               msg_str);
+        goto out;
+    }
+
+    gf_uuid_copy(uuid, priv->uuid);
+    ret = glusterd_uuid_generate_save();
+
+    if (!gf_uuid_compare(uuid, MY_UUID)) {
+        snprintf(msg_str, sizeof(msg_str),
+                 "old uuid and the new uuid"
+                 " are same. Try gluster peer reset again");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_UUIDS_SAME_RETRY, "%s",
+               msg_str);
+        ret = -1;
+        goto out;
+    }
 
-        if (cli_req.dict.dict_len) {
-                /* Unserialize the dictionary */
-                dict  = dict_new ();
+out:
+    if (ret) {
+        rsp.op_ret = -1;
+        if (msg_str[0] == '\0')
+            snprintf(msg_str, sizeof(msg_str),
+                     "Operation "
+                     "failed");
+        rsp.op_errstr = msg_str;
+        ret = 0;
+    } else {
+        rsp.op_errstr = "";
+    }
 
-                ret = dict_unserialize (cli_req.dict.dict_val,
-                                        cli_req.dict.dict_len,
-                                        &dict);
-                if (ret < 0) {
-                        gf_log ("glusterd", GF_LOG_ERROR,
-                                "failed to "
-                                "unserialize req-buffer to dictionary");
-                        goto out;
-                } else {
-                        dict->extra_stdfree = cli_req.dict.dict_val;
-                }
-        }
+    glusterd_to_cli(req, &rsp, NULL, 0, NULL, (xdrproc_t)xdr_gf_cli_rsp, dict);
 
-        ret = glusterd_get_volumes (req, dict, cli_req.flags);
+    return ret;
+}
 
-out:
-        if (dict)
-                dict_unref (dict);
-        return ret;
+int
+glusterd_handle_cli_uuid_reset(rpcsvc_request_t *req)
+{
+    return glusterd_big_locked_handler(req, __glusterd_handle_cli_uuid_reset);
 }
 
 int
-glusterd_handle_create_volume (rpcsvc_request_t *req)
-{
-        int32_t                 ret         = -1;
-        gf1_cli_create_vol_req  cli_req     = {0,};
-        dict_t                 *dict        = NULL;
-        glusterd_brickinfo_t   *brickinfo   = NULL;
-        char                   *brick       = NULL;
-        char                   *bricks      = NULL;
-        char                   *volname     = NULL;
-        int                    brick_count = 0;
-        char                   *tmpptr      = NULL;
-        int                    i           = 0;
-        char                   *brick_list  = NULL;
-        void                   *cli_rsp     = NULL;
-        char                    err_str[2048] = {0,};
-        gf1_cli_create_vol_rsp  rsp         = {0,};
-        glusterd_conf_t        *priv        = NULL;
-        int                     err_ret     = 0;
-        xlator_t               *this        = NULL;
-        char                   *free_ptr    = NULL;
-        char                   *trans_type  = NULL;
-        uuid_t                  volume_id   = {0,};
-        char                    volid[64]   = {0,};
-        glusterd_brickinfo_t    *tmpbrkinfo = NULL;
-        glusterd_volinfo_t      tmpvolinfo = {{0},};
-
-        GF_ASSERT (req);
-
-        this = THIS;
-        GF_ASSERT(this);
-
-        priv = this->private;
-
-        INIT_LIST_HEAD (&tmpvolinfo.bricks);
-
-        if (!gf_xdr_to_cli_create_vol_req (req->msg[0], &cli_req)) {
-                //failed to decode msg;
-                req->rpc_err = GARBAGE_ARGS;
-                goto out;
+__glusterd_handle_cli_uuid_get(rpcsvc_request_t *req)
+{
+    int ret = -1;
+    dict_t *dict = NULL;
+    dict_t *rsp_dict = NULL;
+    xlator_t *this = NULL;
+    glusterd_conf_t *priv = NULL;
+    gf_cli_rsp rsp = {
+        0,
+    };
+    gf_cli_req cli_req = {{
+        0,
+    }};
+    char err_str[64] = {
+        0,
+    };
+    char uuid_str[64] = {
+        0,
+    };
+
+    GF_ASSERT(req);
+
+    this = THIS;
+    priv = this->private;
+    GF_ASSERT(priv);
+
+    ret = xdr_to_generic(req->msg[0], &cli_req, (xdrproc_t)xdr_gf_cli_req);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REQ_DECODE_FAIL,
+               "Failed to decode "
+               "request received from cli");
+        req->rpc_err = GARBAGE_ARGS;
+        goto out;
+    }
+
+    gf_msg_debug("glusterd", 0, "Received uuid get req");
+
+    if (cli_req.dict.dict_len) {
+        dict = dict_new();
+        if (!dict) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_CREATE_FAIL,
+                    NULL);
+            ret = -1;
+            goto out;
         }
 
-        gf_log ("glusterd", GF_LOG_NORMAL, "Received create volume req");
+        ret = dict_unserialize(cli_req.dict.dict_val, cli_req.dict.dict_len,
+                               &dict);
+        if (ret < 0) {
+            gf_msg("glusterd", GF_LOG_ERROR, 0, GD_MSG_DICT_UNSERIALIZE_FAIL,
+                   "failed to "
+                   "unserialize req-buffer to dictionary");
+            snprintf(err_str, sizeof(err_str),
+                     "Unable to decode "
+                     "the buffer");
+            goto out;
 
-        if (cli_req.bricks.bricks_len) {
-                /* Unserialize the dictionary */
-                dict  = dict_new ();
+        } else {
+            dict->extra_stdfree = cli_req.dict.dict_val;
+        }
+    }
+
+    rsp_dict = dict_new();
+    if (!rsp_dict) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_CREATE_FAIL, NULL);
+        ret = -1;
+        goto out;
+    }
+
+    uuid_utoa_r(MY_UUID, uuid_str);
+    ret = dict_set_strn(rsp_dict, "uuid", SLEN("uuid"), uuid_str);
+    if (ret) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED,
+               "Failed to set uuid in "
+               "dictionary.");
+        goto out;
+    }
+
+    ret = dict_allocate_and_serialize(rsp_dict, &rsp.dict.dict_val,
+                                      &rsp.dict.dict_len);
+    if (ret) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                GD_MSG_DICT_ALLOC_AND_SERL_LENGTH_GET_FAIL, NULL);
+        goto out;
+    }
+    ret = 0;
+out:
+    if (ret) {
+        rsp.op_ret = -1;
+        if (err_str[0] == '\0')
+            snprintf(err_str, sizeof(err_str),
+                     "Operation "
+                     "failed");
+        rsp.op_errstr = err_str;
 
-                ret = dict_unserialize (cli_req.bricks.bricks_val,
-                                        cli_req.bricks.bricks_len,
-                                        &dict);
-                if (ret < 0) {
-                        gf_log ("glusterd", GF_LOG_ERROR,
-                                "failed to "
-                                "unserialize req-buffer to dictionary");
-                        goto out;
-                } else {
-                        dict->extra_stdfree = cli_req.bricks.bricks_val;
-                }
-        }
+    } else {
+        rsp.op_errstr = "";
+    }
 
-        ret = dict_get_str (dict, "volname", &volname);
+    glusterd_to_cli(req, &rsp, NULL, 0, NULL, (xdrproc_t)xdr_gf_cli_rsp, dict);
 
-        if (ret) {
-                gf_log ("", GF_LOG_ERROR, "Unable to get volume name");
-                goto out;
-        }
-        gf_cmd_log ("Volume create", "on volname: %s attempted", volname);
+    if (rsp_dict)
+        dict_unref(rsp_dict);
+    GF_FREE(rsp.dict.dict_val);
 
-        if ((ret = glusterd_check_volume_exists (volname))) {
-                snprintf(err_str, 2048, "Volname %s already exists",
-                         volname);
-                gf_log ("glusterd", GF_LOG_ERROR, "%s", err_str);
-                err_ret = 1;
-                goto out;
-        }
-
-        ret = dict_get_int32 (dict, "count", &brick_count);
-        if (ret) {
-                gf_log ("", GF_LOG_ERROR, "Unable to get count");
-                goto out;
-        }
+    return 0;
+}
+int
+glusterd_handle_cli_uuid_get(rpcsvc_request_t *req)
+{
+    return glusterd_big_locked_handler(req, __glusterd_handle_cli_uuid_get);
+}
 
-        ret = dict_get_str (dict, "transport", &trans_type);
-        if (ret) {
-                gf_log ("", GF_LOG_ERROR, "Unable to get transport-type");
-                goto out;
-        }
-        ret = dict_get_str (dict, "bricks", &bricks);
-        if (ret) {
-                gf_log ("", GF_LOG_ERROR, "Unable to get bricks");
-                goto out;
-        }
+int
+__glusterd_handle_cli_list_volume(rpcsvc_request_t *req)
+{
+    int ret = -1;
+    dict_t *dict = NULL;
+    glusterd_conf_t *priv = NULL;
+    glusterd_volinfo_t *volinfo = NULL;
+    int count = 0;
+    char key[64] = {
+        0,
+    };
+    int keylen;
+    gf_cli_rsp rsp = {
+        0,
+    };
+
+    GF_ASSERT(req);
+
+    priv = THIS->private;
+    GF_ASSERT(priv);
+
+    dict = dict_new();
+    if (!dict) {
+        gf_smsg("glusterd", GF_LOG_ERROR, errno, GD_MSG_DICT_CREATE_FAIL, NULL);
+        goto out;
+    }
+
+    cds_list_for_each_entry(volinfo, &priv->volumes, vol_list)
+    {
+        keylen = snprintf(key, sizeof(key), "volume%d", count);
+        ret = dict_set_strn(dict, key, keylen, volinfo->volname);
+        if (ret)
+            goto out;
+        count++;
+    }
 
-        uuid_generate (volume_id);
-        uuid_unparse (volume_id, volid);
-        free_ptr = gf_strdup (volid);
-        ret = dict_set_dynstr (dict, "volume-id", free_ptr);
-        if (ret) {
-                gf_log ("", GF_LOG_ERROR, "unable to set volume-id");
-                goto out;
-        }
-        free_ptr = NULL;
+    ret = dict_set_int32n(dict, "count", SLEN("count"), count);
+    if (ret) {
+        gf_smsg("glusterd", GF_LOG_ERROR, errno, GD_MSG_DICT_SET_FAILED,
+                "Key=count", NULL);
+        goto out;
+    }
 
-        if (bricks) {
-                brick_list = gf_strdup (bricks);
-                free_ptr = brick_list;
-        }
+    ret = dict_allocate_and_serialize(dict, &rsp.dict.dict_val,
+                                      &rsp.dict.dict_len);
+    if (ret)
+        goto out;
 
-        gf_cmd_log ("Volume create", "on volname: %s type:%s count:%d bricks:%s",
-                    cli_req.volname, ((cli_req.type == 0)? "DEFAULT":
-                    ((cli_req.type == 1)? "STRIPE":"REPLICATE")), cli_req.count,
-                    bricks);
+    ret = 0;
 
+out:
+    rsp.op_ret = ret;
+    if (ret)
+        rsp.op_errstr = "Error listing volumes";
+    else
+        rsp.op_errstr = "";
 
-        while ( i < brick_count) {
-                i++;
-                brick= strtok_r (brick_list, " \n", &tmpptr);
-                brick_list = tmpptr;
-                ret = glusterd_brickinfo_from_brick (brick, &brickinfo);
-                if (ret)
-                        goto out;
+    glusterd_submit_reply(req, &rsp, NULL, 0, NULL, (xdrproc_t)xdr_gf_cli_rsp);
+    ret = 0;
 
-                ret = glusterd_new_brick_validate (brick, brickinfo, err_str,
-                                                   sizeof (err_str));
-                if (ret) {
-                        err_ret = 1;
-                        goto out;
-                }
-                ret = glusterd_volume_brickinfo_get (brickinfo->uuid,
-                                                     brickinfo->hostname,
-                                                     brickinfo->path,
-                                                     &tmpvolinfo, &tmpbrkinfo);
-                if (!ret) {
-                        err_ret = 1;
-                        snprintf (err_str, sizeof (err_str), "Brick: %s:%s, %s"
-                                  " in the arguments mean the same",
-                                  tmpbrkinfo->hostname, tmpbrkinfo->path,
-                                  brick);
-                        goto out;
-                }
-                list_add_tail (&brickinfo->brick_list, &tmpvolinfo.bricks);
-                brickinfo = NULL;
-        }
+    if (dict)
+        dict_unref(dict);
 
-        ret = glusterd_create_volume (req, dict);
+    GF_FREE(rsp.dict.dict_val);
 
-        gf_cmd_log ("Volume create", "on volname: %s %s", volname,
-                    ((ret || err_ret) != 0) ? "FAILED": "SUCCESS");
+    glusterd_friend_sm();
+    glusterd_op_sm();
 
-out:
-        if ((err_ret || ret) && dict)
-                dict_unref (dict);
-        if (err_ret) {
-                rsp.op_ret = -1;
-                rsp.op_errno = 0;
-                rsp.volname = "";
-                rsp.op_errstr = err_str;
-                cli_rsp = &rsp;
-                glusterd_submit_reply(req, cli_rsp, NULL, 0, NULL,
-                                      gf_xdr_serialize_cli_create_vol_rsp);
-                if (!glusterd_opinfo_unlock())
-                        gf_log ("glusterd", GF_LOG_ERROR, "Unlock on opinfo"
-                                " failed");
-                ret = 0; //Client response sent, prevent second response
-        }
-
-        if (free_ptr)
-                GF_FREE(free_ptr);
-
-        glusterd_volume_bricks_delete (&tmpvolinfo);
-        if (brickinfo)
-                glusterd_brickinfo_delete (brickinfo);
-        if (cli_req.volname)
-                free (cli_req.volname); // its a malloced by xdr
-        return ret;
+    return ret;
 }
 
 int
-glusterd_handle_cli_start_volume (rpcsvc_request_t *req)
+glusterd_handle_cli_list_volume(rpcsvc_request_t *req)
 {
-        int32_t                         ret = -1;
-        gf1_cli_start_vol_req           cli_req = {0,};
-
-        GF_ASSERT (req);
-
-        if (!gf_xdr_to_cli_start_vol_req (req->msg[0], &cli_req)) {
-                //failed to decode msg;
-                req->rpc_err = GARBAGE_ARGS;
-                goto out;
-        }
-
-        gf_log ("glusterd", GF_LOG_NORMAL, "Received start vol req"
-                "for volume %s", cli_req.volname);
+    return glusterd_big_locked_handler(req, __glusterd_handle_cli_list_volume);
+}
 
-        ret = glusterd_start_volume (req, cli_req.volname, cli_req.flags);
+int32_t
+glusterd_op_begin(rpcsvc_request_t *req, glusterd_op_t op, void *ctx,
+                  char *err_str, size_t err_len)
+{
+    int ret = -1;
 
-        gf_cmd_log ("volume start","on volname: %s %s", cli_req.volname,
-                    ((ret == 0) ? "SUCCESS": "FAILED"));
+    ret = glusterd_op_txn_begin(req, op, ctx, err_str, err_len);
 
-out:
-        if (cli_req.volname)
-                free (cli_req.volname); //its malloced by xdr
-        return ret;
+    return ret;
 }
 
-
 int
-glusterd_handle_cli_stop_volume (rpcsvc_request_t *req)
+__glusterd_handle_ganesha_cmd(rpcsvc_request_t *req)
 {
-        int32_t                         ret = -1;
-        gf1_cli_stop_vol_req           cli_req = {0,};
-
-        GF_ASSERT (req);
-
-        if (!gf_xdr_to_cli_stop_vol_req (req->msg[0], &cli_req)) {
-                //failed to decode msg;
-                req->rpc_err = GARBAGE_ARGS;
-                goto out;
+    int32_t ret = -1;
+    gf_cli_req cli_req = {{
+        0,
+    }};
+    dict_t *dict = NULL;
+    glusterd_op_t cli_op = GD_OP_GANESHA;
+    char *op_errstr = NULL;
+    char err_str[2048] = {
+        0,
+    };
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(req);
+
+    ret = xdr_to_generic(req->msg[0], &cli_req, (xdrproc_t)xdr_gf_cli_req);
+    if (ret < 0) {
+        snprintf(err_str, sizeof(err_str),
+                 "Failed to decode "
+                 "request received from cli");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REQ_DECODE_FAIL, "%s",
+               err_str);
+        req->rpc_err = GARBAGE_ARGS;
+        goto out;
+    }
+
+    if (cli_req.dict.dict_len) {
+        /* Unserialize the dictionary */
+        dict = dict_new();
+        if (!dict) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_CREATE_FAIL,
+                    NULL);
+            ret = -1;
+            goto out;
         }
 
-        gf_log ("glusterd", GF_LOG_NORMAL, "Received stop vol req"
-                "for volume %s", cli_req.volname);
-
-        ret = glusterd_stop_volume (req, cli_req.volname, cli_req.flags);
+        ret = dict_unserialize(cli_req.dict.dict_val, cli_req.dict.dict_len,
+                               &dict);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_UNSERIALIZE_FAIL,
+                   "failed to "
+                   "unserialize req-buffer to dictionary");
+            snprintf(err_str, sizeof(err_str),
+                     "Unable to decode "
+                     "the command");
+            goto out;
+        } else {
+            dict->extra_stdfree = cli_req.dict.dict_val;
+        }
+    }
 
-        gf_cmd_log ("Volume stop","on volname: %s %s", cli_req.volname,
-                    ((ret)?"FAILED":"SUCCESS"));
+    gf_msg_trace(this->name, 0, "Received global option request");
 
+    ret = glusterd_op_begin_synctask(req, GD_OP_GANESHA, dict);
 out:
-        if (cli_req.volname)
-                free (cli_req.volname); //its malloced by xdr
-        return ret;
+    if (ret) {
+        if (err_str[0] == '\0')
+            snprintf(err_str, sizeof(err_str), "Operation failed");
+        ret = glusterd_op_send_cli_response(cli_op, ret, 0, req, dict, err_str);
+    }
+    if (op_errstr)
+        GF_FREE(op_errstr);
+    if (dict)
+        dict_unref(dict);
+
+    return ret;
 }
 
 int
-glusterd_handle_cli_delete_volume (rpcsvc_request_t *req)
+glusterd_handle_ganesha_cmd(rpcsvc_request_t *req)
 {
-        int32_t                         ret = -1;
-        gf1_cli_delete_vol_req          cli_req = {0,};
-        int32_t                         flags = 0;
+    return glusterd_big_locked_handler(req, __glusterd_handle_ganesha_cmd);
+}
 
-        GF_ASSERT (req);
+static int
+__glusterd_handle_reset_volume(rpcsvc_request_t *req)
+{
+    int32_t ret = -1;
+    gf_cli_req cli_req = {{
+        0,
+    }};
+    dict_t *dict = NULL;
+    glusterd_op_t cli_op = GD_OP_RESET_VOLUME;
+    char *volname = NULL;
+    char err_str[64] = {
+        0,
+    };
+    xlator_t *this = NULL;
+
+    GF_ASSERT(req);
+    this = THIS;
+    GF_ASSERT(this);
+
+    gf_msg(this->name, GF_LOG_INFO, 0, 0, "Received reset vol req");
+
+    ret = xdr_to_generic(req->msg[0], &cli_req, (xdrproc_t)xdr_gf_cli_req);
+    if (ret < 0) {
+        snprintf(err_str, sizeof(err_str),
+                 "Failed to decode request "
+                 "received from cli");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REQ_DECODE_FAIL, "%s",
+               err_str);
+        req->rpc_err = GARBAGE_ARGS;
+        goto out;
+    }
+
+    if (cli_req.dict.dict_len) {
+        /* Unserialize the dictionary */
+        dict = dict_new();
+
+        ret = dict_unserialize(cli_req.dict.dict_val, cli_req.dict.dict_len,
+                               &dict);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_UNSERIALIZE_FAIL,
+                   "failed to "
+                   "unserialize req-buffer to dictionary");
+            snprintf(err_str, sizeof(err_str),
+                     "Unable to decode "
+                     "the command");
+            goto out;
+        } else {
+            dict->extra_stdfree = cli_req.dict.dict_val;
+        }
+    }
+
+    ret = dict_get_strn(dict, "volname", SLEN("volname"), &volname);
+    if (ret) {
+        snprintf(err_str, sizeof(err_str),
+                 "Failed to get volume "
+                 "name");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_VOLNAME_NOTFOUND_IN_DICT,
+               "%s", err_str);
+        goto out;
+    }
+    gf_msg_debug(this->name, 0,
+                 "Received volume reset request for "
+                 "volume %s",
+                 volname);
+
+    ret = glusterd_op_begin_synctask(req, GD_OP_RESET_VOLUME, dict);
 
-        if (!gf_xdr_to_cli_delete_vol_req (req->msg[0], &cli_req)) {
-                //failed to decode msg;
-                req->rpc_err = GARBAGE_ARGS;
-                goto out;
-        }
-        gf_cmd_log ("Volume delete","on volname: %s attempted", cli_req.volname);
+out:
+    if (ret) {
+        if (err_str[0] == '\0')
+            snprintf(err_str, sizeof(err_str), "Operation failed");
+        ret = glusterd_op_send_cli_response(cli_op, ret, 0, req, dict, err_str);
+    }
 
-        gf_log ("glusterd", GF_LOG_NORMAL, "Received delete vol req"
-                "for volume %s", cli_req.volname);
+    return ret;
+}
 
-        ret = glusterd_delete_volume (req, cli_req.volname, flags);
+int
+glusterd_handle_reset_volume(rpcsvc_request_t *req)
+{
+    return glusterd_big_locked_handler(req, __glusterd_handle_reset_volume);
+}
 
-        gf_cmd_log ("Volume delete", "on volname: %s %s", cli_req.volname,
-                   ((ret) ? "FAILED" : "SUCCESS"));
+int
+__glusterd_handle_set_volume(rpcsvc_request_t *req)
+{
+    int32_t ret = -1;
+    gf_cli_req cli_req = {{
+        0,
+    }};
+    dict_t *dict = NULL;
+    glusterd_op_t cli_op = GD_OP_SET_VOLUME;
+    char *key = NULL;
+    char *value = NULL;
+    char *volname = NULL;
+    char *op_errstr = NULL;
+    gf_boolean_t help = _gf_false;
+    char err_str[2048] = {
+        0,
+    };
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+
+    GF_ASSERT(req);
+
+    ret = xdr_to_generic(req->msg[0], &cli_req, (xdrproc_t)xdr_gf_cli_req);
+    if (ret < 0) {
+        snprintf(err_str, sizeof(err_str),
+                 "Failed to decode "
+                 "request received from cli");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REQ_DECODE_FAIL, "%s",
+               err_str);
+        req->rpc_err = GARBAGE_ARGS;
+        goto out;
+    }
+
+    if (cli_req.dict.dict_len) {
+        /* Unserialize the dictionary */
+        dict = dict_new();
+
+        ret = dict_unserialize(cli_req.dict.dict_val, cli_req.dict.dict_len,
+                               &dict);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, errno,
+                   GD_MSG_DICT_UNSERIALIZE_FAIL,
+                   "failed to "
+                   "unserialize req-buffer to dictionary");
+            snprintf(err_str, sizeof(err_str),
+                     "Unable to decode "
+                     "the command");
+            goto out;
+        } else {
+            dict->extra_stdfree = cli_req.dict.dict_val;
+        }
+    }
+
+    ret = dict_get_strn(dict, "volname", SLEN("volname"), &volname);
+    if (ret) {
+        snprintf(err_str, sizeof(err_str),
+                 "Failed to get volume "
+                 "name while handling volume set command");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED, "%s",
+               err_str);
+        goto out;
+    }
+
+    if (strcmp(volname, "help") == 0 || strcmp(volname, "help-xml") == 0) {
+        ret = glusterd_volset_help(dict, &op_errstr);
+        help = _gf_true;
+        goto out;
+    }
+
+    ret = dict_get_strn(dict, "key1", SLEN("key1"), &key);
+    if (ret) {
+        snprintf(err_str, sizeof(err_str),
+                 "Failed to get key while"
+                 " handling volume set for %s",
+                 volname);
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED, "%s",
+               err_str);
+        goto out;
+    }
+
+    ret = dict_get_strn(dict, "value1", SLEN("value1"), &value);
+    if (ret) {
+        snprintf(err_str, sizeof(err_str),
+                 "Failed to get value while"
+                 " handling volume set for %s",
+                 volname);
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_GET_FAILED, "%s",
+               err_str);
+        goto out;
+    }
+    gf_msg_debug(this->name, 0,
+                 "Received volume set request for "
+                 "volume %s",
+                 volname);
+
+    ret = glusterd_op_begin_synctask(req, GD_OP_SET_VOLUME, dict);
 
 out:
-        if (cli_req.volname)
-                free (cli_req.volname); //its malloced by xdr
-        return ret;
+    if (help)
+        ret = glusterd_op_send_cli_response(cli_op, ret, 0, req, dict,
+                                            (op_errstr) ? op_errstr : "");
+    else if (ret) {
+        if (err_str[0] == '\0')
+            snprintf(err_str, sizeof(err_str), "Operation failed");
+        ret = glusterd_op_send_cli_response(cli_op, ret, 0, req, dict, err_str);
+    }
+    if (op_errstr)
+        GF_FREE(op_errstr);
+
+    return ret;
 }
 
 int
-glusterd_handle_add_brick (rpcsvc_request_t *req)
+glusterd_handle_set_volume(rpcsvc_request_t *req)
 {
-        int32_t                         ret = -1;
-        gf1_cli_add_brick_req          cli_req = {0,};
-        dict_t                          *dict = NULL;
-        glusterd_brickinfo_t            *brickinfo = NULL;
-        char                            *brick = NULL;
-        char                            *bricks = NULL;
-        char                            *volname = NULL;
-        int                             brick_count = 0;
-        char                            *tmpptr = NULL;
-        int                             i = 0;
-        char                            *brick_list = NULL;
-        void                            *cli_rsp = NULL;
-        char                            err_str[2048] = {0,};
-        gf1_cli_add_brick_rsp           rsp = {0,};
-        glusterd_volinfo_t              *volinfo = NULL;
-        int32_t                         err_ret = 0;
-        glusterd_conf_t                 *priv = NULL;
-        xlator_t                        *this = NULL;
-        char                            *free_ptr = NULL;
-        glusterd_brickinfo_t            *tmpbrkinfo = NULL;
-        glusterd_volinfo_t              tmpvolinfo = {{0},};
-
-        this = THIS;
-        GF_ASSERT(this);
-
-        priv = this->private;
-
-        GF_ASSERT (req);
-
-        INIT_LIST_HEAD (&tmpvolinfo.bricks);
-
-        if (!gf_xdr_to_cli_add_brick_req (req->msg[0], &cli_req)) {
-                //failed to decode msg;
-                req->rpc_err = GARBAGE_ARGS;
-                goto out;
-        }
-
-        gf_cmd_log ("Volume add-brick", "on volname: %s attempted",
-                    cli_req.volname);
-        gf_log ("glusterd", GF_LOG_NORMAL, "Received add brick req");
-
-        if (cli_req.bricks.bricks_len) {
-                /* Unserialize the dictionary */
-                dict  = dict_new ();
+    return glusterd_big_locked_handler(req, __glusterd_handle_set_volume);
+}
 
-                ret = dict_unserialize (cli_req.bricks.bricks_val,
-                                        cli_req.bricks.bricks_len,
-                                        &dict);
-                if (ret < 0) {
-                        gf_log ("glusterd", GF_LOG_ERROR,
-                                "failed to "
-                                "unserialize req-buffer to dictionary");
-                        goto out;
-                } else {
-                        dict->extra_stdfree = cli_req.bricks.bricks_val;
-                }
+int
+__glusterd_handle_sync_volume(rpcsvc_request_t *req)
+{
+    int32_t ret = -1;
+    gf_cli_req cli_req = {{
+        0,
+    }};
+    dict_t *dict = NULL;
+    gf_cli_rsp cli_rsp = {0.};
+    char msg[2048] = {
+        0,
+    };
+    char *volname = NULL;
+    gf1_cli_sync_volume flags = 0;
+    char *hostname = NULL;
+    xlator_t *this = NULL;
+
+    GF_ASSERT(req);
+    this = THIS;
+    GF_ASSERT(this);
+
+    ret = xdr_to_generic(req->msg[0], &cli_req, (xdrproc_t)xdr_gf_cli_req);
+    if (ret < 0) {
+        // failed to decode msg;
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REQ_DECODE_FAIL, "%s",
+               "Failed to decode "
+               "request received from cli");
+        req->rpc_err = GARBAGE_ARGS;
+        goto out;
+    }
+
+    if (cli_req.dict.dict_len) {
+        /* Unserialize the dictionary */
+        dict = dict_new();
+
+        ret = dict_unserialize(cli_req.dict.dict_val, cli_req.dict.dict_len,
+                               &dict);
+        if (ret < 0) {
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_UNSERIALIZE_FAIL,
+                   "failed to "
+                   "unserialize req-buffer to dictionary");
+            snprintf(msg, sizeof(msg),
+                     "Unable to decode the "
+                     "command");
+            goto out;
+        } else {
+            dict->extra_stdfree = cli_req.dict.dict_val;
         }
+    }
 
-        ret = dict_get_str (dict, "volname", &volname);
+    ret = dict_get_strn(dict, "hostname", SLEN("hostname"), &hostname);
+    if (ret) {
+        snprintf(msg, sizeof(msg), "Failed to get hostname");
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_HOSTNAME_NOTFOUND_IN_DICT,
+               "%s", msg);
+        goto out;
+    }
 
+    ret = dict_get_strn(dict, "volname", SLEN("volname"), &volname);
+    if (ret) {
+        ret = dict_get_int32n(dict, "flags", SLEN("flags"), (int32_t *)&flags);
         if (ret) {
-                gf_log ("", GF_LOG_ERROR, "Unable to get volume name");
-                goto out;
-        }
+            snprintf(msg, sizeof(msg), "Failed to get volume name or flags");
+            gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_FLAGS_NOTFOUND_IN_DICT,
+                   "%s", msg);
+            goto out;
+        }
+    }
+
+    gf_msg(this->name, GF_LOG_INFO, 0, GD_MSG_VOL_SYNC_REQ_RCVD,
+           "Received volume sync req "
+           "for volume %s",
+           (flags & GF_CLI_SYNC_ALL) ? "all" : volname);
+
+    if (gf_is_local_addr(hostname)) {
+        ret = -1;
+        snprintf(msg, sizeof(msg),
+                 "sync from localhost"
+                 " not allowed");
+        gf_msg(this->name, GF_LOG_ERROR, 0,
+               GD_MSG_SYNC_FROM_LOCALHOST_UNALLOWED, "%s", msg);
+        goto out;
+    }
+
+    ret = glusterd_op_begin_synctask(req, GD_OP_SYNC_VOLUME, dict);
 
-        if (!(ret = glusterd_check_volume_exists (volname))) {
-                snprintf(err_str, 2048, "Volname %s does not exist",
-                         volname);
-                gf_log ("glusterd", GF_LOG_ERROR, "%s", err_str);
-                err_ret = -1;
-                goto out;
-        }
+out:
+    if (ret) {
+        cli_rsp.op_ret = -1;
+        cli_rsp.op_errstr = msg;
+        if (msg[0] == '\0')
+            snprintf(msg, sizeof(msg), "Operation failed");
+        glusterd_to_cli(req, &cli_rsp, NULL, 0, NULL, (xdrproc_t)xdr_gf_cli_rsp,
+                        dict);
+
+        ret = 0;  // sent error to cli, prevent second reply
+    }
+
+    return ret;
+}
 
-        ret = dict_get_int32 (dict, "count", &brick_count);
-        if (ret) {
-                gf_log ("", GF_LOG_ERROR, "Unable to get count");
-                goto out;
-        }
+int
+glusterd_handle_sync_volume(rpcsvc_request_t *req)
+{
+    return glusterd_big_locked_handler(req, __glusterd_handle_sync_volume);
+}
 
-        if (!(ret = glusterd_volinfo_find (volname, &volinfo))) {
-                if (volinfo->type == GF_CLUSTER_TYPE_NONE)
-                        goto brick_val;
-                if (!brick_count || !volinfo->sub_count)
-                        goto brick_val;
+int
+glusterd_fsm_log_send_resp(rpcsvc_request_t *req, int op_ret, char *op_errstr,
+                           dict_t *dict)
+{
+    int ret = -1;
+    gf1_cli_fsm_log_rsp rsp = {0};
 
-		/* If the brick count is less than sub_count then, allow add-brick only for 
-		   plain replicate volume since in plain stripe brick_count becoming less than 
-		   the sub_count is not allowed */
-                if (volinfo->brick_count < volinfo->sub_count && (volinfo->type == GF_CLUSTER_TYPE_REPLICATE) ) {
-                        if ((volinfo->sub_count - volinfo->brick_count) == brick_count)
-                                goto brick_val;
-                }
+    GF_ASSERT(req);
+    GF_ASSERT(op_errstr);
 
-                if ((brick_count % volinfo->sub_count) != 0) {
-                        snprintf(err_str, 2048, "Incorrect number of bricks"
-                                " supplied %d for type %s with count %d",
-                                brick_count, (volinfo->type == 1)? "STRIPE":
-                                "REPLICATE", volinfo->sub_count);
-                        gf_log("glusterd", GF_LOG_ERROR, "%s", err_str);
-                        err_ret = 1;
-                        goto out;
-                }
-        } else {
-                gf_log("", GF_LOG_ERROR, "Unable to get volinfo for volname"
-                       " %s", volname);
-                goto out;
+    rsp.op_ret = op_ret;
+    rsp.op_errstr = op_errstr;
+    if (rsp.op_ret == 0) {
+        ret = dict_allocate_and_serialize(dict, &rsp.fsm_log.fsm_log_val,
+                                          &rsp.fsm_log.fsm_log_len);
+        if (ret < 0) {
+            gf_smsg("glusterd", GF_LOG_ERROR, errno,
+                    GD_MSG_DICT_ALLOC_AND_SERL_LENGTH_GET_FAIL, NULL);
+            return ret;
         }
+    }
 
-brick_val:
-        ret = dict_get_str (dict, "bricks", &bricks);
-        if (ret) {
-                gf_log ("", GF_LOG_ERROR, "Unable to get bricks");
-                goto out;
-        }
+    ret = glusterd_submit_reply(req, &rsp, NULL, 0, NULL,
+                                (xdrproc_t)xdr_gf1_cli_fsm_log_rsp);
+    GF_FREE(rsp.fsm_log.fsm_log_val);
 
-        if (bricks)
-                brick_list = gf_strdup (bricks);
-        if (!brick_list) {
-                gf_log ("", GF_LOG_ERROR, "Out of memory");
-                ret = -1;
-                goto out;
-        } else {
-                free_ptr = brick_list;
-        }
+    gf_msg_debug("glusterd", 0, "Responded, ret: %d", ret);
 
-        gf_cmd_log ("Volume add-brick", "volname: %s type %s count:%d bricks:%s"
-                    ,volname, ((volinfo->type == 0)? "DEFAULT" : ((volinfo->type
-                    == 1)? "STRIPE": "REPLICATE")), brick_count, brick_list);
+    return 0;
+}
 
+int
+__glusterd_handle_fsm_log(rpcsvc_request_t *req)
+{
+    int32_t ret = -1;
+    gf1_cli_fsm_log_req cli_req = {
+        0,
+    };
+    dict_t *dict = NULL;
+    xlator_t *this = NULL;
+    glusterd_conf_t *conf = NULL;
+    char msg[2048] = {0};
+    glusterd_peerinfo_t *peerinfo = NULL;
+
+    GF_ASSERT(req);
+
+    this = THIS;
+    GF_VALIDATE_OR_GOTO("xlator", (this != NULL), out);
+
+    ret = xdr_to_generic(req->msg[0], &cli_req,
+                         (xdrproc_t)xdr_gf1_cli_fsm_log_req);
+    if (ret < 0) {
+        // failed to decode msg;
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REQ_DECODE_FAIL,
+               "Failed to decode "
+               "request received from client.");
+        req->rpc_err = GARBAGE_ARGS;
+        snprintf(msg, sizeof(msg), "Garbage request");
+        goto out;
+    }
+
+    dict = dict_new();
+    if (!dict) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno, GD_MSG_DICT_CREATE_FAIL, NULL);
+        ret = -1;
+        goto out;
+    }
+
+    if (strcmp("", cli_req.name) == 0) {
+        conf = this->private;
+        ret = glusterd_sm_tr_log_add_to_dict(dict, &conf->op_sm_log);
+    } else {
+        RCU_READ_LOCK;
 
-        while ( i < brick_count) {
-                i++;
-                brick= strtok_r (brick_list, " \n", &tmpptr);
-                brick_list = tmpptr;
-                brickinfo = NULL;
-                ret = glusterd_brickinfo_from_brick (brick, &brickinfo);
-                if (ret)
-                        goto out;
-                ret = glusterd_new_brick_validate (brick, brickinfo, err_str,
-                                                   sizeof (err_str));
-                if (ret) {
-                        err_ret = 1;
-                        goto out;
-                }
-                ret = glusterd_volume_brickinfo_get (brickinfo->uuid,
-                                                     brickinfo->hostname,
-                                                     brickinfo->path,
-                                                     &tmpvolinfo, &tmpbrkinfo);
-                if (!ret) {
-                        err_ret = 1;
-                        snprintf (err_str, sizeof (err_str), "Brick: %s:%s, %s"
-                                  " in the arguments mean the same",
-                                  tmpbrkinfo->hostname, tmpbrkinfo->path,
-                                  brick);
-                        goto out;
-                }
-                list_add_tail (&brickinfo->brick_list, &tmpvolinfo.bricks);
-                brickinfo = NULL;
+        peerinfo = glusterd_peerinfo_find_by_hostname(cli_req.name);
+        if (!peerinfo) {
+            RCU_READ_UNLOCK;
+            ret = -1;
+            snprintf(msg, sizeof(msg), "%s is not a peer", cli_req.name);
+        } else {
+            ret = glusterd_sm_tr_log_add_to_dict(dict, &peerinfo->sm_log);
+            RCU_READ_UNLOCK;
         }
+    }
 
-        ret = glusterd_add_brick (req, dict);
+out:
+    (void)glusterd_fsm_log_send_resp(req, ret, msg, dict);
+    free(cli_req.name);  // malloced by xdr
+    if (dict)
+        dict_unref(dict);
 
-        gf_cmd_log ("Volume add-brick","on volname: %s %s", volname,
-                   ((ret || err_ret) != 0)? "FAILED" : "SUCCESS");
+    glusterd_friend_sm();
+    glusterd_op_sm();
 
-out:
-        if ((err_ret || ret) && dict)
-                dict_unref (dict);
-        if (err_ret) {
-                rsp.op_ret = -1;
-                rsp.op_errno = 0;
-                rsp.volname = "";
-                rsp.op_errstr = err_str;
-                cli_rsp = &rsp;
-                glusterd_submit_reply(req, cli_rsp, NULL, 0, NULL,
-                                      gf_xdr_serialize_cli_add_brick_rsp);
-                if (!glusterd_opinfo_unlock())
-                        gf_log ("glusterd", GF_LOG_ERROR, "Unlock on "
-                               "opinfo failed");
-
-                ret = 0; //sent error to cli, prevent second reply
-        }
-
-        if (free_ptr)
-                GF_FREE (free_ptr);
-        glusterd_volume_bricks_delete (&tmpvolinfo);
-        if (brickinfo)
-                glusterd_brickinfo_delete (brickinfo);
-        if (cli_req.volname)
-                free (cli_req.volname); //its malloced by xdr
-        return ret;
+    return 0;  // send 0 to avoid double reply
 }
 
 int
-glusterd_handle_replace_brick (rpcsvc_request_t *req)
+glusterd_handle_fsm_log(rpcsvc_request_t *req)
 {
-        int32_t                         ret = -1;
-        gf1_cli_replace_brick_req          cli_req = {0,};
-        dict_t                          *dict = NULL;
-        char                            *src_brick = NULL;
-        char                            *dst_brick = NULL;
-        int32_t                         op = 0;
-        char                            operation[256];
-
-        GF_ASSERT (req);
-
-        if (!gf_xdr_to_cli_replace_brick_req (req->msg[0], &cli_req)) {
-                //failed to decode msg;
-                req->rpc_err = GARBAGE_ARGS;
-                goto out;
-        }
+    return glusterd_big_locked_handler(req, __glusterd_handle_fsm_log);
+}
 
-        gf_cmd_log ("Volume replace-brick","on volname: %s attempted", cli_req.volname);
+int
+glusterd_op_lock_send_resp(rpcsvc_request_t *req, int32_t status)
+{
+    gd1_mgmt_cluster_lock_rsp rsp = {
+        {0},
+    };
+    int ret = -1;
 
-        gf_log ("glusterd", GF_LOG_NORMAL, "Received replace brick req");
+    GF_ASSERT(req);
+    glusterd_get_uuid(&rsp.uuid);
+    rsp.op_ret = status;
 
-        if (cli_req.bricks.bricks_len) {
-                /* Unserialize the dictionary */
-                dict  = dict_new ();
+    ret = glusterd_submit_reply(req, &rsp, NULL, 0, NULL,
+                                (xdrproc_t)xdr_gd1_mgmt_cluster_lock_rsp);
 
-                ret = dict_unserialize (cli_req.bricks.bricks_val,
-                                        cli_req.bricks.bricks_len,
-                                        &dict);
-                if (ret < 0) {
-                        gf_log ("glusterd", GF_LOG_ERROR,
-                                "failed to "
-                                "unserialize req-buffer to dictionary");
-                        goto out;
-                } else {
-                        dict->extra_stdfree = cli_req.bricks.bricks_val;
-                }
-        }
+    gf_msg_debug(THIS->name, 0, "Responded to lock, ret: %d", ret);
 
-        ret = dict_get_int32 (dict, "operation", &op);
-        if (ret) {
-                gf_log ("", GF_LOG_DEBUG,
-                        "dict_get on operation failed");
-                goto out;
-        }
+    return 0;
+}
 
-        ret = dict_get_str (dict, "src-brick", &src_brick);
+int
+glusterd_op_unlock_send_resp(rpcsvc_request_t *req, int32_t status)
+{
+    gd1_mgmt_cluster_unlock_rsp rsp = {
+        {0},
+    };
+    int ret = -1;
 
-        if (ret) {
-                gf_log ("", GF_LOG_ERROR, "Unable to get src brick");
-                goto out;
-        }
-        gf_log ("", GF_LOG_DEBUG,
-                "src brick=%s", src_brick);
+    GF_ASSERT(req);
+    rsp.op_ret = status;
+    glusterd_get_uuid(&rsp.uuid);
 
-        ret = dict_get_str (dict, "dst-brick", &dst_brick);
+    ret = glusterd_submit_reply(req, &rsp, NULL, 0, NULL,
+                                (xdrproc_t)xdr_gd1_mgmt_cluster_unlock_rsp);
 
-        if (ret) {
-                gf_log ("", GF_LOG_ERROR, "Unable to get dest brick");
-                goto out;
-        }
+    gf_msg_debug(THIS->name, 0, "Responded to unlock, ret: %d", ret);
 
-        gf_log ("", GF_LOG_DEBUG,
-                "dst brick=%s", dst_brick);
+    return ret;
+}
 
-        switch (op) {
-                case GF_REPLACE_OP_START: strcpy (operation, "start");
-                        break;
-                case GF_REPLACE_OP_COMMIT: strcpy (operation, "commit");
-                        break;
-                case GF_REPLACE_OP_PAUSE:  strcpy (operation, "pause");
-                        break;
-                case GF_REPLACE_OP_ABORT:  strcpy (operation, "abort");
-                        break;
-                case GF_REPLACE_OP_STATUS: strcpy (operation, "status");
-                        break;
-                case GF_REPLACE_OP_COMMIT_FORCE: strcpy (operation, "commit-force");
-                        break;
-                default:strcpy (operation, "unknown");
-                        break;
-        }
+int
+glusterd_op_mgmt_v3_lock_send_resp(rpcsvc_request_t *req, uuid_t *txn_id,
+                                   int32_t status)
+{
+    gd1_mgmt_v3_lock_rsp rsp = {
+        {0},
+    };
+    int ret = -1;
 
-        gf_cmd_log ("Volume replace-brick","volname: %s src_brick:%s"
-                    " dst_brick:%s op:%s",cli_req.volname, src_brick, dst_brick
-                    ,operation);
+    GF_ASSERT(req);
+    GF_ASSERT(txn_id);
+    glusterd_get_uuid(&rsp.uuid);
+    rsp.op_ret = status;
+    if (rsp.op_ret)
+        rsp.op_errno = errno;
+    gf_uuid_copy(rsp.txn_id, *txn_id);
 
-        ret = glusterd_replace_brick (req, dict);
+    ret = glusterd_submit_reply(req, &rsp, NULL, 0, NULL,
+                                (xdrproc_t)xdr_gd1_mgmt_v3_lock_rsp);
 
-        gf_cmd_log ("Volume replace-brick","on volname: %s %s", cli_req.volname,
-                   (ret) ? "FAILED" : "SUCCESS");
+    gf_msg_debug(THIS->name, 0, "Responded to mgmt_v3 lock, ret: %d", ret);
 
-out:
-        if (ret && dict)
-                dict_unref (dict);
-        if (cli_req.volname)
-                free (cli_req.volname);//malloced by xdr
-        return ret;
+    return ret;
 }
 
-
-
-
 int
-glusterd_handle_reset_volume (rpcsvc_request_t *req)
+glusterd_op_mgmt_v3_unlock_send_resp(rpcsvc_request_t *req, uuid_t *txn_id,
+                                     int32_t status)
 {
-        int32_t                           ret = -1;
-        gf1_cli_reset_vol_req           cli_req = {0,};
-        dict_t                          *dict = NULL;
+    gd1_mgmt_v3_unlock_rsp rsp = {
+        {0},
+    };
+    int ret = -1;
 
-        GF_ASSERT (req);
+    GF_ASSERT(req);
+    GF_ASSERT(txn_id);
+    rsp.op_ret = status;
+    if (rsp.op_ret)
+        rsp.op_errno = errno;
+    glusterd_get_uuid(&rsp.uuid);
+    gf_uuid_copy(rsp.txn_id, *txn_id);
 
-        if (!gf_xdr_to_cli_set_vol_req (req->msg[0], &cli_req)) {
-                //failed to decode msg;
-                req->rpc_err = GARBAGE_ARGS;
-                goto out;
-        }
+    ret = glusterd_submit_reply(req, &rsp, NULL, 0, NULL,
+                                (xdrproc_t)xdr_gd1_mgmt_v3_unlock_rsp);
 
-        if (cli_req.dict.dict_len) {
-                /* Unserialize the dictionary */
-                dict  = dict_new ();
+    gf_msg_debug(THIS->name, 0, "Responded to mgmt_v3 unlock, ret: %d", ret);
 
-                ret = dict_unserialize (cli_req.dict.dict_val,
-                                        cli_req.dict.dict_len,
-                                        &dict);
-                if (ret < 0) {
-                        gf_log ("glusterd", GF_LOG_ERROR, "failed to "
-                                    "unserialize req-buffer to dictionary");
-                        goto out;
-                } else {
-                        dict->extra_stdfree = cli_req.dict.dict_val;
-                }
-        }
+    return ret;
+}
 
-        ret = glusterd_reset_volume (req, dict);
+int
+__glusterd_handle_cluster_unlock(rpcsvc_request_t *req)
+{
+    gd1_mgmt_cluster_unlock_req unlock_req = {
+        {0},
+    };
+    int32_t ret = -1;
+    glusterd_op_lock_ctx_t *ctx = NULL;
+    xlator_t *this = NULL;
+    uuid_t *txn_id = NULL;
+    glusterd_conf_t *priv = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+    priv = this->private;
+    GF_ASSERT(priv);
+    GF_ASSERT(req);
+
+    txn_id = &priv->global_txn_id;
+
+    ret = xdr_to_generic(req->msg[0], &unlock_req,
+                         (xdrproc_t)xdr_gd1_mgmt_cluster_unlock_req);
+    if (ret < 0) {
+        gf_msg(this->name, GF_LOG_ERROR, 0, GD_MSG_REQ_DECODE_FAIL,
+               "Failed to decode unlock "
+               "request received from peer");
+        req->rpc_err = GARBAGE_ARGS;
+        goto out;
+    }
+
+    gf_msg_debug(this->name, 0, "Received UNLOCK from uuid: %s",
+                 uuid_utoa(unlock_req.uuid));
+
+    RCU_READ_LOCK;
+    ret = (glusterd_peerinfo_find_by_uuid(unlock_req.uuid) == NULL);
+    RCU_READ_LOCK;
+    if (ret) {
+        gf_msg(this->name, GF_LOG_WARNING, 0, GD_MSG_PEER_NOT_FOUND,
+               "%s doesn't "
+               "belong to the cluster. Ignoring request.",
+               uuid_utoa(unlock_req.uuid));
+        ret = -1;
+        goto out;
+    }
+
+    ctx = GF_CALLOC(1, sizeof(*ctx), gf_gld_mt_op_lock_ctx_t);
+
+    if (!ctx) {
+        // respond here
+        gf_msg(this->name, GF_LOG_ERROR, ENOMEM, GD_MSG_NO_MEMORY,
+               "No memory.");
+        return -1;
+    }
+    gf_uuid_copy(ctx->uuid, unlock_req.uuid);
+    ctx->req = req;
+    ctx->dict = NULL;
+
+    ret = glusterd_op_sm_inject_event(GD_OP_EVENT_UNLOCK, txn_id, ctx);
 
 out:
-                if (cli_req.volname)
-                free (cli_req.volname);//malloced by xdr
+    glusterd_friend_sm();
+    glusterd_op_sm();
 
-        return ret;
+    return ret;
 }
 
 int
-glusterd_handle_set_volume (rpcsvc_request_t *req)
+glusterd_handle_cluster_unlock(rpcsvc_request_t *req)
 {
-        int32_t                           ret = -1;
-        gf1_cli_set_vol_req           cli_req = {0,};
-        dict_t                          *dict = NULL;
-
-        GF_ASSERT (req);
-
-        if (!gf_xdr_to_cli_set_vol_req (req->msg[0], &cli_req)) {
-                //failed to decode msg;
-                req->rpc_err = GARBAGE_ARGS;
-                goto out;
-        }
-
-        if (cli_req.dict.dict_len) {
-                /* Unserialize the dictionary */
-                dict  = dict_new ();
+    return glusterd_big_locked_handler(req, __glusterd_handle_cluster_unlock);
+}
 
-                ret = dict_unserialize (cli_req.dict.dict_val,
-                                        cli_req.dict.dict_len,
-                                        &dict);
-                if (ret < 0) {
-                        gf_log ("glusterd", GF_LOG_ERROR,
-                                "failed to "
-                                "unserialize req-buffer to dictionary");
-                        goto out;
-                } else {
-                        dict->extra_stdfree = cli_req.dict.dict_val;
-                }
-        }
+int
+glusterd_op_stage_send_resp(rpcsvc_request_t *req, int32_t op, int32_t status,
+                            char *op_errstr, dict_t *rsp_dict)
+{
+    gd1_mgmt_stage_op_rsp rsp = {
+        {0},
+    };
+    int ret = -1;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+    GF_ASSERT(req);
+
+    rsp.op_ret = status;
+    glusterd_get_uuid(&rsp.uuid);
+    rsp.op = op;
+    if (op_errstr)
+        rsp.op_errstr = op_errstr;
+    else
+        rsp.op_errstr = "";
+
+    ret = dict_allocate_and_serialize(rsp_dict, &rsp.dict.dict_val,
+                                      &rsp.dict.dict_len);
+    if (ret < 0) {
+        gf_smsg(this->name, GF_LOG_ERROR, errno,
+                GD_MSG_DICT_ALLOC_AND_SERL_LENGTH_GET_FAIL, NULL);
+        return ret;
+    }
 
-        ret = glusterd_set_volume (req, dict);
+    ret = glusterd_submit_reply(req, &rsp, NULL, 0, NULL,
+                                (xdrproc_t)xdr_gd1_mgmt_stage_op_rsp);
 
-out:
-        if (cli_req.volname)
-                free (cli_req.volname);//malloced by xdr
+    gf_msg_debug(this->name, 0, "Responded to stage, ret: %d", ret);
+    GF_FREE(rsp.dict.dict_val);
 
-        return ret;
+    return ret;
 }
 
 int
-glusterd_handle_remove_brick (rpcsvc_request_t *req)
-{
-        int32_t                         ret = -1;
-        gf1_cli_remove_brick_req        cli_req = {0,};
-        dict_t                          *dict = NULL;
-        int32_t                         count = 0;
-        char                            *brick = NULL;
-        char                            key[256] = {0,};
-        char                            *brick_list = NULL;
-        int                             i = 1;
-        glusterd_volinfo_t              *volinfo = NULL;
-        glusterd_brickinfo_t            *brickinfo = NULL;
-        int32_t                         pos = 0;
-        int32_t                         sub_volume = 0;
-        int32_t                         sub_volume_start = 0;
-        int32_t                         sub_volume_end = 0;
-        glusterd_brickinfo_t            *tmp = NULL;
-        int32_t                         err_ret = 0;
-        char                            *err_str = NULL;
-        gf1_cli_remove_brick_rsp        rsp = {0,};
-        void                            *cli_rsp = NULL;
-        char                            vol_type[256] = {0,};
-
-        GF_ASSERT (req);
-
-        if (!gf_xdr_to_cli_remove_brick_req (req->msg[0], &cli_req)) {
-                //failed to decode msg;
-                req->rpc_err = GARBAGE_ARGS;
-                goto out;
+glusterd_op_commit_send_resp(rpcsvc_request_t *req, int32_t op, int32_t status,
+                             char *op_errstr, dict_t *rsp_dict)
+{
+    gd1_mgmt_commit_op_rsp rsp = {
+        {0},
+    };
+    int ret = -1;
+    xlator_t *this = NULL;
+
+    this = THIS;
+    GF_ASSERT(this);
+    GF_ASSERT(req);
+    rsp.op_ret = status;
+    glusterd_get_uuid(&rsp.uuid);
+    rsp.op = op;
+
+    if (op_errstr)
+        rsp.op_errstr = op_errstr;
+    else
+        rsp.op_errstr = "";
+
+    if (rsp_dict) {
+        ret = dict_allocate_and_serialize(rsp_dict, &rsp.dict.dict_val,
+                                          &rsp.dict.dict_len);
+        if (ret < 0) {
+            gf_smsg(this->name, GF_LOG_ERROR, errno,
+                    GD_MSG_DICT_ALLOC_AND_SERL_LENGTH_GET_FAIL, NULL);
+            goto out;
         }
+    }
 
-        gf_cmd_log ("Volume remove-brick","on volname: %s attempted",cli_req.volname);
-        gf_log ("glusterd", GF_LOG_NORMAL, "Received rem brick req");
+    ret = glusterd_submit_reply(req, &rsp, NULL, 0, NULL,
+                                (xdrproc_t)xdr_gd1_mgmt_commit_op_rsp);
 
-        if (cli_req.bricks.bricks_len) {
-                /* Unserialize the dictionary */
-                dict  = dict_new ();
+    gf_msg_debug(this->name, 0, "Responded to commit, ret: %d", ret);
 
-                ret = dict_unserialize (cli_req.bricks.bricks_val,
-                                        cli_req.bricks.bricks_len,
-                                        &dict);
-                if (ret < 0) {
-                        gf_log ("glusterd", GF_LOG_ERROR,
-                                "failed to "
-                                "unserialize req-buffer to dictionary");
-                        goto out;
-                } else {
-                        dict->extra_stdfree = cli_req.bricks.bricks_val;
-                }
-        }
-
-        ret = dict_get_int32 (dict, "count", &count);
-        if (ret) {
-                gf_log ("", GF_LOG_ERROR, "Unable to get count");
-                goto out;
-        }
-
-        err_str = GF_MALLOC (2048 * sizeof(*err_str),gf_common_mt_char);
+out:
+    GF_FREE(rsp.dict.dict_val);
+    return ret;
+}
 
-        if (!err_str) {
-                gf_log ("",GF_LOG_ERROR,"glusterd_handle_remove_brick: "
-                        "Unable to get memory");
-                ret = -1;
-                goto out;
-        }
+int
+__glusterd_handle_incoming_friend_req(rpcsvc_request_t *req)
+{
+    int32_t ret = -1;
+    gd1_mgmt_friend_req friend_req = {
+        {0},
+    };
+    gf_boolean_t run_fsm = _gf_true;
+
+    GF_ASSERT(req);
+    ret = xdr_to_generic(req->msg[0], &friend_req,
+                         (xdrproc_t)xdr_gd1_mgmt_friend_req);
+    if (ret < 0) {
+        // failed to decode msg;
+        gf_msg("glusterd", GF_LOG_ERROR, 0, GD_MSG_REQ_DECODE_FAIL,
+               "Failed to decode "
+               "request received from friend");
+        req->rpc_err = GARBAGE_ARGS;
+        goto out;
+    }
+
+    gf_msg("glusterd", GF_LOG_INFO, 0, GD_MSG_PROBE_RCVD,
+           "Received probe from uuid: %s", uuid_utoa(friend_req.uuid));
+    ret = glusterd_handle_friend_req(req, friend_req.uuid, friend_req.hostname,
+                                     friend_req.port, &friend_req);
+
+    if (ret == GLUSTERD_CONNECTION_AWAITED) {
+        // fsm should be run after connection establishes
+        run_fsm = _gf_false;
+        ret = 0;
+    }
 
-        ret = glusterd_volinfo_find (cli_req.volname, &volinfo);
-        if (ret) {
-                 snprintf (err_str, 2048, "volname %s not found",
-                          cli_req.volname);
-                 gf_log ("", GF_LOG_ERROR, "%s", err_str);
-                 err_ret = 1;
-                 goto out;
-        }
-
-        if (volinfo->type == GF_CLUSTER_TYPE_REPLICATE)
-                strcpy (vol_type, "replica");
-        else if (volinfo->type == GF_CLUSTER_TYPE_STRIPE)
-                strcpy (vol_type, "stripe");
-        else
-                strcpy (vol_type, "distribute");
-
-	/* Do not allow remove-brick if the volume is plain stripe */
-	if ((volinfo->type == GF_CLUSTER_TYPE_STRIPE) && (volinfo->brick_count == volinfo->sub_count)) {
-                snprintf (err_str, 2048, "Removing brick from a plain stripe is not allowed");
-                gf_log ("glusterd", GF_LOG_ERROR, "%s", err_str);
-                err_ret = 1;
-                ret = -1;
-                goto out;
-	}
-
-	/* Do not allow remove-brick if the bricks given is less than the replica count
-	   or stripe count */
-        if (((volinfo->type == GF_CLUSTER_TYPE_REPLICATE) || (volinfo->type == GF_CLUSTER_TYPE_STRIPE)) 
-	    && !(volinfo->brick_count <= volinfo->sub_count)) {
-                if (volinfo->sub_count && (count % volinfo->sub_count != 0)) {
-                        snprintf (err_str, 2048, "Remove brick incorrect"
-                                  " brick count of %d for %s %d",
-                                  count, vol_type, volinfo->sub_count);
-                        gf_log ("", GF_LOG_ERROR, "%s", err_str);
-                        err_ret = 1;
-                        ret = -1;
-                        goto out;
-                }
-        }
+out:
+    free(friend_req.hostname);  // malloced by xdr
 
-        brick_list = GF_MALLOC (120000 * sizeof(*brick_list),gf_common_mt_char);
+    if (run_fsm) {
+        glusterd_friend_sm();
+        glusterd_op_sm();
+    }
 
-        if (!brick_list) {
-                gf_log ("",GF_LOG_ERROR,"glusterd_handle_remove_brick: "
-                        "Unable to get memory");
-                ret = -1;
-                goto out;
-        }
+    return ret;
+}
 
-        strcpy (brick_list, " ");
-        while ( i <= count) {
-                snprintf (key, 256, "brick%d", i);
-                ret = dict_get_str (dict, key, &brick);
-                if (ret) {
-                        gf_log ("", GF_LOG_ERROR, "Unable to get %s", key);
-                        goto out;
-                }
-                gf_log ("", GF_LOG_DEBUG, "Remove brick count %d brick: %s",
-                        i, brick);
+int
+glusterd_handle_incoming_friend_req(rpcsvc_request_t *req)
+{
+    return glusterd_big_locked_handler(req,
+                                       __glusterd_handle_incoming_friend_req);
+}
 
-                ret = glusterd_volume_brickinfo_get_by_brick(brick, volinfo, &brickinfo);
-                if (ret) {
-                        snprintf(err_str, 2048," Incorrect brick %s for volname"
-                                " %s", brick, cli_req.volname);
-                        gf_log ("", GF_LOG_ERROR, "%s", err_str);
-                        err_ret = 1;
-                        goto out;
-                }
-                strcat(brick_list, brick);
-                strcat(brick_list, " ");
-
-                i++;
-                if ((volinfo->type == GF_CLUSTER_TYPE_NONE) ||
-                    (volinfo->brick_count <= volinfo->sub_count))
-                        continue;
- 
-                pos = 0;
-                list_for_each_entry (tmp, &volinfo->bricks, brick_list) {
-
-                        if ((!strcmp (tmp->hostname,brickinfo->hostname)) &&
-                            !strcmp (tmp->path, brickinfo->path)) {
-                                gf_log ("", GF_LOG_NORMAL, "Found brick");
-                                if (!sub_volume && volinfo->sub_count) {
-                                        sub_volume = (pos / volinfo->
-                                                      sub_count) + 1;
-                                        sub_volume_start = volinfo->sub_count *
-                                                           (sub_volume - 1);
-                                        sub_volume_end = (volinfo->sub_count *
-                                                          sub_volume) -1 ;
-                                } else {
-                                        if (pos < sub_volume_start ||
-                                            pos >sub_volume_end) {
-                                                ret = -1;
-                                                snprintf(err_str, 2048,"Bricks"
-                                                         " not from same subvol"
-                                                         " for %s", vol_type);
-                                                gf_log ("",GF_LOG_ERROR,
-                                                        "%s", err_str);
-                                                err_ret = 1;
-                                                goto out;
-                                        }
-                                }
-                                break;
-                        }
-                        pos++;
-                }
-        }
-        gf_cmd_log ("Volume remove-brick","volname: %s count:%d bricks:%s",
-                    cli_req.volname, count, brick_list);
+int
+__glusterd_handle_incoming_unfriend_req(rpcsvc_request_t *req)
+{
+    int32_t ret = -1;
+    gd1_mgmt_friend_req friend_req = {
+        {0},
+    };
+    char remote_hostname[UNIX_PATH_MAX + 1] = {
+        0,
+    };
+
+    GF_ASSERT(req);
+    ret = xdr_to_generic(req->msg[0], &friend_req,
+                         (xdrproc_t)xdr_gd1_mgmt_friend_req);
+    if (ret < 0) {
+        // failed to decode msg;
+        gf_msg("glusterd", GF_LOG_ERROR, 0, GD_MSG_REQ_DECODE_FAIL,
+               "Failed to decode "
+               "request received.");
+        req->rpc_err = GARBAGE_ARGS;
+        goto out;
+    }
+
+    gf_msg("glusterd", GF_LOG_INFO, 0, GD_MSG_UNFRIEND_REQ_RCVD,
+           "Received unfriend from uuid: %s", uuid_utoa(friend_req.uuid));
+
+    ret = glusterd_remote_hostname_get(req, remote_hostname,
+                                       sizeof(remote_hostname));
+    if (ret) {
+        gf_msg("glusterd", GF_LOG_ERROR, 0, GD_MSG_HOSTNAME_RESOLVE_FAIL,
+               "Unable to get the remote hostname");
+        goto out;
+    }
+    ret = glusterd_handle_unfriend_req(req, friend_req.uuid, remote_hostname,
+                                       friend_req.port);
 
-        ret = glusterd_remove_brick (req, dict);
+out:
+    free(friend_req.hostname);       // malloced by xdr
+    free(friend_req.vols.vols_val);  // malloced by xdr
 
-        gf_cmd_log ("Volume remove-brick","on volname: %s %s",cli_req.volname,
-                    (ret) ? "FAILED" : "SUCCESS");
+    glusterd_friend_sm();
+    glusterd_op_sm();
 
-out:
-        if ((ret || err_ret) && dict)
-                dict_unref (dict);
-        if (err_ret) {
-                rsp.op_ret = -1;
-                rsp.op_errno = 0;
-                rsp.volname = "";
-                rsp.op_errstr = err_str;
-                cli_rsp = &rsp;
-                glusterd_submit_reply(req, cli_rsp, NULL, 0, NULL,
-                                      gf_xdr_serialize_cli_remove_brick_rsp);
-                if (!glusterd_opinfo_unlock())
-                        gf_log ("glusterd", GF_LOG_ERROR, "Unlock on "
-                               "opinfo failed");
-
-                ret = 0; //sent error to cli, prevent second reply
-
-        }
-        if (brick_list)
-                GF_FREE (brick_list);
-        if (err_str)
-                GF_FREE (err_str);
-        if (cli_req.volname)
-                free (cli_req.volname); //its malloced by xdr
-        return ret;
+    return ret;
 }
 
 int
-glusterd_handle_log_filename (rpcsvc_request_t *req)
+glusterd_handle_incoming_unfriend_req(rpcsvc_request_t *req)
 {
-        int32_t                   ret     = -1;
-        gf1_cli_log_filename_req  cli_req = {0,};
-        dict_t                   *dict    = NULL;

path: root/cli/src/cli-cmd-parser.c

/*
  Copyright (c) 2010 Gluster, Inc. <http://www.gluster.com>
  This file is part of GlusterFS.

  GlusterFS is free software; you can redistribute it and/or modify
  it under the terms of the GNU Affero General Public License as published
  by the Free Software Foundation; either version 3 of the License,
  or (at your option) any later version.

  GlusterFS is distributed in the hope that it will be useful, but
  WITHOUT ANY WARRANTY; without even the implied warranty of
  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  Affero General Public License for more details.

  You should have received a copy of the GNU Affero General Public License
  along with this program.  If not, see
  <http://www.gnu.org/licenses/>.
*/

#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <stdint.h>
#include <pthread.h>

#ifndef _CONFIG_H
#define _CONFIG_H
#include "config.h"
#endif

#include "cli.h"
#include "cli-cmd.h"
#include "cli-mem-types.h"
#include "dict.h"

#include "protocol-common.h"
#include "cli1-xdr.h"

int32_t
cli_cmd_bricks_parse (const char **words, int wordcount, int brick_index,
                      char **bricks, int *brick_count)
{
        int                    ret = 0;
        char             *tmp_list = NULL;
        char    brick_list[120000] = {0,};
        char                *space = " ";
        char            *delimiter = NULL;
        char            *host_name = NULL;
        char                  *tmp = NULL;
        char        *free_list_ptr = NULL;
        char               *tmpptr = NULL;
        int                      j = 0;
        int         brick_list_len = 0;

        GF_ASSERT (words);
        GF_ASSERT (wordcount);
        GF_ASSERT (bricks);
        GF_ASSERT (brick_index > 0);
        GF_ASSERT (brick_index < wordcount);

        strncpy (brick_list, space, strlen (space));
        brick_list_len++;
        while (brick_index < wordcount) {
                delimiter = strchr (words[brick_index], ':');
                if (!delimiter || delimiter == words[brick_index]
                    || *(delimiter+1) != '/') {
                        cli_out ("wrong brick type: %s, use <HOSTNAME>:"
                                 "<export-dir-abs-path>", words[brick_index]);
                        ret = -1;
                        goto out;
                } else {
                        cli_path_strip_trailing_slashes (delimiter + 1);
                }

                if ((brick_list_len + strlen (words[brick_index]) + 1) > sizeof (brick_list)) {
                        gf_log ("cli", GF_LOG_ERROR,
                                "total brick list is larger than a request "
                                "can take (brick_count %d)", *brick_count);
                        ret = -1;
                        goto out;
                }

                host_name = gf_strdup (words[brick_index]);
                if (!host_name) {
                        ret = -1;
                        gf_log("cli",GF_LOG_ERROR, "Unable to allocate "
                               "memory");
                        goto out;
                }

                strtok_r (host_name, ":", &tmp);
                if (!(strcmp (host_name, "localhost") &&
                      strcmp (host_name, "127.0.0.1"))) {
                        cli_out ("Please provide a valid hostname/ip other "
                                 "than localhost or 127.0.0.1");
                        ret = -1;
                        GF_FREE (host_name);
                        goto out;
                }
                GF_FREE (host_name);
                tmp_list = gf_strdup (brick_list + 1);
                if (free_list_ptr) {
                        GF_FREE (free_list_ptr);
                        free_list_ptr = NULL;
                }
                free_list_ptr = tmp_list;
                j = 0;
                while(j < *brick_count) {
                        strtok_r (tmp_list, " ", &tmpptr);
                        if (!(strcmp (tmp_list, words[brick_index]))) {
                                ret = -1;
                                cli_out ("Found duplicate"
                                         " exports %s",words[brick_index]);
                                goto out;
                       }
                       tmp_list = tmpptr;
                       j++;
                }
                strcat (brick_list, words[brick_index]);
                strcat (brick_list, " ");
                brick_list_len += (strlen (words[brick_index]) + 1);
                ++(*brick_count);
                ++brick_index;
        }

        *bricks = gf_strdup (brick_list);
        if (!*bricks)
                ret = -1;
out:
        if (free_list_ptr)
                GF_FREE (free_list_ptr);
        return ret;
}

int32_t
cli_cmd_volume_create_parse (const char **words, int wordcount, dict_t **options)
{
        dict_t  *dict = NULL;
        char    *volname = NULL;
        int     ret = -1;
        gf1_cluster_type type = GF_CLUSTER_TYPE_NONE;
        int     count = 1;
        int     brick_index = 0;
        int     i = 0;
        char    *trans_type = NULL;
        int32_t index = 0;
        char    *bricks = NULL;
        int32_t brick_count = 0;

        GF_ASSERT (words);
        GF_ASSERT (options);

        GF_ASSERT ((strcmp (words[0], "volume")) == 0);
        GF_ASSERT ((strcmp (words[1], "create")) == 0);

        dict = dict_new ();

        if (!dict)
                goto out;

        if (wordcount < 3)
                goto out;

        volname = (char *)words[2];

        GF_ASSERT (volname);

        /* Validate the volume name here itself */
        {
                if (volname[0] == '-')
                        goto out;

                if (!strcmp (volname, "all")) {
                        cli_out ("\"all\" cannot be the name of a volume.");
                        goto out;
                }

                if (strchr (volname, '/'))
                        goto out;

                if (strlen (volname) > 512)
                        goto out;

                for (i = 0; i < strlen (volname); i++)
                        if (!isalnum (volname[i]) && (volname[i] != '_') && (volname[i] != '-'))
                                goto out;
        }


        ret = dict_set_str (dict, "volname", volname);
        if (ret)
                goto out;

        if (wordcount < 4) {
                ret = -1;
                goto out;
        }
        if ((strcasecmp (words[3], "replica")) == 0) {
                type = GF_CLUSTER_TYPE_REPLICATE;
                if (wordcount < 5) {
                        ret = -1;
                        goto out;
                }
                count = strtol (words[4], NULL, 0);
                if (!count || (count < 2)) {
                        cli_out ("replica count should be greater than 1");
                        ret = -1;
                        goto out;
                }
                ret = dict_set_int32 (dict, "replica-count", count);
                if (ret)
                        goto out;
                brick_index = 5;
        } else if ((strcasecmp (words[3], "stripe")) == 0) {
                type = GF_CLUSTER_TYPE_STRIPE;
                if (wordcount < 5) {
                        ret = -1;
                        goto out;
                }
                count = strtol (words[4], NULL, 0);
                if (!count || (count < 2)) {
                        cli_out ("stripe count should be greater than 1");
                        ret = -1;
                        goto out;
                }
                ret = dict_set_int32 (dict, "stripe-count", count);
                if (ret)
                        goto out;
                brick_index = 5;
        } else {
                type = GF_CLUSTER_TYPE_NONE;
                brick_index = 3;
        }

        ret = dict_set_int32 (dict, "type", type);
        if (ret)
                goto out;

        if (type)
                index = 5;
        else
                index = 3;

        if (wordcount < (index + 1)) {
                ret = -1;
                goto out;
        }

        if (strcasecmp(words[index], "transport") == 0) {
                brick_index = index+2;
                if (wordcount < (index + 2)) {
                        ret = -1;
                        goto out;
                }

                if ((strcasecmp (words[index+1], "tcp") == 0)) {
                        trans_type = gf_strdup ("tcp");
                } else if ((strcasecmp (words[index+1], "rdma") == 0)) {
                        trans_type = gf_strdup ("rdma");
                } else if ((strcasecmp (words[index+1], "tcp,rdma") == 0) ||
                           (strcasecmp (words[index+1], "rdma,tcp") == 0)) {
                        trans_type = gf_strdup ("tcp,rdma");
                } else {
                        gf_log ("", GF_LOG_ERROR, "incorrect transport"
                                       " protocol specified");
                        ret = -1;
                        goto out;
                }
        } else {
                trans_type = gf_strdup ("tcp");
        }

        ret = dict_set_dynstr (dict, "transport", trans_type);
        if (ret)
                goto out;

        ret = cli_cmd_bricks_parse (words, wordcount, brick_index, &bricks,
                                    &brick_count);
        if (ret)
                goto out;

        /* If brick-count is not valid when replica or stripe is
           given, exit here */
        if (!brick_count) {
                cli_out ("No bricks specified");
                ret = -1;
                goto out;
        }

        if (brick_count % count) {
                if (type == GF_CLUSTER_TYPE_STRIPE)
                        cli_out ("number of bricks is not a multiple of "
                                 "stripe count");
                else if (type == GF_CLUSTER_TYPE_REPLICATE)
                        cli_out ("number of bricks is not a multiple of "
                                 "replica count");
                ret = -1;
                goto out;
        }

        ret = dict_set_dynstr (dict, "bricks", bricks);
        if (ret)
                goto out;

        ret = dict_set_int32 (dict, "count", brick_count);
        if (ret)
                goto out;

        *options = dict;

out:
        if (ret) {
                gf_log ("cli", GF_LOG_ERROR, "Unable to parse create volume CLI");
                if (dict)
                        dict_destroy (dict);
        }
        return ret;
}

int32_t
cli_cmd_volume_reset_parse (const char **words, int wordcount, dict_t **options)
{
        dict_t  *dict = NULL;
        char    *volname = NULL;
        int     ret = -1;

        GF_ASSERT (words);
        GF_ASSERT (options);

        GF_ASSERT ((strcmp (words[0], "volume")) == 0);
        GF_ASSERT ((strcmp (words[1], "reset")) == 0);

        dict = dict_new ();

        if (!dict)
                goto out;

        if (wordcount < 3)
                goto out;

        volname = (char *)words[2];

        if (!volname) {
                ret = -1;
                goto out;
        }

        ret = dict_set_str (dict, "volname", volname);

        if (ret)
                goto out;

        *options = dict;

out:
                if (ret) {
        if (dict)
                dict_destroy (dict);
                }

                return ret;
}

int32_t
cli_cmd_quota_parse (const char **words, int wordcount, dict_t **options)
{
        dict_t          *dict    = NULL;
        char            *volname = NULL;
        int              ret     = -1;
        int              i       = 0;
        char             key[20] = {0, };
        gf_quota_type    type    = GF_QUOTA_OPTION_TYPE_NONE;

        GF_ASSERT (words);
        GF_ASSERT (options);

        GF_ASSERT ((strcmp (words[0], "volume")) == 0);
        GF_ASSERT ((strcmp (words[1], "quota")) == 0);

        dict = dict_new ();

        if (!dict)
                goto out;

        if (wordcount < 4)
                goto out;

        volname = (char *)words[2];

        if (!volname) {
                ret = -1;
                goto out;
        }

        /* Validate the volume name here itself */
        {
                if (volname[0] == '-')
                        goto out;

                if (!strcmp (volname, "all")) {
                        cli_out ("\"all\" cannot be the name of a volume.");
                        goto out;
                }

                if (strchr (volname, '/'))
                        goto out;

                if (strlen (volname) > 512)
                        goto out;

                for (i = 0; i < strlen (volname); i++)
                        if (!isalnum (volname[i]) && (volname[i] != '_') && (volname[i] != '-'))
                                goto out;
        }

        ret = dict_set_str (dict, "volname", volname);

        if (ret)
                goto out;


        if ((strcasecmp (words[3], "enable")) == 0) {
                type = GF_QUOTA_OPTION_TYPE_ENABLE;
                goto set_type;
        }

        if (strcasecmp (words[3], "disable") == 0) {
                type = GF_QUOTA_OPTION_TYPE_DISABLE;
                goto set_type;
        }

        if (strcasecmp (words[3], "limit-usage") == 0) {
                if (wordcount != 6) {
                        ret = -1;
                        goto out;
                }

                type = GF_QUOTA_OPTION_TYPE_LIMIT_USAGE;

                if (words[4][0] != '/') {
                        cli_out ("Please enter absolute path");

                        return -2;
                }
                ret = dict_set_str (dict, "path", (char *) words[4]);

                if (ret)
                        goto out;

                if (!words[5]) {
                        gf_log ("cli", GF_LOG_ERROR, "Please enter the limit value "
                                            "to be set");

                        return -2;
                }

                ret = dict_set_str (dict, "limit", (char *) words[5]);
                if (ret)
                        goto out;

                goto set_type;
        }
        if (strcasecmp (words[3], "remove") == 0) {
                if (wordcount != 5) {
                        ret = -1;
                        goto out;
                }

                type = GF_QUOTA_OPTION_TYPE_REMOVE;

                if (words[4][0] != '/') {
                        cli_out ("Please enter absolute path");

                        return -2;
                }

                ret = dict_set_str (dict, "path", (char *) words[4]);
                if (ret)
                        goto out;
                goto set_type;
        }

        if (strcasecmp (words[3], "list") == 0) {
                        if (wordcount < 4) {
                                ret = -1;
                                goto out;
                        }

                        type = GF_QUOTA_OPTION_TYPE_LIST;

                        i = 4;
                        while (i < wordcount) {
                                snprintf (key, 20, "path%d", i-4);

                                ret = dict_set_str (dict, key, (char *) words [i++]);
                                if (ret < 0)
                                        goto out;
                        }

                        ret = dict_set_int32 (dict, "count", i - 4);
                        if (ret < 0)
                                goto out;

                        goto set_type;
        }

        if (strcasecmp (words[3], "version") == 0) {
                type = GF_QUOTA_OPTION_TYPE_VERSION;

        } else {
                ret = -1;
                goto out;
        }

set_type:
        ret = dict_set_int32 (dict, "type", type);

        if (ret)
                goto out;
        *options = dict;
out:
        if (ret) {
                if (dict)
                        dict_destroy (dict);
        }

        return ret;
}

int32_t
cli_cmd_volume_set_parse (const char **words, int wordcount, dict_t **options)
{
        dict_t  *dict = NULL;
        char    *volname = NULL;
        int     ret = -1;
        int     count = 0;
        char    *key = NULL;
        char    *value = NULL;
        int     i = 0;
        char    str[50] = {0,};

        GF_ASSERT (words);
        GF_ASSERT (options);

        GF_ASSERT ((strcmp (words[0], "volume")) == 0);
        GF_ASSERT ((strcmp (words[1], "set")) == 0);

        dict = dict_new ();

        if (!dict)
                goto out;

        if (wordcount < 4)
                goto out;

        volname = (char *)words[2];

        GF_ASSERT (volname);

        ret = dict_set_str (dict, "volname", volname);

        if (ret)
                goto out;


        for (i = 3; i < wordcount; i+=2) {

		key = (char *) words[i];
		value = (char *) words[i+1];

		if ( !key || !value) {
			ret = -1;
			goto out;
        	}

                count++;

                sprintf (str, "key%d", count);
                ret = dict_set_str (dict, str, key);
                if (ret)
                        goto out;

                sprintf (str, "value%d", count);
                ret = dict_set_str (dict, str, value);

                if (ret)
                        goto out;
        }

        ret = dict_set_int32 (dict, "count", wordcount-3);

        if (ret)
                goto out;

        *options = dict;

out:
        if (ret) {
                if (dict)
                        dict_destroy (dict);
        }

        return ret;
}

int32_t
cli_cmd_volume_add_brick_parse (const char **words, int wordcount,
                                dict_t **options)
{
        dict_t  *dict = NULL;
        char    *volname = NULL;
        int     ret = -1;
        int     brick_count = 0, brick_index = 0;
        char    *bricks = NULL;

        GF_ASSERT (words);
        GF_ASSERT (options);

        GF_ASSERT ((strcmp (words[0], "volume")) == 0);
        GF_ASSERT ((strcmp (words[1], "add-brick")) == 0);

        dict = dict_new ();

        if (!dict)
                goto out;

        if (wordcount < 3)
                goto out;

        volname = (char *)words[2];

        GF_ASSERT (volname);

        ret = dict_set_str (dict, "volname", volname);

        if (ret)
                goto out;

        if (wordcount < 4) {
                ret = -1;
                goto out;
        }

        brick_index = 3;
        ret = cli_cmd_bricks_parse (words, wordcount, brick_index, &bricks,
                                    &brick_count);
        if (ret)
                goto out;

        ret = dict_set_dynstr (dict, "bricks", bricks);
        if (ret)
                goto out;

        ret = dict_set_int32 (dict, "count", brick_count);

        if (ret)
                goto out;

        *options = dict;

out:
        if (ret) {
                gf_log ("cli", GF_LOG_ERROR, "Unable to parse add-brick CLI");
                if (dict)
                        dict_destroy (dict);
        }

        return ret;
}


int32_t
cli_cmd_volume_remove_brick_parse (const char **words, int wordcount,
                                   dict_t **options)
{
        dict_t  *dict = NULL;
        char    *volname = NULL;
        char    *delimiter = NULL;
        int     ret = -1;
        char    key[50];
        int     brick_count = 0, brick_index = 0;
        int32_t tmp_index = 0;
        int32_t j = 0;
        char    *tmp_brick = NULL;
        char    *tmp_brick1 = NULL;

        GF_ASSERT (words);
        GF_ASSERT (options);

        GF_ASSERT ((strcmp (words[0], "volume")) == 0);
        GF_ASSERT ((strcmp (words[1], "remove-brick")) == 0);

        dict = dict_new ();

        if (!dict)
                goto out;

        if (wordcount < 3)
                goto out;

        volname = (char *)words[2];

        GF_ASSERT (volname);

        ret = dict_set_str (dict, "volname", volname);

        if (ret)
                goto out;

        if (wordcount < 4) {
                ret = -1;
                goto out;
        }

        brick_index = 3;

        if (ret)
                goto out;

        tmp_index = brick_index;
        tmp_brick = GF_MALLOC(2048 * sizeof(*tmp_brick), gf_common_mt_char);

        if (!tmp_brick) {
                gf_log ("",GF_LOG_ERROR,"cli_cmd_volume_remove_brick_parse: "
                        "Unable to get memory");
                ret = -1;
                goto out;
        }
 
        tmp_brick1 = GF_MALLOC(2048 * sizeof(*tmp_brick1), gf_common_mt_char);

        if (!tmp_brick1) {
                gf_log ("",GF_LOG_ERROR,"cli_cmd_volume_remove_brick_parse: "
                        "Unable to get memory");
                ret = -1;
                goto out;
        }

        while (brick_index < wordcount) {
                delimiter = strchr(words[brick_index], ':');
                if (!delimiter || delimiter == words[brick_index]
                    || *(delimiter+1) != '/') {
                        cli_out ("wrong brick type: %s, use <HOSTNAME>:"
                                 "<export-dir-abs-path>", words[brick_index]);
                        ret = -1;
                        goto out;
                } else {
                        cli_path_strip_trailing_slashes (delimiter + 1);
                }

                j = tmp_index;
                strcpy(tmp_brick, words[brick_index]);
                while ( j < brick_index) {
                        strcpy(tmp_brick1, words[j]);
                        if (!(strcmp (tmp_brick, tmp_brick1))) {
                                gf_log("",GF_LOG_ERROR, "Duplicate bricks"
                                       " found %s", words[brick_index]);
                                cli_out("Duplicate bricks found %s",
                                        words[brick_index]);
                                ret = -1;
                                goto out;
                        }
                        j++;
                }
                snprintf (key, 50, "brick%d", ++brick_count);
                ret = dict_set_str (dict, key, (char *)words[brick_index++]);

                if (ret)
                        goto out;
        }

        ret = dict_set_int32 (dict, "count", brick_count);

        if (ret)
                goto out;

        *options = dict;

out:
        if (ret) {
                gf_log ("cli", GF_LOG_ERROR, "Unable to parse remove-brick CLI");
                if (dict)
                        dict_destroy (dict);
        }

        if (tmp_brick)
                GF_FREE (tmp_brick);
        if (tmp_brick1)
                GF_FREE (tmp_brick1);

        return ret;
}


int32_t
cli_cmd_volume_replace_brick_parse (const char **words, int wordcount,
                                   dict_t **options)
{
        dict_t  *dict = NULL;
        char    *volname = NULL;
        int     ret = -1;
        char    *op = NULL;
        int     op_index = 0;
        char    *delimiter = NULL;
        gf1_cli_replace_op replace_op = GF_REPLACE_OP_NONE;

        GF_ASSERT (words);
        GF_ASSERT (options);

        GF_ASSERT ((strcmp (words[0], "volume")) == 0);
        GF_ASSERT ((strcmp (words[1], "replace-brick")) == 0);

        dict = dict_new ();

        if (!dict)
                goto out;

        if (wordcount < 3)
                goto out;

        volname = (char *)words[2];

        GF_ASSERT (volname);

        ret = dict_set_str (dict, "volname", volname);

        if (ret)
                goto out;

        if (wordcount < 4) {
                ret = -1;
                goto out;
        }

        delimiter = strchr ((char *)words[3], ':');
        if (!delimiter || delimiter == words[3]
            || *(delimiter+1) != '/') {
                cli_out ("wrong brick type: %s, use "
                        "<HOSTNAME>:<export-dir-abs-path>", words[3]);
                ret = -1;
                goto out;
        } else {
                cli_path_strip_trailing_slashes (delimiter + 1);
        }
        ret = dict_set_str (dict, "src-brick", (char *)words[3]);

        if (ret)
                goto out;

        if (wordcount < 5) {
                ret = -1;
                goto out;
        }

        delimiter = strchr ((char *)words[4], ':');
        if (!delimiter || delimiter == words[4]
            || *(delimiter+1) != '/') {
                cli_out ("wrong brick type: %s, use "
                        "<HOSTNAME>:<export-dir-abs-path>", words[4]);
                ret = -1;
                goto out;
        } else {
                cli_path_strip_trailing_slashes (delimiter + 1);
        }


        ret = dict_set_str (dict, "dst-brick", (char *)words[4]);

        if (ret)
                goto out;

        op_index = 5;
        if ((wordcount < (op_index + 1))) {
                ret = -1;
                goto out;
        }

        op = (char *) words[op_index];

        if (!strcasecmp ("start", op)) {
                replace_op = GF_REPLACE_OP_START;
        } else if (!strcasecmp ("commit", op)) {
                replace_op = GF_REPLACE_OP_COMMIT;
        } else if (!strcasecmp ("pause", op)) {
                replace_op = GF_REPLACE_OP_PAUSE;
        } else if (!strcasecmp ("abort", op)) {
                replace_op = GF_REPLACE_OP_ABORT;
        } else if (!strcasecmp ("status", op)) {
                replace_op = GF_REPLACE_OP_STATUS;
        }

        /* commit force option */
        op_index = 6;

        if (wordcount > (op_index + 1)) {
                ret = -1;
                goto out;
        }

        if (wordcount == (op_index + 1)) {
                op = (char *) words[op_index];
                if (!strcasecmp ("force", op)) {
                        replace_op = GF_REPLACE_OP_COMMIT_FORCE;
                }
        }

        if (replace_op == GF_REPLACE_OP_NONE) {
                ret = -1;
                goto out;
        }

        ret = dict_set_int32 (dict, "operation", (int32_t) replace_op);

        if (ret)
                goto out;




        *options = dict;

out:
        if (ret) {
                gf_log ("cli", GF_LOG_ERROR, "Unable to parse remove-brick CLI");
                if (dict)
                        dict_destroy (dict);
        }

        return ret;
}

int32_t
cli_cmd_log_filename_parse (const char **words, int wordcount, dict_t **options)
{
        dict_t  *dict = NULL;
        char    *volname = NULL;
        char    *str = NULL;
        int     ret = -1;
        char    *delimiter = NULL;

        GF_ASSERT (words);
        GF_ASSERT (options);

        GF_ASSERT ((strcmp (words[0], "volume")) == 0);
        GF_ASSERT ((strcmp (words[1], "log")) == 0);
        GF_ASSERT ((strcmp (words[2], "filename")) == 0);

        dict = dict_new ();
        if (!dict)
                goto out;

        volname = (char *)words[3];
        GF_ASSERT (volname);

        ret = dict_set_str (dict, "volname", volname);
        if (ret)
                goto out;

        str = (char *)words[4];
        if (strchr (str, ':')) {
                delimiter = strchr (words[4], ':');
                if (!delimiter || delimiter == words[4]
                    || *(delimiter+1) != '/') {
                        cli_out ("wrong brick type: %s, use <HOSTNAME>:"
                                 "<export-dir-abs-path>", words[4]);
                        ret = -1;
                        goto out;
                } else {
                        cli_path_strip_trailing_slashes (delimiter + 1);
                }
                ret = dict_set_str (dict, "brick", str);
                if (ret)
                        goto out;
                /* Path */
                str = (char *)words[5];
                ret = dict_set_str (dict, "path", str);
                if (ret)
                        goto out;
        } else {
                ret = dict_set_str (dict, "path", str);
                if (ret)
                        goto out;
        }

        *options = dict;

out:
        if (ret && dict)
                dict_destroy (dict);

        return ret;
}

int32_t
cli_cmd_log_locate_parse (const char **words, int wordcount, dict_t **options)
{
        dict_t  *dict = NULL;
        char    *volname = NULL;
        char    *str = NULL;
        int     ret = -1;
        char    *delimiter = NULL;

        GF_ASSERT (words);
        GF_ASSERT (options);

        GF_ASSERT ((strcmp (words[0], "volume")) == 0);
        GF_ASSERT ((strcmp (words[1], "log")) == 0);
        GF_ASSERT ((strcmp (words[2], "locate")) == 0);

        dict = dict_new ();
        if (!dict)
                goto out;

        volname = (char *)words[3];
        GF_ASSERT (volname);

        ret = dict_set_str (dict, "volname", volname);
        if (ret)
                goto out;

        if (words[4]) {
                delimiter = strchr (words[4], ':');
                if (!delimiter || delimiter == words[4]
                    || *(delimiter+1) != '/') {
                        cli_out ("wrong brick type: %s, use <HOSTNAME>:"
                                 "<export-dir-abs-path>", words[4]);
                        ret = -1;
                        goto out;
                } else {
                        cli_path_strip_trailing_slashes (delimiter + 1);
                }
                str = (char *)words[4];
                ret = dict_set_str (dict, "brick", str);
                if (ret)
                        goto out;
        }

        *options = dict;

out:
        if (ret && dict)
                dict_destroy (dict);

        return ret;
}

int32_t
cli_cmd_log_rotate_parse (const char **words, int wordcount, dict_t **options)
{
        dict_t  *dict = NULL;
        char    *volname = NULL;
        char    *str = NULL;
        int     ret = -1;
        char    *delimiter = NULL;

        GF_ASSERT (words);
        GF_ASSERT (options);

        GF_ASSERT ((strcmp (words[0], "volume")) == 0);
        GF_ASSERT ((strcmp (words[1], "log")) == 0);
        GF_ASSERT ((strcmp (words[2], "rotate")) == 0);

        dict = dict_new ();
        if (!dict)
                goto out;

        volname = (char *)words[3];
        GF_ASSERT (volname);

        ret = dict_set_str (dict, "volname", volname);
        if (ret)
                goto out;

        if (words[4]) {
                delimiter = strchr (words[4], ':');
                if (!delimiter || delimiter == words[4]
                    || *(delimiter+1) != '/') {
                        cli_out ("wrong brick type: %s, use <HOSTNAME>:"
                                 "<export-dir-abs-path>", words[4]);
                        ret = -1;
                        goto out;
                } else {
                        cli_path_strip_trailing_slashes (delimiter + 1);
                }
                str = (char *)words[4];
                ret = dict_set_str (dict, "brick", str);
                if (ret)
                        goto out;
        }

        *options = dict;

out:
        if (ret && dict)
                dict_destroy (dict);

        return ret;
}

int32_t
cli_cmd_gsync_set_parse (const char **words, int wordcount, dict_t **options)
{
        int32_t            ret     = 0;
        int32_t            config_type = 0;
        dict_t            *dict   = NULL;
        gf1_cli_gsync_set type    = GF_GSYNC_OPTION_TYPE_NONE;

        GF_ASSERT (words);
        GF_ASSERT (options);

        GF_ASSERT ((strcmp (words[0], "volume")) == 0);
        GF_ASSERT ((strcmp (words[1], "gsync")) == 0);

        dict = dict_new ();
        if (!dict)
                goto out;

        if (wordcount < 5)
                goto out;

        ret = dict_set_str (dict, "master", (char *)words[3]);
        if (ret < 0)
                goto out;

        ret = dict_set_str (dict, "slave", (char *)words[4]);
        if (ret < 0)
                goto out;

        if ((strcmp (words[2], "start")) == 0) {
                type = GF_GSYNC_OPTION_TYPE_START;

                goto set_type;
        }

        if ((strcmp (words[2], "stop")) == 0) {
                type = GF_GSYNC_OPTION_TYPE_STOP;

                goto set_type;
        }

        if ((strcmp (words[2], "configure")) == 0) {
                type = GF_GSYNC_OPTION_TYPE_CONFIGURE;

                if (strcmp (words [5], "config-set") == 0) {
                        config_type = GF_GSYNC_OPTION_TYPE_CONFIG_SET;

                        ret = dict_set_str (dict, "op_name", (char *)words[6]);
                        if (ret < 0)
                                goto out;

                        ret = dict_set_str (dict, "op_value", (char *)words[7]);
                        if (ret < 0)
                                goto out;
                }

                if ((strcmp (words [5], "config-del")) == 0) {
                        config_type = GF_GSYNC_OPTION_TYPE_CONFIG_DEL;

                        ret = dict_set_str (dict, "op_name", (char *)words[6]);
                        if (ret < 0)
                                goto out;
                }

                if ((strcmp (words [5], "config-get")) == 0) {
                        config_type = GF_GSYNC_OPTION_TYPE_CONFIG_GET;

                        ret = dict_set_str (dict, "op_name", (char *)words[6]);
                        if (ret < 0)
                                goto out;
                }

                if ((strcmp (words [5], "config-get-all")) == 0) {
                        config_type = GF_GSYNC_OPTION_TYPE_CONFIG_GET_ALL;
                }

                ret = dict_set_int32 (dict, "config_type", config_type);
                if (ret < 0)
                        goto out;
        }

set_type:
        ret = dict_set_int32 (dict, "type", type);
        if (ret < 0)
                goto out;

        *options = dict;
out:
        if (ret)
                if (dict)
                        dict_destroy (dict);

        return ret;
}

int32_t
cli_cmd_volume_profile_parse (const char **words, int wordcount,
                              dict_t **options)
{
        dict_t    *dict       = NULL;
        char      *volname    = NULL;
        int       ret         = -1;
        gf1_cli_stats_op op = GF_CLI_STATS_NONE;

        GF_ASSERT (words);
        GF_ASSERT (options);

        GF_ASSERT ((strcmp (words[0], "volume")) == 0);
        GF_ASSERT ((strcmp (words[1], "profile")) == 0);

        dict = dict_new ();
        if (!dict)
                goto out;

        if (wordcount != 4)
                goto out;

        volname = (char *)words[2];

        ret = dict_set_str (dict, "volname", volname);
        if (ret)
                goto out;

        if (strcmp (words[3], "start") == 0) {
                op = GF_CLI_STATS_START;
        } else if (strcmp (words[3], "stop") == 0) {
                op = GF_CLI_STATS_STOP;
        } else if (strcmp (words[3], "info") == 0) {
                op = GF_CLI_STATS_INFO;
        } else {
                ret = -1;
                goto out;
        }
        ret = dict_set_int32 (dict, "op", (int32_t)op);
        *options = dict;
out:
        if (ret && dict)
                dict_destroy (dict);
        return ret;
}

int32_t
cli_cmd_volume_top_parse (const char **words, int wordcount,
                              dict_t **options)
{
        dict_t  *dict           = NULL;
        char    *volname        = NULL;
        char    *value          = NULL;
        char    *key            = NULL;
        int      ret            = -1;
        gf1_cli_stats_op op = GF_CLI_STATS_NONE;
        gf1_cli_top_op    top_op = GF_CLI_TOP_NONE;
        int32_t  list_cnt       = 0;
        int      index          = 0;
        int      perf           = 0;
        int32_t  blk_size       = 0;
        int32_t  count          = 0;
        char    *delimiter      = NULL;

        GF_ASSERT (words);
        GF_ASSERT (options);

        GF_ASSERT ((strcmp (words[0], "volume")) == 0);
        GF_ASSERT ((strcmp (words[1], "top")) == 0);

        dict = dict_new ();
        if (!dict)
                goto out;

        if (wordcount < 4)
                goto out;

        volname = (char *)words[2];

        ret = dict_set_str (dict, "volname", volname);
        if (ret)
                goto out;

        op = GF_CLI_STATS_TOP;
        ret = dict_set_int32 (dict, "op", (int32_t)op);
        if (ret)
                goto out;

        if (strcmp (words[3], "open") == 0) {
                top_op = GF_CLI_TOP_OPEN;
        } else if (strcmp (words[3], "read") == 0) {
                top_op = GF_CLI_TOP_READ;
        } else if (strcmp (words[3], "write") == 0) {
                top_op = GF_CLI_TOP_WRITE;
        } else if (strcmp (words[3], "opendir") == 0) {
                top_op = GF_CLI_TOP_OPENDIR;
        } else if (strcmp (words[3], "readdir") == 0) {
                top_op = GF_CLI_TOP_READDIR;
        } else if (strcmp (words[3], "read-perf") == 0) {
                top_op = GF_CLI_TOP_READ_PERF;
                perf = 1;
        } else if (strcmp (words[3], "write-perf") == 0) {
                top_op = GF_CLI_TOP_WRITE_PERF;
                perf = 1;
        } else {
                ret = -1;
                goto out;
        } 
        ret = dict_set_int32 (dict, "top-op", (int32_t)top_op);
        if (ret)
                goto out;

        for (index = 4; index < wordcount; index+=2) {

                key = (char *) words[index];
                value = (char *) words[index+1];

                if ( key && !value ) {
                        ret = -1;
                        goto out;
                }
                if (!strcmp (key, "brick")) {
                        delimiter = strchr (value, ':');
                        if (!delimiter || delimiter == value 
                            || *(delimiter+1) != '/') {
                                cli_out ("wrong brick type: %s, use <HOSTNAME>:"
                                         "<export-dir-abs-path>", value);
                                ret = -1;
                                goto out;
                        } else {
                                cli_path_strip_trailing_slashes (delimiter + 1);
                        }
                        ret = dict_set_str (dict, "brick", value);

                } else if (!strcmp (key, "list-cnt")) {
                        list_cnt = atoi(value);
                        if (!list_cnt) {
                                list_cnt = 100;
                        }
                        if (list_cnt < 1 || list_cnt > 100) {
                                cli_out ("list-cnt should be between 1 to 100");
                                ret = -1;
                                goto out;
                        }
                        ret = dict_set_int32 (dict, "list-cnt", list_cnt);
                } else if (perf && !strcmp (key, "bs")){
                        blk_size = atoi (value);
                        if (blk_size < 0){
                                cli_out ("block size should be an integer "
                                         "greater than zero");
                                ret = -1;
                                goto out;
                        }
                        ret = dict_set_int32 (dict, "blk-size", blk_size);
                } else if (perf && !strcmp (key, "count")) {
                        count = atoi(value);
                        if (count < 0 ){
                                cli_out ("count should be an integer greater "
                                         "zero");
                                ret = -1;
                                goto out;
                        }
                        ret = dict_set_int32 (dict, "blk-cnt", count);
                } else {
                        ret = -1;
                        goto out;
                }
                if (ret) {
                        gf_log ("", GF_LOG_WARNING, "Dict set failed for "
                                "key %s", key);
                        goto out;
                }
        }
        *options = dict;
out:
        if (ret && dict)
                dict_destroy (dict);
        return ret;
}